1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591 1592 1593 1594 1595 1596 1597 1598 1599 1600 1601 1602 1603 1604 1605 1606 1607 1608 1609 1610 1611 1612 1613 1614 1615 1616 1617 1618 1619 1620 1621 1622 1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646 1647 1648 1649 1650 1651 1652 1653 1654 1655 1656 1657 1658 1659 1660 1661 1662 1663 1664 1665 1666 1667 1668 1669 1670 1671 1672 1673 1674 1675 1676 1677 1678 1679 1680 1681 1682 1683 1684 1685 1686 1687 1688 1689 1690 1691 1692 1693 1694 1695 1696 1697 1698 1699 1700 1701 1702 1703 1704 1705 1706 1707 1708 1709 1710 1711 1712 1713 1714 1715 1716 1717 1718 1719 1720 1721 1722 1723 1724 1725 1726 1727 1728 1729 1730 1731 1732 1733 1734 1735 1736 1737 1738 1739 1740 1741 1742 1743 1744 1745 1746 1747 1748 1749 1750 1751 1752 1753 1754 1755 1756 1757 1758 1759 1760 1761 1762 1763 1764 1765 1766 1767 1768 1769 1770 1771 1772 1773 1774 1775 1776 1777 1778 1779 1780 1781 1782 1783 1784 1785 1786 1787 1788 1789 1790 1791 1792 1793 1794 1795 1796 1797 1798 1799 1800 1801 1802 1803 1804 1805 1806 1807 1808 1809 1810 1811 1812 1813 1814 1815 1816 1817 1818 1819 1820 1821 1822 1823 1824 1825 1826 1827 1828 1829 1830 1831 1832 1833 1834 1835 1836 1837 1838 1839 1840 1841 1842 1843 1844 1845 1846 1847 1848 1849 1850 1851 1852 1853 1854 1855 1856 1857 1858 1859 1860 1861 1862 1863 1864 1865 1866 1867 1868 1869 1870 1871 1872 1873 1874 1875 1876 1877 1878 1879 1880 1881 1882 1883 1884 1885 1886 1887 1888 1889 1890 1891 1892 1893 1894 1895 1896 1897 1898 1899 1900 1901 1902 1903 1904 1905 1906 1907 1908 1909 1910 1911 1912 1913 1914 1915 1916 1917 1918 1919 1920 1921 1922 1923 1924 1925 1926 1927 1928 1929 1930 1931 1932 1933 1934 1935 1936 1937 1938 1939 1940 1941 1942 1943 1944 1945 1946 1947 1948 1949 1950 1951 1952 1953 1954 1955 1956 1957 1958 1959 1960 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023 2024 2025 2026 2027 2028 2029 2030 2031 2032 2033 2034 2035 2036 2037 2038 2039 2040 2041 2042 2043 2044 2045 2046 2047 2048 2049 2050 2051 2052 2053 2054 2055 2056 2057 2058 2059 2060 2061 2062 2063 2064 2065 2066 2067 2068 2069 2070 2071 2072 2073 2074 2075 2076 2077 2078 2079 2080 2081 2082 2083 2084 2085 2086 2087 2088 2089 2090 2091 2092 2093 2094 2095 2096 2097 2098 2099 2100 2101 2102 2103 2104 2105 2106 2107 2108 2109 2110 2111 2112 2113 2114 2115 2116 2117 2118 2119 2120 2121 2122 2123 2124 2125 2126 2127 2128 2129 2130 2131 2132 2133 2134 2135 2136 2137 2138 2139 2140 2141 2142 2143 2144 2145 2146 2147 2148 2149 2150 2151 2152 2153 2154 2155 2156 2157 2158 2159 2160 2161 2162 2163 2164 2165 2166 2167 2168 2169 2170 2171 2172 2173 2174 2175 2176 2177 2178 2179 2180 2181 2182 2183 2184 2185 2186 2187 2188 2189 2190 2191 2192 2193 2194 2195 2196 2197 2198 2199 2200 2201 2202 2203 2204 2205 2206 2207 2208 2209 2210 2211 2212 2213 2214 2215 2216 2217 2218 2219 2220 2221 2222 2223 2224 2225 2226 2227 2228 2229 2230 2231 2232 2233 2234 2235 2236 2237 2238 2239 2240 2241 2242 2243 2244 2245 2246 2247 2248 2249 2250 2251 2252 2253 2254 2255 2256 2257 2258 2259 2260 2261 2262 2263 2264 2265 2266 2267 2268 2269 2270 2271 2272 2273 2274 2275 2276 2277 2278 2279 2280 2281 2282 2283 2284 2285 2286 2287 2288 2289 2290 2291 2292 2293 2294 2295 2296 2297 2298 2299 2300 2301 2302 2303 2304 2305 2306 2307 2308 2309 2310 2311 2312 2313 2314 2315 2316 2317 2318 2319 2320 2321 2322 2323 2324 2325 2326 2327 2328 2329 2330 2331 2332 2333 2334 2335 2336 2337 2338 2339 2340 2341 2342 2343 2344 2345 2346 2347 2348 2349 2350 2351 2352 2353 2354 2355 2356 2357 2358 2359 2360 2361 2362 2363 2364 2365 2366 2367 2368 2369 2370 2371 2372 2373 2374 2375 2376 2377 2378 2379 2380 2381 2382 2383 2384 2385 2386 2387 2388 2389 2390 2391 2392 2393 2394 2395 2396 2397 2398 2399 2400 2401 2402 2403 2404 2405 2406 2407 2408 2409 2410 2411 2412 2413 2414 2415 2416 2417 2418 2419 2420 2421 2422 2423 2424 2425 2426 2427 2428 2429 2430 2431 2432 2433 2434 2435 2436 2437 2438 2439 2440 2441 2442 2443 2444 2445 2446 2447 2448 2449 2450 2451 2452 2453 2454 2455 2456 2457 2458 2459 2460 2461 2462 2463 2464 2465 2466 2467 2468 2469 2470 2471 2472 2473 2474 2475 2476 2477 2478 2479 2480 2481 2482 2483 2484 2485 2486 2487 2488 2489 2490 2491 2492 2493 2494 2495 2496 2497 2498 2499 2500 2501 2502 2503 2504 2505 2506 2507 2508 2509 2510 2511 2512 2513 2514 2515 2516 2517 2518 2519 2520 2521 2522 2523 2524 2525 2526 2527 2528 2529 2530 2531 2532 2533 2534 2535 2536 2537 2538 2539 2540 2541 2542 2543 2544 2545 2546 2547 2548 2549 2550 2551 2552 2553 2554 2555 2556 2557 2558 2559 2560 2561 2562 2563 2564 2565 2566 2567 2568 2569 2570 2571 2572 2573 2574 2575 2576 2577 2578 2579 2580 2581 2582 2583 2584 2585 2586 2587 2588 2589 2590 2591 2592 2593 2594 2595 2596 2597 2598 2599 2600 2601 2602 2603 2604 2605 2606 2607 2608 2609 2610 2611 2612 2613 2614 2615 2616 2617 2618 2619 2620 2621 2622 2623 2624 2625 2626 2627 2628 2629 2630 2631 2632 2633 2634 2635 2636 2637 2638 2639 2640 2641 2642 2643 2644 2645 2646 2647 2648 2649 2650 2651 2652 2653 2654 2655 2656 2657 2658 2659 2660 2661 2662 2663 2664 2665 2666 2667 2668 2669 2670 2671 2672 2673 2674 2675 2676 2677 2678 2679 2680 2681 2682 2683 2684 2685 2686 2687 2688 2689 2690 2691 2692 2693 2694 2695 2696 2697 2698 2699 2700 2701 2702 2703 2704 2705 2706 2707 2708 2709 2710 2711 2712 2713 2714 2715 2716 2717 2718 2719 2720 2721 2722 2723 2724 2725 2726 2727 2728 2729 2730 2731 2732 2733 2734 2735 2736 2737 2738 2739 2740 2741 2742 2743 2744 2745 2746 2747 2748 2749 2750 2751 2752 2753 2754 2755 2756 2757 2758 2759 2760 2761 2762 2763 2764 2765 2766 2767 2768 2769 2770 2771 2772 2773 2774 2775 2776 2777 2778 2779 2780 2781 2782 2783 2784 2785 2786 2787 2788 2789 2790 2791 2792 2793 2794 2795 2796 2797 2798 2799 2800 2801 2802 2803 2804 2805 2806 2807 2808 2809 2810 2811 2812 2813 2814 2815 2816 2817 2818 2819 2820 2821 2822 2823 2824 2825 2826 2827 2828 2829 2830 2831 2832 2833 2834 2835 2836 2837 2838 2839 2840 2841 2842 2843 2844 2845 2846 2847 2848 2849 2850 2851 2852 2853 2854 2855 2856 2857 2858 2859 2860 2861 2862 2863 2864 2865 2866 2867 2868 2869 2870 2871 2872 2873 2874 2875 2876 2877 2878 2879 2880 2881 2882 2883 2884 2885 2886 2887 2888 2889 2890 2891 2892 2893 2894 2895 2896 2897 2898 2899 2900 2901 2902 2903 2904 2905 2906 2907 2908 2909 2910 2911 2912 2913 2914 2915 2916 2917 2918 2919 2920 2921 2922 2923 2924 2925 2926 2927 2928 2929 2930 2931 2932 2933 2934 2935 2936 2937 2938 2939 2940 2941 2942 2943 2944 2945 2946 2947 2948 2949 2950 2951 2952 2953 2954 2955 2956 2957 2958 2959 2960 2961 2962 2963 2964 2965 2966 2967 2968 2969 2970 2971 2972 2973 2974 2975 2976 2977 2978 2979 2980 2981 2982 2983 2984 2985 2986 2987 2988 2989 2990 2991 2992 2993 2994 2995 2996 2997 2998 2999 3000 3001 3002 3003 3004 3005 3006 3007 3008 3009 3010 3011 3012 3013 3014 3015 3016 3017 3018 3019 3020 3021 3022 3023 3024 3025 3026 3027 3028 3029 3030 3031 3032 3033 3034 3035 3036 3037 3038 3039 3040 3041 3042 3043 3044 3045 3046 3047 3048 3049 3050 3051 3052 3053 3054 3055 3056 3057 3058 3059 3060 3061 3062 3063 3064 3065 3066 3067 3068 3069 3070 3071 3072 3073 3074 3075 3076 3077 3078 3079 3080 3081 3082 3083 3084 3085 3086 3087 3088 3089 3090 3091 3092 3093 3094 3095 3096 3097 3098 3099 3100 3101 3102 3103 3104 3105 3106 3107 3108 3109 3110 3111 3112 3113 3114 3115 3116 3117 3118 3119 3120 3121 3122 3123 3124 3125 3126 3127 3128 3129 3130 3131 3132 3133 3134 3135 3136 3137 3138 3139 3140 3141 3142 3143 3144 3145 3146 3147 3148 3149 3150 3151 3152 3153 3154 3155 3156 3157 3158 3159 3160 3161 3162 3163 3164 3165 3166 3167 3168 3169 3170 3171 3172 3173 3174 3175 3176 3177 3178 3179 3180 3181 3182 3183 3184 3185 3186 3187 3188 3189 3190 3191 3192 3193 3194 3195 3196 3197 3198 3199 3200 3201 3202 3203 3204 3205 3206 3207 3208 3209 3210 3211 3212 3213 3214 3215 3216 3217 3218 3219 3220 3221 3222 3223 3224 3225 3226 3227 3228 3229 3230 3231 3232 3233 3234 3235 3236 3237 3238 3239 3240 3241 3242 3243 3244 3245 3246 3247 3248 3249 3250 3251 3252 3253 3254 3255 3256 3257 3258 3259 3260 3261 3262 3263 3264 3265 3266 3267 3268 3269 3270 3271 3272 3273 3274 3275 3276 3277 3278 3279 3280 3281 3282 3283 3284 3285 3286 3287 3288 3289 3290 3291 3292 3293 3294 3295 3296 3297 3298 3299 3300 3301 3302 3303 3304 3305 3306 3307 3308 3309 3310 3311 3312 3313 3314 3315 3316 3317 3318 3319 3320 3321 3322 3323 3324 3325 3326 3327 3328 3329 3330 3331 3332 3333 3334 3335 3336 3337 3338 3339 3340 3341 3342 3343 3344 3345 3346 3347 3348 3349 3350 3351 3352 3353 3354 3355 3356 3357 3358 3359 3360 3361 3362 3363 3364 3365 3366 3367 3368 3369 3370 3371 3372 3373 3374 3375 3376 3377 3378 3379 3380 3381 3382 3383 3384 3385 3386 3387 3388 3389 3390 3391 3392 3393 3394 3395 3396 3397 3398 3399 3400 3401 3402 3403 3404 3405 3406 3407 3408 3409 3410 3411 3412 3413 3414 3415 3416 3417 3418 3419 3420 3421 3422 3423 3424 3425 3426 3427 3428 3429 3430 3431 3432 3433 3434 3435 3436 3437 3438 3439 3440 3441 3442 3443 3444 3445 3446 3447 3448 3449 3450 3451 3452 3453 3454 3455 3456 3457 3458 3459 3460 3461 3462 3463 3464 3465 3466 3467 3468 3469 3470 3471 3472 3473 3474 3475 3476 3477 3478 3479 3480 3481 3482 3483 3484 3485 3486 3487 3488 3489 3490 3491 3492 3493 3494 3495 3496 3497 3498 3499 3500 3501 3502 3503 3504 3505 3506 3507 3508 3509 3510 3511 3512 3513 3514 3515 3516 3517 3518 3519 3520 3521 3522 3523 3524 3525 3526 3527 3528 3529 3530 3531 3532 3533 3534 3535 3536 3537 3538 3539 3540 3541 3542 3543 3544 3545 3546 3547 3548 3549 3550 3551 3552 3553 3554 3555 3556 3557 3558 3559 3560 3561 3562 3563 3564 3565 3566 3567 3568 3569 3570 3571 3572 3573 3574 3575 3576 3577 3578 3579 3580 3581 3582 3583 3584 3585 3586 3587 3588 3589 3590 3591 3592 3593 3594 3595 3596 3597 3598 3599 3600 3601 3602 3603 3604 3605 3606 3607 3608 3609 3610 3611 3612 3613 3614 3615 3616 3617 3618 3619 3620 3621 3622 3623 3624 3625 3626 3627 3628 3629 3630 3631 3632 3633 3634 3635 3636 3637 3638 3639 3640 3641 3642 3643 3644 3645 3646 3647 3648 3649 3650 3651 3652 3653 3654 3655 3656 3657 3658 3659 3660 3661 3662 3663 3664 3665 3666 3667 3668 3669 3670 3671 3672 3673 3674 3675 3676 3677 3678 3679 3680 3681 3682 3683 3684 3685 3686 3687 3688 3689 3690 3691 3692 3693 3694 3695 3696 3697 3698 3699 3700 3701 3702 3703 3704 3705 3706 3707 3708 3709 3710 3711 3712 3713 3714 3715 3716 3717 3718 3719 3720 3721 3722 3723 3724 3725 3726 3727 3728 3729 3730 3731 3732 3733 3734 3735 3736 3737 3738 3739 3740 3741 3742 3743 3744 3745 3746 3747 3748 3749 3750 3751 3752 3753 3754 3755 3756 3757 3758 3759 3760 3761 3762 3763 3764 3765 3766 3767 3768 3769 3770 3771 3772 3773 3774 3775 3776 3777 3778 3779 3780 3781 3782 3783 3784 3785 3786 3787 3788 3789 3790 3791 3792 3793 3794 3795 3796 3797 3798 3799 3800 3801 3802 3803 3804 3805 3806 3807 3808 3809 3810 3811 3812 3813 3814 3815 3816 3817 3818 3819 3820 3821 3822 3823 3824 3825 3826 3827 3828 3829 3830 3831 3832 3833 3834 3835 3836 3837 3838 3839 3840 3841 3842 3843 3844 3845 3846 3847 3848 3849 3850 3851 3852 3853 3854 3855 3856 3857 3858 3859 3860 3861 3862 3863 3864 3865 3866 3867 3868 3869 3870 3871 3872 3873 3874 3875 3876 3877 3878 3879 3880 3881 3882 3883 3884 3885 3886 3887 3888 3889 3890 3891 3892 3893 3894 3895 3896 3897 3898 3899 3900 3901 3902 3903 3904 3905 3906 3907 3908 3909 3910 3911 3912 3913 3914 3915 3916 3917 3918 3919 3920 3921 3922 3923 3924 3925 3926 3927 3928 3929 3930 3931 3932 3933 3934 3935 3936 3937 3938 3939 3940 3941 3942 3943 3944 3945 3946 3947 3948 3949 3950 3951 3952 3953 3954 3955 3956 3957 3958 3959 3960 3961 3962 3963 3964 3965 3966 3967 3968 3969 3970 3971 3972 3973 3974 3975 3976 3977 3978 3979 3980 3981 3982 3983 3984 3985 3986 3987 3988 3989 3990 3991 3992 3993 3994 3995 3996 3997 3998 3999 4000 4001 4002 4003 4004 4005 4006 4007 4008 4009 4010 4011 4012 4013 4014 4015 4016 4017 4018 4019 4020 4021 4022 4023 4024 4025 4026 4027 4028 4029 4030 4031 4032 4033 4034 4035 4036 4037 4038 4039 4040 4041 4042 4043 4044 4045 4046 4047 4048 4049 4050 4051 4052 4053 4054 4055 4056 4057 4058 4059 4060 4061 4062 4063 4064 4065 4066 4067 4068 4069 4070 4071 4072 4073 4074 4075 4076 4077 4078 4079 4080 4081 4082 4083 4084 4085 4086 4087 4088 4089 4090 4091 4092 4093 4094 4095 4096 4097 4098 4099 4100 4101 4102 4103 4104 4105 4106 4107 4108 4109 4110 4111 4112 4113 4114 4115 4116 4117 4118 4119 4120 4121 4122 4123 4124 4125 4126 4127 4128 4129 4130 4131 4132 4133 4134 4135 4136 4137 4138 4139 4140 4141 4142 4143 4144 4145 4146 4147 4148 4149 4150 4151 4152 4153 4154 4155 4156 4157 4158 4159 4160 4161 4162 4163 4164 4165 4166 4167 4168 4169 4170 4171 4172 4173 4174 4175 4176 4177 4178 4179 4180 4181 4182 4183 4184 4185 4186 4187 4188 4189 4190 4191 4192 4193 4194 4195 4196 4197 4198 4199 4200 4201 4202 4203 4204 4205 4206 4207 4208 4209 4210 4211 4212 4213 4214 4215 4216 4217 4218 4219 4220 4221 4222 4223 4224 4225 4226 4227 4228 4229 4230 4231 4232 4233 4234 4235 4236 4237 4238 4239 4240 4241 4242 4243 4244 4245 4246 4247 4248 4249 4250 4251 4252 4253 4254 4255 4256 4257 4258 4259 4260 4261 4262 4263 4264 4265 4266 4267 4268 4269 4270 4271 4272 4273 4274 4275 4276 4277 4278 4279 4280 4281 4282 4283 4284 4285 4286 4287 4288 4289 4290 4291 4292 4293 4294 4295 4296 4297 4298 4299 4300 4301 4302 4303 4304 4305 4306 4307 4308 4309 4310 4311 4312 4313 4314 4315 4316 4317 4318 4319 4320 4321 4322 4323 4324 4325 4326 4327 4328 4329 4330 4331 4332 4333 4334 4335 4336 4337 4338 4339 4340 4341 4342 4343 4344 4345 4346 4347 4348 4349 4350 4351 4352 4353 4354 4355 4356 4357 4358 4359 4360 4361 4362 4363 4364 4365 4366 4367 4368 4369 4370 4371 4372 4373 4374 4375 4376 4377 4378 4379 4380 4381 4382 4383 4384 4385 4386 4387 4388 4389 4390 4391 4392 4393 4394 4395 4396 4397 4398 4399 4400 4401 4402 4403 4404 4405 4406 4407 4408 4409 4410 4411 4412 4413 4414 4415 4416 4417 4418 4419 4420 4421 4422 4423 4424 4425 4426 4427 4428 4429 4430 4431 4432 4433 4434 4435 4436 4437 4438 4439 4440 4441 4442 4443 4444 4445 4446 4447 4448 4449 4450 4451 4452 4453 4454 4455 4456 4457 4458 4459 4460 4461 4462 4463 4464 4465 4466 4467 4468 4469 4470 4471 4472 4473 4474 4475 4476 4477 4478 4479 4480 4481 4482 4483 4484 4485 4486 4487 4488 4489 4490 4491 4492 4493 4494 4495 4496 4497 4498 4499 4500 4501 4502 4503 4504 4505 4506 4507 4508 4509 4510 4511 4512 4513 4514 4515 4516 4517 4518 4519 4520 4521 4522 4523 4524 4525 4526 4527 4528 4529 4530 4531 4532 4533 4534 4535 4536 4537 4538 4539 4540 4541 4542 4543 4544 4545 4546 4547 4548 4549 4550 4551 4552 4553 4554 4555 4556 4557 4558 4559 4560 4561 4562 4563 4564 4565 4566 4567 4568 4569 4570 4571 4572 4573 4574 4575 4576 4577 4578 4579 4580 4581 4582 4583 4584 4585 4586 4587 4588 4589 4590 4591 4592 4593 4594 4595 4596 4597 4598 4599 4600 4601 4602 4603 4604 4605 4606 4607 4608 4609 4610 4611 4612 4613 4614 4615 4616 4617 4618 4619 4620 4621 4622 4623 4624 4625 4626 4627 4628 4629 4630 4631 4632 4633 4634 4635 4636 4637 4638 4639 4640 4641 4642 4643 4644 4645 4646 4647 4648 4649 4650 4651 4652 4653 4654 4655 4656 4657 4658 4659 4660 4661 4662 4663 4664 4665 4666 4667 4668 4669 4670 4671 4672 4673 4674 4675 4676 4677 4678 4679 4680 4681 4682 4683 4684 4685 4686 4687 4688 4689 4690 4691 4692 4693 4694 4695 4696 4697 4698 4699 4700 4701 4702 4703 4704 4705 4706 4707 4708 4709 4710 4711 4712 4713 4714 4715 4716 4717 4718 4719 4720 4721 4722 4723 4724 4725 4726 4727 4728 4729 4730 4731 4732 4733 4734 4735 4736 4737 4738 4739 4740 4741 4742 4743 4744 4745 4746 4747 4748 4749 4750 4751 4752 4753 4754 4755 4756 4757 4758 4759 4760 4761 4762 4763 4764 4765 4766 4767 4768 4769 4770 4771 4772 4773 4774 4775 4776 4777 4778 4779 4780 4781 4782 4783 4784 4785 4786 4787 4788 4789 4790 4791 4792 4793 4794 4795 4796 4797 4798 4799 4800 4801 4802 4803 4804 4805 4806 4807 4808 4809 4810 4811 4812 4813 4814 4815 4816 4817 4818 4819 4820 4821 4822 4823 4824 4825 4826 4827 4828 4829 4830 4831 4832 4833 4834 4835 4836 4837 4838 4839 4840 4841 4842 4843 4844 4845 4846 4847 4848 4849 4850 4851 4852 4853 4854 4855 4856 4857 4858 4859 4860 4861 4862 4863 4864 4865 4866 4867 4868 4869 4870 4871 4872 4873 4874 4875 4876 4877 4878 4879 4880 4881 4882 4883 4884 4885 4886 4887 4888 4889 4890 4891 4892 4893 4894 4895 4896 4897 4898 4899 4900 4901 4902 4903 4904 4905 4906 4907 4908 4909 4910 4911 4912 4913 4914 4915 4916 4917 4918 4919 4920 4921 4922 4923 4924 4925 4926 4927 4928 4929 4930 4931 4932 4933 4934 4935 4936 4937 4938 4939 4940 4941 4942 4943 4944 4945 4946 4947 4948 4949 4950 4951 4952 4953 4954 4955 4956 4957 4958 4959 4960 4961 4962 4963 4964 4965 4966 4967 4968 4969 4970 4971 4972 4973 4974 4975 4976 4977 4978 4979 4980 4981 4982 4983 4984 4985 4986 4987 4988 4989 4990 4991 4992 4993 4994 4995 4996 4997 4998 4999 5000 5001 5002 5003 5004 5005 5006 5007 5008 5009 5010 5011 5012 5013 5014 5015 5016 5017 5018 5019 5020 5021 5022 5023 5024 5025 5026 5027 5028 5029 5030 5031 5032 5033 5034 5035 5036 5037 5038 5039 5040 5041 5042 5043 5044 5045 5046 5047 5048 5049 5050 5051 5052 5053 5054 5055 5056 5057 5058 5059 5060 5061 5062 5063 5064 5065 5066 5067 5068 5069 5070 5071 5072 5073 5074 5075 5076 5077 5078 5079 5080 5081 5082 5083 5084 5085 5086 5087 5088 5089 5090 5091 5092 5093 5094 5095 5096 5097 5098 5099 5100 5101 5102 5103 5104 5105 5106 5107 5108 5109 5110 5111 5112 5113 5114 5115 5116 5117 5118 5119 5120 5121 5122 5123 5124 5125 5126 5127 5128 5129 5130 5131 5132 5133 5134 5135 5136 5137 5138 5139 5140 5141 5142 5143 5144 5145 5146 5147 5148 5149 5150 5151 5152 5153 5154 5155 5156 5157 5158 5159 5160 5161 5162 5163 5164 5165 5166 5167 5168 5169 5170 5171 5172 5173 5174 5175 5176 5177 5178 5179 5180 5181 5182 5183 5184 5185 5186 5187 5188 5189 5190 5191 5192 5193 5194 5195 5196 5197 5198 5199 5200 5201 5202 5203 5204 5205 5206 5207 5208 5209 5210 5211 5212 5213 5214 5215 5216 5217 5218 5219 5220 5221 5222 5223 5224 5225 5226 5227 5228 5229 5230 5231 5232 5233 5234 5235 5236 5237 5238 5239 5240 5241 5242 5243 5244 5245 5246 5247 5248 5249 5250 5251 5252 5253 5254 5255 5256 5257 5258 5259 5260 5261 5262 5263 5264 5265 5266 5267 5268 5269 5270 5271 5272 5273 5274 5275 5276 5277 5278 5279 5280 5281 5282 5283 5284 5285 5286 5287 5288 5289 5290 5291 5292 5293 5294 5295 5296 5297 5298 5299 5300 5301 5302 5303 5304 5305 5306 5307 5308 5309 5310 5311 5312 5313 5314 5315 5316 5317 5318 5319 5320 5321 5322 5323 5324 5325 5326 5327 5328 5329 5330 5331 5332 5333 5334 5335 5336 5337 5338 5339 5340 5341 5342 5343 5344 5345 5346 5347 5348 5349 5350 5351 5352 5353 5354 5355 5356 5357 5358 5359 5360 5361 5362 5363 5364 5365 5366 5367 5368 5369 5370 5371 5372 5373 5374 5375 5376 5377 5378 5379 5380 5381 5382 5383 5384 5385 5386 5387 5388 5389 5390 5391 5392 5393 5394 5395 5396 5397 5398 5399 5400 5401 5402 5403 5404 5405 5406 5407 5408 5409 5410 5411 5412 5413 5414 5415 5416 5417 5418 5419 5420 5421 5422 5423 5424 5425 5426 5427 5428 5429 5430 5431 5432 5433 5434 5435 5436 5437 5438 5439 5440 5441 5442 5443 5444 5445 5446 5447 5448 5449 5450 5451 5452 5453 5454 5455 5456 5457 5458 5459 5460 5461 5462 5463 5464 5465 5466 5467 5468 5469 5470 5471 5472 5473 5474 5475 5476 5477 5478 5479 5480 5481 5482 5483 5484 5485 5486 5487 5488 5489 5490 5491 5492 5493 5494 5495 5496 5497 5498 5499 5500 5501 5502 5503 5504 5505 5506 5507 5508 5509 5510 5511 5512 5513 5514 5515 5516 5517 5518 5519 5520 5521 5522 5523 5524 5525 5526 5527 5528 5529 5530 5531 5532 5533 5534 5535 5536 5537 5538 5539 5540 5541 5542 5543 5544 5545 5546 5547 5548 5549 5550 5551 5552 5553 5554 5555 5556 5557 5558 5559 5560 5561 5562 5563 5564 5565 5566 5567 5568 5569 5570 5571 5572 5573 5574 5575 5576 5577 5578 5579 5580 5581 5582 5583 5584 5585 5586 5587 5588 5589 5590 5591 5592 5593 5594 5595 5596 5597 5598 5599 5600 5601 5602 5603 5604 5605 5606 5607 5608 5609 5610 5611 5612 5613 5614 5615 5616 5617 5618 5619 5620 5621 5622 5623 5624 5625 5626 5627 5628 5629 5630 5631 5632 5633 5634 5635 5636 5637 5638 5639 5640 5641 5642 5643 5644 5645 5646 5647 5648 5649 5650 5651 5652 5653 5654 5655 5656 5657 5658 5659 5660 5661 5662 5663 5664 5665 5666 5667 5668 5669 5670 5671 5672 5673 5674 5675 5676 5677 5678 5679 5680 5681 5682 5683 5684 5685 5686 5687 5688 5689 5690 5691 5692 5693 5694 5695 5696 5697 5698 5699 5700 5701 5702 5703 5704 5705 5706 5707 5708 5709 5710 5711 5712 5713 5714 5715 5716 5717 5718 5719 5720 5721 5722 5723 5724 5725 5726 5727 5728 5729 5730 5731 5732 5733 5734 5735 5736 5737 5738 5739 5740 5741 5742 5743 5744 5745 5746 5747 5748 5749 5750 5751 5752 5753 5754 5755 5756 5757 5758 5759 5760 5761 5762 5763 5764 5765 5766 5767 5768 5769 5770 5771 5772 5773 5774 5775 5776 5777 5778 5779 5780 5781 5782 5783 5784 5785 5786 5787 5788 5789 5790 5791 5792 5793 5794 5795 5796 5797 5798 5799 5800 5801 5802 5803 5804 5805 5806 5807 5808 5809 5810 5811 5812 5813 5814 5815 5816 5817 5818 5819 5820 5821 5822 5823 5824 5825 5826 5827 5828 5829 5830 5831 5832 5833 5834 5835 5836 5837 5838 5839 5840 5841 5842 5843 5844 5845 5846 5847 5848 5849 5850 5851 5852 5853 5854 5855 5856 5857 5858 5859 5860 5861 5862 5863 5864 5865 5866 5867 5868 5869 5870 5871 5872 5873 5874 5875 5876 5877 5878 5879 5880 5881 5882 5883 5884 5885 5886 5887 5888 5889 5890 5891 5892 5893 5894 5895 5896 5897 5898 5899 5900 5901 5902 5903 5904 5905 5906 5907 5908 5909 5910 5911 5912 5913 5914 5915 5916 5917 5918 5919 5920 5921 5922 5923 5924 5925 5926 5927 5928 5929 5930 5931 5932 5933 5934 5935 5936 5937 5938 5939 5940 5941 5942 5943 5944 5945 5946 5947 5948 5949 5950 5951 5952 5953 5954 5955 5956 5957 5958 5959 5960 5961 5962 5963 5964 5965 5966 5967 5968 5969 5970 5971 5972 5973 5974 5975 5976 5977 5978 5979 5980 5981 5982 5983 5984 5985 5986 5987 5988 5989 5990 5991 5992 5993 5994 5995 5996 5997 5998 5999 6000 6001 6002 6003 6004 6005 6006 6007 6008 6009 6010 6011 6012 6013 6014 6015 6016 6017 6018 6019 6020 6021 6022 6023 6024 6025 6026 6027 6028 6029 6030 6031 6032 6033 6034 6035 6036 6037 6038 6039 6040 6041 6042 6043 6044 6045 6046 6047 6048 6049 6050 6051 6052 6053 6054 6055 6056 6057 6058 6059 6060 6061 6062 6063 6064 6065 6066 6067 6068 6069 6070 6071 6072 6073 6074 6075 6076 6077 6078 6079 6080 6081 6082 6083 6084 6085 6086 6087 6088 6089 6090 6091 6092 6093 6094 6095 6096 6097 6098 6099 6100 6101 6102 6103 6104 6105 6106 6107 6108 6109 6110 6111 6112 6113 6114 6115 6116 6117 6118 6119 6120 6121 6122 6123 6124 6125 6126 6127 6128 6129 6130 6131 6132 6133 6134 6135 6136 6137 6138 6139 6140 6141 6142 6143 6144 6145 6146 6147 6148 6149 6150 6151 6152 6153 6154 6155 6156 6157 6158 6159 6160 6161 6162 6163 6164 6165 6166 6167 6168 6169 6170 6171 6172 6173 6174 6175 6176 6177 6178 6179 6180 6181 6182 6183 6184 6185 6186 6187 6188 6189 6190 6191 6192 6193 6194 6195 6196 6197 6198 6199 6200 6201 6202 6203 6204 6205 6206 6207 6208 6209 6210 6211 6212 6213 6214 6215 6216 6217 6218 6219 6220 6221 6222 6223 6224 6225 6226 6227 6228 6229 6230 6231 6232 6233 6234 6235 6236 6237 6238 6239 6240 6241 6242 6243 6244 6245 6246 6247 6248 6249 6250 6251 6252 6253 6254 6255 6256 6257 6258 6259 6260 6261 6262 6263 6264 6265 6266 6267 6268 6269 6270 6271 6272 6273 6274 6275 6276 6277 6278 6279 6280 6281 6282 6283 6284 6285 6286 6287 6288 6289 6290 6291 6292 6293 6294 6295 6296 6297 6298 6299 6300 6301 6302 6303 6304 6305 6306 6307 6308 6309 6310 6311 6312 6313 6314 6315 6316 6317 6318 6319 6320 6321 6322 6323 6324 6325 6326 6327 6328 6329 6330 6331 6332 6333 6334 6335 6336 6337 6338 6339 6340 6341 6342 6343 6344 6345 6346 6347 6348 6349 6350 6351 6352 6353 6354 6355 6356 6357 6358 6359 6360 6361 6362 6363 6364 6365 6366 6367 6368 6369 6370 6371 6372 6373 6374 6375 6376 6377 6378 6379 6380 6381 6382 6383 6384 6385 6386 6387 6388 6389 6390 6391 6392 6393 6394 6395 6396 6397 6398 6399 6400 6401 6402 6403 6404 6405 6406 6407 6408 6409 6410 6411 6412 6413 6414 6415 6416 6417 6418 6419 6420 6421 6422 6423 6424 6425 6426 6427 6428 6429 6430 6431 6432 6433 6434 6435 6436 6437 6438 6439 6440 6441 6442 6443 6444 6445 6446 6447 6448 6449 6450 6451 6452 6453 6454 6455 6456 6457 6458 6459 6460 6461 6462 6463 6464 6465 6466 6467 6468 6469 6470 6471 6472 6473 6474 6475 6476 6477 6478 6479 6480 6481 6482 6483 6484 6485 6486 6487 6488 6489 6490 6491 6492 6493 6494 6495 6496 6497 6498 6499 6500 6501 6502 6503 6504 6505 6506 6507 6508 6509 6510 6511 6512 6513 6514 6515 6516 6517 6518 6519 6520 6521 6522 6523 6524 6525 6526 6527 6528 6529 6530 6531 6532 6533 6534 6535 6536 6537 6538 6539 6540 6541 6542 6543 6544 6545 6546 6547 6548 6549 6550 6551 6552 6553 6554 6555 6556 6557 6558 6559 6560 6561 6562 6563 6564 6565 6566 6567 6568 6569 6570 6571 6572 6573 6574 6575 6576 6577 6578 6579 6580 6581 6582 6583 6584 6585 6586 6587 6588 6589 6590 6591 6592 6593 6594 6595 6596 6597 6598 6599 6600 6601 6602 6603 6604 6605 6606 6607 6608 6609 6610 6611 6612 6613 6614 6615 6616 6617 6618 6619 6620 6621 6622 6623 6624 6625 6626 6627 6628 6629 6630 6631 6632 6633 6634 6635 6636 6637 6638 6639 6640 6641 6642 6643 6644 6645 6646 6647 6648 6649 6650 6651 6652 6653 6654 6655 6656 6657 6658 6659 6660 6661 6662 6663 6664 6665 6666 6667 6668 6669 6670 6671 6672 6673 6674 6675 6676 6677 6678 6679 6680 6681 6682 6683 6684 6685 6686 6687 6688 6689 6690 6691 6692 6693 6694 6695 6696 6697 6698 6699 6700 6701 6702 6703 6704 6705 6706 6707 6708 6709 6710 6711 6712 6713 6714 6715 6716 6717 6718 6719 6720 6721 6722 6723 6724 6725 6726 6727 6728 6729 6730 6731 6732 6733 6734 6735 6736 6737 6738 6739 6740 6741 6742 6743 6744 6745 6746 6747 6748 6749 6750 6751 6752 6753 6754 6755 6756 6757 6758 6759 6760 6761 6762 6763 6764 6765 6766 6767 6768 6769 6770 6771 6772 6773 6774 6775 6776 6777 6778 6779 6780 6781 6782 6783 6784 6785 6786 6787 6788 6789 6790 6791 6792 6793 6794 6795 6796 6797 6798 6799 6800 6801 6802 6803 6804 6805 6806 6807 6808 6809 6810 6811 6812 6813 6814 6815 6816 6817 6818 6819 6820 6821 6822 6823 6824 6825 6826 6827 6828 6829 6830 6831 6832 6833 6834 6835 6836 6837 6838 6839 6840 6841 6842 6843 6844 6845 6846 6847 6848 6849 6850 6851 6852 6853 6854 6855 6856 6857 6858 6859 6860 6861 6862 6863 6864 6865 6866 6867 6868 6869 6870 6871 6872 6873 6874 6875 6876 6877 6878 6879 6880 6881 6882 6883 6884 6885 6886 6887 6888 6889 6890 6891 6892 6893 6894 6895 6896 6897 6898 6899 6900 6901 6902 6903 6904 6905 6906 6907 6908 6909 6910 6911 6912 6913 6914 6915 6916 6917 6918 6919 6920 6921 6922 6923 6924 6925 6926 6927 6928 6929 6930 6931 6932 6933 6934 6935 6936 6937 6938 6939 6940 6941 6942 6943 6944 6945 6946 6947 6948 6949 6950 6951 6952 6953 6954 6955 6956 6957 6958 6959 6960 6961 6962 6963 6964 6965 6966 6967 6968 6969 6970 6971 6972 6973 6974 6975 6976 6977 6978 6979 6980 6981 6982 6983 6984 6985 6986 6987 6988 6989 6990 6991 6992 6993 6994 6995 6996 6997 6998 6999 7000 7001 7002 7003 7004 7005 7006 7007 7008 7009 7010 7011 7012 7013 7014 7015 7016 7017 7018 7019 7020 7021 7022 7023 7024 7025 7026 7027 7028 7029 7030 7031 7032 7033 7034 7035 7036 7037 7038 7039 7040 7041 7042 7043 7044 7045 7046 7047 7048 7049 7050 7051 7052 7053 7054 7055 7056 7057 7058 7059 7060 7061 7062 7063 7064 7065 7066 7067 7068 7069 7070 7071 7072 7073 7074 7075 7076 7077 7078 7079 7080 7081 7082 7083 7084 7085 7086 7087 7088 7089 7090 7091 7092 7093 7094 7095 7096 7097 7098 7099 7100 7101 7102 7103 7104 7105 7106 7107 7108 7109 7110 7111 7112 7113 7114 7115 7116 7117 7118 7119 7120 7121 7122 7123 7124 7125 7126 7127 7128 7129 7130 7131 7132 7133 7134 7135 7136 7137 7138 7139 7140 7141 7142 7143 7144 7145 7146 7147 7148 7149 7150 7151 7152 7153 7154 7155 7156 7157 7158 7159 7160 7161 7162 7163 7164 7165 7166 7167 7168 7169 7170 7171 7172 7173 7174 7175 7176 7177 7178 7179 7180 7181 7182 7183 7184 7185 7186 7187 7188 7189 7190 7191 7192 7193 7194 7195 7196 7197 7198 7199 7200 7201 7202 7203 7204 7205 7206 7207 7208 7209 7210 7211 7212 7213 7214 7215 7216 7217 7218 7219 7220 7221 7222 7223 7224 7225 7226 7227 7228 7229 7230 7231 7232 7233 7234 7235 7236 7237 7238 7239 7240 7241 7242 7243 7244 7245 7246 7247 7248 7249 7250 7251 7252 7253 7254 7255 7256 7257 7258 7259 7260 7261 7262 7263 7264 7265 7266 7267 7268 7269 7270 7271 7272 7273 7274 7275 7276 7277 7278 7279 7280 7281 7282 7283 7284 7285 7286 7287 7288 7289 7290 7291 7292 7293 7294 7295 7296 7297 7298 7299 7300 7301 7302 7303 7304 7305 7306 7307 7308 7309 7310 7311 7312 7313 7314 7315 7316 7317 7318 7319 7320 7321 7322 7323 7324 7325 7326 7327 7328 7329 7330 7331 7332 7333 7334 7335 7336 7337 7338 7339 7340 7341 7342 7343 7344 7345 7346 7347 7348 7349 7350 7351 7352 7353 7354 7355 7356 7357 7358 7359 7360 7361 7362 7363 7364 7365 7366 7367 7368 7369 7370 7371 7372 7373 7374 7375 7376 7377 7378 7379 7380 7381 7382 7383 7384 7385 7386 7387 7388 7389 7390 7391 7392 7393 7394 7395 7396 7397 7398 7399 7400 7401 7402 7403 7404 7405 7406 7407 7408 7409 7410 7411 7412 7413 7414 7415 7416 7417 7418 7419 7420 7421 7422 7423 7424 7425 7426 7427 7428 7429 7430 7431 7432 7433 7434 7435 7436 7437 7438 7439 7440 7441 7442 7443 7444 7445 7446 7447 7448 7449 7450 7451 7452 7453 7454 7455 7456 7457 7458 7459 7460 7461 7462 7463 7464 7465 7466 7467 7468 7469 7470 7471 7472 7473 7474 7475 7476 7477 7478 7479 7480 7481 7482 7483 7484 7485 7486 7487 7488 7489 7490 7491 7492 7493 7494 7495 7496 7497 7498 7499 7500 7501 7502 7503 7504 7505 7506 7507 7508 7509 7510 7511 7512 7513 7514 7515 7516 7517 7518 7519 7520 7521 7522 7523 7524 7525 7526 7527 7528 7529 7530 7531 7532 7533 7534 7535 7536 7537 7538 7539 7540 7541 7542 7543 7544 7545 7546 7547 7548 7549 7550 7551 7552 7553 7554 7555 7556 7557 7558 7559 7560 7561 7562 7563 7564 7565 7566 7567 7568 7569 7570 7571 7572 7573 7574 7575 7576 7577 7578 7579 7580 7581 7582 7583 7584 7585 7586 7587 7588 7589 7590 7591 7592 7593 7594 7595 7596 7597 7598 7599 7600 7601 7602 7603 7604 7605 7606 7607 7608 7609 7610 7611 7612 7613 7614 7615 7616 7617 7618 7619 7620 7621 7622 7623 7624 7625 7626 7627 7628 7629 7630 7631 7632 7633 7634 7635 7636 7637 7638 7639 7640 7641 7642 7643 7644 7645 7646 7647 7648 7649 7650 7651 7652 7653 7654 7655 7656 7657 7658 7659 7660 7661 7662 7663 7664 7665 7666 7667 7668 7669 7670 7671 7672 7673 7674 7675 7676 7677 7678 7679 7680 7681 7682 7683 7684 7685 7686 7687 7688 7689 7690 7691 7692 7693 7694 7695 7696 7697 7698 7699 7700 7701 7702 7703 7704 7705 7706 7707 7708 7709 7710 7711 7712 7713 7714 7715 7716 7717 7718 7719 7720 7721 7722 7723 7724 7725 7726 7727 7728 7729 7730 7731 7732 7733 7734 7735 7736 7737 7738 7739 7740 7741 7742 7743 7744 7745 7746 7747 7748 7749 7750 7751 7752 7753 7754 7755 7756 7757 7758 7759 7760 7761 7762 7763 7764 7765 7766 7767 7768 7769 7770 7771 7772 7773 7774 7775 7776 7777 7778 7779 7780 7781 7782 7783 7784 7785 7786 7787 7788 7789 7790 7791 7792 7793 7794 7795 7796 7797 7798 7799 7800 7801 7802 7803 7804 7805 7806 7807 7808 7809 7810 7811 7812 7813 7814 7815 7816 7817 7818 7819 7820 7821 7822 7823 7824 7825 7826 7827 7828 7829 7830 7831 7832 7833 7834 7835 7836 7837 7838 7839 7840 7841 7842 7843 7844 7845 7846 7847 7848 7849 7850 7851 7852 7853 7854 7855 7856 7857 7858 7859 7860 7861 7862 7863 7864 7865 7866 7867 7868 7869 7870 7871 7872 7873 7874 7875 7876 7877 7878 7879 7880 7881 7882 7883 7884 7885 7886 7887 7888 7889 7890 7891 7892 7893 7894 7895 7896 7897 7898 7899 7900 7901 7902 7903 7904 7905 7906 7907 7908 7909 7910 7911 7912 7913 7914 7915 7916 7917 7918 7919 7920 7921 7922 7923 7924 7925 7926 7927 7928 7929 7930 7931 7932 7933 7934 7935 7936 7937 7938 7939 7940 7941 7942 7943 7944 7945 7946 7947 7948 7949 7950 7951 7952 7953 7954 7955 7956 7957 7958 7959 7960 7961 7962 7963 7964 7965 7966 7967 7968 7969 7970 7971 7972 7973 7974 7975 7976 7977 7978 7979 7980 7981 7982 7983 7984 7985 7986 7987 7988 7989 7990 7991 7992 7993 7994 7995 7996 7997 7998 7999 8000 8001 8002 8003 8004 8005 8006 8007 8008 8009 8010 8011 8012 8013 8014 8015 8016 8017 8018 8019 8020 8021 8022 8023 8024 8025 8026 8027 8028 8029 8030 8031 8032 8033 8034 8035 8036 8037 8038 8039 8040 8041 8042 8043 8044 8045 8046 8047 8048 8049 8050 8051 8052 8053 8054 8055 8056 8057 8058 8059 8060 8061 8062 8063 8064 8065 8066 8067 8068 8069 8070 8071 8072 8073 8074 8075 8076 8077 8078 8079 8080 8081 8082 8083 8084 8085 8086 8087 8088 8089 8090 8091 8092 8093 8094 8095 8096 8097 8098 8099 8100 8101 8102 8103 8104 8105 8106 8107 8108 8109 8110 8111 8112 8113 8114 8115 8116 8117 8118 8119 8120 8121 8122 8123 8124 8125 8126 8127 8128 8129 8130 8131 8132 8133 8134 8135 8136 8137 8138 8139 8140 8141 8142 8143 8144 8145 8146 8147 8148 8149 8150 8151 8152 8153 8154 8155 8156 8157 8158 8159 8160 8161 8162 8163 8164 8165 8166 8167 8168 8169 8170 8171 8172 8173 8174 8175 8176 8177 8178 8179 8180 8181 8182 8183 8184 8185 8186 8187 8188 8189 8190 8191 8192 8193 8194 8195 8196 8197 8198 8199 8200 8201 8202 8203 8204 8205 8206 8207 8208 8209 8210 8211 8212 8213 8214 8215 8216 8217 8218 8219 8220 8221 8222 8223 8224 8225 8226 8227 8228 8229 8230 8231 8232 8233 8234 8235 8236 8237 8238 8239 8240 8241 8242 8243 8244 8245 8246 8247 8248 8249 8250 8251 8252 8253 8254 8255 8256 8257 8258 8259 8260 8261 8262 8263 8264 8265 8266 8267 8268 8269 8270 8271 8272 8273 8274 8275 8276 8277 8278 8279 8280 8281 8282 8283 8284 8285 8286 8287 8288 8289 8290 8291 8292 8293 8294 8295 8296 8297 8298 8299 8300 8301 8302 8303 8304 8305 8306 8307 8308 8309 8310 8311 8312 8313 8314 8315 8316 8317 8318 8319 8320 8321 8322 8323 8324 8325 8326 8327 8328 8329 8330 8331 8332 8333 8334 8335 8336 8337 8338 8339 8340 8341 8342 8343 8344 8345 8346 8347 8348 8349 8350 8351 8352 8353 8354 8355 8356 8357 8358 8359 8360 8361 8362 8363 8364 8365 8366 8367 8368 8369 8370 8371 8372 8373 8374 8375 8376 8377 8378 8379 8380 8381 8382 8383 8384 8385 8386 8387 8388 8389 8390 8391 8392 8393 8394 8395 8396 8397 8398 8399 8400 8401 8402 8403 8404 8405 8406 8407 8408 8409 8410 8411 8412 8413 8414 8415 8416 8417 8418 8419 8420 8421 8422 8423 8424 8425 8426 8427 8428 8429 8430 8431 8432 8433 8434 8435 8436 8437 8438 8439 8440 8441 8442 8443 8444 8445 8446 8447 8448 8449 8450 8451 8452 8453 8454 8455 8456 8457 8458 8459 8460 8461 8462 8463 8464 8465 8466 8467 8468 8469 8470 8471 8472 8473 8474 8475 8476 8477 8478 8479 8480 8481 8482 8483 8484 8485 8486 8487 8488 8489 8490 8491 8492 8493 8494 8495 8496 8497 8498 8499 8500 8501 8502 8503 8504 8505 8506 8507 8508 8509 8510 8511 8512 8513 8514 8515 8516 8517 8518 8519 8520 8521 8522 8523 8524 8525 8526 8527 8528 8529 8530 8531 8532 8533 8534 8535 8536 8537 8538 8539 8540 8541 8542 8543 8544 8545 8546 8547 8548 8549 8550 8551 8552 8553 8554 8555 8556 8557 8558 8559 8560 8561 8562 8563 8564 8565 8566 8567 8568 8569 8570 8571 8572 8573 8574 8575 8576 8577 8578 8579 8580 8581 8582 8583 8584 8585 8586 8587 8588 8589 8590 8591 8592 8593 8594 8595 8596 8597 8598 8599 8600 8601 8602 8603 8604 8605 8606 8607 8608 8609 8610 8611 8612 8613 8614 8615 8616 8617 8618 8619 8620 8621 8622 8623 8624 8625 8626 8627 8628 8629 8630 8631 8632 8633 8634 8635 8636 8637 8638 8639 8640 8641 8642 8643 8644 8645 8646 8647 8648 8649 8650 8651 8652 8653 8654 8655 8656 8657 8658 8659 8660 8661 8662 8663 8664 8665 8666 8667 8668 8669 8670 8671 8672 8673 8674 8675 8676 8677 8678 8679 8680 8681 8682 8683 8684 8685 8686 8687 8688 8689 8690 8691 8692 8693 8694 8695 8696 8697 8698 8699 8700 8701 8702 8703 8704 8705 8706 8707 8708 8709 8710 8711 8712 8713 8714 8715 8716 8717 8718 8719 8720 8721 8722 8723 8724 8725 8726 8727 8728 8729 8730 8731 8732 8733 8734 8735 8736 8737 8738 8739 8740 8741 8742 8743 8744 8745 8746 8747 8748 8749 8750 8751 8752 8753 8754 8755 8756 8757 8758 8759 8760 8761 8762 8763 8764 8765 8766 8767 8768 8769 8770 8771 8772 8773 8774 8775 8776 8777 8778 8779 8780 8781 8782 8783 8784 8785 8786 8787 8788 8789 8790 8791 8792 8793 8794 8795 8796 8797 8798 8799 8800 8801 8802 8803 8804 8805 8806 8807 8808 8809 8810 8811 8812 8813 8814 8815 8816 8817 8818 8819 8820 8821 8822 8823 8824 8825 8826 8827 8828 8829 8830 8831 8832 8833 8834 8835 8836 8837 8838 8839 8840 8841 8842 8843 8844 8845 8846 8847 8848 8849 8850 8851 8852 8853 8854 8855 8856 8857 8858 8859 8860 8861 8862 8863 8864 8865 8866 8867 8868 8869 8870 8871 8872 8873 8874 8875 8876 8877 8878 8879 8880 8881 8882 8883 8884 8885 8886 8887 8888 8889 8890 8891 8892 8893 8894 8895 8896 8897 8898 8899 8900 8901 8902 8903 8904 8905 8906 8907 8908 8909 8910 8911 8912 8913 8914 8915 8916 8917 8918 8919 8920 8921 8922 8923 8924 8925 8926 8927 8928 8929 8930 8931 8932 8933 8934 8935 8936 8937 8938 8939 8940 8941 8942 8943 8944 8945 8946 8947 8948 8949 8950 8951 8952 8953 8954 8955 8956 8957 8958 8959 8960 8961 8962 8963 8964 8965 8966 8967 8968 8969 8970 8971 8972 8973 8974 8975 8976 8977 8978 8979 8980 8981 8982 8983 8984 8985 8986 8987 8988 8989 8990 8991 8992 8993 8994 8995 8996 8997 8998 8999 9000 9001 9002 9003 9004 9005 9006 9007 9008 9009 9010 9011 9012 9013 9014 9015 9016 9017 9018 9019 9020 9021 9022 9023 9024 9025 9026 9027 9028 9029 9030 9031 9032 9033 9034 9035 9036 9037 9038 9039 9040 9041 9042 9043 9044 9045 9046 9047 9048 9049 9050 9051 9052 9053 9054 9055 9056 9057 9058 9059 9060 9061 9062 9063 9064 9065 9066 9067 9068 9069 9070 9071 9072 9073 9074 9075 9076 9077 9078 9079 9080 9081 9082 9083 9084 9085 9086 9087 9088 9089 9090 9091 9092 9093 9094 9095 9096 9097 9098 9099 9100 9101 9102 9103 9104 9105 9106 9107 9108 9109 9110 9111 9112 9113 9114 9115 9116 9117 9118 9119 9120 9121 9122 9123 9124 9125 9126 9127 9128 9129 9130 9131 9132 9133 9134 9135 9136 9137 9138 9139 9140 9141 9142 9143 9144 9145 9146 9147 9148 9149 9150 9151 9152 9153 9154 9155 9156 9157 9158 9159 9160 9161 9162 9163 9164 9165 9166 9167 9168 9169 9170 9171 9172 9173 9174 9175 9176 9177 9178 9179 9180 9181 9182 9183 9184 9185 9186 9187 9188 9189 9190 9191 9192 9193 9194 9195 9196 9197 9198 9199 9200 9201 9202 9203 9204 9205 9206 9207 9208 9209 9210 9211 9212 9213 9214 9215 9216 9217 9218 9219 9220 9221 9222 9223 9224 9225 9226 9227 9228 9229 9230 9231 9232 9233 9234 9235 9236 9237 9238 9239 9240 9241 9242 9243 9244 9245 9246 9247 9248 9249 9250 9251 9252 9253 9254 9255 9256 9257 9258 9259 9260 9261 9262 9263 9264 9265 9266 9267 9268 9269 9270 9271 9272 9273 9274 9275 9276 9277 9278 9279 9280 9281 9282 9283 9284 9285 9286 9287 9288 9289 9290 9291 9292 9293 9294 9295 9296 9297 9298 9299 9300 9301 9302 9303 9304 9305 9306 9307 9308 9309 9310 9311 9312 9313 9314 9315 9316 9317 9318 9319 9320 9321 9322 9323 9324 9325 9326 9327 9328 9329 9330 9331 9332 9333 9334 9335 9336 9337 9338 9339 9340 9341 9342 9343 9344 9345 9346 9347 9348 9349 9350 9351 9352 9353 9354 9355 9356 9357 9358 9359 9360 9361 9362 9363 9364 9365 9366 9367 9368 9369 9370 9371 9372 9373 9374 9375 9376 9377 9378 9379 9380 9381 9382 9383 9384 9385 9386 9387 9388 9389 9390 9391 9392 9393 9394 9395 9396 9397 9398 9399 9400 9401 9402 9403 9404 9405 9406 9407 9408 9409 9410 9411 9412 9413 9414 9415 9416 9417 9418 9419 9420 9421 9422 9423 9424 9425 9426 9427 9428 9429 9430 9431 9432 9433 9434 9435 9436 9437 9438 9439 9440 9441 9442 9443 9444 9445 9446 9447 9448 9449 9450 9451 9452 9453 9454 9455 9456 9457 9458 9459 9460 9461 9462 9463 9464 9465 9466 9467 9468 9469 9470 9471 9472 9473 9474 9475 9476 9477 9478 9479 9480 9481 9482 9483 9484 9485 9486 9487 9488 9489 9490 9491 9492 9493 9494 9495 9496 9497 9498 9499 9500 9501 9502 9503 9504 9505 9506 9507 9508 9509 9510 9511 9512 9513 9514 9515 9516 9517 9518 9519 9520 9521 9522 9523 9524 9525 9526 9527 9528 9529 9530 9531 9532 9533 9534 9535 9536 9537 9538 9539 9540 9541 9542 9543 9544 9545 9546 9547 9548 9549 9550 9551 9552 9553 9554 9555 9556 9557 9558 9559 9560 9561 9562 9563 9564 9565 9566 9567 9568 9569 9570 9571 9572 9573 9574 9575 9576 9577 9578 9579 9580 9581 9582 9583 9584 9585 9586 9587 9588 9589 9590 9591 9592 9593 9594 9595 9596 9597 9598 9599 9600 9601 9602 9603 9604 9605 9606 9607 9608 9609 9610 9611 9612 9613 9614 9615 9616 9617 9618 9619 9620 9621 9622 9623 9624 9625 9626 9627 9628 9629 9630 9631 9632 9633 9634 9635 9636 9637 9638 9639 9640 9641 9642 9643 9644 9645 9646 9647 9648 9649 9650 9651 9652 9653 9654 9655 9656 9657 9658 9659 9660 9661 9662 9663 9664 9665 9666 9667 9668 9669 9670 9671 9672 9673 9674 9675 9676 9677 9678 9679 9680 9681 9682 9683 9684 9685 9686 9687 9688 9689 9690 9691 9692 9693 9694 9695 9696 9697 9698 9699 9700 9701 9702 9703 9704 9705 9706 9707 9708 9709 9710 9711 9712 9713 9714 9715 9716 9717 9718 9719 9720 9721 9722 9723 9724 9725 9726 9727 9728 9729 9730 9731 9732 9733 9734 9735 9736 9737 9738 9739 9740 9741 9742 9743 9744 9745 9746 9747 9748 9749 9750 9751 9752 9753 9754 9755 9756 9757 9758 9759 9760 9761 9762 9763 9764 9765 9766 9767 9768 9769 9770 9771 9772 9773 9774 9775 9776 9777 9778 9779 9780 9781 9782 9783 9784 9785 9786 9787 9788 9789 9790 9791 9792 9793 9794 9795 9796 9797 9798 9799 9800 9801 9802 9803 9804 9805 9806 9807 9808 9809 9810 9811 9812 9813 9814 9815 9816 9817 9818 9819 9820 9821 9822 9823 9824 9825 9826 9827 9828 9829 9830 9831 9832 9833 9834 9835 9836 9837 9838 9839 9840 9841 9842 9843 9844 9845 9846 9847 9848 9849 9850 9851 9852 9853 9854 9855 9856 9857 9858 9859 9860 9861 9862 9863 9864 9865 9866 9867 9868 9869 9870 9871 9872 9873 9874 9875 9876 9877 9878 9879 9880 9881 9882 9883 9884 9885 9886 9887 9888 9889 9890 9891 9892 9893 9894 9895 9896 9897 9898 9899 9900 9901 9902 9903 9904 9905 9906 9907 9908 9909 9910 9911 9912 9913 9914 9915 9916 9917 9918 9919 9920 9921 9922 9923 9924 9925 9926 9927 9928 9929 9930 9931 9932 9933 9934 9935 9936 9937 9938 9939 9940 9941 9942 9943 9944 9945 9946 9947 9948 9949 9950 9951 9952 9953 9954 9955 9956 9957 9958 9959 9960 9961 9962 9963 9964 9965 9966 9967 9968 9969 9970 9971 9972 9973 9974 9975 9976 9977 9978 9979 9980 9981 9982 9983 9984 9985 9986 9987 9988 9989 9990 9991 9992 9993 9994 9995 9996 9997 9998 9999 10000 10001 10002 10003 10004 10005 10006 10007 10008 10009 10010 10011 10012 10013 10014 10015 10016 10017 10018 10019 10020 10021 10022 10023 10024 10025 10026 10027 10028 10029 10030 10031 10032 10033 10034 10035 10036 10037 10038 10039 10040 10041 10042 10043 10044 10045 10046 10047 10048 10049 10050 10051 10052 10053 10054 10055 10056 10057 10058 10059 10060 10061 10062 10063 10064 10065 10066 10067 10068 10069 10070 10071 10072 10073 10074 10075 10076 10077 10078 10079 10080 10081 10082 10083 10084 10085 10086 10087 10088 10089 10090 10091 10092 10093 10094 10095 10096 10097 10098 10099 10100 10101 10102 10103 10104 10105 10106 10107 10108 10109 10110 10111 10112 10113 10114 10115 10116 10117 10118 10119 10120 10121 10122 10123 10124 10125 10126 10127 10128 10129 10130 10131 10132 10133 10134 10135 10136 10137 10138 10139 10140 10141 10142 10143 10144 10145 10146 10147 10148 10149 10150 10151 10152 10153 10154 10155 10156 10157 10158 10159 10160 10161 10162 10163 10164 10165 10166 10167 10168 10169 10170 10171 10172 10173 10174 10175 10176 10177 10178 10179 10180 10181 10182 10183 10184 10185 10186 10187 10188 10189 10190 10191 10192 10193 10194 10195 10196 10197 10198 10199 10200 10201 10202 10203 10204 10205 10206 10207 10208 10209 10210 10211 10212 10213 10214 10215 10216 10217 10218 10219 10220 10221 10222 10223 10224 10225 10226 10227 10228 10229 10230 10231 10232 10233 10234 10235 10236 10237 10238 10239 10240 10241 10242 10243 10244 10245 10246 10247 10248 10249 10250 10251 10252 10253 10254 10255 10256 10257 10258 10259 10260 10261 10262 10263 10264 10265 10266 10267 10268 10269 10270 10271 10272 10273 10274 10275 10276 10277 10278 10279 10280 10281 10282 10283 10284 10285 10286 10287 10288 10289 10290 10291 10292 10293 10294 10295 10296 10297 10298 10299 10300 10301 10302 10303 10304 10305 10306 10307 10308 10309 10310 10311 10312 10313 10314 10315 10316 10317 10318 10319 10320 10321 10322 10323 10324 10325 10326 10327 10328 10329 10330 10331 10332 10333 10334 10335 10336 10337 10338 10339 10340 10341 10342 10343 10344 10345 10346 10347 10348 10349 10350 10351 10352 10353 10354 10355 10356 10357 10358 10359 10360 10361 10362 10363 10364 10365 10366 10367 10368 10369 10370 10371 10372 10373 10374 10375 10376 10377 10378 10379 10380 10381 10382 10383 10384 10385 10386 10387 10388 10389 10390 10391 10392 10393 10394 10395 10396 10397 10398 10399 10400 10401 10402 10403 10404 10405 10406 10407 10408 10409 10410 10411 10412 10413 10414 10415 10416 10417 10418 10419 10420 10421 10422 10423 10424 10425 10426 10427 10428 10429 10430 10431 10432 10433 10434 10435 10436 10437 10438 10439 10440 10441 10442 10443 10444 10445 10446 10447 10448 10449 10450 10451 10452 10453 10454 10455 10456 10457 10458 10459 10460 10461 10462 10463 10464 10465 10466 10467 10468 10469 10470 10471 10472 10473 10474 10475 10476 10477 10478 10479 10480 10481 10482 10483 10484 10485 10486 10487 10488 10489 10490 10491 10492 10493 10494 10495 10496 10497 10498 10499 10500 10501 10502 10503 10504 10505 10506 10507 10508 10509 10510 10511 10512 10513 10514 10515 10516 10517 10518 10519 10520 10521 10522 10523 10524 10525 10526 10527 10528 10529 10530 10531 10532 10533 10534 10535 10536 10537 10538 10539 10540 10541 10542 10543 10544 10545 10546 10547 10548 10549 10550 10551 10552 10553 10554 10555 10556 10557 10558 10559 10560 10561 10562 10563 10564 10565 10566 10567 10568 10569 10570 10571 10572 10573 10574 10575 10576 10577 10578 10579 10580 10581 10582 10583 10584 10585 10586 10587 10588 10589 10590 10591 10592 10593 10594 10595 10596 10597 10598 10599 10600 10601 10602 10603 10604 10605 10606 10607 10608 10609 10610 10611 10612 10613 10614 10615 10616 10617 10618 10619 10620 10621 10622 10623 10624 10625 10626 10627 10628 10629 10630 10631 10632 10633 10634 10635 10636 10637 10638 10639 10640 10641 10642 10643 10644 10645 10646 10647 10648 10649 10650 10651 10652 10653 10654 10655 10656 10657 10658 10659 10660 10661 10662 10663 10664 10665 10666 10667 10668 10669 10670 10671 10672 10673 10674 10675 10676 10677 10678 10679 10680 10681 10682 10683 10684 10685 10686 10687 10688 10689 10690 10691 10692 10693 10694 10695 10696 10697 10698 10699 10700 10701 10702 10703 10704 10705 10706 10707 10708 10709 10710 10711 10712 10713 10714 10715 10716 10717 10718 10719 10720 10721 10722 10723 10724 10725 10726 10727 10728 10729 10730 10731 10732 10733 10734 10735 10736 10737 10738 10739 10740 10741 10742 10743 10744 10745 10746 10747 10748 10749 10750 10751 10752 10753 10754 10755 10756 10757 10758 10759 10760 10761 10762 10763 10764 10765 10766 10767 10768 10769 10770 10771 10772 10773 10774 10775 10776 10777 10778 10779 10780 10781 10782 10783 10784 10785 10786 10787 10788 10789 10790 10791 10792 10793 10794 10795 10796 10797 10798 10799 10800 10801 10802 10803 10804 10805 10806 10807 10808 10809 10810 10811 10812 10813 10814 10815 10816 10817 10818 10819 10820 10821 10822 10823 10824 10825 10826 10827 10828 10829 10830 10831 10832 10833 10834 10835 10836 10837 10838 10839 10840 10841 10842 10843 10844 10845 10846 10847 10848 10849 10850 10851 10852 10853 10854 10855 10856 10857 10858 10859 10860 10861 10862 10863 10864 10865 10866 10867 10868 10869 10870 10871 10872 10873 10874 10875 10876 10877 10878 10879 10880 10881 10882 10883 10884 10885 10886 10887 10888 10889 10890 10891 10892 10893 10894 10895 10896 10897 10898 10899 10900 10901 10902 10903 10904 10905 10906 10907 10908 10909 10910 10911 10912 10913 10914 10915 10916 10917 10918 10919 10920 10921 10922 10923 10924 10925 10926 10927 10928 10929 10930 10931 10932 10933 10934 10935 10936 10937 10938 10939 10940 10941 10942 10943 10944 10945 10946 10947 10948 10949 10950 10951 10952 10953 10954 10955 10956 10957 10958 10959 10960 10961 10962 10963 10964 10965 10966 10967 10968 10969 10970 10971 10972 10973 10974 10975 10976 10977 10978 10979 10980 10981 10982 10983 10984 10985 10986 10987 10988 10989 10990 10991 10992 10993 10994 10995 10996 10997 10998 10999 11000 11001 11002 11003 11004 11005 11006 11007 11008 11009 11010 11011 11012 11013 11014 11015 11016 11017 11018 11019 11020 11021 11022 11023 11024 11025 11026 11027 11028 11029 11030 11031 11032 11033 11034 11035 11036 11037 11038 11039 11040 11041 11042 11043 11044 11045 11046 11047 11048 11049 11050 11051 11052 11053 11054 11055 11056 11057 11058 11059 11060 11061 11062 11063 11064 11065 11066 11067 11068 11069 11070 11071 11072 11073 11074 11075 11076 11077 11078 11079 11080 11081 11082 11083 11084 11085 11086 11087 11088 11089 11090 11091 11092 11093 11094 11095 11096 11097 11098 11099 11100 11101 11102 11103 11104 11105 11106 11107 11108 11109 11110 11111 11112 11113 11114 11115 11116 11117 11118 11119 11120 11121 11122 11123 11124 11125 11126 11127 11128 11129 11130 11131 11132 11133 11134 11135 11136 11137 11138 11139 11140 11141 11142 11143 11144 11145 11146 11147 11148 11149 11150 11151 11152 11153 11154 11155 11156 11157 11158 11159 11160 11161 11162 11163 11164 11165 11166 11167 11168 11169 11170 11171 11172 11173 11174 11175 11176 11177 11178 11179 11180 11181 11182 11183 11184 11185 11186 11187 11188 11189 11190 11191 11192 11193 11194 11195 11196 11197 11198 11199 11200 11201 11202 11203 11204 11205 11206 11207 11208 11209 11210 11211 11212 11213 11214 11215 11216 11217 11218 11219 11220 11221 11222 11223 11224 11225 11226 11227 11228 11229 11230 11231 11232 11233 11234 11235 11236 11237 11238 11239 11240 11241 11242 11243 11244 11245 11246 11247 11248 11249 11250 11251 11252 11253 11254 11255 11256 11257 11258 11259 11260 11261 11262 11263 11264 11265 11266 11267 11268 11269 11270 11271 11272 11273 11274 11275 11276 11277 11278 11279 11280 11281 11282 11283 11284 11285 11286 11287 11288 11289 11290 11291 11292 11293 11294 11295 11296 11297 11298 11299 11300 11301 11302 11303 11304 11305 11306 11307 11308 11309 11310 11311 11312 11313 11314 11315 11316 11317 11318 11319 11320 11321 11322 11323 11324 11325 11326 11327 11328 11329 11330 11331 11332 11333 11334 11335 11336 11337 11338 11339 11340 11341 11342 11343 11344 11345 11346 11347 11348 11349 11350 11351 11352 11353 11354 11355 11356 11357 11358 11359 11360 11361 11362 11363 11364 11365 11366 11367 11368 11369 11370 11371 11372 11373 11374 11375 11376 11377 11378 11379 11380 11381 11382 11383 11384 11385 11386 11387 11388 11389 11390 11391 11392 11393 11394 11395 11396 11397 11398 11399 11400 11401 11402 11403 11404 11405 11406 11407 11408 11409 11410 11411 11412 11413 11414 11415 11416 11417 11418 11419 11420 11421 11422 11423 11424 11425 11426 11427 11428 11429 11430 11431 11432 11433 11434 11435 11436 11437 11438 11439 11440 11441 11442 11443 11444 11445 11446 11447 11448 11449 11450 11451 11452 11453 11454 11455 11456 11457 11458 11459 11460 11461 11462 11463 11464 11465 11466 11467 11468 11469 11470 11471 11472 11473 11474 11475 11476 11477 11478 11479 11480 11481 11482 11483 11484 11485 11486 11487 11488 11489 11490 11491 11492 11493 11494 11495 11496 11497 11498 11499 11500 11501 11502 11503 11504 11505 11506 11507 11508 11509 11510 11511 11512 11513 11514 11515 11516 11517 11518 11519 11520 11521 11522 11523 11524 11525 11526 11527 11528 11529 11530 11531 11532 11533 11534 11535 11536 11537 11538 11539 11540 11541 11542 11543 11544 11545 11546 11547 11548 11549 11550 11551 11552 11553 11554 11555 11556 11557 11558 11559 11560 11561 11562 11563 11564 11565 11566 11567 11568 11569 11570 11571 11572 11573 11574 11575 11576 11577 11578 11579 11580 11581 11582 11583 11584 11585 11586 11587 11588 11589 11590 11591 11592 11593 11594 11595 11596 11597 11598 11599 11600 11601 11602 11603 11604 11605 11606 11607 11608 11609 11610 11611 11612 11613 11614 11615 11616 11617 11618 11619 11620 11621 11622 11623 11624 11625 11626 11627 11628 11629 11630 11631 11632 11633 11634 11635 11636 11637 11638 11639 11640 11641 11642 11643 11644 11645 11646 11647 11648 11649 11650 11651 11652 11653 11654 11655 11656 11657 11658 11659 11660 11661 11662 11663 11664 11665 11666 11667 11668 11669 11670 11671 11672 11673 11674 11675 11676 11677 11678 11679 11680 11681 11682 11683 11684 11685 11686 11687 11688 11689 11690 11691 11692 11693 11694 11695 11696 11697 11698 11699 11700 11701 11702 11703 11704 11705 11706 11707 11708 11709 11710 11711 11712 11713 11714 11715 11716 11717 11718 11719 11720 11721 11722 11723 11724 11725 11726 11727 11728 11729 11730 11731 11732 11733 11734 11735 11736 11737 11738 11739 11740 11741 11742 11743 11744 11745 11746 11747 11748 11749 11750 11751 11752 11753 11754 11755 11756 11757 11758 11759 11760 11761 11762 11763 11764 11765 11766 11767 11768 11769 11770 11771 11772 11773 11774 11775 11776 11777 11778 11779 11780 11781 11782 11783 11784 11785 11786 11787 11788 11789 11790 11791 11792 11793 11794 11795 11796 11797 11798 11799 11800 11801 11802 11803 11804 11805 11806 11807 11808 11809 11810 11811 11812 11813 11814 11815 11816 11817 11818 11819 11820 11821 11822 11823 11824 11825 11826 11827 11828 11829 11830 11831 11832 11833 11834 11835 11836 11837 11838 11839 11840 11841 11842 11843 11844 11845 11846 11847 11848 11849 11850 11851 11852 11853 11854 11855 11856 11857 11858 11859 11860 11861 11862 11863 11864 11865 11866 11867 11868 11869 11870 11871 11872 11873 11874 11875 11876 11877 11878 11879 11880 11881 11882 11883 11884 11885 11886 11887 11888 11889 11890 11891 11892 11893 11894 11895 11896 11897 11898 11899 11900 11901 11902 11903 11904 11905 11906 11907 11908 11909 11910 11911 11912 11913 11914 11915 11916 11917 11918 11919 11920 11921 11922 11923 11924 11925 11926 11927 11928 11929 11930 11931 11932 11933 11934 11935 11936 11937 11938 11939 11940 11941 11942 11943 11944 11945 11946 11947 11948 11949 11950 11951 11952 11953 11954 11955 11956 11957 11958 11959 11960 11961 11962 11963 11964 11965 11966 11967 11968 11969 11970 11971 11972 11973 11974 11975 11976 11977 11978 11979 11980 11981 11982 11983 11984 11985 11986 11987 11988 11989 11990 11991 11992 11993 11994 11995 11996 11997 11998 11999 12000 12001 12002 12003 12004 12005 12006 12007 12008 12009 12010 12011 12012 12013 12014 12015 12016 12017 12018 12019 12020 12021 12022 12023 12024 12025 12026 12027 12028 12029 12030 12031 12032 12033 12034 12035 12036 12037 12038 12039 12040 12041 12042 12043 12044 12045 12046 12047 12048 12049 12050 12051 12052 12053 12054 12055 12056 12057 12058 12059 12060 12061 12062 12063 12064 12065 12066 12067 12068 12069 12070 12071 12072 12073 12074 12075 12076 12077 12078 12079 12080 12081 12082 12083 12084 12085 12086 12087 12088 12089 12090 12091 12092 12093 12094 12095 12096 12097 12098 12099 12100 12101 12102 12103 12104 12105 12106 12107 12108 12109 12110 12111 12112 12113 12114 12115 12116 12117 12118 12119 12120 12121 12122 12123 12124 12125 12126 12127 12128 12129 12130 12131 12132 12133 12134 12135 12136 12137 12138 12139 12140 12141 12142 12143 12144 12145 12146 12147 12148 12149 12150 12151 12152 12153 12154 12155 12156 12157 12158 12159 12160 12161 12162 12163 12164 12165 12166 12167 12168 12169 12170 12171 12172 12173 12174 12175 12176 12177 12178 12179 12180 12181 12182 12183 12184 12185 12186 12187 12188 12189 12190 12191 12192 12193 12194 12195 12196 12197 12198 12199 12200 12201 12202 12203 12204 12205 12206 12207 12208 12209 12210 12211 12212 12213 12214 12215 12216 12217 12218 12219 12220 12221 12222 12223 12224 12225 12226 12227 12228 12229 12230 12231 12232 12233 12234 12235 12236 12237 12238 12239 12240 12241 12242 12243 12244 12245 12246 12247 12248 12249 12250 12251 12252 12253 12254 12255 12256 12257 12258 12259 12260 12261 12262 12263 12264 12265 12266 12267 12268 12269 12270 12271 12272 12273 12274 12275 12276 12277 12278 12279 12280 12281 12282 12283 12284 12285 12286 12287 12288 12289 12290 12291 12292 12293 12294 12295 12296 12297 12298 12299 12300 12301 12302 12303 12304 12305 12306 12307 12308 12309 12310 12311 12312 12313 12314 12315 12316 12317 12318 12319 12320 12321 12322 12323 12324 12325 12326 12327 12328 12329 12330 12331 12332 12333 12334 12335 12336 12337 12338 12339 12340 12341 12342 12343 12344 12345 12346 12347 12348 12349 12350 12351 12352 12353 12354 12355 12356 12357 12358 12359 12360 12361 12362 12363 12364 12365 12366 12367 12368 12369 12370 12371 12372 12373 12374 12375 12376 12377 12378 12379 12380 12381 12382 12383 12384 12385 12386 12387 12388 12389 12390 12391 12392 12393 12394 12395 12396 12397 12398 12399 12400 12401 12402 12403 12404 12405 12406 12407 12408 12409 12410 12411 12412 12413 12414 12415 12416 12417 12418 12419 12420 12421 12422 12423 12424 12425 12426 12427 12428 12429 12430 12431 12432 12433 12434 12435 12436 12437 12438 12439 12440 12441 12442 12443 12444 12445 12446 12447 12448 12449 12450 12451 12452 12453 12454 12455 12456 12457 12458 12459 12460 12461 12462 12463 12464 12465 12466 12467 12468 12469 12470 12471 12472 12473 12474 12475 12476 12477 12478 12479 12480 12481 12482 12483 12484 12485 12486 12487 12488 12489 12490 12491 12492 12493 12494 12495 12496 12497 12498 12499 12500 12501 12502 12503 12504 12505 12506 12507 12508 12509 12510 12511 12512 12513 12514 12515 12516 12517 12518 12519 12520 12521 12522 12523 12524 12525 12526 12527 12528 12529 12530 12531 12532 12533 12534 12535 12536 12537 12538 12539 12540 12541 12542 12543 12544 12545 12546 12547 12548 12549 12550 12551 12552 12553 12554 12555 12556 12557 12558 12559 12560 12561 12562 12563 12564 12565 12566 12567 12568 12569 12570 12571 12572 12573 12574 12575 12576 12577 12578 12579 12580 12581 12582 12583 12584 12585 12586 12587 12588 12589 12590 12591 12592 12593 12594 12595 12596 12597 12598 12599 12600 12601 12602 12603 12604 12605 12606 12607 12608 12609 12610 12611 12612 12613 12614 12615 12616 12617 12618 12619 12620 12621 12622 12623 12624 12625 12626 12627 12628 12629 12630 12631 12632 12633 12634 12635 12636 12637 12638 12639 12640 12641 12642 12643 12644 12645 12646 12647 12648 12649 12650 12651 12652 12653 12654 12655 12656 12657 12658 12659 12660 12661 12662 12663 12664 12665 12666 12667 12668 12669 12670 12671 12672 12673 12674 12675 12676 12677 12678 12679 12680 12681 12682 12683 12684 12685 12686 12687 12688 12689 12690 12691 12692 12693 12694 12695 12696 12697 12698 12699 12700 12701 12702 12703 12704 12705 12706 12707 12708 12709 12710 12711 12712 12713 12714 12715 12716 12717 12718 12719 12720 12721 12722 12723 12724 12725 12726 12727 12728 12729 12730 12731 12732 12733 12734 12735 12736 12737 12738 12739 12740 12741 12742 12743 12744 12745 12746 12747 12748 12749 12750 12751 12752 12753 12754 12755 12756 12757 12758 12759 12760 12761 12762 12763 12764 12765 12766 12767 12768 12769 12770 12771 12772 12773 12774 12775 12776 12777 12778 12779 12780 12781 12782 12783 12784 12785 12786 12787 12788 12789 12790 12791 12792 12793 12794 12795 12796 12797 12798 12799 12800 12801 12802 12803 12804 12805 12806 12807 12808 12809 12810 12811 12812 12813 12814 12815 12816 12817 12818 12819 12820 12821 12822 12823 12824 12825 12826 12827 12828 12829 12830 12831 12832 12833 12834 12835 12836 12837 12838 12839 12840 12841 12842 12843 12844 12845 12846 12847 12848 12849 12850 12851 12852 12853 12854 12855 12856 12857 12858 12859 12860 12861 12862 12863 12864 12865 12866 12867 12868 12869 12870 12871 12872 12873 12874 12875 12876 12877 12878 12879 12880 12881 12882 12883 12884 12885 12886 12887 12888 12889 12890 12891 12892 12893 12894 12895 12896 12897 12898 12899 12900 12901 12902 12903 12904 12905 12906 12907 12908 12909 12910 12911 12912 12913 12914 12915 12916 12917 12918 12919 12920 12921 12922 12923 12924 12925 12926 12927 12928 12929 12930 12931 12932 12933 12934 12935 12936 12937 12938 12939 12940 12941 12942 12943 12944 12945 12946 12947 12948 12949 12950 12951 12952 12953 12954 12955 12956 12957 12958 12959 12960 12961 12962 12963 12964 12965 12966 12967 12968 12969 12970 12971 12972 12973 12974 12975 12976 12977 12978 12979 12980 12981 12982 12983 12984 12985 12986 12987 12988 12989 12990 12991 12992 12993 12994 12995 12996 12997 12998 12999 13000 13001 13002 13003 13004 13005 13006 13007 13008 13009 13010 13011 13012 13013 13014 13015 13016 13017 13018 13019 13020 13021 13022 13023 13024 13025 13026 13027 13028 13029 13030 13031 13032 13033 13034 13035 13036 13037 13038 13039 13040 13041 13042 13043 13044 13045 13046 13047 13048 13049 13050 13051 13052 13053 13054 13055 13056 13057 13058 13059 13060 13061 13062 13063 13064 13065 13066 13067 13068 13069 13070 13071 13072 13073 13074 13075 13076 13077 13078 13079 13080 13081 13082 13083 13084 13085 13086 13087 13088 13089 13090 13091 13092 13093 13094 13095 13096 13097 13098 13099 13100 13101 13102 13103 13104 13105 13106 13107 13108 13109 13110 13111 13112 13113 13114 13115 13116 13117 13118 13119 13120 13121 13122 13123 13124 13125 13126 13127 13128 13129 13130 13131 13132 13133 13134 13135 13136 13137 13138 13139 13140 13141 13142 13143 13144 13145 13146 13147 13148 13149 13150 13151 13152 13153 13154 13155 13156 13157 13158 13159 13160 13161 13162 13163 13164 13165 13166 13167 13168 13169 13170 13171 13172 13173 13174 13175 13176 13177 13178 13179 13180 13181 13182 13183 13184 13185 13186 13187 13188 13189 13190 13191 13192 13193 13194 13195 13196 13197 13198 13199 13200 13201 13202 13203 13204 13205 13206 13207 13208 13209 13210 13211 13212 13213 13214 13215 13216 13217 13218 13219 13220 13221 13222 13223 13224 13225 13226 13227 13228 13229 13230 13231 13232 13233 13234 13235 13236 13237 13238 13239 13240 13241 13242 13243 13244 13245 13246 13247 13248 13249 13250 13251 13252 13253 13254 13255 13256 13257 13258 13259 13260 13261 13262 13263 13264 13265 13266 13267 13268 13269 13270 13271 13272 13273 13274 13275 13276 13277 13278 13279 13280 13281 13282 13283 13284 13285 13286 13287 13288 13289 13290 13291 13292 13293 13294 13295 13296 13297 13298 13299 13300 13301 13302 13303 13304 13305 13306 13307 13308 13309 13310 13311 13312 13313 13314 13315 13316 13317 13318 13319 13320 13321 13322 13323 13324 13325 13326 13327 13328 13329 13330 13331 13332 13333 13334 13335 13336 13337 13338 13339 13340 13341 13342 13343 13344 13345 13346 13347 13348 13349 13350 13351 13352 13353 13354 13355 13356 13357 13358 13359 13360 13361 13362 13363 13364 13365 13366 13367 13368 13369 13370 13371 13372 13373 13374
|
# Copyright 2010 United States Government as represented by the
# Administrator of the National Aeronautics and Space Administration.
# All Rights Reserved.
# Copyright (c) 2010 Citrix Systems, Inc.
# Copyright (c) 2011 Piston Cloud Computing, Inc
# Copyright (c) 2012 University Of Minho
# (c) Copyright 2013 Hewlett-Packard Development Company, L.P.
#
# Licensed under the Apache License, Version 2.0 (the "License"); you may
# not use this file except in compliance with the License. You may obtain
# a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS, WITHOUT
# WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the
# License for the specific language governing permissions and limitations
# under the License.
"""
A connection to a hypervisor through libvirt.
Supports KVM, LXC, QEMU, and Parallels.
"""
import binascii
import collections
from collections import deque
import contextlib
import copy
import errno
import functools
import glob
import grp
import itertools
import operator
import os
import pwd
import random
import shutil
import sys
import tempfile
import threading
import time
import typing as ty
import uuid
from castellan import key_manager
from copy import deepcopy
import eventlet
from eventlet import greenthread
from eventlet import tpool
from lxml import etree
from os_brick import encryptors
from os_brick.encryptors import luks as luks_encryptor
from os_brick import exception as brick_exception
from os_brick.initiator import connector
from os_brick.initiator import linuxscsi
import os_resource_classes as orc
import os_traits as ot
from oslo_concurrency import processutils
from oslo_log import log as logging
from oslo_serialization import base64
from oslo_serialization import jsonutils
from oslo_service import loopingcall
from oslo_utils import excutils
from oslo_utils import fileutils
from oslo_utils import importutils
from oslo_utils import netutils as oslo_netutils
from oslo_utils import strutils
from oslo_utils import timeutils
from oslo_utils import units
from oslo_utils import uuidutils
from nova.api.metadata import base as instance_metadata
from nova.api.metadata import password
from nova import block_device
from nova.compute import power_state
from nova.compute import provider_tree
from nova.compute import task_states
from nova.compute import utils as compute_utils
from nova.compute import vm_states
import nova.conf
from nova.console import serial as serial_console
from nova.console import type as ctype
from nova import context as nova_context
from nova import crypto
from nova.db import constants as db_const
from nova import exception
from nova.i18n import _
from nova.image import glance
from nova.network import model as network_model
from nova.network import neutron
from nova import objects
from nova.objects import diagnostics as diagnostics_obj
from nova.objects import fields
from nova.objects import migrate_data as migrate_data_obj
from nova.pci import utils as pci_utils
from nova.pci import whitelist
import nova.privsep.libvirt
import nova.privsep.path
import nova.privsep.utils
from nova.storage import rbd_utils
from nova import utils
from nova import version
from nova.virt import block_device as driver_block_device
from nova.virt import configdrive
from nova.virt.disk import api as disk_api
from nova.virt.disk.vfs import guestfs
from nova.virt import driver
from nova.virt import event as virtevent
from nova.virt import hardware
from nova.virt.image import model as imgmodel
from nova.virt import images
from nova.virt.libvirt import blockinfo
from nova.virt.libvirt import config as vconfig
from nova.virt.libvirt.cpu import api as libvirt_cpu
from nova.virt.libvirt import designer
from nova.virt.libvirt import event as libvirtevent
from nova.virt.libvirt import guest as libvirt_guest
from nova.virt.libvirt import host
from nova.virt.libvirt import imagebackend
from nova.virt.libvirt import imagecache
from nova.virt.libvirt import instancejobtracker
from nova.virt.libvirt import migration as libvirt_migrate
from nova.virt.libvirt.storage import dmcrypt
from nova.virt.libvirt.storage import lvm
from nova.virt.libvirt import utils as libvirt_utils
from nova.virt.libvirt import vif as libvirt_vif
from nova.virt.libvirt.volume import cephfs
from nova.virt.libvirt.volume import fs
from nova.virt.libvirt.volume import mount
from nova.virt.libvirt.volume import nfs
from nova.virt.libvirt.volume import remotefs
from nova.virt.libvirt.volume import volume
from nova.virt import netutils
from nova.volume import cinder
libvirt: ty.Any = None
uefi_logged = False
LOG = logging.getLogger(__name__)
CONF = nova.conf.CONF
MAX_CONSOLE_BYTES = 100 * units.Ki
VALID_DISK_CACHEMODES = [
"default", "none", "writethrough", "writeback", "directsync", "unsafe",
]
# The libvirt driver will prefix any disable reason codes with this string.
DISABLE_PREFIX = 'AUTO: '
# Disable reason for the service which was enabled or disabled without reason
DISABLE_REASON_UNDEFINED = None
# Guest config console string
CONSOLE = "console=tty0 console=ttyS0 console=hvc0"
GuestNumaConfig = collections.namedtuple(
'GuestNumaConfig', ['cpuset', 'cputune', 'numaconfig', 'numatune'])
class InjectionInfo(collections.namedtuple(
'InjectionInfo', ['network_info', 'files', 'admin_pass'])):
__slots__ = ()
def __repr__(self):
return ('InjectionInfo(network_info=%r, files=%r, '
'admin_pass=<SANITIZED>)') % (self.network_info, self.files)
# NOTE(lyarwood): Dict of volume drivers supported by the libvirt driver, keyed
# by the connection_info['driver_volume_type'] returned by Cinder for each
# volume type it supports
# TODO(lyarwood): Add host configurables to allow this list to be changed.
# Allowing native iSCSI to be reintroduced etc.
VOLUME_DRIVERS = {
'iscsi': 'nova.virt.libvirt.volume.iscsi.LibvirtISCSIVolumeDriver',
'iser': 'nova.virt.libvirt.volume.iser.LibvirtISERVolumeDriver',
'local': 'nova.virt.libvirt.volume.volume.LibvirtVolumeDriver',
'fake': 'nova.virt.libvirt.volume.volume.LibvirtFakeVolumeDriver',
'rbd': 'nova.virt.libvirt.volume.net.LibvirtNetVolumeDriver',
'nfs': 'nova.virt.libvirt.volume.nfs.LibvirtNFSVolumeDriver',
'smbfs': 'nova.virt.libvirt.volume.smbfs.LibvirtSMBFSVolumeDriver',
'fibre_channel': 'nova.virt.libvirt.volume.fibrechannel.LibvirtFibreChannelVolumeDriver', # noqa:E501
'gpfs': 'nova.virt.libvirt.volume.gpfs.LibvirtGPFSVolumeDriver',
'quobyte': 'nova.virt.libvirt.volume.quobyte.LibvirtQuobyteVolumeDriver',
'scaleio': 'nova.virt.libvirt.volume.scaleio.LibvirtScaleIOVolumeDriver',
'vzstorage': 'nova.virt.libvirt.volume.vzstorage.LibvirtVZStorageVolumeDriver', # noqa:E501
'storpool': 'nova.virt.libvirt.volume.storpool.LibvirtStorPoolVolumeDriver', # noqa:E501
'nvmeof': 'nova.virt.libvirt.volume.nvme.LibvirtNVMEVolumeDriver',
'lightos': 'nova.virt.libvirt.volume.lightos.LibvirtLightOSVolumeDriver',
}
def patch_tpool_proxy():
"""eventlet.tpool.Proxy doesn't work with old-style class in __str__()
or __repr__() calls. See bug #962840 for details.
We perform a monkey patch to replace those two instance methods.
"""
def str_method(self):
return str(self._obj)
def repr_method(self):
return repr(self._obj)
tpool.Proxy.__str__ = str_method
tpool.Proxy.__repr__ = repr_method
patch_tpool_proxy()
# For information about when MIN_{LIBVIRT,QEMU}_VERSION and
# NEXT_MIN_{LIBVIRT,QEMU}_VERSION can be changed, consult the following:
#
# doc/source/reference/libvirt-distro-support-matrix.rst
#
# DO NOT FORGET to update this document when touching any versions below!
MIN_LIBVIRT_VERSION = (8, 0, 0)
MIN_QEMU_VERSION = (6, 2, 0)
NEXT_MIN_LIBVIRT_VERSION = (10, 0, 0)
NEXT_MIN_QEMU_VERSION = (8, 2, 2)
# vIOMMU model value `virtio` minimal support version
MIN_LIBVIRT_VIOMMU_VIRTIO_MODEL = (8, 3, 0)
MIN_LIBVIRT_TB_CACHE_SIZE = (8, 0, 0)
# Virtuozzo driver support
MIN_VIRTUOZZO_VERSION = (7, 0, 0)
# Names of the types that do not get compressed during migration
NO_COMPRESSION_TYPES = ('qcow2',)
# number of serial console limit
QEMU_MAX_SERIAL_PORTS = 4
# Qemu supports 4 serial consoles, we remove 1 because of the PTY one defined
ALLOWED_QEMU_SERIAL_PORTS = QEMU_MAX_SERIAL_PORTS - 1
VGPU_RESOURCE_SEMAPHORE = 'vgpu_resources'
# Minimum versions supporting mdev live-migration.
MIN_MDEV_LIVEMIG_LIBVIRT_VERSION = (8, 6, 0)
MIN_MDEV_LIVEMIG_QEMU_VERSION = (8, 1, 0)
# Minimum version supporting persistent mdevs.
# https://libvirt.org/drvnodedev.html#mediated-devices-mdevs
MIN_LIBVIRT_PERSISTENT_MDEV = (7, 3, 0)
# Autostart appears to be available starting in 7.8.0
# https://github.com/libvirt/libvirt/commit/c6607a25b93bd6b0188405785d6608fdf71c8e0a
MIN_LIBVIRT_NODEDEV_AUTOSTART = (7, 8, 0)
LIBVIRT_PERF_EVENT_PREFIX = 'VIR_PERF_PARAM_'
# Maxphysaddr minimal support version.
MIN_LIBVIRT_MAXPHYSADDR = (8, 7, 0)
MIN_QEMU_MAXPHYSADDR = (2, 7, 0)
# stateless firmware support
MIN_LIBVIRT_STATELESS_FIRMWARE = (8, 6, 0)
# Minimum versions supporting igb hw_vif_model
MIN_IGB_LIBVIRT_VERSION = (9, 3, 0)
MIN_IGB_QEMU_VERSION = (8, 0, 0)
# Minimum versions supporting vfio-pci variant driver.
MIN_VFIO_PCI_VARIANT_LIBVIRT_VERSION = (10, 0, 0)
MIN_VFIO_PCI_VARIANT_QEMU_VERSION = (8, 2, 2)
REGISTER_IMAGE_PROPERTY_DEFAULTS = [
'hw_machine_type',
'hw_cdrom_bus',
'hw_disk_bus',
'hw_input_bus',
'hw_pointer_model',
'hw_video_model',
'hw_vif_model',
]
class AsyncDeviceEventsHandler:
"""A synchornization point between libvirt events an clients waiting for
such events.
It provides an interface for the clients to wait for one or more libvirt
event types. It implements event delivery by expecting the libvirt driver
to forward libvirt specific events to notify_waiters()
It handles multiple clients for the same instance, device and event
type and delivers the event to each clients.
"""
class Waiter:
def __init__(
self,
instance_uuid: str,
device_name: str,
event_types: ty.Set[ty.Type[libvirtevent.DeviceEvent]]
):
self.instance_uuid = instance_uuid
self.device_name = device_name
self.event_types = event_types
self.threading_event = threading.Event()
self.result: ty.Optional[libvirtevent.DeviceEvent] = None
def matches(self, event: libvirtevent.DeviceEvent) -> bool:
"""Returns true if the event is one of the expected event types
for the given instance and device.
"""
return (
self.instance_uuid == event.uuid and
self.device_name == event.dev and
isinstance(event, tuple(self.event_types)))
def __repr__(self) -> str:
return (
"AsyncDeviceEventsHandler.Waiter("
f"instance_uuid={self.instance_uuid}, "
f"device_name={self.device_name}, "
f"event_types={self.event_types})")
def __init__(self):
self._lock = threading.Lock()
# Ongoing device operations in libvirt where we wait for the events
# about success or failure.
self._waiters: ty.Set[AsyncDeviceEventsHandler.Waiter] = set()
def create_waiter(
self,
instance_uuid: str,
device_name: str,
event_types: ty.Set[ty.Type[libvirtevent.DeviceEvent]]
) -> 'AsyncDeviceEventsHandler.Waiter':
"""Returns an opaque token the caller can use in wait() to
wait for the libvirt event
:param instance_uuid: The UUID of the instance.
:param device_name: The device name alias used by libvirt for this
device.
:param event_type: A set of classes derived from DeviceEvent
specifying which event types the caller waits for. Specifying more
than one event type means waiting for either of the events to be
received.
:returns: an opaque token to be used with wait_for_event().
"""
waiter = AsyncDeviceEventsHandler.Waiter(
instance_uuid, device_name, event_types)
with self._lock:
self._waiters.add(waiter)
return waiter
def delete_waiter(self, token: 'AsyncDeviceEventsHandler.Waiter'):
"""Deletes the waiter
:param token: the opaque token returned by create_waiter() to be
deleted
"""
with self._lock:
self._waiters.remove(token)
def wait(
self, token: 'AsyncDeviceEventsHandler.Waiter', timeout: float,
) -> ty.Optional[libvirtevent.DeviceEvent]:
"""Blocks waiting for the libvirt event represented by the opaque token
:param token: A token created by calling create_waiter()
:param timeout: Maximum number of seconds this call blocks waiting for
the event to be received
:returns: The received libvirt event, or None in case of timeout
"""
token.threading_event.wait(timeout)
with self._lock:
self._waiters.remove(token)
return token.result
def notify_waiters(self, event: libvirtevent.DeviceEvent) -> bool:
"""Unblocks the client waiting for this event.
:param event: the libvirt event that is received
:returns: True if there was a client waiting and False otherwise.
"""
dispatched = False
with self._lock:
for waiter in self._waiters:
if waiter.matches(event):
waiter.result = event
waiter.threading_event.set()
dispatched = True
return dispatched
def cleanup_waiters(self, instance_uuid: str) -> None:
"""Deletes all waiters and unblock all clients related to the specific
instance.
param instance_uuid: The instance UUID for which the cleanup is
requested
"""
with self._lock:
instance_waiters = set()
for waiter in self._waiters:
if waiter.instance_uuid == instance_uuid:
# unblock any waiting thread
waiter.threading_event.set()
instance_waiters.add(waiter)
self._waiters -= instance_waiters
if instance_waiters:
LOG.debug(
'Cleaned up device related libvirt event waiters: %s',
instance_waiters)
class LibvirtDriver(driver.ComputeDriver):
def __init__(self, virtapi, read_only=False):
# NOTE(aspiers) Some of these are dynamic, so putting
# capabilities on the instance rather than on the class.
# This prevents the risk of one test setting a capability
# which bleeds over into other tests.
# LVM and RBD require raw images. If we are not configured to
# force convert images into raw format, then we _require_ raw
# images only.
raw_only = ('rbd', 'lvm')
requires_raw_image = (CONF.libvirt.images_type in raw_only and
not CONF.force_raw_images)
requires_ploop_image = CONF.libvirt.virt_type == 'parallels'
self.image_backend = imagebackend.Backend(CONF.use_cow_images)
self.capabilities = {
"has_imagecache": True,
"supports_evacuate": True,
"supports_migrate_to_same_host": False,
"supports_attach_interface": True,
"supports_device_tagging": True,
"supports_tagged_attach_interface": True,
"supports_tagged_attach_volume": True,
"supports_extend_volume": True,
"supports_multiattach": True,
"supports_trusted_certs": True,
# Supported image types
"supports_image_type_aki": True,
"supports_image_type_ari": True,
"supports_image_type_ami": True,
"supports_image_type_raw": True,
"supports_image_type_iso": True,
# NOTE(danms): Certain backends do not work with complex image
# formats. If we are configured for those backends, then we
# should not expose the corresponding support traits.
"supports_image_type_qcow2": not requires_raw_image,
"supports_image_type_ploop": requires_ploop_image,
"supports_pcpus": True,
"supports_accelerators": True,
"supports_bfv_rescue": True,
"supports_vtpm": CONF.libvirt.swtpm_enabled,
"supports_socket_pci_numa_affinity": True,
"supports_ephemeral_encryption":
self.image_backend.backend().SUPPORTS_LUKS,
"supports_ephemeral_encryption_luks":
self.image_backend.backend().SUPPORTS_LUKS,
}
super(LibvirtDriver, self).__init__(virtapi)
if not sys.platform.startswith('linux'):
raise exception.InternalError(
_('The libvirt driver only works on Linux'))
global libvirt
if libvirt is None:
libvirt = importutils.import_module('libvirt')
libvirt_migrate.libvirt = libvirt
self._host = host.Host(self._uri(), read_only,
lifecycle_event_handler=self.emit_event,
conn_event_handler=self._handle_conn_event)
self._supported_perf_events = []
self.vif_driver = libvirt_vif.LibvirtGenericVIFDriver(self._host)
# NOTE(lyarwood): Volume drivers are loaded on-demand
self.volume_drivers: ty.Dict[str, volume.LibvirtBaseVolumeDriver] = {}
self._disk_cachemode = None
self.image_cache_manager = imagecache.ImageCacheManager()
self.disk_cachemodes = {}
for mode_str in CONF.libvirt.disk_cachemodes:
disk_type, sep, cache_mode = mode_str.partition('=')
if cache_mode not in VALID_DISK_CACHEMODES:
LOG.warning('Invalid cachemode %(cache_mode)s specified '
'for disk type %(disk_type)s.',
{'cache_mode': cache_mode, 'disk_type': disk_type})
continue
self.disk_cachemodes[disk_type] = cache_mode
self._volume_api = cinder.API()
self._image_api = glance.API()
self._network_api = neutron.API()
# The default choice for the sysinfo_serial config option is "unique"
# which does not have a special function since the value is just the
# instance.uuid.
sysinfo_serial_funcs = {
'none': lambda: None,
'hardware': self._get_host_sysinfo_serial_hardware,
'os': self._get_host_sysinfo_serial_os,
'auto': self._get_host_sysinfo_serial_auto,
}
self._sysinfo_serial_func = sysinfo_serial_funcs.get(
CONF.libvirt.sysinfo_serial, lambda: None)
self.job_tracker = instancejobtracker.InstanceJobTracker()
self._remotefs = remotefs.RemoteFilesystem()
self._live_migration_flags = self._block_migration_flags = 0
self.active_migrations = {}
# Compute reserved hugepages from conf file at the very
# beginning to ensure any syntax error will be reported and
# avoid any re-calculation when computing resources.
self._reserved_hugepages = hardware.numa_get_reserved_huge_pages()
# Copy of the compute service ProviderTree object that is updated
# every time update_provider_tree() is called.
# NOTE(sbauza): We only want a read-only cache, this attribute is not
# intended to be updatable directly
self.provider_tree: provider_tree.ProviderTree = None
# driver traits will not change during the runtime of the agent
# so calculate them once and save them
self._static_traits = None
# The CPU models in the configuration are case-insensitive, but the CPU
# model in the libvirt is case-sensitive, therefore create a mapping to
# map the lower case CPU model name to normal CPU model name.
self.cpu_models_mapping = {}
self.cpu_model_flag_mapping = {}
self._vpmems_by_name, self._vpmems_by_rc = self._discover_vpmems(
vpmem_conf=CONF.libvirt.pmem_namespaces)
# We default to not support vGPUs unless the configuration is set.
self.pgpu_type_mapping = collections.defaultdict(str)
# This dict is for knowing which mdev class is supported by a specific
# PCI device like we do (the key being the PCI address and the value
# the mdev class)
self.mdev_class_mapping: ty.Dict[str, str] = (
collections.defaultdict(lambda: orc.VGPU)
)
# This set is for knowing all the mdev classes the operator provides
self.mdev_classes = set([])
# this is for knowing how many mdevs can be created by a type
self.mdev_type_max_mapping = collections.defaultdict(str)
# if we have a wildcard, we default to use this mdev type
self.pgpu_type_default = None
self.supported_vgpu_types = self._get_supported_vgpu_types()
# This dict is for knowing which mdevs are already claimed by some
# instance. This is keyed by instance UUID and the value is a list
# of mediated device UUIDs.
self.instance_claimed_mdevs = {}
# Handles ongoing device manipultion in libvirt where we wait for the
# events about success or failure.
self._device_event_handler = AsyncDeviceEventsHandler()
# NOTE(artom) From a pure functionality point of view, there's no need
# for this to be an attribute of self. However, we want to test power
# management in multinode scenarios (ex: live migration) in our
# functional tests. If the power management code was just a bunch of
# module level functions, the functional tests would not be able to
# distinguish between cores on the source and destination hosts.
# See also nova.virt.libvirt.cpu.api.API.core().
self.cpu_api = libvirt_cpu.API()
def _discover_vpmems(self, vpmem_conf=None):
"""Discover vpmems on host and configuration.
:param vpmem_conf: pmem namespaces configuration from CONF
:returns: a dict of vpmem keyed by name, and
a dict of vpmem list keyed by resource class
:raises: exception.InvalidConfiguration if Libvirt or QEMU version
does not meet requirement.
"""
if not vpmem_conf:
return {}, {}
# vpmem keyed by name {name: objects.LibvirtVPMEMDevice,...}
vpmems_by_name: ty.Dict[str, 'objects.LibvirtVPMEMDevice'] = {}
# vpmem list keyed by resource class
# {'RC_0': [objects.LibvirtVPMEMDevice, ...], 'RC_1': [...]}
vpmems_by_rc: ty.Dict[str, ty.List['objects.LibvirtVPMEMDevice']] = (
collections.defaultdict(list)
)
vpmems_host = self._get_vpmems_on_host()
for ns_conf in vpmem_conf:
try:
ns_label, ns_names = ns_conf.split(":", 1)
except ValueError:
reason = _("The configuration doesn't follow the format")
raise exception.PMEMNamespaceConfigInvalid(
reason=reason)
ns_names = ns_names.split("|")
for ns_name in ns_names:
if ns_name not in vpmems_host:
reason = _("The PMEM namespace %s isn't on host") % ns_name
raise exception.PMEMNamespaceConfigInvalid(
reason=reason)
if ns_name in vpmems_by_name:
reason = (_("Duplicated PMEM namespace %s configured") %
ns_name)
raise exception.PMEMNamespaceConfigInvalid(
reason=reason)
pmem_ns_updated = vpmems_host[ns_name]
pmem_ns_updated.label = ns_label
vpmems_by_name[ns_name] = pmem_ns_updated
rc = orc.normalize_name(
"PMEM_NAMESPACE_%s" % ns_label)
vpmems_by_rc[rc].append(pmem_ns_updated)
return vpmems_by_name, vpmems_by_rc
def _get_vpmems_on_host(self):
"""Get PMEM namespaces on host using ndctl utility."""
try:
output = nova.privsep.libvirt.get_pmem_namespaces()
except Exception as e:
reason = _("Get PMEM namespaces by ndctl utility, "
"please ensure ndctl is installed: %s") % e
raise exception.GetPMEMNamespacesFailed(reason=reason)
if not output:
return {}
namespaces = jsonutils.loads(output)
vpmems_host = {} # keyed by namespace name
for ns in namespaces:
# store namespace info parsed from ndctl utility return
if not ns.get('name'):
# The name is used to identify namespaces, it's optional
# config when creating namespace. If an namespace don't have
# name, it can not be used by Nova, we will skip it.
continue
vpmems_host[ns['name']] = objects.LibvirtVPMEMDevice(
name=ns['name'],
devpath= '/dev/' + ns['daxregion']['devices'][0]['chardev'],
size=ns['size'],
align=ns['daxregion']['align'])
return vpmems_host
@property
def disk_cachemode(self):
# It can be confusing to understand the QEMU cache mode
# behaviour, because each cache=$MODE is a convenient shorthand
# to toggle _three_ cache.* booleans. Consult the below table
# (quoting from the QEMU man page):
#
# | cache.writeback | cache.direct | cache.no-flush
# --------------------------------------------------------------
# writeback | on | off | off
# none | on | on | off
# writethrough | off | off | off
# directsync | off | on | off
# unsafe | on | off | on
#
# Where:
#
# - 'cache.writeback=off' means: QEMU adds an automatic fsync()
# after each write request.
#
# - 'cache.direct=on' means: Use Linux's O_DIRECT, i.e. bypass
# the kernel page cache. Caches in any other layer (disk
# cache, QEMU metadata caches, etc.) can still be present.
#
# - 'cache.no-flush=on' means: Ignore flush requests, i.e.
# never call fsync(), even if the guest explicitly requested
# it.
#
# Use cache mode "none" (cache.writeback=on, cache.direct=on,
# cache.no-flush=off) for consistent performance and
# migration correctness. Some filesystems don't support
# O_DIRECT, though. For those we fallback to the next
# reasonable option that is "writeback" (cache.writeback=on,
# cache.direct=off, cache.no-flush=off).
if self._disk_cachemode is None:
self._disk_cachemode = "none"
if not nova.privsep.utils.supports_direct_io(CONF.instances_path):
self._disk_cachemode = "writeback"
return self._disk_cachemode
def _set_cache_mode(self, conf):
"""Set cache mode on LibvirtConfigGuestDisk object."""
try:
source_type = conf.source_type
driver_cache = conf.driver_cache
except AttributeError:
return
# Shareable disks like for a multi-attach volume need to have the
# driver cache disabled.
if getattr(conf, 'shareable', False):
conf.driver_cache = 'none'
else:
cache_mode = self.disk_cachemodes.get(source_type,
driver_cache)
conf.driver_cache = cache_mode
# NOTE(acewit): If the [libvirt]disk_cachemodes is set as
# `block=writeback` or `block=writethrough` or `block=unsafe`,
# whose corresponding Linux's IO semantic is not O_DIRECT in
# file nova.conf, then it will result in an attachment failure
# because of the libvirt bug
# (https://bugzilla.redhat.com/show_bug.cgi?id=1086704)
if ((getattr(conf, 'driver_io', None) == "native") and
conf.driver_cache not in [None, 'none', 'directsync']):
conf.driver_io = "threads"
LOG.warning("The guest disk driver io mode has fallen back "
"from 'native' to 'threads' because the "
"disk cache mode is set as %(cachemode)s, which does "
"not use O_DIRECT. See the following bug report "
"for more details: https://launchpad.net/bugs/1841363",
{'cachemode': conf.driver_cache})
def _do_quality_warnings(self):
"""Warn about potential configuration issues.
This will log a warning message for things such as untested driver or
host arch configurations in order to indicate potential issues to
administrators.
"""
if CONF.libvirt.virt_type not in ('qemu', 'kvm'):
LOG.warning(
"Support for the '%(type)s' libvirt backend has been "
"deprecated and will be removed in a future release.",
{'type': CONF.libvirt.virt_type},
)
caps = self._host.get_capabilities()
hostarch = caps.host.cpu.arch
if hostarch not in (
fields.Architecture.I686, fields.Architecture.X86_64,
):
LOG.warning(
'The libvirt driver is not tested on %(arch)s by the '
'OpenStack project and thus its quality can not be ensured. '
'For more information, see: https://docs.openstack.org/'
'nova/latest/user/support-matrix.html',
{'arch': hostarch},
)
def _handle_conn_event(self, enabled, reason):
LOG.info("Connection event '%(enabled)d' reason '%(reason)s'",
{'enabled': enabled, 'reason': reason})
self._set_host_enabled(enabled, reason)
def _init_host_topology(self):
"""To work around a bug in libvirt that reports offline CPUs as always
being on socket 0 regardless of their real socket, power up all
dedicated CPUs (the only ones whose socket we actually care about),
then call get_capabilities() to initialize the topology with the
correct socket values. get_capabilities()'s implementation will reuse
these initial socket value, and avoid clobbering them with 0 for
offline CPUs.
"""
cpus = hardware.get_cpu_dedicated_set()
if cpus:
self.cpu_api.power_up(cpus)
self._host.get_capabilities()
def init_host(self, host):
self._host.initialize()
# NOTE(artom) Do this first to make sure our first call to
# get_capabilities() happens with all dedicated CPUs online and caches
# their correct socket ID. Unused dedicated CPUs will be powered down
# further down in this method.
self._check_cpu_set_configuration()
self._init_host_topology()
self._update_host_specific_capabilities()
self._do_quality_warnings()
self._parse_migration_flags()
self._supported_perf_events = self._get_supported_perf_events()
self._check_my_ip()
# TODO(ykarel) This can be dropped when MIN_LIBVIRT_VERSION>=8.0.0
self._supports_tb_cache_size()
if (CONF.libvirt.virt_type == 'lxc' and
not (CONF.libvirt.uid_maps and CONF.libvirt.gid_maps)):
LOG.warning("Running libvirt-lxc without user namespaces is "
"dangerous. Containers spawned by Nova will be run "
"as the host's root user. It is highly suggested "
"that user namespaces be used in a public or "
"multi-tenant environment.")
# Stop libguestfs using KVM unless we're also configured
# to use this. This solves problem where people need to
# stop Nova use of KVM because nested-virt is broken
if CONF.libvirt.virt_type != "kvm":
guestfs.force_tcg()
if not self._host.has_min_version(MIN_LIBVIRT_VERSION):
raise exception.InternalError(
_('Nova requires libvirt version %s or greater.') %
libvirt_utils.version_to_string(MIN_LIBVIRT_VERSION))
if CONF.libvirt.virt_type in ("qemu", "kvm"):
if not self._host.has_min_version(hv_ver=MIN_QEMU_VERSION):
raise exception.InternalError(
_('Nova requires QEMU version %s or greater.') %
libvirt_utils.version_to_string(MIN_QEMU_VERSION))
if CONF.libvirt.virt_type == 'parallels':
if not self._host.has_min_version(hv_ver=MIN_VIRTUOZZO_VERSION):
raise exception.InternalError(
_('Nova requires Virtuozzo version %s or greater.') %
libvirt_utils.version_to_string(MIN_VIRTUOZZO_VERSION))
# Give the cloud admin a heads up if we are intending to
# change the MIN_LIBVIRT_VERSION in the next release.
if not self._host.has_min_version(NEXT_MIN_LIBVIRT_VERSION):
LOG.warning('Running Nova with a libvirt version less than '
'%(version)s is deprecated. The required minimum '
'version of libvirt will be raised to %(version)s '
'in the next release.',
{'version': libvirt_utils.version_to_string(
NEXT_MIN_LIBVIRT_VERSION)})
if (CONF.libvirt.virt_type in ("qemu", "kvm") and
not self._host.has_min_version(hv_ver=NEXT_MIN_QEMU_VERSION)):
LOG.warning('Running Nova with a QEMU version less than '
'%(version)s is deprecated. The required minimum '
'version of QEMU will be raised to %(version)s '
'in the next release.',
{'version': libvirt_utils.version_to_string(
NEXT_MIN_QEMU_VERSION)})
# Allowing both "tunnelling via libvirtd" (which will be
# deprecated once the MIN_{LIBVIRT,QEMU}_VERSION is sufficiently
# new enough) and "native TLS" options at the same time is
# nonsensical.
if (CONF.libvirt.live_migration_tunnelled and
CONF.libvirt.live_migration_with_native_tls):
msg = _("Setting both 'live_migration_tunnelled' and "
"'live_migration_with_native_tls' at the same "
"time is invalid. If you have the relevant "
"libvirt and QEMU versions, and TLS configured "
"in your environment, pick "
"'live_migration_with_native_tls'.")
raise exception.Invalid(msg)
# Some imagebackends are only able to import raw disk images,
# and will fail if given any other format. See the bug
# https://bugs.launchpad.net/nova/+bug/1816686 for more details.
if CONF.libvirt.images_type in ('rbd',):
if not CONF.force_raw_images:
msg = _("'[DEFAULT]/force_raw_images = False' is not "
"allowed with '[libvirt]/images_type = rbd'. "
"Please check the two configs and if you really "
"do want to use rbd as images_type, set "
"force_raw_images to True.")
raise exception.InvalidConfiguration(msg)
# NOTE(sbauza): We verify first if the dedicated CPU performances were
# modified by Nova before. Note that it can provide an exception if
# either the governor strategies are different between the cores or if
# the cores are offline.
self.cpu_api.validate_all_dedicated_cpus()
# NOTE(sbauza): We powerdown all dedicated CPUs but if some instances
# exist that are pinned for some CPUs, then we'll later powerup those
# CPUs when rebooting the instance in _init_instance()
# Note that it can provide an exception if the config options are
# wrongly modified.
self.cpu_api.power_down_all_dedicated_cpus()
if not self._host.has_min_version(MIN_LIBVIRT_PERSISTENT_MDEV):
# TODO(sbauza): Remove this code once mediated devices are
# persisted across reboots.
# TODO(Uggla): Remove in bump cleanup patch
self._recreate_assigned_mediated_devices()
else:
# NOTE(melwitt): We shouldn't need to do this with libvirt 7.8.0
# and newer because we're setting autostart=True on the devices --
# but if that fails for whatever reason and any devices become
# inactive, we can start them here. With libvirt version < 7.8.0,
# this is needed because autostart is not available.
self._start_inactive_mediated_devices()
self._check_cpu_compatibility()
self._check_vtpm_support()
self._check_multipath()
# Even if we already checked the whitelist at startup, this driver
# needs to check specific hypervisor versions
self._check_pci_whitelist()
# Set REGISTER_IMAGE_PROPERTY_DEFAULTS in the instance system_metadata
# to default values for properties that have not already been set.
self._register_all_undefined_instance_details()
def _check_pci_whitelist(self):
need_specific_version = False
if CONF.pci.device_spec:
pci_whitelist = whitelist.Whitelist(CONF.pci.device_spec)
for spec in pci_whitelist.specs:
if spec.tags.get("live_migratable"):
need_specific_version = True
if need_specific_version and not self._host.has_min_version(
lv_ver=MIN_VFIO_PCI_VARIANT_LIBVIRT_VERSION,
hv_ver=MIN_VFIO_PCI_VARIANT_QEMU_VERSION,
hv_type=host.HV_DRIVER_QEMU,
):
msg = _(
"PCI device spec is configured for "
"live_migratable but it's not supported by libvirt."
)
raise exception.InvalidConfiguration(msg)
def _update_host_specific_capabilities(self) -> None:
"""Update driver capabilities based on capabilities of the host."""
# TODO(stephenfin): We should also be reporting e.g. SEV functionality
# or UEFI bootloader support in this manner
self.capabilities.update({
'supports_secure_boot': self._host.supports_secure_boot,
'supports_remote_managed_ports':
self._host.supports_remote_managed_ports,
'supports_virtio_fs': self._host.supports_virtio_fs,
'supports_mem_backing_file': self._host.supports_mem_backing_file
})
supports_maxphysaddr = self._host.has_min_version(
lv_ver=MIN_LIBVIRT_MAXPHYSADDR,
hv_ver=MIN_QEMU_MAXPHYSADDR,
hv_type=host.HV_DRIVER_QEMU,
)
# NOTE(nmiki): Currently libvirt does not provide a distinction
# between passthrough mode and emulated mode support status.
self.capabilities.update({
'supports_address_space_passthrough': supports_maxphysaddr,
'supports_address_space_emulated': supports_maxphysaddr,
})
supports_stateless_firmware = self._host.has_min_version(
lv_ver=MIN_LIBVIRT_STATELESS_FIRMWARE,
)
self.capabilities.update({
'supports_stateless_firmware': supports_stateless_firmware,
})
def _register_all_undefined_instance_details(self) -> None:
"""Register the default image properties of instances on this host
For each instance found on this host by InstanceList.get_by_host ensure
REGISTER_IMAGE_PROPERTY_DEFAULTS are registered within the system
metadata of the instance
"""
context = nova_context.get_admin_context()
hostname = self._host.get_hostname()
for instance in objects.InstanceList.get_by_host(
context, hostname, expected_attrs=['flavor', 'system_metadata']
):
try:
self._register_undefined_instance_details(context, instance)
except Exception:
LOG.exception('Ignoring unknown failure while attempting '
'to save the defaults for unregistered image '
'properties', instance=instance)
def _register_undefined_instance_details(
self,
context: nova_context.RequestContext,
instance: 'objects.Instance',
) -> None:
# Find any unregistered image properties against this instance
unregistered_image_props = [
p for p in REGISTER_IMAGE_PROPERTY_DEFAULTS
if f"image_{p}" not in instance.system_metadata
]
# Return if there's nothing left to register for this instance
if not unregistered_image_props:
return
LOG.debug(f'Attempting to register defaults for the following '
f'image properties: {unregistered_image_props}',
instance=instance)
# NOTE(lyarwood): Only build disk_info once per instance if we need it
# for hw_{disk,cdrom}_bus to avoid pulling bdms from the db etc.
requires_disk_info = ['hw_disk_bus', 'hw_cdrom_bus']
disk_info = None
if set(requires_disk_info) & set(unregistered_image_props):
bdms = objects.BlockDeviceMappingList.get_by_instance_uuid(
context, instance.uuid)
block_device_info = driver.get_block_device_info(instance, bdms)
disk_info = blockinfo.get_disk_info(
CONF.libvirt.virt_type, instance, instance.image_meta,
block_device_info)
# Only pull the guest config once per instance if we need it for
# hw_pointer_model or hw_input_bus.
requires_guest_config = ['hw_pointer_model', 'hw_input_bus']
guest_config = None
if set(requires_guest_config) & set(unregistered_image_props):
guest_config = self._host.get_guest(instance).get_config()
for image_prop in unregistered_image_props:
try:
default_value = self._find_default_for_image_property(
instance, image_prop, disk_info, guest_config)
instance.system_metadata[f"image_{image_prop}"] = default_value
LOG.debug(f'Found default for {image_prop} of {default_value}',
instance=instance)
except Exception:
LOG.exception(f'Ignoring unknown failure while attempting '
f'to find the default of {image_prop}',
instance=instance)
instance.save()
def _find_default_for_image_property(
self,
instance: 'objects.Instance',
image_property: str,
disk_info: ty.Optional[ty.Dict[str, ty.Any]],
guest_config: ty.Optional[vconfig.LibvirtConfigGuest],
) -> ty.Optional[str]:
if image_property == 'hw_machine_type':
return libvirt_utils.get_machine_type(instance.image_meta)
if image_property == 'hw_disk_bus' and disk_info:
return disk_info.get('disk_bus')
if image_property == 'hw_cdrom_bus' and disk_info:
return disk_info.get('cdrom_bus')
if image_property == 'hw_input_bus' and guest_config:
_, default_input_bus = self._get_pointer_bus_and_model(
guest_config, instance.image_meta)
return default_input_bus
if image_property == 'hw_pointer_model' and guest_config:
default_pointer_model, _ = self._get_pointer_bus_and_model(
guest_config, instance.image_meta)
# hw_pointer_model is of type PointerModelType ('usbtablet' instead
# of 'tablet')
if default_pointer_model == 'tablet':
default_pointer_model = 'usbtablet'
return default_pointer_model
if image_property == 'hw_video_model':
return self._get_video_type(instance.image_meta)
if image_property == 'hw_vif_model':
return self.vif_driver.get_vif_model(instance.image_meta)
return None
def _prepare_cpu_flag(self, flag):
# NOTE(kchamart) This helper method will be used while computing
# guest CPU compatibility. It will take into account a
# comma-separated list of CPU flags from
# `[libvirt]cpu_model_extra_flags`. If the CPU flag starts
# with '+', it is enabled for the guest; if it starts with '-',
# it is disabled. If neither '+' nor '-' is specified, the CPU
# flag is enabled.
if flag.startswith('-'):
flag = flag.lstrip('-')
policy_value = 'disable'
else:
flag = flag.lstrip('+')
policy_value = 'require'
cpu_feature = vconfig.LibvirtConfigGuestCPUFeature(
flag, policy=policy_value)
return cpu_feature
def _check_cpu_compatibility(self):
mode = CONF.libvirt.cpu_mode
models = CONF.libvirt.cpu_models
if (CONF.libvirt.virt_type not in ("kvm", "qemu") and
mode not in (None, 'none')):
msg = _("Config requested an explicit CPU model, but "
"the current libvirt hypervisor '%s' does not "
"support selecting CPU models") % CONF.libvirt.virt_type
raise exception.Invalid(msg)
if mode != "custom":
if not models:
return
msg = _("The cpu_models option is not required when "
"cpu_mode!=custom")
raise exception.Invalid(msg)
if not models:
msg = _("The cpu_models option is required when cpu_mode=custom")
raise exception.Invalid(msg)
if not CONF.workarounds.skip_cpu_compare_at_startup:
# Use guest CPU model to check the compatibility between
# guest CPU and configured extra_flags
for model in models:
cpu = vconfig.LibvirtConfigGuestCPU()
cpu.model = self._get_cpu_model_mapping(model)
for flag in set(x.lower() for
x in CONF.libvirt.cpu_model_extra_flags):
cpu_feature = self._prepare_cpu_flag(flag)
cpu.add_feature(cpu_feature)
try:
self._compare_cpu(cpu, self._get_cpu_info(), None)
except exception.InvalidCPUInfo as e:
msg = (_("Configured CPU model: %(model)s "
"and CPU Flags %(flags)s ar not "
"compatible with host CPU. Please correct your "
"config and try again. %(e)s") % {
'model': model, 'e': e,
'flags': CONF.libvirt.cpu_model_extra_flags})
raise exception.InvalidCPUInfo(msg)
def _check_vtpm_support(self) -> None:
# TODO(efried): A key manager must be configured to create/retrieve
# secrets. Is there a way to check that one is set up correctly?
# CONF.key_manager.backend is optional :(
if not CONF.libvirt.swtpm_enabled:
return
if CONF.libvirt.virt_type not in ('qemu', 'kvm'):
msg = _(
"vTPM support requires '[libvirt] virt_type' of 'qemu' or "
"'kvm'; found '%s'.")
raise exception.InvalidConfiguration(msg % CONF.libvirt.virt_type)
vtpm_support = self._host.supports_vtpm
if vtpm_support is not None:
# libvirt >= 8.0.0 presents availability of vTPM support and swtpm
# in domain capabilities
if not vtpm_support:
msg = _(
"vTPM support is configured but it's not supported by "
"libvirt.")
raise exception.InvalidConfiguration(msg)
else:
# These executables need to be installed for libvirt to make use of
# emulated TPM.
# NOTE(stephenfin): This checks using the PATH of the user running
# nova-compute rather than the libvirtd service, meaning it's an
# imperfect check but the best we can do
if not all(shutil.which(cmd) for cmd in (
'swtpm_ioctl', 'swtpm_setup', 'swtpm')):
msg = _(
"vTPM support is configured but some (or all) of "
"the 'swtpm', 'swtpm_setup' and 'swtpm_ioctl' binaries "
"could not be found on PATH.")
raise exception.InvalidConfiguration(msg)
# The user and group must be valid on this host for cold migration and
# resize to function.
try:
pwd.getpwnam(CONF.libvirt.swtpm_user)
except KeyError:
msg = _(
"The user configured in '[libvirt] swtpm_user' does not exist "
"on this host; expected '%s'.")
raise exception.InvalidConfiguration(msg % CONF.libvirt.swtpm_user)
try:
grp.getgrnam(CONF.libvirt.swtpm_group)
except KeyError:
msg = _(
"The group configured in '[libvirt] swtpm_group' does not "
"exist on this host; expected '%s'.")
raise exception.InvalidConfiguration(
msg % CONF.libvirt.swtpm_group)
LOG.debug('Enabling emulated TPM support')
def _check_multipath(self) -> None:
if not CONF.libvirt.volume_enforce_multipath:
return
if not CONF.libvirt.volume_use_multipath:
msg = _("The 'volume_use_multipath' option should be 'True' when "
"the 'volume_enforce_multipath' option is 'True'.")
raise exception.InvalidConfiguration(msg)
multipath_running = linuxscsi.LinuxSCSI.is_multipath_running(
root_helper=utils.get_root_helper())
if not multipath_running:
msg = _("The 'volume_enforce_multipath' option is 'True' but "
"multipathd is not running.")
raise exception.InvalidConfiguration(msg)
def _start_inactive_mediated_devices(self):
# Get a list of inactive mdevs so we can start them and make them
# active. We need to start inactive mdevs even if they are not
# currently assigned to instances because attempting to use an inactive
# mdev when booting a new instance, for example, will raise an error:
# libvirt.libvirtError: device not found: mediated device '<uuid>' not
# found.
# An inactive mdev is an mdev that is defined but not created.
flags = (
libvirt.VIR_CONNECT_LIST_NODE_DEVICES_CAP_MDEV |
libvirt.VIR_CONNECT_LIST_NODE_DEVICES_INACTIVE)
inactive_mdevs = self._host.list_all_devices(flags)
if inactive_mdevs:
names = [mdev.name() for mdev in inactive_mdevs]
LOG.info(f'Found inactive mdevs: {names}')
for mdev in inactive_mdevs:
LOG.info(f'Starting inactive mdev: {mdev.name()}')
self._host.device_start(mdev)
@staticmethod
def _is_existing_mdev(uuid):
# FIXME(sbauza): Some kernel can have a uevent race meaning that the
# libvirt daemon won't know when a mediated device is created unless
# you restart that daemon. Until all kernels we support are not having
# that possible race, check the sysfs directly instead of asking the
# libvirt API.
# See https://bugzilla.redhat.com/show_bug.cgi?id=1376907 for ref.
return os.path.exists('/sys/bus/mdev/devices/{0}'.format(uuid))
def _recreate_assigned_mediated_devices(self):
"""Recreate assigned mdevs that could have disappeared if we reboot
the host.
"""
# NOTE(sbauza): This method just calls sysfs to recreate mediated
# devices by looking up existing guest XMLs and doesn't use
# the Placement API so it works with or without a vGPU reshape.
mdevs = self._get_all_assigned_mediated_devices()
for (mdev_uuid, instance_uuid) in mdevs.items():
if not self._is_existing_mdev(mdev_uuid):
dev_name = libvirt_utils.mdev_uuid2name(mdev_uuid)
dev_info = self._get_mediated_device_information(dev_name)
parent = dev_info['parent']
parent_type = self._get_vgpu_type_per_pgpu(parent)
if dev_info['type'] != parent_type:
# NOTE(sbauza): The mdev was created by using a different
# vGPU type. We can't recreate the mdev until the operator
# modifies the configuration.
parent = "{}:{}:{}.{}".format(*parent[4:].split('_'))
msg = ("The instance UUID %(inst)s uses a mediated device "
"type %(type)s that is no longer supported by the "
"parent PCI device, %(parent)s. Please correct "
"the configuration accordingly." %
{'inst': instance_uuid,
'parent': parent,
'type': dev_info['type']})
raise exception.InvalidLibvirtMdevConfig(reason=msg)
self._create_new_mediated_device(parent, uuid=mdev_uuid)
def _check_my_ip(self):
ips = compute_utils.get_machine_ips()
if CONF.my_ip not in ips:
LOG.warning('my_ip address (%(my_ip)s) was not found on '
'any of the interfaces: %(ifaces)s',
{'my_ip': CONF.my_ip, 'ifaces': ", ".join(ips)})
def _check_cpu_set_configuration(self):
# evaluate these now to force a quick fail if they're invalid
vcpu_pin_set = hardware.get_vcpu_pin_set() or set()
cpu_shared_set = hardware.get_cpu_shared_set() or set()
cpu_dedicated_set = hardware.get_cpu_dedicated_set() or set()
# TODO(stephenfin): Remove this in U once we remove the 'vcpu_pin_set'
# option
if not vcpu_pin_set:
if not (cpu_shared_set or cpu_dedicated_set):
return
if not cpu_dedicated_set.isdisjoint(cpu_shared_set):
msg = _(
"The '[compute] cpu_dedicated_set' and '[compute] "
"cpu_shared_set' configuration options must be "
"disjoint.")
raise exception.InvalidConfiguration(msg)
if CONF.reserved_host_cpus:
msg = _(
"The 'reserved_host_cpus' config option cannot be defined "
"alongside the '[compute] cpu_shared_set' or '[compute] "
"cpu_dedicated_set' options. Unset 'reserved_host_cpus'.")
raise exception.InvalidConfiguration(msg)
return
if cpu_dedicated_set:
# NOTE(stephenfin): This is a new option in Train so it can be
# an error
msg = _(
"The 'vcpu_pin_set' config option has been deprecated and "
"cannot be defined alongside '[compute] cpu_dedicated_set'. "
"Unset 'vcpu_pin_set'.")
raise exception.InvalidConfiguration(msg)
if cpu_shared_set:
LOG.warning(
"The '[compute] cpu_shared_set' and 'vcpu_pin_set' config "
"options have both been defined. While 'vcpu_pin_set' is "
"defined, it will continue to be used to configure the "
"specific host CPUs used for 'VCPU' inventory, while "
"'[compute] cpu_shared_set' will only be used for guest "
"emulator threads when 'hw:emulator_threads_policy=shared' "
"is defined in the flavor. This is legacy behavior and will "
"not be supported in a future release. "
"If you wish to define specific host CPUs to be used for "
"'VCPU' or 'PCPU' inventory, you must migrate the "
"'vcpu_pin_set' config option value to '[compute] "
"cpu_shared_set' and '[compute] cpu_dedicated_set', "
"respectively, and undefine 'vcpu_pin_set'.")
else:
LOG.warning(
"The 'vcpu_pin_set' config option has been deprecated and "
"will be removed in a future release. When defined, "
"'vcpu_pin_set' will be used to calculate 'VCPU' inventory "
"and schedule instances that have 'VCPU' allocations. "
"If you wish to define specific host CPUs to be used for "
"'VCPU' or 'PCPU' inventory, you must migrate the "
"'vcpu_pin_set' config option value to '[compute] "
"cpu_shared_set' and '[compute] cpu_dedicated_set', "
"respectively, and undefine 'vcpu_pin_set'.")
def _supports_tb_cache_size(self):
if (
CONF.libvirt.virt_type == 'qemu' and
CONF.libvirt.tb_cache_size and
CONF.libvirt.tb_cache_size > 0
):
if not self._host.has_min_version(MIN_LIBVIRT_TB_CACHE_SIZE):
raise exception.InvalidConfiguration(
_("Nova requires libvirt version %s or greater "
"with '[libvirt] tb_cache_size' "
"configured.") %
libvirt_utils.version_to_string(MIN_LIBVIRT_TB_CACHE_SIZE))
def _prepare_migration_flags(self):
migration_flags = 0
migration_flags |= libvirt.VIR_MIGRATE_LIVE
# Enable support for p2p migrations
migration_flags |= libvirt.VIR_MIGRATE_PEER2PEER
# Adding VIR_MIGRATE_UNDEFINE_SOURCE because, without it, migrated
# instance will remain defined on the source host
migration_flags |= libvirt.VIR_MIGRATE_UNDEFINE_SOURCE
# Adding VIR_MIGRATE_PERSIST_DEST to persist the VM on the
# destination host
migration_flags |= libvirt.VIR_MIGRATE_PERSIST_DEST
live_migration_flags = block_migration_flags = migration_flags
# Adding VIR_MIGRATE_NON_SHARED_INC, otherwise all block-migrations
# will be live-migrations instead
block_migration_flags |= libvirt.VIR_MIGRATE_NON_SHARED_INC
return (live_migration_flags, block_migration_flags)
# TODO(kchamart) Once the MIN_LIBVIRT_VERSION and MIN_QEMU_VERSION
# reach 4.4.0 and 2.11.0, which provide "native TLS" support by
# default, deprecate and remove the support for "tunnelled live
# migration" (and related config attribute), because:
#
# (a) it cannot handle live migration of disks in a non-shared
# storage setup (a.k.a. "block migration");
#
# (b) has a huge performance overhead and latency, because it burns
# more CPU and memory bandwidth due to increased number of data
# copies on both source and destination hosts.
#
# Both the above limitations are addressed by the QEMU-native TLS
# support (`live_migration_with_native_tls`).
def _handle_live_migration_tunnelled(self, migration_flags):
if CONF.libvirt.live_migration_tunnelled:
migration_flags |= libvirt.VIR_MIGRATE_TUNNELLED
return migration_flags
def _handle_native_tls(self, migration_flags):
if (CONF.libvirt.live_migration_with_native_tls):
migration_flags |= libvirt.VIR_MIGRATE_TLS
return migration_flags
def _handle_live_migration_post_copy(self, migration_flags):
if CONF.libvirt.live_migration_permit_post_copy:
migration_flags |= libvirt.VIR_MIGRATE_POSTCOPY
return migration_flags
def _handle_live_migration_auto_converge(self, migration_flags):
if self._is_post_copy_enabled(migration_flags):
LOG.info('The live_migration_permit_post_copy is set to '
'True and post copy live migration is available '
'so auto-converge will not be in use.')
elif CONF.libvirt.live_migration_permit_auto_converge:
migration_flags |= libvirt.VIR_MIGRATE_AUTO_CONVERGE
return migration_flags
def _parse_migration_flags(self):
(live_migration_flags,
block_migration_flags) = self._prepare_migration_flags()
live_migration_flags = self._handle_live_migration_tunnelled(
live_migration_flags)
block_migration_flags = self._handle_live_migration_tunnelled(
block_migration_flags)
live_migration_flags = self._handle_native_tls(
live_migration_flags)
block_migration_flags = self._handle_native_tls(
block_migration_flags)
live_migration_flags = self._handle_live_migration_post_copy(
live_migration_flags)
block_migration_flags = self._handle_live_migration_post_copy(
block_migration_flags)
live_migration_flags = self._handle_live_migration_auto_converge(
live_migration_flags)
block_migration_flags = self._handle_live_migration_auto_converge(
block_migration_flags)
self._live_migration_flags = live_migration_flags
self._block_migration_flags = block_migration_flags
# TODO(sahid): This method is targeted for removal when the tests
# have been updated to avoid its use
#
# All libvirt API calls on the libvirt.Connect object should be
# encapsulated by methods on the nova.virt.libvirt.host.Host
# object, rather than directly invoking the libvirt APIs. The goal
# is to avoid a direct dependency on the libvirt API from the
# driver.py file.
def _get_connection(self):
return self._host.get_connection()
_conn = property(_get_connection)
@staticmethod
def _uri():
if CONF.libvirt.virt_type == 'lxc':
uri = CONF.libvirt.connection_uri or 'lxc:///'
elif CONF.libvirt.virt_type == 'parallels':
uri = CONF.libvirt.connection_uri or 'parallels:///system'
else:
uri = CONF.libvirt.connection_uri or 'qemu:///system'
return uri
@staticmethod
def _live_migration_uri(dest):
uris = {
'kvm': 'qemu+%(scheme)s://%(dest)s/system',
'qemu': 'qemu+%(scheme)s://%(dest)s/system',
'parallels': 'parallels+tcp://%(dest)s/system',
}
dest = oslo_netutils.escape_ipv6(dest)
virt_type = CONF.libvirt.virt_type
# TODO(pkoniszewski): Remove fetching live_migration_uri in Pike
uri = CONF.libvirt.live_migration_uri
if uri:
return uri % dest
uri = uris.get(virt_type)
if uri is None:
raise exception.LiveMigrationURINotAvailable(virt_type=virt_type)
str_format = {
'dest': dest,
'scheme': CONF.libvirt.live_migration_scheme or 'tcp',
}
return uri % str_format
@staticmethod
def _migrate_uri(dest):
uri = None
dest = oslo_netutils.escape_ipv6(dest)
# Only QEMU live migrations supports migrate-uri parameter
virt_type = CONF.libvirt.virt_type
if virt_type in ('qemu', 'kvm'):
# QEMU accept two schemes: tcp and rdma. By default
# libvirt build the URI using the remote hostname and the
# tcp schema.
uri = 'tcp://%s' % dest
# Because dest might be of type unicode, here we might return value of
# type unicode as well which is not acceptable by libvirt python
# binding when Python 2.7 is in use, so let's convert it explicitly
# back to string. When Python 3.x is in use, libvirt python binding
# accepts unicode type so it is completely fine to do a no-op str(uri)
# conversion which will return value of type unicode.
return uri and str(uri)
def instance_exists(self, instance):
"""Efficient override of base instance_exists method."""
try:
self._host.get_guest(instance)
return True
except (exception.InternalError, exception.InstanceNotFound):
return False
def list_instances(self):
names = []
for guest in self._host.list_guests(only_running=False):
names.append(guest.name)
return names
def list_instance_uuids(self):
uuids = []
for guest in self._host.list_guests(only_running=False):
uuids.append(guest.uuid)
return uuids
def plug_vifs(self, instance, network_info):
"""Plug VIFs into networks."""
for vif in network_info:
self.vif_driver.plug(instance, vif)
def _unplug_vifs(self, instance, network_info, ignore_errors):
"""Unplug VIFs from networks."""
for vif in network_info:
try:
self.vif_driver.unplug(instance, vif)
except exception.NovaException:
if not ignore_errors:
raise
def unplug_vifs(self, instance, network_info):
self._unplug_vifs(instance, network_info, False)
def _teardown_container(self, instance):
inst_path = libvirt_utils.get_instance_path(instance)
container_dir = os.path.join(inst_path, 'rootfs')
rootfs_dev = instance.system_metadata.get('rootfs_device_name')
LOG.debug('Attempting to teardown container at path %(dir)s with '
'root device: %(rootfs_dev)s',
{'dir': container_dir, 'rootfs_dev': rootfs_dev},
instance=instance)
disk_api.teardown_container(container_dir, rootfs_dev)
def _destroy(self, instance):
try:
guest = self._host.get_guest(instance)
if CONF.serial_console.enabled:
# This method is called for several events: destroy,
# rebuild, hard-reboot, power-off - For all of these
# events we want to release the serial ports acquired
# for the guest before destroying it.
serials = self._get_serial_ports_from_guest(guest)
for hostname, port in serials:
serial_console.release_port(host=hostname, port=port)
except exception.InstanceNotFound:
guest = None
# If the instance is already terminated, we're still happy
# Otherwise, destroy it
old_domid = -1
if guest is not None:
try:
old_domid = guest.id
guest.poweroff()
except libvirt.libvirtError as e:
is_okay = False
errcode = e.get_error_code()
if errcode == libvirt.VIR_ERR_NO_DOMAIN:
# Domain already gone. This can safely be ignored.
is_okay = True
elif errcode == libvirt.VIR_ERR_OPERATION_INVALID:
# If the instance is already shut off, we get this:
# Code=55 Error=Requested operation is not valid:
# domain is not running
state = guest.get_power_state(self._host)
if state == power_state.SHUTDOWN:
is_okay = True
elif errcode == libvirt.VIR_ERR_INTERNAL_ERROR:
errmsg = e.get_error_message()
if (CONF.libvirt.virt_type == 'lxc' and
errmsg == 'internal error: '
'Some processes refused to die'):
# Some processes in the container didn't die
# fast enough for libvirt. The container will
# eventually die. For now, move on and let
# the wait_for_destroy logic take over.
is_okay = True
elif errcode == libvirt.VIR_ERR_OPERATION_TIMEOUT:
LOG.warning("Cannot destroy instance, operation time out",
instance=instance)
reason = _("operation time out")
raise exception.InstancePowerOffFailure(reason=reason)
elif errcode == libvirt.VIR_ERR_SYSTEM_ERROR:
with excutils.save_and_reraise_exception():
LOG.warning("Cannot destroy instance, general system "
"call failure", instance=instance)
if not is_okay:
with excutils.save_and_reraise_exception():
LOG.error('Error from libvirt during destroy. '
'Code=%(errcode)s Error=%(e)s',
{'errcode': errcode, 'e': e},
instance=instance)
def _wait_for_destroy(expected_domid):
"""Called at an interval until the VM is gone."""
# NOTE(vish): If the instance disappears during the destroy
# we ignore it so the cleanup can still be
# attempted because we would prefer destroy to
# never fail.
try:
dom_info = self.get_info(instance)
state = dom_info.state
new_domid = dom_info.internal_id
except exception.InstanceNotFound:
LOG.debug("During wait destroy, instance disappeared.",
instance=instance)
state = power_state.SHUTDOWN
if state == power_state.SHUTDOWN:
LOG.info("Instance destroyed successfully.", instance=instance)
raise loopingcall.LoopingCallDone()
# NOTE(wangpan): If the instance was booted again after destroy,
# this may be an endless loop, so check the id of
# domain here, if it changed and the instance is
# still running, we should destroy it again.
# see https://bugs.launchpad.net/nova/+bug/1111213 for more details
if new_domid != expected_domid:
LOG.info("Instance may be started again.", instance=instance)
kwargs['is_running'] = True
raise loopingcall.LoopingCallDone()
kwargs = {'is_running': False}
timer = loopingcall.FixedIntervalLoopingCall(_wait_for_destroy,
old_domid)
timer.start(interval=0.5).wait()
if kwargs['is_running']:
LOG.info("Going to destroy instance again.", instance=instance)
self._destroy(instance)
else:
# NOTE(GuanQiang): teardown container to avoid resource leak
if CONF.libvirt.virt_type == 'lxc':
self._teardown_container(instance)
# We're sure the instance is gone, we can shutdown the core if so
self.cpu_api.power_down_for_instance(instance)
def destroy(self, context, instance, network_info, block_device_info=None,
destroy_disks=True, destroy_secrets=True):
self._destroy(instance)
# NOTE(gibi): if there was device detach in progress then we need to
# unblock the waiting threads and clean up.
self._device_event_handler.cleanup_waiters(instance.uuid)
self.cleanup(context, instance, network_info, block_device_info,
destroy_disks, destroy_secrets=destroy_secrets)
def _undefine_domain(self, instance):
try:
guest = self._host.get_guest(instance)
try:
guest.delete_configuration()
except libvirt.libvirtError as e:
with excutils.save_and_reraise_exception() as ctxt:
errcode = e.get_error_code()
if errcode == libvirt.VIR_ERR_NO_DOMAIN:
LOG.debug("Called undefine, but domain already gone.",
instance=instance)
ctxt.reraise = False
else:
LOG.error('Error from libvirt during undefine. '
'Code=%(errcode)s Error=%(e)s',
{'errcode': errcode,
'e': e},
instance=instance)
except exception.InstanceNotFound:
pass
def cleanup(self, context, instance, network_info, block_device_info=None,
destroy_disks=True, migrate_data=None, destroy_vifs=True,
destroy_secrets=True):
"""Cleanup the instance from the host.
Identify if the instance disks and instance path should be removed
from the host before calling down into the _cleanup method for the
actual removal of resources from the host.
:param context: security context
:param instance: instance object for the instance being cleaned up
:param network_info: instance network information
:param block_device_info: optional instance block device information
:param destroy_disks: if local ephemeral disks should be destroyed
:param migrate_data: optional migrate_data object
:param destroy_vifs: if plugged vifs should be unplugged
:param destroy_secrets: Indicates if secrets should be destroyed
"""
cleanup_instance_dir = False
cleanup_instance_disks = False
# We assume destroy_disks means destroy instance directory and disks
if destroy_disks:
cleanup_instance_dir = True
cleanup_instance_disks = True
else:
# NOTE(mheler): For shared block storage we only need to clean up
# the instance directory when it's not on a shared path.
if migrate_data and 'is_shared_block_storage' in migrate_data:
cleanup_instance_dir = (
migrate_data.is_shared_block_storage and
not migrate_data.is_shared_instance_path)
# NOTE(lyarwood): The following workaround allows operators to
# ensure that non-shared instance directories are removed after an
# evacuation or revert resize when using the shared RBD
# imagebackend. This workaround is not required when cleaning up
# migrations that provide migrate_data to this method as the
# existing is_shared_block_storage conditional will cause the
# instance directory to be removed.
if not cleanup_instance_dir:
if CONF.workarounds.ensure_libvirt_rbd_instance_dir_cleanup:
cleanup_instance_dir = CONF.libvirt.images_type == 'rbd'
return self._cleanup(
context, instance, network_info,
block_device_info=block_device_info,
destroy_vifs=destroy_vifs,
cleanup_instance_dir=cleanup_instance_dir,
cleanup_instance_disks=cleanup_instance_disks,
destroy_secrets=destroy_secrets)
def _cleanup(self, context, instance, network_info, block_device_info=None,
destroy_vifs=True, cleanup_instance_dir=False,
cleanup_instance_disks=False, destroy_secrets=True):
"""Cleanup the domain and any attached resources from the host.
This method cleans up any pmem devices, unplugs VIFs, disconnects
attached volumes and undefines the instance domain within libvirt.
It also optionally removes the ephemeral disks and the instance
directory from the host depending on the cleanup_instance_dir|disks
kwargs provided.
:param context: security context
:param instance: instance object for the instance being cleaned up
:param network_info: instance network information
:param block_device_info: optional instance block device information
:param destroy_vifs: if plugged vifs should be unplugged
:param cleanup_instance_dir: If the instance dir should be removed
:param cleanup_instance_disks: If the instance disks should be removed.
Also removes ephemeral encryption secrets, if present.
:param destroy_secrets: If the cinder volume encryption secrets should
be deleted.
"""
# zero the data on backend pmem device
vpmems = self._get_vpmems(instance)
if vpmems:
self._cleanup_vpmems(vpmems)
if destroy_vifs:
self._unplug_vifs(instance, network_info, True)
# FIXME(wangpan): if the instance is booted again here, such as the
# soft reboot operation boot it here, it will become
# "running deleted", should we check and destroy it
# at the end of this method?
# NOTE(vish): we disconnect from volumes regardless
block_device_mapping = driver.block_device_info_get_mapping(
block_device_info)
for vol in block_device_mapping:
connection_info = vol['connection_info']
if not connection_info:
# if booting from a volume, creation could have failed meaning
# this would be unset
continue
try:
self._disconnect_volume(
context, connection_info, instance,
destroy_secrets=destroy_secrets, force=True)
except Exception as exc:
with excutils.save_and_reraise_exception() as ctxt:
if cleanup_instance_disks:
# Don't block on Volume errors if we're trying to
# delete the instance as we may be partially created
# or deleted
ctxt.reraise = False
LOG.warning(
"Ignoring Volume Error on vol %(vol_id)s "
"during delete %(exc)s",
{'vol_id': vol.get('volume_id'),
'exc': exc},
instance=instance)
if cleanup_instance_disks:
# NOTE(haomai): destroy volumes if needed
if CONF.libvirt.images_type == 'lvm':
self._cleanup_lvm(instance, block_device_info)
if CONF.libvirt.images_type == 'rbd':
self._cleanup_rbd(instance)
if cleanup_instance_dir:
attempts = int(instance.system_metadata.get('clean_attempts',
'0'))
success = self.delete_instance_files(instance)
# NOTE(mriedem): This is used in the _run_pending_deletes periodic
# task in the compute manager. The tight coupling is not great...
instance.system_metadata['clean_attempts'] = str(attempts + 1)
if success:
instance.cleaned = True
try:
instance.save()
except exception.InstanceNotFound:
pass
if cleanup_instance_disks:
crypto.delete_vtpm_secret(context, instance)
# Make sure that the instance directory files were successfully
# deleted before destroying the encryption secrets in the case of
# image backends that are not 'lvm' or 'rbd'. We don't want to
# leave any chance that we delete the secrets if the disks have not
# been deleted.
if CONF.libvirt.images_type in ('lvm', 'rbd') or instance.cleaned:
self._cleanup_ephemeral_encryption_secrets(
context, instance, block_device_info)
self._undefine_domain(instance)
def _cleanup_ephemeral_encryption_secrets(
self, context, instance, block_device_info
):
exception_msgs = []
encrypted_bdms = driver.block_device_info_get_encrypted_disks(
block_device_info)
for driver_bdm in encrypted_bdms:
# NOTE(melwitt): We intentionally only delete libvirt secrets here
# and not secrets in the key manager service (example: barbican).
# Libvirt secrets are local to a compute host and are routinely
# deleted during instance move operations. If we're only moving, we
# don't want to delete the secret in the key manager service. The
# secret in the key manager service should only be deleted when the
# instance is deleted.
secret_usage = f"{instance.uuid}_{driver_bdm['uuid']}"
if self._host.find_secret('volume', secret_usage):
try:
self._host.delete_secret('volume', secret_usage)
except libvirt.libvirtError as e:
msg = (
f'Failed to delete libvirt secret {secret_usage}: ' +
str(e))
LOG.exception(msg, instance=instance)
exception_msgs.append(msg)
if exception_msgs:
msg = '\n'.join(exception_msgs)
raise exception.EphemeralEncryptionCleanupFailed(error=msg)
def cleanup_lingering_instance_resources(self, instance):
# zero the data on backend pmem device, if fails
# it will raise an exception
vpmems = self._get_vpmems(instance)
if vpmems:
self._cleanup_vpmems(vpmems)
# we may have some claimed mdev residue, we need to delete it
mdevs = self.instance_claimed_mdevs.pop(instance.uuid, None)
if mdevs:
# The live migration was aborted, we need to remove the reserved
# values.
LOG.debug("Unclaiming mdevs %s from instance %s",
mdevs, instance.uuid)
def _cleanup_vpmems(self, vpmems):
for vpmem in vpmems:
try:
nova.privsep.libvirt.cleanup_vpmem(vpmem.devpath)
except Exception as e:
raise exception.VPMEMCleanupFailed(dev=vpmem.devpath,
error=e)
def _get_serial_ports_from_guest(self, guest, mode=None):
"""Returns an iterator over serial port(s) configured on guest.
:param mode: Should be a value in (None, bind, connect)
"""
xml = guest.get_xml_desc()
tree = etree.fromstring(xml)
# The 'serial' device is the base for x86 platforms. Other platforms
# (e.g. kvm on system z = S390X) can only use 'console' devices.
xpath_mode = "[@mode='%s']" % mode if mode else ""
serial_tcp = "./devices/serial[@type='tcp']/source" + xpath_mode
console_tcp = "./devices/console[@type='tcp']/source" + xpath_mode
tcp_devices = tree.findall(serial_tcp)
if len(tcp_devices) == 0:
tcp_devices = tree.findall(console_tcp)
for source in tcp_devices:
yield (source.get("host"), int(source.get("service")))
def _get_scsi_controller_next_unit(self, guest):
"""Returns the max disk unit used by scsi controller"""
xml = guest.get_xml_desc()
tree = etree.fromstring(xml)
addrs = "./devices/disk[target/@bus='scsi']/address[@type='drive']"
ret = []
for obj in tree.xpath(addrs):
ret.append(int(obj.get('unit', 0)))
return max(ret) + 1 if ret else 0
def _cleanup_rbd(self, instance):
# NOTE(nic): On revert_resize, the cleanup steps for the root
# volume are handled with an "rbd snap rollback" command,
# and none of this is needed (and is, in fact, harmful) so
# filter out non-ephemerals from the list
if instance.task_state == task_states.RESIZE_REVERTING:
filter_fn = lambda disk: (disk.startswith(instance.uuid) and
disk.endswith('disk.local'))
else:
filter_fn = lambda disk: disk.startswith(instance.uuid)
rbd_utils.RBDDriver().cleanup_volumes(filter_fn)
def _cleanup_lvm(self, instance, block_device_info):
"""Delete all LVM disks for given instance object."""
if instance.get('ephemeral_key_uuid') is not None:
# detach encrypted volumes
disks = self._get_instance_disk_info(instance, block_device_info)
for disk in disks:
if dmcrypt.is_encrypted(disk['path']):
dmcrypt.delete_volume(disk['path'])
disks = self._lvm_disks(instance)
if disks:
lvm.remove_volumes(disks)
def _lvm_disks(self, instance):
"""Returns all LVM disks for given instance object."""
if CONF.libvirt.images_volume_group:
vg = os.path.join('/dev', CONF.libvirt.images_volume_group)
if not os.path.exists(vg):
return []
pattern = '%s_' % instance.uuid
def belongs_to_instance(disk):
return disk.startswith(pattern)
def fullpath(name):
return os.path.join(vg, name)
logical_volumes = lvm.list_volumes(vg)
disks = [fullpath(disk) for disk in logical_volumes
if belongs_to_instance(disk)]
return disks
return []
def get_volume_connector(self, instance):
root_helper = utils.get_root_helper()
return connector.get_connector_properties(
root_helper, CONF.my_block_storage_ip,
CONF.libvirt.volume_use_multipath,
enforce_multipath=True,
host=CONF.host)
def _cleanup_resize_vtpm(
self,
context: nova_context.RequestContext,
instance: 'objects.Instance',
) -> None:
"""Handle vTPM when confirming a migration or resize.
If the old flavor have vTPM and the new one doesn't, there are keys to
be deleted.
"""
old_vtpm_config = hardware.get_vtpm_constraint(
instance.old_flavor, instance.image_meta)
new_vtpm_config = hardware.get_vtpm_constraint(
instance.new_flavor, instance.image_meta)
if old_vtpm_config and not new_vtpm_config:
# the instance no longer cares for its vTPM so delete the related
# secret; the deletion of the instance directory and undefining of
# the domain will take care of the TPM files themselves
LOG.info('New flavor no longer requests vTPM; deleting secret.')
crypto.delete_vtpm_secret(context, instance)
# TODO(stephenfin): Fold this back into its only caller, cleanup_resize
def _cleanup_resize(self, context, instance, network_info):
inst_base = libvirt_utils.get_instance_path(instance)
target = inst_base + '_resize'
# zero the data on backend old pmem device
vpmems = self._get_vpmems(instance, prefix='old')
if vpmems:
self._cleanup_vpmems(vpmems)
# Remove any old vTPM data, if necessary
self._cleanup_resize_vtpm(context, instance)
# Deletion can fail over NFS, so retry the deletion as required.
# Set maximum attempt as 5, most test can remove the directory
# for the second time.
attempts = 0
while os.path.exists(target) and attempts < 5:
shutil.rmtree(target, ignore_errors=True)
if os.path.exists(target):
time.sleep(random.randint(20, 200) / 100.0)
attempts += 1
# NOTE(mriedem): Some image backends will recreate the instance path
# and disk.info during init, and all we need the root disk for
# here is removing cloned snapshots which is backend-specific, so
# check that first before initializing the image backend object. If
# there is ever an image type that supports clone *and* re-creates
# the instance directory and disk.info on init, this condition will
# need to be re-visited to make sure that backend doesn't re-create
# the disk. Refer to bugs: 1666831 1728603 1769131
if self.image_backend.backend(CONF.libvirt.images_type).SUPPORTS_CLONE:
root_disk = self.image_backend.by_name(instance, 'disk')
if root_disk.exists():
root_disk.remove_snap(libvirt_utils.RESIZE_SNAPSHOT_NAME)
if instance.host != CONF.host:
self._undefine_domain(instance)
# TODO(sean-k-mooney): remove this call to unplug_vifs after
# Wallaby is released. VIFs are now unplugged in resize_instance.
try:
self.unplug_vifs(instance, network_info)
except exception.InternalError as e:
LOG.debug(e, instance=instance)
def _get_volume_driver(
self, connection_info: ty.Dict[str, ty.Any]
) -> 'volume.LibvirtBaseVolumeDriver':
"""Fetch the nova.virt.libvirt.volume driver
Based on the provided connection_info return a nova.virt.libvirt.volume
driver. This will call out to os-brick to construct an connector and
check if the connector is valid on the underlying host.
:param connection_info: The connection_info associated with the volume
:raises: VolumeDriverNotFound if no driver is found or if the host
doesn't support the requested driver. This retains legacy behaviour
when only supported drivers were loaded on startup leading to a
VolumeDriverNotFound being raised later if an invalid driver was
requested.
"""
driver_type = connection_info.get('driver_volume_type')
# If the driver_type isn't listed in the supported type list fail
if driver_type not in VOLUME_DRIVERS:
raise exception.VolumeDriverNotFound(driver_type=driver_type)
# Return the cached driver
if driver_type in self.volume_drivers:
return self.volume_drivers.get(driver_type)
@utils.synchronized('cache_volume_driver')
def _cache_volume_driver(driver_type):
# Check if another request cached the driver while we waited
if driver_type in self.volume_drivers:
return self.volume_drivers.get(driver_type)
try:
driver_class = importutils.import_class(
VOLUME_DRIVERS.get(driver_type))
self.volume_drivers[driver_type] = driver_class(self._host)
return self.volume_drivers.get(driver_type)
except brick_exception.InvalidConnectorProtocol:
LOG.debug('Unable to load volume driver %s. It is not '
'supported on this host.', driver_type)
# NOTE(lyarwood): This exception is a subclass of
# VolumeDriverNotFound to ensure no callers have to change
# their error handling code after the move to on-demand loading
# of the volume drivers and associated os-brick connectors.
raise exception.VolumeDriverNotSupported(
volume_driver=VOLUME_DRIVERS.get(driver_type))
# Cache the volume driver if it hasn't already been
return _cache_volume_driver(driver_type)
def _connect_volume(self, context, connection_info, instance,
encryption=None):
vol_driver = self._get_volume_driver(connection_info)
vol_driver.connect_volume(connection_info, instance)
try:
self._attach_encryptor(context, connection_info, encryption)
except Exception:
# Encryption failed so rollback the volume connection.
with excutils.save_and_reraise_exception(logger=LOG):
LOG.exception("Failure attaching encryptor; rolling back "
"volume connection", instance=instance)
vol_driver.disconnect_volume(connection_info, instance)
def _should_disconnect_target(self, context, instance, multiattach,
vol_driver, volume_id):
# NOTE(jdg): Multiattach is a special case (not to be confused
# with shared_targets). With multiattach we may have a single volume
# attached multiple times to *this* compute node (ie Server-1 and
# Server-2). So, if we receive a call to delete the attachment for
# Server-1 we need to take special care to make sure that the Volume
# isn't also attached to another Server on this Node. Otherwise we
# will indiscriminantly delete the connection for all Server and that's
# no good. So check if it's attached multiple times on this node
# if it is we skip the call to brick to delete the connection.
if not multiattach:
return True
# NOTE(deiter): Volume drivers using _HostMountStateManager are another
# special case. _HostMountStateManager ensures that the compute node
# only attempts to mount a single mountpoint in use by multiple
# attachments once, and that it is not unmounted until it is no longer
# in use by any attachments. So we can skip the multiattach check for
# volume drivers that based on LibvirtMountedFileSystemVolumeDriver.
if isinstance(vol_driver, fs.LibvirtMountedFileSystemVolumeDriver):
return True
connection_count = 0
volume = self._volume_api.get(context, volume_id)
attachments = volume.get('attachments', {})
if len(attachments) > 1:
# First we get a list of all Server UUID's associated with
# this Host (Compute Node). We're going to use this to
# determine if the Volume being detached is also in-use by
# another Server on this Host, ie just check to see if more
# than one attachment.server_id for this volume is in our
# list of Server UUID's for this Host
servers_this_host = objects.InstanceList.get_uuids_by_host(
context, instance.host)
# NOTE(jdg): nova.volume.cinder translates the
# volume['attachments'] response into a dict which includes
# the Server UUID as the key, so we're using that
# here to check against our server_this_host list
for server_id, data in attachments.items():
if server_id in servers_this_host:
connection_count += 1
return (False if connection_count > 1 else True)
def _disconnect_volume(self, context, connection_info, instance,
encryption=None, destroy_secrets=True, force=False):
self._detach_encryptor(
context,
connection_info,
encryption=encryption,
destroy_secrets=destroy_secrets
)
vol_driver = self._get_volume_driver(connection_info)
volume_id = driver_block_device.get_volume_id(connection_info)
multiattach = connection_info.get('multiattach', False)
if self._should_disconnect_target(
context, instance, multiattach, vol_driver, volume_id):
vol_driver.disconnect_volume(
connection_info, instance, force=force)
else:
LOG.info('Detected multiple connections on this host for '
'volume: %(volume)s, skipping target disconnect.',
{'volume': volume_id})
def _extend_volume(self, connection_info, instance, requested_size):
vol_driver = self._get_volume_driver(connection_info)
return vol_driver.extend_volume(connection_info, instance,
requested_size)
def _allow_native_luksv1(self, encryption=None):
"""Check if QEMU's native LUKSv1 decryption should be used.
"""
# NOTE(lyarwood): Ensure the LUKSv1 provider is used.
provider = None
if encryption:
provider = encryption.get('provider', None)
if provider in encryptors.LEGACY_PROVIDER_CLASS_TO_FORMAT_MAP:
provider = encryptors.LEGACY_PROVIDER_CLASS_TO_FORMAT_MAP[provider]
return provider == encryptors.LUKS
def _get_volume_config(self, instance, connection_info, disk_info):
vol_driver = self._get_volume_driver(connection_info)
conf = vol_driver.get_config(connection_info, disk_info)
if self._sev_enabled(instance.flavor, instance.image_meta):
designer.set_driver_iommu_for_device(conf)
self._set_cache_mode(conf)
return conf
def _get_volume_encryptor(self, connection_info, encryption):
root_helper = utils.get_root_helper()
return encryptors.get_volume_encryptor(root_helper=root_helper,
keymgr=key_manager.API(CONF),
connection_info=connection_info,
**encryption)
def _get_volume_encryption(self, context, connection_info):
"""Get the encryption metadata dict if it is not provided
"""
encryption = {}
volume_id = driver_block_device.get_volume_id(connection_info)
if volume_id:
encryption = encryptors.get_encryption_metadata(context,
self._volume_api, volume_id, connection_info)
return encryption
def _attach_encryptor(self, context, connection_info, encryption):
"""Attach the frontend encryptor if one is required by the volume.
The request context is only used when an encryption metadata dict is
not provided. The encryption metadata dict being populated is then used
to determine if an attempt to attach the encryptor should be made.
"""
# NOTE(lyarwood): Skip any attempt to fetch encryption metadata or the
# actual passphrase from the key manager if a libvirt secret already
# exists locally for the volume. This suggests that the instance was
# only powered off or the underlying host rebooted.
volume_id = driver_block_device.get_volume_id(connection_info)
if self._host.find_secret('volume', volume_id):
LOG.debug("A libvirt secret for volume %s has been found on the "
"host, skipping any attempt to create another or attach "
"an os-brick encryptor.", volume_id)
return
if encryption is None:
encryption = self._get_volume_encryption(context, connection_info)
if encryption and self._allow_native_luksv1(encryption=encryption):
# NOTE(lyarwood): Fetch the associated key for the volume and
# decode the passphrase from the key.
# FIXME(lyarwood): c-vol currently creates symmetric keys for use
# with volumes, leading to the binary to hex to string conversion
# below.
keymgr = key_manager.API(CONF)
key = keymgr.get(context, encryption['encryption_key_id'])
key_encoded = key.get_encoded()
passphrase = binascii.hexlify(key_encoded).decode('utf-8')
# NOTE(lyarwood): Retain the behaviour of the original os-brick
# encryptors and format any volume that does not identify as
# encrypted with LUKS.
# FIXME(lyarwood): Remove this once c-vol correctly formats
# encrypted volumes during their initial creation:
# https://bugs.launchpad.net/cinder/+bug/1739442
device_path = connection_info.get('data').get('device_path')
if device_path:
root_helper = utils.get_root_helper()
if not luks_encryptor.is_luks(root_helper, device_path):
encryptor = self._get_volume_encryptor(connection_info,
encryption)
encryptor._format_volume(passphrase, **encryption)
# NOTE(lyarwood): Store the passphrase as a libvirt secret locally
# on the compute node. This secret is used later when generating
# the volume config.
self._host.create_secret('volume', volume_id, password=passphrase)
elif encryption:
encryptor = self._get_volume_encryptor(connection_info,
encryption)
encryptor.attach_volume(context, **encryption)
def _detach_encryptor(self, context, connection_info, encryption,
destroy_secrets=True):
"""Detach the frontend encryptor if one is required by the volume.
The request context is only used when an encryption metadata dict is
not provided. The encryption metadata dict being populated is then used
to determine if an attempt to detach the encryptor should be made.
If native LUKS decryption is enabled then delete previously created
Libvirt volume secret from the host.
"""
volume_id = driver_block_device.get_volume_id(connection_info)
if volume_id and self._host.find_secret('volume', volume_id):
if not destroy_secrets:
LOG.debug("Skipping volume secret destruction")
return
return self._host.delete_secret('volume', volume_id)
if encryption is None:
encryption = self._get_volume_encryption(context, connection_info)
# NOTE(lyarwood): Handle bugs #1821696 and #1917619 by avoiding the use
# of the os-brick encryptors if we don't have a device_path. The lack
# of a device_path here suggests the volume was natively attached to
# QEMU anyway as volumes without a device_path are not supported by
# os-brick encryptors. For volumes with a device_path the calls to
# the os-brick encryptors are safe as they are actually idempotent,
# ignoring any failures caused by the volumes actually being natively
# attached previously.
if (encryption and connection_info['data'].get('device_path') is None):
return
if encryption:
encryptor = self._get_volume_encryptor(connection_info,
encryption)
encryptor.detach_volume(**encryption)
def _check_discard_for_attach_volume(self, conf, instance):
"""Perform some checks for volumes configured for discard support.
If discard is configured for the volume, and the guest is using a
configuration known to not work, we will log a message explaining
the reason why.
"""
if conf.driver_discard == 'unmap' and conf.target_bus == 'virtio':
LOG.debug('Attempting to attach volume %(id)s with discard '
'support enabled to an instance using an '
'unsupported configuration. target_bus = '
'%(bus)s. Trim commands will not be issued to '
'the storage device.',
{'bus': conf.target_bus,
'id': conf.serial},
instance=instance)
def attach_volume(self, context, connection_info, instance, mountpoint,
disk_bus=None, device_type=None, encryption=None):
guest = self._host.get_guest(instance)
disk_dev = mountpoint.rpartition("/")[2]
bdm = {
'device_name': disk_dev,
'disk_bus': disk_bus,
'device_type': device_type}
# Note(cfb): If the volume has a custom block size, check that that we
# are using QEMU/KVM. The presence of a block size is considered
# mandatory by cinder so we fail if we can't honor the request.
data = {}
if ('data' in connection_info):
data = connection_info['data']
if ('logical_block_size' in data or 'physical_block_size' in data):
if CONF.libvirt.virt_type not in ["kvm", "qemu"]:
msg = _("Volume sets block size, but the current "
"libvirt hypervisor '%s' does not support custom "
"block size") % CONF.libvirt.virt_type
raise exception.InvalidHypervisorType(msg)
self._connect_volume(context, connection_info, instance,
encryption=encryption)
disk_info = blockinfo.get_info_from_bdm(
instance, CONF.libvirt.virt_type, instance.image_meta, bdm)
if disk_info['bus'] == 'scsi':
disk_info['unit'] = self._get_scsi_controller_next_unit(guest)
conf = self._get_volume_config(instance, connection_info, disk_info)
self._check_discard_for_attach_volume(conf, instance)
try:
state = guest.get_power_state(self._host)
live = state in (power_state.RUNNING, power_state.PAUSED)
guest.attach_device(conf, persistent=True, live=live)
# NOTE(artom) If we're attaching with a device role tag, we need to
# rebuild device_metadata. If we're attaching without a role
# tag, we're rebuilding it here needlessly anyways. This isn't a
# massive deal, and it helps reduce code complexity by not having
# to indicate to the virt driver that the attach is tagged. The
# really important optimization of not calling the database unless
# device_metadata has actually changed is done for us by
# instance.save().
instance.device_metadata = self._build_device_metadata(
context, instance)
instance.save()
except Exception:
LOG.exception('Failed to attach volume at mountpoint: %s',
mountpoint, instance=instance)
with excutils.save_and_reraise_exception():
self._disconnect_volume(context, connection_info, instance,
encryption=encryption)
def _swap_volume(self, guest, disk_dev, conf, resize_to):
"""Swap existing disk with a new block device.
Call virDomainBlockRebase or virDomainBlockCopy with Libvirt >= 6.0.0
to copy and then pivot to a new volume.
:param: guest: Guest object representing the guest domain
:param: disk_dev: Device within the domain that is being swapped
:param: conf: LibvirtConfigGuestDisk object representing the new volume
:param: resize_to: Size of the dst volume, 0 if the same as the src
"""
dev = guest.get_block_device(disk_dev)
# Save a copy of the domain's persistent XML file. We'll use this
# to redefine the domain if anything fails during the volume swap.
xml = guest.get_xml_desc(dump_inactive=True, dump_sensitive=True)
# Abort is an idempotent operation, so make sure any block
# jobs which may have failed are ended.
try:
dev.abort_job()
except Exception:
pass
try:
# NOTE (rmk): virDomainBlockRebase and virDomainBlockCopy cannot be
# executed on persistent domains, so we need to temporarily
# undefine it. If any part of this block fails, the domain is
# re-defined regardless.
if guest.has_persistent_configuration():
guest.delete_configuration()
try:
dev.copy(conf.to_xml(), reuse_ext=True)
while not dev.is_job_complete():
time.sleep(0.5)
dev.abort_job(pivot=True)
except Exception as exc:
# NOTE(lyarwood): conf.source_path is not set for RBD disks so
# fallback to conf.target_dev when None.
new_path = conf.source_path or conf.target_dev
old_path = disk_dev
LOG.exception("Failure rebasing volume %(new_path)s on "
"%(old_path)s.", {'new_path': new_path,
'old_path': old_path})
raise exception.VolumeRebaseFailed(reason=str(exc))
if resize_to:
dev.resize(resize_to * units.Gi)
# Make sure we will redefine the domain using the updated
# configuration after the volume was swapped. The dump_inactive
# keyword arg controls whether we pull the inactive (persistent)
# or active (live) config from the domain. We want to pull the
# live config after the volume was updated to use when we redefine
# the domain.
xml = guest.get_xml_desc(dump_inactive=False, dump_sensitive=True)
finally:
self._host.write_instance_config(xml)
def swap_volume(self, context, old_connection_info,
new_connection_info, instance, mountpoint, resize_to):
# NOTE(lyarwood): https://bugzilla.redhat.com/show_bug.cgi?id=760547
old_encrypt = self._get_volume_encryption(context, old_connection_info)
new_encrypt = self._get_volume_encryption(context, new_connection_info)
if ((old_encrypt and self._allow_native_luksv1(old_encrypt)) or
(new_encrypt and self._allow_native_luksv1(new_encrypt))):
raise NotImplementedError(_("Swap volume is not supported for "
"encrypted volumes when native LUKS decryption is enabled."))
guest = self._host.get_guest(instance)
disk_dev = mountpoint.rpartition("/")[2]
if not guest.get_disk(disk_dev):
raise exception.DiskNotFound(location=disk_dev)
disk_info = {
'dev': disk_dev,
'bus': blockinfo.get_disk_bus_for_disk_dev(
CONF.libvirt.virt_type, disk_dev),
'type': 'disk',
}
# NOTE (lyarwood): new_connection_info will be modified by the
# following _connect_volume call down into the volume drivers. The
# majority of the volume drivers will add a device_path that is in turn
# used by _get_volume_config to set the source_path of the
# LibvirtConfigGuestDisk object it returns. We do not explicitly save
# this to the BDM here as the upper compute swap_volume method will
# eventually do this for us.
self._connect_volume(context, new_connection_info, instance)
conf = self._get_volume_config(
instance, new_connection_info, disk_info)
try:
self._swap_volume(guest, disk_dev, conf, resize_to)
except exception.VolumeRebaseFailed:
with excutils.save_and_reraise_exception():
self._disconnect_volume(context, new_connection_info, instance)
self._disconnect_volume(context, old_connection_info, instance)
def _get_existing_domain_xml(self, instance, network_info,
block_device_info=None, share_info=None):
try:
guest = self._host.get_guest(instance)
xml = guest.get_xml_desc()
except exception.InstanceNotFound:
disk_info = blockinfo.get_disk_info(CONF.libvirt.virt_type,
instance,
instance.image_meta,
block_device_info)
xml = self._get_guest_xml(nova_context.get_admin_context(),
instance, network_info, disk_info,
instance.image_meta,
block_device_info=block_device_info,
share_info=share_info)
return xml
def emit_event(self, event: virtevent.InstanceEvent) -> None:
"""Handles libvirt specific events locally and dispatches the rest to
the compute manager.
"""
if isinstance(event, libvirtevent.LibvirtEvent):
# These are libvirt specific events handled here on the driver
# level instead of propagating them to the compute manager level
if isinstance(event, libvirtevent.DeviceEvent):
had_clients = self._device_event_handler.notify_waiters(event)
if had_clients:
LOG.debug(
"Received event %s from libvirt while the driver is "
"waiting for it; dispatched.",
event,
)
else:
LOG.warning(
"Received event %s from libvirt but the driver is not "
"waiting for it; ignored.",
event,
)
else:
LOG.debug(
"Received event %s from libvirt but no handler is "
"implemented for it in the libvirt driver so it is "
"ignored", event)
else:
# Let the generic driver code dispatch the event to the compute
# manager
super().emit_event(event)
def _detach_with_retry(
self,
guest: libvirt_guest.Guest,
instance_uuid: str,
# to properly typehint this param we would need typing.Protocol but
# that is only available since python 3.8
get_device_conf_func: ty.Callable,
device_name: str,
) -> None:
"""Detaches a device from the guest
If the guest is a running state then the detach is performed on both
the persistent and live domains.
In case of live detach this call will wait for the libvirt event
signalling the end of the detach process.
If the live detach times out then it will retry the detach. Detach from
the persistent config is not retried as it is:
* synchronous and no event is sent from libvirt
* it is always expected to succeed if the device is in the domain
config
:param guest: the guest we are detach the device from
:param instance_uuid: the UUID of the instance we are detaching the
device from
:param get_device_conf_func: function which returns the configuration
for device from the domain, having one optional boolean parameter
`from_persistent_config` to select which domain config to query
:param device_name: This is the name of the device used solely for
error messages. Note that it is not the same as the device alias
used by libvirt to identify the device.
:raises exception.DeviceNotFound: if the device does not exist in the
domain even before we try to detach or if libvirt reported that the
device is missing from the domain synchronously.
:raises exception.DeviceDetachFailed: if libvirt reported error during
detaching from the live domain or we timed out waiting for libvirt
events and run out of retries
:raises libvirt.libvirtError: for any other errors reported by libvirt
synchronously.
"""
state = guest.get_power_state(self._host)
live = state in (power_state.RUNNING, power_state.PAUSED)
persistent = guest.has_persistent_configuration()
if not persistent and not live:
# nothing to do
return
persistent_dev = None
if persistent:
persistent_dev = get_device_conf_func(from_persistent_config=True)
live_dev = None
if live:
live_dev = get_device_conf_func()
# didn't find the device in either domain
if persistent_dev is None and live_dev is None:
raise exception.DeviceNotFound(device=device_name)
if persistent_dev:
try:
self._detach_from_persistent(
guest, instance_uuid, persistent_dev, get_device_conf_func,
device_name)
except exception.DeviceNotFound:
if live_dev:
# ignore the error so that we can do the live detach
LOG.warning(
'Libvirt reported sync error while detaching '
'device %s from instance %s from the persistent '
'domain config. Ignoring the error to proceed with '
'live detach as the device exists in the live domain.',
device_name, instance_uuid)
else:
# if only persistent detach was requested then give up
raise
if live_dev:
self._detach_from_live_with_retry(
guest, instance_uuid, live_dev, get_device_conf_func,
device_name)
def _detach_from_persistent(
self,
guest: libvirt_guest.Guest,
instance_uuid: str,
persistent_dev: ty.Union[
vconfig.LibvirtConfigGuestDisk,
vconfig.LibvirtConfigGuestInterface],
get_device_conf_func,
device_name: str,
):
LOG.debug(
'Attempting to detach device %s from instance %s from '
'the persistent domain config.', device_name, instance_uuid)
self._detach_sync(
persistent_dev, guest, instance_uuid, device_name,
persistent=True, live=False)
# make sure the dev is really gone
persistent_dev = get_device_conf_func(
from_persistent_config=True)
if not persistent_dev:
LOG.info(
'Successfully detached device %s from instance %s '
'from the persistent domain config.',
device_name, instance_uuid)
else:
# Based on the libvirt devs this should never happen
LOG.warning(
'Failed to detach device %s from instance %s '
'from the persistent domain config. Libvirt did not '
'report any error but the device is still in the '
'config.', device_name, instance_uuid)
def _detach_from_live_with_retry(
self,
guest: libvirt_guest.Guest,
instance_uuid: str,
live_dev: ty.Union[
vconfig.LibvirtConfigGuestDisk,
vconfig.LibvirtConfigGuestInterface],
get_device_conf_func,
device_name: str,
):
max_attempts = CONF.libvirt.device_detach_attempts
for attempt in range(max_attempts):
LOG.debug(
'(%s/%s): Attempting to detach device %s with device '
'alias %s from instance %s from the live domain config.',
attempt + 1, max_attempts, device_name, live_dev.alias,
instance_uuid)
self._detach_from_live_and_wait_for_event(
live_dev, guest, instance_uuid, device_name)
# make sure the dev is really gone
live_dev = get_device_conf_func()
if not live_dev:
LOG.info(
'Successfully detached device %s from instance %s '
'from the live domain config.', device_name, instance_uuid)
# we are done
return
LOG.debug(
'Failed to detach device %s with device alias %s from '
'instance %s from the live domain config. Libvirt did not '
'report any error but the device is still in the config.',
device_name, live_dev.alias, instance_uuid)
msg = (
'Run out of retry while detaching device %s with device '
'alias %s from instance %s from the live domain config. '
'Device is still attached to the guest.')
LOG.error(msg, device_name, live_dev.alias, instance_uuid)
raise exception.DeviceDetachFailed(
device=device_name,
reason=msg % (device_name, live_dev.alias, instance_uuid))
def _detach_from_live_and_wait_for_event(
self,
dev: ty.Union[
vconfig.LibvirtConfigGuestDisk,
vconfig.LibvirtConfigGuestInterface],
guest: libvirt_guest.Guest,
instance_uuid: str,
device_name: str,
) -> None:
"""Detaches a device from the live config of the guest and waits for
the libvirt event singling the finish of the detach.
:param dev: the device configuration to be detached
:param guest: the guest we are detach the device from
:param instance_uuid: the UUID of the instance we are detaching the
device from
:param device_name: This is the name of the device used solely for
error messages.
:raises exception.DeviceNotFound: if libvirt reported that the device
is missing from the domain synchronously.
:raises libvirt.libvirtError: for any other errors reported by libvirt
synchronously.
:raises DeviceDetachFailed: if libvirt sent DeviceRemovalFailedEvent
"""
# So we will issue an detach to libvirt and we will wait for an
# event from libvirt about the result. We need to set up the event
# handling before the detach to avoid missing the event if libvirt
# is really fast
# NOTE(gibi): we need to use the alias name of the device as that
# is what libvirt will send back to us in the event
waiter = self._device_event_handler.create_waiter(
instance_uuid, dev.alias,
{libvirtevent.DeviceRemovedEvent,
libvirtevent.DeviceRemovalFailedEvent})
try:
self._detach_sync(
dev, guest, instance_uuid, device_name, persistent=False,
live=True)
except Exception:
# clean up the libvirt event handler as we failed synchronously
self._device_event_handler.delete_waiter(waiter)
raise
LOG.debug(
'Start waiting for the detach event from libvirt for '
'device %s with device alias %s for instance %s',
device_name, dev.alias, instance_uuid)
# We issued the detach without any exception so we can wait for
# a libvirt event to arrive to notify us about the result
# NOTE(gibi): we expect that this call will be unblocked by an
# incoming libvirt DeviceRemovedEvent or DeviceRemovalFailedEvent
event = self._device_event_handler.wait(
waiter, timeout=CONF.libvirt.device_detach_timeout)
if not event:
# This should not happen based on information from the libvirt
# developers. But it does at least during the cleanup of the
# tempest test case
# ServerRescueNegativeTestJSON.test_rescued_vm_detach_volume
# Log a warning and let the upper layer detect that the device is
# still attached and retry
LOG.warning(
'Waiting for libvirt event about the detach of '
'device %s with device alias %s from instance %s is timed '
'out.', device_name, dev.alias, instance_uuid)
if isinstance(event, libvirtevent.DeviceRemovalFailedEvent):
# Based on the libvirt developers this signals a permanent failure
LOG.error(
'Received DeviceRemovalFailedEvent from libvirt for the '
'detach of device %s with device alias %s from instance %s ',
device_name, dev.alias, instance_uuid)
raise exception.DeviceDetachFailed(
device=device_name,
reason="DeviceRemovalFailedEvent received from libvirt")
@staticmethod
def _detach_sync(
dev: ty.Union[
vconfig.LibvirtConfigGuestDisk,
vconfig.LibvirtConfigGuestInterface],
guest: libvirt_guest.Guest,
instance_uuid: str,
device_name: str,
persistent: bool,
live: bool,
):
"""Detaches a device from the guest without waiting for libvirt events
It only handles synchronous errors (i.e. exceptions) but does not wait
for any event from libvirt.
:param dev: the device configuration to be detached
:param guest: the guest we are detach the device from
:param instance_uuid: the UUID of the instance we are detaching the
device from
:param device_name: This is the name of the device used solely for
error messages.
:param live: detach the device from the live domain config only
:param persistent: detach the device from the persistent domain config
only
:raises exception.DeviceNotFound: if libvirt reported that the device
is missing from the domain synchronously.
:raises libvirt.libvirtError: for any other errors reported by libvirt
synchronously.
"""
try:
guest.detach_device(dev, persistent=persistent, live=live)
except libvirt.libvirtError as ex:
code = ex.get_error_code()
msg = ex.get_error_message()
LOG.debug(
"Libvirt returned error while detaching device %s from "
"instance %s. Libvirt error code: %d, error message: %s.",
device_name, instance_uuid, code, msg
)
if (code == libvirt.VIR_ERR_DEVICE_MISSING or
# Libvirt 4.1 improved error code usage but OPERATION_FAILED
# still used in one case during detach:
# https://github.com/libvirt/libvirt/blob/55ea45acc99c549c7757efe954aacc33ad30a8ef/src/qemu/qemu_hotplug.c#L5324-L5328
# TODO(gibi): remove this when a future version of libvirt
# transform this error to VIR_ERR_DEVICE_MISSING too.
(code == libvirt.VIR_ERR_OPERATION_FAILED and
'not found' in msg)
):
LOG.debug(
'Libvirt failed to detach device %s from instance %s '
'synchronously (persistent=%s, live=%s) with error: %s.',
device_name, instance_uuid, persistent, live, str(ex))
raise exception.DeviceNotFound(device=device_name) from ex
# NOTE(lyarwood): https://bugzilla.redhat.com/1878659
# Ignore this known QEMU bug for the time being allowing
# our retry logic to handle it.
# NOTE(gibi): This can only happen in case of detaching from the
# live domain as we never retry a detach from the persistent
# domain so we cannot hit an already running detach there.
# In case of detaching from the live domain this error can happen
# if the caller timed out during the first detach attempt then saw
# that the device is still attached and therefore looped over and
# and retried the detach. In this case the previous attempt stopped
# waiting for the libvirt event. Also libvirt reports that there is
# a detach ongoing, so the current attempt expects that a
# libvirt event will be still emitted. Therefore we simply return
# from here. Then the caller will wait for such event.
if (code == libvirt.VIR_ERR_INTERNAL_ERROR and msg and
'already in the process of unplug' in msg
):
LOG.debug(
'Ignoring QEMU rejecting our request to detach device %s '
'from instance %s as it is caused by a previous request '
'still being in progress.', device_name, instance_uuid)
return
if code == libvirt.VIR_ERR_NO_DOMAIN:
LOG.warning(
"During device detach, instance disappeared.",
instance_uuid=instance_uuid)
# if the domain has disappeared then we have nothing to detach
return
LOG.warning(
'Unexpected libvirt error while detaching device %s from '
'instance %s: %s', device_name, instance_uuid, str(ex))
raise
def _get_guest_disk_device(self, guest, disk_dev, volume_uuid=None,
from_persistent_config=False):
"""Attempt to find the guest disk
If a volume_uuid is provided, we will look for the device based
on the nova-specified alias. If not, or we do not find it that way,
fall back to the old way of using the disk_dev.
"""
if volume_uuid is not None:
dev_alias = vconfig.make_libvirt_device_alias(volume_uuid)
dev = guest.get_device_by_alias(
dev_alias,
from_persistent_config=from_persistent_config)
if dev:
LOG.debug('Found disk %s by alias %s', disk_dev, dev_alias)
return dev
dev = guest.get_disk(disk_dev,
from_persistent_config=from_persistent_config)
if dev:
# NOTE(danms): Only log that we fell back to the old way if it
# worked. Since we call this method after detach is done to
# ensure it is gone, we will always "fall back" to make sure it
# is gone by the "old way" and thus shouldn't announce it.
LOG.info('Device %s not found by alias %s, falling back',
disk_dev, dev_alias)
return dev
def detach_volume(self, context, connection_info, instance, mountpoint,
encryption=None):
disk_dev = mountpoint.rpartition("/")[2]
try:
guest = self._host.get_guest(instance)
# NOTE(lyarwood): The volume must be detached from the VM before
# detaching any attached encryptors or disconnecting the underlying
# volume in _disconnect_volume. Otherwise, the encryptor or volume
# driver may report that the volume is still in use.
volume_id = driver_block_device.get_volume_id(connection_info)
get_dev = functools.partial(self._get_guest_disk_device,
guest,
disk_dev,
volume_uuid=volume_id)
self._detach_with_retry(
guest,
instance.uuid,
get_dev,
device_name=disk_dev,
)
except exception.InstanceNotFound:
# NOTE(zhaoqin): If the instance does not exist, _lookup_by_name()
# will throw InstanceNotFound exception. Need to
# disconnect volume under this circumstance.
LOG.warning("During detach_volume, instance disappeared.",
instance=instance)
except exception.DeviceNotFound:
# We should still try to disconnect logical device from
# host, an error might have happened during a previous
# call.
LOG.info("Device %s not found in instance.",
disk_dev, instance=instance)
self._disconnect_volume(context, connection_info, instance,
encryption=encryption)
def _resize_attached_volume(self, new_size, block_device, instance):
LOG.debug('Resizing target device %(dev)s to %(size)u',
{'dev': block_device._disk, 'size': new_size},
instance=instance)
block_device.resize(new_size)
def _resize_attached_encrypted_volume(self, context, original_new_size,
block_device, instance,
connection_info, encryption):
# TODO(lyarwood): Also handle the dm-crpyt encryption providers of
# plain and LUKSv2, for now just use the original_new_size.
decrypted_device_new_size = original_new_size
# NOTE(lyarwood): original_new_size currently refers to the total size
# of the extended volume in bytes. With natively decrypted LUKSv1
# volumes we need to ensure this now takes the LUKSv1 header and key
# material into account. Otherwise QEMU will attempt and fail to grow
# host block devices and remote RBD volumes.
if self._allow_native_luksv1(encryption):
try:
# NOTE(lyarwood): Find the path to provide to qemu-img
if 'device_path' in connection_info['data']:
path = connection_info['data']['device_path']
elif connection_info['driver_volume_type'] == 'rbd':
volume_name = connection_info['data']['name']
path = f"rbd:{volume_name}"
if connection_info['data'].get('auth_enabled'):
username = connection_info['data']['auth_username']
path = f"rbd:{volume_name}:id={username}"
else:
path = 'unknown'
raise exception.DiskNotFound(location='unknown')
info = images.privileged_qemu_img_info(path)
format_specific_data = info.format_specific['data']
payload_offset = format_specific_data['payload-offset']
# NOTE(lyarwood): Ensure the underlying device is not resized
# by subtracting the LUKSv1 payload_offset (where the users
# encrypted data starts) from the original_new_size (the total
# size of the underlying volume). Both are reported in bytes.
decrypted_device_new_size = original_new_size - payload_offset
except exception.DiskNotFound:
with excutils.save_and_reraise_exception():
LOG.exception('Unable to access the encrypted disk %s.',
path, instance=instance)
except Exception:
with excutils.save_and_reraise_exception():
LOG.exception('Unknown error when attempting to find the '
'payload_offset for LUKSv1 encrypted disk '
'%s.', path, instance=instance)
else: # os-brick encryptor driver
encryptor = self._get_volume_encryptor(connection_info, encryption)
decrypted_device_new_size = encryptor.extend_volume(context,
**encryption)
if decrypted_device_new_size is None:
raise exception.VolumeExtendFailed(
volume_id=block_device._disk,
reason="Encryptor extend failed."
)
# NOTE(lyarwood): Resize the decrypted device within the instance to
# the calculated size as with normal volumes.
self._resize_attached_volume(
decrypted_device_new_size, block_device, instance)
def extend_volume(self, context, connection_info, instance,
requested_size):
volume_id = driver_block_device.get_volume_id(connection_info)
try:
new_size = self._extend_volume(
connection_info, instance, requested_size)
# NOTE(lyarwood): Handle cases where os-brick has ignored failures
# and returned an invalid new_size of None through the vol drivers
if new_size is None:
raise exception.VolumeExtendFailed(
volume_id=volume_id,
reason="Failure to resize underlying volume on compute."
)
except NotImplementedError:
raise exception.ExtendVolumeNotSupported()
# Resize the device in QEMU so its size is updated and
# detected by the instance without rebooting.
try:
guest = self._host.get_guest(instance)
state = guest.get_power_state(self._host)
active_state = state in (power_state.RUNNING, power_state.PAUSED)
if active_state:
if 'device_path' in connection_info['data']:
disk_path = connection_info['data']['device_path']
else:
# Some drivers (eg. net) don't put the device_path
# into the connection_info. Match disks by their serial
# number instead
disk = next(iter([
d for d in guest.get_all_disks()
if d.serial == volume_id
]), None)
if not disk:
raise exception.VolumeNotFound(volume_id=volume_id)
disk_path = disk.target_dev
dev = guest.get_block_device(disk_path)
encryption = encryptors.get_encryption_metadata(
context, self._volume_api, volume_id, connection_info)
if encryption:
self._resize_attached_encrypted_volume(
context, new_size, dev, instance,
connection_info, encryption)
else:
self._resize_attached_volume(
new_size, dev, instance)
else:
LOG.debug('Skipping block device resize, guest is not running',
instance=instance)
except exception.InstanceNotFound:
with excutils.save_and_reraise_exception():
LOG.warning('During extend_volume, instance disappeared.',
instance=instance)
except libvirt.libvirtError:
with excutils.save_and_reraise_exception():
LOG.exception('resizing block device failed.',
instance=instance)
def attach_interface(self, context, instance, image_meta, vif):
guest = self._host.get_guest(instance)
self.vif_driver.plug(instance, vif)
cfg = self.vif_driver.get_config(instance, vif, image_meta,
instance.flavor,
CONF.libvirt.virt_type)
if self._sev_enabled(instance.flavor, image_meta):
designer.set_driver_iommu_for_device(cfg)
try:
state = guest.get_power_state(self._host)
live = state in (power_state.RUNNING, power_state.PAUSED)
guest.attach_device(cfg, persistent=True, live=live)
except libvirt.libvirtError:
LOG.error('attaching network adapter failed.',
instance=instance, exc_info=True)
self.vif_driver.unplug(instance, vif)
raise exception.InterfaceAttachFailed(
instance_uuid=instance.uuid)
try:
# NOTE(artom) If we're attaching with a device role tag, we need to
# rebuild device_metadata. If we're attaching without a role
# tag, we're rebuilding it here needlessly anyways. This isn't a
# massive deal, and it helps reduce code complexity by not having
# to indicate to the virt driver that the attach is tagged. The
# really important optimization of not calling the database unless
# device_metadata has actually changed is done for us by
# instance.save().
instance.device_metadata = self._build_device_metadata(
context, instance)
instance.save()
except Exception:
# NOTE(artom) If we fail here it means the interface attached
# successfully but building and/or saving the device metadata
# failed. Just unplugging the vif is therefore not enough cleanup,
# we need to detach the interface.
with excutils.save_and_reraise_exception(reraise=False):
LOG.error('Interface attached successfully but building '
'and/or saving device metadata failed.',
instance=instance, exc_info=True)
self.detach_interface(context, instance, vif)
raise exception.InterfaceAttachFailed(
instance_uuid=instance.uuid)
try:
guest.set_metadata(
self._get_guest_config_meta(
self.get_instance_driver_metadata(
instance, instance.get_network_info())))
except libvirt.libvirtError:
LOG.warning('updating libvirt metadata failed.', instance=instance)
def detach_interface(self, context, instance, vif):
guest = self._host.get_guest(instance)
cfg = self.vif_driver.get_config(instance, vif,
instance.image_meta,
instance.flavor,
CONF.libvirt.virt_type)
try:
get_dev = functools.partial(guest.get_interface_by_cfg, cfg)
self._detach_with_retry(
guest,
instance.uuid,
get_dev,
device_name=self.vif_driver.get_vif_devname(vif),
)
except exception.DeviceNotFound:
# The interface is gone so just log it as a warning.
LOG.warning('Detaching interface %(mac)s failed because '
'the device is no longer found on the guest.',
{'mac': vif.get('address')}, instance=instance)
finally:
# NOTE(gibi): we need to unplug the vif _after_ the detach is done
# on the libvirt side as otherwise libvirt will still manage the
# device that our unplug code trying to reset. This can cause a
# race and leave the detached device configured. Also even if we
# are failed to detach due to race conditions the unplug is
# necessary for the same reason
self.vif_driver.unplug(instance, vif)
try:
# NOTE(nmiki): In order for the interface to be removed from
# network_info, the nova-compute process need to wait for
# processing on the neutron side.
# Here, I simply exclude the target VIF from metadata.
network_info = list(filter(lambda info: info['id'] != vif['id'],
instance.get_network_info()))
guest.set_metadata(
self._get_guest_config_meta(
self.get_instance_driver_metadata(
instance, network_info)))
except libvirt.libvirtError:
LOG.warning('updating libvirt metadata failed.', instance=instance)
def _create_snapshot_metadata(self, image_meta, instance,
img_fmt, snp_name):
metadata = {'status': 'active',
'name': snp_name,
'properties': {
'kernel_id': instance.kernel_id,
'image_location': 'snapshot',
'image_state': 'available',
'owner_id': instance.project_id,
'ramdisk_id': instance.ramdisk_id,
}
}
if instance.os_type:
metadata['properties']['os_type'] = instance.os_type
metadata['disk_format'] = img_fmt
if image_meta.obj_attr_is_set("container_format"):
metadata['container_format'] = image_meta.container_format
else:
metadata['container_format'] = "bare"
return metadata
def snapshot(self, context, instance, image_id, update_task_state):
"""Create snapshot from a running VM instance.
This command only works with qemu 0.14+
"""
try:
guest = self._host.get_guest(instance)
except exception.InstanceNotFound:
raise exception.InstanceNotRunning(instance_id=instance.uuid)
snapshot = self._image_api.get(context, image_id)
# source_format is an on-disk format
# source_type is a backend type
disk_path, source_format = libvirt_utils.find_disk(guest)
source_type = libvirt_utils.get_disk_type_from_path(disk_path)
# We won't have source_type for raw or qcow2 disks, because we can't
# determine that from the path. We should have it from the libvirt
# xml, though.
if source_type is None:
source_type = source_format
# For lxc instances we won't have it either from libvirt xml
# (because we just gave libvirt the mounted filesystem), or the path,
# so source_type is still going to be None. In this case,
# root_disk is going to default to CONF.libvirt.images_type
# below, which is still safe.
image_format = CONF.libvirt.snapshot_image_format or source_type
# NOTE(bfilippov): save lvm and rbd as raw
if image_format == 'lvm' or image_format == 'rbd':
image_format = 'raw'
metadata = self._create_snapshot_metadata(instance.image_meta,
instance,
image_format,
snapshot['name'])
snapshot_name = uuidutils.generate_uuid(dashed=False)
# store current state so we know what to resume back to if we suspend
original_power_state = guest.get_power_state(self._host)
# NOTE(dgenin): Instances with LVM encrypted ephemeral storage require
# cold snapshots. Currently, checking for encryption is
# redundant because LVM supports only cold snapshots.
# It is necessary in case this situation changes in the
# future.
if (
self._host.has_min_version(hv_type=host.HV_DRIVER_QEMU) and
source_type != 'lvm' and
not CONF.ephemeral_storage_encryption.enabled and
not CONF.workarounds.disable_libvirt_livesnapshot and
# NOTE(stephenfin): Live snapshotting doesn't make sense for
# shutdown instances
original_power_state != power_state.SHUTDOWN
):
live_snapshot = True
else:
live_snapshot = False
self._suspend_guest_for_snapshot(
context, live_snapshot, original_power_state, instance)
root_disk = self.image_backend.by_libvirt_path(
instance, disk_path, image_type=source_type)
if live_snapshot:
LOG.info("Beginning live snapshot process", instance=instance)
else:
LOG.info("Beginning cold snapshot process", instance=instance)
update_task_state(task_state=task_states.IMAGE_PENDING_UPLOAD)
update_task_state(task_state=task_states.IMAGE_UPLOADING,
expected_state=task_states.IMAGE_PENDING_UPLOAD)
try:
metadata['location'] = root_disk.direct_snapshot(
context, snapshot_name, image_format, image_id,
instance.image_ref)
self._resume_guest_after_snapshot(
context, live_snapshot, original_power_state, instance, guest)
self._image_api.update(context, image_id, metadata,
purge_props=False)
except (NotImplementedError, exception.ImageUnacceptable,
exception.Forbidden) as e:
if type(e) is not NotImplementedError:
LOG.warning('Performing standard snapshot because direct '
'snapshot failed: %(error)s',
{'error': e})
failed_snap = metadata.pop('location', None)
if failed_snap:
failed_snap = {'url': str(failed_snap)}
root_disk.cleanup_direct_snapshot(failed_snap,
also_destroy_volume=True,
ignore_errors=True)
update_task_state(task_state=task_states.IMAGE_PENDING_UPLOAD,
expected_state=task_states.IMAGE_UPLOADING)
# TODO(nic): possibly abstract this out to the root_disk
if source_type == 'rbd' and live_snapshot:
# Standard snapshot uses qemu-img convert from RBD which is
# not safe to run with live_snapshot.
live_snapshot = False
# Suspend the guest, so this is no longer a live snapshot
self._suspend_guest_for_snapshot(
context, live_snapshot, original_power_state, instance)
snapshot_directory = CONF.libvirt.snapshots_directory
fileutils.ensure_tree(snapshot_directory)
with utils.tempdir(dir=snapshot_directory) as tmpdir:
try:
out_path = os.path.join(tmpdir, snapshot_name)
if live_snapshot:
# NOTE(xqueralt): libvirt needs o+x in the tempdir
os.chmod(tmpdir, 0o701)
self._live_snapshot(context, instance, guest,
disk_path, out_path, source_format,
image_format, instance.image_meta)
else:
root_disk.snapshot_extract(out_path, image_format)
LOG.info("Snapshot extracted, beginning image upload",
instance=instance)
except libvirt.libvirtError as ex:
error_code = ex.get_error_code()
if error_code == libvirt.VIR_ERR_NO_DOMAIN:
LOG.info('Instance %(instance_name)s disappeared '
'while taking snapshot of it: [Error Code '
'%(error_code)s] %(ex)s',
{'instance_name': instance.name,
'error_code': error_code,
'ex': ex},
instance=instance)
raise exception.InstanceNotFound(
instance_id=instance.uuid)
else:
raise
finally:
self._resume_guest_after_snapshot(
context, live_snapshot, original_power_state, instance,
guest)
# Upload that image to the image service
update_task_state(task_state=task_states.IMAGE_UPLOADING,
expected_state=task_states.IMAGE_PENDING_UPLOAD)
with libvirt_utils.file_open(out_path, 'rb') as image_file:
# execute operation with disk concurrency semaphore
with compute_utils.disk_ops_semaphore:
self._image_api.update(context,
image_id,
metadata,
image_file)
except exception.ImageNotFound:
with excutils.save_and_reraise_exception():
LOG.warning("Failed to snapshot image because it was deleted")
failed_snap = metadata.pop('location', None)
if failed_snap:
failed_snap = {'url': str(failed_snap)}
root_disk.cleanup_direct_snapshot(
failed_snap, also_destroy_volume=True,
ignore_errors=True)
except Exception:
with excutils.save_and_reraise_exception():
LOG.exception("Failed to snapshot image")
failed_snap = metadata.pop('location', None)
if failed_snap:
failed_snap = {'url': str(failed_snap)}
root_disk.cleanup_direct_snapshot(
failed_snap, also_destroy_volume=True,
ignore_errors=True)
LOG.info("Snapshot image upload complete", instance=instance)
def _needs_suspend_resume_for_snapshot(
self,
live_snapshot: bool,
current_power_state: int,
):
# NOTE(dkang): managedSave does not work for LXC
if CONF.libvirt.virt_type == 'lxc':
return False
# Live snapshots do not necessitate suspending the domain
if live_snapshot:
return False
# ...and neither does a non-running domain
return current_power_state in (power_state.RUNNING, power_state.PAUSED)
def _suspend_guest_for_snapshot(
self,
context: nova_context.RequestContext,
live_snapshot: bool,
current_power_state: int,
instance: 'objects.Instance',
):
if self._needs_suspend_resume_for_snapshot(
live_snapshot, current_power_state,
):
self.suspend(context, instance)
def _resume_guest_after_snapshot(
self,
context: nova_context.RequestContext,
live_snapshot: bool,
original_power_state: int,
instance: 'objects.Instance',
guest: libvirt_guest.Guest,
):
if not self._needs_suspend_resume_for_snapshot(
live_snapshot, original_power_state,
):
return
current_power_state = guest.get_power_state(self._host)
self.cpu_api.power_up_for_instance(instance)
# TODO(stephenfin): Any reason we couldn't use 'self.resume' here?
guest.launch(pause=current_power_state == power_state.PAUSED)
self._attach_pci_devices(
guest,
instance.get_pci_devices(
source=objects.InstancePCIRequest.FLAVOR_ALIAS
),
)
self._attach_direct_passthrough_ports(context, instance, guest)
def _can_set_admin_password(self, image_meta):
if CONF.libvirt.virt_type in ('kvm', 'qemu'):
if not image_meta.properties.get('hw_qemu_guest_agent', False):
raise exception.QemuGuestAgentNotEnabled()
elif not CONF.libvirt.virt_type == 'parallels':
raise exception.SetAdminPasswdNotSupported()
def _save_instance_password_if_sshkey_present(self, instance, new_pass):
sshkey = instance.key_data if 'key_data' in instance else None
if sshkey and sshkey.startswith("ssh-rsa"):
enc = crypto.ssh_encrypt_text(sshkey, new_pass)
# NOTE(melwitt): The convert_password method doesn't actually do
# anything with the context argument, so we can pass None.
instance.system_metadata.update(
password.convert_password(None, base64.encode_as_text(enc)))
instance.save()
def set_admin_password(self, instance, new_pass):
self._can_set_admin_password(instance.image_meta)
guest = self._host.get_guest(instance)
user = instance.image_meta.properties.get("os_admin_user")
if not user:
if instance.os_type == "windows":
user = "Administrator"
else:
user = "root"
try:
guest.set_user_password(user, new_pass)
except libvirt.libvirtError as ex:
error_code = ex.get_error_code()
if error_code == libvirt.VIR_ERR_AGENT_UNRESPONSIVE:
LOG.debug('Failed to set password: QEMU agent unresponsive',
instance_uuid=instance.uuid)
raise NotImplementedError()
msg = (_('Error from libvirt while set password for username '
'"%(user)s": [Error Code %(error_code)s] %(ex)s')
% {'user': user, 'error_code': error_code, 'ex': ex})
raise exception.InternalError(msg)
else:
# Save the password in sysmeta so it may be retrieved from the
# metadata service.
self._save_instance_password_if_sshkey_present(instance, new_pass)
def _can_quiesce(self, instance, image_meta):
if CONF.libvirt.virt_type not in ('kvm', 'qemu'):
raise exception.InstanceQuiesceNotSupported(
instance_id=instance.uuid)
if not image_meta.properties.get('hw_qemu_guest_agent', False):
raise exception.QemuGuestAgentNotEnabled()
def _requires_quiesce(self, image_meta):
return image_meta.properties.get('os_require_quiesce', False)
def _set_quiesced(self, context, instance, image_meta, quiesced):
self._can_quiesce(instance, image_meta)
try:
guest = self._host.get_guest(instance)
if quiesced:
guest.freeze_filesystems()
else:
guest.thaw_filesystems()
except libvirt.libvirtError as ex:
error_code = ex.get_error_code()
msg = (_('Error from libvirt while quiescing %(instance_name)s: '
'[Error Code %(error_code)s] %(ex)s')
% {'instance_name': instance.name,
'error_code': error_code, 'ex': ex})
if error_code == libvirt.VIR_ERR_AGENT_UNRESPONSIVE:
msg += (", libvirt cannot connect to the qemu-guest-agent"
" inside the instance.")
raise exception.InstanceQuiesceFailed(reason=msg)
else:
raise exception.InternalError(msg)
def quiesce(self, context, instance, image_meta):
"""Freeze the guest filesystems to prepare for snapshot.
The qemu-guest-agent must be setup to execute fsfreeze.
"""
self._set_quiesced(context, instance, image_meta, True)
def unquiesce(self, context, instance, image_meta):
"""Thaw the guest filesystems after snapshot."""
self._set_quiesced(context, instance, image_meta, False)
def _live_snapshot(self, context, instance, guest, disk_path, out_path,
source_format, image_format, image_meta):
"""Snapshot an instance without downtime."""
dev = guest.get_block_device(disk_path)
# Save a copy of the domain's persistent XML file
xml = guest.get_xml_desc(dump_inactive=True, dump_sensitive=True)
# Abort is an idempotent operation, so make sure any block
# jobs which may have failed are ended.
try:
dev.abort_job()
except Exception:
pass
# NOTE (rmk): We are using shallow rebases as a workaround to a bug
# in QEMU 1.3. In order to do this, we need to create
# a destination image with the original backing file
# and matching size of the instance root disk.
src_disk_size = libvirt_utils.get_disk_size(disk_path,
format=source_format)
src_back_path = libvirt_utils.get_disk_backing_file(disk_path,
format=source_format,
basename=False)
disk_delta = out_path + '.delta'
libvirt_utils.create_image(
disk_delta, 'qcow2', src_disk_size, backing_file=src_back_path)
try:
self._can_quiesce(instance, image_meta)
except exception.NovaException as err:
if image_meta.properties.get('os_require_quiesce', False):
LOG.error('Quiescing instance failed but image property '
'"os_require_quiesce" is set: %(reason)s.',
{'reason': err}, instance=instance)
raise
LOG.info('Quiescing instance not available: %(reason)s.',
{'reason': err}, instance=instance)
try:
# NOTE (rmk): blockRebase cannot be executed on persistent
# domains, so we need to temporarily undefine it.
# If any part of this block fails, the domain is
# re-defined regardless.
if guest.has_persistent_configuration():
guest.delete_configuration()
# NOTE (rmk): Establish a temporary mirror of our root disk and
# issue an abort once we have a complete copy.
dev.rebase(disk_delta, copy=True, reuse_ext=True, shallow=True)
while not dev.is_job_complete():
time.sleep(0.5)
finally:
quiesced = False
try:
# NOTE: The freeze FS is applied after the end of
# the mirroring of the disk to minimize the time of
# the freeze. The mirror between both disks is finished,
# sync continuously, and stopped after abort_job().
self.quiesce(context, instance, image_meta)
quiesced = True
except exception.NovaException as err:
LOG.info('Skipping quiescing instance: %(reason)s.',
{'reason': err}, instance=instance)
dev.abort_job()
nova.privsep.path.chown(disk_delta, uid=os.getuid())
self._host.write_instance_config(xml)
if quiesced:
self.unquiesce(context, instance, image_meta)
# Convert the delta (CoW) image with a backing file to a flat
# image with no backing file.
libvirt_utils.extract_snapshot(disk_delta, 'qcow2',
out_path, image_format)
# Remove the disk_delta file once the snapshot extracted, so that
# it doesn't hang around till the snapshot gets uploaded
fileutils.delete_if_exists(disk_delta)
def _volume_snapshot_update_status(self, context, snapshot_id, status):
"""Send a snapshot status update to Cinder.
This method captures and logs exceptions that occur
since callers cannot do anything useful with these exceptions.
Operations on the Cinder side waiting for this will time out if
a failure occurs sending the update.
:param context: security context
:param snapshot_id: id of snapshot being updated
:param status: new status value
"""
try:
self._volume_api.update_snapshot_status(context,
snapshot_id,
status)
except Exception:
LOG.exception('Failed to send updated snapshot status '
'to volume service.')
def _volume_snapshot_create(self, context, instance, guest,
volume_id, new_file):
"""Perform volume snapshot.
:param guest: VM that volume is attached to
:param volume_id: volume UUID to snapshot
:param new_file: relative path to new qcow2 file present on share
"""
xml = guest.get_xml_desc()
xml_doc = etree.fromstring(xml)
device_info = vconfig.LibvirtConfigGuest()
device_info.parse_dom(xml_doc)
disks_to_snap = [] # to be snapshotted by libvirt
network_disks_to_snap = [] # network disks (netfs, etc.)
disks_to_skip = [] # local disks not snapshotted
for guest_disk in device_info.devices:
if (guest_disk.root_name != 'disk'):
continue
if (guest_disk.target_dev is None):
continue
if (guest_disk.serial is None or guest_disk.serial != volume_id):
disks_to_skip.append(guest_disk.target_dev)
continue
# disk is a Cinder volume with the correct volume_id
disk_info = {
'dev': guest_disk.target_dev,
'serial': guest_disk.serial,
'current_file': guest_disk.source_path,
'source_protocol': guest_disk.source_protocol,
'source_name': guest_disk.source_name,
'source_hosts': guest_disk.source_hosts,
'source_ports': guest_disk.source_ports
}
# Determine path for new_file based on current path
if disk_info['current_file'] is not None:
current_file = disk_info['current_file']
new_file_path = os.path.join(os.path.dirname(current_file),
new_file)
disks_to_snap.append((current_file, new_file_path))
# NOTE(mriedem): This used to include a check for gluster in
# addition to netfs since they were added together. Support for
# gluster was removed in the 16.0.0 Pike release. It is unclear,
# however, if other volume drivers rely on the netfs disk source
# protocol.
elif disk_info['source_protocol'] == 'netfs':
network_disks_to_snap.append((disk_info, new_file))
if not disks_to_snap and not network_disks_to_snap:
msg = _('Found no disk to snapshot.')
raise exception.InternalError(msg)
snapshot = vconfig.LibvirtConfigGuestSnapshot()
for current_name, new_filename in disks_to_snap:
snap_disk = vconfig.LibvirtConfigGuestSnapshotDisk()
snap_disk.name = current_name
snap_disk.source_path = new_filename
snap_disk.source_type = 'file'
snap_disk.snapshot = 'external'
snap_disk.driver_name = 'qcow2'
snapshot.add_disk(snap_disk)
for disk_info, new_filename in network_disks_to_snap:
snap_disk = vconfig.LibvirtConfigGuestSnapshotDisk()
snap_disk.name = disk_info['dev']
snap_disk.source_type = 'network'
snap_disk.source_protocol = disk_info['source_protocol']
snap_disk.snapshot = 'external'
snap_disk.source_path = new_filename
old_dir = disk_info['source_name'].split('/')[0]
snap_disk.source_name = '%s/%s' % (old_dir, new_filename)
snap_disk.source_hosts = disk_info['source_hosts']
snap_disk.source_ports = disk_info['source_ports']
snapshot.add_disk(snap_disk)
for dev in disks_to_skip:
snap_disk = vconfig.LibvirtConfigGuestSnapshotDisk()
snap_disk.name = dev
snap_disk.snapshot = 'no'
snapshot.add_disk(snap_disk)
snapshot_xml = snapshot.to_xml()
LOG.debug("snap xml: %s", snapshot_xml, instance=instance)
image_meta = instance.image_meta
try:
# Check to see if we can quiesce the guest before taking the
# snapshot.
self._can_quiesce(instance, image_meta)
try:
guest.snapshot(snapshot, no_metadata=True, disk_only=True,
reuse_ext=True, quiesce=True)
return
except libvirt.libvirtError:
# If the image says that quiesce is required then we fail.
if self._requires_quiesce(image_meta):
raise
LOG.exception('Unable to create quiesced VM snapshot, '
'attempting again with quiescing disabled.',
instance=instance)
except (exception.InstanceQuiesceNotSupported,
exception.QemuGuestAgentNotEnabled) as err:
# If the image says that quiesce is required then we need to fail.
if self._requires_quiesce(image_meta):
raise
LOG.info('Skipping quiescing instance: %(reason)s.',
{'reason': err}, instance=instance)
try:
guest.snapshot(snapshot, no_metadata=True, disk_only=True,
reuse_ext=True, quiesce=False)
except libvirt.libvirtError:
LOG.exception('Unable to create VM snapshot, '
'failing volume_snapshot operation.',
instance=instance)
raise
def _volume_refresh_connection_info(self, context, instance, volume_id):
bdm = objects.BlockDeviceMapping.get_by_volume_and_instance(
context, volume_id, instance.uuid)
driver_bdm = driver_block_device.convert_volume(bdm)
if driver_bdm:
driver_bdm.refresh_connection_info(context, instance,
self._volume_api, self)
def volume_snapshot_create(self, context, instance, volume_id,
create_info):
"""Create snapshots of a Cinder volume via libvirt.
:param instance: VM instance object reference
:param volume_id: id of volume being snapshotted
:param create_info: dict of information used to create snapshots
- snapshot_id : ID of snapshot
- type : qcow2 / <other>
- new_file : qcow2 file created by Cinder which
becomes the VM's active image after
the snapshot is complete
"""
LOG.debug("volume_snapshot_create: create_info: %(c_info)s",
{'c_info': create_info}, instance=instance)
try:
guest = self._host.get_guest(instance)
except exception.InstanceNotFound:
raise exception.InstanceNotRunning(instance_id=instance.uuid)
if create_info['type'] != 'qcow2':
msg = _('Unknown type: %s') % create_info['type']
raise exception.InternalError(msg)
snapshot_id = create_info.get('snapshot_id', None)
if snapshot_id is None:
msg = _('snapshot_id required in create_info')
raise exception.InternalError(msg)
try:
self._volume_snapshot_create(context, instance, guest,
volume_id, create_info['new_file'])
except Exception:
with excutils.save_and_reraise_exception():
LOG.exception('Error occurred during volume_snapshot_create, '
'sending error status to Cinder.',
instance=instance)
self._volume_snapshot_update_status(
context, snapshot_id, 'error')
self._volume_snapshot_update_status(
context, snapshot_id, 'creating')
def _wait_for_snapshot():
snapshot = self._volume_api.get_snapshot(context, snapshot_id)
if snapshot.get('status') != 'creating':
self._volume_refresh_connection_info(context, instance,
volume_id)
raise loopingcall.LoopingCallDone()
timer = loopingcall.FixedIntervalLoopingCall(_wait_for_snapshot)
timer.start(interval=0.5).wait()
@staticmethod
def _rebase_with_qemu_img(source_path, rebase_base):
"""Rebase a disk using qemu-img.
:param source_path: the disk source path to rebase
:type source_path: string
:param rebase_base: the new parent in the backing chain
:type rebase_base: None or string
"""
if rebase_base is None:
# If backing_file is specified as "" (the empty string), then
# the image is rebased onto no backing file (i.e. it will exist
# independently of any backing file).
backing_file = ""
qemu_img_extra_arg = []
else:
# If the rebased image is going to have a backing file then
# explicitly set the backing file format to avoid any security
# concerns related to file format auto detection.
if os.path.isabs(rebase_base):
backing_file = rebase_base
else:
# this is a probably a volume snapshot case where the
# rebase_base is relative. See bug
# https://bugs.launchpad.net/nova/+bug/1885528
backing_file_name = os.path.basename(rebase_base)
volume_path = os.path.dirname(source_path)
backing_file = os.path.join(volume_path, backing_file_name)
b_file_fmt = images.qemu_img_info(backing_file).file_format
qemu_img_extra_arg = ['-F', b_file_fmt]
qemu_img_extra_arg.append(source_path)
# execute operation with disk concurrency semaphore
with compute_utils.disk_ops_semaphore:
processutils.execute("qemu-img", "rebase", "-b", backing_file,
*qemu_img_extra_arg)
def _volume_snapshot_delete(self, context, instance, volume_id,
snapshot_id, delete_info=None):
"""Note:
if file being merged into == active image:
do a blockRebase (pull) operation
else:
do a blockCommit operation
Files must be adjacent in snap chain.
:param instance: instance object reference
:param volume_id: volume UUID
:param snapshot_id: snapshot UUID (unused currently)
:param delete_info: {
'type': 'qcow2',
'file_to_merge': 'a.img',
'merge_target_file': 'b.img' or None (if merging file_to_merge into
active image)
}
"""
LOG.debug('volume_snapshot_delete: delete_info: %s', delete_info,
instance=instance)
if delete_info['type'] != 'qcow2':
msg = _('Unknown delete_info type %s') % delete_info['type']
raise exception.InternalError(msg)
try:
guest = self._host.get_guest(instance)
except exception.InstanceNotFound:
raise exception.InstanceNotRunning(instance_id=instance.uuid)
# Find dev name
xml = guest.get_xml_desc()
xml_doc = etree.fromstring(xml)
device_info = vconfig.LibvirtConfigGuest()
device_info.parse_dom(xml_doc)
for guest_disk in device_info.devices:
if (guest_disk.root_name != 'disk'):
continue
if (guest_disk.target_dev is None or guest_disk.serial is None):
continue
if (
guest_disk.source_path is None and
guest_disk.source_protocol is None
):
continue
if guest_disk.serial == volume_id:
my_dev = guest_disk.target_dev
active_protocol = guest_disk.source_protocol
active_disk_object = guest_disk
break
else:
LOG.debug('Domain XML: %s', xml, instance=instance)
msg = (_("Disk with id '%s' not found attached to instance.")
% volume_id)
raise exception.InternalError(msg)
LOG.debug("found device at %s", my_dev, instance=instance)
def _get_snap_dev(filename, backing_store):
if filename is None:
msg = _('filename cannot be None')
raise exception.InternalError(msg)
# libgfapi delete
LOG.debug("XML: %s", xml)
LOG.debug("active disk object: %s", active_disk_object)
# determine reference within backing store for desired image
filename_to_merge = filename
matched_name = None
b = backing_store
index = None
current_filename = active_disk_object.source_name.split('/')[1]
if current_filename == filename_to_merge:
return my_dev + '[0]'
while b is not None:
source_filename = b.source_name.split('/')[1]
if source_filename == filename_to_merge:
LOG.debug('found match: %s', b.source_name)
matched_name = b.source_name
index = b.index
break
b = b.backing_store
if matched_name is None:
msg = _('no match found for %s') % (filename_to_merge)
raise exception.InternalError(msg)
LOG.debug('index of match (%s) is %s', b.source_name, index)
my_snap_dev = '%s[%s]' % (my_dev, index)
return my_snap_dev
if delete_info['merge_target_file'] is None:
# pull via blockRebase()
# Merge the most recent snapshot into the active image
rebase_disk = my_dev
rebase_base = delete_info['file_to_merge'] # often None
if (active_protocol is not None) and (rebase_base is not None):
rebase_base = _get_snap_dev(rebase_base,
active_disk_object.backing_store)
relative = rebase_base is not None
LOG.debug(
'disk: %(disk)s, base: %(base)s, '
'bw: %(bw)s, relative: %(relative)s',
{'disk': rebase_disk,
'base': rebase_base,
'bw': libvirt_guest.BlockDevice.REBASE_DEFAULT_BANDWIDTH,
'relative': str(relative)}, instance=instance)
dev = guest.get_block_device(rebase_disk)
if guest.is_active():
result = dev.rebase(rebase_base, relative=relative)
if result == 0:
LOG.debug('blockRebase started successfully',
instance=instance)
while not dev.is_job_complete():
LOG.debug('waiting for blockRebase job completion',
instance=instance)
time.sleep(0.5)
# If the guest is not running libvirt won't do a blockRebase.
# In that case, let's ask qemu-img to rebase the disk.
else:
LOG.debug('Guest is not running so doing a block rebase '
'using "qemu-img rebase"', instance=instance)
# It's unsure how well qemu-img handles network disks for
# every protocol. So let's be safe.
active_protocol = active_disk_object.source_protocol
if active_protocol is not None:
msg = _("Something went wrong when deleting a volume "
"snapshot: rebasing a %(protocol)s network disk "
"using qemu-img has not been fully tested"
) % {'protocol': active_protocol}
LOG.error(msg)
raise exception.InternalError(msg)
self._rebase_with_qemu_img(active_disk_object.source_path,
rebase_base)
else:
# commit with blockCommit()
my_snap_base = None
my_snap_top = None
commit_disk = my_dev
if active_protocol is not None:
my_snap_base = _get_snap_dev(delete_info['merge_target_file'],
active_disk_object.backing_store)
my_snap_top = _get_snap_dev(delete_info['file_to_merge'],
active_disk_object.backing_store)
commit_base = my_snap_base or delete_info['merge_target_file']
commit_top = my_snap_top or delete_info['file_to_merge']
LOG.debug('will call blockCommit with commit_disk=%(commit_disk)s '
'commit_base=%(commit_base)s '
'commit_top=%(commit_top)s ',
{'commit_disk': commit_disk,
'commit_base': commit_base,
'commit_top': commit_top}, instance=instance)
dev = guest.get_block_device(commit_disk)
result = dev.commit(commit_base, commit_top, relative=True)
if result == 0:
LOG.debug('blockCommit started successfully',
instance=instance)
while not dev.is_job_complete():
LOG.debug('waiting for blockCommit job completion',
instance=instance)
time.sleep(0.5)
def volume_snapshot_delete(self, context, instance, volume_id, snapshot_id,
delete_info):
try:
self._volume_snapshot_delete(context, instance, volume_id,
snapshot_id, delete_info=delete_info)
except Exception:
with excutils.save_and_reraise_exception():
LOG.exception('Error occurred during volume_snapshot_delete, '
'sending error status to Cinder.',
instance=instance)
self._volume_snapshot_update_status(
context, snapshot_id, 'error_deleting')
self._volume_snapshot_update_status(context, snapshot_id, 'deleting')
self._volume_refresh_connection_info(context, instance, volume_id)
def reboot(self, context, instance, network_info, reboot_type,
block_device_info=None, bad_volumes_callback=None,
accel_info=None, share_info=None):
"""Reboot a virtual machine, given an instance reference."""
if reboot_type == 'SOFT':
# NOTE(vish): This will attempt to do a graceful shutdown/restart.
try:
soft_reboot_success = self._soft_reboot(instance)
except libvirt.libvirtError as e:
LOG.debug("Instance soft reboot failed: %s",
e,
instance=instance)
soft_reboot_success = False
if soft_reboot_success:
LOG.info("Instance soft rebooted successfully.",
instance=instance)
return
else:
LOG.warning("Failed to soft reboot instance. "
"Trying hard reboot.",
instance=instance)
return self._hard_reboot(context, instance, network_info,
share_info, block_device_info, accel_info
)
def _soft_reboot(self, instance):
"""Attempt to shutdown and restart the instance gracefully.
We use shutdown and create here so we can return if the guest
responded and actually rebooted. Note that this method only
succeeds if the guest responds to acpi. Therefore we return
success or failure so we can fall back to a hard reboot if
necessary.
:returns: True if the reboot succeeded
"""
guest = self._host.get_guest(instance)
state = guest.get_power_state(self._host)
old_domid = guest.id
# NOTE(vish): This check allows us to reboot an instance that
# is already shutdown.
if state == power_state.RUNNING:
guest.shutdown()
# NOTE(vish): This actually could take slightly longer than the
# FLAG defines depending on how long the get_info
# call takes to return.
for x in range(CONF.libvirt.wait_soft_reboot_seconds):
guest = self._host.get_guest(instance)
state = guest.get_power_state(self._host)
new_domid = guest.id
# NOTE(ivoks): By checking domain IDs, we make sure we are
# not recreating domain that's already running.
if old_domid != new_domid:
if state in (power_state.SHUTDOWN, power_state.CRASHED):
LOG.info("Instance shutdown successfully.",
instance=instance)
guest.launch()
timer = loopingcall.FixedIntervalLoopingCall(
self._wait_for_running, instance)
timer.start(interval=0.5).wait()
return True
else:
LOG.info("Instance may have been rebooted during soft "
"reboot, so return now.", instance=instance)
return True
greenthread.sleep(1)
return False
def _hard_reboot(self, context, instance, network_info, share_info,
block_device_info=None, accel_info=None):
"""Reboot a virtual machine, given an instance reference.
Performs a Libvirt reset (if supported) on the domain.
If Libvirt reset is unavailable this method actually destroys and
re-creates the domain to ensure the reboot happens, as the guest
OS cannot ignore this action.
"""
# NOTE(sbauza): Since we undefine the guest XML when destroying, we
# need to remember the existing mdevs for reusing them.
mdevs = self._get_all_assigned_mediated_devices(instance)
mdevs = list(mdevs.keys())
# NOTE(mdbooth): In addition to performing a hard reboot of the domain,
# the hard reboot operation is relied upon by operators to be an
# automated attempt to fix as many things as possible about a
# non-functioning instance before resorting to manual intervention.
# With this goal in mind, we tear down all the aspects of an instance
# we can here without losing data. This allows us to re-initialise from
# scratch, and hopefully fix, most aspects of a non-functioning guest.
self.destroy(context, instance, network_info, destroy_disks=False,
block_device_info=block_device_info,
destroy_secrets=False)
# Convert the system metadata to image metadata
# NOTE(mdbooth): This is a workaround for stateless Nova compute
# https://bugs.launchpad.net/nova/+bug/1349978
instance_dir = libvirt_utils.get_instance_path(instance)
fileutils.ensure_tree(instance_dir)
disk_info = blockinfo.get_disk_info(CONF.libvirt.virt_type,
instance,
instance.image_meta,
block_device_info)
# NOTE(melwitt): It's possible that we lost track of the allocated
# mdevs of an instance if, for example, a libvirt error was encountered
# after the domain XML was undefined in a previous hard reboot.
# Try to get existing mdevs that are created but not assigned so they
# will be added into the generated domain XML.
if instance.flavor.extra_specs.get('resources:VGPU') and not mdevs:
LOG.info(
'The instance flavor requests VGPU but no mdevs are assigned '
'to the instance. Attempting to re-assign mdevs.',
instance=instance)
allocs = self.virtapi.reportclient.get_allocations_for_consumer(
context, instance.uuid)
mdevs = self._allocate_mdevs(allocs)
# NOTE(vish): This could generate the wrong device_format if we are
# using the raw backend and the images don't exist yet.
# The create_images_and_backing below doesn't properly
# regenerate raw backend images, however, so when it
# does we need to (re)generate the xml after the images
# are in place.
xml = self._get_guest_xml(context, instance, network_info, disk_info,
instance.image_meta,
block_device_info=block_device_info,
mdevs=mdevs, accel_info=accel_info,
share_info=share_info)
# NOTE(mdbooth): context.auth_token will not be set when we call
# _hard_reboot from resume_state_on_host_boot()
if context.auth_token is not None:
# NOTE (rmk): Re-populate any missing backing files.
config = vconfig.LibvirtConfigGuest()
config.parse_str(xml)
backing_disk_info = self._get_instance_disk_info_from_config(
config, block_device_info)
self._create_images_and_backing(context, instance, instance_dir,
backing_disk_info)
# Initialize all the necessary networking, block devices and
# start the instance.
# NOTE(melwitt): Pass vifs_already_plugged=True here even though we've
# unplugged vifs earlier. The behavior of neutron plug events depends
# on which vif type we're using and we are working with a stale network
# info cache here, so won't rely on waiting for neutron plug events.
# vifs_already_plugged=True means "do not wait for neutron plug events"
external_events = []
vifs_already_plugged = True
event_expected_for_vnic_types = (
CONF.workarounds.wait_for_vif_plugged_event_during_hard_reboot)
if event_expected_for_vnic_types:
# NOTE(gibi): We unplugged every vif during destroy above and we
# will replug them with _create_guest_with_network. As the
# workaround config has some vnic_types configured we expect
# vif-plugged events for every vif with those vnic_types.
# TODO(gibi): only wait for events if we know that the networking
# backend sends plug time events. For that we need to finish
# https://bugs.launchpad.net/neutron/+bug/1821058 first in Neutron
# then create a driver -> plug-time event mapping in nova.
external_events = [
('network-vif-plugged', vif['id'])
for vif in network_info
if vif['vnic_type'] in event_expected_for_vnic_types
]
vifs_already_plugged = False
# NOTE(efried): The instance should already have a vtpm_secret_uuid
# registered if appropriate.
try:
self._create_guest_with_network(
context, xml, instance, network_info, block_device_info,
vifs_already_plugged=vifs_already_plugged,
external_events=external_events)
except libvirt.libvirtError as e:
errcode = e.get_error_code()
errmsg = e.get_error_message()
# NOTE(melwitt): If we are reassigning mdevs, we might hit the
# following error on the first attempt to create the guest:
# error getting device from group <group>: Input/output error
# Verify all devices in group <group> are bound to vfio-<bus> or
# pci-stub and not already in use
# Retry the guest creation once in this case as it usually succeeds
# on the second try.
if (mdevs and errcode == libvirt.VIR_ERR_INTERNAL_ERROR and
'error getting device from group' in errmsg):
LOG.info(
f'Encountered error {errmsg}, reattempting creation of '
'the guest.', instance=instance)
self._create_guest_with_network(
context, xml, instance, network_info, block_device_info,
vifs_already_plugged=vifs_already_plugged,
external_events=external_events)
else:
raise
def _wait_for_reboot():
"""Called at an interval until the VM is running again."""
state = self.get_info(instance).state
if state == power_state.RUNNING:
LOG.info("Instance rebooted successfully.",
instance=instance)
raise loopingcall.LoopingCallDone()
timer = loopingcall.FixedIntervalLoopingCall(_wait_for_reboot)
timer.start(interval=0.5).wait()
# Rebuild device_metadata to get shares
instance.device_metadata = self._build_device_metadata(
context, instance)
def pause(self, instance):
"""Pause VM instance."""
self._host.get_guest(instance).pause()
def unpause(self, instance):
"""Unpause paused VM instance."""
guest = self._host.get_guest(instance)
guest.resume()
guest.sync_guest_time()
def _clean_shutdown(self, instance, timeout, retry_interval):
"""Attempt to shutdown the instance gracefully.
:param instance: The instance to be shutdown
:param timeout: How long to wait in seconds for the instance to
shutdown
:param retry_interval: How often in seconds to signal the instance
to shutdown while waiting
:returns: True if the shutdown succeeded
"""
# List of states that represent a shutdown instance
SHUTDOWN_STATES = [power_state.SHUTDOWN,
power_state.CRASHED]
try:
guest = self._host.get_guest(instance)
except exception.InstanceNotFound:
# If the instance has gone then we don't need to
# wait for it to shutdown
return True
state = guest.get_power_state(self._host)
if state in SHUTDOWN_STATES:
LOG.info("Instance already shutdown.", instance=instance)
return True
LOG.debug("Shutting down instance from state %s", state,
instance=instance)
try:
guest.shutdown()
except libvirt.libvirtError as e:
LOG.debug("Ignoring libvirt exception from shutdown request: %s",
e,
instance=instance)
retry_countdown = retry_interval
for sec in range(timeout):
guest = self._host.get_guest(instance)
state = guest.get_power_state(self._host)
if state in SHUTDOWN_STATES:
LOG.info("Instance shutdown successfully after %d seconds.",
sec, instance=instance)
return True
# Note(PhilD): We can't assume that the Guest was able to process
# any previous shutdown signal (for example it may
# have still been startingup, so within the overall
# timeout we re-trigger the shutdown every
# retry_interval
if retry_countdown == 0:
retry_countdown = retry_interval
# Instance could shutdown at any time, in which case we
# will get an exception when we call shutdown
try:
LOG.debug("Instance in state %s after %d seconds - "
"resending shutdown", state, sec,
instance=instance)
guest.shutdown()
except libvirt.libvirtError:
# Assume this is because its now shutdown, so loop
# one more time to clean up.
LOG.debug("Ignoring libvirt exception from shutdown "
"request.", instance=instance)
continue
else:
retry_countdown -= 1
time.sleep(1)
LOG.info("Instance failed to shutdown in %d seconds.",
timeout, instance=instance)
return False
def power_off(self, instance, timeout=0, retry_interval=0):
"""Power off the specified instance."""
if timeout:
self._clean_shutdown(instance, timeout, retry_interval)
self._destroy(instance)
def power_on(self, context, instance, network_info,
block_device_info=None, accel_info=None, share_info=None):
"""Power on the specified instance."""
# We use _hard_reboot here to ensure that all backing files,
# network, and block device connections, etc. are established
# and available before we attempt to start the instance.
self._hard_reboot(context, instance, network_info, share_info,
block_device_info, accel_info)
def _get_share_driver_manager(self, host, protocol):
if protocol == fields.ShareMappingProto.NFS:
return nfs.LibvirtNFSVolumeDriver(host)
elif protocol == fields.ShareMappingProto.CEPHFS:
return cephfs.LibvirtCEPHFSVolumeDriver(host)
else:
raise exception.ShareProtocolNotSupported(share_proto=protocol)
def _get_share_connection_info(self, share_mapping):
connection_info = {
"data": {
"export": share_mapping.export_location,
"name": share_mapping.share_id,
}
}
if share_mapping.share_proto == fields.ShareMappingProto.CEPHFS:
if (
"access_to" in share_mapping and
share_mapping.access_to is not None
):
name_opt = "name=" + share_mapping.access_to
secret_opt = "secret=" + share_mapping.access_key
connection_info["data"]["options"] = [name_opt, secret_opt]
return connection_info
def _get_share_mount_path(self, instance, share_mapping):
drv = self._get_share_driver_manager(
instance.host, share_mapping.share_proto)
mount_path = drv._get_mount_path(
self._get_share_connection_info(share_mapping))
return mount_path
def mount_share(self, context, instance, share_mapping):
drv = self._get_share_driver_manager(
instance.host, share_mapping.share_proto)
try:
drv.connect_volume(
self._get_share_connection_info(share_mapping),
instance
)
except processutils.ProcessExecutionError as exc:
raise exception.ShareMountError(
share_id=share_mapping.share_id,
server_id=share_mapping.instance_uuid,
reason=exc
)
def umount_share(self, context, instance, share_mapping):
drv = self._get_share_driver_manager(
instance.host, share_mapping.share_proto)
try:
return drv.disconnect_volume(
self._get_share_connection_info(share_mapping),
instance
)
except processutils.ProcessExecutionError as exc:
raise exception.ShareUmountError(
share_id=share_mapping.share_id,
server_id=share_mapping.instance_uuid,
reason=exc
)
def trigger_crash_dump(self, instance):
"""Trigger crash dump by injecting an NMI to the specified instance."""
try:
self._host.get_guest(instance).inject_nmi()
except libvirt.libvirtError as ex:
error_code = ex.get_error_code()
if error_code == libvirt.VIR_ERR_NO_SUPPORT:
raise exception.TriggerCrashDumpNotSupported()
elif error_code == libvirt.VIR_ERR_OPERATION_INVALID:
raise exception.InstanceNotRunning(instance_id=instance.uuid)
LOG.exception(
'Error from libvirt while injecting an NMI to '
'%(instance_uuid)s: [Error Code %(error_code)s] %(ex)s',
{'instance_uuid': instance.uuid,
'error_code': error_code, 'ex': ex})
raise
def suspend(self, context, instance):
"""Suspend the specified instance."""
guest = self._host.get_guest(instance)
self._detach_pci_devices(
guest,
instance.get_pci_devices(
source=objects.InstancePCIRequest.FLAVOR_ALIAS
),
)
self._detach_direct_passthrough_ports(context, instance, guest)
self._detach_mediated_devices(guest)
guest.save_memory_state()
def resume(
self,
context,
instance,
network_info,
block_device_info=None,
share_info=None
):
"""resume the specified instance."""
if share_info is None:
share_info = objects.ShareMappingList()
xml = self._get_existing_domain_xml(instance, network_info,
block_device_info, share_info)
# NOTE(gsantos): The mediated devices that were removed on suspension
# are still present in the xml. Let's take their references from it
# and re-attach them.
mdevs = self._get_mdevs_from_guest_config(xml)
# NOTE(efried): The instance should already have a vtpm_secret_uuid
# registered if appropriate.
guest = self._create_guest_with_network(
context, xml, instance, network_info, block_device_info,
vifs_already_plugged=True)
self._attach_pci_devices(
guest,
instance.get_pci_devices(
source=objects.InstancePCIRequest.FLAVOR_ALIAS
),
)
self._attach_direct_passthrough_ports(
context, instance, guest, network_info)
self._attach_mediated_devices(guest, mdevs)
timer = loopingcall.FixedIntervalLoopingCall(self._wait_for_running,
instance)
timer.start(interval=0.5).wait()
guest.sync_guest_time()
def resume_state_on_host_boot(self, context, instance, network_info,
share_info, block_device_info=None):
"""resume guest state when a host is booted."""
# Check if the instance is running already and avoid doing
# anything if it is.
try:
guest = self._host.get_guest(instance)
state = guest.get_power_state(self._host)
ignored_states = (power_state.RUNNING,
power_state.SUSPENDED,
power_state.NOSTATE,
power_state.PAUSED)
if state in ignored_states:
return
except (exception.InternalError, exception.InstanceNotFound):
pass
# Instance is not up and could be in an unknown state.
# Be as absolute as possible about getting it back into
# a known and running state.
self._hard_reboot(context, instance, network_info,
share_info, block_device_info
)
def rescue(self, context, instance, network_info, image_meta,
rescue_password, block_device_info, share_info):
"""Loads a VM using rescue images.
A rescue is normally performed when something goes wrong with the
primary images and data needs to be corrected/recovered. Rescuing
should not edit or over-ride the original image, only allow for
data recovery.
Two modes are provided when rescuing an instance with this driver.
The original and default rescue mode, where the rescue boot disk,
original root disk and optional regenerated config drive are attached
to the instance.
A second stable device rescue mode is also provided where all of the
original devices are attached to the instance during the rescue attempt
with the addition of the rescue boot disk. This second mode is
controlled by the hw_rescue_device and hw_rescue_bus image properties
on the rescue image provided to this method via image_meta.
:param nova.context.RequestContext context:
The context for the rescue.
:param nova.objects.instance.Instance instance:
The instance being rescued.
:param nova.network.model.NetworkInfo network_info:
Necessary network information for the resume.
:param nova.objects.ImageMeta image_meta:
The metadata of the image of the instance.
:param rescue_password: new root password to set for rescue.
:param dict block_device_info:
The block device mapping of the instance.
:param nova.objects.ShareMappingList share_info:
list of share_mapping
"""
instance_dir = libvirt_utils.get_instance_path(instance)
unrescue_xml = self._get_existing_domain_xml(
instance, network_info, share_info=share_info)
unrescue_xml_path = os.path.join(instance_dir, 'unrescue.xml')
with open(unrescue_xml_path, 'w') as f:
f.write(unrescue_xml)
rescue_image_id = None
rescue_image_meta = None
if image_meta.obj_attr_is_set("id"):
rescue_image_id = image_meta.id
rescue_images = {
'image_id': (rescue_image_id or
CONF.libvirt.rescue_image_id or instance.image_ref),
'kernel_id': (CONF.libvirt.rescue_kernel_id or
instance.kernel_id),
'ramdisk_id': (CONF.libvirt.rescue_ramdisk_id or
instance.ramdisk_id),
}
virt_type = CONF.libvirt.virt_type
if hardware.check_hw_rescue_props(image_meta):
LOG.info("Attempting a stable device rescue", instance=instance)
# NOTE(lyarwood): Stable device rescue is not supported when using
# the LXC virt_type as it does not support the required
# <boot order=''> definitions allowing an instance to boot from the
# rescue device added as a final device to the domain.
if virt_type == 'lxc':
reason = _(
"Stable device rescue is not supported by virt_type '%s'"
)
raise exception.InstanceNotRescuable(
instance_id=instance.uuid, reason=reason % virt_type)
# NOTE(lyarwood): Stable device rescue provides the original disk
# mapping of the instance with the rescue device appended to the
# end. As a result we need to provide the original image_meta, the
# new rescue_image_meta and block_device_info when calling
# get_disk_info.
rescue_image_meta = image_meta
try:
if instance.image_ref:
image_meta = objects.ImageMeta.from_image_ref(
context, self._image_api, instance.image_ref)
else:
# NOTE(lyarwood): If instance.image_ref isn't set attempt
# to lookup the original image_meta from the bdms. This
# will return an empty dict if no valid image_meta is
# found.
image_meta_dict = block_device.get_bdm_image_metadata(
context, self._image_api, self._volume_api,
block_device_info['block_device_mapping'],
legacy_bdm=False)
image_meta = objects.ImageMeta.from_dict(image_meta_dict)
except exception.ImageNotFound:
image_meta = instance.image_meta
else:
LOG.info("Attempting rescue", instance=instance)
# NOTE(lyarwood): A legacy rescue only provides the rescue device
# and the original root device so we don't need to provide
# block_device_info to the get_disk_info call.
block_device_info = None
disk_info = blockinfo.get_disk_info(virt_type, instance, image_meta,
rescue=True, block_device_info=block_device_info,
rescue_image_meta=rescue_image_meta)
LOG.debug("rescue generated disk_info: %s", disk_info)
injection_info = InjectionInfo(network_info=network_info,
admin_pass=rescue_password,
files=None)
gen_confdrive = functools.partial(self._create_configdrive,
context, instance, injection_info,
rescue=True)
# NOTE(sbauza): Since rescue recreates the guest XML, we need to
# remember the existing mdevs for reusing them.
mdevs = self._get_all_assigned_mediated_devices(instance)
mdevs = list(mdevs.keys())
self._create_image(context, instance, disk_info['mapping'],
injection_info=injection_info, suffix='.rescue',
disk_images=rescue_images)
# NOTE(efried): The instance should already have a vtpm_secret_uuid
# registered if appropriate.
xml = self._get_guest_xml(context, instance, network_info, disk_info,
image_meta, rescue=rescue_images,
mdevs=mdevs,
block_device_info=block_device_info,
share_info=share_info)
self._destroy(instance)
self._create_guest(
context, xml, instance, post_xml_callback=gen_confdrive,
)
def unrescue(
self,
context: nova_context.RequestContext,
instance: 'objects.Instance',
):
"""Reboot the VM which is being rescued back into primary images."""
instance_dir = libvirt_utils.get_instance_path(instance)
unrescue_xml_path = os.path.join(instance_dir, 'unrescue.xml')
# The xml should already contain the secret_uuid if relevant.
xml = libvirt_utils.load_file(unrescue_xml_path)
self._destroy(instance)
self._create_guest(context, xml, instance)
os.unlink(unrescue_xml_path)
rescue_files = os.path.join(instance_dir, "*.rescue")
for rescue_file in glob.iglob(rescue_files):
if os.path.isdir(rescue_file):
shutil.rmtree(rescue_file)
else:
os.unlink(rescue_file)
# cleanup rescue volume
lvm.remove_volumes([lvmdisk for lvmdisk in self._lvm_disks(instance)
if lvmdisk.endswith('.rescue')])
if CONF.libvirt.images_type == 'rbd':
filter_fn = lambda disk: (disk.startswith(instance.uuid) and
disk.endswith('.rescue'))
rbd_utils.RBDDriver().cleanup_volumes(filter_fn)
def poll_rebooting_instances(self, timeout, instances):
pass
@staticmethod
def _get_or_create_encryption_secret(context, instance, driver_bdm):
created = False
secret_uuid = driver_bdm.get('encryption_secret_uuid')
if secret_uuid is None:
# Create a passphrase and stash it in the key manager
secret_uuid, secret = crypto.create_encryption_secret(
context, instance, driver_bdm)
# Stash the UUID of said secret in our driver BDM
driver_bdm['encryption_secret_uuid'] = secret_uuid
created = True
else:
# NOTE(melwitt): In general, we avoid reusing secrets but
# we need to reuse them in the case of shelve/unshelve and
# rebuild. The use case is if an admin user
# shelves/unshelves or rebuilds an instance owned by a
# non-admin user. If we don't reuse the non-admin user's
# secret and instead create a new secret, the new secret
# will be owned by the admin user and will prevent the
# non-admin user from accessing the new secret for their
# instance. There is no way in the barbican API to create a
# secret with a different user/project than the caller, so
# we have to just reuse the secret.
secret = crypto.get_encryption_secret(context, secret_uuid)
if secret is None:
# If we get here, because we know this BDM is supposed
# to have an existing secret, we also know all of the
# other BDMs have existing secrets too. Because we
# didn't create any secrets, we don't need to clean up
# any secrets.
msg = (
f'Failed to find encryption secret {secret_uuid} '
f'in the key manager for driver BDM '
f"{driver_bdm['uuid']}")
raise exception.EphemeralEncryptionSecretNotFound(msg)
return secret_uuid, secret, created
def _add_ephemeral_encryption_driver_bdm_attrs(
self,
context: nova_context.RequestContext,
instance: 'objects.Instance',
block_device_info: ty.Dict[str, ty.Any],
) -> ty.Optional[ty.Dict[str, ty.Any]]:
"""Add ephemeral encryption attributes to driver BDMs before use."""
encrypted_bdms = driver.block_device_info_get_encrypted_disks(
block_device_info)
# Either all of the driver_bdm's should have existing encryption
# secrets (unshelve, rebuild) or none of them should. There should
# never be a mix of both. If there is, something is wrong.
if encrypted_bdms:
bdms_without_secrets = [
driver_bdm for driver_bdm in encrypted_bdms
if not driver_bdm.get('encryption_secret_uuid')]
bdms_with_secrets = [
driver_bdm for driver_bdm in encrypted_bdms
if driver_bdm.get('encryption_secret_uuid')]
if bdms_without_secrets and bdms_with_secrets:
msg = (
f'Found a mix of encrypted BDMs with and without existing '
f'encryption secrets: {encrypted_bdms}')
raise exception.InvalidBDM(msg)
try:
orig_encrypted_bdms = []
created_keymgr_secrets = []
created_libvirt_secrets = []
for driver_bdm in encrypted_bdms:
orig_encrypted_bdms.append(deepcopy(driver_bdm))
# NOTE(lyarwood): Users can request that their ephemeral
# storage be encrypted without providing an encryption format
# to use. If one isn't provided use the host default here and
# record it in the driver BDM.
if driver_bdm.get('encryption_format') is None:
driver_bdm['encryption_format'] = (
CONF.ephemeral_storage_encryption.default_format)
secret_uuid, secret, created = (
self._get_or_create_encryption_secret(
context, instance, driver_bdm))
if created:
created_keymgr_secrets.append(secret_uuid)
# Ensure this is all saved back down in the database via the
# o.vo BlockDeviceMapping object
driver_bdm.save()
# Stash the passphrase itself in a libvirt secret using the
# same UUID as the key manager secret for easy retrieval later
secret_usage = f"{instance.uuid}_{driver_bdm['uuid']}"
# Be extra defensive here and delete any existing libvirt
# secret to ensure we are creating the secret we retrieved or
# created in the key manager just now.
if self._host.find_secret('volume', secret_usage):
self._host.delete_secret('volume', secret_usage)
self._host.create_secret(
'volume', secret_usage, password=secret, uuid=secret_uuid)
created_libvirt_secrets.append(secret_usage)
except Exception:
for secret_uuid in created_keymgr_secrets:
try:
crypto.delete_encryption_secret(
context, instance.uuid, secret_uuid)
except Exception:
LOG.exception(
f'Failed to delete encryption secret '
f'{secret_uuid} in the key manager', instance=instance)
for i, orig_driver_bdm in enumerate(orig_encrypted_bdms):
driver_bdm = encrypted_bdms[i]
for key in ('encryption_format', 'encryption_secret_uuid'):
driver_bdm[key] = orig_driver_bdm[key]
driver_bdm.save()
for secret_usage in created_libvirt_secrets:
try:
if self._host.find_secret('volume', secret_usage):
self._host.delete_secret('volume', secret_usage)
except Exception:
LOG.exception(
f'Failed to delete libvirt secret {secret_usage}',
instance=instance)
raise
return block_device_info
def spawn(self, context, instance, image_meta, injected_files,
admin_password, allocations, network_info=None,
block_device_info=None, power_on=True, accel_info=None):
# NOTE(lyarwood): Before we generate disk_info we need to ensure the
# driver_bdms are populated with any missing encryption attributes such
# as the format to use, associated options and encryption secret uuid.
# This avoids having to pass block_device_info and the driver bdms down
# into the imagebackend later when creating or building the config for
# the disks.
block_device_info = self._add_ephemeral_encryption_driver_bdm_attrs(
context, instance, block_device_info)
disk_info = blockinfo.get_disk_info(CONF.libvirt.virt_type,
instance,
image_meta,
block_device_info)
injection_info = InjectionInfo(network_info=network_info,
files=injected_files,
admin_pass=admin_password)
gen_confdrive = functools.partial(self._create_configdrive,
context, instance,
injection_info)
created_instance_dir, created_disks = self._create_image(
context, instance, disk_info['mapping'],
injection_info=injection_info,
block_device_info=block_device_info)
# Required by Quobyte CI
self._ensure_console_log_for_instance(instance)
# Does the guest need to be assigned some vGPU mediated devices ?
mdevs = self._allocate_mdevs(allocations)
# If the guest needs a vTPM, _get_guest_xml needs its secret to exist
# and its uuid to be registered in the instance prior to _get_guest_xml
if CONF.libvirt.swtpm_enabled and hardware.get_vtpm_constraint(
instance.flavor, image_meta
):
if not instance.system_metadata.get('vtpm_secret_uuid'):
# Create the secret via the key manager service so that we have
# it to hand when generating the XML. This is slightly wasteful
# as we'll perform a redundant key manager API call later when
# we create the domain but the alternative is an ugly mess
crypto.ensure_vtpm_secret(context, instance)
xml = self._get_guest_xml(context, instance, network_info,
disk_info, image_meta,
block_device_info=block_device_info,
mdevs=mdevs, accel_info=accel_info)
self._create_guest_with_network(
context, xml, instance, network_info, block_device_info,
post_xml_callback=gen_confdrive,
power_on=power_on,
cleanup_instance_dir=created_instance_dir,
cleanup_instance_disks=created_disks)
LOG.debug("Guest created on hypervisor", instance=instance)
def _wait_for_boot():
"""Called at an interval until the VM is running."""
state = self.get_info(instance).state
if state == power_state.RUNNING:
LOG.info("Instance spawned successfully.", instance=instance)
raise loopingcall.LoopingCallDone()
if power_on:
timer = loopingcall.FixedIntervalLoopingCall(_wait_for_boot)
timer.start(interval=0.5).wait()
else:
LOG.info("Instance spawned successfully.", instance=instance)
# Finally register defaults for any undefined image properties so that
# future changes by QEMU, libvirt or within this driver don't change
# the ABI of the instance.
self._register_undefined_instance_details(context, instance)
def _get_console_output_file(self, instance, console_log):
bytes_to_read = MAX_CONSOLE_BYTES
log_data = b"" # The last N read bytes
i = 0 # in case there is a log rotation (like "virtlogd")
path = console_log
while bytes_to_read > 0 and os.path.exists(path):
read_log_data, remaining = nova.privsep.path.last_bytes(
path, bytes_to_read)
# We need the log file content in chronological order,
# that's why we *prepend* the log data.
log_data = read_log_data + log_data
# Prep to read the next file in the chain
bytes_to_read -= len(read_log_data)
path = console_log + "." + str(i)
i += 1
if remaining > 0:
LOG.info('Truncated console log returned, '
'%d bytes ignored', remaining, instance=instance)
return log_data
def get_console_output(self, context, instance):
guest = self._host.get_guest(instance)
xml = guest.get_xml_desc()
tree = etree.fromstring(xml)
# check for different types of consoles
path_sources = [
('file', "./devices/console[@type='file']/source[@path]", 'path'),
('tcp', "./devices/console[@type='tcp']/log[@file]", 'file'),
('pty', "./devices/console[@type='pty']/source[@path]", 'path')]
console_type = ""
console_path = ""
for c_type, epath, attrib in path_sources:
node = tree.find(epath)
if (node is not None) and node.get(attrib):
console_type = c_type
console_path = node.get(attrib)
break
# instance has no console at all
if not console_path:
raise exception.ConsoleNotAvailable()
# instance has a console, but file doesn't exist (yet?)
if not os.path.exists(console_path):
LOG.info('console logfile for instance does not exist',
instance=instance)
return ""
# pty consoles need special handling
if console_type == 'pty':
console_log = self._get_console_log_path(instance)
data = nova.privsep.libvirt.readpty(console_path)
# NOTE(markus_z): The virt_types kvm and qemu are the only ones
# which create a dedicated file device for the console logging.
# Other virt_types like lxc and parallels depend on the flush of
# that PTY device into the "console.log" file to ensure that a
# series of "get_console_output" calls return the complete content
# even after rebooting a guest.
nova.privsep.path.writefile(console_log, 'a+', data)
# set console path to logfile, not to pty device
console_path = console_log
# return logfile content
return self._get_console_output_file(instance, console_path)
def get_host_ip_addr(self):
# NOTE(gibi): We should rename this as we might return a hostname
# instead of an IP address. But this is a virt driver interface
# method, so it probably does not worth the hashle. Only the
# resource_tracker use this today outside the virt driver to set up
# the Migration object.
addr = CONF.libvirt.migration_inbound_addr
if "%s" in addr:
addr = addr % self._host.get_hostname()
return addr
def get_vnc_console(self, context, instance):
def get_vnc_port_for_instance(instance_name):
guest = self._host.get_guest(instance)
xml = guest.get_xml_desc()
xml_dom = etree.fromstring(xml)
graphic = xml_dom.find("./devices/graphics[@type='vnc']")
if graphic is not None:
return graphic.get('port')
# NOTE(rmk): We had VNC consoles enabled but the instance in
# question is not actually listening for connections.
raise exception.ConsoleTypeUnavailable(console_type='vnc')
port = get_vnc_port_for_instance(instance.name)
host = CONF.vnc.server_proxyclient_address
return ctype.ConsoleVNC(host=host, port=port)
def get_spice_console(self, context, instance):
def get_spice_ports_for_instance(instance_name):
guest = self._host.get_guest(instance)
xml = guest.get_xml_desc()
xml_dom = etree.fromstring(xml)
graphic = xml_dom.find("./devices/graphics[@type='spice']")
if graphic is not None:
return (graphic.get('port'), graphic.get('tlsPort'))
# NOTE(rmk): We had Spice consoles enabled but the instance in
# question is not actually listening for connections.
raise exception.ConsoleTypeUnavailable(console_type='spice')
ports = get_spice_ports_for_instance(instance.name)
host = CONF.spice.server_proxyclient_address
return ctype.ConsoleSpice(host=host, port=ports[0], tlsPort=ports[1])
def get_serial_console(self, context, instance):
guest = self._host.get_guest(instance)
for hostname, port in self._get_serial_ports_from_guest(
guest, mode='bind'):
return ctype.ConsoleSerial(host=hostname, port=port)
raise exception.ConsoleTypeUnavailable(console_type='serial')
@staticmethod
def _create_ephemeral(target, ephemeral_size,
fs_label, os_type, is_block_dev=False,
context=None, specified_fs=None,
vm_mode=None):
if not is_block_dev:
if (CONF.libvirt.virt_type == "parallels" and
vm_mode == fields.VMMode.EXE):
libvirt_utils.create_ploop_image('expanded', target,
'%dG' % ephemeral_size,
specified_fs)
return
libvirt_utils.create_image(
target, 'raw', f'{ephemeral_size}G', safe=True)
# Run as root only for block devices.
disk_api.mkfs(os_type, fs_label, target, run_as_root=is_block_dev,
specified_fs=specified_fs)
@staticmethod
def _create_swap(target, swap_mb, context=None):
"""Create a swap file of specified size."""
libvirt_utils.create_image(target, 'raw', f'{swap_mb}M')
nova.privsep.fs.unprivileged_mkfs('swap', target)
@staticmethod
def _get_console_log_path(instance):
return os.path.join(libvirt_utils.get_instance_path(instance),
'console.log')
def _ensure_console_log_for_instance(self, instance):
# NOTE(mdbooth): Although libvirt will create this file for us
# automatically when it starts, it will initially create it with
# root ownership and then chown it depending on the configuration of
# the domain it is launching. Quobyte CI explicitly disables the
# chown by setting dynamic_ownership=0 in libvirt's config.
# Consequently when the domain starts it is unable to write to its
# console.log. See bug https://bugs.launchpad.net/nova/+bug/1597644
#
# To work around this, we create the file manually before starting
# the domain so it has the same ownership as Nova. This works
# for Quobyte CI because it is also configured to run qemu as the same
# user as the Nova service. Installations which don't set
# dynamic_ownership=0 are not affected because libvirt will always
# correctly configure permissions regardless of initial ownership.
#
# Setting dynamic_ownership=0 is dubious and potentially broken in
# more ways than console.log (see comment #22 on the above bug), so
# Future Maintainer who finds this code problematic should check to see
# if we still support it.
console_file = self._get_console_log_path(instance)
LOG.debug('Ensure instance console log exists: %s', console_file,
instance=instance)
try:
libvirt_utils.file_open(console_file, 'a').close()
# NOTE(sfinucan): We can safely ignore permission issues here and
# assume that it is libvirt that has taken ownership of this file.
except IOError as ex:
if ex.errno != errno.EACCES:
raise
LOG.debug('Console file already exists: %s.', console_file)
@staticmethod
def _get_disk_config_image_type():
# TODO(mikal): there is a bug here if images_type has
# changed since creation of the instance, but I am pretty
# sure that this bug already exists.
return 'rbd' if CONF.libvirt.images_type == 'rbd' else 'raw'
@staticmethod
def _is_booted_from_volume(block_device_info):
"""Determines whether the VM is booting from volume
Determines whether the block device info indicates that the VM
is booting from a volume.
"""
block_device_mapping = driver.block_device_info_get_mapping(
block_device_info)
return bool(block_device.get_root_bdm(block_device_mapping))
def _inject_data(self, disk, instance, injection_info):
"""Injects data in a disk image
Helper used for injecting data in a disk image file system.
:param disk: The disk we're injecting into (an Image object)
:param instance: The instance we're injecting into
:param injection_info: Injection info
"""
# Handles the partition need to be used.
LOG.debug('Checking root disk injection %s',
str(injection_info), instance=instance)
target_partition = None
if not instance.kernel_id:
target_partition = CONF.libvirt.inject_partition
if target_partition == 0:
target_partition = None
if CONF.libvirt.virt_type == 'lxc':
target_partition = None
# Handles the key injection.
key = None
if CONF.libvirt.inject_key and instance.get('key_data'):
key = str(instance.key_data)
# Handles the admin password injection.
admin_pass = None
if CONF.libvirt.inject_password:
admin_pass = injection_info.admin_pass
# Handles the network injection.
net = netutils.get_injected_network_template(
injection_info.network_info,
libvirt_virt_type=CONF.libvirt.virt_type)
# Handles the metadata injection
metadata = instance.get('metadata')
if any((key, net, metadata, admin_pass, injection_info.files)):
LOG.debug('Injecting %s', str(injection_info),
instance=instance)
img_id = instance.image_ref
try:
disk_api.inject_data(disk.get_model(self._conn),
key, net, metadata, admin_pass,
injection_info.files,
partition=target_partition,
mandatory=('files',))
except Exception as e:
with excutils.save_and_reraise_exception():
LOG.error('Error injecting data into image '
'%(img_id)s (%(e)s)',
{'img_id': img_id, 'e': e},
instance=instance)
# NOTE(sileht): many callers of this method assume that this
# method doesn't fail if an image already exists but instead
# think that it will be reused (ie: (live)-migration/resize)
def _create_image(self, context, instance,
disk_mapping, injection_info=None, suffix='',
disk_images=None, block_device_info=None,
fallback_from_host=None,
ignore_bdi_for_swap=False):
booted_from_volume = self._is_booted_from_volume(block_device_info)
def image(
fname, image_type=CONF.libvirt.images_type, disk_info_mapping=None
):
return self.image_backend.by_name(
instance, fname + suffix, image_type,
disk_info_mapping=disk_info_mapping)
def raw(fname, disk_info_mapping=None):
return image(
fname, image_type='raw', disk_info_mapping=disk_info_mapping)
created_instance_dir = True
# ensure directories exist and are writable
instance_dir = libvirt_utils.get_instance_path(instance)
if os.path.exists(instance_dir):
LOG.debug("Instance directory exists: not creating",
instance=instance)
created_instance_dir = False
else:
LOG.debug("Creating instance directory", instance=instance)
fileutils.ensure_tree(libvirt_utils.get_instance_path(instance))
LOG.info('Creating image(s)', instance=instance)
flavor = instance.get_flavor()
swap_mb = 0
if 'disk.swap' in disk_mapping:
if ignore_bdi_for_swap:
# This is a workaround to support legacy swap resizing,
# which does not touch swap size specified in bdm,
# but works with flavor specified size only.
# In this case we follow the legacy logic and ignore block
# device info completely.
# NOTE(ft): This workaround must be removed when a correct
# implementation of resize operation changing sizes in bdms is
# developed. Also at that stage we probably may get rid of
# the direct usage of flavor swap size here,
# leaving the work with bdm only.
swap_mb = flavor['swap']
else:
disk_info_mapping = disk_mapping['disk.swap']
disk_device = disk_info_mapping['dev']
swap = driver.block_device_info_get_swap(block_device_info)
if driver.swap_is_usable(swap):
swap_mb = swap['swap_size']
elif (
flavor['swap'] > 0 and
not block_device.volume_in_mapping(
disk_device, block_device_info,
)
):
swap_mb = flavor['swap']
if swap_mb > 0:
if (CONF.libvirt.virt_type == "parallels" and
instance.vm_mode == fields.VMMode.EXE):
msg = _("Swap disk is not supported "
"for Virtuozzo container")
raise exception.Invalid(msg)
if not disk_images:
disk_images = {'image_id': instance.image_ref,
'kernel_id': instance.kernel_id,
'ramdisk_id': instance.ramdisk_id}
# NOTE(mdbooth): kernel and ramdisk, if they are defined, are hardcoded
# to use raw, which means they will always be cleaned up with the
# instance directory. We must not consider them for created_disks,
# which may not be using the instance directory.
if disk_images['kernel_id']:
fname = imagecache.get_cache_fname(disk_images['kernel_id'])
raw('kernel').cache(fetch_func=libvirt_utils.fetch_raw_image,
context=context,
filename=fname,
image_id=disk_images['kernel_id'])
if disk_images['ramdisk_id']:
fname = imagecache.get_cache_fname(disk_images['ramdisk_id'])
raw('ramdisk').cache(fetch_func=libvirt_utils.fetch_raw_image,
context=context,
filename=fname,
image_id=disk_images['ramdisk_id'])
created_disks = self._create_and_inject_local_root(
context, instance, disk_mapping, booted_from_volume, suffix,
disk_images, injection_info, fallback_from_host)
# Lookup the filesystem type if required
os_type_with_default = nova.privsep.fs.get_fs_type_for_os_type(
instance.os_type)
# Generate a file extension based on the file system
# type and the mkfs commands configured if any
file_extension = nova.privsep.fs.get_file_extension_for_os_type(
os_type_with_default, CONF.default_ephemeral_format)
vm_mode = fields.VMMode.get_from_instance(instance)
ephemeral_gb = instance.flavor.ephemeral_gb
if 'disk.local' in disk_mapping:
disk_info_mapping = disk_mapping['disk.local']
disk_image = image(
'disk.local', disk_info_mapping=disk_info_mapping)
# Short circuit the exists() tests if we already created a disk
created_disks = created_disks or not disk_image.exists()
fn = functools.partial(self._create_ephemeral,
fs_label='ephemeral0',
os_type=instance.os_type,
is_block_dev=disk_image.is_block_dev,
vm_mode=vm_mode)
fname = "ephemeral_%s_%s" % (ephemeral_gb, file_extension)
size = ephemeral_gb * units.Gi
disk_image.cache(
fetch_func=fn, context=context, filename=fname, size=size,
ephemeral_size=ephemeral_gb, safe=True)
for idx, eph in enumerate(driver.block_device_info_get_ephemerals(
block_device_info)):
disk_name = blockinfo.get_eph_disk(idx)
disk_info_mapping = disk_mapping[disk_name]
disk_image = image(disk_name, disk_info_mapping=disk_info_mapping)
# Short circuit the exists() tests if we already created a disk
created_disks = created_disks or not disk_image.exists()
specified_fs = eph.get('guest_format')
if specified_fs and not self.is_supported_fs_format(specified_fs):
msg = _("%s format is not supported") % specified_fs
raise exception.InvalidBDMFormat(details=msg)
fn = functools.partial(self._create_ephemeral,
fs_label='ephemeral%d' % idx,
os_type=instance.os_type,
is_block_dev=disk_image.is_block_dev,
vm_mode=vm_mode)
size = eph['size'] * units.Gi
fname = "ephemeral_%s_%s" % (eph['size'], file_extension)
disk_image.cache(
fetch_func=fn, context=context, filename=fname, size=size,
ephemeral_size=eph['size'], specified_fs=specified_fs,
safe=True)
if swap_mb > 0:
size = swap_mb * units.Mi
disk_info_mapping = disk_mapping['disk.swap']
swap = image('disk.swap', disk_info_mapping=disk_info_mapping)
# Short circuit the exists() tests if we already created a disk
created_disks = created_disks or not swap.exists()
swap.cache(
fetch_func=self._create_swap, context=context,
filename="swap_%s" % swap_mb, size=size, swap_mb=swap_mb,
safe=True)
if created_disks:
LOG.debug('Created local disks', instance=instance)
else:
LOG.debug('Did not create local disks', instance=instance)
return (created_instance_dir, created_disks)
def _create_and_inject_local_root(self, context, instance, disk_mapping,
booted_from_volume, suffix, disk_images,
injection_info, fallback_from_host):
created_disks = False
# File injection only if needed
need_inject = (not configdrive.required_by(instance) and
injection_info is not None and
CONF.libvirt.inject_partition != -2)
if not booted_from_volume:
root_fname = imagecache.get_cache_fname(disk_images['image_id'])
size = instance.flavor.root_gb * units.Gi
if size == 0 or suffix == '.rescue':
size = None
disk_name = 'disk' + suffix
disk_info_mapping = disk_mapping[disk_name]
backend = self.image_backend.by_name(
instance, disk_name, disk_info_mapping=disk_info_mapping)
created_disks = not backend.exists()
if instance.task_state == task_states.RESIZE_FINISH:
backend.create_snap(libvirt_utils.RESIZE_SNAPSHOT_NAME)
if backend.SUPPORTS_CLONE:
def clone_fallback_to_fetch(
context, target, image_id, trusted_certs=None,
):
refuse_fetch = (
CONF.libvirt.images_type == 'rbd' and
CONF.workarounds.never_download_image_if_on_rbd)
try:
backend.clone(context, disk_images['image_id'])
except exception.ImageUnacceptable:
if refuse_fetch:
# Re-raise the exception from the failed
# ceph clone. The compute manager expects
# ImageUnacceptable as a possible result
# of spawn(), from which this is called.
with excutils.save_and_reraise_exception():
LOG.warning(
'Image %s is not on my ceph and '
'[workarounds]/'
'never_download_image_if_on_rbd=True;'
' refusing to fetch and upload.',
disk_images['image_id'])
libvirt_utils.fetch_image(
context, target, image_id, trusted_certs,
)
fetch_func = clone_fallback_to_fetch
else:
fetch_func = libvirt_utils.fetch_image
self._try_fetch_image_cache(backend, fetch_func, context,
root_fname, disk_images['image_id'],
instance, size, fallback_from_host)
# During unshelve or cross cell resize on Qcow2 backend, we spawn()
# using a snapshot image. Extra work is needed in order to rebase
# disk image to its original image_ref. Disk backing file will
# then represent back image_ref instead of snapshot image.
self._rebase_original_qcow2_image(context, instance, backend)
if need_inject:
self._inject_data(backend, instance, injection_info)
elif need_inject:
LOG.warning('File injection into a boot from volume '
'instance is not supported', instance=instance)
return created_disks
def _needs_rebase_original_qcow2_image(self, instance, backend):
if not isinstance(backend, imagebackend.Qcow2):
return False
if instance.vm_state == vm_states.SHELVED_OFFLOADED:
return True
if instance.task_state == task_states.RESIZE_FINISH:
# We need to distinguish between local versus cross cell resize.
# Rebase is only needed in cross cell case because instance
# is spawn from a snapshot.
base_image_ref = instance.system_metadata.get(
'image_base_image_ref')
if base_image_ref != instance.image_ref:
return True
return False
def _rebase_original_qcow2_image(self, context, instance, backend):
# NOTE(aarents): During qcow2 instance unshelve/cross_cell_resize,
# backing file represents a snapshot image, not original
# instance.image_ref. We rebase here instance disk to original image.
# This second fetch call does nothing except downloading original
# backing file if missing, as image disk have already been
# created/resized by first fetch call.
if not self._needs_rebase_original_qcow2_image(instance, backend):
return
base_dir = self.image_cache_manager.cache_dir
base_image_ref = instance.system_metadata.get('image_base_image_ref')
root_fname = imagecache.get_cache_fname(base_image_ref)
base_backing_fname = os.path.join(base_dir, root_fname)
try:
self._try_fetch_image_cache(backend, libvirt_utils.fetch_image,
context, root_fname, base_image_ref,
instance, None)
except exception.ImageNotFound:
# We must flatten here in order to remove dependency with an orphan
# backing file (as snapshot image will be dropped once
# unshelve/cross_cell_resize is successful).
LOG.warning('Current disk image is created on top of a snapshot '
'image and cannot be rebased to original image '
'because it is no longer available in the image '
'service, disk will be consequently flattened.',
instance=instance)
base_backing_fname = None
LOG.info('Rebasing disk image.', instance=instance)
self._rebase_with_qemu_img(backend.path, base_backing_fname)
def _create_configdrive(self, context, instance, injection_info,
rescue=False):
# As this method being called right after the definition of a
# domain, but before its actual launch, device metadata will be built
# and saved in the instance for it to be used by the config drive and
# the metadata service.
instance.device_metadata = self._build_device_metadata(context,
instance)
if configdrive.required_by(instance):
LOG.info('Using config drive', instance=instance)
name = 'disk.config'
if rescue:
name += '.rescue'
config_disk = self.image_backend.by_name(
instance, name, self._get_disk_config_image_type())
# Don't overwrite an existing config drive
if not config_disk.exists():
extra_md = {}
if injection_info.admin_pass:
extra_md['admin_pass'] = injection_info.admin_pass
inst_md = instance_metadata.InstanceMetadata(
instance, content=injection_info.files, extra_md=extra_md,
network_info=injection_info.network_info)
cdb = configdrive.ConfigDriveBuilder(instance_md=inst_md)
with cdb:
# NOTE(mdbooth): We're hardcoding here the path of the
# config disk when using the flat backend. This isn't
# good, but it's required because we need a local path we
# know we can write to in case we're subsequently
# importing into rbd. This will be cleaned up when we
# replace this with a call to create_from_func, but that
# can't happen until we've updated the backends and we
# teach them not to cache config disks. This isn't
# possible while we're still using cache() under the hood.
config_disk_local_path = os.path.join(
libvirt_utils.get_instance_path(instance), name)
LOG.info('Creating config drive at %(path)s',
{'path': config_disk_local_path},
instance=instance)
try:
cdb.make_drive(config_disk_local_path)
except processutils.ProcessExecutionError as e:
with excutils.save_and_reraise_exception():
LOG.error('Creating config drive failed with '
'error: %s', e, instance=instance)
try:
config_disk.import_file(
instance, config_disk_local_path, name)
finally:
# NOTE(mikal): if the config drive was imported into RBD,
# then we no longer need the local copy
if CONF.libvirt.images_type == 'rbd':
LOG.info('Deleting local config drive %(path)s '
'because it was imported into RBD.',
{'path': config_disk_local_path},
instance=instance)
os.unlink(config_disk_local_path)
def _detach_pci_devices(self, guest, pci_devs):
try:
for dev in pci_devs:
guest.detach_device(self._get_guest_pci_device(dev), live=True)
# after detachDeviceFlags returned, we should check the dom to
# ensure the detaching is finished
xml = guest.get_xml_desc()
xml_doc = etree.fromstring(xml)
guest_config = vconfig.LibvirtConfigGuest()
guest_config.parse_dom(xml_doc)
for hdev in [
d for d in guest_config.devices
if isinstance(d, vconfig.LibvirtConfigGuestHostdevPCI)
]:
hdbsf = [hdev.domain, hdev.bus, hdev.slot, hdev.function]
dbsf = pci_utils.parse_address(dev.address)
if (
[int(x, 16) for x in hdbsf] ==
[int(x, 16) for x in dbsf]
):
raise exception.PciDeviceDetachFailed(
reason="timeout", dev=dev)
except libvirt.libvirtError as ex:
error_code = ex.get_error_code()
if error_code == libvirt.VIR_ERR_NO_DOMAIN:
LOG.warning("Instance disappeared while detaching "
"a PCI device from it.")
else:
raise
def _attach_pci_devices(self, guest, pci_devs):
try:
for dev in pci_devs:
guest.attach_device(self._get_guest_pci_device(dev))
except libvirt.libvirtError:
LOG.error('Attaching PCI devices %(dev)s to %(dom)s failed.',
{'dev': pci_devs, 'dom': guest.id})
raise
@staticmethod
def _has_direct_passthrough_port(network_info):
for vif in network_info:
if (vif['vnic_type'] in
network_model.VNIC_TYPES_DIRECT_PASSTHROUGH):
return True
return False
def _attach_direct_passthrough_ports(
self, context, instance, guest, network_info=None):
if network_info is None:
network_info = instance.info_cache.network_info
if network_info is None:
return
if self._has_direct_passthrough_port(network_info):
for vif in network_info:
if (vif['vnic_type'] in
network_model.VNIC_TYPES_DIRECT_PASSTHROUGH):
cfg = self.vif_driver.get_config(instance,
vif,
instance.image_meta,
instance.flavor,
CONF.libvirt.virt_type)
LOG.debug('Attaching direct passthrough port %(port)s '
'to %(dom)s', {'port': vif, 'dom': guest.id},
instance=instance)
guest.attach_device(cfg)
# TODO(sean-k-mooney): we should try and converge this function with
# _detach_direct_passthrough_vifs which does the same operation correctly
# for live migration
def _detach_direct_passthrough_ports(self, context, instance, guest):
network_info = instance.info_cache.network_info
if network_info is None:
return
if self._has_direct_passthrough_port(network_info):
attached_via_hostdev_element = []
attached_via_interface_element = []
for vif in network_info:
if vif['profile'].get('pci_slot') is None:
# this is not an sriov interface so skip it
continue
if (vif['vnic_type'] not in
network_model.VNIC_TYPES_DIRECT_PASSTHROUGH):
continue
cfg = self.vif_driver.get_config(
instance, vif, instance.image_meta, instance.flavor,
CONF.libvirt.virt_type)
LOG.debug(f'Detaching type: {type(cfg)}, data: {cfg}')
if isinstance(cfg, vconfig.LibvirtConfigGuestHostdevPCI):
attached_via_hostdev_element.append(vif)
else:
attached_via_interface_element.append(vif)
pci_devs = instance.get_pci_devices()
hostdev_pci_addresses = {
vif['profile']['pci_slot']
for vif in attached_via_hostdev_element
}
direct_passthrough_pci_addresses = [
pci_dev for pci_dev in pci_devs
if pci_dev.address in hostdev_pci_addresses
]
# FIXME(sean-k-mooney): i am using _detach_pci_devices because
# of the previous comment introduced by change-id:
# I3a45b1fb41e8e446d1f25d7a1d77991c8bf2a1ed
# in relation to bug 1563874 however i'm not convinced that
# patch was correct so we should reevaluate if we should do this.
# The intent of using _detach_pci_devices is
# to somehow cater for the use case where multiple ports have
# the same MAC address however _detach_pci_device can only remove
# device that are attached as hostdev elements, not via the
# interface element.
# So using it for all devices would break vnic-type direct when
# using the sriov_nic_agent ml2 driver or vif of vnic_type vdpa.
# Since PF ports can't have the same MAC that means that this
# use case was for hardware offloaded OVS? many NICs do not allow
# two VFs to have the same MAC on different VLANs due to the
# ordering of the VLAN and MAC filters in there static packet
# processing pipeline as such its unclear if this will work in any
# non ovs offload case. We should look into this more closely
# as from my testing in this patch we appear to use the interface
# element for hardware offloaded ovs too. Infiniband and vnic_type
# direct-physical port type do need this code path, but those
# can't have duplicate MACs...
self._detach_pci_devices(guest, direct_passthrough_pci_addresses)
# for ports that are attached with interface elements we cannot use
# _detach_pci_devices so we use detach_interface
for vif in attached_via_interface_element:
self.detach_interface(context, instance, vif)
def _update_compute_provider_status(self, context, service):
"""Calls the ComputeVirtAPI.update_compute_provider_status method
:param context: nova auth RequestContext
:param service: nova.objects.Service record for this host which is
expected to only manage a single ComputeNode
"""
rp_uuid = None
try:
rp_uuid = service.compute_node.uuid
self.virtapi.update_compute_provider_status(
context, rp_uuid, enabled=not service.disabled)
except Exception:
# This is best effort so just log the exception but don't fail.
# The update_available_resource periodic task will sync the trait.
LOG.warning(
'An error occurred while updating compute node '
'resource provider status to "%s" for provider: %s',
'disabled' if service.disabled else 'enabled',
rp_uuid or service.host, exc_info=True)
def _set_host_enabled(self, enabled,
disable_reason=DISABLE_REASON_UNDEFINED):
"""Enables / Disables the compute service on this host.
This doesn't override non-automatic disablement with an automatic
setting; thereby permitting operators to keep otherwise
healthy hosts out of rotation.
"""
status_name = {True: 'disabled',
False: 'enabled'}
disable_service = not enabled
ctx = nova_context.get_admin_context()
try:
service = objects.Service.get_by_compute_host(ctx, CONF.host)
if service.disabled != disable_service:
# Note(jang): this is a quick fix to stop operator-
# disabled compute hosts from re-enabling themselves
# automatically. We prefix any automatic reason code
# with a fixed string. We only re-enable a host
# automatically if we find that string in place.
# This should probably be replaced with a separate flag.
if not service.disabled or (
service.disabled_reason and
service.disabled_reason.startswith(DISABLE_PREFIX)):
service.disabled = disable_service
service.disabled_reason = (
DISABLE_PREFIX + disable_reason
if disable_service and disable_reason else
DISABLE_REASON_UNDEFINED)
service.save()
LOG.debug('Updating compute service status to %s',
status_name[disable_service])
# Update the disabled trait status on the corresponding
# compute node resource provider in placement.
self._update_compute_provider_status(ctx, service)
else:
LOG.debug('Not overriding manual compute service '
'status with: %s',
status_name[disable_service])
except exception.ComputeHostNotFound:
LOG.warning('Cannot update service status on host "%s" '
'since it is not registered.', CONF.host)
except Exception:
LOG.warning('Cannot update service status on host "%s" '
'due to an unexpected exception.', CONF.host,
exc_info=True)
if enabled:
mount.get_manager().host_up(self._host)
else:
mount.get_manager().host_down()
def _check_emulation_arch(self, image_meta):
# NOTE(chateaulav) In order to support emulation via qemu,
# there are required metadata properties that need applied
# to the designated glance image. The config drive is not
# supported. This leverages the hw_architecture and
# hw_emulation_architecture image_meta fields to allow for
# emulation to take advantage of all physical multiarch work
# being done.
#
# aarch64 emulation support metadata values:
# 'hw_emulation_architecture=aarch64'
# 'hw_firmware_type=uefi'
# 'hw_machine_type=virt'
#
# ppc64le emulation support metadata values:
# 'hw_emulation_architecture=ppc64le'
# 'hw_machine_type=pseries'
#
# s390x emulation support metadata values:
# 'hw_emulation_architecture=s390x'
# 'hw_machine_type=s390-ccw-virtio'
# 'hw_video_model=virtio'
#
# TODO(chateaulav) Further Work to be done:
# testing mips functionality while waiting on redhat libvirt
# patch https://listman.redhat.com/archives/libvir-list/
# 2016-May/msg00197.html
#
# https://bugzilla.redhat.com/show_bug.cgi?id=1432101
emulation_arch = image_meta.properties.get("hw_emulation_architecture")
if emulation_arch:
arch = emulation_arch
else:
arch = libvirt_utils.get_arch(image_meta)
return arch
def _get_cpu_model_mapping(self, model):
"""Get the CPU model mapping
The CPU models which admin configured are case-insensitive, libvirt is
case-sensitive, therefore build a mapping to get the correct CPU model
name.
:param model: Case-insensitive CPU model name.
:return: It will validate and return the case-sensitive CPU model name
if on a supported platform, otherwise it will just return
what was provided
:raises: exception.InvalidCPUInfo if the CPU model is not supported.
"""
cpu_info = self._get_cpu_info()
if cpu_info['arch'] not in (fields.Architecture.I686,
fields.Architecture.X86_64,
fields.Architecture.PPC64,
fields.Architecture.PPC64LE,
fields.Architecture.PPC):
return model
if not self.cpu_models_mapping:
cpu_models = self._host.get_cpu_model_names()
for cpu_model in cpu_models:
self.cpu_models_mapping[cpu_model.lower()] = cpu_model
if model.lower() not in self.cpu_models_mapping:
msg = (_("Configured CPU model: %(model)s is not correct, "
"or your host CPU arch does not support this "
"model. Please correct your config and try "
"again.") % {'model': model})
raise exception.InvalidCPUInfo(msg)
return self.cpu_models_mapping.get(model.lower())
# TODO(stephenfin): Libvirt exposes information about possible CPU models
# via 'getDomainCapabilities' and we should use it
def _get_guest_cpu_model_config(self, flavor=None, arch=None):
mode = CONF.libvirt.cpu_mode
models = [self._get_cpu_model_mapping(model)
for model in CONF.libvirt.cpu_models]
extra_flags = set([flag.lower() for flag in
CONF.libvirt.cpu_model_extra_flags])
if not arch:
caps = self._host.get_capabilities()
arch = caps.host.cpu.arch
if (
CONF.libvirt.virt_type == "kvm" or
CONF.libvirt.virt_type == "qemu"
):
if mode is None:
# AArch64 lacks 'host-model' support because neither libvirt
# nor QEMU are able to tell what the host CPU model exactly is.
# And there is no CPU description code for ARM(64) at this
# point.
# Also worth noting: 'host-passthrough' mode will completely
# break live migration, *unless* all the Compute nodes (running
# libvirtd) have *identical* CPUs.
if arch == fields.Architecture.AARCH64:
mode = "host-passthrough"
LOG.info('CPU mode "host-passthrough" was chosen. Live '
'migration can break unless all compute nodes '
'have identical cpus. AArch64 does not support '
'other modes.')
else:
mode = "host-model"
if mode == "none":
return vconfig.LibvirtConfigGuestCPU()
# On AArch64 platform the return of _get_cpu_model_mapping will not
# return the default CPU model.
if mode == "custom":
if arch == fields.Architecture.AARCH64:
if not models:
models = ['max']
else:
if mode is None or mode == "none":
return None
cpu = vconfig.LibvirtConfigGuestCPU()
cpu.mode = mode
cpu.model = models[0] if models else None
# compare flavor trait and cpu models, select the first matched model
if flavor and mode == "custom":
flags = libvirt_utils.get_flags_by_flavor_specs(flavor)
if flags:
cpu.model = self._match_cpu_model_by_flags(models, flags)
LOG.debug("CPU mode '%(mode)s' models '%(models)s' was chosen, "
"with extra flags: '%(extra_flags)s'",
{'mode': mode,
'models': (cpu.model or ""),
'extra_flags': (extra_flags or "")})
# NOTE (kchamart): Currently there's no existing way to ask if a
# given CPU model + CPU flags combination is supported by KVM &
# a specific QEMU binary. However, libvirt runs the 'CPUID'
# command upfront -- before even a Nova instance (a QEMU
# process) is launched -- to construct CPU models and check
# their validity; so we are good there. In the long-term,
# upstream libvirt intends to add an additional new API that can
# do fine-grained validation of a certain CPU model + CPU flags
# against a specific QEMU binary (the libvirt RFE bug for that:
# https://bugzilla.redhat.com/show_bug.cgi?id=1559832).
#
# NOTE(kchamart) Similar to what was done in
# _check_cpu_compatibility(), the below parses a comma-separated
# list of CPU flags from `[libvirt]cpu_model_extra_flags` and
# will selectively enable or disable a given CPU flag for the
# guest, before it is launched by Nova.
for flag in extra_flags:
cpu_feature = self._prepare_cpu_flag(flag)
cpu.add_feature(cpu_feature)
return cpu
def _get_guest_cpu_config_maxphysaddr(self, flavor, image_meta):
mode = (flavor.extra_specs.get('hw:maxphysaddr_mode') or
image_meta.properties.get('hw_maxphysaddr_mode'))
bits = (flavor.extra_specs.get('hw:maxphysaddr_bits') or
image_meta.properties.get('hw_maxphysaddr_bits'))
if not mode:
return None
maxphysaddr = vconfig.LibvirtConfigGuestCPUMaxPhysAddr()
maxphysaddr.mode = mode
if bits:
maxphysaddr.bits = int(bits)
return maxphysaddr
def _match_cpu_model_by_flags(self, models, flags):
for model in models:
if flags.issubset(self.cpu_model_flag_mapping.get(model, set([]))):
return model
cpu = vconfig.LibvirtConfigCPU()
cpu.arch = self._host.get_capabilities().host.cpu.arch
cpu.model = model
features_xml = self._get_guest_baseline_cpu_features(cpu.to_xml())
if features_xml:
cpu.parse_str(features_xml)
feature_names = [f.name for f in cpu.features]
self.cpu_model_flag_mapping[model] = feature_names
if flags.issubset(feature_names):
return model
msg = ('No CPU model match traits, models: {models}, required '
'flags: {flags}'.format(models=models, flags=flags))
raise exception.InvalidCPUInfo(msg)
def _get_guest_cpu_config(self, flavor, image_meta,
guest_cpu_numa_config, instance_numa_topology):
arch = self._check_emulation_arch(image_meta)
cpu = self._get_guest_cpu_model_config(flavor, arch)
if cpu is None:
return None
topology = hardware.get_best_cpu_topology(flavor, image_meta)
cpu.sockets = topology.sockets
cpu.cores = topology.cores
cpu.threads = topology.threads
cpu.numa = guest_cpu_numa_config
cpu.maxphysaddr = self._get_guest_cpu_config_maxphysaddr(flavor,
image_meta)
caps = self._host.get_capabilities()
if arch != caps.host.cpu.arch:
# Try emulating. Other arch configs will go here
cpu.mode = None
if arch == fields.Architecture.AARCH64:
cpu.model = "cortex-a57"
elif arch == fields.Architecture.PPC64LE:
cpu.model = "POWER8"
# TODO(chateaulav): re-evaluate when libvirtd adds overall
# RISCV support as a supported architecture, as there is no
# cpu models associated, this simply associates X vcpus to the
# guest according to the flavor. These same issue should be
# present with mipsel due to same limitation, but has not been
# tested.
elif arch == fields.Architecture.MIPSEL:
cpu = None
return cpu
def _get_guest_disk_config(
self, instance, name, disk_mapping, flavor, image_type=None,
boot_order=None,
):
# NOTE(artom) To pass unit tests, wherein the code here is loaded
# *before* any config with self.flags() is done, we need to have the
# default inline in the method, and not in the kwarg declaration.
if image_type is None:
image_type = CONF.libvirt.images_type
disk_unit = None
disk_info_mapping = disk_mapping[name]
disk = self.image_backend.by_name(
instance, name, image_type, disk_info_mapping=disk_info_mapping)
if (name == 'disk.config' and image_type == 'rbd' and
not disk.exists()):
# This is likely an older config drive that has not been migrated
# to rbd yet. Try to fall back on 'flat' image type.
# TODO(melwitt): Add online migration of some sort so we can
# remove this fall back once we know all config drives are in rbd.
# NOTE(vladikr): make sure that the flat image exist, otherwise
# the image will be created after the domain definition.
flat_disk = self.image_backend.by_name(
instance, name, 'flat', disk_info_mapping=disk_info_mapping)
if flat_disk.exists():
disk = flat_disk
LOG.debug('Config drive not found in RBD, falling back to the '
'instance directory', instance=instance)
# The 'unit' key is global to the disk_mapping (rather than for an
# individual disk) because it is used solely to track the incrementing
# unit number.
if 'unit' in disk_mapping and disk_info_mapping['bus'] == 'scsi':
disk_unit = disk_mapping['unit']
disk_mapping['unit'] += 1 # Increments for the next disk
conf = disk.libvirt_info(
self.disk_cachemode, flavor['extra_specs'], disk_unit=disk_unit,
boot_order=boot_order)
return conf
def _get_guest_fs_config(
self, instance, name, image_type=CONF.libvirt.images_type
):
disk = self.image_backend.by_name(instance, name, image_type)
return disk.libvirt_fs_info("/", "ploop")
def _get_guest_storage_config(
self, context, instance, image_meta, disk_info, rescue,
block_device_info, flavor, os_type,
):
devices = []
disk_mapping = disk_info['mapping']
block_device_mapping = driver.block_device_info_get_mapping(
block_device_info)
mount_rootfs = CONF.libvirt.virt_type == "lxc"
scsi_controller = self._get_scsi_controller(image_meta)
if scsi_controller and scsi_controller.model == 'virtio-scsi':
# The virtio-scsi can handle up to 256 devices but the
# optional element "address" must be defined to describe
# where the device is placed on the controller (see:
# LibvirtConfigGuestDeviceAddressDrive).
#
# Note about why it's added in disk_mapping: It's not
# possible to pass an 'int' by reference in Python, so we
# use disk_mapping as container to keep reference of the
# unit added and be able to increment it for each disk
# added.
#
# NOTE(jaypipes,melwitt): If this is a boot-from-volume instance,
# we need to start the disk mapping unit at 1 since we set the
# bootable volume's unit to 0 for the bootable volume.
disk_mapping['unit'] = 0
if self._is_booted_from_volume(block_device_info):
disk_mapping['unit'] = 1
def _get_ephemeral_devices():
eph_devices = []
for idx, eph in enumerate(
driver.block_device_info_get_ephemerals(
block_device_info)):
diskeph = self._get_guest_disk_config(
instance,
blockinfo.get_eph_disk(idx),
disk_mapping, flavor)
eph_devices.append(diskeph)
return eph_devices
if mount_rootfs:
fs = vconfig.LibvirtConfigGuestFilesys()
fs.source_type = "mount"
fs.source_dir = os.path.join(
libvirt_utils.get_instance_path(instance), 'rootfs')
devices.append(fs)
elif (os_type == fields.VMMode.EXE and
CONF.libvirt.virt_type == "parallels"):
if rescue:
fsrescue = self._get_guest_fs_config(instance, "disk.rescue")
devices.append(fsrescue)
fsos = self._get_guest_fs_config(instance, "disk")
fsos.target_dir = "/mnt/rescue"
devices.append(fsos)
else:
if 'disk' in disk_mapping:
fs = self._get_guest_fs_config(instance, "disk")
devices.append(fs)
devices = devices + _get_ephemeral_devices()
else:
if rescue and disk_mapping['disk.rescue'] == disk_mapping['root']:
diskrescue = self._get_guest_disk_config(
instance, 'disk.rescue', disk_mapping, flavor)
devices.append(diskrescue)
diskos = self._get_guest_disk_config(
instance, 'disk', disk_mapping, flavor)
devices.append(diskos)
else:
if 'disk' in disk_mapping:
diskos = self._get_guest_disk_config(
instance, 'disk', disk_mapping, flavor)
devices.append(diskos)
if 'disk.local' in disk_mapping:
disklocal = self._get_guest_disk_config(
instance, 'disk.local', disk_mapping, flavor)
devices.append(disklocal)
instance.default_ephemeral_device = (
block_device.prepend_dev(disklocal.target_dev))
devices = devices + _get_ephemeral_devices()
if 'disk.swap' in disk_mapping:
diskswap = self._get_guest_disk_config(
instance, 'disk.swap', disk_mapping, flavor)
devices.append(diskswap)
instance.default_swap_device = (
block_device.prepend_dev(diskswap.target_dev))
config_name = 'disk.config'
if rescue and disk_mapping['disk.rescue'] == disk_mapping['root']:
config_name = 'disk.config.rescue'
if config_name in disk_mapping:
diskconfig = self._get_guest_disk_config(
instance, config_name, disk_mapping, flavor,
self._get_disk_config_image_type())
devices.append(diskconfig)
for vol in block_device.get_bdms_to_connect(block_device_mapping,
mount_rootfs):
connection_info = vol['connection_info']
vol_dev = block_device.prepend_dev(vol['mount_device'])
info = disk_mapping[vol_dev]
self._connect_volume(context, connection_info, instance)
if scsi_controller and scsi_controller.model == 'virtio-scsi':
# Check if this is the bootable volume when in a
# boot-from-volume instance, and if so, ensure the unit
# attribute is 0.
if vol.get('boot_index') == 0:
info['unit'] = 0
else:
info['unit'] = disk_mapping['unit']
disk_mapping['unit'] += 1
cfg = self._get_volume_config(instance, connection_info, info)
devices.append(cfg)
vol['connection_info'] = connection_info
vol.save()
for d in devices:
self._set_cache_mode(d)
if scsi_controller:
devices.append(scsi_controller)
if rescue and disk_mapping['disk.rescue'] != disk_mapping['root']:
diskrescue = self._get_guest_disk_config(
instance, 'disk.rescue', disk_mapping, flavor, boot_order='1')
devices.append(diskrescue)
return devices
@staticmethod
def _get_scsi_controller(image_meta):
"""Return scsi controller or None based on image meta"""
if image_meta.properties.get('hw_scsi_model'):
hw_scsi_model = image_meta.properties.hw_scsi_model
scsi_controller = vconfig.LibvirtConfigGuestController()
scsi_controller.type = 'scsi'
scsi_controller.model = hw_scsi_model
scsi_controller.index = 0
return scsi_controller
def _get_host_sysinfo_serial_hardware(self):
"""Get a UUID from the host hardware
Get a UUID for the host hardware reported by libvirt.
This is typically from the SMBIOS data, unless it has
been overridden in /etc/libvirt/libvirtd.conf
"""
caps = self._host.get_capabilities()
return caps.host.uuid
def _get_host_sysinfo_serial_os(self):
"""Get a UUID from the host operating system
Get a UUID for the host operating system. Modern Linux
distros based on systemd provide a /etc/machine-id
file containing a UUID. This is also provided inside
systemd based containers and can be provided by other
init systems too, since it is just a plain text file.
"""
if not os.path.exists("/etc/machine-id"):
msg = _("Unable to get host UUID: /etc/machine-id does not exist")
raise exception.InternalError(msg)
with open("/etc/machine-id") as f:
# We want to have '-' in the right place
# so we parse & reformat the value
lines = f.read().split()
if not lines:
msg = _("Unable to get host UUID: /etc/machine-id is empty")
raise exception.InternalError(msg)
return str(uuid.UUID(lines[0]))
def _get_host_sysinfo_serial_auto(self):
if os.path.exists("/etc/machine-id"):
return self._get_host_sysinfo_serial_os()
else:
return self._get_host_sysinfo_serial_hardware()
def _get_guest_config_sysinfo(self, instance):
sysinfo = vconfig.LibvirtConfigGuestSysinfo()
sysinfo.system_manufacturer = version.vendor_string()
sysinfo.system_product = version.product_string()
sysinfo.system_version = version.version_string_with_package()
if CONF.libvirt.sysinfo_serial == 'unique':
sysinfo.system_serial = instance.uuid
else:
sysinfo.system_serial = self._sysinfo_serial_func()
sysinfo.system_uuid = instance.uuid
sysinfo.system_family = "Virtual Machine"
return sysinfo
def _set_managed_mode(self, pcidev, managed):
# only kvm support managed mode
if CONF.libvirt.virt_type in ('parallels',):
pcidev.managed = 'no'
LOG.debug("Managed mode set to '%s' but it is overwritten by "
"parallels hypervisor settings.", managed)
if CONF.libvirt.virt_type in ('kvm', 'qemu'):
pcidev.managed = "yes" if managed == "true" else "no"
def _get_guest_pci_device(self, pci_device):
dbsf = pci_utils.parse_address(pci_device.address)
dev = vconfig.LibvirtConfigGuestHostdevPCI()
dev.domain, dev.bus, dev.slot, dev.function = dbsf
managed = pci_device.extra_info.get('managed', 'true')
self._set_managed_mode(dev, managed)
return dev
def _get_guest_config_meta(self, dmeta: driver.InstanceDriverMetadata):
"""Get metadata config for guest."""
meta = vconfig.LibvirtConfigGuestMetaNovaInstance()
meta.package = dmeta.nova_package
meta.name = dmeta.instance_meta.name
meta.creationTime = dmeta.creation_time
meta.roottype = dmeta.root_type
meta.rootid = dmeta.root_id
ometa = vconfig.LibvirtConfigGuestMetaNovaOwner()
ometa.userid = dmeta.owner.userid
ometa.username = dmeta.owner.username
ometa.projectid = dmeta.owner.projectid
ometa.projectname = dmeta.owner.projectname
meta.owner = ometa
fmeta = vconfig.LibvirtConfigGuestMetaNovaFlavor()
fmeta.name = dmeta.flavor.name
fmeta.memory = dmeta.flavor.memory_mb
fmeta.vcpus = dmeta.flavor.vcpus
fmeta.ephemeral = dmeta.flavor.ephemeral_gb
fmeta.disk = dmeta.flavor.root_gb
fmeta.swap = dmeta.flavor.swap
meta.flavor = fmeta
ports = []
for vif in dmeta.network_info:
ips = []
for subnet in vif.get('network', {}).get('subnets', []):
for ip in subnet.get('ips', []):
ips.append(vconfig.LibvirtConfigGuestMetaNovaIp(
ip.get('type'), ip.get('address'), ip.get('version')))
ports.append(vconfig.LibvirtConfigGuestMetaNovaPort(
vif.get('id'), ips=ips))
meta.ports = vconfig.LibvirtConfigGuestMetaNovaPorts(ports)
return meta
@staticmethod
def _create_idmaps(klass, map_strings):
idmaps = []
if len(map_strings) > 5:
map_strings = map_strings[0:5]
LOG.warning("Too many id maps, only included first five.")
for map_string in map_strings:
try:
idmap = klass()
values = [int(i) for i in map_string.split(":")]
idmap.start = values[0]
idmap.target = values[1]
idmap.count = values[2]
idmaps.append(idmap)
except (ValueError, IndexError):
LOG.warning("Invalid value for id mapping %s", map_string)
return idmaps
def _get_guest_idmaps(self):
id_maps: ty.List[vconfig.LibvirtConfigGuestIDMap] = []
if CONF.libvirt.virt_type == 'lxc' and CONF.libvirt.uid_maps:
uid_maps = self._create_idmaps(vconfig.LibvirtConfigGuestUIDMap,
CONF.libvirt.uid_maps)
id_maps.extend(uid_maps)
if CONF.libvirt.virt_type == 'lxc' and CONF.libvirt.gid_maps:
gid_maps = self._create_idmaps(vconfig.LibvirtConfigGuestGIDMap,
CONF.libvirt.gid_maps)
id_maps.extend(gid_maps)
return id_maps
def _update_guest_cputune(self, guest, flavor):
is_able = self._host.is_cpu_control_policy_capable()
cputuning = ['shares', 'period', 'quota']
wants_cputune = any([k for k in cputuning
if "quota:cpu_" + k in flavor.extra_specs.keys()])
if wants_cputune and not is_able:
raise exception.UnsupportedHostCPUControlPolicy()
if not is_able or CONF.libvirt.virt_type not in ('lxc', 'kvm', 'qemu'):
return
for name in cputuning:
key = "quota:cpu_" + name
if key in flavor.extra_specs:
if guest.cputune is None:
guest.cputune = vconfig.LibvirtConfigGuestCPUTune()
setattr(guest.cputune, name,
int(flavor.extra_specs[key]))
def _get_cpu_numa_config_from_instance(self, instance_numa_topology,
wants_hugepages):
if instance_numa_topology:
guest_cpu_numa = vconfig.LibvirtConfigGuestCPUNUMA()
for instance_cell in instance_numa_topology.cells:
guest_cell = vconfig.LibvirtConfigGuestCPUNUMACell()
guest_cell.id = instance_cell.id
guest_cell.cpus = instance_cell.total_cpus
guest_cell.memory = instance_cell.memory * units.Ki
# The vhost-user network backend requires file backed
# guest memory (ie huge pages) to be marked as shared
# access, not private, so an external process can read
# and write the pages.
#
# You can't change the shared vs private flag for an
# already running guest, and since we can't predict what
# types of NIC may be hotplugged, we have no choice but
# to unconditionally turn on the shared flag. This has
# no real negative functional effect on the guest, so
# is a reasonable approach to take
if wants_hugepages:
guest_cell.memAccess = "shared"
guest_cpu_numa.cells.append(guest_cell)
return guest_cpu_numa
def _wants_hugepages(self, host_topology, instance_topology):
"""Determine if the guest / host topology implies the
use of huge pages for guest RAM backing
"""
if host_topology is None or instance_topology is None:
return False
avail_pagesize = [page.size_kb
for page in host_topology.cells[0].mempages]
avail_pagesize.sort()
# Remove smallest page size as that's not classed as a largepage
avail_pagesize = avail_pagesize[1:]
# See if we have page size set
for cell in instance_topology.cells:
if (cell.pagesize is not None and
cell.pagesize in avail_pagesize):
return True
return False
def _get_cell_pairs(self, guest_cpu_numa_config, host_topology):
"""Returns the lists of pairs(tuple) of an instance cell and
corresponding host cell:
[(LibvirtConfigGuestCPUNUMACell, NUMACell), ...]
"""
cell_pairs = []
for guest_config_cell in guest_cpu_numa_config.cells:
for host_cell in host_topology.cells:
if guest_config_cell.id == host_cell.id:
cell_pairs.append((guest_config_cell, host_cell))
return cell_pairs
def _get_pin_cpuset(self, vcpu, inst_cell, host_cell):
"""Returns the config object of LibvirtConfigGuestCPUTuneVCPUPin.
Prepares vcpupin config for the guest with the following caveats:
a) If the specified instance vCPU is intended to be pinned, we pin
it to the previously selected host CPU.
b) Otherwise we float over the whole host NUMA node
"""
pin_cpuset = vconfig.LibvirtConfigGuestCPUTuneVCPUPin()
pin_cpuset.id = vcpu
# 'InstanceNUMACell.cpu_pinning' tracks the CPU pinning pair for guest
# CPU and host CPU. If the guest CPU is in the keys of 'cpu_pinning',
# fetch the host CPU from it and pin on it, otherwise, let the guest
# CPU be floating on the sharing CPU set belonging to this NUMA cell.
if inst_cell.cpu_pinning and vcpu in inst_cell.cpu_pinning:
pin_cpuset.cpuset = set([inst_cell.cpu_pinning[vcpu]])
else:
pin_cpuset.cpuset = host_cell.cpuset
return pin_cpuset
def _get_emulatorpin_cpuset(self, vcpu, object_numa_cell, vcpus_rt,
emulator_threads_policy,
pin_cpuset):
"""Returns a set of cpu_ids to add to the cpuset for emulator threads
with the following caveats:
a) If emulator threads policy is isolated, we pin emulator threads
to one cpu we have reserved for it.
b) If emulator threads policy is shared and CONF.cpu_shared_set is
defined, we pin emulator threads on the set of pCPUs defined by
CONF.cpu_shared_set
c) Otherwise;
c1) If realtime IS NOT enabled, the emulator threads are
allowed to float cross all the pCPUs associated with
the guest vCPUs.
c2) If realtime IS enabled, at least 1 vCPU is required
to be set aside for non-realtime usage. The emulator
threads are allowed to float across the pCPUs that
are associated with the non-realtime VCPUs.
"""
emulatorpin_cpuset = set([])
shared_ids = hardware.get_cpu_shared_set()
if emulator_threads_policy == fields.CPUEmulatorThreadsPolicy.ISOLATE:
if object_numa_cell.cpuset_reserved:
emulatorpin_cpuset = object_numa_cell.cpuset_reserved
elif ((emulator_threads_policy ==
fields.CPUEmulatorThreadsPolicy.SHARE) and
shared_ids):
online_pcpus = self._host.get_online_cpus()
cpuset = shared_ids & online_pcpus
if not cpuset:
msg = (_("Invalid cpu_shared_set config, one or more of the "
"specified cpuset is not online. Online cpuset(s): "
"%(online)s, requested cpuset(s): %(req)s"),
{'online': sorted(online_pcpus),
'req': sorted(shared_ids)})
raise exception.Invalid(msg)
emulatorpin_cpuset = cpuset
elif not vcpus_rt or vcpu not in vcpus_rt:
emulatorpin_cpuset = pin_cpuset.cpuset
return emulatorpin_cpuset
def _get_guest_numa_config(self, instance_numa_topology, flavor,
image_meta):
"""Returns the config objects for the guest NUMA specs.
Determines the CPUs that the guest can be pinned to if the guest
specifies a cell topology and the host supports it. Constructs the
libvirt XML config object representing the NUMA topology selected
for the guest. Returns a tuple of:
(cpu_set, guest_cpu_tune, guest_cpu_numa, guest_numa_tune)
With the following caveats:
a) If there is no specified guest NUMA topology, then
all tuple elements except cpu_set shall be None. cpu_set
will be populated with the chosen CPUs that the guest
allowed CPUs fit within.
b) If there is a specified guest NUMA topology, then
cpu_set will be None and guest_cpu_numa will be the
LibvirtConfigGuestCPUNUMA object representing the guest's
NUMA topology. If the host supports NUMA, then guest_cpu_tune
will contain a LibvirtConfigGuestCPUTune object representing
the optimized chosen cells that match the host capabilities
with the instance's requested topology. If the host does
not support NUMA, then guest_cpu_tune and guest_numa_tune
will be None.
"""
if (not self._has_numa_support() and
instance_numa_topology is not None):
# We should not get here, since we should have avoided
# reporting NUMA topology from _get_host_numa_topology
# in the first place. Just in case of a scheduler
# mess up though, raise an exception
raise exception.NUMATopologyUnsupported()
# We only pin an instance to some host cores if the user has provided
# configuration to suggest we should.
shared_cpus = None
if CONF.vcpu_pin_set or CONF.compute.cpu_shared_set:
shared_cpus = self._get_vcpu_available()
topology = self._get_host_numa_topology()
# We have instance NUMA so translate it to the config class
guest_cpu_numa_config = self._get_cpu_numa_config_from_instance(
instance_numa_topology,
self._wants_hugepages(topology, instance_numa_topology))
if not guest_cpu_numa_config:
# No NUMA topology defined for instance - let the host kernel deal
# with the NUMA effects.
# TODO(ndipanov): Attempt to spread the instance
# across NUMA nodes and expose the topology to the
# instance as an optimisation
return GuestNumaConfig(shared_cpus, None, None, None)
if not topology:
# No NUMA topology defined for host - This will only happen with
# some libvirt versions and certain platforms.
return GuestNumaConfig(shared_cpus, None,
guest_cpu_numa_config, None)
# Now get configuration from the numa_topology
# Init CPUTune configuration
guest_cpu_tune = vconfig.LibvirtConfigGuestCPUTune()
guest_cpu_tune.emulatorpin = (
vconfig.LibvirtConfigGuestCPUTuneEmulatorPin())
guest_cpu_tune.emulatorpin.cpuset = set([])
# Init NUMATune configuration
guest_numa_tune = vconfig.LibvirtConfigGuestNUMATune()
guest_numa_tune.memory = vconfig.LibvirtConfigGuestNUMATuneMemory()
guest_numa_tune.memnodes = []
emulator_threads_policy = None
if 'emulator_threads_policy' in instance_numa_topology:
emulator_threads_policy = (
instance_numa_topology.emulator_threads_policy)
# Set realtime scheduler for CPUTune
vcpus_rt = hardware.get_realtime_cpu_constraint(flavor, image_meta)
if vcpus_rt:
vcpusched = vconfig.LibvirtConfigGuestCPUTuneVCPUSched()
designer.set_vcpu_realtime_scheduler(
vcpusched, vcpus_rt, CONF.libvirt.realtime_scheduler_priority)
guest_cpu_tune.vcpusched.append(vcpusched)
cell_pairs = self._get_cell_pairs(guest_cpu_numa_config, topology)
for guest_node_id, (guest_config_cell, host_cell) in enumerate(
cell_pairs):
# set NUMATune for the cell
tnode = vconfig.LibvirtConfigGuestNUMATuneMemNode()
designer.set_numa_memnode(tnode, guest_node_id, host_cell.id)
guest_numa_tune.memnodes.append(tnode)
guest_numa_tune.memory.nodeset.append(host_cell.id)
# set CPUTune for the cell
object_numa_cell = instance_numa_topology.cells[guest_node_id]
for cpu in guest_config_cell.cpus:
pin_cpuset = self._get_pin_cpuset(cpu, object_numa_cell,
host_cell)
guest_cpu_tune.vcpupin.append(pin_cpuset)
emu_pin_cpuset = self._get_emulatorpin_cpuset(
cpu, object_numa_cell, vcpus_rt,
emulator_threads_policy, pin_cpuset)
guest_cpu_tune.emulatorpin.cpuset.update(emu_pin_cpuset)
# TODO(berrange) When the guest has >1 NUMA node, it will
# span multiple host NUMA nodes. By pinning emulator threads
# to the union of all nodes, we guarantee there will be
# cross-node memory access by the emulator threads when
# responding to guest I/O operations. The only way to avoid
# this would be to pin emulator threads to a single node and
# tell the guest OS to only do I/O from one of its virtual
# NUMA nodes. This is not even remotely practical.
#
# The long term solution is to make use of a new QEMU feature
# called "I/O Threads" which will let us configure an explicit
# I/O thread for each guest vCPU or guest NUMA node. It is
# still TBD how to make use of this feature though, especially
# how to associate IO threads with guest devices to eliminate
# cross NUMA node traffic. This is an area of investigation
# for QEMU community devs.
# Sort the vcpupin list per vCPU id for human-friendlier XML
guest_cpu_tune.vcpupin.sort(key=operator.attrgetter("id"))
# normalize cell.id
for i, (cell, memnode) in enumerate(zip(guest_cpu_numa_config.cells,
guest_numa_tune.memnodes)):
cell.id = i
memnode.cellid = i
return GuestNumaConfig(None, guest_cpu_tune, guest_cpu_numa_config,
guest_numa_tune)
def _get_guest_os_type(self):
"""Returns the guest OS type based on virt type."""
if CONF.libvirt.virt_type == "lxc":
ret = fields.VMMode.EXE
else:
ret = fields.VMMode.HVM
return ret
def _set_guest_for_rescue(
self, rescue, guest, inst_path, root_device_name,
):
if rescue.get('kernel_id'):
guest.os_kernel = os.path.join(inst_path, "kernel.rescue")
guest.os_cmdline = ("root=%s %s" % (root_device_name, CONSOLE))
if CONF.libvirt.virt_type == "qemu":
guest.os_cmdline += " no_timer_check"
if rescue.get('ramdisk_id'):
guest.os_initrd = os.path.join(inst_path, "ramdisk.rescue")
def _set_guest_for_inst_kernel(
self, instance, guest, inst_path, root_device_name, image_meta,
):
guest.os_kernel = os.path.join(inst_path, "kernel")
guest.os_cmdline = ("root=%s %s" % (root_device_name, CONSOLE))
if CONF.libvirt.virt_type == "qemu":
guest.os_cmdline += " no_timer_check"
if instance.ramdisk_id:
guest.os_initrd = os.path.join(inst_path, "ramdisk")
# we only support os_command_line with images with an explicit
# kernel set and don't want to break nova if there's an
# os_command_line property without a specified kernel_id param
if image_meta.properties.get("os_command_line"):
guest.os_cmdline = image_meta.properties.os_command_line
def _set_clock(self, guest, os_type, image_meta):
# NOTE(mikal): Microsoft Windows expects the clock to be in
# "localtime". If the clock is set to UTC, then you can use a
# registry key to let windows know, but Microsoft says this is
# buggy in http://support.microsoft.com/kb/2687252
clk = vconfig.LibvirtConfigGuestClock()
if os_type == 'windows':
LOG.info('Configuring timezone for windows instance to localtime')
clk.offset = 'localtime'
else:
clk.offset = 'utc'
guest.set_clock(clk)
if CONF.libvirt.virt_type == "kvm":
self._set_kvm_timers(clk, os_type, image_meta)
def _set_kvm_timers(self, clk, os_type, image_meta):
# TODO(berrange) One day this should be per-guest
# OS type configurable
tmpit = vconfig.LibvirtConfigGuestTimer()
tmpit.name = "pit"
tmpit.tickpolicy = "delay"
tmrtc = vconfig.LibvirtConfigGuestTimer()
tmrtc.name = "rtc"
tmrtc.tickpolicy = "catchup"
clk.add_timer(tmpit)
clk.add_timer(tmrtc)
hpet = image_meta.properties.get('hw_time_hpet', False)
guestarch = self._check_emulation_arch(image_meta)
if guestarch in (fields.Architecture.I686,
fields.Architecture.X86_64):
# NOTE(rfolco): HPET is a hardware timer for x86 arch.
# qemu -no-hpet is not supported on non-x86 targets.
tmhpet = vconfig.LibvirtConfigGuestTimer()
tmhpet.name = "hpet"
tmhpet.present = hpet
clk.add_timer(tmhpet)
else:
if hpet:
LOG.warning('HPET is not turned on for non-x86 guests in image'
' %s.', image_meta.id)
# Provide Windows guests with the paravirtualized hyperv timer source.
# This is the windows equiv of kvm-clock, allowing Windows
# guests to accurately keep time.
if os_type == 'windows':
tmhyperv = vconfig.LibvirtConfigGuestTimer()
tmhyperv.name = "hypervclock"
tmhyperv.present = True
clk.add_timer(tmhyperv)
def _set_features(self, guest, os_type, image_meta, flavor):
hide_hypervisor_id = (strutils.bool_from_string(
flavor.extra_specs.get('hide_hypervisor_id')) or
strutils.bool_from_string(
flavor.extra_specs.get('hw:hide_hypervisor_id')) or
image_meta.properties.get('img_hide_hypervisor_id'))
if CONF.libvirt.virt_type in ('qemu', 'kvm'):
guest.add_feature(vconfig.LibvirtConfigGuestFeatureACPI())
if not CONF.workarounds.libvirt_disable_apic:
guest.add_feature(vconfig.LibvirtConfigGuestFeatureAPIC())
if (
CONF.libvirt.virt_type == 'qemu' and
CONF.libvirt.tb_cache_size and
CONF.libvirt.tb_cache_size > 0
):
guest.add_feature(vconfig.LibvirtConfigGuestFeatureTCG(
CONF.libvirt.tb_cache_size))
if CONF.libvirt.virt_type in ('qemu', 'kvm') and os_type == 'windows':
hv = vconfig.LibvirtConfigGuestFeatureHyperV()
hv.relaxed = True
hv.spinlocks = True
# Increase spinlock retries - value recommended by
# KVM maintainers who certify Windows guests
# with Microsoft
hv.spinlock_retries = 8191
hv.vapic = True
hv.vpindex = True
hv.runtime = True
hv.synic = True
hv.reset = True
hv.frequencies = True
hv.tlbflush = True
hv.ipi = True
# NOTE(kosamara): Spoofing the vendor_id aims to allow the nvidia
# driver to work on windows VMs. At the moment, the nvidia driver
# checks for the hyperv vendorid, and if it doesn't find that, it
# works. In the future, its behaviour could become more strict,
# checking for the presence of other hyperv feature flags to
# determine that it's loaded in a VM. If that happens, this
# workaround will not be enough, and we'll need to drop the whole
# hyperv element.
# That would disable some optimizations, reducing the guest's
# performance.
if hide_hypervisor_id:
hv.vendorid_spoof = True
guest.features.append(hv)
if CONF.libvirt.virt_type in ("qemu", "kvm"):
# vmcoreinfo support is x86, ARM-only for now
guestarch = self._check_emulation_arch(image_meta)
if guestarch in (
fields.Architecture.I686, fields.Architecture.X86_64,
fields.Architecture.AARCH64,
):
guest.add_feature(
vconfig.LibvirtConfigGuestFeatureVMCoreInfo())
if hide_hypervisor_id:
guest.add_feature(
vconfig.LibvirtConfigGuestFeatureKvmHidden())
pmu = hardware.get_pmu_constraint(flavor, image_meta)
if pmu is not None:
guest.add_feature(
vconfig.LibvirtConfigGuestFeaturePMU(pmu))
def _check_number_of_serial_console(self, num_ports):
if (
CONF.libvirt.virt_type in ("kvm", "qemu") and
num_ports > ALLOWED_QEMU_SERIAL_PORTS
):
raise exception.SerialPortNumberLimitExceeded(
allowed=ALLOWED_QEMU_SERIAL_PORTS,
virt_type=CONF.libvirt.virt_type)
def _video_model_supported(self, model):
return model in fields.VideoModel.ALL
def _add_video_driver(self, guest, image_meta, flavor):
video = vconfig.LibvirtConfigGuestVideo()
video.type = self._get_video_type(image_meta) or video.type
# Set video memory, only if the flavor's limit is set
video_ram = image_meta.properties.get('hw_video_ram', 0)
max_vram = int(flavor.extra_specs.get('hw_video:ram_max_mb', 0))
if video_ram > max_vram:
raise exception.RequestedVRamTooHigh(req_vram=video_ram,
max_vram=max_vram)
if max_vram and video_ram:
video.vram = video_ram * units.Mi // units.Ki
guest.add_device(video)
# NOTE(sean-k-mooney): return the video device we added
# for simpler testing.
return video
def _get_video_type(
self,
image_meta: objects.ImageMeta,
) -> ty.Optional[str]:
# NOTE(ldbragst): The following logic returns the video type
# depending on supported defaults given the architecture,
# virtualization type, and features. The video type can
# be overridden by the user with image_meta.properties, which
# is carried out first.
if image_meta.properties.get('hw_video_model'):
video_type = image_meta.properties.hw_video_model
if not self._video_model_supported(video_type):
raise exception.InvalidVideoMode(model=video_type)
return video_type
guestarch = self._check_emulation_arch(image_meta)
if CONF.libvirt.virt_type == 'parallels':
return 'vga'
# NOTE(kchamart): 'virtio' is a sensible default whether or not
# the guest has the native kernel driver (called "virtio-gpu" in
# Linux) -- i.e. if the guest has the VirtIO GPU driver, it'll
# be used; otherwise, the 'virtio' model will gracefully
# fallback to VGA compatibility mode.
if (
guestarch in (
fields.Architecture.I686,
fields.Architecture.X86_64
) and not CONF.spice.enabled
):
return 'virtio'
if (
guestarch in (
fields.Architecture.PPC,
fields.Architecture.PPC64,
fields.Architecture.PPC64LE
)
):
# NOTE(ldbragst): PowerKVM doesn't support 'cirrus' be default
# so use 'vga' instead when running on Power hardware.
return 'vga'
if guestarch == fields.Architecture.AARCH64:
# NOTE(kevinz): Only virtio device type is supported by AARCH64
# so use 'virtio' instead when running on AArch64 hardware.
return 'virtio'
elif guestarch == fields.Architecture.MIPSEL:
return 'virtio'
# NOTE(lyarwood): Return None and default to the default of
# LibvirtConfigGuestVideo.type that is currently virtio
return None
def _add_qga_device(self, guest, instance):
qga = vconfig.LibvirtConfigGuestChannel()
qga.type = "unix"
qga.target_name = "org.qemu.guest_agent.0"
qga.source_path = ("/var/lib/libvirt/qemu/%s.%s.sock" %
("org.qemu.guest_agent.0", instance.name))
guest.add_device(qga)
def _add_rng_device(self, guest, flavor, image_meta):
rng_allowed_str = flavor.extra_specs.get('hw_rng:allowed', 'True')
rng_allowed = strutils.bool_from_string(rng_allowed_str)
if not rng_allowed:
return
rng_device = vconfig.LibvirtConfigGuestRng()
rate_bytes = flavor.extra_specs.get('hw_rng:rate_bytes', 0)
period = flavor.extra_specs.get('hw_rng:rate_period', 0)
if rate_bytes:
rng_device.rate_bytes = int(rate_bytes)
rng_device.rate_period = int(period)
rng_path = CONF.libvirt.rng_dev_path
if (rng_path and not os.path.exists(rng_path)):
raise exception.RngDeviceNotExist(path=rng_path)
rng_device.backend = rng_path
guest.add_device(rng_device)
def _add_virtio_serial_controller(self, guest, instance):
virtio_controller = vconfig.LibvirtConfigGuestController()
virtio_controller.type = 'virtio-serial'
guest.add_device(virtio_controller)
def _add_vtpm_device(
self,
guest: vconfig.LibvirtConfigGuest,
flavor: 'objects.Flavor',
instance: 'objects.Instance',
image_meta: 'objects.ImageMeta',
) -> None:
"""Add a vTPM device to the guest, if requested."""
# Enable virtual tpm support if required in the flavor or image.
vtpm_config = hardware.get_vtpm_constraint(flavor, image_meta)
if not vtpm_config:
return None
vtpm_secret_uuid = instance.system_metadata.get('vtpm_secret_uuid')
if not vtpm_secret_uuid:
raise exception.Invalid(
'Refusing to create an emulated TPM with no secret!')
vtpm = vconfig.LibvirtConfigGuestVTPM(vtpm_config, vtpm_secret_uuid)
guest.add_device(vtpm)
def _set_qemu_guest_agent(self, guest, flavor, instance, image_meta):
# Enable qga only if the 'hw_qemu_guest_agent' is equal to yes
if image_meta.properties.get('hw_qemu_guest_agent', False):
# a virtio-serial controller is required for qga. If it is not
# created explicitly, libvirt will do it by itself. But in case
# of AMD SEV, any virtio device should use iommu driver, and
# libvirt does not know about it. That is why the controller
# should be created manually.
if self._sev_enabled(flavor, image_meta):
self._add_virtio_serial_controller(guest, instance)
LOG.debug("Qemu guest agent is enabled through image "
"metadata", instance=instance)
self._add_qga_device(guest, instance)
def _get_guest_memory_backing_config(
self, inst_topology, numatune, flavor, image_meta):
wantsrealtime = hardware.is_realtime_enabled(flavor)
if (
wantsrealtime and
hardware.get_emulator_thread_policy_constraint(flavor) ==
fields.CPUEmulatorThreadsPolicy.SHARE and
not CONF.compute.cpu_shared_set
):
# NOTE(stephenfin) Yes, it's horrible that we're doing this here,
# but the shared policy unfortunately has different behavior
# depending on whether the '[compute] cpu_shared_set' is configured
# or not and we need it to be configured. Also note that we have
# already handled other conditions, such as no emulator thread
# policy being configured whatsoever, at the API level.
LOG.warning(
'Instance is requesting real-time CPUs with pooled '
'emulator threads, but a shared CPU pool has not been '
'configured on this host.'
)
raise exception.RealtimeMaskNotFoundOrInvalid()
wantsmempages = False
if inst_topology:
for cell in inst_topology.cells:
if cell.pagesize:
wantsmempages = True
break
wantsfilebacked = CONF.libvirt.file_backed_memory > 0
if wantsmempages and wantsfilebacked:
# Can't use file-backed memory with hugepages
LOG.warning("Instance requested huge pages, but file-backed "
"memory is enabled, and incompatible with huge pages")
raise exception.MemoryPagesUnsupported()
membacking = None
if wantsmempages:
pages = self._get_memory_backing_hugepages_support(
inst_topology, numatune)
if pages:
membacking = vconfig.LibvirtConfigGuestMemoryBacking()
membacking.hugepages = pages
if wantsrealtime:
if not membacking:
membacking = vconfig.LibvirtConfigGuestMemoryBacking()
membacking.locked = True
membacking.sharedpages = False
if wantsfilebacked:
if not membacking:
membacking = vconfig.LibvirtConfigGuestMemoryBacking()
membacking.filesource = True
membacking.sharedaccess = True
membacking.allocateimmediate = True
membacking.discard = True
if self._sev_enabled(flavor, image_meta):
if not membacking:
membacking = vconfig.LibvirtConfigGuestMemoryBacking()
membacking.locked = True
if hardware.get_locked_memory_constraint(flavor, image_meta):
if not membacking:
membacking = vconfig.LibvirtConfigGuestMemoryBacking()
membacking.locked = True
return membacking
def _get_memory_backing_hugepages_support(self, inst_topology, numatune):
if not self._has_numa_support():
# We should not get here, since we should have avoided
# reporting NUMA topology from _get_host_numa_topology
# in the first place. Just in case of a scheduler
# mess up though, raise an exception
raise exception.MemoryPagesUnsupported()
host_topology = self._get_host_numa_topology()
if host_topology is None:
# As above, we should not get here but just in case...
raise exception.MemoryPagesUnsupported()
# Currently libvirt does not support the smallest
# pagesize set as a backend memory.
# https://bugzilla.redhat.com/show_bug.cgi?id=1173507
avail_pagesize = [page.size_kb
for page in host_topology.cells[0].mempages]
avail_pagesize.sort()
smallest = avail_pagesize[0]
pages = []
for guest_cellid, inst_cell in enumerate(inst_topology.cells):
if inst_cell.pagesize and inst_cell.pagesize > smallest:
for memnode in numatune.memnodes:
if guest_cellid == memnode.cellid:
page = (
vconfig.LibvirtConfigGuestMemoryBackingPage())
page.nodeset = [guest_cellid]
page.size_kb = inst_cell.pagesize
pages.append(page)
break # Quit early...
return pages
def _get_flavor(self, ctxt, instance, flavor):
if flavor is not None:
return flavor
return instance.flavor
def _check_secure_boot_support(
self,
arch: str,
machine_type: str,
firmware_type: str,
) -> bool:
if not self._host.supports_secure_boot:
# secure boot requires host configuration
return False
if firmware_type != fields.FirmwareType.UEFI:
# secure boot is only supported with UEFI
return False
if (
arch == fields.Architecture.X86_64 and
'q35' not in machine_type
):
# secure boot on x86_64 requires the Q35 machine type
return False
return True
def _get_supported_perf_events(self):
if not len(CONF.libvirt.enabled_perf_events):
return []
supported_events = []
for event in CONF.libvirt.enabled_perf_events:
libvirt_perf_event_name = LIBVIRT_PERF_EVENT_PREFIX + event.upper()
if not hasattr(libvirt, libvirt_perf_event_name):
LOG.warning("Libvirt does not support event type '%s'.", event)
continue
if event in ('cmt', 'mbml', 'mbmt'):
LOG.warning(
"Monitoring of Intel CMT `perf` event(s) '%s' is not "
"supported by recent Linux kernels; ignoring.",
event,
)
continue
supported_events.append(event)
return supported_events
def _configure_guest_by_virt_type(
self,
guest: vconfig.LibvirtConfigGuest,
instance: 'objects.Instance',
image_meta: 'objects.ImageMeta',
flavor: 'objects.Flavor',
) -> None:
if CONF.libvirt.virt_type in ("kvm", "qemu"):
caps = self._host.get_capabilities()
host_arch = caps.host.cpu.arch
arch = self._check_emulation_arch(image_meta)
guest.os_arch = self._check_emulation_arch(image_meta)
if arch != host_arch:
# If emulating, downgrade to qemu
guest.virt_type = "qemu"
if arch in (fields.Architecture.I686, fields.Architecture.X86_64):
guest.sysinfo = self._get_guest_config_sysinfo(instance)
guest.os_smbios = vconfig.LibvirtConfigGuestSMBIOS()
mach_type = libvirt_utils.get_machine_type(image_meta)
self._host._check_machine_type(caps, mach_type)
guest.os_mach_type = mach_type
hw_firmware_type = image_meta.properties.get('hw_firmware_type')
hw_firmware_stateless = hardware.get_stateless_firmware_constraint(
image_meta)
if arch == fields.Architecture.AARCH64:
if not hw_firmware_type:
hw_firmware_type = fields.FirmwareType.UEFI
if hw_firmware_type == fields.FirmwareType.UEFI:
global uefi_logged
if not uefi_logged:
LOG.warning("uefi support is without some kind of "
"functional testing and therefore "
"considered experimental.")
uefi_logged = True
if not self._host.supports_uefi:
raise exception.UEFINotSupported()
# TODO(stephenfin): Drop this when we drop support for legacy
# architectures
if not mach_type:
# loaders are specific to arch and machine type - if we
# don't have a machine type here, we're on a legacy
# architecture that we have no default machine type for
raise exception.UEFINotSupported()
os_secure_boot = hardware.get_secure_boot_constraint(
flavor, image_meta)
if os_secure_boot == 'required':
# hard fail if we don't support secure boot and it's
# required
if not self._check_secure_boot_support(
arch, mach_type, hw_firmware_type,
):
raise exception.SecureBootNotSupported()
guest.os_loader_secure = True
elif os_secure_boot == 'optional':
# only enable it if the host is configured appropriately
guest.os_loader_secure = self._check_secure_boot_support(
arch, mach_type, hw_firmware_type,
)
else:
guest.os_loader_secure = False
try:
loader, nvram_template, requires_smm = (
self._host.get_loader(
arch, mach_type,
has_secure_boot=guest.os_loader_secure))
except exception.UEFINotSupported as exc:
if guest.os_loader_secure:
# we raise a specific exception if we requested secure
# boot and couldn't get that
raise exception.SecureBootNotSupported() from exc
raise
guest.os_loader = loader
guest.os_loader_type = 'pflash'
if hw_firmware_stateless:
guest.os_loader_stateless = True
else:
guest.os_nvram_template = nvram_template
# if the feature set says we need SMM then enable it
if requires_smm:
guest.features.append(
vconfig.LibvirtConfigGuestFeatureSMM())
# NOTE(lyarwood): If the machine type isn't recorded in the stashed
# image metadata then record it through the system metadata table.
# This will allow the host configuration to change in the future
# without impacting existing instances.
# NOTE(lyarwood): The value of ``hw_machine_type`` within the
# stashed image metadata of the instance actually comes from the
# system metadata table under the ``image_hw_machine_type`` key via
# nova.objects.ImageMeta.from_instance and the
# nova.utils.get_image_from_system_metadata function.
if image_meta.properties.get('hw_machine_type') is None:
instance.system_metadata['image_hw_machine_type'] = mach_type
if image_meta.properties.get('hw_boot_menu') is None:
guest.os_bootmenu = strutils.bool_from_string(
flavor.extra_specs.get('hw:boot_menu', 'no'))
else:
guest.os_bootmenu = image_meta.properties.hw_boot_menu
elif CONF.libvirt.virt_type == "lxc":
guest.os_init_path = "/sbin/init"
guest.os_cmdline = CONSOLE
guest.os_init_env["product_name"] = "OpenStack Nova"
elif CONF.libvirt.virt_type == "parallels":
if guest.os_type == fields.VMMode.EXE:
guest.os_init_path = "/sbin/init"
return None
def _conf_non_lxc(
self,
guest: vconfig.LibvirtConfigGuest,
root_device_name: str,
rescue: bool,
instance: 'objects.Instance',
inst_path: str,
image_meta: 'objects.ImageMeta',
disk_info: ty.Dict[str, ty.Any],
):
if rescue:
self._set_guest_for_rescue(
rescue, guest, inst_path, root_device_name)
elif instance.kernel_id:
self._set_guest_for_inst_kernel(
instance, guest, inst_path, root_device_name, image_meta)
else:
guest.os_boot_dev = blockinfo.get_boot_order(disk_info)
def _create_consoles(self, guest_cfg, instance, flavor, image_meta):
# NOTE(markus_z): Beware! Below are so many conditionals that it is
# easy to lose track. Use this chart to figure out your case:
#
# case | is serial | is qemu | resulting
# | enabled? | or kvm? | devices
# -------------------------------------------
# 1 | no | no | pty*
# 2 | no | yes | pty with logd
# 3 | yes | no | see case 1
# 4 | yes | yes | tcp with logd
#
# * exception: `virt_type=parallels` doesn't create a device
if CONF.libvirt.virt_type == 'parallels':
pass
elif CONF.libvirt.virt_type == 'lxc':
log_path = self._get_console_log_path(instance)
self._create_pty_device(
guest_cfg, vconfig.LibvirtConfigGuestConsole,
log_path=log_path)
else: # qemu, kvm
if self._is_s390x_guest(image_meta):
self._create_consoles_s390x(
guest_cfg, instance, flavor, image_meta)
else:
self._create_consoles_qemu_kvm(
guest_cfg, instance, flavor, image_meta)
def _is_mipsel_guest(self, image_meta: 'objects.ImageMeta') -> bool:
archs = (fields.Architecture.MIPSEL, fields.Architecture.MIPS64EL)
return self._check_emulation_arch(image_meta) in archs
def _is_s390x_guest(self, image_meta: 'objects.ImageMeta') -> bool:
archs = (fields.Architecture.S390, fields.Architecture.S390X)
return self._check_emulation_arch(image_meta) in archs
def _is_ppc64_guest(self, image_meta: 'objects.ImageMeta') -> bool:
archs = (fields.Architecture.PPC64, fields.Architecture.PPC64LE)
return self._check_emulation_arch(image_meta) in archs
def _is_aarch64_guest(self, image_meta: 'objects.ImageMeta') -> bool:
arch = fields.Architecture.AARCH64
return self._check_emulation_arch(image_meta) == arch
def _is_x86_guest(self, image_meta: 'objects.ImageMeta') -> bool:
archs = (fields.Architecture.I686, fields.Architecture.X86_64)
return self._check_emulation_arch(image_meta) in archs
def _create_consoles_qemu_kvm(self, guest_cfg, instance, flavor,
image_meta):
char_dev_cls = vconfig.LibvirtConfigGuestSerial
log_path = self._get_console_log_path(instance)
if CONF.serial_console.enabled:
if not self._serial_ports_already_defined(instance):
num_ports = hardware.get_number_of_serial_ports(flavor,
image_meta)
self._check_number_of_serial_console(num_ports)
self._create_serial_consoles(guest_cfg, num_ports,
char_dev_cls, log_path)
else:
self._create_pty_device(guest_cfg, char_dev_cls,
log_path=log_path)
def _create_consoles_s390x(self, guest_cfg, instance, flavor, image_meta):
char_dev_cls = vconfig.LibvirtConfigGuestConsole
log_path = self._get_console_log_path(instance)
if CONF.serial_console.enabled:
if not self._serial_ports_already_defined(instance):
num_ports = hardware.get_number_of_serial_ports(flavor,
image_meta)
self._create_serial_consoles(guest_cfg, num_ports,
char_dev_cls, log_path)
else:
self._create_pty_device(guest_cfg, char_dev_cls,
"sclp", log_path)
def _create_pty_device(self, guest_cfg, char_dev_cls, target_type=None,
log_path=None):
consolepty = char_dev_cls()
consolepty.target_type = target_type
consolepty.type = "pty"
log = vconfig.LibvirtConfigGuestCharDeviceLog()
log.file = log_path
consolepty.log = log
guest_cfg.add_device(consolepty)
def _serial_ports_already_defined(self, instance):
try:
guest = self._host.get_guest(instance)
if list(self._get_serial_ports_from_guest(guest)):
# Serial port are already configured for instance that
# means we are in a context of migration.
return True
except exception.InstanceNotFound:
LOG.debug(
"Instance does not exist yet on libvirt, we can "
"safely pass on looking for already defined serial "
"ports in its domain XML", instance=instance)
return False
def _create_serial_consoles(self, guest_cfg, num_ports, char_dev_cls,
log_path):
for port in range(num_ports):
console = char_dev_cls()
console.port = port
console.type = "tcp"
console.listen_host = CONF.serial_console.proxyclient_address
listen_port = serial_console.acquire_port(console.listen_host)
console.listen_port = listen_port
# NOTE: only the first serial console gets the boot messages,
# that's why we attach the logd subdevice only to that.
if port == 0:
log = vconfig.LibvirtConfigGuestCharDeviceLog()
log.file = log_path
console.log = log
guest_cfg.add_device(console)
def _cpu_config_to_vcpu_model(self, cpu_config, vcpu_model):
"""Update VirtCPUModel object according to libvirt CPU config.
:param:cpu_config: vconfig.LibvirtConfigGuestCPU presenting the
instance's virtual cpu configuration.
:param:vcpu_model: VirtCPUModel object. A new object will be created
if None.
:return: Updated VirtCPUModel object, or None if cpu_config is None
"""
if not cpu_config:
return
if not vcpu_model:
vcpu_model = objects.VirtCPUModel()
vcpu_model.arch = cpu_config.arch
vcpu_model.vendor = cpu_config.vendor
vcpu_model.model = cpu_config.model
vcpu_model.mode = cpu_config.mode
vcpu_model.match = cpu_config.match
if cpu_config.sockets:
vcpu_model.topology = objects.VirtCPUTopology(
sockets=cpu_config.sockets,
cores=cpu_config.cores,
threads=cpu_config.threads)
else:
vcpu_model.topology = None
features = [objects.VirtCPUFeature(
name=f.name,
policy=f.policy) for f in cpu_config.features]
vcpu_model.features = features
return vcpu_model
def _vcpu_model_to_cpu_config(self, vcpu_model):
"""Create libvirt CPU config according to VirtCPUModel object.
:param:vcpu_model: VirtCPUModel object.
:return: vconfig.LibvirtConfigGuestCPU.
"""
cpu_config = vconfig.LibvirtConfigGuestCPU()
cpu_config.arch = vcpu_model.arch
cpu_config.model = vcpu_model.model
cpu_config.mode = vcpu_model.mode
cpu_config.match = vcpu_model.match
cpu_config.vendor = vcpu_model.vendor
if vcpu_model.topology:
cpu_config.sockets = vcpu_model.topology.sockets
cpu_config.cores = vcpu_model.topology.cores
cpu_config.threads = vcpu_model.topology.threads
if vcpu_model.features:
for f in vcpu_model.features:
xf = vconfig.LibvirtConfigGuestCPUFeature()
xf.name = f.name
xf.policy = f.policy
cpu_config.features.add(xf)
return cpu_config
def _guest_needs_usb(self, guest, image_meta):
"""Evaluate devices currently attached to the guest."""
if self._is_ppc64_guest(image_meta):
# PPC64 guests get a USB keyboard and mouse automatically
return True
for dev in guest.devices:
if isinstance(dev, vconfig.LibvirtConfigGuestDisk):
if dev.target_bus == 'usb':
return True
if isinstance(dev, vconfig.LibvirtConfigGuestInput):
if dev.bus == 'usb':
return True
return False
def _guest_add_usb_root_controller(self, guest, image_meta):
"""Add USB root controller, if necessary.
Note that these are added by default on x86-64. We add the controller
here explicitly so that we can _disable_ it (by setting the model to
'none') if it's not necessary.
"""
usbhost = vconfig.LibvirtConfigGuestUSBHostController()
usbhost.index = 0
# an unset model means autodetect, while 'none' means don't add a
# controller (x86 gets one by default)
usbhost.model = None
if not self._guest_needs_usb(guest, image_meta):
archs = (
fields.Architecture.PPC,
fields.Architecture.PPC64,
fields.Architecture.PPC64LE,
)
if self._check_emulation_arch(image_meta) in archs:
# NOTE(chateaulav): during actual testing and implementation
# it wanted None for ppc, as this removes it from the domain
# xml, where 'none' adds it but then disables it causing
# libvirt errors and the instances not being able to build
usbhost.model = None
else:
usbhost.model = 'none'
guest.add_device(usbhost)
def _guest_add_pcie_root_ports(self, guest):
"""Add PCI Express root ports.
PCI Express machine can have as many PCIe devices as it has
pcie-root-port controllers (slots in virtual motherboard).
If we want to have more PCIe slots for hotplug then we need to create
whole PCIe structure (libvirt limitation).
"""
pcieroot = vconfig.LibvirtConfigGuestPCIeRootController()
guest.add_device(pcieroot)
for x in range(0, CONF.libvirt.num_pcie_ports):
pcierootport = vconfig.LibvirtConfigGuestPCIeRootPortController()
guest.add_device(pcierootport)
def _guest_needs_pcie(self, guest):
"""Check for prerequisites for adding PCIe root port
controllers
"""
caps = self._host.get_capabilities()
# Add PCIe root port controllers for PCI Express machines
# but only if their amount is configured
if not CONF.libvirt.num_pcie_ports:
return False
# Only certain architectures and machine types can handle PCIe ports;
# the latter will be handled by libvirt.utils.get_machine_type
if (
caps.host.cpu.arch == fields.Architecture.AARCH64 and
guest.os_mach_type.startswith('virt')
):
return True
if (
caps.host.cpu.arch == fields.Architecture.X86_64 and
guest.os_mach_type is not None and
'q35' in guest.os_mach_type
):
return True
return False
def _get_guest_config(self, instance, network_info, image_meta,
disk_info, rescue=None, block_device_info=None,
context=None, mdevs=None, accel_info=None,
share_info=None):
"""Get config data for parameters.
:param rescue: optional dictionary that should contain the key
'ramdisk_id' if a ramdisk is needed for the rescue image and
'kernel_id' if a kernel is needed for the rescue image.
:param mdevs: optional list of mediated devices to assign to the guest.
:param accel_info: optional list of accelerator requests (ARQs)
:param share_info: optional list of share_mapping
"""
flavor = instance.flavor
inst_path = libvirt_utils.get_instance_path(instance)
disk_mapping = disk_info['mapping']
vpmems = self._get_ordered_vpmems(instance, flavor)
guest = vconfig.LibvirtConfigGuest()
guest.virt_type = CONF.libvirt.virt_type
guest.name = instance.name
guest.uuid = instance.uuid
# We are using default unit for memory: KiB
guest.memory = flavor.memory_mb * units.Ki
guest.vcpus = flavor.vcpus
guest_numa_config = self._get_guest_numa_config(
instance.numa_topology, flavor, image_meta)
guest.cpuset = guest_numa_config.cpuset
guest.cputune = guest_numa_config.cputune
guest.numatune = guest_numa_config.numatune
guest.membacking = self._get_guest_memory_backing_config(
instance.numa_topology,
guest_numa_config.numatune,
flavor, image_meta)
guest.metadata.append(
self._get_guest_config_meta(
self.get_instance_driver_metadata(
instance, network_info)))
guest.idmaps = self._get_guest_idmaps()
for event in self._supported_perf_events:
guest.add_perf_event(event)
self._update_guest_cputune(guest, flavor)
guest.cpu = self._get_guest_cpu_config(
flavor, image_meta, guest_numa_config.numaconfig,
instance.numa_topology)
# Notes(yjiang5): we always sync the instance's vcpu model with
# the corresponding config file.
instance.vcpu_model = self._cpu_config_to_vcpu_model(
guest.cpu, instance.vcpu_model)
if 'root' in disk_mapping:
root_device_name = block_device.prepend_dev(
disk_mapping['root']['dev'])
else:
root_device_name = None
guest.os_type = (
fields.VMMode.get_from_instance(instance) or
self._get_guest_os_type()
)
sev_enabled = self._sev_enabled(flavor, image_meta)
self._configure_guest_by_virt_type(guest, instance, image_meta, flavor)
if CONF.libvirt.virt_type != 'lxc':
self._conf_non_lxc(
guest, root_device_name, rescue, instance, inst_path,
image_meta, disk_info)
self._set_features(guest, instance.os_type, image_meta, flavor)
self._set_clock(guest, instance.os_type, image_meta)
storage_configs = self._get_guest_storage_config(context,
instance, image_meta, disk_info, rescue, block_device_info,
flavor, guest.os_type)
for config in storage_configs:
guest.add_device(config)
for vif in network_info:
config = self.vif_driver.get_config(
instance, vif, image_meta, flavor, CONF.libvirt.virt_type,
)
guest.add_device(config)
self._create_consoles(guest, instance, flavor, image_meta)
self._guest_add_spice_channel(guest)
if self._guest_add_video_device(guest):
self._add_video_driver(guest, image_meta, flavor)
self._guest_add_pointer_device(guest, image_meta)
self._guest_add_keyboard_device(guest, image_meta)
# Some features are only supported 'qemu' and 'kvm' hypervisor
if CONF.libvirt.virt_type in ('qemu', 'kvm'):
self._set_qemu_guest_agent(guest, flavor, instance, image_meta)
self._add_rng_device(guest, flavor, image_meta)
self._add_vtpm_device(guest, flavor, instance, image_meta)
if self._guest_needs_pcie(guest):
self._guest_add_pcie_root_ports(guest)
self._guest_add_usb_root_controller(guest, image_meta)
self._guest_add_pci_devices(guest, instance)
pci_arq_list = []
if accel_info:
# NOTE(Sundar): We handle only the case where all attach handles
# are of type 'PCI'. The Cyborg fake driver used for testing
# returns attach handles of type 'TEST_PCI' and so its ARQs will
# not get composed into the VM's domain XML. For now, we do not
# expect a mixture of different attach handles for the same
# instance; but that case also gets ignored by this logic.
ah_types_set = {arq['attach_handle_type'] for arq in accel_info}
supported_types_set = {'PCI'}
if ah_types_set == supported_types_set:
pci_arq_list = accel_info
else:
LOG.info('Ignoring accelerator requests for instance %s. '
'Supported Attach handle types: %s. '
'But got these unsupported types: %s.',
instance.uuid, supported_types_set,
ah_types_set.difference(supported_types_set))
self._guest_add_accel_pci_devices(guest, pci_arq_list)
self._guest_add_virtiofs_for_share(guest, instance, share_info)
self._guest_add_watchdog_action(guest, flavor, image_meta)
self._guest_add_memory_balloon(guest)
if mdevs:
self._guest_add_mdevs(guest, mdevs)
if sev_enabled:
caps = self._host.get_capabilities()
self._guest_configure_sev(guest, caps.host.cpu.arch,
guest.os_mach_type)
if vpmems:
self._guest_add_vpmems(guest, vpmems)
self._guest_add_iommu_device(guest, image_meta, flavor)
return guest
def _get_ordered_vpmems(self, instance, flavor):
resources = self._get_resources(instance)
ordered_vpmem_resources = self._get_ordered_vpmem_resources(
resources, flavor)
ordered_vpmems = [self._vpmems_by_name[resource.identifier]
for resource in ordered_vpmem_resources]
return ordered_vpmems
def _get_vpmems(self, instance, prefix=None):
resources = self._get_resources(instance, prefix=prefix)
vpmem_resources = self._get_vpmem_resources(resources)
vpmems = [self._vpmems_by_name[resource.identifier]
for resource in vpmem_resources]
return vpmems
def _guest_add_vpmems(self, guest, vpmems):
guest.max_memory_size = guest.memory
guest.max_memory_slots = 0
for vpmem in vpmems:
size_kb = vpmem.size // units.Ki
align_kb = vpmem.align // units.Ki
vpmem_config = vconfig.LibvirtConfigGuestVPMEM()
vpmem_config.source_path = vpmem.devpath
vpmem_config.target_size = size_kb
vpmem_config.align_size = align_kb
# max memory size needs contain vpmem size
guest.max_memory_size += size_kb
# one vpmem will occupy one memory slot
guest.max_memory_slots += 1
guest.add_device(vpmem_config)
def _sev_enabled(self, flavor, image_meta):
"""To enable AMD SEV, the following should be true:
a) the supports_amd_sev instance variable in the host is
true,
b) the instance extra specs and/or image properties request
memory encryption to be enabled, and
c) there are no conflicts between extra specs, image properties
and machine type selection.
Most potential conflicts in c) should already be caught in the
API layer. However there is still one remaining case which
needs to be handled here: when the image does not contain an
hw_machine_type property, the machine type will be chosen from
CONF.libvirt.hw_machine_type if configured, otherwise falling
back to the hardcoded value which is currently 'pc'. If it
ends up being 'pc' or another value not in the q35 family, we
need to raise an exception. So calculate the machine type and
pass it to be checked alongside the other sanity checks which
are run while determining whether SEV is selected.
"""
if not self._host.supports_amd_sev:
return False
mach_type = libvirt_utils.get_machine_type(image_meta)
return hardware.get_mem_encryption_constraint(flavor, image_meta,
mach_type)
def _guest_configure_sev(self, guest, arch, mach_type):
sev = self._find_sev_feature(arch, mach_type)
if sev is None:
# In theory this should never happen because it should
# only get called if SEV was requested, in which case the
# guest should only get scheduled on this host if it
# supports SEV, and SEV support is dependent on the
# presence of this <sev> feature. That said, it's
# conceivable that something could get messed up along the
# way, e.g. a mismatch in the choice of machine type. So
# make sure that if it ever does happen, we at least get a
# helpful error rather than something cryptic like
# "AttributeError: 'NoneType' object has no attribute 'cbitpos'
raise exception.MissingDomainCapabilityFeatureException(
feature='sev')
designer.set_driver_iommu_for_all_devices(guest)
self._guest_add_launch_security(guest, sev)
def _guest_add_launch_security(self, guest, sev):
launch_security = vconfig.LibvirtConfigGuestSEVLaunchSecurity()
launch_security.cbitpos = sev.cbitpos
launch_security.reduced_phys_bits = sev.reduced_phys_bits
guest.launch_security = launch_security
def _find_sev_feature(self, arch, mach_type):
"""Search domain capabilities for the given arch and machine type
for the <sev> element under <features>, and return it if found.
"""
domain_caps = self._host.get_domain_capabilities()
if arch not in domain_caps:
LOG.warning(
"Wanted to add SEV to config for guest with arch %(arch)s "
"but only had domain capabilities for: %(archs)s",
{'arch': arch, 'archs': ' '.join(domain_caps)})
return None
if mach_type not in domain_caps[arch]:
LOG.warning(
"Wanted to add SEV to config for guest with machine type "
"%(mtype)s but for arch %(arch)s only had domain capabilities "
"for machine types: %(mtypes)s",
{'mtype': mach_type, 'arch': arch,
'mtypes': ' '.join(domain_caps[arch])})
return None
for feature in domain_caps[arch][mach_type].features:
if feature.root_name == 'sev':
return feature
return None
def _guest_add_mdevs(self, guest, chosen_mdevs):
for chosen_mdev in chosen_mdevs:
mdev = vconfig.LibvirtConfigGuestHostdevMDEV()
mdev.uuid = chosen_mdev
guest.add_device(mdev)
@staticmethod
def _guest_add_spice_channel(guest):
if (
CONF.spice.enabled and CONF.spice.agent_enabled and
CONF.libvirt.virt_type != 'lxc'
):
channel = vconfig.LibvirtConfigGuestChannel()
channel.type = 'spicevmc'
channel.target_name = "com.redhat.spice.0"
guest.add_device(channel)
@staticmethod
def _guest_add_memory_balloon(guest):
# Memory balloon device only support 'qemu/kvm' hypervisor
if (
CONF.libvirt.virt_type in ('qemu', 'kvm') and
CONF.libvirt.mem_stats_period_seconds > 0
):
balloon = vconfig.LibvirtConfigMemoryBalloon()
balloon.model = 'virtio'
balloon.period = CONF.libvirt.mem_stats_period_seconds
guest.add_device(balloon)
@staticmethod
def _guest_add_watchdog_action(guest, flavor, image_meta):
# image meta takes precedence over flavor extra specs; disable the
# watchdog action by default
watchdog_action = (flavor.extra_specs.get('hw:watchdog_action') or
'disabled')
watchdog_action = image_meta.properties.get('hw_watchdog_action',
watchdog_action)
# NB(sross): currently only actually supported by KVM/QEmu
if watchdog_action != 'disabled':
if watchdog_action in fields.WatchdogAction.ALL:
bark = vconfig.LibvirtConfigGuestWatchdog()
bark.action = watchdog_action
guest.add_device(bark)
else:
raise exception.InvalidWatchdogAction(action=watchdog_action)
def _guest_add_pci_devices(self, guest, instance):
if CONF.libvirt.virt_type in ('qemu', 'kvm'):
# Get all generic PCI devices (non-SR-IOV).
for pci_dev in instance.get_pci_devices(
source=objects.InstancePCIRequest.FLAVOR_ALIAS
):
guest.add_device(self._get_guest_pci_device(pci_dev))
else:
# PCI devices is only supported for QEMU/KVM hypervisor
if instance.get_pci_devices():
raise exception.PciDeviceUnsupportedHypervisor(
type=CONF.libvirt.virt_type
)
def _guest_add_accel_pci_devices(self, guest, accel_info):
"""Add all accelerator PCI functions from ARQ list."""
for arq in accel_info:
dev = vconfig.LibvirtConfigGuestHostdevPCI()
pci_addr = arq['attach_handle_info']
dev.domain, dev.bus, dev.slot, dev.function = (
pci_addr['domain'], pci_addr['bus'],
pci_addr['device'], pci_addr['function'])
self._set_managed_mode(dev, "true")
guest.add_device(dev)
@staticmethod
def _guest_add_video_device(guest):
if CONF.libvirt.virt_type == 'lxc':
return False
# NB some versions of libvirt support both SPICE and VNC
# at the same time. We're not trying to second guess which
# those versions are. We'll just let libvirt report the
# errors appropriately if the user enables both.
add_video_driver = False
if CONF.vnc.enabled:
graphics = vconfig.LibvirtConfigGuestGraphics()
graphics.type = "vnc"
graphics.listen = CONF.vnc.server_listen
guest.add_device(graphics)
add_video_driver = True
if CONF.spice.enabled:
graphics = vconfig.LibvirtConfigGuestGraphics()
graphics.type = "spice"
graphics.listen = CONF.spice.server_listen
graphics.image_compression = CONF.spice.image_compression
graphics.jpeg_compression = CONF.spice.jpeg_compression
graphics.zlib_compression = CONF.spice.zlib_compression
graphics.playback_compression = CONF.spice.playback_compression
graphics.streaming_mode = CONF.spice.streaming_mode
graphics.secure = CONF.spice.require_secure
guest.add_device(graphics)
add_video_driver = True
return add_video_driver
def _get_pointer_bus_and_model(
self,
guest: vconfig.LibvirtConfigGuest,
image_meta: objects.ImageMeta,
) -> ty.Tuple[ty.Optional[str], ty.Optional[str]]:
pointer_bus = image_meta.properties.get('hw_input_bus')
pointer_model = image_meta.properties.get('hw_pointer_model')
if pointer_bus:
pointer_model = 'tablet'
pointer_bus = pointer_bus
elif pointer_model or CONF.pointer_model == 'usbtablet':
# Handle the legacy 'hw_pointer_model' image metadata property
pointer_model = 'tablet'
pointer_bus = 'usb'
else:
# If the user hasn't requested anything and the host config says to
# use something other than a USB tablet, there's nothing to do
return None, None
# For backward compatibility, we don't want to error out if the host
# configuration requests a USB tablet but the virtual machine mode is
# not configured as HVM.
if guest.os_type != fields.VMMode.HVM:
LOG.warning(
'USB tablet requested for guests on non-HVM host; '
'in order to accept this request the machine mode should '
'be configured as HVM.')
return None, None
# Ditto for using a USB tablet when the SPICE agent is enabled, since
# that has a paravirt mouse builtin which drastically reduces overhead;
# this only applies if VNC is not also enabled though, since that still
# needs the device
if (
CONF.spice.enabled and CONF.spice.agent_enabled and
not CONF.vnc.enabled
):
LOG.warning(
'USB tablet requested for guests but the SPICE agent is '
'enabled; ignoring request in favour of default '
'configuration.')
return None, None
return pointer_model, pointer_bus
def _guest_add_pointer_device(
self,
guest: vconfig.LibvirtConfigGuest,
image_meta: objects.ImageMeta
) -> None:
"""Build the pointer device to add to the instance.
The configuration is determined by examining the 'hw_input_bus' image
metadata property, the 'hw_pointer_model' image metadata property, and
the '[DEFAULT] pointer_model' config option in that order.
"""
pointer_model, pointer_bus = self._get_pointer_bus_and_model(
guest, image_meta)
if pointer_model and pointer_bus:
pointer = vconfig.LibvirtConfigGuestInput()
pointer.type = pointer_model
pointer.bus = pointer_bus
guest.add_device(pointer)
# returned for unit testing purposes
return pointer
def _guest_add_keyboard_device(self, guest, image_meta):
"""Add keyboard for graphical console use."""
bus = image_meta.properties.get('hw_input_bus')
if not bus:
# AArch64 doesn't provide a default keyboard so we explicitly add
# one; for everything else we rely on default (e.g. for x86,
# libvirt will automatically add a PS2 keyboard)
# TODO(stephenfin): We might want to do this for other non-x86
# architectures
arch = self._check_emulation_arch(image_meta)
if arch != fields.Architecture.AARCH64:
return None
bus = 'usb'
keyboard = vconfig.LibvirtConfigGuestInput()
keyboard.type = 'keyboard'
keyboard.bus = bus
guest.add_device(keyboard)
# returned for unit testing purposes
return keyboard
def _get_iommu_model(
self,
guest: vconfig.LibvirtConfigGuest,
image_meta: 'objects.ImageMeta',
flavor: 'objects.Flavor',
) -> ty.Optional[str]:
model = flavor.extra_specs.get(
'hw:viommu_model') or image_meta.properties.get(
'hw_viommu_model')
if not model:
return None
is_x86 = self._is_x86_guest(image_meta)
is_aarch64 = self._is_aarch64_guest(image_meta)
if is_x86:
if guest.os_mach_type is not None and not (
'q35' in guest.os_mach_type
):
arch = self._check_emulation_arch(image_meta)
mtype = guest.os_mach_type if (
guest.os_mach_type is not None
) else "unknown"
raise exception.InvalidVIOMMUMachineType(
mtype=mtype, arch=arch)
elif is_aarch64:
if guest.os_mach_type is not None and not (
'virt' in guest.os_mach_type
):
arch = self._check_emulation_arch(image_meta)
mtype = guest.os_mach_type if (
guest.os_mach_type is not None
) else "unknown"
raise exception.InvalidVIOMMUMachineType(
mtype=mtype, arch=arch)
else:
raise exception.InvalidVIOMMUArchitecture(
arch=self._check_emulation_arch(image_meta))
if model == fields.VIOMMUModel.AUTO:
if self._host.has_min_version(MIN_LIBVIRT_VIOMMU_VIRTIO_MODEL):
model = fields.VIOMMUModel.VIRTIO
elif self._is_x86_guest(image_meta) and (
guest.os_mach_type is not None and 'q35' in guest.os_mach_type
):
model = fields.VIOMMUModel.INTEL
else:
# AArch64
model = fields.VIOMMUModel.SMMUV3
return model
def _guest_add_iommu_device(
self,
guest: vconfig.LibvirtConfigGuest,
image_meta: 'objects.ImageMeta',
flavor: 'objects.Flavor',
) -> None:
"""Add a virtual IOMMU device to allow e.g. vfio-pci usage."""
if CONF.libvirt.virt_type not in ('qemu', 'kvm'):
# vIOMMU requires QEMU
return
iommu = vconfig.LibvirtConfigGuestIOMMU()
iommu.model = self._get_iommu_model(guest, image_meta, flavor)
if iommu.model is None:
return
iommu.interrupt_remapping = True
iommu.caching_mode = True
iommu.iotlb = True
# As Qemu supported values are 39 and 48, we set this to
# larger width (48) by default and will not exposed to end user.
iommu.aw_bits = 48
if guest.os_mach_type is not None and 'q35' in guest.os_mach_type:
iommu.eim = True
else:
iommu.eim = False
guest.add_device(iommu)
ioapic = vconfig.LibvirtConfigGuestFeatureIOAPIC()
guest.add_feature(ioapic)
def _get_guest_xml(self, context, instance, network_info, disk_info,
image_meta, rescue=None,
block_device_info=None,
mdevs=None, accel_info=None,
share_info=None):
# NOTE(danms): Stringifying a NetworkInfo will take a lock. Do
# this ahead of time so that we don't acquire it while also
# holding the logging lock.
network_info_str = str(network_info)
msg = ('Start _get_guest_xml '
'network_info=%(network_info)s '
'disk_info=%(disk_info)s '
'image_meta=%(image_meta)s rescue=%(rescue)s '
'block_device_info=%(block_device_info)s'
'share_info=%(share_info)s' %
{'network_info': network_info_str, 'disk_info': disk_info,
'image_meta': image_meta, 'rescue': rescue,
'block_device_info': block_device_info,
'share_info': share_info, })
# NOTE(mriedem): block_device_info can contain auth_password so we
# need to sanitize the password in the message.
LOG.debug(strutils.mask_password(msg), instance=instance)
conf = self._get_guest_config(instance, network_info, image_meta,
disk_info, rescue, block_device_info,
context, mdevs, accel_info, share_info)
xml = conf.to_xml()
LOG.debug('End _get_guest_xml xml=%(xml)s',
{'xml': xml}, instance=instance)
return xml
def get_info(self, instance, use_cache=True):
"""Retrieve information from libvirt for a specific instance.
If a libvirt error is encountered during lookup, we might raise a
NotFound exception or Error exception depending on how severe the
libvirt error is.
:param instance: nova.objects.instance.Instance object
:param use_cache: unused in this driver
:returns: An InstanceInfo object
"""
guest = self._host.get_guest(instance)
# Kind of ugly but we need to pass host to get_info as for a
# workaround, see libvirt/compat.py
return guest.get_info(self._host)
def _create_domain_setup_lxc(self, context, instance, image_meta,
block_device_info):
inst_path = libvirt_utils.get_instance_path(instance)
block_device_mapping = driver.block_device_info_get_mapping(
block_device_info)
root_disk = block_device.get_root_bdm(block_device_mapping)
if root_disk:
self._connect_volume(context, root_disk['connection_info'],
instance)
disk_path = root_disk['connection_info']['data']['device_path']
# NOTE(apmelton) - Even though the instance is being booted from a
# cinder volume, it is still presented as a local block device.
# LocalBlockImage is used here to indicate that the instance's
# disk is backed by a local block device.
image_model = imgmodel.LocalBlockImage(disk_path)
else:
root_disk = self.image_backend.by_name(instance, 'disk')
image_model = root_disk.get_model(self._conn)
container_dir = os.path.join(inst_path, 'rootfs')
fileutils.ensure_tree(container_dir)
rootfs_dev = disk_api.setup_container(image_model,
container_dir=container_dir)
try:
# Save rootfs device to disconnect it when deleting the instance
if rootfs_dev:
instance.system_metadata['rootfs_device_name'] = rootfs_dev
if CONF.libvirt.uid_maps or CONF.libvirt.gid_maps:
id_maps = self._get_guest_idmaps()
libvirt_utils.chown_for_id_maps(container_dir, id_maps)
except Exception:
with excutils.save_and_reraise_exception():
self._create_domain_cleanup_lxc(instance)
def _create_domain_cleanup_lxc(self, instance):
inst_path = libvirt_utils.get_instance_path(instance)
container_dir = os.path.join(inst_path, 'rootfs')
try:
state = self.get_info(instance).state
except exception.InstanceNotFound:
# The domain may not be present if the instance failed to start
state = None
if state == power_state.RUNNING:
# NOTE(uni): Now the container is running with its own private
# mount namespace and so there is no need to keep the container
# rootfs mounted in the host namespace
LOG.debug('Attempting to unmount container filesystem: %s',
container_dir, instance=instance)
disk_api.clean_lxc_namespace(container_dir=container_dir)
else:
disk_api.teardown_container(container_dir=container_dir)
@contextlib.contextmanager
def _lxc_disk_handler(self, context, instance, image_meta,
block_device_info):
"""Context manager to handle the pre and post instance boot,
LXC specific disk operations.
An image or a volume path will be prepared and setup to be
used by the container, prior to starting it.
The disk will be disconnected and unmounted if a container has
failed to start.
"""
if CONF.libvirt.virt_type != 'lxc':
yield
return
self._create_domain_setup_lxc(context, instance, image_meta,
block_device_info)
try:
yield
finally:
self._create_domain_cleanup_lxc(instance)
def _create_guest(
self,
context: nova_context.RequestContext,
xml: str,
instance: 'objects.Instance',
power_on: bool = True,
pause: bool = False,
post_xml_callback: ty.Optional[ty.Callable] = None,
) -> libvirt_guest.Guest:
"""Create a Guest from XML.
Create a Guest, which in turn creates a libvirt domain, from XML,
optionally starting it after creation.
:returns guest.Guest: Created guest.
"""
libvirt_secret = None
# determine whether vTPM is in use and, if so, create the secret
if CONF.libvirt.swtpm_enabled and hardware.get_vtpm_constraint(
instance.flavor, instance.image_meta,
):
secret_uuid, passphrase = crypto.ensure_vtpm_secret(
context, instance)
libvirt_secret = self._host.create_secret(
'vtpm', instance.uuid, password=passphrase,
uuid=secret_uuid)
try:
guest = libvirt_guest.Guest.create(xml, self._host)
if post_xml_callback is not None:
post_xml_callback()
if power_on or pause:
self.cpu_api.power_up_for_instance(instance)
guest.launch(pause=pause)
return guest
finally:
if libvirt_secret is not None:
libvirt_secret.undefine()
def _neutron_failed_callback(self, event_name, instance):
LOG.error('Neutron Reported failure on event '
'%(event)s for instance %(uuid)s',
{'event': event_name, 'uuid': instance.uuid},
instance=instance)
if CONF.vif_plugging_is_fatal:
raise exception.VirtualInterfaceCreateException()
def _get_neutron_events(self, network_info):
# NOTE(danms): We need to collect any VIFs that are currently
# down that we expect a down->up event for. Anything that is
# already up will not undergo that transition, and for
# anything that might be stale (cache-wise) assume it's
# already up so we don't block on it.
return [('network-vif-plugged', vif['id'])
for vif in network_info if vif.get('active', True) is False and
vif['vnic_type'] != network_model.VNIC_TYPE_REMOTE_MANAGED]
def _create_guest_with_network(
self,
context: nova_context.RequestContext,
xml: str,
instance: 'objects.Instance',
network_info: network_model.NetworkInfo,
block_device_info: ty.Optional[ty.Dict[str, ty.Any]],
power_on: bool = True,
vifs_already_plugged: bool = False,
post_xml_callback: ty.Optional[ty.Callable] = None,
external_events: ty.Optional[ty.List[ty.Tuple[str, str]]] = None,
cleanup_instance_dir: bool = False,
cleanup_instance_disks: bool = False,
) -> libvirt_guest.Guest:
"""Do required network setup and create domain."""
timeout = CONF.vif_plugging_timeout
if (
CONF.libvirt.virt_type in ('kvm', 'qemu') and
not vifs_already_plugged and power_on and timeout
):
events = (external_events if external_events
else self._get_neutron_events(network_info))
else:
events = []
pause = bool(events)
try:
with self.virtapi.wait_for_instance_event(
instance, events, deadline=timeout,
error_callback=self._neutron_failed_callback,
):
self.plug_vifs(instance, network_info)
with self._lxc_disk_handler(
context, instance, instance.image_meta, block_device_info,
):
guest = self._create_guest(
context, xml, instance,
pause=pause, power_on=power_on,
post_xml_callback=post_xml_callback)
except eventlet.timeout.Timeout:
# We did not receive all expected events from Neutron, a warning
# has already been logged by wait_for_instance_event, but we need
# to decide if the issue is fatal.
if CONF.vif_plugging_is_fatal:
# NOTE(stephenfin): don't worry, guest will be in scope since
# we can only hit this branch if the VIF plug timed out
if guest.is_active():
guest.poweroff()
self._cleanup(
context, instance, network_info, block_device_info,
destroy_vifs=True,
cleanup_instance_dir=cleanup_instance_dir,
cleanup_instance_disks=cleanup_instance_disks)
raise exception.VirtualInterfaceCreateException()
except Exception:
# Any other error, be sure to clean up
LOG.error('Failed to start libvirt guest', instance=instance)
with excutils.save_and_reraise_exception():
self._cleanup(
context, instance, network_info, block_device_info,
destroy_vifs=True,
cleanup_instance_dir=cleanup_instance_dir,
cleanup_instance_disks=cleanup_instance_disks)
# Resume only if domain has been paused
if pause:
guest.resume()
return guest
def _get_pcpu_available(self):
"""Get number of host cores to be used for PCPUs.
:returns: The number of host cores to be used for PCPUs.
"""
if not CONF.compute.cpu_dedicated_set:
return set()
if CONF.libvirt.cpu_power_management:
available_cpus = self._host.get_available_cpus()
else:
available_cpus = self._host.get_online_cpus()
dedicated_cpus = hardware.get_cpu_dedicated_set()
if not dedicated_cpus.issubset(available_cpus):
msg = _("Invalid '[compute] cpu_dedicated_set' config: one or "
"more of the configured CPUs is not available. Available "
"cpuset(s): %(available)s, configured cpuset(s): %(req)s")
raise exception.Invalid(msg % {
'available': sorted(available_cpus),
'req': sorted(dedicated_cpus)})
return dedicated_cpus
def _get_vcpu_available(self):
"""Get host cores to be used for VCPUs.
:returns: A list of host CPU cores that can be used for VCPUs.
"""
online_cpus = self._host.get_online_cpus()
# NOTE(stephenfin): The use of the legacy 'vcpu_pin_set' option happens
# if it's defined, regardless of whether '[compute] cpu_shared_set' is
# also configured. This is legacy behavior required for upgrades that
# should be removed in the future, when we can rely exclusively on
# '[compute] cpu_shared_set'.
if CONF.vcpu_pin_set:
# TODO(stephenfin): Remove this in U
shared_cpus = hardware.get_vcpu_pin_set()
elif CONF.compute.cpu_shared_set:
shared_cpus = hardware.get_cpu_shared_set()
elif CONF.compute.cpu_dedicated_set:
return set()
else:
return online_cpus
if not shared_cpus.issubset(online_cpus):
msg = _("Invalid '%(config_opt)s' config: one or "
"more of the configured CPUs is not online. Online "
"cpuset(s): %(online)s, configured cpuset(s): %(req)s")
if CONF.vcpu_pin_set:
config_opt = 'vcpu_pin_set'
else: # CONF.compute.cpu_shared_set
config_opt = '[compute] cpu_shared_set'
raise exception.Invalid(msg % {
'config_opt': config_opt,
'online': sorted(online_cpus),
'req': sorted(shared_cpus)})
return shared_cpus
@staticmethod
def _get_local_gb_info():
"""Get local storage info of the compute node in GB.
:returns: A dict containing:
:total: How big the overall usable filesystem is (in gigabytes)
:free: How much space is free (in gigabytes)
:used: How much space is used (in gigabytes)
"""
if CONF.libvirt.images_type == 'lvm':
info = lvm.get_volume_group_info(
CONF.libvirt.images_volume_group)
elif CONF.libvirt.images_type == 'rbd':
info = rbd_utils.RBDDriver().get_pool_info()
else:
info = libvirt_utils.get_fs_info(CONF.instances_path)
for (k, v) in info.items():
info[k] = v / units.Gi
return info
def _get_vcpu_used(self):
"""Get vcpu usage number of physical computer.
:returns: The total number of vcpu(s) that are currently being used.
"""
total = 0
# Not all libvirt drivers will support the get_vcpus_info()
#
# For example, LXC does not have a concept of vCPUs, while
# QEMU (TCG) traditionally handles all vCPUs in a single
# thread. So both will report an exception when the vcpus()
# API call is made. In such a case we should report the
# guest as having 1 vCPU, since that lets us still do
# CPU over commit calculations that apply as the total
# guest count scales.
#
# It is also possible that we might see an exception if
# the guest is just in middle of shutting down. Technically
# we should report 0 for vCPU usage in this case, but we
# we can't reliably distinguish the vcpu not supported
# case from the just shutting down case. Thus we don't know
# whether to report 1 or 0 for vCPU count.
#
# Under-reporting vCPUs is bad because it could conceivably
# let the scheduler place too many guests on the host. Over-
# reporting vCPUs is not a problem as it'll auto-correct on
# the next refresh of usage data.
#
# Thus when getting an exception we always report 1 as the
# vCPU count, as the least worst value.
for guest in self._host.list_guests():
try:
vcpus = guest.get_vcpus_info()
total += len(list(vcpus))
except libvirt.libvirtError:
total += 1
# NOTE(gtt116): give other tasks a chance.
greenthread.sleep(0)
return total
def _get_supported_vgpu_types(self):
if not CONF.devices.enabled_mdev_types:
return []
# Make sure we register all the types as the compute service could
# be calling this method before init_host()
nova.conf.devices.register_dynamic_opts(CONF)
enabled_mdev_types = []
for vgpu_type in CONF.devices.enabled_mdev_types:
enabled_mdev_types.append(vgpu_type)
# NOTE(sbauza) group is now always set because we register the
# dynamic options above
group = getattr(CONF, 'mdev_%s' % vgpu_type, None)
if group is None:
# Should never happen but if so, just fails early.
raise exception.InvalidLibvirtMdevConfig(
reason="can't find '[devices]/mdev_%s group' "
"in the configuration" % group
)
mdev_class = group.mdev_class
# By default, max_instances is None
if group.max_instances:
self.mdev_type_max_mapping[vgpu_type] = group.max_instances
if not group.device_addresses:
if not self.pgpu_type_default:
self.pgpu_type_default = vgpu_type
self.mdev_classes.add(mdev_class)
else:
msg = ("Mdev type default already set to "
" %(default_type)s so %(this_type)s will not "
"be used." % {
'default_type': self.pgpu_type_default,
'this_type': vgpu_type})
LOG.warning(msg)
# we remove the type from the supported list.
enabled_mdev_types.remove(vgpu_type)
continue
for device_address in group.device_addresses:
if device_address in self.pgpu_type_mapping:
raise exception.InvalidLibvirtMdevConfig(
reason="duplicate types for PCI ID %s" % device_address
)
# Just checking whether the operator fat-fingered the address.
# If it's wrong, it will return an exception
try:
pci_utils.parse_address(device_address)
except exception.PciDeviceWrongAddressFormat:
raise exception.InvalidLibvirtMdevConfig(
reason="incorrect PCI address: %s" % device_address
)
self.pgpu_type_mapping[device_address] = vgpu_type
self.mdev_class_mapping[device_address] = mdev_class
self.mdev_classes.add(mdev_class)
return enabled_mdev_types
@staticmethod
def _get_pci_id_from_libvirt_name(
libvirt_address: str
) -> ty.Optional[str]:
"""Returns a PCI ID from a libvirt pci address name.
:param libvirt_address: the libvirt PCI device name,
eg.'pci_0000_84_00_0'
"""
try:
device_address = "{}:{}:{}.{}".format(
*libvirt_address[4:].split('_'))
# Validates whether it's a PCI ID...
pci_utils.parse_address(device_address)
# .format() can return IndexError
except (exception.PciDeviceWrongAddressFormat, IndexError):
# this is not a valid PCI address
LOG.warning("The PCI address %s was invalid for getting the "
"related mdev type", libvirt_address)
return None
return device_address
def _get_vgpu_type_per_pgpu(self, device_address):
"""Provides the vGPU type the pGPU supports.
:param device_address: the libvirt PCI device name,
eg.'pci_0000_84_00_0'
"""
# Bail out quickly if we don't support vGPUs
if not self.supported_vgpu_types:
return
device_address = self._get_pci_id_from_libvirt_name(device_address)
if not device_address:
return
mdev_type = self.pgpu_type_mapping.get(device_address)
# if we can't find the mdev type by the config, do we have a default
# type because of a config group not using device_addresses ?
# NOTE(sbauza): By default pgpu_type_default is None if unset
return mdev_type or self.pgpu_type_default
def _get_resource_class_for_device(self, device_address):
"""Returns the resource class for the inventory of this device.
:param device_address: the libvirt PCI device name,
eg.'pci_0000_84_00_0'
"""
device_address = self._get_pci_id_from_libvirt_name(device_address)
if not device_address:
# By default, we should always support VGPU as the standard RC
return orc.VGPU
# Remember, this is a defaultdict with orc.VGPU as the default RC
mdev_class = self.mdev_class_mapping[device_address]
return mdev_class
def _get_supported_mdev_resource_classes(self):
return self.mdev_classes
def _count_mediated_devices(self, enabled_mdev_types):
"""Counts the sysfs objects (handles) that represent a mediated device
and filtered by $enabled_mdev_types.
Those handles can be in use by a libvirt guest or not.
:param enabled_mdev_types: list of enabled VGPU types on this host
:returns: dict, keyed by parent GPU libvirt PCI device ID, of number of
mdev device handles for that GPU
"""
counts_per_parent: ty.Dict[str, int] = collections.defaultdict(int)
mediated_devices = self._get_mediated_devices(types=enabled_mdev_types)
for mdev in mediated_devices:
parent_vgpu_type = self._get_vgpu_type_per_pgpu(mdev['parent'])
if mdev['type'] != parent_vgpu_type:
# Even if some mdev was created for another vGPU type, just
# verify all the mdevs related to the type that their pGPU
# has
continue
counts_per_parent[mdev['parent']] += 1
return counts_per_parent
def _count_mdev_capable_devices(self, enabled_mdev_types):
"""Counts the mdev-capable devices on this host filtered by
$enabled_mdev_types.
:param enabled_mdev_types: list of enabled VGPU types on this host
:returns: dict, keyed by device name, to an integer count of available
instances of each type per device
"""
mdev_capable_devices = self._get_mdev_capable_devices(
types=enabled_mdev_types)
counts_per_dev: ty.Dict[str, int] = collections.defaultdict(int)
for dev in mdev_capable_devices:
# dev_id is the libvirt name for the PCI device,
# eg. pci_0000_84_00_0 which matches a PCI address of 0000:84:00.0
dev_name = dev['dev_id']
dev_supported_type = self._get_vgpu_type_per_pgpu(dev_name)
for _type in dev['types']:
if _type != dev_supported_type:
# This is not the type the operator wanted to support for
# this physical GPU
continue
available = dev['types'][_type]['availableInstances']
# NOTE(sbauza): Even if we support multiple types, Nova will
# only use one per physical GPU.
counts_per_dev[dev_name] += available
return counts_per_dev
def _get_gpu_inventories(self):
"""Returns the inventories for each physical GPU for a specific type
supported by the enabled_mdev_types CONF option.
:returns: dict, keyed by libvirt PCI name, of dicts like:
{'pci_0000_84_00_0':
{'total': $TOTAL,
'min_unit': 1,
'max_unit': $TOTAL,
'step_size': 1,
'reserved': 0,
'allocation_ratio': 1.0,
}
}
"""
# Bail out early if operator doesn't care about providing vGPUs
enabled_mdev_types = self.supported_vgpu_types
if not enabled_mdev_types:
return {}
inventories = {}
# counting how many mdevs we are currently supporting per type
type_limit_mapping: ty.Dict[str, int] = collections.defaultdict(int)
count_per_parent = self._count_mediated_devices(enabled_mdev_types)
for dev_name, count in count_per_parent.items():
mdev_type = self._get_vgpu_type_per_pgpu(dev_name)
type_limit_mapping[mdev_type] += count
inventories[dev_name] = {'total': count}
# Filter how many available mdevs we can create for all the supported
# types.
count_per_dev = self._count_mdev_capable_devices(enabled_mdev_types)
# Combine the counts into the dict that we return to the caller.
for dev_name, count in count_per_dev.items():
mdev_type = self._get_vgpu_type_per_pgpu(dev_name)
mdev_limit = self.mdev_type_max_mapping.get(mdev_type)
# Some GPU types could have defined limits. For the others, say
# they are just unlimited
# NOTE(sbauza): Instead of not accepting GPUs if their capacity is
# more than the limit, we could just accept them by capping their
# total value by the limit.
if (mdev_limit and
type_limit_mapping[mdev_type] + count > mdev_limit):
# We don't have space for creating new mediated devices
LOG.debug("Skipping to update %s as the available count of "
"mediated devices (%s) is above the maximum we can "
"use (%s)",
dev_name, count,
mdev_limit - type_limit_mapping[mdev_type])
# We want the resource provider to be deleted, so we pass the
# inventory with a total of 0 so _ensure_pgpu_providers() will
# delete it.
inventories[dev_name] = {'total': 0}
continue
type_limit_mapping[mdev_type] += count
inv_per_parent = inventories.setdefault(
dev_name, {'total': 0})
inv_per_parent['total'] += count
for dev_name in inventories:
inventories[dev_name].update({
'min_unit': 1,
'step_size': 1,
'reserved': 0,
# NOTE(sbauza): There is no sense to have a ratio but 1.0
# since we can't overallocate vGPU resources
'allocation_ratio': 1.0,
# FIXME(sbauza): Some vendors could support only one
'max_unit': inventories[dev_name]['total'],
})
return inventories
def _get_instance_capabilities(self):
"""Get hypervisor instance capabilities
Returns a list of tuples that describe instances the
hypervisor is capable of hosting. Each tuple consists
of the triplet (arch, hypervisor_type, vm_mode).
:returns: List of tuples describing instance capabilities
"""
caps = self._host.get_capabilities()
instance_caps = list()
for g in caps.guests:
for domain_type in g.domains:
try:
instance_cap = (
fields.Architecture.canonicalize(g.arch),
fields.HVType.canonicalize(domain_type),
fields.VMMode.canonicalize(g.ostype))
instance_caps.append(instance_cap)
except exception.InvalidArchitectureName:
# NOTE(danms): Libvirt is exposing a guest arch that nova
# does not even know about. Avoid aborting here and
# continue to process the rest.
pass
return instance_caps
def _get_cpu_info(self):
"""Get cpuinfo information.
Obtains cpu feature from virConnect.getCapabilities.
:return: see above description
"""
caps = self._host.get_capabilities()
cpu_info = dict()
cpu_info['arch'] = caps.host.cpu.arch
cpu_info['model'] = caps.host.cpu.model
cpu_info['vendor'] = caps.host.cpu.vendor
topology = dict()
topology['cells'] = len(getattr(caps.host.topology, 'cells', [1]))
topology['sockets'] = caps.host.cpu.sockets
topology['cores'] = caps.host.cpu.cores
topology['threads'] = caps.host.cpu.threads
cpu_info['topology'] = topology
if caps.host.cpu.maxphysaddr:
maxphysaddr = dict()
maxphysaddr["mode"] = caps.host.cpu.maxphysaddr.mode
maxphysaddr["bits"] = caps.host.cpu.maxphysaddr.bits
cpu_info["maxphysaddr"] = maxphysaddr
features = set()
for f in caps.host.cpu.features:
features.add(f.name)
cpu_info['features'] = features
return cpu_info
# TODO(stephenfin): Move this to 'host.py'
def _get_pci_passthrough_devices(self):
"""Get host PCI devices information.
Obtains pci devices information from libvirt, and returns
as a JSON string.
Each device information is a dictionary, with mandatory keys
of 'address', 'vendor_id', 'product_id', 'dev_type', 'dev_id',
'label' and other optional device specific information.
Refer to the objects/pci_device.py for more idea of these keys.
:returns: a JSON string containing a list of the assignable PCI
devices information
"""
dev_flags = (
libvirt.VIR_CONNECT_LIST_NODE_DEVICES_CAP_NET |
libvirt.VIR_CONNECT_LIST_NODE_DEVICES_CAP_PCI_DEV |
libvirt.VIR_CONNECT_LIST_NODE_DEVICES_CAP_VDPA
)
devices = {
dev.name(): dev for dev in
self._host.list_all_devices(flags=dev_flags)
}
# NOTE(mnaser): The listCaps() function can raise an exception if the
# device disappeared while we're looping, this method
# returns an empty list rather than raising an exception
# which will remove the device for Nova's resource
# tracker, but that is OK since the device disappeared.
def _safe_list_caps(dev):
try:
return dev.listCaps()
except libvirt.libvirtError:
return []
net_devs = []
vdpa_devs = []
pci_devs = {}
for name, dev in devices.items():
device_caps = _safe_list_caps(dev)
if "net" in device_caps:
net_devs.append(dev)
if "vdpa" in device_caps:
vdpa_devs.append(dev)
if "pci" in device_caps:
pci_devs[name] = dev
# Give other tasks to work during libvirt connection task
greenthread.sleep(0)
pci_info = [
self._host._get_pcidev_info(
name, dev, net_devs,
vdpa_devs, list(pci_devs.values())
)
for name, dev in pci_devs.items()
]
return jsonutils.dumps(pci_info)
def _get_mdev_capabilities_for_dev(self, devname, types=None):
"""Returns a dict of MDEV capable device with the ID as first key
and then a list of supported types, each of them being a dict.
:param types: Only return those specific types.
"""
virtdev = self._host.device_lookup_by_name(devname)
xmlstr = virtdev.XMLDesc(0)
cfgdev = vconfig.LibvirtConfigNodeDevice()
cfgdev.parse_str(xmlstr)
device = {
"dev_id": cfgdev.name,
"types": {},
"vendor_id": cfgdev.pci_capability.vendor_id,
}
for mdev_cap in cfgdev.pci_capability.mdev_capability:
for cap in mdev_cap.mdev_types:
if not types or cap['type'] in types:
device["types"].update({cap['type']: {
'availableInstances': cap['availableInstances'],
# This attribute is optional
'name': cap.get('name'),
'deviceAPI': cap['deviceAPI']}})
return device
def _get_mdev_capable_devices(self, types=None):
"""Get host devices supporting mdev types.
Obtain devices information from libvirt and returns a list of
dictionaries.
:param types: Filter only devices supporting those types.
"""
dev_names = self._host.list_mdev_capable_devices() or []
mdev_capable_devices = []
for name in dev_names:
device = self._get_mdev_capabilities_for_dev(name, types)
if not device["types"]:
continue
mdev_capable_devices.append(device)
return mdev_capable_devices
def _get_mediated_device_information(self, devname):
"""Returns a dict of a mediated device."""
# LP #1951656 - In Libvirt 7.7, the mdev name now includes the PCI
# address of the parent device (e.g. mdev_<uuid>_<pci_address>) due to
# the mdevctl allowing for multiple mediated devs having the same UUID
# defined (only one can be active at a time). Since the guest
# information doesn't have the parent ID, try to lookup which
# mediated device is available that matches the UUID. If multiple
# devices are found that match the UUID, then this is an error
# condition.
try:
virtdev = self._host.device_lookup_by_name(devname)
except libvirt.libvirtError as ex:
if ex.get_error_code() != libvirt.VIR_ERR_NO_NODE_DEVICE:
raise
mdevs = [dev for dev in self._host.list_mediated_devices()
if dev.startswith(devname)]
# If no matching devices are found, simply raise the original
# exception indicating that no devices are found.
if not mdevs:
raise
elif len(mdevs) > 1:
msg = ("The mediated device name %(devname)s refers to a UUID "
"that is present in multiple libvirt mediated devices. "
"Matching libvirt mediated devices are %(devices)s. "
"Mediated device UUIDs must be unique for Nova." %
{'devname': devname,
'devices': ', '.join(mdevs)})
raise exception.InvalidLibvirtMdevConfig(reason=msg)
LOG.debug('Found requested device %s as %s. Using that.',
devname, mdevs[0])
virtdev = self._host.device_lookup_by_name(mdevs[0])
xmlstr = virtdev.XMLDesc(0)
cfgdev = vconfig.LibvirtConfigNodeDevice()
cfgdev.parse_str(xmlstr)
# Starting with Libvirt 7.3, the uuid information is available in the
# node device information. If its there, use that. Otherwise,
# fall back to the previous behavior of parsing the uuid from the
# devname.
if cfgdev.mdev_information.uuid:
mdev_uuid = cfgdev.mdev_information.uuid
else:
mdev_uuid = libvirt_utils.mdev_name2uuid(cfgdev.name)
device = {
"dev_id": cfgdev.name,
"uuid": mdev_uuid,
# the physical GPU PCI device
"parent": cfgdev.parent,
"type": cfgdev.mdev_information.type,
"iommu_group": cfgdev.mdev_information.iommu_group,
}
return device
def _get_mediated_devices(self, types=None):
"""Get host mediated devices.
Obtain devices information from libvirt and returns a list of
dictionaries.
:param types: Filter only devices supporting those types.
"""
dev_names = self._host.list_mediated_devices() or []
mediated_devices = []
for name in dev_names:
device = self._get_mediated_device_information(name)
if not types or device["type"] in types:
mediated_devices.append(device)
return mediated_devices
def _get_mdev_types_from_uuids(self, mdev_uuids):
"""Returns a dict of mdevs and their type from a list of mediated
device UUIDs. If no mdevs are actually using those UUIDs, it returns an
empty dict.
:param mdev_uuids: List of existing mediated device UUIDs.
:returns: dict where key is the mdev UUID and the value is its type.
"""
host_mdevs = self._get_mediated_devices()
inst_dev_infos = filter(lambda dev: dev['uuid'] in mdev_uuids,
host_mdevs)
return {mdev['uuid']: mdev['type'] for mdev in inst_dev_infos}
def _get_all_assigned_mediated_devices(self, instance=None):
"""Lookup all instances from the host and return all the mediated
devices that are assigned to a guest.
:param instance: Only return mediated devices for that instance.
:returns: A dictionary of keys being mediated device UUIDs and their
respective values the instance UUID of the guest using it.
Returns an empty dict if an instance is provided but not
found in the hypervisor.
"""
allocated_mdevs = {}
# Add the reserved mediated devices for live-migration
for instance_uuid, mdev_uuids in self.instance_claimed_mdevs.items():
if instance and instance.uuid != instance_uuid:
continue
for mdev in mdev_uuids:
allocated_mdevs[mdev] = instance_uuid
if instance:
# NOTE(sbauza): In some cases (like a migration issue), the
# instance can exist in the Nova database but libvirt doesn't know
# about it. For such cases, the way to fix that is to hard reboot
# the instance, which will recreate the libvirt guest.
# For that reason, we need to support that case by making sure
# we don't raise an exception if the libvirt guest doesn't exist.
try:
guest = self._host.get_guest(instance)
except exception.InstanceNotFound:
# Bail out early if libvirt doesn't know about it since we
# can't know the existing mediated devices
# Some mdevs could be claimed for that instance
return allocated_mdevs
guests = [guest]
else:
guests = self._host.list_guests(only_running=False)
for guest in guests:
cfg = guest.get_config()
for device in cfg.devices:
if isinstance(device, vconfig.LibvirtConfigGuestHostdevMDEV):
allocated_mdevs[device.uuid] = guest.uuid
return allocated_mdevs
# TODO(sbauza): Rename this method into _mdev_allocations
def _vgpu_allocations(self, allocations):
"""Filtering only the mdev allocations from a list of allocations.
:param allocations: Information about resources allocated to the
instance via placement, of the form returned by
SchedulerReportClient.get_allocations_for_consumer.
"""
if not allocations:
# If no allocations, there is no vGPU request.
return {}
mdev_rcs = self._get_supported_mdev_resource_classes()
vgpu_allocations = {}
for rp in allocations:
res = allocations[rp]['resources']
mdev_resources = {mdev_RC: res[mdev_RC] for mdev_RC in mdev_rcs
if mdev_RC in res and res[mdev_RC] > 0}
if mdev_resources:
vgpu_allocations[rp] = {'resources': mdev_resources}
return vgpu_allocations
def _get_existing_mdevs_not_assigned(self, parent, requested_types=None):
"""Returns the already created mediated devices that are not assigned
to a guest yet.
:param parent: Filter out result for only mdevs from the parent device.
:param requested_types: Filter out the result for only mediated devices
having those types.
"""
LOG.debug('Searching for available mdevs...')
allocated_mdevs = self._get_all_assigned_mediated_devices()
mdevs = self._get_mediated_devices(requested_types)
available_mdevs = set()
for mdev in mdevs:
parent_vgpu_type = self._get_vgpu_type_per_pgpu(mdev['parent'])
if mdev['type'] != parent_vgpu_type:
# This mdev is using a vGPU type that is not supported by the
# configuration option for its pGPU parent, so we can't use it.
continue
# FIXME(sbauza): No longer accept the parent value to be nullable
# once we fix the reshape functional test
if parent is None or mdev['parent'] == parent:
available_mdevs.add(mdev["uuid"])
available_mdevs -= set(allocated_mdevs)
LOG.info('Available mdevs at: %s.', available_mdevs)
return available_mdevs
def _create_mdev(self, dev_name, mdev_type, uuid=None):
if uuid is None:
uuid = uuidutils.generate_uuid()
conf = vconfig.LibvirtConfigNodeDevice()
conf.parent = dev_name
conf.mdev_information = (
vconfig.LibvirtConfigNodeDeviceMdevInformation())
conf.mdev_information.type = mdev_type
conf.mdev_information.uuid = uuid
# Create the transient device.
self._host.device_create(conf)
# Define it to make it persistent.
mdev_dev = self._host.device_define(conf)
# TODO(Uggla): Remove this in the libvirt bump cleanup patch
# As we are not setting autostart anymore, because we are not
# passing in following code.
# It makes test_allocate_mdevs_with_no_mdevs_but_capacity test to fail.
# So removing the tests.
if self._host.has_min_version(MIN_LIBVIRT_NODEDEV_AUTOSTART):
# Set it to automatically start when the compute host boots or the
# parent device becomes available.
# NOTE(melwitt): Make this not fatal because we can try to manually
# start mdevs in init_host() if they didn't start automatically
# after a host reboot.
try:
self._host.device_set_autostart(mdev_dev, autostart=True)
except Exception as e:
LOG.info(
'Failed to set autostart to True for mdev '
f'{mdev_dev.name()} with UUID {uuid}: {str(e)}.')
return uuid
def _create_new_mediated_device(self, parent, uuid=None):
"""Find a physical device that can support a new mediated device and
create it.
:param parent: The libvirt name of the parent GPU, eg. pci_0000_06_00_0
:param uuid: The possible mdev UUID we want to create again
:returns: the newly created mdev UUID or None if not possible
"""
LOG.debug('Attempting to create new mdev...')
supported_types = self.supported_vgpu_types
# Try to see if we can still create a new mediated device
devices = self._get_mdev_capable_devices(supported_types)
for device in devices:
dev_name = device['dev_id']
# FIXME(sbauza): No longer accept the parent value to be nullable
# once we fix the reshape functional test
if parent is not None and dev_name != parent:
# The device is not the one that was called, not creating
# the mdev
continue
LOG.debug('Trying on: %s.', dev_name)
dev_supported_type = self._get_vgpu_type_per_pgpu(dev_name)
if dev_supported_type and device['types'][
dev_supported_type]['availableInstances'] > 0:
# That physical GPU has enough room for a new mdev
# We need the PCI address, not the libvirt name
# The libvirt name is like 'pci_0000_84_00_0'
pci_addr = "{}:{}:{}.{}".format(*dev_name[4:].split('_'))
if not self._host.has_min_version(MIN_LIBVIRT_PERSISTENT_MDEV):
chosen_mdev = nova.privsep.libvirt.create_mdev(
pci_addr, dev_supported_type, uuid=uuid)
else:
chosen_mdev = self._create_mdev(
dev_name, dev_supported_type, uuid=uuid)
LOG.info('Created mdev: %s on pGPU: %s.',
chosen_mdev, pci_addr)
return chosen_mdev
LOG.debug('Failed: No available instances on device.')
LOG.info('Failed to create mdev. '
'No free space found among the following devices: %s.',
[dev['dev_id'] for dev in devices])
@utils.synchronized(VGPU_RESOURCE_SEMAPHORE)
def _allocate_mdevs(self, allocations):
"""Returns a list of mediated device UUIDs corresponding to available
resources we can assign to the guest(s) corresponding to the allocation
requests passed as argument.
That method can either find an existing but unassigned mediated device
it can allocate, or create a new mediated device from a capable
physical device if the latter has enough left capacity.
:param allocations: Information about resources allocated to the
instance via placement, of the form returned by
SchedulerReportClient.get_allocations_for_consumer.
That code is supporting Placement API version 1.12
"""
vgpu_allocations = self._vgpu_allocations(allocations)
if not vgpu_allocations:
return
# TODO(sbauza): For the moment, we only support allocations for only
# one pGPU.
if len(vgpu_allocations) > 1:
LOG.warning('More than one allocation was passed over to libvirt '
'while at the moment libvirt only supports one. Only '
'the first allocation will be looked up.')
rp_uuid, alloc = next(iter(vgpu_allocations.items()))
# We only have one allocation with a supported resource class
vgpus_asked = list(alloc['resources'].values())[0]
# Find if we allocated against a specific pGPU (and then the allocation
# is made against a child RP) or any pGPU (in case the VGPU inventory
# is still on the root RP)
try:
allocated_rp = self.provider_tree.data(rp_uuid)
except ValueError:
# The provider doesn't exist, return a better understandable
# exception
raise exception.ComputeResourcesUnavailable(
reason='mdev-capable resource is not available')
# FIXME(sbauza): The functional reshape test assumes that we could
# run _allocate_mdevs() against non-nested RPs but this is impossible
# as all inventories have been reshaped *before now* since it's done
# on init_host() (when the compute restarts or whatever else calls it).
# That said, since fixing the functional test isn't easy yet, let's
# assume we still support a non-nested RP for now.
if allocated_rp.parent_uuid is None:
# We are on a root RP
parent_device = None
else:
rp_name = allocated_rp.name
# There can be multiple roots, we need to find the root name
# to guess the physical device name
roots = list(self.provider_tree.roots)
for root in roots:
if rp_name.startswith(root.name + '_'):
# The RP name convention is :
# root_name + '_' + parent_device
parent_device = rp_name[len(root.name) + 1:]
break
else:
LOG.warning(
"mdev-capable device name %(name)s can't be guessed from "
"the ProviderTree roots %(roots)s",
{'name': rp_name,
'roots': ', '.join([root.name for root in roots])})
# We f... have no idea what was the parent device
# If we can't find devices having available VGPUs, just raise
raise exception.ComputeResourcesUnavailable(
reason='mdev-capable resource is not available')
supported_types = self.supported_vgpu_types
# Which mediated devices are created but not assigned to a guest ?
mdevs_available = self._get_existing_mdevs_not_assigned(
parent_device, supported_types)
chosen_mdevs = []
for c in range(vgpus_asked):
chosen_mdev = None
if mdevs_available:
# Take the first available mdev
chosen_mdev = mdevs_available.pop()
else:
LOG.debug('No available mdevs where found. '
'Creating an new one...')
chosen_mdev = self._create_new_mediated_device(parent_device)
if not chosen_mdev:
# If we can't find devices having available VGPUs, just raise
raise exception.ComputeResourcesUnavailable(
reason='mdev-capable resource is not available')
else:
chosen_mdevs.append(chosen_mdev)
LOG.info('Allocated mdev: %s.', chosen_mdev)
return chosen_mdevs
def _detach_mediated_devices(self, guest):
mdevs = guest.get_all_devices(
devtype=vconfig.LibvirtConfigGuestHostdevMDEV)
for mdev_cfg in mdevs:
try:
guest.detach_device(mdev_cfg, live=True)
except libvirt.libvirtError as ex:
error_code = ex.get_error_code()
if error_code == libvirt.VIR_ERR_CONFIG_UNSUPPORTED:
reason = _("Suspend is not supported for instances having "
"attached mediated devices.")
raise exception.InstanceFaultRollback(
exception.InstanceSuspendFailure(reason=reason))
else:
raise
def _attach_mediated_devices(self, guest, devs):
for mdev_cfg in devs:
try:
guest.attach_device(mdev_cfg, live=True)
except libvirt.libvirtError as ex:
error_code = ex.get_error_code()
if error_code == libvirt.VIR_ERR_DEVICE_MISSING:
LOG.warning("The mediated device %s was not found and "
"won't be reattached to %s.", mdev_cfg, guest)
else:
raise
def _get_mdevs_from_guest_config(self, xml):
"""Get all libvirt's mediated devices from a guest's config (XML) file.
We don't have to worry about those devices being used by another guest,
since they remain allocated for the current guest as long as they are
present in the XML.
:param xml: The XML from the guest we want to get a list of mdevs from.
:returns: A list containing the objects that represent the mediated
devices attached to the guest's config passed as argument.
"""
config = vconfig.LibvirtConfigGuest()
config.parse_str(xml)
devs = []
for dev in config.devices:
if isinstance(dev, vconfig.LibvirtConfigGuestHostdevMDEV):
devs.append(dev)
return devs
def _has_numa_support(self):
# This means that the host can support LibvirtConfigGuestNUMATune
# and the nodeset field in LibvirtConfigGuestMemoryBackingPage
caps = self._host.get_capabilities()
if (caps.host.cpu.arch in (fields.Architecture.I686,
fields.Architecture.X86_64,
fields.Architecture.AARCH64) and
self._host.has_min_version(hv_type=host.HV_DRIVER_QEMU)):
return True
elif (caps.host.cpu.arch in (fields.Architecture.PPC64,
fields.Architecture.PPC64LE)):
return True
return False
def _get_host_numa_topology(self):
if not self._has_numa_support():
return
caps = self._host.get_capabilities()
topology = caps.host.topology
if topology is None or not topology.cells:
return
cells = []
available_shared_cpus = self._get_vcpu_available()
available_dedicated_cpus = self._get_pcpu_available()
# NOTE(stephenfin): In an ideal world, if the operator had not
# configured this host to report PCPUs using the '[compute]
# cpu_dedicated_set' option, then we should not be able to used pinned
# instances on this host. However, that would force operators to update
# their configuration as part of the Stein -> Train upgrade or be
# unable to schedule instances on the host. As a result, we need to
# revert to legacy behavior and use 'vcpu_pin_set' for both VCPUs and
# PCPUs.
# TODO(stephenfin): Remove this in U
if not available_dedicated_cpus and not (
CONF.compute.cpu_shared_set and not CONF.vcpu_pin_set):
available_dedicated_cpus = available_shared_cpus
def _get_reserved_memory_for_cell(self, cell_id, page_size):
cell = self._reserved_hugepages.get(cell_id, {})
return cell.get(page_size, 0)
def _get_physnet_numa_affinity():
affinities: ty.Dict[int, ty.Set[str]] = {
cell.id: set() for cell in topology.cells
}
for physnet in CONF.neutron.physnets:
# This will error out if the group is not registered, which is
# exactly what we want as that would be a bug
group = getattr(CONF, 'neutron_physnet_%s' % physnet)
if not group.numa_nodes:
msg = ("the physnet '%s' was listed in '[neutron] "
"physnets' but no corresponding "
"'[neutron_physnet_%s] numa_nodes' option was "
"defined." % (physnet, physnet))
raise exception.InvalidNetworkNUMAAffinity(reason=msg)
for node in group.numa_nodes:
if node not in affinities:
msg = ("node %d for physnet %s is not present in host "
"affinity set %r" % (node, physnet, affinities))
# The config option referenced an invalid node
raise exception.InvalidNetworkNUMAAffinity(reason=msg)
affinities[node].add(physnet)
return affinities
def _get_tunnel_numa_affinity():
affinities = {cell.id: False for cell in topology.cells}
for node in CONF.neutron_tunnel.numa_nodes:
if node not in affinities:
msg = ("node %d for tunneled networks is not present "
"in host affinity set %r" % (node, affinities))
# The config option referenced an invalid node
raise exception.InvalidNetworkNUMAAffinity(reason=msg)
affinities[node] = True
return affinities
physnet_affinities = _get_physnet_numa_affinity()
tunnel_affinities = _get_tunnel_numa_affinity()
for cell in topology.cells:
cpus = set(cpu.id for cpu in cell.cpus)
# NOTE(artom) We assume we'll never see hardware with multiple
# sockets in a single NUMA node - IOW, the socket_id for all CPUs
# in a single cell will be the same. To make that assumption
# explicit, we leave the cell's socket_id as None if that's the
# case.
socket_id = None
sockets = set([cpu.socket_id for cpu in cell.cpus])
if len(sockets) == 1:
socket_id = sockets.pop()
else:
LOG.warning('This host appears to have multiple sockets per '
'NUMA node. The `socket` PCI NUMA affinity '
'will not be supported.')
cpuset = cpus & available_shared_cpus
pcpuset = cpus & available_dedicated_cpus
# de-duplicate and sort the list of CPU sibling sets
siblings = sorted(
set(x) for x in set(
tuple(cpu.siblings) or () for cpu in cell.cpus
)
)
cpus &= available_shared_cpus | available_dedicated_cpus
siblings = [sib & cpus for sib in siblings]
# Filter out empty sibling sets that may be left
siblings = [sib for sib in siblings if len(sib) > 0]
mempages = [
objects.NUMAPagesTopology(
size_kb=pages.size,
total=pages.total,
used=0,
reserved=_get_reserved_memory_for_cell(
self, cell.id, pages.size))
for pages in cell.mempages]
network_metadata = objects.NetworkMetadata(
physnets=physnet_affinities[cell.id],
tunneled=tunnel_affinities[cell.id])
# NOTE(stephenfin): Note that we don't actually return any usage
# information here. This is because this is handled by the resource
# tracker via the 'update_available_resource' periodic task, which
# loops through all instances and calculated usage accordingly
cell = objects.NUMACell(
id=cell.id,
socket=socket_id,
cpuset=cpuset,
pcpuset=pcpuset,
memory=cell.memory / units.Ki,
cpu_usage=0,
pinned_cpus=set(),
memory_usage=0,
siblings=siblings,
mempages=mempages,
network_metadata=network_metadata)
cells.append(cell)
return objects.NUMATopology(cells=cells)
def get_all_volume_usage(self, context, compute_host_bdms):
"""Return usage info for volumes attached to vms on
a given host.
"""
vol_usage = []
for instance_bdms in compute_host_bdms:
instance = instance_bdms['instance']
for bdm in instance_bdms['instance_bdms']:
mountpoint = bdm['device_name']
if mountpoint.startswith('/dev/'):
mountpoint = mountpoint[5:]
volume_id = bdm['volume_id']
LOG.debug("Trying to get stats for the volume %s",
volume_id, instance=instance)
vol_stats = self.block_stats(instance, mountpoint)
if vol_stats:
stats = dict(volume=volume_id,
instance=instance,
rd_req=vol_stats[0],
rd_bytes=vol_stats[1],
wr_req=vol_stats[2],
wr_bytes=vol_stats[3])
LOG.debug(
"Got volume usage stats for the volume=%(volume)s,"
" rd_req=%(rd_req)d, rd_bytes=%(rd_bytes)d, "
"wr_req=%(wr_req)d, wr_bytes=%(wr_bytes)d",
stats, instance=instance)
vol_usage.append(stats)
return vol_usage
def block_stats(self, instance, disk_id):
"""Note that this function takes an instance name."""
try:
guest = self._host.get_guest(instance)
dev = guest.get_block_device(disk_id)
return dev.blockStats()
except libvirt.libvirtError as e:
errcode = e.get_error_code()
LOG.info('Getting block stats failed, device might have '
'been detached. Instance=%(instance_name)s '
'Disk=%(disk)s Code=%(errcode)s Error=%(e)s',
{'instance_name': instance.name, 'disk': disk_id,
'errcode': errcode, 'e': e},
instance=instance)
except exception.InstanceNotFound:
LOG.info('Could not find domain in libvirt for instance %s. '
'Cannot get block stats for device', instance.name,
instance=instance)
def update_provider_tree(self, provider_tree, nodename, allocations=None):
"""Update a ProviderTree object with current resource provider,
inventory information and CPU traits.
:param nova.compute.provider_tree.ProviderTree provider_tree:
A nova.compute.provider_tree.ProviderTree object representing all
the providers in the tree associated with the compute node, and any
sharing providers (those with the ``MISC_SHARES_VIA_AGGREGATE``
trait) associated via aggregate with any of those providers (but
not *their* tree- or aggregate-associated providers), as currently
known by placement.
:param nodename:
String name of the compute node (i.e.
ComputeNode.hypervisor_hostname) for which the caller is requesting
updated provider information.
:param allocations:
Dict of allocation data of the form:
{ $CONSUMER_UUID: {
# The shape of each "allocations" dict below is identical
# to the return from GET /allocations/{consumer_uuid}
"allocations": {
$RP_UUID: {
"generation": $RP_GEN,
"resources": {
$RESOURCE_CLASS: $AMOUNT,
...
},
},
...
},
"project_id": $PROJ_ID,
"user_id": $USER_ID,
"consumer_generation": $CONSUMER_GEN,
},
...
}
If None, and the method determines that any inventory needs to be
moved (from one provider to another and/or to a different resource
class), the ReshapeNeeded exception must be raised. Otherwise, this
dict must be edited in place to indicate the desired final state of
allocations.
:raises ReshapeNeeded: If allocations is None and any inventory needs
to be moved from one provider to another and/or to a different
resource class.
:raises: ReshapeFailed if the requested tree reshape fails for
whatever reason.
"""
disk_gb = int(self._get_local_gb_info()['total'])
memory_mb = int(self._host.get_memory_mb_total())
vcpus = len(self._get_vcpu_available())
pcpus = len(self._get_pcpu_available())
memory_enc_slots = self._get_memory_encrypted_slots()
# NOTE(yikun): If the inv record does not exists, the allocation_ratio
# will use the CONF.xxx_allocation_ratio value if xxx_allocation_ratio
# is set, and fallback to use the initial_xxx_allocation_ratio
# otherwise.
inv = provider_tree.data(nodename).inventory
ratios = self._get_allocation_ratios(inv)
resources: ty.Dict[str, ty.Set['objects.Resource']] = (
collections.defaultdict(set)
)
result = {}
if memory_mb:
result[orc.MEMORY_MB] = {
'total': memory_mb,
'min_unit': 1,
'max_unit': memory_mb,
'step_size': 1,
'allocation_ratio': ratios[orc.MEMORY_MB],
'reserved': CONF.reserved_host_memory_mb,
}
# NOTE(stephenfin): We have to optionally report these since placement
# forbids reporting inventory with total=0
if vcpus:
result[orc.VCPU] = {
'total': vcpus,
'min_unit': 1,
'max_unit': vcpus,
'step_size': 1,
'allocation_ratio': ratios[orc.VCPU],
'reserved': CONF.reserved_host_cpus,
}
if pcpus:
result[orc.PCPU] = {
'total': pcpus,
'min_unit': 1,
'max_unit': pcpus,
'step_size': 1,
'allocation_ratio': 1,
'reserved': 0,
}
if memory_enc_slots:
result[orc.MEM_ENCRYPTION_CONTEXT] = {
'total': memory_enc_slots,
'min_unit': 1,
'max_unit': 1,
'step_size': 1,
'allocation_ratio': 1.0,
'reserved': 0,
}
# If a sharing DISK_GB provider exists in the provider tree, then our
# storage is shared, and we should not report the DISK_GB inventory in
# the compute node provider.
# TODO(efried): Reinstate non-reporting of shared resource by the
# compute RP once the issues from bug #1784020 have been resolved.
if provider_tree.has_sharing_provider(orc.DISK_GB):
LOG.debug('Ignoring sharing provider - see bug #1784020')
if disk_gb:
result[orc.DISK_GB] = {
'total': disk_gb,
'min_unit': 1,
'max_unit': disk_gb,
'step_size': 1,
'allocation_ratio': ratios[orc.DISK_GB],
'reserved': (self._get_reserved_host_disk_gb_from_config() +
self._get_disk_size_reserved_for_image_cache()),
}
# TODO(sbauza): Use traits to providing vGPU types. For the moment,
# it will be only documentation support by explaining to use
# osc-placement to create custom traits for each of the pGPU RPs.
self._update_provider_tree_for_vgpu(
provider_tree, nodename, allocations=allocations)
self._update_provider_tree_for_pcpu(
provider_tree, nodename, allocations=allocations)
self._update_provider_tree_for_vpmems(
provider_tree, nodename, result, resources)
provider_tree.update_inventory(nodename, result)
provider_tree.update_resources(nodename, resources)
# Add supported traits i.e. those equal to True to provider tree while
# removing the unsupported ones
traits_to_add = [
t for t in self.static_traits if self.static_traits[t]
]
traits_to_remove = set(self.static_traits) - set(traits_to_add)
provider_tree.add_traits(nodename, *traits_to_add)
provider_tree.remove_traits(nodename, *traits_to_remove)
# Now that we updated the ProviderTree, we want to store it locally
# so that spawn() or other methods can access it thru a getter
self.provider_tree = copy.deepcopy(provider_tree)
def _update_provider_tree_for_vpmems(self, provider_tree, nodename,
inventory, resources):
"""Update resources and inventory for vpmems in provider tree."""
prov_data = provider_tree.data(nodename)
for rc, vpmems in self._vpmems_by_rc.items():
# Skip (and omit) inventories with total=0 because placement does
# not allow setting total=0 for inventory.
if not len(vpmems):
continue
inventory[rc] = {
'total': len(vpmems),
'max_unit': len(vpmems),
'min_unit': 1,
'step_size': 1,
'allocation_ratio': 1.0,
'reserved': 0
}
for vpmem in vpmems:
resource_obj = objects.Resource(
provider_uuid=prov_data.uuid,
resource_class=rc,
identifier=vpmem.name,
metadata=vpmem)
resources[rc].add(resource_obj)
def _get_memory_encrypted_slots(self):
conf_slots = CONF.libvirt.num_memory_encrypted_guests
if not self._host.supports_amd_sev:
if conf_slots and conf_slots > 0:
LOG.warning("Host is configured with "
"libvirt.num_memory_encrypted_guests set to "
"%d, but is not SEV-capable.", conf_slots)
return 0
slots = db_const.MAX_INT
# NOTE(tkajinam): Current nova supports SEV only so we ignore SEV-ES
if self._host.max_sev_guests is not None:
slots = self._host.max_sev_guests
if conf_slots is not None:
if conf_slots > slots:
LOG.warning("Host is configured with "
"libvirt.num_memory_encrypted_guests set to %d, "
"but supports only %d.", conf_slots, slots)
slots = min(slots, conf_slots)
LOG.debug("Available memory encrypted slots: %d", slots)
return slots
@property
def static_traits(self) -> ty.Dict[str, bool]:
if self._static_traits is not None:
return self._static_traits
traits: ty.Dict[str, bool] = {}
traits.update(self._get_cpu_traits())
traits.update(self._get_packed_virtqueue_traits())
traits.update(self._get_storage_bus_traits())
traits.update(self._get_video_model_traits())
traits.update(self._get_vif_model_traits())
traits.update(self._get_iommu_model_traits())
traits.update(self._get_tpm_traits())
_, invalid_traits = ot.check_traits(traits)
for invalid_trait in invalid_traits:
LOG.debug("Trait '%s' is not valid; ignoring.", invalid_trait)
del traits[invalid_trait]
self._static_traits = traits
return self._static_traits
@staticmethod
def _is_reshape_needed_vgpu_on_root(provider_tree, nodename):
"""Determine if root RP has VGPU inventories.
Check to see if the root compute node provider in the tree for
this host already has VGPU inventory because if it does, we either
need to signal for a reshape (if _update_provider_tree_for_vgpu()
has no allocations) or move the allocations within the ProviderTree if
passed.
:param provider_tree: The ProviderTree object for this host.
:param nodename: The ComputeNode.hypervisor_hostname, also known as
the name of the root node provider in the tree for this host.
:returns: boolean, whether we have VGPU root inventory.
"""
root_node = provider_tree.data(nodename)
return orc.VGPU in root_node.inventory
def _ensure_pgpu_providers(self, inventories_dict, provider_tree,
nodename):
"""Ensures GPU inventory providers exist in the tree for $nodename.
GPU providers are named $nodename_$gpu-device-id, e.g.
``somehost.foo.bar.com_pci_0000_84_00_0``.
:param inventories_dict: Dictionary of inventories for VGPU class
directly provided by _get_gpu_inventories() and which looks like:
{'pci_0000_84_00_0':
{'total': $TOTAL,
'min_unit': 1,
'max_unit': $MAX_UNIT, # defaults to $TOTAL
'step_size': 1,
'reserved': 0,
'allocation_ratio': 1.0,
}
}
:param provider_tree: The ProviderTree to update.
:param nodename: The ComputeNode.hypervisor_hostname, also known as
the name of the root node provider in the tree for this host.
:returns: dict, keyed by GPU device ID, to ProviderData object
representing that resource provider in the tree
"""
# Create the VGPU child providers if they do not already exist.
# Dict of PGPU RPs keyed by their libvirt PCI name
pgpu_rps = {}
for pgpu_dev_id, inventory in inventories_dict.items():
# For each physical GPU, we make sure to have a child provider
pgpu_rp_name = '%s_%s' % (nodename, pgpu_dev_id)
# Skip (and omit) inventories with total=0 because placement does
# not allow setting total=0 for inventory. If the inventory already
# exists, we rather delete it.
if not inventory['total']:
if provider_tree.exists(pgpu_rp_name):
LOG.debug('Deleting %s resource provider since it does '
'not longer have any inventory', pgpu_rp_name)
provider_tree.remove(pgpu_rp_name)
continue
if not provider_tree.exists(pgpu_rp_name):
# This is the first time creating the child provider so add
# it to the tree under the root node provider.
provider_tree.new_child(pgpu_rp_name, nodename)
# We want to idempotently return the resource providers with VGPUs
pgpu_rp = provider_tree.data(pgpu_rp_name)
pgpu_rps[pgpu_dev_id] = pgpu_rp
# The VGPU inventory goes on a child provider of the given root
# node, identified by $nodename.
mdev_rc = self._get_resource_class_for_device(pgpu_dev_id)
pgpu_inventory = {mdev_rc: inventory}
provider_tree.update_inventory(pgpu_rp_name, pgpu_inventory)
return pgpu_rps
@staticmethod
def _assert_is_root_provider(
rp_uuid, root_node, consumer_uuid, alloc_data):
"""Asserts during a reshape that rp_uuid is for the root node provider.
When reshaping, inventory and allocations should be on the root node
provider and then moved to child providers.
:param rp_uuid: UUID of the provider that holds inventory/allocations.
:param root_node: ProviderData object representing the root node in a
provider tree.
:param consumer_uuid: UUID of the consumer (instance) holding resource
allocations against the given rp_uuid provider.
:param alloc_data: dict of allocation data for the consumer.
:raises: ReshapeFailed if rp_uuid is not the root node indicating a
reshape was needed but the inventory/allocation structure is not
expected.
"""
if rp_uuid != root_node.uuid:
# Something is wrong - VGPU inventory should
# only be on the root node provider if we are
# reshaping the tree.
msg = (_('Unexpected VGPU resource allocation '
'on provider %(rp_uuid)s for consumer '
'%(consumer_uuid)s: %(alloc_data)s. '
'Expected VGPU allocation to be on root '
'compute node provider %(root_uuid)s.')
% {'rp_uuid': rp_uuid,
'consumer_uuid': consumer_uuid,
'alloc_data': alloc_data,
'root_uuid': root_node.uuid})
raise exception.ReshapeFailed(error=msg)
def _get_assigned_mdevs_for_reshape(
self, instance_uuid, rp_uuid, alloc_data):
"""Gets the mediated devices assigned to the instance during a reshape.
:param instance_uuid: UUID of the instance consuming VGPU resources
on this host.
:param rp_uuid: UUID of the resource provider with VGPU inventory being
consumed by the instance.
:param alloc_data: dict of allocation data for the instance consumer.
:return: list of mediated device UUIDs assigned to the instance
:raises: ReshapeFailed if the instance is not found in the hypervisor
or no mediated devices were found to be assigned to the instance
indicating VGPU allocations are out of sync with the hypervisor
"""
# FIXME(sbauza): We don't really need an Instance
# object, but given some libvirt.host logs needs
# to have an instance name, just provide a fake one
Instance = collections.namedtuple('Instance', ['uuid', 'name'])
instance = Instance(uuid=instance_uuid, name=instance_uuid)
mdevs = self._get_all_assigned_mediated_devices(instance)
# _get_all_assigned_mediated_devices returns {} if the instance is
# not found in the hypervisor
if not mdevs:
# If we found a VGPU allocation against a consumer
# which is not an instance, the only left case for
# Nova would be a migration but we don't support
# this at the moment.
msg = (_('Unexpected VGPU resource allocation on provider '
'%(rp_uuid)s for consumer %(consumer_uuid)s: '
'%(alloc_data)s. The allocation is made against a '
'non-existing instance or there are no devices assigned.')
% {'rp_uuid': rp_uuid, 'consumer_uuid': instance_uuid,
'alloc_data': alloc_data})
raise exception.ReshapeFailed(error=msg)
return mdevs
def _count_vgpus_per_pgpu(self, mdev_uuids):
"""Count the number of VGPUs per physical GPU mediated device.
:param mdev_uuids: List of physical GPU mediated device UUIDs.
:return: dict, keyed by PGPU device ID, to count of VGPUs on that
device
"""
vgpu_count_per_pgpu: ty.Dict[str, int] = collections.defaultdict(int)
for mdev_uuid in mdev_uuids:
# libvirt name is like mdev_00ead764_fdc0_46b6_8db9_2963f5c815b4
dev_name = libvirt_utils.mdev_uuid2name(mdev_uuid)
# Count how many vGPUs are in use for this instance
dev_info = self._get_mediated_device_information(dev_name)
pgpu_dev_id = dev_info['parent']
vgpu_count_per_pgpu[pgpu_dev_id] += 1
return vgpu_count_per_pgpu
@staticmethod
def _check_vgpu_allocations_match_real_use(
vgpu_count_per_pgpu, expected_usage, rp_uuid, consumer_uuid,
alloc_data):
"""Checks that the number of GPU devices assigned to the consumer
matches what is expected from the allocations in the placement service
and logs a warning if there is a mismatch.
:param vgpu_count_per_pgpu: dict, keyed by PGPU device ID, to count of
VGPUs on that device where each device is assigned to the consumer
(guest instance on this hypervisor)
:param expected_usage: The expected usage from placement for the
given resource provider and consumer
:param rp_uuid: UUID of the resource provider with VGPU inventory being
consumed by the instance
:param consumer_uuid: UUID of the consumer (instance) holding resource
allocations against the given rp_uuid provider
:param alloc_data: dict of allocation data for the instance consumer
"""
actual_usage = sum(vgpu_count_per_pgpu.values())
if actual_usage != expected_usage:
# Don't make it blocking, just make sure you actually correctly
# allocate the existing resources
LOG.warning(
'Unexpected VGPU resource allocation on provider %(rp_uuid)s '
'for consumer %(consumer_uuid)s: %(alloc_data)s. Allocations '
'(%(expected_usage)s) differ from actual use '
'(%(actual_usage)s).',
{'rp_uuid': rp_uuid, 'consumer_uuid': consumer_uuid,
'alloc_data': alloc_data, 'expected_usage': expected_usage,
'actual_usage': actual_usage})
def _reshape_vgpu_allocations(
self, rp_uuid, root_node, consumer_uuid, alloc_data, resources,
pgpu_rps):
"""Update existing VGPU allocations by moving them from the root node
provider to the child provider for the given VGPU provider.
:param rp_uuid: UUID of the VGPU resource provider with allocations
from consumer_uuid (should be the root node provider before
reshaping occurs)
:param root_node: ProviderData object for the root compute node
resource provider in the provider tree
:param consumer_uuid: UUID of the consumer (instance) with VGPU
allocations against the resource provider represented by rp_uuid
:param alloc_data: dict of allocation information for consumer_uuid
:param resources: dict, keyed by resource class, of resources allocated
to consumer_uuid from rp_uuid
:param pgpu_rps: dict, keyed by GPU device ID, to ProviderData object
representing that resource provider in the tree
:raises: ReshapeFailed if the reshape fails for whatever reason
"""
# We've found VGPU allocations on a provider. It should be the root
# node provider.
self._assert_is_root_provider(
rp_uuid, root_node, consumer_uuid, alloc_data)
# Find which physical GPU corresponds to this allocation.
mdev_uuids = self._get_assigned_mdevs_for_reshape(
consumer_uuid, rp_uuid, alloc_data)
vgpu_count_per_pgpu = self._count_vgpus_per_pgpu(mdev_uuids)
# We need to make sure we found all the mediated devices that
# correspond to an allocation.
self._check_vgpu_allocations_match_real_use(
vgpu_count_per_pgpu, resources[orc.VGPU],
rp_uuid, consumer_uuid, alloc_data)
# Add the VGPU allocation for each VGPU provider.
allocs = alloc_data['allocations']
for pgpu_dev_id, pgpu_rp in pgpu_rps.items():
vgpu_count = vgpu_count_per_pgpu[pgpu_dev_id]
if vgpu_count:
allocs[pgpu_rp.uuid] = {
'resources': {
orc.VGPU: vgpu_count
}
}
# And remove the VGPU allocation from the root node provider.
del resources[orc.VGPU]
def _reshape_gpu_resources(
self, allocations, root_node, pgpu_rps):
"""Reshapes the provider tree moving VGPU inventory from root to child
:param allocations:
Dict of allocation data of the form:
{ $CONSUMER_UUID: {
# The shape of each "allocations" dict below is identical
# to the return from GET /allocations/{consumer_uuid}
"allocations": {
$RP_UUID: {
"generation": $RP_GEN,
"resources": {
$RESOURCE_CLASS: $AMOUNT,
...
},
},
...
},
"project_id": $PROJ_ID,
"user_id": $USER_ID,
"consumer_generation": $CONSUMER_GEN,
},
...
}
:params root_node: The root node in the provider tree
:params pgpu_rps: dict, keyed by GPU device ID, to ProviderData object
representing that resource provider in the tree
"""
LOG.info('Reshaping tree; moving VGPU allocations from root '
'provider %s to child providers %s.', root_node.uuid,
pgpu_rps.values())
# For each consumer in the allocations dict, look for VGPU
# allocations and move them to the VGPU provider.
for consumer_uuid, alloc_data in allocations.items():
# Copy and iterate over the current set of providers to avoid
# modifying keys while iterating.
allocs = alloc_data['allocations']
for rp_uuid in list(allocs):
resources = allocs[rp_uuid]['resources']
if orc.VGPU in resources:
self._reshape_vgpu_allocations(
rp_uuid, root_node, consumer_uuid, alloc_data,
resources, pgpu_rps)
def _update_provider_tree_for_vgpu(self, provider_tree, nodename,
allocations=None):
"""Updates the provider tree for VGPU inventory.
Before Stein, VGPU inventory and allocations were on the root compute
node provider in the tree. Starting in Stein, the VGPU inventory is
on a child provider in the tree. As a result, this method will
"reshape" the tree if necessary on first start of this compute service
in Stein.
:param provider_tree: The ProviderTree to update.
:param nodename: The ComputeNode.hypervisor_hostname, also known as
the name of the root node provider in the tree for this host.
:param allocations: If not None, indicates a reshape was requested and
should be performed.
:raises: nova.exception.ReshapeNeeded if ``allocations`` is None and
the method determines a reshape of the tree is needed, i.e. VGPU
inventory and allocations must be migrated from the root node
provider to a child provider of VGPU resources in the tree.
:raises: nova.exception.ReshapeFailed if the requested tree reshape
fails for whatever reason.
"""
# First, check if this host actually has vGPU to reshape
inventories_dict = self._get_gpu_inventories()
if not inventories_dict:
return
# Check to see if the root compute node provider in the tree for
# this host already has VGPU inventory because if it does, and
# we're not currently reshaping (allocations is None), we need
# to indicate that a reshape is needed to move the VGPU inventory
# onto a child provider in the tree.
# Ensure GPU providers are in the ProviderTree for the given inventory.
pgpu_rps = self._ensure_pgpu_providers(
inventories_dict, provider_tree, nodename)
if self._is_reshape_needed_vgpu_on_root(provider_tree, nodename):
if allocations is None:
# We have old VGPU inventory on root RP, but we haven't yet
# allocations. That means we need to ask for a reshape.
LOG.info('Requesting provider tree reshape in order to move '
'VGPU inventory from the root compute node provider '
'%s to a child provider.', nodename)
raise exception.ReshapeNeeded()
# We have allocations, that means we already asked for a reshape
# and the Placement API returned us them. We now need to move
# those from the root RP to the needed children RPs.
root_node = provider_tree.data(nodename)
# Reshape VGPU provider inventory and allocations, moving them
# from the root node provider to the child providers.
self._reshape_gpu_resources(allocations, root_node, pgpu_rps)
# Only delete the root inventory once the reshape is done
if orc.VGPU in root_node.inventory:
del root_node.inventory[orc.VGPU]
provider_tree.update_inventory(nodename, root_node.inventory)
def _update_provider_tree_for_pcpu(self, provider_tree, nodename,
allocations=None):
"""Updates the provider tree for PCPU inventory.
Before Train, pinned instances consumed VCPU inventory just like
unpinned instances. Starting in Train, these instances now consume PCPU
inventory. The function can reshape the inventory, changing allocations
of VCPUs to PCPUs.
:param provider_tree: The ProviderTree to update.
:param nodename: The ComputeNode.hypervisor_hostname, also known as
the name of the root node provider in the tree for this host.
:param allocations: A dict, keyed by consumer UUID, of allocation
records, or None::
{
$CONSUMER_UUID: {
"allocations": {
$RP_UUID: {
"generation": $RP_GEN,
"resources": {
$RESOURCE_CLASS: $AMOUNT,
...
},
},
...
},
"project_id": $PROJ_ID,
"user_id": $USER_ID,
"consumer_generation": $CONSUMER_GEN,
},
...
}
If provided, this indicates a reshape was requested and should be
performed.
:raises: nova.exception.ReshapeNeeded if ``allocations`` is None and
the method determines a reshape of the tree is needed, i.e. VCPU
inventory and allocations must be migrated to PCPU resources.
:raises: nova.exception.ReshapeFailed if the requested tree reshape
fails for whatever reason.
"""
# If we're not configuring PCPUs, then we've nothing to worry about
# (yet)
if not CONF.compute.cpu_dedicated_set:
return
root_node = provider_tree.data(nodename)
# Similarly, if PCPU inventories are already reported then there is no
# need to reshape
if orc.PCPU in root_node.inventory:
return
ctx = nova_context.get_admin_context()
compute_node = objects.ComputeNode.get_by_nodename(ctx, nodename)
# Finally, if the compute node doesn't appear to support NUMA, move
# swiftly on
if not compute_node.numa_topology:
return
# The ComputeNode.numa_topology is a StringField, deserialize
numa = objects.NUMATopology.obj_from_db_obj(compute_node.numa_topology)
# If the host doesn't know of any pinned CPUs, we can continue
if not any(cell.pinned_cpus for cell in numa.cells):
return
# At this point, we know there's something to be migrated here but not
# how much. If the allocations are None, we're at the startup of the
# compute node and a Reshape is needed. Indicate this by raising the
# ReshapeNeeded exception
if allocations is None:
LOG.info(
'Requesting provider tree reshape in order to move '
'VCPU to PCPU allocations to the compute node '
'provider %s', nodename)
raise exception.ReshapeNeeded()
# Go figure out how many VCPUs to migrate to PCPUs. We've been telling
# people for years *not* to mix pinned and unpinned instances, meaning
# we should be able to move all VCPUs to PCPUs, but we never actually
# enforced this in code and there's an all-too-high chance someone
# didn't get the memo
allocations_needing_reshape = []
# we need to tackle the allocations against instances on this host...
instances = objects.InstanceList.get_by_host(
ctx, compute_node.host, expected_attrs=['numa_topology'])
for instance in instances:
if not instance.numa_topology:
continue
if instance.numa_topology.cpu_policy != (
fields.CPUAllocationPolicy.DEDICATED
):
continue
allocations_needing_reshape.append(instance.uuid)
# ...and those for any migrations
migrations = objects.MigrationList.get_in_progress_by_host_and_node(
ctx, compute_node.host, compute_node.hypervisor_hostname)
for migration in migrations:
# we don't care about migrations that have landed here, since we
# already have those instances above
if not migration.dest_compute or (
migration.dest_compute == compute_node.host):
continue
instance = objects.Instance.get_by_uuid(
ctx, migration.instance_uuid, expected_attrs=['numa_topology'])
if not instance.numa_topology:
continue
if instance.numa_topology.cpu_policy != (
fields.CPUAllocationPolicy.DEDICATED
):
continue
allocations_needing_reshape.append(migration.uuid)
for allocation_uuid in allocations_needing_reshape:
consumer_allocations = allocations.get(allocation_uuid, {}).get(
'allocations', {})
# TODO(stephenfin): We can probably just check the allocations for
# ComputeNode.uuid since compute nodes are the only (?) provider of
# VCPU and PCPU resources
for rp_uuid in consumer_allocations:
resources = consumer_allocations[rp_uuid]['resources']
if orc.PCPU in resources or orc.VCPU not in resources:
# Either this has been migrated or it's not a compute node
continue
# Switch stuff around. We can do a straight swap since an
# instance is either pinned or unpinned. By doing this, we're
# modifying the provided 'allocations' dict, which will
# eventually be used by the resource tracker to update
# placement
resources['PCPU'] = resources['VCPU']
del resources[orc.VCPU]
def get_available_resource(self, nodename):
"""Retrieve resource information.
This method is called when nova-compute launches, and
as part of a periodic task that records the results in the DB.
:param nodename: unused in this driver
:returns: dictionary containing resource info
"""
disk_info_dict = self._get_local_gb_info()
data = {}
# NOTE(dprince): calling capabilities before getVersion works around
# an initialization issue with some versions of Libvirt (1.0.5.5).
# See: https://bugzilla.redhat.com/show_bug.cgi?id=1000116
# See: https://bugs.launchpad.net/nova/+bug/1215593
data["supported_instances"] = self._get_instance_capabilities()
data["vcpus"] = len(self._get_vcpu_available())
data["memory_mb"] = self._host.get_memory_mb_total()
data["local_gb"] = disk_info_dict['total']
data["vcpus_used"] = self._get_vcpu_used()
data["memory_mb_used"] = self._host.get_memory_mb_used()
data["local_gb_used"] = disk_info_dict['used']
data["hypervisor_type"] = self._host.get_driver_type()
data["hypervisor_version"] = self._host.get_version()
data["hypervisor_hostname"] = self._host.get_hostname()
data["uuid"] = self._host.get_node_uuid()
# TODO(berrange): why do we bother converting the
# libvirt capabilities XML into a special JSON format ?
# The data format is different across all the drivers
# so we could just return the raw capabilities XML
# which 'compare_cpu' could use directly
#
# That said, arch_filter.py now seems to rely on
# the libvirt drivers format which suggests this
# data format needs to be standardized across drivers
data["cpu_info"] = jsonutils.dumps(self._get_cpu_info())
disk_free_gb = disk_info_dict['free']
disk_over_committed = self._get_disk_over_committed_size_total()
available_least = disk_free_gb * units.Gi - disk_over_committed
data['disk_available_least'] = available_least / units.Gi
data['pci_passthrough_devices'] = self._get_pci_passthrough_devices()
numa_topology = self._get_host_numa_topology()
if numa_topology:
data['numa_topology'] = numa_topology._to_json()
else:
data['numa_topology'] = None
return data
def check_instance_shared_storage_local(self, context, instance):
"""Check if instance files located on shared storage.
This runs check on the destination host, and then calls
back to the source host to check the results.
:param context: security context
:param instance: nova.objects.instance.Instance object
:returns:
- tempfile: A dict containing the tempfile info on the destination
host
- None:
1. If the instance path is not existing.
2. If the image backend is shared block storage type.
"""
if self.image_backend.backend().is_shared_block_storage():
return None
dirpath = libvirt_utils.get_instance_path(instance)
if not os.path.exists(dirpath):
return None
fd, tmp_file = tempfile.mkstemp(dir=dirpath)
LOG.debug("Creating tmpfile %s to verify with other "
"compute node that the instance is on "
"the same shared storage.",
tmp_file, instance=instance)
os.close(fd)
return {"filename": tmp_file}
def check_instance_shared_storage_remote(self, context, data):
return os.path.exists(data['filename'])
def check_instance_shared_storage_cleanup(self, context, data):
fileutils.delete_if_exists(data["filename"])
def check_can_live_migrate_destination(self, context, instance,
src_compute_info, dst_compute_info,
block_migration=False,
disk_over_commit=False):
"""Check if it is possible to execute live migration.
This runs checks on the destination host, and then calls
back to the source host to check the results.
:param context: security context
:param instance: nova.db.main.models.Instance
:param block_migration: if true, prepare for block migration
:param disk_over_commit: if true, allow disk over commit
:returns: a LibvirtLiveMigrateData object
"""
if disk_over_commit:
disk_available_gb = dst_compute_info['free_disk_gb']
else:
disk_available_gb = dst_compute_info['disk_available_least']
disk_available_mb = (
(disk_available_gb * units.Ki) - CONF.reserved_host_disk_mb)
if not CONF.workarounds.skip_cpu_compare_on_dest:
# Compare CPU
try:
if not instance.vcpu_model or not instance.vcpu_model.model:
source_cpu_info = src_compute_info['cpu_info']
self._compare_cpu(None, source_cpu_info, instance)
else:
self._compare_cpu(instance.vcpu_model, None, instance)
except exception.InvalidCPUInfo as e:
raise exception.MigrationPreCheckError(reason=e)
# Create file on storage, to be checked on source host
filename = self._create_shared_storage_test_file(instance)
data = objects.LibvirtLiveMigrateData()
data.filename = filename
data.image_type = CONF.libvirt.images_type
data.graphics_listen_addr_vnc = CONF.vnc.server_listen
data.graphics_listen_addr_spice = CONF.spice.server_listen
if CONF.serial_console.enabled:
data.serial_listen_addr = CONF.serial_console.proxyclient_address
else:
data.serial_listen_addr = None
# Notes(eliqiao): block_migration and disk_over_commit are not
# nullable, so just don't set them if they are None
if block_migration is not None:
data.block_migration = block_migration
if disk_over_commit is not None:
data.disk_over_commit = disk_over_commit
data.disk_available_mb = disk_available_mb
data.dst_wants_file_backed_memory = CONF.libvirt.file_backed_memory > 0
# TODO(artom) Set to indicate that the destination (us) can perform a
# NUMA-aware live migration. NUMA-aware live migration will become
# unconditionally supported in RPC 6.0, so this sentinel can be removed
# then.
if instance.numa_topology:
data.dst_supports_numa_live_migration = True
data.dst_cpu_shared_set_info = (
hardware.get_cpu_shared_set() or
hardware.get_vcpu_pin_set() or
set()
)
# NOTE(sean-k-mooney): The migrate_data vifs field is used to signal
# that we are using the multiple port binding workflow so we can only
# populate it if we are using multiple port bindings.
# TODO(stephenfin): Remove once we can do this unconditionally in X or
# later
if self._network_api.has_port_binding_extension(context):
data.vifs = (
migrate_data_obj.VIFMigrateData.create_skeleton_migrate_vifs(
instance.get_network_info()))
for vif in data.vifs:
vif.supports_os_vif_delegation = True
# Just flag the fact we can live-migrate mdevs even if we don't use
# them so the source will know we can use this compute.
if self._host_can_support_mdev_live_migration():
data.dst_supports_mdev_live_migration = True
return data
def check_source_migrate_data_at_dest(self, ctxt, instance, migrate_data,
migration, limits, allocs):
"""Runs the last checks on the destination after the source returned
the migrate_data.
:param ctxt: security context
:param instance: nova.db.main.models.Instance
:param migrate_data: result of check_can_live_migrate_source
:param migration: The Migration object for this live migration
:param limits: The SchedulerLimits object for this live migration
:param allocs: Allocations for this instance
:returns: a LibvirtLiveMigrateData object
:raises: MigrationPreCheckError
"""
if ('source_mdev_types' in migrate_data and
migrate_data.source_mdev_types):
# The instance that needs to be live-migrated has some mdevs
src_mdev_types = migrate_data.source_mdev_types
# As a reminder, src_mdev_types is a dict of mdev UUID and its type
# Are all the types supported by this compute ?
if not all(map(lambda m_type: m_type in self.supported_vgpu_types,
src_mdev_types.values())):
reason = (_('Unable to migrate %(instance_uuid)s: '
'Source mdev types %(src_types)s are not '
'supported by this compute : %(dest_types)s ' %
{'instance_uuid': instance.uuid,
'src_types': list(src_mdev_types.values()),
'dest_types': self.supported_vgpu_types}))
raise exception.MigrationPreCheckError(reason)
dst_mdevs = self._allocate_mdevs(allocs)
dst_mdev_types = self._get_mdev_types_from_uuids(dst_mdevs)
target_mdevs: ty.Dict[str, str] = {}
for src_mdev, src_type in src_mdev_types.items():
for dst_mdev, dst_type in dst_mdev_types.items():
# we want to associate by 1:1 between dst and src mdevs
if (src_type == dst_type and
src_type not in target_mdevs and
dst_mdev not in target_mdevs.values()):
target_mdevs[src_mdev] = dst_mdev
continue
if len(target_mdevs) != len(src_mdev_types):
reason = (_('Unable to migrate %(instance_uuid)s: '
'Source mdevs %(src_mdevs)s are not '
'fully mapped for this compute : %(targets)s ' %
{'instance_uuid': instance.uuid,
'src_mdevs': list(src_mdev_types.keys()),
'targets': target_mdevs}))
raise exception.MigrationPreCheckError(reason)
LOG.debug('Source mediated devices are now associated with those '
'existing mediated devices '
'(source uuid : dest uuid): %s', str(target_mdevs))
migrate_data.target_mdevs = target_mdevs
self.instance_claimed_mdevs[instance.uuid] = dst_mdevs
LOG.info("Current mediated devices reserved by this host "
"(instance UUID: list of reserved mdev UUIDs) : %s ",
self.instance_claimed_mdevs)
return migrate_data
def post_claim_migrate_data(self, context, instance, migrate_data, claim):
migrate_data.dst_numa_info = self._get_live_migrate_numa_info(
claim.claimed_numa_topology, claim.flavor, claim.image_meta)
return migrate_data
def _get_resources(self, instance, prefix=None):
resources: 'objects.ResourceList' = []
if prefix:
migr_context = instance.migration_context
attr_name = prefix + 'resources'
if migr_context and attr_name in migr_context:
resources = getattr(migr_context, attr_name) or []
else:
resources = instance.resources or []
return resources
def _get_vpmem_resources(self, resources):
vpmem_resources = []
for resource in resources:
if 'metadata' in resource and \
isinstance(resource.metadata, objects.LibvirtVPMEMDevice):
vpmem_resources.append(resource)
return vpmem_resources
def _get_ordered_vpmem_resources(self, resources, flavor):
vpmem_resources = self._get_vpmem_resources(resources)
ordered_vpmem_resources = []
labels = hardware.get_vpmems(flavor)
for label in labels:
for vpmem_resource in vpmem_resources:
if vpmem_resource.metadata.label == label:
ordered_vpmem_resources.append(vpmem_resource)
vpmem_resources.remove(vpmem_resource)
break
return ordered_vpmem_resources
def _sorted_migrating_resources(self, instance, flavor):
"""This method is used to sort instance.migration_context.new_resources
claimed on dest host, then the ordered new resources will be used to
update resources info (e.g. vpmems) in the new xml which is used for
live migration.
"""
resources = self._get_resources(instance, prefix='new_')
if not resources:
return
ordered_resources = []
ordered_vpmem_resources = self._get_ordered_vpmem_resources(
resources, flavor)
ordered_resources.extend(ordered_vpmem_resources)
ordered_resources_obj = objects.ResourceList(objects=ordered_resources)
return ordered_resources_obj
def _get_live_migrate_numa_info(self, instance_numa_topology, flavor,
image_meta):
"""Builds a LibvirtLiveMigrateNUMAInfo object to send to the source of
a live migration, containing information about how the instance is to
be pinned on the destination host.
:param instance_numa_topology: The InstanceNUMATopology as fitted to
the destination by the live migration
Claim.
:param flavor: The Flavor object for the instance.
:param image_meta: The ImageMeta object for the instance.
:returns: A LibvirtLiveMigrateNUMAInfo object indicating how to update
the XML for the destination host.
"""
info = objects.LibvirtLiveMigrateNUMAInfo()
cpu_set, guest_cpu_tune, guest_cpu_numa, guest_numa_tune = \
self._get_guest_numa_config(instance_numa_topology, flavor,
image_meta)
# NOTE(artom) These two should always be either None together, or
# truth-y together.
if guest_cpu_tune and guest_numa_tune:
info.cpu_pins = {}
for pin in guest_cpu_tune.vcpupin:
info.cpu_pins[str(pin.id)] = pin.cpuset
info.emulator_pins = guest_cpu_tune.emulatorpin.cpuset
if guest_cpu_tune.vcpusched:
# NOTE(artom) vcpusched is a list, but there's only ever one
# element in it (see _get_guest_numa_config under
# wants_realtime)
info.sched_vcpus = guest_cpu_tune.vcpusched[0].vcpus
info.sched_priority = guest_cpu_tune.vcpusched[0].priority
info.cell_pins = {}
for node in guest_numa_tune.memnodes:
info.cell_pins[str(node.cellid)] = set(node.nodeset)
LOG.debug('Built NUMA live migration info: %s', info)
return info
def cleanup_live_migration_destination_check(self, context,
dest_check_data):
"""Do required cleanup on dest host after check_can_live_migrate calls
:param context: security context
"""
filename = dest_check_data.filename
self._cleanup_shared_storage_test_file(filename)
def check_can_live_migrate_source(self, context, instance,
dest_check_data,
block_device_info=None):
"""Check if it is possible to execute live migration.
This checks if the live migration can succeed, based on the
results from check_can_live_migrate_destination.
:param context: security context
:param instance: nova.db.main.models.Instance
:param dest_check_data: result of check_can_live_migrate_destination
:param block_device_info: result of _get_instance_block_device_info
:returns: a LibvirtLiveMigrateData object
"""
# Checking shared storage connectivity
# if block migration, instances_path should not be on shared storage.
source = CONF.host
dest_check_data.is_shared_instance_path = (
self._check_shared_storage_test_file(
dest_check_data.filename, instance))
dest_check_data.is_shared_block_storage = (
self._is_shared_block_storage(instance, dest_check_data,
block_device_info))
if 'block_migration' not in dest_check_data:
dest_check_data.block_migration = (
not dest_check_data.is_on_shared_storage())
if dest_check_data.block_migration:
# TODO(eliqiao): Once block_migration flag is removed from the API
# we can safely remove the if condition
if dest_check_data.is_on_shared_storage():
reason = _("Block migration can not be used "
"with shared storage.")
raise exception.InvalidLocalStorage(reason=reason, path=source)
if 'disk_over_commit' in dest_check_data:
self._assert_dest_node_has_enough_disk(context, instance,
dest_check_data.disk_available_mb,
dest_check_data.disk_over_commit,
block_device_info)
if block_device_info:
bdm = block_device_info.get('block_device_mapping')
# NOTE(eliqiao): Selective disk migrations are not supported
# with tunnelled block migrations so we can block them early.
if (bdm and
(self._block_migration_flags &
libvirt.VIR_MIGRATE_TUNNELLED != 0)):
msg = (_('Cannot block migrate instance %(uuid)s with'
' mapped volumes. Selective block device'
' migration is not supported with tunnelled'
' block migrations.') % {'uuid': instance.uuid})
LOG.error(msg, instance=instance)
raise exception.MigrationPreCheckError(reason=msg)
elif not (dest_check_data.is_shared_block_storage or
dest_check_data.is_shared_instance_path):
reason = _("Shared storage live-migration requires either shared "
"storage or boot-from-volume with no local disks.")
raise exception.InvalidSharedStorage(reason=reason, path=source)
# NOTE(mikal): include the instance directory name here because it
# doesn't yet exist on the destination but we want to force that
# same name to be used
instance_path = libvirt_utils.get_instance_path(instance,
relative=True)
dest_check_data.instance_relative_path = instance_path
# TODO(artom) Set to indicate that the source (us) can perform a
# NUMA-aware live migration. NUMA-aware live migration will become
# unconditionally supported in RPC 6.0, so this sentinel can be removed
# then.
if instance.numa_topology:
dest_check_data.src_supports_numa_live_migration = True
# If we have mediated devices to live-migrate, just verify we can
# support them.
instance_mdevs = self._get_all_assigned_mediated_devices(instance)
if instance_mdevs:
# This can raise a MigrationPreCheckError if the target is too old
# or if the current QEMU or libvirt versions from this compute are
# too old (only if the current instance uses mdevs)
self._assert_source_can_live_migrate_mdevs(instance,
dest_check_data)
mdev_types = self._get_mdev_types_from_uuids(instance_mdevs.keys())
dest_check_data.source_mdev_types = mdev_types
return dest_check_data
def _host_can_support_mdev_live_migration(self):
return self._host.has_min_version(
lv_ver=MIN_MDEV_LIVEMIG_LIBVIRT_VERSION,
hv_ver=MIN_MDEV_LIVEMIG_QEMU_VERSION,
hv_type=host.HV_DRIVER_QEMU,
)
def _assert_source_can_live_migrate_mdevs(self, instance, dest_check_data):
"""Check if the source can live migrate the instance by looking at the
QEMU and libvirt versions but also at the destination object.
:param instance: nova.objects.instance.Instance object
:param migrate_data: nova.objects.LibvirtLiveMigrateData object
:raises: MigrationPreCheckError if the versions are too old or if the
dst_supports_mdev_live_migration sentinel is not True.
"""
failed = ''
if not self._host_can_support_mdev_live_migration():
failed = 'source'
elif not ('dst_supports_mdev_live_migration' in dest_check_data and
dest_check_data.dst_supports_mdev_live_migration):
failed = 'target'
if failed:
reason = (_('Unable to migrate %(instance_uuid)s: '
'Either libvirt or QEMU version for compute service '
'%(host)s are too old than the supported ones : '
'(QEMU: %(qemu_v)s, libvirt: %(libv_v)s)' %
{'instance_uuid': instance.uuid,
'host': failed,
'qemu_v': libvirt_utils.version_to_string(
MIN_MDEV_LIVEMIG_QEMU_VERSION),
'libv_v': libvirt_utils.version_to_string(
MIN_MDEV_LIVEMIG_LIBVIRT_VERSION)}))
raise exception.MigrationPreCheckError(reason=reason)
def _is_shared_block_storage(self, instance, dest_check_data,
block_device_info=None):
"""Check if all block storage of an instance can be shared
between source and destination of a live migration.
Returns true if the instance is volume backed and has no local disks,
or if the image backend is the same on source and destination and the
backend shares block storage between compute nodes.
:param instance: nova.objects.instance.Instance object
:param dest_check_data: dict with boolean fields image_type,
is_shared_instance_path, and is_volume_backed
"""
if (dest_check_data.obj_attr_is_set('image_type') and
CONF.libvirt.images_type == dest_check_data.image_type and
self.image_backend.backend().is_shared_block_storage()):
# NOTE(dgenin): currently true only for RBD image backend
return True
if (dest_check_data.is_shared_instance_path and
self.image_backend.backend().is_file_in_instance_path()):
# NOTE(angdraug): file based image backends (Flat, Qcow2)
# place block device files under the instance path
return True
if (dest_check_data.is_volume_backed and
not bool(self._get_instance_disk_info(instance,
block_device_info))):
return True
return False
def _assert_dest_node_has_enough_disk(self, context, instance,
available_mb, disk_over_commit,
block_device_info):
"""Checks if destination has enough disk for block migration."""
# Libvirt supports qcow2 disk format,which is usually compressed
# on compute nodes.
# Real disk image (compressed) may enlarged to "virtual disk size",
# that is specified as the maximum disk size.
# (See qemu-img -f path-to-disk)
# Scheduler recognizes destination host still has enough disk space
# if real disk size < available disk size
# if disk_over_commit is True,
# otherwise virtual disk size < available disk size.
available = 0
if available_mb:
available = available_mb * units.Mi
disk_infos = self._get_instance_disk_info(instance, block_device_info)
necessary = 0
if disk_over_commit:
for info in disk_infos:
necessary += int(info['disk_size'])
else:
for info in disk_infos:
necessary += int(info['virt_disk_size'])
# Check that available disk > necessary disk
if (available - necessary) < 0:
reason = (_('Unable to migrate %(instance_uuid)s: '
'Disk of instance is too large(available'
' on destination host:%(available)s '
'< need:%(necessary)s)') %
{'instance_uuid': instance.uuid,
'available': available,
'necessary': necessary})
raise exception.MigrationPreCheckError(reason=reason)
def _compare_cpu(self, guest_cpu, host_cpu_str, instance):
"""Check the host is compatible with the requested CPU
:param guest_cpu: nova.objects.VirtCPUModel
or nova.virt.libvirt.vconfig.LibvirtConfigGuestCPU or None.
:param host_cpu_str: JSON from _get_cpu_info() method
If the 'guest_cpu' parameter is not None, this will be
validated for migration compatibility with the host.
Otherwise the 'host_cpu_str' JSON string will be used for
validation.
:returns:
None. if given cpu info is not compatible to this server,
raise exception.
"""
# NOTE(kchamart): Comparing host to guest CPU model for emulated
# guests (<domain type='qemu'>) should not matter -- in this
# mode (QEMU "TCG") the CPU is fully emulated in software and no
# hardware acceleration, like KVM, is involved. So, skip the CPU
# compatibility check for the QEMU domain type, and retain it for
# KVM guests.
if CONF.libvirt.virt_type not in ['kvm']:
return
if guest_cpu is None:
info = jsonutils.loads(host_cpu_str)
LOG.info('Instance launched has CPU info: %s', host_cpu_str)
cpu = vconfig.LibvirtConfigCPU()
cpu.arch = info['arch']
cpu.model = info['model']
cpu.vendor = info['vendor']
cpu.sockets = info['topology']['sockets']
cpu.cores = info['topology']['cores']
cpu.threads = info['topology']['threads']
for f in info['features']:
cpu.add_feature(vconfig.LibvirtConfigCPUFeature(f))
elif isinstance(guest_cpu, vconfig.LibvirtConfigGuestCPU):
cpu = guest_cpu
else:
cpu = self._vcpu_model_to_cpu_config(guest_cpu)
host_cpu = self._host.get_capabilities().host.cpu
if host_cpu.arch == fields.Architecture.AARCH64:
LOG.debug("On AArch64 hosts, source and destination host "
"CPUs are compared to check if they're compatible"
"(the only use-case supported by libvirt for "
"Arm64/AArch64)")
cpu = host_cpu
u = ("http://libvirt.org/html/libvirt-libvirt-host.html#"
"virCPUCompareResult")
m = _("CPU doesn't have compatibility.\n\n%(ret)s\n\nRefer to %(u)s")
# unknown character exists in xml, then libvirt complains
try:
cpu_xml = cpu.to_xml()
LOG.debug("cpu compare xml: %s", cpu_xml, instance=instance)
ret = self._host.compare_hypervisor_cpu(cpu_xml)
except libvirt.libvirtError as e:
error_code = e.get_error_code()
if error_code == libvirt.VIR_ERR_NO_SUPPORT:
LOG.debug("URI %(uri)s does not support cpu comparison. "
"It will be proceeded though. Error: %(error)s",
{'uri': self._uri(), 'error': e})
return
else:
LOG.error(m, {'ret': e, 'u': u})
raise exception.InvalidCPUInfo(
reason=m % {'ret': e, 'u': u})
if ret <= 0:
LOG.error(m, {'ret': ret, 'u': u})
raise exception.InvalidCPUInfo(reason=m % {'ret': ret, 'u': u})
def _create_shared_storage_test_file(self, instance):
"""Makes tmpfile under CONF.instances_path."""
dirpath = CONF.instances_path
fd, tmp_file = tempfile.mkstemp(dir=dirpath)
LOG.debug("Creating tmpfile %s to notify to other "
"compute nodes that they should mount "
"the same storage.", tmp_file, instance=instance)
os.close(fd)
return os.path.basename(tmp_file)
def _check_shared_storage_test_file(self, filename, instance):
"""Confirms existence of the tmpfile under CONF.instances_path.
Cannot confirm tmpfile return False.
"""
# NOTE(tpatzig): if instances_path is a shared volume that is
# under heavy IO (many instances on many compute nodes),
# then checking the existence of the testfile fails,
# just because it takes longer until the client refreshes and new
# content gets visible.
# os.utime (like touch) on the directory forces the client to refresh.
os.utime(CONF.instances_path, None)
tmp_file = os.path.join(CONF.instances_path, filename)
if not os.path.exists(tmp_file):
exists = False
else:
exists = True
LOG.debug('Check if temp file %s exists to indicate shared storage '
'is being used for migration. Exists? %s', tmp_file, exists,
instance=instance)
return exists
def _cleanup_shared_storage_test_file(self, filename):
"""Removes existence of the tmpfile under CONF.instances_path."""
tmp_file = os.path.join(CONF.instances_path, filename)
os.remove(tmp_file)
def live_migration(self, context, instance, dest,
post_method, recover_method, block_migration=False,
migrate_data=None):
"""Spawning live_migration operation for distributing high-load.
:param context: security context
:param instance:
nova.db.main.models.Instance object
instance object that is migrated.
:param dest: destination host
:param post_method:
post operation method.
expected nova.compute.manager._post_live_migration.
:param recover_method:
recovery method when any exception occurs.
expected nova.compute.manager._rollback_live_migration.
:param block_migration: if true, do block migration.
:param migrate_data: a LibvirtLiveMigrateData object
"""
# 'dest' will be substituted into 'migration_uri' so ensure
# it doesn't contain any characters that could be used to
# exploit the URI accepted by libvirt
if not libvirt_utils.is_valid_hostname(dest):
raise exception.InvalidHostname(hostname=dest)
self._live_migration(context, instance, dest,
post_method, recover_method, block_migration,
migrate_data)
def live_migration_abort(self, instance):
"""Aborting a running live-migration.
:param instance: instance object that is in migration
"""
guest = self._host.get_guest(instance)
dom = guest._domain
try:
dom.abortJob()
except libvirt.libvirtError as e:
LOG.error("Failed to cancel migration %s",
e, instance=instance)
raise
def _verify_serial_console_is_disabled(self):
if CONF.serial_console.enabled:
msg = _('Your destination node does not support'
' retrieving listen addresses. In order'
' for live migration to work properly you'
' must disable serial console.')
raise exception.MigrationError(reason=msg)
def _detach_direct_passthrough_vifs(self, context,
migrate_data, instance):
"""detaches passthrough vif to enable live migration
:param context: security context
:param migrate_data: a LibvirtLiveMigrateData object
:param instance: instance object that is migrated.
"""
# NOTE(sean-k-mooney): if we have vif data available we
# loop over each vif and detach all direct passthrough
# vifs to allow sriov live migration.
direct_vnics = network_model.VNIC_TYPES_DIRECT_PASSTHROUGH
vifs = [vif.source_vif for vif in migrate_data.vifs
if "source_vif" in vif and vif.source_vif]
for vif in vifs:
if vif['vnic_type'] in direct_vnics:
LOG.info("Detaching vif %s from instance "
"%s for live migration", vif['id'], instance.id)
self.detach_interface(context, instance, vif)
def _live_migration_operation(self, context, instance, dest,
block_migration, migrate_data, guest,
device_names):
"""Invoke the live migration operation
:param context: security context
:param instance:
nova.db.main.models.Instance object
instance object that is migrated.
:param dest: destination host
:param block_migration: if true, do block migration.
:param migrate_data: a LibvirtLiveMigrateData object
:param guest: the guest domain object
:param device_names: list of device names that are being migrated with
instance
This method is intended to be run in a background thread and will
block that thread until the migration is finished or failed.
"""
try:
if migrate_data.block_migration:
migration_flags = self._block_migration_flags
else:
migration_flags = self._live_migration_flags
# Note(siva_krishnan): live migrating paused instance fails
# when VIR_MIGRATE_POSTCOPY flag is set. It is unset here
# to permit live migration of paused instance.
if (
instance.vm_state == vm_states.PAUSED and
self._is_post_copy_enabled(migration_flags)
):
LOG.debug('Post-copy flag unset because instance is paused.',
instance=instance)
migration_flags ^= libvirt.VIR_MIGRATE_POSTCOPY
if not migrate_data.serial_listen_addr:
# In this context we want to ensure that serial console is
# disabled on source node. This is because nova couldn't
# retrieve serial listen address from destination node, so we
# consider that destination node might have serial console
# disabled as well.
self._verify_serial_console_is_disabled()
# NOTE(aplanas) migrate_uri will have a value only in the
# case that `live_migration_inbound_addr` parameter is
# set, and we propose a non tunneled migration.
migrate_uri = None
if ('target_connect_addr' in migrate_data and
migrate_data.target_connect_addr is not None):
dest = migrate_data.target_connect_addr
if (migration_flags &
libvirt.VIR_MIGRATE_TUNNELLED == 0):
migrate_uri = self._migrate_uri(dest)
new_xml_str = None
if CONF.libvirt.virt_type != "parallels":
# If the migrate_data has port binding information for the
# destination host, we need to prepare the guest vif config
# for the destination before we start migrating the guest.
get_vif_config = None
if 'vifs' in migrate_data and migrate_data.vifs:
# NOTE(mriedem): The vif kwarg must be built on the fly
# within get_updated_guest_xml based on migrate_data.vifs.
# We could stash the virt_type from the destination host
# into LibvirtLiveMigrateData but the host kwarg is a
# nova.virt.libvirt.host.Host object and is used to check
# information like libvirt version on the destination.
# If this becomes a problem, what we could do is get the
# VIF configs while on the destination host during
# pre_live_migration() and store those in the
# LibvirtLiveMigrateData object. For now we just use the
# source host information for virt_type and
# host (version) since the conductor live_migrate method
# _check_compatible_with_source_hypervisor() ensures that
# the hypervisor types and versions are compatible.
get_vif_config = functools.partial(
self.vif_driver.get_config,
instance=instance,
image_meta=instance.image_meta,
flavor=instance.flavor,
virt_type=CONF.libvirt.virt_type,
)
self._detach_direct_passthrough_vifs(context,
migrate_data, instance)
new_resources = None
if isinstance(instance, objects.Instance):
new_resources = self._sorted_migrating_resources(
instance, instance.flavor)
new_xml_str = libvirt_migrate.get_updated_guest_xml(
# TODO(sahid): It's not a really good idea to pass
# the method _get_volume_config and we should to find
# a way to avoid this in future.
instance, guest, migrate_data, self._get_volume_config,
get_vif_config=get_vif_config, new_resources=new_resources)
# NOTE(pkoniszewski): Because of precheck which blocks
# tunnelled block live migration with mapped volumes we
# can safely remove migrate_disks when tunnelling is on.
# Otherwise we will block all tunnelled block migrations,
# even when an instance does not have volumes mapped.
# This is because selective disk migration is not
# supported in tunnelled block live migration. Also we
# cannot fallback to migrateToURI2 in this case because of
# bug #1398999
#
# TODO(kchamart) Move the following bit to guest.migrate()
if (migration_flags & libvirt.VIR_MIGRATE_TUNNELLED != 0):
device_names = []
# TODO(sahid): This should be in
# post_live_migration_at_source but no way to retrieve
# ports acquired on the host for the guest at this
# step. Since the domain is going to be removed from
# libvird on source host after migration, we backup the
# serial ports to release them if all went well.
serial_ports = []
if CONF.serial_console.enabled:
serial_ports = list(self._get_serial_ports_from_guest(guest))
LOG.debug("About to invoke the migrate API", instance=instance)
guest.migrate(self._live_migration_uri(dest),
migrate_uri=migrate_uri,
flags=migration_flags,
migrate_disks=device_names,
destination_xml=new_xml_str,
bandwidth=CONF.libvirt.live_migration_bandwidth)
LOG.debug("Migrate API has completed", instance=instance)
for hostname, port in serial_ports:
serial_console.release_port(host=hostname, port=port)
except Exception as e:
with excutils.save_and_reraise_exception():
LOG.error("Live Migration failure: %s", e, instance=instance)
# If 'migrateToURI' fails we don't know what state the
# VM instances on each host are in. Possibilities include
#
# 1. src==running, dst==none
#
# Migration failed & rolled back, or never started
#
# 2. src==running, dst==paused
#
# Migration started but is still ongoing
#
# 3. src==paused, dst==paused
#
# Migration data transfer completed, but switchover
# is still ongoing, or failed
#
# 4. src==paused, dst==running
#
# Migration data transfer completed, switchover
# happened but cleanup on source failed
#
# 5. src==none, dst==running
#
# Migration fully succeeded.
#
# Libvirt will aim to complete any migration operation
# or roll it back. So even if the migrateToURI call has
# returned an error, if the migration was not finished
# libvirt should clean up.
#
# So we take the error raise here with a pinch of salt
# and rely on the domain job info status to figure out
# what really happened to the VM, which is a much more
# reliable indicator.
#
# In particular we need to try very hard to ensure that
# Nova does not "forget" about the guest. ie leaving it
# running on a different host to the one recorded in
# the database, as that would be a serious resource leak
LOG.debug("Migration operation thread has finished",
instance=instance)
def _live_migration_copy_disk_paths(self, context, instance, guest):
'''Get list of disks to copy during migration
:param context: security context
:param instance: the instance being migrated
:param guest: the Guest instance being migrated
Get the list of disks to copy during migration.
:returns: a list of local source paths and a list of device names to
copy
'''
disk_paths = []
device_names = []
block_devices = []
if (self._block_migration_flags &
libvirt.VIR_MIGRATE_TUNNELLED == 0):
bdm_list = objects.BlockDeviceMappingList.get_by_instance_uuid(
context, instance.uuid)
block_device_info = driver.get_block_device_info(instance,
bdm_list)
block_device_mappings = driver.block_device_info_get_mapping(
block_device_info)
for bdm in block_device_mappings:
device_name = str(bdm['mount_device'].rsplit('/', 1)[1])
block_devices.append(device_name)
for dev in guest.get_all_disks():
if dev.readonly or dev.shareable:
continue
if dev.source_type not in ["file", "block"]:
continue
if dev.target_dev in block_devices:
continue
disk_paths.append(dev.source_path)
device_names.append(dev.target_dev)
return (disk_paths, device_names)
def _live_migration_data_gb(self, instance, disk_paths):
'''Calculate total amount of data to be transferred
:param instance: the nova.objects.Instance being migrated
:param disk_paths: list of disk paths that are being migrated
with instance
Calculates the total amount of data that needs to be
transferred during the live migration. The actual
amount copied will be larger than this, due to the
guest OS continuing to dirty RAM while the migration
is taking place. So this value represents the minimal
data size possible.
:returns: data size to be copied in GB
'''
ram_gb = instance.flavor.memory_mb * units.Mi / units.Gi
if ram_gb < 2:
ram_gb = 2
disk_gb = 0
for path in disk_paths:
try:
size = os.stat(path).st_size
size_gb = (size / units.Gi)
if size_gb < 2:
size_gb = 2
disk_gb += size_gb
except OSError as e:
LOG.warning("Unable to stat %(disk)s: %(ex)s",
{'disk': path, 'ex': e})
# Ignore error since we don't want to break
# the migration monitoring thread operation
return ram_gb + disk_gb
def _get_migration_flags(self, is_block_migration):
if is_block_migration:
return self._block_migration_flags
return self._live_migration_flags
def _live_migration_monitor(self, context, instance, guest,
dest, post_method,
recover_method, block_migration,
migrate_data, finish_event,
disk_paths):
on_migration_failure: ty.Deque[str] = deque()
data_gb = self._live_migration_data_gb(instance, disk_paths)
downtime_steps = list(libvirt_migrate.downtime_steps(data_gb))
migration = migrate_data.migration
curdowntime = None
migration_flags = self._get_migration_flags(
migrate_data.block_migration)
n = 0
start = time.time()
is_post_copy_enabled = self._is_post_copy_enabled(migration_flags)
# vpmem does not support post copy
is_post_copy_enabled &= not bool(self._get_vpmems(instance))
while True:
info = guest.get_job_info()
if info.type == libvirt.VIR_DOMAIN_JOB_NONE:
# Either still running, or failed or completed,
# lets untangle the mess
if not finish_event.ready():
LOG.debug("Operation thread is still running",
instance=instance)
else:
info.type = libvirt_migrate.find_job_type(guest, instance)
LOG.debug("Fixed incorrect job type to be %d",
info.type, instance=instance)
if info.type == libvirt.VIR_DOMAIN_JOB_NONE:
# Migration is not yet started
LOG.debug("Migration not running yet",
instance=instance)
elif info.type == libvirt.VIR_DOMAIN_JOB_UNBOUNDED:
# Migration is still running
#
# This is where we wire up calls to change live
# migration status. eg change max downtime, cancel
# the operation, change max bandwidth
libvirt_migrate.run_tasks(guest, instance,
self.active_migrations,
on_migration_failure,
migration,
is_post_copy_enabled)
now = time.time()
elapsed = now - start
completion_timeout = int(
CONF.libvirt.live_migration_completion_timeout * data_gb)
# NOTE(yikun): Check the completion timeout to determine
# should trigger the timeout action, and there are two choices
# ``abort`` (default) or ``force_complete``. If the action is
# set to ``force_complete``, the post-copy will be triggered
# if available else the VM will be suspended, otherwise the
# live migrate operation will be aborted.
if libvirt_migrate.should_trigger_timeout_action(
instance, elapsed, completion_timeout,
migration.status):
timeout_act = CONF.libvirt.live_migration_timeout_action
if timeout_act == 'force_complete':
self.live_migration_force_complete(instance)
else:
# timeout action is 'abort'
try:
guest.abort_job()
except libvirt.libvirtError as e:
LOG.warning("Failed to abort migration %s",
e,
instance=instance)
self._clear_empty_migration(instance)
raise
curdowntime = libvirt_migrate.update_downtime(
guest, instance, curdowntime,
downtime_steps, elapsed)
# We loop every 500ms, so don't log on every
# iteration to avoid spamming logs for long
# running migrations. Just once every 5 secs
# is sufficient for developers to debug problems.
# We log once every 30 seconds at info to help
# admins see slow running migration operations
# when debug logs are off.
if (n % 10) == 0:
# Ignoring memory_processed, as due to repeated
# dirtying of data, this can be way larger than
# memory_total. Best to just look at what's
# remaining to copy and ignore what's done already
#
# TODO(berrange) perhaps we could include disk
# transfer stats in the progress too, but it
# might make memory info more obscure as large
# disk sizes might dwarf memory size
remaining = 100
if info.memory_total != 0:
remaining = round(info.memory_remaining *
100 / info.memory_total)
libvirt_migrate.save_stats(instance, migration,
info, remaining)
# NOTE(fanzhang): do not include disk transfer stats in
# the progress percentage calculation but log them.
disk_remaining = 100
if info.disk_total != 0:
disk_remaining = round(info.disk_remaining *
100 / info.disk_total)
lg = LOG.debug
if (n % 60) == 0:
lg = LOG.info
lg("Migration running for %(secs)d secs, "
"memory %(remaining)d%% remaining "
"(bytes processed=%(processed_memory)d, "
"remaining=%(remaining_memory)d, "
"total=%(total_memory)d); "
"disk %(disk_remaining)d%% remaining "
"(bytes processed=%(processed_disk)d, "
"remaining=%(remaining_disk)d, "
"total=%(total_disk)d).",
{"secs": elapsed, "remaining": remaining,
"processed_memory": info.memory_processed,
"remaining_memory": info.memory_remaining,
"total_memory": info.memory_total,
"disk_remaining": disk_remaining,
"processed_disk": info.disk_processed,
"remaining_disk": info.disk_remaining,
"total_disk": info.disk_total}, instance=instance)
n = n + 1
elif info.type == libvirt.VIR_DOMAIN_JOB_COMPLETED:
# Migration is all done
LOG.info("Migration operation has completed",
instance=instance)
post_method(context, instance, dest, block_migration,
migrate_data)
break
elif info.type == libvirt.VIR_DOMAIN_JOB_FAILED:
# Migration did not succeed
LOG.error("Migration operation has aborted", instance=instance)
libvirt_migrate.run_recover_tasks(self._host, guest, instance,
on_migration_failure)
recover_method(context, instance, dest, migrate_data)
break
elif info.type == libvirt.VIR_DOMAIN_JOB_CANCELLED:
# Migration was stopped by admin
LOG.warning("Migration operation was cancelled",
instance=instance)
libvirt_migrate.run_recover_tasks(self._host, guest, instance,
on_migration_failure)
recover_method(context, instance, dest, migrate_data,
migration_status='cancelled')
break
else:
LOG.warning("Unexpected migration job type: %d",
info.type, instance=instance)
time.sleep(0.5)
self._clear_empty_migration(instance)
def _clear_empty_migration(self, instance):
try:
del self.active_migrations[instance.uuid]
except KeyError:
LOG.warning("There are no records in active migrations "
"for instance", instance=instance)
def _live_migration(self, context, instance, dest, post_method,
recover_method, block_migration,
migrate_data):
"""Do live migration.
:param context: security context
:param instance:
nova.db.main.models.Instance object
instance object that is migrated.
:param dest: destination host
:param post_method:
post operation method.
expected nova.compute.manager._post_live_migration.
:param recover_method:
recovery method when any exception occurs.
expected nova.compute.manager._rollback_live_migration.
:param block_migration: if true, do block migration.
:param migrate_data: a LibvirtLiveMigrateData object
This fires off a new thread to run the blocking migration
operation, and then this thread monitors the progress of
migration and controls its operation
"""
guest = self._host.get_guest(instance)
disk_paths = []
device_names = []
if (migrate_data.block_migration and
CONF.libvirt.virt_type != "parallels"):
disk_paths, device_names = self._live_migration_copy_disk_paths(
context, instance, guest)
opthread = utils.spawn(self._live_migration_operation,
context, instance, dest,
block_migration,
migrate_data, guest,
device_names)
finish_event = eventlet.event.Event()
self.active_migrations[instance.uuid] = deque()
def thread_finished(thread, event):
LOG.debug("Migration operation thread notification",
instance=instance)
event.send()
opthread.link(thread_finished, finish_event)
# Let eventlet schedule the new thread right away
time.sleep(0)
try:
LOG.debug("Starting monitoring of live migration",
instance=instance)
self._live_migration_monitor(context, instance, guest, dest,
post_method, recover_method,
block_migration, migrate_data,
finish_event, disk_paths)
except Exception as ex:
LOG.warning("Error monitoring migration: %(ex)s",
{"ex": ex}, instance=instance, exc_info=True)
# NOTE(aarents): Ensure job is aborted if still running before
# raising the exception so this would avoid the migration to be
# done and the libvirt guest to be resumed on the target while
# the instance record would still related to the source host.
try:
# If migration is running in post-copy mode and guest
# already running on dest host, libvirt will refuse to
# cancel migration job.
self.live_migration_abort(instance)
except libvirt.libvirtError:
LOG.warning("Error occurred when trying to abort live ",
"migration job, ignoring it.", instance=instance)
raise
finally:
LOG.debug("Live migration monitoring is all done",
instance=instance)
def _is_post_copy_enabled(self, migration_flags):
return (migration_flags & libvirt.VIR_MIGRATE_POSTCOPY) != 0
def live_migration_force_complete(self, instance):
try:
self.active_migrations[instance.uuid].append('force-complete')
except KeyError:
raise exception.NoActiveMigrationForInstance(
instance_id=instance.uuid)
def _try_fetch_image(self, context, path, image_id, instance,
fallback_from_host=None):
try:
libvirt_utils.fetch_image(context, path, image_id,
instance.trusted_certs)
except exception.ImageNotFound:
if not fallback_from_host:
raise
LOG.debug("Image %(image_id)s doesn't exist anymore on "
"image service, attempting to copy image "
"from %(host)s",
{'image_id': image_id, 'host': fallback_from_host})
libvirt_utils.copy_image(src=path, dest=path,
host=fallback_from_host,
receive=True)
def _fetch_instance_kernel_ramdisk(self, context, instance,
fallback_from_host=None):
"""Download kernel and ramdisk for instance in instance directory."""
instance_dir = libvirt_utils.get_instance_path(instance)
if instance.kernel_id:
kernel_path = os.path.join(instance_dir, 'kernel')
# NOTE(dsanders): only fetch image if it's not available at
# kernel_path. This also avoids ImageNotFound exception if
# the image has been deleted from glance
if not os.path.exists(kernel_path):
self._try_fetch_image(context,
kernel_path,
instance.kernel_id,
instance, fallback_from_host)
if instance.ramdisk_id:
ramdisk_path = os.path.join(instance_dir, 'ramdisk')
# NOTE(dsanders): only fetch image if it's not available at
# ramdisk_path. This also avoids ImageNotFound exception if
# the image has been deleted from glance
if not os.path.exists(ramdisk_path):
self._try_fetch_image(context,
ramdisk_path,
instance.ramdisk_id,
instance, fallback_from_host)
def _reattach_instance_vifs(self, context, instance, network_info):
guest = self._host.get_guest(instance)
# validate that the guest has the expected number of interfaces
# attached.
guest_interfaces = guest.get_interfaces()
# NOTE(sean-k-mooney): In general len(guest_interfaces) will
# be equal to len(network_info) as interfaces will not be hot unplugged
# unless they are SR-IOV direct mode interfaces. As such we do not
# need an else block here as it would be a noop.
if len(guest_interfaces) < len(network_info):
# NOTE(sean-k-mooney): we are doing a post live migration
# for a guest with sriov vif that were detached as part of
# the migration. loop over the vifs and attach the missing
# vif as part of the post live migration phase.
direct_vnics = network_model.VNIC_TYPES_DIRECT_PASSTHROUGH
for vif in network_info:
if vif['vnic_type'] in direct_vnics:
LOG.info("Attaching vif %s to instance %s",
vif['id'], instance.id)
self.attach_interface(context, instance,
instance.image_meta, vif)
def rollback_live_migration_at_source(self, context, instance,
migrate_data):
"""reconnect sriov interfaces after failed live migration
:param context: security context
:param instance: the instance being migrated
:param migrate_date: a LibvirtLiveMigrateData object
"""
# NOTE(artom) migrate_data.vifs might not be set if our Neutron doesn't
# have the multiple port bindings extension.
if 'vifs' in migrate_data and migrate_data.vifs:
network_info = network_model.NetworkInfo(
[vif.source_vif for vif in migrate_data.vifs
if "source_vif" in vif and vif.source_vif])
self._reattach_instance_vifs(context, instance, network_info)
def rollback_live_migration_at_destination(self, context, instance,
network_info,
block_device_info,
destroy_disks=True,
migrate_data=None):
"""Clean up destination node after a failed live migration."""
try:
self.destroy(context, instance, network_info, block_device_info,
destroy_disks)
finally:
# NOTE(gcb): Failed block live migration may leave instance
# directory at destination node, ensure it is always deleted.
is_shared_instance_path = True
if migrate_data:
is_shared_instance_path = migrate_data.is_shared_instance_path
if (migrate_data.obj_attr_is_set("serial_listen_ports") and
migrate_data.serial_listen_ports):
# Releases serial ports reserved.
for port in migrate_data.serial_listen_ports:
serial_console.release_port(
host=migrate_data.serial_listen_addr, port=port)
if (
'dst_numa_info' in migrate_data and
migrate_data.dst_numa_info
):
self.cpu_api.power_down_for_migration(
migrate_data.dst_numa_info)
else:
LOG.debug('No dst_numa_info in migrate_data, '
'no cores to power down in rollback.')
if not is_shared_instance_path:
instance_dir = libvirt_utils.get_instance_path_at_destination(
instance, migrate_data)
if os.path.exists(instance_dir):
shutil.rmtree(instance_dir)
mdevs = self.instance_claimed_mdevs.pop(instance.uuid, None)
if mdevs:
# The live migration is aborted, we need to remove the reserved
# values.
LOG.debug("Unclaiming mdevs %s from instance %s",
mdevs, instance.uuid)
def _pre_live_migration_plug_vifs(self, instance, network_info,
migrate_data):
if 'vifs' in migrate_data and migrate_data.vifs:
LOG.debug('Plugging VIFs using destination host port bindings '
'before live migration.', instance=instance)
vif_plug_nw_info = network_model.NetworkInfo([])
for migrate_vif in migrate_data.vifs:
vif_plug_nw_info.append(migrate_vif.get_dest_vif())
else:
LOG.debug('Plugging VIFs before live migration.',
instance=instance)
vif_plug_nw_info = network_info
# Retry operation is necessary because continuous live migration
# requests to the same host cause concurrent requests to iptables,
# then it complains.
max_retry = CONF.live_migration_retry_count
for cnt in range(max_retry):
try:
self.plug_vifs(instance, vif_plug_nw_info)
break
except processutils.ProcessExecutionError:
if cnt == max_retry - 1:
raise
else:
LOG.warning('plug_vifs() failed %(cnt)d. Retry up to '
'%(max_retry)d.',
{'cnt': cnt, 'max_retry': max_retry},
instance=instance)
greenthread.sleep(1)
def pre_live_migration(self, context, instance, block_device_info,
network_info, disk_info, migrate_data):
"""Preparation live migration."""
if disk_info is not None:
disk_info = jsonutils.loads(disk_info)
LOG.debug('migrate_data in pre_live_migration: %s', migrate_data,
instance=instance)
is_shared_block_storage = migrate_data.is_shared_block_storage
is_shared_instance_path = migrate_data.is_shared_instance_path
is_block_migration = migrate_data.block_migration
if not is_shared_instance_path:
instance_dir = libvirt_utils.get_instance_path_at_destination(
instance, migrate_data)
if os.path.exists(instance_dir):
raise exception.DestinationDiskExists(path=instance_dir)
LOG.debug('Creating instance directory: %s', instance_dir,
instance=instance)
os.mkdir(instance_dir)
# Recreate the disk.info file and in doing so stop the
# imagebackend from recreating it incorrectly by inspecting the
# contents of each file when using the Raw backend.
if disk_info:
image_disk_info = {}
for info in disk_info:
image_file = os.path.basename(info['path'])
image_path = os.path.join(instance_dir, image_file)
image_disk_info[image_path] = info['type']
LOG.debug('Creating disk.info with the contents: %s',
image_disk_info, instance=instance)
image_disk_info_path = os.path.join(instance_dir,
'disk.info')
with open(image_disk_info_path, 'w') as f:
f.write(jsonutils.dumps(image_disk_info))
if not is_shared_block_storage:
# Ensure images and backing files are present.
LOG.debug('Checking to make sure images and backing files are '
'present before live migration.', instance=instance)
self._create_images_and_backing(
context, instance, instance_dir, disk_info,
fallback_from_host=instance.host)
if not is_block_migration:
# NOTE(angdraug): when block storage is shared between source
# and destination and instance path isn't (e.g. volume backed
# or rbd backed instance), instance path on destination has to
# be prepared
# Required by Quobyte CI
self._ensure_console_log_for_instance(instance)
# if image has kernel and ramdisk, just download
# following normal way.
self._fetch_instance_kernel_ramdisk(context, instance)
# Establishing connection to volume server.
block_device_mapping = driver.block_device_info_get_mapping(
block_device_info)
if len(block_device_mapping):
LOG.debug('Connecting volumes before live migration.',
instance=instance)
for bdm in block_device_mapping:
connection_info = bdm['connection_info']
self._connect_volume(context, connection_info, instance)
self._pre_live_migration_plug_vifs(
instance, network_info, migrate_data)
# Store server_listen and latest disk device info
if not migrate_data:
migrate_data = objects.LibvirtLiveMigrateData(bdms=[])
else:
migrate_data.bdms = []
# Store live_migration_inbound_addr
migrate_data.target_connect_addr = \
CONF.libvirt.live_migration_inbound_addr
migrate_data.supported_perf_events = self._supported_perf_events
migrate_data.serial_listen_ports = []
if CONF.serial_console.enabled:
num_ports = hardware.get_number_of_serial_ports(
instance.flavor, instance.image_meta)
for port in range(num_ports):
migrate_data.serial_listen_ports.append(
serial_console.acquire_port(
migrate_data.serial_listen_addr))
for vol in block_device_mapping:
connection_info = vol['connection_info']
if connection_info.get('serial'):
disk_info = blockinfo.get_info_from_bdm(
instance, CONF.libvirt.virt_type,
instance.image_meta, vol)
bdmi = objects.LibvirtLiveMigrateBDMInfo()
bdmi.serial = connection_info['serial']
bdmi.connection_info = connection_info
bdmi.bus = disk_info['bus']
bdmi.dev = disk_info['dev']
bdmi.type = disk_info['type']
bdmi.format = disk_info.get('format')
bdmi.boot_index = disk_info.get('boot_index')
volume_secret = self._host.find_secret('volume', vol.volume_id)
if volume_secret:
bdmi.encryption_secret_uuid = volume_secret.UUIDString()
migrate_data.bdms.append(bdmi)
if 'dst_numa_info' in migrate_data and migrate_data.dst_numa_info:
self.cpu_api.power_up_for_migration(migrate_data.dst_numa_info)
else:
LOG.debug('No dst_numa_info in migrate_data, '
'no cores to power up in pre_live_migration.')
return migrate_data
def _try_fetch_image_cache(self, image, fetch_func, context, filename,
image_id, instance, size,
fallback_from_host=None):
try:
image.cache(fetch_func=fetch_func,
context=context,
filename=filename,
image_id=image_id,
size=size,
trusted_certs=instance.trusted_certs)
except exception.ImageNotFound:
if not fallback_from_host:
raise
LOG.debug("Image %(image_id)s doesn't exist anymore "
"on image service, attempting to copy "
"image from %(host)s",
{'image_id': image_id, 'host': fallback_from_host},
instance=instance)
def copy_from_host(target):
libvirt_utils.copy_image(src=target,
dest=target,
host=fallback_from_host,
receive=True)
image.cache(fetch_func=copy_from_host, size=size,
filename=filename)
# NOTE(lyarwood): If the instance vm_state is shelved offloaded then we
# must be unshelving for _try_fetch_image_cache to be called.
# NOTE(mriedem): Alternatively if we are doing a cross-cell move of a
# non-volume-backed server and finishing (spawning) on the dest host,
# we have to flatten the rbd image so we can delete the temporary
# snapshot in the compute manager.
mig_context = instance.migration_context
cross_cell_move = (
mig_context and mig_context.is_cross_cell_move() or False)
if instance.vm_state == vm_states.SHELVED_OFFLOADED or cross_cell_move:
# NOTE(lyarwood): When using the rbd imagebackend the call to cache
# above will attempt to clone from the shelved snapshot in Glance
# if available from this compute. We then need to flatten the
# resulting image to avoid it still referencing and ultimately
# blocking the removal of the shelved snapshot at the end of the
# unshelve. This is a no-op for all but the rbd imagebackend.
action = (
'migrating instance across cells' if cross_cell_move
else 'unshelving instance')
try:
image.flatten()
LOG.debug('Image %s flattened successfully while %s.',
image.path, action, instance=instance)
except NotImplementedError:
# NOTE(lyarwood): There's an argument to be made for logging
# our inability to call flatten here, however given this isn't
# implemented for most of the backends it may do more harm than
# good, concerning operators etc so for now just pass.
pass
def _create_images_and_backing(self, context, instance, instance_dir,
disk_info, fallback_from_host=None):
""":param context: security context
:param instance:
nova.db.main.models.Instance object
instance object that is migrated.
:param instance_dir:
instance path to use, calculated externally to handle block
migrating an instance with an old style instance path
:param disk_info:
disk info specified in _get_instance_disk_info_from_config
(list of dicts)
:param fallback_from_host:
host where we can retrieve images if the glance images are
not available.
"""
# Virtuozzo containers don't use backing file
if (CONF.libvirt.virt_type == "parallels" and
instance.vm_mode == fields.VMMode.EXE):
return
if not disk_info:
disk_info = []
for info in disk_info:
base = os.path.basename(info['path'])
# Get image type and create empty disk image, and
# create backing file in case of qcow2.
instance_disk = os.path.join(instance_dir, base)
if not info['backing_file'] and not os.path.exists(instance_disk):
libvirt_utils.create_image(
instance_disk, info['type'], info['virt_disk_size'])
elif info['backing_file']:
# Creating backing file follows same way as spawning instances.
cache_name = os.path.basename(info['backing_file'])
disk = self.image_backend.by_name(instance, instance_disk)
if cache_name.startswith('ephemeral'):
# The argument 'size' is used by image.cache to
# validate disk size retrieved from cache against
# the instance disk size (should always return OK)
# and ephemeral_size is used by _create_ephemeral
# to build the image if the disk is not already
# cached.
disk.cache(
fetch_func=self._create_ephemeral,
fs_label=cache_name,
os_type=instance.os_type,
filename=cache_name,
size=info['virt_disk_size'],
ephemeral_size=info['virt_disk_size'] / units.Gi,
safe=True)
elif cache_name.startswith('swap'):
flavor = instance.get_flavor()
swap_mb = flavor.swap
disk.cache(fetch_func=self._create_swap,
filename="swap_%s" % swap_mb,
size=swap_mb * units.Mi,
swap_mb=swap_mb,
safe=True)
else:
self._try_fetch_image_cache(disk,
libvirt_utils.fetch_image,
context, cache_name,
instance.image_ref,
instance,
info['virt_disk_size'],
fallback_from_host)
# if disk has kernel and ramdisk, just download
# following normal way.
self._fetch_instance_kernel_ramdisk(
context, instance, fallback_from_host=fallback_from_host)
def post_live_migration(self, context, instance, block_device_info,
migrate_data=None):
# NOTE(mdbooth): The block_device_info we were passed was initialized
# with BDMs from the source host before they were updated to point to
# the destination. We can safely use this to disconnect the source
# without re-fetching.
block_device_mapping = driver.block_device_info_get_mapping(
block_device_info)
for vol in block_device_mapping:
connection_info = vol['connection_info']
# NOTE(lyarwood): Ignore exceptions here to avoid the instance
# being left in an ERROR state and still marked on the source.
try:
self._disconnect_volume(context, connection_info, instance)
except Exception:
volume_id = driver_block_device.get_volume_id(connection_info)
LOG.exception("Ignoring exception while attempting to "
"disconnect volume %s from the source host "
"during post_live_migration", volume_id,
instance=instance)
def post_live_migration_at_source(self, context, instance, network_info):
"""Unplug VIFs from networks at source.
:param context: security context
:param instance: instance object reference
:param network_info: instance network information
"""
self.unplug_vifs(instance, network_info)
self.cpu_api.power_down_for_instance(instance)
def _qemu_monitor_announce_self(self, instance):
"""Send announce_self command to QEMU monitor.
This is to trigger generation of broadcast RARP frames to
update network switches. This is best effort.
"""
if not CONF.workarounds.enable_qemu_monitor_announce_self:
return
current_attempt = 0
max_attempts = (
CONF.workarounds.qemu_monitor_announce_self_count)
# qemu_monitor_announce_retry_interval specified in seconds
announce_pause = (
CONF.workarounds.qemu_monitor_announce_self_interval)
while current_attempt < max_attempts:
# Increment attempt
current_attempt += 1
# Only use announce_pause after the first attempt to avoid
# pausing before calling announce_self for the first attempt
if current_attempt != 1:
greenthread.sleep(announce_pause)
LOG.info('Sending announce-self command to QEMU monitor. '
'Attempt %(current_attempt)s of %(max_attempts)s',
{'current_attempt': current_attempt,
'max_attempts': max_attempts}, instance=instance)
try:
guest = self._host.get_guest(instance)
guest.announce_self()
except Exception:
LOG.warning('Failed to send announce-self command to '
'QEMU monitor. Attempt %(current_attempt)s of '
'%(max_attempts)s',
{'current_attempt': current_attempt,
'max_attempts': max_attempts}, instance=instance)
LOG.exception()
def post_live_migration_at_destination(self, context,
instance,
network_info,
block_migration=False,
block_device_info=None):
"""Post operation of live migration at destination host.
:param context: security context
:param instance:
nova.db.main.models.Instance object
instance object that is migrated.
:param network_info: instance network information
:param block_migration: if true, post operation of block_migration.
"""
self._reattach_instance_vifs(context, instance, network_info)
self._qemu_monitor_announce_self(instance)
mdevs = self.instance_claimed_mdevs.pop(instance.uuid, None)
if mdevs:
# The live migration is done, the related mdevs are now associated
# to the domain XML so we can remove the reserved values.
LOG.debug("Unclaiming mdevs %s from instance %s",
mdevs, instance.uuid)
def _get_instance_disk_info_from_config(self, guest_config,
block_device_info):
"""Get the non-volume disk information from the domain xml
:param LibvirtConfigGuest guest_config: the libvirt domain config
for the instance
:param dict block_device_info: block device info for BDMs
:returns disk_info: list of dicts with keys:
* 'type': the disk type (str)
* 'path': the disk path (str)
* 'virt_disk_size': the virtual disk size (int)
* 'backing_file': backing file of a disk image (str)
* 'disk_size': physical disk size (int)
* 'over_committed_disk_size': virt_disk_size - disk_size or 0
"""
block_device_mapping = driver.block_device_info_get_mapping(
block_device_info)
volume_devices = set()
for vol in block_device_mapping:
disk_dev = vol['mount_device'].rpartition("/")[2]
volume_devices.add(disk_dev)
disk_info = []
if (
CONF.libvirt.virt_type == 'parallels' and
guest_config.os_type == fields.VMMode.EXE
):
node_type = 'filesystem'
else:
node_type = 'disk'
for device in guest_config.devices:
if device.root_name != node_type:
continue
disk_type = device.source_type
if device.root_name == 'filesystem':
target = device.target_dir
if device.source_type == 'file':
path = device.source_file
elif device.source_type == 'block':
path = device.source_dev
else:
path = None
else:
target = device.target_dev
path = device.source_path
if not path:
LOG.debug('skipping disk for %s as it does not have a path',
guest_config.name)
continue
if disk_type not in ['file', 'block']:
LOG.debug('skipping disk because it looks like a volume', path)
continue
if target in volume_devices:
LOG.debug('skipping disk %(path)s (%(target)s) as it is a '
'volume', {'path': path, 'target': target})
continue
if device.root_name == 'filesystem':
driver_type = device.driver_type
else:
driver_type = device.driver_format
# get the real disk size or
# raise a localized error if image is unavailable
if disk_type == 'file' and driver_type == 'ploop':
dk_size = 0
for dirpath, dirnames, filenames in os.walk(path):
for f in filenames:
fp = os.path.join(dirpath, f)
dk_size += os.path.getsize(fp)
qemu_img_info = disk_api.get_disk_info(path)
virt_size = qemu_img_info.virtual_size
backing_file = libvirt_utils.get_disk_backing_file(path)
over_commit_size = int(virt_size) - dk_size
elif disk_type == 'file' and driver_type == 'qcow2':
qemu_img_info = disk_api.get_disk_info(path)
dk_size = qemu_img_info.disk_size
virt_size = qemu_img_info.virtual_size
backing_file = libvirt_utils.get_disk_backing_file(path)
over_commit_size = max(0, int(virt_size) - dk_size)
elif disk_type == 'file':
dk_size = os.stat(path).st_blocks * 512
virt_size = os.path.getsize(path)
backing_file = ""
over_commit_size = int(virt_size) - dk_size
elif disk_type == 'block' and block_device_info:
dk_size = lvm.get_volume_size(path)
virt_size = dk_size
backing_file = ""
over_commit_size = 0
else:
LOG.debug('skipping disk %(path)s (%(target)s) - unable to '
'determine if volume',
{'path': path, 'target': target})
continue
disk_info.append({'type': driver_type,
'path': path,
'virt_disk_size': virt_size,
'backing_file': backing_file,
'disk_size': dk_size,
'over_committed_disk_size': over_commit_size})
return disk_info
def _get_instance_disk_info(self, instance, block_device_info):
try:
guest = self._host.get_guest(instance)
config = guest.get_config()
except libvirt.libvirtError as ex:
error_code = ex.get_error_code()
LOG.warning('Error from libvirt while getting description of '
'%(instance_name)s: [Error Code %(error_code)s] '
'%(ex)s',
{'instance_name': instance.name,
'error_code': error_code,
'ex': ex},
instance=instance)
raise exception.InstanceNotFound(instance_id=instance.uuid)
return self._get_instance_disk_info_from_config(config,
block_device_info)
def get_instance_disk_info(self, instance,
block_device_info=None):
return jsonutils.dumps(
self._get_instance_disk_info(instance, block_device_info))
def _get_disk_over_committed_size_total(self):
"""Return total over committed disk size for all instances."""
# Disk size that all instance uses : virtual_size - disk_size
disk_over_committed_size = 0
instance_domains = self._host.list_instance_domains(only_running=False)
if not instance_domains:
return disk_over_committed_size
# Get all instance uuids
instance_uuids = [dom.UUIDString() for dom in instance_domains]
ctx = nova_context.get_admin_context()
# Get instance object list by uuid filter
filters = {'uuid': instance_uuids}
# NOTE(ankit): objects.InstanceList.get_by_filters method is
# getting called twice one is here and another in the
# _update_available_resource method of resource_tracker. Since
# _update_available_resource method is synchronized, there is a
# possibility the instances list retrieved here to calculate
# disk_over_committed_size would differ to the list you would get
# in _update_available_resource method for calculating usages based
# on instance utilization.
local_instance_list = objects.InstanceList.get_by_filters(
ctx, filters, use_slave=True)
# Convert instance list to dictionary with instance uuid as key.
local_instances = {inst.uuid: inst for inst in local_instance_list}
# Get bdms by instance uuids
bdms = objects.BlockDeviceMappingList.bdms_by_instance_uuid(
ctx, instance_uuids)
for dom in instance_domains:
try:
guest = libvirt_guest.Guest(dom)
config = guest.get_config()
block_device_info = None
if guest.uuid in local_instances \
and (bdms and guest.uuid in bdms):
# Get block device info for instance
block_device_info = driver.get_block_device_info(
local_instances[guest.uuid], bdms[guest.uuid])
disk_infos = self._get_instance_disk_info_from_config(
config, block_device_info)
if not disk_infos:
continue
for info in disk_infos:
disk_over_committed_size += int(
info['over_committed_disk_size'])
except libvirt.libvirtError as ex:
error_code = ex.get_error_code()
LOG.warning(
'Error from libvirt while getting description of '
'%(instance_name)s: [Error Code %(error_code)s] %(ex)s',
{'instance_name': guest.name,
'error_code': error_code,
'ex': ex})
except OSError as e:
if e.errno in (errno.ENOENT, errno.ESTALE):
LOG.warning('Periodic task is updating the host stat, '
'it is trying to get disk %(i_name)s, '
'but disk file was removed by concurrent '
'operations such as resize.',
{'i_name': guest.name})
elif e.errno == errno.EACCES:
LOG.warning('Periodic task is updating the host stat, '
'it is trying to get disk %(i_name)s, '
'but access is denied. It is most likely '
'due to a VM that exists on the compute '
'node but is not managed by Nova.',
{'i_name': guest.name})
else:
raise
except (exception.VolumeBDMPathNotFound,
exception.DiskNotFound) as e:
if isinstance(e, exception.VolumeBDMPathNotFound):
thing = 'backing volume block device'
elif isinstance(e, exception.DiskNotFound):
thing = 'backing disk storage'
LOG.warning('Periodic task is updating the host stats, '
'it is trying to get disk info for %(i_name)s, '
'but the %(thing)s was removed by a concurrent '
'operation such as resize. Error: %(error)s',
{'i_name': guest.name, 'thing': thing, 'error': e})
# NOTE(gtt116): give other tasks a chance.
greenthread.sleep(0)
return disk_over_committed_size
def get_available_nodes(self, refresh=False):
return [self._host.get_hostname()]
def get_nodenames_by_uuid(self, refresh=False):
return {self._host.get_node_uuid(): self._host.get_hostname()}
def get_host_cpu_stats(self):
"""Return the current CPU state of the host."""
return self._host.get_cpu_stats()
def get_host_uptime(self):
"""Returns the result of calling "uptime"."""
out, err = processutils.execute('env', 'LANG=C', 'uptime')
return out
def manage_image_cache(self, context, all_instances):
"""Manage the local cache of images."""
self.image_cache_manager.update(context, all_instances)
def _cleanup_remote_migration(self, dest, inst_base, inst_base_resize,
shared_storage=False):
"""Used only for cleanup in case migrate_disk_and_power_off fails."""
try:
if os.path.exists(inst_base_resize):
shutil.rmtree(inst_base, ignore_errors=True)
os.rename(inst_base_resize, inst_base)
if not shared_storage:
self._remotefs.remove_dir(dest, inst_base)
except Exception:
pass
def cache_image(self, context, image_id):
cache_dir = os.path.join(CONF.instances_path,
CONF.image_cache.subdirectory_name)
path = os.path.join(cache_dir,
imagecache.get_cache_fname(image_id))
if os.path.exists(path):
LOG.info('Image %(image_id)s already cached; updating timestamp',
{'image_id': image_id})
# NOTE(danms): The regular image cache routines use a wrapper
# (_update_utime_ignore_eacces()) around this to avoid failing
# on permissions (which may or may not be legit due to an NFS
# race). However, since this is best-effort, errors are swallowed
# by compute manager per-image, and we are compelled to report
# errors up our stack, we use the raw method here to avoid the
# silent ignore of the EACCESS.
nova.privsep.path.utime(path)
return False
else:
# NOTE(danms): In case we are running before the first boot, make
# sure the cache directory is created
if not os.path.isdir(cache_dir):
fileutils.ensure_tree(cache_dir)
LOG.info('Caching image %(image_id)s by request',
{'image_id': image_id})
# NOTE(danms): The imagebackend code, as called via spawn() where
# images are normally cached, uses a lock on the root disk it is
# creating at the time, but relies on the
# compute_utils.disk_ops_semaphore for cache fetch mutual
# exclusion, which is grabbed in images.fetch() (which is called
# by images.fetch_to_raw() below). So, by calling fetch_to_raw(),
# we are sharing the same locking for the cache fetch as the
# rest of the code currently called only from spawn().
images.fetch_to_raw(context, image_id, path)
return True
def _get_disk_size_reserved_for_image_cache(self):
"""Return the amount of DISK_GB resource need to be reserved for the
image cache.
:returns: The disk space in GB
"""
if not CONF.workarounds.reserve_disk_resource_for_image_cache:
return 0
return compute_utils.convert_mb_to_ceil_gb(
self.image_cache_manager.get_disk_usage() / 1024.0 / 1024.0)
def _is_path_shared_with(self, dest, path):
# NOTE (rmk): There are two methods of determining whether we are
# on the same filesystem: the source and dest migration
# address are the same, or we create a file on the dest
# system via SSH and check whether the source system can
# also see it.
shared_path = (dest == self.get_host_ip_addr())
if not shared_path:
tmp_file = uuidutils.generate_uuid(dashed=False) + '.tmp'
tmp_path = os.path.join(path, tmp_file)
try:
self._remotefs.create_file(dest, tmp_path)
if os.path.exists(tmp_path):
shared_path = True
os.unlink(tmp_path)
else:
self._remotefs.remove_file(dest, tmp_path)
except Exception:
pass
return shared_path
def migrate_disk_and_power_off(self, context, instance, dest,
flavor, network_info,
block_device_info=None,
timeout=0, retry_interval=0):
LOG.debug("Starting migrate_disk_and_power_off",
instance=instance)
ephemerals = driver.block_device_info_get_ephemerals(block_device_info)
# get_bdm_ephemeral_disk_size() will return 0 if the new
# instance's requested block device mapping contain no
# ephemeral devices. However, we still want to check if
# the original instance's ephemeral_gb property was set and
# ensure that the new requested flavor ephemeral size is greater
eph_size = (block_device.get_bdm_ephemeral_disk_size(ephemerals) or
instance.flavor.ephemeral_gb)
# Checks if the migration needs a disk resize down.
root_down = flavor.root_gb < instance.flavor.root_gb
ephemeral_down = flavor.ephemeral_gb < eph_size
booted_from_volume = self._is_booted_from_volume(block_device_info)
if (root_down and not booted_from_volume) or ephemeral_down:
reason = _("Unable to resize disk down.")
raise exception.InstanceFaultRollback(
exception.ResizeError(reason=reason))
# NOTE(dgenin): Migration is not implemented for LVM backed instances.
if CONF.libvirt.images_type == 'lvm' and not booted_from_volume:
reason = _("Migration is not supported for LVM backed instances")
raise exception.InstanceFaultRollback(
exception.MigrationPreCheckError(reason=reason))
# copy disks to destination
# rename instance dir to +_resize at first for using
# shared storage for instance dir (eg. NFS).
inst_base = libvirt_utils.get_instance_path(instance)
inst_base_resize = inst_base + "_resize"
shared_instance_path = self._is_path_shared_with(dest, inst_base)
# try to create the directory on the remote compute node
# if this fails we pass the exception up the stack so we can catch
# failures here earlier
if not shared_instance_path:
try:
self._remotefs.create_dir(dest, inst_base)
except processutils.ProcessExecutionError as e:
reason = _("not able to execute ssh command: %s") % e
raise exception.InstanceFaultRollback(
exception.ResizeError(reason=reason))
self.power_off(instance, timeout, retry_interval)
self.unplug_vifs(instance, network_info)
block_device_mapping = driver.block_device_info_get_mapping(
block_device_info)
for vol in block_device_mapping:
connection_info = vol['connection_info']
self._disconnect_volume(context, connection_info, instance)
disk_info = self._get_instance_disk_info(instance, block_device_info)
try:
# If cleanup failed in previous resize attempts we try to remedy
# that before a resize is tried again
self._cleanup_failed_instance_base(inst_base_resize)
os.rename(inst_base, inst_base_resize)
# if we are migrating the instance with shared instance path then
# create the directory. If it is a remote node the directory
# has already been created
if shared_instance_path:
dest = None
fileutils.ensure_tree(inst_base)
on_execute = lambda process: \
self.job_tracker.add_job(instance, process.pid)
on_completion = lambda process: \
self.job_tracker.remove_job(instance, process.pid)
for info in disk_info:
# assume inst_base == dirname(info['path'])
img_path = info['path']
fname = os.path.basename(img_path)
from_path = os.path.join(inst_base_resize, fname)
# We will not copy over the swap disk here, and rely on
# finish_migration to re-create it for us. This is ok because
# the OS is shut down, and as recreating a swap disk is very
# cheap it is more efficient than copying either locally or
# over the network. This also means we don't have to resize it.
if fname == 'disk.swap':
continue
compression = info['type'] not in NO_COMPRESSION_TYPES
libvirt_utils.copy_image(from_path, img_path, host=dest,
on_execute=on_execute,
on_completion=on_completion,
compression=compression)
# Ensure disk.info is written to the new path to avoid disks being
# reinspected and potentially changing format.
src_disk_info_path = os.path.join(inst_base_resize, 'disk.info')
if os.path.exists(src_disk_info_path):
dst_disk_info_path = os.path.join(inst_base, 'disk.info')
libvirt_utils.copy_image(src_disk_info_path,
dst_disk_info_path,
host=dest, on_execute=on_execute,
on_completion=on_completion)
# Handle migration of vTPM data if needed
libvirt_utils.save_and_migrate_vtpm_dir(
instance.uuid, inst_base_resize, inst_base, dest,
on_execute, on_completion)
except Exception:
with excutils.save_and_reraise_exception():
self._cleanup_remote_migration(dest, inst_base,
inst_base_resize,
shared_instance_path)
return jsonutils.dumps(disk_info)
def _wait_for_running(self, instance):
state = self.get_info(instance).state
if state == power_state.RUNNING:
LOG.info("Instance running successfully.", instance=instance)
raise loopingcall.LoopingCallDone()
@staticmethod
def _disk_raw_to_qcow2(path):
"""Converts a raw disk to qcow2."""
path_qcow = path + '_qcow'
images.convert_image(path, path_qcow, 'raw', 'qcow2')
os.rename(path_qcow, path)
def _finish_migration_vtpm(
self,
context: nova_context.RequestContext,
instance: 'objects.Instance',
) -> None:
"""Handle vTPM when migrating or resizing an instance.
Handle the case where we're resizing between different versions of TPM,
or enabling/disabling TPM.
"""
old_vtpm_config = hardware.get_vtpm_constraint(
instance.old_flavor, instance.image_meta)
new_vtpm_config = hardware.get_vtpm_constraint(
instance.new_flavor, instance.image_meta)
if old_vtpm_config:
# we had a vTPM in the old flavor; figure out if we need to do
# anything with it
inst_base = libvirt_utils.get_instance_path(instance)
swtpm_dir = os.path.join(inst_base, 'swtpm', instance.uuid)
copy_swtpm_dir = True
if old_vtpm_config != new_vtpm_config:
# we had vTPM in the old flavor but the new flavor either
# doesn't or has different config; delete old TPM data and let
# libvirt create new data
if os.path.exists(swtpm_dir):
LOG.info(
'Old flavor and new flavor have different vTPM '
'configuration; removing existing vTPM data.')
copy_swtpm_dir = False
shutil.rmtree(swtpm_dir)
# apparently shutil.rmtree() isn't reliable on NFS so don't rely
# only on path existence here.
if copy_swtpm_dir and os.path.exists(swtpm_dir):
libvirt_utils.restore_vtpm_dir(swtpm_dir)
elif new_vtpm_config:
# we've requested vTPM in the new flavor and didn't have one
# previously so we need to create a new secret
crypto.ensure_vtpm_secret(context, instance)
def finish_migration(
self,
context: nova_context.RequestContext,
migration: 'objects.Migration',
instance: 'objects.Instance',
disk_info: str,
network_info: network_model.NetworkInfo,
image_meta: 'objects.ImageMeta',
resize_instance: bool,
allocations: ty.Dict[str, ty.Any],
block_device_info: ty.Optional[ty.Dict[str, ty.Any]] = None,
power_on: bool = True,
) -> None:
"""Complete the migration process on the destination host."""
LOG.debug("Starting finish_migration", instance=instance)
block_disk_info = blockinfo.get_disk_info(CONF.libvirt.virt_type,
instance,
image_meta,
block_device_info)
# assume _create_image does nothing if a target file exists.
# NOTE: This has the intended side-effect of fetching a missing
# backing file.
self._create_image(context, instance, block_disk_info['mapping'],
block_device_info=block_device_info,
ignore_bdi_for_swap=True,
fallback_from_host=migration.source_compute)
# Required by Quobyte CI
self._ensure_console_log_for_instance(instance)
gen_confdrive = functools.partial(
self._create_configdrive, context, instance,
InjectionInfo(admin_pass=None, network_info=network_info,
files=None))
# Convert raw disks to qcow2 if migrating to host which uses
# qcow2 from host which uses raw.
for info in jsonutils.loads(disk_info):
path = info['path']
disk_name = os.path.basename(path)
# NOTE(mdbooth): The code below looks wrong, but is actually
# required to prevent a security hole when migrating from a host
# with use_cow_images=False to one with use_cow_images=True.
# Imagebackend uses use_cow_images to select between the
# atrociously-named-Raw and Qcow2 backends. The Qcow2 backend
# writes to disk.info, but does not read it as it assumes qcow2.
# Therefore if we don't convert raw to qcow2 here, a raw disk will
# be incorrectly assumed to be qcow2, which is a severe security
# flaw. The reverse is not true, because the atrociously-named-Raw
# backend supports both qcow2 and raw disks, and will choose
# appropriately between them as long as disk.info exists and is
# correctly populated, which it is because Qcow2 writes to
# disk.info.
#
# In general, we do not yet support format conversion during
# migration. For example:
# * Converting from use_cow_images=True to use_cow_images=False
# isn't handled. This isn't a security bug, but is almost
# certainly buggy in other cases, as the 'Raw' backend doesn't
# expect a backing file.
# * Converting to/from lvm and rbd backends is not supported.
#
# This behaviour is inconsistent, and therefore undesirable for
# users. It is tightly-coupled to implementation quirks of 2
# out of 5 backends in imagebackend and defends against a severe
# security flaw which is not at all obvious without deep analysis,
# and is therefore undesirable to developers. We should aim to
# remove it. This will not be possible, though, until we can
# represent the storage layout of a specific instance
# independent of the default configuration of the local compute
# host.
# Config disks are hard-coded to be raw even when
# use_cow_images=True (see _get_disk_config_image_type),so don't
# need to be converted.
if (disk_name != 'disk.config' and
info['type'] == 'raw' and CONF.use_cow_images):
self._disk_raw_to_qcow2(info['path'])
# Does the guest need to be assigned some vGPU mediated devices ?
mdevs = self._allocate_mdevs(allocations)
# Handle the case where the guest has emulated TPM
self._finish_migration_vtpm(context, instance)
xml = self._get_guest_xml(context, instance, network_info,
block_disk_info, image_meta,
block_device_info=block_device_info,
mdevs=mdevs)
# NOTE(mriedem): vifs_already_plugged=True here, regardless of whether
# or not we've migrated to another host, because we unplug VIFs locally
# and the status change in the port might go undetected by the neutron
# L2 agent (or neutron server) so neutron may not know that the VIF was
# unplugged in the first place and never send an event.
guest = self._create_guest_with_network(
context, xml, instance, network_info, block_device_info,
power_on=power_on, vifs_already_plugged=True,
post_xml_callback=gen_confdrive)
if power_on:
timer = loopingcall.FixedIntervalLoopingCall(
self._wait_for_running,
instance)
timer.start(interval=0.5).wait()
# Sync guest time after migration.
guest.sync_guest_time()
LOG.debug("finish_migration finished successfully.", instance=instance)
def _cleanup_failed_instance_base(self, inst_base):
"""Make sure that a failed migrate or resize doesn't prevent us from
rolling back in a revert or retrying a resize.
"""
try:
shutil.rmtree(inst_base)
except OSError as e:
if e.errno != errno.ENOENT:
raise
def _finish_revert_migration_vtpm(
self,
context: nova_context.RequestContext,
instance: 'objects.Instance',
) -> None:
"""Handle vTPM differences when reverting a migration or resize.
We should either restore any emulated vTPM persistent storage files or
create new ones.
"""
old_vtpm_config = hardware.get_vtpm_constraint(
instance.old_flavor, instance.image_meta)
new_vtpm_config = hardware.get_vtpm_constraint(
instance.new_flavor, instance.image_meta)
if old_vtpm_config:
# the instance had a vTPM before resize and should have one again;
# move the previously-saved vTPM data back to its proper location
inst_base = libvirt_utils.get_instance_path(instance)
swtpm_dir = os.path.join(inst_base, 'swtpm', instance.uuid)
if os.path.exists(swtpm_dir):
libvirt_utils.restore_vtpm_dir(swtpm_dir)
elif new_vtpm_config:
# the instance gained a vTPM and must now lose it; delete the vTPM
# secret, knowing that libvirt will take care of everything else on
# the destination side
crypto.delete_vtpm_secret(context, instance)
def finish_revert_migration(
self,
context: nova.context.RequestContext,
instance: 'objects.Instance',
network_info: network_model.NetworkInfo,
migration: 'objects.Migration',
block_device_info: ty.Optional[ty.Dict[str, ty.Any]] = None,
power_on: bool = True,
) -> None:
"""Finish the second half of reverting a resize on the source host."""
LOG.debug('Starting finish_revert_migration', instance=instance)
inst_base = libvirt_utils.get_instance_path(instance)
inst_base_resize = inst_base + "_resize"
# NOTE(danms): if we're recovering from a failed migration,
# make sure we don't have a left-over same-host base directory
# that would conflict. Also, don't fail on the rename if the
# failure happened early.
if os.path.exists(inst_base_resize):
self._cleanup_failed_instance_base(inst_base)
os.rename(inst_base_resize, inst_base)
root_disk = self.image_backend.by_name(instance, 'disk')
# Once we rollback, the snapshot is no longer needed, so remove it
if root_disk.exists():
root_disk.rollback_to_snap(libvirt_utils.RESIZE_SNAPSHOT_NAME)
root_disk.remove_snap(libvirt_utils.RESIZE_SNAPSHOT_NAME)
self._finish_revert_migration_vtpm(context, instance)
disk_info = blockinfo.get_disk_info(CONF.libvirt.virt_type,
instance,
instance.image_meta,
block_device_info)
# The guest could already have mediated devices, using them for
# the new XML
mdevs = list(self._get_all_assigned_mediated_devices(instance))
xml = self._get_guest_xml(context, instance, network_info, disk_info,
instance.image_meta,
block_device_info=block_device_info,
mdevs=mdevs)
self._create_guest_with_network(
context, xml, instance, network_info, block_device_info,
power_on=power_on)
if power_on:
timer = loopingcall.FixedIntervalLoopingCall(
self._wait_for_running,
instance)
timer.start(interval=0.5).wait()
LOG.debug("finish_revert_migration finished successfully.",
instance=instance)
def confirm_migration(self, context, migration, instance, network_info):
"""Confirms a resize, destroying the source VM."""
self._cleanup_resize(context, instance, network_info)
@staticmethod
def _get_io_devices(xml_doc):
"""get the list of io devices from the xml document."""
result: ty.Dict[str, ty.List[str]] = {"volumes": [], "ifaces": []}
try:
doc = etree.fromstring(xml_doc)
except Exception:
return result
blocks = [('./devices/disk', 'volumes'),
('./devices/interface', 'ifaces')]
for block, key in blocks:
section = doc.findall(block)
for node in section:
for child in node:
if child.tag == 'target' and child.get('dev'):
result[key].append(child.get('dev'))
return result
def get_diagnostics(self, instance):
guest = self._host.get_guest(instance)
# TODO(sahid): We are converting all calls from a
# virDomain object to use nova.virt.libvirt.Guest.
# We should be able to remove domain at the end.
domain = guest._domain
output = {}
# get cpu time, might launch an exception if the method
# is not supported by the underlying hypervisor being
# used by libvirt
try:
for vcpu in guest.get_vcpus_info():
output["cpu" + str(vcpu.id) + "_time"] = vcpu.time
except libvirt.libvirtError:
pass
# get io status
xml = guest.get_xml_desc()
dom_io = LibvirtDriver._get_io_devices(xml)
for guest_disk in dom_io["volumes"]:
try:
# blockStats might launch an exception if the method
# is not supported by the underlying hypervisor being
# used by libvirt
stats = domain.blockStats(guest_disk)
output[guest_disk + "_read_req"] = stats[0]
output[guest_disk + "_read"] = stats[1]
output[guest_disk + "_write_req"] = stats[2]
output[guest_disk + "_write"] = stats[3]
output[guest_disk + "_errors"] = stats[4]
except libvirt.libvirtError:
pass
for interface in dom_io["ifaces"]:
try:
# interfaceStats might launch an exception if the method
# is not supported by the underlying hypervisor being
# used by libvirt
stats = domain.interfaceStats(interface)
output[interface + "_rx"] = stats[0]
output[interface + "_rx_packets"] = stats[1]
output[interface + "_rx_errors"] = stats[2]
output[interface + "_rx_drop"] = stats[3]
output[interface + "_tx"] = stats[4]
output[interface + "_tx_packets"] = stats[5]
output[interface + "_tx_errors"] = stats[6]
output[interface + "_tx_drop"] = stats[7]
except libvirt.libvirtError:
pass
output["memory"] = domain.maxMemory()
# memoryStats might launch an exception if the method
# is not supported by the underlying hypervisor being
# used by libvirt
try:
mem = domain.memoryStats()
for key in mem.keys():
output["memory-" + key] = mem[key]
except (libvirt.libvirtError, AttributeError):
pass
return output
def get_instance_diagnostics(self, instance):
guest = self._host.get_guest(instance)
# TODO(sahid): We are converting all calls from a
# virDomain object to use nova.virt.libvirt.Guest.
# We should be able to remove domain at the end.
domain = guest._domain
xml = guest.get_xml_desc()
xml_doc = etree.fromstring(xml)
# TODO(sahid): Needs to use get_info but more changes have to
# be done since a mapping STATE_MAP LIBVIRT_POWER_STATE is
# needed.
state, max_mem, mem, num_cpu, cpu_time = guest._get_domain_info()
config_drive = configdrive.required_by(instance)
launched_at = timeutils.normalize_time(instance.launched_at)
uptime = timeutils.delta_seconds(launched_at,
timeutils.utcnow())
diags = diagnostics_obj.Diagnostics(state=power_state.STATE_MAP[state],
driver='libvirt',
config_drive=config_drive,
hypervisor=CONF.libvirt.virt_type,
hypervisor_os='linux',
uptime=uptime)
diags.memory_details = diagnostics_obj.MemoryDiagnostics(
maximum=max_mem / units.Mi,
used=mem / units.Mi)
# get cpu time, might launch an exception if the method
# is not supported by the underlying hypervisor being
# used by libvirt
try:
for vcpu in guest.get_vcpus_info():
diags.add_cpu(id=vcpu.id, time=vcpu.time)
except libvirt.libvirtError:
pass
# get io status
dom_io = LibvirtDriver._get_io_devices(xml)
for guest_disk in dom_io["volumes"]:
try:
# blockStats might launch an exception if the method
# is not supported by the underlying hypervisor being
# used by libvirt
stats = domain.blockStats(guest_disk)
diags.add_disk(read_bytes=stats[1],
read_requests=stats[0],
write_bytes=stats[3],
write_requests=stats[2],
errors_count=stats[4])
except libvirt.libvirtError:
pass
for interface in xml_doc.findall('./devices/interface'):
mac_address = interface.find('mac').get('address')
target = interface.find('./target')
# add nic that has no target (therefore no stats)
if target is None:
diags.add_nic(mac_address=mac_address)
continue
# add nic with stats
dev = target.get('dev')
try:
if dev:
# interfaceStats might launch an exception if the
# method is not supported by the underlying hypervisor
# being used by libvirt
stats = domain.interfaceStats(dev)
diags.add_nic(mac_address=mac_address,
rx_octets=stats[0],
rx_errors=stats[2],
rx_drop=stats[3],
rx_packets=stats[1],
tx_octets=stats[4],
tx_errors=stats[6],
tx_drop=stats[7],
tx_packets=stats[5])
except libvirt.libvirtError:
pass
return diags
@staticmethod
def _prepare_device_bus(dev):
"""Determines the device bus and its hypervisor assigned address
"""
bus = None
address = (dev.device_addr.format_address() if
dev.device_addr else None)
if isinstance(dev.device_addr,
vconfig.LibvirtConfigGuestDeviceAddressPCI):
bus = objects.PCIDeviceBus()
elif isinstance(dev, vconfig.LibvirtConfigGuestDisk):
if dev.target_bus == 'scsi':
bus = objects.SCSIDeviceBus()
elif dev.target_bus == 'ide':
bus = objects.IDEDeviceBus()
elif dev.target_bus == 'usb':
bus = objects.USBDeviceBus()
if address is not None and bus is not None:
bus.address = address
return bus
def _build_interface_metadata(self, dev, vifs_to_expose, vlans_by_mac,
trusted_by_mac):
"""Builds a metadata object for a network interface
:param dev: The LibvirtConfigGuestInterface to build metadata for.
:param vifs_to_expose: The list of tagged and/or vlan'ed
VirtualInterface objects.
:param vlans_by_mac: A dictionary of mac address -> vlan associations.
:param trusted_by_mac: A dictionary of mac address -> vf_trusted
associations.
:return: A NetworkInterfaceMetadata object, or None.
"""
vif = vifs_to_expose.get(dev.mac_addr)
if not vif:
LOG.debug('No VIF found with MAC %s, not building metadata',
dev.mac_addr)
return None
bus = self._prepare_device_bus(dev)
device = objects.NetworkInterfaceMetadata(mac=vif.address)
if 'tag' in vif and vif.tag:
device.tags = [vif.tag]
if bus:
device.bus = bus
vlan = vlans_by_mac.get(vif.address)
if vlan:
device.vlan = int(vlan)
device.vf_trusted = trusted_by_mac.get(vif.address, False)
return device
def _build_disk_metadata(self, dev, tagged_bdms):
"""Builds a metadata object for a disk
:param dev: The vconfig.LibvirtConfigGuestDisk to build metadata for.
:param tagged_bdms: The list of tagged BlockDeviceMapping objects.
:return: A DiskMetadata object, or None.
"""
bdm = tagged_bdms.get(dev.target_dev)
if not bdm:
LOG.debug('No BDM found with device name %s, not building '
'metadata.', dev.target_dev)
return None
bus = self._prepare_device_bus(dev)
device = objects.DiskMetadata(tags=[bdm.tag])
# NOTE(artom) Setting the serial (which corresponds to
# volume_id in BlockDeviceMapping) in DiskMetadata allows us to
# find the disks's BlockDeviceMapping object when we detach the
# volume and want to clean up its metadata.
device.serial = bdm.volume_id
if bus:
device.bus = bus
return device
def _build_share_metadata(self, dev, shares):
"""Builds a metadata object for a share
:param dev: The vconfig.LibvirtConfigGuestFilesys to build
metadata for.
:param shares: The list of ShareMapping objects.
:return: A ShareMetadata object, or None.
"""
device = objects.ShareMetadata()
for share in shares:
if dev.driver_type == 'virtiofs' and share.tag == dev.target_dir:
device.share_id = share.share_id
device.tag = share.tag
return device
LOG.warning('Device %s of type filesystem found but it is not '
'linked to any share.', dev)
return None
def _build_hostdev_metadata(self, dev, vifs_to_expose, vlans_by_mac):
"""Builds a metadata object for a hostdev. This can only be a PF, so we
don't need trusted_by_mac like in _build_interface_metadata because
only VFs can be trusted.
:param dev: The LibvirtConfigGuestHostdevPCI to build metadata for.
:param vifs_to_expose: The list of tagged and/or vlan'ed
VirtualInterface objects.
:param vlans_by_mac: A dictionary of mac address -> vlan associations.
:return: A NetworkInterfaceMetadata object, or None.
"""
# Strip out the leading '0x'
pci_address = pci_utils.get_pci_address(
*[x[2:] for x in (dev.domain, dev.bus, dev.slot, dev.function)])
try:
mac = pci_utils.get_mac_by_pci_address(pci_address,
pf_interface=True)
except exception.PciDeviceNotFoundById:
LOG.debug('Not exposing metadata for not found PCI device %s',
pci_address)
return None
vif = vifs_to_expose.get(mac)
if not vif:
LOG.debug('No VIF found with MAC %s, not building metadata', mac)
return None
device = objects.NetworkInterfaceMetadata(mac=mac)
device.bus = objects.PCIDeviceBus(address=pci_address)
if 'tag' in vif and vif.tag:
device.tags = [vif.tag]
vlan = vlans_by_mac.get(mac)
if vlan:
device.vlan = int(vlan)
return device
def _build_device_metadata(self, context, instance):
"""Builds a metadata object for instance devices, that maps the user
provided tag to the hypervisor assigned device address.
"""
def _get_device_name(bdm):
return block_device.strip_dev(bdm.device_name)
network_info = instance.info_cache.network_info
vlans_by_mac = netutils.get_cached_vifs_with_vlan(network_info)
trusted_by_mac = netutils.get_cached_vifs_with_trusted(network_info)
vifs = objects.VirtualInterfaceList.get_by_instance_uuid(context,
instance.uuid)
vifs_to_expose = {vif.address: vif for vif in vifs
if ('tag' in vif and vif.tag) or
vlans_by_mac.get(vif.address)}
# TODO(mriedem): We should be able to avoid the DB query here by using
# block_device_info['block_device_mapping'] which is passed into most
# methods that call this function.
bdms = objects.BlockDeviceMappingList.get_by_instance_uuid(
context, instance.uuid)
tagged_bdms = {_get_device_name(bdm): bdm for bdm in bdms if bdm.tag}
shares = objects.ShareMappingList.get_by_instance_uuid(
context, instance.uuid
)
devices = []
guest = self._host.get_guest(instance)
xml = guest.get_xml_desc()
xml_dom = etree.fromstring(xml)
guest_config = vconfig.LibvirtConfigGuest()
guest_config.parse_dom(xml_dom)
for dev in guest_config.devices:
device = None
if isinstance(dev, vconfig.LibvirtConfigGuestInterface):
device = self._build_interface_metadata(dev, vifs_to_expose,
vlans_by_mac,
trusted_by_mac)
if isinstance(dev, vconfig.LibvirtConfigGuestDisk):
device = self._build_disk_metadata(dev, tagged_bdms)
if isinstance(dev, vconfig.LibvirtConfigGuestHostdevPCI):
device = self._build_hostdev_metadata(dev, vifs_to_expose,
vlans_by_mac)
if isinstance(dev, vconfig.LibvirtConfigGuestFilesys):
device = self._build_share_metadata(dev, shares)
if device:
devices.append(device)
if devices:
dev_meta = objects.InstanceDeviceMetadata(devices=devices)
return dev_meta
def instance_on_disk(self, instance):
# ensure directories exist and are writable
instance_path = libvirt_utils.get_instance_path(instance)
LOG.debug('Checking instance files accessibility %s', instance_path,
instance=instance)
shared_instance_path = os.access(instance_path, os.W_OK)
# NOTE(flwang): For shared block storage scenario, the file system is
# not really shared by the two hosts, but the volume of evacuated
# instance is reachable.
shared_block_storage = (self.image_backend.backend().
is_shared_block_storage())
return shared_instance_path or shared_block_storage
def inject_network_info(self, instance, nw_info):
pass
def delete_instance_files(self, instance):
target = libvirt_utils.get_instance_path(instance)
# A resize may be in progress
target_resize = target + '_resize'
# Other threads may attempt to rename the path, so renaming the path
# to target + '_del' (because it is atomic) and iterating through
# twice in the unlikely event that a concurrent rename occurs between
# the two rename attempts in this method. In general this method
# should be fairly thread-safe without these additional checks, since
# other operations involving renames are not permitted when the task
# state is not None and the task state should be set to something
# other than None by the time this method is invoked.
target_del = target + '_del'
for i in range(2):
try:
os.rename(target, target_del)
break
except Exception:
pass
try:
os.rename(target_resize, target_del)
break
except Exception:
pass
# Either the target or target_resize path may still exist if all
# rename attempts failed.
remaining_path = None
for p in (target, target_resize):
if os.path.exists(p):
remaining_path = p
break
# A previous delete attempt may have been interrupted, so target_del
# may exist even if all rename attempts during the present method
# invocation failed due to the absence of both target and
# target_resize.
if not remaining_path and os.path.exists(target_del):
self.job_tracker.terminate_jobs(instance)
LOG.info('Deleting instance files %s', target_del,
instance=instance)
remaining_path = target_del
try:
shutil.rmtree(target_del)
except OSError as e:
LOG.error('Failed to cleanup directory %(target)s: %(e)s',
{'target': target_del, 'e': e}, instance=instance)
# It is possible that the delete failed, if so don't mark the instance
# as cleaned.
if remaining_path and os.path.exists(remaining_path):
LOG.info('Deletion of %s failed', remaining_path,
instance=instance)
return False
LOG.info('Deletion of %s complete', target_del, instance=instance)
return True
def default_root_device_name(self, instance, image_meta, root_bdm):
disk_bus = blockinfo.get_disk_bus_for_device_type(
instance, CONF.libvirt.virt_type, image_meta, "disk")
cdrom_bus = blockinfo.get_disk_bus_for_device_type(
instance, CONF.libvirt.virt_type, image_meta, "cdrom")
root_info = blockinfo.get_root_info(
instance, CONF.libvirt.virt_type, image_meta,
root_bdm, disk_bus, cdrom_bus)
return block_device.prepend_dev(root_info['dev'])
def default_device_names_for_instance(self, instance, root_device_name,
*block_device_lists):
block_device_mapping = list(itertools.chain(*block_device_lists))
# NOTE(ndipanov): Null out the device names so that blockinfo code
# will assign them
for bdm in block_device_mapping:
if bdm.device_name is not None:
LOG.info(
"Ignoring supplied device name: %(device_name)s. "
"Libvirt can't honour user-supplied dev names",
{'device_name': bdm.device_name}, instance=instance)
bdm.device_name = None
block_device_info = driver.get_block_device_info(instance,
block_device_mapping)
blockinfo.default_device_names(CONF.libvirt.virt_type,
nova_context.get_admin_context(),
instance,
block_device_info,
instance.image_meta)
def get_device_name_for_instance(self, instance, bdms, block_device_obj):
block_device_info = driver.get_block_device_info(instance, bdms)
instance_info = blockinfo.get_disk_info(
CONF.libvirt.virt_type, instance,
instance.image_meta, block_device_info=block_device_info)
suggested_dev_name = block_device_obj.device_name
if suggested_dev_name is not None:
LOG.info(
'Ignoring supplied device name: %(suggested_dev)s',
{'suggested_dev': suggested_dev_name}, instance=instance)
# NOTE(ndipanov): get_info_from_bdm will generate the new device name
# only when it's actually not set on the bd object
block_device_obj.device_name = None
disk_info = blockinfo.get_info_from_bdm(
instance, CONF.libvirt.virt_type, instance.image_meta,
block_device_obj, mapping=instance_info['mapping'])
return block_device.prepend_dev(disk_info['dev'])
def is_supported_fs_format(self, fs_type):
return fs_type in [nova.privsep.fs.FS_FORMAT_EXT2,
nova.privsep.fs.FS_FORMAT_EXT3,
nova.privsep.fs.FS_FORMAT_EXT4,
nova.privsep.fs.FS_FORMAT_XFS]
def _get_tpm_traits(self) -> ty.Dict[str, bool]:
# Assert or deassert TPM support traits
if not CONF.libvirt.swtpm_enabled:
return {
ot.COMPUTE_SECURITY_TPM_2_0: False,
ot.COMPUTE_SECURITY_TPM_1_2: False,
ot.COMPUTE_SECURITY_TPM_TIS: False,
ot.COMPUTE_SECURITY_TPM_CRB: False,
}
tpm_models = self._host.tpm_models
tpm_versions = self._host.tpm_versions
# libvirt < 8.6 does not provide supported versions in domain
# capabilities
tr = {}
if tpm_models is None:
# TODO(tkajinam): Remove this fallback once libvirt>=8.0.0 is
# required.
tr.update({
ot.COMPUTE_SECURITY_TPM_TIS: True,
ot.COMPUTE_SECURITY_TPM_CRB: True,
})
else:
tr.update({
ot.COMPUTE_SECURITY_TPM_TIS: 'tpm-tis' in tpm_models,
ot.COMPUTE_SECURITY_TPM_CRB: 'tpm-crb' in tpm_models,
})
if tpm_versions is None:
# TODO(tkajinam): Remove this fallback once libvirt>=8.6.0 is
# required.
tr.update({
ot.COMPUTE_SECURITY_TPM_2_0: True,
ot.COMPUTE_SECURITY_TPM_1_2: True,
})
else:
tr.update({
ot.COMPUTE_SECURITY_TPM_2_0: '2.0' in tpm_versions,
ot.COMPUTE_SECURITY_TPM_1_2: '1.2' in tpm_versions,
})
return tr
def _get_vif_model_traits(self) -> ty.Dict[str, bool]:
"""Get vif model traits based on the currently enabled virt_type.
Not all traits generated by this function may be valid and the result
should be validated.
:return: A dict of trait names mapped to boolean values.
"""
all_models = set(itertools.chain(
*libvirt_vif.SUPPORTED_VIF_MODELS.values()
))
supported_models = libvirt_vif.SUPPORTED_VIF_MODELS.get(
CONF.libvirt.virt_type, []
)
# remove version dependent vif models if we are on older libvirt/qemu
igb_supported = self._host.has_min_version(
MIN_IGB_LIBVIRT_VERSION, MIN_IGB_QEMU_VERSION)
if not igb_supported:
supported_models = [
model for model in supported_models
if model != network_model.VIF_MODEL_IGB]
# construct the corresponding standard trait from the VIF model name
return {
f'COMPUTE_NET_VIF_MODEL_{model.replace("-", "_").upper()}': model
in supported_models for model in all_models
}
def _get_iommu_model_traits(self) -> ty.Dict[str, bool]:
"""Get iommu model traits based on the currently enabled virt_type.
Not all traits generated by this function may be valid and the result
should be validated.
:return: A dict of trait names mapped to boolean values.
"""
dom_caps = self._host.get_domain_capabilities()
supported_models: ty.Set[str] = {fields.VIOMMUModel.AUTO}
# our min version of qemu/libvirt support q35 and virt machine types.
# They also support the smmuv3 and intel iommu modeles so if the qemu
# binary is available we can report the trait.
if fields.Architecture.AARCH64 in dom_caps:
supported_models.add(fields.VIOMMUModel.SMMUV3)
if fields.Architecture.X86_64 in dom_caps:
supported_models.add(fields.VIOMMUModel.INTEL)
# the virtio iommu model requires a newer libvirt then our min
# libvirt so we need to check the version explicitly.
if self._host.has_min_version(MIN_LIBVIRT_VIOMMU_VIRTIO_MODEL):
supported_models.add(fields.VIOMMUModel.VIRTIO)
return {
f'COMPUTE_VIOMMU_MODEL_{model.replace("-", "_").upper()}': model
in supported_models for model in fields.VIOMMUModel.ALL
}
def _get_storage_bus_traits(self) -> ty.Dict[str, bool]:
"""Get storage bus traits based on the currently enabled virt_type.
For QEMU and KVM this function uses the information returned by the
libvirt domain capabilities API. For other virt types we generate the
traits based on the static information in the blockinfo module.
Not all traits generated by this function may be valid and the result
should be validated.
:return: A dict of trait names mapped to boolean values.
"""
all_buses = set(itertools.chain(
*blockinfo.SUPPORTED_DEVICE_BUSES.values()
))
if CONF.libvirt.virt_type in ('qemu', 'kvm'):
dom_caps = self._host.get_domain_capabilities()
supported_buses: ty.Set[str] = set()
for arch_type in dom_caps:
for machine_type in dom_caps[arch_type]:
supported_buses.update(
dom_caps[arch_type][machine_type].devices.disk.buses
)
else:
supported_buses = blockinfo.SUPPORTED_DEVICE_BUSES.get(
CONF.libvirt.virt_type, []
)
# construct the corresponding standard trait from the storage bus name
return {
f'COMPUTE_STORAGE_BUS_{bus.replace("-", "_").upper()}': bus in
supported_buses for bus in all_buses
}
def _get_video_model_traits(self) -> ty.Dict[str, bool]:
"""Get video model traits from libvirt.
Not all traits generated by this function may be valid and the result
should be validated.
:return: A dict of trait names mapped to boolean values.
"""
all_models = fields.VideoModel.ALL
dom_caps = self._host.get_domain_capabilities()
supported_models: ty.Set[str] = set()
for arch_type in dom_caps:
for machine_type in dom_caps[arch_type]:
supported_models.update(
dom_caps[arch_type][machine_type].devices.video.models
)
# construct the corresponding standard trait from the video model name
return {
f'COMPUTE_GRAPHICS_MODEL_{model.replace("-", "_").upper()}': model
in supported_models for model in all_models
}
def _get_packed_virtqueue_traits(self) -> ty.Dict[str, bool]:
"""Get Virtio Packed Ring traits to be set on the host's
resource provider.
:return: A dict of trait names mapped to boolean values.
"""
return {ot.COMPUTE_NET_VIRTIO_PACKED: True}
def _get_cpu_traits(self) -> ty.Dict[str, bool]:
"""Get CPU-related traits to be set and unset on the host's resource
provider.
:return: A dict of trait names mapped to boolean values.
"""
traits = self._get_cpu_feature_traits()
traits[ot.HW_CPU_X86_AMD_SEV] = self._host.supports_amd_sev
traits[ot.HW_CPU_HYPERTHREADING] = self._host.has_hyperthreading
traits.update(self._get_cpu_arch_traits())
traits.update(self._get_cpu_emulation_arch_traits())
return traits
def _get_cpu_feature_traits(self) -> ty.Dict[str, bool]:
"""Get CPU traits of VMs based on guest CPU model config.
1. If mode is 'host-model' or 'host-passthrough', use host's
CPU features.
2. If mode is None, choose a default CPU model based on CPU
architecture.
3. If mode is 'custom', use cpu_models to generate CPU features.
The code also accounts for cpu_model_extra_flags configuration when
cpu_mode is 'host-model', 'host-passthrough' or 'custom', this
ensures user specified CPU feature flags to be included.
:return: A dict of trait names mapped to boolean values.
"""
cpu = self._get_guest_cpu_model_config()
if not cpu:
LOG.info('The current libvirt hypervisor %(virt_type)s '
'does not support reporting CPU traits.',
{'virt_type': CONF.libvirt.virt_type})
return {}
caps = deepcopy(self._host.get_capabilities())
if cpu.mode in ('host-model', 'host-passthrough'):
# Account for features in cpu_model_extra_flags conf
host_features: ty.Set[str] = {
f.name for f in caps.host.cpu.features | cpu.features
}
return libvirt_utils.cpu_features_to_traits(host_features)
def _resolve_features(cpu):
xml_str = cpu.to_xml()
features_xml = self._get_guest_baseline_cpu_features(xml_str)
feature_names = []
if features_xml:
cpu = vconfig.LibvirtConfigCPU()
cpu.parse_str(features_xml)
feature_names = [f.name for f in cpu.features]
return feature_names
features: ty.Set[str] = set()
# Choose a default CPU model when cpu_mode is not specified
if cpu.mode is None:
caps.host.cpu.model = libvirt_utils.get_cpu_model_from_arch(
caps.host.cpu.arch)
caps.host.cpu.features = set()
features = features.union(_resolve_features(caps.host.cpu))
else:
models = [self._get_cpu_model_mapping(model)
for model in CONF.libvirt.cpu_models]
# Aarch64 platform doesn't return the default CPU models
if caps.host.cpu.arch == fields.Architecture.AARCH64:
if not models:
models = ['max']
# For custom mode, iterate through cpu models
for model in models:
caps.host.cpu.model = model
caps.host.cpu.features = set()
features = features.union(_resolve_features(caps.host.cpu))
# Account for features in cpu_model_extra_flags conf
features = features.union([f.name for f in cpu.features])
return libvirt_utils.cpu_features_to_traits(features)
def _get_cpu_arch_traits(self):
"""Get CPU arch trait based on the host arch.
"""
arch = self._host.get_capabilities().host.cpu.arch.upper()
# we only set for valid arch, rest will be assumed invalid
trait = 'HW_ARCH_' + arch
return {trait: trait in ot.get_traits(prefix='HW_ARCH_')}
def _get_cpu_emulation_arch_traits(self):
"""Get CPU arch emulation traits
"""
# get list of architecture supported by host for
# hw emulation
caps = self._host.get_domain_capabilities().keys()
traits = {}
for arch in caps:
trait = 'COMPUTE_ARCH_' + arch.upper()
if trait in ot.get_traits(prefix='COMPUTE_ARCH_'):
traits[trait] = True
return traits
def _get_guest_baseline_cpu_features(self, xml_str):
"""Calls libvirt's baselineCPU API to compute the biggest set of
CPU features which is compatible with the given host CPU.
:param xml_str: XML description of host CPU
:return: An XML string of the computed CPU, or None on error
"""
LOG.debug("Libvirt baseline CPU %s", xml_str)
# TODO(lei-zh): baselineCPU is not supported on all platforms.
# There is some work going on in the libvirt community to replace the
# baseline call. Consider using the new apis when they are ready. See
# https://www.redhat.com/archives/libvir-list/2018-May/msg01204.html.
try:
if hasattr(libvirt, 'VIR_CONNECT_BASELINE_CPU_EXPAND_FEATURES'):
return self._host.get_connection().baselineCPU(
[xml_str],
libvirt.VIR_CONNECT_BASELINE_CPU_EXPAND_FEATURES)
else:
return self._host.get_connection().baselineCPU([xml_str])
except libvirt.libvirtError as ex:
with excutils.save_and_reraise_exception() as ctxt:
error_code = ex.get_error_code()
if error_code == libvirt.VIR_ERR_NO_SUPPORT:
ctxt.reraise = False
LOG.debug('URI %(uri)s does not support full set'
' of host capabilities: %(error)s',
{'uri': self._host._uri, 'error': ex})
return None
def _guest_add_virtiofs_for_share(self, guest, instance, share_info):
"""Add all share mount point as virtio fs entries."""
if share_info:
for share in share_info:
fs = vconfig.LibvirtConfigGuestFilesys()
fs.source_type = 'mount'
fs.access_mode = 'passthrough'
fs.driver_type = 'virtiofs'
fs.source_dir = self._get_share_mount_path(instance, share)
fs.target_dir = share.tag
guest.add_device(fs)
|