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
|
/****************************************************************
* *
* Copyright (c) 2001-2018 Fidelity National Information *
* Services, Inc. and/or its subsidiaries. All rights reserved. *
* *
* This source code contains the intellectual property *
* of its copyright holder(s), and is made available *
* under a license. If you do not know the terms of *
* the license, please stop and do not read further. *
* *
****************************************************************/
#include "mdef.h"
#include "gtm_inet.h"
#include "gtm_string.h"
#include "gdsroot.h"
#include "gtm_facility.h"
#include "fileinfo.h"
#include "gdsbt.h"
#include "gdsblk.h"
#include "gdsfhead.h"
#include "filestruct.h"
#include "gdscc.h"
#include "min_max.h"
#include "gdsblkops.h"
#include "gdsbml.h"
#include "gdskill.h"
#include "copy.h"
#include "interlock.h"
#include "jnl.h"
#include "buddy_list.h" /* needed for tp.h */
#include "hashtab_int4.h" /* needed for tp.h */
#include "tp.h"
#include "io.h"
#include "gtmsecshr.h"
#include "repl_msg.h"
#include "gtmsource.h"
#include "is_proc_alive.h"
#include "aswp.h"
#include "util.h"
#include "compswap.h"
#include "mutex.h"
#include "repl_instance.h" /* needed for JNLDATA_BASE_OFF macro */
#include "mupipbckup.h" /* needed for backup_block prototype */
#include "cert_blk.h" /* for CERT_BLK_IF_NEEDED macro */
#include "relqueopi.h" /* for INSQTI and INSQHI macros */
#include "caller_id.h"
#include "sec_shr_blk_build.h"
#include "add_inter.h"
#include "send_msg.h" /* for send_msg prototype */
#include "secshr_db_clnup.h"
#include "gdsbgtr.h"
#include "memcoherency.h"
#include "shmpool.h"
#include "wbox_test_init.h"
#include "db_snapshot.h"
#include "muextr.h"
#include "mupip_reorg.h"
#include "op.h"
#include "dpgbldir.h" /* for "get_next_gdr" */
#include "t_abort.h"
#include "have_crit.h"
#include "gtmsource_inline.h"
#define FLUSH 1
GBLREF gd_region *gv_cur_region; /* for the LOCK_HIST macro in the RELEASE_BUFF_UPDATE_LOCK macro */
GBLREF jnlpool_addrs_ptr_t jnlpool;
GBLREF jnlpool_addrs_ptr_t jnlpool_head;
GBLREF sgm_info *first_sgm_info; /* List of all regions (unsorted) in TP transaction */
GBLREF sgm_info *first_tp_si_by_ftok; /* List of READ or UPDATED regions sorted on ftok order */
GBLREF sgmnt_addrs *kip_csa;
GBLREF uint4 process_id;
GBLREF sgmnt_addrs *cs_addrs;
#ifdef DEBUG
GBLREF volatile boolean_t in_wcs_recover; /* TRUE if in "wcs_recover" */
GBLREF uint4 update_trans;
GBLREF boolean_t dse_running;
#endif
error_def(ERR_WCBLOCKED);
/* secshr_db_clnup can be called with one of the following three values for "secshr_state"
*
* a) NORMAL_TERMINATION --> We are called from the exit-handler for precautionary cleanup.
* We should NEVER be in the midst of a database update in this case.
* b) COMMIT_INCOMPLETE --> We are called from t_commit_cleanup.
* We should ALWAYS be in the midst of a database update in this case.
*
* If we are in the midst of a database update, then depending on where we are in the commit logic
* we need to ROLL-BACK (undo the partial commit) or ROLL-FORWARD (complete the partial commit) the database update.
* t_commit_cleanup handles the ROLL-BACK and secshr_db_clnup handles the ROLL-FORWARD
*
* For all error conditions in the database commit logic, t_commit_cleanup gets control first.
* If then determines whether to do a ROLL-BACK or a ROLL-FORWARD.
* If a ROLL-BACK needs to be done, then t_commit_cleanup handles it all by itself and we will not come here.
* If a ROLL-FORWARD needs to be done, then t_commit_cleanup invokes secshr_db_clnup.
* In this case, secshr_db_clnup will be called with a "secshr_state" value of "COMMIT_INCOMPLETE".
*
* Irrespective of whether we are in the midst of a database commit or not, t_commit_cleanup does not get control.
* Since the process can POSSIBLY be in the midst of a database update while it was STOP/IDed,
* the logic for determining whether it is a ROLL-BACK or a ROLL-FORWARD needs to also be in secshr_db_clnup.
* If it is determined that a ROLL-FORWARD needs to be done, secshr_db_clnup takes care of it by itself.
* But if a ROLL-BACK needs to be done, then secshr_db_clnup DOES NOT invoke t_commit_cleanup.
* Instead it sets cnl->wc_blocked to TRUE thereby ensuring the next process that gets CRIT does a cache recovery
* which will take care of doing more than the ROLL-BACK that t_commit_cleanup would have otherwise done.
*
* In order to understand how a ROLL-BACK or ROLL-FORWARD decision is made, the commit flow needs to be understood first.
* The commit logic flow in t_end & tp_tend can be captured as follows.
* Note that
* a) In t_end there is only one region in the "For each region" loop. And "si->update_trans" == "update_trans".
* b) All jnlpool steps below apply only if replication is turned ON.
* c) For MM, the jnl and db commits happen inside crit i.e. no phase2 outside crit like BG.
*
* (CMT01) Get crit on all regions (UPDATED || NON-UPDATED)
* (CMT02) Get crit on jnlpool
* (CMT03) Reserve space in JNLPOOL for journal records (UPDATE_JPL_RSRV_WRITE_ADDR macro)
* For each UPDATED region
* {
* (CMT04) csd->trans_hist.early_tn = csd->trans_hist.curr_tn + 1;
* (CMT05) csa->t_commit_crit = T_COMMIT_CRIT_PHASE0;
* }
* For each UPDATED & JOURNALED region
* {
* (CMT06) Reserve space in JNLBUFF for journal records PHASE1 (UPDATE_JRS_RSRV_FREEADDR macro)
* }
* (CMT06a) If (MM or (BG && jnlbuff-overflow)) Do Step CMT16 here instead of later
* (CMT06b) If (Non-TP && MM) Do Step CMT17 here instead of later
* (CMT07) jnlpool_ctl->jnl_seqno++; jnlpool_ctl->strm_seqno[]++ if supplementary; (SET_JNL_SEQNO macro)
* For each (UPDATED || NON-UPDATED) region
* {
* If this is an UPDATED region
* {
* (CMT08) csa->t_commit_crit = T_COMMIT_CRIT_PHASE1; cnl->update_underway_tn = csd->trans_hist.curr_tn;
* (CMT09) If replication is ON, csd->reg_seqno = jnlpool_ctl->jnl_seqno + 1; csd->strm_reg_seqno[] = xxx
* if (cw_set is non-NULL for this region) // i.e. not duplicate set(s)
* {
* For each cw-se-element
* {
* (CMT10) Commit PHASE1 // bg_update_phase1 or mm_update
* (CMT10a) If (BG && IS_BG_PHASE2_COMMIT_IN_CRIT(cse, mode)) Do Step CMT16 and CMT18 here instead of later
* }
* }
* (CMT11) si->update_trans |= UPDTRNS_TCOMMIT_STARTED_MASK;
* (CMT12) csd->trans_hist.curr_tn++;
* (CMT13) csa->t_commit_crit = T_COMMIT_CRIT_PHASE2;
* }
* (CMT14) Release crit on region
* }
* (CMT15) Release crit on jnlpool
* For each (UPDATED && JOURNALED) region
* {
* (CMT16) If (BG) Write journal records in JNLBUFF & JNLPOOL. PHASE2 (outside crit). Mark write complete in JNLBUFF.
* }
* (CMT17) If (BG) Mark journal record write complete in JNLPOOL.
* For each UPDATED region
* {
* if (cw_set is non-NULL for this region) // i.e. not duplicate set(s)
* {
* (CMT18) Commit all cw-set-elements of this region PHASE2 (outside crit) // bg_update_phase2
* }
* (CMT19) csa->t_commit_crit = FALSE;
* }
*
* If a transaction has proceeded to Step (CMT08) for at least one region, then "tp_update_underway" is set to TRUE
* and the transaction cannot be rolled back but has to be committed. Otherwise the transaction is rolled back.
* This is for the case where the process encounters an error in the midst of commit. In this case, we are guaranteed
* a clean recovery by secshr_db_clnup() & wcs_recover(). Database integrity is guaranteed.
* But if the process in the midst of commit is abnormally killed (e.g. kill -9), then "mutex_salvage",
* "jnl_phase2_salvage" and "repl_phase2_salvage" try to cleanup various pieces of the leftover mess but
* we are not guaranteed db integrity (i.e. CMT18 cannot be currently easily redone by a different process).
*/
void secshr_db_clnup(enum secshr_db_state secshr_state)
{
boolean_t is_bg, is_exiting;
boolean_t update_underway;
cache_que_heads_ptr_t cache_state;
cache_rec_ptr_t cr;
char *wcblocked_ptr;
gd_addr *gd_hdr;
gd_region *reg, *reg_top, *save_gv_cur_region, *repl_reg = NULL;
gtm_uint64_t argarray[SECSHR_ACCOUNTING_MAX_ARGS];
int numargs;
jbuf_rsrv_struct_t *jrs;
jnl_buffer_ptr_t jbp;
node_local_ptr_t cnl;
sm_uc_ptr_t blk_ptr;
blk_hdr_ptr_t blk_hdr_ptr;
jnlpool_addrs_ptr_t save_jnlpool, update_jnlpool = NULL;
jnlpool_ctl_ptr_t jpl;
pid_t pid;
sgm_info *si;
sgmnt_addrs *csa, *repl_csa;
sgmnt_data_ptr_t csd;
# ifdef DEBUG
sgm_info *jnlpool_si;
sgmnt_addrs *jnlpool_csa;
# endif
DCL_THREADGBL_ACCESS;
SETUP_THREADGBL_ACCESS;
save_gv_cur_region = gv_cur_region; /* save it for use at function end in case code in between changes this variable */
assert((NORMAL_TERMINATION == secshr_state) || (COMMIT_INCOMPLETE == secshr_state));
is_exiting = (NORMAL_TERMINATION == secshr_state);
save_jnlpool = jnlpool;
update_underway = FALSE;
if (dollar_tlevel)
{ /* Determine update_underway for TP transaction. A similar check is done in t_commit_cleanup as well.
* Regions are committed in the ftok order using "first_tp_si_by_ftok". Also crit is released on each
* region as the commit completes. Take that into account while determining if update is underway.
*/
for (si = first_tp_si_by_ftok; NULL != si; si = si->next_tp_si_by_ftok)
{
csa = si->tp_csa;
assert(NULL != csa);
if (NULL == csa->nl)
{ /* Possible for example if "mur_close_files" was called in "mupip_exit_handler" (which does
* a "gds_rundown" of all open regions) BEFORE "secshr_db_clnup" gets called. In that case,
* even if "si" and "csa" are accessible, db shared memory is not and so we should not
* attempt to finish any commits that might seem unfinished in case this process got a SIG-15
* and handled it as part of "deferred_signal_handler" in "t_end"/"tp_tend".
*/
continue;
}
assert(csa->regcnt);
if (NULL != si->first_cw_set)
{
if (csa->jnlpool && (jnlpool != csa->jnlpool))
{
assert(!update_jnlpool || (!csa->jnlpool || (update_jnlpool == csa->jnlpool)));
# ifdef DEBUG
jnlpool_si = si;
jnlpool_csa = csa;
# endif
jnlpool = csa->jnlpool;
}
if (!update_jnlpool && REPL_ALLOWED(csa))
update_jnlpool = jnlpool;
if ( T_UPDATE_UNDERWAY(csa))
{
update_underway = TRUE;
break;
}
}
if (UPDTRNS_TCOMMIT_STARTED_MASK & si->update_trans)
{ /* Two possibilities.
* (a) case of duplicate set not creating any cw-sets but updating db curr_tn++.
* (b) Have completed commit for this region and have released crit on this region.
* (in a potentially multi-region TP transaction).
* In either case, update is underway and the transaction cannot be rolled back.
*/
update_underway = TRUE;
break;
}
}
if (!update_jnlpool)
update_jnlpool = jnlpool;
} else if ((NULL != save_gv_cur_region) && save_gv_cur_region->open && (NULL != cs_addrs))
{ /* Determine update_underway for non-TP transaction */
assert(!(cs_addrs->now_crit && (UPDTRNS_TCOMMIT_STARTED_MASK & update_trans))
|| T_UPDATE_UNDERWAY(cs_addrs));
if (T_UPDATE_UNDERWAY(cs_addrs))
update_underway = TRUE;
update_jnlpool = (cs_addrs && cs_addrs->jnlpool) ? cs_addrs->jnlpool : jnlpool;
}
assert(is_exiting || update_underway);
if (update_underway)
{ /* Now that we know we were in the midst of commit (non-TP or TP) when we encountered an error
* (i.e. "is_exiting" == FALSE) OR when we decided to exit (i.e. "is_exiting" == TRUE)
* finish the commit first. Need to go through regions in ftok order to avoid crit deadlocks.
*/
repl_reg = update_jnlpool ? update_jnlpool->jnlpool_dummy_reg : NULL;
repl_csa = ((NULL != repl_reg) && repl_reg->open) ? REG2CSA(repl_reg) : NULL;
if (dollar_tlevel)
for (si = first_tp_si_by_ftok; NULL != si; si = si->next_tp_si_by_ftok)
secshr_finish_CMT08_to_CMT14(si->tp_csa, update_jnlpool); /* Roll forward steps CMT08 thru CMT14 */
else
{
csa = cs_addrs;
secshr_finish_CMT08_to_CMT14(csa, update_jnlpool); /* Roll forward steps CMT08 thru CMT14 */
}
if ((NULL != repl_csa) && repl_csa->now_crit)
secshr_rel_crit(repl_reg, IS_EXITING_FALSE, IS_REPL_REG_TRUE); /* Step CMT15 */
/* In case of DSE, csa->hold_onto_crit is TRUE and so it is possible the above "secshr_rel_crit"
* call did not release crit. Assert accordingly below.
*/
assert((0 == have_crit(CRIT_HAVE_ANY_REG)) || dse_running);
if (dollar_tlevel)
{
for (si = first_tp_si_by_ftok; NULL != si; si = si->next_tp_si_by_ftok)
{
csa = si->tp_csa;
jrs = si->jbuf_rsrv_ptr;
if (NEED_TO_FINISH_JNL_PHASE2(jrs))
FINISH_JNL_PHASE2_IN_JNLBUFF(csa, jrs); /* Roll forward CMT16 */
}
} else
{
jrs = TREF(nontp_jbuf_rsrv);
if (NEED_TO_FINISH_JNL_PHASE2(jrs))
FINISH_JNL_PHASE2_IN_JNLBUFF(csa, jrs); /* Roll forward CMT16 */
}
if (NULL != repl_csa)
FINISH_JNL_PHASE2_IN_JNLPOOL_IF_NEEDED(TRUE, update_jnlpool); /* Roll forward CMT17 */
/* Roll forward Step CMT18 and CMT19 */
if (dollar_tlevel)
for (si = first_tp_si_by_ftok; NULL != si; si = si->next_tp_si_by_ftok)
secshr_finish_CMT18_to_CMT19(si->tp_csa); /* Roll forward Step CMT18 */
else
secshr_finish_CMT18_to_CMT19(csa); /* Roll forward Step CMT18 */
}
if (save_jnlpool != jnlpool)
jnlpool = save_jnlpool;
if (is_exiting)
{
for (gd_hdr = get_next_gdr(NULL); NULL != gd_hdr; gd_hdr = get_next_gdr(gd_hdr))
{
for (reg = gd_hdr->regions, reg_top = reg + gd_hdr->n_regions; reg < reg_top; reg++)
{
if (!reg->open || reg->was_open)
continue;
if (!IS_REG_BG_OR_MM(reg))
continue;
csa = REG2CSA(reg);
if (NULL == csa)
continue;
assert((reg->read_only && !csa->read_write) || (!reg->read_only && csa->read_write));
SECSHR_SET_CSD_CNL_ISBG(csa, csd, cnl, is_bg); /* sets csd/cnl/is_bg */
numargs = 0;
if (csa->now_crit)
{
numargs = 0;
SECSHR_ACCOUNTING(numargs, argarray, __LINE__);
SECSHR_ACCOUNTING(numargs, argarray, sac_secshr_db_clnup);
SECSHR_ACCOUNTING(numargs, argarray, process_id);
SECSHR_ACCOUNTING(numargs, argarray, secshr_state);
SECSHR_ACCOUNTING(numargs, argarray, csd->trans_hist.curr_tn);
secshr_send_DBCLNUPINFO_msg(csa, numargs, argarray);
}
/* If we hold any latches in the node_local area, release them. Note we do not check
* db_latch here because it is never used by the compare and swap logic but rather
* the aswp logic. Since it is only used for the 3 state cache record lock and
* separate recovery exists for it, we do not do anything with it here.
*/
RELEASE_LATCH_IF_OWNER(&cnl->wc_var_lock);
/* Note: In case of "kill -9", cnl->in_wtstart, cnl->wcs_timers & cnl->ref_cnt will stay uncleaned.
* a) cnl->in_wtstart, "wcs_flu" will eventually set cnl->wc_blocked and force cache recovery
* which will clear it.
* b) cnl->wcs_timers decides if a new flush timer is started. If the 2 processes that have
* pending flush timers are killed abnormally, no new processes will start flush timers for
* the lifetime of the database shared memory. Need a recovery scheme for it.
* c) cnl->ref_cnt is mostly for debug purposes (to know an approximate # of processes attached
* to the database and so it is okay if it is not accurate.
* if it finds */
if ((csa->in_wtstart) && (0 < cnl->in_wtstart))
{
DECR_CNT(&cnl->in_wtstart, &cnl->wc_var_lock);
assert(0 < cnl->intent_wtstart);
if (0 < cnl->intent_wtstart)
DECR_CNT(&cnl->intent_wtstart, &cnl->wc_var_lock);
}
csa->in_wtstart = FALSE; /* Let wcs_wtstart run for exit processing */
if (cnl->wcsflu_pid == process_id)
cnl->wcsflu_pid = 0;
if (is_bg)
{
assert(cnl->sec_size);
cache_state = csa->acc_meth.bg.cache_state;
RELEASE_LATCH_IF_OWNER(&cache_state->cacheq_active.latch);
# ifdef DEBUG
if (gtm_white_box_test_case_enabled && (reg == gd_hdr->regions)
&& (WBTEST_SIGTSTP_IN_T_QREAD == gtm_white_box_test_case_number))
{
assert((NULL != TREF(block_now_locked))
&& ((TREF(block_now_locked))->r_epid == process_id));
}
# endif
if (NULL != (cr = TREF(block_now_locked))) /* done by region to ensure access */
{ /* The following is potentially thread-specific rather than process-specific */
TREF(block_now_locked) = NULL;
RELEASE_LATCH_IF_OWNER(&cr->rip_latch);
if ((cr->r_epid == process_id) && (0 == cr->dirty) && (0 == cr->in_cw_set))
{ /* increment cycle for blk number changes (for tp_hist) */
cr->cycle++;
cr->blk = CR_BLKEMPTY;
/* ensure no bt points to this cr for empty blk */
assert(0 == cr->bt_index);
/* Don't mess with ownership the I/O may not yet be cancelled.
* Ownership will be cleared by whoever gets stuck waiting for the buffer.
*/
}
}
}
assert(!T_UPDATE_UNDERWAY(csa));
if (csa->now_crit || csa->t_commit_crit)
{
numargs = 0;
SECSHR_ACCOUNTING(numargs, argarray, __LINE__);
SECSHR_ACCOUNTING(numargs, argarray, sac_secshr_db_clnup);
SECSHR_ACCOUNTING(numargs, argarray, csa->now_crit);
SECSHR_ACCOUNTING(numargs, argarray, csa->t_commit_crit);
SECSHR_ACCOUNTING(numargs, argarray, csd->trans_hist.early_tn);
SECSHR_ACCOUNTING(numargs, argarray, csd->trans_hist.curr_tn);
SECSHR_ACCOUNTING(numargs, argarray, dollar_tlevel);
secshr_send_DBCLNUPINFO_msg(csa, numargs, argarray);
}
/* The process is exiting. If any kills are in progress (bitmap freeup phase of kill), mark
* kill_in_prog as abandoned. Non-TP and TP maintain kill_in_prog information in different
* structures so access them appropriately. Note that even for a TP transaction, the bitmap
* freeup happens as a non-TP transaction so checking dollar_tlevel is not enough to determine
* if we are in TP or non-TP. first_sgm_info is guaranteed to be non-NULL in the case of a TP
* transaction that is temporarily running its bitmap freeup phase as a non-TP transaction.
* And for true non-TP transactions, first_sgm_info is guaranteed to be NULL. So we use this
* for the determination.
*/
if (NULL != first_sgm_info)
{
si = csa->sgm_info_ptr;
/* Note that it is possible "si" is NULL in case of a GTM-E-MEMORY error in gvcst_init.
* Handle that accordingly in the code below.
*/
/* Since the kill process cannot be completed, we need to decerement KIP count
* and increment the abandoned_kills count.
*/
if (si && (NULL != si->kill_set_head) && (NULL != si->kip_csa))
{
assert(csa == si->kip_csa);
DECR_KIP(csd, csa, si->kip_csa);
INCR_ABANDONED_KILLS(csd, csa);
} else
assert((NULL == si) || (NULL == si->kill_set_head) || (NULL == si->kip_csa));
} else if (!dollar_tlevel)
{
if ((NULL != kip_csa) && (csa == kip_csa))
{
assert(0 < kip_csa->hdr->kill_in_prog);
DECR_KIP(csd, csa, kip_csa);
INCR_ABANDONED_KILLS(csd, csa);
}
}
if (JNL_ENABLED(csd))
{
jbp = csa->jnl->jnl_buff;
RELEASE_LATCH_IF_OWNER(&jbp->fsync_in_prog_latch);
if (jbp->io_in_prog_latch.u.parts.latch_pid == process_id)
RELEASE_SWAPLOCK(&jbp->io_in_prog_latch);
if (jbp->blocked == process_id)
{
assert(csa->now_crit);
jbp->blocked = 0;
}
}
if (csa->freeze && (csd->freeze == process_id) && !csa->persistent_freeze)
{
csd->image_count = 0;
csd->freeze = 0;
}
assert(!csa->t_commit_crit);
if (is_bg && (csa->wbuf_dqd || csa->now_crit || csa->t_commit_crit))
{ /* if csa->wbuf_dqd == TRUE, most likely failed during REMQHI in wcs_wtstart
* or db_csh_get. cache corruption is suspected so set wc_blocked.
* if csa->now_crit is TRUE, someone else should clean the cache, so set wc_blocked.
* if csa->t_commit_crit is TRUE, even if csa->now_crit is FALSE, we might need cache
* cleanup (e.g. cleanup of orphaned cnl->wcs_phase2_commit_pidcnt counter in case
* a process gets shot in the midst of DECR_WCS_PHASE2_COMMIT_PIDCNT macro before
* decrementing the shared counter but after committing the transaction otherwise)
* so set wc_blocked. This case is folded into phase2 cleanup case below.
*/
SET_TRACEABLE_VAR(cnl->wc_blocked, TRUE);
if (csa->now_crit)
{
wcblocked_ptr = WCBLOCKED_NOW_CRIT_LIT;
BG_TRACE_PRO_ANY(csa, wcb_secshr_db_clnup_now_crit);
} else if (csa->wbuf_dqd)
{
wcblocked_ptr = WCBLOCKED_WBUF_DQD_LIT;
BG_TRACE_PRO_ANY(csa, wcb_secshr_db_clnup_wbuf_dqd);
} else
{
wcblocked_ptr = WCBLOCKED_PHASE2_CLNUP_LIT;
BG_TRACE_PRO_ANY(csa, wcb_secshr_db_clnup_phase2_clnup);
}
send_msg_csa(CSA_ARG(csa) VARLSTCNT(8) ERR_WCBLOCKED, 6, LEN_AND_STR(wcblocked_ptr),
process_id, &csd->trans_hist.curr_tn, DB_LEN_STR(reg));
}
csa->wbuf_dqd = 0; /* We can clear the flag now */
if (csa->wcs_pidcnt_incremented)
DECR_WCS_PHASE2_COMMIT_PIDCNT(csa, cnl);
if (csa->now_crit)
{
if (csd->trans_hist.curr_tn == csd->trans_hist.early_tn - 1)
{
assert(WBTEST_ENABLED(WBTEST_JNL_CREATE_FAIL)
|| WBTEST_ENABLED(WBTEST_SLEEP_IN_WCS_WTSTART) || in_wcs_recover);
csd->trans_hist.early_tn = csd->trans_hist.curr_tn;
}
assert(csd->trans_hist.early_tn == csd->trans_hist.curr_tn);
secshr_rel_crit(csa->region, IS_EXITING_TRUE, IS_REPL_REG_FALSE);
}
if (NULL != csa->shmpool_buffer)
{
if ((pid = csa->shmpool_buffer->shmpool_crit_latch.u.parts.latch_pid) == process_id)
{ /* Tiz our lock. Force recovery to run and release */
csa->shmpool_buffer->shmpool_blocked = TRUE;
BG_TRACE_PRO_ANY(csa, shmpool_blkd_by_sdc);
SET_LATCH_GLOBAL(&csa->shmpool_buffer->shmpool_crit_latch, LOCK_AVAILABLE);
DEBUG_LATCH(util_out_print("Latch cleaned up", FLUSH));
} else if ((0 != pid) && (FALSE == is_proc_alive(pid, 0)))
{ /* Attempt to make it our lock so we can set blocked */
if (COMPSWAP_LOCK(&csa->shmpool_buffer->shmpool_crit_latch, pid, 0, process_id, 0))
{ /* Now our lock .. set blocked and release. */
csa->shmpool_buffer->shmpool_blocked = TRUE;
BG_TRACE_PRO_ANY(csa, shmpool_blkd_by_sdc);
DEBUG_LATCH(util_out_print("Orphaned latch cleaned up", TRUE));
COMPSWAP_UNLOCK(&csa->shmpool_buffer->shmpool_crit_latch, process_id,
0, LOCK_AVAILABLE, 0);
} /* Else someone else took care of it */
}
}
/* All releases done now. Double check latch is really cleared */
assert(NULL != csa->critical);
/* as long as csa->hold_onto_crit is FALSE, we should have released crit if we held it at entry */
assert(!csa->now_crit || csa->hold_onto_crit);
# ifndef CRIT_USE_PTHREAD_MUTEX
/* Note: Do not release crit if we still hold it. As we want the next process to grab crit to invoke
* "mutex_salvage" (to cleanup stuff) in case we terminate while holding crit. Hence the below line
* is commented out.
*
* RELEASE_LATCH_IF_OWNER(&csa->critical->semaphore);
*/
RELEASE_LATCH_IF_OWNER(&csa->critical->crashcnt_latch);
RELEASE_LATCH_IF_OWNER(&csa->critical->prochead.latch);
RELEASE_LATCH_IF_OWNER(&csa->critical->freehead.latch);
# endif
} /* For all regions */
} /* For all glds */
if (jnlpool_head)
{
save_jnlpool = jnlpool;
for (jnlpool = jnlpool_head; jnlpool; jnlpool = jnlpool->next)
{
if ((NULL != (repl_reg = jnlpool->jnlpool_dummy_reg)) && repl_reg->open) /* assignment */
{
repl_csa = REG2CSA(repl_reg);
assert(repl_csa);
if (repl_csa->now_crit)
secshr_rel_crit(repl_reg, IS_EXITING_TRUE, IS_REPL_REG_TRUE);
}
}
jnlpool = save_jnlpool;
}
/* It is possible we are exiting while in the middle of a transaction (e.g. called through "deferred_signal_handler"
* in the DEFERRED_EXIT_HANDLING_CHECK macro). Since exit handling code can start new non-TP transactions
* (e.g. for statsdb rundown you need to kill ^%YGS node, for recording mprof stats you need to set a global node)
* clean up the effects of any in-progress transaction before the "t_begin" call happens.
*/
if (!dollar_tlevel)
{
assert(save_gv_cur_region == gv_cur_region);
if ((NULL != save_gv_cur_region) && save_gv_cur_region->open && IS_REG_BG_OR_MM(save_gv_cur_region))
t_abort(save_gv_cur_region, &FILE_INFO(save_gv_cur_region)->s_addrs);
} else
OP_TROLLBACK(0);
}
return;
}
|