File: dtprobed.c

package info (click to toggle)
dtrace 2.0.5-1
  • links: PTS
  • area: main
  • in suites: sid
  • size: 24,408 kB
  • sloc: ansic: 61,247; sh: 17,997; asm: 1,717; lex: 947; awk: 754; yacc: 695; perl: 37; sed: 17; makefile: 15
file content (1303 lines) | stat: -rw-r--r-- 30,672 bytes parent folder | download
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
/*
 * Oracle Linux DTrace; DOF-consumption and storage daemon.
 * Copyright (c) 2022, 2026, Oracle and/or its affiliates. All rights reserved.
 * Licensed under the Universal Permissive License v 1.0 as shown at
 * http://oss.oracle.com/licenses/upl.
 */

/*
 * dtprobed's purpose is simple: listen for ioctls on /dev/dtrace/helper and
 * keep track of USDT probes live in running processes.  dtrace(1) cannot do
 * this because it isn't going to be running all the time, and is almost
 * certainly not going to be running in early boot when most daemons start up.
 * It records this DOF in the DOF stash under /run/dtrace (see dof_stash.c in
 * this directory), and also tracks the identity of the probes contained in it.
 *
 * The DOF is recorded in a pre-parsed form, and parsed on receipt by a forked
 * helper jailed by strict-mode seccomp to prevent DOF contributed by hostile
 * binaries from compromising the system.
 */

#include <sys/param.h>
#include <sys/uio.h>
#include <sys/wait.h>
#include <errno.h>
#include <fcntl.h>
#include <malloc.h>
#include <poll.h>
#include <stdarg.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <syslog.h>
#include <time.h>
#include <unistd.h>
#include <config.h>
#include <libelf.h>

#include <linux/seccomp.h>
#include <sys/syscall.h>

/*
 * Compatibility.  With libfuse 3, we use a few newer features that need at
 * least version 31 (3.2.0): with libfuse 2, the default will suffice.  We also
 * use the logging infrastructure added in libfuse 3.7, but provide an alternate
 * implementation if not available.
 */

#if HAVE_LIBFUSE3
#define FUSE_USE_VERSION 31
#else /* libfuse 2 */
/* Use the default (21).  */
#endif

#include <cuse_lowlevel.h>
#include <fuse_lowlevel.h>
#ifdef HAVE_FUSE_LOG
#include <fuse_log.h>
#else
#include "rpl_fuse_log.h"
#endif
#include <port.h>

#include <dtrace/ioctl.h>

#include <dt_list.h>
#include "usdt_parser.h"
#include "dof_stash.h"
#include "libproc.h"

#include "seccomp-assistance.h"

#define DOF_MAXSZ 512 * 1024 * 1024
#define DOF_CHUNKSZ 64 * 1024

static struct fuse_session *cuse_session;

int _dtrace_debug = 0;
static int foreground;
static int testing;
void dt_debug_dump(int unused) {} 		/* For libproc.  */

static pid_t parser_pid;
static int parser_in_pipe;
static int parser_out_pipe;
static int sync_fd = -1;
static int timeout = 5; 			/* In seconds.  */
static int rq_count = 0;
static int cleanup_interval = 128;		/* In requests.  */

static void helper_ioctl(fuse_req_t req, int cmd, void *arg,
			 struct fuse_file_info *fi, unsigned int flags,
			 const void *in_buf, size_t in_bufsz, size_t out_bufsz);

static const struct cuse_lowlevel_ops dtprobed_clop = {
	.ioctl = helper_ioctl,
};

static int
process_dof(pid_t pid, int out, int in, dev_t dev, ino_t inum, dev_t exec_dev,
	    dev_t exec_inum, dof_helper_t *dh, const usdt_data_t *data,
	    int reparsing);

static void
log_msg(enum fuse_log_level level, const char *fmt, va_list ap)
{
	if (!_dtrace_debug && level > FUSE_LOG_INFO)
		return;

	if (foreground) {
		if (testing)
			fprintf(stderr, "dtprobed DEBUG %li: ", time(NULL));
		vfprintf(stderr, fmt, ap);
	} else if (sync_fd >= 0) {
		if (testing)
			daemon_log(sync_fd, "dtprobed DEBUG %li: ", time(NULL));
		daemon_vlog(sync_fd, fmt, ap);
	} else
		vsyslog(level, fmt, ap);
}

/* For libproc */
void
dt_debug_printf(const char *subsys, const char *fmt, va_list ap)
{
	if (!_dtrace_debug)
		return;

	if (foreground) {
		if (testing)
			fprintf(stderr, "%s DEBUG %li: ", subsys, time(NULL));
		vfprintf(stderr, fmt, ap);
	} else if (sync_fd >= 0)
		daemon_vlog(sync_fd, fmt, ap);
	else
		/* Subsystem discarded (it's always 'libproc' anyway).  */
		vsyslog(LOG_DEBUG, fmt, ap);
}

#if HAVE_LIBFUSE3
static int
session_fd(struct fuse_session *cuse_session)
{
	return fuse_session_fd(cuse_session);
}
#else /* libfuse 2 */
static struct fuse_chan *cuse_chan;

static void
init_cuse_chan(struct fuse_session *cuse_session)
{
	cuse_chan = fuse_session_next_chan(cuse_session, NULL);
}

static int
session_fd(struct fuse_session *cuse_session)
{
	return fuse_chan_fd(cuse_chan);
}
#endif

/*
 * States for the ioctl processing loop, which gets repeatedly called due to the
 * request/reply nature of unrestricted FUSE ioctls.
 */
typedef enum dtprobed_fuse_state {
	DTP_IOCTL_START = 0,
	DTP_IOCTL_HDR = 1,
	DTP_IOCTL_DOFHDR = 2,
	DTP_IOCTL_DOFCHUNK = 3,
	DTP_IOCTL_DOF = 4
} dtprobed_fuse_state_t;

/*
 * State crossing calls to CUSE request functions.
 */
typedef struct dtprobed_userdata {
	dt_list_t list;
	pid_t pid;
	dtprobed_fuse_state_t state;
	dof_helper_t dh;
	dof_hdr_t dof_hdr;
	size_t next_chunk;
	char *buf;
} dtprobed_userdata_t;

static dt_list_t all_userdata;

struct fuse_session *
setup_helper_device(int argc, char **argv, char *devname)
{
	struct cuse_info ci;
	struct fuse_session *cs;
	char *args;
	int multithreaded;

	memset(&ci, 0, sizeof(struct cuse_info));

	ci.flags = CUSE_UNRESTRICTED_IOCTL;
	ci.dev_info_argc = 1;
	if (asprintf(&args,"DEVNAME=%s", devname) < 0) {
		perror("allocating helper device");
		exit(2);			/* Allow restarting.  */
	}

	const char *dev_info_argv[] = { args };
	ci.dev_info_argv = dev_info_argv;

	cs = cuse_lowlevel_setup(argc, argv, &ci, &dtprobed_clop,
				 &multithreaded, NULL);

	if (cs == NULL) {
		perror("allocating helper device");
		return NULL;
	}

#ifndef HAVE_LIBFUSE3 /* libfuse 2 */
	init_cuse_chan(cs);
#endif

	if (multithreaded) {
		fprintf(stderr, "CUSE thinks dtprobed is multithreaded!\n");
		fprintf(stderr, "This should never happen.\n");
		errno = EINVAL;
		return NULL;
	}

	free(args);
	return cs;
}

void
teardown_device(void)
{
	/* This is automatically called on SIGTERM.  */
	cuse_lowlevel_teardown(cuse_session);
}

/*
 * Allocate a userdata for the given PID, or return an existing one
 * if possible.  Userdatas are insertion-sorted into descending order
 * to make old stale ones from processes killed at the wrong time a
 * bit less likely to impose overhead.
 */
static dtprobed_userdata_t *
get_userdata(pid_t pid)
{
	dtprobed_userdata_t *userdatap;
	dtprobed_userdata_t *new;
	int prepend = 0;

	for (userdatap = dt_list_next(&all_userdata); userdatap != NULL;
	     userdatap = dt_list_next(userdatap)) {
		if (userdatap->pid == pid)
			return userdatap;

		if (userdatap->pid < pid)
			break;
	}

	if ((new = calloc(1, sizeof(struct dtprobed_userdata))) == NULL)
		return NULL;

	/*
	 * Go back one, to the last list entry with a pid higher than this one,
	 * since dt_list_insert inserts *after* the entry passed in.
	 */
	if (userdatap) {
		userdatap = dt_list_prev(userdatap);
		if (userdatap == NULL)
			prepend = 1;
	}

	new->pid = pid;

	if (!prepend)
		dt_list_insert(&all_userdata, userdatap, new);
	else
		dt_list_prepend(&all_userdata, new);

	return new;
}

/*
 * Clean up userdatas no longer involved in ioctls or no longer corresponding to
 * live pids.
 */
static void
cleanup_userdata(void)
{
	dtprobed_userdata_t *userdatap;
	dtprobed_userdata_t *last_userdatap = NULL;

	for (userdatap = dt_list_next(&all_userdata); userdatap != NULL;
	     userdatap = dt_list_next(userdatap)) {
		if (userdatap->state == DTP_IOCTL_START || kill(userdatap->pid, 0) < 0) {
			dt_list_delete(&all_userdata, userdatap);
			free(userdatap->buf);
			free(last_userdatap);
			last_userdatap = userdatap;
		}
	}
	free(last_userdatap);
}

/*
 * Parse a piece of DOF.  Return 0 iff the pipe has closed or no more parsing
 * is possible.
 */
static int
parse_dof(int in, int out)
{
	int ok;
	dof_helper_t *dh;
	usdt_data_t *data;

	dh = usdt_copyin_helper(in);
	if (!dh)
		return 0;

	data = usdt_copyin_data(in, out, &ok);
	if (!data) {
		free(dh);
		return ok;
	}

	usdt_parse(out, dh, data);

	return ok;
}

/*
 * Kick off the sandboxed USDT parser.  This is run in a seccomp()ed subprocess,
 * and sends a stream of dof_parsed_t back to this process.
 */
static void
usdt_parser_start(void)
{
	int parser_in[2], parser_out[2];
	if ((pipe(parser_in) < 0) ||
	    (pipe(parser_out) < 0)) {
		fuse_log(FUSE_LOG_ERR, "cannot create DOF parser pipes: %s",
			 strerror(errno));
		exit(1);
	}

	parser_out_pipe = parser_in[1];
	parser_in_pipe = parser_out[0];

	switch (parser_pid = fork()) {
	case -1: {
		fuse_log(FUSE_LOG_ERR, "cannot fork DOF parser: %s",
			 strerror(errno));
		exit(1);
	}
	case 0: {
		/*
		 * Sandboxed parser child.  Close unwanted fds and nail into
		 * seccomp jail.
		 */
		close(session_fd(cuse_session));
		close(parser_out_pipe);
		close(parser_in_pipe);
		if (!foreground) {
			close(sync_fd);
			sync_fd = -1;
		}

		/*
		 * Reporting errors at this point is difficult: we have already
		 * closed all pipes that we might use to report it.  Just exit 1
		 * and rely on the admin using strace :(
		 *
		 * Take additional measures to ensure that we can still do
		 * sufficiently-large mallocs in the child without needing to
		 * make any syscalls.
		 *
		 * Don't lock the process in jail if debugging (but still run in
		 * a child process).
		 */
		mallopt(M_MMAP_MAX, 0);
		mallopt(M_TRIM_THRESHOLD, -1);
		seccomp_fill_memory_really_free(seccomp_fill_memory_really_malloc(DOF_MAXSZ * 2));

		if (!_dtrace_debug)
			if (syscall(SYS_seccomp, SECCOMP_SET_MODE_STRICT, 0, NULL) < 0)
				_exit(1);

		while (parse_dof(parser_in[0], parser_out[1]))
			;
		_exit(0);
	}
	}

	close(parser_in[0]);
	close(parser_out[1]);
}

/*
 * Clean up wreckage if the USDT parser dies: optionally restart it.
 */
static void
usdt_parser_tidy(int restart)
{
	int status = 0;

	if (parser_pid == 0)
		return;

	kill(parser_pid, SIGKILL);
	if (errno != ESRCH)
		while (waitpid(parser_pid, &status, 0) < 0 && errno == EINTR);

	close(parser_in_pipe);
	close(parser_out_pipe);

	if (restart)
		usdt_parser_start();
}

static dof_parsed_t *
usdt_read(pid_t pid, int in)
{
	dof_parsed_t *reply = usdt_parser_host_read(in, timeout);

	if (!reply)
		return NULL;

	/*
	 * Log errors.
	 */
	if (reply->type == DIT_ERR) {
		errno = reply->err.err_no;
		fuse_log(FUSE_LOG_WARNING, "%i: dtprobed: DOF parsing error: "
			 "%s\n", pid, reply->err.err);
		free(reply);
		reply = NULL;
	}

	return reply;
}

/*
 * Retrieve and process USDT probe data from a .note.usdt section.
 * The .rodata section is also needed because function names are stored there.
 */
static int
handle_usdt_notes(pid_t pid, uintptr_t addr)
{
	ps_prochandle *P = NULL;
	const prmap_t *mapp, *exec_mapp;
	const prmap_file_t *prf;
	dof_helper_t dh;
	const char *fn, *mod;
	int fd = -1;
	Elf *elf = NULL;
	size_t shstrndx;
	GElf_Shdr shdr;
	size_t nbase, dbase;
	Elf_Scn *scn = NULL, *nscn = NULL, *dscn = NULL;;
	GElf_Ehdr ehdr;
	Elf_Data *elfd, *elfn;
	usdt_data_t ndata, ddata;
	dev_t dev, exec_dev;
	ino_t inum, exec_inum;
	int gen = -1, err;

	/* Grab the process. */
	if ((P = Pgrab(pid, 2, 0, NULL, &err)) == NULL) {
		fuse_log(FUSE_LOG_ERR, "%i: dtprobed: process grab failed: %s\n",
			 pid, strerror(err));
		return -1;
	}

	/* Retrieve mapping information. */
	mapp = Paddr_to_map(P, addr);
	if (mapp == NULL) {
		fuse_log(FUSE_LOG_ERR, "%i: dtprobed: cannot look up mapping (process dead?)\n",
			 pid);
		goto out;
	}

	dev = mapp->pr_dev;
	inum = mapp->pr_inum;

	prf = mapp->pr_file;
	if (prf == NULL || (mapp = prf->first_segment) == NULL) {
		fuse_log(FUSE_LOG_ERR, "%i: dtprobed: cannot look up mapping (process dead?)\n",
			 pid);
		goto out;
	} else if (prf->prf_mapname == NULL ||
		   (fn = Pmap_mapfile_name(P, mapp)) == NULL) {
		fuse_log(FUSE_LOG_ERR, "%i: dtprobed: cannot look up mapname (process dead?)\n",
			 pid);
		goto out;
	}
	mod = strrchr(prf->prf_mapname, '/');
	if (mod)
		mod++;
	else
		mod = prf->prf_mapname;
	snprintf(dh.dofhp_mod, sizeof(dh.dofhp_mod), "%s", mod);

	dh.dofhp_addr = mapp->pr_vaddr;
	dh.dofhp_dof = 0;

	fuse_log(FUSE_LOG_DEBUG, "%i: DOF helper { '%s', %lx, %lx }\n",
		 pid, dh.dofhp_mod, dh.dofhp_addr, dh.dofhp_dof);

	exec_mapp = Plmid_to_map(P, LM_ID_BASE, PR_OBJ_EXEC);
	if (exec_mapp == NULL || (prf = exec_mapp->pr_file) == NULL) {
		fuse_log(FUSE_LOG_ERR, "%i: dtprobed: cannot look up mapping (process dead?)\n",
			 pid);
		goto out;
	}

	exec_dev = exec_mapp->pr_dev;
	exec_inum = exec_mapp->pr_inum;

	/* Open the mapping. */
	if ((fd = open(fn, O_RDONLY)) < 0) {
		fuse_log(FUSE_LOG_ERR, "%i: dtprobed: cannot open %s: %s\n",
			 pid, fn, strerror(errno));
		goto out;
	}

	Prelease(P, PS_RELEASE_NORMAL);
	Pfree(P);
	P = NULL;

	/* Retrieve the .note.usdt ELF section. */
	elf_version(EV_CURRENT);
	if ((elf = elf_begin(fd, ELF_C_READ_MMAP, NULL)) == NULL ||
	     elf_kind(elf) != ELF_K_ELF)
		goto elf_err;

	elf_getshdrstrndx(elf, &shstrndx);
	if (gelf_getehdr(elf, &ehdr) == NULL)
		goto elf_err;
	if (ehdr.e_type == ET_EXEC)
		dh.dofhp_addr = 0;

	while ((scn = elf_nextscn(elf, scn)) != NULL) {
		const char *name;

		if (gelf_getshdr(scn, &shdr) == NULL)
			goto elf_err;

		if (shdr.sh_type == SHT_NOTE &&
		    (name = elf_strptr(elf, shstrndx, shdr.sh_name)) &&
		    strcmp(name, ".note.usdt") == 0) {
			nscn = scn;
			nbase = shdr.sh_addr;
		} else if (shdr.sh_type == SHT_PROGBITS &&
		    (name = elf_strptr(elf, shstrndx, shdr.sh_name)) &&
		    strcmp(name, ".rodata") == 0) {
			dscn = scn;
			dbase = shdr.sh_addr;
		}
	}

	if (nscn == NULL) {
		fuse_log(FUSE_LOG_ERR, "%i: dtprobed: no %s section in %s\n",
			 pid, ".note.usdt", dh.dofhp_mod);
		goto out;
	}
	if (dscn == NULL) {
		fuse_log(FUSE_LOG_ERR, "%i: dtprobed: no %s section in %s\n",
			 pid, ".rodata", dh.dofhp_mod);
		goto out;
	}

	if ((elfn = elf_getdata(nscn, 0)) == NULL ||
	    (elfd = elf_getdata(dscn, 0)) == NULL)
		goto elf_err;

	fuse_log(FUSE_LOG_DEBUG,
		 "%i: %s with %s section (%lu bytes), %s section (%lu bytes)\n",
		 pid, dh.dofhp_mod, ".note.usdt", elfn->d_size, ".rodata",
		 elfd->d_size);

	ndata.base = nbase;
	ndata.size = elfn->d_size;
	ndata.buf = elfn->d_buf;
	ndata.next = &ddata;
	ddata.base = dbase;
	ddata.size = elfd->d_size;
	ddata.buf = elfd->d_buf;
	ddata.next = NULL;
	gen = process_dof(pid, parser_out_pipe, parser_in_pipe, dev, inum,
			  exec_dev, exec_inum, &dh, &ndata, 0);

	goto out;

elf_err:
	fuse_log(FUSE_LOG_ERR, "%i: dtprobed: cannot read ELF %s: %s\n",
		 pid, dh.dofhp_mod, elf_errmsg(elf_errno()));

out:
	if (elf)
		elf_end(elf);
	if (fd >= 0)
		close(fd);
	if (P) {
		Prelease(P, PS_RELEASE_NORMAL);
		Pfree(P);
	}

	return gen;
}

/*
 * Get the (dev, inum) pair for the mapping the passed-in addr belongs to in the
 * given pid.  (If there are multiple, it doesn't matter which we choose as long
 * as we are consistent.)
 */
static const int
mapping_dev_inums(pid_t pid, uintptr_t addr, dev_t *dev, ino_t *inum,
		  dev_t *exec_dev, dev_t *exec_inum)
{
	ps_prochandle *P;
	const prmap_t *mapp, *exec_mapp;
	int err = 0;

	if ((P = Pgrab(pid, 2, 0, NULL, &err)) == NULL) {
		fuse_log(FUSE_LOG_ERR, "%i: dtprobed: process grab failed: %s\n",
			 pid, strerror(err));
		return -1;
	}

	mapp = Paddr_to_map(P, addr);
	exec_mapp = Plmid_to_map(P, LM_ID_BASE, PR_OBJ_EXEC);

	err = -1;
	if (mapp == NULL || exec_mapp == NULL) {
		fuse_log(FUSE_LOG_ERR, "%i: dtprobed: cannot look up mapping (process dead?)\n",
			 pid);
		goto out;
	}

	*dev = mapp->pr_dev;
	*inum = mapp->pr_inum;
	*exec_dev = exec_mapp->pr_dev;
	*exec_inum = exec_mapp->pr_inum;

	err = 0;
out:
	Prelease(P, PS_RELEASE_NORMAL);
	Pfree(P);

	return err;
}

/*
 * Core ioctl() helper.  Repeatedly reinvoked after calls to
 * fuse_reply_ioctl_retry, once per dereference.
 */
static void
helper_ioctl(fuse_req_t req, int cmd, void *arg,
	     struct fuse_file_info *fi, unsigned int flags,
	     const void *in_buf, size_t in_bufsz, size_t out_bufsz)
{
	struct iovec in;
	pid_t pid = fuse_req_ctx(req)->pid;
	dtprobed_userdata_t *userdata = get_userdata(pid);
	const char *errmsg;
	const void *buf;
	dev_t dev = 0, exec_dev = 0;
	ino_t inum = 0, exec_inum = 0;
	int gen;
	usdt_data_t data;

	/*
	 * We can just ignore FUSE_IOCTL_COMPAT: the 32-bit and 64-bit versions
	 * of the DOF structures are intentionally identical.
	 */

	switch (cmd) {
	case DTRACEHIOC_HASUSDT:
		fuse_log(FUSE_LOG_DEBUG, "DTRACEHIOC_HASUSDT from PID %i, addr %lx\n",
			 pid, (uintptr_t) arg);
		if ((gen = handle_usdt_notes(pid, (uintptr_t) arg)) < 0)
			goto process_err;

		goto process_done;
	case DTRACEHIOC_ADDDOF:
		break;
	case DTRACEHIOC_REMOVE:
		fuse_log(FUSE_LOG_DEBUG, "DTRACEHIOC_REMOVE from PID %i, generation %lu\n",
			 pid, (uintptr_t) arg);
		gen = dof_stash_remove(pid, (uintptr_t) arg);
		fuse_reply_ioctl(req, gen, NULL, 0);
		return;
	default: errmsg = "invalid ioctl";
		fuse_log(FUSE_LOG_WARNING, "%i: dtprobed: %s %x\n",
			 pid, errmsg, cmd);
		goto fuse_err;
	}

	/*
	 * First call: get the ioctl arg content, a dof_helper_t.
	 */
	if (userdata->state == DTP_IOCTL_START) {
		in.iov_base = arg;
		in.iov_len = sizeof(dof_helper_t);

		fuse_log(FUSE_LOG_DEBUG, "DTRACEHIOC_ADDDOF from PID %i\n", pid);

		errmsg = "error reading ioctl size";
		if (fuse_reply_ioctl_retry(req, &in, 1, NULL, 0) < 0)
			goto fuse_errmsg;

		/*
		 * userdata->buf should already be freed, but if somehow it is
		 * non-NULL make absolutely sure it is cleared out.
		 */
		userdata->buf = NULL;
		userdata->next_chunk = 0;
		userdata->state = DTP_IOCTL_HDR;
		return;
	}

	/*
	 * Second call: validate the dof_hdr_t length, get the initial DOF.
	 */
	if (userdata->state == DTP_IOCTL_HDR) {
		if (in_bufsz != sizeof(dof_helper_t)) {
			errmsg = "helper size incorrect";
			fuse_log(FUSE_LOG_ERR, "%i: dtprobed: %s: "
				 "expected at least %zi, not %zi\n", pid,
				 errmsg, sizeof(dof_helper_t), in_bufsz);
			goto fuse_err;
		}
		memcpy(&userdata->dh, in_buf, sizeof(dof_helper_t));

		in.iov_base = (void *) userdata->dh.dofhp_dof;
		in.iov_len = sizeof(dof_hdr_t);

		errmsg = "cannot read DOF header";
		if (fuse_reply_ioctl_retry(req, &in, 1, NULL, 0) < 0)
			goto fuse_errmsg;

		userdata->state = DTP_IOCTL_DOFHDR;
		return;
	}

	/*
	 * From here on we are always fetching DOF: the inbound buffer must be
	 * at least as big as the DOF header.
	 */
	if (in_bufsz < sizeof(dof_hdr_t)) {
		errmsg = "DOF too small";
		fuse_log(FUSE_LOG_ERR, "%i: dtprobed: %s: expected at least %zi, "
			 "not %zi\n", pid, errmsg, sizeof(dof_hdr_t), in_bufsz);
		goto fuse_err;
	}

	/*
	 * Third call: validate the DOF length and stash it away.
	 */
	if (userdata->state == DTP_IOCTL_DOFHDR) {
		/*
		 * Too much data is as bad as too little.
		 */
		if (in_bufsz > sizeof(dof_hdr_t)) {
			errmsg = "DOF header size incorrect";
			fuse_log(FUSE_LOG_ERR, "%i: dtprobed: %s: %zi, not %zi\n",
				 pid, errmsg, in_bufsz, sizeof(dof_hdr_t));
			goto fuse_err;
		}
		memcpy(&userdata->dof_hdr, in_buf, sizeof(dof_hdr_t));

		if (userdata->dof_hdr.dofh_loadsz > DOF_MAXSZ)
			fuse_log(FUSE_LOG_WARNING, "%i: dtprobed: DOF size of %zi longer than is sane\n",
				 pid, userdata->dof_hdr.dofh_loadsz);

		/* Fall through. */
	}

	/*
	 * Third and possibly further calls: get the DOF itself, in chunks if
	 * it's too big.
	 */
	if (userdata->state == DTP_IOCTL_DOFHDR ||
		userdata->state == DTP_IOCTL_DOFCHUNK) {
		int by_chunks = 0;

		in.iov_base = (void *) userdata->dh.dofhp_dof;
		in.iov_len = userdata->dof_hdr.dofh_loadsz;

		/*
		 * If the data being read in is too large, read by chunks.  We
		 * cannot determine the chunk size we can use: the first failure
		 * returns -ENOMEM to the caller of ioctl().
		 */
		if (userdata->dof_hdr.dofh_loadsz > DOF_CHUNKSZ) {
			by_chunks = 1;

			if (userdata->state == DTP_IOCTL_DOFHDR) {
				userdata->buf = malloc(userdata->dof_hdr.dofh_loadsz);
				in.iov_len = DOF_CHUNKSZ;

				if (userdata->buf == NULL) {
					errmsg = "out of memory allocating DOF";
					goto fuse_errmsg;
				}
			} else {
				ssize_t remaining;
				size_t to_copy;

				remaining = userdata->dof_hdr.dofh_loadsz -
					    userdata->next_chunk;

				if (remaining < 0)
					remaining = 0;

				/*
				 * We've already read a chunk: preserve it and
				 * go on to the next, until we are out.
				 */

				to_copy = MIN(in_bufsz, remaining);
				memcpy(userdata->buf + userdata->next_chunk, in_buf,
				       to_copy);

				userdata->next_chunk += to_copy;
				remaining -= to_copy;

				if (remaining <= 0) {
					userdata->state = DTP_IOCTL_DOF;
					goto chunks_done;
				}

				in.iov_base = (char *) in.iov_base + userdata->next_chunk;
				in.iov_len = MIN(DOF_CHUNKSZ, remaining);
			}
		}

		errmsg = "cannot read DOF";
		if (fuse_reply_ioctl_retry(req, &in, 1, NULL, 0) < 0)
			goto fuse_errmsg;

		if (by_chunks)
			userdata->state = DTP_IOCTL_DOFCHUNK;
		else
			userdata->state = DTP_IOCTL_DOF;
		return;
	}

chunks_done:
	if (userdata->state != DTP_IOCTL_DOF) {
		errmsg = "FUSE internal state incorrect";
		goto fuse_errmsg;
	}

	/*
	 * Final call: DOF acquired.  Pass to parser for processing, then reply
	 * to unblock the ioctl() caller and return to start state.
	 */

	buf = in_buf;
	if (userdata->buf)
		buf = userdata->buf;

	if ((mapping_dev_inums(pid, userdata->dh.dofhp_dof, &dev, &inum,
		     &exec_dev, &exec_inum)) < 0)
		goto process_err;

	data.base = 0;
	data.size = userdata->dof_hdr.dofh_loadsz;
	data.buf = (void *)buf;
	data.next = NULL;
	if ((gen = process_dof(pid, parser_out_pipe, parser_in_pipe,
			       dev, inum, exec_dev, exec_inum, &userdata->dh,
			       &data, 0)) < 0)
		goto process_err;

process_done:
	if (fuse_reply_ioctl(req, gen, NULL, 0) < 0)
		goto process_err;

	free(userdata->buf);
	userdata->buf = NULL;
	userdata->state = DTP_IOCTL_START;

	/*
	 * Periodically clean up old userdata (applying to pids with no live
	 * transaction, or pids which no longer exist).
	 */
	if (rq_count++ > cleanup_interval) {
		cleanup_userdata();
		dof_stash_prune_dead();
		rq_count = 0;
	}

	return;

fuse_errmsg:
	fuse_log(FUSE_LOG_ERR, "%i: dtprobed: %s\n", pid, errmsg);

fuse_err:
	if (fuse_reply_err(req, EINVAL) < 0)
		fuse_log(FUSE_LOG_ERR, "%i: dtprobed: %s\n", pid,
			 "cannot send error to ioctl caller\n");
	free(userdata->buf);
	userdata->buf = NULL;
	userdata->state = DTP_IOCTL_START;
	return;

process_err:
	if (fuse_reply_err(req, EINVAL) < 0)
		fuse_log(FUSE_LOG_ERR, "%i: cannot unblock caller\n",
			 pid);
	free(userdata->buf);
	userdata->buf = NULL;
	userdata->state = DTP_IOCTL_START;
	return;
}

/*
 * Process some DOF, passing it to the parser and stashing it away for later.
 *
 * If reparsing is set, we are re-parsing existing DOF and should only update
 * the parsed DOF representation.
 */
static int
process_dof(pid_t pid, int out, int in, dev_t dev, ino_t inum, dev_t exec_dev,
	    dev_t exec_inum, dof_helper_t *dh, const usdt_data_t *data,
	    int reparsing)
{
	dof_parsed_t *provider;
	size_t i;
	size_t tries = 0;
	int gen = 0;
	const char *errmsg;
	dt_list_t accum = {0};

	do {
		errmsg = "DOF parser write failed";
		while ((errno = usdt_parser_host_write(out, dh, data)) == EAGAIN);
		if (errno != 0)
			goto err;

		/*
		 * Wait for parsed reply.  If it fails, try once more; possibly
		 * the child was killed due to problems induced by a previous
		 * probe, and a retry will work.
		 */

		errmsg = "parsed DOF read failed";
		provider = usdt_read(pid, in);
		if (!provider) {
			if (tries++ > 0)
				goto err;
			/*
			 * Tidying reopens the parser in and out pipes: catch
			 * up with this.
			 */
			usdt_parser_tidy(1);
			out = parser_out_pipe;
			in = parser_in_pipe;
			continue;
		}
		if (provider->type != DIT_PROVIDER && provider->type != DIT_EOF)
			goto err;
		break;
	} while (!provider);

	while (provider->type != DIT_EOF) {
		if (dof_stash_push_parsed(&accum, provider) < 0)
			goto oom;

		fuse_log(FUSE_LOG_DEBUG, "Parser read: provider %s, %li probes\n",
			 provider->provider.name, provider->provider.nprobes);

		for (i = 0; i < provider->provider.nprobes; i++) {
			dof_parsed_t *probe = usdt_read(pid, in);
			size_t j;

			errmsg = "no probes in this provider, or parse state corrupt";
			if (!probe || probe->type != DIT_PROBE)
				goto err;

			if (_dtrace_debug) {
				const char *mod, *fun, *prb;

				mod = probe->probe.name;
				fun = mod + strlen(mod) + 1;
				prb = fun + strlen(fun) + 1;
				fuse_log(FUSE_LOG_DEBUG,
					 "Parser read: adding %s:%s:%s:%s to stash\n",
					 provider->provider.name,
					 mod, fun, prb);
			}

			if (dof_stash_push_parsed(&accum, probe) < 0)
				goto oom;

			j = 0;
			do {
				dof_parsed_t *tp = usdt_read(pid, in);

				errmsg = "no tracepoints in a probe, or parse state corrupt";
				if (!tp || tp->type == DIT_PROVIDER ||
				    tp->type == DIT_PROBE || tp->type == DIT_EOF)
					goto err;

				if (dof_stash_push_parsed(&accum, tp) < 0)
					goto oom;

				if (tp->type == DIT_TRACEPOINT)
					j++;
			} while (j < probe->probe.ntp);
		}

		errmsg = "subsequent provider read failed, or stream not properly terminated";
		provider = usdt_read(pid, in);
		if (!provider)
			goto err;
	}

	/* Push the EOF on.  */

	if (dof_stash_push_parsed(&accum, provider) < 0)
		goto oom;

	if (!reparsing)
		if ((gen = dof_stash_add(pid, dev, inum, exec_dev, exec_inum,
					 dh, data)) < 0)
			goto fileio;

	if (dof_stash_write_parsed(pid, dev, inum, &accum) < 0) {
		dof_stash_free(&accum);
		dof_stash_remove(pid, gen);
		goto fileio;
	}
	dof_stash_free(&accum);

	return gen;

oom:
	fuse_log(FUSE_LOG_ERR, "%i: out of memory parsing DOF\n", pid);
	goto proc_err;

err:
	fuse_log(FUSE_LOG_ERR, "%i: dtprobed: parser error: %s\n", pid, errmsg);
	goto proc_err;

fileio:
	fuse_log(FUSE_LOG_ERR, "%i: dtprobed: I/O error stashing DOF: %s\n",
		 pid, strerror(errno));

proc_err:
	dof_stash_free(&accum);
	usdt_parser_tidy(1);
	return -1;
}

#if HAVE_LIBFUSE3
static int
loop(void)
{
	struct fuse_buf fbuf = { .mem = NULL };
	struct pollfd fds[1];
	int ret = 0;

	fds[0].fd = fuse_session_fd(cuse_session);
	fds[0].events = POLLIN;

	while (!fuse_session_exited(cuse_session)) {
		if (poll(fds, 1, -1) < 0) {
			if (errno == EINTR)
				continue;
			break;
		}

		if (fds[0].revents != 0) {
			if ((ret = fuse_session_receive_buf(cuse_session,
							    &fbuf)) <= 0) {
				if (ret == -EINTR)
					continue;

				break;
			}

			fuse_session_process_buf(cuse_session, &fbuf);
		}
	}

	free(fbuf.mem);
	fuse_session_reset(cuse_session);
	return ret < 0 ? -1 : 0;
}
#else /* libfuse 2 */
static int
loop(void)
{
	struct pollfd fds[1];
	void *buf;
	size_t bufsize;
	int ret = 0;

	bufsize = fuse_chan_bufsize(cuse_chan);
	buf = malloc(bufsize);
	if (!buf) {
		fprintf(stderr, "Cannot allocate memory for FUSE buffer\n");
		return -1;
	}

	fds[0].fd = fuse_chan_fd(cuse_chan);
	fds[0].events = POLLIN;

	while (!fuse_session_exited(cuse_session)) {
		struct fuse_buf fbuf = { .mem = buf, .size = bufsize };
		struct fuse_chan *tmpch = cuse_chan;

		if (poll(fds, 1, -1) < 0) {
			if (errno == EINTR)
				continue;
			break;
		}

		if (fds[0].revents != 0) {
			if ((ret = fuse_session_receive_buf(cuse_session,
							    &fbuf, &tmpch)) <= 0) {
				if (ret == -EINTR)
					continue;

				break;
			}

#ifdef HAVE_FUSE_NUMA
			fuse_session_process_buf(cuse_session, &fbuf, tmpch, 0);
#else
			fuse_session_process_buf(cuse_session, &fbuf, tmpch);
#endif
		}
	}

	free(buf);
	fuse_session_reset(cuse_session);
	return ret < 0 ? -1 : 0;
}
#endif

/*
 * Force a reparse of all parsed DOF on demand.
 *
 * Only hooked up during in-source-tree testing.
 */
static void
force_reparse(int sig)
{
	fuse_log(FUSE_LOG_DEBUG, "Forced reparse\n");
	reparse_dof(parser_out_pipe, parser_in_pipe, process_dof, 1);
	fuse_log(FUSE_LOG_DEBUG, "Forced reparse complete\n");
}

static void
usage(void)
{
	fprintf(stderr, "Syntax: dtprobed [-F] [-d] [-n devname] [-t timeout]\n");
	exit(1);
}

int
main(int argc, char *argv[])
{
	int opt;
	char *devname = "dtrace/helper";
	char *statedir = NULL;
	int ret;
	struct sigaction sa = {0};

	/*
	 * These are "command-line" arguments to FUSE itself: our args are
	 * different.  The double-NULL allows us to add an arg.
	 */
	char *fuse_argv[] = { argv[0], "-f", "-s", NULL, NULL };
	int fuse_argc = 3;

	while ((opt = getopt(argc, argv, "Fs:dn:t:")) != -1) {
		switch (opt) {
		case 'F':
			foreground = 1;
			break;
		case 'n':
			devname = strdup(optarg);
			break;
		case 'd':
			if (!_dtrace_debug) {
				_dtrace_debug = 1;
				fuse_argv[fuse_argc++] = "-d";
			}
			break;
		case 's':
			/*
			 * This option is purely for the testsuite, so does not
			 * appear in the usage() description.
			 */
			statedir = strdup(optarg);
			break;
		case 't':
			timeout = atoi(optarg);
			if (timeout <= 0) {
				fprintf(stderr, "Error: timeout must be a "
					"positive integer, not %s\n", optarg);
				exit(1);
			}
			break;
		default:
			usage();
		}
	}

	if (optind < argc)
		usage();

	/*
	 * Close all fds before doing anything else: we cannot close them during
	 * daemonization because CUSE opens fds of its own that we want to keep
	 * around.
	 */
	close_range(3, ~0U, 0);

	if ((cuse_session = setup_helper_device(fuse_argc, fuse_argv,
						devname)) == NULL)
		exit(1);

	if (!foreground) {
		if ((sync_fd = daemonize(0)) < 0) {
			teardown_device();
			exit(2);
		}
	}

	/*
	 * We are now daemonized, if we need to be.  Arrange to log to syslog,
	 * fire off the jailed parser subprocess, then report successful startup
	 * down our synchronization pipe (by closing it) and tell systemd (if
	 * present) that we have started.
	 */
	fuse_set_log_func(log_msg);

	/*
	 * Ignore SIGPIPE to allow for a non-hideous way to detect parser
	 * process death.
	 */
	sa.sa_handler = SIG_IGN;
	(void) sigaction(SIGPIPE, &sa, NULL);

	/*
	 * When doing (in-tree) testing, use a shorter cleanup interval, and
	 * hook up a signal allowing the test scripts to force a DOF cleanup.
	 */
	if (getenv("_DTRACE_TESTING")) {
		struct sigaction tmp = {0};

		cleanup_interval = 5;
		tmp.sa_handler = force_reparse;
		tmp.sa_flags = SA_RESTART;
		(void) sigaction(SIGUSR2, &tmp, NULL);
		testing = 1;
	}

	usdt_parser_start();

	if (dof_stash_init(statedir) < 0)
		exit(1);

	/*
	 * Who knows what DOF-containing processes have died since the daemon
	 * was running?  Clean them up.
	 */
	dof_stash_prune_dead();

	/*
	 * Make sure that old parsed DOF is deleted and reparsed: it might come
	 * from a prior daemon invocation, with an older daemon version with a
	 * different parsed DOF representation.  If we can't do this, don't try
	 * restarting the daemon: it'll just fail again in the same way.
	 */
	if (reparse_dof(parser_out_pipe, parser_in_pipe, process_dof, 0) < 0) {
		teardown_device();
		exit(1);
	}

	if (!foreground) {
		close(sync_fd);
		sync_fd = -1;
	}

	systemd_notify("READY=1");

	ret = loop();

	usdt_parser_tidy(0);
	teardown_device();

	if (ret == 0)
		exit(0);
	else
		exit(2);			/* Allow restarting.  */
}