4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2011, 2016 by Delphix. All rights reserved.
24 * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
25 * Copyright (c) 2012 Martin Matuska <mm@FreeBSD.org>. All rights reserved.
26 * Copyright (c) 2013 Steven Hartland. All rights reserved.
27 * Copyright (c) 2014 Integros [integros.com]
31 * The objective of this program is to provide a DMU/ZAP/SPA stress test
32 * that runs entirely in userland, is easy to use, and easy to extend.
34 * The overall design of the ztest program is as follows:
36 * (1) For each major functional area (e.g. adding vdevs to a pool,
37 * creating and destroying datasets, reading and writing objects, etc)
38 * we have a simple routine to test that functionality. These
39 * individual routines do not have to do anything "stressful".
41 * (2) We turn these simple functionality tests into a stress test by
42 * running them all in parallel, with as many threads as desired,
43 * and spread across as many datasets, objects, and vdevs as desired.
45 * (3) While all this is happening, we inject faults into the pool to
46 * verify that self-healing data really works.
48 * (4) Every time we open a dataset, we change its checksum and compression
49 * functions. Thus even individual objects vary from block to block
50 * in which checksum they use and whether they're compressed.
52 * (5) To verify that we never lose on-disk consistency after a crash,
53 * we run the entire test in a child of the main process.
54 * At random times, the child self-immolates with a SIGKILL.
55 * This is the software equivalent of pulling the power cord.
56 * The parent then runs the test again, using the existing
57 * storage pool, as many times as desired. If backwards compatibility
58 * testing is enabled ztest will sometimes run the "older" version
59 * of ztest after a SIGKILL.
61 * (6) To verify that we don't have future leaks or temporal incursions,
62 * many of the functional tests record the transaction group number
63 * as part of their data. When reading old data, they verify that
64 * the transaction group number is less than the current, open txg.
65 * If you add a new test, please do this if applicable.
67 * When run with no arguments, ztest runs for about five minutes and
68 * produces no output if successful. To get a little bit of information,
69 * specify -V. To get more information, specify -VV, and so on.
71 * To turn this into an overnight stress test, use -T to specify run time.
73 * You can ask more more vdevs [-v], datasets [-d], or threads [-t]
74 * to increase the pool capacity, fanout, and overall stress level.
76 * Use the -k option to set the desired frequency of kills.
78 * When ztest invokes itself it passes all relevant information through a
79 * temporary file which is mmap-ed in the child process. This allows shared
80 * memory to survive the exec syscall. The ztest_shared_hdr_t struct is always
81 * stored at offset 0 of this file and contains information on the size and
82 * number of shared structures in the file. The information stored in this file
83 * must remain backwards compatible with older versions of ztest so that
84 * ztest can invoke them during backwards compatibility testing (-B).
87 #include <sys/zfs_context.h>
93 #include <sys/dmu_objset.h>
99 #include <sys/resource.h>
102 #include <sys/zil_impl.h>
103 #include <sys/vdev_impl.h>
104 #include <sys/vdev_file.h>
105 #include <sys/spa_impl.h>
106 #include <sys/metaslab_impl.h>
107 #include <sys/dsl_prop.h>
108 #include <sys/dsl_dataset.h>
109 #include <sys/dsl_destroy.h>
110 #include <sys/dsl_scan.h>
111 #include <sys/zio_checksum.h>
112 #include <sys/refcount.h>
113 #include <sys/zfeature.h>
114 #include <sys/dsl_userhold.h>
116 #include <stdio_ext.h>
125 #include <sys/fs/zfs.h>
126 #include <libnvpair.h>
128 static int ztest_fd_data = -1;
129 static int ztest_fd_rand = -1;
131 typedef struct ztest_shared_hdr {
132 uint64_t zh_hdr_size;
133 uint64_t zh_opts_size;
135 uint64_t zh_stats_size;
136 uint64_t zh_stats_count;
138 uint64_t zh_ds_count;
139 } ztest_shared_hdr_t;
141 static ztest_shared_hdr_t *ztest_shared_hdr;
143 typedef struct ztest_shared_opts {
144 char zo_pool[ZFS_MAX_DATASET_NAME_LEN];
145 char zo_dir[ZFS_MAX_DATASET_NAME_LEN];
146 char zo_alt_ztest[MAXNAMELEN];
147 char zo_alt_libpath[MAXNAMELEN];
149 uint64_t zo_vdevtime;
157 uint64_t zo_passtime;
158 uint64_t zo_killrate;
162 uint64_t zo_maxloops;
163 uint64_t zo_metaslab_gang_bang;
164 } ztest_shared_opts_t;
166 static const ztest_shared_opts_t ztest_opts_defaults = {
167 .zo_pool = { 'z', 't', 'e', 's', 't', '\0' },
168 .zo_dir = { '/', 't', 'm', 'p', '\0' },
169 .zo_alt_ztest = { '\0' },
170 .zo_alt_libpath = { '\0' },
172 .zo_ashift = SPA_MINBLOCKSHIFT,
175 .zo_raidz_parity = 1,
176 .zo_vdev_size = SPA_MINDEVSIZE * 2,
179 .zo_passtime = 60, /* 60 seconds */
180 .zo_killrate = 70, /* 70% kill rate */
183 .zo_time = 300, /* 5 minutes */
184 .zo_maxloops = 50, /* max loops during spa_freeze() */
185 .zo_metaslab_gang_bang = 32 << 10
188 extern uint64_t metaslab_gang_bang;
189 extern uint64_t metaslab_df_alloc_threshold;
190 extern uint64_t zfs_deadman_synctime_ms;
191 extern int metaslab_preload_limit;
192 extern boolean_t zfs_compressed_arc_enabled;
194 static ztest_shared_opts_t *ztest_shared_opts;
195 static ztest_shared_opts_t ztest_opts;
197 typedef struct ztest_shared_ds {
201 static ztest_shared_ds_t *ztest_shared_ds;
202 #define ZTEST_GET_SHARED_DS(d) (&ztest_shared_ds[d])
204 #define BT_MAGIC 0x123456789abcdefULL
205 #define MAXFAULTS() \
206 (MAX(zs->zs_mirrors, 1) * (ztest_opts.zo_raidz_parity + 1) - 1)
210 ZTEST_IO_WRITE_PATTERN,
211 ZTEST_IO_WRITE_ZEROES,
218 typedef struct ztest_block_tag {
228 typedef struct bufwad {
235 * XXX -- fix zfs range locks to be generic so we can use them here.
257 #define ZTEST_RANGE_LOCKS 64
258 #define ZTEST_OBJECT_LOCKS 64
261 * Object descriptor. Used as a template for object lookup/create/remove.
263 typedef struct ztest_od {
266 dmu_object_type_t od_type;
267 dmu_object_type_t od_crtype;
268 uint64_t od_blocksize;
269 uint64_t od_crblocksize;
272 char od_name[ZFS_MAX_DATASET_NAME_LEN];
278 typedef struct ztest_ds {
279 ztest_shared_ds_t *zd_shared;
281 rwlock_t zd_zilog_lock;
283 ztest_od_t *zd_od; /* debugging aid */
284 char zd_name[ZFS_MAX_DATASET_NAME_LEN];
285 mutex_t zd_dirobj_lock;
286 rll_t zd_object_lock[ZTEST_OBJECT_LOCKS];
287 rll_t zd_range_lock[ZTEST_RANGE_LOCKS];
291 * Per-iteration state.
293 typedef void ztest_func_t(ztest_ds_t *zd, uint64_t id);
295 typedef struct ztest_info {
296 ztest_func_t *zi_func; /* test function */
297 uint64_t zi_iters; /* iterations per execution */
298 uint64_t *zi_interval; /* execute every <interval> seconds */
301 typedef struct ztest_shared_callstate {
302 uint64_t zc_count; /* per-pass count */
303 uint64_t zc_time; /* per-pass time */
304 uint64_t zc_next; /* next time to call this function */
305 } ztest_shared_callstate_t;
307 static ztest_shared_callstate_t *ztest_shared_callstate;
308 #define ZTEST_GET_SHARED_CALLSTATE(c) (&ztest_shared_callstate[c])
311 * Note: these aren't static because we want dladdr() to work.
313 ztest_func_t ztest_dmu_read_write;
314 ztest_func_t ztest_dmu_write_parallel;
315 ztest_func_t ztest_dmu_object_alloc_free;
316 ztest_func_t ztest_dmu_commit_callbacks;
317 ztest_func_t ztest_zap;
318 ztest_func_t ztest_zap_parallel;
319 ztest_func_t ztest_zil_commit;
320 ztest_func_t ztest_zil_remount;
321 ztest_func_t ztest_dmu_read_write_zcopy;
322 ztest_func_t ztest_dmu_objset_create_destroy;
323 ztest_func_t ztest_dmu_prealloc;
324 ztest_func_t ztest_fzap;
325 ztest_func_t ztest_dmu_snapshot_create_destroy;
326 ztest_func_t ztest_dsl_prop_get_set;
327 ztest_func_t ztest_spa_prop_get_set;
328 ztest_func_t ztest_spa_create_destroy;
329 ztest_func_t ztest_fault_inject;
330 ztest_func_t ztest_ddt_repair;
331 ztest_func_t ztest_dmu_snapshot_hold;
332 ztest_func_t ztest_spa_rename;
333 ztest_func_t ztest_scrub;
334 ztest_func_t ztest_dsl_dataset_promote_busy;
335 ztest_func_t ztest_vdev_attach_detach;
336 ztest_func_t ztest_vdev_LUN_growth;
337 ztest_func_t ztest_vdev_add_remove;
338 ztest_func_t ztest_vdev_aux_add_remove;
339 ztest_func_t ztest_split_pool;
340 ztest_func_t ztest_reguid;
341 ztest_func_t ztest_spa_upgrade;
343 uint64_t zopt_always = 0ULL * NANOSEC; /* all the time */
344 uint64_t zopt_incessant = 1ULL * NANOSEC / 10; /* every 1/10 second */
345 uint64_t zopt_often = 1ULL * NANOSEC; /* every second */
346 uint64_t zopt_sometimes = 10ULL * NANOSEC; /* every 10 seconds */
347 uint64_t zopt_rarely = 60ULL * NANOSEC; /* every 60 seconds */
349 ztest_info_t ztest_info[] = {
350 { ztest_dmu_read_write, 1, &zopt_always },
351 { ztest_dmu_write_parallel, 10, &zopt_always },
352 { ztest_dmu_object_alloc_free, 1, &zopt_always },
353 { ztest_dmu_commit_callbacks, 1, &zopt_always },
354 { ztest_zap, 30, &zopt_always },
355 { ztest_zap_parallel, 100, &zopt_always },
356 { ztest_split_pool, 1, &zopt_always },
357 { ztest_zil_commit, 1, &zopt_incessant },
358 { ztest_zil_remount, 1, &zopt_sometimes },
359 { ztest_dmu_read_write_zcopy, 1, &zopt_often },
360 { ztest_dmu_objset_create_destroy, 1, &zopt_often },
361 { ztest_dsl_prop_get_set, 1, &zopt_often },
362 { ztest_spa_prop_get_set, 1, &zopt_sometimes },
364 { ztest_dmu_prealloc, 1, &zopt_sometimes },
366 { ztest_fzap, 1, &zopt_sometimes },
367 { ztest_dmu_snapshot_create_destroy, 1, &zopt_sometimes },
368 { ztest_spa_create_destroy, 1, &zopt_sometimes },
369 { ztest_fault_inject, 1, &zopt_sometimes },
370 { ztest_ddt_repair, 1, &zopt_sometimes },
371 { ztest_dmu_snapshot_hold, 1, &zopt_sometimes },
372 { ztest_reguid, 1, &zopt_rarely },
373 { ztest_spa_rename, 1, &zopt_rarely },
374 { ztest_scrub, 1, &zopt_rarely },
375 { ztest_spa_upgrade, 1, &zopt_rarely },
376 { ztest_dsl_dataset_promote_busy, 1, &zopt_rarely },
377 { ztest_vdev_attach_detach, 1, &zopt_sometimes },
378 { ztest_vdev_LUN_growth, 1, &zopt_rarely },
379 { ztest_vdev_add_remove, 1,
380 &ztest_opts.zo_vdevtime },
381 { ztest_vdev_aux_add_remove, 1,
382 &ztest_opts.zo_vdevtime },
385 #define ZTEST_FUNCS (sizeof (ztest_info) / sizeof (ztest_info_t))
388 * The following struct is used to hold a list of uncalled commit callbacks.
389 * The callbacks are ordered by txg number.
391 typedef struct ztest_cb_list {
392 mutex_t zcl_callbacks_lock;
393 list_t zcl_callbacks;
397 * Stuff we need to share writably between parent and child.
399 typedef struct ztest_shared {
400 boolean_t zs_do_init;
401 hrtime_t zs_proc_start;
402 hrtime_t zs_proc_stop;
403 hrtime_t zs_thread_start;
404 hrtime_t zs_thread_stop;
405 hrtime_t zs_thread_kill;
406 uint64_t zs_enospc_count;
407 uint64_t zs_vdev_next_leaf;
408 uint64_t zs_vdev_aux;
413 uint64_t zs_metaslab_sz;
414 uint64_t zs_metaslab_df_alloc_threshold;
418 #define ID_PARALLEL -1ULL
420 static char ztest_dev_template[] = "%s/%s.%llua";
421 static char ztest_aux_template[] = "%s/%s.%s.%llu";
422 ztest_shared_t *ztest_shared;
424 static spa_t *ztest_spa = NULL;
425 static ztest_ds_t *ztest_ds;
427 static mutex_t ztest_vdev_lock;
430 * The ztest_name_lock protects the pool and dataset namespace used by
431 * the individual tests. To modify the namespace, consumers must grab
432 * this lock as writer. Grabbing the lock as reader will ensure that the
433 * namespace does not change while the lock is held.
435 static rwlock_t ztest_name_lock;
437 static boolean_t ztest_dump_core = B_TRUE;
438 static boolean_t ztest_exiting;
440 /* Global commit callback list */
441 static ztest_cb_list_t zcl;
444 ZTEST_META_DNODE = 0,
449 static void usage(boolean_t) __NORETURN;
452 * These libumem hooks provide a reasonable set of defaults for the allocator's
453 * debugging facilities.
458 return ("default,verbose"); /* $UMEM_DEBUG setting */
462 _umem_logging_init(void)
464 return ("fail,contents"); /* $UMEM_LOGGING setting */
467 #define FATAL_MSG_SZ 1024
472 fatal(int do_perror, char *message, ...)
475 int save_errno = errno;
476 char buf[FATAL_MSG_SZ];
478 (void) fflush(stdout);
480 va_start(args, message);
481 (void) sprintf(buf, "ztest: ");
483 (void) vsprintf(buf + strlen(buf), message, args);
486 (void) snprintf(buf + strlen(buf), FATAL_MSG_SZ - strlen(buf),
487 ": %s", strerror(save_errno));
489 (void) fprintf(stderr, "%s\n", buf);
490 fatal_msg = buf; /* to ease debugging */
497 str2shift(const char *buf)
499 const char *ends = "BKMGTPEZ";
504 for (i = 0; i < strlen(ends); i++) {
505 if (toupper(buf[0]) == ends[i])
508 if (i == strlen(ends)) {
509 (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n",
513 if (buf[1] == '\0' || (toupper(buf[1]) == 'B' && buf[2] == '\0')) {
516 (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n", buf);
522 nicenumtoull(const char *buf)
527 val = strtoull(buf, &end, 0);
529 (void) fprintf(stderr, "ztest: bad numeric value: %s\n", buf);
531 } else if (end[0] == '.') {
532 double fval = strtod(buf, &end);
533 fval *= pow(2, str2shift(end));
534 if (fval > UINT64_MAX) {
535 (void) fprintf(stderr, "ztest: value too large: %s\n",
539 val = (uint64_t)fval;
541 int shift = str2shift(end);
542 if (shift >= 64 || (val << shift) >> shift != val) {
543 (void) fprintf(stderr, "ztest: value too large: %s\n",
553 usage(boolean_t requested)
555 const ztest_shared_opts_t *zo = &ztest_opts_defaults;
557 char nice_vdev_size[10];
558 char nice_gang_bang[10];
559 FILE *fp = requested ? stdout : stderr;
561 nicenum(zo->zo_vdev_size, nice_vdev_size);
562 nicenum(zo->zo_metaslab_gang_bang, nice_gang_bang);
564 (void) fprintf(fp, "Usage: %s\n"
565 "\t[-v vdevs (default: %llu)]\n"
566 "\t[-s size_of_each_vdev (default: %s)]\n"
567 "\t[-a alignment_shift (default: %d)] use 0 for random\n"
568 "\t[-m mirror_copies (default: %d)]\n"
569 "\t[-r raidz_disks (default: %d)]\n"
570 "\t[-R raidz_parity (default: %d)]\n"
571 "\t[-d datasets (default: %d)]\n"
572 "\t[-t threads (default: %d)]\n"
573 "\t[-g gang_block_threshold (default: %s)]\n"
574 "\t[-i init_count (default: %d)] initialize pool i times\n"
575 "\t[-k kill_percentage (default: %llu%%)]\n"
576 "\t[-p pool_name (default: %s)]\n"
577 "\t[-f dir (default: %s)] file directory for vdev files\n"
578 "\t[-V] verbose (use multiple times for ever more blather)\n"
579 "\t[-E] use existing pool instead of creating new one\n"
580 "\t[-T time (default: %llu sec)] total run time\n"
581 "\t[-F freezeloops (default: %llu)] max loops in spa_freeze()\n"
582 "\t[-P passtime (default: %llu sec)] time per pass\n"
583 "\t[-B alt_ztest (default: <none>)] alternate ztest path\n"
584 "\t[-h] (print help)\n"
587 (u_longlong_t)zo->zo_vdevs, /* -v */
588 nice_vdev_size, /* -s */
589 zo->zo_ashift, /* -a */
590 zo->zo_mirrors, /* -m */
591 zo->zo_raidz, /* -r */
592 zo->zo_raidz_parity, /* -R */
593 zo->zo_datasets, /* -d */
594 zo->zo_threads, /* -t */
595 nice_gang_bang, /* -g */
596 zo->zo_init, /* -i */
597 (u_longlong_t)zo->zo_killrate, /* -k */
598 zo->zo_pool, /* -p */
600 (u_longlong_t)zo->zo_time, /* -T */
601 (u_longlong_t)zo->zo_maxloops, /* -F */
602 (u_longlong_t)zo->zo_passtime);
603 exit(requested ? 0 : 1);
607 process_options(int argc, char **argv)
610 ztest_shared_opts_t *zo = &ztest_opts;
614 char altdir[MAXNAMELEN] = { 0 };
616 bcopy(&ztest_opts_defaults, zo, sizeof (*zo));
618 while ((opt = getopt(argc, argv,
619 "v:s:a:m:r:R:d:t:g:i:k:p:f:VET:P:hF:B:")) != EOF) {
636 value = nicenumtoull(optarg);
640 zo->zo_vdevs = value;
643 zo->zo_vdev_size = MAX(SPA_MINDEVSIZE, value);
646 zo->zo_ashift = value;
649 zo->zo_mirrors = value;
652 zo->zo_raidz = MAX(1, value);
655 zo->zo_raidz_parity = MIN(MAX(value, 1), 3);
658 zo->zo_datasets = MAX(1, value);
661 zo->zo_threads = MAX(1, value);
664 zo->zo_metaslab_gang_bang = MAX(SPA_MINBLOCKSIZE << 1,
671 zo->zo_killrate = value;
674 (void) strlcpy(zo->zo_pool, optarg,
675 sizeof (zo->zo_pool));
678 path = realpath(optarg, NULL);
680 (void) fprintf(stderr, "error: %s: %s\n",
681 optarg, strerror(errno));
684 (void) strlcpy(zo->zo_dir, path,
685 sizeof (zo->zo_dir));
698 zo->zo_passtime = MAX(1, value);
701 zo->zo_maxloops = MAX(1, value);
704 (void) strlcpy(altdir, optarg, sizeof (altdir));
716 zo->zo_raidz_parity = MIN(zo->zo_raidz_parity, zo->zo_raidz - 1);
719 (zo->zo_vdevs > 0 ? zo->zo_time * NANOSEC / zo->zo_vdevs :
722 if (strlen(altdir) > 0) {
730 cmd = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
731 realaltdir = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
733 VERIFY(NULL != realpath(getexecname(), cmd));
734 if (0 != access(altdir, F_OK)) {
735 ztest_dump_core = B_FALSE;
736 fatal(B_TRUE, "invalid alternate ztest path: %s",
739 VERIFY(NULL != realpath(altdir, realaltdir));
742 * 'cmd' should be of the form "<anything>/usr/bin/<isa>/ztest".
743 * We want to extract <isa> to determine if we should use
744 * 32 or 64 bit binaries.
746 bin = strstr(cmd, "/usr/bin/");
747 ztest = strstr(bin, "/ztest");
749 isalen = ztest - isa;
750 (void) snprintf(zo->zo_alt_ztest, sizeof (zo->zo_alt_ztest),
751 "%s/usr/bin/%.*s/ztest", realaltdir, isalen, isa);
752 (void) snprintf(zo->zo_alt_libpath, sizeof (zo->zo_alt_libpath),
753 "%s/usr/lib/%.*s", realaltdir, isalen, isa);
755 if (0 != access(zo->zo_alt_ztest, X_OK)) {
756 ztest_dump_core = B_FALSE;
757 fatal(B_TRUE, "invalid alternate ztest: %s",
759 } else if (0 != access(zo->zo_alt_libpath, X_OK)) {
760 ztest_dump_core = B_FALSE;
761 fatal(B_TRUE, "invalid alternate lib directory %s",
765 umem_free(cmd, MAXPATHLEN);
766 umem_free(realaltdir, MAXPATHLEN);
771 ztest_kill(ztest_shared_t *zs)
773 zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(ztest_spa));
774 zs->zs_space = metaslab_class_get_space(spa_normal_class(ztest_spa));
777 * Before we kill off ztest, make sure that the config is updated.
778 * See comment above spa_config_sync().
780 mutex_enter(&spa_namespace_lock);
781 spa_config_sync(ztest_spa, B_FALSE, B_FALSE);
782 mutex_exit(&spa_namespace_lock);
784 zfs_dbgmsg_print(FTAG);
785 (void) kill(getpid(), SIGKILL);
789 ztest_random(uint64_t range)
793 ASSERT3S(ztest_fd_rand, >=, 0);
798 if (read(ztest_fd_rand, &r, sizeof (r)) != sizeof (r))
799 fatal(1, "short read from /dev/urandom");
806 ztest_record_enospc(const char *s)
808 ztest_shared->zs_enospc_count++;
812 ztest_get_ashift(void)
814 if (ztest_opts.zo_ashift == 0)
815 return (SPA_MINBLOCKSHIFT + ztest_random(5));
816 return (ztest_opts.zo_ashift);
820 make_vdev_file(char *path, char *aux, char *pool, size_t size, uint64_t ashift)
822 char pathbuf[MAXPATHLEN];
827 ashift = ztest_get_ashift();
833 vdev = ztest_shared->zs_vdev_aux;
834 (void) snprintf(path, sizeof (pathbuf),
835 ztest_aux_template, ztest_opts.zo_dir,
836 pool == NULL ? ztest_opts.zo_pool : pool,
839 vdev = ztest_shared->zs_vdev_next_leaf++;
840 (void) snprintf(path, sizeof (pathbuf),
841 ztest_dev_template, ztest_opts.zo_dir,
842 pool == NULL ? ztest_opts.zo_pool : pool, vdev);
847 int fd = open(path, O_RDWR | O_CREAT | O_TRUNC, 0666);
849 fatal(1, "can't open %s", path);
850 if (ftruncate(fd, size) != 0)
851 fatal(1, "can't ftruncate %s", path);
855 VERIFY(nvlist_alloc(&file, NV_UNIQUE_NAME, 0) == 0);
856 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_TYPE, VDEV_TYPE_FILE) == 0);
857 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_PATH, path) == 0);
858 VERIFY(nvlist_add_uint64(file, ZPOOL_CONFIG_ASHIFT, ashift) == 0);
864 make_vdev_raidz(char *path, char *aux, char *pool, size_t size,
865 uint64_t ashift, int r)
867 nvlist_t *raidz, **child;
871 return (make_vdev_file(path, aux, pool, size, ashift));
872 child = umem_alloc(r * sizeof (nvlist_t *), UMEM_NOFAIL);
874 for (c = 0; c < r; c++)
875 child[c] = make_vdev_file(path, aux, pool, size, ashift);
877 VERIFY(nvlist_alloc(&raidz, NV_UNIQUE_NAME, 0) == 0);
878 VERIFY(nvlist_add_string(raidz, ZPOOL_CONFIG_TYPE,
879 VDEV_TYPE_RAIDZ) == 0);
880 VERIFY(nvlist_add_uint64(raidz, ZPOOL_CONFIG_NPARITY,
881 ztest_opts.zo_raidz_parity) == 0);
882 VERIFY(nvlist_add_nvlist_array(raidz, ZPOOL_CONFIG_CHILDREN,
885 for (c = 0; c < r; c++)
886 nvlist_free(child[c]);
888 umem_free(child, r * sizeof (nvlist_t *));
894 make_vdev_mirror(char *path, char *aux, char *pool, size_t size,
895 uint64_t ashift, int r, int m)
897 nvlist_t *mirror, **child;
901 return (make_vdev_raidz(path, aux, pool, size, ashift, r));
903 child = umem_alloc(m * sizeof (nvlist_t *), UMEM_NOFAIL);
905 for (c = 0; c < m; c++)
906 child[c] = make_vdev_raidz(path, aux, pool, size, ashift, r);
908 VERIFY(nvlist_alloc(&mirror, NV_UNIQUE_NAME, 0) == 0);
909 VERIFY(nvlist_add_string(mirror, ZPOOL_CONFIG_TYPE,
910 VDEV_TYPE_MIRROR) == 0);
911 VERIFY(nvlist_add_nvlist_array(mirror, ZPOOL_CONFIG_CHILDREN,
914 for (c = 0; c < m; c++)
915 nvlist_free(child[c]);
917 umem_free(child, m * sizeof (nvlist_t *));
923 make_vdev_root(char *path, char *aux, char *pool, size_t size, uint64_t ashift,
924 int log, int r, int m, int t)
926 nvlist_t *root, **child;
931 child = umem_alloc(t * sizeof (nvlist_t *), UMEM_NOFAIL);
933 for (c = 0; c < t; c++) {
934 child[c] = make_vdev_mirror(path, aux, pool, size, ashift,
936 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_IS_LOG,
940 VERIFY(nvlist_alloc(&root, NV_UNIQUE_NAME, 0) == 0);
941 VERIFY(nvlist_add_string(root, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) == 0);
942 VERIFY(nvlist_add_nvlist_array(root, aux ? aux : ZPOOL_CONFIG_CHILDREN,
945 for (c = 0; c < t; c++)
946 nvlist_free(child[c]);
948 umem_free(child, t * sizeof (nvlist_t *));
954 * Find a random spa version. Returns back a random spa version in the
955 * range [initial_version, SPA_VERSION_FEATURES].
958 ztest_random_spa_version(uint64_t initial_version)
960 uint64_t version = initial_version;
962 if (version <= SPA_VERSION_BEFORE_FEATURES) {
964 ztest_random(SPA_VERSION_BEFORE_FEATURES - version + 1);
967 if (version > SPA_VERSION_BEFORE_FEATURES)
968 version = SPA_VERSION_FEATURES;
970 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
975 ztest_random_blocksize(void)
977 uint64_t block_shift;
979 * Choose a block size >= the ashift.
980 * If the SPA supports new MAXBLOCKSIZE, test up to 1MB blocks.
982 int maxbs = SPA_OLD_MAXBLOCKSHIFT;
983 if (spa_maxblocksize(ztest_spa) == SPA_MAXBLOCKSIZE)
985 block_shift = ztest_random(maxbs - ztest_spa->spa_max_ashift + 1);
986 return (1 << (SPA_MINBLOCKSHIFT + block_shift));
990 ztest_random_ibshift(void)
992 return (DN_MIN_INDBLKSHIFT +
993 ztest_random(DN_MAX_INDBLKSHIFT - DN_MIN_INDBLKSHIFT + 1));
997 ztest_random_vdev_top(spa_t *spa, boolean_t log_ok)
1000 vdev_t *rvd = spa->spa_root_vdev;
1003 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1006 top = ztest_random(rvd->vdev_children);
1007 tvd = rvd->vdev_child[top];
1008 } while (tvd->vdev_ishole || (tvd->vdev_islog && !log_ok) ||
1009 tvd->vdev_mg == NULL || tvd->vdev_mg->mg_class == NULL);
1015 ztest_random_dsl_prop(zfs_prop_t prop)
1020 value = zfs_prop_random_value(prop, ztest_random(-1ULL));
1021 } while (prop == ZFS_PROP_CHECKSUM && value == ZIO_CHECKSUM_OFF);
1027 ztest_dsl_prop_set_uint64(char *osname, zfs_prop_t prop, uint64_t value,
1030 const char *propname = zfs_prop_to_name(prop);
1031 const char *valname;
1032 char setpoint[MAXPATHLEN];
1036 error = dsl_prop_set_int(osname, propname,
1037 (inherit ? ZPROP_SRC_NONE : ZPROP_SRC_LOCAL), value);
1039 if (error == ENOSPC) {
1040 ztest_record_enospc(FTAG);
1045 VERIFY0(dsl_prop_get_integer(osname, propname, &curval, setpoint));
1047 if (ztest_opts.zo_verbose >= 6) {
1048 VERIFY(zfs_prop_index_to_string(prop, curval, &valname) == 0);
1049 (void) printf("%s %s = %s at '%s'\n",
1050 osname, propname, valname, setpoint);
1057 ztest_spa_prop_set_uint64(zpool_prop_t prop, uint64_t value)
1059 spa_t *spa = ztest_spa;
1060 nvlist_t *props = NULL;
1063 VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
1064 VERIFY(nvlist_add_uint64(props, zpool_prop_to_name(prop), value) == 0);
1066 error = spa_prop_set(spa, props);
1070 if (error == ENOSPC) {
1071 ztest_record_enospc(FTAG);
1080 ztest_rll_init(rll_t *rll)
1082 rll->rll_writer = NULL;
1083 rll->rll_readers = 0;
1084 VERIFY(_mutex_init(&rll->rll_lock, USYNC_THREAD, NULL) == 0);
1085 VERIFY(cond_init(&rll->rll_cv, USYNC_THREAD, NULL) == 0);
1089 ztest_rll_destroy(rll_t *rll)
1091 ASSERT(rll->rll_writer == NULL);
1092 ASSERT(rll->rll_readers == 0);
1093 VERIFY(_mutex_destroy(&rll->rll_lock) == 0);
1094 VERIFY(cond_destroy(&rll->rll_cv) == 0);
1098 ztest_rll_lock(rll_t *rll, rl_type_t type)
1100 VERIFY(mutex_lock(&rll->rll_lock) == 0);
1102 if (type == RL_READER) {
1103 while (rll->rll_writer != NULL)
1104 (void) cond_wait(&rll->rll_cv, &rll->rll_lock);
1107 while (rll->rll_writer != NULL || rll->rll_readers)
1108 (void) cond_wait(&rll->rll_cv, &rll->rll_lock);
1109 rll->rll_writer = curthread;
1112 VERIFY(mutex_unlock(&rll->rll_lock) == 0);
1116 ztest_rll_unlock(rll_t *rll)
1118 VERIFY(mutex_lock(&rll->rll_lock) == 0);
1120 if (rll->rll_writer) {
1121 ASSERT(rll->rll_readers == 0);
1122 rll->rll_writer = NULL;
1124 ASSERT(rll->rll_readers != 0);
1125 ASSERT(rll->rll_writer == NULL);
1129 if (rll->rll_writer == NULL && rll->rll_readers == 0)
1130 VERIFY(cond_broadcast(&rll->rll_cv) == 0);
1132 VERIFY(mutex_unlock(&rll->rll_lock) == 0);
1136 ztest_object_lock(ztest_ds_t *zd, uint64_t object, rl_type_t type)
1138 rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1140 ztest_rll_lock(rll, type);
1144 ztest_object_unlock(ztest_ds_t *zd, uint64_t object)
1146 rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1148 ztest_rll_unlock(rll);
1152 ztest_range_lock(ztest_ds_t *zd, uint64_t object, uint64_t offset,
1153 uint64_t size, rl_type_t type)
1155 uint64_t hash = object ^ (offset % (ZTEST_RANGE_LOCKS + 1));
1156 rll_t *rll = &zd->zd_range_lock[hash & (ZTEST_RANGE_LOCKS - 1)];
1159 rl = umem_alloc(sizeof (*rl), UMEM_NOFAIL);
1160 rl->rl_object = object;
1161 rl->rl_offset = offset;
1165 ztest_rll_lock(rll, type);
1171 ztest_range_unlock(rl_t *rl)
1173 rll_t *rll = rl->rl_lock;
1175 ztest_rll_unlock(rll);
1177 umem_free(rl, sizeof (*rl));
1181 ztest_zd_init(ztest_ds_t *zd, ztest_shared_ds_t *szd, objset_t *os)
1184 zd->zd_zilog = dmu_objset_zil(os);
1185 zd->zd_shared = szd;
1186 dmu_objset_name(os, zd->zd_name);
1188 if (zd->zd_shared != NULL)
1189 zd->zd_shared->zd_seq = 0;
1191 VERIFY(rwlock_init(&zd->zd_zilog_lock, USYNC_THREAD, NULL) == 0);
1192 VERIFY(_mutex_init(&zd->zd_dirobj_lock, USYNC_THREAD, NULL) == 0);
1194 for (int l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1195 ztest_rll_init(&zd->zd_object_lock[l]);
1197 for (int l = 0; l < ZTEST_RANGE_LOCKS; l++)
1198 ztest_rll_init(&zd->zd_range_lock[l]);
1202 ztest_zd_fini(ztest_ds_t *zd)
1204 VERIFY(_mutex_destroy(&zd->zd_dirobj_lock) == 0);
1206 for (int l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1207 ztest_rll_destroy(&zd->zd_object_lock[l]);
1209 for (int l = 0; l < ZTEST_RANGE_LOCKS; l++)
1210 ztest_rll_destroy(&zd->zd_range_lock[l]);
1213 #define TXG_MIGHTWAIT (ztest_random(10) == 0 ? TXG_NOWAIT : TXG_WAIT)
1216 ztest_tx_assign(dmu_tx_t *tx, uint64_t txg_how, const char *tag)
1222 * Attempt to assign tx to some transaction group.
1224 error = dmu_tx_assign(tx, txg_how);
1226 if (error == ERESTART) {
1227 ASSERT(txg_how == TXG_NOWAIT);
1230 ASSERT3U(error, ==, ENOSPC);
1231 ztest_record_enospc(tag);
1236 txg = dmu_tx_get_txg(tx);
1242 ztest_pattern_set(void *buf, uint64_t size, uint64_t value)
1245 uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1252 ztest_pattern_match(void *buf, uint64_t size, uint64_t value)
1255 uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1259 diff |= (value - *ip++);
1265 ztest_bt_generate(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1266 uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg)
1268 bt->bt_magic = BT_MAGIC;
1269 bt->bt_objset = dmu_objset_id(os);
1270 bt->bt_object = object;
1271 bt->bt_offset = offset;
1274 bt->bt_crtxg = crtxg;
1278 ztest_bt_verify(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1279 uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg)
1281 ASSERT3U(bt->bt_magic, ==, BT_MAGIC);
1282 ASSERT3U(bt->bt_objset, ==, dmu_objset_id(os));
1283 ASSERT3U(bt->bt_object, ==, object);
1284 ASSERT3U(bt->bt_offset, ==, offset);
1285 ASSERT3U(bt->bt_gen, <=, gen);
1286 ASSERT3U(bt->bt_txg, <=, txg);
1287 ASSERT3U(bt->bt_crtxg, ==, crtxg);
1290 static ztest_block_tag_t *
1291 ztest_bt_bonus(dmu_buf_t *db)
1293 dmu_object_info_t doi;
1294 ztest_block_tag_t *bt;
1296 dmu_object_info_from_db(db, &doi);
1297 ASSERT3U(doi.doi_bonus_size, <=, db->db_size);
1298 ASSERT3U(doi.doi_bonus_size, >=, sizeof (*bt));
1299 bt = (void *)((char *)db->db_data + doi.doi_bonus_size - sizeof (*bt));
1308 #define lrz_type lr_mode
1309 #define lrz_blocksize lr_uid
1310 #define lrz_ibshift lr_gid
1311 #define lrz_bonustype lr_rdev
1312 #define lrz_bonuslen lr_crtime[1]
1315 ztest_log_create(ztest_ds_t *zd, dmu_tx_t *tx, lr_create_t *lr)
1317 char *name = (void *)(lr + 1); /* name follows lr */
1318 size_t namesize = strlen(name) + 1;
1321 if (zil_replaying(zd->zd_zilog, tx))
1324 itx = zil_itx_create(TX_CREATE, sizeof (*lr) + namesize);
1325 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1326 sizeof (*lr) + namesize - sizeof (lr_t));
1328 zil_itx_assign(zd->zd_zilog, itx, tx);
1332 ztest_log_remove(ztest_ds_t *zd, dmu_tx_t *tx, lr_remove_t *lr, uint64_t object)
1334 char *name = (void *)(lr + 1); /* name follows lr */
1335 size_t namesize = strlen(name) + 1;
1338 if (zil_replaying(zd->zd_zilog, tx))
1341 itx = zil_itx_create(TX_REMOVE, sizeof (*lr) + namesize);
1342 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1343 sizeof (*lr) + namesize - sizeof (lr_t));
1345 itx->itx_oid = object;
1346 zil_itx_assign(zd->zd_zilog, itx, tx);
1350 ztest_log_write(ztest_ds_t *zd, dmu_tx_t *tx, lr_write_t *lr)
1353 itx_wr_state_t write_state = ztest_random(WR_NUM_STATES);
1355 if (zil_replaying(zd->zd_zilog, tx))
1358 if (lr->lr_length > ZIL_MAX_LOG_DATA)
1359 write_state = WR_INDIRECT;
1361 itx = zil_itx_create(TX_WRITE,
1362 sizeof (*lr) + (write_state == WR_COPIED ? lr->lr_length : 0));
1364 if (write_state == WR_COPIED &&
1365 dmu_read(zd->zd_os, lr->lr_foid, lr->lr_offset, lr->lr_length,
1366 ((lr_write_t *)&itx->itx_lr) + 1, DMU_READ_NO_PREFETCH) != 0) {
1367 zil_itx_destroy(itx);
1368 itx = zil_itx_create(TX_WRITE, sizeof (*lr));
1369 write_state = WR_NEED_COPY;
1371 itx->itx_private = zd;
1372 itx->itx_wr_state = write_state;
1373 itx->itx_sync = (ztest_random(8) == 0);
1374 itx->itx_sod += (write_state == WR_NEED_COPY ? lr->lr_length : 0);
1376 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1377 sizeof (*lr) - sizeof (lr_t));
1379 zil_itx_assign(zd->zd_zilog, itx, tx);
1383 ztest_log_truncate(ztest_ds_t *zd, dmu_tx_t *tx, lr_truncate_t *lr)
1387 if (zil_replaying(zd->zd_zilog, tx))
1390 itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr));
1391 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1392 sizeof (*lr) - sizeof (lr_t));
1394 itx->itx_sync = B_FALSE;
1395 zil_itx_assign(zd->zd_zilog, itx, tx);
1399 ztest_log_setattr(ztest_ds_t *zd, dmu_tx_t *tx, lr_setattr_t *lr)
1403 if (zil_replaying(zd->zd_zilog, tx))
1406 itx = zil_itx_create(TX_SETATTR, sizeof (*lr));
1407 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1408 sizeof (*lr) - sizeof (lr_t));
1410 itx->itx_sync = B_FALSE;
1411 zil_itx_assign(zd->zd_zilog, itx, tx);
1418 ztest_replay_create(ztest_ds_t *zd, lr_create_t *lr, boolean_t byteswap)
1420 char *name = (void *)(lr + 1); /* name follows lr */
1421 objset_t *os = zd->zd_os;
1422 ztest_block_tag_t *bbt;
1429 byteswap_uint64_array(lr, sizeof (*lr));
1431 ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1432 ASSERT(name[0] != '\0');
1434 tx = dmu_tx_create(os);
1436 dmu_tx_hold_zap(tx, lr->lr_doid, B_TRUE, name);
1438 if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1439 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, B_TRUE, NULL);
1441 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1444 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1448 ASSERT(dmu_objset_zil(os)->zl_replay == !!lr->lr_foid);
1450 if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1451 if (lr->lr_foid == 0) {
1452 lr->lr_foid = zap_create(os,
1453 lr->lrz_type, lr->lrz_bonustype,
1454 lr->lrz_bonuslen, tx);
1456 error = zap_create_claim(os, lr->lr_foid,
1457 lr->lrz_type, lr->lrz_bonustype,
1458 lr->lrz_bonuslen, tx);
1461 if (lr->lr_foid == 0) {
1462 lr->lr_foid = dmu_object_alloc(os,
1463 lr->lrz_type, 0, lr->lrz_bonustype,
1464 lr->lrz_bonuslen, tx);
1466 error = dmu_object_claim(os, lr->lr_foid,
1467 lr->lrz_type, 0, lr->lrz_bonustype,
1468 lr->lrz_bonuslen, tx);
1473 ASSERT3U(error, ==, EEXIST);
1474 ASSERT(zd->zd_zilog->zl_replay);
1479 ASSERT(lr->lr_foid != 0);
1481 if (lr->lrz_type != DMU_OT_ZAP_OTHER)
1482 VERIFY3U(0, ==, dmu_object_set_blocksize(os, lr->lr_foid,
1483 lr->lrz_blocksize, lr->lrz_ibshift, tx));
1485 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1486 bbt = ztest_bt_bonus(db);
1487 dmu_buf_will_dirty(db, tx);
1488 ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_gen, txg, txg);
1489 dmu_buf_rele(db, FTAG);
1491 VERIFY3U(0, ==, zap_add(os, lr->lr_doid, name, sizeof (uint64_t), 1,
1494 (void) ztest_log_create(zd, tx, lr);
1502 ztest_replay_remove(ztest_ds_t *zd, lr_remove_t *lr, boolean_t byteswap)
1504 char *name = (void *)(lr + 1); /* name follows lr */
1505 objset_t *os = zd->zd_os;
1506 dmu_object_info_t doi;
1508 uint64_t object, txg;
1511 byteswap_uint64_array(lr, sizeof (*lr));
1513 ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1514 ASSERT(name[0] != '\0');
1517 zap_lookup(os, lr->lr_doid, name, sizeof (object), 1, &object));
1518 ASSERT(object != 0);
1520 ztest_object_lock(zd, object, RL_WRITER);
1522 VERIFY3U(0, ==, dmu_object_info(os, object, &doi));
1524 tx = dmu_tx_create(os);
1526 dmu_tx_hold_zap(tx, lr->lr_doid, B_FALSE, name);
1527 dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END);
1529 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1531 ztest_object_unlock(zd, object);
1535 if (doi.doi_type == DMU_OT_ZAP_OTHER) {
1536 VERIFY3U(0, ==, zap_destroy(os, object, tx));
1538 VERIFY3U(0, ==, dmu_object_free(os, object, tx));
1541 VERIFY3U(0, ==, zap_remove(os, lr->lr_doid, name, tx));
1543 (void) ztest_log_remove(zd, tx, lr, object);
1547 ztest_object_unlock(zd, object);
1553 ztest_replay_write(ztest_ds_t *zd, lr_write_t *lr, boolean_t byteswap)
1555 objset_t *os = zd->zd_os;
1556 void *data = lr + 1; /* data follows lr */
1557 uint64_t offset, length;
1558 ztest_block_tag_t *bt = data;
1559 ztest_block_tag_t *bbt;
1560 uint64_t gen, txg, lrtxg, crtxg;
1561 dmu_object_info_t doi;
1564 arc_buf_t *abuf = NULL;
1568 byteswap_uint64_array(lr, sizeof (*lr));
1570 offset = lr->lr_offset;
1571 length = lr->lr_length;
1573 /* If it's a dmu_sync() block, write the whole block */
1574 if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
1575 uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
1576 if (length < blocksize) {
1577 offset -= offset % blocksize;
1582 if (bt->bt_magic == BSWAP_64(BT_MAGIC))
1583 byteswap_uint64_array(bt, sizeof (*bt));
1585 if (bt->bt_magic != BT_MAGIC)
1588 ztest_object_lock(zd, lr->lr_foid, RL_READER);
1589 rl = ztest_range_lock(zd, lr->lr_foid, offset, length, RL_WRITER);
1591 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1593 dmu_object_info_from_db(db, &doi);
1595 bbt = ztest_bt_bonus(db);
1596 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1598 crtxg = bbt->bt_crtxg;
1599 lrtxg = lr->lr_common.lrc_txg;
1601 tx = dmu_tx_create(os);
1603 dmu_tx_hold_write(tx, lr->lr_foid, offset, length);
1605 if (ztest_random(8) == 0 && length == doi.doi_data_block_size &&
1606 P2PHASE(offset, length) == 0)
1607 abuf = dmu_request_arcbuf(db, length);
1609 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1612 dmu_return_arcbuf(abuf);
1613 dmu_buf_rele(db, FTAG);
1614 ztest_range_unlock(rl);
1615 ztest_object_unlock(zd, lr->lr_foid);
1621 * Usually, verify the old data before writing new data --
1622 * but not always, because we also want to verify correct
1623 * behavior when the data was not recently read into cache.
1625 ASSERT(offset % doi.doi_data_block_size == 0);
1626 if (ztest_random(4) != 0) {
1627 int prefetch = ztest_random(2) ?
1628 DMU_READ_PREFETCH : DMU_READ_NO_PREFETCH;
1629 ztest_block_tag_t rbt;
1631 VERIFY(dmu_read(os, lr->lr_foid, offset,
1632 sizeof (rbt), &rbt, prefetch) == 0);
1633 if (rbt.bt_magic == BT_MAGIC) {
1634 ztest_bt_verify(&rbt, os, lr->lr_foid,
1635 offset, gen, txg, crtxg);
1640 * Writes can appear to be newer than the bonus buffer because
1641 * the ztest_get_data() callback does a dmu_read() of the
1642 * open-context data, which may be different than the data
1643 * as it was when the write was generated.
1645 if (zd->zd_zilog->zl_replay) {
1646 ztest_bt_verify(bt, os, lr->lr_foid, offset,
1647 MAX(gen, bt->bt_gen), MAX(txg, lrtxg),
1652 * Set the bt's gen/txg to the bonus buffer's gen/txg
1653 * so that all of the usual ASSERTs will work.
1655 ztest_bt_generate(bt, os, lr->lr_foid, offset, gen, txg, crtxg);
1659 dmu_write(os, lr->lr_foid, offset, length, data, tx);
1661 bcopy(data, abuf->b_data, length);
1662 dmu_assign_arcbuf(db, offset, abuf, tx);
1665 (void) ztest_log_write(zd, tx, lr);
1667 dmu_buf_rele(db, FTAG);
1671 ztest_range_unlock(rl);
1672 ztest_object_unlock(zd, lr->lr_foid);
1678 ztest_replay_truncate(ztest_ds_t *zd, lr_truncate_t *lr, boolean_t byteswap)
1680 objset_t *os = zd->zd_os;
1686 byteswap_uint64_array(lr, sizeof (*lr));
1688 ztest_object_lock(zd, lr->lr_foid, RL_READER);
1689 rl = ztest_range_lock(zd, lr->lr_foid, lr->lr_offset, lr->lr_length,
1692 tx = dmu_tx_create(os);
1694 dmu_tx_hold_free(tx, lr->lr_foid, lr->lr_offset, lr->lr_length);
1696 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1698 ztest_range_unlock(rl);
1699 ztest_object_unlock(zd, lr->lr_foid);
1703 VERIFY(dmu_free_range(os, lr->lr_foid, lr->lr_offset,
1704 lr->lr_length, tx) == 0);
1706 (void) ztest_log_truncate(zd, tx, lr);
1710 ztest_range_unlock(rl);
1711 ztest_object_unlock(zd, lr->lr_foid);
1717 ztest_replay_setattr(ztest_ds_t *zd, lr_setattr_t *lr, boolean_t byteswap)
1719 objset_t *os = zd->zd_os;
1722 ztest_block_tag_t *bbt;
1723 uint64_t txg, lrtxg, crtxg;
1726 byteswap_uint64_array(lr, sizeof (*lr));
1728 ztest_object_lock(zd, lr->lr_foid, RL_WRITER);
1730 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1732 tx = dmu_tx_create(os);
1733 dmu_tx_hold_bonus(tx, lr->lr_foid);
1735 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1737 dmu_buf_rele(db, FTAG);
1738 ztest_object_unlock(zd, lr->lr_foid);
1742 bbt = ztest_bt_bonus(db);
1743 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1744 crtxg = bbt->bt_crtxg;
1745 lrtxg = lr->lr_common.lrc_txg;
1747 if (zd->zd_zilog->zl_replay) {
1748 ASSERT(lr->lr_size != 0);
1749 ASSERT(lr->lr_mode != 0);
1753 * Randomly change the size and increment the generation.
1755 lr->lr_size = (ztest_random(db->db_size / sizeof (*bbt)) + 1) *
1757 lr->lr_mode = bbt->bt_gen + 1;
1762 * Verify that the current bonus buffer is not newer than our txg.
1764 ztest_bt_verify(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode,
1765 MAX(txg, lrtxg), crtxg);
1767 dmu_buf_will_dirty(db, tx);
1769 ASSERT3U(lr->lr_size, >=, sizeof (*bbt));
1770 ASSERT3U(lr->lr_size, <=, db->db_size);
1771 VERIFY0(dmu_set_bonus(db, lr->lr_size, tx));
1772 bbt = ztest_bt_bonus(db);
1774 ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode, txg, crtxg);
1776 dmu_buf_rele(db, FTAG);
1778 (void) ztest_log_setattr(zd, tx, lr);
1782 ztest_object_unlock(zd, lr->lr_foid);
1787 zil_replay_func_t *ztest_replay_vector[TX_MAX_TYPE] = {
1788 NULL, /* 0 no such transaction type */
1789 ztest_replay_create, /* TX_CREATE */
1790 NULL, /* TX_MKDIR */
1791 NULL, /* TX_MKXATTR */
1792 NULL, /* TX_SYMLINK */
1793 ztest_replay_remove, /* TX_REMOVE */
1794 NULL, /* TX_RMDIR */
1796 NULL, /* TX_RENAME */
1797 ztest_replay_write, /* TX_WRITE */
1798 ztest_replay_truncate, /* TX_TRUNCATE */
1799 ztest_replay_setattr, /* TX_SETATTR */
1801 NULL, /* TX_CREATE_ACL */
1802 NULL, /* TX_CREATE_ATTR */
1803 NULL, /* TX_CREATE_ACL_ATTR */
1804 NULL, /* TX_MKDIR_ACL */
1805 NULL, /* TX_MKDIR_ATTR */
1806 NULL, /* TX_MKDIR_ACL_ATTR */
1807 NULL, /* TX_WRITE2 */
1811 * ZIL get_data callbacks
1815 ztest_get_done(zgd_t *zgd, int error)
1817 ztest_ds_t *zd = zgd->zgd_private;
1818 uint64_t object = zgd->zgd_rl->rl_object;
1821 dmu_buf_rele(zgd->zgd_db, zgd);
1823 ztest_range_unlock(zgd->zgd_rl);
1824 ztest_object_unlock(zd, object);
1826 if (error == 0 && zgd->zgd_bp)
1827 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
1829 umem_free(zgd, sizeof (*zgd));
1833 ztest_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
1835 ztest_ds_t *zd = arg;
1836 objset_t *os = zd->zd_os;
1837 uint64_t object = lr->lr_foid;
1838 uint64_t offset = lr->lr_offset;
1839 uint64_t size = lr->lr_length;
1840 blkptr_t *bp = &lr->lr_blkptr;
1841 uint64_t txg = lr->lr_common.lrc_txg;
1843 dmu_object_info_t doi;
1848 ztest_object_lock(zd, object, RL_READER);
1849 error = dmu_bonus_hold(os, object, FTAG, &db);
1851 ztest_object_unlock(zd, object);
1855 crtxg = ztest_bt_bonus(db)->bt_crtxg;
1857 if (crtxg == 0 || crtxg > txg) {
1858 dmu_buf_rele(db, FTAG);
1859 ztest_object_unlock(zd, object);
1863 dmu_object_info_from_db(db, &doi);
1864 dmu_buf_rele(db, FTAG);
1867 zgd = umem_zalloc(sizeof (*zgd), UMEM_NOFAIL);
1868 zgd->zgd_zilog = zd->zd_zilog;
1869 zgd->zgd_private = zd;
1871 if (buf != NULL) { /* immediate write */
1872 zgd->zgd_rl = ztest_range_lock(zd, object, offset, size,
1875 error = dmu_read(os, object, offset, size, buf,
1876 DMU_READ_NO_PREFETCH);
1879 size = doi.doi_data_block_size;
1881 offset = P2ALIGN(offset, size);
1883 ASSERT(offset < size);
1887 zgd->zgd_rl = ztest_range_lock(zd, object, offset, size,
1890 error = dmu_buf_hold(os, object, offset, zgd, &db,
1891 DMU_READ_NO_PREFETCH);
1894 blkptr_t *obp = dmu_buf_get_blkptr(db);
1896 ASSERT(BP_IS_HOLE(bp));
1903 ASSERT(db->db_offset == offset);
1904 ASSERT(db->db_size == size);
1906 error = dmu_sync(zio, lr->lr_common.lrc_txg,
1907 ztest_get_done, zgd);
1914 ztest_get_done(zgd, error);
1920 ztest_lr_alloc(size_t lrsize, char *name)
1923 size_t namesize = name ? strlen(name) + 1 : 0;
1925 lr = umem_zalloc(lrsize + namesize, UMEM_NOFAIL);
1928 bcopy(name, lr + lrsize, namesize);
1934 ztest_lr_free(void *lr, size_t lrsize, char *name)
1936 size_t namesize = name ? strlen(name) + 1 : 0;
1938 umem_free(lr, lrsize + namesize);
1942 * Lookup a bunch of objects. Returns the number of objects not found.
1945 ztest_lookup(ztest_ds_t *zd, ztest_od_t *od, int count)
1950 ASSERT(_mutex_held(&zd->zd_dirobj_lock));
1952 for (int i = 0; i < count; i++, od++) {
1954 error = zap_lookup(zd->zd_os, od->od_dir, od->od_name,
1955 sizeof (uint64_t), 1, &od->od_object);
1957 ASSERT(error == ENOENT);
1958 ASSERT(od->od_object == 0);
1962 ztest_block_tag_t *bbt;
1963 dmu_object_info_t doi;
1965 ASSERT(od->od_object != 0);
1966 ASSERT(missing == 0); /* there should be no gaps */
1968 ztest_object_lock(zd, od->od_object, RL_READER);
1969 VERIFY3U(0, ==, dmu_bonus_hold(zd->zd_os,
1970 od->od_object, FTAG, &db));
1971 dmu_object_info_from_db(db, &doi);
1972 bbt = ztest_bt_bonus(db);
1973 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1974 od->od_type = doi.doi_type;
1975 od->od_blocksize = doi.doi_data_block_size;
1976 od->od_gen = bbt->bt_gen;
1977 dmu_buf_rele(db, FTAG);
1978 ztest_object_unlock(zd, od->od_object);
1986 ztest_create(ztest_ds_t *zd, ztest_od_t *od, int count)
1990 ASSERT(_mutex_held(&zd->zd_dirobj_lock));
1992 for (int i = 0; i < count; i++, od++) {
1999 lr_create_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
2001 lr->lr_doid = od->od_dir;
2002 lr->lr_foid = 0; /* 0 to allocate, > 0 to claim */
2003 lr->lrz_type = od->od_crtype;
2004 lr->lrz_blocksize = od->od_crblocksize;
2005 lr->lrz_ibshift = ztest_random_ibshift();
2006 lr->lrz_bonustype = DMU_OT_UINT64_OTHER;
2007 lr->lrz_bonuslen = dmu_bonus_max();
2008 lr->lr_gen = od->od_crgen;
2009 lr->lr_crtime[0] = time(NULL);
2011 if (ztest_replay_create(zd, lr, B_FALSE) != 0) {
2012 ASSERT(missing == 0);
2016 od->od_object = lr->lr_foid;
2017 od->od_type = od->od_crtype;
2018 od->od_blocksize = od->od_crblocksize;
2019 od->od_gen = od->od_crgen;
2020 ASSERT(od->od_object != 0);
2023 ztest_lr_free(lr, sizeof (*lr), od->od_name);
2030 ztest_remove(ztest_ds_t *zd, ztest_od_t *od, int count)
2035 ASSERT(_mutex_held(&zd->zd_dirobj_lock));
2039 for (int i = count - 1; i >= 0; i--, od--) {
2046 * No object was found.
2048 if (od->od_object == 0)
2051 lr_remove_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
2053 lr->lr_doid = od->od_dir;
2055 if ((error = ztest_replay_remove(zd, lr, B_FALSE)) != 0) {
2056 ASSERT3U(error, ==, ENOSPC);
2061 ztest_lr_free(lr, sizeof (*lr), od->od_name);
2068 ztest_write(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size,
2074 lr = ztest_lr_alloc(sizeof (*lr) + size, NULL);
2076 lr->lr_foid = object;
2077 lr->lr_offset = offset;
2078 lr->lr_length = size;
2080 BP_ZERO(&lr->lr_blkptr);
2082 bcopy(data, lr + 1, size);
2084 error = ztest_replay_write(zd, lr, B_FALSE);
2086 ztest_lr_free(lr, sizeof (*lr) + size, NULL);
2092 ztest_truncate(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2097 lr = ztest_lr_alloc(sizeof (*lr), NULL);
2099 lr->lr_foid = object;
2100 lr->lr_offset = offset;
2101 lr->lr_length = size;
2103 error = ztest_replay_truncate(zd, lr, B_FALSE);
2105 ztest_lr_free(lr, sizeof (*lr), NULL);
2111 ztest_setattr(ztest_ds_t *zd, uint64_t object)
2116 lr = ztest_lr_alloc(sizeof (*lr), NULL);
2118 lr->lr_foid = object;
2122 error = ztest_replay_setattr(zd, lr, B_FALSE);
2124 ztest_lr_free(lr, sizeof (*lr), NULL);
2130 ztest_prealloc(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2132 objset_t *os = zd->zd_os;
2137 txg_wait_synced(dmu_objset_pool(os), 0);
2139 ztest_object_lock(zd, object, RL_READER);
2140 rl = ztest_range_lock(zd, object, offset, size, RL_WRITER);
2142 tx = dmu_tx_create(os);
2144 dmu_tx_hold_write(tx, object, offset, size);
2146 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
2149 dmu_prealloc(os, object, offset, size, tx);
2151 txg_wait_synced(dmu_objset_pool(os), txg);
2153 (void) dmu_free_long_range(os, object, offset, size);
2156 ztest_range_unlock(rl);
2157 ztest_object_unlock(zd, object);
2161 ztest_io(ztest_ds_t *zd, uint64_t object, uint64_t offset)
2164 ztest_block_tag_t wbt;
2165 dmu_object_info_t doi;
2166 enum ztest_io_type io_type;
2170 VERIFY(dmu_object_info(zd->zd_os, object, &doi) == 0);
2171 blocksize = doi.doi_data_block_size;
2172 data = umem_alloc(blocksize, UMEM_NOFAIL);
2175 * Pick an i/o type at random, biased toward writing block tags.
2177 io_type = ztest_random(ZTEST_IO_TYPES);
2178 if (ztest_random(2) == 0)
2179 io_type = ZTEST_IO_WRITE_TAG;
2181 (void) rw_rdlock(&zd->zd_zilog_lock);
2185 case ZTEST_IO_WRITE_TAG:
2186 ztest_bt_generate(&wbt, zd->zd_os, object, offset, 0, 0, 0);
2187 (void) ztest_write(zd, object, offset, sizeof (wbt), &wbt);
2190 case ZTEST_IO_WRITE_PATTERN:
2191 (void) memset(data, 'a' + (object + offset) % 5, blocksize);
2192 if (ztest_random(2) == 0) {
2194 * Induce fletcher2 collisions to ensure that
2195 * zio_ddt_collision() detects and resolves them
2196 * when using fletcher2-verify for deduplication.
2198 ((uint64_t *)data)[0] ^= 1ULL << 63;
2199 ((uint64_t *)data)[4] ^= 1ULL << 63;
2201 (void) ztest_write(zd, object, offset, blocksize, data);
2204 case ZTEST_IO_WRITE_ZEROES:
2205 bzero(data, blocksize);
2206 (void) ztest_write(zd, object, offset, blocksize, data);
2209 case ZTEST_IO_TRUNCATE:
2210 (void) ztest_truncate(zd, object, offset, blocksize);
2213 case ZTEST_IO_SETATTR:
2214 (void) ztest_setattr(zd, object);
2217 case ZTEST_IO_REWRITE:
2218 (void) rw_rdlock(&ztest_name_lock);
2219 err = ztest_dsl_prop_set_uint64(zd->zd_name,
2220 ZFS_PROP_CHECKSUM, spa_dedup_checksum(ztest_spa),
2222 VERIFY(err == 0 || err == ENOSPC);
2223 err = ztest_dsl_prop_set_uint64(zd->zd_name,
2224 ZFS_PROP_COMPRESSION,
2225 ztest_random_dsl_prop(ZFS_PROP_COMPRESSION),
2227 VERIFY(err == 0 || err == ENOSPC);
2228 (void) rw_unlock(&ztest_name_lock);
2230 VERIFY0(dmu_read(zd->zd_os, object, offset, blocksize, data,
2231 DMU_READ_NO_PREFETCH));
2233 (void) ztest_write(zd, object, offset, blocksize, data);
2237 (void) rw_unlock(&zd->zd_zilog_lock);
2239 umem_free(data, blocksize);
2243 * Initialize an object description template.
2246 ztest_od_init(ztest_od_t *od, uint64_t id, char *tag, uint64_t index,
2247 dmu_object_type_t type, uint64_t blocksize, uint64_t gen)
2249 od->od_dir = ZTEST_DIROBJ;
2252 od->od_crtype = type;
2253 od->od_crblocksize = blocksize ? blocksize : ztest_random_blocksize();
2256 od->od_type = DMU_OT_NONE;
2257 od->od_blocksize = 0;
2260 (void) snprintf(od->od_name, sizeof (od->od_name), "%s(%lld)[%llu]",
2261 tag, (int64_t)id, index);
2265 * Lookup or create the objects for a test using the od template.
2266 * If the objects do not all exist, or if 'remove' is specified,
2267 * remove any existing objects and create new ones. Otherwise,
2268 * use the existing objects.
2271 ztest_object_init(ztest_ds_t *zd, ztest_od_t *od, size_t size, boolean_t remove)
2273 int count = size / sizeof (*od);
2276 VERIFY(mutex_lock(&zd->zd_dirobj_lock) == 0);
2277 if ((ztest_lookup(zd, od, count) != 0 || remove) &&
2278 (ztest_remove(zd, od, count) != 0 ||
2279 ztest_create(zd, od, count) != 0))
2282 VERIFY(mutex_unlock(&zd->zd_dirobj_lock) == 0);
2289 ztest_zil_commit(ztest_ds_t *zd, uint64_t id)
2291 zilog_t *zilog = zd->zd_zilog;
2293 (void) rw_rdlock(&zd->zd_zilog_lock);
2295 zil_commit(zilog, ztest_random(ZTEST_OBJECTS));
2298 * Remember the committed values in zd, which is in parent/child
2299 * shared memory. If we die, the next iteration of ztest_run()
2300 * will verify that the log really does contain this record.
2302 mutex_enter(&zilog->zl_lock);
2303 ASSERT(zd->zd_shared != NULL);
2304 ASSERT3U(zd->zd_shared->zd_seq, <=, zilog->zl_commit_lr_seq);
2305 zd->zd_shared->zd_seq = zilog->zl_commit_lr_seq;
2306 mutex_exit(&zilog->zl_lock);
2308 (void) rw_unlock(&zd->zd_zilog_lock);
2312 * This function is designed to simulate the operations that occur during a
2313 * mount/unmount operation. We hold the dataset across these operations in an
2314 * attempt to expose any implicit assumptions about ZIL management.
2318 ztest_zil_remount(ztest_ds_t *zd, uint64_t id)
2320 objset_t *os = zd->zd_os;
2323 * We grab the zd_dirobj_lock to ensure that no other thread is
2324 * updating the zil (i.e. adding in-memory log records) and the
2325 * zd_zilog_lock to block any I/O.
2327 VERIFY0(mutex_lock(&zd->zd_dirobj_lock));
2328 (void) rw_wrlock(&zd->zd_zilog_lock);
2330 /* zfsvfs_teardown() */
2331 zil_close(zd->zd_zilog);
2333 /* zfsvfs_setup() */
2334 VERIFY(zil_open(os, ztest_get_data) == zd->zd_zilog);
2335 zil_replay(os, zd, ztest_replay_vector);
2337 (void) rw_unlock(&zd->zd_zilog_lock);
2338 VERIFY(mutex_unlock(&zd->zd_dirobj_lock) == 0);
2342 * Verify that we can't destroy an active pool, create an existing pool,
2343 * or create a pool with a bad vdev spec.
2347 ztest_spa_create_destroy(ztest_ds_t *zd, uint64_t id)
2349 ztest_shared_opts_t *zo = &ztest_opts;
2354 * Attempt to create using a bad file.
2356 nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1);
2357 VERIFY3U(ENOENT, ==,
2358 spa_create("ztest_bad_file", nvroot, NULL, NULL));
2359 nvlist_free(nvroot);
2362 * Attempt to create using a bad mirror.
2364 nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 2, 1);
2365 VERIFY3U(ENOENT, ==,
2366 spa_create("ztest_bad_mirror", nvroot, NULL, NULL));
2367 nvlist_free(nvroot);
2370 * Attempt to create an existing pool. It shouldn't matter
2371 * what's in the nvroot; we should fail with EEXIST.
2373 (void) rw_rdlock(&ztest_name_lock);
2374 nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1);
2375 VERIFY3U(EEXIST, ==, spa_create(zo->zo_pool, nvroot, NULL, NULL));
2376 nvlist_free(nvroot);
2377 VERIFY3U(0, ==, spa_open(zo->zo_pool, &spa, FTAG));
2378 VERIFY3U(EBUSY, ==, spa_destroy(zo->zo_pool));
2379 spa_close(spa, FTAG);
2381 (void) rw_unlock(&ztest_name_lock);
2386 ztest_spa_upgrade(ztest_ds_t *zd, uint64_t id)
2389 uint64_t initial_version = SPA_VERSION_INITIAL;
2390 uint64_t version, newversion;
2391 nvlist_t *nvroot, *props;
2394 VERIFY0(mutex_lock(&ztest_vdev_lock));
2395 name = kmem_asprintf("%s_upgrade", ztest_opts.zo_pool);
2398 * Clean up from previous runs.
2400 (void) spa_destroy(name);
2402 nvroot = make_vdev_root(NULL, NULL, name, ztest_opts.zo_vdev_size, 0,
2403 0, ztest_opts.zo_raidz, ztest_opts.zo_mirrors, 1);
2406 * If we're configuring a RAIDZ device then make sure that the
2407 * the initial version is capable of supporting that feature.
2409 switch (ztest_opts.zo_raidz_parity) {
2412 initial_version = SPA_VERSION_INITIAL;
2415 initial_version = SPA_VERSION_RAIDZ2;
2418 initial_version = SPA_VERSION_RAIDZ3;
2423 * Create a pool with a spa version that can be upgraded. Pick
2424 * a value between initial_version and SPA_VERSION_BEFORE_FEATURES.
2427 version = ztest_random_spa_version(initial_version);
2428 } while (version > SPA_VERSION_BEFORE_FEATURES);
2430 props = fnvlist_alloc();
2431 fnvlist_add_uint64(props,
2432 zpool_prop_to_name(ZPOOL_PROP_VERSION), version);
2433 VERIFY0(spa_create(name, nvroot, props, NULL));
2434 fnvlist_free(nvroot);
2435 fnvlist_free(props);
2437 VERIFY0(spa_open(name, &spa, FTAG));
2438 VERIFY3U(spa_version(spa), ==, version);
2439 newversion = ztest_random_spa_version(version + 1);
2441 if (ztest_opts.zo_verbose >= 4) {
2442 (void) printf("upgrading spa version from %llu to %llu\n",
2443 (u_longlong_t)version, (u_longlong_t)newversion);
2446 spa_upgrade(spa, newversion);
2447 VERIFY3U(spa_version(spa), >, version);
2448 VERIFY3U(spa_version(spa), ==, fnvlist_lookup_uint64(spa->spa_config,
2449 zpool_prop_to_name(ZPOOL_PROP_VERSION)));
2450 spa_close(spa, FTAG);
2453 VERIFY0(mutex_unlock(&ztest_vdev_lock));
2457 vdev_lookup_by_path(vdev_t *vd, const char *path)
2461 if (vd->vdev_path != NULL && strcmp(path, vd->vdev_path) == 0)
2464 for (int c = 0; c < vd->vdev_children; c++)
2465 if ((mvd = vdev_lookup_by_path(vd->vdev_child[c], path)) !=
2473 * Find the first available hole which can be used as a top-level.
2476 find_vdev_hole(spa_t *spa)
2478 vdev_t *rvd = spa->spa_root_vdev;
2481 ASSERT(spa_config_held(spa, SCL_VDEV, RW_READER) == SCL_VDEV);
2483 for (c = 0; c < rvd->vdev_children; c++) {
2484 vdev_t *cvd = rvd->vdev_child[c];
2486 if (cvd->vdev_ishole)
2493 * Verify that vdev_add() works as expected.
2497 ztest_vdev_add_remove(ztest_ds_t *zd, uint64_t id)
2499 ztest_shared_t *zs = ztest_shared;
2500 spa_t *spa = ztest_spa;
2506 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2507 leaves = MAX(zs->zs_mirrors + zs->zs_splits, 1) * ztest_opts.zo_raidz;
2509 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2511 ztest_shared->zs_vdev_next_leaf = find_vdev_hole(spa) * leaves;
2514 * If we have slogs then remove them 1/4 of the time.
2516 if (spa_has_slogs(spa) && ztest_random(4) == 0) {
2518 * Grab the guid from the head of the log class rotor.
2520 guid = spa_log_class(spa)->mc_rotor->mg_vd->vdev_guid;
2522 spa_config_exit(spa, SCL_VDEV, FTAG);
2525 * We have to grab the zs_name_lock as writer to
2526 * prevent a race between removing a slog (dmu_objset_find)
2527 * and destroying a dataset. Removing the slog will
2528 * grab a reference on the dataset which may cause
2529 * dmu_objset_destroy() to fail with EBUSY thus
2530 * leaving the dataset in an inconsistent state.
2532 VERIFY(rw_wrlock(&ztest_name_lock) == 0);
2533 error = spa_vdev_remove(spa, guid, B_FALSE);
2534 VERIFY(rw_unlock(&ztest_name_lock) == 0);
2536 if (error && error != EEXIST)
2537 fatal(0, "spa_vdev_remove() = %d", error);
2539 spa_config_exit(spa, SCL_VDEV, FTAG);
2542 * Make 1/4 of the devices be log devices.
2544 nvroot = make_vdev_root(NULL, NULL, NULL,
2545 ztest_opts.zo_vdev_size, 0,
2546 ztest_random(4) == 0, ztest_opts.zo_raidz,
2549 error = spa_vdev_add(spa, nvroot);
2550 nvlist_free(nvroot);
2552 if (error == ENOSPC)
2553 ztest_record_enospc("spa_vdev_add");
2554 else if (error != 0)
2555 fatal(0, "spa_vdev_add() = %d", error);
2558 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2562 * Verify that adding/removing aux devices (l2arc, hot spare) works as expected.
2566 ztest_vdev_aux_add_remove(ztest_ds_t *zd, uint64_t id)
2568 ztest_shared_t *zs = ztest_shared;
2569 spa_t *spa = ztest_spa;
2570 vdev_t *rvd = spa->spa_root_vdev;
2571 spa_aux_vdev_t *sav;
2576 if (ztest_random(2) == 0) {
2577 sav = &spa->spa_spares;
2578 aux = ZPOOL_CONFIG_SPARES;
2580 sav = &spa->spa_l2cache;
2581 aux = ZPOOL_CONFIG_L2CACHE;
2584 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2586 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2588 if (sav->sav_count != 0 && ztest_random(4) == 0) {
2590 * Pick a random device to remove.
2592 guid = sav->sav_vdevs[ztest_random(sav->sav_count)]->vdev_guid;
2595 * Find an unused device we can add.
2597 zs->zs_vdev_aux = 0;
2599 char path[MAXPATHLEN];
2601 (void) snprintf(path, sizeof (path), ztest_aux_template,
2602 ztest_opts.zo_dir, ztest_opts.zo_pool, aux,
2604 for (c = 0; c < sav->sav_count; c++)
2605 if (strcmp(sav->sav_vdevs[c]->vdev_path,
2608 if (c == sav->sav_count &&
2609 vdev_lookup_by_path(rvd, path) == NULL)
2615 spa_config_exit(spa, SCL_VDEV, FTAG);
2621 nvlist_t *nvroot = make_vdev_root(NULL, aux, NULL,
2622 (ztest_opts.zo_vdev_size * 5) / 4, 0, 0, 0, 0, 1);
2623 error = spa_vdev_add(spa, nvroot);
2625 fatal(0, "spa_vdev_add(%p) = %d", nvroot, error);
2626 nvlist_free(nvroot);
2629 * Remove an existing device. Sometimes, dirty its
2630 * vdev state first to make sure we handle removal
2631 * of devices that have pending state changes.
2633 if (ztest_random(2) == 0)
2634 (void) vdev_online(spa, guid, 0, NULL);
2636 error = spa_vdev_remove(spa, guid, B_FALSE);
2637 if (error != 0 && error != EBUSY)
2638 fatal(0, "spa_vdev_remove(%llu) = %d", guid, error);
2641 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2645 * split a pool if it has mirror tlvdevs
2649 ztest_split_pool(ztest_ds_t *zd, uint64_t id)
2651 ztest_shared_t *zs = ztest_shared;
2652 spa_t *spa = ztest_spa;
2653 vdev_t *rvd = spa->spa_root_vdev;
2654 nvlist_t *tree, **child, *config, *split, **schild;
2655 uint_t c, children, schildren = 0, lastlogid = 0;
2658 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2660 /* ensure we have a useable config; mirrors of raidz aren't supported */
2661 if (zs->zs_mirrors < 3 || ztest_opts.zo_raidz > 1) {
2662 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2666 /* clean up the old pool, if any */
2667 (void) spa_destroy("splitp");
2669 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2671 /* generate a config from the existing config */
2672 mutex_enter(&spa->spa_props_lock);
2673 VERIFY(nvlist_lookup_nvlist(spa->spa_config, ZPOOL_CONFIG_VDEV_TREE,
2675 mutex_exit(&spa->spa_props_lock);
2677 VERIFY(nvlist_lookup_nvlist_array(tree, ZPOOL_CONFIG_CHILDREN, &child,
2680 schild = malloc(rvd->vdev_children * sizeof (nvlist_t *));
2681 for (c = 0; c < children; c++) {
2682 vdev_t *tvd = rvd->vdev_child[c];
2686 if (tvd->vdev_islog || tvd->vdev_ops == &vdev_hole_ops) {
2687 VERIFY(nvlist_alloc(&schild[schildren], NV_UNIQUE_NAME,
2689 VERIFY(nvlist_add_string(schild[schildren],
2690 ZPOOL_CONFIG_TYPE, VDEV_TYPE_HOLE) == 0);
2691 VERIFY(nvlist_add_uint64(schild[schildren],
2692 ZPOOL_CONFIG_IS_HOLE, 1) == 0);
2694 lastlogid = schildren;
2699 VERIFY(nvlist_lookup_nvlist_array(child[c],
2700 ZPOOL_CONFIG_CHILDREN, &mchild, &mchildren) == 0);
2701 VERIFY(nvlist_dup(mchild[0], &schild[schildren++], 0) == 0);
2704 /* OK, create a config that can be used to split */
2705 VERIFY(nvlist_alloc(&split, NV_UNIQUE_NAME, 0) == 0);
2706 VERIFY(nvlist_add_string(split, ZPOOL_CONFIG_TYPE,
2707 VDEV_TYPE_ROOT) == 0);
2708 VERIFY(nvlist_add_nvlist_array(split, ZPOOL_CONFIG_CHILDREN, schild,
2709 lastlogid != 0 ? lastlogid : schildren) == 0);
2711 VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, 0) == 0);
2712 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, split) == 0);
2714 for (c = 0; c < schildren; c++)
2715 nvlist_free(schild[c]);
2719 spa_config_exit(spa, SCL_VDEV, FTAG);
2721 (void) rw_wrlock(&ztest_name_lock);
2722 error = spa_vdev_split_mirror(spa, "splitp", config, NULL, B_FALSE);
2723 (void) rw_unlock(&ztest_name_lock);
2725 nvlist_free(config);
2728 (void) printf("successful split - results:\n");
2729 mutex_enter(&spa_namespace_lock);
2730 show_pool_stats(spa);
2731 show_pool_stats(spa_lookup("splitp"));
2732 mutex_exit(&spa_namespace_lock);
2736 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2741 * Verify that we can attach and detach devices.
2745 ztest_vdev_attach_detach(ztest_ds_t *zd, uint64_t id)
2747 ztest_shared_t *zs = ztest_shared;
2748 spa_t *spa = ztest_spa;
2749 spa_aux_vdev_t *sav = &spa->spa_spares;
2750 vdev_t *rvd = spa->spa_root_vdev;
2751 vdev_t *oldvd, *newvd, *pvd;
2755 uint64_t ashift = ztest_get_ashift();
2756 uint64_t oldguid, pguid;
2757 uint64_t oldsize, newsize;
2758 char oldpath[MAXPATHLEN], newpath[MAXPATHLEN];
2760 int oldvd_has_siblings = B_FALSE;
2761 int newvd_is_spare = B_FALSE;
2763 int error, expected_error;
2765 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2766 leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz;
2768 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2771 * Decide whether to do an attach or a replace.
2773 replacing = ztest_random(2);
2776 * Pick a random top-level vdev.
2778 top = ztest_random_vdev_top(spa, B_TRUE);
2781 * Pick a random leaf within it.
2783 leaf = ztest_random(leaves);
2788 oldvd = rvd->vdev_child[top];
2789 if (zs->zs_mirrors >= 1) {
2790 ASSERT(oldvd->vdev_ops == &vdev_mirror_ops);
2791 ASSERT(oldvd->vdev_children >= zs->zs_mirrors);
2792 oldvd = oldvd->vdev_child[leaf / ztest_opts.zo_raidz];
2794 if (ztest_opts.zo_raidz > 1) {
2795 ASSERT(oldvd->vdev_ops == &vdev_raidz_ops);
2796 ASSERT(oldvd->vdev_children == ztest_opts.zo_raidz);
2797 oldvd = oldvd->vdev_child[leaf % ztest_opts.zo_raidz];
2801 * If we're already doing an attach or replace, oldvd may be a
2802 * mirror vdev -- in which case, pick a random child.
2804 while (oldvd->vdev_children != 0) {
2805 oldvd_has_siblings = B_TRUE;
2806 ASSERT(oldvd->vdev_children >= 2);
2807 oldvd = oldvd->vdev_child[ztest_random(oldvd->vdev_children)];
2810 oldguid = oldvd->vdev_guid;
2811 oldsize = vdev_get_min_asize(oldvd);
2812 oldvd_is_log = oldvd->vdev_top->vdev_islog;
2813 (void) strcpy(oldpath, oldvd->vdev_path);
2814 pvd = oldvd->vdev_parent;
2815 pguid = pvd->vdev_guid;
2818 * If oldvd has siblings, then half of the time, detach it.
2820 if (oldvd_has_siblings && ztest_random(2) == 0) {
2821 spa_config_exit(spa, SCL_VDEV, FTAG);
2822 error = spa_vdev_detach(spa, oldguid, pguid, B_FALSE);
2823 if (error != 0 && error != ENODEV && error != EBUSY &&
2825 fatal(0, "detach (%s) returned %d", oldpath, error);
2826 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2831 * For the new vdev, choose with equal probability between the two
2832 * standard paths (ending in either 'a' or 'b') or a random hot spare.
2834 if (sav->sav_count != 0 && ztest_random(3) == 0) {
2835 newvd = sav->sav_vdevs[ztest_random(sav->sav_count)];
2836 newvd_is_spare = B_TRUE;
2837 (void) strcpy(newpath, newvd->vdev_path);
2839 (void) snprintf(newpath, sizeof (newpath), ztest_dev_template,
2840 ztest_opts.zo_dir, ztest_opts.zo_pool,
2841 top * leaves + leaf);
2842 if (ztest_random(2) == 0)
2843 newpath[strlen(newpath) - 1] = 'b';
2844 newvd = vdev_lookup_by_path(rvd, newpath);
2848 newsize = vdev_get_min_asize(newvd);
2851 * Make newsize a little bigger or smaller than oldsize.
2852 * If it's smaller, the attach should fail.
2853 * If it's larger, and we're doing a replace,
2854 * we should get dynamic LUN growth when we're done.
2856 newsize = 10 * oldsize / (9 + ztest_random(3));
2860 * If pvd is not a mirror or root, the attach should fail with ENOTSUP,
2861 * unless it's a replace; in that case any non-replacing parent is OK.
2863 * If newvd is already part of the pool, it should fail with EBUSY.
2865 * If newvd is too small, it should fail with EOVERFLOW.
2867 if (pvd->vdev_ops != &vdev_mirror_ops &&
2868 pvd->vdev_ops != &vdev_root_ops && (!replacing ||
2869 pvd->vdev_ops == &vdev_replacing_ops ||
2870 pvd->vdev_ops == &vdev_spare_ops))
2871 expected_error = ENOTSUP;
2872 else if (newvd_is_spare && (!replacing || oldvd_is_log))
2873 expected_error = ENOTSUP;
2874 else if (newvd == oldvd)
2875 expected_error = replacing ? 0 : EBUSY;
2876 else if (vdev_lookup_by_path(rvd, newpath) != NULL)
2877 expected_error = EBUSY;
2878 else if (newsize < oldsize)
2879 expected_error = EOVERFLOW;
2880 else if (ashift > oldvd->vdev_top->vdev_ashift)
2881 expected_error = EDOM;
2885 spa_config_exit(spa, SCL_VDEV, FTAG);
2888 * Build the nvlist describing newpath.
2890 root = make_vdev_root(newpath, NULL, NULL, newvd == NULL ? newsize : 0,
2891 ashift, 0, 0, 0, 1);
2893 error = spa_vdev_attach(spa, oldguid, root, replacing);
2898 * If our parent was the replacing vdev, but the replace completed,
2899 * then instead of failing with ENOTSUP we may either succeed,
2900 * fail with ENODEV, or fail with EOVERFLOW.
2902 if (expected_error == ENOTSUP &&
2903 (error == 0 || error == ENODEV || error == EOVERFLOW))
2904 expected_error = error;
2907 * If someone grew the LUN, the replacement may be too small.
2909 if (error == EOVERFLOW || error == EBUSY)
2910 expected_error = error;
2912 /* XXX workaround 6690467 */
2913 if (error != expected_error && expected_error != EBUSY) {
2914 fatal(0, "attach (%s %llu, %s %llu, %d) "
2915 "returned %d, expected %d",
2916 oldpath, oldsize, newpath,
2917 newsize, replacing, error, expected_error);
2920 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2924 * Callback function which expands the physical size of the vdev.
2927 grow_vdev(vdev_t *vd, void *arg)
2929 spa_t *spa = vd->vdev_spa;
2930 size_t *newsize = arg;
2934 ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
2935 ASSERT(vd->vdev_ops->vdev_op_leaf);
2937 if ((fd = open(vd->vdev_path, O_RDWR)) == -1)
2940 fsize = lseek(fd, 0, SEEK_END);
2941 (void) ftruncate(fd, *newsize);
2943 if (ztest_opts.zo_verbose >= 6) {
2944 (void) printf("%s grew from %lu to %lu bytes\n",
2945 vd->vdev_path, (ulong_t)fsize, (ulong_t)*newsize);
2952 * Callback function which expands a given vdev by calling vdev_online().
2956 online_vdev(vdev_t *vd, void *arg)
2958 spa_t *spa = vd->vdev_spa;
2959 vdev_t *tvd = vd->vdev_top;
2960 uint64_t guid = vd->vdev_guid;
2961 uint64_t generation = spa->spa_config_generation + 1;
2962 vdev_state_t newstate = VDEV_STATE_UNKNOWN;
2965 ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
2966 ASSERT(vd->vdev_ops->vdev_op_leaf);
2968 /* Calling vdev_online will initialize the new metaslabs */
2969 spa_config_exit(spa, SCL_STATE, spa);
2970 error = vdev_online(spa, guid, ZFS_ONLINE_EXPAND, &newstate);
2971 spa_config_enter(spa, SCL_STATE, spa, RW_READER);
2974 * If vdev_online returned an error or the underlying vdev_open
2975 * failed then we abort the expand. The only way to know that
2976 * vdev_open fails is by checking the returned newstate.
2978 if (error || newstate != VDEV_STATE_HEALTHY) {
2979 if (ztest_opts.zo_verbose >= 5) {
2980 (void) printf("Unable to expand vdev, state %llu, "
2981 "error %d\n", (u_longlong_t)newstate, error);
2985 ASSERT3U(newstate, ==, VDEV_STATE_HEALTHY);
2988 * Since we dropped the lock we need to ensure that we're
2989 * still talking to the original vdev. It's possible this
2990 * vdev may have been detached/replaced while we were
2991 * trying to online it.
2993 if (generation != spa->spa_config_generation) {
2994 if (ztest_opts.zo_verbose >= 5) {
2995 (void) printf("vdev configuration has changed, "
2996 "guid %llu, state %llu, expected gen %llu, "
2999 (u_longlong_t)tvd->vdev_state,
3000 (u_longlong_t)generation,
3001 (u_longlong_t)spa->spa_config_generation);
3009 * Traverse the vdev tree calling the supplied function.
3010 * We continue to walk the tree until we either have walked all
3011 * children or we receive a non-NULL return from the callback.
3012 * If a NULL callback is passed, then we just return back the first
3013 * leaf vdev we encounter.
3016 vdev_walk_tree(vdev_t *vd, vdev_t *(*func)(vdev_t *, void *), void *arg)
3018 if (vd->vdev_ops->vdev_op_leaf) {
3022 return (func(vd, arg));
3025 for (uint_t c = 0; c < vd->vdev_children; c++) {
3026 vdev_t *cvd = vd->vdev_child[c];
3027 if ((cvd = vdev_walk_tree(cvd, func, arg)) != NULL)
3034 * Verify that dynamic LUN growth works as expected.
3038 ztest_vdev_LUN_growth(ztest_ds_t *zd, uint64_t id)
3040 spa_t *spa = ztest_spa;
3042 metaslab_class_t *mc;
3043 metaslab_group_t *mg;
3044 size_t psize, newsize;
3046 uint64_t old_class_space, new_class_space, old_ms_count, new_ms_count;
3048 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
3049 spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3051 top = ztest_random_vdev_top(spa, B_TRUE);
3053 tvd = spa->spa_root_vdev->vdev_child[top];
3056 old_ms_count = tvd->vdev_ms_count;
3057 old_class_space = metaslab_class_get_space(mc);
3060 * Determine the size of the first leaf vdev associated with
3061 * our top-level device.
3063 vd = vdev_walk_tree(tvd, NULL, NULL);
3064 ASSERT3P(vd, !=, NULL);
3065 ASSERT(vd->vdev_ops->vdev_op_leaf);
3067 psize = vd->vdev_psize;
3070 * We only try to expand the vdev if it's healthy, less than 4x its
3071 * original size, and it has a valid psize.
3073 if (tvd->vdev_state != VDEV_STATE_HEALTHY ||
3074 psize == 0 || psize >= 4 * ztest_opts.zo_vdev_size) {
3075 spa_config_exit(spa, SCL_STATE, spa);
3076 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3080 newsize = psize + psize / 8;
3081 ASSERT3U(newsize, >, psize);
3083 if (ztest_opts.zo_verbose >= 6) {
3084 (void) printf("Expanding LUN %s from %lu to %lu\n",
3085 vd->vdev_path, (ulong_t)psize, (ulong_t)newsize);
3089 * Growing the vdev is a two step process:
3090 * 1). expand the physical size (i.e. relabel)
3091 * 2). online the vdev to create the new metaslabs
3093 if (vdev_walk_tree(tvd, grow_vdev, &newsize) != NULL ||
3094 vdev_walk_tree(tvd, online_vdev, NULL) != NULL ||
3095 tvd->vdev_state != VDEV_STATE_HEALTHY) {
3096 if (ztest_opts.zo_verbose >= 5) {
3097 (void) printf("Could not expand LUN because "
3098 "the vdev configuration changed.\n");
3100 spa_config_exit(spa, SCL_STATE, spa);
3101 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3105 spa_config_exit(spa, SCL_STATE, spa);
3108 * Expanding the LUN will update the config asynchronously,
3109 * thus we must wait for the async thread to complete any
3110 * pending tasks before proceeding.
3114 mutex_enter(&spa->spa_async_lock);
3115 done = (spa->spa_async_thread == NULL && !spa->spa_async_tasks);
3116 mutex_exit(&spa->spa_async_lock);
3119 txg_wait_synced(spa_get_dsl(spa), 0);
3120 (void) poll(NULL, 0, 100);
3123 spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3125 tvd = spa->spa_root_vdev->vdev_child[top];
3126 new_ms_count = tvd->vdev_ms_count;
3127 new_class_space = metaslab_class_get_space(mc);
3129 if (tvd->vdev_mg != mg || mg->mg_class != mc) {
3130 if (ztest_opts.zo_verbose >= 5) {
3131 (void) printf("Could not verify LUN expansion due to "
3132 "intervening vdev offline or remove.\n");
3134 spa_config_exit(spa, SCL_STATE, spa);
3135 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3140 * Make sure we were able to grow the vdev.
3142 if (new_ms_count <= old_ms_count)
3143 fatal(0, "LUN expansion failed: ms_count %llu <= %llu\n",
3144 old_ms_count, new_ms_count);
3147 * Make sure we were able to grow the pool.
3149 if (new_class_space <= old_class_space)
3150 fatal(0, "LUN expansion failed: class_space %llu <= %llu\n",
3151 old_class_space, new_class_space);
3153 if (ztest_opts.zo_verbose >= 5) {
3154 char oldnumbuf[6], newnumbuf[6];
3156 nicenum(old_class_space, oldnumbuf);
3157 nicenum(new_class_space, newnumbuf);
3158 (void) printf("%s grew from %s to %s\n",
3159 spa->spa_name, oldnumbuf, newnumbuf);
3162 spa_config_exit(spa, SCL_STATE, spa);
3163 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3167 * Verify that dmu_objset_{create,destroy,open,close} work as expected.
3171 ztest_objset_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
3174 * Create the objects common to all ztest datasets.
3176 VERIFY(zap_create_claim(os, ZTEST_DIROBJ,
3177 DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx) == 0);
3181 ztest_dataset_create(char *dsname)
3183 uint64_t zilset = ztest_random(100);
3184 int err = dmu_objset_create(dsname, DMU_OST_OTHER, 0,
3185 ztest_objset_create_cb, NULL);
3187 if (err || zilset < 80)
3190 if (ztest_opts.zo_verbose >= 6)
3191 (void) printf("Setting dataset %s to sync always\n", dsname);
3192 return (ztest_dsl_prop_set_uint64(dsname, ZFS_PROP_SYNC,
3193 ZFS_SYNC_ALWAYS, B_FALSE));
3198 ztest_objset_destroy_cb(const char *name, void *arg)
3201 dmu_object_info_t doi;
3205 * Verify that the dataset contains a directory object.
3207 VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_TRUE, FTAG, &os));
3208 error = dmu_object_info(os, ZTEST_DIROBJ, &doi);
3209 if (error != ENOENT) {
3210 /* We could have crashed in the middle of destroying it */
3212 ASSERT3U(doi.doi_type, ==, DMU_OT_ZAP_OTHER);
3213 ASSERT3S(doi.doi_physical_blocks_512, >=, 0);
3215 dmu_objset_disown(os, FTAG);
3218 * Destroy the dataset.
3220 if (strchr(name, '@') != NULL) {
3221 VERIFY0(dsl_destroy_snapshot(name, B_FALSE));
3223 VERIFY0(dsl_destroy_head(name));
3229 ztest_snapshot_create(char *osname, uint64_t id)
3231 char snapname[ZFS_MAX_DATASET_NAME_LEN];
3234 (void) snprintf(snapname, sizeof (snapname), "%llu", (u_longlong_t)id);
3236 error = dmu_objset_snapshot_one(osname, snapname);
3237 if (error == ENOSPC) {
3238 ztest_record_enospc(FTAG);
3241 if (error != 0 && error != EEXIST) {
3242 fatal(0, "ztest_snapshot_create(%s@%s) = %d", osname,
3249 ztest_snapshot_destroy(char *osname, uint64_t id)
3251 char snapname[ZFS_MAX_DATASET_NAME_LEN];
3254 (void) snprintf(snapname, sizeof (snapname), "%s@%llu", osname,
3257 error = dsl_destroy_snapshot(snapname, B_FALSE);
3258 if (error != 0 && error != ENOENT)
3259 fatal(0, "ztest_snapshot_destroy(%s) = %d", snapname, error);
3265 ztest_dmu_objset_create_destroy(ztest_ds_t *zd, uint64_t id)
3271 char name[ZFS_MAX_DATASET_NAME_LEN];
3274 (void) rw_rdlock(&ztest_name_lock);
3276 (void) snprintf(name, sizeof (name), "%s/temp_%llu",
3277 ztest_opts.zo_pool, (u_longlong_t)id);
3280 * If this dataset exists from a previous run, process its replay log
3281 * half of the time. If we don't replay it, then dmu_objset_destroy()
3282 * (invoked from ztest_objset_destroy_cb()) should just throw it away.
3284 if (ztest_random(2) == 0 &&
3285 dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os) == 0) {
3286 ztest_zd_init(&zdtmp, NULL, os);
3287 zil_replay(os, &zdtmp, ztest_replay_vector);
3288 ztest_zd_fini(&zdtmp);
3289 dmu_objset_disown(os, FTAG);
3293 * There may be an old instance of the dataset we're about to
3294 * create lying around from a previous run. If so, destroy it
3295 * and all of its snapshots.
3297 (void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
3298 DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
3301 * Verify that the destroyed dataset is no longer in the namespace.
3303 VERIFY3U(ENOENT, ==, dmu_objset_own(name, DMU_OST_OTHER, B_TRUE,
3307 * Verify that we can create a new dataset.
3309 error = ztest_dataset_create(name);
3311 if (error == ENOSPC) {
3312 ztest_record_enospc(FTAG);
3313 (void) rw_unlock(&ztest_name_lock);
3316 fatal(0, "dmu_objset_create(%s) = %d", name, error);
3319 VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os));
3321 ztest_zd_init(&zdtmp, NULL, os);
3324 * Open the intent log for it.
3326 zilog = zil_open(os, ztest_get_data);
3329 * Put some objects in there, do a little I/O to them,
3330 * and randomly take a couple of snapshots along the way.
3332 iters = ztest_random(5);
3333 for (int i = 0; i < iters; i++) {
3334 ztest_dmu_object_alloc_free(&zdtmp, id);
3335 if (ztest_random(iters) == 0)
3336 (void) ztest_snapshot_create(name, i);
3340 * Verify that we cannot create an existing dataset.
3342 VERIFY3U(EEXIST, ==,
3343 dmu_objset_create(name, DMU_OST_OTHER, 0, NULL, NULL));
3346 * Verify that we can hold an objset that is also owned.
3348 VERIFY3U(0, ==, dmu_objset_hold(name, FTAG, &os2));
3349 dmu_objset_rele(os2, FTAG);
3352 * Verify that we cannot own an objset that is already owned.
3355 dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os2));
3358 dmu_objset_disown(os, FTAG);
3359 ztest_zd_fini(&zdtmp);
3361 (void) rw_unlock(&ztest_name_lock);
3365 * Verify that dmu_snapshot_{create,destroy,open,close} work as expected.
3368 ztest_dmu_snapshot_create_destroy(ztest_ds_t *zd, uint64_t id)
3370 (void) rw_rdlock(&ztest_name_lock);
3371 (void) ztest_snapshot_destroy(zd->zd_name, id);
3372 (void) ztest_snapshot_create(zd->zd_name, id);
3373 (void) rw_unlock(&ztest_name_lock);
3377 * Cleanup non-standard snapshots and clones.
3380 ztest_dsl_dataset_cleanup(char *osname, uint64_t id)
3382 char snap1name[ZFS_MAX_DATASET_NAME_LEN];
3383 char clone1name[ZFS_MAX_DATASET_NAME_LEN];
3384 char snap2name[ZFS_MAX_DATASET_NAME_LEN];
3385 char clone2name[ZFS_MAX_DATASET_NAME_LEN];
3386 char snap3name[ZFS_MAX_DATASET_NAME_LEN];
3389 (void) snprintf(snap1name, sizeof (snap1name),
3390 "%s@s1_%llu", osname, id);
3391 (void) snprintf(clone1name, sizeof (clone1name),
3392 "%s/c1_%llu", osname, id);
3393 (void) snprintf(snap2name, sizeof (snap2name),
3394 "%s@s2_%llu", clone1name, id);
3395 (void) snprintf(clone2name, sizeof (clone2name),
3396 "%s/c2_%llu", osname, id);
3397 (void) snprintf(snap3name, sizeof (snap3name),
3398 "%s@s3_%llu", clone1name, id);
3400 error = dsl_destroy_head(clone2name);
3401 if (error && error != ENOENT)
3402 fatal(0, "dsl_destroy_head(%s) = %d", clone2name, error);
3403 error = dsl_destroy_snapshot(snap3name, B_FALSE);
3404 if (error && error != ENOENT)
3405 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap3name, error);
3406 error = dsl_destroy_snapshot(snap2name, B_FALSE);
3407 if (error && error != ENOENT)
3408 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap2name, error);
3409 error = dsl_destroy_head(clone1name);
3410 if (error && error != ENOENT)
3411 fatal(0, "dsl_destroy_head(%s) = %d", clone1name, error);
3412 error = dsl_destroy_snapshot(snap1name, B_FALSE);
3413 if (error && error != ENOENT)
3414 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap1name, error);
3418 * Verify dsl_dataset_promote handles EBUSY
3421 ztest_dsl_dataset_promote_busy(ztest_ds_t *zd, uint64_t id)
3424 char snap1name[ZFS_MAX_DATASET_NAME_LEN];
3425 char clone1name[ZFS_MAX_DATASET_NAME_LEN];
3426 char snap2name[ZFS_MAX_DATASET_NAME_LEN];
3427 char clone2name[ZFS_MAX_DATASET_NAME_LEN];
3428 char snap3name[ZFS_MAX_DATASET_NAME_LEN];
3429 char *osname = zd->zd_name;
3432 (void) rw_rdlock(&ztest_name_lock);
3434 ztest_dsl_dataset_cleanup(osname, id);
3436 (void) snprintf(snap1name, sizeof (snap1name),
3437 "%s@s1_%llu", osname, id);
3438 (void) snprintf(clone1name, sizeof (clone1name),
3439 "%s/c1_%llu", osname, id);
3440 (void) snprintf(snap2name, sizeof (snap2name),
3441 "%s@s2_%llu", clone1name, id);
3442 (void) snprintf(clone2name, sizeof (clone2name),
3443 "%s/c2_%llu", osname, id);
3444 (void) snprintf(snap3name, sizeof (snap3name),
3445 "%s@s3_%llu", clone1name, id);
3447 error = dmu_objset_snapshot_one(osname, strchr(snap1name, '@') + 1);
3448 if (error && error != EEXIST) {
3449 if (error == ENOSPC) {
3450 ztest_record_enospc(FTAG);
3453 fatal(0, "dmu_take_snapshot(%s) = %d", snap1name, error);
3456 error = dmu_objset_clone(clone1name, snap1name);
3458 if (error == ENOSPC) {
3459 ztest_record_enospc(FTAG);
3462 fatal(0, "dmu_objset_create(%s) = %d", clone1name, error);
3465 error = dmu_objset_snapshot_one(clone1name, strchr(snap2name, '@') + 1);
3466 if (error && error != EEXIST) {
3467 if (error == ENOSPC) {
3468 ztest_record_enospc(FTAG);
3471 fatal(0, "dmu_open_snapshot(%s) = %d", snap2name, error);
3474 error = dmu_objset_snapshot_one(clone1name, strchr(snap3name, '@') + 1);
3475 if (error && error != EEXIST) {
3476 if (error == ENOSPC) {
3477 ztest_record_enospc(FTAG);
3480 fatal(0, "dmu_open_snapshot(%s) = %d", snap3name, error);
3483 error = dmu_objset_clone(clone2name, snap3name);
3485 if (error == ENOSPC) {
3486 ztest_record_enospc(FTAG);
3489 fatal(0, "dmu_objset_create(%s) = %d", clone2name, error);
3492 error = dmu_objset_own(snap2name, DMU_OST_ANY, B_TRUE, FTAG, &os);
3494 fatal(0, "dmu_objset_own(%s) = %d", snap2name, error);
3495 error = dsl_dataset_promote(clone2name, NULL);
3496 if (error == ENOSPC) {
3497 dmu_objset_disown(os, FTAG);
3498 ztest_record_enospc(FTAG);
3502 fatal(0, "dsl_dataset_promote(%s), %d, not EBUSY", clone2name,
3504 dmu_objset_disown(os, FTAG);
3507 ztest_dsl_dataset_cleanup(osname, id);
3509 (void) rw_unlock(&ztest_name_lock);
3513 * Verify that dmu_object_{alloc,free} work as expected.
3516 ztest_dmu_object_alloc_free(ztest_ds_t *zd, uint64_t id)
3519 int batchsize = sizeof (od) / sizeof (od[0]);
3521 for (int b = 0; b < batchsize; b++)
3522 ztest_od_init(&od[b], id, FTAG, b, DMU_OT_UINT64_OTHER, 0, 0);
3525 * Destroy the previous batch of objects, create a new batch,
3526 * and do some I/O on the new objects.
3528 if (ztest_object_init(zd, od, sizeof (od), B_TRUE) != 0)
3531 while (ztest_random(4 * batchsize) != 0)
3532 ztest_io(zd, od[ztest_random(batchsize)].od_object,
3533 ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
3537 * Verify that dmu_{read,write} work as expected.
3540 ztest_dmu_read_write(ztest_ds_t *zd, uint64_t id)
3542 objset_t *os = zd->zd_os;
3545 int i, freeit, error;
3547 bufwad_t *packbuf, *bigbuf, *pack, *bigH, *bigT;
3548 uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
3549 uint64_t chunksize = (1000 + ztest_random(1000)) * sizeof (uint64_t);
3550 uint64_t regions = 997;
3551 uint64_t stride = 123456789ULL;
3552 uint64_t width = 40;
3553 int free_percent = 5;
3556 * This test uses two objects, packobj and bigobj, that are always
3557 * updated together (i.e. in the same tx) so that their contents are
3558 * in sync and can be compared. Their contents relate to each other
3559 * in a simple way: packobj is a dense array of 'bufwad' structures,
3560 * while bigobj is a sparse array of the same bufwads. Specifically,
3561 * for any index n, there are three bufwads that should be identical:
3563 * packobj, at offset n * sizeof (bufwad_t)
3564 * bigobj, at the head of the nth chunk
3565 * bigobj, at the tail of the nth chunk
3567 * The chunk size is arbitrary. It doesn't have to be a power of two,
3568 * and it doesn't have any relation to the object blocksize.
3569 * The only requirement is that it can hold at least two bufwads.
3571 * Normally, we write the bufwad to each of these locations.
3572 * However, free_percent of the time we instead write zeroes to
3573 * packobj and perform a dmu_free_range() on bigobj. By comparing
3574 * bigobj to packobj, we can verify that the DMU is correctly
3575 * tracking which parts of an object are allocated and free,
3576 * and that the contents of the allocated blocks are correct.
3580 * Read the directory info. If it's the first time, set things up.
3582 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, chunksize);
3583 ztest_od_init(&od[1], id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize);
3585 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
3588 bigobj = od[0].od_object;
3589 packobj = od[1].od_object;
3590 chunksize = od[0].od_gen;
3591 ASSERT(chunksize == od[1].od_gen);
3594 * Prefetch a random chunk of the big object.
3595 * Our aim here is to get some async reads in flight
3596 * for blocks that we may free below; the DMU should
3597 * handle this race correctly.
3599 n = ztest_random(regions) * stride + ztest_random(width);
3600 s = 1 + ztest_random(2 * width - 1);
3601 dmu_prefetch(os, bigobj, 0, n * chunksize, s * chunksize,
3602 ZIO_PRIORITY_SYNC_READ);
3605 * Pick a random index and compute the offsets into packobj and bigobj.
3607 n = ztest_random(regions) * stride + ztest_random(width);
3608 s = 1 + ztest_random(width - 1);
3610 packoff = n * sizeof (bufwad_t);
3611 packsize = s * sizeof (bufwad_t);
3613 bigoff = n * chunksize;
3614 bigsize = s * chunksize;
3616 packbuf = umem_alloc(packsize, UMEM_NOFAIL);
3617 bigbuf = umem_alloc(bigsize, UMEM_NOFAIL);
3620 * free_percent of the time, free a range of bigobj rather than
3623 freeit = (ztest_random(100) < free_percent);
3626 * Read the current contents of our objects.
3628 error = dmu_read(os, packobj, packoff, packsize, packbuf,
3631 error = dmu_read(os, bigobj, bigoff, bigsize, bigbuf,
3636 * Get a tx for the mods to both packobj and bigobj.
3638 tx = dmu_tx_create(os);
3640 dmu_tx_hold_write(tx, packobj, packoff, packsize);
3643 dmu_tx_hold_free(tx, bigobj, bigoff, bigsize);
3645 dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
3647 /* This accounts for setting the checksum/compression. */
3648 dmu_tx_hold_bonus(tx, bigobj);
3650 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3652 umem_free(packbuf, packsize);
3653 umem_free(bigbuf, bigsize);
3657 enum zio_checksum cksum;
3659 cksum = (enum zio_checksum)
3660 ztest_random_dsl_prop(ZFS_PROP_CHECKSUM);
3661 } while (cksum >= ZIO_CHECKSUM_LEGACY_FUNCTIONS);
3662 dmu_object_set_checksum(os, bigobj, cksum, tx);
3664 enum zio_compress comp;
3666 comp = (enum zio_compress)
3667 ztest_random_dsl_prop(ZFS_PROP_COMPRESSION);
3668 } while (comp >= ZIO_COMPRESS_LEGACY_FUNCTIONS);
3669 dmu_object_set_compress(os, bigobj, comp, tx);
3672 * For each index from n to n + s, verify that the existing bufwad
3673 * in packobj matches the bufwads at the head and tail of the
3674 * corresponding chunk in bigobj. Then update all three bufwads
3675 * with the new values we want to write out.
3677 for (i = 0; i < s; i++) {
3679 pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
3681 bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
3683 bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
3685 ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
3686 ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
3688 if (pack->bw_txg > txg)
3689 fatal(0, "future leak: got %llx, open txg is %llx",
3692 if (pack->bw_data != 0 && pack->bw_index != n + i)
3693 fatal(0, "wrong index: got %llx, wanted %llx+%llx",
3694 pack->bw_index, n, i);
3696 if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
3697 fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
3699 if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
3700 fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
3703 bzero(pack, sizeof (bufwad_t));
3705 pack->bw_index = n + i;
3707 pack->bw_data = 1 + ztest_random(-2ULL);
3714 * We've verified all the old bufwads, and made new ones.
3715 * Now write them out.
3717 dmu_write(os, packobj, packoff, packsize, packbuf, tx);
3720 if (ztest_opts.zo_verbose >= 7) {
3721 (void) printf("freeing offset %llx size %llx"
3723 (u_longlong_t)bigoff,
3724 (u_longlong_t)bigsize,
3727 VERIFY(0 == dmu_free_range(os, bigobj, bigoff, bigsize, tx));
3729 if (ztest_opts.zo_verbose >= 7) {
3730 (void) printf("writing offset %llx size %llx"
3732 (u_longlong_t)bigoff,
3733 (u_longlong_t)bigsize,
3736 dmu_write(os, bigobj, bigoff, bigsize, bigbuf, tx);
3742 * Sanity check the stuff we just wrote.
3745 void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
3746 void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
3748 VERIFY(0 == dmu_read(os, packobj, packoff,
3749 packsize, packcheck, DMU_READ_PREFETCH));
3750 VERIFY(0 == dmu_read(os, bigobj, bigoff,
3751 bigsize, bigcheck, DMU_READ_PREFETCH));
3753 ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
3754 ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
3756 umem_free(packcheck, packsize);
3757 umem_free(bigcheck, bigsize);
3760 umem_free(packbuf, packsize);
3761 umem_free(bigbuf, bigsize);
3765 compare_and_update_pbbufs(uint64_t s, bufwad_t *packbuf, bufwad_t *bigbuf,
3766 uint64_t bigsize, uint64_t n, uint64_t chunksize, uint64_t txg)
3774 * For each index from n to n + s, verify that the existing bufwad
3775 * in packobj matches the bufwads at the head and tail of the
3776 * corresponding chunk in bigobj. Then update all three bufwads
3777 * with the new values we want to write out.
3779 for (i = 0; i < s; i++) {
3781 pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
3783 bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
3785 bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
3787 ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
3788 ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
3790 if (pack->bw_txg > txg)
3791 fatal(0, "future leak: got %llx, open txg is %llx",
3794 if (pack->bw_data != 0 && pack->bw_index != n + i)
3795 fatal(0, "wrong index: got %llx, wanted %llx+%llx",
3796 pack->bw_index, n, i);
3798 if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
3799 fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
3801 if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
3802 fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
3804 pack->bw_index = n + i;
3806 pack->bw_data = 1 + ztest_random(-2ULL);
3814 ztest_dmu_read_write_zcopy(ztest_ds_t *zd, uint64_t id)
3816 objset_t *os = zd->zd_os;
3822 bufwad_t *packbuf, *bigbuf;
3823 uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
3824 uint64_t blocksize = ztest_random_blocksize();
3825 uint64_t chunksize = blocksize;
3826 uint64_t regions = 997;
3827 uint64_t stride = 123456789ULL;
3829 dmu_buf_t *bonus_db;
3830 arc_buf_t **bigbuf_arcbufs;
3831 dmu_object_info_t doi;
3834 * This test uses two objects, packobj and bigobj, that are always
3835 * updated together (i.e. in the same tx) so that their contents are
3836 * in sync and can be compared. Their contents relate to each other
3837 * in a simple way: packobj is a dense array of 'bufwad' structures,
3838 * while bigobj is a sparse array of the same bufwads. Specifically,
3839 * for any index n, there are three bufwads that should be identical:
3841 * packobj, at offset n * sizeof (bufwad_t)
3842 * bigobj, at the head of the nth chunk
3843 * bigobj, at the tail of the nth chunk
3845 * The chunk size is set equal to bigobj block size so that
3846 * dmu_assign_arcbuf() can be tested for object updates.
3850 * Read the directory info. If it's the first time, set things up.
3852 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
3853 ztest_od_init(&od[1], id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize);
3855 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
3858 bigobj = od[0].od_object;
3859 packobj = od[1].od_object;
3860 blocksize = od[0].od_blocksize;
3861 chunksize = blocksize;
3862 ASSERT(chunksize == od[1].od_gen);
3864 VERIFY(dmu_object_info(os, bigobj, &doi) == 0);
3865 VERIFY(ISP2(doi.doi_data_block_size));
3866 VERIFY(chunksize == doi.doi_data_block_size);
3867 VERIFY(chunksize >= 2 * sizeof (bufwad_t));
3870 * Pick a random index and compute the offsets into packobj and bigobj.
3872 n = ztest_random(regions) * stride + ztest_random(width);
3873 s = 1 + ztest_random(width - 1);
3875 packoff = n * sizeof (bufwad_t);
3876 packsize = s * sizeof (bufwad_t);
3878 bigoff = n * chunksize;
3879 bigsize = s * chunksize;
3881 packbuf = umem_zalloc(packsize, UMEM_NOFAIL);
3882 bigbuf = umem_zalloc(bigsize, UMEM_NOFAIL);
3884 VERIFY3U(0, ==, dmu_bonus_hold(os, bigobj, FTAG, &bonus_db));
3886 bigbuf_arcbufs = umem_zalloc(2 * s * sizeof (arc_buf_t *), UMEM_NOFAIL);
3889 * Iteration 0 test zcopy for DB_UNCACHED dbufs.
3890 * Iteration 1 test zcopy to already referenced dbufs.
3891 * Iteration 2 test zcopy to dirty dbuf in the same txg.
3892 * Iteration 3 test zcopy to dbuf dirty in previous txg.
3893 * Iteration 4 test zcopy when dbuf is no longer dirty.
3894 * Iteration 5 test zcopy when it can't be done.
3895 * Iteration 6 one more zcopy write.
3897 for (i = 0; i < 7; i++) {
3902 * In iteration 5 (i == 5) use arcbufs
3903 * that don't match bigobj blksz to test
3904 * dmu_assign_arcbuf() when it can't directly
3905 * assign an arcbuf to a dbuf.
3907 for (j = 0; j < s; j++) {
3910 dmu_request_arcbuf(bonus_db, chunksize);
3912 bigbuf_arcbufs[2 * j] =
3913 dmu_request_arcbuf(bonus_db, chunksize / 2);
3914 bigbuf_arcbufs[2 * j + 1] =
3915 dmu_request_arcbuf(bonus_db, chunksize / 2);
3920 * Get a tx for the mods to both packobj and bigobj.
3922 tx = dmu_tx_create(os);
3924 dmu_tx_hold_write(tx, packobj, packoff, packsize);
3925 dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
3927 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3929 umem_free(packbuf, packsize);
3930 umem_free(bigbuf, bigsize);
3931 for (j = 0; j < s; j++) {
3933 dmu_return_arcbuf(bigbuf_arcbufs[j]);
3936 bigbuf_arcbufs[2 * j]);
3938 bigbuf_arcbufs[2 * j + 1]);
3941 umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
3942 dmu_buf_rele(bonus_db, FTAG);
3947 * 50% of the time don't read objects in the 1st iteration to
3948 * test dmu_assign_arcbuf() for the case when there're no
3949 * existing dbufs for the specified offsets.
3951 if (i != 0 || ztest_random(2) != 0) {
3952 error = dmu_read(os, packobj, packoff,
3953 packsize, packbuf, DMU_READ_PREFETCH);
3955 error = dmu_read(os, bigobj, bigoff, bigsize,
3956 bigbuf, DMU_READ_PREFETCH);
3959 compare_and_update_pbbufs(s, packbuf, bigbuf, bigsize,
3963 * We've verified all the old bufwads, and made new ones.
3964 * Now write them out.
3966 dmu_write(os, packobj, packoff, packsize, packbuf, tx);
3967 if (ztest_opts.zo_verbose >= 7) {
3968 (void) printf("writing offset %llx size %llx"
3970 (u_longlong_t)bigoff,
3971 (u_longlong_t)bigsize,
3974 for (off = bigoff, j = 0; j < s; j++, off += chunksize) {
3977 bcopy((caddr_t)bigbuf + (off - bigoff),
3978 bigbuf_arcbufs[j]->b_data, chunksize);
3980 bcopy((caddr_t)bigbuf + (off - bigoff),
3981 bigbuf_arcbufs[2 * j]->b_data,
3983 bcopy((caddr_t)bigbuf + (off - bigoff) +
3985 bigbuf_arcbufs[2 * j + 1]->b_data,
3990 VERIFY(dmu_buf_hold(os, bigobj, off,
3991 FTAG, &dbt, DMU_READ_NO_PREFETCH) == 0);
3994 dmu_assign_arcbuf(bonus_db, off,
3995 bigbuf_arcbufs[j], tx);
3997 dmu_assign_arcbuf(bonus_db, off,
3998 bigbuf_arcbufs[2 * j], tx);
3999 dmu_assign_arcbuf(bonus_db,
4000 off + chunksize / 2,
4001 bigbuf_arcbufs[2 * j + 1], tx);
4004 dmu_buf_rele(dbt, FTAG);
4010 * Sanity check the stuff we just wrote.
4013 void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
4014 void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
4016 VERIFY(0 == dmu_read(os, packobj, packoff,
4017 packsize, packcheck, DMU_READ_PREFETCH));
4018 VERIFY(0 == dmu_read(os, bigobj, bigoff,
4019 bigsize, bigcheck, DMU_READ_PREFETCH));
4021 ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
4022 ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
4024 umem_free(packcheck, packsize);
4025 umem_free(bigcheck, bigsize);
4028 txg_wait_open(dmu_objset_pool(os), 0);
4029 } else if (i == 3) {
4030 txg_wait_synced(dmu_objset_pool(os), 0);
4034 dmu_buf_rele(bonus_db, FTAG);
4035 umem_free(packbuf, packsize);
4036 umem_free(bigbuf, bigsize);
4037 umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
4042 ztest_dmu_write_parallel(ztest_ds_t *zd, uint64_t id)
4045 uint64_t offset = (1ULL << (ztest_random(20) + 43)) +
4046 (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
4049 * Have multiple threads write to large offsets in an object
4050 * to verify that parallel writes to an object -- even to the
4051 * same blocks within the object -- doesn't cause any trouble.
4053 ztest_od_init(&od[0], ID_PARALLEL, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
4055 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4058 while (ztest_random(10) != 0)
4059 ztest_io(zd, od[0].od_object, offset);
4063 ztest_dmu_prealloc(ztest_ds_t *zd, uint64_t id)
4066 uint64_t offset = (1ULL << (ztest_random(4) + SPA_MAXBLOCKSHIFT)) +
4067 (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
4068 uint64_t count = ztest_random(20) + 1;
4069 uint64_t blocksize = ztest_random_blocksize();
4072 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
4074 if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4077 if (ztest_truncate(zd, od[0].od_object, offset, count * blocksize) != 0)
4080 ztest_prealloc(zd, od[0].od_object, offset, count * blocksize);
4082 data = umem_zalloc(blocksize, UMEM_NOFAIL);
4084 while (ztest_random(count) != 0) {
4085 uint64_t randoff = offset + (ztest_random(count) * blocksize);
4086 if (ztest_write(zd, od[0].od_object, randoff, blocksize,
4089 while (ztest_random(4) != 0)
4090 ztest_io(zd, od[0].od_object, randoff);
4093 umem_free(data, blocksize);
4097 * Verify that zap_{create,destroy,add,remove,update} work as expected.
4099 #define ZTEST_ZAP_MIN_INTS 1
4100 #define ZTEST_ZAP_MAX_INTS 4
4101 #define ZTEST_ZAP_MAX_PROPS 1000
4104 ztest_zap(ztest_ds_t *zd, uint64_t id)
4106 objset_t *os = zd->zd_os;
4109 uint64_t txg, last_txg;
4110 uint64_t value[ZTEST_ZAP_MAX_INTS];
4111 uint64_t zl_ints, zl_intsize, prop;
4114 char propname[100], txgname[100];
4116 char *hc[2] = { "s.acl.h", ".s.open.h.hyLZlg" };
4118 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0);
4120 if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4123 object = od[0].od_object;
4126 * Generate a known hash collision, and verify that
4127 * we can lookup and remove both entries.
4129 tx = dmu_tx_create(os);
4130 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4131 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4134 for (i = 0; i < 2; i++) {
4136 VERIFY3U(0, ==, zap_add(os, object, hc[i], sizeof (uint64_t),
4139 for (i = 0; i < 2; i++) {
4140 VERIFY3U(EEXIST, ==, zap_add(os, object, hc[i],
4141 sizeof (uint64_t), 1, &value[i], tx));
4143 zap_length(os, object, hc[i], &zl_intsize, &zl_ints));
4144 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4145 ASSERT3U(zl_ints, ==, 1);
4147 for (i = 0; i < 2; i++) {
4148 VERIFY3U(0, ==, zap_remove(os, object, hc[i], tx));
4153 * Generate a buch of random entries.
4155 ints = MAX(ZTEST_ZAP_MIN_INTS, object % ZTEST_ZAP_MAX_INTS);
4157 prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
4158 (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
4159 (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
4160 bzero(value, sizeof (value));
4164 * If these zap entries already exist, validate their contents.
4166 error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
4168 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4169 ASSERT3U(zl_ints, ==, 1);
4171 VERIFY(zap_lookup(os, object, txgname, zl_intsize,
4172 zl_ints, &last_txg) == 0);
4174 VERIFY(zap_length(os, object, propname, &zl_intsize,
4177 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4178 ASSERT3U(zl_ints, ==, ints);
4180 VERIFY(zap_lookup(os, object, propname, zl_intsize,
4181 zl_ints, value) == 0);
4183 for (i = 0; i < ints; i++) {
4184 ASSERT3U(value[i], ==, last_txg + object + i);
4187 ASSERT3U(error, ==, ENOENT);
4191 * Atomically update two entries in our zap object.
4192 * The first is named txg_%llu, and contains the txg
4193 * in which the property was last updated. The second
4194 * is named prop_%llu, and the nth element of its value
4195 * should be txg + object + n.
4197 tx = dmu_tx_create(os);
4198 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4199 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4204 fatal(0, "zap future leak: old %llu new %llu", last_txg, txg);
4206 for (i = 0; i < ints; i++)
4207 value[i] = txg + object + i;
4209 VERIFY3U(0, ==, zap_update(os, object, txgname, sizeof (uint64_t),
4211 VERIFY3U(0, ==, zap_update(os, object, propname, sizeof (uint64_t),
4217 * Remove a random pair of entries.
4219 prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
4220 (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
4221 (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
4223 error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
4225 if (error == ENOENT)
4230 tx = dmu_tx_create(os);
4231 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4232 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4235 VERIFY3U(0, ==, zap_remove(os, object, txgname, tx));
4236 VERIFY3U(0, ==, zap_remove(os, object, propname, tx));
4241 * Testcase to test the upgrading of a microzap to fatzap.
4244 ztest_fzap(ztest_ds_t *zd, uint64_t id)
4246 objset_t *os = zd->zd_os;
4248 uint64_t object, txg;
4250 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0);
4252 if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4255 object = od[0].od_object;
4258 * Add entries to this ZAP and make sure it spills over
4259 * and gets upgraded to a fatzap. Also, since we are adding
4260 * 2050 entries we should see ptrtbl growth and leaf-block split.
4262 for (int i = 0; i < 2050; i++) {
4263 char name[ZFS_MAX_DATASET_NAME_LEN];
4268 (void) snprintf(name, sizeof (name), "fzap-%llu-%llu",
4271 tx = dmu_tx_create(os);
4272 dmu_tx_hold_zap(tx, object, B_TRUE, name);
4273 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4276 error = zap_add(os, object, name, sizeof (uint64_t), 1,
4278 ASSERT(error == 0 || error == EEXIST);
4285 ztest_zap_parallel(ztest_ds_t *zd, uint64_t id)
4287 objset_t *os = zd->zd_os;
4289 uint64_t txg, object, count, wsize, wc, zl_wsize, zl_wc;
4291 int i, namelen, error;
4292 int micro = ztest_random(2);
4293 char name[20], string_value[20];
4296 ztest_od_init(&od[0], ID_PARALLEL, FTAG, micro, DMU_OT_ZAP_OTHER, 0, 0);
4298 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4301 object = od[0].od_object;
4304 * Generate a random name of the form 'xxx.....' where each
4305 * x is a random printable character and the dots are dots.
4306 * There are 94 such characters, and the name length goes from
4307 * 6 to 20, so there are 94^3 * 15 = 12,458,760 possible names.
4309 namelen = ztest_random(sizeof (name) - 5) + 5 + 1;
4311 for (i = 0; i < 3; i++)
4312 name[i] = '!' + ztest_random('~' - '!' + 1);
4313 for (; i < namelen - 1; i++)
4317 if ((namelen & 1) || micro) {
4318 wsize = sizeof (txg);
4324 data = string_value;
4328 VERIFY0(zap_count(os, object, &count));
4329 ASSERT(count != -1ULL);
4332 * Select an operation: length, lookup, add, update, remove.
4334 i = ztest_random(5);
4337 tx = dmu_tx_create(os);
4338 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4339 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4342 bcopy(name, string_value, namelen);
4346 bzero(string_value, namelen);
4352 error = zap_length(os, object, name, &zl_wsize, &zl_wc);
4354 ASSERT3U(wsize, ==, zl_wsize);
4355 ASSERT3U(wc, ==, zl_wc);
4357 ASSERT3U(error, ==, ENOENT);
4362 error = zap_lookup(os, object, name, wsize, wc, data);
4364 if (data == string_value &&
4365 bcmp(name, data, namelen) != 0)
4366 fatal(0, "name '%s' != val '%s' len %d",
4367 name, data, namelen);
4369 ASSERT3U(error, ==, ENOENT);
4374 error = zap_add(os, object, name, wsize, wc, data, tx);
4375 ASSERT(error == 0 || error == EEXIST);
4379 VERIFY(zap_update(os, object, name, wsize, wc, data, tx) == 0);
4383 error = zap_remove(os, object, name, tx);
4384 ASSERT(error == 0 || error == ENOENT);
4393 * Commit callback data.
4395 typedef struct ztest_cb_data {
4396 list_node_t zcd_node;
4398 int zcd_expected_err;
4399 boolean_t zcd_added;
4400 boolean_t zcd_called;
4404 /* This is the actual commit callback function */
4406 ztest_commit_callback(void *arg, int error)
4408 ztest_cb_data_t *data = arg;
4409 uint64_t synced_txg;
4411 VERIFY(data != NULL);
4412 VERIFY3S(data->zcd_expected_err, ==, error);
4413 VERIFY(!data->zcd_called);
4415 synced_txg = spa_last_synced_txg(data->zcd_spa);
4416 if (data->zcd_txg > synced_txg)
4417 fatal(0, "commit callback of txg %" PRIu64 " called prematurely"
4418 ", last synced txg = %" PRIu64 "\n", data->zcd_txg,
4421 data->zcd_called = B_TRUE;
4423 if (error == ECANCELED) {
4424 ASSERT0(data->zcd_txg);
4425 ASSERT(!data->zcd_added);
4428 * The private callback data should be destroyed here, but
4429 * since we are going to check the zcd_called field after
4430 * dmu_tx_abort(), we will destroy it there.
4435 /* Was this callback added to the global callback list? */
4436 if (!data->zcd_added)
4439 ASSERT3U(data->zcd_txg, !=, 0);
4441 /* Remove our callback from the list */
4442 (void) mutex_lock(&zcl.zcl_callbacks_lock);
4443 list_remove(&zcl.zcl_callbacks, data);
4444 (void) mutex_unlock(&zcl.zcl_callbacks_lock);
4447 umem_free(data, sizeof (ztest_cb_data_t));
4450 /* Allocate and initialize callback data structure */
4451 static ztest_cb_data_t *
4452 ztest_create_cb_data(objset_t *os, uint64_t txg)
4454 ztest_cb_data_t *cb_data;
4456 cb_data = umem_zalloc(sizeof (ztest_cb_data_t), UMEM_NOFAIL);
4458 cb_data->zcd_txg = txg;
4459 cb_data->zcd_spa = dmu_objset_spa(os);
4465 * If a number of txgs equal to this threshold have been created after a commit
4466 * callback has been registered but not called, then we assume there is an
4467 * implementation bug.
4469 #define ZTEST_COMMIT_CALLBACK_THRESH (TXG_CONCURRENT_STATES + 2)
4472 * Commit callback test.
4475 ztest_dmu_commit_callbacks(ztest_ds_t *zd, uint64_t id)
4477 objset_t *os = zd->zd_os;
4480 ztest_cb_data_t *cb_data[3], *tmp_cb;
4481 uint64_t old_txg, txg;
4484 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
4486 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4489 tx = dmu_tx_create(os);
4491 cb_data[0] = ztest_create_cb_data(os, 0);
4492 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[0]);
4494 dmu_tx_hold_write(tx, od[0].od_object, 0, sizeof (uint64_t));
4496 /* Every once in a while, abort the transaction on purpose */
4497 if (ztest_random(100) == 0)
4501 error = dmu_tx_assign(tx, TXG_NOWAIT);
4503 txg = error ? 0 : dmu_tx_get_txg(tx);
4505 cb_data[0]->zcd_txg = txg;
4506 cb_data[1] = ztest_create_cb_data(os, txg);
4507 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[1]);
4511 * It's not a strict requirement to call the registered
4512 * callbacks from inside dmu_tx_abort(), but that's what
4513 * it's supposed to happen in the current implementation
4514 * so we will check for that.
4516 for (i = 0; i < 2; i++) {
4517 cb_data[i]->zcd_expected_err = ECANCELED;
4518 VERIFY(!cb_data[i]->zcd_called);
4523 for (i = 0; i < 2; i++) {
4524 VERIFY(cb_data[i]->zcd_called);
4525 umem_free(cb_data[i], sizeof (ztest_cb_data_t));
4531 cb_data[2] = ztest_create_cb_data(os, txg);
4532 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[2]);
4535 * Read existing data to make sure there isn't a future leak.
4537 VERIFY(0 == dmu_read(os, od[0].od_object, 0, sizeof (uint64_t),
4538 &old_txg, DMU_READ_PREFETCH));
4541 fatal(0, "future leak: got %" PRIu64 ", open txg is %" PRIu64,
4544 dmu_write(os, od[0].od_object, 0, sizeof (uint64_t), &txg, tx);
4546 (void) mutex_lock(&zcl.zcl_callbacks_lock);
4549 * Since commit callbacks don't have any ordering requirement and since
4550 * it is theoretically possible for a commit callback to be called
4551 * after an arbitrary amount of time has elapsed since its txg has been
4552 * synced, it is difficult to reliably determine whether a commit
4553 * callback hasn't been called due to high load or due to a flawed
4556 * In practice, we will assume that if after a certain number of txgs a
4557 * commit callback hasn't been called, then most likely there's an
4558 * implementation bug..
4560 tmp_cb = list_head(&zcl.zcl_callbacks);
4561 if (tmp_cb != NULL &&
4562 (txg - ZTEST_COMMIT_CALLBACK_THRESH) > tmp_cb->zcd_txg) {
4563 fatal(0, "Commit callback threshold exceeded, oldest txg: %"
4564 PRIu64 ", open txg: %" PRIu64 "\n", tmp_cb->zcd_txg, txg);
4568 * Let's find the place to insert our callbacks.
4570 * Even though the list is ordered by txg, it is possible for the
4571 * insertion point to not be the end because our txg may already be
4572 * quiescing at this point and other callbacks in the open txg
4573 * (from other objsets) may have sneaked in.
4575 tmp_cb = list_tail(&zcl.zcl_callbacks);
4576 while (tmp_cb != NULL && tmp_cb->zcd_txg > txg)
4577 tmp_cb = list_prev(&zcl.zcl_callbacks, tmp_cb);
4579 /* Add the 3 callbacks to the list */
4580 for (i = 0; i < 3; i++) {
4582 list_insert_head(&zcl.zcl_callbacks, cb_data[i]);
4584 list_insert_after(&zcl.zcl_callbacks, tmp_cb,
4587 cb_data[i]->zcd_added = B_TRUE;
4588 VERIFY(!cb_data[i]->zcd_called);
4590 tmp_cb = cb_data[i];
4593 (void) mutex_unlock(&zcl.zcl_callbacks_lock);
4600 ztest_dsl_prop_get_set(ztest_ds_t *zd, uint64_t id)
4602 zfs_prop_t proplist[] = {
4604 ZFS_PROP_COMPRESSION,
4609 (void) rw_rdlock(&ztest_name_lock);
4611 for (int p = 0; p < sizeof (proplist) / sizeof (proplist[0]); p++)
4612 (void) ztest_dsl_prop_set_uint64(zd->zd_name, proplist[p],
4613 ztest_random_dsl_prop(proplist[p]), (int)ztest_random(2));
4615 (void) rw_unlock(&ztest_name_lock);
4620 ztest_spa_prop_get_set(ztest_ds_t *zd, uint64_t id)
4622 nvlist_t *props = NULL;
4624 (void) rw_rdlock(&ztest_name_lock);
4626 (void) ztest_spa_prop_set_uint64(ZPOOL_PROP_DEDUPDITTO,
4627 ZIO_DEDUPDITTO_MIN + ztest_random(ZIO_DEDUPDITTO_MIN));
4629 VERIFY0(spa_prop_get(ztest_spa, &props));
4631 if (ztest_opts.zo_verbose >= 6)
4632 dump_nvlist(props, 4);
4636 (void) rw_unlock(&ztest_name_lock);
4640 user_release_one(const char *snapname, const char *holdname)
4642 nvlist_t *snaps, *holds;
4645 snaps = fnvlist_alloc();
4646 holds = fnvlist_alloc();
4647 fnvlist_add_boolean(holds, holdname);
4648 fnvlist_add_nvlist(snaps, snapname, holds);
4649 fnvlist_free(holds);
4650 error = dsl_dataset_user_release(snaps, NULL);
4651 fnvlist_free(snaps);
4656 * Test snapshot hold/release and deferred destroy.
4659 ztest_dmu_snapshot_hold(ztest_ds_t *zd, uint64_t id)
4662 objset_t *os = zd->zd_os;
4666 char clonename[100];
4668 char osname[ZFS_MAX_DATASET_NAME_LEN];
4671 (void) rw_rdlock(&ztest_name_lock);
4673 dmu_objset_name(os, osname);
4675 (void) snprintf(snapname, sizeof (snapname), "sh1_%llu", id);
4676 (void) snprintf(fullname, sizeof (fullname), "%s@%s", osname, snapname);
4677 (void) snprintf(clonename, sizeof (clonename),
4678 "%s/ch1_%llu", osname, id);
4679 (void) snprintf(tag, sizeof (tag), "tag_%llu", id);
4682 * Clean up from any previous run.
4684 error = dsl_destroy_head(clonename);
4685 if (error != ENOENT)
4687 error = user_release_one(fullname, tag);
4688 if (error != ESRCH && error != ENOENT)
4690 error = dsl_destroy_snapshot(fullname, B_FALSE);
4691 if (error != ENOENT)
4695 * Create snapshot, clone it, mark snap for deferred destroy,
4696 * destroy clone, verify snap was also destroyed.
4698 error = dmu_objset_snapshot_one(osname, snapname);
4700 if (error == ENOSPC) {
4701 ztest_record_enospc("dmu_objset_snapshot");
4704 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
4707 error = dmu_objset_clone(clonename, fullname);
4709 if (error == ENOSPC) {
4710 ztest_record_enospc("dmu_objset_clone");
4713 fatal(0, "dmu_objset_clone(%s) = %d", clonename, error);
4716 error = dsl_destroy_snapshot(fullname, B_TRUE);
4718 fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
4722 error = dsl_destroy_head(clonename);
4724 fatal(0, "dsl_destroy_head(%s) = %d", clonename, error);
4726 error = dmu_objset_hold(fullname, FTAG, &origin);
4727 if (error != ENOENT)
4728 fatal(0, "dmu_objset_hold(%s) = %d", fullname, error);
4731 * Create snapshot, add temporary hold, verify that we can't
4732 * destroy a held snapshot, mark for deferred destroy,
4733 * release hold, verify snapshot was destroyed.
4735 error = dmu_objset_snapshot_one(osname, snapname);
4737 if (error == ENOSPC) {
4738 ztest_record_enospc("dmu_objset_snapshot");
4741 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
4744 holds = fnvlist_alloc();
4745 fnvlist_add_string(holds, fullname, tag);
4746 error = dsl_dataset_user_hold(holds, 0, NULL);
4747 fnvlist_free(holds);
4749 if (error == ENOSPC) {
4750 ztest_record_enospc("dsl_dataset_user_hold");
4753 fatal(0, "dsl_dataset_user_hold(%s, %s) = %u",
4754 fullname, tag, error);
4757 error = dsl_destroy_snapshot(fullname, B_FALSE);
4758 if (error != EBUSY) {
4759 fatal(0, "dsl_destroy_snapshot(%s, B_FALSE) = %d",
4763 error = dsl_destroy_snapshot(fullname, B_TRUE);
4765 fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
4769 error = user_release_one(fullname, tag);
4771 fatal(0, "user_release_one(%s, %s) = %d", fullname, tag, error);
4773 VERIFY3U(dmu_objset_hold(fullname, FTAG, &origin), ==, ENOENT);
4776 (void) rw_unlock(&ztest_name_lock);
4780 * Inject random faults into the on-disk data.
4784 ztest_fault_inject(ztest_ds_t *zd, uint64_t id)
4786 ztest_shared_t *zs = ztest_shared;
4787 spa_t *spa = ztest_spa;
4791 uint64_t bad = 0x1990c0ffeedecadeULL;
4793 char path0[MAXPATHLEN];
4794 char pathrand[MAXPATHLEN];
4796 int bshift = SPA_OLD_MAXBLOCKSHIFT + 2; /* don't scrog all labels */
4802 boolean_t islog = B_FALSE;
4804 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
4805 maxfaults = MAXFAULTS();
4806 leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz;
4807 mirror_save = zs->zs_mirrors;
4808 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4810 ASSERT(leaves >= 1);
4813 * Grab the name lock as reader. There are some operations
4814 * which don't like to have their vdevs changed while
4815 * they are in progress (i.e. spa_change_guid). Those
4816 * operations will have grabbed the name lock as writer.
4818 (void) rw_rdlock(&ztest_name_lock);
4821 * We need SCL_STATE here because we're going to look at vd0->vdev_tsd.
4823 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
4825 if (ztest_random(2) == 0) {
4827 * Inject errors on a normal data device or slog device.
4829 top = ztest_random_vdev_top(spa, B_TRUE);
4830 leaf = ztest_random(leaves) + zs->zs_splits;
4833 * Generate paths to the first leaf in this top-level vdev,
4834 * and to the random leaf we selected. We'll induce transient
4835 * write failures and random online/offline activity on leaf 0,
4836 * and we'll write random garbage to the randomly chosen leaf.
4838 (void) snprintf(path0, sizeof (path0), ztest_dev_template,
4839 ztest_opts.zo_dir, ztest_opts.zo_pool,
4840 top * leaves + zs->zs_splits);
4841 (void) snprintf(pathrand, sizeof (pathrand), ztest_dev_template,
4842 ztest_opts.zo_dir, ztest_opts.zo_pool,
4843 top * leaves + leaf);
4845 vd0 = vdev_lookup_by_path(spa->spa_root_vdev, path0);
4846 if (vd0 != NULL && vd0->vdev_top->vdev_islog)
4850 * If the top-level vdev needs to be resilvered
4851 * then we only allow faults on the device that is
4854 if (vd0 != NULL && maxfaults != 1 &&
4855 (!vdev_resilver_needed(vd0->vdev_top, NULL, NULL) ||
4856 vd0->vdev_resilver_txg != 0)) {
4858 * Make vd0 explicitly claim to be unreadable,
4859 * or unwriteable, or reach behind its back
4860 * and close the underlying fd. We can do this if
4861 * maxfaults == 0 because we'll fail and reexecute,
4862 * and we can do it if maxfaults >= 2 because we'll
4863 * have enough redundancy. If maxfaults == 1, the
4864 * combination of this with injection of random data
4865 * corruption below exceeds the pool's fault tolerance.
4867 vdev_file_t *vf = vd0->vdev_tsd;
4869 if (vf != NULL && ztest_random(3) == 0) {
4870 (void) close(vf->vf_vnode->v_fd);
4871 vf->vf_vnode->v_fd = -1;
4872 } else if (ztest_random(2) == 0) {
4873 vd0->vdev_cant_read = B_TRUE;
4875 vd0->vdev_cant_write = B_TRUE;
4877 guid0 = vd0->vdev_guid;
4881 * Inject errors on an l2cache device.
4883 spa_aux_vdev_t *sav = &spa->spa_l2cache;
4885 if (sav->sav_count == 0) {
4886 spa_config_exit(spa, SCL_STATE, FTAG);
4887 (void) rw_unlock(&ztest_name_lock);
4890 vd0 = sav->sav_vdevs[ztest_random(sav->sav_count)];
4891 guid0 = vd0->vdev_guid;
4892 (void) strcpy(path0, vd0->vdev_path);
4893 (void) strcpy(pathrand, vd0->vdev_path);
4897 maxfaults = INT_MAX; /* no limit on cache devices */
4900 spa_config_exit(spa, SCL_STATE, FTAG);
4901 (void) rw_unlock(&ztest_name_lock);
4904 * If we can tolerate two or more faults, or we're dealing
4905 * with a slog, randomly online/offline vd0.
4907 if ((maxfaults >= 2 || islog) && guid0 != 0) {
4908 if (ztest_random(10) < 6) {
4909 int flags = (ztest_random(2) == 0 ?
4910 ZFS_OFFLINE_TEMPORARY : 0);
4913 * We have to grab the zs_name_lock as writer to
4914 * prevent a race between offlining a slog and
4915 * destroying a dataset. Offlining the slog will
4916 * grab a reference on the dataset which may cause
4917 * dmu_objset_destroy() to fail with EBUSY thus
4918 * leaving the dataset in an inconsistent state.
4921 (void) rw_wrlock(&ztest_name_lock);
4923 VERIFY(vdev_offline(spa, guid0, flags) != EBUSY);
4926 (void) rw_unlock(&ztest_name_lock);
4929 * Ideally we would like to be able to randomly
4930 * call vdev_[on|off]line without holding locks
4931 * to force unpredictable failures but the side
4932 * effects of vdev_[on|off]line prevent us from
4933 * doing so. We grab the ztest_vdev_lock here to
4934 * prevent a race between injection testing and
4937 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
4938 (void) vdev_online(spa, guid0, 0, NULL);
4939 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4947 * We have at least single-fault tolerance, so inject data corruption.
4949 fd = open(pathrand, O_RDWR);
4951 if (fd == -1) /* we hit a gap in the device namespace */
4954 fsize = lseek(fd, 0, SEEK_END);
4956 while (--iters != 0) {
4957 offset = ztest_random(fsize / (leaves << bshift)) *
4958 (leaves << bshift) + (leaf << bshift) +
4959 (ztest_random(1ULL << (bshift - 1)) & -8ULL);
4961 if (offset >= fsize)
4964 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
4965 if (mirror_save != zs->zs_mirrors) {
4966 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4971 if (pwrite(fd, &bad, sizeof (bad), offset) != sizeof (bad))
4972 fatal(1, "can't inject bad word at 0x%llx in %s",
4975 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4977 if (ztest_opts.zo_verbose >= 7)
4978 (void) printf("injected bad word into %s,"
4979 " offset 0x%llx\n", pathrand, (u_longlong_t)offset);
4986 * Verify that DDT repair works as expected.
4989 ztest_ddt_repair(ztest_ds_t *zd, uint64_t id)
4991 ztest_shared_t *zs = ztest_shared;
4992 spa_t *spa = ztest_spa;
4993 objset_t *os = zd->zd_os;
4995 uint64_t object, blocksize, txg, pattern, psize;
4996 enum zio_checksum checksum = spa_dedup_checksum(spa);
5001 int copies = 2 * ZIO_DEDUPDITTO_MIN;
5003 blocksize = ztest_random_blocksize();
5004 blocksize = MIN(blocksize, 2048); /* because we write so many */
5006 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
5008 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
5012 * Take the name lock as writer to prevent anyone else from changing
5013 * the pool and dataset properies we need to maintain during this test.
5015 (void) rw_wrlock(&ztest_name_lock);
5017 if (ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_DEDUP, checksum,
5019 ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_COPIES, 1,
5021 (void) rw_unlock(&ztest_name_lock);
5025 dmu_objset_stats_t dds;
5026 dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
5027 dmu_objset_fast_stat(os, &dds);
5028 dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
5030 object = od[0].od_object;
5031 blocksize = od[0].od_blocksize;
5032 pattern = zs->zs_guid ^ dds.dds_guid;
5034 ASSERT(object != 0);
5036 tx = dmu_tx_create(os);
5037 dmu_tx_hold_write(tx, object, 0, copies * blocksize);
5038 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
5040 (void) rw_unlock(&ztest_name_lock);
5045 * Write all the copies of our block.
5047 for (int i = 0; i < copies; i++) {
5048 uint64_t offset = i * blocksize;
5049 int error = dmu_buf_hold(os, object, offset, FTAG, &db,
5050 DMU_READ_NO_PREFETCH);
5052 fatal(B_FALSE, "dmu_buf_hold(%p, %llu, %llu) = %u",
5053 os, (long long)object, (long long) offset, error);
5055 ASSERT(db->db_offset == offset);
5056 ASSERT(db->db_size == blocksize);
5057 ASSERT(ztest_pattern_match(db->db_data, db->db_size, pattern) ||
5058 ztest_pattern_match(db->db_data, db->db_size, 0ULL));
5059 dmu_buf_will_fill(db, tx);
5060 ztest_pattern_set(db->db_data, db->db_size, pattern);
5061 dmu_buf_rele(db, FTAG);
5065 txg_wait_synced(spa_get_dsl(spa), txg);
5068 * Find out what block we got.
5070 VERIFY0(dmu_buf_hold(os, object, 0, FTAG, &db,
5071 DMU_READ_NO_PREFETCH));
5072 blk = *((dmu_buf_impl_t *)db)->db_blkptr;
5073 dmu_buf_rele(db, FTAG);
5076 * Damage the block. Dedup-ditto will save us when we read it later.
5078 psize = BP_GET_PSIZE(&blk);
5079 buf = zio_buf_alloc(psize);
5080 ztest_pattern_set(buf, psize, ~pattern);
5082 (void) zio_wait(zio_rewrite(NULL, spa, 0, &blk,
5083 buf, psize, NULL, NULL, ZIO_PRIORITY_SYNC_WRITE,
5084 ZIO_FLAG_CANFAIL | ZIO_FLAG_INDUCE_DAMAGE, NULL));
5086 zio_buf_free(buf, psize);
5088 (void) rw_unlock(&ztest_name_lock);
5096 ztest_scrub(ztest_ds_t *zd, uint64_t id)
5098 spa_t *spa = ztest_spa;
5100 (void) spa_scan(spa, POOL_SCAN_SCRUB);
5101 (void) poll(NULL, 0, 100); /* wait a moment, then force a restart */
5102 (void) spa_scan(spa, POOL_SCAN_SCRUB);
5106 * Change the guid for the pool.
5110 ztest_reguid(ztest_ds_t *zd, uint64_t id)
5112 spa_t *spa = ztest_spa;
5113 uint64_t orig, load;
5116 orig = spa_guid(spa);
5117 load = spa_load_guid(spa);
5119 (void) rw_wrlock(&ztest_name_lock);
5120 error = spa_change_guid(spa);
5121 (void) rw_unlock(&ztest_name_lock);
5126 if (ztest_opts.zo_verbose >= 4) {
5127 (void) printf("Changed guid old %llu -> %llu\n",
5128 (u_longlong_t)orig, (u_longlong_t)spa_guid(spa));
5131 VERIFY3U(orig, !=, spa_guid(spa));
5132 VERIFY3U(load, ==, spa_load_guid(spa));
5136 * Rename the pool to a different name and then rename it back.
5140 ztest_spa_rename(ztest_ds_t *zd, uint64_t id)
5142 char *oldname, *newname;
5145 (void) rw_wrlock(&ztest_name_lock);
5147 oldname = ztest_opts.zo_pool;
5148 newname = umem_alloc(strlen(oldname) + 5, UMEM_NOFAIL);
5149 (void) strcpy(newname, oldname);
5150 (void) strcat(newname, "_tmp");
5155 VERIFY3U(0, ==, spa_rename(oldname, newname));
5158 * Try to open it under the old name, which shouldn't exist
5160 VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
5163 * Open it under the new name and make sure it's still the same spa_t.
5165 VERIFY3U(0, ==, spa_open(newname, &spa, FTAG));
5167 ASSERT(spa == ztest_spa);
5168 spa_close(spa, FTAG);
5171 * Rename it back to the original
5173 VERIFY3U(0, ==, spa_rename(newname, oldname));
5176 * Make sure it can still be opened
5178 VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG));
5180 ASSERT(spa == ztest_spa);
5181 spa_close(spa, FTAG);
5183 umem_free(newname, strlen(newname) + 1);
5185 (void) rw_unlock(&ztest_name_lock);
5189 * Verify pool integrity by running zdb.
5192 ztest_run_zdb(char *pool)
5195 char zdb[MAXPATHLEN + MAXNAMELEN + 20];
5203 strlcpy(zdb, "/usr/bin/ztest", sizeof(zdb));
5205 /* zdb lives in /usr/sbin, while ztest lives in /usr/bin */
5206 bin = strstr(zdb, "/usr/bin/");
5207 ztest = strstr(bin, "/ztest");
5209 isalen = ztest - isa;
5213 "/usr/sbin%.*s/zdb -bcc%s%s -d -U %s %s",
5216 ztest_opts.zo_verbose >= 3 ? "s" : "",
5217 ztest_opts.zo_verbose >= 4 ? "v" : "",
5222 if (ztest_opts.zo_verbose >= 5)
5223 (void) printf("Executing %s\n", strstr(zdb, "zdb "));
5225 fp = popen(zdb, "r");
5228 while (fgets(zbuf, sizeof (zbuf), fp) != NULL)
5229 if (ztest_opts.zo_verbose >= 3)
5230 (void) printf("%s", zbuf);
5232 status = pclose(fp);
5237 ztest_dump_core = 0;
5238 if (WIFEXITED(status))
5239 fatal(0, "'%s' exit code %d", zdb, WEXITSTATUS(status));
5241 fatal(0, "'%s' died with signal %d", zdb, WTERMSIG(status));
5245 ztest_walk_pool_directory(char *header)
5249 if (ztest_opts.zo_verbose >= 6)
5250 (void) printf("%s\n", header);
5252 mutex_enter(&spa_namespace_lock);
5253 while ((spa = spa_next(spa)) != NULL)
5254 if (ztest_opts.zo_verbose >= 6)
5255 (void) printf("\t%s\n", spa_name(spa));
5256 mutex_exit(&spa_namespace_lock);
5260 ztest_spa_import_export(char *oldname, char *newname)
5262 nvlist_t *config, *newconfig;
5267 if (ztest_opts.zo_verbose >= 4) {
5268 (void) printf("import/export: old = %s, new = %s\n",
5273 * Clean up from previous runs.
5275 (void) spa_destroy(newname);
5278 * Get the pool's configuration and guid.
5280 VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG));
5283 * Kick off a scrub to tickle scrub/export races.
5285 if (ztest_random(2) == 0)
5286 (void) spa_scan(spa, POOL_SCAN_SCRUB);
5288 pool_guid = spa_guid(spa);
5289 spa_close(spa, FTAG);
5291 ztest_walk_pool_directory("pools before export");
5296 VERIFY3U(0, ==, spa_export(oldname, &config, B_FALSE, B_FALSE));
5298 ztest_walk_pool_directory("pools after export");
5303 newconfig = spa_tryimport(config);
5304 ASSERT(newconfig != NULL);
5305 nvlist_free(newconfig);
5308 * Import it under the new name.
5310 error = spa_import(newname, config, NULL, 0);
5312 dump_nvlist(config, 0);
5313 fatal(B_FALSE, "couldn't import pool %s as %s: error %u",
5314 oldname, newname, error);
5317 ztest_walk_pool_directory("pools after import");
5320 * Try to import it again -- should fail with EEXIST.
5322 VERIFY3U(EEXIST, ==, spa_import(newname, config, NULL, 0));
5325 * Try to import it under a different name -- should fail with EEXIST.
5327 VERIFY3U(EEXIST, ==, spa_import(oldname, config, NULL, 0));
5330 * Verify that the pool is no longer visible under the old name.
5332 VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
5335 * Verify that we can open and close the pool using the new name.
5337 VERIFY3U(0, ==, spa_open(newname, &spa, FTAG));
5338 ASSERT(pool_guid == spa_guid(spa));
5339 spa_close(spa, FTAG);
5341 nvlist_free(config);
5345 ztest_resume(spa_t *spa)
5347 if (spa_suspended(spa) && ztest_opts.zo_verbose >= 6)
5348 (void) printf("resuming from suspended state\n");
5349 spa_vdev_state_enter(spa, SCL_NONE);
5350 vdev_clear(spa, NULL);
5351 (void) spa_vdev_state_exit(spa, NULL, 0);
5352 (void) zio_resume(spa);
5356 ztest_resume_thread(void *arg)
5360 while (!ztest_exiting) {
5361 if (spa_suspended(spa))
5363 (void) poll(NULL, 0, 100);
5366 * Periodically change the zfs_compressed_arc_enabled setting.
5368 if (ztest_random(10) == 0)
5369 zfs_compressed_arc_enabled = ztest_random(2);
5375 ztest_deadman_thread(void *arg)
5377 ztest_shared_t *zs = arg;
5378 spa_t *spa = ztest_spa;
5379 hrtime_t delta, total = 0;
5382 delta = zs->zs_thread_stop - zs->zs_thread_start +
5383 MSEC2NSEC(zfs_deadman_synctime_ms);
5385 (void) poll(NULL, 0, (int)NSEC2MSEC(delta));
5388 * If the pool is suspended then fail immediately. Otherwise,
5389 * check to see if the pool is making any progress. If
5390 * vdev_deadman() discovers that there hasn't been any recent
5391 * I/Os then it will end up aborting the tests.
5393 if (spa_suspended(spa) || spa->spa_root_vdev == NULL) {
5394 fatal(0, "aborting test after %llu seconds because "
5395 "pool has transitioned to a suspended state.",
5396 zfs_deadman_synctime_ms / 1000);
5399 vdev_deadman(spa->spa_root_vdev);
5401 total += zfs_deadman_synctime_ms/1000;
5402 (void) printf("ztest has been running for %lld seconds\n",
5408 ztest_execute(int test, ztest_info_t *zi, uint64_t id)
5410 ztest_ds_t *zd = &ztest_ds[id % ztest_opts.zo_datasets];
5411 ztest_shared_callstate_t *zc = ZTEST_GET_SHARED_CALLSTATE(test);
5412 hrtime_t functime = gethrtime();
5414 for (int i = 0; i < zi->zi_iters; i++)
5415 zi->zi_func(zd, id);
5417 functime = gethrtime() - functime;
5419 atomic_add_64(&zc->zc_count, 1);
5420 atomic_add_64(&zc->zc_time, functime);
5422 if (ztest_opts.zo_verbose >= 4) {
5424 (void) dladdr((void *)zi->zi_func, &dli);
5425 (void) printf("%6.2f sec in %s\n",
5426 (double)functime / NANOSEC, dli.dli_sname);
5431 ztest_thread(void *arg)
5434 uint64_t id = (uintptr_t)arg;
5435 ztest_shared_t *zs = ztest_shared;
5439 ztest_shared_callstate_t *zc;
5441 while ((now = gethrtime()) < zs->zs_thread_stop) {
5443 * See if it's time to force a crash.
5445 if (now > zs->zs_thread_kill)
5449 * If we're getting ENOSPC with some regularity, stop.
5451 if (zs->zs_enospc_count > 10)
5455 * Pick a random function to execute.
5457 rand = ztest_random(ZTEST_FUNCS);
5458 zi = &ztest_info[rand];
5459 zc = ZTEST_GET_SHARED_CALLSTATE(rand);
5460 call_next = zc->zc_next;
5462 if (now >= call_next &&
5463 atomic_cas_64(&zc->zc_next, call_next, call_next +
5464 ztest_random(2 * zi->zi_interval[0] + 1)) == call_next) {
5465 ztest_execute(rand, zi, id);
5473 ztest_dataset_name(char *dsname, char *pool, int d)
5475 (void) snprintf(dsname, ZFS_MAX_DATASET_NAME_LEN, "%s/ds_%d", pool, d);
5479 ztest_dataset_destroy(int d)
5481 char name[ZFS_MAX_DATASET_NAME_LEN];
5483 ztest_dataset_name(name, ztest_opts.zo_pool, d);
5485 if (ztest_opts.zo_verbose >= 3)
5486 (void) printf("Destroying %s to free up space\n", name);
5489 * Cleanup any non-standard clones and snapshots. In general,
5490 * ztest thread t operates on dataset (t % zopt_datasets),
5491 * so there may be more than one thing to clean up.
5493 for (int t = d; t < ztest_opts.zo_threads;
5494 t += ztest_opts.zo_datasets) {
5495 ztest_dsl_dataset_cleanup(name, t);
5498 (void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
5499 DS_FIND_SNAPSHOTS | DS_FIND_CHILDREN);
5503 ztest_dataset_dirobj_verify(ztest_ds_t *zd)
5505 uint64_t usedobjs, dirobjs, scratch;
5508 * ZTEST_DIROBJ is the object directory for the entire dataset.
5509 * Therefore, the number of objects in use should equal the
5510 * number of ZTEST_DIROBJ entries, +1 for ZTEST_DIROBJ itself.
5511 * If not, we have an object leak.
5513 * Note that we can only check this in ztest_dataset_open(),
5514 * when the open-context and syncing-context values agree.
5515 * That's because zap_count() returns the open-context value,
5516 * while dmu_objset_space() returns the rootbp fill count.
5518 VERIFY3U(0, ==, zap_count(zd->zd_os, ZTEST_DIROBJ, &dirobjs));
5519 dmu_objset_space(zd->zd_os, &scratch, &scratch, &usedobjs, &scratch);
5520 ASSERT3U(dirobjs + 1, ==, usedobjs);
5524 ztest_dataset_open(int d)
5526 ztest_ds_t *zd = &ztest_ds[d];
5527 uint64_t committed_seq = ZTEST_GET_SHARED_DS(d)->zd_seq;
5530 char name[ZFS_MAX_DATASET_NAME_LEN];
5533 ztest_dataset_name(name, ztest_opts.zo_pool, d);
5535 (void) rw_rdlock(&ztest_name_lock);
5537 error = ztest_dataset_create(name);
5538 if (error == ENOSPC) {
5539 (void) rw_unlock(&ztest_name_lock);
5540 ztest_record_enospc(FTAG);
5543 ASSERT(error == 0 || error == EEXIST);
5545 VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, zd, &os));
5546 (void) rw_unlock(&ztest_name_lock);
5548 ztest_zd_init(zd, ZTEST_GET_SHARED_DS(d), os);
5550 zilog = zd->zd_zilog;
5552 if (zilog->zl_header->zh_claim_lr_seq != 0 &&
5553 zilog->zl_header->zh_claim_lr_seq < committed_seq)
5554 fatal(0, "missing log records: claimed %llu < committed %llu",
5555 zilog->zl_header->zh_claim_lr_seq, committed_seq);
5557 ztest_dataset_dirobj_verify(zd);
5559 zil_replay(os, zd, ztest_replay_vector);
5561 ztest_dataset_dirobj_verify(zd);
5563 if (ztest_opts.zo_verbose >= 6)
5564 (void) printf("%s replay %llu blocks, %llu records, seq %llu\n",
5566 (u_longlong_t)zilog->zl_parse_blk_count,
5567 (u_longlong_t)zilog->zl_parse_lr_count,
5568 (u_longlong_t)zilog->zl_replaying_seq);
5570 zilog = zil_open(os, ztest_get_data);
5572 if (zilog->zl_replaying_seq != 0 &&
5573 zilog->zl_replaying_seq < committed_seq)
5574 fatal(0, "missing log records: replayed %llu < committed %llu",
5575 zilog->zl_replaying_seq, committed_seq);
5581 ztest_dataset_close(int d)
5583 ztest_ds_t *zd = &ztest_ds[d];
5585 zil_close(zd->zd_zilog);
5586 dmu_objset_disown(zd->zd_os, zd);
5592 * Kick off threads to run tests on all datasets in parallel.
5595 ztest_run(ztest_shared_t *zs)
5600 thread_t resume_tid;
5603 ztest_exiting = B_FALSE;
5606 * Initialize parent/child shared state.
5608 VERIFY(_mutex_init(&ztest_vdev_lock, USYNC_THREAD, NULL) == 0);
5609 VERIFY(rwlock_init(&ztest_name_lock, USYNC_THREAD, NULL) == 0);
5611 zs->zs_thread_start = gethrtime();
5612 zs->zs_thread_stop =
5613 zs->zs_thread_start + ztest_opts.zo_passtime * NANOSEC;
5614 zs->zs_thread_stop = MIN(zs->zs_thread_stop, zs->zs_proc_stop);
5615 zs->zs_thread_kill = zs->zs_thread_stop;
5616 if (ztest_random(100) < ztest_opts.zo_killrate) {
5617 zs->zs_thread_kill -=
5618 ztest_random(ztest_opts.zo_passtime * NANOSEC);
5621 (void) _mutex_init(&zcl.zcl_callbacks_lock, USYNC_THREAD, NULL);
5623 list_create(&zcl.zcl_callbacks, sizeof (ztest_cb_data_t),
5624 offsetof(ztest_cb_data_t, zcd_node));
5629 kernel_init(FREAD | FWRITE);
5630 VERIFY0(spa_open(ztest_opts.zo_pool, &spa, FTAG));
5631 spa->spa_debug = B_TRUE;
5632 metaslab_preload_limit = ztest_random(20) + 1;
5635 dmu_objset_stats_t dds;
5636 VERIFY0(dmu_objset_own(ztest_opts.zo_pool,
5637 DMU_OST_ANY, B_TRUE, FTAG, &os));
5638 dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
5639 dmu_objset_fast_stat(os, &dds);
5640 dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
5641 zs->zs_guid = dds.dds_guid;
5642 dmu_objset_disown(os, FTAG);
5644 spa->spa_dedup_ditto = 2 * ZIO_DEDUPDITTO_MIN;
5647 * We don't expect the pool to suspend unless maxfaults == 0,
5648 * in which case ztest_fault_inject() temporarily takes away
5649 * the only valid replica.
5651 if (MAXFAULTS() == 0)
5652 spa->spa_failmode = ZIO_FAILURE_MODE_WAIT;
5654 spa->spa_failmode = ZIO_FAILURE_MODE_PANIC;
5657 * Create a thread to periodically resume suspended I/O.
5659 VERIFY(thr_create(0, 0, ztest_resume_thread, spa, THR_BOUND,
5663 * Create a deadman thread to abort() if we hang.
5665 VERIFY(thr_create(0, 0, ztest_deadman_thread, zs, THR_BOUND,
5669 * Verify that we can safely inquire about about any object,
5670 * whether it's allocated or not. To make it interesting,
5671 * we probe a 5-wide window around each power of two.
5672 * This hits all edge cases, including zero and the max.
5674 for (int t = 0; t < 64; t++) {
5675 for (int d = -5; d <= 5; d++) {
5676 error = dmu_object_info(spa->spa_meta_objset,
5677 (1ULL << t) + d, NULL);
5678 ASSERT(error == 0 || error == ENOENT ||
5684 * If we got any ENOSPC errors on the previous run, destroy something.
5686 if (zs->zs_enospc_count != 0) {
5687 int d = ztest_random(ztest_opts.zo_datasets);
5688 ztest_dataset_destroy(d);
5690 zs->zs_enospc_count = 0;
5692 tid = umem_zalloc(ztest_opts.zo_threads * sizeof (thread_t),
5695 if (ztest_opts.zo_verbose >= 4)
5696 (void) printf("starting main threads...\n");
5699 * Kick off all the tests that run in parallel.
5701 for (int t = 0; t < ztest_opts.zo_threads; t++) {
5702 if (t < ztest_opts.zo_datasets &&
5703 ztest_dataset_open(t) != 0)
5705 VERIFY(thr_create(0, 0, ztest_thread, (void *)(uintptr_t)t,
5706 THR_BOUND, &tid[t]) == 0);
5710 * Wait for all of the tests to complete. We go in reverse order
5711 * so we don't close datasets while threads are still using them.
5713 for (int t = ztest_opts.zo_threads - 1; t >= 0; t--) {
5714 VERIFY(thr_join(tid[t], NULL, NULL) == 0);
5715 if (t < ztest_opts.zo_datasets)
5716 ztest_dataset_close(t);
5719 txg_wait_synced(spa_get_dsl(spa), 0);
5721 zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(spa));
5722 zs->zs_space = metaslab_class_get_space(spa_normal_class(spa));
5723 zfs_dbgmsg_print(FTAG);
5725 umem_free(tid, ztest_opts.zo_threads * sizeof (thread_t));
5727 /* Kill the resume thread */
5728 ztest_exiting = B_TRUE;
5729 VERIFY(thr_join(resume_tid, NULL, NULL) == 0);
5733 * Right before closing the pool, kick off a bunch of async I/O;
5734 * spa_close() should wait for it to complete.
5736 for (uint64_t object = 1; object < 50; object++) {
5737 dmu_prefetch(spa->spa_meta_objset, object, 0, 0, 1ULL << 20,
5738 ZIO_PRIORITY_SYNC_READ);
5741 spa_close(spa, FTAG);
5744 * Verify that we can loop over all pools.
5746 mutex_enter(&spa_namespace_lock);
5747 for (spa = spa_next(NULL); spa != NULL; spa = spa_next(spa))
5748 if (ztest_opts.zo_verbose > 3)
5749 (void) printf("spa_next: found %s\n", spa_name(spa));
5750 mutex_exit(&spa_namespace_lock);
5753 * Verify that we can export the pool and reimport it under a
5756 if (ztest_random(2) == 0) {
5757 char name[ZFS_MAX_DATASET_NAME_LEN];
5758 (void) snprintf(name, sizeof (name), "%s_import",
5759 ztest_opts.zo_pool);
5760 ztest_spa_import_export(ztest_opts.zo_pool, name);
5761 ztest_spa_import_export(name, ztest_opts.zo_pool);
5766 list_destroy(&zcl.zcl_callbacks);
5768 (void) _mutex_destroy(&zcl.zcl_callbacks_lock);
5770 (void) rwlock_destroy(&ztest_name_lock);
5771 (void) _mutex_destroy(&ztest_vdev_lock);
5777 ztest_ds_t *zd = &ztest_ds[0];
5781 if (ztest_opts.zo_verbose >= 3)
5782 (void) printf("testing spa_freeze()...\n");
5784 kernel_init(FREAD | FWRITE);
5785 VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
5786 VERIFY3U(0, ==, ztest_dataset_open(0));
5787 spa->spa_debug = B_TRUE;
5791 * Force the first log block to be transactionally allocated.
5792 * We have to do this before we freeze the pool -- otherwise
5793 * the log chain won't be anchored.
5795 while (BP_IS_HOLE(&zd->zd_zilog->zl_header->zh_log)) {
5796 ztest_dmu_object_alloc_free(zd, 0);
5797 zil_commit(zd->zd_zilog, 0);
5800 txg_wait_synced(spa_get_dsl(spa), 0);
5803 * Freeze the pool. This stops spa_sync() from doing anything,
5804 * so that the only way to record changes from now on is the ZIL.
5809 * Because it is hard to predict how much space a write will actually
5810 * require beforehand, we leave ourselves some fudge space to write over
5813 uint64_t capacity = metaslab_class_get_space(spa_normal_class(spa)) / 2;
5816 * Run tests that generate log records but don't alter the pool config
5817 * or depend on DSL sync tasks (snapshots, objset create/destroy, etc).
5818 * We do a txg_wait_synced() after each iteration to force the txg
5819 * to increase well beyond the last synced value in the uberblock.
5820 * The ZIL should be OK with that.
5822 * Run a random number of times less than zo_maxloops and ensure we do
5823 * not run out of space on the pool.
5825 while (ztest_random(10) != 0 &&
5826 numloops++ < ztest_opts.zo_maxloops &&
5827 metaslab_class_get_alloc(spa_normal_class(spa)) < capacity) {
5829 ztest_od_init(&od, 0, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
5830 VERIFY0(ztest_object_init(zd, &od, sizeof (od), B_FALSE));
5831 ztest_io(zd, od.od_object,
5832 ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
5833 txg_wait_synced(spa_get_dsl(spa), 0);
5837 * Commit all of the changes we just generated.
5839 zil_commit(zd->zd_zilog, 0);
5840 txg_wait_synced(spa_get_dsl(spa), 0);
5843 * Close our dataset and close the pool.
5845 ztest_dataset_close(0);
5846 spa_close(spa, FTAG);
5850 * Open and close the pool and dataset to induce log replay.
5852 kernel_init(FREAD | FWRITE);
5853 VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
5854 ASSERT(spa_freeze_txg(spa) == UINT64_MAX);
5855 VERIFY3U(0, ==, ztest_dataset_open(0));
5856 ztest_dataset_close(0);
5858 spa->spa_debug = B_TRUE;
5860 txg_wait_synced(spa_get_dsl(spa), 0);
5861 ztest_reguid(NULL, 0);
5863 spa_close(spa, FTAG);
5868 print_time(hrtime_t t, char *timebuf)
5870 hrtime_t s = t / NANOSEC;
5871 hrtime_t m = s / 60;
5872 hrtime_t h = m / 60;
5873 hrtime_t d = h / 24;
5882 (void) sprintf(timebuf,
5883 "%llud%02lluh%02llum%02llus", d, h, m, s);
5885 (void) sprintf(timebuf, "%lluh%02llum%02llus", h, m, s);
5887 (void) sprintf(timebuf, "%llum%02llus", m, s);
5889 (void) sprintf(timebuf, "%llus", s);
5897 VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
5898 if (ztest_random(2) == 0)
5900 VERIFY(nvlist_add_uint64(props, "autoreplace", 1) == 0);
5906 * Create a storage pool with the given name and initial vdev size.
5907 * Then test spa_freeze() functionality.
5910 ztest_init(ztest_shared_t *zs)
5913 nvlist_t *nvroot, *props;
5915 VERIFY(_mutex_init(&ztest_vdev_lock, USYNC_THREAD, NULL) == 0);
5916 VERIFY(rwlock_init(&ztest_name_lock, USYNC_THREAD, NULL) == 0);
5918 kernel_init(FREAD | FWRITE);
5921 * Create the storage pool.
5923 (void) spa_destroy(ztest_opts.zo_pool);
5924 ztest_shared->zs_vdev_next_leaf = 0;
5926 zs->zs_mirrors = ztest_opts.zo_mirrors;
5927 nvroot = make_vdev_root(NULL, NULL, NULL, ztest_opts.zo_vdev_size, 0,
5928 0, ztest_opts.zo_raidz, zs->zs_mirrors, 1);
5929 props = make_random_props();
5930 for (int i = 0; i < SPA_FEATURES; i++) {
5932 (void) snprintf(buf, sizeof (buf), "feature@%s",
5933 spa_feature_table[i].fi_uname);
5934 VERIFY3U(0, ==, nvlist_add_uint64(props, buf, 0));
5936 VERIFY3U(0, ==, spa_create(ztest_opts.zo_pool, nvroot, props, NULL));
5937 nvlist_free(nvroot);
5940 VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
5941 zs->zs_metaslab_sz =
5942 1ULL << spa->spa_root_vdev->vdev_child[0]->vdev_ms_shift;
5944 spa_close(spa, FTAG);
5948 ztest_run_zdb(ztest_opts.zo_pool);
5952 ztest_run_zdb(ztest_opts.zo_pool);
5954 (void) rwlock_destroy(&ztest_name_lock);
5955 (void) _mutex_destroy(&ztest_vdev_lock);
5961 static char ztest_name_data[] = "/tmp/ztest.data.XXXXXX";
5963 ztest_fd_data = mkstemp(ztest_name_data);
5964 ASSERT3S(ztest_fd_data, >=, 0);
5965 (void) unlink(ztest_name_data);
5970 shared_data_size(ztest_shared_hdr_t *hdr)
5974 size = hdr->zh_hdr_size;
5975 size += hdr->zh_opts_size;
5976 size += hdr->zh_size;
5977 size += hdr->zh_stats_size * hdr->zh_stats_count;
5978 size += hdr->zh_ds_size * hdr->zh_ds_count;
5987 ztest_shared_hdr_t *hdr;
5989 hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
5990 PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
5991 ASSERT(hdr != MAP_FAILED);
5993 VERIFY3U(0, ==, ftruncate(ztest_fd_data, sizeof (ztest_shared_hdr_t)));
5995 hdr->zh_hdr_size = sizeof (ztest_shared_hdr_t);
5996 hdr->zh_opts_size = sizeof (ztest_shared_opts_t);
5997 hdr->zh_size = sizeof (ztest_shared_t);
5998 hdr->zh_stats_size = sizeof (ztest_shared_callstate_t);
5999 hdr->zh_stats_count = ZTEST_FUNCS;
6000 hdr->zh_ds_size = sizeof (ztest_shared_ds_t);
6001 hdr->zh_ds_count = ztest_opts.zo_datasets;
6003 size = shared_data_size(hdr);
6004 VERIFY3U(0, ==, ftruncate(ztest_fd_data, size));
6006 (void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
6013 ztest_shared_hdr_t *hdr;
6016 hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
6017 PROT_READ, MAP_SHARED, ztest_fd_data, 0);
6018 ASSERT(hdr != MAP_FAILED);
6020 size = shared_data_size(hdr);
6022 (void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
6023 hdr = ztest_shared_hdr = (void *)mmap(0, P2ROUNDUP(size, getpagesize()),
6024 PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
6025 ASSERT(hdr != MAP_FAILED);
6026 buf = (uint8_t *)hdr;
6028 offset = hdr->zh_hdr_size;
6029 ztest_shared_opts = (void *)&buf[offset];
6030 offset += hdr->zh_opts_size;
6031 ztest_shared = (void *)&buf[offset];
6032 offset += hdr->zh_size;
6033 ztest_shared_callstate = (void *)&buf[offset];
6034 offset += hdr->zh_stats_size * hdr->zh_stats_count;
6035 ztest_shared_ds = (void *)&buf[offset];
6039 exec_child(char *cmd, char *libpath, boolean_t ignorekill, int *statusp)
6043 char *cmdbuf = NULL;
6048 cmdbuf = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
6049 (void) strlcpy(cmdbuf, getexecname(), MAXPATHLEN);
6054 fatal(1, "fork failed");
6056 if (pid == 0) { /* child */
6057 char *emptyargv[2] = { cmd, NULL };
6058 char fd_data_str[12];
6060 struct rlimit rl = { 1024, 1024 };
6061 (void) setrlimit(RLIMIT_NOFILE, &rl);
6063 (void) close(ztest_fd_rand);
6065 snprintf(fd_data_str, 12, "%d", ztest_fd_data));
6066 VERIFY0(setenv("ZTEST_FD_DATA", fd_data_str, 1));
6068 (void) enable_extended_FILE_stdio(-1, -1);
6069 if (libpath != NULL)
6070 VERIFY(0 == setenv("LD_LIBRARY_PATH", libpath, 1));
6072 (void) execv(cmd, emptyargv);
6074 (void) execvp(cmd, emptyargv);
6076 ztest_dump_core = B_FALSE;
6077 fatal(B_TRUE, "exec failed: %s", cmd);
6080 if (cmdbuf != NULL) {
6081 umem_free(cmdbuf, MAXPATHLEN);
6085 while (waitpid(pid, &status, 0) != pid)
6087 if (statusp != NULL)
6090 if (WIFEXITED(status)) {
6091 if (WEXITSTATUS(status) != 0) {
6092 (void) fprintf(stderr, "child exited with code %d\n",
6093 WEXITSTATUS(status));
6097 } else if (WIFSIGNALED(status)) {
6098 if (!ignorekill || WTERMSIG(status) != SIGKILL) {
6099 (void) fprintf(stderr, "child died with signal %d\n",
6105 (void) fprintf(stderr, "something strange happened to child\n");
6112 ztest_run_init(void)
6114 ztest_shared_t *zs = ztest_shared;
6116 ASSERT(ztest_opts.zo_init != 0);
6119 * Blow away any existing copy of zpool.cache
6121 (void) remove(spa_config_path);
6124 * Create and initialize our storage pool.
6126 for (int i = 1; i <= ztest_opts.zo_init; i++) {
6127 bzero(zs, sizeof (ztest_shared_t));
6128 if (ztest_opts.zo_verbose >= 3 &&
6129 ztest_opts.zo_init != 1) {
6130 (void) printf("ztest_init(), pass %d\n", i);
6137 main(int argc, char **argv)
6145 ztest_shared_callstate_t *zc;
6151 char *fd_data_str = getenv("ZTEST_FD_DATA");
6153 (void) setvbuf(stdout, NULL, _IOLBF, 0);
6155 dprintf_setup(&argc, argv);
6156 zfs_deadman_synctime_ms = 300000;
6158 ztest_fd_rand = open("/dev/urandom", O_RDONLY);
6159 ASSERT3S(ztest_fd_rand, >=, 0);
6162 process_options(argc, argv);
6167 bcopy(&ztest_opts, ztest_shared_opts,
6168 sizeof (*ztest_shared_opts));
6170 ztest_fd_data = atoi(fd_data_str);
6172 bcopy(ztest_shared_opts, &ztest_opts, sizeof (ztest_opts));
6174 ASSERT3U(ztest_opts.zo_datasets, ==, ztest_shared_hdr->zh_ds_count);
6176 /* Override location of zpool.cache */
6177 VERIFY3U(asprintf((char **)&spa_config_path, "%s/zpool.cache",
6178 ztest_opts.zo_dir), !=, -1);
6180 ztest_ds = umem_alloc(ztest_opts.zo_datasets * sizeof (ztest_ds_t),
6185 metaslab_gang_bang = ztest_opts.zo_metaslab_gang_bang;
6186 metaslab_df_alloc_threshold =
6187 zs->zs_metaslab_df_alloc_threshold;
6196 hasalt = (strlen(ztest_opts.zo_alt_ztest) != 0);
6198 if (ztest_opts.zo_verbose >= 1) {
6199 (void) printf("%llu vdevs, %d datasets, %d threads,"
6200 " %llu seconds...\n",
6201 (u_longlong_t)ztest_opts.zo_vdevs,
6202 ztest_opts.zo_datasets,
6203 ztest_opts.zo_threads,
6204 (u_longlong_t)ztest_opts.zo_time);
6207 cmd = umem_alloc(MAXNAMELEN, UMEM_NOFAIL);
6208 (void) strlcpy(cmd, getexecname(), MAXNAMELEN);
6210 zs->zs_do_init = B_TRUE;
6211 if (strlen(ztest_opts.zo_alt_ztest) != 0) {
6212 if (ztest_opts.zo_verbose >= 1) {
6213 (void) printf("Executing older ztest for "
6214 "initialization: %s\n", ztest_opts.zo_alt_ztest);
6216 VERIFY(!exec_child(ztest_opts.zo_alt_ztest,
6217 ztest_opts.zo_alt_libpath, B_FALSE, NULL));
6219 VERIFY(!exec_child(NULL, NULL, B_FALSE, NULL));
6221 zs->zs_do_init = B_FALSE;
6223 zs->zs_proc_start = gethrtime();
6224 zs->zs_proc_stop = zs->zs_proc_start + ztest_opts.zo_time * NANOSEC;
6226 for (int f = 0; f < ZTEST_FUNCS; f++) {
6227 zi = &ztest_info[f];
6228 zc = ZTEST_GET_SHARED_CALLSTATE(f);
6229 if (zs->zs_proc_start + zi->zi_interval[0] > zs->zs_proc_stop)
6230 zc->zc_next = UINT64_MAX;
6232 zc->zc_next = zs->zs_proc_start +
6233 ztest_random(2 * zi->zi_interval[0] + 1);
6237 * Run the tests in a loop. These tests include fault injection
6238 * to verify that self-healing data works, and forced crashes
6239 * to verify that we never lose on-disk consistency.
6241 while (gethrtime() < zs->zs_proc_stop) {
6246 * Initialize the workload counters for each function.
6248 for (int f = 0; f < ZTEST_FUNCS; f++) {
6249 zc = ZTEST_GET_SHARED_CALLSTATE(f);
6254 /* Set the allocation switch size */
6255 zs->zs_metaslab_df_alloc_threshold =
6256 ztest_random(zs->zs_metaslab_sz / 4) + 1;
6258 if (!hasalt || ztest_random(2) == 0) {
6259 if (hasalt && ztest_opts.zo_verbose >= 1) {
6260 (void) printf("Executing newer ztest: %s\n",
6264 killed = exec_child(cmd, NULL, B_TRUE, &status);
6266 if (hasalt && ztest_opts.zo_verbose >= 1) {
6267 (void) printf("Executing older ztest: %s\n",
6268 ztest_opts.zo_alt_ztest);
6271 killed = exec_child(ztest_opts.zo_alt_ztest,
6272 ztest_opts.zo_alt_libpath, B_TRUE, &status);
6279 if (ztest_opts.zo_verbose >= 1) {
6280 hrtime_t now = gethrtime();
6282 now = MIN(now, zs->zs_proc_stop);
6283 print_time(zs->zs_proc_stop - now, timebuf);
6284 nicenum(zs->zs_space, numbuf);
6286 (void) printf("Pass %3d, %8s, %3llu ENOSPC, "
6287 "%4.1f%% of %5s used, %3.0f%% done, %8s to go\n",
6289 WIFEXITED(status) ? "Complete" : "SIGKILL",
6290 (u_longlong_t)zs->zs_enospc_count,
6291 100.0 * zs->zs_alloc / zs->zs_space,
6293 100.0 * (now - zs->zs_proc_start) /
6294 (ztest_opts.zo_time * NANOSEC), timebuf);
6297 if (ztest_opts.zo_verbose >= 2) {
6298 (void) printf("\nWorkload summary:\n\n");
6299 (void) printf("%7s %9s %s\n",
6300 "Calls", "Time", "Function");
6301 (void) printf("%7s %9s %s\n",
6302 "-----", "----", "--------");
6303 for (int f = 0; f < ZTEST_FUNCS; f++) {
6306 zi = &ztest_info[f];
6307 zc = ZTEST_GET_SHARED_CALLSTATE(f);
6308 print_time(zc->zc_time, timebuf);
6309 (void) dladdr((void *)zi->zi_func, &dli);
6310 (void) printf("%7llu %9s %s\n",
6311 (u_longlong_t)zc->zc_count, timebuf,
6314 (void) printf("\n");
6318 * It's possible that we killed a child during a rename test,
6319 * in which case we'll have a 'ztest_tmp' pool lying around
6320 * instead of 'ztest'. Do a blind rename in case this happened.
6323 if (spa_open(ztest_opts.zo_pool, &spa, FTAG) == 0) {
6324 spa_close(spa, FTAG);
6326 char tmpname[ZFS_MAX_DATASET_NAME_LEN];
6328 kernel_init(FREAD | FWRITE);
6329 (void) snprintf(tmpname, sizeof (tmpname), "%s_tmp",
6330 ztest_opts.zo_pool);
6331 (void) spa_rename(tmpname, ztest_opts.zo_pool);
6335 ztest_run_zdb(ztest_opts.zo_pool);
6338 if (ztest_opts.zo_verbose >= 1) {
6340 (void) printf("%d runs of older ztest: %s\n", older,
6341 ztest_opts.zo_alt_ztest);
6342 (void) printf("%d runs of newer ztest: %s\n", newer,
6345 (void) printf("%d killed, %d completed, %.0f%% kill rate\n",
6346 kills, iters - kills, (100.0 * kills) / MAX(1, iters));
6349 umem_free(cmd, MAXNAMELEN);