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, 2015 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.
30 * The objective of this program is to provide a DMU/ZAP/SPA stress test
31 * that runs entirely in userland, is easy to use, and easy to extend.
33 * The overall design of the ztest program is as follows:
35 * (1) For each major functional area (e.g. adding vdevs to a pool,
36 * creating and destroying datasets, reading and writing objects, etc)
37 * we have a simple routine to test that functionality. These
38 * individual routines do not have to do anything "stressful".
40 * (2) We turn these simple functionality tests into a stress test by
41 * running them all in parallel, with as many threads as desired,
42 * and spread across as many datasets, objects, and vdevs as desired.
44 * (3) While all this is happening, we inject faults into the pool to
45 * verify that self-healing data really works.
47 * (4) Every time we open a dataset, we change its checksum and compression
48 * functions. Thus even individual objects vary from block to block
49 * in which checksum they use and whether they're compressed.
51 * (5) To verify that we never lose on-disk consistency after a crash,
52 * we run the entire test in a child of the main process.
53 * At random times, the child self-immolates with a SIGKILL.
54 * This is the software equivalent of pulling the power cord.
55 * The parent then runs the test again, using the existing
56 * storage pool, as many times as desired. If backwards compatibility
57 * testing is enabled ztest will sometimes run the "older" version
58 * of ztest after a SIGKILL.
60 * (6) To verify that we don't have future leaks or temporal incursions,
61 * many of the functional tests record the transaction group number
62 * as part of their data. When reading old data, they verify that
63 * the transaction group number is less than the current, open txg.
64 * If you add a new test, please do this if applicable.
66 * When run with no arguments, ztest runs for about five minutes and
67 * produces no output if successful. To get a little bit of information,
68 * specify -V. To get more information, specify -VV, and so on.
70 * To turn this into an overnight stress test, use -T to specify run time.
72 * You can ask more more vdevs [-v], datasets [-d], or threads [-t]
73 * to increase the pool capacity, fanout, and overall stress level.
75 * Use the -k option to set the desired frequency of kills.
77 * When ztest invokes itself it passes all relevant information through a
78 * temporary file which is mmap-ed in the child process. This allows shared
79 * memory to survive the exec syscall. The ztest_shared_hdr_t struct is always
80 * stored at offset 0 of this file and contains information on the size and
81 * number of shared structures in the file. The information stored in this file
82 * must remain backwards compatible with older versions of ztest so that
83 * ztest can invoke them during backwards compatibility testing (-B).
86 #include <sys/zfs_context.h>
92 #include <sys/dmu_objset.h>
98 #include <sys/resource.h>
101 #include <sys/zil_impl.h>
102 #include <sys/vdev_impl.h>
103 #include <sys/vdev_file.h>
104 #include <sys/spa_impl.h>
105 #include <sys/metaslab_impl.h>
106 #include <sys/dsl_prop.h>
107 #include <sys/dsl_dataset.h>
108 #include <sys/dsl_destroy.h>
109 #include <sys/dsl_scan.h>
110 #include <sys/zio_checksum.h>
111 #include <sys/refcount.h>
112 #include <sys/zfeature.h>
113 #include <sys/dsl_userhold.h>
115 #include <stdio_ext.h>
124 #include <sys/fs/zfs.h>
125 #include <libnvpair.h>
127 static int ztest_fd_data = -1;
128 static int ztest_fd_rand = -1;
130 typedef struct ztest_shared_hdr {
131 uint64_t zh_hdr_size;
132 uint64_t zh_opts_size;
134 uint64_t zh_stats_size;
135 uint64_t zh_stats_count;
137 uint64_t zh_ds_count;
138 } ztest_shared_hdr_t;
140 static ztest_shared_hdr_t *ztest_shared_hdr;
142 typedef struct ztest_shared_opts {
143 char zo_pool[MAXNAMELEN];
144 char zo_dir[MAXNAMELEN];
145 char zo_alt_ztest[MAXNAMELEN];
146 char zo_alt_libpath[MAXNAMELEN];
148 uint64_t zo_vdevtime;
156 uint64_t zo_passtime;
157 uint64_t zo_killrate;
161 uint64_t zo_maxloops;
162 uint64_t zo_metaslab_gang_bang;
163 } ztest_shared_opts_t;
165 static const ztest_shared_opts_t ztest_opts_defaults = {
166 .zo_pool = { 'z', 't', 'e', 's', 't', '\0' },
167 .zo_dir = { '/', 't', 'm', 'p', '\0' },
168 .zo_alt_ztest = { '\0' },
169 .zo_alt_libpath = { '\0' },
171 .zo_ashift = SPA_MINBLOCKSHIFT,
174 .zo_raidz_parity = 1,
175 .zo_vdev_size = SPA_MINDEVSIZE * 2,
178 .zo_passtime = 60, /* 60 seconds */
179 .zo_killrate = 70, /* 70% kill rate */
182 .zo_time = 300, /* 5 minutes */
183 .zo_maxloops = 50, /* max loops during spa_freeze() */
184 .zo_metaslab_gang_bang = 32 << 10
187 extern uint64_t metaslab_gang_bang;
188 extern uint64_t metaslab_df_alloc_threshold;
189 extern uint64_t zfs_deadman_synctime_ms;
190 extern int metaslab_preload_limit;
192 static ztest_shared_opts_t *ztest_shared_opts;
193 static ztest_shared_opts_t ztest_opts;
195 typedef struct ztest_shared_ds {
199 static ztest_shared_ds_t *ztest_shared_ds;
200 #define ZTEST_GET_SHARED_DS(d) (&ztest_shared_ds[d])
202 #define BT_MAGIC 0x123456789abcdefULL
203 #define MAXFAULTS() \
204 (MAX(zs->zs_mirrors, 1) * (ztest_opts.zo_raidz_parity + 1) - 1)
208 ZTEST_IO_WRITE_PATTERN,
209 ZTEST_IO_WRITE_ZEROES,
216 typedef struct ztest_block_tag {
226 typedef struct bufwad {
233 * XXX -- fix zfs range locks to be generic so we can use them here.
255 #define ZTEST_RANGE_LOCKS 64
256 #define ZTEST_OBJECT_LOCKS 64
259 * Object descriptor. Used as a template for object lookup/create/remove.
261 typedef struct ztest_od {
264 dmu_object_type_t od_type;
265 dmu_object_type_t od_crtype;
266 uint64_t od_blocksize;
267 uint64_t od_crblocksize;
270 char od_name[MAXNAMELEN];
276 typedef struct ztest_ds {
277 ztest_shared_ds_t *zd_shared;
279 rwlock_t zd_zilog_lock;
281 ztest_od_t *zd_od; /* debugging aid */
282 char zd_name[MAXNAMELEN];
283 mutex_t zd_dirobj_lock;
284 rll_t zd_object_lock[ZTEST_OBJECT_LOCKS];
285 rll_t zd_range_lock[ZTEST_RANGE_LOCKS];
289 * Per-iteration state.
291 typedef void ztest_func_t(ztest_ds_t *zd, uint64_t id);
293 typedef struct ztest_info {
294 ztest_func_t *zi_func; /* test function */
295 uint64_t zi_iters; /* iterations per execution */
296 uint64_t *zi_interval; /* execute every <interval> seconds */
299 typedef struct ztest_shared_callstate {
300 uint64_t zc_count; /* per-pass count */
301 uint64_t zc_time; /* per-pass time */
302 uint64_t zc_next; /* next time to call this function */
303 } ztest_shared_callstate_t;
305 static ztest_shared_callstate_t *ztest_shared_callstate;
306 #define ZTEST_GET_SHARED_CALLSTATE(c) (&ztest_shared_callstate[c])
309 * Note: these aren't static because we want dladdr() to work.
311 ztest_func_t ztest_dmu_read_write;
312 ztest_func_t ztest_dmu_write_parallel;
313 ztest_func_t ztest_dmu_object_alloc_free;
314 ztest_func_t ztest_dmu_commit_callbacks;
315 ztest_func_t ztest_zap;
316 ztest_func_t ztest_zap_parallel;
317 ztest_func_t ztest_zil_commit;
318 ztest_func_t ztest_zil_remount;
319 ztest_func_t ztest_dmu_read_write_zcopy;
320 ztest_func_t ztest_dmu_objset_create_destroy;
321 ztest_func_t ztest_dmu_prealloc;
322 ztest_func_t ztest_fzap;
323 ztest_func_t ztest_dmu_snapshot_create_destroy;
324 ztest_func_t ztest_dsl_prop_get_set;
325 ztest_func_t ztest_spa_prop_get_set;
326 ztest_func_t ztest_spa_create_destroy;
327 ztest_func_t ztest_fault_inject;
328 ztest_func_t ztest_ddt_repair;
329 ztest_func_t ztest_dmu_snapshot_hold;
330 ztest_func_t ztest_spa_rename;
331 ztest_func_t ztest_scrub;
332 ztest_func_t ztest_dsl_dataset_promote_busy;
333 ztest_func_t ztest_vdev_attach_detach;
334 ztest_func_t ztest_vdev_LUN_growth;
335 ztest_func_t ztest_vdev_add_remove;
336 ztest_func_t ztest_vdev_aux_add_remove;
337 ztest_func_t ztest_split_pool;
338 ztest_func_t ztest_reguid;
339 ztest_func_t ztest_spa_upgrade;
341 uint64_t zopt_always = 0ULL * NANOSEC; /* all the time */
342 uint64_t zopt_incessant = 1ULL * NANOSEC / 10; /* every 1/10 second */
343 uint64_t zopt_often = 1ULL * NANOSEC; /* every second */
344 uint64_t zopt_sometimes = 10ULL * NANOSEC; /* every 10 seconds */
345 uint64_t zopt_rarely = 60ULL * NANOSEC; /* every 60 seconds */
347 ztest_info_t ztest_info[] = {
348 { ztest_dmu_read_write, 1, &zopt_always },
349 { ztest_dmu_write_parallel, 10, &zopt_always },
350 { ztest_dmu_object_alloc_free, 1, &zopt_always },
351 { ztest_dmu_commit_callbacks, 1, &zopt_always },
352 { ztest_zap, 30, &zopt_always },
353 { ztest_zap_parallel, 100, &zopt_always },
354 { ztest_split_pool, 1, &zopt_always },
355 { ztest_zil_commit, 1, &zopt_incessant },
356 { ztest_zil_remount, 1, &zopt_sometimes },
357 { ztest_dmu_read_write_zcopy, 1, &zopt_often },
358 { ztest_dmu_objset_create_destroy, 1, &zopt_often },
359 { ztest_dsl_prop_get_set, 1, &zopt_often },
360 { ztest_spa_prop_get_set, 1, &zopt_sometimes },
362 { ztest_dmu_prealloc, 1, &zopt_sometimes },
364 { ztest_fzap, 1, &zopt_sometimes },
365 { ztest_dmu_snapshot_create_destroy, 1, &zopt_sometimes },
366 { ztest_spa_create_destroy, 1, &zopt_sometimes },
367 { ztest_fault_inject, 1, &zopt_sometimes },
368 { ztest_ddt_repair, 1, &zopt_sometimes },
369 { ztest_dmu_snapshot_hold, 1, &zopt_sometimes },
370 { ztest_reguid, 1, &zopt_rarely },
371 { ztest_spa_rename, 1, &zopt_rarely },
372 { ztest_scrub, 1, &zopt_rarely },
373 { ztest_spa_upgrade, 1, &zopt_rarely },
374 { ztest_dsl_dataset_promote_busy, 1, &zopt_rarely },
375 { ztest_vdev_attach_detach, 1, &zopt_sometimes },
376 { ztest_vdev_LUN_growth, 1, &zopt_rarely },
377 { ztest_vdev_add_remove, 1,
378 &ztest_opts.zo_vdevtime },
379 { ztest_vdev_aux_add_remove, 1,
380 &ztest_opts.zo_vdevtime },
383 #define ZTEST_FUNCS (sizeof (ztest_info) / sizeof (ztest_info_t))
386 * The following struct is used to hold a list of uncalled commit callbacks.
387 * The callbacks are ordered by txg number.
389 typedef struct ztest_cb_list {
390 mutex_t zcl_callbacks_lock;
391 list_t zcl_callbacks;
395 * Stuff we need to share writably between parent and child.
397 typedef struct ztest_shared {
398 boolean_t zs_do_init;
399 hrtime_t zs_proc_start;
400 hrtime_t zs_proc_stop;
401 hrtime_t zs_thread_start;
402 hrtime_t zs_thread_stop;
403 hrtime_t zs_thread_kill;
404 uint64_t zs_enospc_count;
405 uint64_t zs_vdev_next_leaf;
406 uint64_t zs_vdev_aux;
411 uint64_t zs_metaslab_sz;
412 uint64_t zs_metaslab_df_alloc_threshold;
416 #define ID_PARALLEL -1ULL
418 static char ztest_dev_template[] = "%s/%s.%llua";
419 static char ztest_aux_template[] = "%s/%s.%s.%llu";
420 ztest_shared_t *ztest_shared;
422 static spa_t *ztest_spa = NULL;
423 static ztest_ds_t *ztest_ds;
425 static mutex_t ztest_vdev_lock;
428 * The ztest_name_lock protects the pool and dataset namespace used by
429 * the individual tests. To modify the namespace, consumers must grab
430 * this lock as writer. Grabbing the lock as reader will ensure that the
431 * namespace does not change while the lock is held.
433 static rwlock_t ztest_name_lock;
435 static boolean_t ztest_dump_core = B_TRUE;
436 static boolean_t ztest_exiting;
438 /* Global commit callback list */
439 static ztest_cb_list_t zcl;
442 ZTEST_META_DNODE = 0,
447 static void usage(boolean_t) __NORETURN;
450 * These libumem hooks provide a reasonable set of defaults for the allocator's
451 * debugging facilities.
456 return ("default,verbose"); /* $UMEM_DEBUG setting */
460 _umem_logging_init(void)
462 return ("fail,contents"); /* $UMEM_LOGGING setting */
465 #define FATAL_MSG_SZ 1024
470 fatal(int do_perror, char *message, ...)
473 int save_errno = errno;
474 char buf[FATAL_MSG_SZ];
476 (void) fflush(stdout);
478 va_start(args, message);
479 (void) sprintf(buf, "ztest: ");
481 (void) vsprintf(buf + strlen(buf), message, args);
484 (void) snprintf(buf + strlen(buf), FATAL_MSG_SZ - strlen(buf),
485 ": %s", strerror(save_errno));
487 (void) fprintf(stderr, "%s\n", buf);
488 fatal_msg = buf; /* to ease debugging */
495 str2shift(const char *buf)
497 const char *ends = "BKMGTPEZ";
502 for (i = 0; i < strlen(ends); i++) {
503 if (toupper(buf[0]) == ends[i])
506 if (i == strlen(ends)) {
507 (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n",
511 if (buf[1] == '\0' || (toupper(buf[1]) == 'B' && buf[2] == '\0')) {
514 (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n", buf);
520 nicenumtoull(const char *buf)
525 val = strtoull(buf, &end, 0);
527 (void) fprintf(stderr, "ztest: bad numeric value: %s\n", buf);
529 } else if (end[0] == '.') {
530 double fval = strtod(buf, &end);
531 fval *= pow(2, str2shift(end));
532 if (fval > UINT64_MAX) {
533 (void) fprintf(stderr, "ztest: value too large: %s\n",
537 val = (uint64_t)fval;
539 int shift = str2shift(end);
540 if (shift >= 64 || (val << shift) >> shift != val) {
541 (void) fprintf(stderr, "ztest: value too large: %s\n",
551 usage(boolean_t requested)
553 const ztest_shared_opts_t *zo = &ztest_opts_defaults;
555 char nice_vdev_size[10];
556 char nice_gang_bang[10];
557 FILE *fp = requested ? stdout : stderr;
559 nicenum(zo->zo_vdev_size, nice_vdev_size);
560 nicenum(zo->zo_metaslab_gang_bang, nice_gang_bang);
562 (void) fprintf(fp, "Usage: %s\n"
563 "\t[-v vdevs (default: %llu)]\n"
564 "\t[-s size_of_each_vdev (default: %s)]\n"
565 "\t[-a alignment_shift (default: %d)] use 0 for random\n"
566 "\t[-m mirror_copies (default: %d)]\n"
567 "\t[-r raidz_disks (default: %d)]\n"
568 "\t[-R raidz_parity (default: %d)]\n"
569 "\t[-d datasets (default: %d)]\n"
570 "\t[-t threads (default: %d)]\n"
571 "\t[-g gang_block_threshold (default: %s)]\n"
572 "\t[-i init_count (default: %d)] initialize pool i times\n"
573 "\t[-k kill_percentage (default: %llu%%)]\n"
574 "\t[-p pool_name (default: %s)]\n"
575 "\t[-f dir (default: %s)] file directory for vdev files\n"
576 "\t[-V] verbose (use multiple times for ever more blather)\n"
577 "\t[-E] use existing pool instead of creating new one\n"
578 "\t[-T time (default: %llu sec)] total run time\n"
579 "\t[-F freezeloops (default: %llu)] max loops in spa_freeze()\n"
580 "\t[-P passtime (default: %llu sec)] time per pass\n"
581 "\t[-B alt_ztest (default: <none>)] alternate ztest path\n"
582 "\t[-h] (print help)\n"
585 (u_longlong_t)zo->zo_vdevs, /* -v */
586 nice_vdev_size, /* -s */
587 zo->zo_ashift, /* -a */
588 zo->zo_mirrors, /* -m */
589 zo->zo_raidz, /* -r */
590 zo->zo_raidz_parity, /* -R */
591 zo->zo_datasets, /* -d */
592 zo->zo_threads, /* -t */
593 nice_gang_bang, /* -g */
594 zo->zo_init, /* -i */
595 (u_longlong_t)zo->zo_killrate, /* -k */
596 zo->zo_pool, /* -p */
598 (u_longlong_t)zo->zo_time, /* -T */
599 (u_longlong_t)zo->zo_maxloops, /* -F */
600 (u_longlong_t)zo->zo_passtime);
601 exit(requested ? 0 : 1);
605 process_options(int argc, char **argv)
608 ztest_shared_opts_t *zo = &ztest_opts;
612 char altdir[MAXNAMELEN] = { 0 };
614 bcopy(&ztest_opts_defaults, zo, sizeof (*zo));
616 while ((opt = getopt(argc, argv,
617 "v:s:a:m:r:R:d:t:g:i:k:p:f:VET:P:hF:B:")) != EOF) {
634 value = nicenumtoull(optarg);
638 zo->zo_vdevs = value;
641 zo->zo_vdev_size = MAX(SPA_MINDEVSIZE, value);
644 zo->zo_ashift = value;
647 zo->zo_mirrors = value;
650 zo->zo_raidz = MAX(1, value);
653 zo->zo_raidz_parity = MIN(MAX(value, 1), 3);
656 zo->zo_datasets = MAX(1, value);
659 zo->zo_threads = MAX(1, value);
662 zo->zo_metaslab_gang_bang = MAX(SPA_MINBLOCKSIZE << 1,
669 zo->zo_killrate = value;
672 (void) strlcpy(zo->zo_pool, optarg,
673 sizeof (zo->zo_pool));
676 path = realpath(optarg, NULL);
678 (void) fprintf(stderr, "error: %s: %s\n",
679 optarg, strerror(errno));
682 (void) strlcpy(zo->zo_dir, path,
683 sizeof (zo->zo_dir));
696 zo->zo_passtime = MAX(1, value);
699 zo->zo_maxloops = MAX(1, value);
702 (void) strlcpy(altdir, optarg, sizeof (altdir));
714 zo->zo_raidz_parity = MIN(zo->zo_raidz_parity, zo->zo_raidz - 1);
717 (zo->zo_vdevs > 0 ? zo->zo_time * NANOSEC / zo->zo_vdevs :
720 if (strlen(altdir) > 0) {
728 cmd = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
729 realaltdir = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
731 VERIFY(NULL != realpath(getexecname(), cmd));
732 if (0 != access(altdir, F_OK)) {
733 ztest_dump_core = B_FALSE;
734 fatal(B_TRUE, "invalid alternate ztest path: %s",
737 VERIFY(NULL != realpath(altdir, realaltdir));
740 * 'cmd' should be of the form "<anything>/usr/bin/<isa>/ztest".
741 * We want to extract <isa> to determine if we should use
742 * 32 or 64 bit binaries.
744 bin = strstr(cmd, "/usr/bin/");
745 ztest = strstr(bin, "/ztest");
747 isalen = ztest - isa;
748 (void) snprintf(zo->zo_alt_ztest, sizeof (zo->zo_alt_ztest),
749 "%s/usr/bin/%.*s/ztest", realaltdir, isalen, isa);
750 (void) snprintf(zo->zo_alt_libpath, sizeof (zo->zo_alt_libpath),
751 "%s/usr/lib/%.*s", realaltdir, isalen, isa);
753 if (0 != access(zo->zo_alt_ztest, X_OK)) {
754 ztest_dump_core = B_FALSE;
755 fatal(B_TRUE, "invalid alternate ztest: %s",
757 } else if (0 != access(zo->zo_alt_libpath, X_OK)) {
758 ztest_dump_core = B_FALSE;
759 fatal(B_TRUE, "invalid alternate lib directory %s",
763 umem_free(cmd, MAXPATHLEN);
764 umem_free(realaltdir, MAXPATHLEN);
769 ztest_kill(ztest_shared_t *zs)
771 zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(ztest_spa));
772 zs->zs_space = metaslab_class_get_space(spa_normal_class(ztest_spa));
775 * Before we kill off ztest, make sure that the config is updated.
776 * See comment above spa_config_sync().
778 mutex_enter(&spa_namespace_lock);
779 spa_config_sync(ztest_spa, B_FALSE, B_FALSE);
780 mutex_exit(&spa_namespace_lock);
782 zfs_dbgmsg_print(FTAG);
783 (void) kill(getpid(), SIGKILL);
787 ztest_random(uint64_t range)
791 ASSERT3S(ztest_fd_rand, >=, 0);
796 if (read(ztest_fd_rand, &r, sizeof (r)) != sizeof (r))
797 fatal(1, "short read from /dev/urandom");
804 ztest_record_enospc(const char *s)
806 ztest_shared->zs_enospc_count++;
810 ztest_get_ashift(void)
812 if (ztest_opts.zo_ashift == 0)
813 return (SPA_MINBLOCKSHIFT + ztest_random(5));
814 return (ztest_opts.zo_ashift);
818 make_vdev_file(char *path, char *aux, char *pool, size_t size, uint64_t ashift)
820 char pathbuf[MAXPATHLEN];
825 ashift = ztest_get_ashift();
831 vdev = ztest_shared->zs_vdev_aux;
832 (void) snprintf(path, sizeof (pathbuf),
833 ztest_aux_template, ztest_opts.zo_dir,
834 pool == NULL ? ztest_opts.zo_pool : pool,
837 vdev = ztest_shared->zs_vdev_next_leaf++;
838 (void) snprintf(path, sizeof (pathbuf),
839 ztest_dev_template, ztest_opts.zo_dir,
840 pool == NULL ? ztest_opts.zo_pool : pool, vdev);
845 int fd = open(path, O_RDWR | O_CREAT | O_TRUNC, 0666);
847 fatal(1, "can't open %s", path);
848 if (ftruncate(fd, size) != 0)
849 fatal(1, "can't ftruncate %s", path);
853 VERIFY(nvlist_alloc(&file, NV_UNIQUE_NAME, 0) == 0);
854 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_TYPE, VDEV_TYPE_FILE) == 0);
855 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_PATH, path) == 0);
856 VERIFY(nvlist_add_uint64(file, ZPOOL_CONFIG_ASHIFT, ashift) == 0);
862 make_vdev_raidz(char *path, char *aux, char *pool, size_t size,
863 uint64_t ashift, int r)
865 nvlist_t *raidz, **child;
869 return (make_vdev_file(path, aux, pool, size, ashift));
870 child = umem_alloc(r * sizeof (nvlist_t *), UMEM_NOFAIL);
872 for (c = 0; c < r; c++)
873 child[c] = make_vdev_file(path, aux, pool, size, ashift);
875 VERIFY(nvlist_alloc(&raidz, NV_UNIQUE_NAME, 0) == 0);
876 VERIFY(nvlist_add_string(raidz, ZPOOL_CONFIG_TYPE,
877 VDEV_TYPE_RAIDZ) == 0);
878 VERIFY(nvlist_add_uint64(raidz, ZPOOL_CONFIG_NPARITY,
879 ztest_opts.zo_raidz_parity) == 0);
880 VERIFY(nvlist_add_nvlist_array(raidz, ZPOOL_CONFIG_CHILDREN,
883 for (c = 0; c < r; c++)
884 nvlist_free(child[c]);
886 umem_free(child, r * sizeof (nvlist_t *));
892 make_vdev_mirror(char *path, char *aux, char *pool, size_t size,
893 uint64_t ashift, int r, int m)
895 nvlist_t *mirror, **child;
899 return (make_vdev_raidz(path, aux, pool, size, ashift, r));
901 child = umem_alloc(m * sizeof (nvlist_t *), UMEM_NOFAIL);
903 for (c = 0; c < m; c++)
904 child[c] = make_vdev_raidz(path, aux, pool, size, ashift, r);
906 VERIFY(nvlist_alloc(&mirror, NV_UNIQUE_NAME, 0) == 0);
907 VERIFY(nvlist_add_string(mirror, ZPOOL_CONFIG_TYPE,
908 VDEV_TYPE_MIRROR) == 0);
909 VERIFY(nvlist_add_nvlist_array(mirror, ZPOOL_CONFIG_CHILDREN,
912 for (c = 0; c < m; c++)
913 nvlist_free(child[c]);
915 umem_free(child, m * sizeof (nvlist_t *));
921 make_vdev_root(char *path, char *aux, char *pool, size_t size, uint64_t ashift,
922 int log, int r, int m, int t)
924 nvlist_t *root, **child;
929 child = umem_alloc(t * sizeof (nvlist_t *), UMEM_NOFAIL);
931 for (c = 0; c < t; c++) {
932 child[c] = make_vdev_mirror(path, aux, pool, size, ashift,
934 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_IS_LOG,
938 VERIFY(nvlist_alloc(&root, NV_UNIQUE_NAME, 0) == 0);
939 VERIFY(nvlist_add_string(root, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) == 0);
940 VERIFY(nvlist_add_nvlist_array(root, aux ? aux : ZPOOL_CONFIG_CHILDREN,
943 for (c = 0; c < t; c++)
944 nvlist_free(child[c]);
946 umem_free(child, t * sizeof (nvlist_t *));
952 * Find a random spa version. Returns back a random spa version in the
953 * range [initial_version, SPA_VERSION_FEATURES].
956 ztest_random_spa_version(uint64_t initial_version)
958 uint64_t version = initial_version;
960 if (version <= SPA_VERSION_BEFORE_FEATURES) {
962 ztest_random(SPA_VERSION_BEFORE_FEATURES - version + 1);
965 if (version > SPA_VERSION_BEFORE_FEATURES)
966 version = SPA_VERSION_FEATURES;
968 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
973 ztest_random_blocksize(void)
975 uint64_t block_shift;
977 * Choose a block size >= the ashift.
978 * If the SPA supports new MAXBLOCKSIZE, test up to 1MB blocks.
980 int maxbs = SPA_OLD_MAXBLOCKSHIFT;
981 if (spa_maxblocksize(ztest_spa) == SPA_MAXBLOCKSIZE)
983 block_shift = ztest_random(maxbs - ztest_spa->spa_max_ashift + 1);
984 return (1 << (SPA_MINBLOCKSHIFT + block_shift));
988 ztest_random_ibshift(void)
990 return (DN_MIN_INDBLKSHIFT +
991 ztest_random(DN_MAX_INDBLKSHIFT - DN_MIN_INDBLKSHIFT + 1));
995 ztest_random_vdev_top(spa_t *spa, boolean_t log_ok)
998 vdev_t *rvd = spa->spa_root_vdev;
1001 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1004 top = ztest_random(rvd->vdev_children);
1005 tvd = rvd->vdev_child[top];
1006 } while (tvd->vdev_ishole || (tvd->vdev_islog && !log_ok) ||
1007 tvd->vdev_mg == NULL || tvd->vdev_mg->mg_class == NULL);
1013 ztest_random_dsl_prop(zfs_prop_t prop)
1018 value = zfs_prop_random_value(prop, ztest_random(-1ULL));
1019 } while (prop == ZFS_PROP_CHECKSUM && value == ZIO_CHECKSUM_OFF);
1025 ztest_dsl_prop_set_uint64(char *osname, zfs_prop_t prop, uint64_t value,
1028 const char *propname = zfs_prop_to_name(prop);
1029 const char *valname;
1030 char setpoint[MAXPATHLEN];
1034 error = dsl_prop_set_int(osname, propname,
1035 (inherit ? ZPROP_SRC_NONE : ZPROP_SRC_LOCAL), value);
1037 if (error == ENOSPC) {
1038 ztest_record_enospc(FTAG);
1043 VERIFY0(dsl_prop_get_integer(osname, propname, &curval, setpoint));
1045 if (ztest_opts.zo_verbose >= 6) {
1046 VERIFY(zfs_prop_index_to_string(prop, curval, &valname) == 0);
1047 (void) printf("%s %s = %s at '%s'\n",
1048 osname, propname, valname, setpoint);
1055 ztest_spa_prop_set_uint64(zpool_prop_t prop, uint64_t value)
1057 spa_t *spa = ztest_spa;
1058 nvlist_t *props = NULL;
1061 VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
1062 VERIFY(nvlist_add_uint64(props, zpool_prop_to_name(prop), value) == 0);
1064 error = spa_prop_set(spa, props);
1068 if (error == ENOSPC) {
1069 ztest_record_enospc(FTAG);
1078 ztest_rll_init(rll_t *rll)
1080 rll->rll_writer = NULL;
1081 rll->rll_readers = 0;
1082 VERIFY(_mutex_init(&rll->rll_lock, USYNC_THREAD, NULL) == 0);
1083 VERIFY(cond_init(&rll->rll_cv, USYNC_THREAD, NULL) == 0);
1087 ztest_rll_destroy(rll_t *rll)
1089 ASSERT(rll->rll_writer == NULL);
1090 ASSERT(rll->rll_readers == 0);
1091 VERIFY(_mutex_destroy(&rll->rll_lock) == 0);
1092 VERIFY(cond_destroy(&rll->rll_cv) == 0);
1096 ztest_rll_lock(rll_t *rll, rl_type_t type)
1098 VERIFY(mutex_lock(&rll->rll_lock) == 0);
1100 if (type == RL_READER) {
1101 while (rll->rll_writer != NULL)
1102 (void) cond_wait(&rll->rll_cv, &rll->rll_lock);
1105 while (rll->rll_writer != NULL || rll->rll_readers)
1106 (void) cond_wait(&rll->rll_cv, &rll->rll_lock);
1107 rll->rll_writer = curthread;
1110 VERIFY(mutex_unlock(&rll->rll_lock) == 0);
1114 ztest_rll_unlock(rll_t *rll)
1116 VERIFY(mutex_lock(&rll->rll_lock) == 0);
1118 if (rll->rll_writer) {
1119 ASSERT(rll->rll_readers == 0);
1120 rll->rll_writer = NULL;
1122 ASSERT(rll->rll_readers != 0);
1123 ASSERT(rll->rll_writer == NULL);
1127 if (rll->rll_writer == NULL && rll->rll_readers == 0)
1128 VERIFY(cond_broadcast(&rll->rll_cv) == 0);
1130 VERIFY(mutex_unlock(&rll->rll_lock) == 0);
1134 ztest_object_lock(ztest_ds_t *zd, uint64_t object, rl_type_t type)
1136 rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1138 ztest_rll_lock(rll, type);
1142 ztest_object_unlock(ztest_ds_t *zd, uint64_t object)
1144 rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1146 ztest_rll_unlock(rll);
1150 ztest_range_lock(ztest_ds_t *zd, uint64_t object, uint64_t offset,
1151 uint64_t size, rl_type_t type)
1153 uint64_t hash = object ^ (offset % (ZTEST_RANGE_LOCKS + 1));
1154 rll_t *rll = &zd->zd_range_lock[hash & (ZTEST_RANGE_LOCKS - 1)];
1157 rl = umem_alloc(sizeof (*rl), UMEM_NOFAIL);
1158 rl->rl_object = object;
1159 rl->rl_offset = offset;
1163 ztest_rll_lock(rll, type);
1169 ztest_range_unlock(rl_t *rl)
1171 rll_t *rll = rl->rl_lock;
1173 ztest_rll_unlock(rll);
1175 umem_free(rl, sizeof (*rl));
1179 ztest_zd_init(ztest_ds_t *zd, ztest_shared_ds_t *szd, objset_t *os)
1182 zd->zd_zilog = dmu_objset_zil(os);
1183 zd->zd_shared = szd;
1184 dmu_objset_name(os, zd->zd_name);
1186 if (zd->zd_shared != NULL)
1187 zd->zd_shared->zd_seq = 0;
1189 VERIFY(rwlock_init(&zd->zd_zilog_lock, USYNC_THREAD, NULL) == 0);
1190 VERIFY(_mutex_init(&zd->zd_dirobj_lock, USYNC_THREAD, NULL) == 0);
1192 for (int l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1193 ztest_rll_init(&zd->zd_object_lock[l]);
1195 for (int l = 0; l < ZTEST_RANGE_LOCKS; l++)
1196 ztest_rll_init(&zd->zd_range_lock[l]);
1200 ztest_zd_fini(ztest_ds_t *zd)
1202 VERIFY(_mutex_destroy(&zd->zd_dirobj_lock) == 0);
1204 for (int l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1205 ztest_rll_destroy(&zd->zd_object_lock[l]);
1207 for (int l = 0; l < ZTEST_RANGE_LOCKS; l++)
1208 ztest_rll_destroy(&zd->zd_range_lock[l]);
1211 #define TXG_MIGHTWAIT (ztest_random(10) == 0 ? TXG_NOWAIT : TXG_WAIT)
1214 ztest_tx_assign(dmu_tx_t *tx, uint64_t txg_how, const char *tag)
1220 * Attempt to assign tx to some transaction group.
1222 error = dmu_tx_assign(tx, txg_how);
1224 if (error == ERESTART) {
1225 ASSERT(txg_how == TXG_NOWAIT);
1228 ASSERT3U(error, ==, ENOSPC);
1229 ztest_record_enospc(tag);
1234 txg = dmu_tx_get_txg(tx);
1240 ztest_pattern_set(void *buf, uint64_t size, uint64_t value)
1243 uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1250 ztest_pattern_match(void *buf, uint64_t size, uint64_t value)
1253 uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1257 diff |= (value - *ip++);
1263 ztest_bt_generate(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1264 uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg)
1266 bt->bt_magic = BT_MAGIC;
1267 bt->bt_objset = dmu_objset_id(os);
1268 bt->bt_object = object;
1269 bt->bt_offset = offset;
1272 bt->bt_crtxg = crtxg;
1276 ztest_bt_verify(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1277 uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg)
1279 ASSERT3U(bt->bt_magic, ==, BT_MAGIC);
1280 ASSERT3U(bt->bt_objset, ==, dmu_objset_id(os));
1281 ASSERT3U(bt->bt_object, ==, object);
1282 ASSERT3U(bt->bt_offset, ==, offset);
1283 ASSERT3U(bt->bt_gen, <=, gen);
1284 ASSERT3U(bt->bt_txg, <=, txg);
1285 ASSERT3U(bt->bt_crtxg, ==, crtxg);
1288 static ztest_block_tag_t *
1289 ztest_bt_bonus(dmu_buf_t *db)
1291 dmu_object_info_t doi;
1292 ztest_block_tag_t *bt;
1294 dmu_object_info_from_db(db, &doi);
1295 ASSERT3U(doi.doi_bonus_size, <=, db->db_size);
1296 ASSERT3U(doi.doi_bonus_size, >=, sizeof (*bt));
1297 bt = (void *)((char *)db->db_data + doi.doi_bonus_size - sizeof (*bt));
1306 #define lrz_type lr_mode
1307 #define lrz_blocksize lr_uid
1308 #define lrz_ibshift lr_gid
1309 #define lrz_bonustype lr_rdev
1310 #define lrz_bonuslen lr_crtime[1]
1313 ztest_log_create(ztest_ds_t *zd, dmu_tx_t *tx, lr_create_t *lr)
1315 char *name = (void *)(lr + 1); /* name follows lr */
1316 size_t namesize = strlen(name) + 1;
1319 if (zil_replaying(zd->zd_zilog, tx))
1322 itx = zil_itx_create(TX_CREATE, sizeof (*lr) + namesize);
1323 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1324 sizeof (*lr) + namesize - sizeof (lr_t));
1326 zil_itx_assign(zd->zd_zilog, itx, tx);
1330 ztest_log_remove(ztest_ds_t *zd, dmu_tx_t *tx, lr_remove_t *lr, uint64_t object)
1332 char *name = (void *)(lr + 1); /* name follows lr */
1333 size_t namesize = strlen(name) + 1;
1336 if (zil_replaying(zd->zd_zilog, tx))
1339 itx = zil_itx_create(TX_REMOVE, sizeof (*lr) + namesize);
1340 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1341 sizeof (*lr) + namesize - sizeof (lr_t));
1343 itx->itx_oid = object;
1344 zil_itx_assign(zd->zd_zilog, itx, tx);
1348 ztest_log_write(ztest_ds_t *zd, dmu_tx_t *tx, lr_write_t *lr)
1351 itx_wr_state_t write_state = ztest_random(WR_NUM_STATES);
1353 if (zil_replaying(zd->zd_zilog, tx))
1356 if (lr->lr_length > ZIL_MAX_LOG_DATA)
1357 write_state = WR_INDIRECT;
1359 itx = zil_itx_create(TX_WRITE,
1360 sizeof (*lr) + (write_state == WR_COPIED ? lr->lr_length : 0));
1362 if (write_state == WR_COPIED &&
1363 dmu_read(zd->zd_os, lr->lr_foid, lr->lr_offset, lr->lr_length,
1364 ((lr_write_t *)&itx->itx_lr) + 1, DMU_READ_NO_PREFETCH) != 0) {
1365 zil_itx_destroy(itx);
1366 itx = zil_itx_create(TX_WRITE, sizeof (*lr));
1367 write_state = WR_NEED_COPY;
1369 itx->itx_private = zd;
1370 itx->itx_wr_state = write_state;
1371 itx->itx_sync = (ztest_random(8) == 0);
1372 itx->itx_sod += (write_state == WR_NEED_COPY ? lr->lr_length : 0);
1374 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1375 sizeof (*lr) - sizeof (lr_t));
1377 zil_itx_assign(zd->zd_zilog, itx, tx);
1381 ztest_log_truncate(ztest_ds_t *zd, dmu_tx_t *tx, lr_truncate_t *lr)
1385 if (zil_replaying(zd->zd_zilog, tx))
1388 itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr));
1389 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1390 sizeof (*lr) - sizeof (lr_t));
1392 itx->itx_sync = B_FALSE;
1393 zil_itx_assign(zd->zd_zilog, itx, tx);
1397 ztest_log_setattr(ztest_ds_t *zd, dmu_tx_t *tx, lr_setattr_t *lr)
1401 if (zil_replaying(zd->zd_zilog, tx))
1404 itx = zil_itx_create(TX_SETATTR, sizeof (*lr));
1405 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1406 sizeof (*lr) - sizeof (lr_t));
1408 itx->itx_sync = B_FALSE;
1409 zil_itx_assign(zd->zd_zilog, itx, tx);
1416 ztest_replay_create(ztest_ds_t *zd, lr_create_t *lr, boolean_t byteswap)
1418 char *name = (void *)(lr + 1); /* name follows lr */
1419 objset_t *os = zd->zd_os;
1420 ztest_block_tag_t *bbt;
1427 byteswap_uint64_array(lr, sizeof (*lr));
1429 ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1430 ASSERT(name[0] != '\0');
1432 tx = dmu_tx_create(os);
1434 dmu_tx_hold_zap(tx, lr->lr_doid, B_TRUE, name);
1436 if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1437 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, B_TRUE, NULL);
1439 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1442 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1446 ASSERT(dmu_objset_zil(os)->zl_replay == !!lr->lr_foid);
1448 if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1449 if (lr->lr_foid == 0) {
1450 lr->lr_foid = zap_create(os,
1451 lr->lrz_type, lr->lrz_bonustype,
1452 lr->lrz_bonuslen, tx);
1454 error = zap_create_claim(os, lr->lr_foid,
1455 lr->lrz_type, lr->lrz_bonustype,
1456 lr->lrz_bonuslen, tx);
1459 if (lr->lr_foid == 0) {
1460 lr->lr_foid = dmu_object_alloc(os,
1461 lr->lrz_type, 0, lr->lrz_bonustype,
1462 lr->lrz_bonuslen, tx);
1464 error = dmu_object_claim(os, lr->lr_foid,
1465 lr->lrz_type, 0, lr->lrz_bonustype,
1466 lr->lrz_bonuslen, tx);
1471 ASSERT3U(error, ==, EEXIST);
1472 ASSERT(zd->zd_zilog->zl_replay);
1477 ASSERT(lr->lr_foid != 0);
1479 if (lr->lrz_type != DMU_OT_ZAP_OTHER)
1480 VERIFY3U(0, ==, dmu_object_set_blocksize(os, lr->lr_foid,
1481 lr->lrz_blocksize, lr->lrz_ibshift, tx));
1483 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1484 bbt = ztest_bt_bonus(db);
1485 dmu_buf_will_dirty(db, tx);
1486 ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_gen, txg, txg);
1487 dmu_buf_rele(db, FTAG);
1489 VERIFY3U(0, ==, zap_add(os, lr->lr_doid, name, sizeof (uint64_t), 1,
1492 (void) ztest_log_create(zd, tx, lr);
1500 ztest_replay_remove(ztest_ds_t *zd, lr_remove_t *lr, boolean_t byteswap)
1502 char *name = (void *)(lr + 1); /* name follows lr */
1503 objset_t *os = zd->zd_os;
1504 dmu_object_info_t doi;
1506 uint64_t object, txg;
1509 byteswap_uint64_array(lr, sizeof (*lr));
1511 ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1512 ASSERT(name[0] != '\0');
1515 zap_lookup(os, lr->lr_doid, name, sizeof (object), 1, &object));
1516 ASSERT(object != 0);
1518 ztest_object_lock(zd, object, RL_WRITER);
1520 VERIFY3U(0, ==, dmu_object_info(os, object, &doi));
1522 tx = dmu_tx_create(os);
1524 dmu_tx_hold_zap(tx, lr->lr_doid, B_FALSE, name);
1525 dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END);
1527 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1529 ztest_object_unlock(zd, object);
1533 if (doi.doi_type == DMU_OT_ZAP_OTHER) {
1534 VERIFY3U(0, ==, zap_destroy(os, object, tx));
1536 VERIFY3U(0, ==, dmu_object_free(os, object, tx));
1539 VERIFY3U(0, ==, zap_remove(os, lr->lr_doid, name, tx));
1541 (void) ztest_log_remove(zd, tx, lr, object);
1545 ztest_object_unlock(zd, object);
1551 ztest_replay_write(ztest_ds_t *zd, lr_write_t *lr, boolean_t byteswap)
1553 objset_t *os = zd->zd_os;
1554 void *data = lr + 1; /* data follows lr */
1555 uint64_t offset, length;
1556 ztest_block_tag_t *bt = data;
1557 ztest_block_tag_t *bbt;
1558 uint64_t gen, txg, lrtxg, crtxg;
1559 dmu_object_info_t doi;
1562 arc_buf_t *abuf = NULL;
1566 byteswap_uint64_array(lr, sizeof (*lr));
1568 offset = lr->lr_offset;
1569 length = lr->lr_length;
1571 /* If it's a dmu_sync() block, write the whole block */
1572 if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
1573 uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
1574 if (length < blocksize) {
1575 offset -= offset % blocksize;
1580 if (bt->bt_magic == BSWAP_64(BT_MAGIC))
1581 byteswap_uint64_array(bt, sizeof (*bt));
1583 if (bt->bt_magic != BT_MAGIC)
1586 ztest_object_lock(zd, lr->lr_foid, RL_READER);
1587 rl = ztest_range_lock(zd, lr->lr_foid, offset, length, RL_WRITER);
1589 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1591 dmu_object_info_from_db(db, &doi);
1593 bbt = ztest_bt_bonus(db);
1594 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1596 crtxg = bbt->bt_crtxg;
1597 lrtxg = lr->lr_common.lrc_txg;
1599 tx = dmu_tx_create(os);
1601 dmu_tx_hold_write(tx, lr->lr_foid, offset, length);
1603 if (ztest_random(8) == 0 && length == doi.doi_data_block_size &&
1604 P2PHASE(offset, length) == 0)
1605 abuf = dmu_request_arcbuf(db, length);
1607 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1610 dmu_return_arcbuf(abuf);
1611 dmu_buf_rele(db, FTAG);
1612 ztest_range_unlock(rl);
1613 ztest_object_unlock(zd, lr->lr_foid);
1619 * Usually, verify the old data before writing new data --
1620 * but not always, because we also want to verify correct
1621 * behavior when the data was not recently read into cache.
1623 ASSERT(offset % doi.doi_data_block_size == 0);
1624 if (ztest_random(4) != 0) {
1625 int prefetch = ztest_random(2) ?
1626 DMU_READ_PREFETCH : DMU_READ_NO_PREFETCH;
1627 ztest_block_tag_t rbt;
1629 VERIFY(dmu_read(os, lr->lr_foid, offset,
1630 sizeof (rbt), &rbt, prefetch) == 0);
1631 if (rbt.bt_magic == BT_MAGIC) {
1632 ztest_bt_verify(&rbt, os, lr->lr_foid,
1633 offset, gen, txg, crtxg);
1638 * Writes can appear to be newer than the bonus buffer because
1639 * the ztest_get_data() callback does a dmu_read() of the
1640 * open-context data, which may be different than the data
1641 * as it was when the write was generated.
1643 if (zd->zd_zilog->zl_replay) {
1644 ztest_bt_verify(bt, os, lr->lr_foid, offset,
1645 MAX(gen, bt->bt_gen), MAX(txg, lrtxg),
1650 * Set the bt's gen/txg to the bonus buffer's gen/txg
1651 * so that all of the usual ASSERTs will work.
1653 ztest_bt_generate(bt, os, lr->lr_foid, offset, gen, txg, crtxg);
1657 dmu_write(os, lr->lr_foid, offset, length, data, tx);
1659 bcopy(data, abuf->b_data, length);
1660 dmu_assign_arcbuf(db, offset, abuf, tx);
1663 (void) ztest_log_write(zd, tx, lr);
1665 dmu_buf_rele(db, FTAG);
1669 ztest_range_unlock(rl);
1670 ztest_object_unlock(zd, lr->lr_foid);
1676 ztest_replay_truncate(ztest_ds_t *zd, lr_truncate_t *lr, boolean_t byteswap)
1678 objset_t *os = zd->zd_os;
1684 byteswap_uint64_array(lr, sizeof (*lr));
1686 ztest_object_lock(zd, lr->lr_foid, RL_READER);
1687 rl = ztest_range_lock(zd, lr->lr_foid, lr->lr_offset, lr->lr_length,
1690 tx = dmu_tx_create(os);
1692 dmu_tx_hold_free(tx, lr->lr_foid, lr->lr_offset, lr->lr_length);
1694 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1696 ztest_range_unlock(rl);
1697 ztest_object_unlock(zd, lr->lr_foid);
1701 VERIFY(dmu_free_range(os, lr->lr_foid, lr->lr_offset,
1702 lr->lr_length, tx) == 0);
1704 (void) ztest_log_truncate(zd, tx, lr);
1708 ztest_range_unlock(rl);
1709 ztest_object_unlock(zd, lr->lr_foid);
1715 ztest_replay_setattr(ztest_ds_t *zd, lr_setattr_t *lr, boolean_t byteswap)
1717 objset_t *os = zd->zd_os;
1720 ztest_block_tag_t *bbt;
1721 uint64_t txg, lrtxg, crtxg;
1724 byteswap_uint64_array(lr, sizeof (*lr));
1726 ztest_object_lock(zd, lr->lr_foid, RL_WRITER);
1728 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1730 tx = dmu_tx_create(os);
1731 dmu_tx_hold_bonus(tx, lr->lr_foid);
1733 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1735 dmu_buf_rele(db, FTAG);
1736 ztest_object_unlock(zd, lr->lr_foid);
1740 bbt = ztest_bt_bonus(db);
1741 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1742 crtxg = bbt->bt_crtxg;
1743 lrtxg = lr->lr_common.lrc_txg;
1745 if (zd->zd_zilog->zl_replay) {
1746 ASSERT(lr->lr_size != 0);
1747 ASSERT(lr->lr_mode != 0);
1751 * Randomly change the size and increment the generation.
1753 lr->lr_size = (ztest_random(db->db_size / sizeof (*bbt)) + 1) *
1755 lr->lr_mode = bbt->bt_gen + 1;
1760 * Verify that the current bonus buffer is not newer than our txg.
1762 ztest_bt_verify(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode,
1763 MAX(txg, lrtxg), crtxg);
1765 dmu_buf_will_dirty(db, tx);
1767 ASSERT3U(lr->lr_size, >=, sizeof (*bbt));
1768 ASSERT3U(lr->lr_size, <=, db->db_size);
1769 VERIFY0(dmu_set_bonus(db, lr->lr_size, tx));
1770 bbt = ztest_bt_bonus(db);
1772 ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode, txg, crtxg);
1774 dmu_buf_rele(db, FTAG);
1776 (void) ztest_log_setattr(zd, tx, lr);
1780 ztest_object_unlock(zd, lr->lr_foid);
1785 zil_replay_func_t *ztest_replay_vector[TX_MAX_TYPE] = {
1786 NULL, /* 0 no such transaction type */
1787 ztest_replay_create, /* TX_CREATE */
1788 NULL, /* TX_MKDIR */
1789 NULL, /* TX_MKXATTR */
1790 NULL, /* TX_SYMLINK */
1791 ztest_replay_remove, /* TX_REMOVE */
1792 NULL, /* TX_RMDIR */
1794 NULL, /* TX_RENAME */
1795 ztest_replay_write, /* TX_WRITE */
1796 ztest_replay_truncate, /* TX_TRUNCATE */
1797 ztest_replay_setattr, /* TX_SETATTR */
1799 NULL, /* TX_CREATE_ACL */
1800 NULL, /* TX_CREATE_ATTR */
1801 NULL, /* TX_CREATE_ACL_ATTR */
1802 NULL, /* TX_MKDIR_ACL */
1803 NULL, /* TX_MKDIR_ATTR */
1804 NULL, /* TX_MKDIR_ACL_ATTR */
1805 NULL, /* TX_WRITE2 */
1809 * ZIL get_data callbacks
1813 ztest_get_done(zgd_t *zgd, int error)
1815 ztest_ds_t *zd = zgd->zgd_private;
1816 uint64_t object = zgd->zgd_rl->rl_object;
1819 dmu_buf_rele(zgd->zgd_db, zgd);
1821 ztest_range_unlock(zgd->zgd_rl);
1822 ztest_object_unlock(zd, object);
1824 if (error == 0 && zgd->zgd_bp)
1825 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
1827 umem_free(zgd, sizeof (*zgd));
1831 ztest_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
1833 ztest_ds_t *zd = arg;
1834 objset_t *os = zd->zd_os;
1835 uint64_t object = lr->lr_foid;
1836 uint64_t offset = lr->lr_offset;
1837 uint64_t size = lr->lr_length;
1838 blkptr_t *bp = &lr->lr_blkptr;
1839 uint64_t txg = lr->lr_common.lrc_txg;
1841 dmu_object_info_t doi;
1846 ztest_object_lock(zd, object, RL_READER);
1847 error = dmu_bonus_hold(os, object, FTAG, &db);
1849 ztest_object_unlock(zd, object);
1853 crtxg = ztest_bt_bonus(db)->bt_crtxg;
1855 if (crtxg == 0 || crtxg > txg) {
1856 dmu_buf_rele(db, FTAG);
1857 ztest_object_unlock(zd, object);
1861 dmu_object_info_from_db(db, &doi);
1862 dmu_buf_rele(db, FTAG);
1865 zgd = umem_zalloc(sizeof (*zgd), UMEM_NOFAIL);
1866 zgd->zgd_zilog = zd->zd_zilog;
1867 zgd->zgd_private = zd;
1869 if (buf != NULL) { /* immediate write */
1870 zgd->zgd_rl = ztest_range_lock(zd, object, offset, size,
1873 error = dmu_read(os, object, offset, size, buf,
1874 DMU_READ_NO_PREFETCH);
1877 size = doi.doi_data_block_size;
1879 offset = P2ALIGN(offset, size);
1881 ASSERT(offset < size);
1885 zgd->zgd_rl = ztest_range_lock(zd, object, offset, size,
1888 error = dmu_buf_hold(os, object, offset, zgd, &db,
1889 DMU_READ_NO_PREFETCH);
1892 blkptr_t *obp = dmu_buf_get_blkptr(db);
1894 ASSERT(BP_IS_HOLE(bp));
1901 ASSERT(db->db_offset == offset);
1902 ASSERT(db->db_size == size);
1904 error = dmu_sync(zio, lr->lr_common.lrc_txg,
1905 ztest_get_done, zgd);
1912 ztest_get_done(zgd, error);
1918 ztest_lr_alloc(size_t lrsize, char *name)
1921 size_t namesize = name ? strlen(name) + 1 : 0;
1923 lr = umem_zalloc(lrsize + namesize, UMEM_NOFAIL);
1926 bcopy(name, lr + lrsize, namesize);
1932 ztest_lr_free(void *lr, size_t lrsize, char *name)
1934 size_t namesize = name ? strlen(name) + 1 : 0;
1936 umem_free(lr, lrsize + namesize);
1940 * Lookup a bunch of objects. Returns the number of objects not found.
1943 ztest_lookup(ztest_ds_t *zd, ztest_od_t *od, int count)
1948 ASSERT(_mutex_held(&zd->zd_dirobj_lock));
1950 for (int i = 0; i < count; i++, od++) {
1952 error = zap_lookup(zd->zd_os, od->od_dir, od->od_name,
1953 sizeof (uint64_t), 1, &od->od_object);
1955 ASSERT(error == ENOENT);
1956 ASSERT(od->od_object == 0);
1960 ztest_block_tag_t *bbt;
1961 dmu_object_info_t doi;
1963 ASSERT(od->od_object != 0);
1964 ASSERT(missing == 0); /* there should be no gaps */
1966 ztest_object_lock(zd, od->od_object, RL_READER);
1967 VERIFY3U(0, ==, dmu_bonus_hold(zd->zd_os,
1968 od->od_object, FTAG, &db));
1969 dmu_object_info_from_db(db, &doi);
1970 bbt = ztest_bt_bonus(db);
1971 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1972 od->od_type = doi.doi_type;
1973 od->od_blocksize = doi.doi_data_block_size;
1974 od->od_gen = bbt->bt_gen;
1975 dmu_buf_rele(db, FTAG);
1976 ztest_object_unlock(zd, od->od_object);
1984 ztest_create(ztest_ds_t *zd, ztest_od_t *od, int count)
1988 ASSERT(_mutex_held(&zd->zd_dirobj_lock));
1990 for (int i = 0; i < count; i++, od++) {
1997 lr_create_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
1999 lr->lr_doid = od->od_dir;
2000 lr->lr_foid = 0; /* 0 to allocate, > 0 to claim */
2001 lr->lrz_type = od->od_crtype;
2002 lr->lrz_blocksize = od->od_crblocksize;
2003 lr->lrz_ibshift = ztest_random_ibshift();
2004 lr->lrz_bonustype = DMU_OT_UINT64_OTHER;
2005 lr->lrz_bonuslen = dmu_bonus_max();
2006 lr->lr_gen = od->od_crgen;
2007 lr->lr_crtime[0] = time(NULL);
2009 if (ztest_replay_create(zd, lr, B_FALSE) != 0) {
2010 ASSERT(missing == 0);
2014 od->od_object = lr->lr_foid;
2015 od->od_type = od->od_crtype;
2016 od->od_blocksize = od->od_crblocksize;
2017 od->od_gen = od->od_crgen;
2018 ASSERT(od->od_object != 0);
2021 ztest_lr_free(lr, sizeof (*lr), od->od_name);
2028 ztest_remove(ztest_ds_t *zd, ztest_od_t *od, int count)
2033 ASSERT(_mutex_held(&zd->zd_dirobj_lock));
2037 for (int i = count - 1; i >= 0; i--, od--) {
2044 * No object was found.
2046 if (od->od_object == 0)
2049 lr_remove_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
2051 lr->lr_doid = od->od_dir;
2053 if ((error = ztest_replay_remove(zd, lr, B_FALSE)) != 0) {
2054 ASSERT3U(error, ==, ENOSPC);
2059 ztest_lr_free(lr, sizeof (*lr), od->od_name);
2066 ztest_write(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size,
2072 lr = ztest_lr_alloc(sizeof (*lr) + size, NULL);
2074 lr->lr_foid = object;
2075 lr->lr_offset = offset;
2076 lr->lr_length = size;
2078 BP_ZERO(&lr->lr_blkptr);
2080 bcopy(data, lr + 1, size);
2082 error = ztest_replay_write(zd, lr, B_FALSE);
2084 ztest_lr_free(lr, sizeof (*lr) + size, NULL);
2090 ztest_truncate(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2095 lr = ztest_lr_alloc(sizeof (*lr), NULL);
2097 lr->lr_foid = object;
2098 lr->lr_offset = offset;
2099 lr->lr_length = size;
2101 error = ztest_replay_truncate(zd, lr, B_FALSE);
2103 ztest_lr_free(lr, sizeof (*lr), NULL);
2109 ztest_setattr(ztest_ds_t *zd, uint64_t object)
2114 lr = ztest_lr_alloc(sizeof (*lr), NULL);
2116 lr->lr_foid = object;
2120 error = ztest_replay_setattr(zd, lr, B_FALSE);
2122 ztest_lr_free(lr, sizeof (*lr), NULL);
2128 ztest_prealloc(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2130 objset_t *os = zd->zd_os;
2135 txg_wait_synced(dmu_objset_pool(os), 0);
2137 ztest_object_lock(zd, object, RL_READER);
2138 rl = ztest_range_lock(zd, object, offset, size, RL_WRITER);
2140 tx = dmu_tx_create(os);
2142 dmu_tx_hold_write(tx, object, offset, size);
2144 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
2147 dmu_prealloc(os, object, offset, size, tx);
2149 txg_wait_synced(dmu_objset_pool(os), txg);
2151 (void) dmu_free_long_range(os, object, offset, size);
2154 ztest_range_unlock(rl);
2155 ztest_object_unlock(zd, object);
2159 ztest_io(ztest_ds_t *zd, uint64_t object, uint64_t offset)
2162 ztest_block_tag_t wbt;
2163 dmu_object_info_t doi;
2164 enum ztest_io_type io_type;
2168 VERIFY(dmu_object_info(zd->zd_os, object, &doi) == 0);
2169 blocksize = doi.doi_data_block_size;
2170 data = umem_alloc(blocksize, UMEM_NOFAIL);
2173 * Pick an i/o type at random, biased toward writing block tags.
2175 io_type = ztest_random(ZTEST_IO_TYPES);
2176 if (ztest_random(2) == 0)
2177 io_type = ZTEST_IO_WRITE_TAG;
2179 (void) rw_rdlock(&zd->zd_zilog_lock);
2183 case ZTEST_IO_WRITE_TAG:
2184 ztest_bt_generate(&wbt, zd->zd_os, object, offset, 0, 0, 0);
2185 (void) ztest_write(zd, object, offset, sizeof (wbt), &wbt);
2188 case ZTEST_IO_WRITE_PATTERN:
2189 (void) memset(data, 'a' + (object + offset) % 5, blocksize);
2190 if (ztest_random(2) == 0) {
2192 * Induce fletcher2 collisions to ensure that
2193 * zio_ddt_collision() detects and resolves them
2194 * when using fletcher2-verify for deduplication.
2196 ((uint64_t *)data)[0] ^= 1ULL << 63;
2197 ((uint64_t *)data)[4] ^= 1ULL << 63;
2199 (void) ztest_write(zd, object, offset, blocksize, data);
2202 case ZTEST_IO_WRITE_ZEROES:
2203 bzero(data, blocksize);
2204 (void) ztest_write(zd, object, offset, blocksize, data);
2207 case ZTEST_IO_TRUNCATE:
2208 (void) ztest_truncate(zd, object, offset, blocksize);
2211 case ZTEST_IO_SETATTR:
2212 (void) ztest_setattr(zd, object);
2215 case ZTEST_IO_REWRITE:
2216 (void) rw_rdlock(&ztest_name_lock);
2217 err = ztest_dsl_prop_set_uint64(zd->zd_name,
2218 ZFS_PROP_CHECKSUM, spa_dedup_checksum(ztest_spa),
2220 VERIFY(err == 0 || err == ENOSPC);
2221 err = ztest_dsl_prop_set_uint64(zd->zd_name,
2222 ZFS_PROP_COMPRESSION,
2223 ztest_random_dsl_prop(ZFS_PROP_COMPRESSION),
2225 VERIFY(err == 0 || err == ENOSPC);
2226 (void) rw_unlock(&ztest_name_lock);
2228 VERIFY0(dmu_read(zd->zd_os, object, offset, blocksize, data,
2229 DMU_READ_NO_PREFETCH));
2231 (void) ztest_write(zd, object, offset, blocksize, data);
2235 (void) rw_unlock(&zd->zd_zilog_lock);
2237 umem_free(data, blocksize);
2241 * Initialize an object description template.
2244 ztest_od_init(ztest_od_t *od, uint64_t id, char *tag, uint64_t index,
2245 dmu_object_type_t type, uint64_t blocksize, uint64_t gen)
2247 od->od_dir = ZTEST_DIROBJ;
2250 od->od_crtype = type;
2251 od->od_crblocksize = blocksize ? blocksize : ztest_random_blocksize();
2254 od->od_type = DMU_OT_NONE;
2255 od->od_blocksize = 0;
2258 (void) snprintf(od->od_name, sizeof (od->od_name), "%s(%lld)[%llu]",
2259 tag, (int64_t)id, index);
2263 * Lookup or create the objects for a test using the od template.
2264 * If the objects do not all exist, or if 'remove' is specified,
2265 * remove any existing objects and create new ones. Otherwise,
2266 * use the existing objects.
2269 ztest_object_init(ztest_ds_t *zd, ztest_od_t *od, size_t size, boolean_t remove)
2271 int count = size / sizeof (*od);
2274 VERIFY(mutex_lock(&zd->zd_dirobj_lock) == 0);
2275 if ((ztest_lookup(zd, od, count) != 0 || remove) &&
2276 (ztest_remove(zd, od, count) != 0 ||
2277 ztest_create(zd, od, count) != 0))
2280 VERIFY(mutex_unlock(&zd->zd_dirobj_lock) == 0);
2287 ztest_zil_commit(ztest_ds_t *zd, uint64_t id)
2289 zilog_t *zilog = zd->zd_zilog;
2291 (void) rw_rdlock(&zd->zd_zilog_lock);
2293 zil_commit(zilog, ztest_random(ZTEST_OBJECTS));
2296 * Remember the committed values in zd, which is in parent/child
2297 * shared memory. If we die, the next iteration of ztest_run()
2298 * will verify that the log really does contain this record.
2300 mutex_enter(&zilog->zl_lock);
2301 ASSERT(zd->zd_shared != NULL);
2302 ASSERT3U(zd->zd_shared->zd_seq, <=, zilog->zl_commit_lr_seq);
2303 zd->zd_shared->zd_seq = zilog->zl_commit_lr_seq;
2304 mutex_exit(&zilog->zl_lock);
2306 (void) rw_unlock(&zd->zd_zilog_lock);
2310 * This function is designed to simulate the operations that occur during a
2311 * mount/unmount operation. We hold the dataset across these operations in an
2312 * attempt to expose any implicit assumptions about ZIL management.
2316 ztest_zil_remount(ztest_ds_t *zd, uint64_t id)
2318 objset_t *os = zd->zd_os;
2321 * We grab the zd_dirobj_lock to ensure that no other thread is
2322 * updating the zil (i.e. adding in-memory log records) and the
2323 * zd_zilog_lock to block any I/O.
2325 VERIFY0(mutex_lock(&zd->zd_dirobj_lock));
2326 (void) rw_wrlock(&zd->zd_zilog_lock);
2328 /* zfsvfs_teardown() */
2329 zil_close(zd->zd_zilog);
2331 /* zfsvfs_setup() */
2332 VERIFY(zil_open(os, ztest_get_data) == zd->zd_zilog);
2333 zil_replay(os, zd, ztest_replay_vector);
2335 (void) rw_unlock(&zd->zd_zilog_lock);
2336 VERIFY(mutex_unlock(&zd->zd_dirobj_lock) == 0);
2340 * Verify that we can't destroy an active pool, create an existing pool,
2341 * or create a pool with a bad vdev spec.
2345 ztest_spa_create_destroy(ztest_ds_t *zd, uint64_t id)
2347 ztest_shared_opts_t *zo = &ztest_opts;
2352 * Attempt to create using a bad file.
2354 nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1);
2355 VERIFY3U(ENOENT, ==,
2356 spa_create("ztest_bad_file", nvroot, NULL, NULL));
2357 nvlist_free(nvroot);
2360 * Attempt to create using a bad mirror.
2362 nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 2, 1);
2363 VERIFY3U(ENOENT, ==,
2364 spa_create("ztest_bad_mirror", nvroot, NULL, NULL));
2365 nvlist_free(nvroot);
2368 * Attempt to create an existing pool. It shouldn't matter
2369 * what's in the nvroot; we should fail with EEXIST.
2371 (void) rw_rdlock(&ztest_name_lock);
2372 nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1);
2373 VERIFY3U(EEXIST, ==, spa_create(zo->zo_pool, nvroot, NULL, NULL));
2374 nvlist_free(nvroot);
2375 VERIFY3U(0, ==, spa_open(zo->zo_pool, &spa, FTAG));
2376 VERIFY3U(EBUSY, ==, spa_destroy(zo->zo_pool));
2377 spa_close(spa, FTAG);
2379 (void) rw_unlock(&ztest_name_lock);
2384 ztest_spa_upgrade(ztest_ds_t *zd, uint64_t id)
2387 uint64_t initial_version = SPA_VERSION_INITIAL;
2388 uint64_t version, newversion;
2389 nvlist_t *nvroot, *props;
2392 VERIFY0(mutex_lock(&ztest_vdev_lock));
2393 name = kmem_asprintf("%s_upgrade", ztest_opts.zo_pool);
2396 * Clean up from previous runs.
2398 (void) spa_destroy(name);
2400 nvroot = make_vdev_root(NULL, NULL, name, ztest_opts.zo_vdev_size, 0,
2401 0, ztest_opts.zo_raidz, ztest_opts.zo_mirrors, 1);
2404 * If we're configuring a RAIDZ device then make sure that the
2405 * the initial version is capable of supporting that feature.
2407 switch (ztest_opts.zo_raidz_parity) {
2410 initial_version = SPA_VERSION_INITIAL;
2413 initial_version = SPA_VERSION_RAIDZ2;
2416 initial_version = SPA_VERSION_RAIDZ3;
2421 * Create a pool with a spa version that can be upgraded. Pick
2422 * a value between initial_version and SPA_VERSION_BEFORE_FEATURES.
2425 version = ztest_random_spa_version(initial_version);
2426 } while (version > SPA_VERSION_BEFORE_FEATURES);
2428 props = fnvlist_alloc();
2429 fnvlist_add_uint64(props,
2430 zpool_prop_to_name(ZPOOL_PROP_VERSION), version);
2431 VERIFY0(spa_create(name, nvroot, props, NULL));
2432 fnvlist_free(nvroot);
2433 fnvlist_free(props);
2435 VERIFY0(spa_open(name, &spa, FTAG));
2436 VERIFY3U(spa_version(spa), ==, version);
2437 newversion = ztest_random_spa_version(version + 1);
2439 if (ztest_opts.zo_verbose >= 4) {
2440 (void) printf("upgrading spa version from %llu to %llu\n",
2441 (u_longlong_t)version, (u_longlong_t)newversion);
2444 spa_upgrade(spa, newversion);
2445 VERIFY3U(spa_version(spa), >, version);
2446 VERIFY3U(spa_version(spa), ==, fnvlist_lookup_uint64(spa->spa_config,
2447 zpool_prop_to_name(ZPOOL_PROP_VERSION)));
2448 spa_close(spa, FTAG);
2451 VERIFY0(mutex_unlock(&ztest_vdev_lock));
2455 vdev_lookup_by_path(vdev_t *vd, const char *path)
2459 if (vd->vdev_path != NULL && strcmp(path, vd->vdev_path) == 0)
2462 for (int c = 0; c < vd->vdev_children; c++)
2463 if ((mvd = vdev_lookup_by_path(vd->vdev_child[c], path)) !=
2471 * Find the first available hole which can be used as a top-level.
2474 find_vdev_hole(spa_t *spa)
2476 vdev_t *rvd = spa->spa_root_vdev;
2479 ASSERT(spa_config_held(spa, SCL_VDEV, RW_READER) == SCL_VDEV);
2481 for (c = 0; c < rvd->vdev_children; c++) {
2482 vdev_t *cvd = rvd->vdev_child[c];
2484 if (cvd->vdev_ishole)
2491 * Verify that vdev_add() works as expected.
2495 ztest_vdev_add_remove(ztest_ds_t *zd, uint64_t id)
2497 ztest_shared_t *zs = ztest_shared;
2498 spa_t *spa = ztest_spa;
2504 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2505 leaves = MAX(zs->zs_mirrors + zs->zs_splits, 1) * ztest_opts.zo_raidz;
2507 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2509 ztest_shared->zs_vdev_next_leaf = find_vdev_hole(spa) * leaves;
2512 * If we have slogs then remove them 1/4 of the time.
2514 if (spa_has_slogs(spa) && ztest_random(4) == 0) {
2516 * Grab the guid from the head of the log class rotor.
2518 guid = spa_log_class(spa)->mc_rotor->mg_vd->vdev_guid;
2520 spa_config_exit(spa, SCL_VDEV, FTAG);
2523 * We have to grab the zs_name_lock as writer to
2524 * prevent a race between removing a slog (dmu_objset_find)
2525 * and destroying a dataset. Removing the slog will
2526 * grab a reference on the dataset which may cause
2527 * dmu_objset_destroy() to fail with EBUSY thus
2528 * leaving the dataset in an inconsistent state.
2530 VERIFY(rw_wrlock(&ztest_name_lock) == 0);
2531 error = spa_vdev_remove(spa, guid, B_FALSE);
2532 VERIFY(rw_unlock(&ztest_name_lock) == 0);
2534 if (error && error != EEXIST)
2535 fatal(0, "spa_vdev_remove() = %d", error);
2537 spa_config_exit(spa, SCL_VDEV, FTAG);
2540 * Make 1/4 of the devices be log devices.
2542 nvroot = make_vdev_root(NULL, NULL, NULL,
2543 ztest_opts.zo_vdev_size, 0,
2544 ztest_random(4) == 0, ztest_opts.zo_raidz,
2547 error = spa_vdev_add(spa, nvroot);
2548 nvlist_free(nvroot);
2550 if (error == ENOSPC)
2551 ztest_record_enospc("spa_vdev_add");
2552 else if (error != 0)
2553 fatal(0, "spa_vdev_add() = %d", error);
2556 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2560 * Verify that adding/removing aux devices (l2arc, hot spare) works as expected.
2564 ztest_vdev_aux_add_remove(ztest_ds_t *zd, uint64_t id)
2566 ztest_shared_t *zs = ztest_shared;
2567 spa_t *spa = ztest_spa;
2568 vdev_t *rvd = spa->spa_root_vdev;
2569 spa_aux_vdev_t *sav;
2574 if (ztest_random(2) == 0) {
2575 sav = &spa->spa_spares;
2576 aux = ZPOOL_CONFIG_SPARES;
2578 sav = &spa->spa_l2cache;
2579 aux = ZPOOL_CONFIG_L2CACHE;
2582 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2584 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2586 if (sav->sav_count != 0 && ztest_random(4) == 0) {
2588 * Pick a random device to remove.
2590 guid = sav->sav_vdevs[ztest_random(sav->sav_count)]->vdev_guid;
2593 * Find an unused device we can add.
2595 zs->zs_vdev_aux = 0;
2597 char path[MAXPATHLEN];
2599 (void) snprintf(path, sizeof (path), ztest_aux_template,
2600 ztest_opts.zo_dir, ztest_opts.zo_pool, aux,
2602 for (c = 0; c < sav->sav_count; c++)
2603 if (strcmp(sav->sav_vdevs[c]->vdev_path,
2606 if (c == sav->sav_count &&
2607 vdev_lookup_by_path(rvd, path) == NULL)
2613 spa_config_exit(spa, SCL_VDEV, FTAG);
2619 nvlist_t *nvroot = make_vdev_root(NULL, aux, NULL,
2620 (ztest_opts.zo_vdev_size * 5) / 4, 0, 0, 0, 0, 1);
2621 error = spa_vdev_add(spa, nvroot);
2623 fatal(0, "spa_vdev_add(%p) = %d", nvroot, error);
2624 nvlist_free(nvroot);
2627 * Remove an existing device. Sometimes, dirty its
2628 * vdev state first to make sure we handle removal
2629 * of devices that have pending state changes.
2631 if (ztest_random(2) == 0)
2632 (void) vdev_online(spa, guid, 0, NULL);
2634 error = spa_vdev_remove(spa, guid, B_FALSE);
2635 if (error != 0 && error != EBUSY)
2636 fatal(0, "spa_vdev_remove(%llu) = %d", guid, error);
2639 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2643 * split a pool if it has mirror tlvdevs
2647 ztest_split_pool(ztest_ds_t *zd, uint64_t id)
2649 ztest_shared_t *zs = ztest_shared;
2650 spa_t *spa = ztest_spa;
2651 vdev_t *rvd = spa->spa_root_vdev;
2652 nvlist_t *tree, **child, *config, *split, **schild;
2653 uint_t c, children, schildren = 0, lastlogid = 0;
2656 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2658 /* ensure we have a useable config; mirrors of raidz aren't supported */
2659 if (zs->zs_mirrors < 3 || ztest_opts.zo_raidz > 1) {
2660 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2664 /* clean up the old pool, if any */
2665 (void) spa_destroy("splitp");
2667 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2669 /* generate a config from the existing config */
2670 mutex_enter(&spa->spa_props_lock);
2671 VERIFY(nvlist_lookup_nvlist(spa->spa_config, ZPOOL_CONFIG_VDEV_TREE,
2673 mutex_exit(&spa->spa_props_lock);
2675 VERIFY(nvlist_lookup_nvlist_array(tree, ZPOOL_CONFIG_CHILDREN, &child,
2678 schild = malloc(rvd->vdev_children * sizeof (nvlist_t *));
2679 for (c = 0; c < children; c++) {
2680 vdev_t *tvd = rvd->vdev_child[c];
2684 if (tvd->vdev_islog || tvd->vdev_ops == &vdev_hole_ops) {
2685 VERIFY(nvlist_alloc(&schild[schildren], NV_UNIQUE_NAME,
2687 VERIFY(nvlist_add_string(schild[schildren],
2688 ZPOOL_CONFIG_TYPE, VDEV_TYPE_HOLE) == 0);
2689 VERIFY(nvlist_add_uint64(schild[schildren],
2690 ZPOOL_CONFIG_IS_HOLE, 1) == 0);
2692 lastlogid = schildren;
2697 VERIFY(nvlist_lookup_nvlist_array(child[c],
2698 ZPOOL_CONFIG_CHILDREN, &mchild, &mchildren) == 0);
2699 VERIFY(nvlist_dup(mchild[0], &schild[schildren++], 0) == 0);
2702 /* OK, create a config that can be used to split */
2703 VERIFY(nvlist_alloc(&split, NV_UNIQUE_NAME, 0) == 0);
2704 VERIFY(nvlist_add_string(split, ZPOOL_CONFIG_TYPE,
2705 VDEV_TYPE_ROOT) == 0);
2706 VERIFY(nvlist_add_nvlist_array(split, ZPOOL_CONFIG_CHILDREN, schild,
2707 lastlogid != 0 ? lastlogid : schildren) == 0);
2709 VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, 0) == 0);
2710 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, split) == 0);
2712 for (c = 0; c < schildren; c++)
2713 nvlist_free(schild[c]);
2717 spa_config_exit(spa, SCL_VDEV, FTAG);
2719 (void) rw_wrlock(&ztest_name_lock);
2720 error = spa_vdev_split_mirror(spa, "splitp", config, NULL, B_FALSE);
2721 (void) rw_unlock(&ztest_name_lock);
2723 nvlist_free(config);
2726 (void) printf("successful split - results:\n");
2727 mutex_enter(&spa_namespace_lock);
2728 show_pool_stats(spa);
2729 show_pool_stats(spa_lookup("splitp"));
2730 mutex_exit(&spa_namespace_lock);
2734 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2739 * Verify that we can attach and detach devices.
2743 ztest_vdev_attach_detach(ztest_ds_t *zd, uint64_t id)
2745 ztest_shared_t *zs = ztest_shared;
2746 spa_t *spa = ztest_spa;
2747 spa_aux_vdev_t *sav = &spa->spa_spares;
2748 vdev_t *rvd = spa->spa_root_vdev;
2749 vdev_t *oldvd, *newvd, *pvd;
2753 uint64_t ashift = ztest_get_ashift();
2754 uint64_t oldguid, pguid;
2755 uint64_t oldsize, newsize;
2756 char oldpath[MAXPATHLEN], newpath[MAXPATHLEN];
2758 int oldvd_has_siblings = B_FALSE;
2759 int newvd_is_spare = B_FALSE;
2761 int error, expected_error;
2763 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2764 leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz;
2766 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2769 * Decide whether to do an attach or a replace.
2771 replacing = ztest_random(2);
2774 * Pick a random top-level vdev.
2776 top = ztest_random_vdev_top(spa, B_TRUE);
2779 * Pick a random leaf within it.
2781 leaf = ztest_random(leaves);
2786 oldvd = rvd->vdev_child[top];
2787 if (zs->zs_mirrors >= 1) {
2788 ASSERT(oldvd->vdev_ops == &vdev_mirror_ops);
2789 ASSERT(oldvd->vdev_children >= zs->zs_mirrors);
2790 oldvd = oldvd->vdev_child[leaf / ztest_opts.zo_raidz];
2792 if (ztest_opts.zo_raidz > 1) {
2793 ASSERT(oldvd->vdev_ops == &vdev_raidz_ops);
2794 ASSERT(oldvd->vdev_children == ztest_opts.zo_raidz);
2795 oldvd = oldvd->vdev_child[leaf % ztest_opts.zo_raidz];
2799 * If we're already doing an attach or replace, oldvd may be a
2800 * mirror vdev -- in which case, pick a random child.
2802 while (oldvd->vdev_children != 0) {
2803 oldvd_has_siblings = B_TRUE;
2804 ASSERT(oldvd->vdev_children >= 2);
2805 oldvd = oldvd->vdev_child[ztest_random(oldvd->vdev_children)];
2808 oldguid = oldvd->vdev_guid;
2809 oldsize = vdev_get_min_asize(oldvd);
2810 oldvd_is_log = oldvd->vdev_top->vdev_islog;
2811 (void) strcpy(oldpath, oldvd->vdev_path);
2812 pvd = oldvd->vdev_parent;
2813 pguid = pvd->vdev_guid;
2816 * If oldvd has siblings, then half of the time, detach it.
2818 if (oldvd_has_siblings && ztest_random(2) == 0) {
2819 spa_config_exit(spa, SCL_VDEV, FTAG);
2820 error = spa_vdev_detach(spa, oldguid, pguid, B_FALSE);
2821 if (error != 0 && error != ENODEV && error != EBUSY &&
2823 fatal(0, "detach (%s) returned %d", oldpath, error);
2824 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2829 * For the new vdev, choose with equal probability between the two
2830 * standard paths (ending in either 'a' or 'b') or a random hot spare.
2832 if (sav->sav_count != 0 && ztest_random(3) == 0) {
2833 newvd = sav->sav_vdevs[ztest_random(sav->sav_count)];
2834 newvd_is_spare = B_TRUE;
2835 (void) strcpy(newpath, newvd->vdev_path);
2837 (void) snprintf(newpath, sizeof (newpath), ztest_dev_template,
2838 ztest_opts.zo_dir, ztest_opts.zo_pool,
2839 top * leaves + leaf);
2840 if (ztest_random(2) == 0)
2841 newpath[strlen(newpath) - 1] = 'b';
2842 newvd = vdev_lookup_by_path(rvd, newpath);
2846 newsize = vdev_get_min_asize(newvd);
2849 * Make newsize a little bigger or smaller than oldsize.
2850 * If it's smaller, the attach should fail.
2851 * If it's larger, and we're doing a replace,
2852 * we should get dynamic LUN growth when we're done.
2854 newsize = 10 * oldsize / (9 + ztest_random(3));
2858 * If pvd is not a mirror or root, the attach should fail with ENOTSUP,
2859 * unless it's a replace; in that case any non-replacing parent is OK.
2861 * If newvd is already part of the pool, it should fail with EBUSY.
2863 * If newvd is too small, it should fail with EOVERFLOW.
2865 if (pvd->vdev_ops != &vdev_mirror_ops &&
2866 pvd->vdev_ops != &vdev_root_ops && (!replacing ||
2867 pvd->vdev_ops == &vdev_replacing_ops ||
2868 pvd->vdev_ops == &vdev_spare_ops))
2869 expected_error = ENOTSUP;
2870 else if (newvd_is_spare && (!replacing || oldvd_is_log))
2871 expected_error = ENOTSUP;
2872 else if (newvd == oldvd)
2873 expected_error = replacing ? 0 : EBUSY;
2874 else if (vdev_lookup_by_path(rvd, newpath) != NULL)
2875 expected_error = EBUSY;
2876 else if (newsize < oldsize)
2877 expected_error = EOVERFLOW;
2878 else if (ashift > oldvd->vdev_top->vdev_ashift)
2879 expected_error = EDOM;
2883 spa_config_exit(spa, SCL_VDEV, FTAG);
2886 * Build the nvlist describing newpath.
2888 root = make_vdev_root(newpath, NULL, NULL, newvd == NULL ? newsize : 0,
2889 ashift, 0, 0, 0, 1);
2891 error = spa_vdev_attach(spa, oldguid, root, replacing);
2896 * If our parent was the replacing vdev, but the replace completed,
2897 * then instead of failing with ENOTSUP we may either succeed,
2898 * fail with ENODEV, or fail with EOVERFLOW.
2900 if (expected_error == ENOTSUP &&
2901 (error == 0 || error == ENODEV || error == EOVERFLOW))
2902 expected_error = error;
2905 * If someone grew the LUN, the replacement may be too small.
2907 if (error == EOVERFLOW || error == EBUSY)
2908 expected_error = error;
2910 /* XXX workaround 6690467 */
2911 if (error != expected_error && expected_error != EBUSY) {
2912 fatal(0, "attach (%s %llu, %s %llu, %d) "
2913 "returned %d, expected %d",
2914 oldpath, oldsize, newpath,
2915 newsize, replacing, error, expected_error);
2918 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2922 * Callback function which expands the physical size of the vdev.
2925 grow_vdev(vdev_t *vd, void *arg)
2927 spa_t *spa = vd->vdev_spa;
2928 size_t *newsize = arg;
2932 ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
2933 ASSERT(vd->vdev_ops->vdev_op_leaf);
2935 if ((fd = open(vd->vdev_path, O_RDWR)) == -1)
2938 fsize = lseek(fd, 0, SEEK_END);
2939 (void) ftruncate(fd, *newsize);
2941 if (ztest_opts.zo_verbose >= 6) {
2942 (void) printf("%s grew from %lu to %lu bytes\n",
2943 vd->vdev_path, (ulong_t)fsize, (ulong_t)*newsize);
2950 * Callback function which expands a given vdev by calling vdev_online().
2954 online_vdev(vdev_t *vd, void *arg)
2956 spa_t *spa = vd->vdev_spa;
2957 vdev_t *tvd = vd->vdev_top;
2958 uint64_t guid = vd->vdev_guid;
2959 uint64_t generation = spa->spa_config_generation + 1;
2960 vdev_state_t newstate = VDEV_STATE_UNKNOWN;
2963 ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
2964 ASSERT(vd->vdev_ops->vdev_op_leaf);
2966 /* Calling vdev_online will initialize the new metaslabs */
2967 spa_config_exit(spa, SCL_STATE, spa);
2968 error = vdev_online(spa, guid, ZFS_ONLINE_EXPAND, &newstate);
2969 spa_config_enter(spa, SCL_STATE, spa, RW_READER);
2972 * If vdev_online returned an error or the underlying vdev_open
2973 * failed then we abort the expand. The only way to know that
2974 * vdev_open fails is by checking the returned newstate.
2976 if (error || newstate != VDEV_STATE_HEALTHY) {
2977 if (ztest_opts.zo_verbose >= 5) {
2978 (void) printf("Unable to expand vdev, state %llu, "
2979 "error %d\n", (u_longlong_t)newstate, error);
2983 ASSERT3U(newstate, ==, VDEV_STATE_HEALTHY);
2986 * Since we dropped the lock we need to ensure that we're
2987 * still talking to the original vdev. It's possible this
2988 * vdev may have been detached/replaced while we were
2989 * trying to online it.
2991 if (generation != spa->spa_config_generation) {
2992 if (ztest_opts.zo_verbose >= 5) {
2993 (void) printf("vdev configuration has changed, "
2994 "guid %llu, state %llu, expected gen %llu, "
2997 (u_longlong_t)tvd->vdev_state,
2998 (u_longlong_t)generation,
2999 (u_longlong_t)spa->spa_config_generation);
3007 * Traverse the vdev tree calling the supplied function.
3008 * We continue to walk the tree until we either have walked all
3009 * children or we receive a non-NULL return from the callback.
3010 * If a NULL callback is passed, then we just return back the first
3011 * leaf vdev we encounter.
3014 vdev_walk_tree(vdev_t *vd, vdev_t *(*func)(vdev_t *, void *), void *arg)
3016 if (vd->vdev_ops->vdev_op_leaf) {
3020 return (func(vd, arg));
3023 for (uint_t c = 0; c < vd->vdev_children; c++) {
3024 vdev_t *cvd = vd->vdev_child[c];
3025 if ((cvd = vdev_walk_tree(cvd, func, arg)) != NULL)
3032 * Verify that dynamic LUN growth works as expected.
3036 ztest_vdev_LUN_growth(ztest_ds_t *zd, uint64_t id)
3038 spa_t *spa = ztest_spa;
3040 metaslab_class_t *mc;
3041 metaslab_group_t *mg;
3042 size_t psize, newsize;
3044 uint64_t old_class_space, new_class_space, old_ms_count, new_ms_count;
3046 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
3047 spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3049 top = ztest_random_vdev_top(spa, B_TRUE);
3051 tvd = spa->spa_root_vdev->vdev_child[top];
3054 old_ms_count = tvd->vdev_ms_count;
3055 old_class_space = metaslab_class_get_space(mc);
3058 * Determine the size of the first leaf vdev associated with
3059 * our top-level device.
3061 vd = vdev_walk_tree(tvd, NULL, NULL);
3062 ASSERT3P(vd, !=, NULL);
3063 ASSERT(vd->vdev_ops->vdev_op_leaf);
3065 psize = vd->vdev_psize;
3068 * We only try to expand the vdev if it's healthy, less than 4x its
3069 * original size, and it has a valid psize.
3071 if (tvd->vdev_state != VDEV_STATE_HEALTHY ||
3072 psize == 0 || psize >= 4 * ztest_opts.zo_vdev_size) {
3073 spa_config_exit(spa, SCL_STATE, spa);
3074 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3078 newsize = psize + psize / 8;
3079 ASSERT3U(newsize, >, psize);
3081 if (ztest_opts.zo_verbose >= 6) {
3082 (void) printf("Expanding LUN %s from %lu to %lu\n",
3083 vd->vdev_path, (ulong_t)psize, (ulong_t)newsize);
3087 * Growing the vdev is a two step process:
3088 * 1). expand the physical size (i.e. relabel)
3089 * 2). online the vdev to create the new metaslabs
3091 if (vdev_walk_tree(tvd, grow_vdev, &newsize) != NULL ||
3092 vdev_walk_tree(tvd, online_vdev, NULL) != NULL ||
3093 tvd->vdev_state != VDEV_STATE_HEALTHY) {
3094 if (ztest_opts.zo_verbose >= 5) {
3095 (void) printf("Could not expand LUN because "
3096 "the vdev configuration changed.\n");
3098 spa_config_exit(spa, SCL_STATE, spa);
3099 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3103 spa_config_exit(spa, SCL_STATE, spa);
3106 * Expanding the LUN will update the config asynchronously,
3107 * thus we must wait for the async thread to complete any
3108 * pending tasks before proceeding.
3112 mutex_enter(&spa->spa_async_lock);
3113 done = (spa->spa_async_thread == NULL && !spa->spa_async_tasks);
3114 mutex_exit(&spa->spa_async_lock);
3117 txg_wait_synced(spa_get_dsl(spa), 0);
3118 (void) poll(NULL, 0, 100);
3121 spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3123 tvd = spa->spa_root_vdev->vdev_child[top];
3124 new_ms_count = tvd->vdev_ms_count;
3125 new_class_space = metaslab_class_get_space(mc);
3127 if (tvd->vdev_mg != mg || mg->mg_class != mc) {
3128 if (ztest_opts.zo_verbose >= 5) {
3129 (void) printf("Could not verify LUN expansion due to "
3130 "intervening vdev offline or remove.\n");
3132 spa_config_exit(spa, SCL_STATE, spa);
3133 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3138 * Make sure we were able to grow the vdev.
3140 if (new_ms_count <= old_ms_count)
3141 fatal(0, "LUN expansion failed: ms_count %llu <= %llu\n",
3142 old_ms_count, new_ms_count);
3145 * Make sure we were able to grow the pool.
3147 if (new_class_space <= old_class_space)
3148 fatal(0, "LUN expansion failed: class_space %llu <= %llu\n",
3149 old_class_space, new_class_space);
3151 if (ztest_opts.zo_verbose >= 5) {
3152 char oldnumbuf[6], newnumbuf[6];
3154 nicenum(old_class_space, oldnumbuf);
3155 nicenum(new_class_space, newnumbuf);
3156 (void) printf("%s grew from %s to %s\n",
3157 spa->spa_name, oldnumbuf, newnumbuf);
3160 spa_config_exit(spa, SCL_STATE, spa);
3161 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3165 * Verify that dmu_objset_{create,destroy,open,close} work as expected.
3169 ztest_objset_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
3172 * Create the objects common to all ztest datasets.
3174 VERIFY(zap_create_claim(os, ZTEST_DIROBJ,
3175 DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx) == 0);
3179 ztest_dataset_create(char *dsname)
3181 uint64_t zilset = ztest_random(100);
3182 int err = dmu_objset_create(dsname, DMU_OST_OTHER, 0,
3183 ztest_objset_create_cb, NULL);
3185 if (err || zilset < 80)
3188 if (ztest_opts.zo_verbose >= 6)
3189 (void) printf("Setting dataset %s to sync always\n", dsname);
3190 return (ztest_dsl_prop_set_uint64(dsname, ZFS_PROP_SYNC,
3191 ZFS_SYNC_ALWAYS, B_FALSE));
3196 ztest_objset_destroy_cb(const char *name, void *arg)
3199 dmu_object_info_t doi;
3203 * Verify that the dataset contains a directory object.
3205 VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_TRUE, FTAG, &os));
3206 error = dmu_object_info(os, ZTEST_DIROBJ, &doi);
3207 if (error != ENOENT) {
3208 /* We could have crashed in the middle of destroying it */
3210 ASSERT3U(doi.doi_type, ==, DMU_OT_ZAP_OTHER);
3211 ASSERT3S(doi.doi_physical_blocks_512, >=, 0);
3213 dmu_objset_disown(os, FTAG);
3216 * Destroy the dataset.
3218 if (strchr(name, '@') != NULL) {
3219 VERIFY0(dsl_destroy_snapshot(name, B_FALSE));
3221 VERIFY0(dsl_destroy_head(name));
3227 ztest_snapshot_create(char *osname, uint64_t id)
3229 char snapname[MAXNAMELEN];
3232 (void) snprintf(snapname, sizeof (snapname), "%llu", (u_longlong_t)id);
3234 error = dmu_objset_snapshot_one(osname, snapname);
3235 if (error == ENOSPC) {
3236 ztest_record_enospc(FTAG);
3239 if (error != 0 && error != EEXIST) {
3240 fatal(0, "ztest_snapshot_create(%s@%s) = %d", osname,
3247 ztest_snapshot_destroy(char *osname, uint64_t id)
3249 char snapname[MAXNAMELEN];
3252 (void) snprintf(snapname, MAXNAMELEN, "%s@%llu", osname,
3255 error = dsl_destroy_snapshot(snapname, B_FALSE);
3256 if (error != 0 && error != ENOENT)
3257 fatal(0, "ztest_snapshot_destroy(%s) = %d", snapname, error);
3263 ztest_dmu_objset_create_destroy(ztest_ds_t *zd, uint64_t id)
3269 char name[MAXNAMELEN];
3272 (void) rw_rdlock(&ztest_name_lock);
3274 (void) snprintf(name, MAXNAMELEN, "%s/temp_%llu",
3275 ztest_opts.zo_pool, (u_longlong_t)id);
3278 * If this dataset exists from a previous run, process its replay log
3279 * half of the time. If we don't replay it, then dmu_objset_destroy()
3280 * (invoked from ztest_objset_destroy_cb()) should just throw it away.
3282 if (ztest_random(2) == 0 &&
3283 dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os) == 0) {
3284 ztest_zd_init(&zdtmp, NULL, os);
3285 zil_replay(os, &zdtmp, ztest_replay_vector);
3286 ztest_zd_fini(&zdtmp);
3287 dmu_objset_disown(os, FTAG);
3291 * There may be an old instance of the dataset we're about to
3292 * create lying around from a previous run. If so, destroy it
3293 * and all of its snapshots.
3295 (void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
3296 DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
3299 * Verify that the destroyed dataset is no longer in the namespace.
3301 VERIFY3U(ENOENT, ==, dmu_objset_own(name, DMU_OST_OTHER, B_TRUE,
3305 * Verify that we can create a new dataset.
3307 error = ztest_dataset_create(name);
3309 if (error == ENOSPC) {
3310 ztest_record_enospc(FTAG);
3311 (void) rw_unlock(&ztest_name_lock);
3314 fatal(0, "dmu_objset_create(%s) = %d", name, error);
3317 VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os));
3319 ztest_zd_init(&zdtmp, NULL, os);
3322 * Open the intent log for it.
3324 zilog = zil_open(os, ztest_get_data);
3327 * Put some objects in there, do a little I/O to them,
3328 * and randomly take a couple of snapshots along the way.
3330 iters = ztest_random(5);
3331 for (int i = 0; i < iters; i++) {
3332 ztest_dmu_object_alloc_free(&zdtmp, id);
3333 if (ztest_random(iters) == 0)
3334 (void) ztest_snapshot_create(name, i);
3338 * Verify that we cannot create an existing dataset.
3340 VERIFY3U(EEXIST, ==,
3341 dmu_objset_create(name, DMU_OST_OTHER, 0, NULL, NULL));
3344 * Verify that we can hold an objset that is also owned.
3346 VERIFY3U(0, ==, dmu_objset_hold(name, FTAG, &os2));
3347 dmu_objset_rele(os2, FTAG);
3350 * Verify that we cannot own an objset that is already owned.
3353 dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os2));
3356 dmu_objset_disown(os, FTAG);
3357 ztest_zd_fini(&zdtmp);
3359 (void) rw_unlock(&ztest_name_lock);
3363 * Verify that dmu_snapshot_{create,destroy,open,close} work as expected.
3366 ztest_dmu_snapshot_create_destroy(ztest_ds_t *zd, uint64_t id)
3368 (void) rw_rdlock(&ztest_name_lock);
3369 (void) ztest_snapshot_destroy(zd->zd_name, id);
3370 (void) ztest_snapshot_create(zd->zd_name, id);
3371 (void) rw_unlock(&ztest_name_lock);
3375 * Cleanup non-standard snapshots and clones.
3378 ztest_dsl_dataset_cleanup(char *osname, uint64_t id)
3380 char snap1name[MAXNAMELEN];
3381 char clone1name[MAXNAMELEN];
3382 char snap2name[MAXNAMELEN];
3383 char clone2name[MAXNAMELEN];
3384 char snap3name[MAXNAMELEN];
3387 (void) snprintf(snap1name, MAXNAMELEN, "%s@s1_%llu", osname, id);
3388 (void) snprintf(clone1name, MAXNAMELEN, "%s/c1_%llu", osname, id);
3389 (void) snprintf(snap2name, MAXNAMELEN, "%s@s2_%llu", clone1name, id);
3390 (void) snprintf(clone2name, MAXNAMELEN, "%s/c2_%llu", osname, id);
3391 (void) snprintf(snap3name, MAXNAMELEN, "%s@s3_%llu", clone1name, id);
3393 error = dsl_destroy_head(clone2name);
3394 if (error && error != ENOENT)
3395 fatal(0, "dsl_destroy_head(%s) = %d", clone2name, error);
3396 error = dsl_destroy_snapshot(snap3name, B_FALSE);
3397 if (error && error != ENOENT)
3398 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap3name, error);
3399 error = dsl_destroy_snapshot(snap2name, B_FALSE);
3400 if (error && error != ENOENT)
3401 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap2name, error);
3402 error = dsl_destroy_head(clone1name);
3403 if (error && error != ENOENT)
3404 fatal(0, "dsl_destroy_head(%s) = %d", clone1name, error);
3405 error = dsl_destroy_snapshot(snap1name, B_FALSE);
3406 if (error && error != ENOENT)
3407 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap1name, error);
3411 * Verify dsl_dataset_promote handles EBUSY
3414 ztest_dsl_dataset_promote_busy(ztest_ds_t *zd, uint64_t id)
3417 char snap1name[MAXNAMELEN];
3418 char clone1name[MAXNAMELEN];
3419 char snap2name[MAXNAMELEN];
3420 char clone2name[MAXNAMELEN];
3421 char snap3name[MAXNAMELEN];
3422 char *osname = zd->zd_name;
3425 (void) rw_rdlock(&ztest_name_lock);
3427 ztest_dsl_dataset_cleanup(osname, id);
3429 (void) snprintf(snap1name, MAXNAMELEN, "%s@s1_%llu", osname, id);
3430 (void) snprintf(clone1name, MAXNAMELEN, "%s/c1_%llu", osname, id);
3431 (void) snprintf(snap2name, MAXNAMELEN, "%s@s2_%llu", clone1name, id);
3432 (void) snprintf(clone2name, MAXNAMELEN, "%s/c2_%llu", osname, id);
3433 (void) snprintf(snap3name, MAXNAMELEN, "%s@s3_%llu", clone1name, id);
3435 error = dmu_objset_snapshot_one(osname, strchr(snap1name, '@') + 1);
3436 if (error && error != EEXIST) {
3437 if (error == ENOSPC) {
3438 ztest_record_enospc(FTAG);
3441 fatal(0, "dmu_take_snapshot(%s) = %d", snap1name, error);
3444 error = dmu_objset_clone(clone1name, snap1name);
3446 if (error == ENOSPC) {
3447 ztest_record_enospc(FTAG);
3450 fatal(0, "dmu_objset_create(%s) = %d", clone1name, error);
3453 error = dmu_objset_snapshot_one(clone1name, strchr(snap2name, '@') + 1);
3454 if (error && error != EEXIST) {
3455 if (error == ENOSPC) {
3456 ztest_record_enospc(FTAG);
3459 fatal(0, "dmu_open_snapshot(%s) = %d", snap2name, error);
3462 error = dmu_objset_snapshot_one(clone1name, strchr(snap3name, '@') + 1);
3463 if (error && error != EEXIST) {
3464 if (error == ENOSPC) {
3465 ztest_record_enospc(FTAG);
3468 fatal(0, "dmu_open_snapshot(%s) = %d", snap3name, error);
3471 error = dmu_objset_clone(clone2name, snap3name);
3473 if (error == ENOSPC) {
3474 ztest_record_enospc(FTAG);
3477 fatal(0, "dmu_objset_create(%s) = %d", clone2name, error);
3480 error = dmu_objset_own(snap2name, DMU_OST_ANY, B_TRUE, FTAG, &os);
3482 fatal(0, "dmu_objset_own(%s) = %d", snap2name, error);
3483 error = dsl_dataset_promote(clone2name, NULL);
3484 if (error == ENOSPC) {
3485 dmu_objset_disown(os, FTAG);
3486 ztest_record_enospc(FTAG);
3490 fatal(0, "dsl_dataset_promote(%s), %d, not EBUSY", clone2name,
3492 dmu_objset_disown(os, FTAG);
3495 ztest_dsl_dataset_cleanup(osname, id);
3497 (void) rw_unlock(&ztest_name_lock);
3501 * Verify that dmu_object_{alloc,free} work as expected.
3504 ztest_dmu_object_alloc_free(ztest_ds_t *zd, uint64_t id)
3507 int batchsize = sizeof (od) / sizeof (od[0]);
3509 for (int b = 0; b < batchsize; b++)
3510 ztest_od_init(&od[b], id, FTAG, b, DMU_OT_UINT64_OTHER, 0, 0);
3513 * Destroy the previous batch of objects, create a new batch,
3514 * and do some I/O on the new objects.
3516 if (ztest_object_init(zd, od, sizeof (od), B_TRUE) != 0)
3519 while (ztest_random(4 * batchsize) != 0)
3520 ztest_io(zd, od[ztest_random(batchsize)].od_object,
3521 ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
3525 * Verify that dmu_{read,write} work as expected.
3528 ztest_dmu_read_write(ztest_ds_t *zd, uint64_t id)
3530 objset_t *os = zd->zd_os;
3533 int i, freeit, error;
3535 bufwad_t *packbuf, *bigbuf, *pack, *bigH, *bigT;
3536 uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
3537 uint64_t chunksize = (1000 + ztest_random(1000)) * sizeof (uint64_t);
3538 uint64_t regions = 997;
3539 uint64_t stride = 123456789ULL;
3540 uint64_t width = 40;
3541 int free_percent = 5;
3544 * This test uses two objects, packobj and bigobj, that are always
3545 * updated together (i.e. in the same tx) so that their contents are
3546 * in sync and can be compared. Their contents relate to each other
3547 * in a simple way: packobj is a dense array of 'bufwad' structures,
3548 * while bigobj is a sparse array of the same bufwads. Specifically,
3549 * for any index n, there are three bufwads that should be identical:
3551 * packobj, at offset n * sizeof (bufwad_t)
3552 * bigobj, at the head of the nth chunk
3553 * bigobj, at the tail of the nth chunk
3555 * The chunk size is arbitrary. It doesn't have to be a power of two,
3556 * and it doesn't have any relation to the object blocksize.
3557 * The only requirement is that it can hold at least two bufwads.
3559 * Normally, we write the bufwad to each of these locations.
3560 * However, free_percent of the time we instead write zeroes to
3561 * packobj and perform a dmu_free_range() on bigobj. By comparing
3562 * bigobj to packobj, we can verify that the DMU is correctly
3563 * tracking which parts of an object are allocated and free,
3564 * and that the contents of the allocated blocks are correct.
3568 * Read the directory info. If it's the first time, set things up.
3570 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, chunksize);
3571 ztest_od_init(&od[1], id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize);
3573 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
3576 bigobj = od[0].od_object;
3577 packobj = od[1].od_object;
3578 chunksize = od[0].od_gen;
3579 ASSERT(chunksize == od[1].od_gen);
3582 * Prefetch a random chunk of the big object.
3583 * Our aim here is to get some async reads in flight
3584 * for blocks that we may free below; the DMU should
3585 * handle this race correctly.
3587 n = ztest_random(regions) * stride + ztest_random(width);
3588 s = 1 + ztest_random(2 * width - 1);
3589 dmu_prefetch(os, bigobj, n * chunksize, s * chunksize);
3592 * Pick a random index and compute the offsets into packobj and bigobj.
3594 n = ztest_random(regions) * stride + ztest_random(width);
3595 s = 1 + ztest_random(width - 1);
3597 packoff = n * sizeof (bufwad_t);
3598 packsize = s * sizeof (bufwad_t);
3600 bigoff = n * chunksize;
3601 bigsize = s * chunksize;
3603 packbuf = umem_alloc(packsize, UMEM_NOFAIL);
3604 bigbuf = umem_alloc(bigsize, UMEM_NOFAIL);
3607 * free_percent of the time, free a range of bigobj rather than
3610 freeit = (ztest_random(100) < free_percent);
3613 * Read the current contents of our objects.
3615 error = dmu_read(os, packobj, packoff, packsize, packbuf,
3618 error = dmu_read(os, bigobj, bigoff, bigsize, bigbuf,
3623 * Get a tx for the mods to both packobj and bigobj.
3625 tx = dmu_tx_create(os);
3627 dmu_tx_hold_write(tx, packobj, packoff, packsize);
3630 dmu_tx_hold_free(tx, bigobj, bigoff, bigsize);
3632 dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
3634 /* This accounts for setting the checksum/compression. */
3635 dmu_tx_hold_bonus(tx, bigobj);
3637 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3639 umem_free(packbuf, packsize);
3640 umem_free(bigbuf, bigsize);
3644 enum zio_checksum cksum;
3646 cksum = (enum zio_checksum)
3647 ztest_random_dsl_prop(ZFS_PROP_CHECKSUM);
3648 } while (cksum >= ZIO_CHECKSUM_LEGACY_FUNCTIONS);
3649 dmu_object_set_checksum(os, bigobj, cksum, tx);
3651 enum zio_compress comp;
3653 comp = (enum zio_compress)
3654 ztest_random_dsl_prop(ZFS_PROP_COMPRESSION);
3655 } while (comp >= ZIO_COMPRESS_LEGACY_FUNCTIONS);
3656 dmu_object_set_compress(os, bigobj, comp, tx);
3659 * For each index from n to n + s, verify that the existing bufwad
3660 * in packobj matches the bufwads at the head and tail of the
3661 * corresponding chunk in bigobj. Then update all three bufwads
3662 * with the new values we want to write out.
3664 for (i = 0; i < s; i++) {
3666 pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
3668 bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
3670 bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
3672 ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
3673 ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
3675 if (pack->bw_txg > txg)
3676 fatal(0, "future leak: got %llx, open txg is %llx",
3679 if (pack->bw_data != 0 && pack->bw_index != n + i)
3680 fatal(0, "wrong index: got %llx, wanted %llx+%llx",
3681 pack->bw_index, n, i);
3683 if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
3684 fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
3686 if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
3687 fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
3690 bzero(pack, sizeof (bufwad_t));
3692 pack->bw_index = n + i;
3694 pack->bw_data = 1 + ztest_random(-2ULL);
3701 * We've verified all the old bufwads, and made new ones.
3702 * Now write them out.
3704 dmu_write(os, packobj, packoff, packsize, packbuf, tx);
3707 if (ztest_opts.zo_verbose >= 7) {
3708 (void) printf("freeing offset %llx size %llx"
3710 (u_longlong_t)bigoff,
3711 (u_longlong_t)bigsize,
3714 VERIFY(0 == dmu_free_range(os, bigobj, bigoff, bigsize, tx));
3716 if (ztest_opts.zo_verbose >= 7) {
3717 (void) printf("writing offset %llx size %llx"
3719 (u_longlong_t)bigoff,
3720 (u_longlong_t)bigsize,
3723 dmu_write(os, bigobj, bigoff, bigsize, bigbuf, tx);
3729 * Sanity check the stuff we just wrote.
3732 void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
3733 void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
3735 VERIFY(0 == dmu_read(os, packobj, packoff,
3736 packsize, packcheck, DMU_READ_PREFETCH));
3737 VERIFY(0 == dmu_read(os, bigobj, bigoff,
3738 bigsize, bigcheck, DMU_READ_PREFETCH));
3740 ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
3741 ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
3743 umem_free(packcheck, packsize);
3744 umem_free(bigcheck, bigsize);
3747 umem_free(packbuf, packsize);
3748 umem_free(bigbuf, bigsize);
3752 compare_and_update_pbbufs(uint64_t s, bufwad_t *packbuf, bufwad_t *bigbuf,
3753 uint64_t bigsize, uint64_t n, uint64_t chunksize, uint64_t txg)
3761 * For each index from n to n + s, verify that the existing bufwad
3762 * in packobj matches the bufwads at the head and tail of the
3763 * corresponding chunk in bigobj. Then update all three bufwads
3764 * with the new values we want to write out.
3766 for (i = 0; i < s; i++) {
3768 pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
3770 bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
3772 bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
3774 ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
3775 ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
3777 if (pack->bw_txg > txg)
3778 fatal(0, "future leak: got %llx, open txg is %llx",
3781 if (pack->bw_data != 0 && pack->bw_index != n + i)
3782 fatal(0, "wrong index: got %llx, wanted %llx+%llx",
3783 pack->bw_index, n, i);
3785 if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
3786 fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
3788 if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
3789 fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
3791 pack->bw_index = n + i;
3793 pack->bw_data = 1 + ztest_random(-2ULL);
3801 ztest_dmu_read_write_zcopy(ztest_ds_t *zd, uint64_t id)
3803 objset_t *os = zd->zd_os;
3809 bufwad_t *packbuf, *bigbuf;
3810 uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
3811 uint64_t blocksize = ztest_random_blocksize();
3812 uint64_t chunksize = blocksize;
3813 uint64_t regions = 997;
3814 uint64_t stride = 123456789ULL;
3816 dmu_buf_t *bonus_db;
3817 arc_buf_t **bigbuf_arcbufs;
3818 dmu_object_info_t doi;
3821 * This test uses two objects, packobj and bigobj, that are always
3822 * updated together (i.e. in the same tx) so that their contents are
3823 * in sync and can be compared. Their contents relate to each other
3824 * in a simple way: packobj is a dense array of 'bufwad' structures,
3825 * while bigobj is a sparse array of the same bufwads. Specifically,
3826 * for any index n, there are three bufwads that should be identical:
3828 * packobj, at offset n * sizeof (bufwad_t)
3829 * bigobj, at the head of the nth chunk
3830 * bigobj, at the tail of the nth chunk
3832 * The chunk size is set equal to bigobj block size so that
3833 * dmu_assign_arcbuf() can be tested for object updates.
3837 * Read the directory info. If it's the first time, set things up.
3839 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
3840 ztest_od_init(&od[1], id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize);
3842 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
3845 bigobj = od[0].od_object;
3846 packobj = od[1].od_object;
3847 blocksize = od[0].od_blocksize;
3848 chunksize = blocksize;
3849 ASSERT(chunksize == od[1].od_gen);
3851 VERIFY(dmu_object_info(os, bigobj, &doi) == 0);
3852 VERIFY(ISP2(doi.doi_data_block_size));
3853 VERIFY(chunksize == doi.doi_data_block_size);
3854 VERIFY(chunksize >= 2 * sizeof (bufwad_t));
3857 * Pick a random index and compute the offsets into packobj and bigobj.
3859 n = ztest_random(regions) * stride + ztest_random(width);
3860 s = 1 + ztest_random(width - 1);
3862 packoff = n * sizeof (bufwad_t);
3863 packsize = s * sizeof (bufwad_t);
3865 bigoff = n * chunksize;
3866 bigsize = s * chunksize;
3868 packbuf = umem_zalloc(packsize, UMEM_NOFAIL);
3869 bigbuf = umem_zalloc(bigsize, UMEM_NOFAIL);
3871 VERIFY3U(0, ==, dmu_bonus_hold(os, bigobj, FTAG, &bonus_db));
3873 bigbuf_arcbufs = umem_zalloc(2 * s * sizeof (arc_buf_t *), UMEM_NOFAIL);
3876 * Iteration 0 test zcopy for DB_UNCACHED dbufs.
3877 * Iteration 1 test zcopy to already referenced dbufs.
3878 * Iteration 2 test zcopy to dirty dbuf in the same txg.
3879 * Iteration 3 test zcopy to dbuf dirty in previous txg.
3880 * Iteration 4 test zcopy when dbuf is no longer dirty.
3881 * Iteration 5 test zcopy when it can't be done.
3882 * Iteration 6 one more zcopy write.
3884 for (i = 0; i < 7; i++) {
3889 * In iteration 5 (i == 5) use arcbufs
3890 * that don't match bigobj blksz to test
3891 * dmu_assign_arcbuf() when it can't directly
3892 * assign an arcbuf to a dbuf.
3894 for (j = 0; j < s; j++) {
3897 dmu_request_arcbuf(bonus_db, chunksize);
3899 bigbuf_arcbufs[2 * j] =
3900 dmu_request_arcbuf(bonus_db, chunksize / 2);
3901 bigbuf_arcbufs[2 * j + 1] =
3902 dmu_request_arcbuf(bonus_db, chunksize / 2);
3907 * Get a tx for the mods to both packobj and bigobj.
3909 tx = dmu_tx_create(os);
3911 dmu_tx_hold_write(tx, packobj, packoff, packsize);
3912 dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
3914 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3916 umem_free(packbuf, packsize);
3917 umem_free(bigbuf, bigsize);
3918 for (j = 0; j < s; j++) {
3920 dmu_return_arcbuf(bigbuf_arcbufs[j]);
3923 bigbuf_arcbufs[2 * j]);
3925 bigbuf_arcbufs[2 * j + 1]);
3928 umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
3929 dmu_buf_rele(bonus_db, FTAG);
3934 * 50% of the time don't read objects in the 1st iteration to
3935 * test dmu_assign_arcbuf() for the case when there're no
3936 * existing dbufs for the specified offsets.
3938 if (i != 0 || ztest_random(2) != 0) {
3939 error = dmu_read(os, packobj, packoff,
3940 packsize, packbuf, DMU_READ_PREFETCH);
3942 error = dmu_read(os, bigobj, bigoff, bigsize,
3943 bigbuf, DMU_READ_PREFETCH);
3946 compare_and_update_pbbufs(s, packbuf, bigbuf, bigsize,
3950 * We've verified all the old bufwads, and made new ones.
3951 * Now write them out.
3953 dmu_write(os, packobj, packoff, packsize, packbuf, tx);
3954 if (ztest_opts.zo_verbose >= 7) {
3955 (void) printf("writing offset %llx size %llx"
3957 (u_longlong_t)bigoff,
3958 (u_longlong_t)bigsize,
3961 for (off = bigoff, j = 0; j < s; j++, off += chunksize) {
3964 bcopy((caddr_t)bigbuf + (off - bigoff),
3965 bigbuf_arcbufs[j]->b_data, chunksize);
3967 bcopy((caddr_t)bigbuf + (off - bigoff),
3968 bigbuf_arcbufs[2 * j]->b_data,
3970 bcopy((caddr_t)bigbuf + (off - bigoff) +
3972 bigbuf_arcbufs[2 * j + 1]->b_data,
3977 VERIFY(dmu_buf_hold(os, bigobj, off,
3978 FTAG, &dbt, DMU_READ_NO_PREFETCH) == 0);
3981 dmu_assign_arcbuf(bonus_db, off,
3982 bigbuf_arcbufs[j], tx);
3984 dmu_assign_arcbuf(bonus_db, off,
3985 bigbuf_arcbufs[2 * j], tx);
3986 dmu_assign_arcbuf(bonus_db,
3987 off + chunksize / 2,
3988 bigbuf_arcbufs[2 * j + 1], tx);
3991 dmu_buf_rele(dbt, FTAG);
3997 * Sanity check the stuff we just wrote.
4000 void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
4001 void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
4003 VERIFY(0 == dmu_read(os, packobj, packoff,
4004 packsize, packcheck, DMU_READ_PREFETCH));
4005 VERIFY(0 == dmu_read(os, bigobj, bigoff,
4006 bigsize, bigcheck, DMU_READ_PREFETCH));
4008 ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
4009 ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
4011 umem_free(packcheck, packsize);
4012 umem_free(bigcheck, bigsize);
4015 txg_wait_open(dmu_objset_pool(os), 0);
4016 } else if (i == 3) {
4017 txg_wait_synced(dmu_objset_pool(os), 0);
4021 dmu_buf_rele(bonus_db, FTAG);
4022 umem_free(packbuf, packsize);
4023 umem_free(bigbuf, bigsize);
4024 umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
4029 ztest_dmu_write_parallel(ztest_ds_t *zd, uint64_t id)
4032 uint64_t offset = (1ULL << (ztest_random(20) + 43)) +
4033 (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
4036 * Have multiple threads write to large offsets in an object
4037 * to verify that parallel writes to an object -- even to the
4038 * same blocks within the object -- doesn't cause any trouble.
4040 ztest_od_init(&od[0], ID_PARALLEL, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
4042 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4045 while (ztest_random(10) != 0)
4046 ztest_io(zd, od[0].od_object, offset);
4050 ztest_dmu_prealloc(ztest_ds_t *zd, uint64_t id)
4053 uint64_t offset = (1ULL << (ztest_random(4) + SPA_MAXBLOCKSHIFT)) +
4054 (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
4055 uint64_t count = ztest_random(20) + 1;
4056 uint64_t blocksize = ztest_random_blocksize();
4059 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
4061 if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4064 if (ztest_truncate(zd, od[0].od_object, offset, count * blocksize) != 0)
4067 ztest_prealloc(zd, od[0].od_object, offset, count * blocksize);
4069 data = umem_zalloc(blocksize, UMEM_NOFAIL);
4071 while (ztest_random(count) != 0) {
4072 uint64_t randoff = offset + (ztest_random(count) * blocksize);
4073 if (ztest_write(zd, od[0].od_object, randoff, blocksize,
4076 while (ztest_random(4) != 0)
4077 ztest_io(zd, od[0].od_object, randoff);
4080 umem_free(data, blocksize);
4084 * Verify that zap_{create,destroy,add,remove,update} work as expected.
4086 #define ZTEST_ZAP_MIN_INTS 1
4087 #define ZTEST_ZAP_MAX_INTS 4
4088 #define ZTEST_ZAP_MAX_PROPS 1000
4091 ztest_zap(ztest_ds_t *zd, uint64_t id)
4093 objset_t *os = zd->zd_os;
4096 uint64_t txg, last_txg;
4097 uint64_t value[ZTEST_ZAP_MAX_INTS];
4098 uint64_t zl_ints, zl_intsize, prop;
4101 char propname[100], txgname[100];
4103 char *hc[2] = { "s.acl.h", ".s.open.h.hyLZlg" };
4105 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0);
4107 if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4110 object = od[0].od_object;
4113 * Generate a known hash collision, and verify that
4114 * we can lookup and remove both entries.
4116 tx = dmu_tx_create(os);
4117 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4118 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4121 for (i = 0; i < 2; i++) {
4123 VERIFY3U(0, ==, zap_add(os, object, hc[i], sizeof (uint64_t),
4126 for (i = 0; i < 2; i++) {
4127 VERIFY3U(EEXIST, ==, zap_add(os, object, hc[i],
4128 sizeof (uint64_t), 1, &value[i], tx));
4130 zap_length(os, object, hc[i], &zl_intsize, &zl_ints));
4131 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4132 ASSERT3U(zl_ints, ==, 1);
4134 for (i = 0; i < 2; i++) {
4135 VERIFY3U(0, ==, zap_remove(os, object, hc[i], tx));
4140 * Generate a buch of random entries.
4142 ints = MAX(ZTEST_ZAP_MIN_INTS, object % ZTEST_ZAP_MAX_INTS);
4144 prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
4145 (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
4146 (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
4147 bzero(value, sizeof (value));
4151 * If these zap entries already exist, validate their contents.
4153 error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
4155 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4156 ASSERT3U(zl_ints, ==, 1);
4158 VERIFY(zap_lookup(os, object, txgname, zl_intsize,
4159 zl_ints, &last_txg) == 0);
4161 VERIFY(zap_length(os, object, propname, &zl_intsize,
4164 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4165 ASSERT3U(zl_ints, ==, ints);
4167 VERIFY(zap_lookup(os, object, propname, zl_intsize,
4168 zl_ints, value) == 0);
4170 for (i = 0; i < ints; i++) {
4171 ASSERT3U(value[i], ==, last_txg + object + i);
4174 ASSERT3U(error, ==, ENOENT);
4178 * Atomically update two entries in our zap object.
4179 * The first is named txg_%llu, and contains the txg
4180 * in which the property was last updated. The second
4181 * is named prop_%llu, and the nth element of its value
4182 * should be txg + object + n.
4184 tx = dmu_tx_create(os);
4185 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4186 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4191 fatal(0, "zap future leak: old %llu new %llu", last_txg, txg);
4193 for (i = 0; i < ints; i++)
4194 value[i] = txg + object + i;
4196 VERIFY3U(0, ==, zap_update(os, object, txgname, sizeof (uint64_t),
4198 VERIFY3U(0, ==, zap_update(os, object, propname, sizeof (uint64_t),
4204 * Remove a random pair of entries.
4206 prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
4207 (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
4208 (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
4210 error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
4212 if (error == ENOENT)
4217 tx = dmu_tx_create(os);
4218 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4219 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4222 VERIFY3U(0, ==, zap_remove(os, object, txgname, tx));
4223 VERIFY3U(0, ==, zap_remove(os, object, propname, tx));
4228 * Testcase to test the upgrading of a microzap to fatzap.
4231 ztest_fzap(ztest_ds_t *zd, uint64_t id)
4233 objset_t *os = zd->zd_os;
4235 uint64_t object, txg;
4237 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0);
4239 if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4242 object = od[0].od_object;
4245 * Add entries to this ZAP and make sure it spills over
4246 * and gets upgraded to a fatzap. Also, since we are adding
4247 * 2050 entries we should see ptrtbl growth and leaf-block split.
4249 for (int i = 0; i < 2050; i++) {
4250 char name[MAXNAMELEN];
4255 (void) snprintf(name, sizeof (name), "fzap-%llu-%llu",
4258 tx = dmu_tx_create(os);
4259 dmu_tx_hold_zap(tx, object, B_TRUE, name);
4260 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4263 error = zap_add(os, object, name, sizeof (uint64_t), 1,
4265 ASSERT(error == 0 || error == EEXIST);
4272 ztest_zap_parallel(ztest_ds_t *zd, uint64_t id)
4274 objset_t *os = zd->zd_os;
4276 uint64_t txg, object, count, wsize, wc, zl_wsize, zl_wc;
4278 int i, namelen, error;
4279 int micro = ztest_random(2);
4280 char name[20], string_value[20];
4283 ztest_od_init(&od[0], ID_PARALLEL, FTAG, micro, DMU_OT_ZAP_OTHER, 0, 0);
4285 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4288 object = od[0].od_object;
4291 * Generate a random name of the form 'xxx.....' where each
4292 * x is a random printable character and the dots are dots.
4293 * There are 94 such characters, and the name length goes from
4294 * 6 to 20, so there are 94^3 * 15 = 12,458,760 possible names.
4296 namelen = ztest_random(sizeof (name) - 5) + 5 + 1;
4298 for (i = 0; i < 3; i++)
4299 name[i] = '!' + ztest_random('~' - '!' + 1);
4300 for (; i < namelen - 1; i++)
4304 if ((namelen & 1) || micro) {
4305 wsize = sizeof (txg);
4311 data = string_value;
4315 VERIFY0(zap_count(os, object, &count));
4316 ASSERT(count != -1ULL);
4319 * Select an operation: length, lookup, add, update, remove.
4321 i = ztest_random(5);
4324 tx = dmu_tx_create(os);
4325 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4326 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4329 bcopy(name, string_value, namelen);
4333 bzero(string_value, namelen);
4339 error = zap_length(os, object, name, &zl_wsize, &zl_wc);
4341 ASSERT3U(wsize, ==, zl_wsize);
4342 ASSERT3U(wc, ==, zl_wc);
4344 ASSERT3U(error, ==, ENOENT);
4349 error = zap_lookup(os, object, name, wsize, wc, data);
4351 if (data == string_value &&
4352 bcmp(name, data, namelen) != 0)
4353 fatal(0, "name '%s' != val '%s' len %d",
4354 name, data, namelen);
4356 ASSERT3U(error, ==, ENOENT);
4361 error = zap_add(os, object, name, wsize, wc, data, tx);
4362 ASSERT(error == 0 || error == EEXIST);
4366 VERIFY(zap_update(os, object, name, wsize, wc, data, tx) == 0);
4370 error = zap_remove(os, object, name, tx);
4371 ASSERT(error == 0 || error == ENOENT);
4380 * Commit callback data.
4382 typedef struct ztest_cb_data {
4383 list_node_t zcd_node;
4385 int zcd_expected_err;
4386 boolean_t zcd_added;
4387 boolean_t zcd_called;
4391 /* This is the actual commit callback function */
4393 ztest_commit_callback(void *arg, int error)
4395 ztest_cb_data_t *data = arg;
4396 uint64_t synced_txg;
4398 VERIFY(data != NULL);
4399 VERIFY3S(data->zcd_expected_err, ==, error);
4400 VERIFY(!data->zcd_called);
4402 synced_txg = spa_last_synced_txg(data->zcd_spa);
4403 if (data->zcd_txg > synced_txg)
4404 fatal(0, "commit callback of txg %" PRIu64 " called prematurely"
4405 ", last synced txg = %" PRIu64 "\n", data->zcd_txg,
4408 data->zcd_called = B_TRUE;
4410 if (error == ECANCELED) {
4411 ASSERT0(data->zcd_txg);
4412 ASSERT(!data->zcd_added);
4415 * The private callback data should be destroyed here, but
4416 * since we are going to check the zcd_called field after
4417 * dmu_tx_abort(), we will destroy it there.
4422 /* Was this callback added to the global callback list? */
4423 if (!data->zcd_added)
4426 ASSERT3U(data->zcd_txg, !=, 0);
4428 /* Remove our callback from the list */
4429 (void) mutex_lock(&zcl.zcl_callbacks_lock);
4430 list_remove(&zcl.zcl_callbacks, data);
4431 (void) mutex_unlock(&zcl.zcl_callbacks_lock);
4434 umem_free(data, sizeof (ztest_cb_data_t));
4437 /* Allocate and initialize callback data structure */
4438 static ztest_cb_data_t *
4439 ztest_create_cb_data(objset_t *os, uint64_t txg)
4441 ztest_cb_data_t *cb_data;
4443 cb_data = umem_zalloc(sizeof (ztest_cb_data_t), UMEM_NOFAIL);
4445 cb_data->zcd_txg = txg;
4446 cb_data->zcd_spa = dmu_objset_spa(os);
4452 * If a number of txgs equal to this threshold have been created after a commit
4453 * callback has been registered but not called, then we assume there is an
4454 * implementation bug.
4456 #define ZTEST_COMMIT_CALLBACK_THRESH (TXG_CONCURRENT_STATES + 2)
4459 * Commit callback test.
4462 ztest_dmu_commit_callbacks(ztest_ds_t *zd, uint64_t id)
4464 objset_t *os = zd->zd_os;
4467 ztest_cb_data_t *cb_data[3], *tmp_cb;
4468 uint64_t old_txg, txg;
4471 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
4473 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4476 tx = dmu_tx_create(os);
4478 cb_data[0] = ztest_create_cb_data(os, 0);
4479 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[0]);
4481 dmu_tx_hold_write(tx, od[0].od_object, 0, sizeof (uint64_t));
4483 /* Every once in a while, abort the transaction on purpose */
4484 if (ztest_random(100) == 0)
4488 error = dmu_tx_assign(tx, TXG_NOWAIT);
4490 txg = error ? 0 : dmu_tx_get_txg(tx);
4492 cb_data[0]->zcd_txg = txg;
4493 cb_data[1] = ztest_create_cb_data(os, txg);
4494 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[1]);
4498 * It's not a strict requirement to call the registered
4499 * callbacks from inside dmu_tx_abort(), but that's what
4500 * it's supposed to happen in the current implementation
4501 * so we will check for that.
4503 for (i = 0; i < 2; i++) {
4504 cb_data[i]->zcd_expected_err = ECANCELED;
4505 VERIFY(!cb_data[i]->zcd_called);
4510 for (i = 0; i < 2; i++) {
4511 VERIFY(cb_data[i]->zcd_called);
4512 umem_free(cb_data[i], sizeof (ztest_cb_data_t));
4518 cb_data[2] = ztest_create_cb_data(os, txg);
4519 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[2]);
4522 * Read existing data to make sure there isn't a future leak.
4524 VERIFY(0 == dmu_read(os, od[0].od_object, 0, sizeof (uint64_t),
4525 &old_txg, DMU_READ_PREFETCH));
4528 fatal(0, "future leak: got %" PRIu64 ", open txg is %" PRIu64,
4531 dmu_write(os, od[0].od_object, 0, sizeof (uint64_t), &txg, tx);
4533 (void) mutex_lock(&zcl.zcl_callbacks_lock);
4536 * Since commit callbacks don't have any ordering requirement and since
4537 * it is theoretically possible for a commit callback to be called
4538 * after an arbitrary amount of time has elapsed since its txg has been
4539 * synced, it is difficult to reliably determine whether a commit
4540 * callback hasn't been called due to high load or due to a flawed
4543 * In practice, we will assume that if after a certain number of txgs a
4544 * commit callback hasn't been called, then most likely there's an
4545 * implementation bug..
4547 tmp_cb = list_head(&zcl.zcl_callbacks);
4548 if (tmp_cb != NULL &&
4549 (txg - ZTEST_COMMIT_CALLBACK_THRESH) > tmp_cb->zcd_txg) {
4550 fatal(0, "Commit callback threshold exceeded, oldest txg: %"
4551 PRIu64 ", open txg: %" PRIu64 "\n", tmp_cb->zcd_txg, txg);
4555 * Let's find the place to insert our callbacks.
4557 * Even though the list is ordered by txg, it is possible for the
4558 * insertion point to not be the end because our txg may already be
4559 * quiescing at this point and other callbacks in the open txg
4560 * (from other objsets) may have sneaked in.
4562 tmp_cb = list_tail(&zcl.zcl_callbacks);
4563 while (tmp_cb != NULL && tmp_cb->zcd_txg > txg)
4564 tmp_cb = list_prev(&zcl.zcl_callbacks, tmp_cb);
4566 /* Add the 3 callbacks to the list */
4567 for (i = 0; i < 3; i++) {
4569 list_insert_head(&zcl.zcl_callbacks, cb_data[i]);
4571 list_insert_after(&zcl.zcl_callbacks, tmp_cb,
4574 cb_data[i]->zcd_added = B_TRUE;
4575 VERIFY(!cb_data[i]->zcd_called);
4577 tmp_cb = cb_data[i];
4580 (void) mutex_unlock(&zcl.zcl_callbacks_lock);
4587 ztest_dsl_prop_get_set(ztest_ds_t *zd, uint64_t id)
4589 zfs_prop_t proplist[] = {
4591 ZFS_PROP_COMPRESSION,
4596 (void) rw_rdlock(&ztest_name_lock);
4598 for (int p = 0; p < sizeof (proplist) / sizeof (proplist[0]); p++)
4599 (void) ztest_dsl_prop_set_uint64(zd->zd_name, proplist[p],
4600 ztest_random_dsl_prop(proplist[p]), (int)ztest_random(2));
4602 (void) rw_unlock(&ztest_name_lock);
4607 ztest_spa_prop_get_set(ztest_ds_t *zd, uint64_t id)
4609 nvlist_t *props = NULL;
4611 (void) rw_rdlock(&ztest_name_lock);
4613 (void) ztest_spa_prop_set_uint64(ZPOOL_PROP_DEDUPDITTO,
4614 ZIO_DEDUPDITTO_MIN + ztest_random(ZIO_DEDUPDITTO_MIN));
4616 VERIFY0(spa_prop_get(ztest_spa, &props));
4618 if (ztest_opts.zo_verbose >= 6)
4619 dump_nvlist(props, 4);
4623 (void) rw_unlock(&ztest_name_lock);
4627 user_release_one(const char *snapname, const char *holdname)
4629 nvlist_t *snaps, *holds;
4632 snaps = fnvlist_alloc();
4633 holds = fnvlist_alloc();
4634 fnvlist_add_boolean(holds, holdname);
4635 fnvlist_add_nvlist(snaps, snapname, holds);
4636 fnvlist_free(holds);
4637 error = dsl_dataset_user_release(snaps, NULL);
4638 fnvlist_free(snaps);
4643 * Test snapshot hold/release and deferred destroy.
4646 ztest_dmu_snapshot_hold(ztest_ds_t *zd, uint64_t id)
4649 objset_t *os = zd->zd_os;
4653 char clonename[100];
4655 char osname[MAXNAMELEN];
4658 (void) rw_rdlock(&ztest_name_lock);
4660 dmu_objset_name(os, osname);
4662 (void) snprintf(snapname, sizeof (snapname), "sh1_%llu", id);
4663 (void) snprintf(fullname, sizeof (fullname), "%s@%s", osname, snapname);
4664 (void) snprintf(clonename, sizeof (clonename),
4665 "%s/ch1_%llu", osname, id);
4666 (void) snprintf(tag, sizeof (tag), "tag_%llu", id);
4669 * Clean up from any previous run.
4671 error = dsl_destroy_head(clonename);
4672 if (error != ENOENT)
4674 error = user_release_one(fullname, tag);
4675 if (error != ESRCH && error != ENOENT)
4677 error = dsl_destroy_snapshot(fullname, B_FALSE);
4678 if (error != ENOENT)
4682 * Create snapshot, clone it, mark snap for deferred destroy,
4683 * destroy clone, verify snap was also destroyed.
4685 error = dmu_objset_snapshot_one(osname, snapname);
4687 if (error == ENOSPC) {
4688 ztest_record_enospc("dmu_objset_snapshot");
4691 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
4694 error = dmu_objset_clone(clonename, fullname);
4696 if (error == ENOSPC) {
4697 ztest_record_enospc("dmu_objset_clone");
4700 fatal(0, "dmu_objset_clone(%s) = %d", clonename, error);
4703 error = dsl_destroy_snapshot(fullname, B_TRUE);
4705 fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
4709 error = dsl_destroy_head(clonename);
4711 fatal(0, "dsl_destroy_head(%s) = %d", clonename, error);
4713 error = dmu_objset_hold(fullname, FTAG, &origin);
4714 if (error != ENOENT)
4715 fatal(0, "dmu_objset_hold(%s) = %d", fullname, error);
4718 * Create snapshot, add temporary hold, verify that we can't
4719 * destroy a held snapshot, mark for deferred destroy,
4720 * release hold, verify snapshot was destroyed.
4722 error = dmu_objset_snapshot_one(osname, snapname);
4724 if (error == ENOSPC) {
4725 ztest_record_enospc("dmu_objset_snapshot");
4728 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
4731 holds = fnvlist_alloc();
4732 fnvlist_add_string(holds, fullname, tag);
4733 error = dsl_dataset_user_hold(holds, 0, NULL);
4734 fnvlist_free(holds);
4736 if (error == ENOSPC) {
4737 ztest_record_enospc("dsl_dataset_user_hold");
4740 fatal(0, "dsl_dataset_user_hold(%s, %s) = %u",
4741 fullname, tag, error);
4744 error = dsl_destroy_snapshot(fullname, B_FALSE);
4745 if (error != EBUSY) {
4746 fatal(0, "dsl_destroy_snapshot(%s, B_FALSE) = %d",
4750 error = dsl_destroy_snapshot(fullname, B_TRUE);
4752 fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
4756 error = user_release_one(fullname, tag);
4758 fatal(0, "user_release_one(%s, %s) = %d", fullname, tag, error);
4760 VERIFY3U(dmu_objset_hold(fullname, FTAG, &origin), ==, ENOENT);
4763 (void) rw_unlock(&ztest_name_lock);
4767 * Inject random faults into the on-disk data.
4771 ztest_fault_inject(ztest_ds_t *zd, uint64_t id)
4773 ztest_shared_t *zs = ztest_shared;
4774 spa_t *spa = ztest_spa;
4778 uint64_t bad = 0x1990c0ffeedecadeULL;
4780 char path0[MAXPATHLEN];
4781 char pathrand[MAXPATHLEN];
4783 int bshift = SPA_OLD_MAXBLOCKSHIFT + 2; /* don't scrog all labels */
4789 boolean_t islog = B_FALSE;
4791 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
4792 maxfaults = MAXFAULTS();
4793 leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz;
4794 mirror_save = zs->zs_mirrors;
4795 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4797 ASSERT(leaves >= 1);
4800 * Grab the name lock as reader. There are some operations
4801 * which don't like to have their vdevs changed while
4802 * they are in progress (i.e. spa_change_guid). Those
4803 * operations will have grabbed the name lock as writer.
4805 (void) rw_rdlock(&ztest_name_lock);
4808 * We need SCL_STATE here because we're going to look at vd0->vdev_tsd.
4810 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
4812 if (ztest_random(2) == 0) {
4814 * Inject errors on a normal data device or slog device.
4816 top = ztest_random_vdev_top(spa, B_TRUE);
4817 leaf = ztest_random(leaves) + zs->zs_splits;
4820 * Generate paths to the first leaf in this top-level vdev,
4821 * and to the random leaf we selected. We'll induce transient
4822 * write failures and random online/offline activity on leaf 0,
4823 * and we'll write random garbage to the randomly chosen leaf.
4825 (void) snprintf(path0, sizeof (path0), ztest_dev_template,
4826 ztest_opts.zo_dir, ztest_opts.zo_pool,
4827 top * leaves + zs->zs_splits);
4828 (void) snprintf(pathrand, sizeof (pathrand), ztest_dev_template,
4829 ztest_opts.zo_dir, ztest_opts.zo_pool,
4830 top * leaves + leaf);
4832 vd0 = vdev_lookup_by_path(spa->spa_root_vdev, path0);
4833 if (vd0 != NULL && vd0->vdev_top->vdev_islog)
4837 * If the top-level vdev needs to be resilvered
4838 * then we only allow faults on the device that is
4841 if (vd0 != NULL && maxfaults != 1 &&
4842 (!vdev_resilver_needed(vd0->vdev_top, NULL, NULL) ||
4843 vd0->vdev_resilver_txg != 0)) {
4845 * Make vd0 explicitly claim to be unreadable,
4846 * or unwriteable, or reach behind its back
4847 * and close the underlying fd. We can do this if
4848 * maxfaults == 0 because we'll fail and reexecute,
4849 * and we can do it if maxfaults >= 2 because we'll
4850 * have enough redundancy. If maxfaults == 1, the
4851 * combination of this with injection of random data
4852 * corruption below exceeds the pool's fault tolerance.
4854 vdev_file_t *vf = vd0->vdev_tsd;
4856 if (vf != NULL && ztest_random(3) == 0) {
4857 (void) close(vf->vf_vnode->v_fd);
4858 vf->vf_vnode->v_fd = -1;
4859 } else if (ztest_random(2) == 0) {
4860 vd0->vdev_cant_read = B_TRUE;
4862 vd0->vdev_cant_write = B_TRUE;
4864 guid0 = vd0->vdev_guid;
4868 * Inject errors on an l2cache device.
4870 spa_aux_vdev_t *sav = &spa->spa_l2cache;
4872 if (sav->sav_count == 0) {
4873 spa_config_exit(spa, SCL_STATE, FTAG);
4874 (void) rw_unlock(&ztest_name_lock);
4877 vd0 = sav->sav_vdevs[ztest_random(sav->sav_count)];
4878 guid0 = vd0->vdev_guid;
4879 (void) strcpy(path0, vd0->vdev_path);
4880 (void) strcpy(pathrand, vd0->vdev_path);
4884 maxfaults = INT_MAX; /* no limit on cache devices */
4887 spa_config_exit(spa, SCL_STATE, FTAG);
4888 (void) rw_unlock(&ztest_name_lock);
4891 * If we can tolerate two or more faults, or we're dealing
4892 * with a slog, randomly online/offline vd0.
4894 if ((maxfaults >= 2 || islog) && guid0 != 0) {
4895 if (ztest_random(10) < 6) {
4896 int flags = (ztest_random(2) == 0 ?
4897 ZFS_OFFLINE_TEMPORARY : 0);
4900 * We have to grab the zs_name_lock as writer to
4901 * prevent a race between offlining a slog and
4902 * destroying a dataset. Offlining the slog will
4903 * grab a reference on the dataset which may cause
4904 * dmu_objset_destroy() to fail with EBUSY thus
4905 * leaving the dataset in an inconsistent state.
4908 (void) rw_wrlock(&ztest_name_lock);
4910 VERIFY(vdev_offline(spa, guid0, flags) != EBUSY);
4913 (void) rw_unlock(&ztest_name_lock);
4916 * Ideally we would like to be able to randomly
4917 * call vdev_[on|off]line without holding locks
4918 * to force unpredictable failures but the side
4919 * effects of vdev_[on|off]line prevent us from
4920 * doing so. We grab the ztest_vdev_lock here to
4921 * prevent a race between injection testing and
4924 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
4925 (void) vdev_online(spa, guid0, 0, NULL);
4926 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4934 * We have at least single-fault tolerance, so inject data corruption.
4936 fd = open(pathrand, O_RDWR);
4938 if (fd == -1) /* we hit a gap in the device namespace */
4941 fsize = lseek(fd, 0, SEEK_END);
4943 while (--iters != 0) {
4944 offset = ztest_random(fsize / (leaves << bshift)) *
4945 (leaves << bshift) + (leaf << bshift) +
4946 (ztest_random(1ULL << (bshift - 1)) & -8ULL);
4948 if (offset >= fsize)
4951 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
4952 if (mirror_save != zs->zs_mirrors) {
4953 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4958 if (pwrite(fd, &bad, sizeof (bad), offset) != sizeof (bad))
4959 fatal(1, "can't inject bad word at 0x%llx in %s",
4962 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4964 if (ztest_opts.zo_verbose >= 7)
4965 (void) printf("injected bad word into %s,"
4966 " offset 0x%llx\n", pathrand, (u_longlong_t)offset);
4973 * Verify that DDT repair works as expected.
4976 ztest_ddt_repair(ztest_ds_t *zd, uint64_t id)
4978 ztest_shared_t *zs = ztest_shared;
4979 spa_t *spa = ztest_spa;
4980 objset_t *os = zd->zd_os;
4982 uint64_t object, blocksize, txg, pattern, psize;
4983 enum zio_checksum checksum = spa_dedup_checksum(spa);
4988 int copies = 2 * ZIO_DEDUPDITTO_MIN;
4990 blocksize = ztest_random_blocksize();
4991 blocksize = MIN(blocksize, 2048); /* because we write so many */
4993 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
4995 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4999 * Take the name lock as writer to prevent anyone else from changing
5000 * the pool and dataset properies we need to maintain during this test.
5002 (void) rw_wrlock(&ztest_name_lock);
5004 if (ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_DEDUP, checksum,
5006 ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_COPIES, 1,
5008 (void) rw_unlock(&ztest_name_lock);
5012 object = od[0].od_object;
5013 blocksize = od[0].od_blocksize;
5014 pattern = zs->zs_guid ^ dmu_objset_fsid_guid(os);
5016 ASSERT(object != 0);
5018 tx = dmu_tx_create(os);
5019 dmu_tx_hold_write(tx, object, 0, copies * blocksize);
5020 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
5022 (void) rw_unlock(&ztest_name_lock);
5027 * Write all the copies of our block.
5029 for (int i = 0; i < copies; i++) {
5030 uint64_t offset = i * blocksize;
5031 int error = dmu_buf_hold(os, object, offset, FTAG, &db,
5032 DMU_READ_NO_PREFETCH);
5034 fatal(B_FALSE, "dmu_buf_hold(%p, %llu, %llu) = %u",
5035 os, (long long)object, (long long) offset, error);
5037 ASSERT(db->db_offset == offset);
5038 ASSERT(db->db_size == blocksize);
5039 ASSERT(ztest_pattern_match(db->db_data, db->db_size, pattern) ||
5040 ztest_pattern_match(db->db_data, db->db_size, 0ULL));
5041 dmu_buf_will_fill(db, tx);
5042 ztest_pattern_set(db->db_data, db->db_size, pattern);
5043 dmu_buf_rele(db, FTAG);
5047 txg_wait_synced(spa_get_dsl(spa), txg);
5050 * Find out what block we got.
5052 VERIFY0(dmu_buf_hold(os, object, 0, FTAG, &db,
5053 DMU_READ_NO_PREFETCH));
5054 blk = *((dmu_buf_impl_t *)db)->db_blkptr;
5055 dmu_buf_rele(db, FTAG);
5058 * Damage the block. Dedup-ditto will save us when we read it later.
5060 psize = BP_GET_PSIZE(&blk);
5061 buf = zio_buf_alloc(psize);
5062 ztest_pattern_set(buf, psize, ~pattern);
5064 (void) zio_wait(zio_rewrite(NULL, spa, 0, &blk,
5065 buf, psize, NULL, NULL, ZIO_PRIORITY_SYNC_WRITE,
5066 ZIO_FLAG_CANFAIL | ZIO_FLAG_INDUCE_DAMAGE, NULL));
5068 zio_buf_free(buf, psize);
5070 (void) rw_unlock(&ztest_name_lock);
5078 ztest_scrub(ztest_ds_t *zd, uint64_t id)
5080 spa_t *spa = ztest_spa;
5082 (void) spa_scan(spa, POOL_SCAN_SCRUB);
5083 (void) poll(NULL, 0, 100); /* wait a moment, then force a restart */
5084 (void) spa_scan(spa, POOL_SCAN_SCRUB);
5088 * Change the guid for the pool.
5092 ztest_reguid(ztest_ds_t *zd, uint64_t id)
5094 spa_t *spa = ztest_spa;
5095 uint64_t orig, load;
5098 orig = spa_guid(spa);
5099 load = spa_load_guid(spa);
5101 (void) rw_wrlock(&ztest_name_lock);
5102 error = spa_change_guid(spa);
5103 (void) rw_unlock(&ztest_name_lock);
5108 if (ztest_opts.zo_verbose >= 4) {
5109 (void) printf("Changed guid old %llu -> %llu\n",
5110 (u_longlong_t)orig, (u_longlong_t)spa_guid(spa));
5113 VERIFY3U(orig, !=, spa_guid(spa));
5114 VERIFY3U(load, ==, spa_load_guid(spa));
5118 * Rename the pool to a different name and then rename it back.
5122 ztest_spa_rename(ztest_ds_t *zd, uint64_t id)
5124 char *oldname, *newname;
5127 (void) rw_wrlock(&ztest_name_lock);
5129 oldname = ztest_opts.zo_pool;
5130 newname = umem_alloc(strlen(oldname) + 5, UMEM_NOFAIL);
5131 (void) strcpy(newname, oldname);
5132 (void) strcat(newname, "_tmp");
5137 VERIFY3U(0, ==, spa_rename(oldname, newname));
5140 * Try to open it under the old name, which shouldn't exist
5142 VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
5145 * Open it under the new name and make sure it's still the same spa_t.
5147 VERIFY3U(0, ==, spa_open(newname, &spa, FTAG));
5149 ASSERT(spa == ztest_spa);
5150 spa_close(spa, FTAG);
5153 * Rename it back to the original
5155 VERIFY3U(0, ==, spa_rename(newname, oldname));
5158 * Make sure it can still be opened
5160 VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG));
5162 ASSERT(spa == ztest_spa);
5163 spa_close(spa, FTAG);
5165 umem_free(newname, strlen(newname) + 1);
5167 (void) rw_unlock(&ztest_name_lock);
5171 * Verify pool integrity by running zdb.
5174 ztest_run_zdb(char *pool)
5177 char zdb[MAXPATHLEN + MAXNAMELEN + 20];
5185 strlcpy(zdb, "/usr/bin/ztest", sizeof(zdb));
5187 /* zdb lives in /usr/sbin, while ztest lives in /usr/bin */
5188 bin = strstr(zdb, "/usr/bin/");
5189 ztest = strstr(bin, "/ztest");
5191 isalen = ztest - isa;
5195 "/usr/sbin%.*s/zdb -bcc%s%s -d -U %s %s",
5198 ztest_opts.zo_verbose >= 3 ? "s" : "",
5199 ztest_opts.zo_verbose >= 4 ? "v" : "",
5204 if (ztest_opts.zo_verbose >= 5)
5205 (void) printf("Executing %s\n", strstr(zdb, "zdb "));
5207 fp = popen(zdb, "r");
5210 while (fgets(zbuf, sizeof (zbuf), fp) != NULL)
5211 if (ztest_opts.zo_verbose >= 3)
5212 (void) printf("%s", zbuf);
5214 status = pclose(fp);
5219 ztest_dump_core = 0;
5220 if (WIFEXITED(status))
5221 fatal(0, "'%s' exit code %d", zdb, WEXITSTATUS(status));
5223 fatal(0, "'%s' died with signal %d", zdb, WTERMSIG(status));
5227 ztest_walk_pool_directory(char *header)
5231 if (ztest_opts.zo_verbose >= 6)
5232 (void) printf("%s\n", header);
5234 mutex_enter(&spa_namespace_lock);
5235 while ((spa = spa_next(spa)) != NULL)
5236 if (ztest_opts.zo_verbose >= 6)
5237 (void) printf("\t%s\n", spa_name(spa));
5238 mutex_exit(&spa_namespace_lock);
5242 ztest_spa_import_export(char *oldname, char *newname)
5244 nvlist_t *config, *newconfig;
5249 if (ztest_opts.zo_verbose >= 4) {
5250 (void) printf("import/export: old = %s, new = %s\n",
5255 * Clean up from previous runs.
5257 (void) spa_destroy(newname);
5260 * Get the pool's configuration and guid.
5262 VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG));
5265 * Kick off a scrub to tickle scrub/export races.
5267 if (ztest_random(2) == 0)
5268 (void) spa_scan(spa, POOL_SCAN_SCRUB);
5270 pool_guid = spa_guid(spa);
5271 spa_close(spa, FTAG);
5273 ztest_walk_pool_directory("pools before export");
5278 VERIFY3U(0, ==, spa_export(oldname, &config, B_FALSE, B_FALSE));
5280 ztest_walk_pool_directory("pools after export");
5285 newconfig = spa_tryimport(config);
5286 ASSERT(newconfig != NULL);
5287 nvlist_free(newconfig);
5290 * Import it under the new name.
5292 error = spa_import(newname, config, NULL, 0);
5294 dump_nvlist(config, 0);
5295 fatal(B_FALSE, "couldn't import pool %s as %s: error %u",
5296 oldname, newname, error);
5299 ztest_walk_pool_directory("pools after import");
5302 * Try to import it again -- should fail with EEXIST.
5304 VERIFY3U(EEXIST, ==, spa_import(newname, config, NULL, 0));
5307 * Try to import it under a different name -- should fail with EEXIST.
5309 VERIFY3U(EEXIST, ==, spa_import(oldname, config, NULL, 0));
5312 * Verify that the pool is no longer visible under the old name.
5314 VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
5317 * Verify that we can open and close the pool using the new name.
5319 VERIFY3U(0, ==, spa_open(newname, &spa, FTAG));
5320 ASSERT(pool_guid == spa_guid(spa));
5321 spa_close(spa, FTAG);
5323 nvlist_free(config);
5327 ztest_resume(spa_t *spa)
5329 if (spa_suspended(spa) && ztest_opts.zo_verbose >= 6)
5330 (void) printf("resuming from suspended state\n");
5331 spa_vdev_state_enter(spa, SCL_NONE);
5332 vdev_clear(spa, NULL);
5333 (void) spa_vdev_state_exit(spa, NULL, 0);
5334 (void) zio_resume(spa);
5338 ztest_resume_thread(void *arg)
5342 while (!ztest_exiting) {
5343 if (spa_suspended(spa))
5345 (void) poll(NULL, 0, 100);
5351 ztest_deadman_thread(void *arg)
5353 ztest_shared_t *zs = arg;
5354 spa_t *spa = ztest_spa;
5355 hrtime_t delta, total = 0;
5358 delta = zs->zs_thread_stop - zs->zs_thread_start +
5359 MSEC2NSEC(zfs_deadman_synctime_ms);
5361 (void) poll(NULL, 0, (int)NSEC2MSEC(delta));
5364 * If the pool is suspended then fail immediately. Otherwise,
5365 * check to see if the pool is making any progress. If
5366 * vdev_deadman() discovers that there hasn't been any recent
5367 * I/Os then it will end up aborting the tests.
5369 if (spa_suspended(spa) || spa->spa_root_vdev == NULL) {
5370 fatal(0, "aborting test after %llu seconds because "
5371 "pool has transitioned to a suspended state.",
5372 zfs_deadman_synctime_ms / 1000);
5375 vdev_deadman(spa->spa_root_vdev);
5377 total += zfs_deadman_synctime_ms/1000;
5378 (void) printf("ztest has been running for %lld seconds\n",
5384 ztest_execute(int test, ztest_info_t *zi, uint64_t id)
5386 ztest_ds_t *zd = &ztest_ds[id % ztest_opts.zo_datasets];
5387 ztest_shared_callstate_t *zc = ZTEST_GET_SHARED_CALLSTATE(test);
5388 hrtime_t functime = gethrtime();
5390 for (int i = 0; i < zi->zi_iters; i++)
5391 zi->zi_func(zd, id);
5393 functime = gethrtime() - functime;
5395 atomic_add_64(&zc->zc_count, 1);
5396 atomic_add_64(&zc->zc_time, functime);
5398 if (ztest_opts.zo_verbose >= 4) {
5400 (void) dladdr((void *)zi->zi_func, &dli);
5401 (void) printf("%6.2f sec in %s\n",
5402 (double)functime / NANOSEC, dli.dli_sname);
5407 ztest_thread(void *arg)
5410 uint64_t id = (uintptr_t)arg;
5411 ztest_shared_t *zs = ztest_shared;
5415 ztest_shared_callstate_t *zc;
5417 while ((now = gethrtime()) < zs->zs_thread_stop) {
5419 * See if it's time to force a crash.
5421 if (now > zs->zs_thread_kill)
5425 * If we're getting ENOSPC with some regularity, stop.
5427 if (zs->zs_enospc_count > 10)
5431 * Pick a random function to execute.
5433 rand = ztest_random(ZTEST_FUNCS);
5434 zi = &ztest_info[rand];
5435 zc = ZTEST_GET_SHARED_CALLSTATE(rand);
5436 call_next = zc->zc_next;
5438 if (now >= call_next &&
5439 atomic_cas_64(&zc->zc_next, call_next, call_next +
5440 ztest_random(2 * zi->zi_interval[0] + 1)) == call_next) {
5441 ztest_execute(rand, zi, id);
5449 ztest_dataset_name(char *dsname, char *pool, int d)
5451 (void) snprintf(dsname, MAXNAMELEN, "%s/ds_%d", pool, d);
5455 ztest_dataset_destroy(int d)
5457 char name[MAXNAMELEN];
5459 ztest_dataset_name(name, ztest_opts.zo_pool, d);
5461 if (ztest_opts.zo_verbose >= 3)
5462 (void) printf("Destroying %s to free up space\n", name);
5465 * Cleanup any non-standard clones and snapshots. In general,
5466 * ztest thread t operates on dataset (t % zopt_datasets),
5467 * so there may be more than one thing to clean up.
5469 for (int t = d; t < ztest_opts.zo_threads;
5470 t += ztest_opts.zo_datasets) {
5471 ztest_dsl_dataset_cleanup(name, t);
5474 (void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
5475 DS_FIND_SNAPSHOTS | DS_FIND_CHILDREN);
5479 ztest_dataset_dirobj_verify(ztest_ds_t *zd)
5481 uint64_t usedobjs, dirobjs, scratch;
5484 * ZTEST_DIROBJ is the object directory for the entire dataset.
5485 * Therefore, the number of objects in use should equal the
5486 * number of ZTEST_DIROBJ entries, +1 for ZTEST_DIROBJ itself.
5487 * If not, we have an object leak.
5489 * Note that we can only check this in ztest_dataset_open(),
5490 * when the open-context and syncing-context values agree.
5491 * That's because zap_count() returns the open-context value,
5492 * while dmu_objset_space() returns the rootbp fill count.
5494 VERIFY3U(0, ==, zap_count(zd->zd_os, ZTEST_DIROBJ, &dirobjs));
5495 dmu_objset_space(zd->zd_os, &scratch, &scratch, &usedobjs, &scratch);
5496 ASSERT3U(dirobjs + 1, ==, usedobjs);
5500 ztest_dataset_open(int d)
5502 ztest_ds_t *zd = &ztest_ds[d];
5503 uint64_t committed_seq = ZTEST_GET_SHARED_DS(d)->zd_seq;
5506 char name[MAXNAMELEN];
5509 ztest_dataset_name(name, ztest_opts.zo_pool, d);
5511 (void) rw_rdlock(&ztest_name_lock);
5513 error = ztest_dataset_create(name);
5514 if (error == ENOSPC) {
5515 (void) rw_unlock(&ztest_name_lock);
5516 ztest_record_enospc(FTAG);
5519 ASSERT(error == 0 || error == EEXIST);
5521 VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, zd, &os));
5522 (void) rw_unlock(&ztest_name_lock);
5524 ztest_zd_init(zd, ZTEST_GET_SHARED_DS(d), os);
5526 zilog = zd->zd_zilog;
5528 if (zilog->zl_header->zh_claim_lr_seq != 0 &&
5529 zilog->zl_header->zh_claim_lr_seq < committed_seq)
5530 fatal(0, "missing log records: claimed %llu < committed %llu",
5531 zilog->zl_header->zh_claim_lr_seq, committed_seq);
5533 ztest_dataset_dirobj_verify(zd);
5535 zil_replay(os, zd, ztest_replay_vector);
5537 ztest_dataset_dirobj_verify(zd);
5539 if (ztest_opts.zo_verbose >= 6)
5540 (void) printf("%s replay %llu blocks, %llu records, seq %llu\n",
5542 (u_longlong_t)zilog->zl_parse_blk_count,
5543 (u_longlong_t)zilog->zl_parse_lr_count,
5544 (u_longlong_t)zilog->zl_replaying_seq);
5546 zilog = zil_open(os, ztest_get_data);
5548 if (zilog->zl_replaying_seq != 0 &&
5549 zilog->zl_replaying_seq < committed_seq)
5550 fatal(0, "missing log records: replayed %llu < committed %llu",
5551 zilog->zl_replaying_seq, committed_seq);
5557 ztest_dataset_close(int d)
5559 ztest_ds_t *zd = &ztest_ds[d];
5561 zil_close(zd->zd_zilog);
5562 dmu_objset_disown(zd->zd_os, zd);
5568 * Kick off threads to run tests on all datasets in parallel.
5571 ztest_run(ztest_shared_t *zs)
5576 thread_t resume_tid;
5579 ztest_exiting = B_FALSE;
5582 * Initialize parent/child shared state.
5584 VERIFY(_mutex_init(&ztest_vdev_lock, USYNC_THREAD, NULL) == 0);
5585 VERIFY(rwlock_init(&ztest_name_lock, USYNC_THREAD, NULL) == 0);
5587 zs->zs_thread_start = gethrtime();
5588 zs->zs_thread_stop =
5589 zs->zs_thread_start + ztest_opts.zo_passtime * NANOSEC;
5590 zs->zs_thread_stop = MIN(zs->zs_thread_stop, zs->zs_proc_stop);
5591 zs->zs_thread_kill = zs->zs_thread_stop;
5592 if (ztest_random(100) < ztest_opts.zo_killrate) {
5593 zs->zs_thread_kill -=
5594 ztest_random(ztest_opts.zo_passtime * NANOSEC);
5597 (void) _mutex_init(&zcl.zcl_callbacks_lock, USYNC_THREAD, NULL);
5599 list_create(&zcl.zcl_callbacks, sizeof (ztest_cb_data_t),
5600 offsetof(ztest_cb_data_t, zcd_node));
5605 kernel_init(FREAD | FWRITE);
5606 VERIFY0(spa_open(ztest_opts.zo_pool, &spa, FTAG));
5607 spa->spa_debug = B_TRUE;
5608 metaslab_preload_limit = ztest_random(20) + 1;
5611 VERIFY0(dmu_objset_own(ztest_opts.zo_pool,
5612 DMU_OST_ANY, B_TRUE, FTAG, &os));
5613 zs->zs_guid = dmu_objset_fsid_guid(os);
5614 dmu_objset_disown(os, FTAG);
5616 spa->spa_dedup_ditto = 2 * ZIO_DEDUPDITTO_MIN;
5619 * We don't expect the pool to suspend unless maxfaults == 0,
5620 * in which case ztest_fault_inject() temporarily takes away
5621 * the only valid replica.
5623 if (MAXFAULTS() == 0)
5624 spa->spa_failmode = ZIO_FAILURE_MODE_WAIT;
5626 spa->spa_failmode = ZIO_FAILURE_MODE_PANIC;
5629 * Create a thread to periodically resume suspended I/O.
5631 VERIFY(thr_create(0, 0, ztest_resume_thread, spa, THR_BOUND,
5635 * Create a deadman thread to abort() if we hang.
5637 VERIFY(thr_create(0, 0, ztest_deadman_thread, zs, THR_BOUND,
5641 * Verify that we can safely inquire about about any object,
5642 * whether it's allocated or not. To make it interesting,
5643 * we probe a 5-wide window around each power of two.
5644 * This hits all edge cases, including zero and the max.
5646 for (int t = 0; t < 64; t++) {
5647 for (int d = -5; d <= 5; d++) {
5648 error = dmu_object_info(spa->spa_meta_objset,
5649 (1ULL << t) + d, NULL);
5650 ASSERT(error == 0 || error == ENOENT ||
5656 * If we got any ENOSPC errors on the previous run, destroy something.
5658 if (zs->zs_enospc_count != 0) {
5659 int d = ztest_random(ztest_opts.zo_datasets);
5660 ztest_dataset_destroy(d);
5662 zs->zs_enospc_count = 0;
5664 tid = umem_zalloc(ztest_opts.zo_threads * sizeof (thread_t),
5667 if (ztest_opts.zo_verbose >= 4)
5668 (void) printf("starting main threads...\n");
5671 * Kick off all the tests that run in parallel.
5673 for (int t = 0; t < ztest_opts.zo_threads; t++) {
5674 if (t < ztest_opts.zo_datasets &&
5675 ztest_dataset_open(t) != 0)
5677 VERIFY(thr_create(0, 0, ztest_thread, (void *)(uintptr_t)t,
5678 THR_BOUND, &tid[t]) == 0);
5682 * Wait for all of the tests to complete. We go in reverse order
5683 * so we don't close datasets while threads are still using them.
5685 for (int t = ztest_opts.zo_threads - 1; t >= 0; t--) {
5686 VERIFY(thr_join(tid[t], NULL, NULL) == 0);
5687 if (t < ztest_opts.zo_datasets)
5688 ztest_dataset_close(t);
5691 txg_wait_synced(spa_get_dsl(spa), 0);
5693 zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(spa));
5694 zs->zs_space = metaslab_class_get_space(spa_normal_class(spa));
5695 zfs_dbgmsg_print(FTAG);
5697 umem_free(tid, ztest_opts.zo_threads * sizeof (thread_t));
5699 /* Kill the resume thread */
5700 ztest_exiting = B_TRUE;
5701 VERIFY(thr_join(resume_tid, NULL, NULL) == 0);
5705 * Right before closing the pool, kick off a bunch of async I/O;
5706 * spa_close() should wait for it to complete.
5708 for (uint64_t object = 1; object < 50; object++)
5709 dmu_prefetch(spa->spa_meta_objset, object, 0, 1ULL << 20);
5711 spa_close(spa, FTAG);
5714 * Verify that we can loop over all pools.
5716 mutex_enter(&spa_namespace_lock);
5717 for (spa = spa_next(NULL); spa != NULL; spa = spa_next(spa))
5718 if (ztest_opts.zo_verbose > 3)
5719 (void) printf("spa_next: found %s\n", spa_name(spa));
5720 mutex_exit(&spa_namespace_lock);
5723 * Verify that we can export the pool and reimport it under a
5726 if (ztest_random(2) == 0) {
5727 char name[MAXNAMELEN];
5728 (void) snprintf(name, MAXNAMELEN, "%s_import",
5729 ztest_opts.zo_pool);
5730 ztest_spa_import_export(ztest_opts.zo_pool, name);
5731 ztest_spa_import_export(name, ztest_opts.zo_pool);
5736 list_destroy(&zcl.zcl_callbacks);
5738 (void) _mutex_destroy(&zcl.zcl_callbacks_lock);
5740 (void) rwlock_destroy(&ztest_name_lock);
5741 (void) _mutex_destroy(&ztest_vdev_lock);
5747 ztest_ds_t *zd = &ztest_ds[0];
5751 if (ztest_opts.zo_verbose >= 3)
5752 (void) printf("testing spa_freeze()...\n");
5754 kernel_init(FREAD | FWRITE);
5755 VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
5756 VERIFY3U(0, ==, ztest_dataset_open(0));
5757 spa->spa_debug = B_TRUE;
5761 * Force the first log block to be transactionally allocated.
5762 * We have to do this before we freeze the pool -- otherwise
5763 * the log chain won't be anchored.
5765 while (BP_IS_HOLE(&zd->zd_zilog->zl_header->zh_log)) {
5766 ztest_dmu_object_alloc_free(zd, 0);
5767 zil_commit(zd->zd_zilog, 0);
5770 txg_wait_synced(spa_get_dsl(spa), 0);
5773 * Freeze the pool. This stops spa_sync() from doing anything,
5774 * so that the only way to record changes from now on is the ZIL.
5779 * Because it is hard to predict how much space a write will actually
5780 * require beforehand, we leave ourselves some fudge space to write over
5783 uint64_t capacity = metaslab_class_get_space(spa_normal_class(spa)) / 2;
5786 * Run tests that generate log records but don't alter the pool config
5787 * or depend on DSL sync tasks (snapshots, objset create/destroy, etc).
5788 * We do a txg_wait_synced() after each iteration to force the txg
5789 * to increase well beyond the last synced value in the uberblock.
5790 * The ZIL should be OK with that.
5792 * Run a random number of times less than zo_maxloops and ensure we do
5793 * not run out of space on the pool.
5795 while (ztest_random(10) != 0 &&
5796 numloops++ < ztest_opts.zo_maxloops &&
5797 metaslab_class_get_alloc(spa_normal_class(spa)) < capacity) {
5799 ztest_od_init(&od, 0, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
5800 VERIFY0(ztest_object_init(zd, &od, sizeof (od), B_FALSE));
5801 ztest_io(zd, od.od_object,
5802 ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
5803 txg_wait_synced(spa_get_dsl(spa), 0);
5807 * Commit all of the changes we just generated.
5809 zil_commit(zd->zd_zilog, 0);
5810 txg_wait_synced(spa_get_dsl(spa), 0);
5813 * Close our dataset and close the pool.
5815 ztest_dataset_close(0);
5816 spa_close(spa, FTAG);
5820 * Open and close the pool and dataset to induce log replay.
5822 kernel_init(FREAD | FWRITE);
5823 VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
5824 ASSERT(spa_freeze_txg(spa) == UINT64_MAX);
5825 VERIFY3U(0, ==, ztest_dataset_open(0));
5826 ztest_dataset_close(0);
5828 spa->spa_debug = B_TRUE;
5830 txg_wait_synced(spa_get_dsl(spa), 0);
5831 ztest_reguid(NULL, 0);
5833 spa_close(spa, FTAG);
5838 print_time(hrtime_t t, char *timebuf)
5840 hrtime_t s = t / NANOSEC;
5841 hrtime_t m = s / 60;
5842 hrtime_t h = m / 60;
5843 hrtime_t d = h / 24;
5852 (void) sprintf(timebuf,
5853 "%llud%02lluh%02llum%02llus", d, h, m, s);
5855 (void) sprintf(timebuf, "%lluh%02llum%02llus", h, m, s);
5857 (void) sprintf(timebuf, "%llum%02llus", m, s);
5859 (void) sprintf(timebuf, "%llus", s);
5867 VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
5868 if (ztest_random(2) == 0)
5870 VERIFY(nvlist_add_uint64(props, "autoreplace", 1) == 0);
5876 * Create a storage pool with the given name and initial vdev size.
5877 * Then test spa_freeze() functionality.
5880 ztest_init(ztest_shared_t *zs)
5883 nvlist_t *nvroot, *props;
5885 VERIFY(_mutex_init(&ztest_vdev_lock, USYNC_THREAD, NULL) == 0);
5886 VERIFY(rwlock_init(&ztest_name_lock, USYNC_THREAD, NULL) == 0);
5888 kernel_init(FREAD | FWRITE);
5891 * Create the storage pool.
5893 (void) spa_destroy(ztest_opts.zo_pool);
5894 ztest_shared->zs_vdev_next_leaf = 0;
5896 zs->zs_mirrors = ztest_opts.zo_mirrors;
5897 nvroot = make_vdev_root(NULL, NULL, NULL, ztest_opts.zo_vdev_size, 0,
5898 0, ztest_opts.zo_raidz, zs->zs_mirrors, 1);
5899 props = make_random_props();
5900 for (int i = 0; i < SPA_FEATURES; i++) {
5902 (void) snprintf(buf, sizeof (buf), "feature@%s",
5903 spa_feature_table[i].fi_uname);
5904 VERIFY3U(0, ==, nvlist_add_uint64(props, buf, 0));
5906 VERIFY3U(0, ==, spa_create(ztest_opts.zo_pool, nvroot, props, NULL));
5907 nvlist_free(nvroot);
5910 VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
5911 zs->zs_metaslab_sz =
5912 1ULL << spa->spa_root_vdev->vdev_child[0]->vdev_ms_shift;
5914 spa_close(spa, FTAG);
5918 ztest_run_zdb(ztest_opts.zo_pool);
5922 ztest_run_zdb(ztest_opts.zo_pool);
5924 (void) rwlock_destroy(&ztest_name_lock);
5925 (void) _mutex_destroy(&ztest_vdev_lock);
5931 static char ztest_name_data[] = "/tmp/ztest.data.XXXXXX";
5933 ztest_fd_data = mkstemp(ztest_name_data);
5934 ASSERT3S(ztest_fd_data, >=, 0);
5935 (void) unlink(ztest_name_data);
5940 shared_data_size(ztest_shared_hdr_t *hdr)
5944 size = hdr->zh_hdr_size;
5945 size += hdr->zh_opts_size;
5946 size += hdr->zh_size;
5947 size += hdr->zh_stats_size * hdr->zh_stats_count;
5948 size += hdr->zh_ds_size * hdr->zh_ds_count;
5957 ztest_shared_hdr_t *hdr;
5959 hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
5960 PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
5961 ASSERT(hdr != MAP_FAILED);
5963 VERIFY3U(0, ==, ftruncate(ztest_fd_data, sizeof (ztest_shared_hdr_t)));
5965 hdr->zh_hdr_size = sizeof (ztest_shared_hdr_t);
5966 hdr->zh_opts_size = sizeof (ztest_shared_opts_t);
5967 hdr->zh_size = sizeof (ztest_shared_t);
5968 hdr->zh_stats_size = sizeof (ztest_shared_callstate_t);
5969 hdr->zh_stats_count = ZTEST_FUNCS;
5970 hdr->zh_ds_size = sizeof (ztest_shared_ds_t);
5971 hdr->zh_ds_count = ztest_opts.zo_datasets;
5973 size = shared_data_size(hdr);
5974 VERIFY3U(0, ==, ftruncate(ztest_fd_data, size));
5976 (void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
5983 ztest_shared_hdr_t *hdr;
5986 hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
5987 PROT_READ, MAP_SHARED, ztest_fd_data, 0);
5988 ASSERT(hdr != MAP_FAILED);
5990 size = shared_data_size(hdr);
5992 (void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
5993 hdr = ztest_shared_hdr = (void *)mmap(0, P2ROUNDUP(size, getpagesize()),
5994 PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
5995 ASSERT(hdr != MAP_FAILED);
5996 buf = (uint8_t *)hdr;
5998 offset = hdr->zh_hdr_size;
5999 ztest_shared_opts = (void *)&buf[offset];
6000 offset += hdr->zh_opts_size;
6001 ztest_shared = (void *)&buf[offset];
6002 offset += hdr->zh_size;
6003 ztest_shared_callstate = (void *)&buf[offset];
6004 offset += hdr->zh_stats_size * hdr->zh_stats_count;
6005 ztest_shared_ds = (void *)&buf[offset];
6009 exec_child(char *cmd, char *libpath, boolean_t ignorekill, int *statusp)
6013 char *cmdbuf = NULL;
6018 cmdbuf = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
6019 (void) strlcpy(cmdbuf, getexecname(), MAXPATHLEN);
6024 fatal(1, "fork failed");
6026 if (pid == 0) { /* child */
6027 char *emptyargv[2] = { cmd, NULL };
6028 char fd_data_str[12];
6030 struct rlimit rl = { 1024, 1024 };
6031 (void) setrlimit(RLIMIT_NOFILE, &rl);
6033 (void) close(ztest_fd_rand);
6035 snprintf(fd_data_str, 12, "%d", ztest_fd_data));
6036 VERIFY0(setenv("ZTEST_FD_DATA", fd_data_str, 1));
6038 (void) enable_extended_FILE_stdio(-1, -1);
6039 if (libpath != NULL)
6040 VERIFY(0 == setenv("LD_LIBRARY_PATH", libpath, 1));
6042 (void) execv(cmd, emptyargv);
6044 (void) execvp(cmd, emptyargv);
6046 ztest_dump_core = B_FALSE;
6047 fatal(B_TRUE, "exec failed: %s", cmd);
6050 if (cmdbuf != NULL) {
6051 umem_free(cmdbuf, MAXPATHLEN);
6055 while (waitpid(pid, &status, 0) != pid)
6057 if (statusp != NULL)
6060 if (WIFEXITED(status)) {
6061 if (WEXITSTATUS(status) != 0) {
6062 (void) fprintf(stderr, "child exited with code %d\n",
6063 WEXITSTATUS(status));
6067 } else if (WIFSIGNALED(status)) {
6068 if (!ignorekill || WTERMSIG(status) != SIGKILL) {
6069 (void) fprintf(stderr, "child died with signal %d\n",
6075 (void) fprintf(stderr, "something strange happened to child\n");
6082 ztest_run_init(void)
6084 ztest_shared_t *zs = ztest_shared;
6086 ASSERT(ztest_opts.zo_init != 0);
6089 * Blow away any existing copy of zpool.cache
6091 (void) remove(spa_config_path);
6094 * Create and initialize our storage pool.
6096 for (int i = 1; i <= ztest_opts.zo_init; i++) {
6097 bzero(zs, sizeof (ztest_shared_t));
6098 if (ztest_opts.zo_verbose >= 3 &&
6099 ztest_opts.zo_init != 1) {
6100 (void) printf("ztest_init(), pass %d\n", i);
6107 main(int argc, char **argv)
6115 ztest_shared_callstate_t *zc;
6121 char *fd_data_str = getenv("ZTEST_FD_DATA");
6123 (void) setvbuf(stdout, NULL, _IOLBF, 0);
6125 dprintf_setup(&argc, argv);
6126 zfs_deadman_synctime_ms = 300000;
6128 ztest_fd_rand = open("/dev/urandom", O_RDONLY);
6129 ASSERT3S(ztest_fd_rand, >=, 0);
6132 process_options(argc, argv);
6137 bcopy(&ztest_opts, ztest_shared_opts,
6138 sizeof (*ztest_shared_opts));
6140 ztest_fd_data = atoi(fd_data_str);
6142 bcopy(ztest_shared_opts, &ztest_opts, sizeof (ztest_opts));
6144 ASSERT3U(ztest_opts.zo_datasets, ==, ztest_shared_hdr->zh_ds_count);
6146 /* Override location of zpool.cache */
6147 VERIFY3U(asprintf((char **)&spa_config_path, "%s/zpool.cache",
6148 ztest_opts.zo_dir), !=, -1);
6150 ztest_ds = umem_alloc(ztest_opts.zo_datasets * sizeof (ztest_ds_t),
6155 metaslab_gang_bang = ztest_opts.zo_metaslab_gang_bang;
6156 metaslab_df_alloc_threshold =
6157 zs->zs_metaslab_df_alloc_threshold;
6166 hasalt = (strlen(ztest_opts.zo_alt_ztest) != 0);
6168 if (ztest_opts.zo_verbose >= 1) {
6169 (void) printf("%llu vdevs, %d datasets, %d threads,"
6170 " %llu seconds...\n",
6171 (u_longlong_t)ztest_opts.zo_vdevs,
6172 ztest_opts.zo_datasets,
6173 ztest_opts.zo_threads,
6174 (u_longlong_t)ztest_opts.zo_time);
6177 cmd = umem_alloc(MAXNAMELEN, UMEM_NOFAIL);
6178 (void) strlcpy(cmd, getexecname(), MAXNAMELEN);
6180 zs->zs_do_init = B_TRUE;
6181 if (strlen(ztest_opts.zo_alt_ztest) != 0) {
6182 if (ztest_opts.zo_verbose >= 1) {
6183 (void) printf("Executing older ztest for "
6184 "initialization: %s\n", ztest_opts.zo_alt_ztest);
6186 VERIFY(!exec_child(ztest_opts.zo_alt_ztest,
6187 ztest_opts.zo_alt_libpath, B_FALSE, NULL));
6189 VERIFY(!exec_child(NULL, NULL, B_FALSE, NULL));
6191 zs->zs_do_init = B_FALSE;
6193 zs->zs_proc_start = gethrtime();
6194 zs->zs_proc_stop = zs->zs_proc_start + ztest_opts.zo_time * NANOSEC;
6196 for (int f = 0; f < ZTEST_FUNCS; f++) {
6197 zi = &ztest_info[f];
6198 zc = ZTEST_GET_SHARED_CALLSTATE(f);
6199 if (zs->zs_proc_start + zi->zi_interval[0] > zs->zs_proc_stop)
6200 zc->zc_next = UINT64_MAX;
6202 zc->zc_next = zs->zs_proc_start +
6203 ztest_random(2 * zi->zi_interval[0] + 1);
6207 * Run the tests in a loop. These tests include fault injection
6208 * to verify that self-healing data works, and forced crashes
6209 * to verify that we never lose on-disk consistency.
6211 while (gethrtime() < zs->zs_proc_stop) {
6216 * Initialize the workload counters for each function.
6218 for (int f = 0; f < ZTEST_FUNCS; f++) {
6219 zc = ZTEST_GET_SHARED_CALLSTATE(f);
6224 /* Set the allocation switch size */
6225 zs->zs_metaslab_df_alloc_threshold =
6226 ztest_random(zs->zs_metaslab_sz / 4) + 1;
6228 if (!hasalt || ztest_random(2) == 0) {
6229 if (hasalt && ztest_opts.zo_verbose >= 1) {
6230 (void) printf("Executing newer ztest: %s\n",
6234 killed = exec_child(cmd, NULL, B_TRUE, &status);
6236 if (hasalt && ztest_opts.zo_verbose >= 1) {
6237 (void) printf("Executing older ztest: %s\n",
6238 ztest_opts.zo_alt_ztest);
6241 killed = exec_child(ztest_opts.zo_alt_ztest,
6242 ztest_opts.zo_alt_libpath, B_TRUE, &status);
6249 if (ztest_opts.zo_verbose >= 1) {
6250 hrtime_t now = gethrtime();
6252 now = MIN(now, zs->zs_proc_stop);
6253 print_time(zs->zs_proc_stop - now, timebuf);
6254 nicenum(zs->zs_space, numbuf);
6256 (void) printf("Pass %3d, %8s, %3llu ENOSPC, "
6257 "%4.1f%% of %5s used, %3.0f%% done, %8s to go\n",
6259 WIFEXITED(status) ? "Complete" : "SIGKILL",
6260 (u_longlong_t)zs->zs_enospc_count,
6261 100.0 * zs->zs_alloc / zs->zs_space,
6263 100.0 * (now - zs->zs_proc_start) /
6264 (ztest_opts.zo_time * NANOSEC), timebuf);
6267 if (ztest_opts.zo_verbose >= 2) {
6268 (void) printf("\nWorkload summary:\n\n");
6269 (void) printf("%7s %9s %s\n",
6270 "Calls", "Time", "Function");
6271 (void) printf("%7s %9s %s\n",
6272 "-----", "----", "--------");
6273 for (int f = 0; f < ZTEST_FUNCS; f++) {
6276 zi = &ztest_info[f];
6277 zc = ZTEST_GET_SHARED_CALLSTATE(f);
6278 print_time(zc->zc_time, timebuf);
6279 (void) dladdr((void *)zi->zi_func, &dli);
6280 (void) printf("%7llu %9s %s\n",
6281 (u_longlong_t)zc->zc_count, timebuf,
6284 (void) printf("\n");
6288 * It's possible that we killed a child during a rename test,
6289 * in which case we'll have a 'ztest_tmp' pool lying around
6290 * instead of 'ztest'. Do a blind rename in case this happened.
6293 if (spa_open(ztest_opts.zo_pool, &spa, FTAG) == 0) {
6294 spa_close(spa, FTAG);
6296 char tmpname[MAXNAMELEN];
6298 kernel_init(FREAD | FWRITE);
6299 (void) snprintf(tmpname, sizeof (tmpname), "%s_tmp",
6300 ztest_opts.zo_pool);
6301 (void) spa_rename(tmpname, ztest_opts.zo_pool);
6305 ztest_run_zdb(ztest_opts.zo_pool);
6308 if (ztest_opts.zo_verbose >= 1) {
6310 (void) printf("%d runs of older ztest: %s\n", older,
6311 ztest_opts.zo_alt_ztest);
6312 (void) printf("%d runs of newer ztest: %s\n", newer,
6315 (void) printf("%d killed, %d completed, %.0f%% kill rate\n",
6316 kills, iters - kills, (100.0 * kills) / MAX(1, iters));
6319 umem_free(cmd, MAXNAMELEN);