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) 2012 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.
29 * The objective of this program is to provide a DMU/ZAP/SPA stress test
30 * that runs entirely in userland, is easy to use, and easy to extend.
32 * The overall design of the ztest program is as follows:
34 * (1) For each major functional area (e.g. adding vdevs to a pool,
35 * creating and destroying datasets, reading and writing objects, etc)
36 * we have a simple routine to test that functionality. These
37 * individual routines do not have to do anything "stressful".
39 * (2) We turn these simple functionality tests into a stress test by
40 * running them all in parallel, with as many threads as desired,
41 * and spread across as many datasets, objects, and vdevs as desired.
43 * (3) While all this is happening, we inject faults into the pool to
44 * verify that self-healing data really works.
46 * (4) Every time we open a dataset, we change its checksum and compression
47 * functions. Thus even individual objects vary from block to block
48 * in which checksum they use and whether they're compressed.
50 * (5) To verify that we never lose on-disk consistency after a crash,
51 * we run the entire test in a child of the main process.
52 * At random times, the child self-immolates with a SIGKILL.
53 * This is the software equivalent of pulling the power cord.
54 * The parent then runs the test again, using the existing
55 * storage pool, as many times as desired. If backwards compatability
56 * testing is enabled ztest will sometimes run the "older" version
57 * of ztest after a SIGKILL.
59 * (6) To verify that we don't have future leaks or temporal incursions,
60 * many of the functional tests record the transaction group number
61 * as part of their data. When reading old data, they verify that
62 * the transaction group number is less than the current, open txg.
63 * If you add a new test, please do this if applicable.
65 * When run with no arguments, ztest runs for about five minutes and
66 * produces no output if successful. To get a little bit of information,
67 * specify -V. To get more information, specify -VV, and so on.
69 * To turn this into an overnight stress test, use -T to specify run time.
71 * You can ask more more vdevs [-v], datasets [-d], or threads [-t]
72 * to increase the pool capacity, fanout, and overall stress level.
74 * Use the -k option to set the desired frequency of kills.
76 * When ztest invokes itself it passes all relevant information through a
77 * temporary file which is mmap-ed in the child process. This allows shared
78 * memory to survive the exec syscall. The ztest_shared_hdr_t struct is always
79 * stored at offset 0 of this file and contains information on the size and
80 * number of shared structures in the file. The information stored in this file
81 * must remain backwards compatible with older versions of ztest so that
82 * ztest can invoke them during backwards compatibility testing (-B).
85 #include <sys/zfs_context.h>
91 #include <sys/dmu_objset.h>
97 #include <sys/resource.h>
100 #include <sys/zil_impl.h>
101 #include <sys/vdev_impl.h>
102 #include <sys/vdev_file.h>
103 #include <sys/spa_impl.h>
104 #include <sys/metaslab_impl.h>
105 #include <sys/dsl_prop.h>
106 #include <sys/dsl_dataset.h>
107 #include <sys/dsl_scan.h>
108 #include <sys/zio_checksum.h>
109 #include <sys/refcount.h>
110 #include <sys/zfeature.h>
112 #include <stdio_ext.h>
121 #include <sys/fs/zfs.h>
122 #include <libnvpair.h>
124 static int ztest_fd_data = -1;
125 static int ztest_fd_rand = -1;
127 typedef struct ztest_shared_hdr {
128 uint64_t zh_hdr_size;
129 uint64_t zh_opts_size;
131 uint64_t zh_stats_size;
132 uint64_t zh_stats_count;
134 uint64_t zh_ds_count;
135 } ztest_shared_hdr_t;
137 static ztest_shared_hdr_t *ztest_shared_hdr;
139 typedef struct ztest_shared_opts {
140 char zo_pool[MAXNAMELEN];
141 char zo_dir[MAXNAMELEN];
142 char zo_alt_ztest[MAXNAMELEN];
143 char zo_alt_libpath[MAXNAMELEN];
145 uint64_t zo_vdevtime;
153 uint64_t zo_passtime;
154 uint64_t zo_killrate;
158 uint64_t zo_maxloops;
159 uint64_t zo_metaslab_gang_bang;
160 } ztest_shared_opts_t;
162 static const ztest_shared_opts_t ztest_opts_defaults = {
163 .zo_pool = { 'z', 't', 'e', 's', 't', '\0' },
164 .zo_dir = { '/', 't', 'm', 'p', '\0' },
165 .zo_alt_ztest = { '\0' },
166 .zo_alt_libpath = { '\0' },
168 .zo_ashift = SPA_MINBLOCKSHIFT,
171 .zo_raidz_parity = 1,
172 .zo_vdev_size = SPA_MINDEVSIZE,
175 .zo_passtime = 60, /* 60 seconds */
176 .zo_killrate = 70, /* 70% kill rate */
179 .zo_time = 300, /* 5 minutes */
180 .zo_maxloops = 50, /* max loops during spa_freeze() */
181 .zo_metaslab_gang_bang = 32 << 10
184 extern uint64_t metaslab_gang_bang;
185 extern uint64_t metaslab_df_alloc_threshold;
187 static ztest_shared_opts_t *ztest_shared_opts;
188 static ztest_shared_opts_t ztest_opts;
190 typedef struct ztest_shared_ds {
194 static ztest_shared_ds_t *ztest_shared_ds;
195 #define ZTEST_GET_SHARED_DS(d) (&ztest_shared_ds[d])
197 #define BT_MAGIC 0x123456789abcdefULL
198 #define MAXFAULTS() \
199 (MAX(zs->zs_mirrors, 1) * (ztest_opts.zo_raidz_parity + 1) - 1)
203 ZTEST_IO_WRITE_PATTERN,
204 ZTEST_IO_WRITE_ZEROES,
211 typedef struct ztest_block_tag {
221 typedef struct bufwad {
228 * XXX -- fix zfs range locks to be generic so we can use them here.
250 #define ZTEST_RANGE_LOCKS 64
251 #define ZTEST_OBJECT_LOCKS 64
254 * Object descriptor. Used as a template for object lookup/create/remove.
256 typedef struct ztest_od {
259 dmu_object_type_t od_type;
260 dmu_object_type_t od_crtype;
261 uint64_t od_blocksize;
262 uint64_t od_crblocksize;
265 char od_name[MAXNAMELEN];
271 typedef struct ztest_ds {
272 ztest_shared_ds_t *zd_shared;
274 rwlock_t zd_zilog_lock;
276 ztest_od_t *zd_od; /* debugging aid */
277 char zd_name[MAXNAMELEN];
278 mutex_t zd_dirobj_lock;
279 rll_t zd_object_lock[ZTEST_OBJECT_LOCKS];
280 rll_t zd_range_lock[ZTEST_RANGE_LOCKS];
284 * Per-iteration state.
286 typedef void ztest_func_t(ztest_ds_t *zd, uint64_t id);
288 typedef struct ztest_info {
289 ztest_func_t *zi_func; /* test function */
290 uint64_t zi_iters; /* iterations per execution */
291 uint64_t *zi_interval; /* execute every <interval> seconds */
294 typedef struct ztest_shared_callstate {
295 uint64_t zc_count; /* per-pass count */
296 uint64_t zc_time; /* per-pass time */
297 uint64_t zc_next; /* next time to call this function */
298 } ztest_shared_callstate_t;
300 static ztest_shared_callstate_t *ztest_shared_callstate;
301 #define ZTEST_GET_SHARED_CALLSTATE(c) (&ztest_shared_callstate[c])
304 * Note: these aren't static because we want dladdr() to work.
306 ztest_func_t ztest_dmu_read_write;
307 ztest_func_t ztest_dmu_write_parallel;
308 ztest_func_t ztest_dmu_object_alloc_free;
309 ztest_func_t ztest_dmu_commit_callbacks;
310 ztest_func_t ztest_zap;
311 ztest_func_t ztest_zap_parallel;
312 ztest_func_t ztest_zil_commit;
313 ztest_func_t ztest_zil_remount;
314 ztest_func_t ztest_dmu_read_write_zcopy;
315 ztest_func_t ztest_dmu_objset_create_destroy;
316 ztest_func_t ztest_dmu_prealloc;
317 ztest_func_t ztest_fzap;
318 ztest_func_t ztest_dmu_snapshot_create_destroy;
319 ztest_func_t ztest_dsl_prop_get_set;
320 ztest_func_t ztest_spa_prop_get_set;
321 ztest_func_t ztest_spa_create_destroy;
322 ztest_func_t ztest_fault_inject;
323 ztest_func_t ztest_ddt_repair;
324 ztest_func_t ztest_dmu_snapshot_hold;
325 ztest_func_t ztest_spa_rename;
326 ztest_func_t ztest_scrub;
327 ztest_func_t ztest_dsl_dataset_promote_busy;
328 ztest_func_t ztest_vdev_attach_detach;
329 ztest_func_t ztest_vdev_LUN_growth;
330 ztest_func_t ztest_vdev_add_remove;
331 ztest_func_t ztest_vdev_aux_add_remove;
332 ztest_func_t ztest_split_pool;
333 ztest_func_t ztest_reguid;
334 ztest_func_t ztest_spa_upgrade;
336 uint64_t zopt_always = 0ULL * NANOSEC; /* all the time */
337 uint64_t zopt_incessant = 1ULL * NANOSEC / 10; /* every 1/10 second */
338 uint64_t zopt_often = 1ULL * NANOSEC; /* every second */
339 uint64_t zopt_sometimes = 10ULL * NANOSEC; /* every 10 seconds */
340 uint64_t zopt_rarely = 60ULL * NANOSEC; /* every 60 seconds */
342 ztest_info_t ztest_info[] = {
343 { ztest_dmu_read_write, 1, &zopt_always },
344 { ztest_dmu_write_parallel, 10, &zopt_always },
345 { ztest_dmu_object_alloc_free, 1, &zopt_always },
346 { ztest_dmu_commit_callbacks, 1, &zopt_always },
347 { ztest_zap, 30, &zopt_always },
348 { ztest_zap_parallel, 100, &zopt_always },
349 { ztest_split_pool, 1, &zopt_always },
350 { ztest_zil_commit, 1, &zopt_incessant },
351 { ztest_zil_remount, 1, &zopt_sometimes },
352 { ztest_dmu_read_write_zcopy, 1, &zopt_often },
353 { ztest_dmu_objset_create_destroy, 1, &zopt_often },
354 { ztest_dsl_prop_get_set, 1, &zopt_often },
355 { ztest_spa_prop_get_set, 1, &zopt_sometimes },
357 { ztest_dmu_prealloc, 1, &zopt_sometimes },
359 { ztest_fzap, 1, &zopt_sometimes },
360 { ztest_dmu_snapshot_create_destroy, 1, &zopt_sometimes },
361 { ztest_spa_create_destroy, 1, &zopt_sometimes },
362 { ztest_fault_inject, 1, &zopt_sometimes },
363 { ztest_ddt_repair, 1, &zopt_sometimes },
364 { ztest_dmu_snapshot_hold, 1, &zopt_sometimes },
365 { ztest_reguid, 1, &zopt_sometimes },
366 { ztest_spa_rename, 1, &zopt_rarely },
367 { ztest_scrub, 1, &zopt_rarely },
368 { ztest_spa_upgrade, 1, &zopt_rarely },
369 { ztest_dsl_dataset_promote_busy, 1, &zopt_rarely },
370 { ztest_vdev_attach_detach, 1, &zopt_rarely },
371 { ztest_vdev_LUN_growth, 1, &zopt_rarely },
372 { ztest_vdev_add_remove, 1,
373 &ztest_opts.zo_vdevtime },
374 { ztest_vdev_aux_add_remove, 1,
375 &ztest_opts.zo_vdevtime },
378 #define ZTEST_FUNCS (sizeof (ztest_info) / sizeof (ztest_info_t))
381 * The following struct is used to hold a list of uncalled commit callbacks.
382 * The callbacks are ordered by txg number.
384 typedef struct ztest_cb_list {
385 mutex_t zcl_callbacks_lock;
386 list_t zcl_callbacks;
390 * Stuff we need to share writably between parent and child.
392 typedef struct ztest_shared {
393 boolean_t zs_do_init;
394 hrtime_t zs_proc_start;
395 hrtime_t zs_proc_stop;
396 hrtime_t zs_thread_start;
397 hrtime_t zs_thread_stop;
398 hrtime_t zs_thread_kill;
399 uint64_t zs_enospc_count;
400 uint64_t zs_vdev_next_leaf;
401 uint64_t zs_vdev_aux;
406 uint64_t zs_metaslab_sz;
407 uint64_t zs_metaslab_df_alloc_threshold;
411 #define ID_PARALLEL -1ULL
413 static char ztest_dev_template[] = "%s/%s.%llua";
414 static char ztest_aux_template[] = "%s/%s.%s.%llu";
415 ztest_shared_t *ztest_shared;
417 static spa_t *ztest_spa = NULL;
418 static ztest_ds_t *ztest_ds;
420 static mutex_t ztest_vdev_lock;
423 * The ztest_name_lock protects the pool and dataset namespace used by
424 * the individual tests. To modify the namespace, consumers must grab
425 * this lock as writer. Grabbing the lock as reader will ensure that the
426 * namespace does not change while the lock is held.
428 static rwlock_t ztest_name_lock;
430 static boolean_t ztest_dump_core = B_TRUE;
431 static boolean_t ztest_exiting;
433 /* Global commit callback list */
434 static ztest_cb_list_t zcl;
437 ZTEST_META_DNODE = 0,
442 static void usage(boolean_t) __NORETURN;
445 * These libumem hooks provide a reasonable set of defaults for the allocator's
446 * debugging facilities.
451 return ("default,verbose"); /* $UMEM_DEBUG setting */
455 _umem_logging_init(void)
457 return ("fail,contents"); /* $UMEM_LOGGING setting */
460 #define FATAL_MSG_SZ 1024
465 fatal(int do_perror, char *message, ...)
468 int save_errno = errno;
469 char buf[FATAL_MSG_SZ];
471 (void) fflush(stdout);
473 va_start(args, message);
474 (void) sprintf(buf, "ztest: ");
476 (void) vsprintf(buf + strlen(buf), message, args);
479 (void) snprintf(buf + strlen(buf), FATAL_MSG_SZ - strlen(buf),
480 ": %s", strerror(save_errno));
482 (void) fprintf(stderr, "%s\n", buf);
483 fatal_msg = buf; /* to ease debugging */
490 str2shift(const char *buf)
492 const char *ends = "BKMGTPEZ";
497 for (i = 0; i < strlen(ends); i++) {
498 if (toupper(buf[0]) == ends[i])
501 if (i == strlen(ends)) {
502 (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n",
506 if (buf[1] == '\0' || (toupper(buf[1]) == 'B' && buf[2] == '\0')) {
509 (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n", buf);
515 nicenumtoull(const char *buf)
520 val = strtoull(buf, &end, 0);
522 (void) fprintf(stderr, "ztest: bad numeric value: %s\n", buf);
524 } else if (end[0] == '.') {
525 double fval = strtod(buf, &end);
526 fval *= pow(2, str2shift(end));
527 if (fval > UINT64_MAX) {
528 (void) fprintf(stderr, "ztest: value too large: %s\n",
532 val = (uint64_t)fval;
534 int shift = str2shift(end);
535 if (shift >= 64 || (val << shift) >> shift != val) {
536 (void) fprintf(stderr, "ztest: value too large: %s\n",
546 usage(boolean_t requested)
548 const ztest_shared_opts_t *zo = &ztest_opts_defaults;
550 char nice_vdev_size[10];
551 char nice_gang_bang[10];
552 FILE *fp = requested ? stdout : stderr;
554 nicenum(zo->zo_vdev_size, nice_vdev_size);
555 nicenum(zo->zo_metaslab_gang_bang, nice_gang_bang);
557 (void) fprintf(fp, "Usage: %s\n"
558 "\t[-v vdevs (default: %llu)]\n"
559 "\t[-s size_of_each_vdev (default: %s)]\n"
560 "\t[-a alignment_shift (default: %d)] use 0 for random\n"
561 "\t[-m mirror_copies (default: %d)]\n"
562 "\t[-r raidz_disks (default: %d)]\n"
563 "\t[-R raidz_parity (default: %d)]\n"
564 "\t[-d datasets (default: %d)]\n"
565 "\t[-t threads (default: %d)]\n"
566 "\t[-g gang_block_threshold (default: %s)]\n"
567 "\t[-i init_count (default: %d)] initialize pool i times\n"
568 "\t[-k kill_percentage (default: %llu%%)]\n"
569 "\t[-p pool_name (default: %s)]\n"
570 "\t[-f dir (default: %s)] file directory for vdev files\n"
571 "\t[-V] verbose (use multiple times for ever more blather)\n"
572 "\t[-E] use existing pool instead of creating new one\n"
573 "\t[-T time (default: %llu sec)] total run time\n"
574 "\t[-F freezeloops (default: %llu)] max loops in spa_freeze()\n"
575 "\t[-P passtime (default: %llu sec)] time per pass\n"
576 "\t[-B alt_ztest (default: <none>)] alternate ztest path\n"
577 "\t[-h] (print help)\n"
580 (u_longlong_t)zo->zo_vdevs, /* -v */
581 nice_vdev_size, /* -s */
582 zo->zo_ashift, /* -a */
583 zo->zo_mirrors, /* -m */
584 zo->zo_raidz, /* -r */
585 zo->zo_raidz_parity, /* -R */
586 zo->zo_datasets, /* -d */
587 zo->zo_threads, /* -t */
588 nice_gang_bang, /* -g */
589 zo->zo_init, /* -i */
590 (u_longlong_t)zo->zo_killrate, /* -k */
591 zo->zo_pool, /* -p */
593 (u_longlong_t)zo->zo_time, /* -T */
594 (u_longlong_t)zo->zo_maxloops, /* -F */
595 (u_longlong_t)zo->zo_passtime);
596 exit(requested ? 0 : 1);
600 process_options(int argc, char **argv)
603 ztest_shared_opts_t *zo = &ztest_opts;
607 char altdir[MAXNAMELEN] = { 0 };
609 bcopy(&ztest_opts_defaults, zo, sizeof (*zo));
611 while ((opt = getopt(argc, argv,
612 "v:s:a:m:r:R:d:t:g:i:k:p:f:VET:P:hF:B:")) != EOF) {
629 value = nicenumtoull(optarg);
633 zo->zo_vdevs = value;
636 zo->zo_vdev_size = MAX(SPA_MINDEVSIZE, value);
639 zo->zo_ashift = value;
642 zo->zo_mirrors = value;
645 zo->zo_raidz = MAX(1, value);
648 zo->zo_raidz_parity = MIN(MAX(value, 1), 3);
651 zo->zo_datasets = MAX(1, value);
654 zo->zo_threads = MAX(1, value);
657 zo->zo_metaslab_gang_bang = MAX(SPA_MINBLOCKSIZE << 1,
664 zo->zo_killrate = value;
667 (void) strlcpy(zo->zo_pool, optarg,
668 sizeof (zo->zo_pool));
671 path = realpath(optarg, NULL);
673 (void) fprintf(stderr, "error: %s: %s\n",
674 optarg, strerror(errno));
677 (void) strlcpy(zo->zo_dir, path,
678 sizeof (zo->zo_dir));
691 zo->zo_passtime = MAX(1, value);
694 zo->zo_maxloops = MAX(1, value);
697 (void) strlcpy(altdir, optarg, sizeof (altdir));
709 zo->zo_raidz_parity = MIN(zo->zo_raidz_parity, zo->zo_raidz - 1);
712 (zo->zo_vdevs > 0 ? zo->zo_time * NANOSEC / zo->zo_vdevs :
715 if (strlen(altdir) > 0) {
723 cmd = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
724 realaltdir = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
726 VERIFY(NULL != realpath(getexecname(), cmd));
727 if (0 != access(altdir, F_OK)) {
728 ztest_dump_core = B_FALSE;
729 fatal(B_TRUE, "invalid alternate ztest path: %s",
732 VERIFY(NULL != realpath(altdir, realaltdir));
735 * 'cmd' should be of the form "<anything>/usr/bin/<isa>/ztest".
736 * We want to extract <isa> to determine if we should use
737 * 32 or 64 bit binaries.
739 bin = strstr(cmd, "/usr/bin/");
740 ztest = strstr(bin, "/ztest");
742 isalen = ztest - isa;
743 (void) snprintf(zo->zo_alt_ztest, sizeof (zo->zo_alt_ztest),
744 "%s/usr/bin/%.*s/ztest", realaltdir, isalen, isa);
745 (void) snprintf(zo->zo_alt_libpath, sizeof (zo->zo_alt_libpath),
746 "%s/usr/lib/%.*s", realaltdir, isalen, isa);
748 if (0 != access(zo->zo_alt_ztest, X_OK)) {
749 ztest_dump_core = B_FALSE;
750 fatal(B_TRUE, "invalid alternate ztest: %s",
752 } else if (0 != access(zo->zo_alt_libpath, X_OK)) {
753 ztest_dump_core = B_FALSE;
754 fatal(B_TRUE, "invalid alternate lib directory %s",
758 umem_free(cmd, MAXPATHLEN);
759 umem_free(realaltdir, MAXPATHLEN);
764 ztest_kill(ztest_shared_t *zs)
766 zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(ztest_spa));
767 zs->zs_space = metaslab_class_get_space(spa_normal_class(ztest_spa));
768 (void) kill(getpid(), SIGKILL);
772 ztest_random(uint64_t range)
776 ASSERT3S(ztest_fd_rand, >=, 0);
781 if (read(ztest_fd_rand, &r, sizeof (r)) != sizeof (r))
782 fatal(1, "short read from /dev/urandom");
789 ztest_record_enospc(const char *s)
791 ztest_shared->zs_enospc_count++;
795 ztest_get_ashift(void)
797 if (ztest_opts.zo_ashift == 0)
798 return (SPA_MINBLOCKSHIFT + ztest_random(3));
799 return (ztest_opts.zo_ashift);
803 make_vdev_file(char *path, char *aux, char *pool, size_t size, uint64_t ashift)
805 char pathbuf[MAXPATHLEN];
810 ashift = ztest_get_ashift();
816 vdev = ztest_shared->zs_vdev_aux;
817 (void) snprintf(path, sizeof (pathbuf),
818 ztest_aux_template, ztest_opts.zo_dir,
819 pool == NULL ? ztest_opts.zo_pool : pool,
822 vdev = ztest_shared->zs_vdev_next_leaf++;
823 (void) snprintf(path, sizeof (pathbuf),
824 ztest_dev_template, ztest_opts.zo_dir,
825 pool == NULL ? ztest_opts.zo_pool : pool, vdev);
830 int fd = open(path, O_RDWR | O_CREAT | O_TRUNC, 0666);
832 fatal(1, "can't open %s", path);
833 if (ftruncate(fd, size) != 0)
834 fatal(1, "can't ftruncate %s", path);
838 VERIFY(nvlist_alloc(&file, NV_UNIQUE_NAME, 0) == 0);
839 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_TYPE, VDEV_TYPE_FILE) == 0);
840 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_PATH, path) == 0);
841 VERIFY(nvlist_add_uint64(file, ZPOOL_CONFIG_ASHIFT, ashift) == 0);
847 make_vdev_raidz(char *path, char *aux, char *pool, size_t size,
848 uint64_t ashift, int r)
850 nvlist_t *raidz, **child;
854 return (make_vdev_file(path, aux, pool, size, ashift));
855 child = umem_alloc(r * sizeof (nvlist_t *), UMEM_NOFAIL);
857 for (c = 0; c < r; c++)
858 child[c] = make_vdev_file(path, aux, pool, size, ashift);
860 VERIFY(nvlist_alloc(&raidz, NV_UNIQUE_NAME, 0) == 0);
861 VERIFY(nvlist_add_string(raidz, ZPOOL_CONFIG_TYPE,
862 VDEV_TYPE_RAIDZ) == 0);
863 VERIFY(nvlist_add_uint64(raidz, ZPOOL_CONFIG_NPARITY,
864 ztest_opts.zo_raidz_parity) == 0);
865 VERIFY(nvlist_add_nvlist_array(raidz, ZPOOL_CONFIG_CHILDREN,
868 for (c = 0; c < r; c++)
869 nvlist_free(child[c]);
871 umem_free(child, r * sizeof (nvlist_t *));
877 make_vdev_mirror(char *path, char *aux, char *pool, size_t size,
878 uint64_t ashift, int r, int m)
880 nvlist_t *mirror, **child;
884 return (make_vdev_raidz(path, aux, pool, size, ashift, r));
886 child = umem_alloc(m * sizeof (nvlist_t *), UMEM_NOFAIL);
888 for (c = 0; c < m; c++)
889 child[c] = make_vdev_raidz(path, aux, pool, size, ashift, r);
891 VERIFY(nvlist_alloc(&mirror, NV_UNIQUE_NAME, 0) == 0);
892 VERIFY(nvlist_add_string(mirror, ZPOOL_CONFIG_TYPE,
893 VDEV_TYPE_MIRROR) == 0);
894 VERIFY(nvlist_add_nvlist_array(mirror, ZPOOL_CONFIG_CHILDREN,
897 for (c = 0; c < m; c++)
898 nvlist_free(child[c]);
900 umem_free(child, m * sizeof (nvlist_t *));
906 make_vdev_root(char *path, char *aux, char *pool, size_t size, uint64_t ashift,
907 int log, int r, int m, int t)
909 nvlist_t *root, **child;
914 child = umem_alloc(t * sizeof (nvlist_t *), UMEM_NOFAIL);
916 for (c = 0; c < t; c++) {
917 child[c] = make_vdev_mirror(path, aux, pool, size, ashift,
919 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_IS_LOG,
923 VERIFY(nvlist_alloc(&root, NV_UNIQUE_NAME, 0) == 0);
924 VERIFY(nvlist_add_string(root, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) == 0);
925 VERIFY(nvlist_add_nvlist_array(root, aux ? aux : ZPOOL_CONFIG_CHILDREN,
928 for (c = 0; c < t; c++)
929 nvlist_free(child[c]);
931 umem_free(child, t * sizeof (nvlist_t *));
937 * Find a random spa version. Returns back a random spa version in the
938 * range [initial_version, SPA_VERSION_FEATURES].
941 ztest_random_spa_version(uint64_t initial_version)
943 uint64_t version = initial_version;
945 if (version <= SPA_VERSION_BEFORE_FEATURES) {
947 ztest_random(SPA_VERSION_BEFORE_FEATURES - version + 1);
950 if (version > SPA_VERSION_BEFORE_FEATURES)
951 version = SPA_VERSION_FEATURES;
953 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
958 ztest_random_blocksize(void)
960 return (1 << (SPA_MINBLOCKSHIFT +
961 ztest_random(SPA_MAXBLOCKSHIFT - SPA_MINBLOCKSHIFT + 1)));
965 ztest_random_ibshift(void)
967 return (DN_MIN_INDBLKSHIFT +
968 ztest_random(DN_MAX_INDBLKSHIFT - DN_MIN_INDBLKSHIFT + 1));
972 ztest_random_vdev_top(spa_t *spa, boolean_t log_ok)
975 vdev_t *rvd = spa->spa_root_vdev;
978 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
981 top = ztest_random(rvd->vdev_children);
982 tvd = rvd->vdev_child[top];
983 } while (tvd->vdev_ishole || (tvd->vdev_islog && !log_ok) ||
984 tvd->vdev_mg == NULL || tvd->vdev_mg->mg_class == NULL);
990 ztest_random_dsl_prop(zfs_prop_t prop)
995 value = zfs_prop_random_value(prop, ztest_random(-1ULL));
996 } while (prop == ZFS_PROP_CHECKSUM && value == ZIO_CHECKSUM_OFF);
1002 ztest_dsl_prop_set_uint64(char *osname, zfs_prop_t prop, uint64_t value,
1005 const char *propname = zfs_prop_to_name(prop);
1006 const char *valname;
1007 char setpoint[MAXPATHLEN];
1011 error = dsl_prop_set(osname, propname,
1012 (inherit ? ZPROP_SRC_NONE : ZPROP_SRC_LOCAL),
1013 sizeof (value), 1, &value);
1015 if (error == ENOSPC) {
1016 ztest_record_enospc(FTAG);
1021 VERIFY3U(dsl_prop_get(osname, propname, sizeof (curval),
1022 1, &curval, setpoint), ==, 0);
1024 if (ztest_opts.zo_verbose >= 6) {
1025 VERIFY(zfs_prop_index_to_string(prop, curval, &valname) == 0);
1026 (void) printf("%s %s = %s at '%s'\n",
1027 osname, propname, valname, setpoint);
1034 ztest_spa_prop_set_uint64(zpool_prop_t prop, uint64_t value)
1036 spa_t *spa = ztest_spa;
1037 nvlist_t *props = NULL;
1040 VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
1041 VERIFY(nvlist_add_uint64(props, zpool_prop_to_name(prop), value) == 0);
1043 error = spa_prop_set(spa, props);
1047 if (error == ENOSPC) {
1048 ztest_record_enospc(FTAG);
1057 ztest_rll_init(rll_t *rll)
1059 rll->rll_writer = NULL;
1060 rll->rll_readers = 0;
1061 VERIFY(_mutex_init(&rll->rll_lock, USYNC_THREAD, NULL) == 0);
1062 VERIFY(cond_init(&rll->rll_cv, USYNC_THREAD, NULL) == 0);
1066 ztest_rll_destroy(rll_t *rll)
1068 ASSERT(rll->rll_writer == NULL);
1069 ASSERT(rll->rll_readers == 0);
1070 VERIFY(_mutex_destroy(&rll->rll_lock) == 0);
1071 VERIFY(cond_destroy(&rll->rll_cv) == 0);
1075 ztest_rll_lock(rll_t *rll, rl_type_t type)
1077 VERIFY(mutex_lock(&rll->rll_lock) == 0);
1079 if (type == RL_READER) {
1080 while (rll->rll_writer != NULL)
1081 (void) cond_wait(&rll->rll_cv, &rll->rll_lock);
1084 while (rll->rll_writer != NULL || rll->rll_readers)
1085 (void) cond_wait(&rll->rll_cv, &rll->rll_lock);
1086 rll->rll_writer = curthread;
1089 VERIFY(mutex_unlock(&rll->rll_lock) == 0);
1093 ztest_rll_unlock(rll_t *rll)
1095 VERIFY(mutex_lock(&rll->rll_lock) == 0);
1097 if (rll->rll_writer) {
1098 ASSERT(rll->rll_readers == 0);
1099 rll->rll_writer = NULL;
1101 ASSERT(rll->rll_readers != 0);
1102 ASSERT(rll->rll_writer == NULL);
1106 if (rll->rll_writer == NULL && rll->rll_readers == 0)
1107 VERIFY(cond_broadcast(&rll->rll_cv) == 0);
1109 VERIFY(mutex_unlock(&rll->rll_lock) == 0);
1113 ztest_object_lock(ztest_ds_t *zd, uint64_t object, rl_type_t type)
1115 rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1117 ztest_rll_lock(rll, type);
1121 ztest_object_unlock(ztest_ds_t *zd, uint64_t object)
1123 rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1125 ztest_rll_unlock(rll);
1129 ztest_range_lock(ztest_ds_t *zd, uint64_t object, uint64_t offset,
1130 uint64_t size, rl_type_t type)
1132 uint64_t hash = object ^ (offset % (ZTEST_RANGE_LOCKS + 1));
1133 rll_t *rll = &zd->zd_range_lock[hash & (ZTEST_RANGE_LOCKS - 1)];
1136 rl = umem_alloc(sizeof (*rl), UMEM_NOFAIL);
1137 rl->rl_object = object;
1138 rl->rl_offset = offset;
1142 ztest_rll_lock(rll, type);
1148 ztest_range_unlock(rl_t *rl)
1150 rll_t *rll = rl->rl_lock;
1152 ztest_rll_unlock(rll);
1154 umem_free(rl, sizeof (*rl));
1158 ztest_zd_init(ztest_ds_t *zd, ztest_shared_ds_t *szd, objset_t *os)
1161 zd->zd_zilog = dmu_objset_zil(os);
1162 zd->zd_shared = szd;
1163 dmu_objset_name(os, zd->zd_name);
1165 if (zd->zd_shared != NULL)
1166 zd->zd_shared->zd_seq = 0;
1168 VERIFY(rwlock_init(&zd->zd_zilog_lock, USYNC_THREAD, NULL) == 0);
1169 VERIFY(_mutex_init(&zd->zd_dirobj_lock, USYNC_THREAD, NULL) == 0);
1171 for (int l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1172 ztest_rll_init(&zd->zd_object_lock[l]);
1174 for (int l = 0; l < ZTEST_RANGE_LOCKS; l++)
1175 ztest_rll_init(&zd->zd_range_lock[l]);
1179 ztest_zd_fini(ztest_ds_t *zd)
1181 VERIFY(_mutex_destroy(&zd->zd_dirobj_lock) == 0);
1183 for (int l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1184 ztest_rll_destroy(&zd->zd_object_lock[l]);
1186 for (int l = 0; l < ZTEST_RANGE_LOCKS; l++)
1187 ztest_rll_destroy(&zd->zd_range_lock[l]);
1190 #define TXG_MIGHTWAIT (ztest_random(10) == 0 ? TXG_NOWAIT : TXG_WAIT)
1193 ztest_tx_assign(dmu_tx_t *tx, uint64_t txg_how, const char *tag)
1199 * Attempt to assign tx to some transaction group.
1201 error = dmu_tx_assign(tx, txg_how);
1203 if (error == ERESTART) {
1204 ASSERT(txg_how == TXG_NOWAIT);
1207 ASSERT3U(error, ==, ENOSPC);
1208 ztest_record_enospc(tag);
1213 txg = dmu_tx_get_txg(tx);
1219 ztest_pattern_set(void *buf, uint64_t size, uint64_t value)
1222 uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1229 ztest_pattern_match(void *buf, uint64_t size, uint64_t value)
1232 uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1236 diff |= (value - *ip++);
1242 ztest_bt_generate(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1243 uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg)
1245 bt->bt_magic = BT_MAGIC;
1246 bt->bt_objset = dmu_objset_id(os);
1247 bt->bt_object = object;
1248 bt->bt_offset = offset;
1251 bt->bt_crtxg = crtxg;
1255 ztest_bt_verify(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1256 uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg)
1258 ASSERT(bt->bt_magic == BT_MAGIC);
1259 ASSERT(bt->bt_objset == dmu_objset_id(os));
1260 ASSERT(bt->bt_object == object);
1261 ASSERT(bt->bt_offset == offset);
1262 ASSERT(bt->bt_gen <= gen);
1263 ASSERT(bt->bt_txg <= txg);
1264 ASSERT(bt->bt_crtxg == crtxg);
1267 static ztest_block_tag_t *
1268 ztest_bt_bonus(dmu_buf_t *db)
1270 dmu_object_info_t doi;
1271 ztest_block_tag_t *bt;
1273 dmu_object_info_from_db(db, &doi);
1274 ASSERT3U(doi.doi_bonus_size, <=, db->db_size);
1275 ASSERT3U(doi.doi_bonus_size, >=, sizeof (*bt));
1276 bt = (void *)((char *)db->db_data + doi.doi_bonus_size - sizeof (*bt));
1285 #define lrz_type lr_mode
1286 #define lrz_blocksize lr_uid
1287 #define lrz_ibshift lr_gid
1288 #define lrz_bonustype lr_rdev
1289 #define lrz_bonuslen lr_crtime[1]
1292 ztest_log_create(ztest_ds_t *zd, dmu_tx_t *tx, lr_create_t *lr)
1294 char *name = (void *)(lr + 1); /* name follows lr */
1295 size_t namesize = strlen(name) + 1;
1298 if (zil_replaying(zd->zd_zilog, tx))
1301 itx = zil_itx_create(TX_CREATE, sizeof (*lr) + namesize);
1302 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1303 sizeof (*lr) + namesize - sizeof (lr_t));
1305 zil_itx_assign(zd->zd_zilog, itx, tx);
1309 ztest_log_remove(ztest_ds_t *zd, dmu_tx_t *tx, lr_remove_t *lr, uint64_t object)
1311 char *name = (void *)(lr + 1); /* name follows lr */
1312 size_t namesize = strlen(name) + 1;
1315 if (zil_replaying(zd->zd_zilog, tx))
1318 itx = zil_itx_create(TX_REMOVE, sizeof (*lr) + namesize);
1319 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1320 sizeof (*lr) + namesize - sizeof (lr_t));
1322 itx->itx_oid = object;
1323 zil_itx_assign(zd->zd_zilog, itx, tx);
1327 ztest_log_write(ztest_ds_t *zd, dmu_tx_t *tx, lr_write_t *lr)
1330 itx_wr_state_t write_state = ztest_random(WR_NUM_STATES);
1332 if (zil_replaying(zd->zd_zilog, tx))
1335 if (lr->lr_length > ZIL_MAX_LOG_DATA)
1336 write_state = WR_INDIRECT;
1338 itx = zil_itx_create(TX_WRITE,
1339 sizeof (*lr) + (write_state == WR_COPIED ? lr->lr_length : 0));
1341 if (write_state == WR_COPIED &&
1342 dmu_read(zd->zd_os, lr->lr_foid, lr->lr_offset, lr->lr_length,
1343 ((lr_write_t *)&itx->itx_lr) + 1, DMU_READ_NO_PREFETCH) != 0) {
1344 zil_itx_destroy(itx);
1345 itx = zil_itx_create(TX_WRITE, sizeof (*lr));
1346 write_state = WR_NEED_COPY;
1348 itx->itx_private = zd;
1349 itx->itx_wr_state = write_state;
1350 itx->itx_sync = (ztest_random(8) == 0);
1351 itx->itx_sod += (write_state == WR_NEED_COPY ? lr->lr_length : 0);
1353 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1354 sizeof (*lr) - sizeof (lr_t));
1356 zil_itx_assign(zd->zd_zilog, itx, tx);
1360 ztest_log_truncate(ztest_ds_t *zd, dmu_tx_t *tx, lr_truncate_t *lr)
1364 if (zil_replaying(zd->zd_zilog, tx))
1367 itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr));
1368 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1369 sizeof (*lr) - sizeof (lr_t));
1371 itx->itx_sync = B_FALSE;
1372 zil_itx_assign(zd->zd_zilog, itx, tx);
1376 ztest_log_setattr(ztest_ds_t *zd, dmu_tx_t *tx, lr_setattr_t *lr)
1380 if (zil_replaying(zd->zd_zilog, tx))
1383 itx = zil_itx_create(TX_SETATTR, sizeof (*lr));
1384 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1385 sizeof (*lr) - sizeof (lr_t));
1387 itx->itx_sync = B_FALSE;
1388 zil_itx_assign(zd->zd_zilog, itx, tx);
1395 ztest_replay_create(ztest_ds_t *zd, lr_create_t *lr, boolean_t byteswap)
1397 char *name = (void *)(lr + 1); /* name follows lr */
1398 objset_t *os = zd->zd_os;
1399 ztest_block_tag_t *bbt;
1406 byteswap_uint64_array(lr, sizeof (*lr));
1408 ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1409 ASSERT(name[0] != '\0');
1411 tx = dmu_tx_create(os);
1413 dmu_tx_hold_zap(tx, lr->lr_doid, B_TRUE, name);
1415 if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1416 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, B_TRUE, NULL);
1418 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1421 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1425 ASSERT(dmu_objset_zil(os)->zl_replay == !!lr->lr_foid);
1427 if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1428 if (lr->lr_foid == 0) {
1429 lr->lr_foid = zap_create(os,
1430 lr->lrz_type, lr->lrz_bonustype,
1431 lr->lrz_bonuslen, tx);
1433 error = zap_create_claim(os, lr->lr_foid,
1434 lr->lrz_type, lr->lrz_bonustype,
1435 lr->lrz_bonuslen, tx);
1438 if (lr->lr_foid == 0) {
1439 lr->lr_foid = dmu_object_alloc(os,
1440 lr->lrz_type, 0, lr->lrz_bonustype,
1441 lr->lrz_bonuslen, tx);
1443 error = dmu_object_claim(os, lr->lr_foid,
1444 lr->lrz_type, 0, lr->lrz_bonustype,
1445 lr->lrz_bonuslen, tx);
1450 ASSERT3U(error, ==, EEXIST);
1451 ASSERT(zd->zd_zilog->zl_replay);
1456 ASSERT(lr->lr_foid != 0);
1458 if (lr->lrz_type != DMU_OT_ZAP_OTHER)
1459 VERIFY3U(0, ==, dmu_object_set_blocksize(os, lr->lr_foid,
1460 lr->lrz_blocksize, lr->lrz_ibshift, tx));
1462 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1463 bbt = ztest_bt_bonus(db);
1464 dmu_buf_will_dirty(db, tx);
1465 ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_gen, txg, txg);
1466 dmu_buf_rele(db, FTAG);
1468 VERIFY3U(0, ==, zap_add(os, lr->lr_doid, name, sizeof (uint64_t), 1,
1471 (void) ztest_log_create(zd, tx, lr);
1479 ztest_replay_remove(ztest_ds_t *zd, lr_remove_t *lr, boolean_t byteswap)
1481 char *name = (void *)(lr + 1); /* name follows lr */
1482 objset_t *os = zd->zd_os;
1483 dmu_object_info_t doi;
1485 uint64_t object, txg;
1488 byteswap_uint64_array(lr, sizeof (*lr));
1490 ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1491 ASSERT(name[0] != '\0');
1494 zap_lookup(os, lr->lr_doid, name, sizeof (object), 1, &object));
1495 ASSERT(object != 0);
1497 ztest_object_lock(zd, object, RL_WRITER);
1499 VERIFY3U(0, ==, dmu_object_info(os, object, &doi));
1501 tx = dmu_tx_create(os);
1503 dmu_tx_hold_zap(tx, lr->lr_doid, B_FALSE, name);
1504 dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END);
1506 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1508 ztest_object_unlock(zd, object);
1512 if (doi.doi_type == DMU_OT_ZAP_OTHER) {
1513 VERIFY3U(0, ==, zap_destroy(os, object, tx));
1515 VERIFY3U(0, ==, dmu_object_free(os, object, tx));
1518 VERIFY3U(0, ==, zap_remove(os, lr->lr_doid, name, tx));
1520 (void) ztest_log_remove(zd, tx, lr, object);
1524 ztest_object_unlock(zd, object);
1530 ztest_replay_write(ztest_ds_t *zd, lr_write_t *lr, boolean_t byteswap)
1532 objset_t *os = zd->zd_os;
1533 void *data = lr + 1; /* data follows lr */
1534 uint64_t offset, length;
1535 ztest_block_tag_t *bt = data;
1536 ztest_block_tag_t *bbt;
1537 uint64_t gen, txg, lrtxg, crtxg;
1538 dmu_object_info_t doi;
1541 arc_buf_t *abuf = NULL;
1545 byteswap_uint64_array(lr, sizeof (*lr));
1547 offset = lr->lr_offset;
1548 length = lr->lr_length;
1550 /* If it's a dmu_sync() block, write the whole block */
1551 if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
1552 uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
1553 if (length < blocksize) {
1554 offset -= offset % blocksize;
1559 if (bt->bt_magic == BSWAP_64(BT_MAGIC))
1560 byteswap_uint64_array(bt, sizeof (*bt));
1562 if (bt->bt_magic != BT_MAGIC)
1565 ztest_object_lock(zd, lr->lr_foid, RL_READER);
1566 rl = ztest_range_lock(zd, lr->lr_foid, offset, length, RL_WRITER);
1568 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1570 dmu_object_info_from_db(db, &doi);
1572 bbt = ztest_bt_bonus(db);
1573 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1575 crtxg = bbt->bt_crtxg;
1576 lrtxg = lr->lr_common.lrc_txg;
1578 tx = dmu_tx_create(os);
1580 dmu_tx_hold_write(tx, lr->lr_foid, offset, length);
1582 if (ztest_random(8) == 0 && length == doi.doi_data_block_size &&
1583 P2PHASE(offset, length) == 0)
1584 abuf = dmu_request_arcbuf(db, length);
1586 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1589 dmu_return_arcbuf(abuf);
1590 dmu_buf_rele(db, FTAG);
1591 ztest_range_unlock(rl);
1592 ztest_object_unlock(zd, lr->lr_foid);
1598 * Usually, verify the old data before writing new data --
1599 * but not always, because we also want to verify correct
1600 * behavior when the data was not recently read into cache.
1602 ASSERT(offset % doi.doi_data_block_size == 0);
1603 if (ztest_random(4) != 0) {
1604 int prefetch = ztest_random(2) ?
1605 DMU_READ_PREFETCH : DMU_READ_NO_PREFETCH;
1606 ztest_block_tag_t rbt;
1608 VERIFY(dmu_read(os, lr->lr_foid, offset,
1609 sizeof (rbt), &rbt, prefetch) == 0);
1610 if (rbt.bt_magic == BT_MAGIC) {
1611 ztest_bt_verify(&rbt, os, lr->lr_foid,
1612 offset, gen, txg, crtxg);
1617 * Writes can appear to be newer than the bonus buffer because
1618 * the ztest_get_data() callback does a dmu_read() of the
1619 * open-context data, which may be different than the data
1620 * as it was when the write was generated.
1622 if (zd->zd_zilog->zl_replay) {
1623 ztest_bt_verify(bt, os, lr->lr_foid, offset,
1624 MAX(gen, bt->bt_gen), MAX(txg, lrtxg),
1629 * Set the bt's gen/txg to the bonus buffer's gen/txg
1630 * so that all of the usual ASSERTs will work.
1632 ztest_bt_generate(bt, os, lr->lr_foid, offset, gen, txg, crtxg);
1636 dmu_write(os, lr->lr_foid, offset, length, data, tx);
1638 bcopy(data, abuf->b_data, length);
1639 dmu_assign_arcbuf(db, offset, abuf, tx);
1642 (void) ztest_log_write(zd, tx, lr);
1644 dmu_buf_rele(db, FTAG);
1648 ztest_range_unlock(rl);
1649 ztest_object_unlock(zd, lr->lr_foid);
1655 ztest_replay_truncate(ztest_ds_t *zd, lr_truncate_t *lr, boolean_t byteswap)
1657 objset_t *os = zd->zd_os;
1663 byteswap_uint64_array(lr, sizeof (*lr));
1665 ztest_object_lock(zd, lr->lr_foid, RL_READER);
1666 rl = ztest_range_lock(zd, lr->lr_foid, lr->lr_offset, lr->lr_length,
1669 tx = dmu_tx_create(os);
1671 dmu_tx_hold_free(tx, lr->lr_foid, lr->lr_offset, lr->lr_length);
1673 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1675 ztest_range_unlock(rl);
1676 ztest_object_unlock(zd, lr->lr_foid);
1680 VERIFY(dmu_free_range(os, lr->lr_foid, lr->lr_offset,
1681 lr->lr_length, tx) == 0);
1683 (void) ztest_log_truncate(zd, tx, lr);
1687 ztest_range_unlock(rl);
1688 ztest_object_unlock(zd, lr->lr_foid);
1694 ztest_replay_setattr(ztest_ds_t *zd, lr_setattr_t *lr, boolean_t byteswap)
1696 objset_t *os = zd->zd_os;
1699 ztest_block_tag_t *bbt;
1700 uint64_t txg, lrtxg, crtxg;
1703 byteswap_uint64_array(lr, sizeof (*lr));
1705 ztest_object_lock(zd, lr->lr_foid, RL_WRITER);
1707 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1709 tx = dmu_tx_create(os);
1710 dmu_tx_hold_bonus(tx, lr->lr_foid);
1712 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1714 dmu_buf_rele(db, FTAG);
1715 ztest_object_unlock(zd, lr->lr_foid);
1719 bbt = ztest_bt_bonus(db);
1720 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1721 crtxg = bbt->bt_crtxg;
1722 lrtxg = lr->lr_common.lrc_txg;
1724 if (zd->zd_zilog->zl_replay) {
1725 ASSERT(lr->lr_size != 0);
1726 ASSERT(lr->lr_mode != 0);
1730 * Randomly change the size and increment the generation.
1732 lr->lr_size = (ztest_random(db->db_size / sizeof (*bbt)) + 1) *
1734 lr->lr_mode = bbt->bt_gen + 1;
1739 * Verify that the current bonus buffer is not newer than our txg.
1741 ztest_bt_verify(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode,
1742 MAX(txg, lrtxg), crtxg);
1744 dmu_buf_will_dirty(db, tx);
1746 ASSERT3U(lr->lr_size, >=, sizeof (*bbt));
1747 ASSERT3U(lr->lr_size, <=, db->db_size);
1748 VERIFY0(dmu_set_bonus(db, lr->lr_size, tx));
1749 bbt = ztest_bt_bonus(db);
1751 ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode, txg, crtxg);
1753 dmu_buf_rele(db, FTAG);
1755 (void) ztest_log_setattr(zd, tx, lr);
1759 ztest_object_unlock(zd, lr->lr_foid);
1764 zil_replay_func_t *ztest_replay_vector[TX_MAX_TYPE] = {
1765 NULL, /* 0 no such transaction type */
1766 ztest_replay_create, /* TX_CREATE */
1767 NULL, /* TX_MKDIR */
1768 NULL, /* TX_MKXATTR */
1769 NULL, /* TX_SYMLINK */
1770 ztest_replay_remove, /* TX_REMOVE */
1771 NULL, /* TX_RMDIR */
1773 NULL, /* TX_RENAME */
1774 ztest_replay_write, /* TX_WRITE */
1775 ztest_replay_truncate, /* TX_TRUNCATE */
1776 ztest_replay_setattr, /* TX_SETATTR */
1778 NULL, /* TX_CREATE_ACL */
1779 NULL, /* TX_CREATE_ATTR */
1780 NULL, /* TX_CREATE_ACL_ATTR */
1781 NULL, /* TX_MKDIR_ACL */
1782 NULL, /* TX_MKDIR_ATTR */
1783 NULL, /* TX_MKDIR_ACL_ATTR */
1784 NULL, /* TX_WRITE2 */
1788 * ZIL get_data callbacks
1792 ztest_get_done(zgd_t *zgd, int error)
1794 ztest_ds_t *zd = zgd->zgd_private;
1795 uint64_t object = zgd->zgd_rl->rl_object;
1798 dmu_buf_rele(zgd->zgd_db, zgd);
1800 ztest_range_unlock(zgd->zgd_rl);
1801 ztest_object_unlock(zd, object);
1803 if (error == 0 && zgd->zgd_bp)
1804 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
1806 umem_free(zgd, sizeof (*zgd));
1810 ztest_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
1812 ztest_ds_t *zd = arg;
1813 objset_t *os = zd->zd_os;
1814 uint64_t object = lr->lr_foid;
1815 uint64_t offset = lr->lr_offset;
1816 uint64_t size = lr->lr_length;
1817 blkptr_t *bp = &lr->lr_blkptr;
1818 uint64_t txg = lr->lr_common.lrc_txg;
1820 dmu_object_info_t doi;
1825 ztest_object_lock(zd, object, RL_READER);
1826 error = dmu_bonus_hold(os, object, FTAG, &db);
1828 ztest_object_unlock(zd, object);
1832 crtxg = ztest_bt_bonus(db)->bt_crtxg;
1834 if (crtxg == 0 || crtxg > txg) {
1835 dmu_buf_rele(db, FTAG);
1836 ztest_object_unlock(zd, object);
1840 dmu_object_info_from_db(db, &doi);
1841 dmu_buf_rele(db, FTAG);
1844 zgd = umem_zalloc(sizeof (*zgd), UMEM_NOFAIL);
1845 zgd->zgd_zilog = zd->zd_zilog;
1846 zgd->zgd_private = zd;
1848 if (buf != NULL) { /* immediate write */
1849 zgd->zgd_rl = ztest_range_lock(zd, object, offset, size,
1852 error = dmu_read(os, object, offset, size, buf,
1853 DMU_READ_NO_PREFETCH);
1856 size = doi.doi_data_block_size;
1858 offset = P2ALIGN(offset, size);
1860 ASSERT(offset < size);
1864 zgd->zgd_rl = ztest_range_lock(zd, object, offset, size,
1867 error = dmu_buf_hold(os, object, offset, zgd, &db,
1868 DMU_READ_NO_PREFETCH);
1871 blkptr_t *obp = dmu_buf_get_blkptr(db);
1873 ASSERT(BP_IS_HOLE(bp));
1880 ASSERT(db->db_offset == offset);
1881 ASSERT(db->db_size == size);
1883 error = dmu_sync(zio, lr->lr_common.lrc_txg,
1884 ztest_get_done, zgd);
1891 ztest_get_done(zgd, error);
1897 ztest_lr_alloc(size_t lrsize, char *name)
1900 size_t namesize = name ? strlen(name) + 1 : 0;
1902 lr = umem_zalloc(lrsize + namesize, UMEM_NOFAIL);
1905 bcopy(name, lr + lrsize, namesize);
1911 ztest_lr_free(void *lr, size_t lrsize, char *name)
1913 size_t namesize = name ? strlen(name) + 1 : 0;
1915 umem_free(lr, lrsize + namesize);
1919 * Lookup a bunch of objects. Returns the number of objects not found.
1922 ztest_lookup(ztest_ds_t *zd, ztest_od_t *od, int count)
1927 ASSERT(_mutex_held(&zd->zd_dirobj_lock));
1929 for (int i = 0; i < count; i++, od++) {
1931 error = zap_lookup(zd->zd_os, od->od_dir, od->od_name,
1932 sizeof (uint64_t), 1, &od->od_object);
1934 ASSERT(error == ENOENT);
1935 ASSERT(od->od_object == 0);
1939 ztest_block_tag_t *bbt;
1940 dmu_object_info_t doi;
1942 ASSERT(od->od_object != 0);
1943 ASSERT(missing == 0); /* there should be no gaps */
1945 ztest_object_lock(zd, od->od_object, RL_READER);
1946 VERIFY3U(0, ==, dmu_bonus_hold(zd->zd_os,
1947 od->od_object, FTAG, &db));
1948 dmu_object_info_from_db(db, &doi);
1949 bbt = ztest_bt_bonus(db);
1950 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1951 od->od_type = doi.doi_type;
1952 od->od_blocksize = doi.doi_data_block_size;
1953 od->od_gen = bbt->bt_gen;
1954 dmu_buf_rele(db, FTAG);
1955 ztest_object_unlock(zd, od->od_object);
1963 ztest_create(ztest_ds_t *zd, ztest_od_t *od, int count)
1967 ASSERT(_mutex_held(&zd->zd_dirobj_lock));
1969 for (int i = 0; i < count; i++, od++) {
1976 lr_create_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
1978 lr->lr_doid = od->od_dir;
1979 lr->lr_foid = 0; /* 0 to allocate, > 0 to claim */
1980 lr->lrz_type = od->od_crtype;
1981 lr->lrz_blocksize = od->od_crblocksize;
1982 lr->lrz_ibshift = ztest_random_ibshift();
1983 lr->lrz_bonustype = DMU_OT_UINT64_OTHER;
1984 lr->lrz_bonuslen = dmu_bonus_max();
1985 lr->lr_gen = od->od_crgen;
1986 lr->lr_crtime[0] = time(NULL);
1988 if (ztest_replay_create(zd, lr, B_FALSE) != 0) {
1989 ASSERT(missing == 0);
1993 od->od_object = lr->lr_foid;
1994 od->od_type = od->od_crtype;
1995 od->od_blocksize = od->od_crblocksize;
1996 od->od_gen = od->od_crgen;
1997 ASSERT(od->od_object != 0);
2000 ztest_lr_free(lr, sizeof (*lr), od->od_name);
2007 ztest_remove(ztest_ds_t *zd, ztest_od_t *od, int count)
2012 ASSERT(_mutex_held(&zd->zd_dirobj_lock));
2016 for (int i = count - 1; i >= 0; i--, od--) {
2023 * No object was found.
2025 if (od->od_object == 0)
2028 lr_remove_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
2030 lr->lr_doid = od->od_dir;
2032 if ((error = ztest_replay_remove(zd, lr, B_FALSE)) != 0) {
2033 ASSERT3U(error, ==, ENOSPC);
2038 ztest_lr_free(lr, sizeof (*lr), od->od_name);
2045 ztest_write(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size,
2051 lr = ztest_lr_alloc(sizeof (*lr) + size, NULL);
2053 lr->lr_foid = object;
2054 lr->lr_offset = offset;
2055 lr->lr_length = size;
2057 BP_ZERO(&lr->lr_blkptr);
2059 bcopy(data, lr + 1, size);
2061 error = ztest_replay_write(zd, lr, B_FALSE);
2063 ztest_lr_free(lr, sizeof (*lr) + size, NULL);
2069 ztest_truncate(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2074 lr = ztest_lr_alloc(sizeof (*lr), NULL);
2076 lr->lr_foid = object;
2077 lr->lr_offset = offset;
2078 lr->lr_length = size;
2080 error = ztest_replay_truncate(zd, lr, B_FALSE);
2082 ztest_lr_free(lr, sizeof (*lr), NULL);
2088 ztest_setattr(ztest_ds_t *zd, uint64_t object)
2093 lr = ztest_lr_alloc(sizeof (*lr), NULL);
2095 lr->lr_foid = object;
2099 error = ztest_replay_setattr(zd, lr, B_FALSE);
2101 ztest_lr_free(lr, sizeof (*lr), NULL);
2107 ztest_prealloc(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2109 objset_t *os = zd->zd_os;
2114 txg_wait_synced(dmu_objset_pool(os), 0);
2116 ztest_object_lock(zd, object, RL_READER);
2117 rl = ztest_range_lock(zd, object, offset, size, RL_WRITER);
2119 tx = dmu_tx_create(os);
2121 dmu_tx_hold_write(tx, object, offset, size);
2123 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
2126 dmu_prealloc(os, object, offset, size, tx);
2128 txg_wait_synced(dmu_objset_pool(os), txg);
2130 (void) dmu_free_long_range(os, object, offset, size);
2133 ztest_range_unlock(rl);
2134 ztest_object_unlock(zd, object);
2138 ztest_io(ztest_ds_t *zd, uint64_t object, uint64_t offset)
2141 ztest_block_tag_t wbt;
2142 dmu_object_info_t doi;
2143 enum ztest_io_type io_type;
2147 VERIFY(dmu_object_info(zd->zd_os, object, &doi) == 0);
2148 blocksize = doi.doi_data_block_size;
2149 data = umem_alloc(blocksize, UMEM_NOFAIL);
2152 * Pick an i/o type at random, biased toward writing block tags.
2154 io_type = ztest_random(ZTEST_IO_TYPES);
2155 if (ztest_random(2) == 0)
2156 io_type = ZTEST_IO_WRITE_TAG;
2158 (void) rw_rdlock(&zd->zd_zilog_lock);
2162 case ZTEST_IO_WRITE_TAG:
2163 ztest_bt_generate(&wbt, zd->zd_os, object, offset, 0, 0, 0);
2164 (void) ztest_write(zd, object, offset, sizeof (wbt), &wbt);
2167 case ZTEST_IO_WRITE_PATTERN:
2168 (void) memset(data, 'a' + (object + offset) % 5, blocksize);
2169 if (ztest_random(2) == 0) {
2171 * Induce fletcher2 collisions to ensure that
2172 * zio_ddt_collision() detects and resolves them
2173 * when using fletcher2-verify for deduplication.
2175 ((uint64_t *)data)[0] ^= 1ULL << 63;
2176 ((uint64_t *)data)[4] ^= 1ULL << 63;
2178 (void) ztest_write(zd, object, offset, blocksize, data);
2181 case ZTEST_IO_WRITE_ZEROES:
2182 bzero(data, blocksize);
2183 (void) ztest_write(zd, object, offset, blocksize, data);
2186 case ZTEST_IO_TRUNCATE:
2187 (void) ztest_truncate(zd, object, offset, blocksize);
2190 case ZTEST_IO_SETATTR:
2191 (void) ztest_setattr(zd, object);
2194 case ZTEST_IO_REWRITE:
2195 (void) rw_rdlock(&ztest_name_lock);
2196 err = ztest_dsl_prop_set_uint64(zd->zd_name,
2197 ZFS_PROP_CHECKSUM, spa_dedup_checksum(ztest_spa),
2199 VERIFY(err == 0 || err == ENOSPC);
2200 err = ztest_dsl_prop_set_uint64(zd->zd_name,
2201 ZFS_PROP_COMPRESSION,
2202 ztest_random_dsl_prop(ZFS_PROP_COMPRESSION),
2204 VERIFY(err == 0 || err == ENOSPC);
2205 (void) rw_unlock(&ztest_name_lock);
2207 VERIFY0(dmu_read(zd->zd_os, object, offset, blocksize, data,
2208 DMU_READ_NO_PREFETCH));
2210 (void) ztest_write(zd, object, offset, blocksize, data);
2214 (void) rw_unlock(&zd->zd_zilog_lock);
2216 umem_free(data, blocksize);
2220 * Initialize an object description template.
2223 ztest_od_init(ztest_od_t *od, uint64_t id, char *tag, uint64_t index,
2224 dmu_object_type_t type, uint64_t blocksize, uint64_t gen)
2226 od->od_dir = ZTEST_DIROBJ;
2229 od->od_crtype = type;
2230 od->od_crblocksize = blocksize ? blocksize : ztest_random_blocksize();
2233 od->od_type = DMU_OT_NONE;
2234 od->od_blocksize = 0;
2237 (void) snprintf(od->od_name, sizeof (od->od_name), "%s(%lld)[%llu]",
2238 tag, (int64_t)id, index);
2242 * Lookup or create the objects for a test using the od template.
2243 * If the objects do not all exist, or if 'remove' is specified,
2244 * remove any existing objects and create new ones. Otherwise,
2245 * use the existing objects.
2248 ztest_object_init(ztest_ds_t *zd, ztest_od_t *od, size_t size, boolean_t remove)
2250 int count = size / sizeof (*od);
2253 VERIFY(mutex_lock(&zd->zd_dirobj_lock) == 0);
2254 if ((ztest_lookup(zd, od, count) != 0 || remove) &&
2255 (ztest_remove(zd, od, count) != 0 ||
2256 ztest_create(zd, od, count) != 0))
2259 VERIFY(mutex_unlock(&zd->zd_dirobj_lock) == 0);
2266 ztest_zil_commit(ztest_ds_t *zd, uint64_t id)
2268 zilog_t *zilog = zd->zd_zilog;
2270 (void) rw_rdlock(&zd->zd_zilog_lock);
2272 zil_commit(zilog, ztest_random(ZTEST_OBJECTS));
2275 * Remember the committed values in zd, which is in parent/child
2276 * shared memory. If we die, the next iteration of ztest_run()
2277 * will verify that the log really does contain this record.
2279 mutex_enter(&zilog->zl_lock);
2280 ASSERT(zd->zd_shared != NULL);
2281 ASSERT3U(zd->zd_shared->zd_seq, <=, zilog->zl_commit_lr_seq);
2282 zd->zd_shared->zd_seq = zilog->zl_commit_lr_seq;
2283 mutex_exit(&zilog->zl_lock);
2285 (void) rw_unlock(&zd->zd_zilog_lock);
2289 * This function is designed to simulate the operations that occur during a
2290 * mount/unmount operation. We hold the dataset across these operations in an
2291 * attempt to expose any implicit assumptions about ZIL management.
2295 ztest_zil_remount(ztest_ds_t *zd, uint64_t id)
2297 objset_t *os = zd->zd_os;
2300 * We grab the zd_dirobj_lock to ensure that no other thread is
2301 * updating the zil (i.e. adding in-memory log records) and the
2302 * zd_zilog_lock to block any I/O.
2304 VERIFY0(mutex_lock(&zd->zd_dirobj_lock));
2305 (void) rw_wrlock(&zd->zd_zilog_lock);
2307 /* zfsvfs_teardown() */
2308 zil_close(zd->zd_zilog);
2310 /* zfsvfs_setup() */
2311 VERIFY(zil_open(os, ztest_get_data) == zd->zd_zilog);
2312 zil_replay(os, zd, ztest_replay_vector);
2314 (void) rw_unlock(&zd->zd_zilog_lock);
2315 VERIFY(mutex_unlock(&zd->zd_dirobj_lock) == 0);
2319 * Verify that we can't destroy an active pool, create an existing pool,
2320 * or create a pool with a bad vdev spec.
2324 ztest_spa_create_destroy(ztest_ds_t *zd, uint64_t id)
2326 ztest_shared_opts_t *zo = &ztest_opts;
2331 * Attempt to create using a bad file.
2333 nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1);
2334 VERIFY3U(ENOENT, ==,
2335 spa_create("ztest_bad_file", nvroot, NULL, NULL, NULL));
2336 nvlist_free(nvroot);
2339 * Attempt to create using a bad mirror.
2341 nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 2, 1);
2342 VERIFY3U(ENOENT, ==,
2343 spa_create("ztest_bad_mirror", nvroot, NULL, NULL, NULL));
2344 nvlist_free(nvroot);
2347 * Attempt to create an existing pool. It shouldn't matter
2348 * what's in the nvroot; we should fail with EEXIST.
2350 (void) rw_rdlock(&ztest_name_lock);
2351 nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1);
2352 VERIFY3U(EEXIST, ==, spa_create(zo->zo_pool, nvroot, NULL, NULL, NULL));
2353 nvlist_free(nvroot);
2354 VERIFY3U(0, ==, spa_open(zo->zo_pool, &spa, FTAG));
2355 VERIFY3U(EBUSY, ==, spa_destroy(zo->zo_pool));
2356 spa_close(spa, FTAG);
2358 (void) rw_unlock(&ztest_name_lock);
2363 ztest_spa_upgrade(ztest_ds_t *zd, uint64_t id)
2366 uint64_t initial_version = SPA_VERSION_INITIAL;
2367 uint64_t version, newversion;
2368 nvlist_t *nvroot, *props;
2371 VERIFY0(mutex_lock(&ztest_vdev_lock));
2372 name = kmem_asprintf("%s_upgrade", ztest_opts.zo_pool);
2375 * Clean up from previous runs.
2377 (void) spa_destroy(name);
2379 nvroot = make_vdev_root(NULL, NULL, name, ztest_opts.zo_vdev_size, 0,
2380 0, ztest_opts.zo_raidz, ztest_opts.zo_mirrors, 1);
2383 * If we're configuring a RAIDZ device then make sure that the
2384 * the initial version is capable of supporting that feature.
2386 switch (ztest_opts.zo_raidz_parity) {
2389 initial_version = SPA_VERSION_INITIAL;
2392 initial_version = SPA_VERSION_RAIDZ2;
2395 initial_version = SPA_VERSION_RAIDZ3;
2400 * Create a pool with a spa version that can be upgraded. Pick
2401 * a value between initial_version and SPA_VERSION_BEFORE_FEATURES.
2404 version = ztest_random_spa_version(initial_version);
2405 } while (version > SPA_VERSION_BEFORE_FEATURES);
2407 props = fnvlist_alloc();
2408 fnvlist_add_uint64(props,
2409 zpool_prop_to_name(ZPOOL_PROP_VERSION), version);
2410 VERIFY0(spa_create(name, nvroot, props, NULL, NULL));
2411 fnvlist_free(nvroot);
2412 fnvlist_free(props);
2414 VERIFY0(spa_open(name, &spa, FTAG));
2415 VERIFY3U(spa_version(spa), ==, version);
2416 newversion = ztest_random_spa_version(version + 1);
2418 if (ztest_opts.zo_verbose >= 4) {
2419 (void) printf("upgrading spa version from %llu to %llu\n",
2420 (u_longlong_t)version, (u_longlong_t)newversion);
2423 spa_upgrade(spa, newversion);
2424 VERIFY3U(spa_version(spa), >, version);
2425 VERIFY3U(spa_version(spa), ==, fnvlist_lookup_uint64(spa->spa_config,
2426 zpool_prop_to_name(ZPOOL_PROP_VERSION)));
2427 spa_close(spa, FTAG);
2430 VERIFY0(mutex_unlock(&ztest_vdev_lock));
2434 vdev_lookup_by_path(vdev_t *vd, const char *path)
2438 if (vd->vdev_path != NULL && strcmp(path, vd->vdev_path) == 0)
2441 for (int c = 0; c < vd->vdev_children; c++)
2442 if ((mvd = vdev_lookup_by_path(vd->vdev_child[c], path)) !=
2450 * Find the first available hole which can be used as a top-level.
2453 find_vdev_hole(spa_t *spa)
2455 vdev_t *rvd = spa->spa_root_vdev;
2458 ASSERT(spa_config_held(spa, SCL_VDEV, RW_READER) == SCL_VDEV);
2460 for (c = 0; c < rvd->vdev_children; c++) {
2461 vdev_t *cvd = rvd->vdev_child[c];
2463 if (cvd->vdev_ishole)
2470 * Verify that vdev_add() works as expected.
2474 ztest_vdev_add_remove(ztest_ds_t *zd, uint64_t id)
2476 ztest_shared_t *zs = ztest_shared;
2477 spa_t *spa = ztest_spa;
2483 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2485 MAX(zs->zs_mirrors + zs->zs_splits, 1) * ztest_opts.zo_raidz;
2487 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2489 ztest_shared->zs_vdev_next_leaf = find_vdev_hole(spa) * leaves;
2492 * If we have slogs then remove them 1/4 of the time.
2494 if (spa_has_slogs(spa) && ztest_random(4) == 0) {
2496 * Grab the guid from the head of the log class rotor.
2498 guid = spa_log_class(spa)->mc_rotor->mg_vd->vdev_guid;
2500 spa_config_exit(spa, SCL_VDEV, FTAG);
2503 * We have to grab the zs_name_lock as writer to
2504 * prevent a race between removing a slog (dmu_objset_find)
2505 * and destroying a dataset. Removing the slog will
2506 * grab a reference on the dataset which may cause
2507 * dmu_objset_destroy() to fail with EBUSY thus
2508 * leaving the dataset in an inconsistent state.
2510 VERIFY(rw_wrlock(&ztest_name_lock) == 0);
2511 error = spa_vdev_remove(spa, guid, B_FALSE);
2512 VERIFY(rw_unlock(&ztest_name_lock) == 0);
2514 if (error && error != EEXIST)
2515 fatal(0, "spa_vdev_remove() = %d", error);
2517 spa_config_exit(spa, SCL_VDEV, FTAG);
2520 * Make 1/4 of the devices be log devices.
2522 nvroot = make_vdev_root(NULL, NULL, NULL,
2523 ztest_opts.zo_vdev_size, 0,
2524 ztest_random(4) == 0, ztest_opts.zo_raidz,
2527 error = spa_vdev_add(spa, nvroot);
2528 nvlist_free(nvroot);
2530 if (error == ENOSPC)
2531 ztest_record_enospc("spa_vdev_add");
2532 else if (error != 0)
2533 fatal(0, "spa_vdev_add() = %d", error);
2536 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2540 * Verify that adding/removing aux devices (l2arc, hot spare) works as expected.
2544 ztest_vdev_aux_add_remove(ztest_ds_t *zd, uint64_t id)
2546 ztest_shared_t *zs = ztest_shared;
2547 spa_t *spa = ztest_spa;
2548 vdev_t *rvd = spa->spa_root_vdev;
2549 spa_aux_vdev_t *sav;
2554 if (ztest_random(2) == 0) {
2555 sav = &spa->spa_spares;
2556 aux = ZPOOL_CONFIG_SPARES;
2558 sav = &spa->spa_l2cache;
2559 aux = ZPOOL_CONFIG_L2CACHE;
2562 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2564 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2566 if (sav->sav_count != 0 && ztest_random(4) == 0) {
2568 * Pick a random device to remove.
2570 guid = sav->sav_vdevs[ztest_random(sav->sav_count)]->vdev_guid;
2573 * Find an unused device we can add.
2575 zs->zs_vdev_aux = 0;
2577 char path[MAXPATHLEN];
2579 (void) snprintf(path, sizeof (path), ztest_aux_template,
2580 ztest_opts.zo_dir, ztest_opts.zo_pool, aux,
2582 for (c = 0; c < sav->sav_count; c++)
2583 if (strcmp(sav->sav_vdevs[c]->vdev_path,
2586 if (c == sav->sav_count &&
2587 vdev_lookup_by_path(rvd, path) == NULL)
2593 spa_config_exit(spa, SCL_VDEV, FTAG);
2599 nvlist_t *nvroot = make_vdev_root(NULL, aux, NULL,
2600 (ztest_opts.zo_vdev_size * 5) / 4, 0, 0, 0, 0, 1);
2601 error = spa_vdev_add(spa, nvroot);
2603 fatal(0, "spa_vdev_add(%p) = %d", nvroot, error);
2604 nvlist_free(nvroot);
2607 * Remove an existing device. Sometimes, dirty its
2608 * vdev state first to make sure we handle removal
2609 * of devices that have pending state changes.
2611 if (ztest_random(2) == 0)
2612 (void) vdev_online(spa, guid, 0, NULL);
2614 error = spa_vdev_remove(spa, guid, B_FALSE);
2615 if (error != 0 && error != EBUSY)
2616 fatal(0, "spa_vdev_remove(%llu) = %d", guid, error);
2619 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2623 * split a pool if it has mirror tlvdevs
2627 ztest_split_pool(ztest_ds_t *zd, uint64_t id)
2629 ztest_shared_t *zs = ztest_shared;
2630 spa_t *spa = ztest_spa;
2631 vdev_t *rvd = spa->spa_root_vdev;
2632 nvlist_t *tree, **child, *config, *split, **schild;
2633 uint_t c, children, schildren = 0, lastlogid = 0;
2636 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2638 /* ensure we have a useable config; mirrors of raidz aren't supported */
2639 if (zs->zs_mirrors < 3 || ztest_opts.zo_raidz > 1) {
2640 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2644 /* clean up the old pool, if any */
2645 (void) spa_destroy("splitp");
2647 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2649 /* generate a config from the existing config */
2650 mutex_enter(&spa->spa_props_lock);
2651 VERIFY(nvlist_lookup_nvlist(spa->spa_config, ZPOOL_CONFIG_VDEV_TREE,
2653 mutex_exit(&spa->spa_props_lock);
2655 VERIFY(nvlist_lookup_nvlist_array(tree, ZPOOL_CONFIG_CHILDREN, &child,
2658 schild = malloc(rvd->vdev_children * sizeof (nvlist_t *));
2659 for (c = 0; c < children; c++) {
2660 vdev_t *tvd = rvd->vdev_child[c];
2664 if (tvd->vdev_islog || tvd->vdev_ops == &vdev_hole_ops) {
2665 VERIFY(nvlist_alloc(&schild[schildren], NV_UNIQUE_NAME,
2667 VERIFY(nvlist_add_string(schild[schildren],
2668 ZPOOL_CONFIG_TYPE, VDEV_TYPE_HOLE) == 0);
2669 VERIFY(nvlist_add_uint64(schild[schildren],
2670 ZPOOL_CONFIG_IS_HOLE, 1) == 0);
2672 lastlogid = schildren;
2677 VERIFY(nvlist_lookup_nvlist_array(child[c],
2678 ZPOOL_CONFIG_CHILDREN, &mchild, &mchildren) == 0);
2679 VERIFY(nvlist_dup(mchild[0], &schild[schildren++], 0) == 0);
2682 /* OK, create a config that can be used to split */
2683 VERIFY(nvlist_alloc(&split, NV_UNIQUE_NAME, 0) == 0);
2684 VERIFY(nvlist_add_string(split, ZPOOL_CONFIG_TYPE,
2685 VDEV_TYPE_ROOT) == 0);
2686 VERIFY(nvlist_add_nvlist_array(split, ZPOOL_CONFIG_CHILDREN, schild,
2687 lastlogid != 0 ? lastlogid : schildren) == 0);
2689 VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, 0) == 0);
2690 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, split) == 0);
2692 for (c = 0; c < schildren; c++)
2693 nvlist_free(schild[c]);
2697 spa_config_exit(spa, SCL_VDEV, FTAG);
2699 (void) rw_wrlock(&ztest_name_lock);
2700 error = spa_vdev_split_mirror(spa, "splitp", config, NULL, B_FALSE);
2701 (void) rw_unlock(&ztest_name_lock);
2703 nvlist_free(config);
2706 (void) printf("successful split - results:\n");
2707 mutex_enter(&spa_namespace_lock);
2708 show_pool_stats(spa);
2709 show_pool_stats(spa_lookup("splitp"));
2710 mutex_exit(&spa_namespace_lock);
2714 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2719 * Verify that we can attach and detach devices.
2723 ztest_vdev_attach_detach(ztest_ds_t *zd, uint64_t id)
2725 ztest_shared_t *zs = ztest_shared;
2726 spa_t *spa = ztest_spa;
2727 spa_aux_vdev_t *sav = &spa->spa_spares;
2728 vdev_t *rvd = spa->spa_root_vdev;
2729 vdev_t *oldvd, *newvd, *pvd;
2733 uint64_t ashift = ztest_get_ashift();
2734 uint64_t oldguid, pguid;
2735 size_t oldsize, newsize;
2736 char oldpath[MAXPATHLEN], newpath[MAXPATHLEN];
2738 int oldvd_has_siblings = B_FALSE;
2739 int newvd_is_spare = B_FALSE;
2741 int error, expected_error;
2743 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2744 leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz;
2746 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2749 * Decide whether to do an attach or a replace.
2751 replacing = ztest_random(2);
2754 * Pick a random top-level vdev.
2756 top = ztest_random_vdev_top(spa, B_TRUE);
2759 * Pick a random leaf within it.
2761 leaf = ztest_random(leaves);
2766 oldvd = rvd->vdev_child[top];
2767 if (zs->zs_mirrors >= 1) {
2768 ASSERT(oldvd->vdev_ops == &vdev_mirror_ops);
2769 ASSERT(oldvd->vdev_children >= zs->zs_mirrors);
2770 oldvd = oldvd->vdev_child[leaf / ztest_opts.zo_raidz];
2772 if (ztest_opts.zo_raidz > 1) {
2773 ASSERT(oldvd->vdev_ops == &vdev_raidz_ops);
2774 ASSERT(oldvd->vdev_children == ztest_opts.zo_raidz);
2775 oldvd = oldvd->vdev_child[leaf % ztest_opts.zo_raidz];
2779 * If we're already doing an attach or replace, oldvd may be a
2780 * mirror vdev -- in which case, pick a random child.
2782 while (oldvd->vdev_children != 0) {
2783 oldvd_has_siblings = B_TRUE;
2784 ASSERT(oldvd->vdev_children >= 2);
2785 oldvd = oldvd->vdev_child[ztest_random(oldvd->vdev_children)];
2788 oldguid = oldvd->vdev_guid;
2789 oldsize = vdev_get_min_asize(oldvd);
2790 oldvd_is_log = oldvd->vdev_top->vdev_islog;
2791 (void) strcpy(oldpath, oldvd->vdev_path);
2792 pvd = oldvd->vdev_parent;
2793 pguid = pvd->vdev_guid;
2796 * If oldvd has siblings, then half of the time, detach it.
2798 if (oldvd_has_siblings && ztest_random(2) == 0) {
2799 spa_config_exit(spa, SCL_VDEV, FTAG);
2800 error = spa_vdev_detach(spa, oldguid, pguid, B_FALSE);
2801 if (error != 0 && error != ENODEV && error != EBUSY &&
2803 fatal(0, "detach (%s) returned %d", oldpath, error);
2804 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2809 * For the new vdev, choose with equal probability between the two
2810 * standard paths (ending in either 'a' or 'b') or a random hot spare.
2812 if (sav->sav_count != 0 && ztest_random(3) == 0) {
2813 newvd = sav->sav_vdevs[ztest_random(sav->sav_count)];
2814 newvd_is_spare = B_TRUE;
2815 (void) strcpy(newpath, newvd->vdev_path);
2817 (void) snprintf(newpath, sizeof (newpath), ztest_dev_template,
2818 ztest_opts.zo_dir, ztest_opts.zo_pool,
2819 top * leaves + leaf);
2820 if (ztest_random(2) == 0)
2821 newpath[strlen(newpath) - 1] = 'b';
2822 newvd = vdev_lookup_by_path(rvd, newpath);
2826 newsize = vdev_get_min_asize(newvd);
2829 * Make newsize a little bigger or smaller than oldsize.
2830 * If it's smaller, the attach should fail.
2831 * If it's larger, and we're doing a replace,
2832 * we should get dynamic LUN growth when we're done.
2834 newsize = 10 * oldsize / (9 + ztest_random(3));
2838 * If pvd is not a mirror or root, the attach should fail with ENOTSUP,
2839 * unless it's a replace; in that case any non-replacing parent is OK.
2841 * If newvd is already part of the pool, it should fail with EBUSY.
2843 * If newvd is too small, it should fail with EOVERFLOW.
2845 if (pvd->vdev_ops != &vdev_mirror_ops &&
2846 pvd->vdev_ops != &vdev_root_ops && (!replacing ||
2847 pvd->vdev_ops == &vdev_replacing_ops ||
2848 pvd->vdev_ops == &vdev_spare_ops))
2849 expected_error = ENOTSUP;
2850 else if (newvd_is_spare && (!replacing || oldvd_is_log))
2851 expected_error = ENOTSUP;
2852 else if (newvd == oldvd)
2853 expected_error = replacing ? 0 : EBUSY;
2854 else if (vdev_lookup_by_path(rvd, newpath) != NULL)
2855 expected_error = EBUSY;
2856 else if (newsize < oldsize)
2857 expected_error = EOVERFLOW;
2858 else if (ashift > oldvd->vdev_top->vdev_ashift)
2859 expected_error = EDOM;
2863 spa_config_exit(spa, SCL_VDEV, FTAG);
2866 * Build the nvlist describing newpath.
2868 root = make_vdev_root(newpath, NULL, NULL, newvd == NULL ? newsize : 0,
2869 ashift, 0, 0, 0, 1);
2871 error = spa_vdev_attach(spa, oldguid, root, replacing);
2876 * If our parent was the replacing vdev, but the replace completed,
2877 * then instead of failing with ENOTSUP we may either succeed,
2878 * fail with ENODEV, or fail with EOVERFLOW.
2880 if (expected_error == ENOTSUP &&
2881 (error == 0 || error == ENODEV || error == EOVERFLOW))
2882 expected_error = error;
2885 * If someone grew the LUN, the replacement may be too small.
2887 if (error == EOVERFLOW || error == EBUSY)
2888 expected_error = error;
2890 /* XXX workaround 6690467 */
2891 if (error != expected_error && expected_error != EBUSY) {
2892 fatal(0, "attach (%s %llu, %s %llu, %d) "
2893 "returned %d, expected %d",
2894 oldpath, (longlong_t)oldsize, newpath,
2895 (longlong_t)newsize, replacing, error, expected_error);
2898 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2902 * Callback function which expands the physical size of the vdev.
2905 grow_vdev(vdev_t *vd, void *arg)
2907 spa_t *spa = vd->vdev_spa;
2908 size_t *newsize = arg;
2912 ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
2913 ASSERT(vd->vdev_ops->vdev_op_leaf);
2915 if ((fd = open(vd->vdev_path, O_RDWR)) == -1)
2918 fsize = lseek(fd, 0, SEEK_END);
2919 (void) ftruncate(fd, *newsize);
2921 if (ztest_opts.zo_verbose >= 6) {
2922 (void) printf("%s grew from %lu to %lu bytes\n",
2923 vd->vdev_path, (ulong_t)fsize, (ulong_t)*newsize);
2930 * Callback function which expands a given vdev by calling vdev_online().
2934 online_vdev(vdev_t *vd, void *arg)
2936 spa_t *spa = vd->vdev_spa;
2937 vdev_t *tvd = vd->vdev_top;
2938 uint64_t guid = vd->vdev_guid;
2939 uint64_t generation = spa->spa_config_generation + 1;
2940 vdev_state_t newstate = VDEV_STATE_UNKNOWN;
2943 ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
2944 ASSERT(vd->vdev_ops->vdev_op_leaf);
2946 /* Calling vdev_online will initialize the new metaslabs */
2947 spa_config_exit(spa, SCL_STATE, spa);
2948 error = vdev_online(spa, guid, ZFS_ONLINE_EXPAND, &newstate);
2949 spa_config_enter(spa, SCL_STATE, spa, RW_READER);
2952 * If vdev_online returned an error or the underlying vdev_open
2953 * failed then we abort the expand. The only way to know that
2954 * vdev_open fails is by checking the returned newstate.
2956 if (error || newstate != VDEV_STATE_HEALTHY) {
2957 if (ztest_opts.zo_verbose >= 5) {
2958 (void) printf("Unable to expand vdev, state %llu, "
2959 "error %d\n", (u_longlong_t)newstate, error);
2963 ASSERT3U(newstate, ==, VDEV_STATE_HEALTHY);
2966 * Since we dropped the lock we need to ensure that we're
2967 * still talking to the original vdev. It's possible this
2968 * vdev may have been detached/replaced while we were
2969 * trying to online it.
2971 if (generation != spa->spa_config_generation) {
2972 if (ztest_opts.zo_verbose >= 5) {
2973 (void) printf("vdev configuration has changed, "
2974 "guid %llu, state %llu, expected gen %llu, "
2977 (u_longlong_t)tvd->vdev_state,
2978 (u_longlong_t)generation,
2979 (u_longlong_t)spa->spa_config_generation);
2987 * Traverse the vdev tree calling the supplied function.
2988 * We continue to walk the tree until we either have walked all
2989 * children or we receive a non-NULL return from the callback.
2990 * If a NULL callback is passed, then we just return back the first
2991 * leaf vdev we encounter.
2994 vdev_walk_tree(vdev_t *vd, vdev_t *(*func)(vdev_t *, void *), void *arg)
2996 if (vd->vdev_ops->vdev_op_leaf) {
3000 return (func(vd, arg));
3003 for (uint_t c = 0; c < vd->vdev_children; c++) {
3004 vdev_t *cvd = vd->vdev_child[c];
3005 if ((cvd = vdev_walk_tree(cvd, func, arg)) != NULL)
3012 * Verify that dynamic LUN growth works as expected.
3016 ztest_vdev_LUN_growth(ztest_ds_t *zd, uint64_t id)
3018 spa_t *spa = ztest_spa;
3020 metaslab_class_t *mc;
3021 metaslab_group_t *mg;
3022 size_t psize, newsize;
3024 uint64_t old_class_space, new_class_space, old_ms_count, new_ms_count;
3026 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
3027 spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3029 top = ztest_random_vdev_top(spa, B_TRUE);
3031 tvd = spa->spa_root_vdev->vdev_child[top];
3034 old_ms_count = tvd->vdev_ms_count;
3035 old_class_space = metaslab_class_get_space(mc);
3038 * Determine the size of the first leaf vdev associated with
3039 * our top-level device.
3041 vd = vdev_walk_tree(tvd, NULL, NULL);
3042 ASSERT3P(vd, !=, NULL);
3043 ASSERT(vd->vdev_ops->vdev_op_leaf);
3045 psize = vd->vdev_psize;
3048 * We only try to expand the vdev if it's healthy, less than 4x its
3049 * original size, and it has a valid psize.
3051 if (tvd->vdev_state != VDEV_STATE_HEALTHY ||
3052 psize == 0 || psize >= 4 * ztest_opts.zo_vdev_size) {
3053 spa_config_exit(spa, SCL_STATE, spa);
3054 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3058 newsize = psize + psize / 8;
3059 ASSERT3U(newsize, >, psize);
3061 if (ztest_opts.zo_verbose >= 6) {
3062 (void) printf("Expanding LUN %s from %lu to %lu\n",
3063 vd->vdev_path, (ulong_t)psize, (ulong_t)newsize);
3067 * Growing the vdev is a two step process:
3068 * 1). expand the physical size (i.e. relabel)
3069 * 2). online the vdev to create the new metaslabs
3071 if (vdev_walk_tree(tvd, grow_vdev, &newsize) != NULL ||
3072 vdev_walk_tree(tvd, online_vdev, NULL) != NULL ||
3073 tvd->vdev_state != VDEV_STATE_HEALTHY) {
3074 if (ztest_opts.zo_verbose >= 5) {
3075 (void) printf("Could not expand LUN because "
3076 "the vdev configuration changed.\n");
3078 spa_config_exit(spa, SCL_STATE, spa);
3079 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3083 spa_config_exit(spa, SCL_STATE, spa);
3086 * Expanding the LUN will update the config asynchronously,
3087 * thus we must wait for the async thread to complete any
3088 * pending tasks before proceeding.
3092 mutex_enter(&spa->spa_async_lock);
3093 done = (spa->spa_async_thread == NULL && !spa->spa_async_tasks);
3094 mutex_exit(&spa->spa_async_lock);
3097 txg_wait_synced(spa_get_dsl(spa), 0);
3098 (void) poll(NULL, 0, 100);
3101 spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3103 tvd = spa->spa_root_vdev->vdev_child[top];
3104 new_ms_count = tvd->vdev_ms_count;
3105 new_class_space = metaslab_class_get_space(mc);
3107 if (tvd->vdev_mg != mg || mg->mg_class != mc) {
3108 if (ztest_opts.zo_verbose >= 5) {
3109 (void) printf("Could not verify LUN expansion due to "
3110 "intervening vdev offline or remove.\n");
3112 spa_config_exit(spa, SCL_STATE, spa);
3113 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3118 * Make sure we were able to grow the vdev.
3120 if (new_ms_count <= old_ms_count)
3121 fatal(0, "LUN expansion failed: ms_count %llu <= %llu\n",
3122 old_ms_count, new_ms_count);
3125 * Make sure we were able to grow the pool.
3127 if (new_class_space <= old_class_space)
3128 fatal(0, "LUN expansion failed: class_space %llu <= %llu\n",
3129 old_class_space, new_class_space);
3131 if (ztest_opts.zo_verbose >= 5) {
3132 char oldnumbuf[6], newnumbuf[6];
3134 nicenum(old_class_space, oldnumbuf);
3135 nicenum(new_class_space, newnumbuf);
3136 (void) printf("%s grew from %s to %s\n",
3137 spa->spa_name, oldnumbuf, newnumbuf);
3140 spa_config_exit(spa, SCL_STATE, spa);
3141 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3145 * Verify that dmu_objset_{create,destroy,open,close} work as expected.
3149 ztest_objset_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
3152 * Create the objects common to all ztest datasets.
3154 VERIFY(zap_create_claim(os, ZTEST_DIROBJ,
3155 DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx) == 0);
3159 ztest_dataset_create(char *dsname)
3161 uint64_t zilset = ztest_random(100);
3162 int err = dmu_objset_create(dsname, DMU_OST_OTHER, 0,
3163 ztest_objset_create_cb, NULL);
3165 if (err || zilset < 80)
3168 if (ztest_opts.zo_verbose >= 6)
3169 (void) printf("Setting dataset %s to sync always\n", dsname);
3170 return (ztest_dsl_prop_set_uint64(dsname, ZFS_PROP_SYNC,
3171 ZFS_SYNC_ALWAYS, B_FALSE));
3176 ztest_objset_destroy_cb(const char *name, void *arg)
3179 dmu_object_info_t doi;
3183 * Verify that the dataset contains a directory object.
3185 VERIFY3U(0, ==, dmu_objset_hold(name, FTAG, &os));
3186 error = dmu_object_info(os, ZTEST_DIROBJ, &doi);
3187 if (error != ENOENT) {
3188 /* We could have crashed in the middle of destroying it */
3190 ASSERT3U(doi.doi_type, ==, DMU_OT_ZAP_OTHER);
3191 ASSERT3S(doi.doi_physical_blocks_512, >=, 0);
3193 dmu_objset_rele(os, FTAG);
3196 * Destroy the dataset.
3198 VERIFY3U(0, ==, dmu_objset_destroy(name, B_FALSE));
3203 ztest_snapshot_create(char *osname, uint64_t id)
3205 char snapname[MAXNAMELEN];
3208 (void) snprintf(snapname, MAXNAMELEN, "%s@%llu", osname,
3211 error = dmu_objset_snapshot(osname, strchr(snapname, '@') + 1,
3212 NULL, NULL, B_FALSE, B_FALSE, -1);
3213 if (error == ENOSPC) {
3214 ztest_record_enospc(FTAG);
3217 if (error != 0 && error != EEXIST)
3218 fatal(0, "ztest_snapshot_create(%s) = %d", snapname, error);
3223 ztest_snapshot_destroy(char *osname, uint64_t id)
3225 char snapname[MAXNAMELEN];
3228 (void) snprintf(snapname, MAXNAMELEN, "%s@%llu", osname,
3231 error = dmu_objset_destroy(snapname, B_FALSE);
3232 if (error != 0 && error != ENOENT)
3233 fatal(0, "ztest_snapshot_destroy(%s) = %d", snapname, error);
3239 ztest_dmu_objset_create_destroy(ztest_ds_t *zd, uint64_t id)
3245 char name[MAXNAMELEN];
3248 (void) rw_rdlock(&ztest_name_lock);
3250 (void) snprintf(name, MAXNAMELEN, "%s/temp_%llu",
3251 ztest_opts.zo_pool, (u_longlong_t)id);
3254 * If this dataset exists from a previous run, process its replay log
3255 * half of the time. If we don't replay it, then dmu_objset_destroy()
3256 * (invoked from ztest_objset_destroy_cb()) should just throw it away.
3258 if (ztest_random(2) == 0 &&
3259 dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os) == 0) {
3260 ztest_zd_init(&zdtmp, NULL, os);
3261 zil_replay(os, &zdtmp, ztest_replay_vector);
3262 ztest_zd_fini(&zdtmp);
3263 dmu_objset_disown(os, FTAG);
3267 * There may be an old instance of the dataset we're about to
3268 * create lying around from a previous run. If so, destroy it
3269 * and all of its snapshots.
3271 (void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
3272 DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
3275 * Verify that the destroyed dataset is no longer in the namespace.
3277 VERIFY3U(ENOENT, ==, dmu_objset_hold(name, FTAG, &os));
3280 * Verify that we can create a new dataset.
3282 error = ztest_dataset_create(name);
3284 if (error == ENOSPC) {
3285 ztest_record_enospc(FTAG);
3286 (void) rw_unlock(&ztest_name_lock);
3289 fatal(0, "dmu_objset_create(%s) = %d", name, error);
3293 dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os));
3295 ztest_zd_init(&zdtmp, NULL, os);
3298 * Open the intent log for it.
3300 zilog = zil_open(os, ztest_get_data);
3303 * Put some objects in there, do a little I/O to them,
3304 * and randomly take a couple of snapshots along the way.
3306 iters = ztest_random(5);
3307 for (int i = 0; i < iters; i++) {
3308 ztest_dmu_object_alloc_free(&zdtmp, id);
3309 if (ztest_random(iters) == 0)
3310 (void) ztest_snapshot_create(name, i);
3314 * Verify that we cannot create an existing dataset.
3316 VERIFY3U(EEXIST, ==,
3317 dmu_objset_create(name, DMU_OST_OTHER, 0, NULL, NULL));
3320 * Verify that we can hold an objset that is also owned.
3322 VERIFY3U(0, ==, dmu_objset_hold(name, FTAG, &os2));
3323 dmu_objset_rele(os2, FTAG);
3326 * Verify that we cannot own an objset that is already owned.
3329 dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os2));
3332 dmu_objset_disown(os, FTAG);
3333 ztest_zd_fini(&zdtmp);
3335 (void) rw_unlock(&ztest_name_lock);
3339 * Verify that dmu_snapshot_{create,destroy,open,close} work as expected.
3342 ztest_dmu_snapshot_create_destroy(ztest_ds_t *zd, uint64_t id)
3344 (void) rw_rdlock(&ztest_name_lock);
3345 (void) ztest_snapshot_destroy(zd->zd_name, id);
3346 (void) ztest_snapshot_create(zd->zd_name, id);
3347 (void) rw_unlock(&ztest_name_lock);
3351 * Cleanup non-standard snapshots and clones.
3354 ztest_dsl_dataset_cleanup(char *osname, uint64_t id)
3356 char snap1name[MAXNAMELEN];
3357 char clone1name[MAXNAMELEN];
3358 char snap2name[MAXNAMELEN];
3359 char clone2name[MAXNAMELEN];
3360 char snap3name[MAXNAMELEN];
3363 (void) snprintf(snap1name, MAXNAMELEN, "%s@s1_%llu", osname, id);
3364 (void) snprintf(clone1name, MAXNAMELEN, "%s/c1_%llu", osname, id);
3365 (void) snprintf(snap2name, MAXNAMELEN, "%s@s2_%llu", clone1name, id);
3366 (void) snprintf(clone2name, MAXNAMELEN, "%s/c2_%llu", osname, id);
3367 (void) snprintf(snap3name, MAXNAMELEN, "%s@s3_%llu", clone1name, id);
3369 error = dmu_objset_destroy(clone2name, B_FALSE);
3370 if (error && error != ENOENT)
3371 fatal(0, "dmu_objset_destroy(%s) = %d", clone2name, error);
3372 error = dmu_objset_destroy(snap3name, B_FALSE);
3373 if (error && error != ENOENT)
3374 fatal(0, "dmu_objset_destroy(%s) = %d", snap3name, error);
3375 error = dmu_objset_destroy(snap2name, B_FALSE);
3376 if (error && error != ENOENT)
3377 fatal(0, "dmu_objset_destroy(%s) = %d", snap2name, error);
3378 error = dmu_objset_destroy(clone1name, B_FALSE);
3379 if (error && error != ENOENT)
3380 fatal(0, "dmu_objset_destroy(%s) = %d", clone1name, error);
3381 error = dmu_objset_destroy(snap1name, B_FALSE);
3382 if (error && error != ENOENT)
3383 fatal(0, "dmu_objset_destroy(%s) = %d", snap1name, error);
3387 * Verify dsl_dataset_promote handles EBUSY
3390 ztest_dsl_dataset_promote_busy(ztest_ds_t *zd, uint64_t id)
3394 char snap1name[MAXNAMELEN];
3395 char clone1name[MAXNAMELEN];
3396 char snap2name[MAXNAMELEN];
3397 char clone2name[MAXNAMELEN];
3398 char snap3name[MAXNAMELEN];
3399 char *osname = zd->zd_name;
3402 (void) rw_rdlock(&ztest_name_lock);
3404 ztest_dsl_dataset_cleanup(osname, id);
3406 (void) snprintf(snap1name, MAXNAMELEN, "%s@s1_%llu", osname, id);
3407 (void) snprintf(clone1name, MAXNAMELEN, "%s/c1_%llu", osname, id);
3408 (void) snprintf(snap2name, MAXNAMELEN, "%s@s2_%llu", clone1name, id);
3409 (void) snprintf(clone2name, MAXNAMELEN, "%s/c2_%llu", osname, id);
3410 (void) snprintf(snap3name, MAXNAMELEN, "%s@s3_%llu", clone1name, id);
3412 error = dmu_objset_snapshot(osname, strchr(snap1name, '@')+1,
3413 NULL, NULL, B_FALSE, B_FALSE, -1);
3414 if (error && error != EEXIST) {
3415 if (error == ENOSPC) {
3416 ztest_record_enospc(FTAG);
3419 fatal(0, "dmu_take_snapshot(%s) = %d", snap1name, error);
3422 error = dmu_objset_hold(snap1name, FTAG, &clone);
3424 fatal(0, "dmu_open_snapshot(%s) = %d", snap1name, error);
3426 error = dmu_objset_clone(clone1name, dmu_objset_ds(clone), 0);
3427 dmu_objset_rele(clone, FTAG);
3429 if (error == ENOSPC) {
3430 ztest_record_enospc(FTAG);
3433 fatal(0, "dmu_objset_create(%s) = %d", clone1name, error);
3436 error = dmu_objset_snapshot(clone1name, strchr(snap2name, '@')+1,
3437 NULL, NULL, B_FALSE, B_FALSE, -1);
3438 if (error && error != EEXIST) {
3439 if (error == ENOSPC) {
3440 ztest_record_enospc(FTAG);
3443 fatal(0, "dmu_open_snapshot(%s) = %d", snap2name, error);
3446 error = dmu_objset_snapshot(clone1name, strchr(snap3name, '@')+1,
3447 NULL, NULL, B_FALSE, B_FALSE, -1);
3448 if (error && error != EEXIST) {
3449 if (error == ENOSPC) {
3450 ztest_record_enospc(FTAG);
3453 fatal(0, "dmu_open_snapshot(%s) = %d", snap3name, error);
3456 error = dmu_objset_hold(snap3name, FTAG, &clone);
3458 fatal(0, "dmu_open_snapshot(%s) = %d", snap3name, error);
3460 error = dmu_objset_clone(clone2name, dmu_objset_ds(clone), 0);
3461 dmu_objset_rele(clone, FTAG);
3463 if (error == ENOSPC) {
3464 ztest_record_enospc(FTAG);
3467 fatal(0, "dmu_objset_create(%s) = %d", clone2name, error);
3470 error = dsl_dataset_own(snap2name, B_FALSE, FTAG, &ds);
3472 fatal(0, "dsl_dataset_own(%s) = %d", snap2name, error);
3473 error = dsl_dataset_promote(clone2name, NULL);
3475 fatal(0, "dsl_dataset_promote(%s), %d, not EBUSY", clone2name,
3477 dsl_dataset_disown(ds, FTAG);
3480 ztest_dsl_dataset_cleanup(osname, id);
3482 (void) rw_unlock(&ztest_name_lock);
3486 * Verify that dmu_object_{alloc,free} work as expected.
3489 ztest_dmu_object_alloc_free(ztest_ds_t *zd, uint64_t id)
3492 int batchsize = sizeof (od) / sizeof (od[0]);
3494 for (int b = 0; b < batchsize; b++)
3495 ztest_od_init(&od[b], id, FTAG, b, DMU_OT_UINT64_OTHER, 0, 0);
3498 * Destroy the previous batch of objects, create a new batch,
3499 * and do some I/O on the new objects.
3501 if (ztest_object_init(zd, od, sizeof (od), B_TRUE) != 0)
3504 while (ztest_random(4 * batchsize) != 0)
3505 ztest_io(zd, od[ztest_random(batchsize)].od_object,
3506 ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
3510 * Verify that dmu_{read,write} work as expected.
3513 ztest_dmu_read_write(ztest_ds_t *zd, uint64_t id)
3515 objset_t *os = zd->zd_os;
3518 int i, freeit, error;
3520 bufwad_t *packbuf, *bigbuf, *pack, *bigH, *bigT;
3521 uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
3522 uint64_t chunksize = (1000 + ztest_random(1000)) * sizeof (uint64_t);
3523 uint64_t regions = 997;
3524 uint64_t stride = 123456789ULL;
3525 uint64_t width = 40;
3526 int free_percent = 5;
3529 * This test uses two objects, packobj and bigobj, that are always
3530 * updated together (i.e. in the same tx) so that their contents are
3531 * in sync and can be compared. Their contents relate to each other
3532 * in a simple way: packobj is a dense array of 'bufwad' structures,
3533 * while bigobj is a sparse array of the same bufwads. Specifically,
3534 * for any index n, there are three bufwads that should be identical:
3536 * packobj, at offset n * sizeof (bufwad_t)
3537 * bigobj, at the head of the nth chunk
3538 * bigobj, at the tail of the nth chunk
3540 * The chunk size is arbitrary. It doesn't have to be a power of two,
3541 * and it doesn't have any relation to the object blocksize.
3542 * The only requirement is that it can hold at least two bufwads.
3544 * Normally, we write the bufwad to each of these locations.
3545 * However, free_percent of the time we instead write zeroes to
3546 * packobj and perform a dmu_free_range() on bigobj. By comparing
3547 * bigobj to packobj, we can verify that the DMU is correctly
3548 * tracking which parts of an object are allocated and free,
3549 * and that the contents of the allocated blocks are correct.
3553 * Read the directory info. If it's the first time, set things up.
3555 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, chunksize);
3556 ztest_od_init(&od[1], id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize);
3558 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
3561 bigobj = od[0].od_object;
3562 packobj = od[1].od_object;
3563 chunksize = od[0].od_gen;
3564 ASSERT(chunksize == od[1].od_gen);
3567 * Prefetch a random chunk of the big object.
3568 * Our aim here is to get some async reads in flight
3569 * for blocks that we may free below; the DMU should
3570 * handle this race correctly.
3572 n = ztest_random(regions) * stride + ztest_random(width);
3573 s = 1 + ztest_random(2 * width - 1);
3574 dmu_prefetch(os, bigobj, n * chunksize, s * chunksize);
3577 * Pick a random index and compute the offsets into packobj and bigobj.
3579 n = ztest_random(regions) * stride + ztest_random(width);
3580 s = 1 + ztest_random(width - 1);
3582 packoff = n * sizeof (bufwad_t);
3583 packsize = s * sizeof (bufwad_t);
3585 bigoff = n * chunksize;
3586 bigsize = s * chunksize;
3588 packbuf = umem_alloc(packsize, UMEM_NOFAIL);
3589 bigbuf = umem_alloc(bigsize, UMEM_NOFAIL);
3592 * free_percent of the time, free a range of bigobj rather than
3595 freeit = (ztest_random(100) < free_percent);
3598 * Read the current contents of our objects.
3600 error = dmu_read(os, packobj, packoff, packsize, packbuf,
3603 error = dmu_read(os, bigobj, bigoff, bigsize, bigbuf,
3608 * Get a tx for the mods to both packobj and bigobj.
3610 tx = dmu_tx_create(os);
3612 dmu_tx_hold_write(tx, packobj, packoff, packsize);
3615 dmu_tx_hold_free(tx, bigobj, bigoff, bigsize);
3617 dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
3619 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3621 umem_free(packbuf, packsize);
3622 umem_free(bigbuf, bigsize);
3626 dmu_object_set_checksum(os, bigobj,
3627 (enum zio_checksum)ztest_random_dsl_prop(ZFS_PROP_CHECKSUM), tx);
3629 dmu_object_set_compress(os, bigobj,
3630 (enum zio_compress)ztest_random_dsl_prop(ZFS_PROP_COMPRESSION), tx);
3633 * For each index from n to n + s, verify that the existing bufwad
3634 * in packobj matches the bufwads at the head and tail of the
3635 * corresponding chunk in bigobj. Then update all three bufwads
3636 * with the new values we want to write out.
3638 for (i = 0; i < s; i++) {
3640 pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
3642 bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
3644 bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
3646 ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
3647 ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
3649 if (pack->bw_txg > txg)
3650 fatal(0, "future leak: got %llx, open txg is %llx",
3653 if (pack->bw_data != 0 && pack->bw_index != n + i)
3654 fatal(0, "wrong index: got %llx, wanted %llx+%llx",
3655 pack->bw_index, n, i);
3657 if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
3658 fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
3660 if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
3661 fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
3664 bzero(pack, sizeof (bufwad_t));
3666 pack->bw_index = n + i;
3668 pack->bw_data = 1 + ztest_random(-2ULL);
3675 * We've verified all the old bufwads, and made new ones.
3676 * Now write them out.
3678 dmu_write(os, packobj, packoff, packsize, packbuf, tx);
3681 if (ztest_opts.zo_verbose >= 7) {
3682 (void) printf("freeing offset %llx size %llx"
3684 (u_longlong_t)bigoff,
3685 (u_longlong_t)bigsize,
3688 VERIFY(0 == dmu_free_range(os, bigobj, bigoff, bigsize, tx));
3690 if (ztest_opts.zo_verbose >= 7) {
3691 (void) printf("writing offset %llx size %llx"
3693 (u_longlong_t)bigoff,
3694 (u_longlong_t)bigsize,
3697 dmu_write(os, bigobj, bigoff, bigsize, bigbuf, tx);
3703 * Sanity check the stuff we just wrote.
3706 void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
3707 void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
3709 VERIFY(0 == dmu_read(os, packobj, packoff,
3710 packsize, packcheck, DMU_READ_PREFETCH));
3711 VERIFY(0 == dmu_read(os, bigobj, bigoff,
3712 bigsize, bigcheck, DMU_READ_PREFETCH));
3714 ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
3715 ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
3717 umem_free(packcheck, packsize);
3718 umem_free(bigcheck, bigsize);
3721 umem_free(packbuf, packsize);
3722 umem_free(bigbuf, bigsize);
3726 compare_and_update_pbbufs(uint64_t s, bufwad_t *packbuf, bufwad_t *bigbuf,
3727 uint64_t bigsize, uint64_t n, uint64_t chunksize, uint64_t txg)
3735 * For each index from n to n + s, verify that the existing bufwad
3736 * in packobj matches the bufwads at the head and tail of the
3737 * corresponding chunk in bigobj. Then update all three bufwads
3738 * with the new values we want to write out.
3740 for (i = 0; i < s; i++) {
3742 pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
3744 bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
3746 bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
3748 ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
3749 ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
3751 if (pack->bw_txg > txg)
3752 fatal(0, "future leak: got %llx, open txg is %llx",
3755 if (pack->bw_data != 0 && pack->bw_index != n + i)
3756 fatal(0, "wrong index: got %llx, wanted %llx+%llx",
3757 pack->bw_index, n, i);
3759 if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
3760 fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
3762 if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
3763 fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
3765 pack->bw_index = n + i;
3767 pack->bw_data = 1 + ztest_random(-2ULL);
3775 ztest_dmu_read_write_zcopy(ztest_ds_t *zd, uint64_t id)
3777 objset_t *os = zd->zd_os;
3783 bufwad_t *packbuf, *bigbuf;
3784 uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
3785 uint64_t blocksize = ztest_random_blocksize();
3786 uint64_t chunksize = blocksize;
3787 uint64_t regions = 997;
3788 uint64_t stride = 123456789ULL;
3790 dmu_buf_t *bonus_db;
3791 arc_buf_t **bigbuf_arcbufs;
3792 dmu_object_info_t doi;
3795 * This test uses two objects, packobj and bigobj, that are always
3796 * updated together (i.e. in the same tx) so that their contents are
3797 * in sync and can be compared. Their contents relate to each other
3798 * in a simple way: packobj is a dense array of 'bufwad' structures,
3799 * while bigobj is a sparse array of the same bufwads. Specifically,
3800 * for any index n, there are three bufwads that should be identical:
3802 * packobj, at offset n * sizeof (bufwad_t)
3803 * bigobj, at the head of the nth chunk
3804 * bigobj, at the tail of the nth chunk
3806 * The chunk size is set equal to bigobj block size so that
3807 * dmu_assign_arcbuf() can be tested for object updates.
3811 * Read the directory info. If it's the first time, set things up.
3813 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
3814 ztest_od_init(&od[1], id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize);
3816 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
3819 bigobj = od[0].od_object;
3820 packobj = od[1].od_object;
3821 blocksize = od[0].od_blocksize;
3822 chunksize = blocksize;
3823 ASSERT(chunksize == od[1].od_gen);
3825 VERIFY(dmu_object_info(os, bigobj, &doi) == 0);
3826 VERIFY(ISP2(doi.doi_data_block_size));
3827 VERIFY(chunksize == doi.doi_data_block_size);
3828 VERIFY(chunksize >= 2 * sizeof (bufwad_t));
3831 * Pick a random index and compute the offsets into packobj and bigobj.
3833 n = ztest_random(regions) * stride + ztest_random(width);
3834 s = 1 + ztest_random(width - 1);
3836 packoff = n * sizeof (bufwad_t);
3837 packsize = s * sizeof (bufwad_t);
3839 bigoff = n * chunksize;
3840 bigsize = s * chunksize;
3842 packbuf = umem_zalloc(packsize, UMEM_NOFAIL);
3843 bigbuf = umem_zalloc(bigsize, UMEM_NOFAIL);
3845 VERIFY3U(0, ==, dmu_bonus_hold(os, bigobj, FTAG, &bonus_db));
3847 bigbuf_arcbufs = umem_zalloc(2 * s * sizeof (arc_buf_t *), UMEM_NOFAIL);
3850 * Iteration 0 test zcopy for DB_UNCACHED dbufs.
3851 * Iteration 1 test zcopy to already referenced dbufs.
3852 * Iteration 2 test zcopy to dirty dbuf in the same txg.
3853 * Iteration 3 test zcopy to dbuf dirty in previous txg.
3854 * Iteration 4 test zcopy when dbuf is no longer dirty.
3855 * Iteration 5 test zcopy when it can't be done.
3856 * Iteration 6 one more zcopy write.
3858 for (i = 0; i < 7; i++) {
3863 * In iteration 5 (i == 5) use arcbufs
3864 * that don't match bigobj blksz to test
3865 * dmu_assign_arcbuf() when it can't directly
3866 * assign an arcbuf to a dbuf.
3868 for (j = 0; j < s; j++) {
3871 dmu_request_arcbuf(bonus_db, chunksize);
3873 bigbuf_arcbufs[2 * j] =
3874 dmu_request_arcbuf(bonus_db, chunksize / 2);
3875 bigbuf_arcbufs[2 * j + 1] =
3876 dmu_request_arcbuf(bonus_db, chunksize / 2);
3881 * Get a tx for the mods to both packobj and bigobj.
3883 tx = dmu_tx_create(os);
3885 dmu_tx_hold_write(tx, packobj, packoff, packsize);
3886 dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
3888 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3890 umem_free(packbuf, packsize);
3891 umem_free(bigbuf, bigsize);
3892 for (j = 0; j < s; j++) {
3894 dmu_return_arcbuf(bigbuf_arcbufs[j]);
3897 bigbuf_arcbufs[2 * j]);
3899 bigbuf_arcbufs[2 * j + 1]);
3902 umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
3903 dmu_buf_rele(bonus_db, FTAG);
3908 * 50% of the time don't read objects in the 1st iteration to
3909 * test dmu_assign_arcbuf() for the case when there're no
3910 * existing dbufs for the specified offsets.
3912 if (i != 0 || ztest_random(2) != 0) {
3913 error = dmu_read(os, packobj, packoff,
3914 packsize, packbuf, DMU_READ_PREFETCH);
3916 error = dmu_read(os, bigobj, bigoff, bigsize,
3917 bigbuf, DMU_READ_PREFETCH);
3920 compare_and_update_pbbufs(s, packbuf, bigbuf, bigsize,
3924 * We've verified all the old bufwads, and made new ones.
3925 * Now write them out.
3927 dmu_write(os, packobj, packoff, packsize, packbuf, tx);
3928 if (ztest_opts.zo_verbose >= 7) {
3929 (void) printf("writing offset %llx size %llx"
3931 (u_longlong_t)bigoff,
3932 (u_longlong_t)bigsize,
3935 for (off = bigoff, j = 0; j < s; j++, off += chunksize) {
3938 bcopy((caddr_t)bigbuf + (off - bigoff),
3939 bigbuf_arcbufs[j]->b_data, chunksize);
3941 bcopy((caddr_t)bigbuf + (off - bigoff),
3942 bigbuf_arcbufs[2 * j]->b_data,
3944 bcopy((caddr_t)bigbuf + (off - bigoff) +
3946 bigbuf_arcbufs[2 * j + 1]->b_data,
3951 VERIFY(dmu_buf_hold(os, bigobj, off,
3952 FTAG, &dbt, DMU_READ_NO_PREFETCH) == 0);
3955 dmu_assign_arcbuf(bonus_db, off,
3956 bigbuf_arcbufs[j], tx);
3958 dmu_assign_arcbuf(bonus_db, off,
3959 bigbuf_arcbufs[2 * j], tx);
3960 dmu_assign_arcbuf(bonus_db,
3961 off + chunksize / 2,
3962 bigbuf_arcbufs[2 * j + 1], tx);
3965 dmu_buf_rele(dbt, FTAG);
3971 * Sanity check the stuff we just wrote.
3974 void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
3975 void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
3977 VERIFY(0 == dmu_read(os, packobj, packoff,
3978 packsize, packcheck, DMU_READ_PREFETCH));
3979 VERIFY(0 == dmu_read(os, bigobj, bigoff,
3980 bigsize, bigcheck, DMU_READ_PREFETCH));
3982 ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
3983 ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
3985 umem_free(packcheck, packsize);
3986 umem_free(bigcheck, bigsize);
3989 txg_wait_open(dmu_objset_pool(os), 0);
3990 } else if (i == 3) {
3991 txg_wait_synced(dmu_objset_pool(os), 0);
3995 dmu_buf_rele(bonus_db, FTAG);
3996 umem_free(packbuf, packsize);
3997 umem_free(bigbuf, bigsize);
3998 umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
4003 ztest_dmu_write_parallel(ztest_ds_t *zd, uint64_t id)
4006 uint64_t offset = (1ULL << (ztest_random(20) + 43)) +
4007 (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
4010 * Have multiple threads write to large offsets in an object
4011 * to verify that parallel writes to an object -- even to the
4012 * same blocks within the object -- doesn't cause any trouble.
4014 ztest_od_init(&od[0], ID_PARALLEL, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
4016 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4019 while (ztest_random(10) != 0)
4020 ztest_io(zd, od[0].od_object, offset);
4024 ztest_dmu_prealloc(ztest_ds_t *zd, uint64_t id)
4027 uint64_t offset = (1ULL << (ztest_random(4) + SPA_MAXBLOCKSHIFT)) +
4028 (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
4029 uint64_t count = ztest_random(20) + 1;
4030 uint64_t blocksize = ztest_random_blocksize();
4033 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
4035 if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4038 if (ztest_truncate(zd, od[0].od_object, offset, count * blocksize) != 0)
4041 ztest_prealloc(zd, od[0].od_object, offset, count * blocksize);
4043 data = umem_zalloc(blocksize, UMEM_NOFAIL);
4045 while (ztest_random(count) != 0) {
4046 uint64_t randoff = offset + (ztest_random(count) * blocksize);
4047 if (ztest_write(zd, od[0].od_object, randoff, blocksize,
4050 while (ztest_random(4) != 0)
4051 ztest_io(zd, od[0].od_object, randoff);
4054 umem_free(data, blocksize);
4058 * Verify that zap_{create,destroy,add,remove,update} work as expected.
4060 #define ZTEST_ZAP_MIN_INTS 1
4061 #define ZTEST_ZAP_MAX_INTS 4
4062 #define ZTEST_ZAP_MAX_PROPS 1000
4065 ztest_zap(ztest_ds_t *zd, uint64_t id)
4067 objset_t *os = zd->zd_os;
4070 uint64_t txg, last_txg;
4071 uint64_t value[ZTEST_ZAP_MAX_INTS];
4072 uint64_t zl_ints, zl_intsize, prop;
4075 char propname[100], txgname[100];
4077 char *hc[2] = { "s.acl.h", ".s.open.h.hyLZlg" };
4079 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0);
4081 if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4084 object = od[0].od_object;
4087 * Generate a known hash collision, and verify that
4088 * we can lookup and remove both entries.
4090 tx = dmu_tx_create(os);
4091 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4092 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4095 for (i = 0; i < 2; i++) {
4097 VERIFY3U(0, ==, zap_add(os, object, hc[i], sizeof (uint64_t),
4100 for (i = 0; i < 2; i++) {
4101 VERIFY3U(EEXIST, ==, zap_add(os, object, hc[i],
4102 sizeof (uint64_t), 1, &value[i], tx));
4104 zap_length(os, object, hc[i], &zl_intsize, &zl_ints));
4105 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4106 ASSERT3U(zl_ints, ==, 1);
4108 for (i = 0; i < 2; i++) {
4109 VERIFY3U(0, ==, zap_remove(os, object, hc[i], tx));
4114 * Generate a buch of random entries.
4116 ints = MAX(ZTEST_ZAP_MIN_INTS, object % ZTEST_ZAP_MAX_INTS);
4118 prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
4119 (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
4120 (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
4121 bzero(value, sizeof (value));
4125 * If these zap entries already exist, validate their contents.
4127 error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
4129 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4130 ASSERT3U(zl_ints, ==, 1);
4132 VERIFY(zap_lookup(os, object, txgname, zl_intsize,
4133 zl_ints, &last_txg) == 0);
4135 VERIFY(zap_length(os, object, propname, &zl_intsize,
4138 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4139 ASSERT3U(zl_ints, ==, ints);
4141 VERIFY(zap_lookup(os, object, propname, zl_intsize,
4142 zl_ints, value) == 0);
4144 for (i = 0; i < ints; i++) {
4145 ASSERT3U(value[i], ==, last_txg + object + i);
4148 ASSERT3U(error, ==, ENOENT);
4152 * Atomically update two entries in our zap object.
4153 * The first is named txg_%llu, and contains the txg
4154 * in which the property was last updated. The second
4155 * is named prop_%llu, and the nth element of its value
4156 * should be txg + object + n.
4158 tx = dmu_tx_create(os);
4159 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4160 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4165 fatal(0, "zap future leak: old %llu new %llu", last_txg, txg);
4167 for (i = 0; i < ints; i++)
4168 value[i] = txg + object + i;
4170 VERIFY3U(0, ==, zap_update(os, object, txgname, sizeof (uint64_t),
4172 VERIFY3U(0, ==, zap_update(os, object, propname, sizeof (uint64_t),
4178 * Remove a random pair of entries.
4180 prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
4181 (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
4182 (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
4184 error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
4186 if (error == ENOENT)
4191 tx = dmu_tx_create(os);
4192 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4193 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4196 VERIFY3U(0, ==, zap_remove(os, object, txgname, tx));
4197 VERIFY3U(0, ==, zap_remove(os, object, propname, tx));
4202 * Testcase to test the upgrading of a microzap to fatzap.
4205 ztest_fzap(ztest_ds_t *zd, uint64_t id)
4207 objset_t *os = zd->zd_os;
4209 uint64_t object, txg;
4211 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0);
4213 if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4216 object = od[0].od_object;
4219 * Add entries to this ZAP and make sure it spills over
4220 * and gets upgraded to a fatzap. Also, since we are adding
4221 * 2050 entries we should see ptrtbl growth and leaf-block split.
4223 for (int i = 0; i < 2050; i++) {
4224 char name[MAXNAMELEN];
4229 (void) snprintf(name, sizeof (name), "fzap-%llu-%llu",
4232 tx = dmu_tx_create(os);
4233 dmu_tx_hold_zap(tx, object, B_TRUE, name);
4234 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4237 error = zap_add(os, object, name, sizeof (uint64_t), 1,
4239 ASSERT(error == 0 || error == EEXIST);
4246 ztest_zap_parallel(ztest_ds_t *zd, uint64_t id)
4248 objset_t *os = zd->zd_os;
4250 uint64_t txg, object, count, wsize, wc, zl_wsize, zl_wc;
4252 int i, namelen, error;
4253 int micro = ztest_random(2);
4254 char name[20], string_value[20];
4257 ztest_od_init(&od[0], ID_PARALLEL, FTAG, micro, DMU_OT_ZAP_OTHER, 0, 0);
4259 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4262 object = od[0].od_object;
4265 * Generate a random name of the form 'xxx.....' where each
4266 * x is a random printable character and the dots are dots.
4267 * There are 94 such characters, and the name length goes from
4268 * 6 to 20, so there are 94^3 * 15 = 12,458,760 possible names.
4270 namelen = ztest_random(sizeof (name) - 5) + 5 + 1;
4272 for (i = 0; i < 3; i++)
4273 name[i] = '!' + ztest_random('~' - '!' + 1);
4274 for (; i < namelen - 1; i++)
4278 if ((namelen & 1) || micro) {
4279 wsize = sizeof (txg);
4285 data = string_value;
4289 VERIFY(zap_count(os, object, &count) == 0);
4290 ASSERT(count != -1ULL);
4293 * Select an operation: length, lookup, add, update, remove.
4295 i = ztest_random(5);
4298 tx = dmu_tx_create(os);
4299 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4300 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4303 bcopy(name, string_value, namelen);
4307 bzero(string_value, namelen);
4313 error = zap_length(os, object, name, &zl_wsize, &zl_wc);
4315 ASSERT3U(wsize, ==, zl_wsize);
4316 ASSERT3U(wc, ==, zl_wc);
4318 ASSERT3U(error, ==, ENOENT);
4323 error = zap_lookup(os, object, name, wsize, wc, data);
4325 if (data == string_value &&
4326 bcmp(name, data, namelen) != 0)
4327 fatal(0, "name '%s' != val '%s' len %d",
4328 name, data, namelen);
4330 ASSERT3U(error, ==, ENOENT);
4335 error = zap_add(os, object, name, wsize, wc, data, tx);
4336 ASSERT(error == 0 || error == EEXIST);
4340 VERIFY(zap_update(os, object, name, wsize, wc, data, tx) == 0);
4344 error = zap_remove(os, object, name, tx);
4345 ASSERT(error == 0 || error == ENOENT);
4354 * Commit callback data.
4356 typedef struct ztest_cb_data {
4357 list_node_t zcd_node;
4359 int zcd_expected_err;
4360 boolean_t zcd_added;
4361 boolean_t zcd_called;
4365 /* This is the actual commit callback function */
4367 ztest_commit_callback(void *arg, int error)
4369 ztest_cb_data_t *data = arg;
4370 uint64_t synced_txg;
4372 VERIFY(data != NULL);
4373 VERIFY3S(data->zcd_expected_err, ==, error);
4374 VERIFY(!data->zcd_called);
4376 synced_txg = spa_last_synced_txg(data->zcd_spa);
4377 if (data->zcd_txg > synced_txg)
4378 fatal(0, "commit callback of txg %" PRIu64 " called prematurely"
4379 ", last synced txg = %" PRIu64 "\n", data->zcd_txg,
4382 data->zcd_called = B_TRUE;
4384 if (error == ECANCELED) {
4385 ASSERT0(data->zcd_txg);
4386 ASSERT(!data->zcd_added);
4389 * The private callback data should be destroyed here, but
4390 * since we are going to check the zcd_called field after
4391 * dmu_tx_abort(), we will destroy it there.
4396 /* Was this callback added to the global callback list? */
4397 if (!data->zcd_added)
4400 ASSERT3U(data->zcd_txg, !=, 0);
4402 /* Remove our callback from the list */
4403 (void) mutex_lock(&zcl.zcl_callbacks_lock);
4404 list_remove(&zcl.zcl_callbacks, data);
4405 (void) mutex_unlock(&zcl.zcl_callbacks_lock);
4408 umem_free(data, sizeof (ztest_cb_data_t));
4411 /* Allocate and initialize callback data structure */
4412 static ztest_cb_data_t *
4413 ztest_create_cb_data(objset_t *os, uint64_t txg)
4415 ztest_cb_data_t *cb_data;
4417 cb_data = umem_zalloc(sizeof (ztest_cb_data_t), UMEM_NOFAIL);
4419 cb_data->zcd_txg = txg;
4420 cb_data->zcd_spa = dmu_objset_spa(os);
4426 * If a number of txgs equal to this threshold have been created after a commit
4427 * callback has been registered but not called, then we assume there is an
4428 * implementation bug.
4430 #define ZTEST_COMMIT_CALLBACK_THRESH (TXG_CONCURRENT_STATES + 2)
4433 * Commit callback test.
4436 ztest_dmu_commit_callbacks(ztest_ds_t *zd, uint64_t id)
4438 objset_t *os = zd->zd_os;
4441 ztest_cb_data_t *cb_data[3], *tmp_cb;
4442 uint64_t old_txg, txg;
4445 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
4447 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4450 tx = dmu_tx_create(os);
4452 cb_data[0] = ztest_create_cb_data(os, 0);
4453 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[0]);
4455 dmu_tx_hold_write(tx, od[0].od_object, 0, sizeof (uint64_t));
4457 /* Every once in a while, abort the transaction on purpose */
4458 if (ztest_random(100) == 0)
4462 error = dmu_tx_assign(tx, TXG_NOWAIT);
4464 txg = error ? 0 : dmu_tx_get_txg(tx);
4466 cb_data[0]->zcd_txg = txg;
4467 cb_data[1] = ztest_create_cb_data(os, txg);
4468 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[1]);
4472 * It's not a strict requirement to call the registered
4473 * callbacks from inside dmu_tx_abort(), but that's what
4474 * it's supposed to happen in the current implementation
4475 * so we will check for that.
4477 for (i = 0; i < 2; i++) {
4478 cb_data[i]->zcd_expected_err = ECANCELED;
4479 VERIFY(!cb_data[i]->zcd_called);
4484 for (i = 0; i < 2; i++) {
4485 VERIFY(cb_data[i]->zcd_called);
4486 umem_free(cb_data[i], sizeof (ztest_cb_data_t));
4492 cb_data[2] = ztest_create_cb_data(os, txg);
4493 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[2]);
4496 * Read existing data to make sure there isn't a future leak.
4498 VERIFY(0 == dmu_read(os, od[0].od_object, 0, sizeof (uint64_t),
4499 &old_txg, DMU_READ_PREFETCH));
4502 fatal(0, "future leak: got %" PRIu64 ", open txg is %" PRIu64,
4505 dmu_write(os, od[0].od_object, 0, sizeof (uint64_t), &txg, tx);
4507 (void) mutex_lock(&zcl.zcl_callbacks_lock);
4510 * Since commit callbacks don't have any ordering requirement and since
4511 * it is theoretically possible for a commit callback to be called
4512 * after an arbitrary amount of time has elapsed since its txg has been
4513 * synced, it is difficult to reliably determine whether a commit
4514 * callback hasn't been called due to high load or due to a flawed
4517 * In practice, we will assume that if after a certain number of txgs a
4518 * commit callback hasn't been called, then most likely there's an
4519 * implementation bug..
4521 tmp_cb = list_head(&zcl.zcl_callbacks);
4522 if (tmp_cb != NULL &&
4523 tmp_cb->zcd_txg > txg - ZTEST_COMMIT_CALLBACK_THRESH) {
4524 fatal(0, "Commit callback threshold exceeded, oldest txg: %"
4525 PRIu64 ", open txg: %" PRIu64 "\n", tmp_cb->zcd_txg, txg);
4529 * Let's find the place to insert our callbacks.
4531 * Even though the list is ordered by txg, it is possible for the
4532 * insertion point to not be the end because our txg may already be
4533 * quiescing at this point and other callbacks in the open txg
4534 * (from other objsets) may have sneaked in.
4536 tmp_cb = list_tail(&zcl.zcl_callbacks);
4537 while (tmp_cb != NULL && tmp_cb->zcd_txg > txg)
4538 tmp_cb = list_prev(&zcl.zcl_callbacks, tmp_cb);
4540 /* Add the 3 callbacks to the list */
4541 for (i = 0; i < 3; i++) {
4543 list_insert_head(&zcl.zcl_callbacks, cb_data[i]);
4545 list_insert_after(&zcl.zcl_callbacks, tmp_cb,
4548 cb_data[i]->zcd_added = B_TRUE;
4549 VERIFY(!cb_data[i]->zcd_called);
4551 tmp_cb = cb_data[i];
4554 (void) mutex_unlock(&zcl.zcl_callbacks_lock);
4561 ztest_dsl_prop_get_set(ztest_ds_t *zd, uint64_t id)
4563 zfs_prop_t proplist[] = {
4565 ZFS_PROP_COMPRESSION,
4570 (void) rw_rdlock(&ztest_name_lock);
4572 for (int p = 0; p < sizeof (proplist) / sizeof (proplist[0]); p++)
4573 (void) ztest_dsl_prop_set_uint64(zd->zd_name, proplist[p],
4574 ztest_random_dsl_prop(proplist[p]), (int)ztest_random(2));
4576 (void) rw_unlock(&ztest_name_lock);
4581 ztest_spa_prop_get_set(ztest_ds_t *zd, uint64_t id)
4583 nvlist_t *props = NULL;
4585 (void) rw_rdlock(&ztest_name_lock);
4587 (void) ztest_spa_prop_set_uint64(ZPOOL_PROP_DEDUPDITTO,
4588 ZIO_DEDUPDITTO_MIN + ztest_random(ZIO_DEDUPDITTO_MIN));
4590 VERIFY0(spa_prop_get(ztest_spa, &props));
4592 if (ztest_opts.zo_verbose >= 6)
4593 dump_nvlist(props, 4);
4597 (void) rw_unlock(&ztest_name_lock);
4601 * Test snapshot hold/release and deferred destroy.
4604 ztest_dmu_snapshot_hold(ztest_ds_t *zd, uint64_t id)
4607 objset_t *os = zd->zd_os;
4611 char clonename[100];
4613 char osname[MAXNAMELEN];
4615 (void) rw_rdlock(&ztest_name_lock);
4617 dmu_objset_name(os, osname);
4619 (void) snprintf(snapname, 100, "sh1_%llu", id);
4620 (void) snprintf(fullname, 100, "%s@%s", osname, snapname);
4621 (void) snprintf(clonename, 100, "%s/ch1_%llu", osname, id);
4622 (void) snprintf(tag, 100, "%tag_%llu", id);
4625 * Clean up from any previous run.
4627 (void) dmu_objset_destroy(clonename, B_FALSE);
4628 (void) dsl_dataset_user_release(osname, snapname, tag, B_FALSE);
4629 (void) dmu_objset_destroy(fullname, B_FALSE);
4632 * Create snapshot, clone it, mark snap for deferred destroy,
4633 * destroy clone, verify snap was also destroyed.
4635 error = dmu_objset_snapshot(osname, snapname, NULL, NULL, FALSE,
4638 if (error == ENOSPC) {
4639 ztest_record_enospc("dmu_objset_snapshot");
4642 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
4645 error = dmu_objset_hold(fullname, FTAG, &origin);
4647 fatal(0, "dmu_objset_hold(%s) = %d", fullname, error);
4649 error = dmu_objset_clone(clonename, dmu_objset_ds(origin), 0);
4650 dmu_objset_rele(origin, FTAG);
4652 if (error == ENOSPC) {
4653 ztest_record_enospc("dmu_objset_clone");
4656 fatal(0, "dmu_objset_clone(%s) = %d", clonename, error);
4659 error = dmu_objset_destroy(fullname, B_TRUE);
4661 fatal(0, "dmu_objset_destroy(%s, B_TRUE) = %d",
4665 error = dmu_objset_destroy(clonename, B_FALSE);
4667 fatal(0, "dmu_objset_destroy(%s) = %d", clonename, error);
4669 error = dmu_objset_hold(fullname, FTAG, &origin);
4670 if (error != ENOENT)
4671 fatal(0, "dmu_objset_hold(%s) = %d", fullname, error);
4674 * Create snapshot, add temporary hold, verify that we can't
4675 * destroy a held snapshot, mark for deferred destroy,
4676 * release hold, verify snapshot was destroyed.
4678 error = dmu_objset_snapshot(osname, snapname, NULL, NULL, FALSE,
4681 if (error == ENOSPC) {
4682 ztest_record_enospc("dmu_objset_snapshot");
4685 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
4688 error = dsl_dataset_user_hold(osname, snapname, tag, B_FALSE,
4691 fatal(0, "dsl_dataset_user_hold(%s)", fullname, tag);
4693 error = dmu_objset_destroy(fullname, B_FALSE);
4694 if (error != EBUSY) {
4695 fatal(0, "dmu_objset_destroy(%s, B_FALSE) = %d",
4699 error = dmu_objset_destroy(fullname, B_TRUE);
4701 fatal(0, "dmu_objset_destroy(%s, B_TRUE) = %d",
4705 error = dsl_dataset_user_release(osname, snapname, tag, B_FALSE);
4707 fatal(0, "dsl_dataset_user_release(%s)", fullname, tag);
4709 VERIFY(dmu_objset_hold(fullname, FTAG, &origin) == ENOENT);
4712 (void) rw_unlock(&ztest_name_lock);
4716 * Inject random faults into the on-disk data.
4720 ztest_fault_inject(ztest_ds_t *zd, uint64_t id)
4722 ztest_shared_t *zs = ztest_shared;
4723 spa_t *spa = ztest_spa;
4727 uint64_t bad = 0x1990c0ffeedecadeULL;
4729 char path0[MAXPATHLEN];
4730 char pathrand[MAXPATHLEN];
4732 int bshift = SPA_MAXBLOCKSHIFT + 2; /* don't scrog all labels */
4738 boolean_t islog = B_FALSE;
4740 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
4741 maxfaults = MAXFAULTS();
4742 leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz;
4743 mirror_save = zs->zs_mirrors;
4744 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4746 ASSERT(leaves >= 1);
4749 * We need SCL_STATE here because we're going to look at vd0->vdev_tsd.
4751 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
4753 if (ztest_random(2) == 0) {
4755 * Inject errors on a normal data device or slog device.
4757 top = ztest_random_vdev_top(spa, B_TRUE);
4758 leaf = ztest_random(leaves) + zs->zs_splits;
4761 * Generate paths to the first leaf in this top-level vdev,
4762 * and to the random leaf we selected. We'll induce transient
4763 * write failures and random online/offline activity on leaf 0,
4764 * and we'll write random garbage to the randomly chosen leaf.
4766 (void) snprintf(path0, sizeof (path0), ztest_dev_template,
4767 ztest_opts.zo_dir, ztest_opts.zo_pool,
4768 top * leaves + zs->zs_splits);
4769 (void) snprintf(pathrand, sizeof (pathrand), ztest_dev_template,
4770 ztest_opts.zo_dir, ztest_opts.zo_pool,
4771 top * leaves + leaf);
4773 vd0 = vdev_lookup_by_path(spa->spa_root_vdev, path0);
4774 if (vd0 != NULL && vd0->vdev_top->vdev_islog)
4777 if (vd0 != NULL && maxfaults != 1) {
4779 * Make vd0 explicitly claim to be unreadable,
4780 * or unwriteable, or reach behind its back
4781 * and close the underlying fd. We can do this if
4782 * maxfaults == 0 because we'll fail and reexecute,
4783 * and we can do it if maxfaults >= 2 because we'll
4784 * have enough redundancy. If maxfaults == 1, the
4785 * combination of this with injection of random data
4786 * corruption below exceeds the pool's fault tolerance.
4788 vdev_file_t *vf = vd0->vdev_tsd;
4790 if (vf != NULL && ztest_random(3) == 0) {
4791 (void) close(vf->vf_vnode->v_fd);
4792 vf->vf_vnode->v_fd = -1;
4793 } else if (ztest_random(2) == 0) {
4794 vd0->vdev_cant_read = B_TRUE;
4796 vd0->vdev_cant_write = B_TRUE;
4798 guid0 = vd0->vdev_guid;
4802 * Inject errors on an l2cache device.
4804 spa_aux_vdev_t *sav = &spa->spa_l2cache;
4806 if (sav->sav_count == 0) {
4807 spa_config_exit(spa, SCL_STATE, FTAG);
4810 vd0 = sav->sav_vdevs[ztest_random(sav->sav_count)];
4811 guid0 = vd0->vdev_guid;
4812 (void) strcpy(path0, vd0->vdev_path);
4813 (void) strcpy(pathrand, vd0->vdev_path);
4817 maxfaults = INT_MAX; /* no limit on cache devices */
4820 spa_config_exit(spa, SCL_STATE, FTAG);
4823 * If we can tolerate two or more faults, or we're dealing
4824 * with a slog, randomly online/offline vd0.
4826 if ((maxfaults >= 2 || islog) && guid0 != 0) {
4827 if (ztest_random(10) < 6) {
4828 int flags = (ztest_random(2) == 0 ?
4829 ZFS_OFFLINE_TEMPORARY : 0);
4832 * We have to grab the zs_name_lock as writer to
4833 * prevent a race between offlining a slog and
4834 * destroying a dataset. Offlining the slog will
4835 * grab a reference on the dataset which may cause
4836 * dmu_objset_destroy() to fail with EBUSY thus
4837 * leaving the dataset in an inconsistent state.
4840 (void) rw_wrlock(&ztest_name_lock);
4842 VERIFY(vdev_offline(spa, guid0, flags) != EBUSY);
4845 (void) rw_unlock(&ztest_name_lock);
4848 * Ideally we would like to be able to randomly
4849 * call vdev_[on|off]line without holding locks
4850 * to force unpredictable failures but the side
4851 * effects of vdev_[on|off]line prevent us from
4852 * doing so. We grab the ztest_vdev_lock here to
4853 * prevent a race between injection testing and
4856 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
4857 (void) vdev_online(spa, guid0, 0, NULL);
4858 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4866 * We have at least single-fault tolerance, so inject data corruption.
4868 fd = open(pathrand, O_RDWR);
4870 if (fd == -1) /* we hit a gap in the device namespace */
4873 fsize = lseek(fd, 0, SEEK_END);
4875 while (--iters != 0) {
4876 offset = ztest_random(fsize / (leaves << bshift)) *
4877 (leaves << bshift) + (leaf << bshift) +
4878 (ztest_random(1ULL << (bshift - 1)) & -8ULL);
4880 if (offset >= fsize)
4883 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
4884 if (mirror_save != zs->zs_mirrors) {
4885 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4890 if (pwrite(fd, &bad, sizeof (bad), offset) != sizeof (bad))
4891 fatal(1, "can't inject bad word at 0x%llx in %s",
4894 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4896 if (ztest_opts.zo_verbose >= 7)
4897 (void) printf("injected bad word into %s,"
4898 " offset 0x%llx\n", pathrand, (u_longlong_t)offset);
4905 * Verify that DDT repair works as expected.
4908 ztest_ddt_repair(ztest_ds_t *zd, uint64_t id)
4910 ztest_shared_t *zs = ztest_shared;
4911 spa_t *spa = ztest_spa;
4912 objset_t *os = zd->zd_os;
4914 uint64_t object, blocksize, txg, pattern, psize;
4915 enum zio_checksum checksum = spa_dedup_checksum(spa);
4920 int copies = 2 * ZIO_DEDUPDITTO_MIN;
4922 blocksize = ztest_random_blocksize();
4923 blocksize = MIN(blocksize, 2048); /* because we write so many */
4925 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
4927 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4931 * Take the name lock as writer to prevent anyone else from changing
4932 * the pool and dataset properies we need to maintain during this test.
4934 (void) rw_wrlock(&ztest_name_lock);
4936 if (ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_DEDUP, checksum,
4938 ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_COPIES, 1,
4940 (void) rw_unlock(&ztest_name_lock);
4944 object = od[0].od_object;
4945 blocksize = od[0].od_blocksize;
4946 pattern = zs->zs_guid ^ dmu_objset_fsid_guid(os);
4948 ASSERT(object != 0);
4950 tx = dmu_tx_create(os);
4951 dmu_tx_hold_write(tx, object, 0, copies * blocksize);
4952 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
4954 (void) rw_unlock(&ztest_name_lock);
4959 * Write all the copies of our block.
4961 for (int i = 0; i < copies; i++) {
4962 uint64_t offset = i * blocksize;
4963 VERIFY0(dmu_buf_hold(os, object, offset, FTAG, &db,
4964 DMU_READ_NO_PREFETCH));
4965 ASSERT(db->db_offset == offset);
4966 ASSERT(db->db_size == blocksize);
4967 ASSERT(ztest_pattern_match(db->db_data, db->db_size, pattern) ||
4968 ztest_pattern_match(db->db_data, db->db_size, 0ULL));
4969 dmu_buf_will_fill(db, tx);
4970 ztest_pattern_set(db->db_data, db->db_size, pattern);
4971 dmu_buf_rele(db, FTAG);
4975 txg_wait_synced(spa_get_dsl(spa), txg);
4978 * Find out what block we got.
4980 VERIFY0(dmu_buf_hold(os, object, 0, FTAG, &db,
4981 DMU_READ_NO_PREFETCH));
4982 blk = *((dmu_buf_impl_t *)db)->db_blkptr;
4983 dmu_buf_rele(db, FTAG);
4986 * Damage the block. Dedup-ditto will save us when we read it later.
4988 psize = BP_GET_PSIZE(&blk);
4989 buf = zio_buf_alloc(psize);
4990 ztest_pattern_set(buf, psize, ~pattern);
4992 (void) zio_wait(zio_rewrite(NULL, spa, 0, &blk,
4993 buf, psize, NULL, NULL, ZIO_PRIORITY_SYNC_WRITE,
4994 ZIO_FLAG_CANFAIL | ZIO_FLAG_INDUCE_DAMAGE, NULL));
4996 zio_buf_free(buf, psize);
4998 (void) rw_unlock(&ztest_name_lock);
5006 ztest_scrub(ztest_ds_t *zd, uint64_t id)
5008 spa_t *spa = ztest_spa;
5010 (void) spa_scan(spa, POOL_SCAN_SCRUB);
5011 (void) poll(NULL, 0, 100); /* wait a moment, then force a restart */
5012 (void) spa_scan(spa, POOL_SCAN_SCRUB);
5016 * Change the guid for the pool.
5020 ztest_reguid(ztest_ds_t *zd, uint64_t id)
5022 spa_t *spa = ztest_spa;
5023 uint64_t orig, load;
5026 orig = spa_guid(spa);
5027 load = spa_load_guid(spa);
5029 (void) rw_wrlock(&ztest_name_lock);
5030 error = spa_change_guid(spa);
5031 (void) rw_unlock(&ztest_name_lock);
5036 if (ztest_opts.zo_verbose >= 4) {
5037 (void) printf("Changed guid old %llu -> %llu\n",
5038 (u_longlong_t)orig, (u_longlong_t)spa_guid(spa));
5041 VERIFY3U(orig, !=, spa_guid(spa));
5042 VERIFY3U(load, ==, spa_load_guid(spa));
5046 * Rename the pool to a different name and then rename it back.
5050 ztest_spa_rename(ztest_ds_t *zd, uint64_t id)
5052 char *oldname, *newname;
5055 (void) rw_wrlock(&ztest_name_lock);
5057 oldname = ztest_opts.zo_pool;
5058 newname = umem_alloc(strlen(oldname) + 5, UMEM_NOFAIL);
5059 (void) strcpy(newname, oldname);
5060 (void) strcat(newname, "_tmp");
5065 VERIFY3U(0, ==, spa_rename(oldname, newname));
5068 * Try to open it under the old name, which shouldn't exist
5070 VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
5073 * Open it under the new name and make sure it's still the same spa_t.
5075 VERIFY3U(0, ==, spa_open(newname, &spa, FTAG));
5077 ASSERT(spa == ztest_spa);
5078 spa_close(spa, FTAG);
5081 * Rename it back to the original
5083 VERIFY3U(0, ==, spa_rename(newname, oldname));
5086 * Make sure it can still be opened
5088 VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG));
5090 ASSERT(spa == ztest_spa);
5091 spa_close(spa, FTAG);
5093 umem_free(newname, strlen(newname) + 1);
5095 (void) rw_unlock(&ztest_name_lock);
5099 * Verify pool integrity by running zdb.
5102 ztest_run_zdb(char *pool)
5105 char zdb[MAXPATHLEN + MAXNAMELEN + 20];
5113 strlcpy(zdb, "/usr/bin/ztest", sizeof(zdb));
5115 /* zdb lives in /usr/sbin, while ztest lives in /usr/bin */
5116 bin = strstr(zdb, "/usr/bin/");
5117 ztest = strstr(bin, "/ztest");
5119 isalen = ztest - isa;
5123 "/usr/sbin%.*s/zdb -bcc%s%s -U %s %s",
5126 ztest_opts.zo_verbose >= 3 ? "s" : "",
5127 ztest_opts.zo_verbose >= 4 ? "v" : "",
5132 if (ztest_opts.zo_verbose >= 5)
5133 (void) printf("Executing %s\n", strstr(zdb, "zdb "));
5135 fp = popen(zdb, "r");
5138 while (fgets(zbuf, sizeof (zbuf), fp) != NULL)
5139 if (ztest_opts.zo_verbose >= 3)
5140 (void) printf("%s", zbuf);
5142 status = pclose(fp);
5147 ztest_dump_core = 0;
5148 if (WIFEXITED(status))
5149 fatal(0, "'%s' exit code %d", zdb, WEXITSTATUS(status));
5151 fatal(0, "'%s' died with signal %d", zdb, WTERMSIG(status));
5155 ztest_walk_pool_directory(char *header)
5159 if (ztest_opts.zo_verbose >= 6)
5160 (void) printf("%s\n", header);
5162 mutex_enter(&spa_namespace_lock);
5163 while ((spa = spa_next(spa)) != NULL)
5164 if (ztest_opts.zo_verbose >= 6)
5165 (void) printf("\t%s\n", spa_name(spa));
5166 mutex_exit(&spa_namespace_lock);
5170 ztest_spa_import_export(char *oldname, char *newname)
5172 nvlist_t *config, *newconfig;
5176 if (ztest_opts.zo_verbose >= 4) {
5177 (void) printf("import/export: old = %s, new = %s\n",
5182 * Clean up from previous runs.
5184 (void) spa_destroy(newname);
5187 * Get the pool's configuration and guid.
5189 VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG));
5192 * Kick off a scrub to tickle scrub/export races.
5194 if (ztest_random(2) == 0)
5195 (void) spa_scan(spa, POOL_SCAN_SCRUB);
5197 pool_guid = spa_guid(spa);
5198 spa_close(spa, FTAG);
5200 ztest_walk_pool_directory("pools before export");
5205 VERIFY3U(0, ==, spa_export(oldname, &config, B_FALSE, B_FALSE));
5207 ztest_walk_pool_directory("pools after export");
5212 newconfig = spa_tryimport(config);
5213 ASSERT(newconfig != NULL);
5214 nvlist_free(newconfig);
5217 * Import it under the new name.
5219 VERIFY3U(0, ==, spa_import(newname, config, NULL, 0));
5221 ztest_walk_pool_directory("pools after import");
5224 * Try to import it again -- should fail with EEXIST.
5226 VERIFY3U(EEXIST, ==, spa_import(newname, config, NULL, 0));
5229 * Try to import it under a different name -- should fail with EEXIST.
5231 VERIFY3U(EEXIST, ==, spa_import(oldname, config, NULL, 0));
5234 * Verify that the pool is no longer visible under the old name.
5236 VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
5239 * Verify that we can open and close the pool using the new name.
5241 VERIFY3U(0, ==, spa_open(newname, &spa, FTAG));
5242 ASSERT(pool_guid == spa_guid(spa));
5243 spa_close(spa, FTAG);
5245 nvlist_free(config);
5249 ztest_resume(spa_t *spa)
5251 if (spa_suspended(spa) && ztest_opts.zo_verbose >= 6)
5252 (void) printf("resuming from suspended state\n");
5253 spa_vdev_state_enter(spa, SCL_NONE);
5254 vdev_clear(spa, NULL);
5255 (void) spa_vdev_state_exit(spa, NULL, 0);
5256 (void) zio_resume(spa);
5260 ztest_resume_thread(void *arg)
5264 while (!ztest_exiting) {
5265 if (spa_suspended(spa))
5267 (void) poll(NULL, 0, 100);
5273 ztest_deadman_thread(void *arg)
5275 ztest_shared_t *zs = arg;
5279 delta = (zs->zs_thread_stop - zs->zs_thread_start) / NANOSEC + grace;
5281 (void) poll(NULL, 0, (int)(1000 * delta));
5283 fatal(0, "failed to complete within %d seconds of deadline", grace);
5289 ztest_execute(int test, ztest_info_t *zi, uint64_t id)
5291 ztest_ds_t *zd = &ztest_ds[id % ztest_opts.zo_datasets];
5292 ztest_shared_callstate_t *zc = ZTEST_GET_SHARED_CALLSTATE(test);
5293 hrtime_t functime = gethrtime();
5295 for (int i = 0; i < zi->zi_iters; i++)
5296 zi->zi_func(zd, id);
5298 functime = gethrtime() - functime;
5300 atomic_add_64(&zc->zc_count, 1);
5301 atomic_add_64(&zc->zc_time, functime);
5303 if (ztest_opts.zo_verbose >= 4) {
5305 (void) dladdr((void *)zi->zi_func, &dli);
5306 (void) printf("%6.2f sec in %s\n",
5307 (double)functime / NANOSEC, dli.dli_sname);
5312 ztest_thread(void *arg)
5315 uint64_t id = (uintptr_t)arg;
5316 ztest_shared_t *zs = ztest_shared;
5320 ztest_shared_callstate_t *zc;
5322 while ((now = gethrtime()) < zs->zs_thread_stop) {
5324 * See if it's time to force a crash.
5326 if (now > zs->zs_thread_kill)
5330 * If we're getting ENOSPC with some regularity, stop.
5332 if (zs->zs_enospc_count > 10)
5336 * Pick a random function to execute.
5338 rand = ztest_random(ZTEST_FUNCS);
5339 zi = &ztest_info[rand];
5340 zc = ZTEST_GET_SHARED_CALLSTATE(rand);
5341 call_next = zc->zc_next;
5343 if (now >= call_next &&
5344 atomic_cas_64(&zc->zc_next, call_next, call_next +
5345 ztest_random(2 * zi->zi_interval[0] + 1)) == call_next) {
5346 ztest_execute(rand, zi, id);
5354 ztest_dataset_name(char *dsname, char *pool, int d)
5356 (void) snprintf(dsname, MAXNAMELEN, "%s/ds_%d", pool, d);
5360 ztest_dataset_destroy(int d)
5362 char name[MAXNAMELEN];
5364 ztest_dataset_name(name, ztest_opts.zo_pool, d);
5366 if (ztest_opts.zo_verbose >= 3)
5367 (void) printf("Destroying %s to free up space\n", name);
5370 * Cleanup any non-standard clones and snapshots. In general,
5371 * ztest thread t operates on dataset (t % zopt_datasets),
5372 * so there may be more than one thing to clean up.
5374 for (int t = d; t < ztest_opts.zo_threads;
5375 t += ztest_opts.zo_datasets) {
5376 ztest_dsl_dataset_cleanup(name, t);
5379 (void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
5380 DS_FIND_SNAPSHOTS | DS_FIND_CHILDREN);
5384 ztest_dataset_dirobj_verify(ztest_ds_t *zd)
5386 uint64_t usedobjs, dirobjs, scratch;
5389 * ZTEST_DIROBJ is the object directory for the entire dataset.
5390 * Therefore, the number of objects in use should equal the
5391 * number of ZTEST_DIROBJ entries, +1 for ZTEST_DIROBJ itself.
5392 * If not, we have an object leak.
5394 * Note that we can only check this in ztest_dataset_open(),
5395 * when the open-context and syncing-context values agree.
5396 * That's because zap_count() returns the open-context value,
5397 * while dmu_objset_space() returns the rootbp fill count.
5399 VERIFY3U(0, ==, zap_count(zd->zd_os, ZTEST_DIROBJ, &dirobjs));
5400 dmu_objset_space(zd->zd_os, &scratch, &scratch, &usedobjs, &scratch);
5401 ASSERT3U(dirobjs + 1, ==, usedobjs);
5405 ztest_dataset_open(int d)
5407 ztest_ds_t *zd = &ztest_ds[d];
5408 uint64_t committed_seq = ZTEST_GET_SHARED_DS(d)->zd_seq;
5411 char name[MAXNAMELEN];
5414 ztest_dataset_name(name, ztest_opts.zo_pool, d);
5416 (void) rw_rdlock(&ztest_name_lock);
5418 error = ztest_dataset_create(name);
5419 if (error == ENOSPC) {
5420 (void) rw_unlock(&ztest_name_lock);
5421 ztest_record_enospc(FTAG);
5424 ASSERT(error == 0 || error == EEXIST);
5426 VERIFY0(dmu_objset_hold(name, zd, &os));
5427 (void) rw_unlock(&ztest_name_lock);
5429 ztest_zd_init(zd, ZTEST_GET_SHARED_DS(d), os);
5431 zilog = zd->zd_zilog;
5433 if (zilog->zl_header->zh_claim_lr_seq != 0 &&
5434 zilog->zl_header->zh_claim_lr_seq < committed_seq)
5435 fatal(0, "missing log records: claimed %llu < committed %llu",
5436 zilog->zl_header->zh_claim_lr_seq, committed_seq);
5438 ztest_dataset_dirobj_verify(zd);
5440 zil_replay(os, zd, ztest_replay_vector);
5442 ztest_dataset_dirobj_verify(zd);
5444 if (ztest_opts.zo_verbose >= 6)
5445 (void) printf("%s replay %llu blocks, %llu records, seq %llu\n",
5447 (u_longlong_t)zilog->zl_parse_blk_count,
5448 (u_longlong_t)zilog->zl_parse_lr_count,
5449 (u_longlong_t)zilog->zl_replaying_seq);
5451 zilog = zil_open(os, ztest_get_data);
5453 if (zilog->zl_replaying_seq != 0 &&
5454 zilog->zl_replaying_seq < committed_seq)
5455 fatal(0, "missing log records: replayed %llu < committed %llu",
5456 zilog->zl_replaying_seq, committed_seq);
5462 ztest_dataset_close(int d)
5464 ztest_ds_t *zd = &ztest_ds[d];
5466 zil_close(zd->zd_zilog);
5467 dmu_objset_rele(zd->zd_os, zd);
5473 * Kick off threads to run tests on all datasets in parallel.
5476 ztest_run(ztest_shared_t *zs)
5481 thread_t resume_tid;
5484 ztest_exiting = B_FALSE;
5487 * Initialize parent/child shared state.
5489 VERIFY(_mutex_init(&ztest_vdev_lock, USYNC_THREAD, NULL) == 0);
5490 VERIFY(rwlock_init(&ztest_name_lock, USYNC_THREAD, NULL) == 0);
5492 zs->zs_thread_start = gethrtime();
5493 zs->zs_thread_stop =
5494 zs->zs_thread_start + ztest_opts.zo_passtime * NANOSEC;
5495 zs->zs_thread_stop = MIN(zs->zs_thread_stop, zs->zs_proc_stop);
5496 zs->zs_thread_kill = zs->zs_thread_stop;
5497 if (ztest_random(100) < ztest_opts.zo_killrate) {
5498 zs->zs_thread_kill -=
5499 ztest_random(ztest_opts.zo_passtime * NANOSEC);
5502 (void) _mutex_init(&zcl.zcl_callbacks_lock, USYNC_THREAD, NULL);
5504 list_create(&zcl.zcl_callbacks, sizeof (ztest_cb_data_t),
5505 offsetof(ztest_cb_data_t, zcd_node));
5510 kernel_init(FREAD | FWRITE);
5511 VERIFY(spa_open(ztest_opts.zo_pool, &spa, FTAG) == 0);
5512 spa->spa_debug = B_TRUE;
5515 VERIFY3U(0, ==, dmu_objset_hold(ztest_opts.zo_pool, FTAG, &os));
5516 zs->zs_guid = dmu_objset_fsid_guid(os);
5517 dmu_objset_rele(os, FTAG);
5519 spa->spa_dedup_ditto = 2 * ZIO_DEDUPDITTO_MIN;
5522 * We don't expect the pool to suspend unless maxfaults == 0,
5523 * in which case ztest_fault_inject() temporarily takes away
5524 * the only valid replica.
5526 if (MAXFAULTS() == 0)
5527 spa->spa_failmode = ZIO_FAILURE_MODE_WAIT;
5529 spa->spa_failmode = ZIO_FAILURE_MODE_PANIC;
5532 * Create a thread to periodically resume suspended I/O.
5534 VERIFY(thr_create(0, 0, ztest_resume_thread, spa, THR_BOUND,
5538 * Create a deadman thread to abort() if we hang.
5540 VERIFY(thr_create(0, 0, ztest_deadman_thread, zs, THR_BOUND,
5544 * Verify that we can safely inquire about about any object,
5545 * whether it's allocated or not. To make it interesting,
5546 * we probe a 5-wide window around each power of two.
5547 * This hits all edge cases, including zero and the max.
5549 for (int t = 0; t < 64; t++) {
5550 for (int d = -5; d <= 5; d++) {
5551 error = dmu_object_info(spa->spa_meta_objset,
5552 (1ULL << t) + d, NULL);
5553 ASSERT(error == 0 || error == ENOENT ||
5559 * If we got any ENOSPC errors on the previous run, destroy something.
5561 if (zs->zs_enospc_count != 0) {
5562 int d = ztest_random(ztest_opts.zo_datasets);
5563 ztest_dataset_destroy(d);
5565 zs->zs_enospc_count = 0;
5567 tid = umem_zalloc(ztest_opts.zo_threads * sizeof (thread_t),
5570 if (ztest_opts.zo_verbose >= 4)
5571 (void) printf("starting main threads...\n");
5574 * Kick off all the tests that run in parallel.
5576 for (int t = 0; t < ztest_opts.zo_threads; t++) {
5577 if (t < ztest_opts.zo_datasets &&
5578 ztest_dataset_open(t) != 0)
5580 VERIFY(thr_create(0, 0, ztest_thread, (void *)(uintptr_t)t,
5581 THR_BOUND, &tid[t]) == 0);
5585 * Wait for all of the tests to complete. We go in reverse order
5586 * so we don't close datasets while threads are still using them.
5588 for (int t = ztest_opts.zo_threads - 1; t >= 0; t--) {
5589 VERIFY(thr_join(tid[t], NULL, NULL) == 0);
5590 if (t < ztest_opts.zo_datasets)
5591 ztest_dataset_close(t);
5594 txg_wait_synced(spa_get_dsl(spa), 0);
5596 zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(spa));
5597 zs->zs_space = metaslab_class_get_space(spa_normal_class(spa));
5599 umem_free(tid, ztest_opts.zo_threads * sizeof (thread_t));
5601 /* Kill the resume thread */
5602 ztest_exiting = B_TRUE;
5603 VERIFY(thr_join(resume_tid, NULL, NULL) == 0);
5607 * Right before closing the pool, kick off a bunch of async I/O;
5608 * spa_close() should wait for it to complete.
5610 for (uint64_t object = 1; object < 50; object++)
5611 dmu_prefetch(spa->spa_meta_objset, object, 0, 1ULL << 20);
5613 spa_close(spa, FTAG);
5616 * Verify that we can loop over all pools.
5618 mutex_enter(&spa_namespace_lock);
5619 for (spa = spa_next(NULL); spa != NULL; spa = spa_next(spa))
5620 if (ztest_opts.zo_verbose > 3)
5621 (void) printf("spa_next: found %s\n", spa_name(spa));
5622 mutex_exit(&spa_namespace_lock);
5625 * Verify that we can export the pool and reimport it under a
5628 if (ztest_random(2) == 0) {
5629 char name[MAXNAMELEN];
5630 (void) snprintf(name, MAXNAMELEN, "%s_import",
5631 ztest_opts.zo_pool);
5632 ztest_spa_import_export(ztest_opts.zo_pool, name);
5633 ztest_spa_import_export(name, ztest_opts.zo_pool);
5638 list_destroy(&zcl.zcl_callbacks);
5640 (void) _mutex_destroy(&zcl.zcl_callbacks_lock);
5642 (void) rwlock_destroy(&ztest_name_lock);
5643 (void) _mutex_destroy(&ztest_vdev_lock);
5649 ztest_ds_t *zd = &ztest_ds[0];
5653 if (ztest_opts.zo_verbose >= 3)
5654 (void) printf("testing spa_freeze()...\n");
5656 kernel_init(FREAD | FWRITE);
5657 VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
5658 VERIFY3U(0, ==, ztest_dataset_open(0));
5659 spa->spa_debug = B_TRUE;
5663 * Force the first log block to be transactionally allocated.
5664 * We have to do this before we freeze the pool -- otherwise
5665 * the log chain won't be anchored.
5667 while (BP_IS_HOLE(&zd->zd_zilog->zl_header->zh_log)) {
5668 ztest_dmu_object_alloc_free(zd, 0);
5669 zil_commit(zd->zd_zilog, 0);
5672 txg_wait_synced(spa_get_dsl(spa), 0);
5675 * Freeze the pool. This stops spa_sync() from doing anything,
5676 * so that the only way to record changes from now on is the ZIL.
5681 * Run tests that generate log records but don't alter the pool config
5682 * or depend on DSL sync tasks (snapshots, objset create/destroy, etc).
5683 * We do a txg_wait_synced() after each iteration to force the txg
5684 * to increase well beyond the last synced value in the uberblock.
5685 * The ZIL should be OK with that.
5687 while (ztest_random(10) != 0 &&
5688 numloops++ < ztest_opts.zo_maxloops) {
5689 ztest_dmu_write_parallel(zd, 0);
5690 ztest_dmu_object_alloc_free(zd, 0);
5691 txg_wait_synced(spa_get_dsl(spa), 0);
5695 * Commit all of the changes we just generated.
5697 zil_commit(zd->zd_zilog, 0);
5698 txg_wait_synced(spa_get_dsl(spa), 0);
5701 * Close our dataset and close the pool.
5703 ztest_dataset_close(0);
5704 spa_close(spa, FTAG);
5708 * Open and close the pool and dataset to induce log replay.
5710 kernel_init(FREAD | FWRITE);
5711 VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
5712 ASSERT(spa_freeze_txg(spa) == UINT64_MAX);
5713 VERIFY3U(0, ==, ztest_dataset_open(0));
5714 ztest_dataset_close(0);
5716 spa->spa_debug = B_TRUE;
5718 txg_wait_synced(spa_get_dsl(spa), 0);
5719 ztest_reguid(NULL, 0);
5721 spa_close(spa, FTAG);
5726 print_time(hrtime_t t, char *timebuf)
5728 hrtime_t s = t / NANOSEC;
5729 hrtime_t m = s / 60;
5730 hrtime_t h = m / 60;
5731 hrtime_t d = h / 24;
5740 (void) sprintf(timebuf,
5741 "%llud%02lluh%02llum%02llus", d, h, m, s);
5743 (void) sprintf(timebuf, "%lluh%02llum%02llus", h, m, s);
5745 (void) sprintf(timebuf, "%llum%02llus", m, s);
5747 (void) sprintf(timebuf, "%llus", s);
5755 VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
5756 if (ztest_random(2) == 0)
5758 VERIFY(nvlist_add_uint64(props, "autoreplace", 1) == 0);
5764 * Create a storage pool with the given name and initial vdev size.
5765 * Then test spa_freeze() functionality.
5768 ztest_init(ztest_shared_t *zs)
5771 nvlist_t *nvroot, *props;
5773 VERIFY(_mutex_init(&ztest_vdev_lock, USYNC_THREAD, NULL) == 0);
5774 VERIFY(rwlock_init(&ztest_name_lock, USYNC_THREAD, NULL) == 0);
5776 kernel_init(FREAD | FWRITE);
5779 * Create the storage pool.
5781 (void) spa_destroy(ztest_opts.zo_pool);
5782 ztest_shared->zs_vdev_next_leaf = 0;
5784 zs->zs_mirrors = ztest_opts.zo_mirrors;
5785 nvroot = make_vdev_root(NULL, NULL, NULL, ztest_opts.zo_vdev_size, 0,
5786 0, ztest_opts.zo_raidz, zs->zs_mirrors, 1);
5787 props = make_random_props();
5788 for (int i = 0; i < SPA_FEATURES; i++) {
5790 (void) snprintf(buf, sizeof (buf), "feature@%s",
5791 spa_feature_table[i].fi_uname);
5792 VERIFY3U(0, ==, nvlist_add_uint64(props, buf, 0));
5794 VERIFY3U(0, ==, spa_create(ztest_opts.zo_pool, nvroot, props,
5796 nvlist_free(nvroot);
5798 VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
5799 zs->zs_metaslab_sz =
5800 1ULL << spa->spa_root_vdev->vdev_child[0]->vdev_ms_shift;
5802 spa_close(spa, FTAG);
5806 ztest_run_zdb(ztest_opts.zo_pool);
5810 ztest_run_zdb(ztest_opts.zo_pool);
5812 (void) rwlock_destroy(&ztest_name_lock);
5813 (void) _mutex_destroy(&ztest_vdev_lock);
5819 static char ztest_name_data[] = "/tmp/ztest.data.XXXXXX";
5821 ztest_fd_data = mkstemp(ztest_name_data);
5822 ASSERT3S(ztest_fd_data, >=, 0);
5823 (void) unlink(ztest_name_data);
5828 shared_data_size(ztest_shared_hdr_t *hdr)
5832 size = hdr->zh_hdr_size;
5833 size += hdr->zh_opts_size;
5834 size += hdr->zh_size;
5835 size += hdr->zh_stats_size * hdr->zh_stats_count;
5836 size += hdr->zh_ds_size * hdr->zh_ds_count;
5845 ztest_shared_hdr_t *hdr;
5847 hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
5848 PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
5849 ASSERT(hdr != MAP_FAILED);
5851 VERIFY3U(0, ==, ftruncate(ztest_fd_data, sizeof (ztest_shared_hdr_t)));
5853 hdr->zh_hdr_size = sizeof (ztest_shared_hdr_t);
5854 hdr->zh_opts_size = sizeof (ztest_shared_opts_t);
5855 hdr->zh_size = sizeof (ztest_shared_t);
5856 hdr->zh_stats_size = sizeof (ztest_shared_callstate_t);
5857 hdr->zh_stats_count = ZTEST_FUNCS;
5858 hdr->zh_ds_size = sizeof (ztest_shared_ds_t);
5859 hdr->zh_ds_count = ztest_opts.zo_datasets;
5861 size = shared_data_size(hdr);
5862 VERIFY3U(0, ==, ftruncate(ztest_fd_data, size));
5864 (void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
5871 ztest_shared_hdr_t *hdr;
5874 hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
5875 PROT_READ, MAP_SHARED, ztest_fd_data, 0);
5876 ASSERT(hdr != MAP_FAILED);
5878 size = shared_data_size(hdr);
5880 (void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
5881 hdr = ztest_shared_hdr = (void *)mmap(0, P2ROUNDUP(size, getpagesize()),
5882 PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
5883 ASSERT(hdr != MAP_FAILED);
5884 buf = (uint8_t *)hdr;
5886 offset = hdr->zh_hdr_size;
5887 ztest_shared_opts = (void *)&buf[offset];
5888 offset += hdr->zh_opts_size;
5889 ztest_shared = (void *)&buf[offset];
5890 offset += hdr->zh_size;
5891 ztest_shared_callstate = (void *)&buf[offset];
5892 offset += hdr->zh_stats_size * hdr->zh_stats_count;
5893 ztest_shared_ds = (void *)&buf[offset];
5897 exec_child(char *cmd, char *libpath, boolean_t ignorekill, int *statusp)
5901 char *cmdbuf = NULL;
5906 cmdbuf = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
5907 (void) strlcpy(cmdbuf, getexecname(), MAXPATHLEN);
5912 fatal(1, "fork failed");
5914 if (pid == 0) { /* child */
5915 char *emptyargv[2] = { cmd, NULL };
5916 char fd_data_str[12];
5918 struct rlimit rl = { 1024, 1024 };
5919 (void) setrlimit(RLIMIT_NOFILE, &rl);
5921 (void) close(ztest_fd_rand);
5923 snprintf(fd_data_str, 12, "%d", ztest_fd_data));
5924 VERIFY0(setenv("ZTEST_FD_DATA", fd_data_str, 1));
5926 (void) enable_extended_FILE_stdio(-1, -1);
5927 if (libpath != NULL)
5928 VERIFY(0 == setenv("LD_LIBRARY_PATH", libpath, 1));
5930 (void) execv(cmd, emptyargv);
5932 (void) execvp(cmd, emptyargv);
5934 ztest_dump_core = B_FALSE;
5935 fatal(B_TRUE, "exec failed: %s", cmd);
5938 if (cmdbuf != NULL) {
5939 umem_free(cmdbuf, MAXPATHLEN);
5943 while (waitpid(pid, &status, 0) != pid)
5945 if (statusp != NULL)
5948 if (WIFEXITED(status)) {
5949 if (WEXITSTATUS(status) != 0) {
5950 (void) fprintf(stderr, "child exited with code %d\n",
5951 WEXITSTATUS(status));
5955 } else if (WIFSIGNALED(status)) {
5956 if (!ignorekill || WTERMSIG(status) != SIGKILL) {
5957 (void) fprintf(stderr, "child died with signal %d\n",
5963 (void) fprintf(stderr, "something strange happened to child\n");
5970 ztest_run_init(void)
5972 ztest_shared_t *zs = ztest_shared;
5974 ASSERT(ztest_opts.zo_init != 0);
5977 * Blow away any existing copy of zpool.cache
5979 (void) remove(spa_config_path);
5982 * Create and initialize our storage pool.
5984 for (int i = 1; i <= ztest_opts.zo_init; i++) {
5985 bzero(zs, sizeof (ztest_shared_t));
5986 if (ztest_opts.zo_verbose >= 3 &&
5987 ztest_opts.zo_init != 1) {
5988 (void) printf("ztest_init(), pass %d\n", i);
5995 main(int argc, char **argv)
6003 ztest_shared_callstate_t *zc;
6009 char *fd_data_str = getenv("ZTEST_FD_DATA");
6011 (void) setvbuf(stdout, NULL, _IOLBF, 0);
6013 dprintf_setup(&argc, argv);
6015 ztest_fd_rand = open("/dev/urandom", O_RDONLY);
6016 ASSERT3S(ztest_fd_rand, >=, 0);
6019 process_options(argc, argv);
6024 bcopy(&ztest_opts, ztest_shared_opts,
6025 sizeof (*ztest_shared_opts));
6027 ztest_fd_data = atoi(fd_data_str);
6029 bcopy(ztest_shared_opts, &ztest_opts, sizeof (ztest_opts));
6031 ASSERT3U(ztest_opts.zo_datasets, ==, ztest_shared_hdr->zh_ds_count);
6033 /* Override location of zpool.cache */
6034 VERIFY3U(asprintf((char **)&spa_config_path, "%s/zpool.cache",
6035 ztest_opts.zo_dir), !=, -1);
6037 ztest_ds = umem_alloc(ztest_opts.zo_datasets * sizeof (ztest_ds_t),
6042 metaslab_gang_bang = ztest_opts.zo_metaslab_gang_bang;
6043 metaslab_df_alloc_threshold =
6044 zs->zs_metaslab_df_alloc_threshold;
6053 hasalt = (strlen(ztest_opts.zo_alt_ztest) != 0);
6055 if (ztest_opts.zo_verbose >= 1) {
6056 (void) printf("%llu vdevs, %d datasets, %d threads,"
6057 " %llu seconds...\n",
6058 (u_longlong_t)ztest_opts.zo_vdevs,
6059 ztest_opts.zo_datasets,
6060 ztest_opts.zo_threads,
6061 (u_longlong_t)ztest_opts.zo_time);
6064 cmd = umem_alloc(MAXNAMELEN, UMEM_NOFAIL);
6065 (void) strlcpy(cmd, getexecname(), MAXNAMELEN);
6067 zs->zs_do_init = B_TRUE;
6068 if (strlen(ztest_opts.zo_alt_ztest) != 0) {
6069 if (ztest_opts.zo_verbose >= 1) {
6070 (void) printf("Executing older ztest for "
6071 "initialization: %s\n", ztest_opts.zo_alt_ztest);
6073 VERIFY(!exec_child(ztest_opts.zo_alt_ztest,
6074 ztest_opts.zo_alt_libpath, B_FALSE, NULL));
6076 VERIFY(!exec_child(NULL, NULL, B_FALSE, NULL));
6078 zs->zs_do_init = B_FALSE;
6080 zs->zs_proc_start = gethrtime();
6081 zs->zs_proc_stop = zs->zs_proc_start + ztest_opts.zo_time * NANOSEC;
6083 for (int f = 0; f < ZTEST_FUNCS; f++) {
6084 zi = &ztest_info[f];
6085 zc = ZTEST_GET_SHARED_CALLSTATE(f);
6086 if (zs->zs_proc_start + zi->zi_interval[0] > zs->zs_proc_stop)
6087 zc->zc_next = UINT64_MAX;
6089 zc->zc_next = zs->zs_proc_start +
6090 ztest_random(2 * zi->zi_interval[0] + 1);
6094 * Run the tests in a loop. These tests include fault injection
6095 * to verify that self-healing data works, and forced crashes
6096 * to verify that we never lose on-disk consistency.
6098 while (gethrtime() < zs->zs_proc_stop) {
6103 * Initialize the workload counters for each function.
6105 for (int f = 0; f < ZTEST_FUNCS; f++) {
6106 zc = ZTEST_GET_SHARED_CALLSTATE(f);
6111 /* Set the allocation switch size */
6112 zs->zs_metaslab_df_alloc_threshold =
6113 ztest_random(zs->zs_metaslab_sz / 4) + 1;
6115 if (!hasalt || ztest_random(2) == 0) {
6116 if (hasalt && ztest_opts.zo_verbose >= 1) {
6117 (void) printf("Executing newer ztest: %s\n",
6121 killed = exec_child(cmd, NULL, B_TRUE, &status);
6123 if (hasalt && ztest_opts.zo_verbose >= 1) {
6124 (void) printf("Executing older ztest: %s\n",
6125 ztest_opts.zo_alt_ztest);
6128 killed = exec_child(ztest_opts.zo_alt_ztest,
6129 ztest_opts.zo_alt_libpath, B_TRUE, &status);
6136 if (ztest_opts.zo_verbose >= 1) {
6137 hrtime_t now = gethrtime();
6139 now = MIN(now, zs->zs_proc_stop);
6140 print_time(zs->zs_proc_stop - now, timebuf);
6141 nicenum(zs->zs_space, numbuf);
6143 (void) printf("Pass %3d, %8s, %3llu ENOSPC, "
6144 "%4.1f%% of %5s used, %3.0f%% done, %8s to go\n",
6146 WIFEXITED(status) ? "Complete" : "SIGKILL",
6147 (u_longlong_t)zs->zs_enospc_count,
6148 100.0 * zs->zs_alloc / zs->zs_space,
6150 100.0 * (now - zs->zs_proc_start) /
6151 (ztest_opts.zo_time * NANOSEC), timebuf);
6154 if (ztest_opts.zo_verbose >= 2) {
6155 (void) printf("\nWorkload summary:\n\n");
6156 (void) printf("%7s %9s %s\n",
6157 "Calls", "Time", "Function");
6158 (void) printf("%7s %9s %s\n",
6159 "-----", "----", "--------");
6160 for (int f = 0; f < ZTEST_FUNCS; f++) {
6163 zi = &ztest_info[f];
6164 zc = ZTEST_GET_SHARED_CALLSTATE(f);
6165 print_time(zc->zc_time, timebuf);
6166 (void) dladdr((void *)zi->zi_func, &dli);
6167 (void) printf("%7llu %9s %s\n",
6168 (u_longlong_t)zc->zc_count, timebuf,
6171 (void) printf("\n");
6175 * It's possible that we killed a child during a rename test,
6176 * in which case we'll have a 'ztest_tmp' pool lying around
6177 * instead of 'ztest'. Do a blind rename in case this happened.
6180 if (spa_open(ztest_opts.zo_pool, &spa, FTAG) == 0) {
6181 spa_close(spa, FTAG);
6183 char tmpname[MAXNAMELEN];
6185 kernel_init(FREAD | FWRITE);
6186 (void) snprintf(tmpname, sizeof (tmpname), "%s_tmp",
6187 ztest_opts.zo_pool);
6188 (void) spa_rename(tmpname, ztest_opts.zo_pool);
6192 ztest_run_zdb(ztest_opts.zo_pool);
6195 if (ztest_opts.zo_verbose >= 1) {
6197 (void) printf("%d runs of older ztest: %s\n", older,
6198 ztest_opts.zo_alt_ztest);
6199 (void) printf("%d runs of newer ztest: %s\n", newer,
6202 (void) printf("%d killed, %d completed, %.0f%% kill rate\n",
6203 kills, iters - kills, (100.0 * kills) / MAX(1, iters));
6206 umem_free(cmd, MAXNAMELEN);