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.
26 * Copyright (c) 2013 Steven Hartland. All rights reserved.
30 * The objective of this program is to provide a DMU/ZAP/SPA stress test
31 * that runs entirely in userland, is easy to use, and easy to extend.
33 * The overall design of the ztest program is as follows:
35 * (1) For each major functional area (e.g. adding vdevs to a pool,
36 * creating and destroying datasets, reading and writing objects, etc)
37 * we have a simple routine to test that functionality. These
38 * individual routines do not have to do anything "stressful".
40 * (2) We turn these simple functionality tests into a stress test by
41 * running them all in parallel, with as many threads as desired,
42 * and spread across as many datasets, objects, and vdevs as desired.
44 * (3) While all this is happening, we inject faults into the pool to
45 * verify that self-healing data really works.
47 * (4) Every time we open a dataset, we change its checksum and compression
48 * functions. Thus even individual objects vary from block to block
49 * in which checksum they use and whether they're compressed.
51 * (5) To verify that we never lose on-disk consistency after a crash,
52 * we run the entire test in a child of the main process.
53 * At random times, the child self-immolates with a SIGKILL.
54 * This is the software equivalent of pulling the power cord.
55 * The parent then runs the test again, using the existing
56 * storage pool, as many times as desired. If backwards compatability
57 * testing is enabled ztest will sometimes run the "older" version
58 * of ztest after a SIGKILL.
60 * (6) To verify that we don't have future leaks or temporal incursions,
61 * many of the functional tests record the transaction group number
62 * as part of their data. When reading old data, they verify that
63 * the transaction group number is less than the current, open txg.
64 * If you add a new test, please do this if applicable.
66 * When run with no arguments, ztest runs for about five minutes and
67 * produces no output if successful. To get a little bit of information,
68 * specify -V. To get more information, specify -VV, and so on.
70 * To turn this into an overnight stress test, use -T to specify run time.
72 * You can ask more more vdevs [-v], datasets [-d], or threads [-t]
73 * to increase the pool capacity, fanout, and overall stress level.
75 * Use the -k option to set the desired frequency of kills.
77 * When ztest invokes itself it passes all relevant information through a
78 * temporary file which is mmap-ed in the child process. This allows shared
79 * memory to survive the exec syscall. The ztest_shared_hdr_t struct is always
80 * stored at offset 0 of this file and contains information on the size and
81 * number of shared structures in the file. The information stored in this file
82 * must remain backwards compatible with older versions of ztest so that
83 * ztest can invoke them during backwards compatibility testing (-B).
86 #include <sys/zfs_context.h>
92 #include <sys/dmu_objset.h>
98 #include <sys/resource.h>
101 #include <sys/zil_impl.h>
102 #include <sys/vdev_impl.h>
103 #include <sys/vdev_file.h>
104 #include <sys/spa_impl.h>
105 #include <sys/metaslab_impl.h>
106 #include <sys/dsl_prop.h>
107 #include <sys/dsl_dataset.h>
108 #include <sys/dsl_destroy.h>
109 #include <sys/dsl_scan.h>
110 #include <sys/zio_checksum.h>
111 #include <sys/refcount.h>
112 #include <sys/zfeature.h>
113 #include <sys/dsl_userhold.h>
115 #include <stdio_ext.h>
124 #include <sys/fs/zfs.h>
125 #include <libnvpair.h>
127 static int ztest_fd_data = -1;
128 static int ztest_fd_rand = -1;
130 typedef struct ztest_shared_hdr {
131 uint64_t zh_hdr_size;
132 uint64_t zh_opts_size;
134 uint64_t zh_stats_size;
135 uint64_t zh_stats_count;
137 uint64_t zh_ds_count;
138 } ztest_shared_hdr_t;
140 static ztest_shared_hdr_t *ztest_shared_hdr;
142 typedef struct ztest_shared_opts {
143 char zo_pool[MAXNAMELEN];
144 char zo_dir[MAXNAMELEN];
145 char zo_alt_ztest[MAXNAMELEN];
146 char zo_alt_libpath[MAXNAMELEN];
148 uint64_t zo_vdevtime;
156 uint64_t zo_passtime;
157 uint64_t zo_killrate;
161 uint64_t zo_maxloops;
162 uint64_t zo_metaslab_gang_bang;
163 } ztest_shared_opts_t;
165 static const ztest_shared_opts_t ztest_opts_defaults = {
166 .zo_pool = { 'z', 't', 'e', 's', 't', '\0' },
167 .zo_dir = { '/', 't', 'm', 'p', '\0' },
168 .zo_alt_ztest = { '\0' },
169 .zo_alt_libpath = { '\0' },
171 .zo_ashift = SPA_MINBLOCKSHIFT,
174 .zo_raidz_parity = 1,
175 .zo_vdev_size = SPA_MINDEVSIZE,
178 .zo_passtime = 60, /* 60 seconds */
179 .zo_killrate = 70, /* 70% kill rate */
182 .zo_time = 300, /* 5 minutes */
183 .zo_maxloops = 50, /* max loops during spa_freeze() */
184 .zo_metaslab_gang_bang = 32 << 10
187 extern uint64_t metaslab_gang_bang;
188 extern uint64_t metaslab_df_alloc_threshold;
189 extern uint64_t zfs_deadman_synctime;
191 static ztest_shared_opts_t *ztest_shared_opts;
192 static ztest_shared_opts_t ztest_opts;
194 typedef struct ztest_shared_ds {
198 static ztest_shared_ds_t *ztest_shared_ds;
199 #define ZTEST_GET_SHARED_DS(d) (&ztest_shared_ds[d])
201 #define BT_MAGIC 0x123456789abcdefULL
202 #define MAXFAULTS() \
203 (MAX(zs->zs_mirrors, 1) * (ztest_opts.zo_raidz_parity + 1) - 1)
207 ZTEST_IO_WRITE_PATTERN,
208 ZTEST_IO_WRITE_ZEROES,
215 typedef struct ztest_block_tag {
225 typedef struct bufwad {
232 * XXX -- fix zfs range locks to be generic so we can use them here.
254 #define ZTEST_RANGE_LOCKS 64
255 #define ZTEST_OBJECT_LOCKS 64
258 * Object descriptor. Used as a template for object lookup/create/remove.
260 typedef struct ztest_od {
263 dmu_object_type_t od_type;
264 dmu_object_type_t od_crtype;
265 uint64_t od_blocksize;
266 uint64_t od_crblocksize;
269 char od_name[MAXNAMELEN];
275 typedef struct ztest_ds {
276 ztest_shared_ds_t *zd_shared;
278 rwlock_t zd_zilog_lock;
280 ztest_od_t *zd_od; /* debugging aid */
281 char zd_name[MAXNAMELEN];
282 mutex_t zd_dirobj_lock;
283 rll_t zd_object_lock[ZTEST_OBJECT_LOCKS];
284 rll_t zd_range_lock[ZTEST_RANGE_LOCKS];
288 * Per-iteration state.
290 typedef void ztest_func_t(ztest_ds_t *zd, uint64_t id);
292 typedef struct ztest_info {
293 ztest_func_t *zi_func; /* test function */
294 uint64_t zi_iters; /* iterations per execution */
295 uint64_t *zi_interval; /* execute every <interval> seconds */
298 typedef struct ztest_shared_callstate {
299 uint64_t zc_count; /* per-pass count */
300 uint64_t zc_time; /* per-pass time */
301 uint64_t zc_next; /* next time to call this function */
302 } ztest_shared_callstate_t;
304 static ztest_shared_callstate_t *ztest_shared_callstate;
305 #define ZTEST_GET_SHARED_CALLSTATE(c) (&ztest_shared_callstate[c])
308 * Note: these aren't static because we want dladdr() to work.
310 ztest_func_t ztest_dmu_read_write;
311 ztest_func_t ztest_dmu_write_parallel;
312 ztest_func_t ztest_dmu_object_alloc_free;
313 ztest_func_t ztest_dmu_commit_callbacks;
314 ztest_func_t ztest_zap;
315 ztest_func_t ztest_zap_parallel;
316 ztest_func_t ztest_zil_commit;
317 ztest_func_t ztest_zil_remount;
318 ztest_func_t ztest_dmu_read_write_zcopy;
319 ztest_func_t ztest_dmu_objset_create_destroy;
320 ztest_func_t ztest_dmu_prealloc;
321 ztest_func_t ztest_fzap;
322 ztest_func_t ztest_dmu_snapshot_create_destroy;
323 ztest_func_t ztest_dsl_prop_get_set;
324 ztest_func_t ztest_spa_prop_get_set;
325 ztest_func_t ztest_spa_create_destroy;
326 ztest_func_t ztest_fault_inject;
327 ztest_func_t ztest_ddt_repair;
328 ztest_func_t ztest_dmu_snapshot_hold;
329 ztest_func_t ztest_spa_rename;
330 ztest_func_t ztest_scrub;
331 ztest_func_t ztest_dsl_dataset_promote_busy;
332 ztest_func_t ztest_vdev_attach_detach;
333 ztest_func_t ztest_vdev_LUN_growth;
334 ztest_func_t ztest_vdev_add_remove;
335 ztest_func_t ztest_vdev_aux_add_remove;
336 ztest_func_t ztest_split_pool;
337 ztest_func_t ztest_reguid;
338 ztest_func_t ztest_spa_upgrade;
340 uint64_t zopt_always = 0ULL * NANOSEC; /* all the time */
341 uint64_t zopt_incessant = 1ULL * NANOSEC / 10; /* every 1/10 second */
342 uint64_t zopt_often = 1ULL * NANOSEC; /* every second */
343 uint64_t zopt_sometimes = 10ULL * NANOSEC; /* every 10 seconds */
344 uint64_t zopt_rarely = 60ULL * NANOSEC; /* every 60 seconds */
346 ztest_info_t ztest_info[] = {
347 { ztest_dmu_read_write, 1, &zopt_always },
348 { ztest_dmu_write_parallel, 10, &zopt_always },
349 { ztest_dmu_object_alloc_free, 1, &zopt_always },
350 { ztest_dmu_commit_callbacks, 1, &zopt_always },
351 { ztest_zap, 30, &zopt_always },
352 { ztest_zap_parallel, 100, &zopt_always },
353 { ztest_split_pool, 1, &zopt_always },
354 { ztest_zil_commit, 1, &zopt_incessant },
355 { ztest_zil_remount, 1, &zopt_sometimes },
356 { ztest_dmu_read_write_zcopy, 1, &zopt_often },
357 { ztest_dmu_objset_create_destroy, 1, &zopt_often },
358 { ztest_dsl_prop_get_set, 1, &zopt_often },
359 { ztest_spa_prop_get_set, 1, &zopt_sometimes },
361 { ztest_dmu_prealloc, 1, &zopt_sometimes },
363 { ztest_fzap, 1, &zopt_sometimes },
364 { ztest_dmu_snapshot_create_destroy, 1, &zopt_sometimes },
365 { ztest_spa_create_destroy, 1, &zopt_sometimes },
366 { ztest_fault_inject, 1, &zopt_sometimes },
367 { ztest_ddt_repair, 1, &zopt_sometimes },
368 { ztest_dmu_snapshot_hold, 1, &zopt_sometimes },
369 { ztest_reguid, 1, &zopt_rarely },
370 { ztest_spa_rename, 1, &zopt_rarely },
371 { ztest_scrub, 1, &zopt_rarely },
372 { ztest_spa_upgrade, 1, &zopt_rarely },
373 { ztest_dsl_dataset_promote_busy, 1, &zopt_rarely },
374 { ztest_vdev_attach_detach, 1, &zopt_sometimes },
375 { ztest_vdev_LUN_growth, 1, &zopt_rarely },
376 { ztest_vdev_add_remove, 1,
377 &ztest_opts.zo_vdevtime },
378 { ztest_vdev_aux_add_remove, 1,
379 &ztest_opts.zo_vdevtime },
382 #define ZTEST_FUNCS (sizeof (ztest_info) / sizeof (ztest_info_t))
385 * The following struct is used to hold a list of uncalled commit callbacks.
386 * The callbacks are ordered by txg number.
388 typedef struct ztest_cb_list {
389 mutex_t zcl_callbacks_lock;
390 list_t zcl_callbacks;
394 * Stuff we need to share writably between parent and child.
396 typedef struct ztest_shared {
397 boolean_t zs_do_init;
398 hrtime_t zs_proc_start;
399 hrtime_t zs_proc_stop;
400 hrtime_t zs_thread_start;
401 hrtime_t zs_thread_stop;
402 hrtime_t zs_thread_kill;
403 uint64_t zs_enospc_count;
404 uint64_t zs_vdev_next_leaf;
405 uint64_t zs_vdev_aux;
410 uint64_t zs_metaslab_sz;
411 uint64_t zs_metaslab_df_alloc_threshold;
415 #define ID_PARALLEL -1ULL
417 static char ztest_dev_template[] = "%s/%s.%llua";
418 static char ztest_aux_template[] = "%s/%s.%s.%llu";
419 ztest_shared_t *ztest_shared;
421 static spa_t *ztest_spa = NULL;
422 static ztest_ds_t *ztest_ds;
424 static mutex_t ztest_vdev_lock;
427 * The ztest_name_lock protects the pool and dataset namespace used by
428 * the individual tests. To modify the namespace, consumers must grab
429 * this lock as writer. Grabbing the lock as reader will ensure that the
430 * namespace does not change while the lock is held.
432 static rwlock_t ztest_name_lock;
434 static boolean_t ztest_dump_core = B_TRUE;
435 static boolean_t ztest_exiting;
437 /* Global commit callback list */
438 static ztest_cb_list_t zcl;
441 ZTEST_META_DNODE = 0,
446 static void usage(boolean_t) __NORETURN;
449 * These libumem hooks provide a reasonable set of defaults for the allocator's
450 * debugging facilities.
455 return ("default,verbose"); /* $UMEM_DEBUG setting */
459 _umem_logging_init(void)
461 return ("fail,contents"); /* $UMEM_LOGGING setting */
464 #define FATAL_MSG_SZ 1024
469 fatal(int do_perror, char *message, ...)
472 int save_errno = errno;
473 char buf[FATAL_MSG_SZ];
475 (void) fflush(stdout);
477 va_start(args, message);
478 (void) sprintf(buf, "ztest: ");
480 (void) vsprintf(buf + strlen(buf), message, args);
483 (void) snprintf(buf + strlen(buf), FATAL_MSG_SZ - strlen(buf),
484 ": %s", strerror(save_errno));
486 (void) fprintf(stderr, "%s\n", buf);
487 fatal_msg = buf; /* to ease debugging */
494 str2shift(const char *buf)
496 const char *ends = "BKMGTPEZ";
501 for (i = 0; i < strlen(ends); i++) {
502 if (toupper(buf[0]) == ends[i])
505 if (i == strlen(ends)) {
506 (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n",
510 if (buf[1] == '\0' || (toupper(buf[1]) == 'B' && buf[2] == '\0')) {
513 (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n", buf);
519 nicenumtoull(const char *buf)
524 val = strtoull(buf, &end, 0);
526 (void) fprintf(stderr, "ztest: bad numeric value: %s\n", buf);
528 } else if (end[0] == '.') {
529 double fval = strtod(buf, &end);
530 fval *= pow(2, str2shift(end));
531 if (fval > UINT64_MAX) {
532 (void) fprintf(stderr, "ztest: value too large: %s\n",
536 val = (uint64_t)fval;
538 int shift = str2shift(end);
539 if (shift >= 64 || (val << shift) >> shift != val) {
540 (void) fprintf(stderr, "ztest: value too large: %s\n",
550 usage(boolean_t requested)
552 const ztest_shared_opts_t *zo = &ztest_opts_defaults;
554 char nice_vdev_size[10];
555 char nice_gang_bang[10];
556 FILE *fp = requested ? stdout : stderr;
558 nicenum(zo->zo_vdev_size, nice_vdev_size);
559 nicenum(zo->zo_metaslab_gang_bang, nice_gang_bang);
561 (void) fprintf(fp, "Usage: %s\n"
562 "\t[-v vdevs (default: %llu)]\n"
563 "\t[-s size_of_each_vdev (default: %s)]\n"
564 "\t[-a alignment_shift (default: %d)] use 0 for random\n"
565 "\t[-m mirror_copies (default: %d)]\n"
566 "\t[-r raidz_disks (default: %d)]\n"
567 "\t[-R raidz_parity (default: %d)]\n"
568 "\t[-d datasets (default: %d)]\n"
569 "\t[-t threads (default: %d)]\n"
570 "\t[-g gang_block_threshold (default: %s)]\n"
571 "\t[-i init_count (default: %d)] initialize pool i times\n"
572 "\t[-k kill_percentage (default: %llu%%)]\n"
573 "\t[-p pool_name (default: %s)]\n"
574 "\t[-f dir (default: %s)] file directory for vdev files\n"
575 "\t[-V] verbose (use multiple times for ever more blather)\n"
576 "\t[-E] use existing pool instead of creating new one\n"
577 "\t[-T time (default: %llu sec)] total run time\n"
578 "\t[-F freezeloops (default: %llu)] max loops in spa_freeze()\n"
579 "\t[-P passtime (default: %llu sec)] time per pass\n"
580 "\t[-B alt_ztest (default: <none>)] alternate ztest path\n"
581 "\t[-h] (print help)\n"
584 (u_longlong_t)zo->zo_vdevs, /* -v */
585 nice_vdev_size, /* -s */
586 zo->zo_ashift, /* -a */
587 zo->zo_mirrors, /* -m */
588 zo->zo_raidz, /* -r */
589 zo->zo_raidz_parity, /* -R */
590 zo->zo_datasets, /* -d */
591 zo->zo_threads, /* -t */
592 nice_gang_bang, /* -g */
593 zo->zo_init, /* -i */
594 (u_longlong_t)zo->zo_killrate, /* -k */
595 zo->zo_pool, /* -p */
597 (u_longlong_t)zo->zo_time, /* -T */
598 (u_longlong_t)zo->zo_maxloops, /* -F */
599 (u_longlong_t)zo->zo_passtime);
600 exit(requested ? 0 : 1);
604 process_options(int argc, char **argv)
607 ztest_shared_opts_t *zo = &ztest_opts;
611 char altdir[MAXNAMELEN] = { 0 };
613 bcopy(&ztest_opts_defaults, zo, sizeof (*zo));
615 while ((opt = getopt(argc, argv,
616 "v:s:a:m:r:R:d:t:g:i:k:p:f:VET:P:hF:B:")) != EOF) {
633 value = nicenumtoull(optarg);
637 zo->zo_vdevs = value;
640 zo->zo_vdev_size = MAX(SPA_MINDEVSIZE, value);
643 zo->zo_ashift = value;
646 zo->zo_mirrors = value;
649 zo->zo_raidz = MAX(1, value);
652 zo->zo_raidz_parity = MIN(MAX(value, 1), 3);
655 zo->zo_datasets = MAX(1, value);
658 zo->zo_threads = MAX(1, value);
661 zo->zo_metaslab_gang_bang = MAX(SPA_MINBLOCKSIZE << 1,
668 zo->zo_killrate = value;
671 (void) strlcpy(zo->zo_pool, optarg,
672 sizeof (zo->zo_pool));
675 path = realpath(optarg, NULL);
677 (void) fprintf(stderr, "error: %s: %s\n",
678 optarg, strerror(errno));
681 (void) strlcpy(zo->zo_dir, path,
682 sizeof (zo->zo_dir));
695 zo->zo_passtime = MAX(1, value);
698 zo->zo_maxloops = MAX(1, value);
701 (void) strlcpy(altdir, optarg, sizeof (altdir));
713 zo->zo_raidz_parity = MIN(zo->zo_raidz_parity, zo->zo_raidz - 1);
716 (zo->zo_vdevs > 0 ? zo->zo_time * NANOSEC / zo->zo_vdevs :
719 if (strlen(altdir) > 0) {
727 cmd = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
728 realaltdir = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
730 VERIFY(NULL != realpath(getexecname(), cmd));
731 if (0 != access(altdir, F_OK)) {
732 ztest_dump_core = B_FALSE;
733 fatal(B_TRUE, "invalid alternate ztest path: %s",
736 VERIFY(NULL != realpath(altdir, realaltdir));
739 * 'cmd' should be of the form "<anything>/usr/bin/<isa>/ztest".
740 * We want to extract <isa> to determine if we should use
741 * 32 or 64 bit binaries.
743 bin = strstr(cmd, "/usr/bin/");
744 ztest = strstr(bin, "/ztest");
746 isalen = ztest - isa;
747 (void) snprintf(zo->zo_alt_ztest, sizeof (zo->zo_alt_ztest),
748 "%s/usr/bin/%.*s/ztest", realaltdir, isalen, isa);
749 (void) snprintf(zo->zo_alt_libpath, sizeof (zo->zo_alt_libpath),
750 "%s/usr/lib/%.*s", realaltdir, isalen, isa);
752 if (0 != access(zo->zo_alt_ztest, X_OK)) {
753 ztest_dump_core = B_FALSE;
754 fatal(B_TRUE, "invalid alternate ztest: %s",
756 } else if (0 != access(zo->zo_alt_libpath, X_OK)) {
757 ztest_dump_core = B_FALSE;
758 fatal(B_TRUE, "invalid alternate lib directory %s",
762 umem_free(cmd, MAXPATHLEN);
763 umem_free(realaltdir, MAXPATHLEN);
768 ztest_kill(ztest_shared_t *zs)
770 zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(ztest_spa));
771 zs->zs_space = metaslab_class_get_space(spa_normal_class(ztest_spa));
772 (void) kill(getpid(), SIGKILL);
776 ztest_random(uint64_t range)
780 ASSERT3S(ztest_fd_rand, >=, 0);
785 if (read(ztest_fd_rand, &r, sizeof (r)) != sizeof (r))
786 fatal(1, "short read from /dev/urandom");
793 ztest_record_enospc(const char *s)
795 ztest_shared->zs_enospc_count++;
799 ztest_get_ashift(void)
801 if (ztest_opts.zo_ashift == 0)
802 return (SPA_MINBLOCKSHIFT + ztest_random(3));
803 return (ztest_opts.zo_ashift);
807 make_vdev_file(char *path, char *aux, char *pool, size_t size, uint64_t ashift)
809 char pathbuf[MAXPATHLEN];
814 ashift = ztest_get_ashift();
820 vdev = ztest_shared->zs_vdev_aux;
821 (void) snprintf(path, sizeof (pathbuf),
822 ztest_aux_template, ztest_opts.zo_dir,
823 pool == NULL ? ztest_opts.zo_pool : pool,
826 vdev = ztest_shared->zs_vdev_next_leaf++;
827 (void) snprintf(path, sizeof (pathbuf),
828 ztest_dev_template, ztest_opts.zo_dir,
829 pool == NULL ? ztest_opts.zo_pool : pool, vdev);
834 int fd = open(path, O_RDWR | O_CREAT | O_TRUNC, 0666);
836 fatal(1, "can't open %s", path);
837 if (ftruncate(fd, size) != 0)
838 fatal(1, "can't ftruncate %s", path);
842 VERIFY(nvlist_alloc(&file, NV_UNIQUE_NAME, 0) == 0);
843 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_TYPE, VDEV_TYPE_FILE) == 0);
844 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_PATH, path) == 0);
845 VERIFY(nvlist_add_uint64(file, ZPOOL_CONFIG_ASHIFT, ashift) == 0);
851 make_vdev_raidz(char *path, char *aux, char *pool, size_t size,
852 uint64_t ashift, int r)
854 nvlist_t *raidz, **child;
858 return (make_vdev_file(path, aux, pool, size, ashift));
859 child = umem_alloc(r * sizeof (nvlist_t *), UMEM_NOFAIL);
861 for (c = 0; c < r; c++)
862 child[c] = make_vdev_file(path, aux, pool, size, ashift);
864 VERIFY(nvlist_alloc(&raidz, NV_UNIQUE_NAME, 0) == 0);
865 VERIFY(nvlist_add_string(raidz, ZPOOL_CONFIG_TYPE,
866 VDEV_TYPE_RAIDZ) == 0);
867 VERIFY(nvlist_add_uint64(raidz, ZPOOL_CONFIG_NPARITY,
868 ztest_opts.zo_raidz_parity) == 0);
869 VERIFY(nvlist_add_nvlist_array(raidz, ZPOOL_CONFIG_CHILDREN,
872 for (c = 0; c < r; c++)
873 nvlist_free(child[c]);
875 umem_free(child, r * sizeof (nvlist_t *));
881 make_vdev_mirror(char *path, char *aux, char *pool, size_t size,
882 uint64_t ashift, int r, int m)
884 nvlist_t *mirror, **child;
888 return (make_vdev_raidz(path, aux, pool, size, ashift, r));
890 child = umem_alloc(m * sizeof (nvlist_t *), UMEM_NOFAIL);
892 for (c = 0; c < m; c++)
893 child[c] = make_vdev_raidz(path, aux, pool, size, ashift, r);
895 VERIFY(nvlist_alloc(&mirror, NV_UNIQUE_NAME, 0) == 0);
896 VERIFY(nvlist_add_string(mirror, ZPOOL_CONFIG_TYPE,
897 VDEV_TYPE_MIRROR) == 0);
898 VERIFY(nvlist_add_nvlist_array(mirror, ZPOOL_CONFIG_CHILDREN,
901 for (c = 0; c < m; c++)
902 nvlist_free(child[c]);
904 umem_free(child, m * sizeof (nvlist_t *));
910 make_vdev_root(char *path, char *aux, char *pool, size_t size, uint64_t ashift,
911 int log, int r, int m, int t)
913 nvlist_t *root, **child;
918 child = umem_alloc(t * sizeof (nvlist_t *), UMEM_NOFAIL);
920 for (c = 0; c < t; c++) {
921 child[c] = make_vdev_mirror(path, aux, pool, size, ashift,
923 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_IS_LOG,
927 VERIFY(nvlist_alloc(&root, NV_UNIQUE_NAME, 0) == 0);
928 VERIFY(nvlist_add_string(root, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) == 0);
929 VERIFY(nvlist_add_nvlist_array(root, aux ? aux : ZPOOL_CONFIG_CHILDREN,
932 for (c = 0; c < t; c++)
933 nvlist_free(child[c]);
935 umem_free(child, t * sizeof (nvlist_t *));
941 * Find a random spa version. Returns back a random spa version in the
942 * range [initial_version, SPA_VERSION_FEATURES].
945 ztest_random_spa_version(uint64_t initial_version)
947 uint64_t version = initial_version;
949 if (version <= SPA_VERSION_BEFORE_FEATURES) {
951 ztest_random(SPA_VERSION_BEFORE_FEATURES - version + 1);
954 if (version > SPA_VERSION_BEFORE_FEATURES)
955 version = SPA_VERSION_FEATURES;
957 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
962 ztest_random_blocksize(void)
964 return (1 << (SPA_MINBLOCKSHIFT +
965 ztest_random(SPA_MAXBLOCKSHIFT - SPA_MINBLOCKSHIFT + 1)));
969 ztest_random_ibshift(void)
971 return (DN_MIN_INDBLKSHIFT +
972 ztest_random(DN_MAX_INDBLKSHIFT - DN_MIN_INDBLKSHIFT + 1));
976 ztest_random_vdev_top(spa_t *spa, boolean_t log_ok)
979 vdev_t *rvd = spa->spa_root_vdev;
982 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
985 top = ztest_random(rvd->vdev_children);
986 tvd = rvd->vdev_child[top];
987 } while (tvd->vdev_ishole || (tvd->vdev_islog && !log_ok) ||
988 tvd->vdev_mg == NULL || tvd->vdev_mg->mg_class == NULL);
994 ztest_random_dsl_prop(zfs_prop_t prop)
999 value = zfs_prop_random_value(prop, ztest_random(-1ULL));
1000 } while (prop == ZFS_PROP_CHECKSUM && value == ZIO_CHECKSUM_OFF);
1006 ztest_dsl_prop_set_uint64(char *osname, zfs_prop_t prop, uint64_t value,
1009 const char *propname = zfs_prop_to_name(prop);
1010 const char *valname;
1011 char setpoint[MAXPATHLEN];
1015 error = dsl_prop_set_int(osname, propname,
1016 (inherit ? ZPROP_SRC_NONE : ZPROP_SRC_LOCAL), value);
1018 if (error == ENOSPC) {
1019 ztest_record_enospc(FTAG);
1024 VERIFY0(dsl_prop_get_integer(osname, propname, &curval, setpoint));
1026 if (ztest_opts.zo_verbose >= 6) {
1027 VERIFY(zfs_prop_index_to_string(prop, curval, &valname) == 0);
1028 (void) printf("%s %s = %s at '%s'\n",
1029 osname, propname, valname, setpoint);
1036 ztest_spa_prop_set_uint64(zpool_prop_t prop, uint64_t value)
1038 spa_t *spa = ztest_spa;
1039 nvlist_t *props = NULL;
1042 VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
1043 VERIFY(nvlist_add_uint64(props, zpool_prop_to_name(prop), value) == 0);
1045 error = spa_prop_set(spa, props);
1049 if (error == ENOSPC) {
1050 ztest_record_enospc(FTAG);
1059 ztest_rll_init(rll_t *rll)
1061 rll->rll_writer = NULL;
1062 rll->rll_readers = 0;
1063 VERIFY(_mutex_init(&rll->rll_lock, USYNC_THREAD, NULL) == 0);
1064 VERIFY(cond_init(&rll->rll_cv, USYNC_THREAD, NULL) == 0);
1068 ztest_rll_destroy(rll_t *rll)
1070 ASSERT(rll->rll_writer == NULL);
1071 ASSERT(rll->rll_readers == 0);
1072 VERIFY(_mutex_destroy(&rll->rll_lock) == 0);
1073 VERIFY(cond_destroy(&rll->rll_cv) == 0);
1077 ztest_rll_lock(rll_t *rll, rl_type_t type)
1079 VERIFY(mutex_lock(&rll->rll_lock) == 0);
1081 if (type == RL_READER) {
1082 while (rll->rll_writer != NULL)
1083 (void) cond_wait(&rll->rll_cv, &rll->rll_lock);
1086 while (rll->rll_writer != NULL || rll->rll_readers)
1087 (void) cond_wait(&rll->rll_cv, &rll->rll_lock);
1088 rll->rll_writer = curthread;
1091 VERIFY(mutex_unlock(&rll->rll_lock) == 0);
1095 ztest_rll_unlock(rll_t *rll)
1097 VERIFY(mutex_lock(&rll->rll_lock) == 0);
1099 if (rll->rll_writer) {
1100 ASSERT(rll->rll_readers == 0);
1101 rll->rll_writer = NULL;
1103 ASSERT(rll->rll_readers != 0);
1104 ASSERT(rll->rll_writer == NULL);
1108 if (rll->rll_writer == NULL && rll->rll_readers == 0)
1109 VERIFY(cond_broadcast(&rll->rll_cv) == 0);
1111 VERIFY(mutex_unlock(&rll->rll_lock) == 0);
1115 ztest_object_lock(ztest_ds_t *zd, uint64_t object, rl_type_t type)
1117 rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1119 ztest_rll_lock(rll, type);
1123 ztest_object_unlock(ztest_ds_t *zd, uint64_t object)
1125 rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1127 ztest_rll_unlock(rll);
1131 ztest_range_lock(ztest_ds_t *zd, uint64_t object, uint64_t offset,
1132 uint64_t size, rl_type_t type)
1134 uint64_t hash = object ^ (offset % (ZTEST_RANGE_LOCKS + 1));
1135 rll_t *rll = &zd->zd_range_lock[hash & (ZTEST_RANGE_LOCKS - 1)];
1138 rl = umem_alloc(sizeof (*rl), UMEM_NOFAIL);
1139 rl->rl_object = object;
1140 rl->rl_offset = offset;
1144 ztest_rll_lock(rll, type);
1150 ztest_range_unlock(rl_t *rl)
1152 rll_t *rll = rl->rl_lock;
1154 ztest_rll_unlock(rll);
1156 umem_free(rl, sizeof (*rl));
1160 ztest_zd_init(ztest_ds_t *zd, ztest_shared_ds_t *szd, objset_t *os)
1163 zd->zd_zilog = dmu_objset_zil(os);
1164 zd->zd_shared = szd;
1165 dmu_objset_name(os, zd->zd_name);
1167 if (zd->zd_shared != NULL)
1168 zd->zd_shared->zd_seq = 0;
1170 VERIFY(rwlock_init(&zd->zd_zilog_lock, USYNC_THREAD, NULL) == 0);
1171 VERIFY(_mutex_init(&zd->zd_dirobj_lock, USYNC_THREAD, NULL) == 0);
1173 for (int l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1174 ztest_rll_init(&zd->zd_object_lock[l]);
1176 for (int l = 0; l < ZTEST_RANGE_LOCKS; l++)
1177 ztest_rll_init(&zd->zd_range_lock[l]);
1181 ztest_zd_fini(ztest_ds_t *zd)
1183 VERIFY(_mutex_destroy(&zd->zd_dirobj_lock) == 0);
1185 for (int l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1186 ztest_rll_destroy(&zd->zd_object_lock[l]);
1188 for (int l = 0; l < ZTEST_RANGE_LOCKS; l++)
1189 ztest_rll_destroy(&zd->zd_range_lock[l]);
1192 #define TXG_MIGHTWAIT (ztest_random(10) == 0 ? TXG_NOWAIT : TXG_WAIT)
1195 ztest_tx_assign(dmu_tx_t *tx, uint64_t txg_how, const char *tag)
1201 * Attempt to assign tx to some transaction group.
1203 error = dmu_tx_assign(tx, txg_how);
1205 if (error == ERESTART) {
1206 ASSERT(txg_how == TXG_NOWAIT);
1209 ASSERT3U(error, ==, ENOSPC);
1210 ztest_record_enospc(tag);
1215 txg = dmu_tx_get_txg(tx);
1221 ztest_pattern_set(void *buf, uint64_t size, uint64_t value)
1224 uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1231 ztest_pattern_match(void *buf, uint64_t size, uint64_t value)
1234 uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1238 diff |= (value - *ip++);
1244 ztest_bt_generate(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1245 uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg)
1247 bt->bt_magic = BT_MAGIC;
1248 bt->bt_objset = dmu_objset_id(os);
1249 bt->bt_object = object;
1250 bt->bt_offset = offset;
1253 bt->bt_crtxg = crtxg;
1257 ztest_bt_verify(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1258 uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg)
1260 ASSERT(bt->bt_magic == BT_MAGIC);
1261 ASSERT(bt->bt_objset == dmu_objset_id(os));
1262 ASSERT(bt->bt_object == object);
1263 ASSERT(bt->bt_offset == offset);
1264 ASSERT(bt->bt_gen <= gen);
1265 ASSERT(bt->bt_txg <= txg);
1266 ASSERT(bt->bt_crtxg == crtxg);
1269 static ztest_block_tag_t *
1270 ztest_bt_bonus(dmu_buf_t *db)
1272 dmu_object_info_t doi;
1273 ztest_block_tag_t *bt;
1275 dmu_object_info_from_db(db, &doi);
1276 ASSERT3U(doi.doi_bonus_size, <=, db->db_size);
1277 ASSERT3U(doi.doi_bonus_size, >=, sizeof (*bt));
1278 bt = (void *)((char *)db->db_data + doi.doi_bonus_size - sizeof (*bt));
1287 #define lrz_type lr_mode
1288 #define lrz_blocksize lr_uid
1289 #define lrz_ibshift lr_gid
1290 #define lrz_bonustype lr_rdev
1291 #define lrz_bonuslen lr_crtime[1]
1294 ztest_log_create(ztest_ds_t *zd, dmu_tx_t *tx, lr_create_t *lr)
1296 char *name = (void *)(lr + 1); /* name follows lr */
1297 size_t namesize = strlen(name) + 1;
1300 if (zil_replaying(zd->zd_zilog, tx))
1303 itx = zil_itx_create(TX_CREATE, sizeof (*lr) + namesize);
1304 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1305 sizeof (*lr) + namesize - sizeof (lr_t));
1307 zil_itx_assign(zd->zd_zilog, itx, tx);
1311 ztest_log_remove(ztest_ds_t *zd, dmu_tx_t *tx, lr_remove_t *lr, uint64_t object)
1313 char *name = (void *)(lr + 1); /* name follows lr */
1314 size_t namesize = strlen(name) + 1;
1317 if (zil_replaying(zd->zd_zilog, tx))
1320 itx = zil_itx_create(TX_REMOVE, sizeof (*lr) + namesize);
1321 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1322 sizeof (*lr) + namesize - sizeof (lr_t));
1324 itx->itx_oid = object;
1325 zil_itx_assign(zd->zd_zilog, itx, tx);
1329 ztest_log_write(ztest_ds_t *zd, dmu_tx_t *tx, lr_write_t *lr)
1332 itx_wr_state_t write_state = ztest_random(WR_NUM_STATES);
1334 if (zil_replaying(zd->zd_zilog, tx))
1337 if (lr->lr_length > ZIL_MAX_LOG_DATA)
1338 write_state = WR_INDIRECT;
1340 itx = zil_itx_create(TX_WRITE,
1341 sizeof (*lr) + (write_state == WR_COPIED ? lr->lr_length : 0));
1343 if (write_state == WR_COPIED &&
1344 dmu_read(zd->zd_os, lr->lr_foid, lr->lr_offset, lr->lr_length,
1345 ((lr_write_t *)&itx->itx_lr) + 1, DMU_READ_NO_PREFETCH) != 0) {
1346 zil_itx_destroy(itx);
1347 itx = zil_itx_create(TX_WRITE, sizeof (*lr));
1348 write_state = WR_NEED_COPY;
1350 itx->itx_private = zd;
1351 itx->itx_wr_state = write_state;
1352 itx->itx_sync = (ztest_random(8) == 0);
1353 itx->itx_sod += (write_state == WR_NEED_COPY ? lr->lr_length : 0);
1355 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1356 sizeof (*lr) - sizeof (lr_t));
1358 zil_itx_assign(zd->zd_zilog, itx, tx);
1362 ztest_log_truncate(ztest_ds_t *zd, dmu_tx_t *tx, lr_truncate_t *lr)
1366 if (zil_replaying(zd->zd_zilog, tx))
1369 itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr));
1370 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1371 sizeof (*lr) - sizeof (lr_t));
1373 itx->itx_sync = B_FALSE;
1374 zil_itx_assign(zd->zd_zilog, itx, tx);
1378 ztest_log_setattr(ztest_ds_t *zd, dmu_tx_t *tx, lr_setattr_t *lr)
1382 if (zil_replaying(zd->zd_zilog, tx))
1385 itx = zil_itx_create(TX_SETATTR, sizeof (*lr));
1386 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1387 sizeof (*lr) - sizeof (lr_t));
1389 itx->itx_sync = B_FALSE;
1390 zil_itx_assign(zd->zd_zilog, itx, tx);
1397 ztest_replay_create(ztest_ds_t *zd, lr_create_t *lr, boolean_t byteswap)
1399 char *name = (void *)(lr + 1); /* name follows lr */
1400 objset_t *os = zd->zd_os;
1401 ztest_block_tag_t *bbt;
1408 byteswap_uint64_array(lr, sizeof (*lr));
1410 ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1411 ASSERT(name[0] != '\0');
1413 tx = dmu_tx_create(os);
1415 dmu_tx_hold_zap(tx, lr->lr_doid, B_TRUE, name);
1417 if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1418 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, B_TRUE, NULL);
1420 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1423 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1427 ASSERT(dmu_objset_zil(os)->zl_replay == !!lr->lr_foid);
1429 if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1430 if (lr->lr_foid == 0) {
1431 lr->lr_foid = zap_create(os,
1432 lr->lrz_type, lr->lrz_bonustype,
1433 lr->lrz_bonuslen, tx);
1435 error = zap_create_claim(os, lr->lr_foid,
1436 lr->lrz_type, lr->lrz_bonustype,
1437 lr->lrz_bonuslen, tx);
1440 if (lr->lr_foid == 0) {
1441 lr->lr_foid = dmu_object_alloc(os,
1442 lr->lrz_type, 0, lr->lrz_bonustype,
1443 lr->lrz_bonuslen, tx);
1445 error = dmu_object_claim(os, lr->lr_foid,
1446 lr->lrz_type, 0, lr->lrz_bonustype,
1447 lr->lrz_bonuslen, tx);
1452 ASSERT3U(error, ==, EEXIST);
1453 ASSERT(zd->zd_zilog->zl_replay);
1458 ASSERT(lr->lr_foid != 0);
1460 if (lr->lrz_type != DMU_OT_ZAP_OTHER)
1461 VERIFY3U(0, ==, dmu_object_set_blocksize(os, lr->lr_foid,
1462 lr->lrz_blocksize, lr->lrz_ibshift, tx));
1464 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1465 bbt = ztest_bt_bonus(db);
1466 dmu_buf_will_dirty(db, tx);
1467 ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_gen, txg, txg);
1468 dmu_buf_rele(db, FTAG);
1470 VERIFY3U(0, ==, zap_add(os, lr->lr_doid, name, sizeof (uint64_t), 1,
1473 (void) ztest_log_create(zd, tx, lr);
1481 ztest_replay_remove(ztest_ds_t *zd, lr_remove_t *lr, boolean_t byteswap)
1483 char *name = (void *)(lr + 1); /* name follows lr */
1484 objset_t *os = zd->zd_os;
1485 dmu_object_info_t doi;
1487 uint64_t object, txg;
1490 byteswap_uint64_array(lr, sizeof (*lr));
1492 ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1493 ASSERT(name[0] != '\0');
1496 zap_lookup(os, lr->lr_doid, name, sizeof (object), 1, &object));
1497 ASSERT(object != 0);
1499 ztest_object_lock(zd, object, RL_WRITER);
1501 VERIFY3U(0, ==, dmu_object_info(os, object, &doi));
1503 tx = dmu_tx_create(os);
1505 dmu_tx_hold_zap(tx, lr->lr_doid, B_FALSE, name);
1506 dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END);
1508 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1510 ztest_object_unlock(zd, object);
1514 if (doi.doi_type == DMU_OT_ZAP_OTHER) {
1515 VERIFY3U(0, ==, zap_destroy(os, object, tx));
1517 VERIFY3U(0, ==, dmu_object_free(os, object, tx));
1520 VERIFY3U(0, ==, zap_remove(os, lr->lr_doid, name, tx));
1522 (void) ztest_log_remove(zd, tx, lr, object);
1526 ztest_object_unlock(zd, object);
1532 ztest_replay_write(ztest_ds_t *zd, lr_write_t *lr, boolean_t byteswap)
1534 objset_t *os = zd->zd_os;
1535 void *data = lr + 1; /* data follows lr */
1536 uint64_t offset, length;
1537 ztest_block_tag_t *bt = data;
1538 ztest_block_tag_t *bbt;
1539 uint64_t gen, txg, lrtxg, crtxg;
1540 dmu_object_info_t doi;
1543 arc_buf_t *abuf = NULL;
1547 byteswap_uint64_array(lr, sizeof (*lr));
1549 offset = lr->lr_offset;
1550 length = lr->lr_length;
1552 /* If it's a dmu_sync() block, write the whole block */
1553 if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
1554 uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
1555 if (length < blocksize) {
1556 offset -= offset % blocksize;
1561 if (bt->bt_magic == BSWAP_64(BT_MAGIC))
1562 byteswap_uint64_array(bt, sizeof (*bt));
1564 if (bt->bt_magic != BT_MAGIC)
1567 ztest_object_lock(zd, lr->lr_foid, RL_READER);
1568 rl = ztest_range_lock(zd, lr->lr_foid, offset, length, RL_WRITER);
1570 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1572 dmu_object_info_from_db(db, &doi);
1574 bbt = ztest_bt_bonus(db);
1575 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1577 crtxg = bbt->bt_crtxg;
1578 lrtxg = lr->lr_common.lrc_txg;
1580 tx = dmu_tx_create(os);
1582 dmu_tx_hold_write(tx, lr->lr_foid, offset, length);
1584 if (ztest_random(8) == 0 && length == doi.doi_data_block_size &&
1585 P2PHASE(offset, length) == 0)
1586 abuf = dmu_request_arcbuf(db, length);
1588 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1591 dmu_return_arcbuf(abuf);
1592 dmu_buf_rele(db, FTAG);
1593 ztest_range_unlock(rl);
1594 ztest_object_unlock(zd, lr->lr_foid);
1600 * Usually, verify the old data before writing new data --
1601 * but not always, because we also want to verify correct
1602 * behavior when the data was not recently read into cache.
1604 ASSERT(offset % doi.doi_data_block_size == 0);
1605 if (ztest_random(4) != 0) {
1606 int prefetch = ztest_random(2) ?
1607 DMU_READ_PREFETCH : DMU_READ_NO_PREFETCH;
1608 ztest_block_tag_t rbt;
1610 VERIFY(dmu_read(os, lr->lr_foid, offset,
1611 sizeof (rbt), &rbt, prefetch) == 0);
1612 if (rbt.bt_magic == BT_MAGIC) {
1613 ztest_bt_verify(&rbt, os, lr->lr_foid,
1614 offset, gen, txg, crtxg);
1619 * Writes can appear to be newer than the bonus buffer because
1620 * the ztest_get_data() callback does a dmu_read() of the
1621 * open-context data, which may be different than the data
1622 * as it was when the write was generated.
1624 if (zd->zd_zilog->zl_replay) {
1625 ztest_bt_verify(bt, os, lr->lr_foid, offset,
1626 MAX(gen, bt->bt_gen), MAX(txg, lrtxg),
1631 * Set the bt's gen/txg to the bonus buffer's gen/txg
1632 * so that all of the usual ASSERTs will work.
1634 ztest_bt_generate(bt, os, lr->lr_foid, offset, gen, txg, crtxg);
1638 dmu_write(os, lr->lr_foid, offset, length, data, tx);
1640 bcopy(data, abuf->b_data, length);
1641 dmu_assign_arcbuf(db, offset, abuf, tx);
1644 (void) ztest_log_write(zd, tx, lr);
1646 dmu_buf_rele(db, FTAG);
1650 ztest_range_unlock(rl);
1651 ztest_object_unlock(zd, lr->lr_foid);
1657 ztest_replay_truncate(ztest_ds_t *zd, lr_truncate_t *lr, boolean_t byteswap)
1659 objset_t *os = zd->zd_os;
1665 byteswap_uint64_array(lr, sizeof (*lr));
1667 ztest_object_lock(zd, lr->lr_foid, RL_READER);
1668 rl = ztest_range_lock(zd, lr->lr_foid, lr->lr_offset, lr->lr_length,
1671 tx = dmu_tx_create(os);
1673 dmu_tx_hold_free(tx, lr->lr_foid, lr->lr_offset, lr->lr_length);
1675 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1677 ztest_range_unlock(rl);
1678 ztest_object_unlock(zd, lr->lr_foid);
1682 VERIFY(dmu_free_range(os, lr->lr_foid, lr->lr_offset,
1683 lr->lr_length, tx) == 0);
1685 (void) ztest_log_truncate(zd, tx, lr);
1689 ztest_range_unlock(rl);
1690 ztest_object_unlock(zd, lr->lr_foid);
1696 ztest_replay_setattr(ztest_ds_t *zd, lr_setattr_t *lr, boolean_t byteswap)
1698 objset_t *os = zd->zd_os;
1701 ztest_block_tag_t *bbt;
1702 uint64_t txg, lrtxg, crtxg;
1705 byteswap_uint64_array(lr, sizeof (*lr));
1707 ztest_object_lock(zd, lr->lr_foid, RL_WRITER);
1709 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1711 tx = dmu_tx_create(os);
1712 dmu_tx_hold_bonus(tx, lr->lr_foid);
1714 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1716 dmu_buf_rele(db, FTAG);
1717 ztest_object_unlock(zd, lr->lr_foid);
1721 bbt = ztest_bt_bonus(db);
1722 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1723 crtxg = bbt->bt_crtxg;
1724 lrtxg = lr->lr_common.lrc_txg;
1726 if (zd->zd_zilog->zl_replay) {
1727 ASSERT(lr->lr_size != 0);
1728 ASSERT(lr->lr_mode != 0);
1732 * Randomly change the size and increment the generation.
1734 lr->lr_size = (ztest_random(db->db_size / sizeof (*bbt)) + 1) *
1736 lr->lr_mode = bbt->bt_gen + 1;
1741 * Verify that the current bonus buffer is not newer than our txg.
1743 ztest_bt_verify(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode,
1744 MAX(txg, lrtxg), crtxg);
1746 dmu_buf_will_dirty(db, tx);
1748 ASSERT3U(lr->lr_size, >=, sizeof (*bbt));
1749 ASSERT3U(lr->lr_size, <=, db->db_size);
1750 VERIFY0(dmu_set_bonus(db, lr->lr_size, tx));
1751 bbt = ztest_bt_bonus(db);
1753 ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode, txg, crtxg);
1755 dmu_buf_rele(db, FTAG);
1757 (void) ztest_log_setattr(zd, tx, lr);
1761 ztest_object_unlock(zd, lr->lr_foid);
1766 zil_replay_func_t *ztest_replay_vector[TX_MAX_TYPE] = {
1767 NULL, /* 0 no such transaction type */
1768 ztest_replay_create, /* TX_CREATE */
1769 NULL, /* TX_MKDIR */
1770 NULL, /* TX_MKXATTR */
1771 NULL, /* TX_SYMLINK */
1772 ztest_replay_remove, /* TX_REMOVE */
1773 NULL, /* TX_RMDIR */
1775 NULL, /* TX_RENAME */
1776 ztest_replay_write, /* TX_WRITE */
1777 ztest_replay_truncate, /* TX_TRUNCATE */
1778 ztest_replay_setattr, /* TX_SETATTR */
1780 NULL, /* TX_CREATE_ACL */
1781 NULL, /* TX_CREATE_ATTR */
1782 NULL, /* TX_CREATE_ACL_ATTR */
1783 NULL, /* TX_MKDIR_ACL */
1784 NULL, /* TX_MKDIR_ATTR */
1785 NULL, /* TX_MKDIR_ACL_ATTR */
1786 NULL, /* TX_WRITE2 */
1790 * ZIL get_data callbacks
1794 ztest_get_done(zgd_t *zgd, int error)
1796 ztest_ds_t *zd = zgd->zgd_private;
1797 uint64_t object = zgd->zgd_rl->rl_object;
1800 dmu_buf_rele(zgd->zgd_db, zgd);
1802 ztest_range_unlock(zgd->zgd_rl);
1803 ztest_object_unlock(zd, object);
1805 if (error == 0 && zgd->zgd_bp)
1806 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
1808 umem_free(zgd, sizeof (*zgd));
1812 ztest_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
1814 ztest_ds_t *zd = arg;
1815 objset_t *os = zd->zd_os;
1816 uint64_t object = lr->lr_foid;
1817 uint64_t offset = lr->lr_offset;
1818 uint64_t size = lr->lr_length;
1819 blkptr_t *bp = &lr->lr_blkptr;
1820 uint64_t txg = lr->lr_common.lrc_txg;
1822 dmu_object_info_t doi;
1827 ztest_object_lock(zd, object, RL_READER);
1828 error = dmu_bonus_hold(os, object, FTAG, &db);
1830 ztest_object_unlock(zd, object);
1834 crtxg = ztest_bt_bonus(db)->bt_crtxg;
1836 if (crtxg == 0 || crtxg > txg) {
1837 dmu_buf_rele(db, FTAG);
1838 ztest_object_unlock(zd, object);
1842 dmu_object_info_from_db(db, &doi);
1843 dmu_buf_rele(db, FTAG);
1846 zgd = umem_zalloc(sizeof (*zgd), UMEM_NOFAIL);
1847 zgd->zgd_zilog = zd->zd_zilog;
1848 zgd->zgd_private = zd;
1850 if (buf != NULL) { /* immediate write */
1851 zgd->zgd_rl = ztest_range_lock(zd, object, offset, size,
1854 error = dmu_read(os, object, offset, size, buf,
1855 DMU_READ_NO_PREFETCH);
1858 size = doi.doi_data_block_size;
1860 offset = P2ALIGN(offset, size);
1862 ASSERT(offset < size);
1866 zgd->zgd_rl = ztest_range_lock(zd, object, offset, size,
1869 error = dmu_buf_hold(os, object, offset, zgd, &db,
1870 DMU_READ_NO_PREFETCH);
1873 blkptr_t *obp = dmu_buf_get_blkptr(db);
1875 ASSERT(BP_IS_HOLE(bp));
1882 ASSERT(db->db_offset == offset);
1883 ASSERT(db->db_size == size);
1885 error = dmu_sync(zio, lr->lr_common.lrc_txg,
1886 ztest_get_done, zgd);
1893 ztest_get_done(zgd, error);
1899 ztest_lr_alloc(size_t lrsize, char *name)
1902 size_t namesize = name ? strlen(name) + 1 : 0;
1904 lr = umem_zalloc(lrsize + namesize, UMEM_NOFAIL);
1907 bcopy(name, lr + lrsize, namesize);
1913 ztest_lr_free(void *lr, size_t lrsize, char *name)
1915 size_t namesize = name ? strlen(name) + 1 : 0;
1917 umem_free(lr, lrsize + namesize);
1921 * Lookup a bunch of objects. Returns the number of objects not found.
1924 ztest_lookup(ztest_ds_t *zd, ztest_od_t *od, int count)
1929 ASSERT(_mutex_held(&zd->zd_dirobj_lock));
1931 for (int i = 0; i < count; i++, od++) {
1933 error = zap_lookup(zd->zd_os, od->od_dir, od->od_name,
1934 sizeof (uint64_t), 1, &od->od_object);
1936 ASSERT(error == ENOENT);
1937 ASSERT(od->od_object == 0);
1941 ztest_block_tag_t *bbt;
1942 dmu_object_info_t doi;
1944 ASSERT(od->od_object != 0);
1945 ASSERT(missing == 0); /* there should be no gaps */
1947 ztest_object_lock(zd, od->od_object, RL_READER);
1948 VERIFY3U(0, ==, dmu_bonus_hold(zd->zd_os,
1949 od->od_object, FTAG, &db));
1950 dmu_object_info_from_db(db, &doi);
1951 bbt = ztest_bt_bonus(db);
1952 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1953 od->od_type = doi.doi_type;
1954 od->od_blocksize = doi.doi_data_block_size;
1955 od->od_gen = bbt->bt_gen;
1956 dmu_buf_rele(db, FTAG);
1957 ztest_object_unlock(zd, od->od_object);
1965 ztest_create(ztest_ds_t *zd, ztest_od_t *od, int count)
1969 ASSERT(_mutex_held(&zd->zd_dirobj_lock));
1971 for (int i = 0; i < count; i++, od++) {
1978 lr_create_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
1980 lr->lr_doid = od->od_dir;
1981 lr->lr_foid = 0; /* 0 to allocate, > 0 to claim */
1982 lr->lrz_type = od->od_crtype;
1983 lr->lrz_blocksize = od->od_crblocksize;
1984 lr->lrz_ibshift = ztest_random_ibshift();
1985 lr->lrz_bonustype = DMU_OT_UINT64_OTHER;
1986 lr->lrz_bonuslen = dmu_bonus_max();
1987 lr->lr_gen = od->od_crgen;
1988 lr->lr_crtime[0] = time(NULL);
1990 if (ztest_replay_create(zd, lr, B_FALSE) != 0) {
1991 ASSERT(missing == 0);
1995 od->od_object = lr->lr_foid;
1996 od->od_type = od->od_crtype;
1997 od->od_blocksize = od->od_crblocksize;
1998 od->od_gen = od->od_crgen;
1999 ASSERT(od->od_object != 0);
2002 ztest_lr_free(lr, sizeof (*lr), od->od_name);
2009 ztest_remove(ztest_ds_t *zd, ztest_od_t *od, int count)
2014 ASSERT(_mutex_held(&zd->zd_dirobj_lock));
2018 for (int i = count - 1; i >= 0; i--, od--) {
2025 * No object was found.
2027 if (od->od_object == 0)
2030 lr_remove_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
2032 lr->lr_doid = od->od_dir;
2034 if ((error = ztest_replay_remove(zd, lr, B_FALSE)) != 0) {
2035 ASSERT3U(error, ==, ENOSPC);
2040 ztest_lr_free(lr, sizeof (*lr), od->od_name);
2047 ztest_write(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size,
2053 lr = ztest_lr_alloc(sizeof (*lr) + size, NULL);
2055 lr->lr_foid = object;
2056 lr->lr_offset = offset;
2057 lr->lr_length = size;
2059 BP_ZERO(&lr->lr_blkptr);
2061 bcopy(data, lr + 1, size);
2063 error = ztest_replay_write(zd, lr, B_FALSE);
2065 ztest_lr_free(lr, sizeof (*lr) + size, NULL);
2071 ztest_truncate(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2076 lr = ztest_lr_alloc(sizeof (*lr), NULL);
2078 lr->lr_foid = object;
2079 lr->lr_offset = offset;
2080 lr->lr_length = size;
2082 error = ztest_replay_truncate(zd, lr, B_FALSE);
2084 ztest_lr_free(lr, sizeof (*lr), NULL);
2090 ztest_setattr(ztest_ds_t *zd, uint64_t object)
2095 lr = ztest_lr_alloc(sizeof (*lr), NULL);
2097 lr->lr_foid = object;
2101 error = ztest_replay_setattr(zd, lr, B_FALSE);
2103 ztest_lr_free(lr, sizeof (*lr), NULL);
2109 ztest_prealloc(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2111 objset_t *os = zd->zd_os;
2116 txg_wait_synced(dmu_objset_pool(os), 0);
2118 ztest_object_lock(zd, object, RL_READER);
2119 rl = ztest_range_lock(zd, object, offset, size, RL_WRITER);
2121 tx = dmu_tx_create(os);
2123 dmu_tx_hold_write(tx, object, offset, size);
2125 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
2128 dmu_prealloc(os, object, offset, size, tx);
2130 txg_wait_synced(dmu_objset_pool(os), txg);
2132 (void) dmu_free_long_range(os, object, offset, size);
2135 ztest_range_unlock(rl);
2136 ztest_object_unlock(zd, object);
2140 ztest_io(ztest_ds_t *zd, uint64_t object, uint64_t offset)
2143 ztest_block_tag_t wbt;
2144 dmu_object_info_t doi;
2145 enum ztest_io_type io_type;
2149 VERIFY(dmu_object_info(zd->zd_os, object, &doi) == 0);
2150 blocksize = doi.doi_data_block_size;
2151 data = umem_alloc(blocksize, UMEM_NOFAIL);
2154 * Pick an i/o type at random, biased toward writing block tags.
2156 io_type = ztest_random(ZTEST_IO_TYPES);
2157 if (ztest_random(2) == 0)
2158 io_type = ZTEST_IO_WRITE_TAG;
2160 (void) rw_rdlock(&zd->zd_zilog_lock);
2164 case ZTEST_IO_WRITE_TAG:
2165 ztest_bt_generate(&wbt, zd->zd_os, object, offset, 0, 0, 0);
2166 (void) ztest_write(zd, object, offset, sizeof (wbt), &wbt);
2169 case ZTEST_IO_WRITE_PATTERN:
2170 (void) memset(data, 'a' + (object + offset) % 5, blocksize);
2171 if (ztest_random(2) == 0) {
2173 * Induce fletcher2 collisions to ensure that
2174 * zio_ddt_collision() detects and resolves them
2175 * when using fletcher2-verify for deduplication.
2177 ((uint64_t *)data)[0] ^= 1ULL << 63;
2178 ((uint64_t *)data)[4] ^= 1ULL << 63;
2180 (void) ztest_write(zd, object, offset, blocksize, data);
2183 case ZTEST_IO_WRITE_ZEROES:
2184 bzero(data, blocksize);
2185 (void) ztest_write(zd, object, offset, blocksize, data);
2188 case ZTEST_IO_TRUNCATE:
2189 (void) ztest_truncate(zd, object, offset, blocksize);
2192 case ZTEST_IO_SETATTR:
2193 (void) ztest_setattr(zd, object);
2196 case ZTEST_IO_REWRITE:
2197 (void) rw_rdlock(&ztest_name_lock);
2198 err = ztest_dsl_prop_set_uint64(zd->zd_name,
2199 ZFS_PROP_CHECKSUM, spa_dedup_checksum(ztest_spa),
2201 VERIFY(err == 0 || err == ENOSPC);
2202 err = ztest_dsl_prop_set_uint64(zd->zd_name,
2203 ZFS_PROP_COMPRESSION,
2204 ztest_random_dsl_prop(ZFS_PROP_COMPRESSION),
2206 VERIFY(err == 0 || err == ENOSPC);
2207 (void) rw_unlock(&ztest_name_lock);
2209 VERIFY0(dmu_read(zd->zd_os, object, offset, blocksize, data,
2210 DMU_READ_NO_PREFETCH));
2212 (void) ztest_write(zd, object, offset, blocksize, data);
2216 (void) rw_unlock(&zd->zd_zilog_lock);
2218 umem_free(data, blocksize);
2222 * Initialize an object description template.
2225 ztest_od_init(ztest_od_t *od, uint64_t id, char *tag, uint64_t index,
2226 dmu_object_type_t type, uint64_t blocksize, uint64_t gen)
2228 od->od_dir = ZTEST_DIROBJ;
2231 od->od_crtype = type;
2232 od->od_crblocksize = blocksize ? blocksize : ztest_random_blocksize();
2235 od->od_type = DMU_OT_NONE;
2236 od->od_blocksize = 0;
2239 (void) snprintf(od->od_name, sizeof (od->od_name), "%s(%lld)[%llu]",
2240 tag, (int64_t)id, index);
2244 * Lookup or create the objects for a test using the od template.
2245 * If the objects do not all exist, or if 'remove' is specified,
2246 * remove any existing objects and create new ones. Otherwise,
2247 * use the existing objects.
2250 ztest_object_init(ztest_ds_t *zd, ztest_od_t *od, size_t size, boolean_t remove)
2252 int count = size / sizeof (*od);
2255 VERIFY(mutex_lock(&zd->zd_dirobj_lock) == 0);
2256 if ((ztest_lookup(zd, od, count) != 0 || remove) &&
2257 (ztest_remove(zd, od, count) != 0 ||
2258 ztest_create(zd, od, count) != 0))
2261 VERIFY(mutex_unlock(&zd->zd_dirobj_lock) == 0);
2268 ztest_zil_commit(ztest_ds_t *zd, uint64_t id)
2270 zilog_t *zilog = zd->zd_zilog;
2272 (void) rw_rdlock(&zd->zd_zilog_lock);
2274 zil_commit(zilog, ztest_random(ZTEST_OBJECTS));
2277 * Remember the committed values in zd, which is in parent/child
2278 * shared memory. If we die, the next iteration of ztest_run()
2279 * will verify that the log really does contain this record.
2281 mutex_enter(&zilog->zl_lock);
2282 ASSERT(zd->zd_shared != NULL);
2283 ASSERT3U(zd->zd_shared->zd_seq, <=, zilog->zl_commit_lr_seq);
2284 zd->zd_shared->zd_seq = zilog->zl_commit_lr_seq;
2285 mutex_exit(&zilog->zl_lock);
2287 (void) rw_unlock(&zd->zd_zilog_lock);
2291 * This function is designed to simulate the operations that occur during a
2292 * mount/unmount operation. We hold the dataset across these operations in an
2293 * attempt to expose any implicit assumptions about ZIL management.
2297 ztest_zil_remount(ztest_ds_t *zd, uint64_t id)
2299 objset_t *os = zd->zd_os;
2302 * We grab the zd_dirobj_lock to ensure that no other thread is
2303 * updating the zil (i.e. adding in-memory log records) and the
2304 * zd_zilog_lock to block any I/O.
2306 VERIFY0(mutex_lock(&zd->zd_dirobj_lock));
2307 (void) rw_wrlock(&zd->zd_zilog_lock);
2309 /* zfsvfs_teardown() */
2310 zil_close(zd->zd_zilog);
2312 /* zfsvfs_setup() */
2313 VERIFY(zil_open(os, ztest_get_data) == zd->zd_zilog);
2314 zil_replay(os, zd, ztest_replay_vector);
2316 (void) rw_unlock(&zd->zd_zilog_lock);
2317 VERIFY(mutex_unlock(&zd->zd_dirobj_lock) == 0);
2321 * Verify that we can't destroy an active pool, create an existing pool,
2322 * or create a pool with a bad vdev spec.
2326 ztest_spa_create_destroy(ztest_ds_t *zd, uint64_t id)
2328 ztest_shared_opts_t *zo = &ztest_opts;
2333 * Attempt to create using a bad file.
2335 nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1);
2336 VERIFY3U(ENOENT, ==,
2337 spa_create("ztest_bad_file", nvroot, NULL, NULL));
2338 nvlist_free(nvroot);
2341 * Attempt to create using a bad mirror.
2343 nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 2, 1);
2344 VERIFY3U(ENOENT, ==,
2345 spa_create("ztest_bad_mirror", nvroot, NULL, NULL));
2346 nvlist_free(nvroot);
2349 * Attempt to create an existing pool. It shouldn't matter
2350 * what's in the nvroot; we should fail with EEXIST.
2352 (void) rw_rdlock(&ztest_name_lock);
2353 nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1);
2354 VERIFY3U(EEXIST, ==, spa_create(zo->zo_pool, nvroot, NULL, NULL));
2355 nvlist_free(nvroot);
2356 VERIFY3U(0, ==, spa_open(zo->zo_pool, &spa, FTAG));
2357 VERIFY3U(EBUSY, ==, spa_destroy(zo->zo_pool));
2358 spa_close(spa, FTAG);
2360 (void) rw_unlock(&ztest_name_lock);
2365 ztest_spa_upgrade(ztest_ds_t *zd, uint64_t id)
2368 uint64_t initial_version = SPA_VERSION_INITIAL;
2369 uint64_t version, newversion;
2370 nvlist_t *nvroot, *props;
2373 VERIFY0(mutex_lock(&ztest_vdev_lock));
2374 name = kmem_asprintf("%s_upgrade", ztest_opts.zo_pool);
2377 * Clean up from previous runs.
2379 (void) spa_destroy(name);
2381 nvroot = make_vdev_root(NULL, NULL, name, ztest_opts.zo_vdev_size, 0,
2382 0, ztest_opts.zo_raidz, ztest_opts.zo_mirrors, 1);
2385 * If we're configuring a RAIDZ device then make sure that the
2386 * the initial version is capable of supporting that feature.
2388 switch (ztest_opts.zo_raidz_parity) {
2391 initial_version = SPA_VERSION_INITIAL;
2394 initial_version = SPA_VERSION_RAIDZ2;
2397 initial_version = SPA_VERSION_RAIDZ3;
2402 * Create a pool with a spa version that can be upgraded. Pick
2403 * a value between initial_version and SPA_VERSION_BEFORE_FEATURES.
2406 version = ztest_random_spa_version(initial_version);
2407 } while (version > SPA_VERSION_BEFORE_FEATURES);
2409 props = fnvlist_alloc();
2410 fnvlist_add_uint64(props,
2411 zpool_prop_to_name(ZPOOL_PROP_VERSION), version);
2412 VERIFY0(spa_create(name, nvroot, props, NULL));
2413 fnvlist_free(nvroot);
2414 fnvlist_free(props);
2416 VERIFY0(spa_open(name, &spa, FTAG));
2417 VERIFY3U(spa_version(spa), ==, version);
2418 newversion = ztest_random_spa_version(version + 1);
2420 if (ztest_opts.zo_verbose >= 4) {
2421 (void) printf("upgrading spa version from %llu to %llu\n",
2422 (u_longlong_t)version, (u_longlong_t)newversion);
2425 spa_upgrade(spa, newversion);
2426 VERIFY3U(spa_version(spa), >, version);
2427 VERIFY3U(spa_version(spa), ==, fnvlist_lookup_uint64(spa->spa_config,
2428 zpool_prop_to_name(ZPOOL_PROP_VERSION)));
2429 spa_close(spa, FTAG);
2432 VERIFY0(mutex_unlock(&ztest_vdev_lock));
2436 vdev_lookup_by_path(vdev_t *vd, const char *path)
2440 if (vd->vdev_path != NULL && strcmp(path, vd->vdev_path) == 0)
2443 for (int c = 0; c < vd->vdev_children; c++)
2444 if ((mvd = vdev_lookup_by_path(vd->vdev_child[c], path)) !=
2452 * Find the first available hole which can be used as a top-level.
2455 find_vdev_hole(spa_t *spa)
2457 vdev_t *rvd = spa->spa_root_vdev;
2460 ASSERT(spa_config_held(spa, SCL_VDEV, RW_READER) == SCL_VDEV);
2462 for (c = 0; c < rvd->vdev_children; c++) {
2463 vdev_t *cvd = rvd->vdev_child[c];
2465 if (cvd->vdev_ishole)
2472 * Verify that vdev_add() works as expected.
2476 ztest_vdev_add_remove(ztest_ds_t *zd, uint64_t id)
2478 ztest_shared_t *zs = ztest_shared;
2479 spa_t *spa = ztest_spa;
2485 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2486 leaves = MAX(zs->zs_mirrors + zs->zs_splits, 1) * ztest_opts.zo_raidz;
2488 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2490 ztest_shared->zs_vdev_next_leaf = find_vdev_hole(spa) * leaves;
2493 * If we have slogs then remove them 1/4 of the time.
2495 if (spa_has_slogs(spa) && ztest_random(4) == 0) {
2497 * Grab the guid from the head of the log class rotor.
2499 guid = spa_log_class(spa)->mc_rotor->mg_vd->vdev_guid;
2501 spa_config_exit(spa, SCL_VDEV, FTAG);
2504 * We have to grab the zs_name_lock as writer to
2505 * prevent a race between removing a slog (dmu_objset_find)
2506 * and destroying a dataset. Removing the slog will
2507 * grab a reference on the dataset which may cause
2508 * dmu_objset_destroy() to fail with EBUSY thus
2509 * leaving the dataset in an inconsistent state.
2511 VERIFY(rw_wrlock(&ztest_name_lock) == 0);
2512 error = spa_vdev_remove(spa, guid, B_FALSE);
2513 VERIFY(rw_unlock(&ztest_name_lock) == 0);
2515 if (error && error != EEXIST)
2516 fatal(0, "spa_vdev_remove() = %d", error);
2518 spa_config_exit(spa, SCL_VDEV, FTAG);
2521 * Make 1/4 of the devices be log devices.
2523 nvroot = make_vdev_root(NULL, NULL, NULL,
2524 ztest_opts.zo_vdev_size, 0,
2525 ztest_random(4) == 0, ztest_opts.zo_raidz,
2528 error = spa_vdev_add(spa, nvroot);
2529 nvlist_free(nvroot);
2531 if (error == ENOSPC)
2532 ztest_record_enospc("spa_vdev_add");
2533 else if (error != 0)
2534 fatal(0, "spa_vdev_add() = %d", error);
2537 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2541 * Verify that adding/removing aux devices (l2arc, hot spare) works as expected.
2545 ztest_vdev_aux_add_remove(ztest_ds_t *zd, uint64_t id)
2547 ztest_shared_t *zs = ztest_shared;
2548 spa_t *spa = ztest_spa;
2549 vdev_t *rvd = spa->spa_root_vdev;
2550 spa_aux_vdev_t *sav;
2555 if (ztest_random(2) == 0) {
2556 sav = &spa->spa_spares;
2557 aux = ZPOOL_CONFIG_SPARES;
2559 sav = &spa->spa_l2cache;
2560 aux = ZPOOL_CONFIG_L2CACHE;
2563 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2565 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2567 if (sav->sav_count != 0 && ztest_random(4) == 0) {
2569 * Pick a random device to remove.
2571 guid = sav->sav_vdevs[ztest_random(sav->sav_count)]->vdev_guid;
2574 * Find an unused device we can add.
2576 zs->zs_vdev_aux = 0;
2578 char path[MAXPATHLEN];
2580 (void) snprintf(path, sizeof (path), ztest_aux_template,
2581 ztest_opts.zo_dir, ztest_opts.zo_pool, aux,
2583 for (c = 0; c < sav->sav_count; c++)
2584 if (strcmp(sav->sav_vdevs[c]->vdev_path,
2587 if (c == sav->sav_count &&
2588 vdev_lookup_by_path(rvd, path) == NULL)
2594 spa_config_exit(spa, SCL_VDEV, FTAG);
2600 nvlist_t *nvroot = make_vdev_root(NULL, aux, NULL,
2601 (ztest_opts.zo_vdev_size * 5) / 4, 0, 0, 0, 0, 1);
2602 error = spa_vdev_add(spa, nvroot);
2604 fatal(0, "spa_vdev_add(%p) = %d", nvroot, error);
2605 nvlist_free(nvroot);
2608 * Remove an existing device. Sometimes, dirty its
2609 * vdev state first to make sure we handle removal
2610 * of devices that have pending state changes.
2612 if (ztest_random(2) == 0)
2613 (void) vdev_online(spa, guid, 0, NULL);
2615 error = spa_vdev_remove(spa, guid, B_FALSE);
2616 if (error != 0 && error != EBUSY)
2617 fatal(0, "spa_vdev_remove(%llu) = %d", guid, error);
2620 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2624 * split a pool if it has mirror tlvdevs
2628 ztest_split_pool(ztest_ds_t *zd, uint64_t id)
2630 ztest_shared_t *zs = ztest_shared;
2631 spa_t *spa = ztest_spa;
2632 vdev_t *rvd = spa->spa_root_vdev;
2633 nvlist_t *tree, **child, *config, *split, **schild;
2634 uint_t c, children, schildren = 0, lastlogid = 0;
2637 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2639 /* ensure we have a useable config; mirrors of raidz aren't supported */
2640 if (zs->zs_mirrors < 3 || ztest_opts.zo_raidz > 1) {
2641 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2645 /* clean up the old pool, if any */
2646 (void) spa_destroy("splitp");
2648 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2650 /* generate a config from the existing config */
2651 mutex_enter(&spa->spa_props_lock);
2652 VERIFY(nvlist_lookup_nvlist(spa->spa_config, ZPOOL_CONFIG_VDEV_TREE,
2654 mutex_exit(&spa->spa_props_lock);
2656 VERIFY(nvlist_lookup_nvlist_array(tree, ZPOOL_CONFIG_CHILDREN, &child,
2659 schild = malloc(rvd->vdev_children * sizeof (nvlist_t *));
2660 for (c = 0; c < children; c++) {
2661 vdev_t *tvd = rvd->vdev_child[c];
2665 if (tvd->vdev_islog || tvd->vdev_ops == &vdev_hole_ops) {
2666 VERIFY(nvlist_alloc(&schild[schildren], NV_UNIQUE_NAME,
2668 VERIFY(nvlist_add_string(schild[schildren],
2669 ZPOOL_CONFIG_TYPE, VDEV_TYPE_HOLE) == 0);
2670 VERIFY(nvlist_add_uint64(schild[schildren],
2671 ZPOOL_CONFIG_IS_HOLE, 1) == 0);
2673 lastlogid = schildren;
2678 VERIFY(nvlist_lookup_nvlist_array(child[c],
2679 ZPOOL_CONFIG_CHILDREN, &mchild, &mchildren) == 0);
2680 VERIFY(nvlist_dup(mchild[0], &schild[schildren++], 0) == 0);
2683 /* OK, create a config that can be used to split */
2684 VERIFY(nvlist_alloc(&split, NV_UNIQUE_NAME, 0) == 0);
2685 VERIFY(nvlist_add_string(split, ZPOOL_CONFIG_TYPE,
2686 VDEV_TYPE_ROOT) == 0);
2687 VERIFY(nvlist_add_nvlist_array(split, ZPOOL_CONFIG_CHILDREN, schild,
2688 lastlogid != 0 ? lastlogid : schildren) == 0);
2690 VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, 0) == 0);
2691 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, split) == 0);
2693 for (c = 0; c < schildren; c++)
2694 nvlist_free(schild[c]);
2698 spa_config_exit(spa, SCL_VDEV, FTAG);
2700 (void) rw_wrlock(&ztest_name_lock);
2701 error = spa_vdev_split_mirror(spa, "splitp", config, NULL, B_FALSE);
2702 (void) rw_unlock(&ztest_name_lock);
2704 nvlist_free(config);
2707 (void) printf("successful split - results:\n");
2708 mutex_enter(&spa_namespace_lock);
2709 show_pool_stats(spa);
2710 show_pool_stats(spa_lookup("splitp"));
2711 mutex_exit(&spa_namespace_lock);
2715 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2720 * Verify that we can attach and detach devices.
2724 ztest_vdev_attach_detach(ztest_ds_t *zd, uint64_t id)
2726 ztest_shared_t *zs = ztest_shared;
2727 spa_t *spa = ztest_spa;
2728 spa_aux_vdev_t *sav = &spa->spa_spares;
2729 vdev_t *rvd = spa->spa_root_vdev;
2730 vdev_t *oldvd, *newvd, *pvd;
2734 uint64_t ashift = ztest_get_ashift();
2735 uint64_t oldguid, pguid;
2736 size_t oldsize, newsize;
2737 char oldpath[MAXPATHLEN], newpath[MAXPATHLEN];
2739 int oldvd_has_siblings = B_FALSE;
2740 int newvd_is_spare = B_FALSE;
2742 int error, expected_error;
2744 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2745 leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz;
2747 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2750 * Decide whether to do an attach or a replace.
2752 replacing = ztest_random(2);
2755 * Pick a random top-level vdev.
2757 top = ztest_random_vdev_top(spa, B_TRUE);
2760 * Pick a random leaf within it.
2762 leaf = ztest_random(leaves);
2767 oldvd = rvd->vdev_child[top];
2768 if (zs->zs_mirrors >= 1) {
2769 ASSERT(oldvd->vdev_ops == &vdev_mirror_ops);
2770 ASSERT(oldvd->vdev_children >= zs->zs_mirrors);
2771 oldvd = oldvd->vdev_child[leaf / ztest_opts.zo_raidz];
2773 if (ztest_opts.zo_raidz > 1) {
2774 ASSERT(oldvd->vdev_ops == &vdev_raidz_ops);
2775 ASSERT(oldvd->vdev_children == ztest_opts.zo_raidz);
2776 oldvd = oldvd->vdev_child[leaf % ztest_opts.zo_raidz];
2780 * If we're already doing an attach or replace, oldvd may be a
2781 * mirror vdev -- in which case, pick a random child.
2783 while (oldvd->vdev_children != 0) {
2784 oldvd_has_siblings = B_TRUE;
2785 ASSERT(oldvd->vdev_children >= 2);
2786 oldvd = oldvd->vdev_child[ztest_random(oldvd->vdev_children)];
2789 oldguid = oldvd->vdev_guid;
2790 oldsize = vdev_get_min_asize(oldvd);
2791 oldvd_is_log = oldvd->vdev_top->vdev_islog;
2792 (void) strcpy(oldpath, oldvd->vdev_path);
2793 pvd = oldvd->vdev_parent;
2794 pguid = pvd->vdev_guid;
2797 * If oldvd has siblings, then half of the time, detach it.
2799 if (oldvd_has_siblings && ztest_random(2) == 0) {
2800 spa_config_exit(spa, SCL_VDEV, FTAG);
2801 error = spa_vdev_detach(spa, oldguid, pguid, B_FALSE);
2802 if (error != 0 && error != ENODEV && error != EBUSY &&
2804 fatal(0, "detach (%s) returned %d", oldpath, error);
2805 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2810 * For the new vdev, choose with equal probability between the two
2811 * standard paths (ending in either 'a' or 'b') or a random hot spare.
2813 if (sav->sav_count != 0 && ztest_random(3) == 0) {
2814 newvd = sav->sav_vdevs[ztest_random(sav->sav_count)];
2815 newvd_is_spare = B_TRUE;
2816 (void) strcpy(newpath, newvd->vdev_path);
2818 (void) snprintf(newpath, sizeof (newpath), ztest_dev_template,
2819 ztest_opts.zo_dir, ztest_opts.zo_pool,
2820 top * leaves + leaf);
2821 if (ztest_random(2) == 0)
2822 newpath[strlen(newpath) - 1] = 'b';
2823 newvd = vdev_lookup_by_path(rvd, newpath);
2827 newsize = vdev_get_min_asize(newvd);
2830 * Make newsize a little bigger or smaller than oldsize.
2831 * If it's smaller, the attach should fail.
2832 * If it's larger, and we're doing a replace,
2833 * we should get dynamic LUN growth when we're done.
2835 newsize = 10 * oldsize / (9 + ztest_random(3));
2839 * If pvd is not a mirror or root, the attach should fail with ENOTSUP,
2840 * unless it's a replace; in that case any non-replacing parent is OK.
2842 * If newvd is already part of the pool, it should fail with EBUSY.
2844 * If newvd is too small, it should fail with EOVERFLOW.
2846 if (pvd->vdev_ops != &vdev_mirror_ops &&
2847 pvd->vdev_ops != &vdev_root_ops && (!replacing ||
2848 pvd->vdev_ops == &vdev_replacing_ops ||
2849 pvd->vdev_ops == &vdev_spare_ops))
2850 expected_error = ENOTSUP;
2851 else if (newvd_is_spare && (!replacing || oldvd_is_log))
2852 expected_error = ENOTSUP;
2853 else if (newvd == oldvd)
2854 expected_error = replacing ? 0 : EBUSY;
2855 else if (vdev_lookup_by_path(rvd, newpath) != NULL)
2856 expected_error = EBUSY;
2857 else if (newsize < oldsize)
2858 expected_error = EOVERFLOW;
2859 else if (ashift > oldvd->vdev_top->vdev_ashift)
2860 expected_error = EDOM;
2864 spa_config_exit(spa, SCL_VDEV, FTAG);
2867 * Build the nvlist describing newpath.
2869 root = make_vdev_root(newpath, NULL, NULL, newvd == NULL ? newsize : 0,
2870 ashift, 0, 0, 0, 1);
2872 error = spa_vdev_attach(spa, oldguid, root, replacing);
2877 * If our parent was the replacing vdev, but the replace completed,
2878 * then instead of failing with ENOTSUP we may either succeed,
2879 * fail with ENODEV, or fail with EOVERFLOW.
2881 if (expected_error == ENOTSUP &&
2882 (error == 0 || error == ENODEV || error == EOVERFLOW))
2883 expected_error = error;
2886 * If someone grew the LUN, the replacement may be too small.
2888 if (error == EOVERFLOW || error == EBUSY)
2889 expected_error = error;
2891 /* XXX workaround 6690467 */
2892 if (error != expected_error && expected_error != EBUSY) {
2893 fatal(0, "attach (%s %llu, %s %llu, %d) "
2894 "returned %d, expected %d",
2895 oldpath, (longlong_t)oldsize, newpath,
2896 (longlong_t)newsize, replacing, error, expected_error);
2899 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2903 * Callback function which expands the physical size of the vdev.
2906 grow_vdev(vdev_t *vd, void *arg)
2908 spa_t *spa = vd->vdev_spa;
2909 size_t *newsize = arg;
2913 ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
2914 ASSERT(vd->vdev_ops->vdev_op_leaf);
2916 if ((fd = open(vd->vdev_path, O_RDWR)) == -1)
2919 fsize = lseek(fd, 0, SEEK_END);
2920 (void) ftruncate(fd, *newsize);
2922 if (ztest_opts.zo_verbose >= 6) {
2923 (void) printf("%s grew from %lu to %lu bytes\n",
2924 vd->vdev_path, (ulong_t)fsize, (ulong_t)*newsize);
2931 * Callback function which expands a given vdev by calling vdev_online().
2935 online_vdev(vdev_t *vd, void *arg)
2937 spa_t *spa = vd->vdev_spa;
2938 vdev_t *tvd = vd->vdev_top;
2939 uint64_t guid = vd->vdev_guid;
2940 uint64_t generation = spa->spa_config_generation + 1;
2941 vdev_state_t newstate = VDEV_STATE_UNKNOWN;
2944 ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
2945 ASSERT(vd->vdev_ops->vdev_op_leaf);
2947 /* Calling vdev_online will initialize the new metaslabs */
2948 spa_config_exit(spa, SCL_STATE, spa);
2949 error = vdev_online(spa, guid, ZFS_ONLINE_EXPAND, &newstate);
2950 spa_config_enter(spa, SCL_STATE, spa, RW_READER);
2953 * If vdev_online returned an error or the underlying vdev_open
2954 * failed then we abort the expand. The only way to know that
2955 * vdev_open fails is by checking the returned newstate.
2957 if (error || newstate != VDEV_STATE_HEALTHY) {
2958 if (ztest_opts.zo_verbose >= 5) {
2959 (void) printf("Unable to expand vdev, state %llu, "
2960 "error %d\n", (u_longlong_t)newstate, error);
2964 ASSERT3U(newstate, ==, VDEV_STATE_HEALTHY);
2967 * Since we dropped the lock we need to ensure that we're
2968 * still talking to the original vdev. It's possible this
2969 * vdev may have been detached/replaced while we were
2970 * trying to online it.
2972 if (generation != spa->spa_config_generation) {
2973 if (ztest_opts.zo_verbose >= 5) {
2974 (void) printf("vdev configuration has changed, "
2975 "guid %llu, state %llu, expected gen %llu, "
2978 (u_longlong_t)tvd->vdev_state,
2979 (u_longlong_t)generation,
2980 (u_longlong_t)spa->spa_config_generation);
2988 * Traverse the vdev tree calling the supplied function.
2989 * We continue to walk the tree until we either have walked all
2990 * children or we receive a non-NULL return from the callback.
2991 * If a NULL callback is passed, then we just return back the first
2992 * leaf vdev we encounter.
2995 vdev_walk_tree(vdev_t *vd, vdev_t *(*func)(vdev_t *, void *), void *arg)
2997 if (vd->vdev_ops->vdev_op_leaf) {
3001 return (func(vd, arg));
3004 for (uint_t c = 0; c < vd->vdev_children; c++) {
3005 vdev_t *cvd = vd->vdev_child[c];
3006 if ((cvd = vdev_walk_tree(cvd, func, arg)) != NULL)
3013 * Verify that dynamic LUN growth works as expected.
3017 ztest_vdev_LUN_growth(ztest_ds_t *zd, uint64_t id)
3019 spa_t *spa = ztest_spa;
3021 metaslab_class_t *mc;
3022 metaslab_group_t *mg;
3023 size_t psize, newsize;
3025 uint64_t old_class_space, new_class_space, old_ms_count, new_ms_count;
3027 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
3028 spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3030 top = ztest_random_vdev_top(spa, B_TRUE);
3032 tvd = spa->spa_root_vdev->vdev_child[top];
3035 old_ms_count = tvd->vdev_ms_count;
3036 old_class_space = metaslab_class_get_space(mc);
3039 * Determine the size of the first leaf vdev associated with
3040 * our top-level device.
3042 vd = vdev_walk_tree(tvd, NULL, NULL);
3043 ASSERT3P(vd, !=, NULL);
3044 ASSERT(vd->vdev_ops->vdev_op_leaf);
3046 psize = vd->vdev_psize;
3049 * We only try to expand the vdev if it's healthy, less than 4x its
3050 * original size, and it has a valid psize.
3052 if (tvd->vdev_state != VDEV_STATE_HEALTHY ||
3053 psize == 0 || psize >= 4 * ztest_opts.zo_vdev_size) {
3054 spa_config_exit(spa, SCL_STATE, spa);
3055 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3059 newsize = psize + psize / 8;
3060 ASSERT3U(newsize, >, psize);
3062 if (ztest_opts.zo_verbose >= 6) {
3063 (void) printf("Expanding LUN %s from %lu to %lu\n",
3064 vd->vdev_path, (ulong_t)psize, (ulong_t)newsize);
3068 * Growing the vdev is a two step process:
3069 * 1). expand the physical size (i.e. relabel)
3070 * 2). online the vdev to create the new metaslabs
3072 if (vdev_walk_tree(tvd, grow_vdev, &newsize) != NULL ||
3073 vdev_walk_tree(tvd, online_vdev, NULL) != NULL ||
3074 tvd->vdev_state != VDEV_STATE_HEALTHY) {
3075 if (ztest_opts.zo_verbose >= 5) {
3076 (void) printf("Could not expand LUN because "
3077 "the vdev configuration changed.\n");
3079 spa_config_exit(spa, SCL_STATE, spa);
3080 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3084 spa_config_exit(spa, SCL_STATE, spa);
3087 * Expanding the LUN will update the config asynchronously,
3088 * thus we must wait for the async thread to complete any
3089 * pending tasks before proceeding.
3093 mutex_enter(&spa->spa_async_lock);
3094 done = (spa->spa_async_thread == NULL && !spa->spa_async_tasks);
3095 mutex_exit(&spa->spa_async_lock);
3098 txg_wait_synced(spa_get_dsl(spa), 0);
3099 (void) poll(NULL, 0, 100);
3102 spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3104 tvd = spa->spa_root_vdev->vdev_child[top];
3105 new_ms_count = tvd->vdev_ms_count;
3106 new_class_space = metaslab_class_get_space(mc);
3108 if (tvd->vdev_mg != mg || mg->mg_class != mc) {
3109 if (ztest_opts.zo_verbose >= 5) {
3110 (void) printf("Could not verify LUN expansion due to "
3111 "intervening vdev offline or remove.\n");
3113 spa_config_exit(spa, SCL_STATE, spa);
3114 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3119 * Make sure we were able to grow the vdev.
3121 if (new_ms_count <= old_ms_count)
3122 fatal(0, "LUN expansion failed: ms_count %llu <= %llu\n",
3123 old_ms_count, new_ms_count);
3126 * Make sure we were able to grow the pool.
3128 if (new_class_space <= old_class_space)
3129 fatal(0, "LUN expansion failed: class_space %llu <= %llu\n",
3130 old_class_space, new_class_space);
3132 if (ztest_opts.zo_verbose >= 5) {
3133 char oldnumbuf[6], newnumbuf[6];
3135 nicenum(old_class_space, oldnumbuf);
3136 nicenum(new_class_space, newnumbuf);
3137 (void) printf("%s grew from %s to %s\n",
3138 spa->spa_name, oldnumbuf, newnumbuf);
3141 spa_config_exit(spa, SCL_STATE, spa);
3142 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3146 * Verify that dmu_objset_{create,destroy,open,close} work as expected.
3150 ztest_objset_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
3153 * Create the objects common to all ztest datasets.
3155 VERIFY(zap_create_claim(os, ZTEST_DIROBJ,
3156 DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx) == 0);
3160 ztest_dataset_create(char *dsname)
3162 uint64_t zilset = ztest_random(100);
3163 int err = dmu_objset_create(dsname, DMU_OST_OTHER, 0,
3164 ztest_objset_create_cb, NULL);
3166 if (err || zilset < 80)
3169 if (ztest_opts.zo_verbose >= 6)
3170 (void) printf("Setting dataset %s to sync always\n", dsname);
3171 return (ztest_dsl_prop_set_uint64(dsname, ZFS_PROP_SYNC,
3172 ZFS_SYNC_ALWAYS, B_FALSE));
3177 ztest_objset_destroy_cb(const char *name, void *arg)
3180 dmu_object_info_t doi;
3184 * Verify that the dataset contains a directory object.
3186 VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_TRUE, FTAG, &os));
3187 error = dmu_object_info(os, ZTEST_DIROBJ, &doi);
3188 if (error != ENOENT) {
3189 /* We could have crashed in the middle of destroying it */
3191 ASSERT3U(doi.doi_type, ==, DMU_OT_ZAP_OTHER);
3192 ASSERT3S(doi.doi_physical_blocks_512, >=, 0);
3194 dmu_objset_disown(os, FTAG);
3197 * Destroy the dataset.
3199 if (strchr(name, '@') != NULL) {
3200 VERIFY0(dsl_destroy_snapshot(name, B_FALSE));
3202 VERIFY0(dsl_destroy_head(name));
3208 ztest_snapshot_create(char *osname, uint64_t id)
3210 char snapname[MAXNAMELEN];
3213 (void) snprintf(snapname, sizeof (snapname), "%llu", (u_longlong_t)id);
3215 error = dmu_objset_snapshot_one(osname, snapname);
3216 if (error == ENOSPC) {
3217 ztest_record_enospc(FTAG);
3220 if (error != 0 && error != EEXIST) {
3221 fatal(0, "ztest_snapshot_create(%s@%s) = %d", osname,
3228 ztest_snapshot_destroy(char *osname, uint64_t id)
3230 char snapname[MAXNAMELEN];
3233 (void) snprintf(snapname, MAXNAMELEN, "%s@%llu", osname,
3236 error = dsl_destroy_snapshot(snapname, B_FALSE);
3237 if (error != 0 && error != ENOENT)
3238 fatal(0, "ztest_snapshot_destroy(%s) = %d", snapname, error);
3244 ztest_dmu_objset_create_destroy(ztest_ds_t *zd, uint64_t id)
3250 char name[MAXNAMELEN];
3253 (void) rw_rdlock(&ztest_name_lock);
3255 (void) snprintf(name, MAXNAMELEN, "%s/temp_%llu",
3256 ztest_opts.zo_pool, (u_longlong_t)id);
3259 * If this dataset exists from a previous run, process its replay log
3260 * half of the time. If we don't replay it, then dmu_objset_destroy()
3261 * (invoked from ztest_objset_destroy_cb()) should just throw it away.
3263 if (ztest_random(2) == 0 &&
3264 dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os) == 0) {
3265 ztest_zd_init(&zdtmp, NULL, os);
3266 zil_replay(os, &zdtmp, ztest_replay_vector);
3267 ztest_zd_fini(&zdtmp);
3268 dmu_objset_disown(os, FTAG);
3272 * There may be an old instance of the dataset we're about to
3273 * create lying around from a previous run. If so, destroy it
3274 * and all of its snapshots.
3276 (void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
3277 DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
3280 * Verify that the destroyed dataset is no longer in the namespace.
3282 VERIFY3U(ENOENT, ==, dmu_objset_own(name, DMU_OST_OTHER, B_TRUE,
3286 * Verify that we can create a new dataset.
3288 error = ztest_dataset_create(name);
3290 if (error == ENOSPC) {
3291 ztest_record_enospc(FTAG);
3292 (void) rw_unlock(&ztest_name_lock);
3295 fatal(0, "dmu_objset_create(%s) = %d", name, error);
3298 VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os));
3300 ztest_zd_init(&zdtmp, NULL, os);
3303 * Open the intent log for it.
3305 zilog = zil_open(os, ztest_get_data);
3308 * Put some objects in there, do a little I/O to them,
3309 * and randomly take a couple of snapshots along the way.
3311 iters = ztest_random(5);
3312 for (int i = 0; i < iters; i++) {
3313 ztest_dmu_object_alloc_free(&zdtmp, id);
3314 if (ztest_random(iters) == 0)
3315 (void) ztest_snapshot_create(name, i);
3319 * Verify that we cannot create an existing dataset.
3321 VERIFY3U(EEXIST, ==,
3322 dmu_objset_create(name, DMU_OST_OTHER, 0, NULL, NULL));
3325 * Verify that we can hold an objset that is also owned.
3327 VERIFY3U(0, ==, dmu_objset_hold(name, FTAG, &os2));
3328 dmu_objset_rele(os2, FTAG);
3331 * Verify that we cannot own an objset that is already owned.
3334 dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os2));
3337 dmu_objset_disown(os, FTAG);
3338 ztest_zd_fini(&zdtmp);
3340 (void) rw_unlock(&ztest_name_lock);
3344 * Verify that dmu_snapshot_{create,destroy,open,close} work as expected.
3347 ztest_dmu_snapshot_create_destroy(ztest_ds_t *zd, uint64_t id)
3349 (void) rw_rdlock(&ztest_name_lock);
3350 (void) ztest_snapshot_destroy(zd->zd_name, id);
3351 (void) ztest_snapshot_create(zd->zd_name, id);
3352 (void) rw_unlock(&ztest_name_lock);
3356 * Cleanup non-standard snapshots and clones.
3359 ztest_dsl_dataset_cleanup(char *osname, uint64_t id)
3361 char snap1name[MAXNAMELEN];
3362 char clone1name[MAXNAMELEN];
3363 char snap2name[MAXNAMELEN];
3364 char clone2name[MAXNAMELEN];
3365 char snap3name[MAXNAMELEN];
3368 (void) snprintf(snap1name, MAXNAMELEN, "%s@s1_%llu", osname, id);
3369 (void) snprintf(clone1name, MAXNAMELEN, "%s/c1_%llu", osname, id);
3370 (void) snprintf(snap2name, MAXNAMELEN, "%s@s2_%llu", clone1name, id);
3371 (void) snprintf(clone2name, MAXNAMELEN, "%s/c2_%llu", osname, id);
3372 (void) snprintf(snap3name, MAXNAMELEN, "%s@s3_%llu", clone1name, id);
3374 error = dsl_destroy_head(clone2name);
3375 if (error && error != ENOENT)
3376 fatal(0, "dsl_destroy_head(%s) = %d", clone2name, error);
3377 error = dsl_destroy_snapshot(snap3name, B_FALSE);
3378 if (error && error != ENOENT)
3379 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap3name, error);
3380 error = dsl_destroy_snapshot(snap2name, B_FALSE);
3381 if (error && error != ENOENT)
3382 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap2name, error);
3383 error = dsl_destroy_head(clone1name);
3384 if (error && error != ENOENT)
3385 fatal(0, "dsl_destroy_head(%s) = %d", clone1name, error);
3386 error = dsl_destroy_snapshot(snap1name, B_FALSE);
3387 if (error && error != ENOENT)
3388 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap1name, error);
3392 * Verify dsl_dataset_promote handles EBUSY
3395 ztest_dsl_dataset_promote_busy(ztest_ds_t *zd, uint64_t id)
3398 char snap1name[MAXNAMELEN];
3399 char clone1name[MAXNAMELEN];
3400 char snap2name[MAXNAMELEN];
3401 char clone2name[MAXNAMELEN];
3402 char snap3name[MAXNAMELEN];
3403 char *osname = zd->zd_name;
3406 (void) rw_rdlock(&ztest_name_lock);
3408 ztest_dsl_dataset_cleanup(osname, id);
3410 (void) snprintf(snap1name, MAXNAMELEN, "%s@s1_%llu", osname, id);
3411 (void) snprintf(clone1name, MAXNAMELEN, "%s/c1_%llu", osname, id);
3412 (void) snprintf(snap2name, MAXNAMELEN, "%s@s2_%llu", clone1name, id);
3413 (void) snprintf(clone2name, MAXNAMELEN, "%s/c2_%llu", osname, id);
3414 (void) snprintf(snap3name, MAXNAMELEN, "%s@s3_%llu", clone1name, id);
3416 error = dmu_objset_snapshot_one(osname, strchr(snap1name, '@') + 1);
3417 if (error && error != EEXIST) {
3418 if (error == ENOSPC) {
3419 ztest_record_enospc(FTAG);
3422 fatal(0, "dmu_take_snapshot(%s) = %d", snap1name, error);
3425 error = dmu_objset_clone(clone1name, snap1name);
3427 if (error == ENOSPC) {
3428 ztest_record_enospc(FTAG);
3431 fatal(0, "dmu_objset_create(%s) = %d", clone1name, error);
3434 error = dmu_objset_snapshot_one(clone1name, strchr(snap2name, '@') + 1);
3435 if (error && error != EEXIST) {
3436 if (error == ENOSPC) {
3437 ztest_record_enospc(FTAG);
3440 fatal(0, "dmu_open_snapshot(%s) = %d", snap2name, error);
3443 error = dmu_objset_snapshot_one(clone1name, strchr(snap3name, '@') + 1);
3444 if (error && error != EEXIST) {
3445 if (error == ENOSPC) {
3446 ztest_record_enospc(FTAG);
3449 fatal(0, "dmu_open_snapshot(%s) = %d", snap3name, error);
3452 error = dmu_objset_clone(clone2name, snap3name);
3454 if (error == ENOSPC) {
3455 ztest_record_enospc(FTAG);
3458 fatal(0, "dmu_objset_create(%s) = %d", clone2name, error);
3461 error = dmu_objset_own(snap2name, DMU_OST_ANY, B_TRUE, FTAG, &os);
3463 fatal(0, "dmu_objset_own(%s) = %d", snap2name, error);
3464 error = dsl_dataset_promote(clone2name, NULL);
3466 fatal(0, "dsl_dataset_promote(%s), %d, not EBUSY", clone2name,
3468 dmu_objset_disown(os, FTAG);
3471 ztest_dsl_dataset_cleanup(osname, id);
3473 (void) rw_unlock(&ztest_name_lock);
3477 * Verify that dmu_object_{alloc,free} work as expected.
3480 ztest_dmu_object_alloc_free(ztest_ds_t *zd, uint64_t id)
3483 int batchsize = sizeof (od) / sizeof (od[0]);
3485 for (int b = 0; b < batchsize; b++)
3486 ztest_od_init(&od[b], id, FTAG, b, DMU_OT_UINT64_OTHER, 0, 0);
3489 * Destroy the previous batch of objects, create a new batch,
3490 * and do some I/O on the new objects.
3492 if (ztest_object_init(zd, od, sizeof (od), B_TRUE) != 0)
3495 while (ztest_random(4 * batchsize) != 0)
3496 ztest_io(zd, od[ztest_random(batchsize)].od_object,
3497 ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
3501 * Verify that dmu_{read,write} work as expected.
3504 ztest_dmu_read_write(ztest_ds_t *zd, uint64_t id)
3506 objset_t *os = zd->zd_os;
3509 int i, freeit, error;
3511 bufwad_t *packbuf, *bigbuf, *pack, *bigH, *bigT;
3512 uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
3513 uint64_t chunksize = (1000 + ztest_random(1000)) * sizeof (uint64_t);
3514 uint64_t regions = 997;
3515 uint64_t stride = 123456789ULL;
3516 uint64_t width = 40;
3517 int free_percent = 5;
3520 * This test uses two objects, packobj and bigobj, that are always
3521 * updated together (i.e. in the same tx) so that their contents are
3522 * in sync and can be compared. Their contents relate to each other
3523 * in a simple way: packobj is a dense array of 'bufwad' structures,
3524 * while bigobj is a sparse array of the same bufwads. Specifically,
3525 * for any index n, there are three bufwads that should be identical:
3527 * packobj, at offset n * sizeof (bufwad_t)
3528 * bigobj, at the head of the nth chunk
3529 * bigobj, at the tail of the nth chunk
3531 * The chunk size is arbitrary. It doesn't have to be a power of two,
3532 * and it doesn't have any relation to the object blocksize.
3533 * The only requirement is that it can hold at least two bufwads.
3535 * Normally, we write the bufwad to each of these locations.
3536 * However, free_percent of the time we instead write zeroes to
3537 * packobj and perform a dmu_free_range() on bigobj. By comparing
3538 * bigobj to packobj, we can verify that the DMU is correctly
3539 * tracking which parts of an object are allocated and free,
3540 * and that the contents of the allocated blocks are correct.
3544 * Read the directory info. If it's the first time, set things up.
3546 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, chunksize);
3547 ztest_od_init(&od[1], id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize);
3549 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
3552 bigobj = od[0].od_object;
3553 packobj = od[1].od_object;
3554 chunksize = od[0].od_gen;
3555 ASSERT(chunksize == od[1].od_gen);
3558 * Prefetch a random chunk of the big object.
3559 * Our aim here is to get some async reads in flight
3560 * for blocks that we may free below; the DMU should
3561 * handle this race correctly.
3563 n = ztest_random(regions) * stride + ztest_random(width);
3564 s = 1 + ztest_random(2 * width - 1);
3565 dmu_prefetch(os, bigobj, n * chunksize, s * chunksize);
3568 * Pick a random index and compute the offsets into packobj and bigobj.
3570 n = ztest_random(regions) * stride + ztest_random(width);
3571 s = 1 + ztest_random(width - 1);
3573 packoff = n * sizeof (bufwad_t);
3574 packsize = s * sizeof (bufwad_t);
3576 bigoff = n * chunksize;
3577 bigsize = s * chunksize;
3579 packbuf = umem_alloc(packsize, UMEM_NOFAIL);
3580 bigbuf = umem_alloc(bigsize, UMEM_NOFAIL);
3583 * free_percent of the time, free a range of bigobj rather than
3586 freeit = (ztest_random(100) < free_percent);
3589 * Read the current contents of our objects.
3591 error = dmu_read(os, packobj, packoff, packsize, packbuf,
3594 error = dmu_read(os, bigobj, bigoff, bigsize, bigbuf,
3599 * Get a tx for the mods to both packobj and bigobj.
3601 tx = dmu_tx_create(os);
3603 dmu_tx_hold_write(tx, packobj, packoff, packsize);
3606 dmu_tx_hold_free(tx, bigobj, bigoff, bigsize);
3608 dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
3610 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3612 umem_free(packbuf, packsize);
3613 umem_free(bigbuf, bigsize);
3617 dmu_object_set_checksum(os, bigobj,
3618 (enum zio_checksum)ztest_random_dsl_prop(ZFS_PROP_CHECKSUM), tx);
3620 dmu_object_set_compress(os, bigobj,
3621 (enum zio_compress)ztest_random_dsl_prop(ZFS_PROP_COMPRESSION), tx);
3624 * For each index from n to n + s, verify that the existing bufwad
3625 * in packobj matches the bufwads at the head and tail of the
3626 * corresponding chunk in bigobj. Then update all three bufwads
3627 * with the new values we want to write out.
3629 for (i = 0; i < s; i++) {
3631 pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
3633 bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
3635 bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
3637 ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
3638 ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
3640 if (pack->bw_txg > txg)
3641 fatal(0, "future leak: got %llx, open txg is %llx",
3644 if (pack->bw_data != 0 && pack->bw_index != n + i)
3645 fatal(0, "wrong index: got %llx, wanted %llx+%llx",
3646 pack->bw_index, n, i);
3648 if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
3649 fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
3651 if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
3652 fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
3655 bzero(pack, sizeof (bufwad_t));
3657 pack->bw_index = n + i;
3659 pack->bw_data = 1 + ztest_random(-2ULL);
3666 * We've verified all the old bufwads, and made new ones.
3667 * Now write them out.
3669 dmu_write(os, packobj, packoff, packsize, packbuf, tx);
3672 if (ztest_opts.zo_verbose >= 7) {
3673 (void) printf("freeing offset %llx size %llx"
3675 (u_longlong_t)bigoff,
3676 (u_longlong_t)bigsize,
3679 VERIFY(0 == dmu_free_range(os, bigobj, bigoff, bigsize, tx));
3681 if (ztest_opts.zo_verbose >= 7) {
3682 (void) printf("writing offset %llx size %llx"
3684 (u_longlong_t)bigoff,
3685 (u_longlong_t)bigsize,
3688 dmu_write(os, bigobj, bigoff, bigsize, bigbuf, tx);
3694 * Sanity check the stuff we just wrote.
3697 void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
3698 void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
3700 VERIFY(0 == dmu_read(os, packobj, packoff,
3701 packsize, packcheck, DMU_READ_PREFETCH));
3702 VERIFY(0 == dmu_read(os, bigobj, bigoff,
3703 bigsize, bigcheck, DMU_READ_PREFETCH));
3705 ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
3706 ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
3708 umem_free(packcheck, packsize);
3709 umem_free(bigcheck, bigsize);
3712 umem_free(packbuf, packsize);
3713 umem_free(bigbuf, bigsize);
3717 compare_and_update_pbbufs(uint64_t s, bufwad_t *packbuf, bufwad_t *bigbuf,
3718 uint64_t bigsize, uint64_t n, uint64_t chunksize, uint64_t txg)
3726 * For each index from n to n + s, verify that the existing bufwad
3727 * in packobj matches the bufwads at the head and tail of the
3728 * corresponding chunk in bigobj. Then update all three bufwads
3729 * with the new values we want to write out.
3731 for (i = 0; i < s; i++) {
3733 pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
3735 bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
3737 bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
3739 ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
3740 ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
3742 if (pack->bw_txg > txg)
3743 fatal(0, "future leak: got %llx, open txg is %llx",
3746 if (pack->bw_data != 0 && pack->bw_index != n + i)
3747 fatal(0, "wrong index: got %llx, wanted %llx+%llx",
3748 pack->bw_index, n, i);
3750 if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
3751 fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
3753 if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
3754 fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
3756 pack->bw_index = n + i;
3758 pack->bw_data = 1 + ztest_random(-2ULL);
3766 ztest_dmu_read_write_zcopy(ztest_ds_t *zd, uint64_t id)
3768 objset_t *os = zd->zd_os;
3774 bufwad_t *packbuf, *bigbuf;
3775 uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
3776 uint64_t blocksize = ztest_random_blocksize();
3777 uint64_t chunksize = blocksize;
3778 uint64_t regions = 997;
3779 uint64_t stride = 123456789ULL;
3781 dmu_buf_t *bonus_db;
3782 arc_buf_t **bigbuf_arcbufs;
3783 dmu_object_info_t doi;
3786 * This test uses two objects, packobj and bigobj, that are always
3787 * updated together (i.e. in the same tx) so that their contents are
3788 * in sync and can be compared. Their contents relate to each other
3789 * in a simple way: packobj is a dense array of 'bufwad' structures,
3790 * while bigobj is a sparse array of the same bufwads. Specifically,
3791 * for any index n, there are three bufwads that should be identical:
3793 * packobj, at offset n * sizeof (bufwad_t)
3794 * bigobj, at the head of the nth chunk
3795 * bigobj, at the tail of the nth chunk
3797 * The chunk size is set equal to bigobj block size so that
3798 * dmu_assign_arcbuf() can be tested for object updates.
3802 * Read the directory info. If it's the first time, set things up.
3804 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
3805 ztest_od_init(&od[1], id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize);
3807 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
3810 bigobj = od[0].od_object;
3811 packobj = od[1].od_object;
3812 blocksize = od[0].od_blocksize;
3813 chunksize = blocksize;
3814 ASSERT(chunksize == od[1].od_gen);
3816 VERIFY(dmu_object_info(os, bigobj, &doi) == 0);
3817 VERIFY(ISP2(doi.doi_data_block_size));
3818 VERIFY(chunksize == doi.doi_data_block_size);
3819 VERIFY(chunksize >= 2 * sizeof (bufwad_t));
3822 * Pick a random index and compute the offsets into packobj and bigobj.
3824 n = ztest_random(regions) * stride + ztest_random(width);
3825 s = 1 + ztest_random(width - 1);
3827 packoff = n * sizeof (bufwad_t);
3828 packsize = s * sizeof (bufwad_t);
3830 bigoff = n * chunksize;
3831 bigsize = s * chunksize;
3833 packbuf = umem_zalloc(packsize, UMEM_NOFAIL);
3834 bigbuf = umem_zalloc(bigsize, UMEM_NOFAIL);
3836 VERIFY3U(0, ==, dmu_bonus_hold(os, bigobj, FTAG, &bonus_db));
3838 bigbuf_arcbufs = umem_zalloc(2 * s * sizeof (arc_buf_t *), UMEM_NOFAIL);
3841 * Iteration 0 test zcopy for DB_UNCACHED dbufs.
3842 * Iteration 1 test zcopy to already referenced dbufs.
3843 * Iteration 2 test zcopy to dirty dbuf in the same txg.
3844 * Iteration 3 test zcopy to dbuf dirty in previous txg.
3845 * Iteration 4 test zcopy when dbuf is no longer dirty.
3846 * Iteration 5 test zcopy when it can't be done.
3847 * Iteration 6 one more zcopy write.
3849 for (i = 0; i < 7; i++) {
3854 * In iteration 5 (i == 5) use arcbufs
3855 * that don't match bigobj blksz to test
3856 * dmu_assign_arcbuf() when it can't directly
3857 * assign an arcbuf to a dbuf.
3859 for (j = 0; j < s; j++) {
3862 dmu_request_arcbuf(bonus_db, chunksize);
3864 bigbuf_arcbufs[2 * j] =
3865 dmu_request_arcbuf(bonus_db, chunksize / 2);
3866 bigbuf_arcbufs[2 * j + 1] =
3867 dmu_request_arcbuf(bonus_db, chunksize / 2);
3872 * Get a tx for the mods to both packobj and bigobj.
3874 tx = dmu_tx_create(os);
3876 dmu_tx_hold_write(tx, packobj, packoff, packsize);
3877 dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
3879 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3881 umem_free(packbuf, packsize);
3882 umem_free(bigbuf, bigsize);
3883 for (j = 0; j < s; j++) {
3885 dmu_return_arcbuf(bigbuf_arcbufs[j]);
3888 bigbuf_arcbufs[2 * j]);
3890 bigbuf_arcbufs[2 * j + 1]);
3893 umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
3894 dmu_buf_rele(bonus_db, FTAG);
3899 * 50% of the time don't read objects in the 1st iteration to
3900 * test dmu_assign_arcbuf() for the case when there're no
3901 * existing dbufs for the specified offsets.
3903 if (i != 0 || ztest_random(2) != 0) {
3904 error = dmu_read(os, packobj, packoff,
3905 packsize, packbuf, DMU_READ_PREFETCH);
3907 error = dmu_read(os, bigobj, bigoff, bigsize,
3908 bigbuf, DMU_READ_PREFETCH);
3911 compare_and_update_pbbufs(s, packbuf, bigbuf, bigsize,
3915 * We've verified all the old bufwads, and made new ones.
3916 * Now write them out.
3918 dmu_write(os, packobj, packoff, packsize, packbuf, tx);
3919 if (ztest_opts.zo_verbose >= 7) {
3920 (void) printf("writing offset %llx size %llx"
3922 (u_longlong_t)bigoff,
3923 (u_longlong_t)bigsize,
3926 for (off = bigoff, j = 0; j < s; j++, off += chunksize) {
3929 bcopy((caddr_t)bigbuf + (off - bigoff),
3930 bigbuf_arcbufs[j]->b_data, chunksize);
3932 bcopy((caddr_t)bigbuf + (off - bigoff),
3933 bigbuf_arcbufs[2 * j]->b_data,
3935 bcopy((caddr_t)bigbuf + (off - bigoff) +
3937 bigbuf_arcbufs[2 * j + 1]->b_data,
3942 VERIFY(dmu_buf_hold(os, bigobj, off,
3943 FTAG, &dbt, DMU_READ_NO_PREFETCH) == 0);
3946 dmu_assign_arcbuf(bonus_db, off,
3947 bigbuf_arcbufs[j], tx);
3949 dmu_assign_arcbuf(bonus_db, off,
3950 bigbuf_arcbufs[2 * j], tx);
3951 dmu_assign_arcbuf(bonus_db,
3952 off + chunksize / 2,
3953 bigbuf_arcbufs[2 * j + 1], tx);
3956 dmu_buf_rele(dbt, FTAG);
3962 * Sanity check the stuff we just wrote.
3965 void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
3966 void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
3968 VERIFY(0 == dmu_read(os, packobj, packoff,
3969 packsize, packcheck, DMU_READ_PREFETCH));
3970 VERIFY(0 == dmu_read(os, bigobj, bigoff,
3971 bigsize, bigcheck, DMU_READ_PREFETCH));
3973 ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
3974 ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
3976 umem_free(packcheck, packsize);
3977 umem_free(bigcheck, bigsize);
3980 txg_wait_open(dmu_objset_pool(os), 0);
3981 } else if (i == 3) {
3982 txg_wait_synced(dmu_objset_pool(os), 0);
3986 dmu_buf_rele(bonus_db, FTAG);
3987 umem_free(packbuf, packsize);
3988 umem_free(bigbuf, bigsize);
3989 umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
3994 ztest_dmu_write_parallel(ztest_ds_t *zd, uint64_t id)
3997 uint64_t offset = (1ULL << (ztest_random(20) + 43)) +
3998 (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
4001 * Have multiple threads write to large offsets in an object
4002 * to verify that parallel writes to an object -- even to the
4003 * same blocks within the object -- doesn't cause any trouble.
4005 ztest_od_init(&od[0], ID_PARALLEL, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
4007 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4010 while (ztest_random(10) != 0)
4011 ztest_io(zd, od[0].od_object, offset);
4015 ztest_dmu_prealloc(ztest_ds_t *zd, uint64_t id)
4018 uint64_t offset = (1ULL << (ztest_random(4) + SPA_MAXBLOCKSHIFT)) +
4019 (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
4020 uint64_t count = ztest_random(20) + 1;
4021 uint64_t blocksize = ztest_random_blocksize();
4024 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
4026 if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4029 if (ztest_truncate(zd, od[0].od_object, offset, count * blocksize) != 0)
4032 ztest_prealloc(zd, od[0].od_object, offset, count * blocksize);
4034 data = umem_zalloc(blocksize, UMEM_NOFAIL);
4036 while (ztest_random(count) != 0) {
4037 uint64_t randoff = offset + (ztest_random(count) * blocksize);
4038 if (ztest_write(zd, od[0].od_object, randoff, blocksize,
4041 while (ztest_random(4) != 0)
4042 ztest_io(zd, od[0].od_object, randoff);
4045 umem_free(data, blocksize);
4049 * Verify that zap_{create,destroy,add,remove,update} work as expected.
4051 #define ZTEST_ZAP_MIN_INTS 1
4052 #define ZTEST_ZAP_MAX_INTS 4
4053 #define ZTEST_ZAP_MAX_PROPS 1000
4056 ztest_zap(ztest_ds_t *zd, uint64_t id)
4058 objset_t *os = zd->zd_os;
4061 uint64_t txg, last_txg;
4062 uint64_t value[ZTEST_ZAP_MAX_INTS];
4063 uint64_t zl_ints, zl_intsize, prop;
4066 char propname[100], txgname[100];
4068 char *hc[2] = { "s.acl.h", ".s.open.h.hyLZlg" };
4070 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0);
4072 if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4075 object = od[0].od_object;
4078 * Generate a known hash collision, and verify that
4079 * we can lookup and remove both entries.
4081 tx = dmu_tx_create(os);
4082 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4083 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4086 for (i = 0; i < 2; i++) {
4088 VERIFY3U(0, ==, zap_add(os, object, hc[i], sizeof (uint64_t),
4091 for (i = 0; i < 2; i++) {
4092 VERIFY3U(EEXIST, ==, zap_add(os, object, hc[i],
4093 sizeof (uint64_t), 1, &value[i], tx));
4095 zap_length(os, object, hc[i], &zl_intsize, &zl_ints));
4096 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4097 ASSERT3U(zl_ints, ==, 1);
4099 for (i = 0; i < 2; i++) {
4100 VERIFY3U(0, ==, zap_remove(os, object, hc[i], tx));
4105 * Generate a buch of random entries.
4107 ints = MAX(ZTEST_ZAP_MIN_INTS, object % ZTEST_ZAP_MAX_INTS);
4109 prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
4110 (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
4111 (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
4112 bzero(value, sizeof (value));
4116 * If these zap entries already exist, validate their contents.
4118 error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
4120 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4121 ASSERT3U(zl_ints, ==, 1);
4123 VERIFY(zap_lookup(os, object, txgname, zl_intsize,
4124 zl_ints, &last_txg) == 0);
4126 VERIFY(zap_length(os, object, propname, &zl_intsize,
4129 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4130 ASSERT3U(zl_ints, ==, ints);
4132 VERIFY(zap_lookup(os, object, propname, zl_intsize,
4133 zl_ints, value) == 0);
4135 for (i = 0; i < ints; i++) {
4136 ASSERT3U(value[i], ==, last_txg + object + i);
4139 ASSERT3U(error, ==, ENOENT);
4143 * Atomically update two entries in our zap object.
4144 * The first is named txg_%llu, and contains the txg
4145 * in which the property was last updated. The second
4146 * is named prop_%llu, and the nth element of its value
4147 * should be txg + object + n.
4149 tx = dmu_tx_create(os);
4150 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4151 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4156 fatal(0, "zap future leak: old %llu new %llu", last_txg, txg);
4158 for (i = 0; i < ints; i++)
4159 value[i] = txg + object + i;
4161 VERIFY3U(0, ==, zap_update(os, object, txgname, sizeof (uint64_t),
4163 VERIFY3U(0, ==, zap_update(os, object, propname, sizeof (uint64_t),
4169 * Remove a random pair of entries.
4171 prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
4172 (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
4173 (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
4175 error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
4177 if (error == ENOENT)
4182 tx = dmu_tx_create(os);
4183 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4184 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4187 VERIFY3U(0, ==, zap_remove(os, object, txgname, tx));
4188 VERIFY3U(0, ==, zap_remove(os, object, propname, tx));
4193 * Testcase to test the upgrading of a microzap to fatzap.
4196 ztest_fzap(ztest_ds_t *zd, uint64_t id)
4198 objset_t *os = zd->zd_os;
4200 uint64_t object, txg;
4202 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0);
4204 if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4207 object = od[0].od_object;
4210 * Add entries to this ZAP and make sure it spills over
4211 * and gets upgraded to a fatzap. Also, since we are adding
4212 * 2050 entries we should see ptrtbl growth and leaf-block split.
4214 for (int i = 0; i < 2050; i++) {
4215 char name[MAXNAMELEN];
4220 (void) snprintf(name, sizeof (name), "fzap-%llu-%llu",
4223 tx = dmu_tx_create(os);
4224 dmu_tx_hold_zap(tx, object, B_TRUE, name);
4225 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4228 error = zap_add(os, object, name, sizeof (uint64_t), 1,
4230 ASSERT(error == 0 || error == EEXIST);
4237 ztest_zap_parallel(ztest_ds_t *zd, uint64_t id)
4239 objset_t *os = zd->zd_os;
4241 uint64_t txg, object, count, wsize, wc, zl_wsize, zl_wc;
4243 int i, namelen, error;
4244 int micro = ztest_random(2);
4245 char name[20], string_value[20];
4248 ztest_od_init(&od[0], ID_PARALLEL, FTAG, micro, DMU_OT_ZAP_OTHER, 0, 0);
4250 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4253 object = od[0].od_object;
4256 * Generate a random name of the form 'xxx.....' where each
4257 * x is a random printable character and the dots are dots.
4258 * There are 94 such characters, and the name length goes from
4259 * 6 to 20, so there are 94^3 * 15 = 12,458,760 possible names.
4261 namelen = ztest_random(sizeof (name) - 5) + 5 + 1;
4263 for (i = 0; i < 3; i++)
4264 name[i] = '!' + ztest_random('~' - '!' + 1);
4265 for (; i < namelen - 1; i++)
4269 if ((namelen & 1) || micro) {
4270 wsize = sizeof (txg);
4276 data = string_value;
4280 VERIFY0(zap_count(os, object, &count));
4281 ASSERT(count != -1ULL);
4284 * Select an operation: length, lookup, add, update, remove.
4286 i = ztest_random(5);
4289 tx = dmu_tx_create(os);
4290 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4291 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4294 bcopy(name, string_value, namelen);
4298 bzero(string_value, namelen);
4304 error = zap_length(os, object, name, &zl_wsize, &zl_wc);
4306 ASSERT3U(wsize, ==, zl_wsize);
4307 ASSERT3U(wc, ==, zl_wc);
4309 ASSERT3U(error, ==, ENOENT);
4314 error = zap_lookup(os, object, name, wsize, wc, data);
4316 if (data == string_value &&
4317 bcmp(name, data, namelen) != 0)
4318 fatal(0, "name '%s' != val '%s' len %d",
4319 name, data, namelen);
4321 ASSERT3U(error, ==, ENOENT);
4326 error = zap_add(os, object, name, wsize, wc, data, tx);
4327 ASSERT(error == 0 || error == EEXIST);
4331 VERIFY(zap_update(os, object, name, wsize, wc, data, tx) == 0);
4335 error = zap_remove(os, object, name, tx);
4336 ASSERT(error == 0 || error == ENOENT);
4345 * Commit callback data.
4347 typedef struct ztest_cb_data {
4348 list_node_t zcd_node;
4350 int zcd_expected_err;
4351 boolean_t zcd_added;
4352 boolean_t zcd_called;
4356 /* This is the actual commit callback function */
4358 ztest_commit_callback(void *arg, int error)
4360 ztest_cb_data_t *data = arg;
4361 uint64_t synced_txg;
4363 VERIFY(data != NULL);
4364 VERIFY3S(data->zcd_expected_err, ==, error);
4365 VERIFY(!data->zcd_called);
4367 synced_txg = spa_last_synced_txg(data->zcd_spa);
4368 if (data->zcd_txg > synced_txg)
4369 fatal(0, "commit callback of txg %" PRIu64 " called prematurely"
4370 ", last synced txg = %" PRIu64 "\n", data->zcd_txg,
4373 data->zcd_called = B_TRUE;
4375 if (error == ECANCELED) {
4376 ASSERT0(data->zcd_txg);
4377 ASSERT(!data->zcd_added);
4380 * The private callback data should be destroyed here, but
4381 * since we are going to check the zcd_called field after
4382 * dmu_tx_abort(), we will destroy it there.
4387 /* Was this callback added to the global callback list? */
4388 if (!data->zcd_added)
4391 ASSERT3U(data->zcd_txg, !=, 0);
4393 /* Remove our callback from the list */
4394 (void) mutex_lock(&zcl.zcl_callbacks_lock);
4395 list_remove(&zcl.zcl_callbacks, data);
4396 (void) mutex_unlock(&zcl.zcl_callbacks_lock);
4399 umem_free(data, sizeof (ztest_cb_data_t));
4402 /* Allocate and initialize callback data structure */
4403 static ztest_cb_data_t *
4404 ztest_create_cb_data(objset_t *os, uint64_t txg)
4406 ztest_cb_data_t *cb_data;
4408 cb_data = umem_zalloc(sizeof (ztest_cb_data_t), UMEM_NOFAIL);
4410 cb_data->zcd_txg = txg;
4411 cb_data->zcd_spa = dmu_objset_spa(os);
4417 * If a number of txgs equal to this threshold have been created after a commit
4418 * callback has been registered but not called, then we assume there is an
4419 * implementation bug.
4421 #define ZTEST_COMMIT_CALLBACK_THRESH (TXG_CONCURRENT_STATES + 2)
4424 * Commit callback test.
4427 ztest_dmu_commit_callbacks(ztest_ds_t *zd, uint64_t id)
4429 objset_t *os = zd->zd_os;
4432 ztest_cb_data_t *cb_data[3], *tmp_cb;
4433 uint64_t old_txg, txg;
4436 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
4438 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4441 tx = dmu_tx_create(os);
4443 cb_data[0] = ztest_create_cb_data(os, 0);
4444 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[0]);
4446 dmu_tx_hold_write(tx, od[0].od_object, 0, sizeof (uint64_t));
4448 /* Every once in a while, abort the transaction on purpose */
4449 if (ztest_random(100) == 0)
4453 error = dmu_tx_assign(tx, TXG_NOWAIT);
4455 txg = error ? 0 : dmu_tx_get_txg(tx);
4457 cb_data[0]->zcd_txg = txg;
4458 cb_data[1] = ztest_create_cb_data(os, txg);
4459 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[1]);
4463 * It's not a strict requirement to call the registered
4464 * callbacks from inside dmu_tx_abort(), but that's what
4465 * it's supposed to happen in the current implementation
4466 * so we will check for that.
4468 for (i = 0; i < 2; i++) {
4469 cb_data[i]->zcd_expected_err = ECANCELED;
4470 VERIFY(!cb_data[i]->zcd_called);
4475 for (i = 0; i < 2; i++) {
4476 VERIFY(cb_data[i]->zcd_called);
4477 umem_free(cb_data[i], sizeof (ztest_cb_data_t));
4483 cb_data[2] = ztest_create_cb_data(os, txg);
4484 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[2]);
4487 * Read existing data to make sure there isn't a future leak.
4489 VERIFY(0 == dmu_read(os, od[0].od_object, 0, sizeof (uint64_t),
4490 &old_txg, DMU_READ_PREFETCH));
4493 fatal(0, "future leak: got %" PRIu64 ", open txg is %" PRIu64,
4496 dmu_write(os, od[0].od_object, 0, sizeof (uint64_t), &txg, tx);
4498 (void) mutex_lock(&zcl.zcl_callbacks_lock);
4501 * Since commit callbacks don't have any ordering requirement and since
4502 * it is theoretically possible for a commit callback to be called
4503 * after an arbitrary amount of time has elapsed since its txg has been
4504 * synced, it is difficult to reliably determine whether a commit
4505 * callback hasn't been called due to high load or due to a flawed
4508 * In practice, we will assume that if after a certain number of txgs a
4509 * commit callback hasn't been called, then most likely there's an
4510 * implementation bug..
4512 tmp_cb = list_head(&zcl.zcl_callbacks);
4513 if (tmp_cb != NULL &&
4514 (txg - ZTEST_COMMIT_CALLBACK_THRESH) > tmp_cb->zcd_txg) {
4515 fatal(0, "Commit callback threshold exceeded, oldest txg: %"
4516 PRIu64 ", open txg: %" PRIu64 "\n", tmp_cb->zcd_txg, txg);
4520 * Let's find the place to insert our callbacks.
4522 * Even though the list is ordered by txg, it is possible for the
4523 * insertion point to not be the end because our txg may already be
4524 * quiescing at this point and other callbacks in the open txg
4525 * (from other objsets) may have sneaked in.
4527 tmp_cb = list_tail(&zcl.zcl_callbacks);
4528 while (tmp_cb != NULL && tmp_cb->zcd_txg > txg)
4529 tmp_cb = list_prev(&zcl.zcl_callbacks, tmp_cb);
4531 /* Add the 3 callbacks to the list */
4532 for (i = 0; i < 3; i++) {
4534 list_insert_head(&zcl.zcl_callbacks, cb_data[i]);
4536 list_insert_after(&zcl.zcl_callbacks, tmp_cb,
4539 cb_data[i]->zcd_added = B_TRUE;
4540 VERIFY(!cb_data[i]->zcd_called);
4542 tmp_cb = cb_data[i];
4545 (void) mutex_unlock(&zcl.zcl_callbacks_lock);
4552 ztest_dsl_prop_get_set(ztest_ds_t *zd, uint64_t id)
4554 zfs_prop_t proplist[] = {
4556 ZFS_PROP_COMPRESSION,
4561 (void) rw_rdlock(&ztest_name_lock);
4563 for (int p = 0; p < sizeof (proplist) / sizeof (proplist[0]); p++)
4564 (void) ztest_dsl_prop_set_uint64(zd->zd_name, proplist[p],
4565 ztest_random_dsl_prop(proplist[p]), (int)ztest_random(2));
4567 (void) rw_unlock(&ztest_name_lock);
4572 ztest_spa_prop_get_set(ztest_ds_t *zd, uint64_t id)
4574 nvlist_t *props = NULL;
4576 (void) rw_rdlock(&ztest_name_lock);
4578 (void) ztest_spa_prop_set_uint64(ZPOOL_PROP_DEDUPDITTO,
4579 ZIO_DEDUPDITTO_MIN + ztest_random(ZIO_DEDUPDITTO_MIN));
4581 VERIFY0(spa_prop_get(ztest_spa, &props));
4583 if (ztest_opts.zo_verbose >= 6)
4584 dump_nvlist(props, 4);
4588 (void) rw_unlock(&ztest_name_lock);
4592 user_release_one(const char *snapname, const char *holdname)
4594 nvlist_t *snaps, *holds;
4597 snaps = fnvlist_alloc();
4598 holds = fnvlist_alloc();
4599 fnvlist_add_boolean(holds, holdname);
4600 fnvlist_add_nvlist(snaps, snapname, holds);
4601 fnvlist_free(holds);
4602 error = dsl_dataset_user_release(snaps, NULL);
4603 fnvlist_free(snaps);
4608 * Test snapshot hold/release and deferred destroy.
4611 ztest_dmu_snapshot_hold(ztest_ds_t *zd, uint64_t id)
4614 objset_t *os = zd->zd_os;
4618 char clonename[100];
4620 char osname[MAXNAMELEN];
4623 (void) rw_rdlock(&ztest_name_lock);
4625 dmu_objset_name(os, osname);
4627 (void) snprintf(snapname, sizeof (snapname), "sh1_%llu", id);
4628 (void) snprintf(fullname, sizeof (fullname), "%s@%s", osname, snapname);
4629 (void) snprintf(clonename, sizeof (clonename),
4630 "%s/ch1_%llu", osname, id);
4631 (void) snprintf(tag, sizeof (tag), "tag_%llu", id);
4634 * Clean up from any previous run.
4636 error = dsl_destroy_head(clonename);
4637 if (error != ENOENT)
4639 error = user_release_one(fullname, tag);
4640 if (error != ESRCH && error != ENOENT)
4642 error = dsl_destroy_snapshot(fullname, B_FALSE);
4643 if (error != ENOENT)
4647 * Create snapshot, clone it, mark snap for deferred destroy,
4648 * destroy clone, verify snap was also destroyed.
4650 error = dmu_objset_snapshot_one(osname, snapname);
4652 if (error == ENOSPC) {
4653 ztest_record_enospc("dmu_objset_snapshot");
4656 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
4659 error = dmu_objset_clone(clonename, fullname);
4661 if (error == ENOSPC) {
4662 ztest_record_enospc("dmu_objset_clone");
4665 fatal(0, "dmu_objset_clone(%s) = %d", clonename, error);
4668 error = dsl_destroy_snapshot(fullname, B_TRUE);
4670 fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
4674 error = dsl_destroy_head(clonename);
4676 fatal(0, "dsl_destroy_head(%s) = %d", clonename, error);
4678 error = dmu_objset_hold(fullname, FTAG, &origin);
4679 if (error != ENOENT)
4680 fatal(0, "dmu_objset_hold(%s) = %d", fullname, error);
4683 * Create snapshot, add temporary hold, verify that we can't
4684 * destroy a held snapshot, mark for deferred destroy,
4685 * release hold, verify snapshot was destroyed.
4687 error = dmu_objset_snapshot_one(osname, snapname);
4689 if (error == ENOSPC) {
4690 ztest_record_enospc("dmu_objset_snapshot");
4693 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
4696 holds = fnvlist_alloc();
4697 fnvlist_add_string(holds, fullname, tag);
4698 error = dsl_dataset_user_hold(holds, 0, NULL);
4699 fnvlist_free(holds);
4702 fatal(0, "dsl_dataset_user_hold(%s)", fullname, tag);
4704 error = dsl_destroy_snapshot(fullname, B_FALSE);
4705 if (error != EBUSY) {
4706 fatal(0, "dsl_destroy_snapshot(%s, B_FALSE) = %d",
4710 error = dsl_destroy_snapshot(fullname, B_TRUE);
4712 fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
4716 error = user_release_one(fullname, tag);
4718 fatal(0, "user_release_one(%s, %s) = %d", fullname, tag, error);
4720 VERIFY3U(dmu_objset_hold(fullname, FTAG, &origin), ==, ENOENT);
4723 (void) rw_unlock(&ztest_name_lock);
4727 * Inject random faults into the on-disk data.
4731 ztest_fault_inject(ztest_ds_t *zd, uint64_t id)
4733 ztest_shared_t *zs = ztest_shared;
4734 spa_t *spa = ztest_spa;
4738 uint64_t bad = 0x1990c0ffeedecadeULL;
4740 char path0[MAXPATHLEN];
4741 char pathrand[MAXPATHLEN];
4743 int bshift = SPA_MAXBLOCKSHIFT + 2; /* don't scrog all labels */
4749 boolean_t islog = B_FALSE;
4751 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
4752 maxfaults = MAXFAULTS();
4753 leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz;
4754 mirror_save = zs->zs_mirrors;
4755 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4757 ASSERT(leaves >= 1);
4760 * Grab the name lock as reader. There are some operations
4761 * which don't like to have their vdevs changed while
4762 * they are in progress (i.e. spa_change_guid). Those
4763 * operations will have grabbed the name lock as writer.
4765 (void) rw_rdlock(&ztest_name_lock);
4768 * We need SCL_STATE here because we're going to look at vd0->vdev_tsd.
4770 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
4772 if (ztest_random(2) == 0) {
4774 * Inject errors on a normal data device or slog device.
4776 top = ztest_random_vdev_top(spa, B_TRUE);
4777 leaf = ztest_random(leaves) + zs->zs_splits;
4780 * Generate paths to the first leaf in this top-level vdev,
4781 * and to the random leaf we selected. We'll induce transient
4782 * write failures and random online/offline activity on leaf 0,
4783 * and we'll write random garbage to the randomly chosen leaf.
4785 (void) snprintf(path0, sizeof (path0), ztest_dev_template,
4786 ztest_opts.zo_dir, ztest_opts.zo_pool,
4787 top * leaves + zs->zs_splits);
4788 (void) snprintf(pathrand, sizeof (pathrand), ztest_dev_template,
4789 ztest_opts.zo_dir, ztest_opts.zo_pool,
4790 top * leaves + leaf);
4792 vd0 = vdev_lookup_by_path(spa->spa_root_vdev, path0);
4793 if (vd0 != NULL && vd0->vdev_top->vdev_islog)
4797 * If the top-level vdev needs to be resilvered
4798 * then we only allow faults on the device that is
4801 if (vd0 != NULL && maxfaults != 1 &&
4802 (!vdev_resilver_needed(vd0->vdev_top, NULL, NULL) ||
4803 vd0->vdev_resilvering)) {
4805 * Make vd0 explicitly claim to be unreadable,
4806 * or unwriteable, or reach behind its back
4807 * and close the underlying fd. We can do this if
4808 * maxfaults == 0 because we'll fail and reexecute,
4809 * and we can do it if maxfaults >= 2 because we'll
4810 * have enough redundancy. If maxfaults == 1, the
4811 * combination of this with injection of random data
4812 * corruption below exceeds the pool's fault tolerance.
4814 vdev_file_t *vf = vd0->vdev_tsd;
4816 if (vf != NULL && ztest_random(3) == 0) {
4817 (void) close(vf->vf_vnode->v_fd);
4818 vf->vf_vnode->v_fd = -1;
4819 } else if (ztest_random(2) == 0) {
4820 vd0->vdev_cant_read = B_TRUE;
4822 vd0->vdev_cant_write = B_TRUE;
4824 guid0 = vd0->vdev_guid;
4828 * Inject errors on an l2cache device.
4830 spa_aux_vdev_t *sav = &spa->spa_l2cache;
4832 if (sav->sav_count == 0) {
4833 spa_config_exit(spa, SCL_STATE, FTAG);
4834 (void) rw_unlock(&ztest_name_lock);
4837 vd0 = sav->sav_vdevs[ztest_random(sav->sav_count)];
4838 guid0 = vd0->vdev_guid;
4839 (void) strcpy(path0, vd0->vdev_path);
4840 (void) strcpy(pathrand, vd0->vdev_path);
4844 maxfaults = INT_MAX; /* no limit on cache devices */
4847 spa_config_exit(spa, SCL_STATE, FTAG);
4848 (void) rw_unlock(&ztest_name_lock);
4851 * If we can tolerate two or more faults, or we're dealing
4852 * with a slog, randomly online/offline vd0.
4854 if ((maxfaults >= 2 || islog) && guid0 != 0) {
4855 if (ztest_random(10) < 6) {
4856 int flags = (ztest_random(2) == 0 ?
4857 ZFS_OFFLINE_TEMPORARY : 0);
4860 * We have to grab the zs_name_lock as writer to
4861 * prevent a race between offlining a slog and
4862 * destroying a dataset. Offlining the slog will
4863 * grab a reference on the dataset which may cause
4864 * dmu_objset_destroy() to fail with EBUSY thus
4865 * leaving the dataset in an inconsistent state.
4868 (void) rw_wrlock(&ztest_name_lock);
4870 VERIFY(vdev_offline(spa, guid0, flags) != EBUSY);
4873 (void) rw_unlock(&ztest_name_lock);
4876 * Ideally we would like to be able to randomly
4877 * call vdev_[on|off]line without holding locks
4878 * to force unpredictable failures but the side
4879 * effects of vdev_[on|off]line prevent us from
4880 * doing so. We grab the ztest_vdev_lock here to
4881 * prevent a race between injection testing and
4884 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
4885 (void) vdev_online(spa, guid0, 0, NULL);
4886 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4894 * We have at least single-fault tolerance, so inject data corruption.
4896 fd = open(pathrand, O_RDWR);
4898 if (fd == -1) /* we hit a gap in the device namespace */
4901 fsize = lseek(fd, 0, SEEK_END);
4903 while (--iters != 0) {
4904 offset = ztest_random(fsize / (leaves << bshift)) *
4905 (leaves << bshift) + (leaf << bshift) +
4906 (ztest_random(1ULL << (bshift - 1)) & -8ULL);
4908 if (offset >= fsize)
4911 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
4912 if (mirror_save != zs->zs_mirrors) {
4913 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4918 if (pwrite(fd, &bad, sizeof (bad), offset) != sizeof (bad))
4919 fatal(1, "can't inject bad word at 0x%llx in %s",
4922 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4924 if (ztest_opts.zo_verbose >= 7)
4925 (void) printf("injected bad word into %s,"
4926 " offset 0x%llx\n", pathrand, (u_longlong_t)offset);
4933 * Verify that DDT repair works as expected.
4936 ztest_ddt_repair(ztest_ds_t *zd, uint64_t id)
4938 ztest_shared_t *zs = ztest_shared;
4939 spa_t *spa = ztest_spa;
4940 objset_t *os = zd->zd_os;
4942 uint64_t object, blocksize, txg, pattern, psize;
4943 enum zio_checksum checksum = spa_dedup_checksum(spa);
4948 int copies = 2 * ZIO_DEDUPDITTO_MIN;
4950 blocksize = ztest_random_blocksize();
4951 blocksize = MIN(blocksize, 2048); /* because we write so many */
4953 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
4955 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4959 * Take the name lock as writer to prevent anyone else from changing
4960 * the pool and dataset properies we need to maintain during this test.
4962 (void) rw_wrlock(&ztest_name_lock);
4964 if (ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_DEDUP, checksum,
4966 ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_COPIES, 1,
4968 (void) rw_unlock(&ztest_name_lock);
4972 object = od[0].od_object;
4973 blocksize = od[0].od_blocksize;
4974 pattern = zs->zs_guid ^ dmu_objset_fsid_guid(os);
4976 ASSERT(object != 0);
4978 tx = dmu_tx_create(os);
4979 dmu_tx_hold_write(tx, object, 0, copies * blocksize);
4980 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
4982 (void) rw_unlock(&ztest_name_lock);
4987 * Write all the copies of our block.
4989 for (int i = 0; i < copies; i++) {
4990 uint64_t offset = i * blocksize;
4991 int error = dmu_buf_hold(os, object, offset, FTAG, &db,
4992 DMU_READ_NO_PREFETCH);
4994 fatal(B_FALSE, "dmu_buf_hold(%p, %llu, %llu) = %u",
4995 os, (long long)object, (long long) offset, error);
4997 ASSERT(db->db_offset == offset);
4998 ASSERT(db->db_size == blocksize);
4999 ASSERT(ztest_pattern_match(db->db_data, db->db_size, pattern) ||
5000 ztest_pattern_match(db->db_data, db->db_size, 0ULL));
5001 dmu_buf_will_fill(db, tx);
5002 ztest_pattern_set(db->db_data, db->db_size, pattern);
5003 dmu_buf_rele(db, FTAG);
5007 txg_wait_synced(spa_get_dsl(spa), txg);
5010 * Find out what block we got.
5012 VERIFY0(dmu_buf_hold(os, object, 0, FTAG, &db,
5013 DMU_READ_NO_PREFETCH));
5014 blk = *((dmu_buf_impl_t *)db)->db_blkptr;
5015 dmu_buf_rele(db, FTAG);
5018 * Damage the block. Dedup-ditto will save us when we read it later.
5020 psize = BP_GET_PSIZE(&blk);
5021 buf = zio_buf_alloc(psize);
5022 ztest_pattern_set(buf, psize, ~pattern);
5024 (void) zio_wait(zio_rewrite(NULL, spa, 0, &blk,
5025 buf, psize, NULL, NULL, ZIO_PRIORITY_SYNC_WRITE,
5026 ZIO_FLAG_CANFAIL | ZIO_FLAG_INDUCE_DAMAGE, NULL));
5028 zio_buf_free(buf, psize);
5030 (void) rw_unlock(&ztest_name_lock);
5038 ztest_scrub(ztest_ds_t *zd, uint64_t id)
5040 spa_t *spa = ztest_spa;
5042 (void) spa_scan(spa, POOL_SCAN_SCRUB);
5043 (void) poll(NULL, 0, 100); /* wait a moment, then force a restart */
5044 (void) spa_scan(spa, POOL_SCAN_SCRUB);
5048 * Change the guid for the pool.
5052 ztest_reguid(ztest_ds_t *zd, uint64_t id)
5054 spa_t *spa = ztest_spa;
5055 uint64_t orig, load;
5058 orig = spa_guid(spa);
5059 load = spa_load_guid(spa);
5061 (void) rw_wrlock(&ztest_name_lock);
5062 error = spa_change_guid(spa);
5063 (void) rw_unlock(&ztest_name_lock);
5068 if (ztest_opts.zo_verbose >= 4) {
5069 (void) printf("Changed guid old %llu -> %llu\n",
5070 (u_longlong_t)orig, (u_longlong_t)spa_guid(spa));
5073 VERIFY3U(orig, !=, spa_guid(spa));
5074 VERIFY3U(load, ==, spa_load_guid(spa));
5078 * Rename the pool to a different name and then rename it back.
5082 ztest_spa_rename(ztest_ds_t *zd, uint64_t id)
5084 char *oldname, *newname;
5087 (void) rw_wrlock(&ztest_name_lock);
5089 oldname = ztest_opts.zo_pool;
5090 newname = umem_alloc(strlen(oldname) + 5, UMEM_NOFAIL);
5091 (void) strcpy(newname, oldname);
5092 (void) strcat(newname, "_tmp");
5097 VERIFY3U(0, ==, spa_rename(oldname, newname));
5100 * Try to open it under the old name, which shouldn't exist
5102 VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
5105 * Open it under the new name and make sure it's still the same spa_t.
5107 VERIFY3U(0, ==, spa_open(newname, &spa, FTAG));
5109 ASSERT(spa == ztest_spa);
5110 spa_close(spa, FTAG);
5113 * Rename it back to the original
5115 VERIFY3U(0, ==, spa_rename(newname, oldname));
5118 * Make sure it can still be opened
5120 VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG));
5122 ASSERT(spa == ztest_spa);
5123 spa_close(spa, FTAG);
5125 umem_free(newname, strlen(newname) + 1);
5127 (void) rw_unlock(&ztest_name_lock);
5131 * Verify pool integrity by running zdb.
5134 ztest_run_zdb(char *pool)
5137 char zdb[MAXPATHLEN + MAXNAMELEN + 20];
5145 strlcpy(zdb, "/usr/bin/ztest", sizeof(zdb));
5147 /* zdb lives in /usr/sbin, while ztest lives in /usr/bin */
5148 bin = strstr(zdb, "/usr/bin/");
5149 ztest = strstr(bin, "/ztest");
5151 isalen = ztest - isa;
5155 "/usr/sbin%.*s/zdb -bcc%s%s -U %s %s",
5158 ztest_opts.zo_verbose >= 3 ? "s" : "",
5159 ztest_opts.zo_verbose >= 4 ? "v" : "",
5164 if (ztest_opts.zo_verbose >= 5)
5165 (void) printf("Executing %s\n", strstr(zdb, "zdb "));
5167 fp = popen(zdb, "r");
5170 while (fgets(zbuf, sizeof (zbuf), fp) != NULL)
5171 if (ztest_opts.zo_verbose >= 3)
5172 (void) printf("%s", zbuf);
5174 status = pclose(fp);
5179 ztest_dump_core = 0;
5180 if (WIFEXITED(status))
5181 fatal(0, "'%s' exit code %d", zdb, WEXITSTATUS(status));
5183 fatal(0, "'%s' died with signal %d", zdb, WTERMSIG(status));
5187 ztest_walk_pool_directory(char *header)
5191 if (ztest_opts.zo_verbose >= 6)
5192 (void) printf("%s\n", header);
5194 mutex_enter(&spa_namespace_lock);
5195 while ((spa = spa_next(spa)) != NULL)
5196 if (ztest_opts.zo_verbose >= 6)
5197 (void) printf("\t%s\n", spa_name(spa));
5198 mutex_exit(&spa_namespace_lock);
5202 ztest_spa_import_export(char *oldname, char *newname)
5204 nvlist_t *config, *newconfig;
5209 if (ztest_opts.zo_verbose >= 4) {
5210 (void) printf("import/export: old = %s, new = %s\n",
5215 * Clean up from previous runs.
5217 (void) spa_destroy(newname);
5220 * Get the pool's configuration and guid.
5222 VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG));
5225 * Kick off a scrub to tickle scrub/export races.
5227 if (ztest_random(2) == 0)
5228 (void) spa_scan(spa, POOL_SCAN_SCRUB);
5230 pool_guid = spa_guid(spa);
5231 spa_close(spa, FTAG);
5233 ztest_walk_pool_directory("pools before export");
5238 VERIFY3U(0, ==, spa_export(oldname, &config, B_FALSE, B_FALSE));
5240 ztest_walk_pool_directory("pools after export");
5245 newconfig = spa_tryimport(config);
5246 ASSERT(newconfig != NULL);
5247 nvlist_free(newconfig);
5250 * Import it under the new name.
5252 error = spa_import(newname, config, NULL, 0);
5254 dump_nvlist(config, 0);
5255 fatal(B_FALSE, "couldn't import pool %s as %s: error %u",
5256 oldname, newname, error);
5259 ztest_walk_pool_directory("pools after import");
5262 * Try to import it again -- should fail with EEXIST.
5264 VERIFY3U(EEXIST, ==, spa_import(newname, config, NULL, 0));
5267 * Try to import it under a different name -- should fail with EEXIST.
5269 VERIFY3U(EEXIST, ==, spa_import(oldname, config, NULL, 0));
5272 * Verify that the pool is no longer visible under the old name.
5274 VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
5277 * Verify that we can open and close the pool using the new name.
5279 VERIFY3U(0, ==, spa_open(newname, &spa, FTAG));
5280 ASSERT(pool_guid == spa_guid(spa));
5281 spa_close(spa, FTAG);
5283 nvlist_free(config);
5287 ztest_resume(spa_t *spa)
5289 if (spa_suspended(spa) && ztest_opts.zo_verbose >= 6)
5290 (void) printf("resuming from suspended state\n");
5291 spa_vdev_state_enter(spa, SCL_NONE);
5292 vdev_clear(spa, NULL);
5293 (void) spa_vdev_state_exit(spa, NULL, 0);
5294 (void) zio_resume(spa);
5298 ztest_resume_thread(void *arg)
5302 while (!ztest_exiting) {
5303 if (spa_suspended(spa))
5305 (void) poll(NULL, 0, 100);
5311 ztest_deadman_thread(void *arg)
5313 ztest_shared_t *zs = arg;
5314 spa_t *spa = ztest_spa;
5315 hrtime_t delta, total = 0;
5318 delta = (zs->zs_thread_stop - zs->zs_thread_start) /
5319 NANOSEC + zfs_deadman_synctime;
5321 (void) poll(NULL, 0, (int)(1000 * delta));
5324 * If the pool is suspended then fail immediately. Otherwise,
5325 * check to see if the pool is making any progress. If
5326 * vdev_deadman() discovers that there hasn't been any recent
5327 * I/Os then it will end up aborting the tests.
5329 if (spa_suspended(spa)) {
5330 fatal(0, "aborting test after %llu seconds because "
5331 "pool has transitioned to a suspended state.",
5332 zfs_deadman_synctime);
5335 vdev_deadman(spa->spa_root_vdev);
5337 total += zfs_deadman_synctime;
5338 (void) printf("ztest has been running for %lld seconds\n",
5344 ztest_execute(int test, ztest_info_t *zi, uint64_t id)
5346 ztest_ds_t *zd = &ztest_ds[id % ztest_opts.zo_datasets];
5347 ztest_shared_callstate_t *zc = ZTEST_GET_SHARED_CALLSTATE(test);
5348 hrtime_t functime = gethrtime();
5350 for (int i = 0; i < zi->zi_iters; i++)
5351 zi->zi_func(zd, id);
5353 functime = gethrtime() - functime;
5355 atomic_add_64(&zc->zc_count, 1);
5356 atomic_add_64(&zc->zc_time, functime);
5358 if (ztest_opts.zo_verbose >= 4) {
5360 (void) dladdr((void *)zi->zi_func, &dli);
5361 (void) printf("%6.2f sec in %s\n",
5362 (double)functime / NANOSEC, dli.dli_sname);
5367 ztest_thread(void *arg)
5370 uint64_t id = (uintptr_t)arg;
5371 ztest_shared_t *zs = ztest_shared;
5375 ztest_shared_callstate_t *zc;
5377 while ((now = gethrtime()) < zs->zs_thread_stop) {
5379 * See if it's time to force a crash.
5381 if (now > zs->zs_thread_kill)
5385 * If we're getting ENOSPC with some regularity, stop.
5387 if (zs->zs_enospc_count > 10)
5391 * Pick a random function to execute.
5393 rand = ztest_random(ZTEST_FUNCS);
5394 zi = &ztest_info[rand];
5395 zc = ZTEST_GET_SHARED_CALLSTATE(rand);
5396 call_next = zc->zc_next;
5398 if (now >= call_next &&
5399 atomic_cas_64(&zc->zc_next, call_next, call_next +
5400 ztest_random(2 * zi->zi_interval[0] + 1)) == call_next) {
5401 ztest_execute(rand, zi, id);
5409 ztest_dataset_name(char *dsname, char *pool, int d)
5411 (void) snprintf(dsname, MAXNAMELEN, "%s/ds_%d", pool, d);
5415 ztest_dataset_destroy(int d)
5417 char name[MAXNAMELEN];
5419 ztest_dataset_name(name, ztest_opts.zo_pool, d);
5421 if (ztest_opts.zo_verbose >= 3)
5422 (void) printf("Destroying %s to free up space\n", name);
5425 * Cleanup any non-standard clones and snapshots. In general,
5426 * ztest thread t operates on dataset (t % zopt_datasets),
5427 * so there may be more than one thing to clean up.
5429 for (int t = d; t < ztest_opts.zo_threads;
5430 t += ztest_opts.zo_datasets) {
5431 ztest_dsl_dataset_cleanup(name, t);
5434 (void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
5435 DS_FIND_SNAPSHOTS | DS_FIND_CHILDREN);
5439 ztest_dataset_dirobj_verify(ztest_ds_t *zd)
5441 uint64_t usedobjs, dirobjs, scratch;
5444 * ZTEST_DIROBJ is the object directory for the entire dataset.
5445 * Therefore, the number of objects in use should equal the
5446 * number of ZTEST_DIROBJ entries, +1 for ZTEST_DIROBJ itself.
5447 * If not, we have an object leak.
5449 * Note that we can only check this in ztest_dataset_open(),
5450 * when the open-context and syncing-context values agree.
5451 * That's because zap_count() returns the open-context value,
5452 * while dmu_objset_space() returns the rootbp fill count.
5454 VERIFY3U(0, ==, zap_count(zd->zd_os, ZTEST_DIROBJ, &dirobjs));
5455 dmu_objset_space(zd->zd_os, &scratch, &scratch, &usedobjs, &scratch);
5456 ASSERT3U(dirobjs + 1, ==, usedobjs);
5460 ztest_dataset_open(int d)
5462 ztest_ds_t *zd = &ztest_ds[d];
5463 uint64_t committed_seq = ZTEST_GET_SHARED_DS(d)->zd_seq;
5466 char name[MAXNAMELEN];
5469 ztest_dataset_name(name, ztest_opts.zo_pool, d);
5471 (void) rw_rdlock(&ztest_name_lock);
5473 error = ztest_dataset_create(name);
5474 if (error == ENOSPC) {
5475 (void) rw_unlock(&ztest_name_lock);
5476 ztest_record_enospc(FTAG);
5479 ASSERT(error == 0 || error == EEXIST);
5481 VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, zd, &os));
5482 (void) rw_unlock(&ztest_name_lock);
5484 ztest_zd_init(zd, ZTEST_GET_SHARED_DS(d), os);
5486 zilog = zd->zd_zilog;
5488 if (zilog->zl_header->zh_claim_lr_seq != 0 &&
5489 zilog->zl_header->zh_claim_lr_seq < committed_seq)
5490 fatal(0, "missing log records: claimed %llu < committed %llu",
5491 zilog->zl_header->zh_claim_lr_seq, committed_seq);
5493 ztest_dataset_dirobj_verify(zd);
5495 zil_replay(os, zd, ztest_replay_vector);
5497 ztest_dataset_dirobj_verify(zd);
5499 if (ztest_opts.zo_verbose >= 6)
5500 (void) printf("%s replay %llu blocks, %llu records, seq %llu\n",
5502 (u_longlong_t)zilog->zl_parse_blk_count,
5503 (u_longlong_t)zilog->zl_parse_lr_count,
5504 (u_longlong_t)zilog->zl_replaying_seq);
5506 zilog = zil_open(os, ztest_get_data);
5508 if (zilog->zl_replaying_seq != 0 &&
5509 zilog->zl_replaying_seq < committed_seq)
5510 fatal(0, "missing log records: replayed %llu < committed %llu",
5511 zilog->zl_replaying_seq, committed_seq);
5517 ztest_dataset_close(int d)
5519 ztest_ds_t *zd = &ztest_ds[d];
5521 zil_close(zd->zd_zilog);
5522 dmu_objset_disown(zd->zd_os, zd);
5528 * Kick off threads to run tests on all datasets in parallel.
5531 ztest_run(ztest_shared_t *zs)
5536 thread_t resume_tid;
5539 ztest_exiting = B_FALSE;
5542 * Initialize parent/child shared state.
5544 VERIFY(_mutex_init(&ztest_vdev_lock, USYNC_THREAD, NULL) == 0);
5545 VERIFY(rwlock_init(&ztest_name_lock, USYNC_THREAD, NULL) == 0);
5547 zs->zs_thread_start = gethrtime();
5548 zs->zs_thread_stop =
5549 zs->zs_thread_start + ztest_opts.zo_passtime * NANOSEC;
5550 zs->zs_thread_stop = MIN(zs->zs_thread_stop, zs->zs_proc_stop);
5551 zs->zs_thread_kill = zs->zs_thread_stop;
5552 if (ztest_random(100) < ztest_opts.zo_killrate) {
5553 zs->zs_thread_kill -=
5554 ztest_random(ztest_opts.zo_passtime * NANOSEC);
5557 (void) _mutex_init(&zcl.zcl_callbacks_lock, USYNC_THREAD, NULL);
5559 list_create(&zcl.zcl_callbacks, sizeof (ztest_cb_data_t),
5560 offsetof(ztest_cb_data_t, zcd_node));
5565 kernel_init(FREAD | FWRITE);
5566 VERIFY0(spa_open(ztest_opts.zo_pool, &spa, FTAG));
5567 spa->spa_debug = B_TRUE;
5570 VERIFY0(dmu_objset_own(ztest_opts.zo_pool,
5571 DMU_OST_ANY, B_TRUE, FTAG, &os));
5572 zs->zs_guid = dmu_objset_fsid_guid(os);
5573 dmu_objset_disown(os, FTAG);
5575 spa->spa_dedup_ditto = 2 * ZIO_DEDUPDITTO_MIN;
5578 * We don't expect the pool to suspend unless maxfaults == 0,
5579 * in which case ztest_fault_inject() temporarily takes away
5580 * the only valid replica.
5582 if (MAXFAULTS() == 0)
5583 spa->spa_failmode = ZIO_FAILURE_MODE_WAIT;
5585 spa->spa_failmode = ZIO_FAILURE_MODE_PANIC;
5588 * Create a thread to periodically resume suspended I/O.
5590 VERIFY(thr_create(0, 0, ztest_resume_thread, spa, THR_BOUND,
5594 * Create a deadman thread to abort() if we hang.
5596 VERIFY(thr_create(0, 0, ztest_deadman_thread, zs, THR_BOUND,
5600 * Verify that we can safely inquire about about any object,
5601 * whether it's allocated or not. To make it interesting,
5602 * we probe a 5-wide window around each power of two.
5603 * This hits all edge cases, including zero and the max.
5605 for (int t = 0; t < 64; t++) {
5606 for (int d = -5; d <= 5; d++) {
5607 error = dmu_object_info(spa->spa_meta_objset,
5608 (1ULL << t) + d, NULL);
5609 ASSERT(error == 0 || error == ENOENT ||
5615 * If we got any ENOSPC errors on the previous run, destroy something.
5617 if (zs->zs_enospc_count != 0) {
5618 int d = ztest_random(ztest_opts.zo_datasets);
5619 ztest_dataset_destroy(d);
5621 zs->zs_enospc_count = 0;
5623 tid = umem_zalloc(ztest_opts.zo_threads * sizeof (thread_t),
5626 if (ztest_opts.zo_verbose >= 4)
5627 (void) printf("starting main threads...\n");
5630 * Kick off all the tests that run in parallel.
5632 for (int t = 0; t < ztest_opts.zo_threads; t++) {
5633 if (t < ztest_opts.zo_datasets &&
5634 ztest_dataset_open(t) != 0)
5636 VERIFY(thr_create(0, 0, ztest_thread, (void *)(uintptr_t)t,
5637 THR_BOUND, &tid[t]) == 0);
5641 * Wait for all of the tests to complete. We go in reverse order
5642 * so we don't close datasets while threads are still using them.
5644 for (int t = ztest_opts.zo_threads - 1; t >= 0; t--) {
5645 VERIFY(thr_join(tid[t], NULL, NULL) == 0);
5646 if (t < ztest_opts.zo_datasets)
5647 ztest_dataset_close(t);
5650 txg_wait_synced(spa_get_dsl(spa), 0);
5652 zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(spa));
5653 zs->zs_space = metaslab_class_get_space(spa_normal_class(spa));
5655 umem_free(tid, ztest_opts.zo_threads * sizeof (thread_t));
5657 /* Kill the resume thread */
5658 ztest_exiting = B_TRUE;
5659 VERIFY(thr_join(resume_tid, NULL, NULL) == 0);
5663 * Right before closing the pool, kick off a bunch of async I/O;
5664 * spa_close() should wait for it to complete.
5666 for (uint64_t object = 1; object < 50; object++)
5667 dmu_prefetch(spa->spa_meta_objset, object, 0, 1ULL << 20);
5669 spa_close(spa, FTAG);
5672 * Verify that we can loop over all pools.
5674 mutex_enter(&spa_namespace_lock);
5675 for (spa = spa_next(NULL); spa != NULL; spa = spa_next(spa))
5676 if (ztest_opts.zo_verbose > 3)
5677 (void) printf("spa_next: found %s\n", spa_name(spa));
5678 mutex_exit(&spa_namespace_lock);
5681 * Verify that we can export the pool and reimport it under a
5684 if (ztest_random(2) == 0) {
5685 char name[MAXNAMELEN];
5686 (void) snprintf(name, MAXNAMELEN, "%s_import",
5687 ztest_opts.zo_pool);
5688 ztest_spa_import_export(ztest_opts.zo_pool, name);
5689 ztest_spa_import_export(name, ztest_opts.zo_pool);
5694 list_destroy(&zcl.zcl_callbacks);
5696 (void) _mutex_destroy(&zcl.zcl_callbacks_lock);
5698 (void) rwlock_destroy(&ztest_name_lock);
5699 (void) _mutex_destroy(&ztest_vdev_lock);
5705 ztest_ds_t *zd = &ztest_ds[0];
5709 if (ztest_opts.zo_verbose >= 3)
5710 (void) printf("testing spa_freeze()...\n");
5712 kernel_init(FREAD | FWRITE);
5713 VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
5714 VERIFY3U(0, ==, ztest_dataset_open(0));
5715 spa->spa_debug = B_TRUE;
5719 * Force the first log block to be transactionally allocated.
5720 * We have to do this before we freeze the pool -- otherwise
5721 * the log chain won't be anchored.
5723 while (BP_IS_HOLE(&zd->zd_zilog->zl_header->zh_log)) {
5724 ztest_dmu_object_alloc_free(zd, 0);
5725 zil_commit(zd->zd_zilog, 0);
5728 txg_wait_synced(spa_get_dsl(spa), 0);
5731 * Freeze the pool. This stops spa_sync() from doing anything,
5732 * so that the only way to record changes from now on is the ZIL.
5737 * Run tests that generate log records but don't alter the pool config
5738 * or depend on DSL sync tasks (snapshots, objset create/destroy, etc).
5739 * We do a txg_wait_synced() after each iteration to force the txg
5740 * to increase well beyond the last synced value in the uberblock.
5741 * The ZIL should be OK with that.
5743 while (ztest_random(10) != 0 &&
5744 numloops++ < ztest_opts.zo_maxloops) {
5745 ztest_dmu_write_parallel(zd, 0);
5746 ztest_dmu_object_alloc_free(zd, 0);
5747 txg_wait_synced(spa_get_dsl(spa), 0);
5751 * Commit all of the changes we just generated.
5753 zil_commit(zd->zd_zilog, 0);
5754 txg_wait_synced(spa_get_dsl(spa), 0);
5757 * Close our dataset and close the pool.
5759 ztest_dataset_close(0);
5760 spa_close(spa, FTAG);
5764 * Open and close the pool and dataset to induce log replay.
5766 kernel_init(FREAD | FWRITE);
5767 VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
5768 ASSERT(spa_freeze_txg(spa) == UINT64_MAX);
5769 VERIFY3U(0, ==, ztest_dataset_open(0));
5770 ztest_dataset_close(0);
5772 spa->spa_debug = B_TRUE;
5774 txg_wait_synced(spa_get_dsl(spa), 0);
5775 ztest_reguid(NULL, 0);
5777 spa_close(spa, FTAG);
5782 print_time(hrtime_t t, char *timebuf)
5784 hrtime_t s = t / NANOSEC;
5785 hrtime_t m = s / 60;
5786 hrtime_t h = m / 60;
5787 hrtime_t d = h / 24;
5796 (void) sprintf(timebuf,
5797 "%llud%02lluh%02llum%02llus", d, h, m, s);
5799 (void) sprintf(timebuf, "%lluh%02llum%02llus", h, m, s);
5801 (void) sprintf(timebuf, "%llum%02llus", m, s);
5803 (void) sprintf(timebuf, "%llus", s);
5811 VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
5812 if (ztest_random(2) == 0)
5814 VERIFY(nvlist_add_uint64(props, "autoreplace", 1) == 0);
5820 * Create a storage pool with the given name and initial vdev size.
5821 * Then test spa_freeze() functionality.
5824 ztest_init(ztest_shared_t *zs)
5827 nvlist_t *nvroot, *props;
5829 VERIFY(_mutex_init(&ztest_vdev_lock, USYNC_THREAD, NULL) == 0);
5830 VERIFY(rwlock_init(&ztest_name_lock, USYNC_THREAD, NULL) == 0);
5832 kernel_init(FREAD | FWRITE);
5835 * Create the storage pool.
5837 (void) spa_destroy(ztest_opts.zo_pool);
5838 ztest_shared->zs_vdev_next_leaf = 0;
5840 zs->zs_mirrors = ztest_opts.zo_mirrors;
5841 nvroot = make_vdev_root(NULL, NULL, NULL, ztest_opts.zo_vdev_size, 0,
5842 0, ztest_opts.zo_raidz, zs->zs_mirrors, 1);
5843 props = make_random_props();
5844 for (int i = 0; i < SPA_FEATURES; i++) {
5846 (void) snprintf(buf, sizeof (buf), "feature@%s",
5847 spa_feature_table[i].fi_uname);
5848 VERIFY3U(0, ==, nvlist_add_uint64(props, buf, 0));
5850 VERIFY3U(0, ==, spa_create(ztest_opts.zo_pool, nvroot, props, NULL));
5851 nvlist_free(nvroot);
5853 VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
5854 zs->zs_metaslab_sz =
5855 1ULL << spa->spa_root_vdev->vdev_child[0]->vdev_ms_shift;
5857 spa_close(spa, FTAG);
5861 ztest_run_zdb(ztest_opts.zo_pool);
5865 ztest_run_zdb(ztest_opts.zo_pool);
5867 (void) rwlock_destroy(&ztest_name_lock);
5868 (void) _mutex_destroy(&ztest_vdev_lock);
5874 static char ztest_name_data[] = "/tmp/ztest.data.XXXXXX";
5876 ztest_fd_data = mkstemp(ztest_name_data);
5877 ASSERT3S(ztest_fd_data, >=, 0);
5878 (void) unlink(ztest_name_data);
5883 shared_data_size(ztest_shared_hdr_t *hdr)
5887 size = hdr->zh_hdr_size;
5888 size += hdr->zh_opts_size;
5889 size += hdr->zh_size;
5890 size += hdr->zh_stats_size * hdr->zh_stats_count;
5891 size += hdr->zh_ds_size * hdr->zh_ds_count;
5900 ztest_shared_hdr_t *hdr;
5902 hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
5903 PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
5904 ASSERT(hdr != MAP_FAILED);
5906 VERIFY3U(0, ==, ftruncate(ztest_fd_data, sizeof (ztest_shared_hdr_t)));
5908 hdr->zh_hdr_size = sizeof (ztest_shared_hdr_t);
5909 hdr->zh_opts_size = sizeof (ztest_shared_opts_t);
5910 hdr->zh_size = sizeof (ztest_shared_t);
5911 hdr->zh_stats_size = sizeof (ztest_shared_callstate_t);
5912 hdr->zh_stats_count = ZTEST_FUNCS;
5913 hdr->zh_ds_size = sizeof (ztest_shared_ds_t);
5914 hdr->zh_ds_count = ztest_opts.zo_datasets;
5916 size = shared_data_size(hdr);
5917 VERIFY3U(0, ==, ftruncate(ztest_fd_data, size));
5919 (void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
5926 ztest_shared_hdr_t *hdr;
5929 hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
5930 PROT_READ, MAP_SHARED, ztest_fd_data, 0);
5931 ASSERT(hdr != MAP_FAILED);
5933 size = shared_data_size(hdr);
5935 (void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
5936 hdr = ztest_shared_hdr = (void *)mmap(0, P2ROUNDUP(size, getpagesize()),
5937 PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
5938 ASSERT(hdr != MAP_FAILED);
5939 buf = (uint8_t *)hdr;
5941 offset = hdr->zh_hdr_size;
5942 ztest_shared_opts = (void *)&buf[offset];
5943 offset += hdr->zh_opts_size;
5944 ztest_shared = (void *)&buf[offset];
5945 offset += hdr->zh_size;
5946 ztest_shared_callstate = (void *)&buf[offset];
5947 offset += hdr->zh_stats_size * hdr->zh_stats_count;
5948 ztest_shared_ds = (void *)&buf[offset];
5952 exec_child(char *cmd, char *libpath, boolean_t ignorekill, int *statusp)
5956 char *cmdbuf = NULL;
5961 cmdbuf = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
5962 (void) strlcpy(cmdbuf, getexecname(), MAXPATHLEN);
5967 fatal(1, "fork failed");
5969 if (pid == 0) { /* child */
5970 char *emptyargv[2] = { cmd, NULL };
5971 char fd_data_str[12];
5973 struct rlimit rl = { 1024, 1024 };
5974 (void) setrlimit(RLIMIT_NOFILE, &rl);
5976 (void) close(ztest_fd_rand);
5978 snprintf(fd_data_str, 12, "%d", ztest_fd_data));
5979 VERIFY0(setenv("ZTEST_FD_DATA", fd_data_str, 1));
5981 (void) enable_extended_FILE_stdio(-1, -1);
5982 if (libpath != NULL)
5983 VERIFY(0 == setenv("LD_LIBRARY_PATH", libpath, 1));
5985 (void) execv(cmd, emptyargv);
5987 (void) execvp(cmd, emptyargv);
5989 ztest_dump_core = B_FALSE;
5990 fatal(B_TRUE, "exec failed: %s", cmd);
5993 if (cmdbuf != NULL) {
5994 umem_free(cmdbuf, MAXPATHLEN);
5998 while (waitpid(pid, &status, 0) != pid)
6000 if (statusp != NULL)
6003 if (WIFEXITED(status)) {
6004 if (WEXITSTATUS(status) != 0) {
6005 (void) fprintf(stderr, "child exited with code %d\n",
6006 WEXITSTATUS(status));
6010 } else if (WIFSIGNALED(status)) {
6011 if (!ignorekill || WTERMSIG(status) != SIGKILL) {
6012 (void) fprintf(stderr, "child died with signal %d\n",
6018 (void) fprintf(stderr, "something strange happened to child\n");
6025 ztest_run_init(void)
6027 ztest_shared_t *zs = ztest_shared;
6029 ASSERT(ztest_opts.zo_init != 0);
6032 * Blow away any existing copy of zpool.cache
6034 (void) remove(spa_config_path);
6037 * Create and initialize our storage pool.
6039 for (int i = 1; i <= ztest_opts.zo_init; i++) {
6040 bzero(zs, sizeof (ztest_shared_t));
6041 if (ztest_opts.zo_verbose >= 3 &&
6042 ztest_opts.zo_init != 1) {
6043 (void) printf("ztest_init(), pass %d\n", i);
6050 main(int argc, char **argv)
6058 ztest_shared_callstate_t *zc;
6064 char *fd_data_str = getenv("ZTEST_FD_DATA");
6066 (void) setvbuf(stdout, NULL, _IOLBF, 0);
6068 dprintf_setup(&argc, argv);
6069 zfs_deadman_synctime = 300;
6071 ztest_fd_rand = open("/dev/urandom", O_RDONLY);
6072 ASSERT3S(ztest_fd_rand, >=, 0);
6075 process_options(argc, argv);
6080 bcopy(&ztest_opts, ztest_shared_opts,
6081 sizeof (*ztest_shared_opts));
6083 ztest_fd_data = atoi(fd_data_str);
6085 bcopy(ztest_shared_opts, &ztest_opts, sizeof (ztest_opts));
6087 ASSERT3U(ztest_opts.zo_datasets, ==, ztest_shared_hdr->zh_ds_count);
6089 /* Override location of zpool.cache */
6090 VERIFY3U(asprintf((char **)&spa_config_path, "%s/zpool.cache",
6091 ztest_opts.zo_dir), !=, -1);
6093 ztest_ds = umem_alloc(ztest_opts.zo_datasets * sizeof (ztest_ds_t),
6098 metaslab_gang_bang = ztest_opts.zo_metaslab_gang_bang;
6099 metaslab_df_alloc_threshold =
6100 zs->zs_metaslab_df_alloc_threshold;
6109 hasalt = (strlen(ztest_opts.zo_alt_ztest) != 0);
6111 if (ztest_opts.zo_verbose >= 1) {
6112 (void) printf("%llu vdevs, %d datasets, %d threads,"
6113 " %llu seconds...\n",
6114 (u_longlong_t)ztest_opts.zo_vdevs,
6115 ztest_opts.zo_datasets,
6116 ztest_opts.zo_threads,
6117 (u_longlong_t)ztest_opts.zo_time);
6120 cmd = umem_alloc(MAXNAMELEN, UMEM_NOFAIL);
6121 (void) strlcpy(cmd, getexecname(), MAXNAMELEN);
6123 zs->zs_do_init = B_TRUE;
6124 if (strlen(ztest_opts.zo_alt_ztest) != 0) {
6125 if (ztest_opts.zo_verbose >= 1) {
6126 (void) printf("Executing older ztest for "
6127 "initialization: %s\n", ztest_opts.zo_alt_ztest);
6129 VERIFY(!exec_child(ztest_opts.zo_alt_ztest,
6130 ztest_opts.zo_alt_libpath, B_FALSE, NULL));
6132 VERIFY(!exec_child(NULL, NULL, B_FALSE, NULL));
6134 zs->zs_do_init = B_FALSE;
6136 zs->zs_proc_start = gethrtime();
6137 zs->zs_proc_stop = zs->zs_proc_start + ztest_opts.zo_time * NANOSEC;
6139 for (int f = 0; f < ZTEST_FUNCS; f++) {
6140 zi = &ztest_info[f];
6141 zc = ZTEST_GET_SHARED_CALLSTATE(f);
6142 if (zs->zs_proc_start + zi->zi_interval[0] > zs->zs_proc_stop)
6143 zc->zc_next = UINT64_MAX;
6145 zc->zc_next = zs->zs_proc_start +
6146 ztest_random(2 * zi->zi_interval[0] + 1);
6150 * Run the tests in a loop. These tests include fault injection
6151 * to verify that self-healing data works, and forced crashes
6152 * to verify that we never lose on-disk consistency.
6154 while (gethrtime() < zs->zs_proc_stop) {
6159 * Initialize the workload counters for each function.
6161 for (int f = 0; f < ZTEST_FUNCS; f++) {
6162 zc = ZTEST_GET_SHARED_CALLSTATE(f);
6167 /* Set the allocation switch size */
6168 zs->zs_metaslab_df_alloc_threshold =
6169 ztest_random(zs->zs_metaslab_sz / 4) + 1;
6171 if (!hasalt || ztest_random(2) == 0) {
6172 if (hasalt && ztest_opts.zo_verbose >= 1) {
6173 (void) printf("Executing newer ztest: %s\n",
6177 killed = exec_child(cmd, NULL, B_TRUE, &status);
6179 if (hasalt && ztest_opts.zo_verbose >= 1) {
6180 (void) printf("Executing older ztest: %s\n",
6181 ztest_opts.zo_alt_ztest);
6184 killed = exec_child(ztest_opts.zo_alt_ztest,
6185 ztest_opts.zo_alt_libpath, B_TRUE, &status);
6192 if (ztest_opts.zo_verbose >= 1) {
6193 hrtime_t now = gethrtime();
6195 now = MIN(now, zs->zs_proc_stop);
6196 print_time(zs->zs_proc_stop - now, timebuf);
6197 nicenum(zs->zs_space, numbuf);
6199 (void) printf("Pass %3d, %8s, %3llu ENOSPC, "
6200 "%4.1f%% of %5s used, %3.0f%% done, %8s to go\n",
6202 WIFEXITED(status) ? "Complete" : "SIGKILL",
6203 (u_longlong_t)zs->zs_enospc_count,
6204 100.0 * zs->zs_alloc / zs->zs_space,
6206 100.0 * (now - zs->zs_proc_start) /
6207 (ztest_opts.zo_time * NANOSEC), timebuf);
6210 if (ztest_opts.zo_verbose >= 2) {
6211 (void) printf("\nWorkload summary:\n\n");
6212 (void) printf("%7s %9s %s\n",
6213 "Calls", "Time", "Function");
6214 (void) printf("%7s %9s %s\n",
6215 "-----", "----", "--------");
6216 for (int f = 0; f < ZTEST_FUNCS; f++) {
6219 zi = &ztest_info[f];
6220 zc = ZTEST_GET_SHARED_CALLSTATE(f);
6221 print_time(zc->zc_time, timebuf);
6222 (void) dladdr((void *)zi->zi_func, &dli);
6223 (void) printf("%7llu %9s %s\n",
6224 (u_longlong_t)zc->zc_count, timebuf,
6227 (void) printf("\n");
6231 * It's possible that we killed a child during a rename test,
6232 * in which case we'll have a 'ztest_tmp' pool lying around
6233 * instead of 'ztest'. Do a blind rename in case this happened.
6236 if (spa_open(ztest_opts.zo_pool, &spa, FTAG) == 0) {
6237 spa_close(spa, FTAG);
6239 char tmpname[MAXNAMELEN];
6241 kernel_init(FREAD | FWRITE);
6242 (void) snprintf(tmpname, sizeof (tmpname), "%s_tmp",
6243 ztest_opts.zo_pool);
6244 (void) spa_rename(tmpname, ztest_opts.zo_pool);
6248 ztest_run_zdb(ztest_opts.zo_pool);
6251 if (ztest_opts.zo_verbose >= 1) {
6253 (void) printf("%d runs of older ztest: %s\n", older,
6254 ztest_opts.zo_alt_ztest);
6255 (void) printf("%d runs of newer ztest: %s\n", newer,
6258 (void) printf("%d killed, %d completed, %.0f%% kill rate\n",
6259 kills, iters - kills, (100.0 * kills) / MAX(1, iters));
6262 umem_free(cmd, MAXNAMELEN);