4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2011, 2016 by Delphix. All rights reserved.
24 * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
25 * Copyright (c) 2012 Martin Matuska <mm@FreeBSD.org>. All rights reserved.
26 * Copyright (c) 2013 Steven Hartland. All rights reserved.
27 * Copyright (c) 2014 Integros [integros.com]
28 * Copyright 2017 Joyent, Inc.
32 * The objective of this program is to provide a DMU/ZAP/SPA stress test
33 * that runs entirely in userland, is easy to use, and easy to extend.
35 * The overall design of the ztest program is as follows:
37 * (1) For each major functional area (e.g. adding vdevs to a pool,
38 * creating and destroying datasets, reading and writing objects, etc)
39 * we have a simple routine to test that functionality. These
40 * individual routines do not have to do anything "stressful".
42 * (2) We turn these simple functionality tests into a stress test by
43 * running them all in parallel, with as many threads as desired,
44 * and spread across as many datasets, objects, and vdevs as desired.
46 * (3) While all this is happening, we inject faults into the pool to
47 * verify that self-healing data really works.
49 * (4) Every time we open a dataset, we change its checksum and compression
50 * functions. Thus even individual objects vary from block to block
51 * in which checksum they use and whether they're compressed.
53 * (5) To verify that we never lose on-disk consistency after a crash,
54 * we run the entire test in a child of the main process.
55 * At random times, the child self-immolates with a SIGKILL.
56 * This is the software equivalent of pulling the power cord.
57 * The parent then runs the test again, using the existing
58 * storage pool, as many times as desired. If backwards compatibility
59 * testing is enabled ztest will sometimes run the "older" version
60 * of ztest after a SIGKILL.
62 * (6) To verify that we don't have future leaks or temporal incursions,
63 * many of the functional tests record the transaction group number
64 * as part of their data. When reading old data, they verify that
65 * the transaction group number is less than the current, open txg.
66 * If you add a new test, please do this if applicable.
68 * When run with no arguments, ztest runs for about five minutes and
69 * produces no output if successful. To get a little bit of information,
70 * specify -V. To get more information, specify -VV, and so on.
72 * To turn this into an overnight stress test, use -T to specify run time.
74 * You can ask more more vdevs [-v], datasets [-d], or threads [-t]
75 * to increase the pool capacity, fanout, and overall stress level.
77 * Use the -k option to set the desired frequency of kills.
79 * When ztest invokes itself it passes all relevant information through a
80 * temporary file which is mmap-ed in the child process. This allows shared
81 * memory to survive the exec syscall. The ztest_shared_hdr_t struct is always
82 * stored at offset 0 of this file and contains information on the size and
83 * number of shared structures in the file. The information stored in this file
84 * must remain backwards compatible with older versions of ztest so that
85 * ztest can invoke them during backwards compatibility testing (-B).
88 #include <sys/zfs_context.h>
94 #include <sys/dmu_objset.h>
100 #include <sys/resource.h>
103 #include <sys/zil_impl.h>
104 #include <sys/vdev_impl.h>
105 #include <sys/vdev_file.h>
106 #include <sys/spa_impl.h>
107 #include <sys/metaslab_impl.h>
108 #include <sys/dsl_prop.h>
109 #include <sys/dsl_dataset.h>
110 #include <sys/dsl_destroy.h>
111 #include <sys/dsl_scan.h>
112 #include <sys/zio_checksum.h>
113 #include <sys/refcount.h>
114 #include <sys/zfeature.h>
115 #include <sys/dsl_userhold.h>
117 #include <stdio_ext.h>
126 #include <sys/fs/zfs.h>
127 #include <libnvpair.h>
128 #include <libcmdutils.h>
130 static int ztest_fd_data = -1;
131 static int ztest_fd_rand = -1;
133 typedef struct ztest_shared_hdr {
134 uint64_t zh_hdr_size;
135 uint64_t zh_opts_size;
137 uint64_t zh_stats_size;
138 uint64_t zh_stats_count;
140 uint64_t zh_ds_count;
141 } ztest_shared_hdr_t;
143 static ztest_shared_hdr_t *ztest_shared_hdr;
145 typedef struct ztest_shared_opts {
146 char zo_pool[ZFS_MAX_DATASET_NAME_LEN];
147 char zo_dir[ZFS_MAX_DATASET_NAME_LEN];
148 char zo_alt_ztest[MAXNAMELEN];
149 char zo_alt_libpath[MAXNAMELEN];
151 uint64_t zo_vdevtime;
159 uint64_t zo_passtime;
160 uint64_t zo_killrate;
164 uint64_t zo_maxloops;
165 uint64_t zo_metaslab_gang_bang;
166 } ztest_shared_opts_t;
168 static const ztest_shared_opts_t ztest_opts_defaults = {
169 .zo_pool = { 'z', 't', 'e', 's', 't', '\0' },
170 .zo_dir = { '/', 't', 'm', 'p', '\0' },
171 .zo_alt_ztest = { '\0' },
172 .zo_alt_libpath = { '\0' },
174 .zo_ashift = SPA_MINBLOCKSHIFT,
177 .zo_raidz_parity = 1,
178 .zo_vdev_size = SPA_MINDEVSIZE * 2,
181 .zo_passtime = 60, /* 60 seconds */
182 .zo_killrate = 70, /* 70% kill rate */
185 .zo_time = 300, /* 5 minutes */
186 .zo_maxloops = 50, /* max loops during spa_freeze() */
187 .zo_metaslab_gang_bang = 32 << 10
190 extern uint64_t metaslab_gang_bang;
191 extern uint64_t metaslab_df_alloc_threshold;
192 extern uint64_t zfs_deadman_synctime_ms;
193 extern int metaslab_preload_limit;
194 extern boolean_t zfs_compressed_arc_enabled;
196 static ztest_shared_opts_t *ztest_shared_opts;
197 static ztest_shared_opts_t ztest_opts;
199 typedef struct ztest_shared_ds {
203 static ztest_shared_ds_t *ztest_shared_ds;
204 #define ZTEST_GET_SHARED_DS(d) (&ztest_shared_ds[d])
206 #define BT_MAGIC 0x123456789abcdefULL
207 #define MAXFAULTS() \
208 (MAX(zs->zs_mirrors, 1) * (ztest_opts.zo_raidz_parity + 1) - 1)
212 ZTEST_IO_WRITE_PATTERN,
213 ZTEST_IO_WRITE_ZEROES,
220 typedef struct ztest_block_tag {
230 typedef struct bufwad {
237 * XXX -- fix zfs range locks to be generic so we can use them here.
259 #define ZTEST_RANGE_LOCKS 64
260 #define ZTEST_OBJECT_LOCKS 64
263 * Object descriptor. Used as a template for object lookup/create/remove.
265 typedef struct ztest_od {
268 dmu_object_type_t od_type;
269 dmu_object_type_t od_crtype;
270 uint64_t od_blocksize;
271 uint64_t od_crblocksize;
274 char od_name[ZFS_MAX_DATASET_NAME_LEN];
280 typedef struct ztest_ds {
281 ztest_shared_ds_t *zd_shared;
283 rwlock_t zd_zilog_lock;
285 ztest_od_t *zd_od; /* debugging aid */
286 char zd_name[ZFS_MAX_DATASET_NAME_LEN];
287 mutex_t zd_dirobj_lock;
288 rll_t zd_object_lock[ZTEST_OBJECT_LOCKS];
289 rll_t zd_range_lock[ZTEST_RANGE_LOCKS];
293 * Per-iteration state.
295 typedef void ztest_func_t(ztest_ds_t *zd, uint64_t id);
297 typedef struct ztest_info {
298 ztest_func_t *zi_func; /* test function */
299 uint64_t zi_iters; /* iterations per execution */
300 uint64_t *zi_interval; /* execute every <interval> seconds */
303 typedef struct ztest_shared_callstate {
304 uint64_t zc_count; /* per-pass count */
305 uint64_t zc_time; /* per-pass time */
306 uint64_t zc_next; /* next time to call this function */
307 } ztest_shared_callstate_t;
309 static ztest_shared_callstate_t *ztest_shared_callstate;
310 #define ZTEST_GET_SHARED_CALLSTATE(c) (&ztest_shared_callstate[c])
313 * Note: these aren't static because we want dladdr() to work.
315 ztest_func_t ztest_dmu_read_write;
316 ztest_func_t ztest_dmu_write_parallel;
317 ztest_func_t ztest_dmu_object_alloc_free;
318 ztest_func_t ztest_dmu_commit_callbacks;
319 ztest_func_t ztest_zap;
320 ztest_func_t ztest_zap_parallel;
321 ztest_func_t ztest_zil_commit;
322 ztest_func_t ztest_zil_remount;
323 ztest_func_t ztest_dmu_read_write_zcopy;
324 ztest_func_t ztest_dmu_objset_create_destroy;
325 ztest_func_t ztest_dmu_prealloc;
326 ztest_func_t ztest_fzap;
327 ztest_func_t ztest_dmu_snapshot_create_destroy;
328 ztest_func_t ztest_dsl_prop_get_set;
329 ztest_func_t ztest_spa_prop_get_set;
330 ztest_func_t ztest_spa_create_destroy;
331 ztest_func_t ztest_fault_inject;
332 ztest_func_t ztest_ddt_repair;
333 ztest_func_t ztest_dmu_snapshot_hold;
334 ztest_func_t ztest_spa_rename;
335 ztest_func_t ztest_scrub;
336 ztest_func_t ztest_dsl_dataset_promote_busy;
337 ztest_func_t ztest_vdev_attach_detach;
338 ztest_func_t ztest_vdev_LUN_growth;
339 ztest_func_t ztest_vdev_add_remove;
340 ztest_func_t ztest_vdev_aux_add_remove;
341 ztest_func_t ztest_split_pool;
342 ztest_func_t ztest_reguid;
343 ztest_func_t ztest_spa_upgrade;
345 uint64_t zopt_always = 0ULL * NANOSEC; /* all the time */
346 uint64_t zopt_incessant = 1ULL * NANOSEC / 10; /* every 1/10 second */
347 uint64_t zopt_often = 1ULL * NANOSEC; /* every second */
348 uint64_t zopt_sometimes = 10ULL * NANOSEC; /* every 10 seconds */
349 uint64_t zopt_rarely = 60ULL * NANOSEC; /* every 60 seconds */
351 ztest_info_t ztest_info[] = {
352 { ztest_dmu_read_write, 1, &zopt_always },
353 { ztest_dmu_write_parallel, 10, &zopt_always },
354 { ztest_dmu_object_alloc_free, 1, &zopt_always },
355 { ztest_dmu_commit_callbacks, 1, &zopt_always },
356 { ztest_zap, 30, &zopt_always },
357 { ztest_zap_parallel, 100, &zopt_always },
358 { ztest_split_pool, 1, &zopt_always },
359 { ztest_zil_commit, 1, &zopt_incessant },
360 { ztest_zil_remount, 1, &zopt_sometimes },
361 { ztest_dmu_read_write_zcopy, 1, &zopt_often },
362 { ztest_dmu_objset_create_destroy, 1, &zopt_often },
363 { ztest_dsl_prop_get_set, 1, &zopt_often },
364 { ztest_spa_prop_get_set, 1, &zopt_sometimes },
366 { ztest_dmu_prealloc, 1, &zopt_sometimes },
368 { ztest_fzap, 1, &zopt_sometimes },
369 { ztest_dmu_snapshot_create_destroy, 1, &zopt_sometimes },
370 { ztest_spa_create_destroy, 1, &zopt_sometimes },
371 { ztest_fault_inject, 1, &zopt_sometimes },
372 { ztest_ddt_repair, 1, &zopt_sometimes },
373 { ztest_dmu_snapshot_hold, 1, &zopt_sometimes },
374 { ztest_reguid, 1, &zopt_rarely },
375 { ztest_spa_rename, 1, &zopt_rarely },
376 { ztest_scrub, 1, &zopt_rarely },
377 { ztest_spa_upgrade, 1, &zopt_rarely },
378 { ztest_dsl_dataset_promote_busy, 1, &zopt_rarely },
379 { ztest_vdev_attach_detach, 1, &zopt_sometimes },
380 { ztest_vdev_LUN_growth, 1, &zopt_rarely },
381 { ztest_vdev_add_remove, 1,
382 &ztest_opts.zo_vdevtime },
383 { ztest_vdev_aux_add_remove, 1,
384 &ztest_opts.zo_vdevtime },
387 #define ZTEST_FUNCS (sizeof (ztest_info) / sizeof (ztest_info_t))
390 * The following struct is used to hold a list of uncalled commit callbacks.
391 * The callbacks are ordered by txg number.
393 typedef struct ztest_cb_list {
394 mutex_t zcl_callbacks_lock;
395 list_t zcl_callbacks;
399 * Stuff we need to share writably between parent and child.
401 typedef struct ztest_shared {
402 boolean_t zs_do_init;
403 hrtime_t zs_proc_start;
404 hrtime_t zs_proc_stop;
405 hrtime_t zs_thread_start;
406 hrtime_t zs_thread_stop;
407 hrtime_t zs_thread_kill;
408 uint64_t zs_enospc_count;
409 uint64_t zs_vdev_next_leaf;
410 uint64_t zs_vdev_aux;
415 uint64_t zs_metaslab_sz;
416 uint64_t zs_metaslab_df_alloc_threshold;
420 #define ID_PARALLEL -1ULL
422 static char ztest_dev_template[] = "%s/%s.%llua";
423 static char ztest_aux_template[] = "%s/%s.%s.%llu";
424 ztest_shared_t *ztest_shared;
426 static spa_t *ztest_spa = NULL;
427 static ztest_ds_t *ztest_ds;
429 static mutex_t ztest_vdev_lock;
432 * The ztest_name_lock protects the pool and dataset namespace used by
433 * the individual tests. To modify the namespace, consumers must grab
434 * this lock as writer. Grabbing the lock as reader will ensure that the
435 * namespace does not change while the lock is held.
437 static rwlock_t ztest_name_lock;
439 static boolean_t ztest_dump_core = B_TRUE;
440 static boolean_t ztest_exiting;
442 /* Global commit callback list */
443 static ztest_cb_list_t zcl;
446 ZTEST_META_DNODE = 0,
451 static void usage(boolean_t) __NORETURN;
454 * These libumem hooks provide a reasonable set of defaults for the allocator's
455 * debugging facilities.
460 return ("default,verbose"); /* $UMEM_DEBUG setting */
464 _umem_logging_init(void)
466 return ("fail,contents"); /* $UMEM_LOGGING setting */
469 #define FATAL_MSG_SZ 1024
474 fatal(int do_perror, char *message, ...)
477 int save_errno = errno;
478 char buf[FATAL_MSG_SZ];
480 (void) fflush(stdout);
482 va_start(args, message);
483 (void) sprintf(buf, "ztest: ");
485 (void) vsprintf(buf + strlen(buf), message, args);
488 (void) snprintf(buf + strlen(buf), FATAL_MSG_SZ - strlen(buf),
489 ": %s", strerror(save_errno));
491 (void) fprintf(stderr, "%s\n", buf);
492 fatal_msg = buf; /* to ease debugging */
499 str2shift(const char *buf)
501 const char *ends = "BKMGTPEZ";
506 for (i = 0; i < strlen(ends); i++) {
507 if (toupper(buf[0]) == ends[i])
510 if (i == strlen(ends)) {
511 (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n",
515 if (buf[1] == '\0' || (toupper(buf[1]) == 'B' && buf[2] == '\0')) {
518 (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n", buf);
524 nicenumtoull(const char *buf)
529 val = strtoull(buf, &end, 0);
531 (void) fprintf(stderr, "ztest: bad numeric value: %s\n", buf);
533 } else if (end[0] == '.') {
534 double fval = strtod(buf, &end);
535 fval *= pow(2, str2shift(end));
536 if (fval > UINT64_MAX) {
537 (void) fprintf(stderr, "ztest: value too large: %s\n",
541 val = (uint64_t)fval;
543 int shift = str2shift(end);
544 if (shift >= 64 || (val << shift) >> shift != val) {
545 (void) fprintf(stderr, "ztest: value too large: %s\n",
555 usage(boolean_t requested)
557 const ztest_shared_opts_t *zo = &ztest_opts_defaults;
559 char nice_vdev_size[NN_NUMBUF_SZ];
560 char nice_gang_bang[NN_NUMBUF_SZ];
561 FILE *fp = requested ? stdout : stderr;
563 nicenum(zo->zo_vdev_size, nice_vdev_size, sizeof (nice_vdev_size));
564 nicenum(zo->zo_metaslab_gang_bang, nice_gang_bang,
565 sizeof (nice_gang_bang));
567 (void) fprintf(fp, "Usage: %s\n"
568 "\t[-v vdevs (default: %llu)]\n"
569 "\t[-s size_of_each_vdev (default: %s)]\n"
570 "\t[-a alignment_shift (default: %d)] use 0 for random\n"
571 "\t[-m mirror_copies (default: %d)]\n"
572 "\t[-r raidz_disks (default: %d)]\n"
573 "\t[-R raidz_parity (default: %d)]\n"
574 "\t[-d datasets (default: %d)]\n"
575 "\t[-t threads (default: %d)]\n"
576 "\t[-g gang_block_threshold (default: %s)]\n"
577 "\t[-i init_count (default: %d)] initialize pool i times\n"
578 "\t[-k kill_percentage (default: %llu%%)]\n"
579 "\t[-p pool_name (default: %s)]\n"
580 "\t[-f dir (default: %s)] file directory for vdev files\n"
581 "\t[-V] verbose (use multiple times for ever more blather)\n"
582 "\t[-E] use existing pool instead of creating new one\n"
583 "\t[-T time (default: %llu sec)] total run time\n"
584 "\t[-F freezeloops (default: %llu)] max loops in spa_freeze()\n"
585 "\t[-P passtime (default: %llu sec)] time per pass\n"
586 "\t[-B alt_ztest (default: <none>)] alternate ztest path\n"
587 "\t[-o variable=value] ... set global variable to an unsigned\n"
588 "\t 32-bit integer value\n"
589 "\t[-h] (print help)\n"
592 (u_longlong_t)zo->zo_vdevs, /* -v */
593 nice_vdev_size, /* -s */
594 zo->zo_ashift, /* -a */
595 zo->zo_mirrors, /* -m */
596 zo->zo_raidz, /* -r */
597 zo->zo_raidz_parity, /* -R */
598 zo->zo_datasets, /* -d */
599 zo->zo_threads, /* -t */
600 nice_gang_bang, /* -g */
601 zo->zo_init, /* -i */
602 (u_longlong_t)zo->zo_killrate, /* -k */
603 zo->zo_pool, /* -p */
605 (u_longlong_t)zo->zo_time, /* -T */
606 (u_longlong_t)zo->zo_maxloops, /* -F */
607 (u_longlong_t)zo->zo_passtime);
608 exit(requested ? 0 : 1);
612 process_options(int argc, char **argv)
615 ztest_shared_opts_t *zo = &ztest_opts;
619 char altdir[MAXNAMELEN] = { 0 };
621 bcopy(&ztest_opts_defaults, zo, sizeof (*zo));
623 while ((opt = getopt(argc, argv,
624 "v:s:a:m:r:R:d:t:g:i:k:p:f:VET:P:hF:B:o:")) != EOF) {
641 value = nicenumtoull(optarg);
645 zo->zo_vdevs = value;
648 zo->zo_vdev_size = MAX(SPA_MINDEVSIZE, value);
651 zo->zo_ashift = value;
654 zo->zo_mirrors = value;
657 zo->zo_raidz = MAX(1, value);
660 zo->zo_raidz_parity = MIN(MAX(value, 1), 3);
663 zo->zo_datasets = MAX(1, value);
666 zo->zo_threads = MAX(1, value);
669 zo->zo_metaslab_gang_bang = MAX(SPA_MINBLOCKSIZE << 1,
676 zo->zo_killrate = value;
679 (void) strlcpy(zo->zo_pool, optarg,
680 sizeof (zo->zo_pool));
683 path = realpath(optarg, NULL);
685 (void) fprintf(stderr, "error: %s: %s\n",
686 optarg, strerror(errno));
689 (void) strlcpy(zo->zo_dir, path,
690 sizeof (zo->zo_dir));
703 zo->zo_passtime = MAX(1, value);
706 zo->zo_maxloops = MAX(1, value);
709 (void) strlcpy(altdir, optarg, sizeof (altdir));
712 if (set_global_var(optarg) != 0)
725 zo->zo_raidz_parity = MIN(zo->zo_raidz_parity, zo->zo_raidz - 1);
728 (zo->zo_vdevs > 0 ? zo->zo_time * NANOSEC / zo->zo_vdevs :
731 if (strlen(altdir) > 0) {
739 cmd = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
740 realaltdir = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
742 VERIFY(NULL != realpath(getexecname(), cmd));
743 if (0 != access(altdir, F_OK)) {
744 ztest_dump_core = B_FALSE;
745 fatal(B_TRUE, "invalid alternate ztest path: %s",
748 VERIFY(NULL != realpath(altdir, realaltdir));
751 * 'cmd' should be of the form "<anything>/usr/bin/<isa>/ztest".
752 * We want to extract <isa> to determine if we should use
753 * 32 or 64 bit binaries.
755 bin = strstr(cmd, "/usr/bin/");
756 ztest = strstr(bin, "/ztest");
758 isalen = ztest - isa;
759 (void) snprintf(zo->zo_alt_ztest, sizeof (zo->zo_alt_ztest),
760 "%s/usr/bin/%.*s/ztest", realaltdir, isalen, isa);
761 (void) snprintf(zo->zo_alt_libpath, sizeof (zo->zo_alt_libpath),
762 "%s/usr/lib/%.*s", realaltdir, isalen, isa);
764 if (0 != access(zo->zo_alt_ztest, X_OK)) {
765 ztest_dump_core = B_FALSE;
766 fatal(B_TRUE, "invalid alternate ztest: %s",
768 } else if (0 != access(zo->zo_alt_libpath, X_OK)) {
769 ztest_dump_core = B_FALSE;
770 fatal(B_TRUE, "invalid alternate lib directory %s",
774 umem_free(cmd, MAXPATHLEN);
775 umem_free(realaltdir, MAXPATHLEN);
780 ztest_kill(ztest_shared_t *zs)
782 zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(ztest_spa));
783 zs->zs_space = metaslab_class_get_space(spa_normal_class(ztest_spa));
786 * Before we kill off ztest, make sure that the config is updated.
787 * See comment above spa_config_sync().
789 mutex_enter(&spa_namespace_lock);
790 spa_config_sync(ztest_spa, B_FALSE, B_FALSE);
791 mutex_exit(&spa_namespace_lock);
793 zfs_dbgmsg_print(FTAG);
794 (void) kill(getpid(), SIGKILL);
798 ztest_random(uint64_t range)
802 ASSERT3S(ztest_fd_rand, >=, 0);
807 if (read(ztest_fd_rand, &r, sizeof (r)) != sizeof (r))
808 fatal(1, "short read from /dev/urandom");
815 ztest_record_enospc(const char *s)
817 ztest_shared->zs_enospc_count++;
821 ztest_get_ashift(void)
823 if (ztest_opts.zo_ashift == 0)
824 return (SPA_MINBLOCKSHIFT + ztest_random(5));
825 return (ztest_opts.zo_ashift);
829 make_vdev_file(char *path, char *aux, char *pool, size_t size, uint64_t ashift)
831 char pathbuf[MAXPATHLEN];
836 ashift = ztest_get_ashift();
842 vdev = ztest_shared->zs_vdev_aux;
843 (void) snprintf(path, sizeof (pathbuf),
844 ztest_aux_template, ztest_opts.zo_dir,
845 pool == NULL ? ztest_opts.zo_pool : pool,
848 vdev = ztest_shared->zs_vdev_next_leaf++;
849 (void) snprintf(path, sizeof (pathbuf),
850 ztest_dev_template, ztest_opts.zo_dir,
851 pool == NULL ? ztest_opts.zo_pool : pool, vdev);
856 int fd = open(path, O_RDWR | O_CREAT | O_TRUNC, 0666);
858 fatal(1, "can't open %s", path);
859 if (ftruncate(fd, size) != 0)
860 fatal(1, "can't ftruncate %s", path);
864 VERIFY(nvlist_alloc(&file, NV_UNIQUE_NAME, 0) == 0);
865 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_TYPE, VDEV_TYPE_FILE) == 0);
866 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_PATH, path) == 0);
867 VERIFY(nvlist_add_uint64(file, ZPOOL_CONFIG_ASHIFT, ashift) == 0);
873 make_vdev_raidz(char *path, char *aux, char *pool, size_t size,
874 uint64_t ashift, int r)
876 nvlist_t *raidz, **child;
880 return (make_vdev_file(path, aux, pool, size, ashift));
881 child = umem_alloc(r * sizeof (nvlist_t *), UMEM_NOFAIL);
883 for (c = 0; c < r; c++)
884 child[c] = make_vdev_file(path, aux, pool, size, ashift);
886 VERIFY(nvlist_alloc(&raidz, NV_UNIQUE_NAME, 0) == 0);
887 VERIFY(nvlist_add_string(raidz, ZPOOL_CONFIG_TYPE,
888 VDEV_TYPE_RAIDZ) == 0);
889 VERIFY(nvlist_add_uint64(raidz, ZPOOL_CONFIG_NPARITY,
890 ztest_opts.zo_raidz_parity) == 0);
891 VERIFY(nvlist_add_nvlist_array(raidz, ZPOOL_CONFIG_CHILDREN,
894 for (c = 0; c < r; c++)
895 nvlist_free(child[c]);
897 umem_free(child, r * sizeof (nvlist_t *));
903 make_vdev_mirror(char *path, char *aux, char *pool, size_t size,
904 uint64_t ashift, int r, int m)
906 nvlist_t *mirror, **child;
910 return (make_vdev_raidz(path, aux, pool, size, ashift, r));
912 child = umem_alloc(m * sizeof (nvlist_t *), UMEM_NOFAIL);
914 for (c = 0; c < m; c++)
915 child[c] = make_vdev_raidz(path, aux, pool, size, ashift, r);
917 VERIFY(nvlist_alloc(&mirror, NV_UNIQUE_NAME, 0) == 0);
918 VERIFY(nvlist_add_string(mirror, ZPOOL_CONFIG_TYPE,
919 VDEV_TYPE_MIRROR) == 0);
920 VERIFY(nvlist_add_nvlist_array(mirror, ZPOOL_CONFIG_CHILDREN,
923 for (c = 0; c < m; c++)
924 nvlist_free(child[c]);
926 umem_free(child, m * sizeof (nvlist_t *));
932 make_vdev_root(char *path, char *aux, char *pool, size_t size, uint64_t ashift,
933 int log, int r, int m, int t)
935 nvlist_t *root, **child;
940 child = umem_alloc(t * sizeof (nvlist_t *), UMEM_NOFAIL);
942 for (c = 0; c < t; c++) {
943 child[c] = make_vdev_mirror(path, aux, pool, size, ashift,
945 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_IS_LOG,
949 VERIFY(nvlist_alloc(&root, NV_UNIQUE_NAME, 0) == 0);
950 VERIFY(nvlist_add_string(root, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) == 0);
951 VERIFY(nvlist_add_nvlist_array(root, aux ? aux : ZPOOL_CONFIG_CHILDREN,
954 for (c = 0; c < t; c++)
955 nvlist_free(child[c]);
957 umem_free(child, t * sizeof (nvlist_t *));
963 * Find a random spa version. Returns back a random spa version in the
964 * range [initial_version, SPA_VERSION_FEATURES].
967 ztest_random_spa_version(uint64_t initial_version)
969 uint64_t version = initial_version;
971 if (version <= SPA_VERSION_BEFORE_FEATURES) {
973 ztest_random(SPA_VERSION_BEFORE_FEATURES - version + 1);
976 if (version > SPA_VERSION_BEFORE_FEATURES)
977 version = SPA_VERSION_FEATURES;
979 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
984 ztest_random_blocksize(void)
986 uint64_t block_shift;
988 * Choose a block size >= the ashift.
989 * If the SPA supports new MAXBLOCKSIZE, test up to 1MB blocks.
991 int maxbs = SPA_OLD_MAXBLOCKSHIFT;
992 if (spa_maxblocksize(ztest_spa) == SPA_MAXBLOCKSIZE)
994 block_shift = ztest_random(maxbs - ztest_spa->spa_max_ashift + 1);
995 return (1 << (SPA_MINBLOCKSHIFT + block_shift));
999 ztest_random_ibshift(void)
1001 return (DN_MIN_INDBLKSHIFT +
1002 ztest_random(DN_MAX_INDBLKSHIFT - DN_MIN_INDBLKSHIFT + 1));
1006 ztest_random_vdev_top(spa_t *spa, boolean_t log_ok)
1009 vdev_t *rvd = spa->spa_root_vdev;
1012 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1015 top = ztest_random(rvd->vdev_children);
1016 tvd = rvd->vdev_child[top];
1017 } while (tvd->vdev_ishole || (tvd->vdev_islog && !log_ok) ||
1018 tvd->vdev_mg == NULL || tvd->vdev_mg->mg_class == NULL);
1024 ztest_random_dsl_prop(zfs_prop_t prop)
1029 value = zfs_prop_random_value(prop, ztest_random(-1ULL));
1030 } while (prop == ZFS_PROP_CHECKSUM && value == ZIO_CHECKSUM_OFF);
1036 ztest_dsl_prop_set_uint64(char *osname, zfs_prop_t prop, uint64_t value,
1039 const char *propname = zfs_prop_to_name(prop);
1040 const char *valname;
1041 char setpoint[MAXPATHLEN];
1045 error = dsl_prop_set_int(osname, propname,
1046 (inherit ? ZPROP_SRC_NONE : ZPROP_SRC_LOCAL), value);
1048 if (error == ENOSPC) {
1049 ztest_record_enospc(FTAG);
1054 VERIFY0(dsl_prop_get_integer(osname, propname, &curval, setpoint));
1056 if (ztest_opts.zo_verbose >= 6) {
1057 VERIFY(zfs_prop_index_to_string(prop, curval, &valname) == 0);
1058 (void) printf("%s %s = %s at '%s'\n",
1059 osname, propname, valname, setpoint);
1066 ztest_spa_prop_set_uint64(zpool_prop_t prop, uint64_t value)
1068 spa_t *spa = ztest_spa;
1069 nvlist_t *props = NULL;
1072 VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
1073 VERIFY(nvlist_add_uint64(props, zpool_prop_to_name(prop), value) == 0);
1075 error = spa_prop_set(spa, props);
1079 if (error == ENOSPC) {
1080 ztest_record_enospc(FTAG);
1089 ztest_rll_init(rll_t *rll)
1091 rll->rll_writer = NULL;
1092 rll->rll_readers = 0;
1093 VERIFY(_mutex_init(&rll->rll_lock, USYNC_THREAD, NULL) == 0);
1094 VERIFY(cond_init(&rll->rll_cv, USYNC_THREAD, NULL) == 0);
1098 ztest_rll_destroy(rll_t *rll)
1100 ASSERT(rll->rll_writer == NULL);
1101 ASSERT(rll->rll_readers == 0);
1102 VERIFY(_mutex_destroy(&rll->rll_lock) == 0);
1103 VERIFY(cond_destroy(&rll->rll_cv) == 0);
1107 ztest_rll_lock(rll_t *rll, rl_type_t type)
1109 VERIFY(mutex_lock(&rll->rll_lock) == 0);
1111 if (type == RL_READER) {
1112 while (rll->rll_writer != NULL)
1113 (void) cond_wait(&rll->rll_cv, &rll->rll_lock);
1116 while (rll->rll_writer != NULL || rll->rll_readers)
1117 (void) cond_wait(&rll->rll_cv, &rll->rll_lock);
1118 rll->rll_writer = curthread;
1121 VERIFY(mutex_unlock(&rll->rll_lock) == 0);
1125 ztest_rll_unlock(rll_t *rll)
1127 VERIFY(mutex_lock(&rll->rll_lock) == 0);
1129 if (rll->rll_writer) {
1130 ASSERT(rll->rll_readers == 0);
1131 rll->rll_writer = NULL;
1133 ASSERT(rll->rll_readers != 0);
1134 ASSERT(rll->rll_writer == NULL);
1138 if (rll->rll_writer == NULL && rll->rll_readers == 0)
1139 VERIFY(cond_broadcast(&rll->rll_cv) == 0);
1141 VERIFY(mutex_unlock(&rll->rll_lock) == 0);
1145 ztest_object_lock(ztest_ds_t *zd, uint64_t object, rl_type_t type)
1147 rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1149 ztest_rll_lock(rll, type);
1153 ztest_object_unlock(ztest_ds_t *zd, uint64_t object)
1155 rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1157 ztest_rll_unlock(rll);
1161 ztest_range_lock(ztest_ds_t *zd, uint64_t object, uint64_t offset,
1162 uint64_t size, rl_type_t type)
1164 uint64_t hash = object ^ (offset % (ZTEST_RANGE_LOCKS + 1));
1165 rll_t *rll = &zd->zd_range_lock[hash & (ZTEST_RANGE_LOCKS - 1)];
1168 rl = umem_alloc(sizeof (*rl), UMEM_NOFAIL);
1169 rl->rl_object = object;
1170 rl->rl_offset = offset;
1174 ztest_rll_lock(rll, type);
1180 ztest_range_unlock(rl_t *rl)
1182 rll_t *rll = rl->rl_lock;
1184 ztest_rll_unlock(rll);
1186 umem_free(rl, sizeof (*rl));
1190 ztest_zd_init(ztest_ds_t *zd, ztest_shared_ds_t *szd, objset_t *os)
1193 zd->zd_zilog = dmu_objset_zil(os);
1194 zd->zd_shared = szd;
1195 dmu_objset_name(os, zd->zd_name);
1197 if (zd->zd_shared != NULL)
1198 zd->zd_shared->zd_seq = 0;
1200 VERIFY(rwlock_init(&zd->zd_zilog_lock, USYNC_THREAD, NULL) == 0);
1201 VERIFY(_mutex_init(&zd->zd_dirobj_lock, USYNC_THREAD, NULL) == 0);
1203 for (int l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1204 ztest_rll_init(&zd->zd_object_lock[l]);
1206 for (int l = 0; l < ZTEST_RANGE_LOCKS; l++)
1207 ztest_rll_init(&zd->zd_range_lock[l]);
1211 ztest_zd_fini(ztest_ds_t *zd)
1213 VERIFY(_mutex_destroy(&zd->zd_dirobj_lock) == 0);
1215 for (int l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1216 ztest_rll_destroy(&zd->zd_object_lock[l]);
1218 for (int l = 0; l < ZTEST_RANGE_LOCKS; l++)
1219 ztest_rll_destroy(&zd->zd_range_lock[l]);
1222 #define TXG_MIGHTWAIT (ztest_random(10) == 0 ? TXG_NOWAIT : TXG_WAIT)
1225 ztest_tx_assign(dmu_tx_t *tx, uint64_t txg_how, const char *tag)
1231 * Attempt to assign tx to some transaction group.
1233 error = dmu_tx_assign(tx, txg_how);
1235 if (error == ERESTART) {
1236 ASSERT(txg_how == TXG_NOWAIT);
1239 ASSERT3U(error, ==, ENOSPC);
1240 ztest_record_enospc(tag);
1245 txg = dmu_tx_get_txg(tx);
1251 ztest_pattern_set(void *buf, uint64_t size, uint64_t value)
1254 uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1261 ztest_pattern_match(void *buf, uint64_t size, uint64_t value)
1264 uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1268 diff |= (value - *ip++);
1274 ztest_bt_generate(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1275 uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg)
1277 bt->bt_magic = BT_MAGIC;
1278 bt->bt_objset = dmu_objset_id(os);
1279 bt->bt_object = object;
1280 bt->bt_offset = offset;
1283 bt->bt_crtxg = crtxg;
1287 ztest_bt_verify(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1288 uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg)
1290 ASSERT3U(bt->bt_magic, ==, BT_MAGIC);
1291 ASSERT3U(bt->bt_objset, ==, dmu_objset_id(os));
1292 ASSERT3U(bt->bt_object, ==, object);
1293 ASSERT3U(bt->bt_offset, ==, offset);
1294 ASSERT3U(bt->bt_gen, <=, gen);
1295 ASSERT3U(bt->bt_txg, <=, txg);
1296 ASSERT3U(bt->bt_crtxg, ==, crtxg);
1299 static ztest_block_tag_t *
1300 ztest_bt_bonus(dmu_buf_t *db)
1302 dmu_object_info_t doi;
1303 ztest_block_tag_t *bt;
1305 dmu_object_info_from_db(db, &doi);
1306 ASSERT3U(doi.doi_bonus_size, <=, db->db_size);
1307 ASSERT3U(doi.doi_bonus_size, >=, sizeof (*bt));
1308 bt = (void *)((char *)db->db_data + doi.doi_bonus_size - sizeof (*bt));
1317 #define lrz_type lr_mode
1318 #define lrz_blocksize lr_uid
1319 #define lrz_ibshift lr_gid
1320 #define lrz_bonustype lr_rdev
1321 #define lrz_bonuslen lr_crtime[1]
1324 ztest_log_create(ztest_ds_t *zd, dmu_tx_t *tx, lr_create_t *lr)
1326 char *name = (void *)(lr + 1); /* name follows lr */
1327 size_t namesize = strlen(name) + 1;
1330 if (zil_replaying(zd->zd_zilog, tx))
1333 itx = zil_itx_create(TX_CREATE, sizeof (*lr) + namesize);
1334 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1335 sizeof (*lr) + namesize - sizeof (lr_t));
1337 zil_itx_assign(zd->zd_zilog, itx, tx);
1341 ztest_log_remove(ztest_ds_t *zd, dmu_tx_t *tx, lr_remove_t *lr, uint64_t object)
1343 char *name = (void *)(lr + 1); /* name follows lr */
1344 size_t namesize = strlen(name) + 1;
1347 if (zil_replaying(zd->zd_zilog, tx))
1350 itx = zil_itx_create(TX_REMOVE, sizeof (*lr) + namesize);
1351 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1352 sizeof (*lr) + namesize - sizeof (lr_t));
1354 itx->itx_oid = object;
1355 zil_itx_assign(zd->zd_zilog, itx, tx);
1359 ztest_log_write(ztest_ds_t *zd, dmu_tx_t *tx, lr_write_t *lr)
1362 itx_wr_state_t write_state = ztest_random(WR_NUM_STATES);
1364 if (zil_replaying(zd->zd_zilog, tx))
1367 if (lr->lr_length > ZIL_MAX_LOG_DATA)
1368 write_state = WR_INDIRECT;
1370 itx = zil_itx_create(TX_WRITE,
1371 sizeof (*lr) + (write_state == WR_COPIED ? lr->lr_length : 0));
1373 if (write_state == WR_COPIED &&
1374 dmu_read(zd->zd_os, lr->lr_foid, lr->lr_offset, lr->lr_length,
1375 ((lr_write_t *)&itx->itx_lr) + 1, DMU_READ_NO_PREFETCH) != 0) {
1376 zil_itx_destroy(itx);
1377 itx = zil_itx_create(TX_WRITE, sizeof (*lr));
1378 write_state = WR_NEED_COPY;
1380 itx->itx_private = zd;
1381 itx->itx_wr_state = write_state;
1382 itx->itx_sync = (ztest_random(8) == 0);
1384 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1385 sizeof (*lr) - sizeof (lr_t));
1387 zil_itx_assign(zd->zd_zilog, itx, tx);
1391 ztest_log_truncate(ztest_ds_t *zd, dmu_tx_t *tx, lr_truncate_t *lr)
1395 if (zil_replaying(zd->zd_zilog, tx))
1398 itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr));
1399 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1400 sizeof (*lr) - sizeof (lr_t));
1402 itx->itx_sync = B_FALSE;
1403 zil_itx_assign(zd->zd_zilog, itx, tx);
1407 ztest_log_setattr(ztest_ds_t *zd, dmu_tx_t *tx, lr_setattr_t *lr)
1411 if (zil_replaying(zd->zd_zilog, tx))
1414 itx = zil_itx_create(TX_SETATTR, sizeof (*lr));
1415 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1416 sizeof (*lr) - sizeof (lr_t));
1418 itx->itx_sync = B_FALSE;
1419 zil_itx_assign(zd->zd_zilog, itx, tx);
1426 ztest_replay_create(ztest_ds_t *zd, lr_create_t *lr, boolean_t byteswap)
1428 char *name = (void *)(lr + 1); /* name follows lr */
1429 objset_t *os = zd->zd_os;
1430 ztest_block_tag_t *bbt;
1437 byteswap_uint64_array(lr, sizeof (*lr));
1439 ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1440 ASSERT(name[0] != '\0');
1442 tx = dmu_tx_create(os);
1444 dmu_tx_hold_zap(tx, lr->lr_doid, B_TRUE, name);
1446 if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1447 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, B_TRUE, NULL);
1449 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1452 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1456 ASSERT(dmu_objset_zil(os)->zl_replay == !!lr->lr_foid);
1458 if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1459 if (lr->lr_foid == 0) {
1460 lr->lr_foid = zap_create(os,
1461 lr->lrz_type, lr->lrz_bonustype,
1462 lr->lrz_bonuslen, tx);
1464 error = zap_create_claim(os, lr->lr_foid,
1465 lr->lrz_type, lr->lrz_bonustype,
1466 lr->lrz_bonuslen, tx);
1469 if (lr->lr_foid == 0) {
1470 lr->lr_foid = dmu_object_alloc(os,
1471 lr->lrz_type, 0, lr->lrz_bonustype,
1472 lr->lrz_bonuslen, tx);
1474 error = dmu_object_claim(os, lr->lr_foid,
1475 lr->lrz_type, 0, lr->lrz_bonustype,
1476 lr->lrz_bonuslen, tx);
1481 ASSERT3U(error, ==, EEXIST);
1482 ASSERT(zd->zd_zilog->zl_replay);
1487 ASSERT(lr->lr_foid != 0);
1489 if (lr->lrz_type != DMU_OT_ZAP_OTHER)
1490 VERIFY3U(0, ==, dmu_object_set_blocksize(os, lr->lr_foid,
1491 lr->lrz_blocksize, lr->lrz_ibshift, tx));
1493 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1494 bbt = ztest_bt_bonus(db);
1495 dmu_buf_will_dirty(db, tx);
1496 ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_gen, txg, txg);
1497 dmu_buf_rele(db, FTAG);
1499 VERIFY3U(0, ==, zap_add(os, lr->lr_doid, name, sizeof (uint64_t), 1,
1502 (void) ztest_log_create(zd, tx, lr);
1510 ztest_replay_remove(ztest_ds_t *zd, lr_remove_t *lr, boolean_t byteswap)
1512 char *name = (void *)(lr + 1); /* name follows lr */
1513 objset_t *os = zd->zd_os;
1514 dmu_object_info_t doi;
1516 uint64_t object, txg;
1519 byteswap_uint64_array(lr, sizeof (*lr));
1521 ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1522 ASSERT(name[0] != '\0');
1525 zap_lookup(os, lr->lr_doid, name, sizeof (object), 1, &object));
1526 ASSERT(object != 0);
1528 ztest_object_lock(zd, object, RL_WRITER);
1530 VERIFY3U(0, ==, dmu_object_info(os, object, &doi));
1532 tx = dmu_tx_create(os);
1534 dmu_tx_hold_zap(tx, lr->lr_doid, B_FALSE, name);
1535 dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END);
1537 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1539 ztest_object_unlock(zd, object);
1543 if (doi.doi_type == DMU_OT_ZAP_OTHER) {
1544 VERIFY3U(0, ==, zap_destroy(os, object, tx));
1546 VERIFY3U(0, ==, dmu_object_free(os, object, tx));
1549 VERIFY3U(0, ==, zap_remove(os, lr->lr_doid, name, tx));
1551 (void) ztest_log_remove(zd, tx, lr, object);
1555 ztest_object_unlock(zd, object);
1561 ztest_replay_write(ztest_ds_t *zd, lr_write_t *lr, boolean_t byteswap)
1563 objset_t *os = zd->zd_os;
1564 void *data = lr + 1; /* data follows lr */
1565 uint64_t offset, length;
1566 ztest_block_tag_t *bt = data;
1567 ztest_block_tag_t *bbt;
1568 uint64_t gen, txg, lrtxg, crtxg;
1569 dmu_object_info_t doi;
1572 arc_buf_t *abuf = NULL;
1576 byteswap_uint64_array(lr, sizeof (*lr));
1578 offset = lr->lr_offset;
1579 length = lr->lr_length;
1581 /* If it's a dmu_sync() block, write the whole block */
1582 if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
1583 uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
1584 if (length < blocksize) {
1585 offset -= offset % blocksize;
1590 if (bt->bt_magic == BSWAP_64(BT_MAGIC))
1591 byteswap_uint64_array(bt, sizeof (*bt));
1593 if (bt->bt_magic != BT_MAGIC)
1596 ztest_object_lock(zd, lr->lr_foid, RL_READER);
1597 rl = ztest_range_lock(zd, lr->lr_foid, offset, length, RL_WRITER);
1599 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1601 dmu_object_info_from_db(db, &doi);
1603 bbt = ztest_bt_bonus(db);
1604 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1606 crtxg = bbt->bt_crtxg;
1607 lrtxg = lr->lr_common.lrc_txg;
1609 tx = dmu_tx_create(os);
1611 dmu_tx_hold_write(tx, lr->lr_foid, offset, length);
1613 if (ztest_random(8) == 0 && length == doi.doi_data_block_size &&
1614 P2PHASE(offset, length) == 0)
1615 abuf = dmu_request_arcbuf(db, length);
1617 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1620 dmu_return_arcbuf(abuf);
1621 dmu_buf_rele(db, FTAG);
1622 ztest_range_unlock(rl);
1623 ztest_object_unlock(zd, lr->lr_foid);
1629 * Usually, verify the old data before writing new data --
1630 * but not always, because we also want to verify correct
1631 * behavior when the data was not recently read into cache.
1633 ASSERT(offset % doi.doi_data_block_size == 0);
1634 if (ztest_random(4) != 0) {
1635 int prefetch = ztest_random(2) ?
1636 DMU_READ_PREFETCH : DMU_READ_NO_PREFETCH;
1637 ztest_block_tag_t rbt;
1639 VERIFY(dmu_read(os, lr->lr_foid, offset,
1640 sizeof (rbt), &rbt, prefetch) == 0);
1641 if (rbt.bt_magic == BT_MAGIC) {
1642 ztest_bt_verify(&rbt, os, lr->lr_foid,
1643 offset, gen, txg, crtxg);
1648 * Writes can appear to be newer than the bonus buffer because
1649 * the ztest_get_data() callback does a dmu_read() of the
1650 * open-context data, which may be different than the data
1651 * as it was when the write was generated.
1653 if (zd->zd_zilog->zl_replay) {
1654 ztest_bt_verify(bt, os, lr->lr_foid, offset,
1655 MAX(gen, bt->bt_gen), MAX(txg, lrtxg),
1660 * Set the bt's gen/txg to the bonus buffer's gen/txg
1661 * so that all of the usual ASSERTs will work.
1663 ztest_bt_generate(bt, os, lr->lr_foid, offset, gen, txg, crtxg);
1667 dmu_write(os, lr->lr_foid, offset, length, data, tx);
1669 bcopy(data, abuf->b_data, length);
1670 dmu_assign_arcbuf(db, offset, abuf, tx);
1673 (void) ztest_log_write(zd, tx, lr);
1675 dmu_buf_rele(db, FTAG);
1679 ztest_range_unlock(rl);
1680 ztest_object_unlock(zd, lr->lr_foid);
1686 ztest_replay_truncate(ztest_ds_t *zd, lr_truncate_t *lr, boolean_t byteswap)
1688 objset_t *os = zd->zd_os;
1694 byteswap_uint64_array(lr, sizeof (*lr));
1696 ztest_object_lock(zd, lr->lr_foid, RL_READER);
1697 rl = ztest_range_lock(zd, lr->lr_foid, lr->lr_offset, lr->lr_length,
1700 tx = dmu_tx_create(os);
1702 dmu_tx_hold_free(tx, lr->lr_foid, lr->lr_offset, lr->lr_length);
1704 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1706 ztest_range_unlock(rl);
1707 ztest_object_unlock(zd, lr->lr_foid);
1711 VERIFY(dmu_free_range(os, lr->lr_foid, lr->lr_offset,
1712 lr->lr_length, tx) == 0);
1714 (void) ztest_log_truncate(zd, tx, lr);
1718 ztest_range_unlock(rl);
1719 ztest_object_unlock(zd, lr->lr_foid);
1725 ztest_replay_setattr(ztest_ds_t *zd, lr_setattr_t *lr, boolean_t byteswap)
1727 objset_t *os = zd->zd_os;
1730 ztest_block_tag_t *bbt;
1731 uint64_t txg, lrtxg, crtxg;
1734 byteswap_uint64_array(lr, sizeof (*lr));
1736 ztest_object_lock(zd, lr->lr_foid, RL_WRITER);
1738 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1740 tx = dmu_tx_create(os);
1741 dmu_tx_hold_bonus(tx, lr->lr_foid);
1743 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1745 dmu_buf_rele(db, FTAG);
1746 ztest_object_unlock(zd, lr->lr_foid);
1750 bbt = ztest_bt_bonus(db);
1751 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1752 crtxg = bbt->bt_crtxg;
1753 lrtxg = lr->lr_common.lrc_txg;
1755 if (zd->zd_zilog->zl_replay) {
1756 ASSERT(lr->lr_size != 0);
1757 ASSERT(lr->lr_mode != 0);
1761 * Randomly change the size and increment the generation.
1763 lr->lr_size = (ztest_random(db->db_size / sizeof (*bbt)) + 1) *
1765 lr->lr_mode = bbt->bt_gen + 1;
1770 * Verify that the current bonus buffer is not newer than our txg.
1772 ztest_bt_verify(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode,
1773 MAX(txg, lrtxg), crtxg);
1775 dmu_buf_will_dirty(db, tx);
1777 ASSERT3U(lr->lr_size, >=, sizeof (*bbt));
1778 ASSERT3U(lr->lr_size, <=, db->db_size);
1779 VERIFY0(dmu_set_bonus(db, lr->lr_size, tx));
1780 bbt = ztest_bt_bonus(db);
1782 ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode, txg, crtxg);
1784 dmu_buf_rele(db, FTAG);
1786 (void) ztest_log_setattr(zd, tx, lr);
1790 ztest_object_unlock(zd, lr->lr_foid);
1795 zil_replay_func_t *ztest_replay_vector[TX_MAX_TYPE] = {
1796 NULL, /* 0 no such transaction type */
1797 ztest_replay_create, /* TX_CREATE */
1798 NULL, /* TX_MKDIR */
1799 NULL, /* TX_MKXATTR */
1800 NULL, /* TX_SYMLINK */
1801 ztest_replay_remove, /* TX_REMOVE */
1802 NULL, /* TX_RMDIR */
1804 NULL, /* TX_RENAME */
1805 ztest_replay_write, /* TX_WRITE */
1806 ztest_replay_truncate, /* TX_TRUNCATE */
1807 ztest_replay_setattr, /* TX_SETATTR */
1809 NULL, /* TX_CREATE_ACL */
1810 NULL, /* TX_CREATE_ATTR */
1811 NULL, /* TX_CREATE_ACL_ATTR */
1812 NULL, /* TX_MKDIR_ACL */
1813 NULL, /* TX_MKDIR_ATTR */
1814 NULL, /* TX_MKDIR_ACL_ATTR */
1815 NULL, /* TX_WRITE2 */
1819 * ZIL get_data callbacks
1823 ztest_get_done(zgd_t *zgd, int error)
1825 ztest_ds_t *zd = zgd->zgd_private;
1826 uint64_t object = zgd->zgd_rl->rl_object;
1829 dmu_buf_rele(zgd->zgd_db, zgd);
1831 ztest_range_unlock(zgd->zgd_rl);
1832 ztest_object_unlock(zd, object);
1834 if (error == 0 && zgd->zgd_bp)
1835 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
1837 umem_free(zgd, sizeof (*zgd));
1841 ztest_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
1843 ztest_ds_t *zd = arg;
1844 objset_t *os = zd->zd_os;
1845 uint64_t object = lr->lr_foid;
1846 uint64_t offset = lr->lr_offset;
1847 uint64_t size = lr->lr_length;
1848 blkptr_t *bp = &lr->lr_blkptr;
1849 uint64_t txg = lr->lr_common.lrc_txg;
1851 dmu_object_info_t doi;
1856 ztest_object_lock(zd, object, RL_READER);
1857 error = dmu_bonus_hold(os, object, FTAG, &db);
1859 ztest_object_unlock(zd, object);
1863 crtxg = ztest_bt_bonus(db)->bt_crtxg;
1865 if (crtxg == 0 || crtxg > txg) {
1866 dmu_buf_rele(db, FTAG);
1867 ztest_object_unlock(zd, object);
1871 dmu_object_info_from_db(db, &doi);
1872 dmu_buf_rele(db, FTAG);
1875 zgd = umem_zalloc(sizeof (*zgd), UMEM_NOFAIL);
1876 zgd->zgd_zilog = zd->zd_zilog;
1877 zgd->zgd_private = zd;
1879 if (buf != NULL) { /* immediate write */
1880 zgd->zgd_rl = ztest_range_lock(zd, object, offset, size,
1883 error = dmu_read(os, object, offset, size, buf,
1884 DMU_READ_NO_PREFETCH);
1887 size = doi.doi_data_block_size;
1889 offset = P2ALIGN(offset, size);
1891 ASSERT(offset < size);
1895 zgd->zgd_rl = ztest_range_lock(zd, object, offset, size,
1898 error = dmu_buf_hold(os, object, offset, zgd, &db,
1899 DMU_READ_NO_PREFETCH);
1902 blkptr_t *obp = dmu_buf_get_blkptr(db);
1904 ASSERT(BP_IS_HOLE(bp));
1911 ASSERT(db->db_offset == offset);
1912 ASSERT(db->db_size == size);
1914 error = dmu_sync(zio, lr->lr_common.lrc_txg,
1915 ztest_get_done, zgd);
1922 ztest_get_done(zgd, error);
1928 ztest_lr_alloc(size_t lrsize, char *name)
1931 size_t namesize = name ? strlen(name) + 1 : 0;
1933 lr = umem_zalloc(lrsize + namesize, UMEM_NOFAIL);
1936 bcopy(name, lr + lrsize, namesize);
1942 ztest_lr_free(void *lr, size_t lrsize, char *name)
1944 size_t namesize = name ? strlen(name) + 1 : 0;
1946 umem_free(lr, lrsize + namesize);
1950 * Lookup a bunch of objects. Returns the number of objects not found.
1953 ztest_lookup(ztest_ds_t *zd, ztest_od_t *od, int count)
1958 ASSERT(_mutex_held(&zd->zd_dirobj_lock));
1960 for (int i = 0; i < count; i++, od++) {
1962 error = zap_lookup(zd->zd_os, od->od_dir, od->od_name,
1963 sizeof (uint64_t), 1, &od->od_object);
1965 ASSERT(error == ENOENT);
1966 ASSERT(od->od_object == 0);
1970 ztest_block_tag_t *bbt;
1971 dmu_object_info_t doi;
1973 ASSERT(od->od_object != 0);
1974 ASSERT(missing == 0); /* there should be no gaps */
1976 ztest_object_lock(zd, od->od_object, RL_READER);
1977 VERIFY3U(0, ==, dmu_bonus_hold(zd->zd_os,
1978 od->od_object, FTAG, &db));
1979 dmu_object_info_from_db(db, &doi);
1980 bbt = ztest_bt_bonus(db);
1981 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1982 od->od_type = doi.doi_type;
1983 od->od_blocksize = doi.doi_data_block_size;
1984 od->od_gen = bbt->bt_gen;
1985 dmu_buf_rele(db, FTAG);
1986 ztest_object_unlock(zd, od->od_object);
1994 ztest_create(ztest_ds_t *zd, ztest_od_t *od, int count)
1998 ASSERT(_mutex_held(&zd->zd_dirobj_lock));
2000 for (int i = 0; i < count; i++, od++) {
2007 lr_create_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
2009 lr->lr_doid = od->od_dir;
2010 lr->lr_foid = 0; /* 0 to allocate, > 0 to claim */
2011 lr->lrz_type = od->od_crtype;
2012 lr->lrz_blocksize = od->od_crblocksize;
2013 lr->lrz_ibshift = ztest_random_ibshift();
2014 lr->lrz_bonustype = DMU_OT_UINT64_OTHER;
2015 lr->lrz_bonuslen = dmu_bonus_max();
2016 lr->lr_gen = od->od_crgen;
2017 lr->lr_crtime[0] = time(NULL);
2019 if (ztest_replay_create(zd, lr, B_FALSE) != 0) {
2020 ASSERT(missing == 0);
2024 od->od_object = lr->lr_foid;
2025 od->od_type = od->od_crtype;
2026 od->od_blocksize = od->od_crblocksize;
2027 od->od_gen = od->od_crgen;
2028 ASSERT(od->od_object != 0);
2031 ztest_lr_free(lr, sizeof (*lr), od->od_name);
2038 ztest_remove(ztest_ds_t *zd, ztest_od_t *od, int count)
2043 ASSERT(_mutex_held(&zd->zd_dirobj_lock));
2047 for (int i = count - 1; i >= 0; i--, od--) {
2054 * No object was found.
2056 if (od->od_object == 0)
2059 lr_remove_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
2061 lr->lr_doid = od->od_dir;
2063 if ((error = ztest_replay_remove(zd, lr, B_FALSE)) != 0) {
2064 ASSERT3U(error, ==, ENOSPC);
2069 ztest_lr_free(lr, sizeof (*lr), od->od_name);
2076 ztest_write(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size,
2082 lr = ztest_lr_alloc(sizeof (*lr) + size, NULL);
2084 lr->lr_foid = object;
2085 lr->lr_offset = offset;
2086 lr->lr_length = size;
2088 BP_ZERO(&lr->lr_blkptr);
2090 bcopy(data, lr + 1, size);
2092 error = ztest_replay_write(zd, lr, B_FALSE);
2094 ztest_lr_free(lr, sizeof (*lr) + size, NULL);
2100 ztest_truncate(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2105 lr = ztest_lr_alloc(sizeof (*lr), NULL);
2107 lr->lr_foid = object;
2108 lr->lr_offset = offset;
2109 lr->lr_length = size;
2111 error = ztest_replay_truncate(zd, lr, B_FALSE);
2113 ztest_lr_free(lr, sizeof (*lr), NULL);
2119 ztest_setattr(ztest_ds_t *zd, uint64_t object)
2124 lr = ztest_lr_alloc(sizeof (*lr), NULL);
2126 lr->lr_foid = object;
2130 error = ztest_replay_setattr(zd, lr, B_FALSE);
2132 ztest_lr_free(lr, sizeof (*lr), NULL);
2138 ztest_prealloc(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2140 objset_t *os = zd->zd_os;
2145 txg_wait_synced(dmu_objset_pool(os), 0);
2147 ztest_object_lock(zd, object, RL_READER);
2148 rl = ztest_range_lock(zd, object, offset, size, RL_WRITER);
2150 tx = dmu_tx_create(os);
2152 dmu_tx_hold_write(tx, object, offset, size);
2154 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
2157 dmu_prealloc(os, object, offset, size, tx);
2159 txg_wait_synced(dmu_objset_pool(os), txg);
2161 (void) dmu_free_long_range(os, object, offset, size);
2164 ztest_range_unlock(rl);
2165 ztest_object_unlock(zd, object);
2169 ztest_io(ztest_ds_t *zd, uint64_t object, uint64_t offset)
2172 ztest_block_tag_t wbt;
2173 dmu_object_info_t doi;
2174 enum ztest_io_type io_type;
2178 VERIFY(dmu_object_info(zd->zd_os, object, &doi) == 0);
2179 blocksize = doi.doi_data_block_size;
2180 data = umem_alloc(blocksize, UMEM_NOFAIL);
2183 * Pick an i/o type at random, biased toward writing block tags.
2185 io_type = ztest_random(ZTEST_IO_TYPES);
2186 if (ztest_random(2) == 0)
2187 io_type = ZTEST_IO_WRITE_TAG;
2189 (void) rw_rdlock(&zd->zd_zilog_lock);
2193 case ZTEST_IO_WRITE_TAG:
2194 ztest_bt_generate(&wbt, zd->zd_os, object, offset, 0, 0, 0);
2195 (void) ztest_write(zd, object, offset, sizeof (wbt), &wbt);
2198 case ZTEST_IO_WRITE_PATTERN:
2199 (void) memset(data, 'a' + (object + offset) % 5, blocksize);
2200 if (ztest_random(2) == 0) {
2202 * Induce fletcher2 collisions to ensure that
2203 * zio_ddt_collision() detects and resolves them
2204 * when using fletcher2-verify for deduplication.
2206 ((uint64_t *)data)[0] ^= 1ULL << 63;
2207 ((uint64_t *)data)[4] ^= 1ULL << 63;
2209 (void) ztest_write(zd, object, offset, blocksize, data);
2212 case ZTEST_IO_WRITE_ZEROES:
2213 bzero(data, blocksize);
2214 (void) ztest_write(zd, object, offset, blocksize, data);
2217 case ZTEST_IO_TRUNCATE:
2218 (void) ztest_truncate(zd, object, offset, blocksize);
2221 case ZTEST_IO_SETATTR:
2222 (void) ztest_setattr(zd, object);
2225 case ZTEST_IO_REWRITE:
2226 (void) rw_rdlock(&ztest_name_lock);
2227 err = ztest_dsl_prop_set_uint64(zd->zd_name,
2228 ZFS_PROP_CHECKSUM, spa_dedup_checksum(ztest_spa),
2230 VERIFY(err == 0 || err == ENOSPC);
2231 err = ztest_dsl_prop_set_uint64(zd->zd_name,
2232 ZFS_PROP_COMPRESSION,
2233 ztest_random_dsl_prop(ZFS_PROP_COMPRESSION),
2235 VERIFY(err == 0 || err == ENOSPC);
2236 (void) rw_unlock(&ztest_name_lock);
2238 VERIFY0(dmu_read(zd->zd_os, object, offset, blocksize, data,
2239 DMU_READ_NO_PREFETCH));
2241 (void) ztest_write(zd, object, offset, blocksize, data);
2245 (void) rw_unlock(&zd->zd_zilog_lock);
2247 umem_free(data, blocksize);
2251 * Initialize an object description template.
2254 ztest_od_init(ztest_od_t *od, uint64_t id, char *tag, uint64_t index,
2255 dmu_object_type_t type, uint64_t blocksize, uint64_t gen)
2257 od->od_dir = ZTEST_DIROBJ;
2260 od->od_crtype = type;
2261 od->od_crblocksize = blocksize ? blocksize : ztest_random_blocksize();
2264 od->od_type = DMU_OT_NONE;
2265 od->od_blocksize = 0;
2268 (void) snprintf(od->od_name, sizeof (od->od_name), "%s(%lld)[%llu]",
2269 tag, (int64_t)id, index);
2273 * Lookup or create the objects for a test using the od template.
2274 * If the objects do not all exist, or if 'remove' is specified,
2275 * remove any existing objects and create new ones. Otherwise,
2276 * use the existing objects.
2279 ztest_object_init(ztest_ds_t *zd, ztest_od_t *od, size_t size, boolean_t remove)
2281 int count = size / sizeof (*od);
2284 VERIFY(mutex_lock(&zd->zd_dirobj_lock) == 0);
2285 if ((ztest_lookup(zd, od, count) != 0 || remove) &&
2286 (ztest_remove(zd, od, count) != 0 ||
2287 ztest_create(zd, od, count) != 0))
2290 VERIFY(mutex_unlock(&zd->zd_dirobj_lock) == 0);
2297 ztest_zil_commit(ztest_ds_t *zd, uint64_t id)
2299 zilog_t *zilog = zd->zd_zilog;
2301 (void) rw_rdlock(&zd->zd_zilog_lock);
2303 zil_commit(zilog, ztest_random(ZTEST_OBJECTS));
2306 * Remember the committed values in zd, which is in parent/child
2307 * shared memory. If we die, the next iteration of ztest_run()
2308 * will verify that the log really does contain this record.
2310 mutex_enter(&zilog->zl_lock);
2311 ASSERT(zd->zd_shared != NULL);
2312 ASSERT3U(zd->zd_shared->zd_seq, <=, zilog->zl_commit_lr_seq);
2313 zd->zd_shared->zd_seq = zilog->zl_commit_lr_seq;
2314 mutex_exit(&zilog->zl_lock);
2316 (void) rw_unlock(&zd->zd_zilog_lock);
2320 * This function is designed to simulate the operations that occur during a
2321 * mount/unmount operation. We hold the dataset across these operations in an
2322 * attempt to expose any implicit assumptions about ZIL management.
2326 ztest_zil_remount(ztest_ds_t *zd, uint64_t id)
2328 objset_t *os = zd->zd_os;
2331 * We grab the zd_dirobj_lock to ensure that no other thread is
2332 * updating the zil (i.e. adding in-memory log records) and the
2333 * zd_zilog_lock to block any I/O.
2335 VERIFY0(mutex_lock(&zd->zd_dirobj_lock));
2336 (void) rw_wrlock(&zd->zd_zilog_lock);
2338 /* zfsvfs_teardown() */
2339 zil_close(zd->zd_zilog);
2341 /* zfsvfs_setup() */
2342 VERIFY(zil_open(os, ztest_get_data) == zd->zd_zilog);
2343 zil_replay(os, zd, ztest_replay_vector);
2345 (void) rw_unlock(&zd->zd_zilog_lock);
2346 VERIFY(mutex_unlock(&zd->zd_dirobj_lock) == 0);
2350 * Verify that we can't destroy an active pool, create an existing pool,
2351 * or create a pool with a bad vdev spec.
2355 ztest_spa_create_destroy(ztest_ds_t *zd, uint64_t id)
2357 ztest_shared_opts_t *zo = &ztest_opts;
2362 * Attempt to create using a bad file.
2364 nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1);
2365 VERIFY3U(ENOENT, ==,
2366 spa_create("ztest_bad_file", nvroot, NULL, NULL));
2367 nvlist_free(nvroot);
2370 * Attempt to create using a bad mirror.
2372 nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 2, 1);
2373 VERIFY3U(ENOENT, ==,
2374 spa_create("ztest_bad_mirror", nvroot, NULL, NULL));
2375 nvlist_free(nvroot);
2378 * Attempt to create an existing pool. It shouldn't matter
2379 * what's in the nvroot; we should fail with EEXIST.
2381 (void) rw_rdlock(&ztest_name_lock);
2382 nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1);
2383 VERIFY3U(EEXIST, ==, spa_create(zo->zo_pool, nvroot, NULL, NULL));
2384 nvlist_free(nvroot);
2385 VERIFY3U(0, ==, spa_open(zo->zo_pool, &spa, FTAG));
2386 VERIFY3U(EBUSY, ==, spa_destroy(zo->zo_pool));
2387 spa_close(spa, FTAG);
2389 (void) rw_unlock(&ztest_name_lock);
2394 ztest_spa_upgrade(ztest_ds_t *zd, uint64_t id)
2397 uint64_t initial_version = SPA_VERSION_INITIAL;
2398 uint64_t version, newversion;
2399 nvlist_t *nvroot, *props;
2402 VERIFY0(mutex_lock(&ztest_vdev_lock));
2403 name = kmem_asprintf("%s_upgrade", ztest_opts.zo_pool);
2406 * Clean up from previous runs.
2408 (void) spa_destroy(name);
2410 nvroot = make_vdev_root(NULL, NULL, name, ztest_opts.zo_vdev_size, 0,
2411 0, ztest_opts.zo_raidz, ztest_opts.zo_mirrors, 1);
2414 * If we're configuring a RAIDZ device then make sure that the
2415 * the initial version is capable of supporting that feature.
2417 switch (ztest_opts.zo_raidz_parity) {
2420 initial_version = SPA_VERSION_INITIAL;
2423 initial_version = SPA_VERSION_RAIDZ2;
2426 initial_version = SPA_VERSION_RAIDZ3;
2431 * Create a pool with a spa version that can be upgraded. Pick
2432 * a value between initial_version and SPA_VERSION_BEFORE_FEATURES.
2435 version = ztest_random_spa_version(initial_version);
2436 } while (version > SPA_VERSION_BEFORE_FEATURES);
2438 props = fnvlist_alloc();
2439 fnvlist_add_uint64(props,
2440 zpool_prop_to_name(ZPOOL_PROP_VERSION), version);
2441 VERIFY0(spa_create(name, nvroot, props, NULL));
2442 fnvlist_free(nvroot);
2443 fnvlist_free(props);
2445 VERIFY0(spa_open(name, &spa, FTAG));
2446 VERIFY3U(spa_version(spa), ==, version);
2447 newversion = ztest_random_spa_version(version + 1);
2449 if (ztest_opts.zo_verbose >= 4) {
2450 (void) printf("upgrading spa version from %llu to %llu\n",
2451 (u_longlong_t)version, (u_longlong_t)newversion);
2454 spa_upgrade(spa, newversion);
2455 VERIFY3U(spa_version(spa), >, version);
2456 VERIFY3U(spa_version(spa), ==, fnvlist_lookup_uint64(spa->spa_config,
2457 zpool_prop_to_name(ZPOOL_PROP_VERSION)));
2458 spa_close(spa, FTAG);
2461 VERIFY0(mutex_unlock(&ztest_vdev_lock));
2465 vdev_lookup_by_path(vdev_t *vd, const char *path)
2469 if (vd->vdev_path != NULL && strcmp(path, vd->vdev_path) == 0)
2472 for (int c = 0; c < vd->vdev_children; c++)
2473 if ((mvd = vdev_lookup_by_path(vd->vdev_child[c], path)) !=
2481 * Find the first available hole which can be used as a top-level.
2484 find_vdev_hole(spa_t *spa)
2486 vdev_t *rvd = spa->spa_root_vdev;
2489 ASSERT(spa_config_held(spa, SCL_VDEV, RW_READER) == SCL_VDEV);
2491 for (c = 0; c < rvd->vdev_children; c++) {
2492 vdev_t *cvd = rvd->vdev_child[c];
2494 if (cvd->vdev_ishole)
2501 * Verify that vdev_add() works as expected.
2505 ztest_vdev_add_remove(ztest_ds_t *zd, uint64_t id)
2507 ztest_shared_t *zs = ztest_shared;
2508 spa_t *spa = ztest_spa;
2514 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2515 leaves = MAX(zs->zs_mirrors + zs->zs_splits, 1) * ztest_opts.zo_raidz;
2517 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2519 ztest_shared->zs_vdev_next_leaf = find_vdev_hole(spa) * leaves;
2522 * If we have slogs then remove them 1/4 of the time.
2524 if (spa_has_slogs(spa) && ztest_random(4) == 0) {
2526 * Grab the guid from the head of the log class rotor.
2528 guid = spa_log_class(spa)->mc_rotor->mg_vd->vdev_guid;
2530 spa_config_exit(spa, SCL_VDEV, FTAG);
2533 * We have to grab the zs_name_lock as writer to
2534 * prevent a race between removing a slog (dmu_objset_find)
2535 * and destroying a dataset. Removing the slog will
2536 * grab a reference on the dataset which may cause
2537 * dmu_objset_destroy() to fail with EBUSY thus
2538 * leaving the dataset in an inconsistent state.
2540 VERIFY(rw_wrlock(&ztest_name_lock) == 0);
2541 error = spa_vdev_remove(spa, guid, B_FALSE);
2542 VERIFY(rw_unlock(&ztest_name_lock) == 0);
2544 if (error && error != EEXIST)
2545 fatal(0, "spa_vdev_remove() = %d", error);
2547 spa_config_exit(spa, SCL_VDEV, FTAG);
2550 * Make 1/4 of the devices be log devices.
2552 nvroot = make_vdev_root(NULL, NULL, NULL,
2553 ztest_opts.zo_vdev_size, 0,
2554 ztest_random(4) == 0, ztest_opts.zo_raidz,
2557 error = spa_vdev_add(spa, nvroot);
2558 nvlist_free(nvroot);
2560 if (error == ENOSPC)
2561 ztest_record_enospc("spa_vdev_add");
2562 else if (error != 0)
2563 fatal(0, "spa_vdev_add() = %d", error);
2566 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2570 * Verify that adding/removing aux devices (l2arc, hot spare) works as expected.
2574 ztest_vdev_aux_add_remove(ztest_ds_t *zd, uint64_t id)
2576 ztest_shared_t *zs = ztest_shared;
2577 spa_t *spa = ztest_spa;
2578 vdev_t *rvd = spa->spa_root_vdev;
2579 spa_aux_vdev_t *sav;
2584 if (ztest_random(2) == 0) {
2585 sav = &spa->spa_spares;
2586 aux = ZPOOL_CONFIG_SPARES;
2588 sav = &spa->spa_l2cache;
2589 aux = ZPOOL_CONFIG_L2CACHE;
2592 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2594 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2596 if (sav->sav_count != 0 && ztest_random(4) == 0) {
2598 * Pick a random device to remove.
2600 guid = sav->sav_vdevs[ztest_random(sav->sav_count)]->vdev_guid;
2603 * Find an unused device we can add.
2605 zs->zs_vdev_aux = 0;
2607 char path[MAXPATHLEN];
2609 (void) snprintf(path, sizeof (path), ztest_aux_template,
2610 ztest_opts.zo_dir, ztest_opts.zo_pool, aux,
2612 for (c = 0; c < sav->sav_count; c++)
2613 if (strcmp(sav->sav_vdevs[c]->vdev_path,
2616 if (c == sav->sav_count &&
2617 vdev_lookup_by_path(rvd, path) == NULL)
2623 spa_config_exit(spa, SCL_VDEV, FTAG);
2629 nvlist_t *nvroot = make_vdev_root(NULL, aux, NULL,
2630 (ztest_opts.zo_vdev_size * 5) / 4, 0, 0, 0, 0, 1);
2631 error = spa_vdev_add(spa, nvroot);
2633 fatal(0, "spa_vdev_add(%p) = %d", nvroot, error);
2634 nvlist_free(nvroot);
2637 * Remove an existing device. Sometimes, dirty its
2638 * vdev state first to make sure we handle removal
2639 * of devices that have pending state changes.
2641 if (ztest_random(2) == 0)
2642 (void) vdev_online(spa, guid, 0, NULL);
2644 error = spa_vdev_remove(spa, guid, B_FALSE);
2645 if (error != 0 && error != EBUSY)
2646 fatal(0, "spa_vdev_remove(%llu) = %d", guid, error);
2649 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2653 * split a pool if it has mirror tlvdevs
2657 ztest_split_pool(ztest_ds_t *zd, uint64_t id)
2659 ztest_shared_t *zs = ztest_shared;
2660 spa_t *spa = ztest_spa;
2661 vdev_t *rvd = spa->spa_root_vdev;
2662 nvlist_t *tree, **child, *config, *split, **schild;
2663 uint_t c, children, schildren = 0, lastlogid = 0;
2666 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2668 /* ensure we have a useable config; mirrors of raidz aren't supported */
2669 if (zs->zs_mirrors < 3 || ztest_opts.zo_raidz > 1) {
2670 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2674 /* clean up the old pool, if any */
2675 (void) spa_destroy("splitp");
2677 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2679 /* generate a config from the existing config */
2680 mutex_enter(&spa->spa_props_lock);
2681 VERIFY(nvlist_lookup_nvlist(spa->spa_config, ZPOOL_CONFIG_VDEV_TREE,
2683 mutex_exit(&spa->spa_props_lock);
2685 VERIFY(nvlist_lookup_nvlist_array(tree, ZPOOL_CONFIG_CHILDREN, &child,
2688 schild = malloc(rvd->vdev_children * sizeof (nvlist_t *));
2689 for (c = 0; c < children; c++) {
2690 vdev_t *tvd = rvd->vdev_child[c];
2694 if (tvd->vdev_islog || tvd->vdev_ops == &vdev_hole_ops) {
2695 VERIFY(nvlist_alloc(&schild[schildren], NV_UNIQUE_NAME,
2697 VERIFY(nvlist_add_string(schild[schildren],
2698 ZPOOL_CONFIG_TYPE, VDEV_TYPE_HOLE) == 0);
2699 VERIFY(nvlist_add_uint64(schild[schildren],
2700 ZPOOL_CONFIG_IS_HOLE, 1) == 0);
2702 lastlogid = schildren;
2707 VERIFY(nvlist_lookup_nvlist_array(child[c],
2708 ZPOOL_CONFIG_CHILDREN, &mchild, &mchildren) == 0);
2709 VERIFY(nvlist_dup(mchild[0], &schild[schildren++], 0) == 0);
2712 /* OK, create a config that can be used to split */
2713 VERIFY(nvlist_alloc(&split, NV_UNIQUE_NAME, 0) == 0);
2714 VERIFY(nvlist_add_string(split, ZPOOL_CONFIG_TYPE,
2715 VDEV_TYPE_ROOT) == 0);
2716 VERIFY(nvlist_add_nvlist_array(split, ZPOOL_CONFIG_CHILDREN, schild,
2717 lastlogid != 0 ? lastlogid : schildren) == 0);
2719 VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, 0) == 0);
2720 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, split) == 0);
2722 for (c = 0; c < schildren; c++)
2723 nvlist_free(schild[c]);
2727 spa_config_exit(spa, SCL_VDEV, FTAG);
2729 (void) rw_wrlock(&ztest_name_lock);
2730 error = spa_vdev_split_mirror(spa, "splitp", config, NULL, B_FALSE);
2731 (void) rw_unlock(&ztest_name_lock);
2733 nvlist_free(config);
2736 (void) printf("successful split - results:\n");
2737 mutex_enter(&spa_namespace_lock);
2738 show_pool_stats(spa);
2739 show_pool_stats(spa_lookup("splitp"));
2740 mutex_exit(&spa_namespace_lock);
2744 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2749 * Verify that we can attach and detach devices.
2753 ztest_vdev_attach_detach(ztest_ds_t *zd, uint64_t id)
2755 ztest_shared_t *zs = ztest_shared;
2756 spa_t *spa = ztest_spa;
2757 spa_aux_vdev_t *sav = &spa->spa_spares;
2758 vdev_t *rvd = spa->spa_root_vdev;
2759 vdev_t *oldvd, *newvd, *pvd;
2763 uint64_t ashift = ztest_get_ashift();
2764 uint64_t oldguid, pguid;
2765 uint64_t oldsize, newsize;
2766 char oldpath[MAXPATHLEN], newpath[MAXPATHLEN];
2768 int oldvd_has_siblings = B_FALSE;
2769 int newvd_is_spare = B_FALSE;
2771 int error, expected_error;
2773 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2774 leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz;
2776 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2779 * Decide whether to do an attach or a replace.
2781 replacing = ztest_random(2);
2784 * Pick a random top-level vdev.
2786 top = ztest_random_vdev_top(spa, B_TRUE);
2789 * Pick a random leaf within it.
2791 leaf = ztest_random(leaves);
2796 oldvd = rvd->vdev_child[top];
2797 if (zs->zs_mirrors >= 1) {
2798 ASSERT(oldvd->vdev_ops == &vdev_mirror_ops);
2799 ASSERT(oldvd->vdev_children >= zs->zs_mirrors);
2800 oldvd = oldvd->vdev_child[leaf / ztest_opts.zo_raidz];
2802 if (ztest_opts.zo_raidz > 1) {
2803 ASSERT(oldvd->vdev_ops == &vdev_raidz_ops);
2804 ASSERT(oldvd->vdev_children == ztest_opts.zo_raidz);
2805 oldvd = oldvd->vdev_child[leaf % ztest_opts.zo_raidz];
2809 * If we're already doing an attach or replace, oldvd may be a
2810 * mirror vdev -- in which case, pick a random child.
2812 while (oldvd->vdev_children != 0) {
2813 oldvd_has_siblings = B_TRUE;
2814 ASSERT(oldvd->vdev_children >= 2);
2815 oldvd = oldvd->vdev_child[ztest_random(oldvd->vdev_children)];
2818 oldguid = oldvd->vdev_guid;
2819 oldsize = vdev_get_min_asize(oldvd);
2820 oldvd_is_log = oldvd->vdev_top->vdev_islog;
2821 (void) strcpy(oldpath, oldvd->vdev_path);
2822 pvd = oldvd->vdev_parent;
2823 pguid = pvd->vdev_guid;
2826 * If oldvd has siblings, then half of the time, detach it.
2828 if (oldvd_has_siblings && ztest_random(2) == 0) {
2829 spa_config_exit(spa, SCL_VDEV, FTAG);
2830 error = spa_vdev_detach(spa, oldguid, pguid, B_FALSE);
2831 if (error != 0 && error != ENODEV && error != EBUSY &&
2833 fatal(0, "detach (%s) returned %d", oldpath, error);
2834 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2839 * For the new vdev, choose with equal probability between the two
2840 * standard paths (ending in either 'a' or 'b') or a random hot spare.
2842 if (sav->sav_count != 0 && ztest_random(3) == 0) {
2843 newvd = sav->sav_vdevs[ztest_random(sav->sav_count)];
2844 newvd_is_spare = B_TRUE;
2845 (void) strcpy(newpath, newvd->vdev_path);
2847 (void) snprintf(newpath, sizeof (newpath), ztest_dev_template,
2848 ztest_opts.zo_dir, ztest_opts.zo_pool,
2849 top * leaves + leaf);
2850 if (ztest_random(2) == 0)
2851 newpath[strlen(newpath) - 1] = 'b';
2852 newvd = vdev_lookup_by_path(rvd, newpath);
2856 newsize = vdev_get_min_asize(newvd);
2859 * Make newsize a little bigger or smaller than oldsize.
2860 * If it's smaller, the attach should fail.
2861 * If it's larger, and we're doing a replace,
2862 * we should get dynamic LUN growth when we're done.
2864 newsize = 10 * oldsize / (9 + ztest_random(3));
2868 * If pvd is not a mirror or root, the attach should fail with ENOTSUP,
2869 * unless it's a replace; in that case any non-replacing parent is OK.
2871 * If newvd is already part of the pool, it should fail with EBUSY.
2873 * If newvd is too small, it should fail with EOVERFLOW.
2875 if (pvd->vdev_ops != &vdev_mirror_ops &&
2876 pvd->vdev_ops != &vdev_root_ops && (!replacing ||
2877 pvd->vdev_ops == &vdev_replacing_ops ||
2878 pvd->vdev_ops == &vdev_spare_ops))
2879 expected_error = ENOTSUP;
2880 else if (newvd_is_spare && (!replacing || oldvd_is_log))
2881 expected_error = ENOTSUP;
2882 else if (newvd == oldvd)
2883 expected_error = replacing ? 0 : EBUSY;
2884 else if (vdev_lookup_by_path(rvd, newpath) != NULL)
2885 expected_error = EBUSY;
2886 else if (newsize < oldsize)
2887 expected_error = EOVERFLOW;
2888 else if (ashift > oldvd->vdev_top->vdev_ashift)
2889 expected_error = EDOM;
2893 spa_config_exit(spa, SCL_VDEV, FTAG);
2896 * Build the nvlist describing newpath.
2898 root = make_vdev_root(newpath, NULL, NULL, newvd == NULL ? newsize : 0,
2899 ashift, 0, 0, 0, 1);
2901 error = spa_vdev_attach(spa, oldguid, root, replacing);
2906 * If our parent was the replacing vdev, but the replace completed,
2907 * then instead of failing with ENOTSUP we may either succeed,
2908 * fail with ENODEV, or fail with EOVERFLOW.
2910 if (expected_error == ENOTSUP &&
2911 (error == 0 || error == ENODEV || error == EOVERFLOW))
2912 expected_error = error;
2915 * If someone grew the LUN, the replacement may be too small.
2917 if (error == EOVERFLOW || error == EBUSY)
2918 expected_error = error;
2920 /* XXX workaround 6690467 */
2921 if (error != expected_error && expected_error != EBUSY) {
2922 fatal(0, "attach (%s %llu, %s %llu, %d) "
2923 "returned %d, expected %d",
2924 oldpath, oldsize, newpath,
2925 newsize, replacing, error, expected_error);
2928 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2932 * Callback function which expands the physical size of the vdev.
2935 grow_vdev(vdev_t *vd, void *arg)
2937 spa_t *spa = vd->vdev_spa;
2938 size_t *newsize = arg;
2942 ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
2943 ASSERT(vd->vdev_ops->vdev_op_leaf);
2945 if ((fd = open(vd->vdev_path, O_RDWR)) == -1)
2948 fsize = lseek(fd, 0, SEEK_END);
2949 (void) ftruncate(fd, *newsize);
2951 if (ztest_opts.zo_verbose >= 6) {
2952 (void) printf("%s grew from %lu to %lu bytes\n",
2953 vd->vdev_path, (ulong_t)fsize, (ulong_t)*newsize);
2960 * Callback function which expands a given vdev by calling vdev_online().
2964 online_vdev(vdev_t *vd, void *arg)
2966 spa_t *spa = vd->vdev_spa;
2967 vdev_t *tvd = vd->vdev_top;
2968 uint64_t guid = vd->vdev_guid;
2969 uint64_t generation = spa->spa_config_generation + 1;
2970 vdev_state_t newstate = VDEV_STATE_UNKNOWN;
2973 ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
2974 ASSERT(vd->vdev_ops->vdev_op_leaf);
2976 /* Calling vdev_online will initialize the new metaslabs */
2977 spa_config_exit(spa, SCL_STATE, spa);
2978 error = vdev_online(spa, guid, ZFS_ONLINE_EXPAND, &newstate);
2979 spa_config_enter(spa, SCL_STATE, spa, RW_READER);
2982 * If vdev_online returned an error or the underlying vdev_open
2983 * failed then we abort the expand. The only way to know that
2984 * vdev_open fails is by checking the returned newstate.
2986 if (error || newstate != VDEV_STATE_HEALTHY) {
2987 if (ztest_opts.zo_verbose >= 5) {
2988 (void) printf("Unable to expand vdev, state %llu, "
2989 "error %d\n", (u_longlong_t)newstate, error);
2993 ASSERT3U(newstate, ==, VDEV_STATE_HEALTHY);
2996 * Since we dropped the lock we need to ensure that we're
2997 * still talking to the original vdev. It's possible this
2998 * vdev may have been detached/replaced while we were
2999 * trying to online it.
3001 if (generation != spa->spa_config_generation) {
3002 if (ztest_opts.zo_verbose >= 5) {
3003 (void) printf("vdev configuration has changed, "
3004 "guid %llu, state %llu, expected gen %llu, "
3007 (u_longlong_t)tvd->vdev_state,
3008 (u_longlong_t)generation,
3009 (u_longlong_t)spa->spa_config_generation);
3017 * Traverse the vdev tree calling the supplied function.
3018 * We continue to walk the tree until we either have walked all
3019 * children or we receive a non-NULL return from the callback.
3020 * If a NULL callback is passed, then we just return back the first
3021 * leaf vdev we encounter.
3024 vdev_walk_tree(vdev_t *vd, vdev_t *(*func)(vdev_t *, void *), void *arg)
3026 if (vd->vdev_ops->vdev_op_leaf) {
3030 return (func(vd, arg));
3033 for (uint_t c = 0; c < vd->vdev_children; c++) {
3034 vdev_t *cvd = vd->vdev_child[c];
3035 if ((cvd = vdev_walk_tree(cvd, func, arg)) != NULL)
3042 * Verify that dynamic LUN growth works as expected.
3046 ztest_vdev_LUN_growth(ztest_ds_t *zd, uint64_t id)
3048 spa_t *spa = ztest_spa;
3050 metaslab_class_t *mc;
3051 metaslab_group_t *mg;
3052 size_t psize, newsize;
3054 uint64_t old_class_space, new_class_space, old_ms_count, new_ms_count;
3056 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
3057 spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3059 top = ztest_random_vdev_top(spa, B_TRUE);
3061 tvd = spa->spa_root_vdev->vdev_child[top];
3064 old_ms_count = tvd->vdev_ms_count;
3065 old_class_space = metaslab_class_get_space(mc);
3068 * Determine the size of the first leaf vdev associated with
3069 * our top-level device.
3071 vd = vdev_walk_tree(tvd, NULL, NULL);
3072 ASSERT3P(vd, !=, NULL);
3073 ASSERT(vd->vdev_ops->vdev_op_leaf);
3075 psize = vd->vdev_psize;
3078 * We only try to expand the vdev if it's healthy, less than 4x its
3079 * original size, and it has a valid psize.
3081 if (tvd->vdev_state != VDEV_STATE_HEALTHY ||
3082 psize == 0 || psize >= 4 * ztest_opts.zo_vdev_size) {
3083 spa_config_exit(spa, SCL_STATE, spa);
3084 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3088 newsize = psize + psize / 8;
3089 ASSERT3U(newsize, >, psize);
3091 if (ztest_opts.zo_verbose >= 6) {
3092 (void) printf("Expanding LUN %s from %lu to %lu\n",
3093 vd->vdev_path, (ulong_t)psize, (ulong_t)newsize);
3097 * Growing the vdev is a two step process:
3098 * 1). expand the physical size (i.e. relabel)
3099 * 2). online the vdev to create the new metaslabs
3101 if (vdev_walk_tree(tvd, grow_vdev, &newsize) != NULL ||
3102 vdev_walk_tree(tvd, online_vdev, NULL) != NULL ||
3103 tvd->vdev_state != VDEV_STATE_HEALTHY) {
3104 if (ztest_opts.zo_verbose >= 5) {
3105 (void) printf("Could not expand LUN because "
3106 "the vdev configuration changed.\n");
3108 spa_config_exit(spa, SCL_STATE, spa);
3109 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3113 spa_config_exit(spa, SCL_STATE, spa);
3116 * Expanding the LUN will update the config asynchronously,
3117 * thus we must wait for the async thread to complete any
3118 * pending tasks before proceeding.
3122 mutex_enter(&spa->spa_async_lock);
3123 done = (spa->spa_async_thread == NULL && !spa->spa_async_tasks);
3124 mutex_exit(&spa->spa_async_lock);
3127 txg_wait_synced(spa_get_dsl(spa), 0);
3128 (void) poll(NULL, 0, 100);
3131 spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3133 tvd = spa->spa_root_vdev->vdev_child[top];
3134 new_ms_count = tvd->vdev_ms_count;
3135 new_class_space = metaslab_class_get_space(mc);
3137 if (tvd->vdev_mg != mg || mg->mg_class != mc) {
3138 if (ztest_opts.zo_verbose >= 5) {
3139 (void) printf("Could not verify LUN expansion due to "
3140 "intervening vdev offline or remove.\n");
3142 spa_config_exit(spa, SCL_STATE, spa);
3143 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3148 * Make sure we were able to grow the vdev.
3150 if (new_ms_count <= old_ms_count)
3151 fatal(0, "LUN expansion failed: ms_count %llu <= %llu\n",
3152 old_ms_count, new_ms_count);
3155 * Make sure we were able to grow the pool.
3157 if (new_class_space <= old_class_space)
3158 fatal(0, "LUN expansion failed: class_space %llu <= %llu\n",
3159 old_class_space, new_class_space);
3161 if (ztest_opts.zo_verbose >= 5) {
3162 char oldnumbuf[NN_NUMBUF_SZ], newnumbuf[NN_NUMBUF_SZ];
3164 nicenum(old_class_space, oldnumbuf, sizeof (oldnumbuf));
3165 nicenum(new_class_space, newnumbuf, sizeof (newnumbuf));
3166 (void) printf("%s grew from %s to %s\n",
3167 spa->spa_name, oldnumbuf, newnumbuf);
3170 spa_config_exit(spa, SCL_STATE, spa);
3171 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3175 * Verify that dmu_objset_{create,destroy,open,close} work as expected.
3179 ztest_objset_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
3182 * Create the objects common to all ztest datasets.
3184 VERIFY(zap_create_claim(os, ZTEST_DIROBJ,
3185 DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx) == 0);
3189 ztest_dataset_create(char *dsname)
3191 uint64_t zilset = ztest_random(100);
3192 int err = dmu_objset_create(dsname, DMU_OST_OTHER, 0,
3193 ztest_objset_create_cb, NULL);
3195 if (err || zilset < 80)
3198 if (ztest_opts.zo_verbose >= 6)
3199 (void) printf("Setting dataset %s to sync always\n", dsname);
3200 return (ztest_dsl_prop_set_uint64(dsname, ZFS_PROP_SYNC,
3201 ZFS_SYNC_ALWAYS, B_FALSE));
3206 ztest_objset_destroy_cb(const char *name, void *arg)
3209 dmu_object_info_t doi;
3213 * Verify that the dataset contains a directory object.
3215 VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_TRUE, FTAG, &os));
3216 error = dmu_object_info(os, ZTEST_DIROBJ, &doi);
3217 if (error != ENOENT) {
3218 /* We could have crashed in the middle of destroying it */
3220 ASSERT3U(doi.doi_type, ==, DMU_OT_ZAP_OTHER);
3221 ASSERT3S(doi.doi_physical_blocks_512, >=, 0);
3223 dmu_objset_disown(os, FTAG);
3226 * Destroy the dataset.
3228 if (strchr(name, '@') != NULL) {
3229 VERIFY0(dsl_destroy_snapshot(name, B_FALSE));
3231 VERIFY0(dsl_destroy_head(name));
3237 ztest_snapshot_create(char *osname, uint64_t id)
3239 char snapname[ZFS_MAX_DATASET_NAME_LEN];
3242 (void) snprintf(snapname, sizeof (snapname), "%llu", (u_longlong_t)id);
3244 error = dmu_objset_snapshot_one(osname, snapname);
3245 if (error == ENOSPC) {
3246 ztest_record_enospc(FTAG);
3249 if (error != 0 && error != EEXIST) {
3250 fatal(0, "ztest_snapshot_create(%s@%s) = %d", osname,
3257 ztest_snapshot_destroy(char *osname, uint64_t id)
3259 char snapname[ZFS_MAX_DATASET_NAME_LEN];
3262 (void) snprintf(snapname, sizeof (snapname), "%s@%llu", osname,
3265 error = dsl_destroy_snapshot(snapname, B_FALSE);
3266 if (error != 0 && error != ENOENT)
3267 fatal(0, "ztest_snapshot_destroy(%s) = %d", snapname, error);
3273 ztest_dmu_objset_create_destroy(ztest_ds_t *zd, uint64_t id)
3279 char name[ZFS_MAX_DATASET_NAME_LEN];
3282 (void) rw_rdlock(&ztest_name_lock);
3284 (void) snprintf(name, sizeof (name), "%s/temp_%llu",
3285 ztest_opts.zo_pool, (u_longlong_t)id);
3288 * If this dataset exists from a previous run, process its replay log
3289 * half of the time. If we don't replay it, then dmu_objset_destroy()
3290 * (invoked from ztest_objset_destroy_cb()) should just throw it away.
3292 if (ztest_random(2) == 0 &&
3293 dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os) == 0) {
3294 ztest_zd_init(&zdtmp, NULL, os);
3295 zil_replay(os, &zdtmp, ztest_replay_vector);
3296 ztest_zd_fini(&zdtmp);
3297 dmu_objset_disown(os, FTAG);
3301 * There may be an old instance of the dataset we're about to
3302 * create lying around from a previous run. If so, destroy it
3303 * and all of its snapshots.
3305 (void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
3306 DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
3309 * Verify that the destroyed dataset is no longer in the namespace.
3311 VERIFY3U(ENOENT, ==, dmu_objset_own(name, DMU_OST_OTHER, B_TRUE,
3315 * Verify that we can create a new dataset.
3317 error = ztest_dataset_create(name);
3319 if (error == ENOSPC) {
3320 ztest_record_enospc(FTAG);
3321 (void) rw_unlock(&ztest_name_lock);
3324 fatal(0, "dmu_objset_create(%s) = %d", name, error);
3327 VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os));
3329 ztest_zd_init(&zdtmp, NULL, os);
3332 * Open the intent log for it.
3334 zilog = zil_open(os, ztest_get_data);
3337 * Put some objects in there, do a little I/O to them,
3338 * and randomly take a couple of snapshots along the way.
3340 iters = ztest_random(5);
3341 for (int i = 0; i < iters; i++) {
3342 ztest_dmu_object_alloc_free(&zdtmp, id);
3343 if (ztest_random(iters) == 0)
3344 (void) ztest_snapshot_create(name, i);
3348 * Verify that we cannot create an existing dataset.
3350 VERIFY3U(EEXIST, ==,
3351 dmu_objset_create(name, DMU_OST_OTHER, 0, NULL, NULL));
3354 * Verify that we can hold an objset that is also owned.
3356 VERIFY3U(0, ==, dmu_objset_hold(name, FTAG, &os2));
3357 dmu_objset_rele(os2, FTAG);
3360 * Verify that we cannot own an objset that is already owned.
3363 dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os2));
3366 dmu_objset_disown(os, FTAG);
3367 ztest_zd_fini(&zdtmp);
3369 (void) rw_unlock(&ztest_name_lock);
3373 * Verify that dmu_snapshot_{create,destroy,open,close} work as expected.
3376 ztest_dmu_snapshot_create_destroy(ztest_ds_t *zd, uint64_t id)
3378 (void) rw_rdlock(&ztest_name_lock);
3379 (void) ztest_snapshot_destroy(zd->zd_name, id);
3380 (void) ztest_snapshot_create(zd->zd_name, id);
3381 (void) rw_unlock(&ztest_name_lock);
3385 * Cleanup non-standard snapshots and clones.
3388 ztest_dsl_dataset_cleanup(char *osname, uint64_t id)
3390 char snap1name[ZFS_MAX_DATASET_NAME_LEN];
3391 char clone1name[ZFS_MAX_DATASET_NAME_LEN];
3392 char snap2name[ZFS_MAX_DATASET_NAME_LEN];
3393 char clone2name[ZFS_MAX_DATASET_NAME_LEN];
3394 char snap3name[ZFS_MAX_DATASET_NAME_LEN];
3397 (void) snprintf(snap1name, sizeof (snap1name),
3398 "%s@s1_%llu", osname, id);
3399 (void) snprintf(clone1name, sizeof (clone1name),
3400 "%s/c1_%llu", osname, id);
3401 (void) snprintf(snap2name, sizeof (snap2name),
3402 "%s@s2_%llu", clone1name, id);
3403 (void) snprintf(clone2name, sizeof (clone2name),
3404 "%s/c2_%llu", osname, id);
3405 (void) snprintf(snap3name, sizeof (snap3name),
3406 "%s@s3_%llu", clone1name, id);
3408 error = dsl_destroy_head(clone2name);
3409 if (error && error != ENOENT)
3410 fatal(0, "dsl_destroy_head(%s) = %d", clone2name, error);
3411 error = dsl_destroy_snapshot(snap3name, B_FALSE);
3412 if (error && error != ENOENT)
3413 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap3name, error);
3414 error = dsl_destroy_snapshot(snap2name, B_FALSE);
3415 if (error && error != ENOENT)
3416 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap2name, error);
3417 error = dsl_destroy_head(clone1name);
3418 if (error && error != ENOENT)
3419 fatal(0, "dsl_destroy_head(%s) = %d", clone1name, error);
3420 error = dsl_destroy_snapshot(snap1name, B_FALSE);
3421 if (error && error != ENOENT)
3422 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap1name, error);
3426 * Verify dsl_dataset_promote handles EBUSY
3429 ztest_dsl_dataset_promote_busy(ztest_ds_t *zd, uint64_t id)
3432 char snap1name[ZFS_MAX_DATASET_NAME_LEN];
3433 char clone1name[ZFS_MAX_DATASET_NAME_LEN];
3434 char snap2name[ZFS_MAX_DATASET_NAME_LEN];
3435 char clone2name[ZFS_MAX_DATASET_NAME_LEN];
3436 char snap3name[ZFS_MAX_DATASET_NAME_LEN];
3437 char *osname = zd->zd_name;
3440 (void) rw_rdlock(&ztest_name_lock);
3442 ztest_dsl_dataset_cleanup(osname, id);
3444 (void) snprintf(snap1name, sizeof (snap1name),
3445 "%s@s1_%llu", osname, id);
3446 (void) snprintf(clone1name, sizeof (clone1name),
3447 "%s/c1_%llu", osname, id);
3448 (void) snprintf(snap2name, sizeof (snap2name),
3449 "%s@s2_%llu", clone1name, id);
3450 (void) snprintf(clone2name, sizeof (clone2name),
3451 "%s/c2_%llu", osname, id);
3452 (void) snprintf(snap3name, sizeof (snap3name),
3453 "%s@s3_%llu", clone1name, id);
3455 error = dmu_objset_snapshot_one(osname, strchr(snap1name, '@') + 1);
3456 if (error && error != EEXIST) {
3457 if (error == ENOSPC) {
3458 ztest_record_enospc(FTAG);
3461 fatal(0, "dmu_take_snapshot(%s) = %d", snap1name, error);
3464 error = dmu_objset_clone(clone1name, snap1name);
3466 if (error == ENOSPC) {
3467 ztest_record_enospc(FTAG);
3470 fatal(0, "dmu_objset_create(%s) = %d", clone1name, error);
3473 error = dmu_objset_snapshot_one(clone1name, strchr(snap2name, '@') + 1);
3474 if (error && error != EEXIST) {
3475 if (error == ENOSPC) {
3476 ztest_record_enospc(FTAG);
3479 fatal(0, "dmu_open_snapshot(%s) = %d", snap2name, error);
3482 error = dmu_objset_snapshot_one(clone1name, strchr(snap3name, '@') + 1);
3483 if (error && error != EEXIST) {
3484 if (error == ENOSPC) {
3485 ztest_record_enospc(FTAG);
3488 fatal(0, "dmu_open_snapshot(%s) = %d", snap3name, error);
3491 error = dmu_objset_clone(clone2name, snap3name);
3493 if (error == ENOSPC) {
3494 ztest_record_enospc(FTAG);
3497 fatal(0, "dmu_objset_create(%s) = %d", clone2name, error);
3500 error = dmu_objset_own(snap2name, DMU_OST_ANY, B_TRUE, FTAG, &os);
3502 fatal(0, "dmu_objset_own(%s) = %d", snap2name, error);
3503 error = dsl_dataset_promote(clone2name, NULL);
3504 if (error == ENOSPC) {
3505 dmu_objset_disown(os, FTAG);
3506 ztest_record_enospc(FTAG);
3510 fatal(0, "dsl_dataset_promote(%s), %d, not EBUSY", clone2name,
3512 dmu_objset_disown(os, FTAG);
3515 ztest_dsl_dataset_cleanup(osname, id);
3517 (void) rw_unlock(&ztest_name_lock);
3521 * Verify that dmu_object_{alloc,free} work as expected.
3524 ztest_dmu_object_alloc_free(ztest_ds_t *zd, uint64_t id)
3527 int batchsize = sizeof (od) / sizeof (od[0]);
3529 for (int b = 0; b < batchsize; b++)
3530 ztest_od_init(&od[b], id, FTAG, b, DMU_OT_UINT64_OTHER, 0, 0);
3533 * Destroy the previous batch of objects, create a new batch,
3534 * and do some I/O on the new objects.
3536 if (ztest_object_init(zd, od, sizeof (od), B_TRUE) != 0)
3539 while (ztest_random(4 * batchsize) != 0)
3540 ztest_io(zd, od[ztest_random(batchsize)].od_object,
3541 ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
3545 * Verify that dmu_{read,write} work as expected.
3548 ztest_dmu_read_write(ztest_ds_t *zd, uint64_t id)
3550 objset_t *os = zd->zd_os;
3553 int i, freeit, error;
3555 bufwad_t *packbuf, *bigbuf, *pack, *bigH, *bigT;
3556 uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
3557 uint64_t chunksize = (1000 + ztest_random(1000)) * sizeof (uint64_t);
3558 uint64_t regions = 997;
3559 uint64_t stride = 123456789ULL;
3560 uint64_t width = 40;
3561 int free_percent = 5;
3564 * This test uses two objects, packobj and bigobj, that are always
3565 * updated together (i.e. in the same tx) so that their contents are
3566 * in sync and can be compared. Their contents relate to each other
3567 * in a simple way: packobj is a dense array of 'bufwad' structures,
3568 * while bigobj is a sparse array of the same bufwads. Specifically,
3569 * for any index n, there are three bufwads that should be identical:
3571 * packobj, at offset n * sizeof (bufwad_t)
3572 * bigobj, at the head of the nth chunk
3573 * bigobj, at the tail of the nth chunk
3575 * The chunk size is arbitrary. It doesn't have to be a power of two,
3576 * and it doesn't have any relation to the object blocksize.
3577 * The only requirement is that it can hold at least two bufwads.
3579 * Normally, we write the bufwad to each of these locations.
3580 * However, free_percent of the time we instead write zeroes to
3581 * packobj and perform a dmu_free_range() on bigobj. By comparing
3582 * bigobj to packobj, we can verify that the DMU is correctly
3583 * tracking which parts of an object are allocated and free,
3584 * and that the contents of the allocated blocks are correct.
3588 * Read the directory info. If it's the first time, set things up.
3590 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, chunksize);
3591 ztest_od_init(&od[1], id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize);
3593 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
3596 bigobj = od[0].od_object;
3597 packobj = od[1].od_object;
3598 chunksize = od[0].od_gen;
3599 ASSERT(chunksize == od[1].od_gen);
3602 * Prefetch a random chunk of the big object.
3603 * Our aim here is to get some async reads in flight
3604 * for blocks that we may free below; the DMU should
3605 * handle this race correctly.
3607 n = ztest_random(regions) * stride + ztest_random(width);
3608 s = 1 + ztest_random(2 * width - 1);
3609 dmu_prefetch(os, bigobj, 0, n * chunksize, s * chunksize,
3610 ZIO_PRIORITY_SYNC_READ);
3613 * Pick a random index and compute the offsets into packobj and bigobj.
3615 n = ztest_random(regions) * stride + ztest_random(width);
3616 s = 1 + ztest_random(width - 1);
3618 packoff = n * sizeof (bufwad_t);
3619 packsize = s * sizeof (bufwad_t);
3621 bigoff = n * chunksize;
3622 bigsize = s * chunksize;
3624 packbuf = umem_alloc(packsize, UMEM_NOFAIL);
3625 bigbuf = umem_alloc(bigsize, UMEM_NOFAIL);
3628 * free_percent of the time, free a range of bigobj rather than
3631 freeit = (ztest_random(100) < free_percent);
3634 * Read the current contents of our objects.
3636 error = dmu_read(os, packobj, packoff, packsize, packbuf,
3639 error = dmu_read(os, bigobj, bigoff, bigsize, bigbuf,
3644 * Get a tx for the mods to both packobj and bigobj.
3646 tx = dmu_tx_create(os);
3648 dmu_tx_hold_write(tx, packobj, packoff, packsize);
3651 dmu_tx_hold_free(tx, bigobj, bigoff, bigsize);
3653 dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
3655 /* This accounts for setting the checksum/compression. */
3656 dmu_tx_hold_bonus(tx, bigobj);
3658 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3660 umem_free(packbuf, packsize);
3661 umem_free(bigbuf, bigsize);
3665 enum zio_checksum cksum;
3667 cksum = (enum zio_checksum)
3668 ztest_random_dsl_prop(ZFS_PROP_CHECKSUM);
3669 } while (cksum >= ZIO_CHECKSUM_LEGACY_FUNCTIONS);
3670 dmu_object_set_checksum(os, bigobj, cksum, tx);
3672 enum zio_compress comp;
3674 comp = (enum zio_compress)
3675 ztest_random_dsl_prop(ZFS_PROP_COMPRESSION);
3676 } while (comp >= ZIO_COMPRESS_LEGACY_FUNCTIONS);
3677 dmu_object_set_compress(os, bigobj, comp, tx);
3680 * For each index from n to n + s, verify that the existing bufwad
3681 * in packobj matches the bufwads at the head and tail of the
3682 * corresponding chunk in bigobj. Then update all three bufwads
3683 * with the new values we want to write out.
3685 for (i = 0; i < s; i++) {
3687 pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
3689 bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
3691 bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
3693 ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
3694 ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
3696 if (pack->bw_txg > txg)
3697 fatal(0, "future leak: got %llx, open txg is %llx",
3700 if (pack->bw_data != 0 && pack->bw_index != n + i)
3701 fatal(0, "wrong index: got %llx, wanted %llx+%llx",
3702 pack->bw_index, n, i);
3704 if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
3705 fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
3707 if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
3708 fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
3711 bzero(pack, sizeof (bufwad_t));
3713 pack->bw_index = n + i;
3715 pack->bw_data = 1 + ztest_random(-2ULL);
3722 * We've verified all the old bufwads, and made new ones.
3723 * Now write them out.
3725 dmu_write(os, packobj, packoff, packsize, packbuf, tx);
3728 if (ztest_opts.zo_verbose >= 7) {
3729 (void) printf("freeing offset %llx size %llx"
3731 (u_longlong_t)bigoff,
3732 (u_longlong_t)bigsize,
3735 VERIFY(0 == dmu_free_range(os, bigobj, bigoff, bigsize, tx));
3737 if (ztest_opts.zo_verbose >= 7) {
3738 (void) printf("writing offset %llx size %llx"
3740 (u_longlong_t)bigoff,
3741 (u_longlong_t)bigsize,
3744 dmu_write(os, bigobj, bigoff, bigsize, bigbuf, tx);
3750 * Sanity check the stuff we just wrote.
3753 void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
3754 void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
3756 VERIFY(0 == dmu_read(os, packobj, packoff,
3757 packsize, packcheck, DMU_READ_PREFETCH));
3758 VERIFY(0 == dmu_read(os, bigobj, bigoff,
3759 bigsize, bigcheck, DMU_READ_PREFETCH));
3761 ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
3762 ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
3764 umem_free(packcheck, packsize);
3765 umem_free(bigcheck, bigsize);
3768 umem_free(packbuf, packsize);
3769 umem_free(bigbuf, bigsize);
3773 compare_and_update_pbbufs(uint64_t s, bufwad_t *packbuf, bufwad_t *bigbuf,
3774 uint64_t bigsize, uint64_t n, uint64_t chunksize, uint64_t txg)
3782 * For each index from n to n + s, verify that the existing bufwad
3783 * in packobj matches the bufwads at the head and tail of the
3784 * corresponding chunk in bigobj. Then update all three bufwads
3785 * with the new values we want to write out.
3787 for (i = 0; i < s; i++) {
3789 pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
3791 bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
3793 bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
3795 ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
3796 ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
3798 if (pack->bw_txg > txg)
3799 fatal(0, "future leak: got %llx, open txg is %llx",
3802 if (pack->bw_data != 0 && pack->bw_index != n + i)
3803 fatal(0, "wrong index: got %llx, wanted %llx+%llx",
3804 pack->bw_index, n, i);
3806 if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
3807 fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
3809 if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
3810 fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
3812 pack->bw_index = n + i;
3814 pack->bw_data = 1 + ztest_random(-2ULL);
3822 ztest_dmu_read_write_zcopy(ztest_ds_t *zd, uint64_t id)
3824 objset_t *os = zd->zd_os;
3830 bufwad_t *packbuf, *bigbuf;
3831 uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
3832 uint64_t blocksize = ztest_random_blocksize();
3833 uint64_t chunksize = blocksize;
3834 uint64_t regions = 997;
3835 uint64_t stride = 123456789ULL;
3837 dmu_buf_t *bonus_db;
3838 arc_buf_t **bigbuf_arcbufs;
3839 dmu_object_info_t doi;
3842 * This test uses two objects, packobj and bigobj, that are always
3843 * updated together (i.e. in the same tx) so that their contents are
3844 * in sync and can be compared. Their contents relate to each other
3845 * in a simple way: packobj is a dense array of 'bufwad' structures,
3846 * while bigobj is a sparse array of the same bufwads. Specifically,
3847 * for any index n, there are three bufwads that should be identical:
3849 * packobj, at offset n * sizeof (bufwad_t)
3850 * bigobj, at the head of the nth chunk
3851 * bigobj, at the tail of the nth chunk
3853 * The chunk size is set equal to bigobj block size so that
3854 * dmu_assign_arcbuf() can be tested for object updates.
3858 * Read the directory info. If it's the first time, set things up.
3860 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
3861 ztest_od_init(&od[1], id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize);
3863 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
3866 bigobj = od[0].od_object;
3867 packobj = od[1].od_object;
3868 blocksize = od[0].od_blocksize;
3869 chunksize = blocksize;
3870 ASSERT(chunksize == od[1].od_gen);
3872 VERIFY(dmu_object_info(os, bigobj, &doi) == 0);
3873 VERIFY(ISP2(doi.doi_data_block_size));
3874 VERIFY(chunksize == doi.doi_data_block_size);
3875 VERIFY(chunksize >= 2 * sizeof (bufwad_t));
3878 * Pick a random index and compute the offsets into packobj and bigobj.
3880 n = ztest_random(regions) * stride + ztest_random(width);
3881 s = 1 + ztest_random(width - 1);
3883 packoff = n * sizeof (bufwad_t);
3884 packsize = s * sizeof (bufwad_t);
3886 bigoff = n * chunksize;
3887 bigsize = s * chunksize;
3889 packbuf = umem_zalloc(packsize, UMEM_NOFAIL);
3890 bigbuf = umem_zalloc(bigsize, UMEM_NOFAIL);
3892 VERIFY3U(0, ==, dmu_bonus_hold(os, bigobj, FTAG, &bonus_db));
3894 bigbuf_arcbufs = umem_zalloc(2 * s * sizeof (arc_buf_t *), UMEM_NOFAIL);
3897 * Iteration 0 test zcopy for DB_UNCACHED dbufs.
3898 * Iteration 1 test zcopy to already referenced dbufs.
3899 * Iteration 2 test zcopy to dirty dbuf in the same txg.
3900 * Iteration 3 test zcopy to dbuf dirty in previous txg.
3901 * Iteration 4 test zcopy when dbuf is no longer dirty.
3902 * Iteration 5 test zcopy when it can't be done.
3903 * Iteration 6 one more zcopy write.
3905 for (i = 0; i < 7; i++) {
3910 * In iteration 5 (i == 5) use arcbufs
3911 * that don't match bigobj blksz to test
3912 * dmu_assign_arcbuf() when it can't directly
3913 * assign an arcbuf to a dbuf.
3915 for (j = 0; j < s; j++) {
3918 dmu_request_arcbuf(bonus_db, chunksize);
3920 bigbuf_arcbufs[2 * j] =
3921 dmu_request_arcbuf(bonus_db, chunksize / 2);
3922 bigbuf_arcbufs[2 * j + 1] =
3923 dmu_request_arcbuf(bonus_db, chunksize / 2);
3928 * Get a tx for the mods to both packobj and bigobj.
3930 tx = dmu_tx_create(os);
3932 dmu_tx_hold_write(tx, packobj, packoff, packsize);
3933 dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
3935 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3937 umem_free(packbuf, packsize);
3938 umem_free(bigbuf, bigsize);
3939 for (j = 0; j < s; j++) {
3941 dmu_return_arcbuf(bigbuf_arcbufs[j]);
3944 bigbuf_arcbufs[2 * j]);
3946 bigbuf_arcbufs[2 * j + 1]);
3949 umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
3950 dmu_buf_rele(bonus_db, FTAG);
3955 * 50% of the time don't read objects in the 1st iteration to
3956 * test dmu_assign_arcbuf() for the case when there're no
3957 * existing dbufs for the specified offsets.
3959 if (i != 0 || ztest_random(2) != 0) {
3960 error = dmu_read(os, packobj, packoff,
3961 packsize, packbuf, DMU_READ_PREFETCH);
3963 error = dmu_read(os, bigobj, bigoff, bigsize,
3964 bigbuf, DMU_READ_PREFETCH);
3967 compare_and_update_pbbufs(s, packbuf, bigbuf, bigsize,
3971 * We've verified all the old bufwads, and made new ones.
3972 * Now write them out.
3974 dmu_write(os, packobj, packoff, packsize, packbuf, tx);
3975 if (ztest_opts.zo_verbose >= 7) {
3976 (void) printf("writing offset %llx size %llx"
3978 (u_longlong_t)bigoff,
3979 (u_longlong_t)bigsize,
3982 for (off = bigoff, j = 0; j < s; j++, off += chunksize) {
3985 bcopy((caddr_t)bigbuf + (off - bigoff),
3986 bigbuf_arcbufs[j]->b_data, chunksize);
3988 bcopy((caddr_t)bigbuf + (off - bigoff),
3989 bigbuf_arcbufs[2 * j]->b_data,
3991 bcopy((caddr_t)bigbuf + (off - bigoff) +
3993 bigbuf_arcbufs[2 * j + 1]->b_data,
3998 VERIFY(dmu_buf_hold(os, bigobj, off,
3999 FTAG, &dbt, DMU_READ_NO_PREFETCH) == 0);
4002 dmu_assign_arcbuf(bonus_db, off,
4003 bigbuf_arcbufs[j], tx);
4005 dmu_assign_arcbuf(bonus_db, off,
4006 bigbuf_arcbufs[2 * j], tx);
4007 dmu_assign_arcbuf(bonus_db,
4008 off + chunksize / 2,
4009 bigbuf_arcbufs[2 * j + 1], tx);
4012 dmu_buf_rele(dbt, FTAG);
4018 * Sanity check the stuff we just wrote.
4021 void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
4022 void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
4024 VERIFY(0 == dmu_read(os, packobj, packoff,
4025 packsize, packcheck, DMU_READ_PREFETCH));
4026 VERIFY(0 == dmu_read(os, bigobj, bigoff,
4027 bigsize, bigcheck, DMU_READ_PREFETCH));
4029 ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
4030 ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
4032 umem_free(packcheck, packsize);
4033 umem_free(bigcheck, bigsize);
4036 txg_wait_open(dmu_objset_pool(os), 0);
4037 } else if (i == 3) {
4038 txg_wait_synced(dmu_objset_pool(os), 0);
4042 dmu_buf_rele(bonus_db, FTAG);
4043 umem_free(packbuf, packsize);
4044 umem_free(bigbuf, bigsize);
4045 umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
4050 ztest_dmu_write_parallel(ztest_ds_t *zd, uint64_t id)
4053 uint64_t offset = (1ULL << (ztest_random(20) + 43)) +
4054 (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
4057 * Have multiple threads write to large offsets in an object
4058 * to verify that parallel writes to an object -- even to the
4059 * same blocks within the object -- doesn't cause any trouble.
4061 ztest_od_init(&od[0], ID_PARALLEL, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
4063 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4066 while (ztest_random(10) != 0)
4067 ztest_io(zd, od[0].od_object, offset);
4071 ztest_dmu_prealloc(ztest_ds_t *zd, uint64_t id)
4074 uint64_t offset = (1ULL << (ztest_random(4) + SPA_MAXBLOCKSHIFT)) +
4075 (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
4076 uint64_t count = ztest_random(20) + 1;
4077 uint64_t blocksize = ztest_random_blocksize();
4080 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
4082 if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4085 if (ztest_truncate(zd, od[0].od_object, offset, count * blocksize) != 0)
4088 ztest_prealloc(zd, od[0].od_object, offset, count * blocksize);
4090 data = umem_zalloc(blocksize, UMEM_NOFAIL);
4092 while (ztest_random(count) != 0) {
4093 uint64_t randoff = offset + (ztest_random(count) * blocksize);
4094 if (ztest_write(zd, od[0].od_object, randoff, blocksize,
4097 while (ztest_random(4) != 0)
4098 ztest_io(zd, od[0].od_object, randoff);
4101 umem_free(data, blocksize);
4105 * Verify that zap_{create,destroy,add,remove,update} work as expected.
4107 #define ZTEST_ZAP_MIN_INTS 1
4108 #define ZTEST_ZAP_MAX_INTS 4
4109 #define ZTEST_ZAP_MAX_PROPS 1000
4112 ztest_zap(ztest_ds_t *zd, uint64_t id)
4114 objset_t *os = zd->zd_os;
4117 uint64_t txg, last_txg;
4118 uint64_t value[ZTEST_ZAP_MAX_INTS];
4119 uint64_t zl_ints, zl_intsize, prop;
4122 char propname[100], txgname[100];
4124 char *hc[2] = { "s.acl.h", ".s.open.h.hyLZlg" };
4126 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0);
4128 if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4131 object = od[0].od_object;
4134 * Generate a known hash collision, and verify that
4135 * we can lookup and remove both entries.
4137 tx = dmu_tx_create(os);
4138 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4139 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4142 for (i = 0; i < 2; i++) {
4144 VERIFY3U(0, ==, zap_add(os, object, hc[i], sizeof (uint64_t),
4147 for (i = 0; i < 2; i++) {
4148 VERIFY3U(EEXIST, ==, zap_add(os, object, hc[i],
4149 sizeof (uint64_t), 1, &value[i], tx));
4151 zap_length(os, object, hc[i], &zl_intsize, &zl_ints));
4152 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4153 ASSERT3U(zl_ints, ==, 1);
4155 for (i = 0; i < 2; i++) {
4156 VERIFY3U(0, ==, zap_remove(os, object, hc[i], tx));
4161 * Generate a buch of random entries.
4163 ints = MAX(ZTEST_ZAP_MIN_INTS, object % ZTEST_ZAP_MAX_INTS);
4165 prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
4166 (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
4167 (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
4168 bzero(value, sizeof (value));
4172 * If these zap entries already exist, validate their contents.
4174 error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
4176 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4177 ASSERT3U(zl_ints, ==, 1);
4179 VERIFY(zap_lookup(os, object, txgname, zl_intsize,
4180 zl_ints, &last_txg) == 0);
4182 VERIFY(zap_length(os, object, propname, &zl_intsize,
4185 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4186 ASSERT3U(zl_ints, ==, ints);
4188 VERIFY(zap_lookup(os, object, propname, zl_intsize,
4189 zl_ints, value) == 0);
4191 for (i = 0; i < ints; i++) {
4192 ASSERT3U(value[i], ==, last_txg + object + i);
4195 ASSERT3U(error, ==, ENOENT);
4199 * Atomically update two entries in our zap object.
4200 * The first is named txg_%llu, and contains the txg
4201 * in which the property was last updated. The second
4202 * is named prop_%llu, and the nth element of its value
4203 * should be txg + object + n.
4205 tx = dmu_tx_create(os);
4206 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4207 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4212 fatal(0, "zap future leak: old %llu new %llu", last_txg, txg);
4214 for (i = 0; i < ints; i++)
4215 value[i] = txg + object + i;
4217 VERIFY3U(0, ==, zap_update(os, object, txgname, sizeof (uint64_t),
4219 VERIFY3U(0, ==, zap_update(os, object, propname, sizeof (uint64_t),
4225 * Remove a random pair of entries.
4227 prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
4228 (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
4229 (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
4231 error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
4233 if (error == ENOENT)
4238 tx = dmu_tx_create(os);
4239 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4240 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4243 VERIFY3U(0, ==, zap_remove(os, object, txgname, tx));
4244 VERIFY3U(0, ==, zap_remove(os, object, propname, tx));
4249 * Testcase to test the upgrading of a microzap to fatzap.
4252 ztest_fzap(ztest_ds_t *zd, uint64_t id)
4254 objset_t *os = zd->zd_os;
4256 uint64_t object, txg;
4258 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0);
4260 if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4263 object = od[0].od_object;
4266 * Add entries to this ZAP and make sure it spills over
4267 * and gets upgraded to a fatzap. Also, since we are adding
4268 * 2050 entries we should see ptrtbl growth and leaf-block split.
4270 for (int i = 0; i < 2050; i++) {
4271 char name[ZFS_MAX_DATASET_NAME_LEN];
4276 (void) snprintf(name, sizeof (name), "fzap-%llu-%llu",
4279 tx = dmu_tx_create(os);
4280 dmu_tx_hold_zap(tx, object, B_TRUE, name);
4281 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4284 error = zap_add(os, object, name, sizeof (uint64_t), 1,
4286 ASSERT(error == 0 || error == EEXIST);
4293 ztest_zap_parallel(ztest_ds_t *zd, uint64_t id)
4295 objset_t *os = zd->zd_os;
4297 uint64_t txg, object, count, wsize, wc, zl_wsize, zl_wc;
4299 int i, namelen, error;
4300 int micro = ztest_random(2);
4301 char name[20], string_value[20];
4304 ztest_od_init(&od[0], ID_PARALLEL, FTAG, micro, DMU_OT_ZAP_OTHER, 0, 0);
4306 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4309 object = od[0].od_object;
4312 * Generate a random name of the form 'xxx.....' where each
4313 * x is a random printable character and the dots are dots.
4314 * There are 94 such characters, and the name length goes from
4315 * 6 to 20, so there are 94^3 * 15 = 12,458,760 possible names.
4317 namelen = ztest_random(sizeof (name) - 5) + 5 + 1;
4319 for (i = 0; i < 3; i++)
4320 name[i] = '!' + ztest_random('~' - '!' + 1);
4321 for (; i < namelen - 1; i++)
4325 if ((namelen & 1) || micro) {
4326 wsize = sizeof (txg);
4332 data = string_value;
4336 VERIFY0(zap_count(os, object, &count));
4337 ASSERT(count != -1ULL);
4340 * Select an operation: length, lookup, add, update, remove.
4342 i = ztest_random(5);
4345 tx = dmu_tx_create(os);
4346 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4347 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4350 bcopy(name, string_value, namelen);
4354 bzero(string_value, namelen);
4360 error = zap_length(os, object, name, &zl_wsize, &zl_wc);
4362 ASSERT3U(wsize, ==, zl_wsize);
4363 ASSERT3U(wc, ==, zl_wc);
4365 ASSERT3U(error, ==, ENOENT);
4370 error = zap_lookup(os, object, name, wsize, wc, data);
4372 if (data == string_value &&
4373 bcmp(name, data, namelen) != 0)
4374 fatal(0, "name '%s' != val '%s' len %d",
4375 name, data, namelen);
4377 ASSERT3U(error, ==, ENOENT);
4382 error = zap_add(os, object, name, wsize, wc, data, tx);
4383 ASSERT(error == 0 || error == EEXIST);
4387 VERIFY(zap_update(os, object, name, wsize, wc, data, tx) == 0);
4391 error = zap_remove(os, object, name, tx);
4392 ASSERT(error == 0 || error == ENOENT);
4401 * Commit callback data.
4403 typedef struct ztest_cb_data {
4404 list_node_t zcd_node;
4406 int zcd_expected_err;
4407 boolean_t zcd_added;
4408 boolean_t zcd_called;
4412 /* This is the actual commit callback function */
4414 ztest_commit_callback(void *arg, int error)
4416 ztest_cb_data_t *data = arg;
4417 uint64_t synced_txg;
4419 VERIFY(data != NULL);
4420 VERIFY3S(data->zcd_expected_err, ==, error);
4421 VERIFY(!data->zcd_called);
4423 synced_txg = spa_last_synced_txg(data->zcd_spa);
4424 if (data->zcd_txg > synced_txg)
4425 fatal(0, "commit callback of txg %" PRIu64 " called prematurely"
4426 ", last synced txg = %" PRIu64 "\n", data->zcd_txg,
4429 data->zcd_called = B_TRUE;
4431 if (error == ECANCELED) {
4432 ASSERT0(data->zcd_txg);
4433 ASSERT(!data->zcd_added);
4436 * The private callback data should be destroyed here, but
4437 * since we are going to check the zcd_called field after
4438 * dmu_tx_abort(), we will destroy it there.
4443 /* Was this callback added to the global callback list? */
4444 if (!data->zcd_added)
4447 ASSERT3U(data->zcd_txg, !=, 0);
4449 /* Remove our callback from the list */
4450 (void) mutex_lock(&zcl.zcl_callbacks_lock);
4451 list_remove(&zcl.zcl_callbacks, data);
4452 (void) mutex_unlock(&zcl.zcl_callbacks_lock);
4455 umem_free(data, sizeof (ztest_cb_data_t));
4458 /* Allocate and initialize callback data structure */
4459 static ztest_cb_data_t *
4460 ztest_create_cb_data(objset_t *os, uint64_t txg)
4462 ztest_cb_data_t *cb_data;
4464 cb_data = umem_zalloc(sizeof (ztest_cb_data_t), UMEM_NOFAIL);
4466 cb_data->zcd_txg = txg;
4467 cb_data->zcd_spa = dmu_objset_spa(os);
4473 * If a number of txgs equal to this threshold have been created after a commit
4474 * callback has been registered but not called, then we assume there is an
4475 * implementation bug.
4477 #define ZTEST_COMMIT_CALLBACK_THRESH (TXG_CONCURRENT_STATES + 2)
4480 * Commit callback test.
4483 ztest_dmu_commit_callbacks(ztest_ds_t *zd, uint64_t id)
4485 objset_t *os = zd->zd_os;
4488 ztest_cb_data_t *cb_data[3], *tmp_cb;
4489 uint64_t old_txg, txg;
4492 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
4494 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4497 tx = dmu_tx_create(os);
4499 cb_data[0] = ztest_create_cb_data(os, 0);
4500 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[0]);
4502 dmu_tx_hold_write(tx, od[0].od_object, 0, sizeof (uint64_t));
4504 /* Every once in a while, abort the transaction on purpose */
4505 if (ztest_random(100) == 0)
4509 error = dmu_tx_assign(tx, TXG_NOWAIT);
4511 txg = error ? 0 : dmu_tx_get_txg(tx);
4513 cb_data[0]->zcd_txg = txg;
4514 cb_data[1] = ztest_create_cb_data(os, txg);
4515 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[1]);
4519 * It's not a strict requirement to call the registered
4520 * callbacks from inside dmu_tx_abort(), but that's what
4521 * it's supposed to happen in the current implementation
4522 * so we will check for that.
4524 for (i = 0; i < 2; i++) {
4525 cb_data[i]->zcd_expected_err = ECANCELED;
4526 VERIFY(!cb_data[i]->zcd_called);
4531 for (i = 0; i < 2; i++) {
4532 VERIFY(cb_data[i]->zcd_called);
4533 umem_free(cb_data[i], sizeof (ztest_cb_data_t));
4539 cb_data[2] = ztest_create_cb_data(os, txg);
4540 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[2]);
4543 * Read existing data to make sure there isn't a future leak.
4545 VERIFY(0 == dmu_read(os, od[0].od_object, 0, sizeof (uint64_t),
4546 &old_txg, DMU_READ_PREFETCH));
4549 fatal(0, "future leak: got %" PRIu64 ", open txg is %" PRIu64,
4552 dmu_write(os, od[0].od_object, 0, sizeof (uint64_t), &txg, tx);
4554 (void) mutex_lock(&zcl.zcl_callbacks_lock);
4557 * Since commit callbacks don't have any ordering requirement and since
4558 * it is theoretically possible for a commit callback to be called
4559 * after an arbitrary amount of time has elapsed since its txg has been
4560 * synced, it is difficult to reliably determine whether a commit
4561 * callback hasn't been called due to high load or due to a flawed
4564 * In practice, we will assume that if after a certain number of txgs a
4565 * commit callback hasn't been called, then most likely there's an
4566 * implementation bug..
4568 tmp_cb = list_head(&zcl.zcl_callbacks);
4569 if (tmp_cb != NULL &&
4570 (txg - ZTEST_COMMIT_CALLBACK_THRESH) > tmp_cb->zcd_txg) {
4571 fatal(0, "Commit callback threshold exceeded, oldest txg: %"
4572 PRIu64 ", open txg: %" PRIu64 "\n", tmp_cb->zcd_txg, txg);
4576 * Let's find the place to insert our callbacks.
4578 * Even though the list is ordered by txg, it is possible for the
4579 * insertion point to not be the end because our txg may already be
4580 * quiescing at this point and other callbacks in the open txg
4581 * (from other objsets) may have sneaked in.
4583 tmp_cb = list_tail(&zcl.zcl_callbacks);
4584 while (tmp_cb != NULL && tmp_cb->zcd_txg > txg)
4585 tmp_cb = list_prev(&zcl.zcl_callbacks, tmp_cb);
4587 /* Add the 3 callbacks to the list */
4588 for (i = 0; i < 3; i++) {
4590 list_insert_head(&zcl.zcl_callbacks, cb_data[i]);
4592 list_insert_after(&zcl.zcl_callbacks, tmp_cb,
4595 cb_data[i]->zcd_added = B_TRUE;
4596 VERIFY(!cb_data[i]->zcd_called);
4598 tmp_cb = cb_data[i];
4601 (void) mutex_unlock(&zcl.zcl_callbacks_lock);
4608 ztest_dsl_prop_get_set(ztest_ds_t *zd, uint64_t id)
4610 zfs_prop_t proplist[] = {
4612 ZFS_PROP_COMPRESSION,
4617 (void) rw_rdlock(&ztest_name_lock);
4619 for (int p = 0; p < sizeof (proplist) / sizeof (proplist[0]); p++)
4620 (void) ztest_dsl_prop_set_uint64(zd->zd_name, proplist[p],
4621 ztest_random_dsl_prop(proplist[p]), (int)ztest_random(2));
4623 (void) rw_unlock(&ztest_name_lock);
4628 ztest_spa_prop_get_set(ztest_ds_t *zd, uint64_t id)
4630 nvlist_t *props = NULL;
4632 (void) rw_rdlock(&ztest_name_lock);
4634 (void) ztest_spa_prop_set_uint64(ZPOOL_PROP_DEDUPDITTO,
4635 ZIO_DEDUPDITTO_MIN + ztest_random(ZIO_DEDUPDITTO_MIN));
4637 VERIFY0(spa_prop_get(ztest_spa, &props));
4639 if (ztest_opts.zo_verbose >= 6)
4640 dump_nvlist(props, 4);
4644 (void) rw_unlock(&ztest_name_lock);
4648 user_release_one(const char *snapname, const char *holdname)
4650 nvlist_t *snaps, *holds;
4653 snaps = fnvlist_alloc();
4654 holds = fnvlist_alloc();
4655 fnvlist_add_boolean(holds, holdname);
4656 fnvlist_add_nvlist(snaps, snapname, holds);
4657 fnvlist_free(holds);
4658 error = dsl_dataset_user_release(snaps, NULL);
4659 fnvlist_free(snaps);
4664 * Test snapshot hold/release and deferred destroy.
4667 ztest_dmu_snapshot_hold(ztest_ds_t *zd, uint64_t id)
4670 objset_t *os = zd->zd_os;
4674 char clonename[100];
4676 char osname[ZFS_MAX_DATASET_NAME_LEN];
4679 (void) rw_rdlock(&ztest_name_lock);
4681 dmu_objset_name(os, osname);
4683 (void) snprintf(snapname, sizeof (snapname), "sh1_%llu", id);
4684 (void) snprintf(fullname, sizeof (fullname), "%s@%s", osname, snapname);
4685 (void) snprintf(clonename, sizeof (clonename),
4686 "%s/ch1_%llu", osname, id);
4687 (void) snprintf(tag, sizeof (tag), "tag_%llu", id);
4690 * Clean up from any previous run.
4692 error = dsl_destroy_head(clonename);
4693 if (error != ENOENT)
4695 error = user_release_one(fullname, tag);
4696 if (error != ESRCH && error != ENOENT)
4698 error = dsl_destroy_snapshot(fullname, B_FALSE);
4699 if (error != ENOENT)
4703 * Create snapshot, clone it, mark snap for deferred destroy,
4704 * destroy clone, verify snap was also destroyed.
4706 error = dmu_objset_snapshot_one(osname, snapname);
4708 if (error == ENOSPC) {
4709 ztest_record_enospc("dmu_objset_snapshot");
4712 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
4715 error = dmu_objset_clone(clonename, fullname);
4717 if (error == ENOSPC) {
4718 ztest_record_enospc("dmu_objset_clone");
4721 fatal(0, "dmu_objset_clone(%s) = %d", clonename, error);
4724 error = dsl_destroy_snapshot(fullname, B_TRUE);
4726 fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
4730 error = dsl_destroy_head(clonename);
4732 fatal(0, "dsl_destroy_head(%s) = %d", clonename, error);
4734 error = dmu_objset_hold(fullname, FTAG, &origin);
4735 if (error != ENOENT)
4736 fatal(0, "dmu_objset_hold(%s) = %d", fullname, error);
4739 * Create snapshot, add temporary hold, verify that we can't
4740 * destroy a held snapshot, mark for deferred destroy,
4741 * release hold, verify snapshot was destroyed.
4743 error = dmu_objset_snapshot_one(osname, snapname);
4745 if (error == ENOSPC) {
4746 ztest_record_enospc("dmu_objset_snapshot");
4749 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
4752 holds = fnvlist_alloc();
4753 fnvlist_add_string(holds, fullname, tag);
4754 error = dsl_dataset_user_hold(holds, 0, NULL);
4755 fnvlist_free(holds);
4757 if (error == ENOSPC) {
4758 ztest_record_enospc("dsl_dataset_user_hold");
4761 fatal(0, "dsl_dataset_user_hold(%s, %s) = %u",
4762 fullname, tag, error);
4765 error = dsl_destroy_snapshot(fullname, B_FALSE);
4766 if (error != EBUSY) {
4767 fatal(0, "dsl_destroy_snapshot(%s, B_FALSE) = %d",
4771 error = dsl_destroy_snapshot(fullname, B_TRUE);
4773 fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
4777 error = user_release_one(fullname, tag);
4779 fatal(0, "user_release_one(%s, %s) = %d", fullname, tag, error);
4781 VERIFY3U(dmu_objset_hold(fullname, FTAG, &origin), ==, ENOENT);
4784 (void) rw_unlock(&ztest_name_lock);
4788 * Inject random faults into the on-disk data.
4792 ztest_fault_inject(ztest_ds_t *zd, uint64_t id)
4794 ztest_shared_t *zs = ztest_shared;
4795 spa_t *spa = ztest_spa;
4799 uint64_t bad = 0x1990c0ffeedecadeULL;
4801 char path0[MAXPATHLEN];
4802 char pathrand[MAXPATHLEN];
4804 int bshift = SPA_MAXBLOCKSHIFT + 2;
4810 boolean_t islog = B_FALSE;
4812 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
4813 maxfaults = MAXFAULTS();
4814 leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz;
4815 mirror_save = zs->zs_mirrors;
4816 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4818 ASSERT(leaves >= 1);
4821 * Grab the name lock as reader. There are some operations
4822 * which don't like to have their vdevs changed while
4823 * they are in progress (i.e. spa_change_guid). Those
4824 * operations will have grabbed the name lock as writer.
4826 (void) rw_rdlock(&ztest_name_lock);
4829 * We need SCL_STATE here because we're going to look at vd0->vdev_tsd.
4831 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
4833 if (ztest_random(2) == 0) {
4835 * Inject errors on a normal data device or slog device.
4837 top = ztest_random_vdev_top(spa, B_TRUE);
4838 leaf = ztest_random(leaves) + zs->zs_splits;
4841 * Generate paths to the first leaf in this top-level vdev,
4842 * and to the random leaf we selected. We'll induce transient
4843 * write failures and random online/offline activity on leaf 0,
4844 * and we'll write random garbage to the randomly chosen leaf.
4846 (void) snprintf(path0, sizeof (path0), ztest_dev_template,
4847 ztest_opts.zo_dir, ztest_opts.zo_pool,
4848 top * leaves + zs->zs_splits);
4849 (void) snprintf(pathrand, sizeof (pathrand), ztest_dev_template,
4850 ztest_opts.zo_dir, ztest_opts.zo_pool,
4851 top * leaves + leaf);
4853 vd0 = vdev_lookup_by_path(spa->spa_root_vdev, path0);
4854 if (vd0 != NULL && vd0->vdev_top->vdev_islog)
4858 * If the top-level vdev needs to be resilvered
4859 * then we only allow faults on the device that is
4862 if (vd0 != NULL && maxfaults != 1 &&
4863 (!vdev_resilver_needed(vd0->vdev_top, NULL, NULL) ||
4864 vd0->vdev_resilver_txg != 0)) {
4866 * Make vd0 explicitly claim to be unreadable,
4867 * or unwriteable, or reach behind its back
4868 * and close the underlying fd. We can do this if
4869 * maxfaults == 0 because we'll fail and reexecute,
4870 * and we can do it if maxfaults >= 2 because we'll
4871 * have enough redundancy. If maxfaults == 1, the
4872 * combination of this with injection of random data
4873 * corruption below exceeds the pool's fault tolerance.
4875 vdev_file_t *vf = vd0->vdev_tsd;
4877 if (vf != NULL && ztest_random(3) == 0) {
4878 (void) close(vf->vf_vnode->v_fd);
4879 vf->vf_vnode->v_fd = -1;
4880 } else if (ztest_random(2) == 0) {
4881 vd0->vdev_cant_read = B_TRUE;
4883 vd0->vdev_cant_write = B_TRUE;
4885 guid0 = vd0->vdev_guid;
4889 * Inject errors on an l2cache device.
4891 spa_aux_vdev_t *sav = &spa->spa_l2cache;
4893 if (sav->sav_count == 0) {
4894 spa_config_exit(spa, SCL_STATE, FTAG);
4895 (void) rw_unlock(&ztest_name_lock);
4898 vd0 = sav->sav_vdevs[ztest_random(sav->sav_count)];
4899 guid0 = vd0->vdev_guid;
4900 (void) strcpy(path0, vd0->vdev_path);
4901 (void) strcpy(pathrand, vd0->vdev_path);
4905 maxfaults = INT_MAX; /* no limit on cache devices */
4908 spa_config_exit(spa, SCL_STATE, FTAG);
4909 (void) rw_unlock(&ztest_name_lock);
4912 * If we can tolerate two or more faults, or we're dealing
4913 * with a slog, randomly online/offline vd0.
4915 if ((maxfaults >= 2 || islog) && guid0 != 0) {
4916 if (ztest_random(10) < 6) {
4917 int flags = (ztest_random(2) == 0 ?
4918 ZFS_OFFLINE_TEMPORARY : 0);
4921 * We have to grab the zs_name_lock as writer to
4922 * prevent a race between offlining a slog and
4923 * destroying a dataset. Offlining the slog will
4924 * grab a reference on the dataset which may cause
4925 * dmu_objset_destroy() to fail with EBUSY thus
4926 * leaving the dataset in an inconsistent state.
4929 (void) rw_wrlock(&ztest_name_lock);
4931 VERIFY(vdev_offline(spa, guid0, flags) != EBUSY);
4934 (void) rw_unlock(&ztest_name_lock);
4937 * Ideally we would like to be able to randomly
4938 * call vdev_[on|off]line without holding locks
4939 * to force unpredictable failures but the side
4940 * effects of vdev_[on|off]line prevent us from
4941 * doing so. We grab the ztest_vdev_lock here to
4942 * prevent a race between injection testing and
4945 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
4946 (void) vdev_online(spa, guid0, 0, NULL);
4947 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4955 * We have at least single-fault tolerance, so inject data corruption.
4957 fd = open(pathrand, O_RDWR);
4959 if (fd == -1) /* we hit a gap in the device namespace */
4962 fsize = lseek(fd, 0, SEEK_END);
4964 while (--iters != 0) {
4966 * The offset must be chosen carefully to ensure that
4967 * we do not inject a given logical block with errors
4968 * on two different leaf devices, because ZFS can not
4969 * tolerate that (if maxfaults==1).
4971 * We divide each leaf into chunks of size
4972 * (# leaves * SPA_MAXBLOCKSIZE * 4). Within each chunk
4973 * there is a series of ranges to which we can inject errors.
4974 * Each range can accept errors on only a single leaf vdev.
4975 * The error injection ranges are separated by ranges
4976 * which we will not inject errors on any device (DMZs).
4977 * Each DMZ must be large enough such that a single block
4978 * can not straddle it, so that a single block can not be
4979 * a target in two different injection ranges (on different
4982 * For example, with 3 leaves, each chunk looks like:
4983 * 0 to 32M: injection range for leaf 0
4984 * 32M to 64M: DMZ - no injection allowed
4985 * 64M to 96M: injection range for leaf 1
4986 * 96M to 128M: DMZ - no injection allowed
4987 * 128M to 160M: injection range for leaf 2
4988 * 160M to 192M: DMZ - no injection allowed
4990 offset = ztest_random(fsize / (leaves << bshift)) *
4991 (leaves << bshift) + (leaf << bshift) +
4992 (ztest_random(1ULL << (bshift - 1)) & -8ULL);
4995 * Only allow damage to the labels at one end of the vdev.
4997 * If all labels are damaged, the device will be totally
4998 * inaccessible, which will result in loss of data,
4999 * because we also damage (parts of) the other side of
5002 * Additionally, we will always have both an even and an
5003 * odd label, so that we can handle crashes in the
5004 * middle of vdev_config_sync().
5006 if ((leaf & 1) == 0 && offset < VDEV_LABEL_START_SIZE)
5010 * The two end labels are stored at the "end" of the disk, but
5011 * the end of the disk (vdev_psize) is aligned to
5012 * sizeof (vdev_label_t).
5014 uint64_t psize = P2ALIGN(fsize, sizeof (vdev_label_t));
5015 if ((leaf & 1) == 1 &&
5016 offset + sizeof (bad) > psize - VDEV_LABEL_END_SIZE)
5019 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
5020 if (mirror_save != zs->zs_mirrors) {
5021 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
5026 if (pwrite(fd, &bad, sizeof (bad), offset) != sizeof (bad))
5027 fatal(1, "can't inject bad word at 0x%llx in %s",
5030 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
5032 if (ztest_opts.zo_verbose >= 7)
5033 (void) printf("injected bad word into %s,"
5034 " offset 0x%llx\n", pathrand, (u_longlong_t)offset);
5041 * Verify that DDT repair works as expected.
5044 ztest_ddt_repair(ztest_ds_t *zd, uint64_t id)
5046 ztest_shared_t *zs = ztest_shared;
5047 spa_t *spa = ztest_spa;
5048 objset_t *os = zd->zd_os;
5050 uint64_t object, blocksize, txg, pattern, psize;
5051 enum zio_checksum checksum = spa_dedup_checksum(spa);
5056 int copies = 2 * ZIO_DEDUPDITTO_MIN;
5058 blocksize = ztest_random_blocksize();
5059 blocksize = MIN(blocksize, 2048); /* because we write so many */
5061 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
5063 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
5067 * Take the name lock as writer to prevent anyone else from changing
5068 * the pool and dataset properies we need to maintain during this test.
5070 (void) rw_wrlock(&ztest_name_lock);
5072 if (ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_DEDUP, checksum,
5074 ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_COPIES, 1,
5076 (void) rw_unlock(&ztest_name_lock);
5080 dmu_objset_stats_t dds;
5081 dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
5082 dmu_objset_fast_stat(os, &dds);
5083 dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
5085 object = od[0].od_object;
5086 blocksize = od[0].od_blocksize;
5087 pattern = zs->zs_guid ^ dds.dds_guid;
5089 ASSERT(object != 0);
5091 tx = dmu_tx_create(os);
5092 dmu_tx_hold_write(tx, object, 0, copies * blocksize);
5093 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
5095 (void) rw_unlock(&ztest_name_lock);
5100 * Write all the copies of our block.
5102 for (int i = 0; i < copies; i++) {
5103 uint64_t offset = i * blocksize;
5104 int error = dmu_buf_hold(os, object, offset, FTAG, &db,
5105 DMU_READ_NO_PREFETCH);
5107 fatal(B_FALSE, "dmu_buf_hold(%p, %llu, %llu) = %u",
5108 os, (long long)object, (long long) offset, error);
5110 ASSERT(db->db_offset == offset);
5111 ASSERT(db->db_size == blocksize);
5112 ASSERT(ztest_pattern_match(db->db_data, db->db_size, pattern) ||
5113 ztest_pattern_match(db->db_data, db->db_size, 0ULL));
5114 dmu_buf_will_fill(db, tx);
5115 ztest_pattern_set(db->db_data, db->db_size, pattern);
5116 dmu_buf_rele(db, FTAG);
5120 txg_wait_synced(spa_get_dsl(spa), txg);
5123 * Find out what block we got.
5125 VERIFY0(dmu_buf_hold(os, object, 0, FTAG, &db,
5126 DMU_READ_NO_PREFETCH));
5127 blk = *((dmu_buf_impl_t *)db)->db_blkptr;
5128 dmu_buf_rele(db, FTAG);
5131 * Damage the block. Dedup-ditto will save us when we read it later.
5133 psize = BP_GET_PSIZE(&blk);
5134 buf = zio_buf_alloc(psize);
5135 ztest_pattern_set(buf, psize, ~pattern);
5137 (void) zio_wait(zio_rewrite(NULL, spa, 0, &blk,
5138 buf, psize, NULL, NULL, ZIO_PRIORITY_SYNC_WRITE,
5139 ZIO_FLAG_CANFAIL | ZIO_FLAG_INDUCE_DAMAGE, NULL));
5141 zio_buf_free(buf, psize);
5143 (void) rw_unlock(&ztest_name_lock);
5151 ztest_scrub(ztest_ds_t *zd, uint64_t id)
5153 spa_t *spa = ztest_spa;
5155 (void) spa_scan(spa, POOL_SCAN_SCRUB);
5156 (void) poll(NULL, 0, 100); /* wait a moment, then force a restart */
5157 (void) spa_scan(spa, POOL_SCAN_SCRUB);
5161 * Change the guid for the pool.
5165 ztest_reguid(ztest_ds_t *zd, uint64_t id)
5167 spa_t *spa = ztest_spa;
5168 uint64_t orig, load;
5171 orig = spa_guid(spa);
5172 load = spa_load_guid(spa);
5174 (void) rw_wrlock(&ztest_name_lock);
5175 error = spa_change_guid(spa);
5176 (void) rw_unlock(&ztest_name_lock);
5181 if (ztest_opts.zo_verbose >= 4) {
5182 (void) printf("Changed guid old %llu -> %llu\n",
5183 (u_longlong_t)orig, (u_longlong_t)spa_guid(spa));
5186 VERIFY3U(orig, !=, spa_guid(spa));
5187 VERIFY3U(load, ==, spa_load_guid(spa));
5191 * Rename the pool to a different name and then rename it back.
5195 ztest_spa_rename(ztest_ds_t *zd, uint64_t id)
5197 char *oldname, *newname;
5200 (void) rw_wrlock(&ztest_name_lock);
5202 oldname = ztest_opts.zo_pool;
5203 newname = umem_alloc(strlen(oldname) + 5, UMEM_NOFAIL);
5204 (void) strcpy(newname, oldname);
5205 (void) strcat(newname, "_tmp");
5210 VERIFY3U(0, ==, spa_rename(oldname, newname));
5213 * Try to open it under the old name, which shouldn't exist
5215 VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
5218 * Open it under the new name and make sure it's still the same spa_t.
5220 VERIFY3U(0, ==, spa_open(newname, &spa, FTAG));
5222 ASSERT(spa == ztest_spa);
5223 spa_close(spa, FTAG);
5226 * Rename it back to the original
5228 VERIFY3U(0, ==, spa_rename(newname, oldname));
5231 * Make sure it can still be opened
5233 VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG));
5235 ASSERT(spa == ztest_spa);
5236 spa_close(spa, FTAG);
5238 umem_free(newname, strlen(newname) + 1);
5240 (void) rw_unlock(&ztest_name_lock);
5244 * Verify pool integrity by running zdb.
5247 ztest_run_zdb(char *pool)
5250 char zdb[MAXPATHLEN + MAXNAMELEN + 20];
5258 strlcpy(zdb, "/usr/bin/ztest", sizeof(zdb));
5260 /* zdb lives in /usr/sbin, while ztest lives in /usr/bin */
5261 bin = strstr(zdb, "/usr/bin/");
5262 ztest = strstr(bin, "/ztest");
5264 isalen = ztest - isa;
5268 "/usr/sbin%.*s/zdb -bcc%s%s -d -U %s %s",
5271 ztest_opts.zo_verbose >= 3 ? "s" : "",
5272 ztest_opts.zo_verbose >= 4 ? "v" : "",
5277 if (ztest_opts.zo_verbose >= 5)
5278 (void) printf("Executing %s\n", strstr(zdb, "zdb "));
5280 fp = popen(zdb, "r");
5283 while (fgets(zbuf, sizeof (zbuf), fp) != NULL)
5284 if (ztest_opts.zo_verbose >= 3)
5285 (void) printf("%s", zbuf);
5287 status = pclose(fp);
5292 ztest_dump_core = 0;
5293 if (WIFEXITED(status))
5294 fatal(0, "'%s' exit code %d", zdb, WEXITSTATUS(status));
5296 fatal(0, "'%s' died with signal %d", zdb, WTERMSIG(status));
5300 ztest_walk_pool_directory(char *header)
5304 if (ztest_opts.zo_verbose >= 6)
5305 (void) printf("%s\n", header);
5307 mutex_enter(&spa_namespace_lock);
5308 while ((spa = spa_next(spa)) != NULL)
5309 if (ztest_opts.zo_verbose >= 6)
5310 (void) printf("\t%s\n", spa_name(spa));
5311 mutex_exit(&spa_namespace_lock);
5315 ztest_spa_import_export(char *oldname, char *newname)
5317 nvlist_t *config, *newconfig;
5322 if (ztest_opts.zo_verbose >= 4) {
5323 (void) printf("import/export: old = %s, new = %s\n",
5328 * Clean up from previous runs.
5330 (void) spa_destroy(newname);
5333 * Get the pool's configuration and guid.
5335 VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG));
5338 * Kick off a scrub to tickle scrub/export races.
5340 if (ztest_random(2) == 0)
5341 (void) spa_scan(spa, POOL_SCAN_SCRUB);
5343 pool_guid = spa_guid(spa);
5344 spa_close(spa, FTAG);
5346 ztest_walk_pool_directory("pools before export");
5351 VERIFY3U(0, ==, spa_export(oldname, &config, B_FALSE, B_FALSE));
5353 ztest_walk_pool_directory("pools after export");
5358 newconfig = spa_tryimport(config);
5359 ASSERT(newconfig != NULL);
5360 nvlist_free(newconfig);
5363 * Import it under the new name.
5365 error = spa_import(newname, config, NULL, 0);
5367 dump_nvlist(config, 0);
5368 fatal(B_FALSE, "couldn't import pool %s as %s: error %u",
5369 oldname, newname, error);
5372 ztest_walk_pool_directory("pools after import");
5375 * Try to import it again -- should fail with EEXIST.
5377 VERIFY3U(EEXIST, ==, spa_import(newname, config, NULL, 0));
5380 * Try to import it under a different name -- should fail with EEXIST.
5382 VERIFY3U(EEXIST, ==, spa_import(oldname, config, NULL, 0));
5385 * Verify that the pool is no longer visible under the old name.
5387 VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
5390 * Verify that we can open and close the pool using the new name.
5392 VERIFY3U(0, ==, spa_open(newname, &spa, FTAG));
5393 ASSERT(pool_guid == spa_guid(spa));
5394 spa_close(spa, FTAG);
5396 nvlist_free(config);
5400 ztest_resume(spa_t *spa)
5402 if (spa_suspended(spa) && ztest_opts.zo_verbose >= 6)
5403 (void) printf("resuming from suspended state\n");
5404 spa_vdev_state_enter(spa, SCL_NONE);
5405 vdev_clear(spa, NULL);
5406 (void) spa_vdev_state_exit(spa, NULL, 0);
5407 (void) zio_resume(spa);
5411 ztest_resume_thread(void *arg)
5415 while (!ztest_exiting) {
5416 if (spa_suspended(spa))
5418 (void) poll(NULL, 0, 100);
5421 * Periodically change the zfs_compressed_arc_enabled setting.
5423 if (ztest_random(10) == 0)
5424 zfs_compressed_arc_enabled = ztest_random(2);
5430 ztest_deadman_thread(void *arg)
5432 ztest_shared_t *zs = arg;
5433 spa_t *spa = ztest_spa;
5434 hrtime_t delta, total = 0;
5437 delta = zs->zs_thread_stop - zs->zs_thread_start +
5438 MSEC2NSEC(zfs_deadman_synctime_ms);
5440 (void) poll(NULL, 0, (int)NSEC2MSEC(delta));
5443 * If the pool is suspended then fail immediately. Otherwise,
5444 * check to see if the pool is making any progress. If
5445 * vdev_deadman() discovers that there hasn't been any recent
5446 * I/Os then it will end up aborting the tests.
5448 if (spa_suspended(spa) || spa->spa_root_vdev == NULL) {
5449 fatal(0, "aborting test after %llu seconds because "
5450 "pool has transitioned to a suspended state.",
5451 zfs_deadman_synctime_ms / 1000);
5454 vdev_deadman(spa->spa_root_vdev);
5456 total += zfs_deadman_synctime_ms/1000;
5457 (void) printf("ztest has been running for %lld seconds\n",
5463 ztest_execute(int test, ztest_info_t *zi, uint64_t id)
5465 ztest_ds_t *zd = &ztest_ds[id % ztest_opts.zo_datasets];
5466 ztest_shared_callstate_t *zc = ZTEST_GET_SHARED_CALLSTATE(test);
5467 hrtime_t functime = gethrtime();
5469 for (int i = 0; i < zi->zi_iters; i++)
5470 zi->zi_func(zd, id);
5472 functime = gethrtime() - functime;
5474 atomic_add_64(&zc->zc_count, 1);
5475 atomic_add_64(&zc->zc_time, functime);
5477 if (ztest_opts.zo_verbose >= 4) {
5479 (void) dladdr((void *)zi->zi_func, &dli);
5480 (void) printf("%6.2f sec in %s\n",
5481 (double)functime / NANOSEC, dli.dli_sname);
5486 ztest_thread(void *arg)
5489 uint64_t id = (uintptr_t)arg;
5490 ztest_shared_t *zs = ztest_shared;
5494 ztest_shared_callstate_t *zc;
5496 while ((now = gethrtime()) < zs->zs_thread_stop) {
5498 * See if it's time to force a crash.
5500 if (now > zs->zs_thread_kill)
5504 * If we're getting ENOSPC with some regularity, stop.
5506 if (zs->zs_enospc_count > 10)
5510 * Pick a random function to execute.
5512 rand = ztest_random(ZTEST_FUNCS);
5513 zi = &ztest_info[rand];
5514 zc = ZTEST_GET_SHARED_CALLSTATE(rand);
5515 call_next = zc->zc_next;
5517 if (now >= call_next &&
5518 atomic_cas_64(&zc->zc_next, call_next, call_next +
5519 ztest_random(2 * zi->zi_interval[0] + 1)) == call_next) {
5520 ztest_execute(rand, zi, id);
5528 ztest_dataset_name(char *dsname, char *pool, int d)
5530 (void) snprintf(dsname, ZFS_MAX_DATASET_NAME_LEN, "%s/ds_%d", pool, d);
5534 ztest_dataset_destroy(int d)
5536 char name[ZFS_MAX_DATASET_NAME_LEN];
5538 ztest_dataset_name(name, ztest_opts.zo_pool, d);
5540 if (ztest_opts.zo_verbose >= 3)
5541 (void) printf("Destroying %s to free up space\n", name);
5544 * Cleanup any non-standard clones and snapshots. In general,
5545 * ztest thread t operates on dataset (t % zopt_datasets),
5546 * so there may be more than one thing to clean up.
5548 for (int t = d; t < ztest_opts.zo_threads;
5549 t += ztest_opts.zo_datasets) {
5550 ztest_dsl_dataset_cleanup(name, t);
5553 (void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
5554 DS_FIND_SNAPSHOTS | DS_FIND_CHILDREN);
5558 ztest_dataset_dirobj_verify(ztest_ds_t *zd)
5560 uint64_t usedobjs, dirobjs, scratch;
5563 * ZTEST_DIROBJ is the object directory for the entire dataset.
5564 * Therefore, the number of objects in use should equal the
5565 * number of ZTEST_DIROBJ entries, +1 for ZTEST_DIROBJ itself.
5566 * If not, we have an object leak.
5568 * Note that we can only check this in ztest_dataset_open(),
5569 * when the open-context and syncing-context values agree.
5570 * That's because zap_count() returns the open-context value,
5571 * while dmu_objset_space() returns the rootbp fill count.
5573 VERIFY3U(0, ==, zap_count(zd->zd_os, ZTEST_DIROBJ, &dirobjs));
5574 dmu_objset_space(zd->zd_os, &scratch, &scratch, &usedobjs, &scratch);
5575 ASSERT3U(dirobjs + 1, ==, usedobjs);
5579 ztest_dataset_open(int d)
5581 ztest_ds_t *zd = &ztest_ds[d];
5582 uint64_t committed_seq = ZTEST_GET_SHARED_DS(d)->zd_seq;
5585 char name[ZFS_MAX_DATASET_NAME_LEN];
5588 ztest_dataset_name(name, ztest_opts.zo_pool, d);
5590 (void) rw_rdlock(&ztest_name_lock);
5592 error = ztest_dataset_create(name);
5593 if (error == ENOSPC) {
5594 (void) rw_unlock(&ztest_name_lock);
5595 ztest_record_enospc(FTAG);
5598 ASSERT(error == 0 || error == EEXIST);
5600 VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, zd, &os));
5601 (void) rw_unlock(&ztest_name_lock);
5603 ztest_zd_init(zd, ZTEST_GET_SHARED_DS(d), os);
5605 zilog = zd->zd_zilog;
5607 if (zilog->zl_header->zh_claim_lr_seq != 0 &&
5608 zilog->zl_header->zh_claim_lr_seq < committed_seq)
5609 fatal(0, "missing log records: claimed %llu < committed %llu",
5610 zilog->zl_header->zh_claim_lr_seq, committed_seq);
5612 ztest_dataset_dirobj_verify(zd);
5614 zil_replay(os, zd, ztest_replay_vector);
5616 ztest_dataset_dirobj_verify(zd);
5618 if (ztest_opts.zo_verbose >= 6)
5619 (void) printf("%s replay %llu blocks, %llu records, seq %llu\n",
5621 (u_longlong_t)zilog->zl_parse_blk_count,
5622 (u_longlong_t)zilog->zl_parse_lr_count,
5623 (u_longlong_t)zilog->zl_replaying_seq);
5625 zilog = zil_open(os, ztest_get_data);
5627 if (zilog->zl_replaying_seq != 0 &&
5628 zilog->zl_replaying_seq < committed_seq)
5629 fatal(0, "missing log records: replayed %llu < committed %llu",
5630 zilog->zl_replaying_seq, committed_seq);
5636 ztest_dataset_close(int d)
5638 ztest_ds_t *zd = &ztest_ds[d];
5640 zil_close(zd->zd_zilog);
5641 dmu_objset_disown(zd->zd_os, zd);
5647 * Kick off threads to run tests on all datasets in parallel.
5650 ztest_run(ztest_shared_t *zs)
5655 thread_t resume_tid;
5658 ztest_exiting = B_FALSE;
5661 * Initialize parent/child shared state.
5663 VERIFY(_mutex_init(&ztest_vdev_lock, USYNC_THREAD, NULL) == 0);
5664 VERIFY(rwlock_init(&ztest_name_lock, USYNC_THREAD, NULL) == 0);
5666 zs->zs_thread_start = gethrtime();
5667 zs->zs_thread_stop =
5668 zs->zs_thread_start + ztest_opts.zo_passtime * NANOSEC;
5669 zs->zs_thread_stop = MIN(zs->zs_thread_stop, zs->zs_proc_stop);
5670 zs->zs_thread_kill = zs->zs_thread_stop;
5671 if (ztest_random(100) < ztest_opts.zo_killrate) {
5672 zs->zs_thread_kill -=
5673 ztest_random(ztest_opts.zo_passtime * NANOSEC);
5676 (void) _mutex_init(&zcl.zcl_callbacks_lock, USYNC_THREAD, NULL);
5678 list_create(&zcl.zcl_callbacks, sizeof (ztest_cb_data_t),
5679 offsetof(ztest_cb_data_t, zcd_node));
5684 kernel_init(FREAD | FWRITE);
5685 VERIFY0(spa_open(ztest_opts.zo_pool, &spa, FTAG));
5686 spa->spa_debug = B_TRUE;
5687 metaslab_preload_limit = ztest_random(20) + 1;
5690 dmu_objset_stats_t dds;
5691 VERIFY0(dmu_objset_own(ztest_opts.zo_pool,
5692 DMU_OST_ANY, B_TRUE, FTAG, &os));
5693 dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
5694 dmu_objset_fast_stat(os, &dds);
5695 dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
5696 zs->zs_guid = dds.dds_guid;
5697 dmu_objset_disown(os, FTAG);
5699 spa->spa_dedup_ditto = 2 * ZIO_DEDUPDITTO_MIN;
5702 * We don't expect the pool to suspend unless maxfaults == 0,
5703 * in which case ztest_fault_inject() temporarily takes away
5704 * the only valid replica.
5706 if (MAXFAULTS() == 0)
5707 spa->spa_failmode = ZIO_FAILURE_MODE_WAIT;
5709 spa->spa_failmode = ZIO_FAILURE_MODE_PANIC;
5712 * Create a thread to periodically resume suspended I/O.
5714 VERIFY(thr_create(0, 0, ztest_resume_thread, spa, THR_BOUND,
5718 * Create a deadman thread to abort() if we hang.
5720 VERIFY(thr_create(0, 0, ztest_deadman_thread, zs, THR_BOUND,
5724 * Verify that we can safely inquire about about any object,
5725 * whether it's allocated or not. To make it interesting,
5726 * we probe a 5-wide window around each power of two.
5727 * This hits all edge cases, including zero and the max.
5729 for (int t = 0; t < 64; t++) {
5730 for (int d = -5; d <= 5; d++) {
5731 error = dmu_object_info(spa->spa_meta_objset,
5732 (1ULL << t) + d, NULL);
5733 ASSERT(error == 0 || error == ENOENT ||
5739 * If we got any ENOSPC errors on the previous run, destroy something.
5741 if (zs->zs_enospc_count != 0) {
5742 int d = ztest_random(ztest_opts.zo_datasets);
5743 ztest_dataset_destroy(d);
5745 zs->zs_enospc_count = 0;
5747 tid = umem_zalloc(ztest_opts.zo_threads * sizeof (thread_t),
5750 if (ztest_opts.zo_verbose >= 4)
5751 (void) printf("starting main threads...\n");
5754 * Kick off all the tests that run in parallel.
5756 for (int t = 0; t < ztest_opts.zo_threads; t++) {
5757 if (t < ztest_opts.zo_datasets &&
5758 ztest_dataset_open(t) != 0)
5760 VERIFY(thr_create(0, 0, ztest_thread, (void *)(uintptr_t)t,
5761 THR_BOUND, &tid[t]) == 0);
5765 * Wait for all of the tests to complete. We go in reverse order
5766 * so we don't close datasets while threads are still using them.
5768 for (int t = ztest_opts.zo_threads - 1; t >= 0; t--) {
5769 VERIFY(thr_join(tid[t], NULL, NULL) == 0);
5770 if (t < ztest_opts.zo_datasets)
5771 ztest_dataset_close(t);
5774 txg_wait_synced(spa_get_dsl(spa), 0);
5776 zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(spa));
5777 zs->zs_space = metaslab_class_get_space(spa_normal_class(spa));
5778 zfs_dbgmsg_print(FTAG);
5780 umem_free(tid, ztest_opts.zo_threads * sizeof (thread_t));
5782 /* Kill the resume thread */
5783 ztest_exiting = B_TRUE;
5784 VERIFY(thr_join(resume_tid, NULL, NULL) == 0);
5788 * Right before closing the pool, kick off a bunch of async I/O;
5789 * spa_close() should wait for it to complete.
5791 for (uint64_t object = 1; object < 50; object++) {
5792 dmu_prefetch(spa->spa_meta_objset, object, 0, 0, 1ULL << 20,
5793 ZIO_PRIORITY_SYNC_READ);
5796 spa_close(spa, FTAG);
5799 * Verify that we can loop over all pools.
5801 mutex_enter(&spa_namespace_lock);
5802 for (spa = spa_next(NULL); spa != NULL; spa = spa_next(spa))
5803 if (ztest_opts.zo_verbose > 3)
5804 (void) printf("spa_next: found %s\n", spa_name(spa));
5805 mutex_exit(&spa_namespace_lock);
5808 * Verify that we can export the pool and reimport it under a
5811 if (ztest_random(2) == 0) {
5812 char name[ZFS_MAX_DATASET_NAME_LEN];
5813 (void) snprintf(name, sizeof (name), "%s_import",
5814 ztest_opts.zo_pool);
5815 ztest_spa_import_export(ztest_opts.zo_pool, name);
5816 ztest_spa_import_export(name, ztest_opts.zo_pool);
5821 list_destroy(&zcl.zcl_callbacks);
5823 (void) _mutex_destroy(&zcl.zcl_callbacks_lock);
5825 (void) rwlock_destroy(&ztest_name_lock);
5826 (void) _mutex_destroy(&ztest_vdev_lock);
5832 ztest_ds_t *zd = &ztest_ds[0];
5836 if (ztest_opts.zo_verbose >= 3)
5837 (void) printf("testing spa_freeze()...\n");
5839 kernel_init(FREAD | FWRITE);
5840 VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
5841 VERIFY3U(0, ==, ztest_dataset_open(0));
5842 spa->spa_debug = B_TRUE;
5846 * Force the first log block to be transactionally allocated.
5847 * We have to do this before we freeze the pool -- otherwise
5848 * the log chain won't be anchored.
5850 while (BP_IS_HOLE(&zd->zd_zilog->zl_header->zh_log)) {
5851 ztest_dmu_object_alloc_free(zd, 0);
5852 zil_commit(zd->zd_zilog, 0);
5855 txg_wait_synced(spa_get_dsl(spa), 0);
5858 * Freeze the pool. This stops spa_sync() from doing anything,
5859 * so that the only way to record changes from now on is the ZIL.
5864 * Because it is hard to predict how much space a write will actually
5865 * require beforehand, we leave ourselves some fudge space to write over
5868 uint64_t capacity = metaslab_class_get_space(spa_normal_class(spa)) / 2;
5871 * Run tests that generate log records but don't alter the pool config
5872 * or depend on DSL sync tasks (snapshots, objset create/destroy, etc).
5873 * We do a txg_wait_synced() after each iteration to force the txg
5874 * to increase well beyond the last synced value in the uberblock.
5875 * The ZIL should be OK with that.
5877 * Run a random number of times less than zo_maxloops and ensure we do
5878 * not run out of space on the pool.
5880 while (ztest_random(10) != 0 &&
5881 numloops++ < ztest_opts.zo_maxloops &&
5882 metaslab_class_get_alloc(spa_normal_class(spa)) < capacity) {
5884 ztest_od_init(&od, 0, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
5885 VERIFY0(ztest_object_init(zd, &od, sizeof (od), B_FALSE));
5886 ztest_io(zd, od.od_object,
5887 ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
5888 txg_wait_synced(spa_get_dsl(spa), 0);
5892 * Commit all of the changes we just generated.
5894 zil_commit(zd->zd_zilog, 0);
5895 txg_wait_synced(spa_get_dsl(spa), 0);
5898 * Close our dataset and close the pool.
5900 ztest_dataset_close(0);
5901 spa_close(spa, FTAG);
5905 * Open and close the pool and dataset to induce log replay.
5907 kernel_init(FREAD | FWRITE);
5908 VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
5909 ASSERT(spa_freeze_txg(spa) == UINT64_MAX);
5910 VERIFY3U(0, ==, ztest_dataset_open(0));
5911 ztest_dataset_close(0);
5913 spa->spa_debug = B_TRUE;
5915 txg_wait_synced(spa_get_dsl(spa), 0);
5916 ztest_reguid(NULL, 0);
5918 spa_close(spa, FTAG);
5923 print_time(hrtime_t t, char *timebuf)
5925 hrtime_t s = t / NANOSEC;
5926 hrtime_t m = s / 60;
5927 hrtime_t h = m / 60;
5928 hrtime_t d = h / 24;
5937 (void) sprintf(timebuf,
5938 "%llud%02lluh%02llum%02llus", d, h, m, s);
5940 (void) sprintf(timebuf, "%lluh%02llum%02llus", h, m, s);
5942 (void) sprintf(timebuf, "%llum%02llus", m, s);
5944 (void) sprintf(timebuf, "%llus", s);
5952 VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
5953 if (ztest_random(2) == 0)
5955 VERIFY(nvlist_add_uint64(props, "autoreplace", 1) == 0);
5961 * Create a storage pool with the given name and initial vdev size.
5962 * Then test spa_freeze() functionality.
5965 ztest_init(ztest_shared_t *zs)
5968 nvlist_t *nvroot, *props;
5970 VERIFY(_mutex_init(&ztest_vdev_lock, USYNC_THREAD, NULL) == 0);
5971 VERIFY(rwlock_init(&ztest_name_lock, USYNC_THREAD, NULL) == 0);
5973 kernel_init(FREAD | FWRITE);
5976 * Create the storage pool.
5978 (void) spa_destroy(ztest_opts.zo_pool);
5979 ztest_shared->zs_vdev_next_leaf = 0;
5981 zs->zs_mirrors = ztest_opts.zo_mirrors;
5982 nvroot = make_vdev_root(NULL, NULL, NULL, ztest_opts.zo_vdev_size, 0,
5983 0, ztest_opts.zo_raidz, zs->zs_mirrors, 1);
5984 props = make_random_props();
5985 for (int i = 0; i < SPA_FEATURES; i++) {
5987 (void) snprintf(buf, sizeof (buf), "feature@%s",
5988 spa_feature_table[i].fi_uname);
5989 VERIFY3U(0, ==, nvlist_add_uint64(props, buf, 0));
5991 VERIFY3U(0, ==, spa_create(ztest_opts.zo_pool, nvroot, props, NULL));
5992 nvlist_free(nvroot);
5995 VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
5996 zs->zs_metaslab_sz =
5997 1ULL << spa->spa_root_vdev->vdev_child[0]->vdev_ms_shift;
5999 spa_close(spa, FTAG);
6003 ztest_run_zdb(ztest_opts.zo_pool);
6007 ztest_run_zdb(ztest_opts.zo_pool);
6009 (void) rwlock_destroy(&ztest_name_lock);
6010 (void) _mutex_destroy(&ztest_vdev_lock);
6016 static char ztest_name_data[] = "/tmp/ztest.data.XXXXXX";
6018 ztest_fd_data = mkstemp(ztest_name_data);
6019 ASSERT3S(ztest_fd_data, >=, 0);
6020 (void) unlink(ztest_name_data);
6025 shared_data_size(ztest_shared_hdr_t *hdr)
6029 size = hdr->zh_hdr_size;
6030 size += hdr->zh_opts_size;
6031 size += hdr->zh_size;
6032 size += hdr->zh_stats_size * hdr->zh_stats_count;
6033 size += hdr->zh_ds_size * hdr->zh_ds_count;
6042 ztest_shared_hdr_t *hdr;
6044 hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
6045 PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
6046 ASSERT(hdr != MAP_FAILED);
6048 VERIFY3U(0, ==, ftruncate(ztest_fd_data, sizeof (ztest_shared_hdr_t)));
6050 hdr->zh_hdr_size = sizeof (ztest_shared_hdr_t);
6051 hdr->zh_opts_size = sizeof (ztest_shared_opts_t);
6052 hdr->zh_size = sizeof (ztest_shared_t);
6053 hdr->zh_stats_size = sizeof (ztest_shared_callstate_t);
6054 hdr->zh_stats_count = ZTEST_FUNCS;
6055 hdr->zh_ds_size = sizeof (ztest_shared_ds_t);
6056 hdr->zh_ds_count = ztest_opts.zo_datasets;
6058 size = shared_data_size(hdr);
6059 VERIFY3U(0, ==, ftruncate(ztest_fd_data, size));
6061 (void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
6068 ztest_shared_hdr_t *hdr;
6071 hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
6072 PROT_READ, MAP_SHARED, ztest_fd_data, 0);
6073 ASSERT(hdr != MAP_FAILED);
6075 size = shared_data_size(hdr);
6077 (void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
6078 hdr = ztest_shared_hdr = (void *)mmap(0, P2ROUNDUP(size, getpagesize()),
6079 PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
6080 ASSERT(hdr != MAP_FAILED);
6081 buf = (uint8_t *)hdr;
6083 offset = hdr->zh_hdr_size;
6084 ztest_shared_opts = (void *)&buf[offset];
6085 offset += hdr->zh_opts_size;
6086 ztest_shared = (void *)&buf[offset];
6087 offset += hdr->zh_size;
6088 ztest_shared_callstate = (void *)&buf[offset];
6089 offset += hdr->zh_stats_size * hdr->zh_stats_count;
6090 ztest_shared_ds = (void *)&buf[offset];
6094 exec_child(char *cmd, char *libpath, boolean_t ignorekill, int *statusp)
6098 char *cmdbuf = NULL;
6103 cmdbuf = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
6104 (void) strlcpy(cmdbuf, getexecname(), MAXPATHLEN);
6109 fatal(1, "fork failed");
6111 if (pid == 0) { /* child */
6112 char *emptyargv[2] = { cmd, NULL };
6113 char fd_data_str[12];
6115 struct rlimit rl = { 1024, 1024 };
6116 (void) setrlimit(RLIMIT_NOFILE, &rl);
6118 (void) close(ztest_fd_rand);
6120 snprintf(fd_data_str, 12, "%d", ztest_fd_data));
6121 VERIFY0(setenv("ZTEST_FD_DATA", fd_data_str, 1));
6123 (void) enable_extended_FILE_stdio(-1, -1);
6124 if (libpath != NULL)
6125 VERIFY(0 == setenv("LD_LIBRARY_PATH", libpath, 1));
6127 (void) execv(cmd, emptyargv);
6129 (void) execvp(cmd, emptyargv);
6131 ztest_dump_core = B_FALSE;
6132 fatal(B_TRUE, "exec failed: %s", cmd);
6135 if (cmdbuf != NULL) {
6136 umem_free(cmdbuf, MAXPATHLEN);
6140 while (waitpid(pid, &status, 0) != pid)
6142 if (statusp != NULL)
6145 if (WIFEXITED(status)) {
6146 if (WEXITSTATUS(status) != 0) {
6147 (void) fprintf(stderr, "child exited with code %d\n",
6148 WEXITSTATUS(status));
6152 } else if (WIFSIGNALED(status)) {
6153 if (!ignorekill || WTERMSIG(status) != SIGKILL) {
6154 (void) fprintf(stderr, "child died with signal %d\n",
6160 (void) fprintf(stderr, "something strange happened to child\n");
6167 ztest_run_init(void)
6169 ztest_shared_t *zs = ztest_shared;
6171 ASSERT(ztest_opts.zo_init != 0);
6174 * Blow away any existing copy of zpool.cache
6176 (void) remove(spa_config_path);
6179 * Create and initialize our storage pool.
6181 for (int i = 1; i <= ztest_opts.zo_init; i++) {
6182 bzero(zs, sizeof (ztest_shared_t));
6183 if (ztest_opts.zo_verbose >= 3 &&
6184 ztest_opts.zo_init != 1) {
6185 (void) printf("ztest_init(), pass %d\n", i);
6192 main(int argc, char **argv)
6200 ztest_shared_callstate_t *zc;
6202 char numbuf[NN_NUMBUF_SZ];
6206 char *fd_data_str = getenv("ZTEST_FD_DATA");
6208 (void) setvbuf(stdout, NULL, _IOLBF, 0);
6210 dprintf_setup(&argc, argv);
6211 zfs_deadman_synctime_ms = 300000;
6213 ztest_fd_rand = open("/dev/urandom", O_RDONLY);
6214 ASSERT3S(ztest_fd_rand, >=, 0);
6217 process_options(argc, argv);
6222 bcopy(&ztest_opts, ztest_shared_opts,
6223 sizeof (*ztest_shared_opts));
6225 ztest_fd_data = atoi(fd_data_str);
6227 bcopy(ztest_shared_opts, &ztest_opts, sizeof (ztest_opts));
6229 ASSERT3U(ztest_opts.zo_datasets, ==, ztest_shared_hdr->zh_ds_count);
6231 /* Override location of zpool.cache */
6232 VERIFY3U(asprintf((char **)&spa_config_path, "%s/zpool.cache",
6233 ztest_opts.zo_dir), !=, -1);
6235 ztest_ds = umem_alloc(ztest_opts.zo_datasets * sizeof (ztest_ds_t),
6240 metaslab_gang_bang = ztest_opts.zo_metaslab_gang_bang;
6241 metaslab_df_alloc_threshold =
6242 zs->zs_metaslab_df_alloc_threshold;
6251 hasalt = (strlen(ztest_opts.zo_alt_ztest) != 0);
6253 if (ztest_opts.zo_verbose >= 1) {
6254 (void) printf("%llu vdevs, %d datasets, %d threads,"
6255 " %llu seconds...\n",
6256 (u_longlong_t)ztest_opts.zo_vdevs,
6257 ztest_opts.zo_datasets,
6258 ztest_opts.zo_threads,
6259 (u_longlong_t)ztest_opts.zo_time);
6262 cmd = umem_alloc(MAXNAMELEN, UMEM_NOFAIL);
6263 (void) strlcpy(cmd, getexecname(), MAXNAMELEN);
6265 zs->zs_do_init = B_TRUE;
6266 if (strlen(ztest_opts.zo_alt_ztest) != 0) {
6267 if (ztest_opts.zo_verbose >= 1) {
6268 (void) printf("Executing older ztest for "
6269 "initialization: %s\n", ztest_opts.zo_alt_ztest);
6271 VERIFY(!exec_child(ztest_opts.zo_alt_ztest,
6272 ztest_opts.zo_alt_libpath, B_FALSE, NULL));
6274 VERIFY(!exec_child(NULL, NULL, B_FALSE, NULL));
6276 zs->zs_do_init = B_FALSE;
6278 zs->zs_proc_start = gethrtime();
6279 zs->zs_proc_stop = zs->zs_proc_start + ztest_opts.zo_time * NANOSEC;
6281 for (int f = 0; f < ZTEST_FUNCS; f++) {
6282 zi = &ztest_info[f];
6283 zc = ZTEST_GET_SHARED_CALLSTATE(f);
6284 if (zs->zs_proc_start + zi->zi_interval[0] > zs->zs_proc_stop)
6285 zc->zc_next = UINT64_MAX;
6287 zc->zc_next = zs->zs_proc_start +
6288 ztest_random(2 * zi->zi_interval[0] + 1);
6292 * Run the tests in a loop. These tests include fault injection
6293 * to verify that self-healing data works, and forced crashes
6294 * to verify that we never lose on-disk consistency.
6296 while (gethrtime() < zs->zs_proc_stop) {
6301 * Initialize the workload counters for each function.
6303 for (int f = 0; f < ZTEST_FUNCS; f++) {
6304 zc = ZTEST_GET_SHARED_CALLSTATE(f);
6309 /* Set the allocation switch size */
6310 zs->zs_metaslab_df_alloc_threshold =
6311 ztest_random(zs->zs_metaslab_sz / 4) + 1;
6313 if (!hasalt || ztest_random(2) == 0) {
6314 if (hasalt && ztest_opts.zo_verbose >= 1) {
6315 (void) printf("Executing newer ztest: %s\n",
6319 killed = exec_child(cmd, NULL, B_TRUE, &status);
6321 if (hasalt && ztest_opts.zo_verbose >= 1) {
6322 (void) printf("Executing older ztest: %s\n",
6323 ztest_opts.zo_alt_ztest);
6326 killed = exec_child(ztest_opts.zo_alt_ztest,
6327 ztest_opts.zo_alt_libpath, B_TRUE, &status);
6334 if (ztest_opts.zo_verbose >= 1) {
6335 hrtime_t now = gethrtime();
6337 now = MIN(now, zs->zs_proc_stop);
6338 print_time(zs->zs_proc_stop - now, timebuf);
6339 nicenum(zs->zs_space, numbuf, sizeof (numbuf));
6341 (void) printf("Pass %3d, %8s, %3llu ENOSPC, "
6342 "%4.1f%% of %5s used, %3.0f%% done, %8s to go\n",
6344 WIFEXITED(status) ? "Complete" : "SIGKILL",
6345 (u_longlong_t)zs->zs_enospc_count,
6346 100.0 * zs->zs_alloc / zs->zs_space,
6348 100.0 * (now - zs->zs_proc_start) /
6349 (ztest_opts.zo_time * NANOSEC), timebuf);
6352 if (ztest_opts.zo_verbose >= 2) {
6353 (void) printf("\nWorkload summary:\n\n");
6354 (void) printf("%7s %9s %s\n",
6355 "Calls", "Time", "Function");
6356 (void) printf("%7s %9s %s\n",
6357 "-----", "----", "--------");
6358 for (int f = 0; f < ZTEST_FUNCS; f++) {
6361 zi = &ztest_info[f];
6362 zc = ZTEST_GET_SHARED_CALLSTATE(f);
6363 print_time(zc->zc_time, timebuf);
6364 (void) dladdr((void *)zi->zi_func, &dli);
6365 (void) printf("%7llu %9s %s\n",
6366 (u_longlong_t)zc->zc_count, timebuf,
6369 (void) printf("\n");
6373 * It's possible that we killed a child during a rename test,
6374 * in which case we'll have a 'ztest_tmp' pool lying around
6375 * instead of 'ztest'. Do a blind rename in case this happened.
6378 if (spa_open(ztest_opts.zo_pool, &spa, FTAG) == 0) {
6379 spa_close(spa, FTAG);
6381 char tmpname[ZFS_MAX_DATASET_NAME_LEN];
6383 kernel_init(FREAD | FWRITE);
6384 (void) snprintf(tmpname, sizeof (tmpname), "%s_tmp",
6385 ztest_opts.zo_pool);
6386 (void) spa_rename(tmpname, ztest_opts.zo_pool);
6390 ztest_run_zdb(ztest_opts.zo_pool);
6393 if (ztest_opts.zo_verbose >= 1) {
6395 (void) printf("%d runs of older ztest: %s\n", older,
6396 ztest_opts.zo_alt_ztest);
6397 (void) printf("%d runs of newer ztest: %s\n", newer,
6400 (void) printf("%d killed, %d completed, %.0f%% kill rate\n",
6401 kills, iters - kills, (100.0 * kills) / MAX(1, iters));
6404 umem_free(cmd, MAXNAMELEN);