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, 2014 by Delphix. All rights reserved.
24 * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
25 * Copyright (c) 2012 Martin Matuska <mm@FreeBSD.org>. All rights reserved.
26 * Copyright (c) 2013 Steven Hartland. All rights reserved.
30 * The objective of this program is to provide a DMU/ZAP/SPA stress test
31 * that runs entirely in userland, is easy to use, and easy to extend.
33 * The overall design of the ztest program is as follows:
35 * (1) For each major functional area (e.g. adding vdevs to a pool,
36 * creating and destroying datasets, reading and writing objects, etc)
37 * we have a simple routine to test that functionality. These
38 * individual routines do not have to do anything "stressful".
40 * (2) We turn these simple functionality tests into a stress test by
41 * running them all in parallel, with as many threads as desired,
42 * and spread across as many datasets, objects, and vdevs as desired.
44 * (3) While all this is happening, we inject faults into the pool to
45 * verify that self-healing data really works.
47 * (4) Every time we open a dataset, we change its checksum and compression
48 * functions. Thus even individual objects vary from block to block
49 * in which checksum they use and whether they're compressed.
51 * (5) To verify that we never lose on-disk consistency after a crash,
52 * we run the entire test in a child of the main process.
53 * At random times, the child self-immolates with a SIGKILL.
54 * This is the software equivalent of pulling the power cord.
55 * The parent then runs the test again, using the existing
56 * storage pool, as many times as desired. If backwards compatibility
57 * testing is enabled ztest will sometimes run the "older" version
58 * of ztest after a SIGKILL.
60 * (6) To verify that we don't have future leaks or temporal incursions,
61 * many of the functional tests record the transaction group number
62 * as part of their data. When reading old data, they verify that
63 * the transaction group number is less than the current, open txg.
64 * If you add a new test, please do this if applicable.
66 * When run with no arguments, ztest runs for about five minutes and
67 * produces no output if successful. To get a little bit of information,
68 * specify -V. To get more information, specify -VV, and so on.
70 * To turn this into an overnight stress test, use -T to specify run time.
72 * You can ask more more vdevs [-v], datasets [-d], or threads [-t]
73 * to increase the pool capacity, fanout, and overall stress level.
75 * Use the -k option to set the desired frequency of kills.
77 * When ztest invokes itself it passes all relevant information through a
78 * temporary file which is mmap-ed in the child process. This allows shared
79 * memory to survive the exec syscall. The ztest_shared_hdr_t struct is always
80 * stored at offset 0 of this file and contains information on the size and
81 * number of shared structures in the file. The information stored in this file
82 * must remain backwards compatible with older versions of ztest so that
83 * ztest can invoke them during backwards compatibility testing (-B).
86 #include <sys/zfs_context.h>
92 #include <sys/dmu_objset.h>
98 #include <sys/resource.h>
101 #include <sys/zil_impl.h>
102 #include <sys/vdev_impl.h>
103 #include <sys/vdev_file.h>
104 #include <sys/spa_impl.h>
105 #include <sys/metaslab_impl.h>
106 #include <sys/dsl_prop.h>
107 #include <sys/dsl_dataset.h>
108 #include <sys/dsl_destroy.h>
109 #include <sys/dsl_scan.h>
110 #include <sys/zio_checksum.h>
111 #include <sys/refcount.h>
112 #include <sys/zfeature.h>
113 #include <sys/dsl_userhold.h>
115 #include <stdio_ext.h>
124 #include <sys/fs/zfs.h>
125 #include <libnvpair.h>
127 static int ztest_fd_data = -1;
128 static int ztest_fd_rand = -1;
130 typedef struct ztest_shared_hdr {
131 uint64_t zh_hdr_size;
132 uint64_t zh_opts_size;
134 uint64_t zh_stats_size;
135 uint64_t zh_stats_count;
137 uint64_t zh_ds_count;
138 } ztest_shared_hdr_t;
140 static ztest_shared_hdr_t *ztest_shared_hdr;
142 typedef struct ztest_shared_opts {
143 char zo_pool[MAXNAMELEN];
144 char zo_dir[MAXNAMELEN];
145 char zo_alt_ztest[MAXNAMELEN];
146 char zo_alt_libpath[MAXNAMELEN];
148 uint64_t zo_vdevtime;
156 uint64_t zo_passtime;
157 uint64_t zo_killrate;
161 uint64_t zo_maxloops;
162 uint64_t zo_metaslab_gang_bang;
163 } ztest_shared_opts_t;
165 static const ztest_shared_opts_t ztest_opts_defaults = {
166 .zo_pool = { 'z', 't', 'e', 's', 't', '\0' },
167 .zo_dir = { '/', 't', 'm', 'p', '\0' },
168 .zo_alt_ztest = { '\0' },
169 .zo_alt_libpath = { '\0' },
171 .zo_ashift = SPA_MINBLOCKSHIFT,
174 .zo_raidz_parity = 1,
175 .zo_vdev_size = SPA_MINDEVSIZE * 2,
178 .zo_passtime = 60, /* 60 seconds */
179 .zo_killrate = 70, /* 70% kill rate */
182 .zo_time = 300, /* 5 minutes */
183 .zo_maxloops = 50, /* max loops during spa_freeze() */
184 .zo_metaslab_gang_bang = 32 << 10
187 extern uint64_t metaslab_gang_bang;
188 extern uint64_t metaslab_df_alloc_threshold;
189 extern uint64_t zfs_deadman_synctime_ms;
190 extern int metaslab_preload_limit;
192 static ztest_shared_opts_t *ztest_shared_opts;
193 static ztest_shared_opts_t ztest_opts;
195 typedef struct ztest_shared_ds {
199 static ztest_shared_ds_t *ztest_shared_ds;
200 #define ZTEST_GET_SHARED_DS(d) (&ztest_shared_ds[d])
202 #define BT_MAGIC 0x123456789abcdefULL
203 #define MAXFAULTS() \
204 (MAX(zs->zs_mirrors, 1) * (ztest_opts.zo_raidz_parity + 1) - 1)
208 ZTEST_IO_WRITE_PATTERN,
209 ZTEST_IO_WRITE_ZEROES,
216 typedef struct ztest_block_tag {
226 typedef struct bufwad {
233 * XXX -- fix zfs range locks to be generic so we can use them here.
255 #define ZTEST_RANGE_LOCKS 64
256 #define ZTEST_OBJECT_LOCKS 64
259 * Object descriptor. Used as a template for object lookup/create/remove.
261 typedef struct ztest_od {
264 dmu_object_type_t od_type;
265 dmu_object_type_t od_crtype;
266 uint64_t od_blocksize;
267 uint64_t od_crblocksize;
270 char od_name[MAXNAMELEN];
276 typedef struct ztest_ds {
277 ztest_shared_ds_t *zd_shared;
279 rwlock_t zd_zilog_lock;
281 ztest_od_t *zd_od; /* debugging aid */
282 char zd_name[MAXNAMELEN];
283 mutex_t zd_dirobj_lock;
284 rll_t zd_object_lock[ZTEST_OBJECT_LOCKS];
285 rll_t zd_range_lock[ZTEST_RANGE_LOCKS];
289 * Per-iteration state.
291 typedef void ztest_func_t(ztest_ds_t *zd, uint64_t id);
293 typedef struct ztest_info {
294 ztest_func_t *zi_func; /* test function */
295 uint64_t zi_iters; /* iterations per execution */
296 uint64_t *zi_interval; /* execute every <interval> seconds */
299 typedef struct ztest_shared_callstate {
300 uint64_t zc_count; /* per-pass count */
301 uint64_t zc_time; /* per-pass time */
302 uint64_t zc_next; /* next time to call this function */
303 } ztest_shared_callstate_t;
305 static ztest_shared_callstate_t *ztest_shared_callstate;
306 #define ZTEST_GET_SHARED_CALLSTATE(c) (&ztest_shared_callstate[c])
309 * Note: these aren't static because we want dladdr() to work.
311 ztest_func_t ztest_dmu_read_write;
312 ztest_func_t ztest_dmu_write_parallel;
313 ztest_func_t ztest_dmu_object_alloc_free;
314 ztest_func_t ztest_dmu_commit_callbacks;
315 ztest_func_t ztest_zap;
316 ztest_func_t ztest_zap_parallel;
317 ztest_func_t ztest_zil_commit;
318 ztest_func_t ztest_zil_remount;
319 ztest_func_t ztest_dmu_read_write_zcopy;
320 ztest_func_t ztest_dmu_objset_create_destroy;
321 ztest_func_t ztest_dmu_prealloc;
322 ztest_func_t ztest_fzap;
323 ztest_func_t ztest_dmu_snapshot_create_destroy;
324 ztest_func_t ztest_dsl_prop_get_set;
325 ztest_func_t ztest_spa_prop_get_set;
326 ztest_func_t ztest_spa_create_destroy;
327 ztest_func_t ztest_fault_inject;
328 ztest_func_t ztest_ddt_repair;
329 ztest_func_t ztest_dmu_snapshot_hold;
330 ztest_func_t ztest_spa_rename;
331 ztest_func_t ztest_scrub;
332 ztest_func_t ztest_dsl_dataset_promote_busy;
333 ztest_func_t ztest_vdev_attach_detach;
334 ztest_func_t ztest_vdev_LUN_growth;
335 ztest_func_t ztest_vdev_add_remove;
336 ztest_func_t ztest_vdev_aux_add_remove;
337 ztest_func_t ztest_split_pool;
338 ztest_func_t ztest_reguid;
339 ztest_func_t ztest_spa_upgrade;
341 uint64_t zopt_always = 0ULL * NANOSEC; /* all the time */
342 uint64_t zopt_incessant = 1ULL * NANOSEC / 10; /* every 1/10 second */
343 uint64_t zopt_often = 1ULL * NANOSEC; /* every second */
344 uint64_t zopt_sometimes = 10ULL * NANOSEC; /* every 10 seconds */
345 uint64_t zopt_rarely = 60ULL * NANOSEC; /* every 60 seconds */
347 ztest_info_t ztest_info[] = {
348 { ztest_dmu_read_write, 1, &zopt_always },
349 { ztest_dmu_write_parallel, 10, &zopt_always },
350 { ztest_dmu_object_alloc_free, 1, &zopt_always },
351 { ztest_dmu_commit_callbacks, 1, &zopt_always },
352 { ztest_zap, 30, &zopt_always },
353 { ztest_zap_parallel, 100, &zopt_always },
354 { ztest_split_pool, 1, &zopt_always },
355 { ztest_zil_commit, 1, &zopt_incessant },
356 { ztest_zil_remount, 1, &zopt_sometimes },
357 { ztest_dmu_read_write_zcopy, 1, &zopt_often },
358 { ztest_dmu_objset_create_destroy, 1, &zopt_often },
359 { ztest_dsl_prop_get_set, 1, &zopt_often },
360 { ztest_spa_prop_get_set, 1, &zopt_sometimes },
362 { ztest_dmu_prealloc, 1, &zopt_sometimes },
364 { ztest_fzap, 1, &zopt_sometimes },
365 { ztest_dmu_snapshot_create_destroy, 1, &zopt_sometimes },
366 { ztest_spa_create_destroy, 1, &zopt_sometimes },
367 { ztest_fault_inject, 1, &zopt_sometimes },
368 { ztest_ddt_repair, 1, &zopt_sometimes },
369 { ztest_dmu_snapshot_hold, 1, &zopt_sometimes },
370 { ztest_reguid, 1, &zopt_rarely },
371 { ztest_spa_rename, 1, &zopt_rarely },
372 { ztest_scrub, 1, &zopt_rarely },
373 { ztest_spa_upgrade, 1, &zopt_rarely },
374 { ztest_dsl_dataset_promote_busy, 1, &zopt_rarely },
375 { ztest_vdev_attach_detach, 1, &zopt_sometimes },
376 { ztest_vdev_LUN_growth, 1, &zopt_rarely },
377 { ztest_vdev_add_remove, 1,
378 &ztest_opts.zo_vdevtime },
379 { ztest_vdev_aux_add_remove, 1,
380 &ztest_opts.zo_vdevtime },
383 #define ZTEST_FUNCS (sizeof (ztest_info) / sizeof (ztest_info_t))
386 * The following struct is used to hold a list of uncalled commit callbacks.
387 * The callbacks are ordered by txg number.
389 typedef struct ztest_cb_list {
390 mutex_t zcl_callbacks_lock;
391 list_t zcl_callbacks;
395 * Stuff we need to share writably between parent and child.
397 typedef struct ztest_shared {
398 boolean_t zs_do_init;
399 hrtime_t zs_proc_start;
400 hrtime_t zs_proc_stop;
401 hrtime_t zs_thread_start;
402 hrtime_t zs_thread_stop;
403 hrtime_t zs_thread_kill;
404 uint64_t zs_enospc_count;
405 uint64_t zs_vdev_next_leaf;
406 uint64_t zs_vdev_aux;
411 uint64_t zs_metaslab_sz;
412 uint64_t zs_metaslab_df_alloc_threshold;
416 #define ID_PARALLEL -1ULL
418 static char ztest_dev_template[] = "%s/%s.%llua";
419 static char ztest_aux_template[] = "%s/%s.%s.%llu";
420 ztest_shared_t *ztest_shared;
422 static spa_t *ztest_spa = NULL;
423 static ztest_ds_t *ztest_ds;
425 static mutex_t ztest_vdev_lock;
428 * The ztest_name_lock protects the pool and dataset namespace used by
429 * the individual tests. To modify the namespace, consumers must grab
430 * this lock as writer. Grabbing the lock as reader will ensure that the
431 * namespace does not change while the lock is held.
433 static rwlock_t ztest_name_lock;
435 static boolean_t ztest_dump_core = B_TRUE;
436 static boolean_t ztest_exiting;
438 /* Global commit callback list */
439 static ztest_cb_list_t zcl;
442 ZTEST_META_DNODE = 0,
447 static void usage(boolean_t) __NORETURN;
450 * These libumem hooks provide a reasonable set of defaults for the allocator's
451 * debugging facilities.
456 return ("default,verbose"); /* $UMEM_DEBUG setting */
460 _umem_logging_init(void)
462 return ("fail,contents"); /* $UMEM_LOGGING setting */
465 #define FATAL_MSG_SZ 1024
470 fatal(int do_perror, char *message, ...)
473 int save_errno = errno;
474 char buf[FATAL_MSG_SZ];
476 (void) fflush(stdout);
478 va_start(args, message);
479 (void) sprintf(buf, "ztest: ");
481 (void) vsprintf(buf + strlen(buf), message, args);
484 (void) snprintf(buf + strlen(buf), FATAL_MSG_SZ - strlen(buf),
485 ": %s", strerror(save_errno));
487 (void) fprintf(stderr, "%s\n", buf);
488 fatal_msg = buf; /* to ease debugging */
495 str2shift(const char *buf)
497 const char *ends = "BKMGTPEZ";
502 for (i = 0; i < strlen(ends); i++) {
503 if (toupper(buf[0]) == ends[i])
506 if (i == strlen(ends)) {
507 (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n",
511 if (buf[1] == '\0' || (toupper(buf[1]) == 'B' && buf[2] == '\0')) {
514 (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n", buf);
520 nicenumtoull(const char *buf)
525 val = strtoull(buf, &end, 0);
527 (void) fprintf(stderr, "ztest: bad numeric value: %s\n", buf);
529 } else if (end[0] == '.') {
530 double fval = strtod(buf, &end);
531 fval *= pow(2, str2shift(end));
532 if (fval > UINT64_MAX) {
533 (void) fprintf(stderr, "ztest: value too large: %s\n",
537 val = (uint64_t)fval;
539 int shift = str2shift(end);
540 if (shift >= 64 || (val << shift) >> shift != val) {
541 (void) fprintf(stderr, "ztest: value too large: %s\n",
551 usage(boolean_t requested)
553 const ztest_shared_opts_t *zo = &ztest_opts_defaults;
555 char nice_vdev_size[10];
556 char nice_gang_bang[10];
557 FILE *fp = requested ? stdout : stderr;
559 nicenum(zo->zo_vdev_size, nice_vdev_size);
560 nicenum(zo->zo_metaslab_gang_bang, nice_gang_bang);
562 (void) fprintf(fp, "Usage: %s\n"
563 "\t[-v vdevs (default: %llu)]\n"
564 "\t[-s size_of_each_vdev (default: %s)]\n"
565 "\t[-a alignment_shift (default: %d)] use 0 for random\n"
566 "\t[-m mirror_copies (default: %d)]\n"
567 "\t[-r raidz_disks (default: %d)]\n"
568 "\t[-R raidz_parity (default: %d)]\n"
569 "\t[-d datasets (default: %d)]\n"
570 "\t[-t threads (default: %d)]\n"
571 "\t[-g gang_block_threshold (default: %s)]\n"
572 "\t[-i init_count (default: %d)] initialize pool i times\n"
573 "\t[-k kill_percentage (default: %llu%%)]\n"
574 "\t[-p pool_name (default: %s)]\n"
575 "\t[-f dir (default: %s)] file directory for vdev files\n"
576 "\t[-V] verbose (use multiple times for ever more blather)\n"
577 "\t[-E] use existing pool instead of creating new one\n"
578 "\t[-T time (default: %llu sec)] total run time\n"
579 "\t[-F freezeloops (default: %llu)] max loops in spa_freeze()\n"
580 "\t[-P passtime (default: %llu sec)] time per pass\n"
581 "\t[-B alt_ztest (default: <none>)] alternate ztest path\n"
582 "\t[-h] (print help)\n"
585 (u_longlong_t)zo->zo_vdevs, /* -v */
586 nice_vdev_size, /* -s */
587 zo->zo_ashift, /* -a */
588 zo->zo_mirrors, /* -m */
589 zo->zo_raidz, /* -r */
590 zo->zo_raidz_parity, /* -R */
591 zo->zo_datasets, /* -d */
592 zo->zo_threads, /* -t */
593 nice_gang_bang, /* -g */
594 zo->zo_init, /* -i */
595 (u_longlong_t)zo->zo_killrate, /* -k */
596 zo->zo_pool, /* -p */
598 (u_longlong_t)zo->zo_time, /* -T */
599 (u_longlong_t)zo->zo_maxloops, /* -F */
600 (u_longlong_t)zo->zo_passtime);
601 exit(requested ? 0 : 1);
605 process_options(int argc, char **argv)
608 ztest_shared_opts_t *zo = &ztest_opts;
612 char altdir[MAXNAMELEN] = { 0 };
614 bcopy(&ztest_opts_defaults, zo, sizeof (*zo));
616 while ((opt = getopt(argc, argv,
617 "v:s:a:m:r:R:d:t:g:i:k:p:f:VET:P:hF:B:")) != EOF) {
634 value = nicenumtoull(optarg);
638 zo->zo_vdevs = value;
641 zo->zo_vdev_size = MAX(SPA_MINDEVSIZE, value);
644 zo->zo_ashift = value;
647 zo->zo_mirrors = value;
650 zo->zo_raidz = MAX(1, value);
653 zo->zo_raidz_parity = MIN(MAX(value, 1), 3);
656 zo->zo_datasets = MAX(1, value);
659 zo->zo_threads = MAX(1, value);
662 zo->zo_metaslab_gang_bang = MAX(SPA_MINBLOCKSIZE << 1,
669 zo->zo_killrate = value;
672 (void) strlcpy(zo->zo_pool, optarg,
673 sizeof (zo->zo_pool));
676 path = realpath(optarg, NULL);
678 (void) fprintf(stderr, "error: %s: %s\n",
679 optarg, strerror(errno));
682 (void) strlcpy(zo->zo_dir, path,
683 sizeof (zo->zo_dir));
696 zo->zo_passtime = MAX(1, value);
699 zo->zo_maxloops = MAX(1, value);
702 (void) strlcpy(altdir, optarg, sizeof (altdir));
714 zo->zo_raidz_parity = MIN(zo->zo_raidz_parity, zo->zo_raidz - 1);
717 (zo->zo_vdevs > 0 ? zo->zo_time * NANOSEC / zo->zo_vdevs :
720 if (strlen(altdir) > 0) {
728 cmd = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
729 realaltdir = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
731 VERIFY(NULL != realpath(getexecname(), cmd));
732 if (0 != access(altdir, F_OK)) {
733 ztest_dump_core = B_FALSE;
734 fatal(B_TRUE, "invalid alternate ztest path: %s",
737 VERIFY(NULL != realpath(altdir, realaltdir));
740 * 'cmd' should be of the form "<anything>/usr/bin/<isa>/ztest".
741 * We want to extract <isa> to determine if we should use
742 * 32 or 64 bit binaries.
744 bin = strstr(cmd, "/usr/bin/");
745 ztest = strstr(bin, "/ztest");
747 isalen = ztest - isa;
748 (void) snprintf(zo->zo_alt_ztest, sizeof (zo->zo_alt_ztest),
749 "%s/usr/bin/%.*s/ztest", realaltdir, isalen, isa);
750 (void) snprintf(zo->zo_alt_libpath, sizeof (zo->zo_alt_libpath),
751 "%s/usr/lib/%.*s", realaltdir, isalen, isa);
753 if (0 != access(zo->zo_alt_ztest, X_OK)) {
754 ztest_dump_core = B_FALSE;
755 fatal(B_TRUE, "invalid alternate ztest: %s",
757 } else if (0 != access(zo->zo_alt_libpath, X_OK)) {
758 ztest_dump_core = B_FALSE;
759 fatal(B_TRUE, "invalid alternate lib directory %s",
763 umem_free(cmd, MAXPATHLEN);
764 umem_free(realaltdir, MAXPATHLEN);
769 ztest_kill(ztest_shared_t *zs)
771 zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(ztest_spa));
772 zs->zs_space = metaslab_class_get_space(spa_normal_class(ztest_spa));
775 * Before we kill off ztest, make sure that the config is updated.
776 * See comment above spa_config_sync().
778 mutex_enter(&spa_namespace_lock);
779 spa_config_sync(ztest_spa, B_FALSE, B_FALSE);
780 mutex_exit(&spa_namespace_lock);
782 zfs_dbgmsg_print(FTAG);
783 (void) kill(getpid(), SIGKILL);
787 ztest_random(uint64_t range)
791 ASSERT3S(ztest_fd_rand, >=, 0);
796 if (read(ztest_fd_rand, &r, sizeof (r)) != sizeof (r))
797 fatal(1, "short read from /dev/urandom");
804 ztest_record_enospc(const char *s)
806 ztest_shared->zs_enospc_count++;
810 ztest_get_ashift(void)
812 if (ztest_opts.zo_ashift == 0)
813 return (SPA_MINBLOCKSHIFT + ztest_random(5));
814 return (ztest_opts.zo_ashift);
818 make_vdev_file(char *path, char *aux, char *pool, size_t size, uint64_t ashift)
820 char pathbuf[MAXPATHLEN];
825 ashift = ztest_get_ashift();
831 vdev = ztest_shared->zs_vdev_aux;
832 (void) snprintf(path, sizeof (pathbuf),
833 ztest_aux_template, ztest_opts.zo_dir,
834 pool == NULL ? ztest_opts.zo_pool : pool,
837 vdev = ztest_shared->zs_vdev_next_leaf++;
838 (void) snprintf(path, sizeof (pathbuf),
839 ztest_dev_template, ztest_opts.zo_dir,
840 pool == NULL ? ztest_opts.zo_pool : pool, vdev);
845 int fd = open(path, O_RDWR | O_CREAT | O_TRUNC, 0666);
847 fatal(1, "can't open %s", path);
848 if (ftruncate(fd, size) != 0)
849 fatal(1, "can't ftruncate %s", path);
853 VERIFY(nvlist_alloc(&file, NV_UNIQUE_NAME, 0) == 0);
854 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_TYPE, VDEV_TYPE_FILE) == 0);
855 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_PATH, path) == 0);
856 VERIFY(nvlist_add_uint64(file, ZPOOL_CONFIG_ASHIFT, ashift) == 0);
862 make_vdev_raidz(char *path, char *aux, char *pool, size_t size,
863 uint64_t ashift, int r)
865 nvlist_t *raidz, **child;
869 return (make_vdev_file(path, aux, pool, size, ashift));
870 child = umem_alloc(r * sizeof (nvlist_t *), UMEM_NOFAIL);
872 for (c = 0; c < r; c++)
873 child[c] = make_vdev_file(path, aux, pool, size, ashift);
875 VERIFY(nvlist_alloc(&raidz, NV_UNIQUE_NAME, 0) == 0);
876 VERIFY(nvlist_add_string(raidz, ZPOOL_CONFIG_TYPE,
877 VDEV_TYPE_RAIDZ) == 0);
878 VERIFY(nvlist_add_uint64(raidz, ZPOOL_CONFIG_NPARITY,
879 ztest_opts.zo_raidz_parity) == 0);
880 VERIFY(nvlist_add_nvlist_array(raidz, ZPOOL_CONFIG_CHILDREN,
883 for (c = 0; c < r; c++)
884 nvlist_free(child[c]);
886 umem_free(child, r * sizeof (nvlist_t *));
892 make_vdev_mirror(char *path, char *aux, char *pool, size_t size,
893 uint64_t ashift, int r, int m)
895 nvlist_t *mirror, **child;
899 return (make_vdev_raidz(path, aux, pool, size, ashift, r));
901 child = umem_alloc(m * sizeof (nvlist_t *), UMEM_NOFAIL);
903 for (c = 0; c < m; c++)
904 child[c] = make_vdev_raidz(path, aux, pool, size, ashift, r);
906 VERIFY(nvlist_alloc(&mirror, NV_UNIQUE_NAME, 0) == 0);
907 VERIFY(nvlist_add_string(mirror, ZPOOL_CONFIG_TYPE,
908 VDEV_TYPE_MIRROR) == 0);
909 VERIFY(nvlist_add_nvlist_array(mirror, ZPOOL_CONFIG_CHILDREN,
912 for (c = 0; c < m; c++)
913 nvlist_free(child[c]);
915 umem_free(child, m * sizeof (nvlist_t *));
921 make_vdev_root(char *path, char *aux, char *pool, size_t size, uint64_t ashift,
922 int log, int r, int m, int t)
924 nvlist_t *root, **child;
929 child = umem_alloc(t * sizeof (nvlist_t *), UMEM_NOFAIL);
931 for (c = 0; c < t; c++) {
932 child[c] = make_vdev_mirror(path, aux, pool, size, ashift,
934 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_IS_LOG,
938 VERIFY(nvlist_alloc(&root, NV_UNIQUE_NAME, 0) == 0);
939 VERIFY(nvlist_add_string(root, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) == 0);
940 VERIFY(nvlist_add_nvlist_array(root, aux ? aux : ZPOOL_CONFIG_CHILDREN,
943 for (c = 0; c < t; c++)
944 nvlist_free(child[c]);
946 umem_free(child, t * sizeof (nvlist_t *));
952 * Find a random spa version. Returns back a random spa version in the
953 * range [initial_version, SPA_VERSION_FEATURES].
956 ztest_random_spa_version(uint64_t initial_version)
958 uint64_t version = initial_version;
960 if (version <= SPA_VERSION_BEFORE_FEATURES) {
962 ztest_random(SPA_VERSION_BEFORE_FEATURES - version + 1);
965 if (version > SPA_VERSION_BEFORE_FEATURES)
966 version = SPA_VERSION_FEATURES;
968 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
973 * Find the largest ashift used
976 ztest_spa_get_ashift() {
978 uint64_t ashift = SPA_MINBLOCKSHIFT;
979 vdev_t *rvd = ztest_spa->spa_root_vdev;
981 for (i = 0; i < rvd->vdev_children; i++) {
982 ashift = MAX(ashift, rvd->vdev_child[i]->vdev_ashift);
988 ztest_random_blocksize(void)
990 uint64_t block_shift;
992 * Choose a block size >= the ashift.
993 * If the SPA supports new MAXBLOCKSIZE, test up to 1MB blocks.
995 int maxbs = SPA_OLD_MAXBLOCKSHIFT;
996 if (spa_maxblocksize(ztest_spa) == SPA_MAXBLOCKSIZE)
998 block_shift = ztest_random(maxbs - ztest_spa_get_ashift() + 1);
999 return (1 << (SPA_MINBLOCKSHIFT + block_shift));
1003 ztest_random_ibshift(void)
1005 return (DN_MIN_INDBLKSHIFT +
1006 ztest_random(DN_MAX_INDBLKSHIFT - DN_MIN_INDBLKSHIFT + 1));
1010 ztest_random_vdev_top(spa_t *spa, boolean_t log_ok)
1013 vdev_t *rvd = spa->spa_root_vdev;
1016 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1019 top = ztest_random(rvd->vdev_children);
1020 tvd = rvd->vdev_child[top];
1021 } while (tvd->vdev_ishole || (tvd->vdev_islog && !log_ok) ||
1022 tvd->vdev_mg == NULL || tvd->vdev_mg->mg_class == NULL);
1028 ztest_random_dsl_prop(zfs_prop_t prop)
1033 value = zfs_prop_random_value(prop, ztest_random(-1ULL));
1034 } while (prop == ZFS_PROP_CHECKSUM && value == ZIO_CHECKSUM_OFF);
1040 ztest_dsl_prop_set_uint64(char *osname, zfs_prop_t prop, uint64_t value,
1043 const char *propname = zfs_prop_to_name(prop);
1044 const char *valname;
1045 char setpoint[MAXPATHLEN];
1049 error = dsl_prop_set_int(osname, propname,
1050 (inherit ? ZPROP_SRC_NONE : ZPROP_SRC_LOCAL), value);
1052 if (error == ENOSPC) {
1053 ztest_record_enospc(FTAG);
1058 VERIFY0(dsl_prop_get_integer(osname, propname, &curval, setpoint));
1060 if (ztest_opts.zo_verbose >= 6) {
1061 VERIFY(zfs_prop_index_to_string(prop, curval, &valname) == 0);
1062 (void) printf("%s %s = %s at '%s'\n",
1063 osname, propname, valname, setpoint);
1070 ztest_spa_prop_set_uint64(zpool_prop_t prop, uint64_t value)
1072 spa_t *spa = ztest_spa;
1073 nvlist_t *props = NULL;
1076 VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
1077 VERIFY(nvlist_add_uint64(props, zpool_prop_to_name(prop), value) == 0);
1079 error = spa_prop_set(spa, props);
1083 if (error == ENOSPC) {
1084 ztest_record_enospc(FTAG);
1093 ztest_rll_init(rll_t *rll)
1095 rll->rll_writer = NULL;
1096 rll->rll_readers = 0;
1097 VERIFY(_mutex_init(&rll->rll_lock, USYNC_THREAD, NULL) == 0);
1098 VERIFY(cond_init(&rll->rll_cv, USYNC_THREAD, NULL) == 0);
1102 ztest_rll_destroy(rll_t *rll)
1104 ASSERT(rll->rll_writer == NULL);
1105 ASSERT(rll->rll_readers == 0);
1106 VERIFY(_mutex_destroy(&rll->rll_lock) == 0);
1107 VERIFY(cond_destroy(&rll->rll_cv) == 0);
1111 ztest_rll_lock(rll_t *rll, rl_type_t type)
1113 VERIFY(mutex_lock(&rll->rll_lock) == 0);
1115 if (type == RL_READER) {
1116 while (rll->rll_writer != NULL)
1117 (void) cond_wait(&rll->rll_cv, &rll->rll_lock);
1120 while (rll->rll_writer != NULL || rll->rll_readers)
1121 (void) cond_wait(&rll->rll_cv, &rll->rll_lock);
1122 rll->rll_writer = curthread;
1125 VERIFY(mutex_unlock(&rll->rll_lock) == 0);
1129 ztest_rll_unlock(rll_t *rll)
1131 VERIFY(mutex_lock(&rll->rll_lock) == 0);
1133 if (rll->rll_writer) {
1134 ASSERT(rll->rll_readers == 0);
1135 rll->rll_writer = NULL;
1137 ASSERT(rll->rll_readers != 0);
1138 ASSERT(rll->rll_writer == NULL);
1142 if (rll->rll_writer == NULL && rll->rll_readers == 0)
1143 VERIFY(cond_broadcast(&rll->rll_cv) == 0);
1145 VERIFY(mutex_unlock(&rll->rll_lock) == 0);
1149 ztest_object_lock(ztest_ds_t *zd, uint64_t object, rl_type_t type)
1151 rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1153 ztest_rll_lock(rll, type);
1157 ztest_object_unlock(ztest_ds_t *zd, uint64_t object)
1159 rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1161 ztest_rll_unlock(rll);
1165 ztest_range_lock(ztest_ds_t *zd, uint64_t object, uint64_t offset,
1166 uint64_t size, rl_type_t type)
1168 uint64_t hash = object ^ (offset % (ZTEST_RANGE_LOCKS + 1));
1169 rll_t *rll = &zd->zd_range_lock[hash & (ZTEST_RANGE_LOCKS - 1)];
1172 rl = umem_alloc(sizeof (*rl), UMEM_NOFAIL);
1173 rl->rl_object = object;
1174 rl->rl_offset = offset;
1178 ztest_rll_lock(rll, type);
1184 ztest_range_unlock(rl_t *rl)
1186 rll_t *rll = rl->rl_lock;
1188 ztest_rll_unlock(rll);
1190 umem_free(rl, sizeof (*rl));
1194 ztest_zd_init(ztest_ds_t *zd, ztest_shared_ds_t *szd, objset_t *os)
1197 zd->zd_zilog = dmu_objset_zil(os);
1198 zd->zd_shared = szd;
1199 dmu_objset_name(os, zd->zd_name);
1201 if (zd->zd_shared != NULL)
1202 zd->zd_shared->zd_seq = 0;
1204 VERIFY(rwlock_init(&zd->zd_zilog_lock, USYNC_THREAD, NULL) == 0);
1205 VERIFY(_mutex_init(&zd->zd_dirobj_lock, USYNC_THREAD, NULL) == 0);
1207 for (int l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1208 ztest_rll_init(&zd->zd_object_lock[l]);
1210 for (int l = 0; l < ZTEST_RANGE_LOCKS; l++)
1211 ztest_rll_init(&zd->zd_range_lock[l]);
1215 ztest_zd_fini(ztest_ds_t *zd)
1217 VERIFY(_mutex_destroy(&zd->zd_dirobj_lock) == 0);
1219 for (int l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1220 ztest_rll_destroy(&zd->zd_object_lock[l]);
1222 for (int l = 0; l < ZTEST_RANGE_LOCKS; l++)
1223 ztest_rll_destroy(&zd->zd_range_lock[l]);
1226 #define TXG_MIGHTWAIT (ztest_random(10) == 0 ? TXG_NOWAIT : TXG_WAIT)
1229 ztest_tx_assign(dmu_tx_t *tx, uint64_t txg_how, const char *tag)
1235 * Attempt to assign tx to some transaction group.
1237 error = dmu_tx_assign(tx, txg_how);
1239 if (error == ERESTART) {
1240 ASSERT(txg_how == TXG_NOWAIT);
1243 ASSERT3U(error, ==, ENOSPC);
1244 ztest_record_enospc(tag);
1249 txg = dmu_tx_get_txg(tx);
1255 ztest_pattern_set(void *buf, uint64_t size, uint64_t value)
1258 uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1265 ztest_pattern_match(void *buf, uint64_t size, uint64_t value)
1268 uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1272 diff |= (value - *ip++);
1278 ztest_bt_generate(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1279 uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg)
1281 bt->bt_magic = BT_MAGIC;
1282 bt->bt_objset = dmu_objset_id(os);
1283 bt->bt_object = object;
1284 bt->bt_offset = offset;
1287 bt->bt_crtxg = crtxg;
1291 ztest_bt_verify(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1292 uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg)
1294 ASSERT3U(bt->bt_magic, ==, BT_MAGIC);
1295 ASSERT3U(bt->bt_objset, ==, dmu_objset_id(os));
1296 ASSERT3U(bt->bt_object, ==, object);
1297 ASSERT3U(bt->bt_offset, ==, offset);
1298 ASSERT3U(bt->bt_gen, <=, gen);
1299 ASSERT3U(bt->bt_txg, <=, txg);
1300 ASSERT3U(bt->bt_crtxg, ==, crtxg);
1303 static ztest_block_tag_t *
1304 ztest_bt_bonus(dmu_buf_t *db)
1306 dmu_object_info_t doi;
1307 ztest_block_tag_t *bt;
1309 dmu_object_info_from_db(db, &doi);
1310 ASSERT3U(doi.doi_bonus_size, <=, db->db_size);
1311 ASSERT3U(doi.doi_bonus_size, >=, sizeof (*bt));
1312 bt = (void *)((char *)db->db_data + doi.doi_bonus_size - sizeof (*bt));
1321 #define lrz_type lr_mode
1322 #define lrz_blocksize lr_uid
1323 #define lrz_ibshift lr_gid
1324 #define lrz_bonustype lr_rdev
1325 #define lrz_bonuslen lr_crtime[1]
1328 ztest_log_create(ztest_ds_t *zd, dmu_tx_t *tx, lr_create_t *lr)
1330 char *name = (void *)(lr + 1); /* name follows lr */
1331 size_t namesize = strlen(name) + 1;
1334 if (zil_replaying(zd->zd_zilog, tx))
1337 itx = zil_itx_create(TX_CREATE, sizeof (*lr) + namesize);
1338 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1339 sizeof (*lr) + namesize - sizeof (lr_t));
1341 zil_itx_assign(zd->zd_zilog, itx, tx);
1345 ztest_log_remove(ztest_ds_t *zd, dmu_tx_t *tx, lr_remove_t *lr, uint64_t object)
1347 char *name = (void *)(lr + 1); /* name follows lr */
1348 size_t namesize = strlen(name) + 1;
1351 if (zil_replaying(zd->zd_zilog, tx))
1354 itx = zil_itx_create(TX_REMOVE, sizeof (*lr) + namesize);
1355 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1356 sizeof (*lr) + namesize - sizeof (lr_t));
1358 itx->itx_oid = object;
1359 zil_itx_assign(zd->zd_zilog, itx, tx);
1363 ztest_log_write(ztest_ds_t *zd, dmu_tx_t *tx, lr_write_t *lr)
1366 itx_wr_state_t write_state = ztest_random(WR_NUM_STATES);
1368 if (zil_replaying(zd->zd_zilog, tx))
1371 if (lr->lr_length > ZIL_MAX_LOG_DATA)
1372 write_state = WR_INDIRECT;
1374 itx = zil_itx_create(TX_WRITE,
1375 sizeof (*lr) + (write_state == WR_COPIED ? lr->lr_length : 0));
1377 if (write_state == WR_COPIED &&
1378 dmu_read(zd->zd_os, lr->lr_foid, lr->lr_offset, lr->lr_length,
1379 ((lr_write_t *)&itx->itx_lr) + 1, DMU_READ_NO_PREFETCH) != 0) {
1380 zil_itx_destroy(itx);
1381 itx = zil_itx_create(TX_WRITE, sizeof (*lr));
1382 write_state = WR_NEED_COPY;
1384 itx->itx_private = zd;
1385 itx->itx_wr_state = write_state;
1386 itx->itx_sync = (ztest_random(8) == 0);
1387 itx->itx_sod += (write_state == WR_NEED_COPY ? lr->lr_length : 0);
1389 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1390 sizeof (*lr) - sizeof (lr_t));
1392 zil_itx_assign(zd->zd_zilog, itx, tx);
1396 ztest_log_truncate(ztest_ds_t *zd, dmu_tx_t *tx, lr_truncate_t *lr)
1400 if (zil_replaying(zd->zd_zilog, tx))
1403 itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr));
1404 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1405 sizeof (*lr) - sizeof (lr_t));
1407 itx->itx_sync = B_FALSE;
1408 zil_itx_assign(zd->zd_zilog, itx, tx);
1412 ztest_log_setattr(ztest_ds_t *zd, dmu_tx_t *tx, lr_setattr_t *lr)
1416 if (zil_replaying(zd->zd_zilog, tx))
1419 itx = zil_itx_create(TX_SETATTR, sizeof (*lr));
1420 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1421 sizeof (*lr) - sizeof (lr_t));
1423 itx->itx_sync = B_FALSE;
1424 zil_itx_assign(zd->zd_zilog, itx, tx);
1431 ztest_replay_create(ztest_ds_t *zd, lr_create_t *lr, boolean_t byteswap)
1433 char *name = (void *)(lr + 1); /* name follows lr */
1434 objset_t *os = zd->zd_os;
1435 ztest_block_tag_t *bbt;
1442 byteswap_uint64_array(lr, sizeof (*lr));
1444 ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1445 ASSERT(name[0] != '\0');
1447 tx = dmu_tx_create(os);
1449 dmu_tx_hold_zap(tx, lr->lr_doid, B_TRUE, name);
1451 if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1452 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, B_TRUE, NULL);
1454 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1457 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1461 ASSERT(dmu_objset_zil(os)->zl_replay == !!lr->lr_foid);
1463 if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1464 if (lr->lr_foid == 0) {
1465 lr->lr_foid = zap_create(os,
1466 lr->lrz_type, lr->lrz_bonustype,
1467 lr->lrz_bonuslen, tx);
1469 error = zap_create_claim(os, lr->lr_foid,
1470 lr->lrz_type, lr->lrz_bonustype,
1471 lr->lrz_bonuslen, tx);
1474 if (lr->lr_foid == 0) {
1475 lr->lr_foid = dmu_object_alloc(os,
1476 lr->lrz_type, 0, lr->lrz_bonustype,
1477 lr->lrz_bonuslen, tx);
1479 error = dmu_object_claim(os, lr->lr_foid,
1480 lr->lrz_type, 0, lr->lrz_bonustype,
1481 lr->lrz_bonuslen, tx);
1486 ASSERT3U(error, ==, EEXIST);
1487 ASSERT(zd->zd_zilog->zl_replay);
1492 ASSERT(lr->lr_foid != 0);
1494 if (lr->lrz_type != DMU_OT_ZAP_OTHER)
1495 VERIFY3U(0, ==, dmu_object_set_blocksize(os, lr->lr_foid,
1496 lr->lrz_blocksize, lr->lrz_ibshift, tx));
1498 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1499 bbt = ztest_bt_bonus(db);
1500 dmu_buf_will_dirty(db, tx);
1501 ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_gen, txg, txg);
1502 dmu_buf_rele(db, FTAG);
1504 VERIFY3U(0, ==, zap_add(os, lr->lr_doid, name, sizeof (uint64_t), 1,
1507 (void) ztest_log_create(zd, tx, lr);
1515 ztest_replay_remove(ztest_ds_t *zd, lr_remove_t *lr, boolean_t byteswap)
1517 char *name = (void *)(lr + 1); /* name follows lr */
1518 objset_t *os = zd->zd_os;
1519 dmu_object_info_t doi;
1521 uint64_t object, txg;
1524 byteswap_uint64_array(lr, sizeof (*lr));
1526 ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1527 ASSERT(name[0] != '\0');
1530 zap_lookup(os, lr->lr_doid, name, sizeof (object), 1, &object));
1531 ASSERT(object != 0);
1533 ztest_object_lock(zd, object, RL_WRITER);
1535 VERIFY3U(0, ==, dmu_object_info(os, object, &doi));
1537 tx = dmu_tx_create(os);
1539 dmu_tx_hold_zap(tx, lr->lr_doid, B_FALSE, name);
1540 dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END);
1542 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1544 ztest_object_unlock(zd, object);
1548 if (doi.doi_type == DMU_OT_ZAP_OTHER) {
1549 VERIFY3U(0, ==, zap_destroy(os, object, tx));
1551 VERIFY3U(0, ==, dmu_object_free(os, object, tx));
1554 VERIFY3U(0, ==, zap_remove(os, lr->lr_doid, name, tx));
1556 (void) ztest_log_remove(zd, tx, lr, object);
1560 ztest_object_unlock(zd, object);
1566 ztest_replay_write(ztest_ds_t *zd, lr_write_t *lr, boolean_t byteswap)
1568 objset_t *os = zd->zd_os;
1569 void *data = lr + 1; /* data follows lr */
1570 uint64_t offset, length;
1571 ztest_block_tag_t *bt = data;
1572 ztest_block_tag_t *bbt;
1573 uint64_t gen, txg, lrtxg, crtxg;
1574 dmu_object_info_t doi;
1577 arc_buf_t *abuf = NULL;
1581 byteswap_uint64_array(lr, sizeof (*lr));
1583 offset = lr->lr_offset;
1584 length = lr->lr_length;
1586 /* If it's a dmu_sync() block, write the whole block */
1587 if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
1588 uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
1589 if (length < blocksize) {
1590 offset -= offset % blocksize;
1595 if (bt->bt_magic == BSWAP_64(BT_MAGIC))
1596 byteswap_uint64_array(bt, sizeof (*bt));
1598 if (bt->bt_magic != BT_MAGIC)
1601 ztest_object_lock(zd, lr->lr_foid, RL_READER);
1602 rl = ztest_range_lock(zd, lr->lr_foid, offset, length, RL_WRITER);
1604 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1606 dmu_object_info_from_db(db, &doi);
1608 bbt = ztest_bt_bonus(db);
1609 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1611 crtxg = bbt->bt_crtxg;
1612 lrtxg = lr->lr_common.lrc_txg;
1614 tx = dmu_tx_create(os);
1616 dmu_tx_hold_write(tx, lr->lr_foid, offset, length);
1618 if (ztest_random(8) == 0 && length == doi.doi_data_block_size &&
1619 P2PHASE(offset, length) == 0)
1620 abuf = dmu_request_arcbuf(db, length);
1622 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1625 dmu_return_arcbuf(abuf);
1626 dmu_buf_rele(db, FTAG);
1627 ztest_range_unlock(rl);
1628 ztest_object_unlock(zd, lr->lr_foid);
1634 * Usually, verify the old data before writing new data --
1635 * but not always, because we also want to verify correct
1636 * behavior when the data was not recently read into cache.
1638 ASSERT(offset % doi.doi_data_block_size == 0);
1639 if (ztest_random(4) != 0) {
1640 int prefetch = ztest_random(2) ?
1641 DMU_READ_PREFETCH : DMU_READ_NO_PREFETCH;
1642 ztest_block_tag_t rbt;
1644 VERIFY(dmu_read(os, lr->lr_foid, offset,
1645 sizeof (rbt), &rbt, prefetch) == 0);
1646 if (rbt.bt_magic == BT_MAGIC) {
1647 ztest_bt_verify(&rbt, os, lr->lr_foid,
1648 offset, gen, txg, crtxg);
1653 * Writes can appear to be newer than the bonus buffer because
1654 * the ztest_get_data() callback does a dmu_read() of the
1655 * open-context data, which may be different than the data
1656 * as it was when the write was generated.
1658 if (zd->zd_zilog->zl_replay) {
1659 ztest_bt_verify(bt, os, lr->lr_foid, offset,
1660 MAX(gen, bt->bt_gen), MAX(txg, lrtxg),
1665 * Set the bt's gen/txg to the bonus buffer's gen/txg
1666 * so that all of the usual ASSERTs will work.
1668 ztest_bt_generate(bt, os, lr->lr_foid, offset, gen, txg, crtxg);
1672 dmu_write(os, lr->lr_foid, offset, length, data, tx);
1674 bcopy(data, abuf->b_data, length);
1675 dmu_assign_arcbuf(db, offset, abuf, tx);
1678 (void) ztest_log_write(zd, tx, lr);
1680 dmu_buf_rele(db, FTAG);
1684 ztest_range_unlock(rl);
1685 ztest_object_unlock(zd, lr->lr_foid);
1691 ztest_replay_truncate(ztest_ds_t *zd, lr_truncate_t *lr, boolean_t byteswap)
1693 objset_t *os = zd->zd_os;
1699 byteswap_uint64_array(lr, sizeof (*lr));
1701 ztest_object_lock(zd, lr->lr_foid, RL_READER);
1702 rl = ztest_range_lock(zd, lr->lr_foid, lr->lr_offset, lr->lr_length,
1705 tx = dmu_tx_create(os);
1707 dmu_tx_hold_free(tx, lr->lr_foid, lr->lr_offset, lr->lr_length);
1709 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1711 ztest_range_unlock(rl);
1712 ztest_object_unlock(zd, lr->lr_foid);
1716 VERIFY(dmu_free_range(os, lr->lr_foid, lr->lr_offset,
1717 lr->lr_length, tx) == 0);
1719 (void) ztest_log_truncate(zd, tx, lr);
1723 ztest_range_unlock(rl);
1724 ztest_object_unlock(zd, lr->lr_foid);
1730 ztest_replay_setattr(ztest_ds_t *zd, lr_setattr_t *lr, boolean_t byteswap)
1732 objset_t *os = zd->zd_os;
1735 ztest_block_tag_t *bbt;
1736 uint64_t txg, lrtxg, crtxg;
1739 byteswap_uint64_array(lr, sizeof (*lr));
1741 ztest_object_lock(zd, lr->lr_foid, RL_WRITER);
1743 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1745 tx = dmu_tx_create(os);
1746 dmu_tx_hold_bonus(tx, lr->lr_foid);
1748 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1750 dmu_buf_rele(db, FTAG);
1751 ztest_object_unlock(zd, lr->lr_foid);
1755 bbt = ztest_bt_bonus(db);
1756 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1757 crtxg = bbt->bt_crtxg;
1758 lrtxg = lr->lr_common.lrc_txg;
1760 if (zd->zd_zilog->zl_replay) {
1761 ASSERT(lr->lr_size != 0);
1762 ASSERT(lr->lr_mode != 0);
1766 * Randomly change the size and increment the generation.
1768 lr->lr_size = (ztest_random(db->db_size / sizeof (*bbt)) + 1) *
1770 lr->lr_mode = bbt->bt_gen + 1;
1775 * Verify that the current bonus buffer is not newer than our txg.
1777 ztest_bt_verify(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode,
1778 MAX(txg, lrtxg), crtxg);
1780 dmu_buf_will_dirty(db, tx);
1782 ASSERT3U(lr->lr_size, >=, sizeof (*bbt));
1783 ASSERT3U(lr->lr_size, <=, db->db_size);
1784 VERIFY0(dmu_set_bonus(db, lr->lr_size, tx));
1785 bbt = ztest_bt_bonus(db);
1787 ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode, txg, crtxg);
1789 dmu_buf_rele(db, FTAG);
1791 (void) ztest_log_setattr(zd, tx, lr);
1795 ztest_object_unlock(zd, lr->lr_foid);
1800 zil_replay_func_t *ztest_replay_vector[TX_MAX_TYPE] = {
1801 NULL, /* 0 no such transaction type */
1802 ztest_replay_create, /* TX_CREATE */
1803 NULL, /* TX_MKDIR */
1804 NULL, /* TX_MKXATTR */
1805 NULL, /* TX_SYMLINK */
1806 ztest_replay_remove, /* TX_REMOVE */
1807 NULL, /* TX_RMDIR */
1809 NULL, /* TX_RENAME */
1810 ztest_replay_write, /* TX_WRITE */
1811 ztest_replay_truncate, /* TX_TRUNCATE */
1812 ztest_replay_setattr, /* TX_SETATTR */
1814 NULL, /* TX_CREATE_ACL */
1815 NULL, /* TX_CREATE_ATTR */
1816 NULL, /* TX_CREATE_ACL_ATTR */
1817 NULL, /* TX_MKDIR_ACL */
1818 NULL, /* TX_MKDIR_ATTR */
1819 NULL, /* TX_MKDIR_ACL_ATTR */
1820 NULL, /* TX_WRITE2 */
1824 * ZIL get_data callbacks
1828 ztest_get_done(zgd_t *zgd, int error)
1830 ztest_ds_t *zd = zgd->zgd_private;
1831 uint64_t object = zgd->zgd_rl->rl_object;
1834 dmu_buf_rele(zgd->zgd_db, zgd);
1836 ztest_range_unlock(zgd->zgd_rl);
1837 ztest_object_unlock(zd, object);
1839 if (error == 0 && zgd->zgd_bp)
1840 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
1842 umem_free(zgd, sizeof (*zgd));
1846 ztest_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
1848 ztest_ds_t *zd = arg;
1849 objset_t *os = zd->zd_os;
1850 uint64_t object = lr->lr_foid;
1851 uint64_t offset = lr->lr_offset;
1852 uint64_t size = lr->lr_length;
1853 blkptr_t *bp = &lr->lr_blkptr;
1854 uint64_t txg = lr->lr_common.lrc_txg;
1856 dmu_object_info_t doi;
1861 ztest_object_lock(zd, object, RL_READER);
1862 error = dmu_bonus_hold(os, object, FTAG, &db);
1864 ztest_object_unlock(zd, object);
1868 crtxg = ztest_bt_bonus(db)->bt_crtxg;
1870 if (crtxg == 0 || crtxg > txg) {
1871 dmu_buf_rele(db, FTAG);
1872 ztest_object_unlock(zd, object);
1876 dmu_object_info_from_db(db, &doi);
1877 dmu_buf_rele(db, FTAG);
1880 zgd = umem_zalloc(sizeof (*zgd), UMEM_NOFAIL);
1881 zgd->zgd_zilog = zd->zd_zilog;
1882 zgd->zgd_private = zd;
1884 if (buf != NULL) { /* immediate write */
1885 zgd->zgd_rl = ztest_range_lock(zd, object, offset, size,
1888 error = dmu_read(os, object, offset, size, buf,
1889 DMU_READ_NO_PREFETCH);
1892 size = doi.doi_data_block_size;
1894 offset = P2ALIGN(offset, size);
1896 ASSERT(offset < size);
1900 zgd->zgd_rl = ztest_range_lock(zd, object, offset, size,
1903 error = dmu_buf_hold(os, object, offset, zgd, &db,
1904 DMU_READ_NO_PREFETCH);
1907 blkptr_t *obp = dmu_buf_get_blkptr(db);
1909 ASSERT(BP_IS_HOLE(bp));
1916 ASSERT(db->db_offset == offset);
1917 ASSERT(db->db_size == size);
1919 error = dmu_sync(zio, lr->lr_common.lrc_txg,
1920 ztest_get_done, zgd);
1927 ztest_get_done(zgd, error);
1933 ztest_lr_alloc(size_t lrsize, char *name)
1936 size_t namesize = name ? strlen(name) + 1 : 0;
1938 lr = umem_zalloc(lrsize + namesize, UMEM_NOFAIL);
1941 bcopy(name, lr + lrsize, namesize);
1947 ztest_lr_free(void *lr, size_t lrsize, char *name)
1949 size_t namesize = name ? strlen(name) + 1 : 0;
1951 umem_free(lr, lrsize + namesize);
1955 * Lookup a bunch of objects. Returns the number of objects not found.
1958 ztest_lookup(ztest_ds_t *zd, ztest_od_t *od, int count)
1963 ASSERT(_mutex_held(&zd->zd_dirobj_lock));
1965 for (int i = 0; i < count; i++, od++) {
1967 error = zap_lookup(zd->zd_os, od->od_dir, od->od_name,
1968 sizeof (uint64_t), 1, &od->od_object);
1970 ASSERT(error == ENOENT);
1971 ASSERT(od->od_object == 0);
1975 ztest_block_tag_t *bbt;
1976 dmu_object_info_t doi;
1978 ASSERT(od->od_object != 0);
1979 ASSERT(missing == 0); /* there should be no gaps */
1981 ztest_object_lock(zd, od->od_object, RL_READER);
1982 VERIFY3U(0, ==, dmu_bonus_hold(zd->zd_os,
1983 od->od_object, FTAG, &db));
1984 dmu_object_info_from_db(db, &doi);
1985 bbt = ztest_bt_bonus(db);
1986 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1987 od->od_type = doi.doi_type;
1988 od->od_blocksize = doi.doi_data_block_size;
1989 od->od_gen = bbt->bt_gen;
1990 dmu_buf_rele(db, FTAG);
1991 ztest_object_unlock(zd, od->od_object);
1999 ztest_create(ztest_ds_t *zd, ztest_od_t *od, int count)
2003 ASSERT(_mutex_held(&zd->zd_dirobj_lock));
2005 for (int i = 0; i < count; i++, od++) {
2012 lr_create_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
2014 lr->lr_doid = od->od_dir;
2015 lr->lr_foid = 0; /* 0 to allocate, > 0 to claim */
2016 lr->lrz_type = od->od_crtype;
2017 lr->lrz_blocksize = od->od_crblocksize;
2018 lr->lrz_ibshift = ztest_random_ibshift();
2019 lr->lrz_bonustype = DMU_OT_UINT64_OTHER;
2020 lr->lrz_bonuslen = dmu_bonus_max();
2021 lr->lr_gen = od->od_crgen;
2022 lr->lr_crtime[0] = time(NULL);
2024 if (ztest_replay_create(zd, lr, B_FALSE) != 0) {
2025 ASSERT(missing == 0);
2029 od->od_object = lr->lr_foid;
2030 od->od_type = od->od_crtype;
2031 od->od_blocksize = od->od_crblocksize;
2032 od->od_gen = od->od_crgen;
2033 ASSERT(od->od_object != 0);
2036 ztest_lr_free(lr, sizeof (*lr), od->od_name);
2043 ztest_remove(ztest_ds_t *zd, ztest_od_t *od, int count)
2048 ASSERT(_mutex_held(&zd->zd_dirobj_lock));
2052 for (int i = count - 1; i >= 0; i--, od--) {
2059 * No object was found.
2061 if (od->od_object == 0)
2064 lr_remove_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
2066 lr->lr_doid = od->od_dir;
2068 if ((error = ztest_replay_remove(zd, lr, B_FALSE)) != 0) {
2069 ASSERT3U(error, ==, ENOSPC);
2074 ztest_lr_free(lr, sizeof (*lr), od->od_name);
2081 ztest_write(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size,
2087 lr = ztest_lr_alloc(sizeof (*lr) + size, NULL);
2089 lr->lr_foid = object;
2090 lr->lr_offset = offset;
2091 lr->lr_length = size;
2093 BP_ZERO(&lr->lr_blkptr);
2095 bcopy(data, lr + 1, size);
2097 error = ztest_replay_write(zd, lr, B_FALSE);
2099 ztest_lr_free(lr, sizeof (*lr) + size, NULL);
2105 ztest_truncate(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2110 lr = ztest_lr_alloc(sizeof (*lr), NULL);
2112 lr->lr_foid = object;
2113 lr->lr_offset = offset;
2114 lr->lr_length = size;
2116 error = ztest_replay_truncate(zd, lr, B_FALSE);
2118 ztest_lr_free(lr, sizeof (*lr), NULL);
2124 ztest_setattr(ztest_ds_t *zd, uint64_t object)
2129 lr = ztest_lr_alloc(sizeof (*lr), NULL);
2131 lr->lr_foid = object;
2135 error = ztest_replay_setattr(zd, lr, B_FALSE);
2137 ztest_lr_free(lr, sizeof (*lr), NULL);
2143 ztest_prealloc(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2145 objset_t *os = zd->zd_os;
2150 txg_wait_synced(dmu_objset_pool(os), 0);
2152 ztest_object_lock(zd, object, RL_READER);
2153 rl = ztest_range_lock(zd, object, offset, size, RL_WRITER);
2155 tx = dmu_tx_create(os);
2157 dmu_tx_hold_write(tx, object, offset, size);
2159 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
2162 dmu_prealloc(os, object, offset, size, tx);
2164 txg_wait_synced(dmu_objset_pool(os), txg);
2166 (void) dmu_free_long_range(os, object, offset, size);
2169 ztest_range_unlock(rl);
2170 ztest_object_unlock(zd, object);
2174 ztest_io(ztest_ds_t *zd, uint64_t object, uint64_t offset)
2177 ztest_block_tag_t wbt;
2178 dmu_object_info_t doi;
2179 enum ztest_io_type io_type;
2183 VERIFY(dmu_object_info(zd->zd_os, object, &doi) == 0);
2184 blocksize = doi.doi_data_block_size;
2185 data = umem_alloc(blocksize, UMEM_NOFAIL);
2188 * Pick an i/o type at random, biased toward writing block tags.
2190 io_type = ztest_random(ZTEST_IO_TYPES);
2191 if (ztest_random(2) == 0)
2192 io_type = ZTEST_IO_WRITE_TAG;
2194 (void) rw_rdlock(&zd->zd_zilog_lock);
2198 case ZTEST_IO_WRITE_TAG:
2199 ztest_bt_generate(&wbt, zd->zd_os, object, offset, 0, 0, 0);
2200 (void) ztest_write(zd, object, offset, sizeof (wbt), &wbt);
2203 case ZTEST_IO_WRITE_PATTERN:
2204 (void) memset(data, 'a' + (object + offset) % 5, blocksize);
2205 if (ztest_random(2) == 0) {
2207 * Induce fletcher2 collisions to ensure that
2208 * zio_ddt_collision() detects and resolves them
2209 * when using fletcher2-verify for deduplication.
2211 ((uint64_t *)data)[0] ^= 1ULL << 63;
2212 ((uint64_t *)data)[4] ^= 1ULL << 63;
2214 (void) ztest_write(zd, object, offset, blocksize, data);
2217 case ZTEST_IO_WRITE_ZEROES:
2218 bzero(data, blocksize);
2219 (void) ztest_write(zd, object, offset, blocksize, data);
2222 case ZTEST_IO_TRUNCATE:
2223 (void) ztest_truncate(zd, object, offset, blocksize);
2226 case ZTEST_IO_SETATTR:
2227 (void) ztest_setattr(zd, object);
2230 case ZTEST_IO_REWRITE:
2231 (void) rw_rdlock(&ztest_name_lock);
2232 err = ztest_dsl_prop_set_uint64(zd->zd_name,
2233 ZFS_PROP_CHECKSUM, spa_dedup_checksum(ztest_spa),
2235 VERIFY(err == 0 || err == ENOSPC);
2236 err = ztest_dsl_prop_set_uint64(zd->zd_name,
2237 ZFS_PROP_COMPRESSION,
2238 ztest_random_dsl_prop(ZFS_PROP_COMPRESSION),
2240 VERIFY(err == 0 || err == ENOSPC);
2241 (void) rw_unlock(&ztest_name_lock);
2243 VERIFY0(dmu_read(zd->zd_os, object, offset, blocksize, data,
2244 DMU_READ_NO_PREFETCH));
2246 (void) ztest_write(zd, object, offset, blocksize, data);
2250 (void) rw_unlock(&zd->zd_zilog_lock);
2252 umem_free(data, blocksize);
2256 * Initialize an object description template.
2259 ztest_od_init(ztest_od_t *od, uint64_t id, char *tag, uint64_t index,
2260 dmu_object_type_t type, uint64_t blocksize, uint64_t gen)
2262 od->od_dir = ZTEST_DIROBJ;
2265 od->od_crtype = type;
2266 od->od_crblocksize = blocksize ? blocksize : ztest_random_blocksize();
2269 od->od_type = DMU_OT_NONE;
2270 od->od_blocksize = 0;
2273 (void) snprintf(od->od_name, sizeof (od->od_name), "%s(%lld)[%llu]",
2274 tag, (int64_t)id, index);
2278 * Lookup or create the objects for a test using the od template.
2279 * If the objects do not all exist, or if 'remove' is specified,
2280 * remove any existing objects and create new ones. Otherwise,
2281 * use the existing objects.
2284 ztest_object_init(ztest_ds_t *zd, ztest_od_t *od, size_t size, boolean_t remove)
2286 int count = size / sizeof (*od);
2289 VERIFY(mutex_lock(&zd->zd_dirobj_lock) == 0);
2290 if ((ztest_lookup(zd, od, count) != 0 || remove) &&
2291 (ztest_remove(zd, od, count) != 0 ||
2292 ztest_create(zd, od, count) != 0))
2295 VERIFY(mutex_unlock(&zd->zd_dirobj_lock) == 0);
2302 ztest_zil_commit(ztest_ds_t *zd, uint64_t id)
2304 zilog_t *zilog = zd->zd_zilog;
2306 (void) rw_rdlock(&zd->zd_zilog_lock);
2308 zil_commit(zilog, ztest_random(ZTEST_OBJECTS));
2311 * Remember the committed values in zd, which is in parent/child
2312 * shared memory. If we die, the next iteration of ztest_run()
2313 * will verify that the log really does contain this record.
2315 mutex_enter(&zilog->zl_lock);
2316 ASSERT(zd->zd_shared != NULL);
2317 ASSERT3U(zd->zd_shared->zd_seq, <=, zilog->zl_commit_lr_seq);
2318 zd->zd_shared->zd_seq = zilog->zl_commit_lr_seq;
2319 mutex_exit(&zilog->zl_lock);
2321 (void) rw_unlock(&zd->zd_zilog_lock);
2325 * This function is designed to simulate the operations that occur during a
2326 * mount/unmount operation. We hold the dataset across these operations in an
2327 * attempt to expose any implicit assumptions about ZIL management.
2331 ztest_zil_remount(ztest_ds_t *zd, uint64_t id)
2333 objset_t *os = zd->zd_os;
2336 * We grab the zd_dirobj_lock to ensure that no other thread is
2337 * updating the zil (i.e. adding in-memory log records) and the
2338 * zd_zilog_lock to block any I/O.
2340 VERIFY0(mutex_lock(&zd->zd_dirobj_lock));
2341 (void) rw_wrlock(&zd->zd_zilog_lock);
2343 /* zfsvfs_teardown() */
2344 zil_close(zd->zd_zilog);
2346 /* zfsvfs_setup() */
2347 VERIFY(zil_open(os, ztest_get_data) == zd->zd_zilog);
2348 zil_replay(os, zd, ztest_replay_vector);
2350 (void) rw_unlock(&zd->zd_zilog_lock);
2351 VERIFY(mutex_unlock(&zd->zd_dirobj_lock) == 0);
2355 * Verify that we can't destroy an active pool, create an existing pool,
2356 * or create a pool with a bad vdev spec.
2360 ztest_spa_create_destroy(ztest_ds_t *zd, uint64_t id)
2362 ztest_shared_opts_t *zo = &ztest_opts;
2367 * Attempt to create using a bad file.
2369 nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1);
2370 VERIFY3U(ENOENT, ==,
2371 spa_create("ztest_bad_file", nvroot, NULL, NULL));
2372 nvlist_free(nvroot);
2375 * Attempt to create using a bad mirror.
2377 nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 2, 1);
2378 VERIFY3U(ENOENT, ==,
2379 spa_create("ztest_bad_mirror", nvroot, NULL, NULL));
2380 nvlist_free(nvroot);
2383 * Attempt to create an existing pool. It shouldn't matter
2384 * what's in the nvroot; we should fail with EEXIST.
2386 (void) rw_rdlock(&ztest_name_lock);
2387 nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1);
2388 VERIFY3U(EEXIST, ==, spa_create(zo->zo_pool, nvroot, NULL, NULL));
2389 nvlist_free(nvroot);
2390 VERIFY3U(0, ==, spa_open(zo->zo_pool, &spa, FTAG));
2391 VERIFY3U(EBUSY, ==, spa_destroy(zo->zo_pool));
2392 spa_close(spa, FTAG);
2394 (void) rw_unlock(&ztest_name_lock);
2399 ztest_spa_upgrade(ztest_ds_t *zd, uint64_t id)
2402 uint64_t initial_version = SPA_VERSION_INITIAL;
2403 uint64_t version, newversion;
2404 nvlist_t *nvroot, *props;
2407 VERIFY0(mutex_lock(&ztest_vdev_lock));
2408 name = kmem_asprintf("%s_upgrade", ztest_opts.zo_pool);
2411 * Clean up from previous runs.
2413 (void) spa_destroy(name);
2415 nvroot = make_vdev_root(NULL, NULL, name, ztest_opts.zo_vdev_size, 0,
2416 0, ztest_opts.zo_raidz, ztest_opts.zo_mirrors, 1);
2419 * If we're configuring a RAIDZ device then make sure that the
2420 * the initial version is capable of supporting that feature.
2422 switch (ztest_opts.zo_raidz_parity) {
2425 initial_version = SPA_VERSION_INITIAL;
2428 initial_version = SPA_VERSION_RAIDZ2;
2431 initial_version = SPA_VERSION_RAIDZ3;
2436 * Create a pool with a spa version that can be upgraded. Pick
2437 * a value between initial_version and SPA_VERSION_BEFORE_FEATURES.
2440 version = ztest_random_spa_version(initial_version);
2441 } while (version > SPA_VERSION_BEFORE_FEATURES);
2443 props = fnvlist_alloc();
2444 fnvlist_add_uint64(props,
2445 zpool_prop_to_name(ZPOOL_PROP_VERSION), version);
2446 VERIFY0(spa_create(name, nvroot, props, NULL));
2447 fnvlist_free(nvroot);
2448 fnvlist_free(props);
2450 VERIFY0(spa_open(name, &spa, FTAG));
2451 VERIFY3U(spa_version(spa), ==, version);
2452 newversion = ztest_random_spa_version(version + 1);
2454 if (ztest_opts.zo_verbose >= 4) {
2455 (void) printf("upgrading spa version from %llu to %llu\n",
2456 (u_longlong_t)version, (u_longlong_t)newversion);
2459 spa_upgrade(spa, newversion);
2460 VERIFY3U(spa_version(spa), >, version);
2461 VERIFY3U(spa_version(spa), ==, fnvlist_lookup_uint64(spa->spa_config,
2462 zpool_prop_to_name(ZPOOL_PROP_VERSION)));
2463 spa_close(spa, FTAG);
2466 VERIFY0(mutex_unlock(&ztest_vdev_lock));
2470 vdev_lookup_by_path(vdev_t *vd, const char *path)
2474 if (vd->vdev_path != NULL && strcmp(path, vd->vdev_path) == 0)
2477 for (int c = 0; c < vd->vdev_children; c++)
2478 if ((mvd = vdev_lookup_by_path(vd->vdev_child[c], path)) !=
2486 * Find the first available hole which can be used as a top-level.
2489 find_vdev_hole(spa_t *spa)
2491 vdev_t *rvd = spa->spa_root_vdev;
2494 ASSERT(spa_config_held(spa, SCL_VDEV, RW_READER) == SCL_VDEV);
2496 for (c = 0; c < rvd->vdev_children; c++) {
2497 vdev_t *cvd = rvd->vdev_child[c];
2499 if (cvd->vdev_ishole)
2506 * Verify that vdev_add() works as expected.
2510 ztest_vdev_add_remove(ztest_ds_t *zd, uint64_t id)
2512 ztest_shared_t *zs = ztest_shared;
2513 spa_t *spa = ztest_spa;
2519 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2520 leaves = MAX(zs->zs_mirrors + zs->zs_splits, 1) * ztest_opts.zo_raidz;
2522 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2524 ztest_shared->zs_vdev_next_leaf = find_vdev_hole(spa) * leaves;
2527 * If we have slogs then remove them 1/4 of the time.
2529 if (spa_has_slogs(spa) && ztest_random(4) == 0) {
2531 * Grab the guid from the head of the log class rotor.
2533 guid = spa_log_class(spa)->mc_rotor->mg_vd->vdev_guid;
2535 spa_config_exit(spa, SCL_VDEV, FTAG);
2538 * We have to grab the zs_name_lock as writer to
2539 * prevent a race between removing a slog (dmu_objset_find)
2540 * and destroying a dataset. Removing the slog will
2541 * grab a reference on the dataset which may cause
2542 * dmu_objset_destroy() to fail with EBUSY thus
2543 * leaving the dataset in an inconsistent state.
2545 VERIFY(rw_wrlock(&ztest_name_lock) == 0);
2546 error = spa_vdev_remove(spa, guid, B_FALSE);
2547 VERIFY(rw_unlock(&ztest_name_lock) == 0);
2549 if (error && error != EEXIST)
2550 fatal(0, "spa_vdev_remove() = %d", error);
2552 spa_config_exit(spa, SCL_VDEV, FTAG);
2555 * Make 1/4 of the devices be log devices.
2557 nvroot = make_vdev_root(NULL, NULL, NULL,
2558 ztest_opts.zo_vdev_size, 0,
2559 ztest_random(4) == 0, ztest_opts.zo_raidz,
2562 error = spa_vdev_add(spa, nvroot);
2563 nvlist_free(nvroot);
2565 if (error == ENOSPC)
2566 ztest_record_enospc("spa_vdev_add");
2567 else if (error != 0)
2568 fatal(0, "spa_vdev_add() = %d", error);
2571 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2575 * Verify that adding/removing aux devices (l2arc, hot spare) works as expected.
2579 ztest_vdev_aux_add_remove(ztest_ds_t *zd, uint64_t id)
2581 ztest_shared_t *zs = ztest_shared;
2582 spa_t *spa = ztest_spa;
2583 vdev_t *rvd = spa->spa_root_vdev;
2584 spa_aux_vdev_t *sav;
2589 if (ztest_random(2) == 0) {
2590 sav = &spa->spa_spares;
2591 aux = ZPOOL_CONFIG_SPARES;
2593 sav = &spa->spa_l2cache;
2594 aux = ZPOOL_CONFIG_L2CACHE;
2597 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2599 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2601 if (sav->sav_count != 0 && ztest_random(4) == 0) {
2603 * Pick a random device to remove.
2605 guid = sav->sav_vdevs[ztest_random(sav->sav_count)]->vdev_guid;
2608 * Find an unused device we can add.
2610 zs->zs_vdev_aux = 0;
2612 char path[MAXPATHLEN];
2614 (void) snprintf(path, sizeof (path), ztest_aux_template,
2615 ztest_opts.zo_dir, ztest_opts.zo_pool, aux,
2617 for (c = 0; c < sav->sav_count; c++)
2618 if (strcmp(sav->sav_vdevs[c]->vdev_path,
2621 if (c == sav->sav_count &&
2622 vdev_lookup_by_path(rvd, path) == NULL)
2628 spa_config_exit(spa, SCL_VDEV, FTAG);
2634 nvlist_t *nvroot = make_vdev_root(NULL, aux, NULL,
2635 (ztest_opts.zo_vdev_size * 5) / 4, 0, 0, 0, 0, 1);
2636 error = spa_vdev_add(spa, nvroot);
2638 fatal(0, "spa_vdev_add(%p) = %d", nvroot, error);
2639 nvlist_free(nvroot);
2642 * Remove an existing device. Sometimes, dirty its
2643 * vdev state first to make sure we handle removal
2644 * of devices that have pending state changes.
2646 if (ztest_random(2) == 0)
2647 (void) vdev_online(spa, guid, 0, NULL);
2649 error = spa_vdev_remove(spa, guid, B_FALSE);
2650 if (error != 0 && error != EBUSY)
2651 fatal(0, "spa_vdev_remove(%llu) = %d", guid, error);
2654 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2658 * split a pool if it has mirror tlvdevs
2662 ztest_split_pool(ztest_ds_t *zd, uint64_t id)
2664 ztest_shared_t *zs = ztest_shared;
2665 spa_t *spa = ztest_spa;
2666 vdev_t *rvd = spa->spa_root_vdev;
2667 nvlist_t *tree, **child, *config, *split, **schild;
2668 uint_t c, children, schildren = 0, lastlogid = 0;
2671 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2673 /* ensure we have a useable config; mirrors of raidz aren't supported */
2674 if (zs->zs_mirrors < 3 || ztest_opts.zo_raidz > 1) {
2675 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2679 /* clean up the old pool, if any */
2680 (void) spa_destroy("splitp");
2682 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2684 /* generate a config from the existing config */
2685 mutex_enter(&spa->spa_props_lock);
2686 VERIFY(nvlist_lookup_nvlist(spa->spa_config, ZPOOL_CONFIG_VDEV_TREE,
2688 mutex_exit(&spa->spa_props_lock);
2690 VERIFY(nvlist_lookup_nvlist_array(tree, ZPOOL_CONFIG_CHILDREN, &child,
2693 schild = malloc(rvd->vdev_children * sizeof (nvlist_t *));
2694 for (c = 0; c < children; c++) {
2695 vdev_t *tvd = rvd->vdev_child[c];
2699 if (tvd->vdev_islog || tvd->vdev_ops == &vdev_hole_ops) {
2700 VERIFY(nvlist_alloc(&schild[schildren], NV_UNIQUE_NAME,
2702 VERIFY(nvlist_add_string(schild[schildren],
2703 ZPOOL_CONFIG_TYPE, VDEV_TYPE_HOLE) == 0);
2704 VERIFY(nvlist_add_uint64(schild[schildren],
2705 ZPOOL_CONFIG_IS_HOLE, 1) == 0);
2707 lastlogid = schildren;
2712 VERIFY(nvlist_lookup_nvlist_array(child[c],
2713 ZPOOL_CONFIG_CHILDREN, &mchild, &mchildren) == 0);
2714 VERIFY(nvlist_dup(mchild[0], &schild[schildren++], 0) == 0);
2717 /* OK, create a config that can be used to split */
2718 VERIFY(nvlist_alloc(&split, NV_UNIQUE_NAME, 0) == 0);
2719 VERIFY(nvlist_add_string(split, ZPOOL_CONFIG_TYPE,
2720 VDEV_TYPE_ROOT) == 0);
2721 VERIFY(nvlist_add_nvlist_array(split, ZPOOL_CONFIG_CHILDREN, schild,
2722 lastlogid != 0 ? lastlogid : schildren) == 0);
2724 VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, 0) == 0);
2725 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, split) == 0);
2727 for (c = 0; c < schildren; c++)
2728 nvlist_free(schild[c]);
2732 spa_config_exit(spa, SCL_VDEV, FTAG);
2734 (void) rw_wrlock(&ztest_name_lock);
2735 error = spa_vdev_split_mirror(spa, "splitp", config, NULL, B_FALSE);
2736 (void) rw_unlock(&ztest_name_lock);
2738 nvlist_free(config);
2741 (void) printf("successful split - results:\n");
2742 mutex_enter(&spa_namespace_lock);
2743 show_pool_stats(spa);
2744 show_pool_stats(spa_lookup("splitp"));
2745 mutex_exit(&spa_namespace_lock);
2749 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2754 * Verify that we can attach and detach devices.
2758 ztest_vdev_attach_detach(ztest_ds_t *zd, uint64_t id)
2760 ztest_shared_t *zs = ztest_shared;
2761 spa_t *spa = ztest_spa;
2762 spa_aux_vdev_t *sav = &spa->spa_spares;
2763 vdev_t *rvd = spa->spa_root_vdev;
2764 vdev_t *oldvd, *newvd, *pvd;
2768 uint64_t ashift = ztest_get_ashift();
2769 uint64_t oldguid, pguid;
2770 uint64_t oldsize, newsize;
2771 char oldpath[MAXPATHLEN], newpath[MAXPATHLEN];
2773 int oldvd_has_siblings = B_FALSE;
2774 int newvd_is_spare = B_FALSE;
2776 int error, expected_error;
2778 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2779 leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz;
2781 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2784 * Decide whether to do an attach or a replace.
2786 replacing = ztest_random(2);
2789 * Pick a random top-level vdev.
2791 top = ztest_random_vdev_top(spa, B_TRUE);
2794 * Pick a random leaf within it.
2796 leaf = ztest_random(leaves);
2801 oldvd = rvd->vdev_child[top];
2802 if (zs->zs_mirrors >= 1) {
2803 ASSERT(oldvd->vdev_ops == &vdev_mirror_ops);
2804 ASSERT(oldvd->vdev_children >= zs->zs_mirrors);
2805 oldvd = oldvd->vdev_child[leaf / ztest_opts.zo_raidz];
2807 if (ztest_opts.zo_raidz > 1) {
2808 ASSERT(oldvd->vdev_ops == &vdev_raidz_ops);
2809 ASSERT(oldvd->vdev_children == ztest_opts.zo_raidz);
2810 oldvd = oldvd->vdev_child[leaf % ztest_opts.zo_raidz];
2814 * If we're already doing an attach or replace, oldvd may be a
2815 * mirror vdev -- in which case, pick a random child.
2817 while (oldvd->vdev_children != 0) {
2818 oldvd_has_siblings = B_TRUE;
2819 ASSERT(oldvd->vdev_children >= 2);
2820 oldvd = oldvd->vdev_child[ztest_random(oldvd->vdev_children)];
2823 oldguid = oldvd->vdev_guid;
2824 oldsize = vdev_get_min_asize(oldvd);
2825 oldvd_is_log = oldvd->vdev_top->vdev_islog;
2826 (void) strcpy(oldpath, oldvd->vdev_path);
2827 pvd = oldvd->vdev_parent;
2828 pguid = pvd->vdev_guid;
2831 * If oldvd has siblings, then half of the time, detach it.
2833 if (oldvd_has_siblings && ztest_random(2) == 0) {
2834 spa_config_exit(spa, SCL_VDEV, FTAG);
2835 error = spa_vdev_detach(spa, oldguid, pguid, B_FALSE);
2836 if (error != 0 && error != ENODEV && error != EBUSY &&
2838 fatal(0, "detach (%s) returned %d", oldpath, error);
2839 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2844 * For the new vdev, choose with equal probability between the two
2845 * standard paths (ending in either 'a' or 'b') or a random hot spare.
2847 if (sav->sav_count != 0 && ztest_random(3) == 0) {
2848 newvd = sav->sav_vdevs[ztest_random(sav->sav_count)];
2849 newvd_is_spare = B_TRUE;
2850 (void) strcpy(newpath, newvd->vdev_path);
2852 (void) snprintf(newpath, sizeof (newpath), ztest_dev_template,
2853 ztest_opts.zo_dir, ztest_opts.zo_pool,
2854 top * leaves + leaf);
2855 if (ztest_random(2) == 0)
2856 newpath[strlen(newpath) - 1] = 'b';
2857 newvd = vdev_lookup_by_path(rvd, newpath);
2861 newsize = vdev_get_min_asize(newvd);
2864 * Make newsize a little bigger or smaller than oldsize.
2865 * If it's smaller, the attach should fail.
2866 * If it's larger, and we're doing a replace,
2867 * we should get dynamic LUN growth when we're done.
2869 newsize = 10 * oldsize / (9 + ztest_random(3));
2873 * If pvd is not a mirror or root, the attach should fail with ENOTSUP,
2874 * unless it's a replace; in that case any non-replacing parent is OK.
2876 * If newvd is already part of the pool, it should fail with EBUSY.
2878 * If newvd is too small, it should fail with EOVERFLOW.
2880 if (pvd->vdev_ops != &vdev_mirror_ops &&
2881 pvd->vdev_ops != &vdev_root_ops && (!replacing ||
2882 pvd->vdev_ops == &vdev_replacing_ops ||
2883 pvd->vdev_ops == &vdev_spare_ops))
2884 expected_error = ENOTSUP;
2885 else if (newvd_is_spare && (!replacing || oldvd_is_log))
2886 expected_error = ENOTSUP;
2887 else if (newvd == oldvd)
2888 expected_error = replacing ? 0 : EBUSY;
2889 else if (vdev_lookup_by_path(rvd, newpath) != NULL)
2890 expected_error = EBUSY;
2891 else if (newsize < oldsize)
2892 expected_error = EOVERFLOW;
2893 else if (ashift > oldvd->vdev_top->vdev_ashift)
2894 expected_error = EDOM;
2898 spa_config_exit(spa, SCL_VDEV, FTAG);
2901 * Build the nvlist describing newpath.
2903 root = make_vdev_root(newpath, NULL, NULL, newvd == NULL ? newsize : 0,
2904 ashift, 0, 0, 0, 1);
2906 error = spa_vdev_attach(spa, oldguid, root, replacing);
2911 * If our parent was the replacing vdev, but the replace completed,
2912 * then instead of failing with ENOTSUP we may either succeed,
2913 * fail with ENODEV, or fail with EOVERFLOW.
2915 if (expected_error == ENOTSUP &&
2916 (error == 0 || error == ENODEV || error == EOVERFLOW))
2917 expected_error = error;
2920 * If someone grew the LUN, the replacement may be too small.
2922 if (error == EOVERFLOW || error == EBUSY)
2923 expected_error = error;
2925 /* XXX workaround 6690467 */
2926 if (error != expected_error && expected_error != EBUSY) {
2927 fatal(0, "attach (%s %llu, %s %llu, %d) "
2928 "returned %d, expected %d",
2929 oldpath, oldsize, newpath,
2930 newsize, replacing, error, expected_error);
2933 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2937 * Callback function which expands the physical size of the vdev.
2940 grow_vdev(vdev_t *vd, void *arg)
2942 spa_t *spa = vd->vdev_spa;
2943 size_t *newsize = arg;
2947 ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
2948 ASSERT(vd->vdev_ops->vdev_op_leaf);
2950 if ((fd = open(vd->vdev_path, O_RDWR)) == -1)
2953 fsize = lseek(fd, 0, SEEK_END);
2954 (void) ftruncate(fd, *newsize);
2956 if (ztest_opts.zo_verbose >= 6) {
2957 (void) printf("%s grew from %lu to %lu bytes\n",
2958 vd->vdev_path, (ulong_t)fsize, (ulong_t)*newsize);
2965 * Callback function which expands a given vdev by calling vdev_online().
2969 online_vdev(vdev_t *vd, void *arg)
2971 spa_t *spa = vd->vdev_spa;
2972 vdev_t *tvd = vd->vdev_top;
2973 uint64_t guid = vd->vdev_guid;
2974 uint64_t generation = spa->spa_config_generation + 1;
2975 vdev_state_t newstate = VDEV_STATE_UNKNOWN;
2978 ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
2979 ASSERT(vd->vdev_ops->vdev_op_leaf);
2981 /* Calling vdev_online will initialize the new metaslabs */
2982 spa_config_exit(spa, SCL_STATE, spa);
2983 error = vdev_online(spa, guid, ZFS_ONLINE_EXPAND, &newstate);
2984 spa_config_enter(spa, SCL_STATE, spa, RW_READER);
2987 * If vdev_online returned an error or the underlying vdev_open
2988 * failed then we abort the expand. The only way to know that
2989 * vdev_open fails is by checking the returned newstate.
2991 if (error || newstate != VDEV_STATE_HEALTHY) {
2992 if (ztest_opts.zo_verbose >= 5) {
2993 (void) printf("Unable to expand vdev, state %llu, "
2994 "error %d\n", (u_longlong_t)newstate, error);
2998 ASSERT3U(newstate, ==, VDEV_STATE_HEALTHY);
3001 * Since we dropped the lock we need to ensure that we're
3002 * still talking to the original vdev. It's possible this
3003 * vdev may have been detached/replaced while we were
3004 * trying to online it.
3006 if (generation != spa->spa_config_generation) {
3007 if (ztest_opts.zo_verbose >= 5) {
3008 (void) printf("vdev configuration has changed, "
3009 "guid %llu, state %llu, expected gen %llu, "
3012 (u_longlong_t)tvd->vdev_state,
3013 (u_longlong_t)generation,
3014 (u_longlong_t)spa->spa_config_generation);
3022 * Traverse the vdev tree calling the supplied function.
3023 * We continue to walk the tree until we either have walked all
3024 * children or we receive a non-NULL return from the callback.
3025 * If a NULL callback is passed, then we just return back the first
3026 * leaf vdev we encounter.
3029 vdev_walk_tree(vdev_t *vd, vdev_t *(*func)(vdev_t *, void *), void *arg)
3031 if (vd->vdev_ops->vdev_op_leaf) {
3035 return (func(vd, arg));
3038 for (uint_t c = 0; c < vd->vdev_children; c++) {
3039 vdev_t *cvd = vd->vdev_child[c];
3040 if ((cvd = vdev_walk_tree(cvd, func, arg)) != NULL)
3047 * Verify that dynamic LUN growth works as expected.
3051 ztest_vdev_LUN_growth(ztest_ds_t *zd, uint64_t id)
3053 spa_t *spa = ztest_spa;
3055 metaslab_class_t *mc;
3056 metaslab_group_t *mg;
3057 size_t psize, newsize;
3059 uint64_t old_class_space, new_class_space, old_ms_count, new_ms_count;
3061 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
3062 spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3064 top = ztest_random_vdev_top(spa, B_TRUE);
3066 tvd = spa->spa_root_vdev->vdev_child[top];
3069 old_ms_count = tvd->vdev_ms_count;
3070 old_class_space = metaslab_class_get_space(mc);
3073 * Determine the size of the first leaf vdev associated with
3074 * our top-level device.
3076 vd = vdev_walk_tree(tvd, NULL, NULL);
3077 ASSERT3P(vd, !=, NULL);
3078 ASSERT(vd->vdev_ops->vdev_op_leaf);
3080 psize = vd->vdev_psize;
3083 * We only try to expand the vdev if it's healthy, less than 4x its
3084 * original size, and it has a valid psize.
3086 if (tvd->vdev_state != VDEV_STATE_HEALTHY ||
3087 psize == 0 || psize >= 4 * ztest_opts.zo_vdev_size) {
3088 spa_config_exit(spa, SCL_STATE, spa);
3089 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3093 newsize = psize + psize / 8;
3094 ASSERT3U(newsize, >, psize);
3096 if (ztest_opts.zo_verbose >= 6) {
3097 (void) printf("Expanding LUN %s from %lu to %lu\n",
3098 vd->vdev_path, (ulong_t)psize, (ulong_t)newsize);
3102 * Growing the vdev is a two step process:
3103 * 1). expand the physical size (i.e. relabel)
3104 * 2). online the vdev to create the new metaslabs
3106 if (vdev_walk_tree(tvd, grow_vdev, &newsize) != NULL ||
3107 vdev_walk_tree(tvd, online_vdev, NULL) != NULL ||
3108 tvd->vdev_state != VDEV_STATE_HEALTHY) {
3109 if (ztest_opts.zo_verbose >= 5) {
3110 (void) printf("Could not expand LUN because "
3111 "the vdev configuration changed.\n");
3113 spa_config_exit(spa, SCL_STATE, spa);
3114 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3118 spa_config_exit(spa, SCL_STATE, spa);
3121 * Expanding the LUN will update the config asynchronously,
3122 * thus we must wait for the async thread to complete any
3123 * pending tasks before proceeding.
3127 mutex_enter(&spa->spa_async_lock);
3128 done = (spa->spa_async_thread == NULL && !spa->spa_async_tasks);
3129 mutex_exit(&spa->spa_async_lock);
3132 txg_wait_synced(spa_get_dsl(spa), 0);
3133 (void) poll(NULL, 0, 100);
3136 spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3138 tvd = spa->spa_root_vdev->vdev_child[top];
3139 new_ms_count = tvd->vdev_ms_count;
3140 new_class_space = metaslab_class_get_space(mc);
3142 if (tvd->vdev_mg != mg || mg->mg_class != mc) {
3143 if (ztest_opts.zo_verbose >= 5) {
3144 (void) printf("Could not verify LUN expansion due to "
3145 "intervening vdev offline or remove.\n");
3147 spa_config_exit(spa, SCL_STATE, spa);
3148 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3153 * Make sure we were able to grow the vdev.
3155 if (new_ms_count <= old_ms_count)
3156 fatal(0, "LUN expansion failed: ms_count %llu <= %llu\n",
3157 old_ms_count, new_ms_count);
3160 * Make sure we were able to grow the pool.
3162 if (new_class_space <= old_class_space)
3163 fatal(0, "LUN expansion failed: class_space %llu <= %llu\n",
3164 old_class_space, new_class_space);
3166 if (ztest_opts.zo_verbose >= 5) {
3167 char oldnumbuf[6], newnumbuf[6];
3169 nicenum(old_class_space, oldnumbuf);
3170 nicenum(new_class_space, newnumbuf);
3171 (void) printf("%s grew from %s to %s\n",
3172 spa->spa_name, oldnumbuf, newnumbuf);
3175 spa_config_exit(spa, SCL_STATE, spa);
3176 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3180 * Verify that dmu_objset_{create,destroy,open,close} work as expected.
3184 ztest_objset_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
3187 * Create the objects common to all ztest datasets.
3189 VERIFY(zap_create_claim(os, ZTEST_DIROBJ,
3190 DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx) == 0);
3194 ztest_dataset_create(char *dsname)
3196 uint64_t zilset = ztest_random(100);
3197 int err = dmu_objset_create(dsname, DMU_OST_OTHER, 0,
3198 ztest_objset_create_cb, NULL);
3200 if (err || zilset < 80)
3203 if (ztest_opts.zo_verbose >= 6)
3204 (void) printf("Setting dataset %s to sync always\n", dsname);
3205 return (ztest_dsl_prop_set_uint64(dsname, ZFS_PROP_SYNC,
3206 ZFS_SYNC_ALWAYS, B_FALSE));
3211 ztest_objset_destroy_cb(const char *name, void *arg)
3214 dmu_object_info_t doi;
3218 * Verify that the dataset contains a directory object.
3220 VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_TRUE, FTAG, &os));
3221 error = dmu_object_info(os, ZTEST_DIROBJ, &doi);
3222 if (error != ENOENT) {
3223 /* We could have crashed in the middle of destroying it */
3225 ASSERT3U(doi.doi_type, ==, DMU_OT_ZAP_OTHER);
3226 ASSERT3S(doi.doi_physical_blocks_512, >=, 0);
3228 dmu_objset_disown(os, FTAG);
3231 * Destroy the dataset.
3233 if (strchr(name, '@') != NULL) {
3234 VERIFY0(dsl_destroy_snapshot(name, B_FALSE));
3236 VERIFY0(dsl_destroy_head(name));
3242 ztest_snapshot_create(char *osname, uint64_t id)
3244 char snapname[MAXNAMELEN];
3247 (void) snprintf(snapname, sizeof (snapname), "%llu", (u_longlong_t)id);
3249 error = dmu_objset_snapshot_one(osname, snapname);
3250 if (error == ENOSPC) {
3251 ztest_record_enospc(FTAG);
3254 if (error != 0 && error != EEXIST) {
3255 fatal(0, "ztest_snapshot_create(%s@%s) = %d", osname,
3262 ztest_snapshot_destroy(char *osname, uint64_t id)
3264 char snapname[MAXNAMELEN];
3267 (void) snprintf(snapname, MAXNAMELEN, "%s@%llu", osname,
3270 error = dsl_destroy_snapshot(snapname, B_FALSE);
3271 if (error != 0 && error != ENOENT)
3272 fatal(0, "ztest_snapshot_destroy(%s) = %d", snapname, error);
3278 ztest_dmu_objset_create_destroy(ztest_ds_t *zd, uint64_t id)
3284 char name[MAXNAMELEN];
3287 (void) rw_rdlock(&ztest_name_lock);
3289 (void) snprintf(name, MAXNAMELEN, "%s/temp_%llu",
3290 ztest_opts.zo_pool, (u_longlong_t)id);
3293 * If this dataset exists from a previous run, process its replay log
3294 * half of the time. If we don't replay it, then dmu_objset_destroy()
3295 * (invoked from ztest_objset_destroy_cb()) should just throw it away.
3297 if (ztest_random(2) == 0 &&
3298 dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os) == 0) {
3299 ztest_zd_init(&zdtmp, NULL, os);
3300 zil_replay(os, &zdtmp, ztest_replay_vector);
3301 ztest_zd_fini(&zdtmp);
3302 dmu_objset_disown(os, FTAG);
3306 * There may be an old instance of the dataset we're about to
3307 * create lying around from a previous run. If so, destroy it
3308 * and all of its snapshots.
3310 (void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
3311 DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
3314 * Verify that the destroyed dataset is no longer in the namespace.
3316 VERIFY3U(ENOENT, ==, dmu_objset_own(name, DMU_OST_OTHER, B_TRUE,
3320 * Verify that we can create a new dataset.
3322 error = ztest_dataset_create(name);
3324 if (error == ENOSPC) {
3325 ztest_record_enospc(FTAG);
3326 (void) rw_unlock(&ztest_name_lock);
3329 fatal(0, "dmu_objset_create(%s) = %d", name, error);
3332 VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os));
3334 ztest_zd_init(&zdtmp, NULL, os);
3337 * Open the intent log for it.
3339 zilog = zil_open(os, ztest_get_data);
3342 * Put some objects in there, do a little I/O to them,
3343 * and randomly take a couple of snapshots along the way.
3345 iters = ztest_random(5);
3346 for (int i = 0; i < iters; i++) {
3347 ztest_dmu_object_alloc_free(&zdtmp, id);
3348 if (ztest_random(iters) == 0)
3349 (void) ztest_snapshot_create(name, i);
3353 * Verify that we cannot create an existing dataset.
3355 VERIFY3U(EEXIST, ==,
3356 dmu_objset_create(name, DMU_OST_OTHER, 0, NULL, NULL));
3359 * Verify that we can hold an objset that is also owned.
3361 VERIFY3U(0, ==, dmu_objset_hold(name, FTAG, &os2));
3362 dmu_objset_rele(os2, FTAG);
3365 * Verify that we cannot own an objset that is already owned.
3368 dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os2));
3371 dmu_objset_disown(os, FTAG);
3372 ztest_zd_fini(&zdtmp);
3374 (void) rw_unlock(&ztest_name_lock);
3378 * Verify that dmu_snapshot_{create,destroy,open,close} work as expected.
3381 ztest_dmu_snapshot_create_destroy(ztest_ds_t *zd, uint64_t id)
3383 (void) rw_rdlock(&ztest_name_lock);
3384 (void) ztest_snapshot_destroy(zd->zd_name, id);
3385 (void) ztest_snapshot_create(zd->zd_name, id);
3386 (void) rw_unlock(&ztest_name_lock);
3390 * Cleanup non-standard snapshots and clones.
3393 ztest_dsl_dataset_cleanup(char *osname, uint64_t id)
3395 char snap1name[MAXNAMELEN];
3396 char clone1name[MAXNAMELEN];
3397 char snap2name[MAXNAMELEN];
3398 char clone2name[MAXNAMELEN];
3399 char snap3name[MAXNAMELEN];
3402 (void) snprintf(snap1name, MAXNAMELEN, "%s@s1_%llu", osname, id);
3403 (void) snprintf(clone1name, MAXNAMELEN, "%s/c1_%llu", osname, id);
3404 (void) snprintf(snap2name, MAXNAMELEN, "%s@s2_%llu", clone1name, id);
3405 (void) snprintf(clone2name, MAXNAMELEN, "%s/c2_%llu", osname, id);
3406 (void) snprintf(snap3name, MAXNAMELEN, "%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[MAXNAMELEN];
3433 char clone1name[MAXNAMELEN];
3434 char snap2name[MAXNAMELEN];
3435 char clone2name[MAXNAMELEN];
3436 char snap3name[MAXNAMELEN];
3437 char *osname = zd->zd_name;
3440 (void) rw_rdlock(&ztest_name_lock);
3442 ztest_dsl_dataset_cleanup(osname, id);
3444 (void) snprintf(snap1name, MAXNAMELEN, "%s@s1_%llu", osname, id);
3445 (void) snprintf(clone1name, MAXNAMELEN, "%s/c1_%llu", osname, id);
3446 (void) snprintf(snap2name, MAXNAMELEN, "%s@s2_%llu", clone1name, id);
3447 (void) snprintf(clone2name, MAXNAMELEN, "%s/c2_%llu", osname, id);
3448 (void) snprintf(snap3name, MAXNAMELEN, "%s@s3_%llu", clone1name, id);
3450 error = dmu_objset_snapshot_one(osname, strchr(snap1name, '@') + 1);
3451 if (error && error != EEXIST) {
3452 if (error == ENOSPC) {
3453 ztest_record_enospc(FTAG);
3456 fatal(0, "dmu_take_snapshot(%s) = %d", snap1name, error);
3459 error = dmu_objset_clone(clone1name, snap1name);
3461 if (error == ENOSPC) {
3462 ztest_record_enospc(FTAG);
3465 fatal(0, "dmu_objset_create(%s) = %d", clone1name, error);
3468 error = dmu_objset_snapshot_one(clone1name, strchr(snap2name, '@') + 1);
3469 if (error && error != EEXIST) {
3470 if (error == ENOSPC) {
3471 ztest_record_enospc(FTAG);
3474 fatal(0, "dmu_open_snapshot(%s) = %d", snap2name, error);
3477 error = dmu_objset_snapshot_one(clone1name, strchr(snap3name, '@') + 1);
3478 if (error && error != EEXIST) {
3479 if (error == ENOSPC) {
3480 ztest_record_enospc(FTAG);
3483 fatal(0, "dmu_open_snapshot(%s) = %d", snap3name, error);
3486 error = dmu_objset_clone(clone2name, snap3name);
3488 if (error == ENOSPC) {
3489 ztest_record_enospc(FTAG);
3492 fatal(0, "dmu_objset_create(%s) = %d", clone2name, error);
3495 error = dmu_objset_own(snap2name, DMU_OST_ANY, B_TRUE, FTAG, &os);
3497 fatal(0, "dmu_objset_own(%s) = %d", snap2name, error);
3498 error = dsl_dataset_promote(clone2name, NULL);
3499 if (error == ENOSPC) {
3500 dmu_objset_disown(os, FTAG);
3501 ztest_record_enospc(FTAG);
3505 fatal(0, "dsl_dataset_promote(%s), %d, not EBUSY", clone2name,
3507 dmu_objset_disown(os, FTAG);
3510 ztest_dsl_dataset_cleanup(osname, id);
3512 (void) rw_unlock(&ztest_name_lock);
3516 * Verify that dmu_object_{alloc,free} work as expected.
3519 ztest_dmu_object_alloc_free(ztest_ds_t *zd, uint64_t id)
3522 int batchsize = sizeof (od) / sizeof (od[0]);
3524 for (int b = 0; b < batchsize; b++)
3525 ztest_od_init(&od[b], id, FTAG, b, DMU_OT_UINT64_OTHER, 0, 0);
3528 * Destroy the previous batch of objects, create a new batch,
3529 * and do some I/O on the new objects.
3531 if (ztest_object_init(zd, od, sizeof (od), B_TRUE) != 0)
3534 while (ztest_random(4 * batchsize) != 0)
3535 ztest_io(zd, od[ztest_random(batchsize)].od_object,
3536 ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
3540 * Verify that dmu_{read,write} work as expected.
3543 ztest_dmu_read_write(ztest_ds_t *zd, uint64_t id)
3545 objset_t *os = zd->zd_os;
3548 int i, freeit, error;
3550 bufwad_t *packbuf, *bigbuf, *pack, *bigH, *bigT;
3551 uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
3552 uint64_t chunksize = (1000 + ztest_random(1000)) * sizeof (uint64_t);
3553 uint64_t regions = 997;
3554 uint64_t stride = 123456789ULL;
3555 uint64_t width = 40;
3556 int free_percent = 5;
3559 * This test uses two objects, packobj and bigobj, that are always
3560 * updated together (i.e. in the same tx) so that their contents are
3561 * in sync and can be compared. Their contents relate to each other
3562 * in a simple way: packobj is a dense array of 'bufwad' structures,
3563 * while bigobj is a sparse array of the same bufwads. Specifically,
3564 * for any index n, there are three bufwads that should be identical:
3566 * packobj, at offset n * sizeof (bufwad_t)
3567 * bigobj, at the head of the nth chunk
3568 * bigobj, at the tail of the nth chunk
3570 * The chunk size is arbitrary. It doesn't have to be a power of two,
3571 * and it doesn't have any relation to the object blocksize.
3572 * The only requirement is that it can hold at least two bufwads.
3574 * Normally, we write the bufwad to each of these locations.
3575 * However, free_percent of the time we instead write zeroes to
3576 * packobj and perform a dmu_free_range() on bigobj. By comparing
3577 * bigobj to packobj, we can verify that the DMU is correctly
3578 * tracking which parts of an object are allocated and free,
3579 * and that the contents of the allocated blocks are correct.
3583 * Read the directory info. If it's the first time, set things up.
3585 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, chunksize);
3586 ztest_od_init(&od[1], id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize);
3588 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
3591 bigobj = od[0].od_object;
3592 packobj = od[1].od_object;
3593 chunksize = od[0].od_gen;
3594 ASSERT(chunksize == od[1].od_gen);
3597 * Prefetch a random chunk of the big object.
3598 * Our aim here is to get some async reads in flight
3599 * for blocks that we may free below; the DMU should
3600 * handle this race correctly.
3602 n = ztest_random(regions) * stride + ztest_random(width);
3603 s = 1 + ztest_random(2 * width - 1);
3604 dmu_prefetch(os, bigobj, n * chunksize, s * chunksize);
3607 * Pick a random index and compute the offsets into packobj and bigobj.
3609 n = ztest_random(regions) * stride + ztest_random(width);
3610 s = 1 + ztest_random(width - 1);
3612 packoff = n * sizeof (bufwad_t);
3613 packsize = s * sizeof (bufwad_t);
3615 bigoff = n * chunksize;
3616 bigsize = s * chunksize;
3618 packbuf = umem_alloc(packsize, UMEM_NOFAIL);
3619 bigbuf = umem_alloc(bigsize, UMEM_NOFAIL);
3622 * free_percent of the time, free a range of bigobj rather than
3625 freeit = (ztest_random(100) < free_percent);
3628 * Read the current contents of our objects.
3630 error = dmu_read(os, packobj, packoff, packsize, packbuf,
3633 error = dmu_read(os, bigobj, bigoff, bigsize, bigbuf,
3638 * Get a tx for the mods to both packobj and bigobj.
3640 tx = dmu_tx_create(os);
3642 dmu_tx_hold_write(tx, packobj, packoff, packsize);
3645 dmu_tx_hold_free(tx, bigobj, bigoff, bigsize);
3647 dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
3649 /* This accounts for setting the checksum/compression. */
3650 dmu_tx_hold_bonus(tx, bigobj);
3652 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3654 umem_free(packbuf, packsize);
3655 umem_free(bigbuf, bigsize);
3659 enum zio_checksum cksum;
3661 cksum = (enum zio_checksum)
3662 ztest_random_dsl_prop(ZFS_PROP_CHECKSUM);
3663 } while (cksum >= ZIO_CHECKSUM_LEGACY_FUNCTIONS);
3664 dmu_object_set_checksum(os, bigobj, cksum, tx);
3666 enum zio_compress comp;
3668 comp = (enum zio_compress)
3669 ztest_random_dsl_prop(ZFS_PROP_COMPRESSION);
3670 } while (comp >= ZIO_COMPRESS_LEGACY_FUNCTIONS);
3671 dmu_object_set_compress(os, bigobj, comp, tx);
3674 * For each index from n to n + s, verify that the existing bufwad
3675 * in packobj matches the bufwads at the head and tail of the
3676 * corresponding chunk in bigobj. Then update all three bufwads
3677 * with the new values we want to write out.
3679 for (i = 0; i < s; i++) {
3681 pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
3683 bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
3685 bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
3687 ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
3688 ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
3690 if (pack->bw_txg > txg)
3691 fatal(0, "future leak: got %llx, open txg is %llx",
3694 if (pack->bw_data != 0 && pack->bw_index != n + i)
3695 fatal(0, "wrong index: got %llx, wanted %llx+%llx",
3696 pack->bw_index, n, i);
3698 if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
3699 fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
3701 if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
3702 fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
3705 bzero(pack, sizeof (bufwad_t));
3707 pack->bw_index = n + i;
3709 pack->bw_data = 1 + ztest_random(-2ULL);
3716 * We've verified all the old bufwads, and made new ones.
3717 * Now write them out.
3719 dmu_write(os, packobj, packoff, packsize, packbuf, tx);
3722 if (ztest_opts.zo_verbose >= 7) {
3723 (void) printf("freeing offset %llx size %llx"
3725 (u_longlong_t)bigoff,
3726 (u_longlong_t)bigsize,
3729 VERIFY(0 == dmu_free_range(os, bigobj, bigoff, bigsize, tx));
3731 if (ztest_opts.zo_verbose >= 7) {
3732 (void) printf("writing offset %llx size %llx"
3734 (u_longlong_t)bigoff,
3735 (u_longlong_t)bigsize,
3738 dmu_write(os, bigobj, bigoff, bigsize, bigbuf, tx);
3744 * Sanity check the stuff we just wrote.
3747 void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
3748 void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
3750 VERIFY(0 == dmu_read(os, packobj, packoff,
3751 packsize, packcheck, DMU_READ_PREFETCH));
3752 VERIFY(0 == dmu_read(os, bigobj, bigoff,
3753 bigsize, bigcheck, DMU_READ_PREFETCH));
3755 ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
3756 ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
3758 umem_free(packcheck, packsize);
3759 umem_free(bigcheck, bigsize);
3762 umem_free(packbuf, packsize);
3763 umem_free(bigbuf, bigsize);
3767 compare_and_update_pbbufs(uint64_t s, bufwad_t *packbuf, bufwad_t *bigbuf,
3768 uint64_t bigsize, uint64_t n, uint64_t chunksize, uint64_t txg)
3776 * For each index from n to n + s, verify that the existing bufwad
3777 * in packobj matches the bufwads at the head and tail of the
3778 * corresponding chunk in bigobj. Then update all three bufwads
3779 * with the new values we want to write out.
3781 for (i = 0; i < s; i++) {
3783 pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
3785 bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
3787 bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
3789 ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
3790 ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
3792 if (pack->bw_txg > txg)
3793 fatal(0, "future leak: got %llx, open txg is %llx",
3796 if (pack->bw_data != 0 && pack->bw_index != n + i)
3797 fatal(0, "wrong index: got %llx, wanted %llx+%llx",
3798 pack->bw_index, n, i);
3800 if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
3801 fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
3803 if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
3804 fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
3806 pack->bw_index = n + i;
3808 pack->bw_data = 1 + ztest_random(-2ULL);
3816 ztest_dmu_read_write_zcopy(ztest_ds_t *zd, uint64_t id)
3818 objset_t *os = zd->zd_os;
3824 bufwad_t *packbuf, *bigbuf;
3825 uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
3826 uint64_t blocksize = ztest_random_blocksize();
3827 uint64_t chunksize = blocksize;
3828 uint64_t regions = 997;
3829 uint64_t stride = 123456789ULL;
3831 dmu_buf_t *bonus_db;
3832 arc_buf_t **bigbuf_arcbufs;
3833 dmu_object_info_t doi;
3836 * This test uses two objects, packobj and bigobj, that are always
3837 * updated together (i.e. in the same tx) so that their contents are
3838 * in sync and can be compared. Their contents relate to each other
3839 * in a simple way: packobj is a dense array of 'bufwad' structures,
3840 * while bigobj is a sparse array of the same bufwads. Specifically,
3841 * for any index n, there are three bufwads that should be identical:
3843 * packobj, at offset n * sizeof (bufwad_t)
3844 * bigobj, at the head of the nth chunk
3845 * bigobj, at the tail of the nth chunk
3847 * The chunk size is set equal to bigobj block size so that
3848 * dmu_assign_arcbuf() can be tested for object updates.
3852 * Read the directory info. If it's the first time, set things up.
3854 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
3855 ztest_od_init(&od[1], id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize);
3857 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
3860 bigobj = od[0].od_object;
3861 packobj = od[1].od_object;
3862 blocksize = od[0].od_blocksize;
3863 chunksize = blocksize;
3864 ASSERT(chunksize == od[1].od_gen);
3866 VERIFY(dmu_object_info(os, bigobj, &doi) == 0);
3867 VERIFY(ISP2(doi.doi_data_block_size));
3868 VERIFY(chunksize == doi.doi_data_block_size);
3869 VERIFY(chunksize >= 2 * sizeof (bufwad_t));
3872 * Pick a random index and compute the offsets into packobj and bigobj.
3874 n = ztest_random(regions) * stride + ztest_random(width);
3875 s = 1 + ztest_random(width - 1);
3877 packoff = n * sizeof (bufwad_t);
3878 packsize = s * sizeof (bufwad_t);
3880 bigoff = n * chunksize;
3881 bigsize = s * chunksize;
3883 packbuf = umem_zalloc(packsize, UMEM_NOFAIL);
3884 bigbuf = umem_zalloc(bigsize, UMEM_NOFAIL);
3886 VERIFY3U(0, ==, dmu_bonus_hold(os, bigobj, FTAG, &bonus_db));
3888 bigbuf_arcbufs = umem_zalloc(2 * s * sizeof (arc_buf_t *), UMEM_NOFAIL);
3891 * Iteration 0 test zcopy for DB_UNCACHED dbufs.
3892 * Iteration 1 test zcopy to already referenced dbufs.
3893 * Iteration 2 test zcopy to dirty dbuf in the same txg.
3894 * Iteration 3 test zcopy to dbuf dirty in previous txg.
3895 * Iteration 4 test zcopy when dbuf is no longer dirty.
3896 * Iteration 5 test zcopy when it can't be done.
3897 * Iteration 6 one more zcopy write.
3899 for (i = 0; i < 7; i++) {
3904 * In iteration 5 (i == 5) use arcbufs
3905 * that don't match bigobj blksz to test
3906 * dmu_assign_arcbuf() when it can't directly
3907 * assign an arcbuf to a dbuf.
3909 for (j = 0; j < s; j++) {
3912 dmu_request_arcbuf(bonus_db, chunksize);
3914 bigbuf_arcbufs[2 * j] =
3915 dmu_request_arcbuf(bonus_db, chunksize / 2);
3916 bigbuf_arcbufs[2 * j + 1] =
3917 dmu_request_arcbuf(bonus_db, chunksize / 2);
3922 * Get a tx for the mods to both packobj and bigobj.
3924 tx = dmu_tx_create(os);
3926 dmu_tx_hold_write(tx, packobj, packoff, packsize);
3927 dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
3929 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3931 umem_free(packbuf, packsize);
3932 umem_free(bigbuf, bigsize);
3933 for (j = 0; j < s; j++) {
3935 dmu_return_arcbuf(bigbuf_arcbufs[j]);
3938 bigbuf_arcbufs[2 * j]);
3940 bigbuf_arcbufs[2 * j + 1]);
3943 umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
3944 dmu_buf_rele(bonus_db, FTAG);
3949 * 50% of the time don't read objects in the 1st iteration to
3950 * test dmu_assign_arcbuf() for the case when there're no
3951 * existing dbufs for the specified offsets.
3953 if (i != 0 || ztest_random(2) != 0) {
3954 error = dmu_read(os, packobj, packoff,
3955 packsize, packbuf, DMU_READ_PREFETCH);
3957 error = dmu_read(os, bigobj, bigoff, bigsize,
3958 bigbuf, DMU_READ_PREFETCH);
3961 compare_and_update_pbbufs(s, packbuf, bigbuf, bigsize,
3965 * We've verified all the old bufwads, and made new ones.
3966 * Now write them out.
3968 dmu_write(os, packobj, packoff, packsize, packbuf, tx);
3969 if (ztest_opts.zo_verbose >= 7) {
3970 (void) printf("writing offset %llx size %llx"
3972 (u_longlong_t)bigoff,
3973 (u_longlong_t)bigsize,
3976 for (off = bigoff, j = 0; j < s; j++, off += chunksize) {
3979 bcopy((caddr_t)bigbuf + (off - bigoff),
3980 bigbuf_arcbufs[j]->b_data, chunksize);
3982 bcopy((caddr_t)bigbuf + (off - bigoff),
3983 bigbuf_arcbufs[2 * j]->b_data,
3985 bcopy((caddr_t)bigbuf + (off - bigoff) +
3987 bigbuf_arcbufs[2 * j + 1]->b_data,
3992 VERIFY(dmu_buf_hold(os, bigobj, off,
3993 FTAG, &dbt, DMU_READ_NO_PREFETCH) == 0);
3996 dmu_assign_arcbuf(bonus_db, off,
3997 bigbuf_arcbufs[j], tx);
3999 dmu_assign_arcbuf(bonus_db, off,
4000 bigbuf_arcbufs[2 * j], tx);
4001 dmu_assign_arcbuf(bonus_db,
4002 off + chunksize / 2,
4003 bigbuf_arcbufs[2 * j + 1], tx);
4006 dmu_buf_rele(dbt, FTAG);
4012 * Sanity check the stuff we just wrote.
4015 void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
4016 void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
4018 VERIFY(0 == dmu_read(os, packobj, packoff,
4019 packsize, packcheck, DMU_READ_PREFETCH));
4020 VERIFY(0 == dmu_read(os, bigobj, bigoff,
4021 bigsize, bigcheck, DMU_READ_PREFETCH));
4023 ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
4024 ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
4026 umem_free(packcheck, packsize);
4027 umem_free(bigcheck, bigsize);
4030 txg_wait_open(dmu_objset_pool(os), 0);
4031 } else if (i == 3) {
4032 txg_wait_synced(dmu_objset_pool(os), 0);
4036 dmu_buf_rele(bonus_db, FTAG);
4037 umem_free(packbuf, packsize);
4038 umem_free(bigbuf, bigsize);
4039 umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
4044 ztest_dmu_write_parallel(ztest_ds_t *zd, uint64_t id)
4047 uint64_t offset = (1ULL << (ztest_random(20) + 43)) +
4048 (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
4051 * Have multiple threads write to large offsets in an object
4052 * to verify that parallel writes to an object -- even to the
4053 * same blocks within the object -- doesn't cause any trouble.
4055 ztest_od_init(&od[0], ID_PARALLEL, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
4057 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4060 while (ztest_random(10) != 0)
4061 ztest_io(zd, od[0].od_object, offset);
4065 ztest_dmu_prealloc(ztest_ds_t *zd, uint64_t id)
4068 uint64_t offset = (1ULL << (ztest_random(4) + SPA_MAXBLOCKSHIFT)) +
4069 (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
4070 uint64_t count = ztest_random(20) + 1;
4071 uint64_t blocksize = ztest_random_blocksize();
4074 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
4076 if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4079 if (ztest_truncate(zd, od[0].od_object, offset, count * blocksize) != 0)
4082 ztest_prealloc(zd, od[0].od_object, offset, count * blocksize);
4084 data = umem_zalloc(blocksize, UMEM_NOFAIL);
4086 while (ztest_random(count) != 0) {
4087 uint64_t randoff = offset + (ztest_random(count) * blocksize);
4088 if (ztest_write(zd, od[0].od_object, randoff, blocksize,
4091 while (ztest_random(4) != 0)
4092 ztest_io(zd, od[0].od_object, randoff);
4095 umem_free(data, blocksize);
4099 * Verify that zap_{create,destroy,add,remove,update} work as expected.
4101 #define ZTEST_ZAP_MIN_INTS 1
4102 #define ZTEST_ZAP_MAX_INTS 4
4103 #define ZTEST_ZAP_MAX_PROPS 1000
4106 ztest_zap(ztest_ds_t *zd, uint64_t id)
4108 objset_t *os = zd->zd_os;
4111 uint64_t txg, last_txg;
4112 uint64_t value[ZTEST_ZAP_MAX_INTS];
4113 uint64_t zl_ints, zl_intsize, prop;
4116 char propname[100], txgname[100];
4118 char *hc[2] = { "s.acl.h", ".s.open.h.hyLZlg" };
4120 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0);
4122 if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4125 object = od[0].od_object;
4128 * Generate a known hash collision, and verify that
4129 * we can lookup and remove both entries.
4131 tx = dmu_tx_create(os);
4132 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4133 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4136 for (i = 0; i < 2; i++) {
4138 VERIFY3U(0, ==, zap_add(os, object, hc[i], sizeof (uint64_t),
4141 for (i = 0; i < 2; i++) {
4142 VERIFY3U(EEXIST, ==, zap_add(os, object, hc[i],
4143 sizeof (uint64_t), 1, &value[i], tx));
4145 zap_length(os, object, hc[i], &zl_intsize, &zl_ints));
4146 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4147 ASSERT3U(zl_ints, ==, 1);
4149 for (i = 0; i < 2; i++) {
4150 VERIFY3U(0, ==, zap_remove(os, object, hc[i], tx));
4155 * Generate a buch of random entries.
4157 ints = MAX(ZTEST_ZAP_MIN_INTS, object % ZTEST_ZAP_MAX_INTS);
4159 prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
4160 (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
4161 (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
4162 bzero(value, sizeof (value));
4166 * If these zap entries already exist, validate their contents.
4168 error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
4170 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4171 ASSERT3U(zl_ints, ==, 1);
4173 VERIFY(zap_lookup(os, object, txgname, zl_intsize,
4174 zl_ints, &last_txg) == 0);
4176 VERIFY(zap_length(os, object, propname, &zl_intsize,
4179 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4180 ASSERT3U(zl_ints, ==, ints);
4182 VERIFY(zap_lookup(os, object, propname, zl_intsize,
4183 zl_ints, value) == 0);
4185 for (i = 0; i < ints; i++) {
4186 ASSERT3U(value[i], ==, last_txg + object + i);
4189 ASSERT3U(error, ==, ENOENT);
4193 * Atomically update two entries in our zap object.
4194 * The first is named txg_%llu, and contains the txg
4195 * in which the property was last updated. The second
4196 * is named prop_%llu, and the nth element of its value
4197 * should be txg + object + n.
4199 tx = dmu_tx_create(os);
4200 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4201 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4206 fatal(0, "zap future leak: old %llu new %llu", last_txg, txg);
4208 for (i = 0; i < ints; i++)
4209 value[i] = txg + object + i;
4211 VERIFY3U(0, ==, zap_update(os, object, txgname, sizeof (uint64_t),
4213 VERIFY3U(0, ==, zap_update(os, object, propname, sizeof (uint64_t),
4219 * Remove a random pair of entries.
4221 prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
4222 (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
4223 (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
4225 error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
4227 if (error == ENOENT)
4232 tx = dmu_tx_create(os);
4233 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4234 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4237 VERIFY3U(0, ==, zap_remove(os, object, txgname, tx));
4238 VERIFY3U(0, ==, zap_remove(os, object, propname, tx));
4243 * Testcase to test the upgrading of a microzap to fatzap.
4246 ztest_fzap(ztest_ds_t *zd, uint64_t id)
4248 objset_t *os = zd->zd_os;
4250 uint64_t object, txg;
4252 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0);
4254 if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4257 object = od[0].od_object;
4260 * Add entries to this ZAP and make sure it spills over
4261 * and gets upgraded to a fatzap. Also, since we are adding
4262 * 2050 entries we should see ptrtbl growth and leaf-block split.
4264 for (int i = 0; i < 2050; i++) {
4265 char name[MAXNAMELEN];
4270 (void) snprintf(name, sizeof (name), "fzap-%llu-%llu",
4273 tx = dmu_tx_create(os);
4274 dmu_tx_hold_zap(tx, object, B_TRUE, name);
4275 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4278 error = zap_add(os, object, name, sizeof (uint64_t), 1,
4280 ASSERT(error == 0 || error == EEXIST);
4287 ztest_zap_parallel(ztest_ds_t *zd, uint64_t id)
4289 objset_t *os = zd->zd_os;
4291 uint64_t txg, object, count, wsize, wc, zl_wsize, zl_wc;
4293 int i, namelen, error;
4294 int micro = ztest_random(2);
4295 char name[20], string_value[20];
4298 ztest_od_init(&od[0], ID_PARALLEL, FTAG, micro, DMU_OT_ZAP_OTHER, 0, 0);
4300 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4303 object = od[0].od_object;
4306 * Generate a random name of the form 'xxx.....' where each
4307 * x is a random printable character and the dots are dots.
4308 * There are 94 such characters, and the name length goes from
4309 * 6 to 20, so there are 94^3 * 15 = 12,458,760 possible names.
4311 namelen = ztest_random(sizeof (name) - 5) + 5 + 1;
4313 for (i = 0; i < 3; i++)
4314 name[i] = '!' + ztest_random('~' - '!' + 1);
4315 for (; i < namelen - 1; i++)
4319 if ((namelen & 1) || micro) {
4320 wsize = sizeof (txg);
4326 data = string_value;
4330 VERIFY0(zap_count(os, object, &count));
4331 ASSERT(count != -1ULL);
4334 * Select an operation: length, lookup, add, update, remove.
4336 i = ztest_random(5);
4339 tx = dmu_tx_create(os);
4340 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4341 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4344 bcopy(name, string_value, namelen);
4348 bzero(string_value, namelen);
4354 error = zap_length(os, object, name, &zl_wsize, &zl_wc);
4356 ASSERT3U(wsize, ==, zl_wsize);
4357 ASSERT3U(wc, ==, zl_wc);
4359 ASSERT3U(error, ==, ENOENT);
4364 error = zap_lookup(os, object, name, wsize, wc, data);
4366 if (data == string_value &&
4367 bcmp(name, data, namelen) != 0)
4368 fatal(0, "name '%s' != val '%s' len %d",
4369 name, data, namelen);
4371 ASSERT3U(error, ==, ENOENT);
4376 error = zap_add(os, object, name, wsize, wc, data, tx);
4377 ASSERT(error == 0 || error == EEXIST);
4381 VERIFY(zap_update(os, object, name, wsize, wc, data, tx) == 0);
4385 error = zap_remove(os, object, name, tx);
4386 ASSERT(error == 0 || error == ENOENT);
4395 * Commit callback data.
4397 typedef struct ztest_cb_data {
4398 list_node_t zcd_node;
4400 int zcd_expected_err;
4401 boolean_t zcd_added;
4402 boolean_t zcd_called;
4406 /* This is the actual commit callback function */
4408 ztest_commit_callback(void *arg, int error)
4410 ztest_cb_data_t *data = arg;
4411 uint64_t synced_txg;
4413 VERIFY(data != NULL);
4414 VERIFY3S(data->zcd_expected_err, ==, error);
4415 VERIFY(!data->zcd_called);
4417 synced_txg = spa_last_synced_txg(data->zcd_spa);
4418 if (data->zcd_txg > synced_txg)
4419 fatal(0, "commit callback of txg %" PRIu64 " called prematurely"
4420 ", last synced txg = %" PRIu64 "\n", data->zcd_txg,
4423 data->zcd_called = B_TRUE;
4425 if (error == ECANCELED) {
4426 ASSERT0(data->zcd_txg);
4427 ASSERT(!data->zcd_added);
4430 * The private callback data should be destroyed here, but
4431 * since we are going to check the zcd_called field after
4432 * dmu_tx_abort(), we will destroy it there.
4437 /* Was this callback added to the global callback list? */
4438 if (!data->zcd_added)
4441 ASSERT3U(data->zcd_txg, !=, 0);
4443 /* Remove our callback from the list */
4444 (void) mutex_lock(&zcl.zcl_callbacks_lock);
4445 list_remove(&zcl.zcl_callbacks, data);
4446 (void) mutex_unlock(&zcl.zcl_callbacks_lock);
4449 umem_free(data, sizeof (ztest_cb_data_t));
4452 /* Allocate and initialize callback data structure */
4453 static ztest_cb_data_t *
4454 ztest_create_cb_data(objset_t *os, uint64_t txg)
4456 ztest_cb_data_t *cb_data;
4458 cb_data = umem_zalloc(sizeof (ztest_cb_data_t), UMEM_NOFAIL);
4460 cb_data->zcd_txg = txg;
4461 cb_data->zcd_spa = dmu_objset_spa(os);
4467 * If a number of txgs equal to this threshold have been created after a commit
4468 * callback has been registered but not called, then we assume there is an
4469 * implementation bug.
4471 #define ZTEST_COMMIT_CALLBACK_THRESH (TXG_CONCURRENT_STATES + 2)
4474 * Commit callback test.
4477 ztest_dmu_commit_callbacks(ztest_ds_t *zd, uint64_t id)
4479 objset_t *os = zd->zd_os;
4482 ztest_cb_data_t *cb_data[3], *tmp_cb;
4483 uint64_t old_txg, txg;
4486 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
4488 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4491 tx = dmu_tx_create(os);
4493 cb_data[0] = ztest_create_cb_data(os, 0);
4494 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[0]);
4496 dmu_tx_hold_write(tx, od[0].od_object, 0, sizeof (uint64_t));
4498 /* Every once in a while, abort the transaction on purpose */
4499 if (ztest_random(100) == 0)
4503 error = dmu_tx_assign(tx, TXG_NOWAIT);
4505 txg = error ? 0 : dmu_tx_get_txg(tx);
4507 cb_data[0]->zcd_txg = txg;
4508 cb_data[1] = ztest_create_cb_data(os, txg);
4509 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[1]);
4513 * It's not a strict requirement to call the registered
4514 * callbacks from inside dmu_tx_abort(), but that's what
4515 * it's supposed to happen in the current implementation
4516 * so we will check for that.
4518 for (i = 0; i < 2; i++) {
4519 cb_data[i]->zcd_expected_err = ECANCELED;
4520 VERIFY(!cb_data[i]->zcd_called);
4525 for (i = 0; i < 2; i++) {
4526 VERIFY(cb_data[i]->zcd_called);
4527 umem_free(cb_data[i], sizeof (ztest_cb_data_t));
4533 cb_data[2] = ztest_create_cb_data(os, txg);
4534 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[2]);
4537 * Read existing data to make sure there isn't a future leak.
4539 VERIFY(0 == dmu_read(os, od[0].od_object, 0, sizeof (uint64_t),
4540 &old_txg, DMU_READ_PREFETCH));
4543 fatal(0, "future leak: got %" PRIu64 ", open txg is %" PRIu64,
4546 dmu_write(os, od[0].od_object, 0, sizeof (uint64_t), &txg, tx);
4548 (void) mutex_lock(&zcl.zcl_callbacks_lock);
4551 * Since commit callbacks don't have any ordering requirement and since
4552 * it is theoretically possible for a commit callback to be called
4553 * after an arbitrary amount of time has elapsed since its txg has been
4554 * synced, it is difficult to reliably determine whether a commit
4555 * callback hasn't been called due to high load or due to a flawed
4558 * In practice, we will assume that if after a certain number of txgs a
4559 * commit callback hasn't been called, then most likely there's an
4560 * implementation bug..
4562 tmp_cb = list_head(&zcl.zcl_callbacks);
4563 if (tmp_cb != NULL &&
4564 (txg - ZTEST_COMMIT_CALLBACK_THRESH) > tmp_cb->zcd_txg) {
4565 fatal(0, "Commit callback threshold exceeded, oldest txg: %"
4566 PRIu64 ", open txg: %" PRIu64 "\n", tmp_cb->zcd_txg, txg);
4570 * Let's find the place to insert our callbacks.
4572 * Even though the list is ordered by txg, it is possible for the
4573 * insertion point to not be the end because our txg may already be
4574 * quiescing at this point and other callbacks in the open txg
4575 * (from other objsets) may have sneaked in.
4577 tmp_cb = list_tail(&zcl.zcl_callbacks);
4578 while (tmp_cb != NULL && tmp_cb->zcd_txg > txg)
4579 tmp_cb = list_prev(&zcl.zcl_callbacks, tmp_cb);
4581 /* Add the 3 callbacks to the list */
4582 for (i = 0; i < 3; i++) {
4584 list_insert_head(&zcl.zcl_callbacks, cb_data[i]);
4586 list_insert_after(&zcl.zcl_callbacks, tmp_cb,
4589 cb_data[i]->zcd_added = B_TRUE;
4590 VERIFY(!cb_data[i]->zcd_called);
4592 tmp_cb = cb_data[i];
4595 (void) mutex_unlock(&zcl.zcl_callbacks_lock);
4602 ztest_dsl_prop_get_set(ztest_ds_t *zd, uint64_t id)
4604 zfs_prop_t proplist[] = {
4606 ZFS_PROP_COMPRESSION,
4611 (void) rw_rdlock(&ztest_name_lock);
4613 for (int p = 0; p < sizeof (proplist) / sizeof (proplist[0]); p++)
4614 (void) ztest_dsl_prop_set_uint64(zd->zd_name, proplist[p],
4615 ztest_random_dsl_prop(proplist[p]), (int)ztest_random(2));
4617 (void) rw_unlock(&ztest_name_lock);
4622 ztest_spa_prop_get_set(ztest_ds_t *zd, uint64_t id)
4624 nvlist_t *props = NULL;
4626 (void) rw_rdlock(&ztest_name_lock);
4628 (void) ztest_spa_prop_set_uint64(ZPOOL_PROP_DEDUPDITTO,
4629 ZIO_DEDUPDITTO_MIN + ztest_random(ZIO_DEDUPDITTO_MIN));
4631 VERIFY0(spa_prop_get(ztest_spa, &props));
4633 if (ztest_opts.zo_verbose >= 6)
4634 dump_nvlist(props, 4);
4638 (void) rw_unlock(&ztest_name_lock);
4642 user_release_one(const char *snapname, const char *holdname)
4644 nvlist_t *snaps, *holds;
4647 snaps = fnvlist_alloc();
4648 holds = fnvlist_alloc();
4649 fnvlist_add_boolean(holds, holdname);
4650 fnvlist_add_nvlist(snaps, snapname, holds);
4651 fnvlist_free(holds);
4652 error = dsl_dataset_user_release(snaps, NULL);
4653 fnvlist_free(snaps);
4658 * Test snapshot hold/release and deferred destroy.
4661 ztest_dmu_snapshot_hold(ztest_ds_t *zd, uint64_t id)
4664 objset_t *os = zd->zd_os;
4668 char clonename[100];
4670 char osname[MAXNAMELEN];
4673 (void) rw_rdlock(&ztest_name_lock);
4675 dmu_objset_name(os, osname);
4677 (void) snprintf(snapname, sizeof (snapname), "sh1_%llu", id);
4678 (void) snprintf(fullname, sizeof (fullname), "%s@%s", osname, snapname);
4679 (void) snprintf(clonename, sizeof (clonename),
4680 "%s/ch1_%llu", osname, id);
4681 (void) snprintf(tag, sizeof (tag), "tag_%llu", id);
4684 * Clean up from any previous run.
4686 error = dsl_destroy_head(clonename);
4687 if (error != ENOENT)
4689 error = user_release_one(fullname, tag);
4690 if (error != ESRCH && error != ENOENT)
4692 error = dsl_destroy_snapshot(fullname, B_FALSE);
4693 if (error != ENOENT)
4697 * Create snapshot, clone it, mark snap for deferred destroy,
4698 * destroy clone, verify snap was also destroyed.
4700 error = dmu_objset_snapshot_one(osname, snapname);
4702 if (error == ENOSPC) {
4703 ztest_record_enospc("dmu_objset_snapshot");
4706 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
4709 error = dmu_objset_clone(clonename, fullname);
4711 if (error == ENOSPC) {
4712 ztest_record_enospc("dmu_objset_clone");
4715 fatal(0, "dmu_objset_clone(%s) = %d", clonename, error);
4718 error = dsl_destroy_snapshot(fullname, B_TRUE);
4720 fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
4724 error = dsl_destroy_head(clonename);
4726 fatal(0, "dsl_destroy_head(%s) = %d", clonename, error);
4728 error = dmu_objset_hold(fullname, FTAG, &origin);
4729 if (error != ENOENT)
4730 fatal(0, "dmu_objset_hold(%s) = %d", fullname, error);
4733 * Create snapshot, add temporary hold, verify that we can't
4734 * destroy a held snapshot, mark for deferred destroy,
4735 * release hold, verify snapshot was destroyed.
4737 error = dmu_objset_snapshot_one(osname, snapname);
4739 if (error == ENOSPC) {
4740 ztest_record_enospc("dmu_objset_snapshot");
4743 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
4746 holds = fnvlist_alloc();
4747 fnvlist_add_string(holds, fullname, tag);
4748 error = dsl_dataset_user_hold(holds, 0, NULL);
4749 fnvlist_free(holds);
4751 if (error == ENOSPC) {
4752 ztest_record_enospc("dsl_dataset_user_hold");
4755 fatal(0, "dsl_dataset_user_hold(%s, %s) = %u",
4756 fullname, tag, error);
4759 error = dsl_destroy_snapshot(fullname, B_FALSE);
4760 if (error != EBUSY) {
4761 fatal(0, "dsl_destroy_snapshot(%s, B_FALSE) = %d",
4765 error = dsl_destroy_snapshot(fullname, B_TRUE);
4767 fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
4771 error = user_release_one(fullname, tag);
4773 fatal(0, "user_release_one(%s, %s) = %d", fullname, tag, error);
4775 VERIFY3U(dmu_objset_hold(fullname, FTAG, &origin), ==, ENOENT);
4778 (void) rw_unlock(&ztest_name_lock);
4782 * Inject random faults into the on-disk data.
4786 ztest_fault_inject(ztest_ds_t *zd, uint64_t id)
4788 ztest_shared_t *zs = ztest_shared;
4789 spa_t *spa = ztest_spa;
4793 uint64_t bad = 0x1990c0ffeedecadeULL;
4795 char path0[MAXPATHLEN];
4796 char pathrand[MAXPATHLEN];
4798 int bshift = SPA_OLD_MAXBLOCKSHIFT + 2; /* don't scrog all labels */
4804 boolean_t islog = B_FALSE;
4806 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
4807 maxfaults = MAXFAULTS();
4808 leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz;
4809 mirror_save = zs->zs_mirrors;
4810 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4812 ASSERT(leaves >= 1);
4815 * Grab the name lock as reader. There are some operations
4816 * which don't like to have their vdevs changed while
4817 * they are in progress (i.e. spa_change_guid). Those
4818 * operations will have grabbed the name lock as writer.
4820 (void) rw_rdlock(&ztest_name_lock);
4823 * We need SCL_STATE here because we're going to look at vd0->vdev_tsd.
4825 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
4827 if (ztest_random(2) == 0) {
4829 * Inject errors on a normal data device or slog device.
4831 top = ztest_random_vdev_top(spa, B_TRUE);
4832 leaf = ztest_random(leaves) + zs->zs_splits;
4835 * Generate paths to the first leaf in this top-level vdev,
4836 * and to the random leaf we selected. We'll induce transient
4837 * write failures and random online/offline activity on leaf 0,
4838 * and we'll write random garbage to the randomly chosen leaf.
4840 (void) snprintf(path0, sizeof (path0), ztest_dev_template,
4841 ztest_opts.zo_dir, ztest_opts.zo_pool,
4842 top * leaves + zs->zs_splits);
4843 (void) snprintf(pathrand, sizeof (pathrand), ztest_dev_template,
4844 ztest_opts.zo_dir, ztest_opts.zo_pool,
4845 top * leaves + leaf);
4847 vd0 = vdev_lookup_by_path(spa->spa_root_vdev, path0);
4848 if (vd0 != NULL && vd0->vdev_top->vdev_islog)
4852 * If the top-level vdev needs to be resilvered
4853 * then we only allow faults on the device that is
4856 if (vd0 != NULL && maxfaults != 1 &&
4857 (!vdev_resilver_needed(vd0->vdev_top, NULL, NULL) ||
4858 vd0->vdev_resilver_txg != 0)) {
4860 * Make vd0 explicitly claim to be unreadable,
4861 * or unwriteable, or reach behind its back
4862 * and close the underlying fd. We can do this if
4863 * maxfaults == 0 because we'll fail and reexecute,
4864 * and we can do it if maxfaults >= 2 because we'll
4865 * have enough redundancy. If maxfaults == 1, the
4866 * combination of this with injection of random data
4867 * corruption below exceeds the pool's fault tolerance.
4869 vdev_file_t *vf = vd0->vdev_tsd;
4871 if (vf != NULL && ztest_random(3) == 0) {
4872 (void) close(vf->vf_vnode->v_fd);
4873 vf->vf_vnode->v_fd = -1;
4874 } else if (ztest_random(2) == 0) {
4875 vd0->vdev_cant_read = B_TRUE;
4877 vd0->vdev_cant_write = B_TRUE;
4879 guid0 = vd0->vdev_guid;
4883 * Inject errors on an l2cache device.
4885 spa_aux_vdev_t *sav = &spa->spa_l2cache;
4887 if (sav->sav_count == 0) {
4888 spa_config_exit(spa, SCL_STATE, FTAG);
4889 (void) rw_unlock(&ztest_name_lock);
4892 vd0 = sav->sav_vdevs[ztest_random(sav->sav_count)];
4893 guid0 = vd0->vdev_guid;
4894 (void) strcpy(path0, vd0->vdev_path);
4895 (void) strcpy(pathrand, vd0->vdev_path);
4899 maxfaults = INT_MAX; /* no limit on cache devices */
4902 spa_config_exit(spa, SCL_STATE, FTAG);
4903 (void) rw_unlock(&ztest_name_lock);
4906 * If we can tolerate two or more faults, or we're dealing
4907 * with a slog, randomly online/offline vd0.
4909 if ((maxfaults >= 2 || islog) && guid0 != 0) {
4910 if (ztest_random(10) < 6) {
4911 int flags = (ztest_random(2) == 0 ?
4912 ZFS_OFFLINE_TEMPORARY : 0);
4915 * We have to grab the zs_name_lock as writer to
4916 * prevent a race between offlining a slog and
4917 * destroying a dataset. Offlining the slog will
4918 * grab a reference on the dataset which may cause
4919 * dmu_objset_destroy() to fail with EBUSY thus
4920 * leaving the dataset in an inconsistent state.
4923 (void) rw_wrlock(&ztest_name_lock);
4925 VERIFY(vdev_offline(spa, guid0, flags) != EBUSY);
4928 (void) rw_unlock(&ztest_name_lock);
4931 * Ideally we would like to be able to randomly
4932 * call vdev_[on|off]line without holding locks
4933 * to force unpredictable failures but the side
4934 * effects of vdev_[on|off]line prevent us from
4935 * doing so. We grab the ztest_vdev_lock here to
4936 * prevent a race between injection testing and
4939 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
4940 (void) vdev_online(spa, guid0, 0, NULL);
4941 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4949 * We have at least single-fault tolerance, so inject data corruption.
4951 fd = open(pathrand, O_RDWR);
4953 if (fd == -1) /* we hit a gap in the device namespace */
4956 fsize = lseek(fd, 0, SEEK_END);
4958 while (--iters != 0) {
4959 offset = ztest_random(fsize / (leaves << bshift)) *
4960 (leaves << bshift) + (leaf << bshift) +
4961 (ztest_random(1ULL << (bshift - 1)) & -8ULL);
4963 if (offset >= fsize)
4966 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
4967 if (mirror_save != zs->zs_mirrors) {
4968 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4973 if (pwrite(fd, &bad, sizeof (bad), offset) != sizeof (bad))
4974 fatal(1, "can't inject bad word at 0x%llx in %s",
4977 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4979 if (ztest_opts.zo_verbose >= 7)
4980 (void) printf("injected bad word into %s,"
4981 " offset 0x%llx\n", pathrand, (u_longlong_t)offset);
4988 * Verify that DDT repair works as expected.
4991 ztest_ddt_repair(ztest_ds_t *zd, uint64_t id)
4993 ztest_shared_t *zs = ztest_shared;
4994 spa_t *spa = ztest_spa;
4995 objset_t *os = zd->zd_os;
4997 uint64_t object, blocksize, txg, pattern, psize;
4998 enum zio_checksum checksum = spa_dedup_checksum(spa);
5003 int copies = 2 * ZIO_DEDUPDITTO_MIN;
5005 blocksize = ztest_random_blocksize();
5006 blocksize = MIN(blocksize, 2048); /* because we write so many */
5008 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
5010 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
5014 * Take the name lock as writer to prevent anyone else from changing
5015 * the pool and dataset properies we need to maintain during this test.
5017 (void) rw_wrlock(&ztest_name_lock);
5019 if (ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_DEDUP, checksum,
5021 ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_COPIES, 1,
5023 (void) rw_unlock(&ztest_name_lock);
5027 object = od[0].od_object;
5028 blocksize = od[0].od_blocksize;
5029 pattern = zs->zs_guid ^ dmu_objset_fsid_guid(os);
5031 ASSERT(object != 0);
5033 tx = dmu_tx_create(os);
5034 dmu_tx_hold_write(tx, object, 0, copies * blocksize);
5035 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
5037 (void) rw_unlock(&ztest_name_lock);
5042 * Write all the copies of our block.
5044 for (int i = 0; i < copies; i++) {
5045 uint64_t offset = i * blocksize;
5046 int error = dmu_buf_hold(os, object, offset, FTAG, &db,
5047 DMU_READ_NO_PREFETCH);
5049 fatal(B_FALSE, "dmu_buf_hold(%p, %llu, %llu) = %u",
5050 os, (long long)object, (long long) offset, error);
5052 ASSERT(db->db_offset == offset);
5053 ASSERT(db->db_size == blocksize);
5054 ASSERT(ztest_pattern_match(db->db_data, db->db_size, pattern) ||
5055 ztest_pattern_match(db->db_data, db->db_size, 0ULL));
5056 dmu_buf_will_fill(db, tx);
5057 ztest_pattern_set(db->db_data, db->db_size, pattern);
5058 dmu_buf_rele(db, FTAG);
5062 txg_wait_synced(spa_get_dsl(spa), txg);
5065 * Find out what block we got.
5067 VERIFY0(dmu_buf_hold(os, object, 0, FTAG, &db,
5068 DMU_READ_NO_PREFETCH));
5069 blk = *((dmu_buf_impl_t *)db)->db_blkptr;
5070 dmu_buf_rele(db, FTAG);
5073 * Damage the block. Dedup-ditto will save us when we read it later.
5075 psize = BP_GET_PSIZE(&blk);
5076 buf = zio_buf_alloc(psize);
5077 ztest_pattern_set(buf, psize, ~pattern);
5079 (void) zio_wait(zio_rewrite(NULL, spa, 0, &blk,
5080 buf, psize, NULL, NULL, ZIO_PRIORITY_SYNC_WRITE,
5081 ZIO_FLAG_CANFAIL | ZIO_FLAG_INDUCE_DAMAGE, NULL));
5083 zio_buf_free(buf, psize);
5085 (void) rw_unlock(&ztest_name_lock);
5093 ztest_scrub(ztest_ds_t *zd, uint64_t id)
5095 spa_t *spa = ztest_spa;
5097 (void) spa_scan(spa, POOL_SCAN_SCRUB);
5098 (void) poll(NULL, 0, 100); /* wait a moment, then force a restart */
5099 (void) spa_scan(spa, POOL_SCAN_SCRUB);
5103 * Change the guid for the pool.
5107 ztest_reguid(ztest_ds_t *zd, uint64_t id)
5109 spa_t *spa = ztest_spa;
5110 uint64_t orig, load;
5113 orig = spa_guid(spa);
5114 load = spa_load_guid(spa);
5116 (void) rw_wrlock(&ztest_name_lock);
5117 error = spa_change_guid(spa);
5118 (void) rw_unlock(&ztest_name_lock);
5123 if (ztest_opts.zo_verbose >= 4) {
5124 (void) printf("Changed guid old %llu -> %llu\n",
5125 (u_longlong_t)orig, (u_longlong_t)spa_guid(spa));
5128 VERIFY3U(orig, !=, spa_guid(spa));
5129 VERIFY3U(load, ==, spa_load_guid(spa));
5133 * Rename the pool to a different name and then rename it back.
5137 ztest_spa_rename(ztest_ds_t *zd, uint64_t id)
5139 char *oldname, *newname;
5142 (void) rw_wrlock(&ztest_name_lock);
5144 oldname = ztest_opts.zo_pool;
5145 newname = umem_alloc(strlen(oldname) + 5, UMEM_NOFAIL);
5146 (void) strcpy(newname, oldname);
5147 (void) strcat(newname, "_tmp");
5152 VERIFY3U(0, ==, spa_rename(oldname, newname));
5155 * Try to open it under the old name, which shouldn't exist
5157 VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
5160 * Open it under the new name and make sure it's still the same spa_t.
5162 VERIFY3U(0, ==, spa_open(newname, &spa, FTAG));
5164 ASSERT(spa == ztest_spa);
5165 spa_close(spa, FTAG);
5168 * Rename it back to the original
5170 VERIFY3U(0, ==, spa_rename(newname, oldname));
5173 * Make sure it can still be opened
5175 VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG));
5177 ASSERT(spa == ztest_spa);
5178 spa_close(spa, FTAG);
5180 umem_free(newname, strlen(newname) + 1);
5182 (void) rw_unlock(&ztest_name_lock);
5186 * Verify pool integrity by running zdb.
5189 ztest_run_zdb(char *pool)
5192 char zdb[MAXPATHLEN + MAXNAMELEN + 20];
5200 strlcpy(zdb, "/usr/bin/ztest", sizeof(zdb));
5202 /* zdb lives in /usr/sbin, while ztest lives in /usr/bin */
5203 bin = strstr(zdb, "/usr/bin/");
5204 ztest = strstr(bin, "/ztest");
5206 isalen = ztest - isa;
5210 "/usr/sbin%.*s/zdb -bcc%s%s -d -U %s %s",
5213 ztest_opts.zo_verbose >= 3 ? "s" : "",
5214 ztest_opts.zo_verbose >= 4 ? "v" : "",
5219 if (ztest_opts.zo_verbose >= 5)
5220 (void) printf("Executing %s\n", strstr(zdb, "zdb "));
5222 fp = popen(zdb, "r");
5225 while (fgets(zbuf, sizeof (zbuf), fp) != NULL)
5226 if (ztest_opts.zo_verbose >= 3)
5227 (void) printf("%s", zbuf);
5229 status = pclose(fp);
5234 ztest_dump_core = 0;
5235 if (WIFEXITED(status))
5236 fatal(0, "'%s' exit code %d", zdb, WEXITSTATUS(status));
5238 fatal(0, "'%s' died with signal %d", zdb, WTERMSIG(status));
5242 ztest_walk_pool_directory(char *header)
5246 if (ztest_opts.zo_verbose >= 6)
5247 (void) printf("%s\n", header);
5249 mutex_enter(&spa_namespace_lock);
5250 while ((spa = spa_next(spa)) != NULL)
5251 if (ztest_opts.zo_verbose >= 6)
5252 (void) printf("\t%s\n", spa_name(spa));
5253 mutex_exit(&spa_namespace_lock);
5257 ztest_spa_import_export(char *oldname, char *newname)
5259 nvlist_t *config, *newconfig;
5264 if (ztest_opts.zo_verbose >= 4) {
5265 (void) printf("import/export: old = %s, new = %s\n",
5270 * Clean up from previous runs.
5272 (void) spa_destroy(newname);
5275 * Get the pool's configuration and guid.
5277 VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG));
5280 * Kick off a scrub to tickle scrub/export races.
5282 if (ztest_random(2) == 0)
5283 (void) spa_scan(spa, POOL_SCAN_SCRUB);
5285 pool_guid = spa_guid(spa);
5286 spa_close(spa, FTAG);
5288 ztest_walk_pool_directory("pools before export");
5293 VERIFY3U(0, ==, spa_export(oldname, &config, B_FALSE, B_FALSE));
5295 ztest_walk_pool_directory("pools after export");
5300 newconfig = spa_tryimport(config);
5301 ASSERT(newconfig != NULL);
5302 nvlist_free(newconfig);
5305 * Import it under the new name.
5307 error = spa_import(newname, config, NULL, 0);
5309 dump_nvlist(config, 0);
5310 fatal(B_FALSE, "couldn't import pool %s as %s: error %u",
5311 oldname, newname, error);
5314 ztest_walk_pool_directory("pools after import");
5317 * Try to import it again -- should fail with EEXIST.
5319 VERIFY3U(EEXIST, ==, spa_import(newname, config, NULL, 0));
5322 * Try to import it under a different name -- should fail with EEXIST.
5324 VERIFY3U(EEXIST, ==, spa_import(oldname, config, NULL, 0));
5327 * Verify that the pool is no longer visible under the old name.
5329 VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
5332 * Verify that we can open and close the pool using the new name.
5334 VERIFY3U(0, ==, spa_open(newname, &spa, FTAG));
5335 ASSERT(pool_guid == spa_guid(spa));
5336 spa_close(spa, FTAG);
5338 nvlist_free(config);
5342 ztest_resume(spa_t *spa)
5344 if (spa_suspended(spa) && ztest_opts.zo_verbose >= 6)
5345 (void) printf("resuming from suspended state\n");
5346 spa_vdev_state_enter(spa, SCL_NONE);
5347 vdev_clear(spa, NULL);
5348 (void) spa_vdev_state_exit(spa, NULL, 0);
5349 (void) zio_resume(spa);
5353 ztest_resume_thread(void *arg)
5357 while (!ztest_exiting) {
5358 if (spa_suspended(spa))
5360 (void) poll(NULL, 0, 100);
5366 ztest_deadman_thread(void *arg)
5368 ztest_shared_t *zs = arg;
5369 spa_t *spa = ztest_spa;
5370 hrtime_t delta, total = 0;
5373 delta = zs->zs_thread_stop - zs->zs_thread_start +
5374 MSEC2NSEC(zfs_deadman_synctime_ms);
5376 (void) poll(NULL, 0, (int)NSEC2MSEC(delta));
5379 * If the pool is suspended then fail immediately. Otherwise,
5380 * check to see if the pool is making any progress. If
5381 * vdev_deadman() discovers that there hasn't been any recent
5382 * I/Os then it will end up aborting the tests.
5384 if (spa_suspended(spa) || spa->spa_root_vdev == NULL) {
5385 fatal(0, "aborting test after %llu seconds because "
5386 "pool has transitioned to a suspended state.",
5387 zfs_deadman_synctime_ms / 1000);
5390 vdev_deadman(spa->spa_root_vdev);
5392 total += zfs_deadman_synctime_ms/1000;
5393 (void) printf("ztest has been running for %lld seconds\n",
5399 ztest_execute(int test, ztest_info_t *zi, uint64_t id)
5401 ztest_ds_t *zd = &ztest_ds[id % ztest_opts.zo_datasets];
5402 ztest_shared_callstate_t *zc = ZTEST_GET_SHARED_CALLSTATE(test);
5403 hrtime_t functime = gethrtime();
5405 for (int i = 0; i < zi->zi_iters; i++)
5406 zi->zi_func(zd, id);
5408 functime = gethrtime() - functime;
5410 atomic_add_64(&zc->zc_count, 1);
5411 atomic_add_64(&zc->zc_time, functime);
5413 if (ztest_opts.zo_verbose >= 4) {
5415 (void) dladdr((void *)zi->zi_func, &dli);
5416 (void) printf("%6.2f sec in %s\n",
5417 (double)functime / NANOSEC, dli.dli_sname);
5422 ztest_thread(void *arg)
5425 uint64_t id = (uintptr_t)arg;
5426 ztest_shared_t *zs = ztest_shared;
5430 ztest_shared_callstate_t *zc;
5432 while ((now = gethrtime()) < zs->zs_thread_stop) {
5434 * See if it's time to force a crash.
5436 if (now > zs->zs_thread_kill)
5440 * If we're getting ENOSPC with some regularity, stop.
5442 if (zs->zs_enospc_count > 10)
5446 * Pick a random function to execute.
5448 rand = ztest_random(ZTEST_FUNCS);
5449 zi = &ztest_info[rand];
5450 zc = ZTEST_GET_SHARED_CALLSTATE(rand);
5451 call_next = zc->zc_next;
5453 if (now >= call_next &&
5454 atomic_cas_64(&zc->zc_next, call_next, call_next +
5455 ztest_random(2 * zi->zi_interval[0] + 1)) == call_next) {
5456 ztest_execute(rand, zi, id);
5464 ztest_dataset_name(char *dsname, char *pool, int d)
5466 (void) snprintf(dsname, MAXNAMELEN, "%s/ds_%d", pool, d);
5470 ztest_dataset_destroy(int d)
5472 char name[MAXNAMELEN];
5474 ztest_dataset_name(name, ztest_opts.zo_pool, d);
5476 if (ztest_opts.zo_verbose >= 3)
5477 (void) printf("Destroying %s to free up space\n", name);
5480 * Cleanup any non-standard clones and snapshots. In general,
5481 * ztest thread t operates on dataset (t % zopt_datasets),
5482 * so there may be more than one thing to clean up.
5484 for (int t = d; t < ztest_opts.zo_threads;
5485 t += ztest_opts.zo_datasets) {
5486 ztest_dsl_dataset_cleanup(name, t);
5489 (void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
5490 DS_FIND_SNAPSHOTS | DS_FIND_CHILDREN);
5494 ztest_dataset_dirobj_verify(ztest_ds_t *zd)
5496 uint64_t usedobjs, dirobjs, scratch;
5499 * ZTEST_DIROBJ is the object directory for the entire dataset.
5500 * Therefore, the number of objects in use should equal the
5501 * number of ZTEST_DIROBJ entries, +1 for ZTEST_DIROBJ itself.
5502 * If not, we have an object leak.
5504 * Note that we can only check this in ztest_dataset_open(),
5505 * when the open-context and syncing-context values agree.
5506 * That's because zap_count() returns the open-context value,
5507 * while dmu_objset_space() returns the rootbp fill count.
5509 VERIFY3U(0, ==, zap_count(zd->zd_os, ZTEST_DIROBJ, &dirobjs));
5510 dmu_objset_space(zd->zd_os, &scratch, &scratch, &usedobjs, &scratch);
5511 ASSERT3U(dirobjs + 1, ==, usedobjs);
5515 ztest_dataset_open(int d)
5517 ztest_ds_t *zd = &ztest_ds[d];
5518 uint64_t committed_seq = ZTEST_GET_SHARED_DS(d)->zd_seq;
5521 char name[MAXNAMELEN];
5524 ztest_dataset_name(name, ztest_opts.zo_pool, d);
5526 (void) rw_rdlock(&ztest_name_lock);
5528 error = ztest_dataset_create(name);
5529 if (error == ENOSPC) {
5530 (void) rw_unlock(&ztest_name_lock);
5531 ztest_record_enospc(FTAG);
5534 ASSERT(error == 0 || error == EEXIST);
5536 VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, zd, &os));
5537 (void) rw_unlock(&ztest_name_lock);
5539 ztest_zd_init(zd, ZTEST_GET_SHARED_DS(d), os);
5541 zilog = zd->zd_zilog;
5543 if (zilog->zl_header->zh_claim_lr_seq != 0 &&
5544 zilog->zl_header->zh_claim_lr_seq < committed_seq)
5545 fatal(0, "missing log records: claimed %llu < committed %llu",
5546 zilog->zl_header->zh_claim_lr_seq, committed_seq);
5548 ztest_dataset_dirobj_verify(zd);
5550 zil_replay(os, zd, ztest_replay_vector);
5552 ztest_dataset_dirobj_verify(zd);
5554 if (ztest_opts.zo_verbose >= 6)
5555 (void) printf("%s replay %llu blocks, %llu records, seq %llu\n",
5557 (u_longlong_t)zilog->zl_parse_blk_count,
5558 (u_longlong_t)zilog->zl_parse_lr_count,
5559 (u_longlong_t)zilog->zl_replaying_seq);
5561 zilog = zil_open(os, ztest_get_data);
5563 if (zilog->zl_replaying_seq != 0 &&
5564 zilog->zl_replaying_seq < committed_seq)
5565 fatal(0, "missing log records: replayed %llu < committed %llu",
5566 zilog->zl_replaying_seq, committed_seq);
5572 ztest_dataset_close(int d)
5574 ztest_ds_t *zd = &ztest_ds[d];
5576 zil_close(zd->zd_zilog);
5577 dmu_objset_disown(zd->zd_os, zd);
5583 * Kick off threads to run tests on all datasets in parallel.
5586 ztest_run(ztest_shared_t *zs)
5591 thread_t resume_tid;
5594 ztest_exiting = B_FALSE;
5597 * Initialize parent/child shared state.
5599 VERIFY(_mutex_init(&ztest_vdev_lock, USYNC_THREAD, NULL) == 0);
5600 VERIFY(rwlock_init(&ztest_name_lock, USYNC_THREAD, NULL) == 0);
5602 zs->zs_thread_start = gethrtime();
5603 zs->zs_thread_stop =
5604 zs->zs_thread_start + ztest_opts.zo_passtime * NANOSEC;
5605 zs->zs_thread_stop = MIN(zs->zs_thread_stop, zs->zs_proc_stop);
5606 zs->zs_thread_kill = zs->zs_thread_stop;
5607 if (ztest_random(100) < ztest_opts.zo_killrate) {
5608 zs->zs_thread_kill -=
5609 ztest_random(ztest_opts.zo_passtime * NANOSEC);
5612 (void) _mutex_init(&zcl.zcl_callbacks_lock, USYNC_THREAD, NULL);
5614 list_create(&zcl.zcl_callbacks, sizeof (ztest_cb_data_t),
5615 offsetof(ztest_cb_data_t, zcd_node));
5620 kernel_init(FREAD | FWRITE);
5621 VERIFY0(spa_open(ztest_opts.zo_pool, &spa, FTAG));
5622 spa->spa_debug = B_TRUE;
5623 metaslab_preload_limit = ztest_random(20) + 1;
5626 VERIFY0(dmu_objset_own(ztest_opts.zo_pool,
5627 DMU_OST_ANY, B_TRUE, FTAG, &os));
5628 zs->zs_guid = dmu_objset_fsid_guid(os);
5629 dmu_objset_disown(os, FTAG);
5631 spa->spa_dedup_ditto = 2 * ZIO_DEDUPDITTO_MIN;
5634 * We don't expect the pool to suspend unless maxfaults == 0,
5635 * in which case ztest_fault_inject() temporarily takes away
5636 * the only valid replica.
5638 if (MAXFAULTS() == 0)
5639 spa->spa_failmode = ZIO_FAILURE_MODE_WAIT;
5641 spa->spa_failmode = ZIO_FAILURE_MODE_PANIC;
5644 * Create a thread to periodically resume suspended I/O.
5646 VERIFY(thr_create(0, 0, ztest_resume_thread, spa, THR_BOUND,
5650 * Create a deadman thread to abort() if we hang.
5652 VERIFY(thr_create(0, 0, ztest_deadman_thread, zs, THR_BOUND,
5656 * Verify that we can safely inquire about about any object,
5657 * whether it's allocated or not. To make it interesting,
5658 * we probe a 5-wide window around each power of two.
5659 * This hits all edge cases, including zero and the max.
5661 for (int t = 0; t < 64; t++) {
5662 for (int d = -5; d <= 5; d++) {
5663 error = dmu_object_info(spa->spa_meta_objset,
5664 (1ULL << t) + d, NULL);
5665 ASSERT(error == 0 || error == ENOENT ||
5671 * If we got any ENOSPC errors on the previous run, destroy something.
5673 if (zs->zs_enospc_count != 0) {
5674 int d = ztest_random(ztest_opts.zo_datasets);
5675 ztest_dataset_destroy(d);
5677 zs->zs_enospc_count = 0;
5679 tid = umem_zalloc(ztest_opts.zo_threads * sizeof (thread_t),
5682 if (ztest_opts.zo_verbose >= 4)
5683 (void) printf("starting main threads...\n");
5686 * Kick off all the tests that run in parallel.
5688 for (int t = 0; t < ztest_opts.zo_threads; t++) {
5689 if (t < ztest_opts.zo_datasets &&
5690 ztest_dataset_open(t) != 0)
5692 VERIFY(thr_create(0, 0, ztest_thread, (void *)(uintptr_t)t,
5693 THR_BOUND, &tid[t]) == 0);
5697 * Wait for all of the tests to complete. We go in reverse order
5698 * so we don't close datasets while threads are still using them.
5700 for (int t = ztest_opts.zo_threads - 1; t >= 0; t--) {
5701 VERIFY(thr_join(tid[t], NULL, NULL) == 0);
5702 if (t < ztest_opts.zo_datasets)
5703 ztest_dataset_close(t);
5706 txg_wait_synced(spa_get_dsl(spa), 0);
5708 zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(spa));
5709 zs->zs_space = metaslab_class_get_space(spa_normal_class(spa));
5710 zfs_dbgmsg_print(FTAG);
5712 umem_free(tid, ztest_opts.zo_threads * sizeof (thread_t));
5714 /* Kill the resume thread */
5715 ztest_exiting = B_TRUE;
5716 VERIFY(thr_join(resume_tid, NULL, NULL) == 0);
5720 * Right before closing the pool, kick off a bunch of async I/O;
5721 * spa_close() should wait for it to complete.
5723 for (uint64_t object = 1; object < 50; object++)
5724 dmu_prefetch(spa->spa_meta_objset, object, 0, 1ULL << 20);
5726 spa_close(spa, FTAG);
5729 * Verify that we can loop over all pools.
5731 mutex_enter(&spa_namespace_lock);
5732 for (spa = spa_next(NULL); spa != NULL; spa = spa_next(spa))
5733 if (ztest_opts.zo_verbose > 3)
5734 (void) printf("spa_next: found %s\n", spa_name(spa));
5735 mutex_exit(&spa_namespace_lock);
5738 * Verify that we can export the pool and reimport it under a
5741 if (ztest_random(2) == 0) {
5742 char name[MAXNAMELEN];
5743 (void) snprintf(name, MAXNAMELEN, "%s_import",
5744 ztest_opts.zo_pool);
5745 ztest_spa_import_export(ztest_opts.zo_pool, name);
5746 ztest_spa_import_export(name, ztest_opts.zo_pool);
5751 list_destroy(&zcl.zcl_callbacks);
5753 (void) _mutex_destroy(&zcl.zcl_callbacks_lock);
5755 (void) rwlock_destroy(&ztest_name_lock);
5756 (void) _mutex_destroy(&ztest_vdev_lock);
5762 ztest_ds_t *zd = &ztest_ds[0];
5766 if (ztest_opts.zo_verbose >= 3)
5767 (void) printf("testing spa_freeze()...\n");
5769 kernel_init(FREAD | FWRITE);
5770 VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
5771 VERIFY3U(0, ==, ztest_dataset_open(0));
5772 spa->spa_debug = B_TRUE;
5776 * Force the first log block to be transactionally allocated.
5777 * We have to do this before we freeze the pool -- otherwise
5778 * the log chain won't be anchored.
5780 while (BP_IS_HOLE(&zd->zd_zilog->zl_header->zh_log)) {
5781 ztest_dmu_object_alloc_free(zd, 0);
5782 zil_commit(zd->zd_zilog, 0);
5785 txg_wait_synced(spa_get_dsl(spa), 0);
5788 * Freeze the pool. This stops spa_sync() from doing anything,
5789 * so that the only way to record changes from now on is the ZIL.
5794 * Because it is hard to predict how much space a write will actually
5795 * require beforehand, we leave ourselves some fudge space to write over
5798 uint64_t capacity = metaslab_class_get_space(spa_normal_class(spa)) / 2;
5801 * Run tests that generate log records but don't alter the pool config
5802 * or depend on DSL sync tasks (snapshots, objset create/destroy, etc).
5803 * We do a txg_wait_synced() after each iteration to force the txg
5804 * to increase well beyond the last synced value in the uberblock.
5805 * The ZIL should be OK with that.
5807 * Run a random number of times less than zo_maxloops and ensure we do
5808 * not run out of space on the pool.
5810 while (ztest_random(10) != 0 &&
5811 numloops++ < ztest_opts.zo_maxloops &&
5812 metaslab_class_get_alloc(spa_normal_class(spa)) < capacity) {
5814 ztest_od_init(&od, 0, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
5815 VERIFY0(ztest_object_init(zd, &od, sizeof (od), B_FALSE));
5816 ztest_io(zd, od.od_object,
5817 ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
5818 txg_wait_synced(spa_get_dsl(spa), 0);
5822 * Commit all of the changes we just generated.
5824 zil_commit(zd->zd_zilog, 0);
5825 txg_wait_synced(spa_get_dsl(spa), 0);
5828 * Close our dataset and close the pool.
5830 ztest_dataset_close(0);
5831 spa_close(spa, FTAG);
5835 * Open and close the pool and dataset to induce log replay.
5837 kernel_init(FREAD | FWRITE);
5838 VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
5839 ASSERT(spa_freeze_txg(spa) == UINT64_MAX);
5840 VERIFY3U(0, ==, ztest_dataset_open(0));
5841 ztest_dataset_close(0);
5843 spa->spa_debug = B_TRUE;
5845 txg_wait_synced(spa_get_dsl(spa), 0);
5846 ztest_reguid(NULL, 0);
5848 spa_close(spa, FTAG);
5853 print_time(hrtime_t t, char *timebuf)
5855 hrtime_t s = t / NANOSEC;
5856 hrtime_t m = s / 60;
5857 hrtime_t h = m / 60;
5858 hrtime_t d = h / 24;
5867 (void) sprintf(timebuf,
5868 "%llud%02lluh%02llum%02llus", d, h, m, s);
5870 (void) sprintf(timebuf, "%lluh%02llum%02llus", h, m, s);
5872 (void) sprintf(timebuf, "%llum%02llus", m, s);
5874 (void) sprintf(timebuf, "%llus", s);
5882 VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
5883 if (ztest_random(2) == 0)
5885 VERIFY(nvlist_add_uint64(props, "autoreplace", 1) == 0);
5891 * Create a storage pool with the given name and initial vdev size.
5892 * Then test spa_freeze() functionality.
5895 ztest_init(ztest_shared_t *zs)
5898 nvlist_t *nvroot, *props;
5900 VERIFY(_mutex_init(&ztest_vdev_lock, USYNC_THREAD, NULL) == 0);
5901 VERIFY(rwlock_init(&ztest_name_lock, USYNC_THREAD, NULL) == 0);
5903 kernel_init(FREAD | FWRITE);
5906 * Create the storage pool.
5908 (void) spa_destroy(ztest_opts.zo_pool);
5909 ztest_shared->zs_vdev_next_leaf = 0;
5911 zs->zs_mirrors = ztest_opts.zo_mirrors;
5912 nvroot = make_vdev_root(NULL, NULL, NULL, ztest_opts.zo_vdev_size, 0,
5913 0, ztest_opts.zo_raidz, zs->zs_mirrors, 1);
5914 props = make_random_props();
5915 for (int i = 0; i < SPA_FEATURES; i++) {
5917 (void) snprintf(buf, sizeof (buf), "feature@%s",
5918 spa_feature_table[i].fi_uname);
5919 VERIFY3U(0, ==, nvlist_add_uint64(props, buf, 0));
5921 VERIFY3U(0, ==, spa_create(ztest_opts.zo_pool, nvroot, props, NULL));
5922 nvlist_free(nvroot);
5924 VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
5925 zs->zs_metaslab_sz =
5926 1ULL << spa->spa_root_vdev->vdev_child[0]->vdev_ms_shift;
5928 spa_close(spa, FTAG);
5932 ztest_run_zdb(ztest_opts.zo_pool);
5936 ztest_run_zdb(ztest_opts.zo_pool);
5938 (void) rwlock_destroy(&ztest_name_lock);
5939 (void) _mutex_destroy(&ztest_vdev_lock);
5945 static char ztest_name_data[] = "/tmp/ztest.data.XXXXXX";
5947 ztest_fd_data = mkstemp(ztest_name_data);
5948 ASSERT3S(ztest_fd_data, >=, 0);
5949 (void) unlink(ztest_name_data);
5954 shared_data_size(ztest_shared_hdr_t *hdr)
5958 size = hdr->zh_hdr_size;
5959 size += hdr->zh_opts_size;
5960 size += hdr->zh_size;
5961 size += hdr->zh_stats_size * hdr->zh_stats_count;
5962 size += hdr->zh_ds_size * hdr->zh_ds_count;
5971 ztest_shared_hdr_t *hdr;
5973 hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
5974 PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
5975 ASSERT(hdr != MAP_FAILED);
5977 VERIFY3U(0, ==, ftruncate(ztest_fd_data, sizeof (ztest_shared_hdr_t)));
5979 hdr->zh_hdr_size = sizeof (ztest_shared_hdr_t);
5980 hdr->zh_opts_size = sizeof (ztest_shared_opts_t);
5981 hdr->zh_size = sizeof (ztest_shared_t);
5982 hdr->zh_stats_size = sizeof (ztest_shared_callstate_t);
5983 hdr->zh_stats_count = ZTEST_FUNCS;
5984 hdr->zh_ds_size = sizeof (ztest_shared_ds_t);
5985 hdr->zh_ds_count = ztest_opts.zo_datasets;
5987 size = shared_data_size(hdr);
5988 VERIFY3U(0, ==, ftruncate(ztest_fd_data, size));
5990 (void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
5997 ztest_shared_hdr_t *hdr;
6000 hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
6001 PROT_READ, MAP_SHARED, ztest_fd_data, 0);
6002 ASSERT(hdr != MAP_FAILED);
6004 size = shared_data_size(hdr);
6006 (void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
6007 hdr = ztest_shared_hdr = (void *)mmap(0, P2ROUNDUP(size, getpagesize()),
6008 PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
6009 ASSERT(hdr != MAP_FAILED);
6010 buf = (uint8_t *)hdr;
6012 offset = hdr->zh_hdr_size;
6013 ztest_shared_opts = (void *)&buf[offset];
6014 offset += hdr->zh_opts_size;
6015 ztest_shared = (void *)&buf[offset];
6016 offset += hdr->zh_size;
6017 ztest_shared_callstate = (void *)&buf[offset];
6018 offset += hdr->zh_stats_size * hdr->zh_stats_count;
6019 ztest_shared_ds = (void *)&buf[offset];
6023 exec_child(char *cmd, char *libpath, boolean_t ignorekill, int *statusp)
6027 char *cmdbuf = NULL;
6032 cmdbuf = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
6033 (void) strlcpy(cmdbuf, getexecname(), MAXPATHLEN);
6038 fatal(1, "fork failed");
6040 if (pid == 0) { /* child */
6041 char *emptyargv[2] = { cmd, NULL };
6042 char fd_data_str[12];
6044 struct rlimit rl = { 1024, 1024 };
6045 (void) setrlimit(RLIMIT_NOFILE, &rl);
6047 (void) close(ztest_fd_rand);
6049 snprintf(fd_data_str, 12, "%d", ztest_fd_data));
6050 VERIFY0(setenv("ZTEST_FD_DATA", fd_data_str, 1));
6052 (void) enable_extended_FILE_stdio(-1, -1);
6053 if (libpath != NULL)
6054 VERIFY(0 == setenv("LD_LIBRARY_PATH", libpath, 1));
6056 (void) execv(cmd, emptyargv);
6058 (void) execvp(cmd, emptyargv);
6060 ztest_dump_core = B_FALSE;
6061 fatal(B_TRUE, "exec failed: %s", cmd);
6064 if (cmdbuf != NULL) {
6065 umem_free(cmdbuf, MAXPATHLEN);
6069 while (waitpid(pid, &status, 0) != pid)
6071 if (statusp != NULL)
6074 if (WIFEXITED(status)) {
6075 if (WEXITSTATUS(status) != 0) {
6076 (void) fprintf(stderr, "child exited with code %d\n",
6077 WEXITSTATUS(status));
6081 } else if (WIFSIGNALED(status)) {
6082 if (!ignorekill || WTERMSIG(status) != SIGKILL) {
6083 (void) fprintf(stderr, "child died with signal %d\n",
6089 (void) fprintf(stderr, "something strange happened to child\n");
6096 ztest_run_init(void)
6098 ztest_shared_t *zs = ztest_shared;
6100 ASSERT(ztest_opts.zo_init != 0);
6103 * Blow away any existing copy of zpool.cache
6105 (void) remove(spa_config_path);
6108 * Create and initialize our storage pool.
6110 for (int i = 1; i <= ztest_opts.zo_init; i++) {
6111 bzero(zs, sizeof (ztest_shared_t));
6112 if (ztest_opts.zo_verbose >= 3 &&
6113 ztest_opts.zo_init != 1) {
6114 (void) printf("ztest_init(), pass %d\n", i);
6121 main(int argc, char **argv)
6129 ztest_shared_callstate_t *zc;
6135 char *fd_data_str = getenv("ZTEST_FD_DATA");
6137 (void) setvbuf(stdout, NULL, _IOLBF, 0);
6139 dprintf_setup(&argc, argv);
6140 zfs_deadman_synctime_ms = 300000;
6142 ztest_fd_rand = open("/dev/urandom", O_RDONLY);
6143 ASSERT3S(ztest_fd_rand, >=, 0);
6146 process_options(argc, argv);
6151 bcopy(&ztest_opts, ztest_shared_opts,
6152 sizeof (*ztest_shared_opts));
6154 ztest_fd_data = atoi(fd_data_str);
6156 bcopy(ztest_shared_opts, &ztest_opts, sizeof (ztest_opts));
6158 ASSERT3U(ztest_opts.zo_datasets, ==, ztest_shared_hdr->zh_ds_count);
6160 /* Override location of zpool.cache */
6161 VERIFY3U(asprintf((char **)&spa_config_path, "%s/zpool.cache",
6162 ztest_opts.zo_dir), !=, -1);
6164 ztest_ds = umem_alloc(ztest_opts.zo_datasets * sizeof (ztest_ds_t),
6169 metaslab_gang_bang = ztest_opts.zo_metaslab_gang_bang;
6170 metaslab_df_alloc_threshold =
6171 zs->zs_metaslab_df_alloc_threshold;
6180 hasalt = (strlen(ztest_opts.zo_alt_ztest) != 0);
6182 if (ztest_opts.zo_verbose >= 1) {
6183 (void) printf("%llu vdevs, %d datasets, %d threads,"
6184 " %llu seconds...\n",
6185 (u_longlong_t)ztest_opts.zo_vdevs,
6186 ztest_opts.zo_datasets,
6187 ztest_opts.zo_threads,
6188 (u_longlong_t)ztest_opts.zo_time);
6191 cmd = umem_alloc(MAXNAMELEN, UMEM_NOFAIL);
6192 (void) strlcpy(cmd, getexecname(), MAXNAMELEN);
6194 zs->zs_do_init = B_TRUE;
6195 if (strlen(ztest_opts.zo_alt_ztest) != 0) {
6196 if (ztest_opts.zo_verbose >= 1) {
6197 (void) printf("Executing older ztest for "
6198 "initialization: %s\n", ztest_opts.zo_alt_ztest);
6200 VERIFY(!exec_child(ztest_opts.zo_alt_ztest,
6201 ztest_opts.zo_alt_libpath, B_FALSE, NULL));
6203 VERIFY(!exec_child(NULL, NULL, B_FALSE, NULL));
6205 zs->zs_do_init = B_FALSE;
6207 zs->zs_proc_start = gethrtime();
6208 zs->zs_proc_stop = zs->zs_proc_start + ztest_opts.zo_time * NANOSEC;
6210 for (int f = 0; f < ZTEST_FUNCS; f++) {
6211 zi = &ztest_info[f];
6212 zc = ZTEST_GET_SHARED_CALLSTATE(f);
6213 if (zs->zs_proc_start + zi->zi_interval[0] > zs->zs_proc_stop)
6214 zc->zc_next = UINT64_MAX;
6216 zc->zc_next = zs->zs_proc_start +
6217 ztest_random(2 * zi->zi_interval[0] + 1);
6221 * Run the tests in a loop. These tests include fault injection
6222 * to verify that self-healing data works, and forced crashes
6223 * to verify that we never lose on-disk consistency.
6225 while (gethrtime() < zs->zs_proc_stop) {
6230 * Initialize the workload counters for each function.
6232 for (int f = 0; f < ZTEST_FUNCS; f++) {
6233 zc = ZTEST_GET_SHARED_CALLSTATE(f);
6238 /* Set the allocation switch size */
6239 zs->zs_metaslab_df_alloc_threshold =
6240 ztest_random(zs->zs_metaslab_sz / 4) + 1;
6242 if (!hasalt || ztest_random(2) == 0) {
6243 if (hasalt && ztest_opts.zo_verbose >= 1) {
6244 (void) printf("Executing newer ztest: %s\n",
6248 killed = exec_child(cmd, NULL, B_TRUE, &status);
6250 if (hasalt && ztest_opts.zo_verbose >= 1) {
6251 (void) printf("Executing older ztest: %s\n",
6252 ztest_opts.zo_alt_ztest);
6255 killed = exec_child(ztest_opts.zo_alt_ztest,
6256 ztest_opts.zo_alt_libpath, B_TRUE, &status);
6263 if (ztest_opts.zo_verbose >= 1) {
6264 hrtime_t now = gethrtime();
6266 now = MIN(now, zs->zs_proc_stop);
6267 print_time(zs->zs_proc_stop - now, timebuf);
6268 nicenum(zs->zs_space, numbuf);
6270 (void) printf("Pass %3d, %8s, %3llu ENOSPC, "
6271 "%4.1f%% of %5s used, %3.0f%% done, %8s to go\n",
6273 WIFEXITED(status) ? "Complete" : "SIGKILL",
6274 (u_longlong_t)zs->zs_enospc_count,
6275 100.0 * zs->zs_alloc / zs->zs_space,
6277 100.0 * (now - zs->zs_proc_start) /
6278 (ztest_opts.zo_time * NANOSEC), timebuf);
6281 if (ztest_opts.zo_verbose >= 2) {
6282 (void) printf("\nWorkload summary:\n\n");
6283 (void) printf("%7s %9s %s\n",
6284 "Calls", "Time", "Function");
6285 (void) printf("%7s %9s %s\n",
6286 "-----", "----", "--------");
6287 for (int f = 0; f < ZTEST_FUNCS; f++) {
6290 zi = &ztest_info[f];
6291 zc = ZTEST_GET_SHARED_CALLSTATE(f);
6292 print_time(zc->zc_time, timebuf);
6293 (void) dladdr((void *)zi->zi_func, &dli);
6294 (void) printf("%7llu %9s %s\n",
6295 (u_longlong_t)zc->zc_count, timebuf,
6298 (void) printf("\n");
6302 * It's possible that we killed a child during a rename test,
6303 * in which case we'll have a 'ztest_tmp' pool lying around
6304 * instead of 'ztest'. Do a blind rename in case this happened.
6307 if (spa_open(ztest_opts.zo_pool, &spa, FTAG) == 0) {
6308 spa_close(spa, FTAG);
6310 char tmpname[MAXNAMELEN];
6312 kernel_init(FREAD | FWRITE);
6313 (void) snprintf(tmpname, sizeof (tmpname), "%s_tmp",
6314 ztest_opts.zo_pool);
6315 (void) spa_rename(tmpname, ztest_opts.zo_pool);
6319 ztest_run_zdb(ztest_opts.zo_pool);
6322 if (ztest_opts.zo_verbose >= 1) {
6324 (void) printf("%d runs of older ztest: %s\n", older,
6325 ztest_opts.zo_alt_ztest);
6326 (void) printf("%d runs of newer ztest: %s\n", newer,
6329 (void) printf("%d killed, %d completed, %.0f%% kill rate\n",
6330 kills, iters - kills, (100.0 * kills) / MAX(1, iters));
6333 umem_free(cmd, MAXNAMELEN);