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.
26 * The objective of this program is to provide a DMU/ZAP/SPA stress test
27 * that runs entirely in userland, is easy to use, and easy to extend.
29 * The overall design of the ztest program is as follows:
31 * (1) For each major functional area (e.g. adding vdevs to a pool,
32 * creating and destroying datasets, reading and writing objects, etc)
33 * we have a simple routine to test that functionality. These
34 * individual routines do not have to do anything "stressful".
36 * (2) We turn these simple functionality tests into a stress test by
37 * running them all in parallel, with as many threads as desired,
38 * and spread across as many datasets, objects, and vdevs as desired.
40 * (3) While all this is happening, we inject faults into the pool to
41 * verify that self-healing data really works.
43 * (4) Every time we open a dataset, we change its checksum and compression
44 * functions. Thus even individual objects vary from block to block
45 * in which checksum they use and whether they're compressed.
47 * (5) To verify that we never lose on-disk consistency after a crash,
48 * we run the entire test in a child of the main process.
49 * At random times, the child self-immolates with a SIGKILL.
50 * This is the software equivalent of pulling the power cord.
51 * The parent then runs the test again, using the existing
52 * storage pool, as many times as desired.
54 * (6) To verify that we don't have future leaks or temporal incursions,
55 * many of the functional tests record the transaction group number
56 * as part of their data. When reading old data, they verify that
57 * the transaction group number is less than the current, open txg.
58 * If you add a new test, please do this if applicable.
60 * When run with no arguments, ztest runs for about five minutes and
61 * produces no output if successful. To get a little bit of information,
62 * specify -V. To get more information, specify -VV, and so on.
64 * To turn this into an overnight stress test, use -T to specify run time.
66 * You can ask more more vdevs [-v], datasets [-d], or threads [-t]
67 * to increase the pool capacity, fanout, and overall stress level.
69 * The -N(okill) option will suppress kills, so each child runs to completion.
70 * This can be useful when you're trying to distinguish temporal incursions
71 * from plain old race conditions.
74 #include <sys/zfs_context.h>
80 #include <sys/dmu_objset.h>
86 #include <sys/resource.h>
89 #include <sys/zil_impl.h>
90 #include <sys/vdev_impl.h>
91 #include <sys/vdev_file.h>
92 #include <sys/spa_impl.h>
93 #include <sys/metaslab_impl.h>
94 #include <sys/dsl_prop.h>
95 #include <sys/dsl_dataset.h>
96 #include <sys/dsl_scan.h>
97 #include <sys/zio_checksum.h>
98 #include <sys/refcount.h>
100 #include <stdio_ext.h>
108 #include <sys/fs/zfs.h>
109 #include <libnvpair.h>
111 static char cmdname[] = "ztest";
112 static char *zopt_pool = cmdname;
114 static uint64_t zopt_vdevs = 5;
115 static uint64_t zopt_vdevtime;
116 static int zopt_ashift = SPA_MINBLOCKSHIFT;
117 static int zopt_mirrors = 2;
118 static int zopt_raidz = 4;
119 static int zopt_raidz_parity = 1;
120 static size_t zopt_vdev_size = SPA_MINDEVSIZE;
121 static int zopt_datasets = 7;
122 static int zopt_threads = 23;
123 static uint64_t zopt_passtime = 60; /* 60 seconds */
124 static uint64_t zopt_killrate = 70; /* 70% kill rate */
125 static int zopt_verbose = 0;
126 static int zopt_init = 1;
127 static char *zopt_dir = "/tmp";
128 static uint64_t zopt_time = 300; /* 5 minutes */
129 static uint64_t zopt_maxloops = 50; /* max loops during spa_freeze() */
131 #define BT_MAGIC 0x123456789abcdefULL
132 #define MAXFAULTS() (MAX(zs->zs_mirrors, 1) * (zopt_raidz_parity + 1) - 1)
136 ZTEST_IO_WRITE_PATTERN,
137 ZTEST_IO_WRITE_ZEROES,
143 typedef struct ztest_block_tag {
153 typedef struct bufwad {
160 * XXX -- fix zfs range locks to be generic so we can use them here.
182 #define ZTEST_RANGE_LOCKS 64
183 #define ZTEST_OBJECT_LOCKS 64
186 * Object descriptor. Used as a template for object lookup/create/remove.
188 typedef struct ztest_od {
191 dmu_object_type_t od_type;
192 dmu_object_type_t od_crtype;
193 uint64_t od_blocksize;
194 uint64_t od_crblocksize;
197 char od_name[MAXNAMELEN];
203 typedef struct ztest_ds {
207 ztest_od_t *zd_od; /* debugging aid */
208 char zd_name[MAXNAMELEN];
209 mutex_t zd_dirobj_lock;
210 rll_t zd_object_lock[ZTEST_OBJECT_LOCKS];
211 rll_t zd_range_lock[ZTEST_RANGE_LOCKS];
215 * Per-iteration state.
217 typedef void ztest_func_t(ztest_ds_t *zd, uint64_t id);
219 typedef struct ztest_info {
220 ztest_func_t *zi_func; /* test function */
221 uint64_t zi_iters; /* iterations per execution */
222 uint64_t *zi_interval; /* execute every <interval> seconds */
223 uint64_t zi_call_count; /* per-pass count */
224 uint64_t zi_call_time; /* per-pass time */
225 uint64_t zi_call_next; /* next time to call this function */
229 * Note: these aren't static because we want dladdr() to work.
231 ztest_func_t ztest_dmu_read_write;
232 ztest_func_t ztest_dmu_write_parallel;
233 ztest_func_t ztest_dmu_object_alloc_free;
234 ztest_func_t ztest_dmu_commit_callbacks;
235 ztest_func_t ztest_zap;
236 ztest_func_t ztest_zap_parallel;
237 ztest_func_t ztest_zil_commit;
238 ztest_func_t ztest_dmu_read_write_zcopy;
239 ztest_func_t ztest_dmu_objset_create_destroy;
240 ztest_func_t ztest_dmu_prealloc;
241 ztest_func_t ztest_fzap;
242 ztest_func_t ztest_dmu_snapshot_create_destroy;
243 ztest_func_t ztest_dsl_prop_get_set;
244 ztest_func_t ztest_spa_prop_get_set;
245 ztest_func_t ztest_spa_create_destroy;
246 ztest_func_t ztest_fault_inject;
247 ztest_func_t ztest_ddt_repair;
248 ztest_func_t ztest_dmu_snapshot_hold;
249 ztest_func_t ztest_spa_rename;
250 ztest_func_t ztest_scrub;
251 ztest_func_t ztest_dsl_dataset_promote_busy;
252 ztest_func_t ztest_vdev_attach_detach;
253 ztest_func_t ztest_vdev_LUN_growth;
254 ztest_func_t ztest_vdev_add_remove;
255 ztest_func_t ztest_vdev_aux_add_remove;
256 ztest_func_t ztest_split_pool;
258 uint64_t zopt_always = 0ULL * NANOSEC; /* all the time */
259 uint64_t zopt_incessant = 1ULL * NANOSEC / 10; /* every 1/10 second */
260 uint64_t zopt_often = 1ULL * NANOSEC; /* every second */
261 uint64_t zopt_sometimes = 10ULL * NANOSEC; /* every 10 seconds */
262 uint64_t zopt_rarely = 60ULL * NANOSEC; /* every 60 seconds */
264 ztest_info_t ztest_info[] = {
265 { ztest_dmu_read_write, 1, &zopt_always },
266 { ztest_dmu_write_parallel, 10, &zopt_always },
267 { ztest_dmu_object_alloc_free, 1, &zopt_always },
268 { ztest_dmu_commit_callbacks, 1, &zopt_always },
269 { ztest_zap, 30, &zopt_always },
270 { ztest_zap_parallel, 100, &zopt_always },
271 { ztest_split_pool, 1, &zopt_always },
272 { ztest_zil_commit, 1, &zopt_incessant },
273 { ztest_dmu_read_write_zcopy, 1, &zopt_often },
274 { ztest_dmu_objset_create_destroy, 1, &zopt_often },
275 { ztest_dsl_prop_get_set, 1, &zopt_often },
276 { ztest_spa_prop_get_set, 1, &zopt_sometimes },
278 { ztest_dmu_prealloc, 1, &zopt_sometimes },
280 { ztest_fzap, 1, &zopt_sometimes },
281 { ztest_dmu_snapshot_create_destroy, 1, &zopt_sometimes },
282 { ztest_spa_create_destroy, 1, &zopt_sometimes },
283 { ztest_fault_inject, 1, &zopt_sometimes },
284 { ztest_ddt_repair, 1, &zopt_sometimes },
285 { ztest_dmu_snapshot_hold, 1, &zopt_sometimes },
286 { ztest_spa_rename, 1, &zopt_rarely },
287 { ztest_scrub, 1, &zopt_rarely },
288 { ztest_dsl_dataset_promote_busy, 1, &zopt_rarely },
289 { ztest_vdev_attach_detach, 1, &zopt_rarely },
290 { ztest_vdev_LUN_growth, 1, &zopt_rarely },
291 { ztest_vdev_add_remove, 1, &zopt_vdevtime },
292 { ztest_vdev_aux_add_remove, 1, &zopt_vdevtime },
295 #define ZTEST_FUNCS (sizeof (ztest_info) / sizeof (ztest_info_t))
298 * The following struct is used to hold a list of uncalled commit callbacks.
299 * The callbacks are ordered by txg number.
301 typedef struct ztest_cb_list {
302 mutex_t zcl_callbacks_lock;
303 list_t zcl_callbacks;
307 * Stuff we need to share writably between parent and child.
309 typedef struct ztest_shared {
312 hrtime_t zs_proc_start;
313 hrtime_t zs_proc_stop;
314 hrtime_t zs_thread_start;
315 hrtime_t zs_thread_stop;
316 hrtime_t zs_thread_kill;
317 uint64_t zs_enospc_count;
318 uint64_t zs_vdev_next_leaf;
319 uint64_t zs_vdev_aux;
322 mutex_t zs_vdev_lock;
323 rwlock_t zs_name_lock;
324 ztest_info_t zs_info[ZTEST_FUNCS];
330 #define ID_PARALLEL -1ULL
332 static char ztest_dev_template[] = "%s/%s.%llua";
333 static char ztest_aux_template[] = "%s/%s.%s.%llu";
334 ztest_shared_t *ztest_shared;
337 static int ztest_random_fd;
338 static int ztest_dump_core = 1;
340 static boolean_t ztest_exiting;
342 /* Global commit callback list */
343 static ztest_cb_list_t zcl;
345 extern uint64_t metaslab_gang_bang;
346 extern uint64_t metaslab_df_alloc_threshold;
347 static uint64_t metaslab_sz;
350 ZTEST_META_DNODE = 0,
355 static void usage(boolean_t) __NORETURN;
358 * These libumem hooks provide a reasonable set of defaults for the allocator's
359 * debugging facilities.
364 return ("default,verbose"); /* $UMEM_DEBUG setting */
368 _umem_logging_init(void)
370 return ("fail,contents"); /* $UMEM_LOGGING setting */
373 #define FATAL_MSG_SZ 1024
378 fatal(int do_perror, char *message, ...)
381 int save_errno = errno;
382 char buf[FATAL_MSG_SZ];
384 (void) fflush(stdout);
386 va_start(args, message);
387 (void) sprintf(buf, "ztest: ");
389 (void) vsprintf(buf + strlen(buf), message, args);
392 (void) snprintf(buf + strlen(buf), FATAL_MSG_SZ - strlen(buf),
393 ": %s", strerror(save_errno));
395 (void) fprintf(stderr, "%s\n", buf);
396 fatal_msg = buf; /* to ease debugging */
403 str2shift(const char *buf)
405 const char *ends = "BKMGTPEZ";
410 for (i = 0; i < strlen(ends); i++) {
411 if (toupper(buf[0]) == ends[i])
414 if (i == strlen(ends)) {
415 (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n",
419 if (buf[1] == '\0' || (toupper(buf[1]) == 'B' && buf[2] == '\0')) {
422 (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n", buf);
428 nicenumtoull(const char *buf)
433 val = strtoull(buf, &end, 0);
435 (void) fprintf(stderr, "ztest: bad numeric value: %s\n", buf);
437 } else if (end[0] == '.') {
438 double fval = strtod(buf, &end);
439 fval *= pow(2, str2shift(end));
440 if (fval > UINT64_MAX) {
441 (void) fprintf(stderr, "ztest: value too large: %s\n",
445 val = (uint64_t)fval;
447 int shift = str2shift(end);
448 if (shift >= 64 || (val << shift) >> shift != val) {
449 (void) fprintf(stderr, "ztest: value too large: %s\n",
459 usage(boolean_t requested)
461 char nice_vdev_size[10];
462 char nice_gang_bang[10];
463 FILE *fp = requested ? stdout : stderr;
465 nicenum(zopt_vdev_size, nice_vdev_size);
466 nicenum(metaslab_gang_bang, nice_gang_bang);
468 (void) fprintf(fp, "Usage: %s\n"
469 "\t[-v vdevs (default: %llu)]\n"
470 "\t[-s size_of_each_vdev (default: %s)]\n"
471 "\t[-a alignment_shift (default: %d)] use 0 for random\n"
472 "\t[-m mirror_copies (default: %d)]\n"
473 "\t[-r raidz_disks (default: %d)]\n"
474 "\t[-R raidz_parity (default: %d)]\n"
475 "\t[-d datasets (default: %d)]\n"
476 "\t[-t threads (default: %d)]\n"
477 "\t[-g gang_block_threshold (default: %s)]\n"
478 "\t[-i init_count (default: %d)] initialize pool i times\n"
479 "\t[-k kill_percentage (default: %llu%%)]\n"
480 "\t[-p pool_name (default: %s)]\n"
481 "\t[-f dir (default: %s)] file directory for vdev files\n"
482 "\t[-V] verbose (use multiple times for ever more blather)\n"
483 "\t[-E] use existing pool instead of creating new one\n"
484 "\t[-T time (default: %llu sec)] total run time\n"
485 "\t[-F freezeloops (default: %llu)] max loops in spa_freeze()\n"
486 "\t[-P passtime (default: %llu sec)] time per pass\n"
487 "\t[-h] (print help)\n"
490 (u_longlong_t)zopt_vdevs, /* -v */
491 nice_vdev_size, /* -s */
492 zopt_ashift, /* -a */
493 zopt_mirrors, /* -m */
495 zopt_raidz_parity, /* -R */
496 zopt_datasets, /* -d */
497 zopt_threads, /* -t */
498 nice_gang_bang, /* -g */
500 (u_longlong_t)zopt_killrate, /* -k */
503 (u_longlong_t)zopt_time, /* -T */
504 (u_longlong_t)zopt_maxloops, /* -F */
505 (u_longlong_t)zopt_passtime); /* -P */
506 exit(requested ? 0 : 1);
510 process_options(int argc, char **argv)
515 /* By default, test gang blocks for blocks 32K and greater */
516 metaslab_gang_bang = 32 << 10;
518 while ((opt = getopt(argc, argv,
519 "v:s:a:m:r:R:d:t:g:i:k:p:f:VET:P:hF:")) != EOF) {
536 value = nicenumtoull(optarg);
543 zopt_vdev_size = MAX(SPA_MINDEVSIZE, value);
549 zopt_mirrors = value;
552 zopt_raidz = MAX(1, value);
555 zopt_raidz_parity = MIN(MAX(value, 1), 3);
558 zopt_datasets = MAX(1, value);
561 zopt_threads = MAX(1, value);
564 metaslab_gang_bang = MAX(SPA_MINBLOCKSIZE << 1, value);
570 zopt_killrate = value;
573 zopt_pool = strdup(optarg);
576 zopt_dir = strdup(optarg);
588 zopt_passtime = MAX(1, value);
591 zopt_maxloops = MAX(1, value);
603 zopt_raidz_parity = MIN(zopt_raidz_parity, zopt_raidz - 1);
605 zopt_vdevtime = (zopt_vdevs > 0 ? zopt_time * NANOSEC / zopt_vdevs :
610 ztest_kill(ztest_shared_t *zs)
612 zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(zs->zs_spa));
613 zs->zs_space = metaslab_class_get_space(spa_normal_class(zs->zs_spa));
614 (void) kill(getpid(), SIGKILL);
618 ztest_random(uint64_t range)
625 if (read(ztest_random_fd, &r, sizeof (r)) != sizeof (r))
626 fatal(1, "short read from /dev/urandom");
633 ztest_record_enospc(const char *s)
635 ztest_shared->zs_enospc_count++;
639 ztest_get_ashift(void)
641 if (zopt_ashift == 0)
642 return (SPA_MINBLOCKSHIFT + ztest_random(3));
643 return (zopt_ashift);
647 make_vdev_file(char *path, char *aux, size_t size, uint64_t ashift)
649 char pathbuf[MAXPATHLEN];
654 ashift = ztest_get_ashift();
660 vdev = ztest_shared->zs_vdev_aux;
661 (void) sprintf(path, ztest_aux_template,
662 zopt_dir, zopt_pool, aux, vdev);
664 vdev = ztest_shared->zs_vdev_next_leaf++;
665 (void) sprintf(path, ztest_dev_template,
666 zopt_dir, zopt_pool, vdev);
671 int fd = open(path, O_RDWR | O_CREAT | O_TRUNC, 0666);
673 fatal(1, "can't open %s", path);
674 if (ftruncate(fd, size) != 0)
675 fatal(1, "can't ftruncate %s", path);
679 VERIFY(nvlist_alloc(&file, NV_UNIQUE_NAME, 0) == 0);
680 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_TYPE, VDEV_TYPE_FILE) == 0);
681 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_PATH, path) == 0);
682 VERIFY(nvlist_add_uint64(file, ZPOOL_CONFIG_ASHIFT, ashift) == 0);
688 make_vdev_raidz(char *path, char *aux, size_t size, uint64_t ashift, int r)
690 nvlist_t *raidz, **child;
694 return (make_vdev_file(path, aux, size, ashift));
695 child = umem_alloc(r * sizeof (nvlist_t *), UMEM_NOFAIL);
697 for (c = 0; c < r; c++)
698 child[c] = make_vdev_file(path, aux, size, ashift);
700 VERIFY(nvlist_alloc(&raidz, NV_UNIQUE_NAME, 0) == 0);
701 VERIFY(nvlist_add_string(raidz, ZPOOL_CONFIG_TYPE,
702 VDEV_TYPE_RAIDZ) == 0);
703 VERIFY(nvlist_add_uint64(raidz, ZPOOL_CONFIG_NPARITY,
704 zopt_raidz_parity) == 0);
705 VERIFY(nvlist_add_nvlist_array(raidz, ZPOOL_CONFIG_CHILDREN,
708 for (c = 0; c < r; c++)
709 nvlist_free(child[c]);
711 umem_free(child, r * sizeof (nvlist_t *));
717 make_vdev_mirror(char *path, char *aux, size_t size, uint64_t ashift,
720 nvlist_t *mirror, **child;
724 return (make_vdev_raidz(path, aux, size, ashift, r));
726 child = umem_alloc(m * sizeof (nvlist_t *), UMEM_NOFAIL);
728 for (c = 0; c < m; c++)
729 child[c] = make_vdev_raidz(path, aux, size, ashift, r);
731 VERIFY(nvlist_alloc(&mirror, NV_UNIQUE_NAME, 0) == 0);
732 VERIFY(nvlist_add_string(mirror, ZPOOL_CONFIG_TYPE,
733 VDEV_TYPE_MIRROR) == 0);
734 VERIFY(nvlist_add_nvlist_array(mirror, ZPOOL_CONFIG_CHILDREN,
737 for (c = 0; c < m; c++)
738 nvlist_free(child[c]);
740 umem_free(child, m * sizeof (nvlist_t *));
746 make_vdev_root(char *path, char *aux, size_t size, uint64_t ashift,
747 int log, int r, int m, int t)
749 nvlist_t *root, **child;
754 child = umem_alloc(t * sizeof (nvlist_t *), UMEM_NOFAIL);
756 for (c = 0; c < t; c++) {
757 child[c] = make_vdev_mirror(path, aux, size, ashift, r, m);
758 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_IS_LOG,
762 VERIFY(nvlist_alloc(&root, NV_UNIQUE_NAME, 0) == 0);
763 VERIFY(nvlist_add_string(root, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) == 0);
764 VERIFY(nvlist_add_nvlist_array(root, aux ? aux : ZPOOL_CONFIG_CHILDREN,
767 for (c = 0; c < t; c++)
768 nvlist_free(child[c]);
770 umem_free(child, t * sizeof (nvlist_t *));
776 ztest_random_blocksize(void)
778 return (1 << (SPA_MINBLOCKSHIFT +
779 ztest_random(SPA_MAXBLOCKSHIFT - SPA_MINBLOCKSHIFT + 1)));
783 ztest_random_ibshift(void)
785 return (DN_MIN_INDBLKSHIFT +
786 ztest_random(DN_MAX_INDBLKSHIFT - DN_MIN_INDBLKSHIFT + 1));
790 ztest_random_vdev_top(spa_t *spa, boolean_t log_ok)
793 vdev_t *rvd = spa->spa_root_vdev;
796 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
799 top = ztest_random(rvd->vdev_children);
800 tvd = rvd->vdev_child[top];
801 } while (tvd->vdev_ishole || (tvd->vdev_islog && !log_ok) ||
802 tvd->vdev_mg == NULL || tvd->vdev_mg->mg_class == NULL);
808 ztest_random_dsl_prop(zfs_prop_t prop)
813 value = zfs_prop_random_value(prop, ztest_random(-1ULL));
814 } while (prop == ZFS_PROP_CHECKSUM && value == ZIO_CHECKSUM_OFF);
820 ztest_dsl_prop_set_uint64(char *osname, zfs_prop_t prop, uint64_t value,
823 const char *propname = zfs_prop_to_name(prop);
825 char setpoint[MAXPATHLEN];
829 error = dsl_prop_set(osname, propname,
830 (inherit ? ZPROP_SRC_NONE : ZPROP_SRC_LOCAL),
831 sizeof (value), 1, &value);
833 if (error == ENOSPC) {
834 ztest_record_enospc(FTAG);
837 ASSERT3U(error, ==, 0);
839 VERIFY3U(dsl_prop_get(osname, propname, sizeof (curval),
840 1, &curval, setpoint), ==, 0);
842 if (zopt_verbose >= 6) {
843 VERIFY(zfs_prop_index_to_string(prop, curval, &valname) == 0);
844 (void) printf("%s %s = %s at '%s'\n",
845 osname, propname, valname, setpoint);
852 ztest_spa_prop_set_uint64(ztest_shared_t *zs, zpool_prop_t prop, uint64_t value)
854 spa_t *spa = zs->zs_spa;
855 nvlist_t *props = NULL;
858 VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
859 VERIFY(nvlist_add_uint64(props, zpool_prop_to_name(prop), value) == 0);
861 error = spa_prop_set(spa, props);
865 if (error == ENOSPC) {
866 ztest_record_enospc(FTAG);
869 ASSERT3U(error, ==, 0);
875 ztest_rll_init(rll_t *rll)
877 rll->rll_writer = NULL;
878 rll->rll_readers = 0;
879 VERIFY(_mutex_init(&rll->rll_lock, USYNC_THREAD, NULL) == 0);
880 VERIFY(cond_init(&rll->rll_cv, USYNC_THREAD, NULL) == 0);
884 ztest_rll_destroy(rll_t *rll)
886 ASSERT(rll->rll_writer == NULL);
887 ASSERT(rll->rll_readers == 0);
888 VERIFY(_mutex_destroy(&rll->rll_lock) == 0);
889 VERIFY(cond_destroy(&rll->rll_cv) == 0);
893 ztest_rll_lock(rll_t *rll, rl_type_t type)
895 VERIFY(mutex_lock(&rll->rll_lock) == 0);
897 if (type == RL_READER) {
898 while (rll->rll_writer != NULL)
899 (void) cond_wait(&rll->rll_cv, &rll->rll_lock);
902 while (rll->rll_writer != NULL || rll->rll_readers)
903 (void) cond_wait(&rll->rll_cv, &rll->rll_lock);
904 rll->rll_writer = curthread;
907 VERIFY(mutex_unlock(&rll->rll_lock) == 0);
911 ztest_rll_unlock(rll_t *rll)
913 VERIFY(mutex_lock(&rll->rll_lock) == 0);
915 if (rll->rll_writer) {
916 ASSERT(rll->rll_readers == 0);
917 rll->rll_writer = NULL;
919 ASSERT(rll->rll_readers != 0);
920 ASSERT(rll->rll_writer == NULL);
924 if (rll->rll_writer == NULL && rll->rll_readers == 0)
925 VERIFY(cond_broadcast(&rll->rll_cv) == 0);
927 VERIFY(mutex_unlock(&rll->rll_lock) == 0);
931 ztest_object_lock(ztest_ds_t *zd, uint64_t object, rl_type_t type)
933 rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
935 ztest_rll_lock(rll, type);
939 ztest_object_unlock(ztest_ds_t *zd, uint64_t object)
941 rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
943 ztest_rll_unlock(rll);
947 ztest_range_lock(ztest_ds_t *zd, uint64_t object, uint64_t offset,
948 uint64_t size, rl_type_t type)
950 uint64_t hash = object ^ (offset % (ZTEST_RANGE_LOCKS + 1));
951 rll_t *rll = &zd->zd_range_lock[hash & (ZTEST_RANGE_LOCKS - 1)];
954 rl = umem_alloc(sizeof (*rl), UMEM_NOFAIL);
955 rl->rl_object = object;
956 rl->rl_offset = offset;
960 ztest_rll_lock(rll, type);
966 ztest_range_unlock(rl_t *rl)
968 rll_t *rll = rl->rl_lock;
970 ztest_rll_unlock(rll);
972 umem_free(rl, sizeof (*rl));
976 ztest_zd_init(ztest_ds_t *zd, objset_t *os)
979 zd->zd_zilog = dmu_objset_zil(os);
981 dmu_objset_name(os, zd->zd_name);
984 VERIFY(_mutex_init(&zd->zd_dirobj_lock, USYNC_THREAD, NULL) == 0);
986 for (l = 0; l < ZTEST_OBJECT_LOCKS; l++)
987 ztest_rll_init(&zd->zd_object_lock[l]);
989 for (l = 0; l < ZTEST_RANGE_LOCKS; l++)
990 ztest_rll_init(&zd->zd_range_lock[l]);
994 ztest_zd_fini(ztest_ds_t *zd)
998 VERIFY(_mutex_destroy(&zd->zd_dirobj_lock) == 0);
1000 for (l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1001 ztest_rll_destroy(&zd->zd_object_lock[l]);
1003 for (l = 0; l < ZTEST_RANGE_LOCKS; l++)
1004 ztest_rll_destroy(&zd->zd_range_lock[l]);
1007 #define TXG_MIGHTWAIT (ztest_random(10) == 0 ? TXG_NOWAIT : TXG_WAIT)
1010 ztest_tx_assign(dmu_tx_t *tx, uint64_t txg_how, const char *tag)
1016 * Attempt to assign tx to some transaction group.
1018 error = dmu_tx_assign(tx, txg_how);
1020 if (error == ERESTART) {
1021 ASSERT(txg_how == TXG_NOWAIT);
1024 ASSERT3U(error, ==, ENOSPC);
1025 ztest_record_enospc(tag);
1030 txg = dmu_tx_get_txg(tx);
1036 ztest_pattern_set(void *buf, uint64_t size, uint64_t value)
1039 uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1046 ztest_pattern_match(void *buf, uint64_t size, uint64_t value)
1049 uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1053 diff |= (value - *ip++);
1059 ztest_bt_generate(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1060 uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg)
1062 bt->bt_magic = BT_MAGIC;
1063 bt->bt_objset = dmu_objset_id(os);
1064 bt->bt_object = object;
1065 bt->bt_offset = offset;
1068 bt->bt_crtxg = crtxg;
1072 ztest_bt_verify(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1073 uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg)
1075 ASSERT(bt->bt_magic == BT_MAGIC);
1076 ASSERT(bt->bt_objset == dmu_objset_id(os));
1077 ASSERT(bt->bt_object == object);
1078 ASSERT(bt->bt_offset == offset);
1079 ASSERT(bt->bt_gen <= gen);
1080 ASSERT(bt->bt_txg <= txg);
1081 ASSERT(bt->bt_crtxg == crtxg);
1084 static ztest_block_tag_t *
1085 ztest_bt_bonus(dmu_buf_t *db)
1087 dmu_object_info_t doi;
1088 ztest_block_tag_t *bt;
1090 dmu_object_info_from_db(db, &doi);
1091 ASSERT3U(doi.doi_bonus_size, <=, db->db_size);
1092 ASSERT3U(doi.doi_bonus_size, >=, sizeof (*bt));
1093 bt = (void *)((char *)db->db_data + doi.doi_bonus_size - sizeof (*bt));
1102 #define lrz_type lr_mode
1103 #define lrz_blocksize lr_uid
1104 #define lrz_ibshift lr_gid
1105 #define lrz_bonustype lr_rdev
1106 #define lrz_bonuslen lr_crtime[1]
1109 ztest_log_create(ztest_ds_t *zd, dmu_tx_t *tx, lr_create_t *lr)
1111 char *name = (void *)(lr + 1); /* name follows lr */
1112 size_t namesize = strlen(name) + 1;
1115 if (zil_replaying(zd->zd_zilog, tx))
1118 itx = zil_itx_create(TX_CREATE, sizeof (*lr) + namesize);
1119 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1120 sizeof (*lr) + namesize - sizeof (lr_t));
1122 zil_itx_assign(zd->zd_zilog, itx, tx);
1126 ztest_log_remove(ztest_ds_t *zd, dmu_tx_t *tx, lr_remove_t *lr, uint64_t object)
1128 char *name = (void *)(lr + 1); /* name follows lr */
1129 size_t namesize = strlen(name) + 1;
1132 if (zil_replaying(zd->zd_zilog, tx))
1135 itx = zil_itx_create(TX_REMOVE, sizeof (*lr) + namesize);
1136 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1137 sizeof (*lr) + namesize - sizeof (lr_t));
1139 itx->itx_oid = object;
1140 zil_itx_assign(zd->zd_zilog, itx, tx);
1144 ztest_log_write(ztest_ds_t *zd, dmu_tx_t *tx, lr_write_t *lr)
1147 itx_wr_state_t write_state = ztest_random(WR_NUM_STATES);
1149 if (zil_replaying(zd->zd_zilog, tx))
1152 if (lr->lr_length > ZIL_MAX_LOG_DATA)
1153 write_state = WR_INDIRECT;
1155 itx = zil_itx_create(TX_WRITE,
1156 sizeof (*lr) + (write_state == WR_COPIED ? lr->lr_length : 0));
1158 if (write_state == WR_COPIED &&
1159 dmu_read(zd->zd_os, lr->lr_foid, lr->lr_offset, lr->lr_length,
1160 ((lr_write_t *)&itx->itx_lr) + 1, DMU_READ_NO_PREFETCH) != 0) {
1161 zil_itx_destroy(itx);
1162 itx = zil_itx_create(TX_WRITE, sizeof (*lr));
1163 write_state = WR_NEED_COPY;
1165 itx->itx_private = zd;
1166 itx->itx_wr_state = write_state;
1167 itx->itx_sync = (ztest_random(8) == 0);
1168 itx->itx_sod += (write_state == WR_NEED_COPY ? lr->lr_length : 0);
1170 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1171 sizeof (*lr) - sizeof (lr_t));
1173 zil_itx_assign(zd->zd_zilog, itx, tx);
1177 ztest_log_truncate(ztest_ds_t *zd, dmu_tx_t *tx, lr_truncate_t *lr)
1181 if (zil_replaying(zd->zd_zilog, tx))
1184 itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr));
1185 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1186 sizeof (*lr) - sizeof (lr_t));
1188 itx->itx_sync = B_FALSE;
1189 zil_itx_assign(zd->zd_zilog, itx, tx);
1193 ztest_log_setattr(ztest_ds_t *zd, dmu_tx_t *tx, lr_setattr_t *lr)
1197 if (zil_replaying(zd->zd_zilog, tx))
1200 itx = zil_itx_create(TX_SETATTR, sizeof (*lr));
1201 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1202 sizeof (*lr) - sizeof (lr_t));
1204 itx->itx_sync = B_FALSE;
1205 zil_itx_assign(zd->zd_zilog, itx, tx);
1212 ztest_replay_create(ztest_ds_t *zd, lr_create_t *lr, boolean_t byteswap)
1214 char *name = (void *)(lr + 1); /* name follows lr */
1215 objset_t *os = zd->zd_os;
1216 ztest_block_tag_t *bbt;
1223 byteswap_uint64_array(lr, sizeof (*lr));
1225 ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1226 ASSERT(name[0] != '\0');
1228 tx = dmu_tx_create(os);
1230 dmu_tx_hold_zap(tx, lr->lr_doid, B_TRUE, name);
1232 if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1233 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, B_TRUE, NULL);
1235 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1238 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1242 ASSERT(dmu_objset_zil(os)->zl_replay == !!lr->lr_foid);
1244 if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1245 if (lr->lr_foid == 0) {
1246 lr->lr_foid = zap_create(os,
1247 lr->lrz_type, lr->lrz_bonustype,
1248 lr->lrz_bonuslen, tx);
1250 error = zap_create_claim(os, lr->lr_foid,
1251 lr->lrz_type, lr->lrz_bonustype,
1252 lr->lrz_bonuslen, tx);
1255 if (lr->lr_foid == 0) {
1256 lr->lr_foid = dmu_object_alloc(os,
1257 lr->lrz_type, 0, lr->lrz_bonustype,
1258 lr->lrz_bonuslen, tx);
1260 error = dmu_object_claim(os, lr->lr_foid,
1261 lr->lrz_type, 0, lr->lrz_bonustype,
1262 lr->lrz_bonuslen, tx);
1267 ASSERT3U(error, ==, EEXIST);
1268 ASSERT(zd->zd_zilog->zl_replay);
1273 ASSERT(lr->lr_foid != 0);
1275 if (lr->lrz_type != DMU_OT_ZAP_OTHER)
1276 VERIFY3U(0, ==, dmu_object_set_blocksize(os, lr->lr_foid,
1277 lr->lrz_blocksize, lr->lrz_ibshift, tx));
1279 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1280 bbt = ztest_bt_bonus(db);
1281 dmu_buf_will_dirty(db, tx);
1282 ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_gen, txg, txg);
1283 dmu_buf_rele(db, FTAG);
1285 VERIFY3U(0, ==, zap_add(os, lr->lr_doid, name, sizeof (uint64_t), 1,
1288 (void) ztest_log_create(zd, tx, lr);
1296 ztest_replay_remove(ztest_ds_t *zd, lr_remove_t *lr, boolean_t byteswap)
1298 char *name = (void *)(lr + 1); /* name follows lr */
1299 objset_t *os = zd->zd_os;
1300 dmu_object_info_t doi;
1302 uint64_t object, txg;
1305 byteswap_uint64_array(lr, sizeof (*lr));
1307 ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1308 ASSERT(name[0] != '\0');
1311 zap_lookup(os, lr->lr_doid, name, sizeof (object), 1, &object));
1312 ASSERT(object != 0);
1314 ztest_object_lock(zd, object, RL_WRITER);
1316 VERIFY3U(0, ==, dmu_object_info(os, object, &doi));
1318 tx = dmu_tx_create(os);
1320 dmu_tx_hold_zap(tx, lr->lr_doid, B_FALSE, name);
1321 dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END);
1323 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1325 ztest_object_unlock(zd, object);
1329 if (doi.doi_type == DMU_OT_ZAP_OTHER) {
1330 VERIFY3U(0, ==, zap_destroy(os, object, tx));
1332 VERIFY3U(0, ==, dmu_object_free(os, object, tx));
1335 VERIFY3U(0, ==, zap_remove(os, lr->lr_doid, name, tx));
1337 (void) ztest_log_remove(zd, tx, lr, object);
1341 ztest_object_unlock(zd, object);
1347 ztest_replay_write(ztest_ds_t *zd, lr_write_t *lr, boolean_t byteswap)
1349 objset_t *os = zd->zd_os;
1350 void *data = lr + 1; /* data follows lr */
1351 uint64_t offset, length;
1352 ztest_block_tag_t *bt = data;
1353 ztest_block_tag_t *bbt;
1354 uint64_t gen, txg, lrtxg, crtxg;
1355 dmu_object_info_t doi;
1358 arc_buf_t *abuf = NULL;
1362 byteswap_uint64_array(lr, sizeof (*lr));
1364 offset = lr->lr_offset;
1365 length = lr->lr_length;
1367 /* If it's a dmu_sync() block, write the whole block */
1368 if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
1369 uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
1370 if (length < blocksize) {
1371 offset -= offset % blocksize;
1376 if (bt->bt_magic == BSWAP_64(BT_MAGIC))
1377 byteswap_uint64_array(bt, sizeof (*bt));
1379 if (bt->bt_magic != BT_MAGIC)
1382 ztest_object_lock(zd, lr->lr_foid, RL_READER);
1383 rl = ztest_range_lock(zd, lr->lr_foid, offset, length, RL_WRITER);
1385 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1387 dmu_object_info_from_db(db, &doi);
1389 bbt = ztest_bt_bonus(db);
1390 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1392 crtxg = bbt->bt_crtxg;
1393 lrtxg = lr->lr_common.lrc_txg;
1395 tx = dmu_tx_create(os);
1397 dmu_tx_hold_write(tx, lr->lr_foid, offset, length);
1399 if (ztest_random(8) == 0 && length == doi.doi_data_block_size &&
1400 P2PHASE(offset, length) == 0)
1401 abuf = dmu_request_arcbuf(db, length);
1403 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1406 dmu_return_arcbuf(abuf);
1407 dmu_buf_rele(db, FTAG);
1408 ztest_range_unlock(rl);
1409 ztest_object_unlock(zd, lr->lr_foid);
1415 * Usually, verify the old data before writing new data --
1416 * but not always, because we also want to verify correct
1417 * behavior when the data was not recently read into cache.
1419 ASSERT(offset % doi.doi_data_block_size == 0);
1420 if (ztest_random(4) != 0) {
1421 int prefetch = ztest_random(2) ?
1422 DMU_READ_PREFETCH : DMU_READ_NO_PREFETCH;
1423 ztest_block_tag_t rbt;
1425 VERIFY(dmu_read(os, lr->lr_foid, offset,
1426 sizeof (rbt), &rbt, prefetch) == 0);
1427 if (rbt.bt_magic == BT_MAGIC) {
1428 ztest_bt_verify(&rbt, os, lr->lr_foid,
1429 offset, gen, txg, crtxg);
1434 * Writes can appear to be newer than the bonus buffer because
1435 * the ztest_get_data() callback does a dmu_read() of the
1436 * open-context data, which may be different than the data
1437 * as it was when the write was generated.
1439 if (zd->zd_zilog->zl_replay) {
1440 ztest_bt_verify(bt, os, lr->lr_foid, offset,
1441 MAX(gen, bt->bt_gen), MAX(txg, lrtxg),
1446 * Set the bt's gen/txg to the bonus buffer's gen/txg
1447 * so that all of the usual ASSERTs will work.
1449 ztest_bt_generate(bt, os, lr->lr_foid, offset, gen, txg, crtxg);
1453 dmu_write(os, lr->lr_foid, offset, length, data, tx);
1455 bcopy(data, abuf->b_data, length);
1456 dmu_assign_arcbuf(db, offset, abuf, tx);
1459 (void) ztest_log_write(zd, tx, lr);
1461 dmu_buf_rele(db, FTAG);
1465 ztest_range_unlock(rl);
1466 ztest_object_unlock(zd, lr->lr_foid);
1472 ztest_replay_truncate(ztest_ds_t *zd, lr_truncate_t *lr, boolean_t byteswap)
1474 objset_t *os = zd->zd_os;
1480 byteswap_uint64_array(lr, sizeof (*lr));
1482 ztest_object_lock(zd, lr->lr_foid, RL_READER);
1483 rl = ztest_range_lock(zd, lr->lr_foid, lr->lr_offset, lr->lr_length,
1486 tx = dmu_tx_create(os);
1488 dmu_tx_hold_free(tx, lr->lr_foid, lr->lr_offset, lr->lr_length);
1490 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1492 ztest_range_unlock(rl);
1493 ztest_object_unlock(zd, lr->lr_foid);
1497 VERIFY(dmu_free_range(os, lr->lr_foid, lr->lr_offset,
1498 lr->lr_length, tx) == 0);
1500 (void) ztest_log_truncate(zd, tx, lr);
1504 ztest_range_unlock(rl);
1505 ztest_object_unlock(zd, lr->lr_foid);
1511 ztest_replay_setattr(ztest_ds_t *zd, lr_setattr_t *lr, boolean_t byteswap)
1513 objset_t *os = zd->zd_os;
1516 ztest_block_tag_t *bbt;
1517 uint64_t txg, lrtxg, crtxg;
1520 byteswap_uint64_array(lr, sizeof (*lr));
1522 ztest_object_lock(zd, lr->lr_foid, RL_WRITER);
1524 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1526 tx = dmu_tx_create(os);
1527 dmu_tx_hold_bonus(tx, lr->lr_foid);
1529 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1531 dmu_buf_rele(db, FTAG);
1532 ztest_object_unlock(zd, lr->lr_foid);
1536 bbt = ztest_bt_bonus(db);
1537 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1538 crtxg = bbt->bt_crtxg;
1539 lrtxg = lr->lr_common.lrc_txg;
1541 if (zd->zd_zilog->zl_replay) {
1542 ASSERT(lr->lr_size != 0);
1543 ASSERT(lr->lr_mode != 0);
1547 * Randomly change the size and increment the generation.
1549 lr->lr_size = (ztest_random(db->db_size / sizeof (*bbt)) + 1) *
1551 lr->lr_mode = bbt->bt_gen + 1;
1556 * Verify that the current bonus buffer is not newer than our txg.
1558 ztest_bt_verify(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode,
1559 MAX(txg, lrtxg), crtxg);
1561 dmu_buf_will_dirty(db, tx);
1563 ASSERT3U(lr->lr_size, >=, sizeof (*bbt));
1564 ASSERT3U(lr->lr_size, <=, db->db_size);
1565 VERIFY3U(dmu_set_bonus(db, lr->lr_size, tx), ==, 0);
1566 bbt = ztest_bt_bonus(db);
1568 ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode, txg, crtxg);
1570 dmu_buf_rele(db, FTAG);
1572 (void) ztest_log_setattr(zd, tx, lr);
1576 ztest_object_unlock(zd, lr->lr_foid);
1581 zil_replay_func_t *ztest_replay_vector[TX_MAX_TYPE] = {
1582 NULL, /* 0 no such transaction type */
1583 ztest_replay_create, /* TX_CREATE */
1584 NULL, /* TX_MKDIR */
1585 NULL, /* TX_MKXATTR */
1586 NULL, /* TX_SYMLINK */
1587 ztest_replay_remove, /* TX_REMOVE */
1588 NULL, /* TX_RMDIR */
1590 NULL, /* TX_RENAME */
1591 ztest_replay_write, /* TX_WRITE */
1592 ztest_replay_truncate, /* TX_TRUNCATE */
1593 ztest_replay_setattr, /* TX_SETATTR */
1595 NULL, /* TX_CREATE_ACL */
1596 NULL, /* TX_CREATE_ATTR */
1597 NULL, /* TX_CREATE_ACL_ATTR */
1598 NULL, /* TX_MKDIR_ACL */
1599 NULL, /* TX_MKDIR_ATTR */
1600 NULL, /* TX_MKDIR_ACL_ATTR */
1601 NULL, /* TX_WRITE2 */
1605 * ZIL get_data callbacks
1609 ztest_get_done(zgd_t *zgd, int error)
1611 ztest_ds_t *zd = zgd->zgd_private;
1612 uint64_t object = zgd->zgd_rl->rl_object;
1615 dmu_buf_rele(zgd->zgd_db, zgd);
1617 ztest_range_unlock(zgd->zgd_rl);
1618 ztest_object_unlock(zd, object);
1620 if (error == 0 && zgd->zgd_bp)
1621 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
1623 umem_free(zgd, sizeof (*zgd));
1627 ztest_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
1629 ztest_ds_t *zd = arg;
1630 objset_t *os = zd->zd_os;
1631 uint64_t object = lr->lr_foid;
1632 uint64_t offset = lr->lr_offset;
1633 uint64_t size = lr->lr_length;
1634 blkptr_t *bp = &lr->lr_blkptr;
1635 uint64_t txg = lr->lr_common.lrc_txg;
1637 dmu_object_info_t doi;
1642 ztest_object_lock(zd, object, RL_READER);
1643 error = dmu_bonus_hold(os, object, FTAG, &db);
1645 ztest_object_unlock(zd, object);
1649 crtxg = ztest_bt_bonus(db)->bt_crtxg;
1651 if (crtxg == 0 || crtxg > txg) {
1652 dmu_buf_rele(db, FTAG);
1653 ztest_object_unlock(zd, object);
1657 dmu_object_info_from_db(db, &doi);
1658 dmu_buf_rele(db, FTAG);
1661 zgd = umem_zalloc(sizeof (*zgd), UMEM_NOFAIL);
1662 zgd->zgd_zilog = zd->zd_zilog;
1663 zgd->zgd_private = zd;
1665 if (buf != NULL) { /* immediate write */
1666 zgd->zgd_rl = ztest_range_lock(zd, object, offset, size,
1669 error = dmu_read(os, object, offset, size, buf,
1670 DMU_READ_NO_PREFETCH);
1673 size = doi.doi_data_block_size;
1675 offset = P2ALIGN(offset, size);
1677 ASSERT(offset < size);
1681 zgd->zgd_rl = ztest_range_lock(zd, object, offset, size,
1684 error = dmu_buf_hold(os, object, offset, zgd, &db,
1685 DMU_READ_NO_PREFETCH);
1691 ASSERT(db->db_offset == offset);
1692 ASSERT(db->db_size == size);
1694 error = dmu_sync(zio, lr->lr_common.lrc_txg,
1695 ztest_get_done, zgd);
1702 ztest_get_done(zgd, error);
1708 ztest_lr_alloc(size_t lrsize, char *name)
1711 size_t namesize = name ? strlen(name) + 1 : 0;
1713 lr = umem_zalloc(lrsize + namesize, UMEM_NOFAIL);
1716 bcopy(name, lr + lrsize, namesize);
1722 ztest_lr_free(void *lr, size_t lrsize, char *name)
1724 size_t namesize = name ? strlen(name) + 1 : 0;
1726 umem_free(lr, lrsize + namesize);
1730 * Lookup a bunch of objects. Returns the number of objects not found.
1733 ztest_lookup(ztest_ds_t *zd, ztest_od_t *od, int count)
1739 ASSERT(_mutex_held(&zd->zd_dirobj_lock));
1741 for (i = 0; i < count; i++, od++) {
1743 error = zap_lookup(zd->zd_os, od->od_dir, od->od_name,
1744 sizeof (uint64_t), 1, &od->od_object);
1746 ASSERT(error == ENOENT);
1747 ASSERT(od->od_object == 0);
1751 ztest_block_tag_t *bbt;
1752 dmu_object_info_t doi;
1754 ASSERT(od->od_object != 0);
1755 ASSERT(missing == 0); /* there should be no gaps */
1757 ztest_object_lock(zd, od->od_object, RL_READER);
1758 VERIFY3U(0, ==, dmu_bonus_hold(zd->zd_os,
1759 od->od_object, FTAG, &db));
1760 dmu_object_info_from_db(db, &doi);
1761 bbt = ztest_bt_bonus(db);
1762 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1763 od->od_type = doi.doi_type;
1764 od->od_blocksize = doi.doi_data_block_size;
1765 od->od_gen = bbt->bt_gen;
1766 dmu_buf_rele(db, FTAG);
1767 ztest_object_unlock(zd, od->od_object);
1775 ztest_create(ztest_ds_t *zd, ztest_od_t *od, int count)
1780 ASSERT(_mutex_held(&zd->zd_dirobj_lock));
1782 for (i = 0; i < count; i++, od++) {
1789 lr_create_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
1791 lr->lr_doid = od->od_dir;
1792 lr->lr_foid = 0; /* 0 to allocate, > 0 to claim */
1793 lr->lrz_type = od->od_crtype;
1794 lr->lrz_blocksize = od->od_crblocksize;
1795 lr->lrz_ibshift = ztest_random_ibshift();
1796 lr->lrz_bonustype = DMU_OT_UINT64_OTHER;
1797 lr->lrz_bonuslen = dmu_bonus_max();
1798 lr->lr_gen = od->od_crgen;
1799 lr->lr_crtime[0] = time(NULL);
1801 if (ztest_replay_create(zd, lr, B_FALSE) != 0) {
1802 ASSERT(missing == 0);
1806 od->od_object = lr->lr_foid;
1807 od->od_type = od->od_crtype;
1808 od->od_blocksize = od->od_crblocksize;
1809 od->od_gen = od->od_crgen;
1810 ASSERT(od->od_object != 0);
1813 ztest_lr_free(lr, sizeof (*lr), od->od_name);
1820 ztest_remove(ztest_ds_t *zd, ztest_od_t *od, int count)
1826 ASSERT(_mutex_held(&zd->zd_dirobj_lock));
1830 for (i = count - 1; i >= 0; i--, od--) {
1836 if (od->od_object == 0)
1839 lr_remove_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
1841 lr->lr_doid = od->od_dir;
1843 if ((error = ztest_replay_remove(zd, lr, B_FALSE)) != 0) {
1844 ASSERT3U(error, ==, ENOSPC);
1849 ztest_lr_free(lr, sizeof (*lr), od->od_name);
1856 ztest_write(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size,
1862 lr = ztest_lr_alloc(sizeof (*lr) + size, NULL);
1864 lr->lr_foid = object;
1865 lr->lr_offset = offset;
1866 lr->lr_length = size;
1868 BP_ZERO(&lr->lr_blkptr);
1870 bcopy(data, lr + 1, size);
1872 error = ztest_replay_write(zd, lr, B_FALSE);
1874 ztest_lr_free(lr, sizeof (*lr) + size, NULL);
1880 ztest_truncate(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
1885 lr = ztest_lr_alloc(sizeof (*lr), NULL);
1887 lr->lr_foid = object;
1888 lr->lr_offset = offset;
1889 lr->lr_length = size;
1891 error = ztest_replay_truncate(zd, lr, B_FALSE);
1893 ztest_lr_free(lr, sizeof (*lr), NULL);
1899 ztest_setattr(ztest_ds_t *zd, uint64_t object)
1904 lr = ztest_lr_alloc(sizeof (*lr), NULL);
1906 lr->lr_foid = object;
1910 error = ztest_replay_setattr(zd, lr, B_FALSE);
1912 ztest_lr_free(lr, sizeof (*lr), NULL);
1918 ztest_prealloc(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
1920 objset_t *os = zd->zd_os;
1925 txg_wait_synced(dmu_objset_pool(os), 0);
1927 ztest_object_lock(zd, object, RL_READER);
1928 rl = ztest_range_lock(zd, object, offset, size, RL_WRITER);
1930 tx = dmu_tx_create(os);
1932 dmu_tx_hold_write(tx, object, offset, size);
1934 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1937 dmu_prealloc(os, object, offset, size, tx);
1939 txg_wait_synced(dmu_objset_pool(os), txg);
1941 (void) dmu_free_long_range(os, object, offset, size);
1944 ztest_range_unlock(rl);
1945 ztest_object_unlock(zd, object);
1949 ztest_io(ztest_ds_t *zd, uint64_t object, uint64_t offset)
1951 ztest_block_tag_t wbt;
1952 dmu_object_info_t doi;
1953 enum ztest_io_type io_type;
1957 VERIFY(dmu_object_info(zd->zd_os, object, &doi) == 0);
1958 blocksize = doi.doi_data_block_size;
1959 data = umem_alloc(blocksize, UMEM_NOFAIL);
1962 * Pick an i/o type at random, biased toward writing block tags.
1964 io_type = ztest_random(ZTEST_IO_TYPES);
1965 if (ztest_random(2) == 0)
1966 io_type = ZTEST_IO_WRITE_TAG;
1970 case ZTEST_IO_WRITE_TAG:
1971 ztest_bt_generate(&wbt, zd->zd_os, object, offset, 0, 0, 0);
1972 (void) ztest_write(zd, object, offset, sizeof (wbt), &wbt);
1975 case ZTEST_IO_WRITE_PATTERN:
1976 (void) memset(data, 'a' + (object + offset) % 5, blocksize);
1977 if (ztest_random(2) == 0) {
1979 * Induce fletcher2 collisions to ensure that
1980 * zio_ddt_collision() detects and resolves them
1981 * when using fletcher2-verify for deduplication.
1983 ((uint64_t *)data)[0] ^= 1ULL << 63;
1984 ((uint64_t *)data)[4] ^= 1ULL << 63;
1986 (void) ztest_write(zd, object, offset, blocksize, data);
1989 case ZTEST_IO_WRITE_ZEROES:
1990 bzero(data, blocksize);
1991 (void) ztest_write(zd, object, offset, blocksize, data);
1994 case ZTEST_IO_TRUNCATE:
1995 (void) ztest_truncate(zd, object, offset, blocksize);
1998 case ZTEST_IO_SETATTR:
1999 (void) ztest_setattr(zd, object);
2003 umem_free(data, blocksize);
2007 * Initialize an object description template.
2010 ztest_od_init(ztest_od_t *od, uint64_t id, char *tag, uint64_t index,
2011 dmu_object_type_t type, uint64_t blocksize, uint64_t gen)
2013 od->od_dir = ZTEST_DIROBJ;
2016 od->od_crtype = type;
2017 od->od_crblocksize = blocksize ? blocksize : ztest_random_blocksize();
2020 od->od_type = DMU_OT_NONE;
2021 od->od_blocksize = 0;
2024 (void) snprintf(od->od_name, sizeof (od->od_name), "%s(%lld)[%llu]",
2025 tag, (longlong_t)id, (u_longlong_t)index);
2029 * Lookup or create the objects for a test using the od template.
2030 * If the objects do not all exist, or if 'remove' is specified,
2031 * remove any existing objects and create new ones. Otherwise,
2032 * use the existing objects.
2035 ztest_object_init(ztest_ds_t *zd, ztest_od_t *od, size_t size, boolean_t remove)
2037 int count = size / sizeof (*od);
2040 VERIFY(mutex_lock(&zd->zd_dirobj_lock) == 0);
2041 if ((ztest_lookup(zd, od, count) != 0 || remove) &&
2042 (ztest_remove(zd, od, count) != 0 ||
2043 ztest_create(zd, od, count) != 0))
2046 VERIFY(mutex_unlock(&zd->zd_dirobj_lock) == 0);
2053 ztest_zil_commit(ztest_ds_t *zd, uint64_t id)
2055 zilog_t *zilog = zd->zd_zilog;
2057 zil_commit(zilog, ztest_random(ZTEST_OBJECTS));
2060 * Remember the committed values in zd, which is in parent/child
2061 * shared memory. If we die, the next iteration of ztest_run()
2062 * will verify that the log really does contain this record.
2064 mutex_enter(&zilog->zl_lock);
2065 ASSERT(zd->zd_seq <= zilog->zl_commit_lr_seq);
2066 zd->zd_seq = zilog->zl_commit_lr_seq;
2067 mutex_exit(&zilog->zl_lock);
2071 * Verify that we can't destroy an active pool, create an existing pool,
2072 * or create a pool with a bad vdev spec.
2076 ztest_spa_create_destroy(ztest_ds_t *zd, uint64_t id)
2078 ztest_shared_t *zs = ztest_shared;
2083 * Attempt to create using a bad file.
2085 nvroot = make_vdev_root("/dev/bogus", NULL, 0, 0, 0, 0, 0, 1);
2086 VERIFY3U(ENOENT, ==,
2087 spa_create("ztest_bad_file", nvroot, NULL, NULL, NULL));
2088 nvlist_free(nvroot);
2091 * Attempt to create using a bad mirror.
2093 nvroot = make_vdev_root("/dev/bogus", NULL, 0, 0, 0, 0, 2, 1);
2094 VERIFY3U(ENOENT, ==,
2095 spa_create("ztest_bad_mirror", nvroot, NULL, NULL, NULL));
2096 nvlist_free(nvroot);
2099 * Attempt to create an existing pool. It shouldn't matter
2100 * what's in the nvroot; we should fail with EEXIST.
2102 (void) rw_rdlock(&zs->zs_name_lock);
2103 nvroot = make_vdev_root("/dev/bogus", NULL, 0, 0, 0, 0, 0, 1);
2104 VERIFY3U(EEXIST, ==, spa_create(zs->zs_pool, nvroot, NULL, NULL, NULL));
2105 nvlist_free(nvroot);
2106 VERIFY3U(0, ==, spa_open(zs->zs_pool, &spa, FTAG));
2107 VERIFY3U(EBUSY, ==, spa_destroy(zs->zs_pool));
2108 spa_close(spa, FTAG);
2110 (void) rw_unlock(&zs->zs_name_lock);
2114 vdev_lookup_by_path(vdev_t *vd, const char *path)
2119 if (vd->vdev_path != NULL && strcmp(path, vd->vdev_path) == 0)
2122 for (c = 0; c < vd->vdev_children; c++)
2123 if ((mvd = vdev_lookup_by_path(vd->vdev_child[c], path)) !=
2131 * Find the first available hole which can be used as a top-level.
2134 find_vdev_hole(spa_t *spa)
2136 vdev_t *rvd = spa->spa_root_vdev;
2139 ASSERT(spa_config_held(spa, SCL_VDEV, RW_READER) == SCL_VDEV);
2141 for (c = 0; c < rvd->vdev_children; c++) {
2142 vdev_t *cvd = rvd->vdev_child[c];
2144 if (cvd->vdev_ishole)
2151 * Verify that vdev_add() works as expected.
2155 ztest_vdev_add_remove(ztest_ds_t *zd, uint64_t id)
2157 ztest_shared_t *zs = ztest_shared;
2158 spa_t *spa = zs->zs_spa;
2164 VERIFY(mutex_lock(&zs->zs_vdev_lock) == 0);
2165 leaves = MAX(zs->zs_mirrors + zs->zs_splits, 1) * zopt_raidz;
2167 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2169 ztest_shared->zs_vdev_next_leaf = find_vdev_hole(spa) * leaves;
2172 * If we have slogs then remove them 1/4 of the time.
2174 if (spa_has_slogs(spa) && ztest_random(4) == 0) {
2176 * Grab the guid from the head of the log class rotor.
2178 guid = spa_log_class(spa)->mc_rotor->mg_vd->vdev_guid;
2180 spa_config_exit(spa, SCL_VDEV, FTAG);
2183 * We have to grab the zs_name_lock as writer to
2184 * prevent a race between removing a slog (dmu_objset_find)
2185 * and destroying a dataset. Removing the slog will
2186 * grab a reference on the dataset which may cause
2187 * dmu_objset_destroy() to fail with EBUSY thus
2188 * leaving the dataset in an inconsistent state.
2190 VERIFY(rw_wrlock(&ztest_shared->zs_name_lock) == 0);
2191 error = spa_vdev_remove(spa, guid, B_FALSE);
2192 VERIFY(rw_unlock(&ztest_shared->zs_name_lock) == 0);
2194 if (error && error != EEXIST)
2195 fatal(0, "spa_vdev_remove() = %d", error);
2197 spa_config_exit(spa, SCL_VDEV, FTAG);
2200 * Make 1/4 of the devices be log devices.
2202 nvroot = make_vdev_root(NULL, NULL, zopt_vdev_size, 0,
2203 ztest_random(4) == 0, zopt_raidz, zs->zs_mirrors, 1);
2205 error = spa_vdev_add(spa, nvroot);
2206 nvlist_free(nvroot);
2208 if (error == ENOSPC)
2209 ztest_record_enospc("spa_vdev_add");
2210 else if (error != 0)
2211 fatal(0, "spa_vdev_add() = %d", error);
2214 VERIFY(mutex_unlock(&ztest_shared->zs_vdev_lock) == 0);
2218 * Verify that adding/removing aux devices (l2arc, hot spare) works as expected.
2222 ztest_vdev_aux_add_remove(ztest_ds_t *zd, uint64_t id)
2224 ztest_shared_t *zs = ztest_shared;
2225 spa_t *spa = zs->zs_spa;
2226 vdev_t *rvd = spa->spa_root_vdev;
2227 spa_aux_vdev_t *sav;
2232 if (ztest_random(2) == 0) {
2233 sav = &spa->spa_spares;
2234 aux = ZPOOL_CONFIG_SPARES;
2236 sav = &spa->spa_l2cache;
2237 aux = ZPOOL_CONFIG_L2CACHE;
2240 VERIFY(mutex_lock(&zs->zs_vdev_lock) == 0);
2242 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2244 if (sav->sav_count != 0 && ztest_random(4) == 0) {
2246 * Pick a random device to remove.
2248 guid = sav->sav_vdevs[ztest_random(sav->sav_count)]->vdev_guid;
2251 * Find an unused device we can add.
2253 zs->zs_vdev_aux = 0;
2255 char path[MAXPATHLEN];
2257 (void) sprintf(path, ztest_aux_template, zopt_dir,
2258 zopt_pool, aux, zs->zs_vdev_aux);
2259 for (c = 0; c < sav->sav_count; c++)
2260 if (strcmp(sav->sav_vdevs[c]->vdev_path,
2263 if (c == sav->sav_count &&
2264 vdev_lookup_by_path(rvd, path) == NULL)
2270 spa_config_exit(spa, SCL_VDEV, FTAG);
2276 nvlist_t *nvroot = make_vdev_root(NULL, aux,
2277 (zopt_vdev_size * 5) / 4, 0, 0, 0, 0, 1);
2278 error = spa_vdev_add(spa, nvroot);
2280 fatal(0, "spa_vdev_add(%p) = %d", nvroot, error);
2281 nvlist_free(nvroot);
2284 * Remove an existing device. Sometimes, dirty its
2285 * vdev state first to make sure we handle removal
2286 * of devices that have pending state changes.
2288 if (ztest_random(2) == 0)
2289 (void) vdev_online(spa, guid, 0, NULL);
2291 error = spa_vdev_remove(spa, guid, B_FALSE);
2292 if (error != 0 && error != EBUSY)
2293 fatal(0, "spa_vdev_remove(%llu) = %d", guid, error);
2296 VERIFY(mutex_unlock(&zs->zs_vdev_lock) == 0);
2300 * split a pool if it has mirror tlvdevs
2304 ztest_split_pool(ztest_ds_t *zd, uint64_t id)
2306 ztest_shared_t *zs = ztest_shared;
2307 spa_t *spa = zs->zs_spa;
2308 vdev_t *rvd = spa->spa_root_vdev;
2309 nvlist_t *tree, **child, *config, *split, **schild;
2310 uint_t c, children, schildren = 0, lastlogid = 0;
2313 VERIFY(mutex_lock(&zs->zs_vdev_lock) == 0);
2315 /* ensure we have a useable config; mirrors of raidz aren't supported */
2316 if (zs->zs_mirrors < 3 || zopt_raidz > 1) {
2317 VERIFY(mutex_unlock(&zs->zs_vdev_lock) == 0);
2321 /* clean up the old pool, if any */
2322 (void) spa_destroy("splitp");
2324 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2326 /* generate a config from the existing config */
2327 mutex_enter(&spa->spa_props_lock);
2328 VERIFY(nvlist_lookup_nvlist(spa->spa_config, ZPOOL_CONFIG_VDEV_TREE,
2330 mutex_exit(&spa->spa_props_lock);
2332 VERIFY(nvlist_lookup_nvlist_array(tree, ZPOOL_CONFIG_CHILDREN, &child,
2335 schild = malloc(rvd->vdev_children * sizeof (nvlist_t *));
2336 for (c = 0; c < children; c++) {
2337 vdev_t *tvd = rvd->vdev_child[c];
2341 if (tvd->vdev_islog || tvd->vdev_ops == &vdev_hole_ops) {
2342 VERIFY(nvlist_alloc(&schild[schildren], NV_UNIQUE_NAME,
2344 VERIFY(nvlist_add_string(schild[schildren],
2345 ZPOOL_CONFIG_TYPE, VDEV_TYPE_HOLE) == 0);
2346 VERIFY(nvlist_add_uint64(schild[schildren],
2347 ZPOOL_CONFIG_IS_HOLE, 1) == 0);
2349 lastlogid = schildren;
2354 VERIFY(nvlist_lookup_nvlist_array(child[c],
2355 ZPOOL_CONFIG_CHILDREN, &mchild, &mchildren) == 0);
2356 VERIFY(nvlist_dup(mchild[0], &schild[schildren++], 0) == 0);
2359 /* OK, create a config that can be used to split */
2360 VERIFY(nvlist_alloc(&split, NV_UNIQUE_NAME, 0) == 0);
2361 VERIFY(nvlist_add_string(split, ZPOOL_CONFIG_TYPE,
2362 VDEV_TYPE_ROOT) == 0);
2363 VERIFY(nvlist_add_nvlist_array(split, ZPOOL_CONFIG_CHILDREN, schild,
2364 lastlogid != 0 ? lastlogid : schildren) == 0);
2366 VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, 0) == 0);
2367 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, split) == 0);
2369 for (c = 0; c < schildren; c++)
2370 nvlist_free(schild[c]);
2374 spa_config_exit(spa, SCL_VDEV, FTAG);
2376 (void) rw_wrlock(&zs->zs_name_lock);
2377 error = spa_vdev_split_mirror(spa, "splitp", config, NULL, B_FALSE);
2378 (void) rw_unlock(&zs->zs_name_lock);
2380 nvlist_free(config);
2383 (void) printf("successful split - results:\n");
2384 mutex_enter(&spa_namespace_lock);
2385 show_pool_stats(spa);
2386 show_pool_stats(spa_lookup("splitp"));
2387 mutex_exit(&spa_namespace_lock);
2391 VERIFY(mutex_unlock(&zs->zs_vdev_lock) == 0);
2396 * Verify that we can attach and detach devices.
2400 ztest_vdev_attach_detach(ztest_ds_t *zd, uint64_t id)
2402 ztest_shared_t *zs = ztest_shared;
2403 spa_t *spa = zs->zs_spa;
2404 spa_aux_vdev_t *sav = &spa->spa_spares;
2405 vdev_t *rvd = spa->spa_root_vdev;
2406 vdev_t *oldvd, *newvd, *pvd;
2410 uint64_t ashift = ztest_get_ashift();
2411 uint64_t oldguid, pguid;
2412 size_t oldsize, newsize;
2413 char oldpath[MAXPATHLEN], newpath[MAXPATHLEN];
2415 int oldvd_has_siblings = B_FALSE;
2416 int newvd_is_spare = B_FALSE;
2418 int error, expected_error;
2420 VERIFY(mutex_lock(&zs->zs_vdev_lock) == 0);
2421 leaves = MAX(zs->zs_mirrors, 1) * zopt_raidz;
2423 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2426 * Decide whether to do an attach or a replace.
2428 replacing = ztest_random(2);
2431 * Pick a random top-level vdev.
2433 top = ztest_random_vdev_top(spa, B_TRUE);
2436 * Pick a random leaf within it.
2438 leaf = ztest_random(leaves);
2443 oldvd = rvd->vdev_child[top];
2444 if (zs->zs_mirrors >= 1) {
2445 ASSERT(oldvd->vdev_ops == &vdev_mirror_ops);
2446 ASSERT(oldvd->vdev_children >= zs->zs_mirrors);
2447 oldvd = oldvd->vdev_child[leaf / zopt_raidz];
2449 if (zopt_raidz > 1) {
2450 ASSERT(oldvd->vdev_ops == &vdev_raidz_ops);
2451 ASSERT(oldvd->vdev_children == zopt_raidz);
2452 oldvd = oldvd->vdev_child[leaf % zopt_raidz];
2456 * If we're already doing an attach or replace, oldvd may be a
2457 * mirror vdev -- in which case, pick a random child.
2459 while (oldvd->vdev_children != 0) {
2460 oldvd_has_siblings = B_TRUE;
2461 ASSERT(oldvd->vdev_children >= 2);
2462 oldvd = oldvd->vdev_child[ztest_random(oldvd->vdev_children)];
2465 oldguid = oldvd->vdev_guid;
2466 oldsize = vdev_get_min_asize(oldvd);
2467 oldvd_is_log = oldvd->vdev_top->vdev_islog;
2468 (void) strcpy(oldpath, oldvd->vdev_path);
2469 pvd = oldvd->vdev_parent;
2470 pguid = pvd->vdev_guid;
2473 * If oldvd has siblings, then half of the time, detach it.
2475 if (oldvd_has_siblings && ztest_random(2) == 0) {
2476 spa_config_exit(spa, SCL_VDEV, FTAG);
2477 error = spa_vdev_detach(spa, oldguid, pguid, B_FALSE);
2478 if (error != 0 && error != ENODEV && error != EBUSY &&
2480 fatal(0, "detach (%s) returned %d", oldpath, error);
2481 VERIFY(mutex_unlock(&zs->zs_vdev_lock) == 0);
2486 * For the new vdev, choose with equal probability between the two
2487 * standard paths (ending in either 'a' or 'b') or a random hot spare.
2489 if (sav->sav_count != 0 && ztest_random(3) == 0) {
2490 newvd = sav->sav_vdevs[ztest_random(sav->sav_count)];
2491 newvd_is_spare = B_TRUE;
2492 (void) strcpy(newpath, newvd->vdev_path);
2494 (void) snprintf(newpath, sizeof (newpath), ztest_dev_template,
2495 zopt_dir, zopt_pool, top * leaves + leaf);
2496 if (ztest_random(2) == 0)
2497 newpath[strlen(newpath) - 1] = 'b';
2498 newvd = vdev_lookup_by_path(rvd, newpath);
2502 newsize = vdev_get_min_asize(newvd);
2505 * Make newsize a little bigger or smaller than oldsize.
2506 * If it's smaller, the attach should fail.
2507 * If it's larger, and we're doing a replace,
2508 * we should get dynamic LUN growth when we're done.
2510 newsize = 10 * oldsize / (9 + ztest_random(3));
2514 * If pvd is not a mirror or root, the attach should fail with ENOTSUP,
2515 * unless it's a replace; in that case any non-replacing parent is OK.
2517 * If newvd is already part of the pool, it should fail with EBUSY.
2519 * If newvd is too small, it should fail with EOVERFLOW.
2521 if (pvd->vdev_ops != &vdev_mirror_ops &&
2522 pvd->vdev_ops != &vdev_root_ops && (!replacing ||
2523 pvd->vdev_ops == &vdev_replacing_ops ||
2524 pvd->vdev_ops == &vdev_spare_ops))
2525 expected_error = ENOTSUP;
2526 else if (newvd_is_spare && (!replacing || oldvd_is_log))
2527 expected_error = ENOTSUP;
2528 else if (newvd == oldvd)
2529 expected_error = replacing ? 0 : EBUSY;
2530 else if (vdev_lookup_by_path(rvd, newpath) != NULL)
2531 expected_error = EBUSY;
2532 else if (newsize < oldsize)
2533 expected_error = EOVERFLOW;
2534 else if (ashift > oldvd->vdev_top->vdev_ashift)
2535 expected_error = EDOM;
2539 spa_config_exit(spa, SCL_VDEV, FTAG);
2542 * Build the nvlist describing newpath.
2544 root = make_vdev_root(newpath, NULL, newvd == NULL ? newsize : 0,
2545 ashift, 0, 0, 0, 1);
2547 error = spa_vdev_attach(spa, oldguid, root, replacing);
2552 * If our parent was the replacing vdev, but the replace completed,
2553 * then instead of failing with ENOTSUP we may either succeed,
2554 * fail with ENODEV, or fail with EOVERFLOW.
2556 if (expected_error == ENOTSUP &&
2557 (error == 0 || error == ENODEV || error == EOVERFLOW))
2558 expected_error = error;
2561 * If someone grew the LUN, the replacement may be too small.
2563 if (error == EOVERFLOW || error == EBUSY)
2564 expected_error = error;
2566 /* XXX workaround 6690467 */
2567 if (error != expected_error && expected_error != EBUSY) {
2568 fatal(0, "attach (%s %llu, %s %llu, %d) "
2569 "returned %d, expected %d",
2570 oldpath, (longlong_t)oldsize, newpath,
2571 (longlong_t)newsize, replacing, error, expected_error);
2574 VERIFY(mutex_unlock(&zs->zs_vdev_lock) == 0);
2578 * Callback function which expands the physical size of the vdev.
2581 grow_vdev(vdev_t *vd, void *arg)
2583 spa_t *spa = vd->vdev_spa;
2584 size_t *newsize = arg;
2588 ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
2589 ASSERT(vd->vdev_ops->vdev_op_leaf);
2591 if ((fd = open(vd->vdev_path, O_RDWR)) == -1)
2594 fsize = lseek(fd, 0, SEEK_END);
2595 VERIFY(ftruncate(fd, *newsize) == 0);
2597 if (zopt_verbose >= 6) {
2598 (void) printf("%s grew from %lu to %lu bytes\n",
2599 vd->vdev_path, (ulong_t)fsize, (ulong_t)*newsize);
2606 * Callback function which expands a given vdev by calling vdev_online().
2610 online_vdev(vdev_t *vd, void *arg)
2612 spa_t *spa = vd->vdev_spa;
2613 vdev_t *tvd = vd->vdev_top;
2614 uint64_t guid = vd->vdev_guid;
2615 uint64_t generation = spa->spa_config_generation + 1;
2616 vdev_state_t newstate = VDEV_STATE_UNKNOWN;
2619 ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
2620 ASSERT(vd->vdev_ops->vdev_op_leaf);
2622 /* Calling vdev_online will initialize the new metaslabs */
2623 spa_config_exit(spa, SCL_STATE, spa);
2624 error = vdev_online(spa, guid, ZFS_ONLINE_EXPAND, &newstate);
2625 spa_config_enter(spa, SCL_STATE, spa, RW_READER);
2628 * If vdev_online returned an error or the underlying vdev_open
2629 * failed then we abort the expand. The only way to know that
2630 * vdev_open fails is by checking the returned newstate.
2632 if (error || newstate != VDEV_STATE_HEALTHY) {
2633 if (zopt_verbose >= 5) {
2634 (void) printf("Unable to expand vdev, state %llu, "
2635 "error %d\n", (u_longlong_t)newstate, error);
2639 ASSERT3U(newstate, ==, VDEV_STATE_HEALTHY);
2642 * Since we dropped the lock we need to ensure that we're
2643 * still talking to the original vdev. It's possible this
2644 * vdev may have been detached/replaced while we were
2645 * trying to online it.
2647 if (generation != spa->spa_config_generation) {
2648 if (zopt_verbose >= 5) {
2649 (void) printf("vdev configuration has changed, "
2650 "guid %llu, state %llu, expected gen %llu, "
2653 (u_longlong_t)tvd->vdev_state,
2654 (u_longlong_t)generation,
2655 (u_longlong_t)spa->spa_config_generation);
2663 * Traverse the vdev tree calling the supplied function.
2664 * We continue to walk the tree until we either have walked all
2665 * children or we receive a non-NULL return from the callback.
2666 * If a NULL callback is passed, then we just return back the first
2667 * leaf vdev we encounter.
2670 vdev_walk_tree(vdev_t *vd, vdev_t *(*func)(vdev_t *, void *), void *arg)
2674 if (vd->vdev_ops->vdev_op_leaf) {
2678 return (func(vd, arg));
2681 for (c = 0; c < vd->vdev_children; c++) {
2682 vdev_t *cvd = vd->vdev_child[c];
2683 if ((cvd = vdev_walk_tree(cvd, func, arg)) != NULL)
2690 * Verify that dynamic LUN growth works as expected.
2694 ztest_vdev_LUN_growth(ztest_ds_t *zd, uint64_t id)
2696 ztest_shared_t *zs = ztest_shared;
2697 spa_t *spa = zs->zs_spa;
2699 metaslab_class_t *mc;
2700 metaslab_group_t *mg;
2701 size_t psize, newsize;
2703 uint64_t old_class_space, new_class_space, old_ms_count, new_ms_count;
2705 VERIFY(mutex_lock(&zs->zs_vdev_lock) == 0);
2706 spa_config_enter(spa, SCL_STATE, spa, RW_READER);
2708 top = ztest_random_vdev_top(spa, B_TRUE);
2710 tvd = spa->spa_root_vdev->vdev_child[top];
2713 old_ms_count = tvd->vdev_ms_count;
2714 old_class_space = metaslab_class_get_space(mc);
2717 * Determine the size of the first leaf vdev associated with
2718 * our top-level device.
2720 vd = vdev_walk_tree(tvd, NULL, NULL);
2721 ASSERT3P(vd, !=, NULL);
2722 ASSERT(vd->vdev_ops->vdev_op_leaf);
2724 psize = vd->vdev_psize;
2727 * We only try to expand the vdev if it's healthy, less than 4x its
2728 * original size, and it has a valid psize.
2730 if (tvd->vdev_state != VDEV_STATE_HEALTHY ||
2731 psize == 0 || psize >= 4 * zopt_vdev_size) {
2732 spa_config_exit(spa, SCL_STATE, spa);
2733 VERIFY(mutex_unlock(&zs->zs_vdev_lock) == 0);
2737 newsize = psize + psize / 8;
2738 ASSERT3U(newsize, >, psize);
2740 if (zopt_verbose >= 6) {
2741 (void) printf("Expanding LUN %s from %lu to %lu\n",
2742 vd->vdev_path, (ulong_t)psize, (ulong_t)newsize);
2746 * Growing the vdev is a two step process:
2747 * 1). expand the physical size (i.e. relabel)
2748 * 2). online the vdev to create the new metaslabs
2750 if (vdev_walk_tree(tvd, grow_vdev, &newsize) != NULL ||
2751 vdev_walk_tree(tvd, online_vdev, NULL) != NULL ||
2752 tvd->vdev_state != VDEV_STATE_HEALTHY) {
2753 if (zopt_verbose >= 5) {
2754 (void) printf("Could not expand LUN because "
2755 "the vdev configuration changed.\n");
2757 spa_config_exit(spa, SCL_STATE, spa);
2758 VERIFY(mutex_unlock(&zs->zs_vdev_lock) == 0);
2762 spa_config_exit(spa, SCL_STATE, spa);
2765 * Expanding the LUN will update the config asynchronously,
2766 * thus we must wait for the async thread to complete any
2767 * pending tasks before proceeding.
2771 mutex_enter(&spa->spa_async_lock);
2772 done = (spa->spa_async_thread == NULL && !spa->spa_async_tasks);
2773 mutex_exit(&spa->spa_async_lock);
2776 txg_wait_synced(spa_get_dsl(spa), 0);
2777 (void) poll(NULL, 0, 100);
2780 spa_config_enter(spa, SCL_STATE, spa, RW_READER);
2782 tvd = spa->spa_root_vdev->vdev_child[top];
2783 new_ms_count = tvd->vdev_ms_count;
2784 new_class_space = metaslab_class_get_space(mc);
2786 if (tvd->vdev_mg != mg || mg->mg_class != mc) {
2787 if (zopt_verbose >= 5) {
2788 (void) printf("Could not verify LUN expansion due to "
2789 "intervening vdev offline or remove.\n");
2791 spa_config_exit(spa, SCL_STATE, spa);
2792 VERIFY(mutex_unlock(&zs->zs_vdev_lock) == 0);
2797 * Make sure we were able to grow the vdev.
2799 if (new_ms_count <= old_ms_count)
2800 fatal(0, "LUN expansion failed: ms_count %llu <= %llu\n",
2801 old_ms_count, new_ms_count);
2804 * Make sure we were able to grow the pool.
2806 if (new_class_space <= old_class_space)
2807 fatal(0, "LUN expansion failed: class_space %llu <= %llu\n",
2808 old_class_space, new_class_space);
2810 if (zopt_verbose >= 5) {
2811 char oldnumbuf[6], newnumbuf[6];
2813 nicenum(old_class_space, oldnumbuf);
2814 nicenum(new_class_space, newnumbuf);
2815 (void) printf("%s grew from %s to %s\n",
2816 spa->spa_name, oldnumbuf, newnumbuf);
2819 spa_config_exit(spa, SCL_STATE, spa);
2820 VERIFY(mutex_unlock(&zs->zs_vdev_lock) == 0);
2824 * Verify that dmu_objset_{create,destroy,open,close} work as expected.
2828 ztest_objset_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
2831 * Create the objects common to all ztest datasets.
2833 VERIFY(zap_create_claim(os, ZTEST_DIROBJ,
2834 DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx) == 0);
2838 ztest_dataset_create(char *dsname)
2840 uint64_t zilset = ztest_random(100);
2841 int err = dmu_objset_create(dsname, DMU_OST_OTHER, 0,
2842 ztest_objset_create_cb, NULL);
2844 if (err || zilset < 80)
2847 (void) printf("Setting dataset %s to sync always\n", dsname);
2848 return (ztest_dsl_prop_set_uint64(dsname, ZFS_PROP_SYNC,
2849 ZFS_SYNC_ALWAYS, B_FALSE));
2854 ztest_objset_destroy_cb(const char *name, void *arg)
2857 dmu_object_info_t doi;
2861 * Verify that the dataset contains a directory object.
2863 VERIFY3U(0, ==, dmu_objset_hold(name, FTAG, &os));
2864 error = dmu_object_info(os, ZTEST_DIROBJ, &doi);
2865 if (error != ENOENT) {
2866 /* We could have crashed in the middle of destroying it */
2867 ASSERT3U(error, ==, 0);
2868 ASSERT3U(doi.doi_type, ==, DMU_OT_ZAP_OTHER);
2869 ASSERT3S(doi.doi_physical_blocks_512, >=, 0);
2871 dmu_objset_rele(os, FTAG);
2874 * Destroy the dataset.
2876 VERIFY3U(0, ==, dmu_objset_destroy(name, B_FALSE));
2881 ztest_snapshot_create(char *osname, uint64_t id)
2883 char snapname[MAXNAMELEN];
2886 (void) snprintf(snapname, MAXNAMELEN, "%s@%llu", osname,
2889 error = dmu_objset_snapshot(osname, strchr(snapname, '@') + 1,
2890 NULL, NULL, B_FALSE, B_FALSE, -1);
2891 if (error == ENOSPC) {
2892 ztest_record_enospc(FTAG);
2895 if (error != 0 && error != EEXIST)
2896 fatal(0, "ztest_snapshot_create(%s) = %d", snapname, error);
2901 ztest_snapshot_destroy(char *osname, uint64_t id)
2903 char snapname[MAXNAMELEN];
2906 (void) snprintf(snapname, MAXNAMELEN, "%s@%llu", osname,
2909 error = dmu_objset_destroy(snapname, B_FALSE);
2910 if (error != 0 && error != ENOENT)
2911 fatal(0, "ztest_snapshot_destroy(%s) = %d", snapname, error);
2917 ztest_dmu_objset_create_destroy(ztest_ds_t *zd, uint64_t id)
2919 ztest_shared_t *zs = ztest_shared;
2924 char name[MAXNAMELEN];
2928 (void) rw_rdlock(&zs->zs_name_lock);
2930 (void) snprintf(name, MAXNAMELEN, "%s/temp_%llu",
2931 zs->zs_pool, (u_longlong_t)id);
2934 * If this dataset exists from a previous run, process its replay log
2935 * half of the time. If we don't replay it, then dmu_objset_destroy()
2936 * (invoked from ztest_objset_destroy_cb()) should just throw it away.
2938 if (ztest_random(2) == 0 &&
2939 dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os) == 0) {
2940 ztest_zd_init(&zdtmp, os);
2941 zil_replay(os, &zdtmp, ztest_replay_vector);
2942 ztest_zd_fini(&zdtmp);
2943 dmu_objset_disown(os, FTAG);
2947 * There may be an old instance of the dataset we're about to
2948 * create lying around from a previous run. If so, destroy it
2949 * and all of its snapshots.
2951 (void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
2952 DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
2955 * Verify that the destroyed dataset is no longer in the namespace.
2957 VERIFY3U(ENOENT, ==, dmu_objset_hold(name, FTAG, &os));
2960 * Verify that we can create a new dataset.
2962 error = ztest_dataset_create(name);
2964 if (error == ENOSPC) {
2965 ztest_record_enospc(FTAG);
2966 (void) rw_unlock(&zs->zs_name_lock);
2969 fatal(0, "dmu_objset_create(%s) = %d", name, error);
2973 dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os));
2975 ztest_zd_init(&zdtmp, os);
2978 * Open the intent log for it.
2980 zilog = zil_open(os, ztest_get_data);
2983 * Put some objects in there, do a little I/O to them,
2984 * and randomly take a couple of snapshots along the way.
2986 iters = ztest_random(5);
2987 for (i = 0; i < iters; i++) {
2988 ztest_dmu_object_alloc_free(&zdtmp, id);
2989 if (ztest_random(iters) == 0)
2990 (void) ztest_snapshot_create(name, i);
2994 * Verify that we cannot create an existing dataset.
2996 VERIFY3U(EEXIST, ==,
2997 dmu_objset_create(name, DMU_OST_OTHER, 0, NULL, NULL));
3000 * Verify that we can hold an objset that is also owned.
3002 VERIFY3U(0, ==, dmu_objset_hold(name, FTAG, &os2));
3003 dmu_objset_rele(os2, FTAG);
3006 * Verify that we cannot own an objset that is already owned.
3009 dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os2));
3012 dmu_objset_disown(os, FTAG);
3013 ztest_zd_fini(&zdtmp);
3015 (void) rw_unlock(&zs->zs_name_lock);
3019 * Verify that dmu_snapshot_{create,destroy,open,close} work as expected.
3022 ztest_dmu_snapshot_create_destroy(ztest_ds_t *zd, uint64_t id)
3024 ztest_shared_t *zs = ztest_shared;
3026 (void) rw_rdlock(&zs->zs_name_lock);
3027 (void) ztest_snapshot_destroy(zd->zd_name, id);
3028 (void) ztest_snapshot_create(zd->zd_name, id);
3029 (void) rw_unlock(&zs->zs_name_lock);
3033 * Cleanup non-standard snapshots and clones.
3036 ztest_dsl_dataset_cleanup(char *osname, uint64_t id)
3038 char snap1name[MAXNAMELEN];
3039 char clone1name[MAXNAMELEN];
3040 char snap2name[MAXNAMELEN];
3041 char clone2name[MAXNAMELEN];
3042 char snap3name[MAXNAMELEN];
3045 (void) snprintf(snap1name, MAXNAMELEN, "%s@s1_%llu",
3046 osname, (u_longlong_t)id);
3047 (void) snprintf(clone1name, MAXNAMELEN, "%s/c1_%llu",
3048 osname, (u_longlong_t)id);
3049 (void) snprintf(snap2name, MAXNAMELEN, "%s@s2_%llu",
3050 clone1name, (u_longlong_t)id);
3051 (void) snprintf(clone2name, MAXNAMELEN, "%s/c2_%llu",
3052 osname, (u_longlong_t)id);
3053 (void) snprintf(snap3name, MAXNAMELEN, "%s@s3_%llu",
3054 clone1name, (u_longlong_t)id);
3056 error = dmu_objset_destroy(clone2name, B_FALSE);
3057 if (error && error != ENOENT)
3058 fatal(0, "dmu_objset_destroy(%s) = %d", clone2name, error);
3059 error = dmu_objset_destroy(snap3name, B_FALSE);
3060 if (error && error != ENOENT)
3061 fatal(0, "dmu_objset_destroy(%s) = %d", snap3name, error);
3062 error = dmu_objset_destroy(snap2name, B_FALSE);
3063 if (error && error != ENOENT)
3064 fatal(0, "dmu_objset_destroy(%s) = %d", snap2name, error);
3065 error = dmu_objset_destroy(clone1name, B_FALSE);
3066 if (error && error != ENOENT)
3067 fatal(0, "dmu_objset_destroy(%s) = %d", clone1name, error);
3068 error = dmu_objset_destroy(snap1name, B_FALSE);
3069 if (error && error != ENOENT)
3070 fatal(0, "dmu_objset_destroy(%s) = %d", snap1name, error);
3074 * Verify dsl_dataset_promote handles EBUSY
3077 ztest_dsl_dataset_promote_busy(ztest_ds_t *zd, uint64_t id)
3079 ztest_shared_t *zs = ztest_shared;
3082 char snap1name[MAXNAMELEN];
3083 char clone1name[MAXNAMELEN];
3084 char snap2name[MAXNAMELEN];
3085 char clone2name[MAXNAMELEN];
3086 char snap3name[MAXNAMELEN];
3087 char *osname = zd->zd_name;
3090 (void) rw_rdlock(&zs->zs_name_lock);
3092 ztest_dsl_dataset_cleanup(osname, id);
3094 (void) snprintf(snap1name, MAXNAMELEN, "%s@s1_%llu",
3095 osname, (u_longlong_t)id);
3096 (void) snprintf(clone1name, MAXNAMELEN, "%s/c1_%llu",
3097 osname, (u_longlong_t)id);
3098 (void) snprintf(snap2name, MAXNAMELEN, "%s@s2_%llu",
3099 clone1name, (u_longlong_t)id);
3100 (void) snprintf(clone2name, MAXNAMELEN, "%s/c2_%llu",
3101 osname, (u_longlong_t)id);
3102 (void) snprintf(snap3name, MAXNAMELEN, "%s@s3_%llu",
3103 clone1name, (u_longlong_t)id);
3105 error = dmu_objset_snapshot(osname, strchr(snap1name, '@')+1,
3106 NULL, NULL, B_FALSE, B_FALSE, -1);
3107 if (error && error != EEXIST) {
3108 if (error == ENOSPC) {
3109 ztest_record_enospc(FTAG);
3112 fatal(0, "dmu_take_snapshot(%s) = %d", snap1name, error);
3115 error = dmu_objset_hold(snap1name, FTAG, &clone);
3117 fatal(0, "dmu_open_snapshot(%s) = %d", snap1name, error);
3119 error = dmu_objset_clone(clone1name, dmu_objset_ds(clone), 0);
3120 dmu_objset_rele(clone, FTAG);
3122 if (error == ENOSPC) {
3123 ztest_record_enospc(FTAG);
3126 fatal(0, "dmu_objset_create(%s) = %d", clone1name, error);
3129 error = dmu_objset_snapshot(clone1name, strchr(snap2name, '@')+1,
3130 NULL, NULL, B_FALSE, B_FALSE, -1);
3131 if (error && error != EEXIST) {
3132 if (error == ENOSPC) {
3133 ztest_record_enospc(FTAG);
3136 fatal(0, "dmu_open_snapshot(%s) = %d", snap2name, error);
3139 error = dmu_objset_snapshot(clone1name, strchr(snap3name, '@')+1,
3140 NULL, NULL, B_FALSE, B_FALSE, -1);
3141 if (error && error != EEXIST) {
3142 if (error == ENOSPC) {
3143 ztest_record_enospc(FTAG);
3146 fatal(0, "dmu_open_snapshot(%s) = %d", snap3name, error);
3149 error = dmu_objset_hold(snap3name, FTAG, &clone);
3151 fatal(0, "dmu_open_snapshot(%s) = %d", snap3name, error);
3153 error = dmu_objset_clone(clone2name, dmu_objset_ds(clone), 0);
3154 dmu_objset_rele(clone, FTAG);
3156 if (error == ENOSPC) {
3157 ztest_record_enospc(FTAG);
3160 fatal(0, "dmu_objset_create(%s) = %d", clone2name, error);
3163 error = dsl_dataset_own(snap2name, B_FALSE, FTAG, &ds);
3165 fatal(0, "dsl_dataset_own(%s) = %d", snap2name, error);
3166 error = dsl_dataset_promote(clone2name, NULL);
3168 fatal(0, "dsl_dataset_promote(%s), %d, not EBUSY", clone2name,
3170 dsl_dataset_disown(ds, FTAG);
3173 ztest_dsl_dataset_cleanup(osname, id);
3175 (void) rw_unlock(&zs->zs_name_lock);
3179 * Verify that dmu_object_{alloc,free} work as expected.
3182 ztest_dmu_object_alloc_free(ztest_ds_t *zd, uint64_t id)
3185 int batchsize = sizeof (od) / sizeof (od[0]);
3188 for (b = 0; b < batchsize; b++)
3189 ztest_od_init(&od[b], id, FTAG, b, DMU_OT_UINT64_OTHER, 0, 0);
3192 * Destroy the previous batch of objects, create a new batch,
3193 * and do some I/O on the new objects.
3195 if (ztest_object_init(zd, od, sizeof (od), B_TRUE) != 0)
3198 while (ztest_random(4 * batchsize) != 0)
3199 ztest_io(zd, od[ztest_random(batchsize)].od_object,
3200 ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
3204 * Verify that dmu_{read,write} work as expected.
3207 ztest_dmu_read_write(ztest_ds_t *zd, uint64_t id)
3209 objset_t *os = zd->zd_os;
3212 int i, freeit, error;
3214 bufwad_t *packbuf, *bigbuf, *pack, *bigH, *bigT;
3215 uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
3216 uint64_t chunksize = (1000 + ztest_random(1000)) * sizeof (uint64_t);
3217 uint64_t regions = 997;
3218 uint64_t stride = 123456789ULL;
3219 uint64_t width = 40;
3220 int free_percent = 5;
3223 * This test uses two objects, packobj and bigobj, that are always
3224 * updated together (i.e. in the same tx) so that their contents are
3225 * in sync and can be compared. Their contents relate to each other
3226 * in a simple way: packobj is a dense array of 'bufwad' structures,
3227 * while bigobj is a sparse array of the same bufwads. Specifically,
3228 * for any index n, there are three bufwads that should be identical:
3230 * packobj, at offset n * sizeof (bufwad_t)
3231 * bigobj, at the head of the nth chunk
3232 * bigobj, at the tail of the nth chunk
3234 * The chunk size is arbitrary. It doesn't have to be a power of two,
3235 * and it doesn't have any relation to the object blocksize.
3236 * The only requirement is that it can hold at least two bufwads.
3238 * Normally, we write the bufwad to each of these locations.
3239 * However, free_percent of the time we instead write zeroes to
3240 * packobj and perform a dmu_free_range() on bigobj. By comparing
3241 * bigobj to packobj, we can verify that the DMU is correctly
3242 * tracking which parts of an object are allocated and free,
3243 * and that the contents of the allocated blocks are correct.
3247 * Read the directory info. If it's the first time, set things up.
3249 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, chunksize);
3250 ztest_od_init(&od[1], id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize);
3252 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
3255 bigobj = od[0].od_object;
3256 packobj = od[1].od_object;
3257 chunksize = od[0].od_gen;
3258 ASSERT(chunksize == od[1].od_gen);
3261 * Prefetch a random chunk of the big object.
3262 * Our aim here is to get some async reads in flight
3263 * for blocks that we may free below; the DMU should
3264 * handle this race correctly.
3266 n = ztest_random(regions) * stride + ztest_random(width);
3267 s = 1 + ztest_random(2 * width - 1);
3268 dmu_prefetch(os, bigobj, n * chunksize, s * chunksize);
3271 * Pick a random index and compute the offsets into packobj and bigobj.
3273 n = ztest_random(regions) * stride + ztest_random(width);
3274 s = 1 + ztest_random(width - 1);
3276 packoff = n * sizeof (bufwad_t);
3277 packsize = s * sizeof (bufwad_t);
3279 bigoff = n * chunksize;
3280 bigsize = s * chunksize;
3282 packbuf = umem_alloc(packsize, UMEM_NOFAIL);
3283 bigbuf = umem_alloc(bigsize, UMEM_NOFAIL);
3286 * free_percent of the time, free a range of bigobj rather than
3289 freeit = (ztest_random(100) < free_percent);
3292 * Read the current contents of our objects.
3294 error = dmu_read(os, packobj, packoff, packsize, packbuf,
3296 ASSERT3U(error, ==, 0);
3297 error = dmu_read(os, bigobj, bigoff, bigsize, bigbuf,
3299 ASSERT3U(error, ==, 0);
3302 * Get a tx for the mods to both packobj and bigobj.
3304 tx = dmu_tx_create(os);
3306 dmu_tx_hold_write(tx, packobj, packoff, packsize);
3309 dmu_tx_hold_free(tx, bigobj, bigoff, bigsize);
3311 dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
3313 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3315 umem_free(packbuf, packsize);
3316 umem_free(bigbuf, bigsize);
3320 dmu_object_set_checksum(os, bigobj,
3321 (enum zio_checksum)ztest_random_dsl_prop(ZFS_PROP_CHECKSUM), tx);
3323 dmu_object_set_compress(os, bigobj,
3324 (enum zio_compress)ztest_random_dsl_prop(ZFS_PROP_COMPRESSION), tx);
3327 * For each index from n to n + s, verify that the existing bufwad
3328 * in packobj matches the bufwads at the head and tail of the
3329 * corresponding chunk in bigobj. Then update all three bufwads
3330 * with the new values we want to write out.
3332 for (i = 0; i < s; i++) {
3334 pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
3336 bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
3338 bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
3340 ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
3341 ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
3343 if (pack->bw_txg > txg)
3344 fatal(0, "future leak: got %llx, open txg is %llx",
3347 if (pack->bw_data != 0 && pack->bw_index != n + i)
3348 fatal(0, "wrong index: got %llx, wanted %llx+%llx",
3349 pack->bw_index, n, i);
3351 if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
3352 fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
3354 if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
3355 fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
3358 bzero(pack, sizeof (bufwad_t));
3360 pack->bw_index = n + i;
3362 pack->bw_data = 1 + ztest_random(-2ULL);
3369 * We've verified all the old bufwads, and made new ones.
3370 * Now write them out.
3372 dmu_write(os, packobj, packoff, packsize, packbuf, tx);
3375 if (zopt_verbose >= 7) {
3376 (void) printf("freeing offset %llx size %llx"
3378 (u_longlong_t)bigoff,
3379 (u_longlong_t)bigsize,
3382 VERIFY(0 == dmu_free_range(os, bigobj, bigoff, bigsize, tx));
3384 if (zopt_verbose >= 7) {
3385 (void) printf("writing offset %llx size %llx"
3387 (u_longlong_t)bigoff,
3388 (u_longlong_t)bigsize,
3391 dmu_write(os, bigobj, bigoff, bigsize, bigbuf, tx);
3397 * Sanity check the stuff we just wrote.
3400 void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
3401 void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
3403 VERIFY(0 == dmu_read(os, packobj, packoff,
3404 packsize, packcheck, DMU_READ_PREFETCH));
3405 VERIFY(0 == dmu_read(os, bigobj, bigoff,
3406 bigsize, bigcheck, DMU_READ_PREFETCH));
3408 ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
3409 ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
3411 umem_free(packcheck, packsize);
3412 umem_free(bigcheck, bigsize);
3415 umem_free(packbuf, packsize);
3416 umem_free(bigbuf, bigsize);
3420 compare_and_update_pbbufs(uint64_t s, bufwad_t *packbuf, bufwad_t *bigbuf,
3421 uint64_t bigsize, uint64_t n, uint64_t chunksize, uint64_t txg)
3429 * For each index from n to n + s, verify that the existing bufwad
3430 * in packobj matches the bufwads at the head and tail of the
3431 * corresponding chunk in bigobj. Then update all three bufwads
3432 * with the new values we want to write out.
3434 for (i = 0; i < s; i++) {
3436 pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
3438 bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
3440 bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
3442 ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
3443 ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
3445 if (pack->bw_txg > txg)
3446 fatal(0, "future leak: got %llx, open txg is %llx",
3449 if (pack->bw_data != 0 && pack->bw_index != n + i)
3450 fatal(0, "wrong index: got %llx, wanted %llx+%llx",
3451 pack->bw_index, n, i);
3453 if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
3454 fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
3456 if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
3457 fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
3459 pack->bw_index = n + i;
3461 pack->bw_data = 1 + ztest_random(-2ULL);
3469 ztest_dmu_read_write_zcopy(ztest_ds_t *zd, uint64_t id)
3471 objset_t *os = zd->zd_os;
3477 bufwad_t *packbuf, *bigbuf;
3478 uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
3479 uint64_t blocksize = ztest_random_blocksize();
3480 uint64_t chunksize = blocksize;
3481 uint64_t regions = 997;
3482 uint64_t stride = 123456789ULL;
3484 dmu_buf_t *bonus_db;
3485 arc_buf_t **bigbuf_arcbufs;
3486 dmu_object_info_t doi;
3489 * This test uses two objects, packobj and bigobj, that are always
3490 * updated together (i.e. in the same tx) so that their contents are
3491 * in sync and can be compared. Their contents relate to each other
3492 * in a simple way: packobj is a dense array of 'bufwad' structures,
3493 * while bigobj is a sparse array of the same bufwads. Specifically,
3494 * for any index n, there are three bufwads that should be identical:
3496 * packobj, at offset n * sizeof (bufwad_t)
3497 * bigobj, at the head of the nth chunk
3498 * bigobj, at the tail of the nth chunk
3500 * The chunk size is set equal to bigobj block size so that
3501 * dmu_assign_arcbuf() can be tested for object updates.
3505 * Read the directory info. If it's the first time, set things up.
3507 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
3508 ztest_od_init(&od[1], id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize);
3510 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
3513 bigobj = od[0].od_object;
3514 packobj = od[1].od_object;
3515 blocksize = od[0].od_blocksize;
3516 chunksize = blocksize;
3517 ASSERT(chunksize == od[1].od_gen);
3519 VERIFY(dmu_object_info(os, bigobj, &doi) == 0);
3520 VERIFY(ISP2(doi.doi_data_block_size));
3521 VERIFY(chunksize == doi.doi_data_block_size);
3522 VERIFY(chunksize >= 2 * sizeof (bufwad_t));
3525 * Pick a random index and compute the offsets into packobj and bigobj.
3527 n = ztest_random(regions) * stride + ztest_random(width);
3528 s = 1 + ztest_random(width - 1);
3530 packoff = n * sizeof (bufwad_t);
3531 packsize = s * sizeof (bufwad_t);
3533 bigoff = n * chunksize;
3534 bigsize = s * chunksize;
3536 packbuf = umem_zalloc(packsize, UMEM_NOFAIL);
3537 bigbuf = umem_zalloc(bigsize, UMEM_NOFAIL);
3539 VERIFY3U(0, ==, dmu_bonus_hold(os, bigobj, FTAG, &bonus_db));
3541 bigbuf_arcbufs = umem_zalloc(2 * s * sizeof (arc_buf_t *), UMEM_NOFAIL);
3544 * Iteration 0 test zcopy for DB_UNCACHED dbufs.
3545 * Iteration 1 test zcopy to already referenced dbufs.
3546 * Iteration 2 test zcopy to dirty dbuf in the same txg.
3547 * Iteration 3 test zcopy to dbuf dirty in previous txg.
3548 * Iteration 4 test zcopy when dbuf is no longer dirty.
3549 * Iteration 5 test zcopy when it can't be done.
3550 * Iteration 6 one more zcopy write.
3552 for (i = 0; i < 7; i++) {
3557 * In iteration 5 (i == 5) use arcbufs
3558 * that don't match bigobj blksz to test
3559 * dmu_assign_arcbuf() when it can't directly
3560 * assign an arcbuf to a dbuf.
3562 for (j = 0; j < s; j++) {
3565 dmu_request_arcbuf(bonus_db, chunksize);
3567 bigbuf_arcbufs[2 * j] =
3568 dmu_request_arcbuf(bonus_db, chunksize / 2);
3569 bigbuf_arcbufs[2 * j + 1] =
3570 dmu_request_arcbuf(bonus_db, chunksize / 2);
3575 * Get a tx for the mods to both packobj and bigobj.
3577 tx = dmu_tx_create(os);
3579 dmu_tx_hold_write(tx, packobj, packoff, packsize);
3580 dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
3582 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3584 umem_free(packbuf, packsize);
3585 umem_free(bigbuf, bigsize);
3586 for (j = 0; j < s; j++) {
3588 dmu_return_arcbuf(bigbuf_arcbufs[j]);
3591 bigbuf_arcbufs[2 * j]);
3593 bigbuf_arcbufs[2 * j + 1]);
3596 umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
3597 dmu_buf_rele(bonus_db, FTAG);
3602 * 50% of the time don't read objects in the 1st iteration to
3603 * test dmu_assign_arcbuf() for the case when there're no
3604 * existing dbufs for the specified offsets.
3606 if (i != 0 || ztest_random(2) != 0) {
3607 error = dmu_read(os, packobj, packoff,
3608 packsize, packbuf, DMU_READ_PREFETCH);
3609 ASSERT3U(error, ==, 0);
3610 error = dmu_read(os, bigobj, bigoff, bigsize,
3611 bigbuf, DMU_READ_PREFETCH);
3612 ASSERT3U(error, ==, 0);
3614 compare_and_update_pbbufs(s, packbuf, bigbuf, bigsize,
3618 * We've verified all the old bufwads, and made new ones.
3619 * Now write them out.
3621 dmu_write(os, packobj, packoff, packsize, packbuf, tx);
3622 if (zopt_verbose >= 7) {
3623 (void) printf("writing offset %llx size %llx"
3625 (u_longlong_t)bigoff,
3626 (u_longlong_t)bigsize,
3629 for (off = bigoff, j = 0; j < s; j++, off += chunksize) {
3632 bcopy((caddr_t)bigbuf + (off - bigoff),
3633 bigbuf_arcbufs[j]->b_data, chunksize);
3635 bcopy((caddr_t)bigbuf + (off - bigoff),
3636 bigbuf_arcbufs[2 * j]->b_data,
3638 bcopy((caddr_t)bigbuf + (off - bigoff) +
3640 bigbuf_arcbufs[2 * j + 1]->b_data,
3645 VERIFY(dmu_buf_hold(os, bigobj, off,
3646 FTAG, &dbt, DMU_READ_NO_PREFETCH) == 0);
3649 dmu_assign_arcbuf(bonus_db, off,
3650 bigbuf_arcbufs[j], tx);
3652 dmu_assign_arcbuf(bonus_db, off,
3653 bigbuf_arcbufs[2 * j], tx);
3654 dmu_assign_arcbuf(bonus_db,
3655 off + chunksize / 2,
3656 bigbuf_arcbufs[2 * j + 1], tx);
3659 dmu_buf_rele(dbt, FTAG);
3665 * Sanity check the stuff we just wrote.
3668 void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
3669 void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
3671 VERIFY(0 == dmu_read(os, packobj, packoff,
3672 packsize, packcheck, DMU_READ_PREFETCH));
3673 VERIFY(0 == dmu_read(os, bigobj, bigoff,
3674 bigsize, bigcheck, DMU_READ_PREFETCH));
3676 ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
3677 ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
3679 umem_free(packcheck, packsize);
3680 umem_free(bigcheck, bigsize);
3683 txg_wait_open(dmu_objset_pool(os), 0);
3684 } else if (i == 3) {
3685 txg_wait_synced(dmu_objset_pool(os), 0);
3689 dmu_buf_rele(bonus_db, FTAG);
3690 umem_free(packbuf, packsize);
3691 umem_free(bigbuf, bigsize);
3692 umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
3697 ztest_dmu_write_parallel(ztest_ds_t *zd, uint64_t id)
3700 uint64_t offset = (1ULL << (ztest_random(20) + 43)) +
3701 (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
3704 * Have multiple threads write to large offsets in an object
3705 * to verify that parallel writes to an object -- even to the
3706 * same blocks within the object -- doesn't cause any trouble.
3708 ztest_od_init(&od[0], ID_PARALLEL, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
3710 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
3713 while (ztest_random(10) != 0)
3714 ztest_io(zd, od[0].od_object, offset);
3718 ztest_dmu_prealloc(ztest_ds_t *zd, uint64_t id)
3721 uint64_t offset = (1ULL << (ztest_random(4) + SPA_MAXBLOCKSHIFT)) +
3722 (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
3723 uint64_t count = ztest_random(20) + 1;
3724 uint64_t blocksize = ztest_random_blocksize();
3727 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
3729 if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
3732 if (ztest_truncate(zd, od[0].od_object, offset, count * blocksize) != 0)
3735 ztest_prealloc(zd, od[0].od_object, offset, count * blocksize);
3737 data = umem_zalloc(blocksize, UMEM_NOFAIL);
3739 while (ztest_random(count) != 0) {
3740 uint64_t randoff = offset + (ztest_random(count) * blocksize);
3741 if (ztest_write(zd, od[0].od_object, randoff, blocksize,
3744 while (ztest_random(4) != 0)
3745 ztest_io(zd, od[0].od_object, randoff);
3748 umem_free(data, blocksize);
3752 * Verify that zap_{create,destroy,add,remove,update} work as expected.
3754 #define ZTEST_ZAP_MIN_INTS 1
3755 #define ZTEST_ZAP_MAX_INTS 4
3756 #define ZTEST_ZAP_MAX_PROPS 1000
3759 ztest_zap(ztest_ds_t *zd, uint64_t id)
3761 objset_t *os = zd->zd_os;
3764 uint64_t txg, last_txg;
3765 uint64_t value[ZTEST_ZAP_MAX_INTS];
3766 uint64_t zl_ints, zl_intsize, prop;
3769 char propname[100], txgname[100];
3771 char *hc[2] = { "s.acl.h", ".s.open.h.hyLZlg" };
3773 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0);
3775 if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
3778 object = od[0].od_object;
3781 * Generate a known hash collision, and verify that
3782 * we can lookup and remove both entries.
3784 tx = dmu_tx_create(os);
3785 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
3786 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3789 for (i = 0; i < 2; i++) {
3791 VERIFY3U(0, ==, zap_add(os, object, hc[i], sizeof (uint64_t),
3794 for (i = 0; i < 2; i++) {
3795 VERIFY3U(EEXIST, ==, zap_add(os, object, hc[i],
3796 sizeof (uint64_t), 1, &value[i], tx));
3798 zap_length(os, object, hc[i], &zl_intsize, &zl_ints));
3799 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
3800 ASSERT3U(zl_ints, ==, 1);
3802 for (i = 0; i < 2; i++) {
3803 VERIFY3U(0, ==, zap_remove(os, object, hc[i], tx));
3808 * Generate a buch of random entries.
3810 ints = MAX(ZTEST_ZAP_MIN_INTS, object % ZTEST_ZAP_MAX_INTS);
3812 prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
3813 (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
3814 (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
3815 bzero(value, sizeof (value));
3819 * If these zap entries already exist, validate their contents.
3821 error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
3823 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
3824 ASSERT3U(zl_ints, ==, 1);
3826 VERIFY(zap_lookup(os, object, txgname, zl_intsize,
3827 zl_ints, &last_txg) == 0);
3829 VERIFY(zap_length(os, object, propname, &zl_intsize,
3832 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
3833 ASSERT3U(zl_ints, ==, ints);
3835 VERIFY(zap_lookup(os, object, propname, zl_intsize,
3836 zl_ints, value) == 0);
3838 for (i = 0; i < ints; i++) {
3839 ASSERT3U(value[i], ==, last_txg + object + i);
3842 ASSERT3U(error, ==, ENOENT);
3846 * Atomically update two entries in our zap object.
3847 * The first is named txg_%llu, and contains the txg
3848 * in which the property was last updated. The second
3849 * is named prop_%llu, and the nth element of its value
3850 * should be txg + object + n.
3852 tx = dmu_tx_create(os);
3853 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
3854 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3859 fatal(0, "zap future leak: old %llu new %llu", last_txg, txg);
3861 for (i = 0; i < ints; i++)
3862 value[i] = txg + object + i;
3864 VERIFY3U(0, ==, zap_update(os, object, txgname, sizeof (uint64_t),
3866 VERIFY3U(0, ==, zap_update(os, object, propname, sizeof (uint64_t),
3872 * Remove a random pair of entries.
3874 prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
3875 (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
3876 (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
3878 error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
3880 if (error == ENOENT)
3883 ASSERT3U(error, ==, 0);
3885 tx = dmu_tx_create(os);
3886 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
3887 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3890 VERIFY3U(0, ==, zap_remove(os, object, txgname, tx));
3891 VERIFY3U(0, ==, zap_remove(os, object, propname, tx));
3896 * Testcase to test the upgrading of a microzap to fatzap.
3899 ztest_fzap(ztest_ds_t *zd, uint64_t id)
3901 objset_t *os = zd->zd_os;
3903 uint64_t object, txg;
3906 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0);
3908 if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
3911 object = od[0].od_object;
3914 * Add entries to this ZAP and make sure it spills over
3915 * and gets upgraded to a fatzap. Also, since we are adding
3916 * 2050 entries we should see ptrtbl growth and leaf-block split.
3918 for (i = 0; i < 2050; i++) {
3919 char name[MAXNAMELEN];
3924 (void) snprintf(name, sizeof (name), "fzap-%llu-%llu",
3925 (u_longlong_t)id, (u_longlong_t)value);
3927 tx = dmu_tx_create(os);
3928 dmu_tx_hold_zap(tx, object, B_TRUE, name);
3929 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3932 error = zap_add(os, object, name, sizeof (uint64_t), 1,
3934 ASSERT(error == 0 || error == EEXIST);
3941 ztest_zap_parallel(ztest_ds_t *zd, uint64_t id)
3943 objset_t *os = zd->zd_os;
3945 uint64_t txg, object, count, wsize, wc, zl_wsize, zl_wc;
3947 int i, namelen, error;
3948 int micro = ztest_random(2);
3949 char name[20], string_value[20];
3952 ztest_od_init(&od[0], ID_PARALLEL, FTAG, micro, DMU_OT_ZAP_OTHER, 0, 0);
3954 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
3957 object = od[0].od_object;
3960 * Generate a random name of the form 'xxx.....' where each
3961 * x is a random printable character and the dots are dots.
3962 * There are 94 such characters, and the name length goes from
3963 * 6 to 20, so there are 94^3 * 15 = 12,458,760 possible names.
3965 namelen = ztest_random(sizeof (name) - 5) + 5 + 1;
3967 for (i = 0; i < 3; i++)
3968 name[i] = '!' + ztest_random('~' - '!' + 1);
3969 for (; i < namelen - 1; i++)
3973 if ((namelen & 1) || micro) {
3974 wsize = sizeof (txg);
3980 data = string_value;
3984 VERIFY(zap_count(os, object, &count) == 0);
3985 ASSERT(count != -1ULL);
3988 * Select an operation: length, lookup, add, update, remove.
3990 i = ztest_random(5);
3993 tx = dmu_tx_create(os);
3994 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
3995 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3998 bcopy(name, string_value, namelen);
4002 bzero(string_value, namelen);
4008 error = zap_length(os, object, name, &zl_wsize, &zl_wc);
4010 ASSERT3U(wsize, ==, zl_wsize);
4011 ASSERT3U(wc, ==, zl_wc);
4013 ASSERT3U(error, ==, ENOENT);
4018 error = zap_lookup(os, object, name, wsize, wc, data);
4020 if (data == string_value &&
4021 bcmp(name, data, namelen) != 0)
4022 fatal(0, "name '%s' != val '%s' len %d",
4023 name, data, namelen);
4025 ASSERT3U(error, ==, ENOENT);
4030 error = zap_add(os, object, name, wsize, wc, data, tx);
4031 ASSERT(error == 0 || error == EEXIST);
4035 VERIFY(zap_update(os, object, name, wsize, wc, data, tx) == 0);
4039 error = zap_remove(os, object, name, tx);
4040 ASSERT(error == 0 || error == ENOENT);
4049 * Commit callback data.
4051 typedef struct ztest_cb_data {
4052 list_node_t zcd_node;
4054 int zcd_expected_err;
4055 boolean_t zcd_added;
4056 boolean_t zcd_called;
4060 /* This is the actual commit callback function */
4062 ztest_commit_callback(void *arg, int error)
4064 ztest_cb_data_t *data = arg;
4065 uint64_t synced_txg;
4067 VERIFY(data != NULL);
4068 VERIFY3S(data->zcd_expected_err, ==, error);
4069 VERIFY(!data->zcd_called);
4071 synced_txg = spa_last_synced_txg(data->zcd_spa);
4072 if (data->zcd_txg > synced_txg)
4073 fatal(0, "commit callback of txg %" PRIu64 " called prematurely"
4074 ", last synced txg = %" PRIu64 "\n", data->zcd_txg,
4077 data->zcd_called = B_TRUE;
4079 if (error == ECANCELED) {
4080 ASSERT3U(data->zcd_txg, ==, 0);
4081 ASSERT(!data->zcd_added);
4084 * The private callback data should be destroyed here, but
4085 * since we are going to check the zcd_called field after
4086 * dmu_tx_abort(), we will destroy it there.
4091 /* Was this callback added to the global callback list? */
4092 if (!data->zcd_added)
4095 ASSERT3U(data->zcd_txg, !=, 0);
4097 /* Remove our callback from the list */
4098 (void) mutex_lock(&zcl.zcl_callbacks_lock);
4099 list_remove(&zcl.zcl_callbacks, data);
4100 (void) mutex_unlock(&zcl.zcl_callbacks_lock);
4103 umem_free(data, sizeof (ztest_cb_data_t));
4106 /* Allocate and initialize callback data structure */
4107 static ztest_cb_data_t *
4108 ztest_create_cb_data(objset_t *os, uint64_t txg)
4110 ztest_cb_data_t *cb_data;
4112 cb_data = umem_zalloc(sizeof (ztest_cb_data_t), UMEM_NOFAIL);
4114 cb_data->zcd_txg = txg;
4115 cb_data->zcd_spa = dmu_objset_spa(os);
4121 * If a number of txgs equal to this threshold have been created after a commit
4122 * callback has been registered but not called, then we assume there is an
4123 * implementation bug.
4125 #define ZTEST_COMMIT_CALLBACK_THRESH (TXG_CONCURRENT_STATES + 2)
4128 * Commit callback test.
4131 ztest_dmu_commit_callbacks(ztest_ds_t *zd, uint64_t id)
4133 objset_t *os = zd->zd_os;
4136 ztest_cb_data_t *cb_data[3], *tmp_cb;
4137 uint64_t old_txg, txg;
4140 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
4142 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4145 tx = dmu_tx_create(os);
4147 cb_data[0] = ztest_create_cb_data(os, 0);
4148 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[0]);
4150 dmu_tx_hold_write(tx, od[0].od_object, 0, sizeof (uint64_t));
4152 /* Every once in a while, abort the transaction on purpose */
4153 if (ztest_random(100) == 0)
4157 error = dmu_tx_assign(tx, TXG_NOWAIT);
4159 txg = error ? 0 : dmu_tx_get_txg(tx);
4161 cb_data[0]->zcd_txg = txg;
4162 cb_data[1] = ztest_create_cb_data(os, txg);
4163 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[1]);
4167 * It's not a strict requirement to call the registered
4168 * callbacks from inside dmu_tx_abort(), but that's what
4169 * it's supposed to happen in the current implementation
4170 * so we will check for that.
4172 for (i = 0; i < 2; i++) {
4173 cb_data[i]->zcd_expected_err = ECANCELED;
4174 VERIFY(!cb_data[i]->zcd_called);
4179 for (i = 0; i < 2; i++) {
4180 VERIFY(cb_data[i]->zcd_called);
4181 umem_free(cb_data[i], sizeof (ztest_cb_data_t));
4187 cb_data[2] = ztest_create_cb_data(os, txg);
4188 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[2]);
4191 * Read existing data to make sure there isn't a future leak.
4193 VERIFY(0 == dmu_read(os, od[0].od_object, 0, sizeof (uint64_t),
4194 &old_txg, DMU_READ_PREFETCH));
4197 fatal(0, "future leak: got %" PRIu64 ", open txg is %" PRIu64,
4200 dmu_write(os, od[0].od_object, 0, sizeof (uint64_t), &txg, tx);
4202 (void) mutex_lock(&zcl.zcl_callbacks_lock);
4205 * Since commit callbacks don't have any ordering requirement and since
4206 * it is theoretically possible for a commit callback to be called
4207 * after an arbitrary amount of time has elapsed since its txg has been
4208 * synced, it is difficult to reliably determine whether a commit
4209 * callback hasn't been called due to high load or due to a flawed
4212 * In practice, we will assume that if after a certain number of txgs a
4213 * commit callback hasn't been called, then most likely there's an
4214 * implementation bug..
4216 tmp_cb = list_head(&zcl.zcl_callbacks);
4217 if (tmp_cb != NULL &&
4218 tmp_cb->zcd_txg > txg - ZTEST_COMMIT_CALLBACK_THRESH) {
4219 fatal(0, "Commit callback threshold exceeded, oldest txg: %"
4220 PRIu64 ", open txg: %" PRIu64 "\n", tmp_cb->zcd_txg, txg);
4224 * Let's find the place to insert our callbacks.
4226 * Even though the list is ordered by txg, it is possible for the
4227 * insertion point to not be the end because our txg may already be
4228 * quiescing at this point and other callbacks in the open txg
4229 * (from other objsets) may have sneaked in.
4231 tmp_cb = list_tail(&zcl.zcl_callbacks);
4232 while (tmp_cb != NULL && tmp_cb->zcd_txg > txg)
4233 tmp_cb = list_prev(&zcl.zcl_callbacks, tmp_cb);
4235 /* Add the 3 callbacks to the list */
4236 for (i = 0; i < 3; i++) {
4238 list_insert_head(&zcl.zcl_callbacks, cb_data[i]);
4240 list_insert_after(&zcl.zcl_callbacks, tmp_cb,
4243 cb_data[i]->zcd_added = B_TRUE;
4244 VERIFY(!cb_data[i]->zcd_called);
4246 tmp_cb = cb_data[i];
4249 (void) mutex_unlock(&zcl.zcl_callbacks_lock);
4256 ztest_dsl_prop_get_set(ztest_ds_t *zd, uint64_t id)
4258 zfs_prop_t proplist[] = {
4260 ZFS_PROP_COMPRESSION,
4264 ztest_shared_t *zs = ztest_shared;
4267 (void) rw_rdlock(&zs->zs_name_lock);
4269 for (p = 0; p < sizeof (proplist) / sizeof (proplist[0]); p++)
4270 (void) ztest_dsl_prop_set_uint64(zd->zd_name, proplist[p],
4271 ztest_random_dsl_prop(proplist[p]), (int)ztest_random(2));
4273 (void) rw_unlock(&zs->zs_name_lock);
4278 ztest_spa_prop_get_set(ztest_ds_t *zd, uint64_t id)
4280 ztest_shared_t *zs = ztest_shared;
4281 nvlist_t *props = NULL;
4283 (void) rw_rdlock(&zs->zs_name_lock);
4285 (void) ztest_spa_prop_set_uint64(zs, ZPOOL_PROP_DEDUPDITTO,
4286 ZIO_DEDUPDITTO_MIN + ztest_random(ZIO_DEDUPDITTO_MIN));
4288 VERIFY3U(spa_prop_get(zs->zs_spa, &props), ==, 0);
4290 if (zopt_verbose >= 6)
4291 dump_nvlist(props, 4);
4295 (void) rw_unlock(&zs->zs_name_lock);
4299 * Test snapshot hold/release and deferred destroy.
4302 ztest_dmu_snapshot_hold(ztest_ds_t *zd, uint64_t id)
4305 objset_t *os = zd->zd_os;
4309 char clonename[100];
4311 char osname[MAXNAMELEN];
4313 (void) rw_rdlock(&ztest_shared->zs_name_lock);
4315 dmu_objset_name(os, osname);
4317 (void) snprintf(snapname, 100, "sh1_%llu", id);
4318 (void) snprintf(fullname, 100, "%s@%s", osname, snapname);
4319 (void) snprintf(clonename, 100, "%s/ch1_%llu", osname, id);
4320 (void) snprintf(tag, 100, "%tag_%llu", id);
4323 * Clean up from any previous run.
4325 (void) dmu_objset_destroy(clonename, B_FALSE);
4326 (void) dsl_dataset_user_release(osname, snapname, tag, B_FALSE);
4327 (void) dmu_objset_destroy(fullname, B_FALSE);
4330 * Create snapshot, clone it, mark snap for deferred destroy,
4331 * destroy clone, verify snap was also destroyed.
4333 error = dmu_objset_snapshot(osname, snapname, NULL, NULL, FALSE,
4336 if (error == ENOSPC) {
4337 ztest_record_enospc("dmu_objset_snapshot");
4340 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
4343 error = dmu_objset_hold(fullname, FTAG, &origin);
4345 fatal(0, "dmu_objset_hold(%s) = %d", fullname, error);
4347 error = dmu_objset_clone(clonename, dmu_objset_ds(origin), 0);
4348 dmu_objset_rele(origin, FTAG);
4350 if (error == ENOSPC) {
4351 ztest_record_enospc("dmu_objset_clone");
4354 fatal(0, "dmu_objset_clone(%s) = %d", clonename, error);
4357 error = dmu_objset_destroy(fullname, B_TRUE);
4359 fatal(0, "dmu_objset_destroy(%s, B_TRUE) = %d",
4363 error = dmu_objset_destroy(clonename, B_FALSE);
4365 fatal(0, "dmu_objset_destroy(%s) = %d", clonename, error);
4367 error = dmu_objset_hold(fullname, FTAG, &origin);
4368 if (error != ENOENT)
4369 fatal(0, "dmu_objset_hold(%s) = %d", fullname, error);
4372 * Create snapshot, add temporary hold, verify that we can't
4373 * destroy a held snapshot, mark for deferred destroy,
4374 * release hold, verify snapshot was destroyed.
4376 error = dmu_objset_snapshot(osname, snapname, NULL, NULL, FALSE,
4379 if (error == ENOSPC) {
4380 ztest_record_enospc("dmu_objset_snapshot");
4383 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
4386 error = dsl_dataset_user_hold(osname, snapname, tag, B_FALSE,
4389 fatal(0, "dsl_dataset_user_hold(%s)", fullname, tag);
4391 error = dmu_objset_destroy(fullname, B_FALSE);
4392 if (error != EBUSY) {
4393 fatal(0, "dmu_objset_destroy(%s, B_FALSE) = %d",
4397 error = dmu_objset_destroy(fullname, B_TRUE);
4399 fatal(0, "dmu_objset_destroy(%s, B_TRUE) = %d",
4403 error = dsl_dataset_user_release(osname, snapname, tag, B_FALSE);
4405 fatal(0, "dsl_dataset_user_release(%s)", fullname, tag);
4407 VERIFY(dmu_objset_hold(fullname, FTAG, &origin) == ENOENT);
4410 (void) rw_unlock(&ztest_shared->zs_name_lock);
4414 * Inject random faults into the on-disk data.
4418 ztest_fault_inject(ztest_ds_t *zd, uint64_t id)
4420 ztest_shared_t *zs = ztest_shared;
4421 spa_t *spa = zs->zs_spa;
4425 uint64_t bad = 0x1990c0ffeedecadeull;
4427 char path0[MAXPATHLEN];
4428 char pathrand[MAXPATHLEN];
4430 int bshift = SPA_MAXBLOCKSHIFT + 2; /* don't scrog all labels */
4436 boolean_t islog = B_FALSE;
4438 VERIFY(mutex_lock(&zs->zs_vdev_lock) == 0);
4439 maxfaults = MAXFAULTS();
4440 leaves = MAX(zs->zs_mirrors, 1) * zopt_raidz;
4441 mirror_save = zs->zs_mirrors;
4442 VERIFY(mutex_unlock(&zs->zs_vdev_lock) == 0);
4444 ASSERT(leaves >= 1);
4447 * We need SCL_STATE here because we're going to look at vd0->vdev_tsd.
4449 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
4451 if (ztest_random(2) == 0) {
4453 * Inject errors on a normal data device or slog device.
4455 top = ztest_random_vdev_top(spa, B_TRUE);
4456 leaf = ztest_random(leaves) + zs->zs_splits;
4459 * Generate paths to the first leaf in this top-level vdev,
4460 * and to the random leaf we selected. We'll induce transient
4461 * write failures and random online/offline activity on leaf 0,
4462 * and we'll write random garbage to the randomly chosen leaf.
4464 (void) snprintf(path0, sizeof (path0), ztest_dev_template,
4465 zopt_dir, zopt_pool, top * leaves + zs->zs_splits);
4466 (void) snprintf(pathrand, sizeof (pathrand), ztest_dev_template,
4467 zopt_dir, zopt_pool, top * leaves + leaf);
4469 vd0 = vdev_lookup_by_path(spa->spa_root_vdev, path0);
4470 if (vd0 != NULL && vd0->vdev_top->vdev_islog)
4473 if (vd0 != NULL && maxfaults != 1) {
4475 * Make vd0 explicitly claim to be unreadable,
4476 * or unwriteable, or reach behind its back
4477 * and close the underlying fd. We can do this if
4478 * maxfaults == 0 because we'll fail and reexecute,
4479 * and we can do it if maxfaults >= 2 because we'll
4480 * have enough redundancy. If maxfaults == 1, the
4481 * combination of this with injection of random data
4482 * corruption below exceeds the pool's fault tolerance.
4484 vdev_file_t *vf = vd0->vdev_tsd;
4486 if (vf != NULL && ztest_random(3) == 0) {
4487 (void) close(vf->vf_vnode->v_fd);
4488 vf->vf_vnode->v_fd = -1;
4489 } else if (ztest_random(2) == 0) {
4490 vd0->vdev_cant_read = B_TRUE;
4492 vd0->vdev_cant_write = B_TRUE;
4494 guid0 = vd0->vdev_guid;
4498 * Inject errors on an l2cache device.
4500 spa_aux_vdev_t *sav = &spa->spa_l2cache;
4502 if (sav->sav_count == 0) {
4503 spa_config_exit(spa, SCL_STATE, FTAG);
4506 vd0 = sav->sav_vdevs[ztest_random(sav->sav_count)];
4507 guid0 = vd0->vdev_guid;
4508 (void) strcpy(path0, vd0->vdev_path);
4509 (void) strcpy(pathrand, vd0->vdev_path);
4513 maxfaults = INT_MAX; /* no limit on cache devices */
4516 spa_config_exit(spa, SCL_STATE, FTAG);
4519 * If we can tolerate two or more faults, or we're dealing
4520 * with a slog, randomly online/offline vd0.
4522 if ((maxfaults >= 2 || islog) && guid0 != 0) {
4523 if (ztest_random(10) < 6) {
4524 int flags = (ztest_random(2) == 0 ?
4525 ZFS_OFFLINE_TEMPORARY : 0);
4528 * We have to grab the zs_name_lock as writer to
4529 * prevent a race between offlining a slog and
4530 * destroying a dataset. Offlining the slog will
4531 * grab a reference on the dataset which may cause
4532 * dmu_objset_destroy() to fail with EBUSY thus
4533 * leaving the dataset in an inconsistent state.
4536 (void) rw_wrlock(&ztest_shared->zs_name_lock);
4538 VERIFY(vdev_offline(spa, guid0, flags) != EBUSY);
4541 (void) rw_unlock(&ztest_shared->zs_name_lock);
4543 (void) vdev_online(spa, guid0, 0, NULL);
4551 * We have at least single-fault tolerance, so inject data corruption.
4553 fd = open(pathrand, O_RDWR);
4555 if (fd == -1) /* we hit a gap in the device namespace */
4558 fsize = lseek(fd, 0, SEEK_END);
4560 while (--iters != 0) {
4561 offset = ztest_random(fsize / (leaves << bshift)) *
4562 (leaves << bshift) + (leaf << bshift) +
4563 (ztest_random(1ULL << (bshift - 1)) & -8ULL);
4565 if (offset >= fsize)
4568 VERIFY(mutex_lock(&zs->zs_vdev_lock) == 0);
4569 if (mirror_save != zs->zs_mirrors) {
4570 VERIFY(mutex_unlock(&zs->zs_vdev_lock) == 0);
4575 if (pwrite(fd, &bad, sizeof (bad), offset) != sizeof (bad))
4576 fatal(1, "can't inject bad word at 0x%llx in %s",
4579 VERIFY(mutex_unlock(&zs->zs_vdev_lock) == 0);
4581 if (zopt_verbose >= 7)
4582 (void) printf("injected bad word into %s,"
4583 " offset 0x%llx\n", pathrand, (u_longlong_t)offset);
4590 * Verify that DDT repair works as expected.
4593 ztest_ddt_repair(ztest_ds_t *zd, uint64_t id)
4595 ztest_shared_t *zs = ztest_shared;
4596 spa_t *spa = zs->zs_spa;
4597 objset_t *os = zd->zd_os;
4599 uint64_t object, blocksize, txg, pattern, psize;
4600 enum zio_checksum checksum = spa_dedup_checksum(spa);
4605 int copies = 2 * ZIO_DEDUPDITTO_MIN;
4608 blocksize = ztest_random_blocksize();
4609 blocksize = MIN(blocksize, 2048); /* because we write so many */
4611 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
4613 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4617 * Take the name lock as writer to prevent anyone else from changing
4618 * the pool and dataset properies we need to maintain during this test.
4620 (void) rw_wrlock(&zs->zs_name_lock);
4622 if (ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_DEDUP, checksum,
4624 ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_COPIES, 1,
4626 (void) rw_unlock(&zs->zs_name_lock);
4630 object = od[0].od_object;
4631 blocksize = od[0].od_blocksize;
4632 pattern = spa_guid(spa) ^ dmu_objset_fsid_guid(os);
4634 ASSERT(object != 0);
4636 tx = dmu_tx_create(os);
4637 dmu_tx_hold_write(tx, object, 0, copies * blocksize);
4638 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
4640 (void) rw_unlock(&zs->zs_name_lock);
4645 * Write all the copies of our block.
4647 for (i = 0; i < copies; i++) {
4648 uint64_t offset = i * blocksize;
4649 VERIFY(dmu_buf_hold(os, object, offset, FTAG, &db,
4650 DMU_READ_NO_PREFETCH) == 0);
4651 ASSERT(db->db_offset == offset);
4652 ASSERT(db->db_size == blocksize);
4653 ASSERT(ztest_pattern_match(db->db_data, db->db_size, pattern) ||
4654 ztest_pattern_match(db->db_data, db->db_size, 0ULL));
4655 dmu_buf_will_fill(db, tx);
4656 ztest_pattern_set(db->db_data, db->db_size, pattern);
4657 dmu_buf_rele(db, FTAG);
4661 txg_wait_synced(spa_get_dsl(spa), txg);
4664 * Find out what block we got.
4666 VERIFY(dmu_buf_hold(os, object, 0, FTAG, &db,
4667 DMU_READ_NO_PREFETCH) == 0);
4668 blk = *((dmu_buf_impl_t *)db)->db_blkptr;
4669 dmu_buf_rele(db, FTAG);
4672 * Damage the block. Dedup-ditto will save us when we read it later.
4674 psize = BP_GET_PSIZE(&blk);
4675 buf = zio_buf_alloc(psize);
4676 ztest_pattern_set(buf, psize, ~pattern);
4678 (void) zio_wait(zio_rewrite(NULL, spa, 0, &blk,
4679 buf, psize, NULL, NULL, ZIO_PRIORITY_SYNC_WRITE,
4680 ZIO_FLAG_CANFAIL | ZIO_FLAG_INDUCE_DAMAGE, NULL));
4682 zio_buf_free(buf, psize);
4684 (void) rw_unlock(&zs->zs_name_lock);
4692 ztest_scrub(ztest_ds_t *zd, uint64_t id)
4694 ztest_shared_t *zs = ztest_shared;
4695 spa_t *spa = zs->zs_spa;
4697 (void) spa_scan(spa, POOL_SCAN_SCRUB);
4698 (void) poll(NULL, 0, 100); /* wait a moment, then force a restart */
4699 (void) spa_scan(spa, POOL_SCAN_SCRUB);
4703 * Rename the pool to a different name and then rename it back.
4707 ztest_spa_rename(ztest_ds_t *zd, uint64_t id)
4709 ztest_shared_t *zs = ztest_shared;
4710 char *oldname, *newname;
4713 (void) rw_wrlock(&zs->zs_name_lock);
4715 oldname = zs->zs_pool;
4716 newname = umem_alloc(strlen(oldname) + 5, UMEM_NOFAIL);
4717 (void) strcpy(newname, oldname);
4718 (void) strcat(newname, "_tmp");
4723 VERIFY3U(0, ==, spa_rename(oldname, newname));
4726 * Try to open it under the old name, which shouldn't exist
4728 VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
4731 * Open it under the new name and make sure it's still the same spa_t.
4733 VERIFY3U(0, ==, spa_open(newname, &spa, FTAG));
4735 ASSERT(spa == zs->zs_spa);
4736 spa_close(spa, FTAG);
4739 * Rename it back to the original
4741 VERIFY3U(0, ==, spa_rename(newname, oldname));
4744 * Make sure it can still be opened
4746 VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG));
4748 ASSERT(spa == zs->zs_spa);
4749 spa_close(spa, FTAG);
4751 umem_free(newname, strlen(newname) + 1);
4753 (void) rw_unlock(&zs->zs_name_lock);
4757 * Verify pool integrity by running zdb.
4760 ztest_run_zdb(char *pool)
4763 char zdb[MAXPATHLEN + MAXNAMELEN + 20];
4771 (void) realpath(getexecname(), zdb);
4773 /* zdb lives in /usr/sbin, while ztest lives in /usr/bin */
4774 bin = strstr(zdb, "/usr/bin/");
4775 ztest = strstr(bin, "/ztest");
4777 isalen = ztest - isa;
4781 "/usr/sbin%.*s/zdb -bcc%s%s -U %s %s",
4784 zopt_verbose >= 3 ? "s" : "",
4785 zopt_verbose >= 4 ? "v" : "",
4790 if (zopt_verbose >= 5)
4791 (void) printf("Executing %s\n", strstr(zdb, "zdb "));
4793 fp = popen(zdb, "r");
4795 while (fgets(zbuf, sizeof (zbuf), fp) != NULL)
4796 if (zopt_verbose >= 3)
4797 (void) printf("%s", zbuf);
4799 status = pclose(fp);
4804 ztest_dump_core = 0;
4805 if (WIFEXITED(status))
4806 fatal(0, "'%s' exit code %d", zdb, WEXITSTATUS(status));
4808 fatal(0, "'%s' died with signal %d", zdb, WTERMSIG(status));
4812 ztest_walk_pool_directory(char *header)
4816 if (zopt_verbose >= 6)
4817 (void) printf("%s\n", header);
4819 mutex_enter(&spa_namespace_lock);
4820 while ((spa = spa_next(spa)) != NULL)
4821 if (zopt_verbose >= 6)
4822 (void) printf("\t%s\n", spa_name(spa));
4823 mutex_exit(&spa_namespace_lock);
4827 ztest_spa_import_export(char *oldname, char *newname)
4829 nvlist_t *config, *newconfig;
4833 if (zopt_verbose >= 4) {
4834 (void) printf("import/export: old = %s, new = %s\n",
4839 * Clean up from previous runs.
4841 (void) spa_destroy(newname);
4844 * Get the pool's configuration and guid.
4846 VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG));
4849 * Kick off a scrub to tickle scrub/export races.
4851 if (ztest_random(2) == 0)
4852 (void) spa_scan(spa, POOL_SCAN_SCRUB);
4854 pool_guid = spa_guid(spa);
4855 spa_close(spa, FTAG);
4857 ztest_walk_pool_directory("pools before export");
4862 VERIFY3U(0, ==, spa_export(oldname, &config, B_FALSE, B_FALSE));
4864 ztest_walk_pool_directory("pools after export");
4869 newconfig = spa_tryimport(config);
4870 ASSERT(newconfig != NULL);
4871 nvlist_free(newconfig);
4874 * Import it under the new name.
4876 VERIFY3U(0, ==, spa_import(newname, config, NULL, 0));
4878 ztest_walk_pool_directory("pools after import");
4881 * Try to import it again -- should fail with EEXIST.
4883 VERIFY3U(EEXIST, ==, spa_import(newname, config, NULL, 0));
4886 * Try to import it under a different name -- should fail with EEXIST.
4888 VERIFY3U(EEXIST, ==, spa_import(oldname, config, NULL, 0));
4891 * Verify that the pool is no longer visible under the old name.
4893 VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
4896 * Verify that we can open and close the pool using the new name.
4898 VERIFY3U(0, ==, spa_open(newname, &spa, FTAG));
4899 ASSERT(pool_guid == spa_guid(spa));
4900 spa_close(spa, FTAG);
4902 nvlist_free(config);
4906 ztest_resume(spa_t *spa)
4908 if (spa_suspended(spa) && zopt_verbose >= 6)
4909 (void) printf("resuming from suspended state\n");
4910 spa_vdev_state_enter(spa, SCL_NONE);
4911 vdev_clear(spa, NULL);
4912 (void) spa_vdev_state_exit(spa, NULL, 0);
4913 (void) zio_resume(spa);
4917 ztest_resume_thread(void *arg)
4921 while (!ztest_exiting) {
4922 if (spa_suspended(spa))
4924 (void) poll(NULL, 0, 100);
4930 ztest_deadman_thread(void *arg)
4932 ztest_shared_t *zs = arg;
4936 delta = (zs->zs_thread_stop - zs->zs_thread_start) / NANOSEC + grace;
4938 (void) poll(NULL, 0, (int)(1000 * delta));
4940 fatal(0, "failed to complete within %d seconds of deadline", grace);
4946 ztest_execute(ztest_info_t *zi, uint64_t id)
4948 ztest_shared_t *zs = ztest_shared;
4949 ztest_ds_t *zd = &zs->zs_zd[id % zopt_datasets];
4950 hrtime_t functime = gethrtime();
4953 for (i = 0; i < zi->zi_iters; i++)
4954 zi->zi_func(zd, id);
4956 functime = gethrtime() - functime;
4958 atomic_add_64(&zi->zi_call_count, 1);
4959 atomic_add_64(&zi->zi_call_time, functime);
4961 if (zopt_verbose >= 4) {
4963 (void) dladdr((void *)zi->zi_func, &dli);
4964 (void) printf("%6.2f sec in %s\n",
4965 (double)functime / NANOSEC, dli.dli_sname);
4970 ztest_thread(void *arg)
4972 uint64_t id = (uintptr_t)arg;
4973 ztest_shared_t *zs = ztest_shared;
4978 while ((now = gethrtime()) < zs->zs_thread_stop) {
4980 * See if it's time to force a crash.
4982 if (now > zs->zs_thread_kill)
4986 * If we're getting ENOSPC with some regularity, stop.
4988 if (zs->zs_enospc_count > 10)
4992 * Pick a random function to execute.
4994 zi = &zs->zs_info[ztest_random(ZTEST_FUNCS)];
4995 call_next = zi->zi_call_next;
4997 if (now >= call_next &&
4998 atomic_cas_64(&zi->zi_call_next, call_next, call_next +
4999 ztest_random(2 * zi->zi_interval[0] + 1)) == call_next)
5000 ztest_execute(zi, id);
5007 ztest_dataset_name(char *dsname, char *pool, int d)
5009 (void) snprintf(dsname, MAXNAMELEN, "%s/ds_%d", pool, d);
5013 ztest_dataset_destroy(ztest_shared_t *zs, int d)
5015 char name[MAXNAMELEN];
5018 ztest_dataset_name(name, zs->zs_pool, d);
5020 if (zopt_verbose >= 3)
5021 (void) printf("Destroying %s to free up space\n", name);
5024 * Cleanup any non-standard clones and snapshots. In general,
5025 * ztest thread t operates on dataset (t % zopt_datasets),
5026 * so there may be more than one thing to clean up.
5028 for (t = d; t < zopt_threads; t += zopt_datasets)
5029 ztest_dsl_dataset_cleanup(name, t);
5031 (void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
5032 DS_FIND_SNAPSHOTS | DS_FIND_CHILDREN);
5036 ztest_dataset_dirobj_verify(ztest_ds_t *zd)
5038 uint64_t usedobjs, dirobjs, scratch;
5041 * ZTEST_DIROBJ is the object directory for the entire dataset.
5042 * Therefore, the number of objects in use should equal the
5043 * number of ZTEST_DIROBJ entries, +1 for ZTEST_DIROBJ itself.
5044 * If not, we have an object leak.
5046 * Note that we can only check this in ztest_dataset_open(),
5047 * when the open-context and syncing-context values agree.
5048 * That's because zap_count() returns the open-context value,
5049 * while dmu_objset_space() returns the rootbp fill count.
5051 VERIFY3U(0, ==, zap_count(zd->zd_os, ZTEST_DIROBJ, &dirobjs));
5052 dmu_objset_space(zd->zd_os, &scratch, &scratch, &usedobjs, &scratch);
5053 ASSERT3U(dirobjs + 1, ==, usedobjs);
5057 ztest_dataset_open(ztest_shared_t *zs, int d)
5059 ztest_ds_t *zd = &zs->zs_zd[d];
5060 uint64_t committed_seq = zd->zd_seq;
5063 char name[MAXNAMELEN];
5066 ztest_dataset_name(name, zs->zs_pool, d);
5068 (void) rw_rdlock(&zs->zs_name_lock);
5070 error = ztest_dataset_create(name);
5071 if (error == ENOSPC) {
5072 (void) rw_unlock(&zs->zs_name_lock);
5073 ztest_record_enospc(FTAG);
5076 ASSERT(error == 0 || error == EEXIST);
5078 VERIFY3U(dmu_objset_hold(name, zd, &os), ==, 0);
5079 (void) rw_unlock(&zs->zs_name_lock);
5081 ztest_zd_init(zd, os);
5083 zilog = zd->zd_zilog;
5085 if (zilog->zl_header->zh_claim_lr_seq != 0 &&
5086 zilog->zl_header->zh_claim_lr_seq < committed_seq)
5087 fatal(0, "missing log records: claimed %llu < committed %llu",
5088 zilog->zl_header->zh_claim_lr_seq, committed_seq);
5090 ztest_dataset_dirobj_verify(zd);
5092 zil_replay(os, zd, ztest_replay_vector);
5094 ztest_dataset_dirobj_verify(zd);
5096 if (zopt_verbose >= 6)
5097 (void) printf("%s replay %llu blocks, %llu records, seq %llu\n",
5099 (u_longlong_t)zilog->zl_parse_blk_count,
5100 (u_longlong_t)zilog->zl_parse_lr_count,
5101 (u_longlong_t)zilog->zl_replaying_seq);
5103 zilog = zil_open(os, ztest_get_data);
5105 if (zilog->zl_replaying_seq != 0 &&
5106 zilog->zl_replaying_seq < committed_seq)
5107 fatal(0, "missing log records: replayed %llu < committed %llu",
5108 zilog->zl_replaying_seq, committed_seq);
5114 ztest_dataset_close(ztest_shared_t *zs, int d)
5116 ztest_ds_t *zd = &zs->zs_zd[d];
5118 zil_close(zd->zd_zilog);
5119 dmu_objset_rele(zd->zd_os, zd);
5125 * Kick off threads to run tests on all datasets in parallel.
5128 ztest_run(ztest_shared_t *zs)
5132 thread_t resume_tid;
5136 ztest_exiting = B_FALSE;
5139 * Initialize parent/child shared state.
5141 VERIFY(_mutex_init(&zs->zs_vdev_lock, USYNC_THREAD, NULL) == 0);
5142 VERIFY(rwlock_init(&zs->zs_name_lock, USYNC_THREAD, NULL) == 0);
5144 zs->zs_thread_start = gethrtime();
5145 zs->zs_thread_stop = zs->zs_thread_start + zopt_passtime * NANOSEC;
5146 zs->zs_thread_stop = MIN(zs->zs_thread_stop, zs->zs_proc_stop);
5147 zs->zs_thread_kill = zs->zs_thread_stop;
5148 if (ztest_random(100) < zopt_killrate)
5149 zs->zs_thread_kill -= ztest_random(zopt_passtime * NANOSEC);
5151 (void) _mutex_init(&zcl.zcl_callbacks_lock, USYNC_THREAD, NULL);
5153 list_create(&zcl.zcl_callbacks, sizeof (ztest_cb_data_t),
5154 offsetof(ztest_cb_data_t, zcd_node));
5159 kernel_init(FREAD | FWRITE);
5160 VERIFY(spa_open(zs->zs_pool, &spa, FTAG) == 0);
5163 spa->spa_dedup_ditto = 2 * ZIO_DEDUPDITTO_MIN;
5166 * We don't expect the pool to suspend unless maxfaults == 0,
5167 * in which case ztest_fault_inject() temporarily takes away
5168 * the only valid replica.
5170 if (MAXFAULTS() == 0)
5171 spa->spa_failmode = ZIO_FAILURE_MODE_WAIT;
5173 spa->spa_failmode = ZIO_FAILURE_MODE_PANIC;
5176 * Create a thread to periodically resume suspended I/O.
5178 VERIFY(thr_create(0, 0, ztest_resume_thread, spa, THR_BOUND,
5182 * Create a deadman thread to abort() if we hang.
5184 VERIFY(thr_create(0, 0, ztest_deadman_thread, zs, THR_BOUND,
5188 * Verify that we can safely inquire about about any object,
5189 * whether it's allocated or not. To make it interesting,
5190 * we probe a 5-wide window around each power of two.
5191 * This hits all edge cases, including zero and the max.
5193 for (t = 0; t < 64; t++) {
5194 for (d = -5; d <= 5; d++) {
5195 error = dmu_object_info(spa->spa_meta_objset,
5196 (1ULL << t) + d, NULL);
5197 ASSERT(error == 0 || error == ENOENT ||
5203 * If we got any ENOSPC errors on the previous run, destroy something.
5205 if (zs->zs_enospc_count != 0) {
5206 int d = ztest_random(zopt_datasets);
5207 ztest_dataset_destroy(zs, d);
5209 zs->zs_enospc_count = 0;
5211 tid = umem_zalloc(zopt_threads * sizeof (thread_t), UMEM_NOFAIL);
5213 if (zopt_verbose >= 4)
5214 (void) printf("starting main threads...\n");
5217 * Kick off all the tests that run in parallel.
5219 for (t = 0; t < zopt_threads; t++) {
5220 if (t < zopt_datasets && ztest_dataset_open(zs, t) != 0)
5222 VERIFY(thr_create(0, 0, ztest_thread, (void *)(uintptr_t)t,
5223 THR_BOUND, &tid[t]) == 0);
5227 * Wait for all of the tests to complete. We go in reverse order
5228 * so we don't close datasets while threads are still using them.
5230 for (t = zopt_threads - 1; t >= 0; t--) {
5231 VERIFY(thr_join(tid[t], NULL, NULL) == 0);
5232 if (t < zopt_datasets)
5233 ztest_dataset_close(zs, t);
5236 txg_wait_synced(spa_get_dsl(spa), 0);
5238 zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(spa));
5239 zs->zs_space = metaslab_class_get_space(spa_normal_class(spa));
5241 umem_free(tid, zopt_threads * sizeof (thread_t));
5243 /* Kill the resume thread */
5244 ztest_exiting = B_TRUE;
5245 VERIFY(thr_join(resume_tid, NULL, NULL) == 0);
5249 * Right before closing the pool, kick off a bunch of async I/O;
5250 * spa_close() should wait for it to complete.
5252 for (uint64_t object = 1; object < 50; object++)
5253 dmu_prefetch(spa->spa_meta_objset, object, 0, 1ULL << 20);
5255 spa_close(spa, FTAG);
5258 * Verify that we can loop over all pools.
5260 mutex_enter(&spa_namespace_lock);
5261 for (spa = spa_next(NULL); spa != NULL; spa = spa_next(spa))
5262 if (zopt_verbose > 3)
5263 (void) printf("spa_next: found %s\n", spa_name(spa));
5264 mutex_exit(&spa_namespace_lock);
5267 * Verify that we can export the pool and reimport it under a
5270 if (ztest_random(2) == 0) {
5271 char name[MAXNAMELEN];
5272 (void) snprintf(name, MAXNAMELEN, "%s_import", zs->zs_pool);
5273 ztest_spa_import_export(zs->zs_pool, name);
5274 ztest_spa_import_export(name, zs->zs_pool);
5279 list_destroy(&zcl.zcl_callbacks);
5281 (void) _mutex_destroy(&zcl.zcl_callbacks_lock);
5283 (void) rwlock_destroy(&zs->zs_name_lock);
5284 (void) _mutex_destroy(&zs->zs_vdev_lock);
5288 ztest_freeze(ztest_shared_t *zs)
5290 ztest_ds_t *zd = &zs->zs_zd[0];
5294 if (zopt_verbose >= 3)
5295 (void) printf("testing spa_freeze()...\n");
5297 kernel_init(FREAD | FWRITE);
5298 VERIFY3U(0, ==, spa_open(zs->zs_pool, &spa, FTAG));
5299 VERIFY3U(0, ==, ztest_dataset_open(zs, 0));
5302 * Force the first log block to be transactionally allocated.
5303 * We have to do this before we freeze the pool -- otherwise
5304 * the log chain won't be anchored.
5306 while (BP_IS_HOLE(&zd->zd_zilog->zl_header->zh_log)) {
5307 ztest_dmu_object_alloc_free(zd, 0);
5308 zil_commit(zd->zd_zilog, 0);
5311 txg_wait_synced(spa_get_dsl(spa), 0);
5314 * Freeze the pool. This stops spa_sync() from doing anything,
5315 * so that the only way to record changes from now on is the ZIL.
5320 * Run tests that generate log records but don't alter the pool config
5321 * or depend on DSL sync tasks (snapshots, objset create/destroy, etc).
5322 * We do a txg_wait_synced() after each iteration to force the txg
5323 * to increase well beyond the last synced value in the uberblock.
5324 * The ZIL should be OK with that.
5326 while (ztest_random(10) != 0 && numloops++ < zopt_maxloops) {
5327 ztest_dmu_write_parallel(zd, 0);
5328 ztest_dmu_object_alloc_free(zd, 0);
5329 txg_wait_synced(spa_get_dsl(spa), 0);
5333 * Commit all of the changes we just generated.
5335 zil_commit(zd->zd_zilog, 0);
5336 txg_wait_synced(spa_get_dsl(spa), 0);
5339 * Close our dataset and close the pool.
5341 ztest_dataset_close(zs, 0);
5342 spa_close(spa, FTAG);
5346 * Open and close the pool and dataset to induce log replay.
5348 kernel_init(FREAD | FWRITE);
5349 VERIFY3U(0, ==, spa_open(zs->zs_pool, &spa, FTAG));
5350 VERIFY3U(0, ==, ztest_dataset_open(zs, 0));
5351 ztest_dataset_close(zs, 0);
5352 spa_close(spa, FTAG);
5357 print_time(hrtime_t t, char *timebuf)
5359 hrtime_t s = t / NANOSEC;
5360 hrtime_t m = s / 60;
5361 hrtime_t h = m / 60;
5362 hrtime_t d = h / 24;
5371 (void) sprintf(timebuf,
5372 "%llud%02lluh%02llum%02llus", d, h, m, s);
5374 (void) sprintf(timebuf, "%lluh%02llum%02llus", h, m, s);
5376 (void) sprintf(timebuf, "%llum%02llus", m, s);
5378 (void) sprintf(timebuf, "%llus", s);
5386 if (ztest_random(2) == 0)
5389 VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
5390 VERIFY(nvlist_add_uint64(props, "autoreplace", 1) == 0);
5392 (void) printf("props:\n");
5393 dump_nvlist(props, 4);
5399 * Create a storage pool with the given name and initial vdev size.
5400 * Then test spa_freeze() functionality.
5403 ztest_init(ztest_shared_t *zs)
5406 nvlist_t *nvroot, *props;
5408 VERIFY(_mutex_init(&zs->zs_vdev_lock, USYNC_THREAD, NULL) == 0);
5409 VERIFY(rwlock_init(&zs->zs_name_lock, USYNC_THREAD, NULL) == 0);
5411 kernel_init(FREAD | FWRITE);
5414 * Create the storage pool.
5416 (void) spa_destroy(zs->zs_pool);
5417 ztest_shared->zs_vdev_next_leaf = 0;
5419 zs->zs_mirrors = zopt_mirrors;
5420 nvroot = make_vdev_root(NULL, NULL, zopt_vdev_size, 0,
5421 0, zopt_raidz, zs->zs_mirrors, 1);
5422 props = make_random_props();
5423 VERIFY3U(0, ==, spa_create(zs->zs_pool, nvroot, props, NULL, NULL));
5424 nvlist_free(nvroot);
5426 VERIFY3U(0, ==, spa_open(zs->zs_pool, &spa, FTAG));
5427 metaslab_sz = 1ULL << spa->spa_root_vdev->vdev_child[0]->vdev_ms_shift;
5428 spa_close(spa, FTAG);
5432 ztest_run_zdb(zs->zs_pool);
5436 ztest_run_zdb(zs->zs_pool);
5438 (void) rwlock_destroy(&zs->zs_name_lock);
5439 (void) _mutex_destroy(&zs->zs_vdev_lock);
5443 main(int argc, char **argv)
5455 (void) setvbuf(stdout, NULL, _IOLBF, 0);
5457 ztest_random_fd = open("/dev/urandom", O_RDONLY);
5459 process_options(argc, argv);
5461 /* Override location of zpool.cache */
5462 VERIFY(asprintf((char **)&spa_config_path, "%s/zpool.cache",
5466 * Blow away any existing copy of zpool.cache
5469 (void) remove(spa_config_path);
5471 shared_size = sizeof (*zs) + zopt_datasets * sizeof (ztest_ds_t);
5473 zs = ztest_shared = (void *)mmap(0,
5474 P2ROUNDUP(shared_size, getpagesize()),
5475 PROT_READ | PROT_WRITE, MAP_SHARED | MAP_ANON, -1, 0);
5477 if (zopt_verbose >= 1) {
5478 (void) printf("%llu vdevs, %d datasets, %d threads,"
5479 " %llu seconds...\n",
5480 (u_longlong_t)zopt_vdevs, zopt_datasets, zopt_threads,
5481 (u_longlong_t)zopt_time);
5485 * Create and initialize our storage pool.
5487 for (i = 1; i <= zopt_init; i++) {
5488 bzero(zs, sizeof (ztest_shared_t));
5489 if (zopt_verbose >= 3 && zopt_init != 1)
5490 (void) printf("ztest_init(), pass %d\n", i);
5491 zs->zs_pool = zopt_pool;
5495 zs->zs_pool = zopt_pool;
5496 zs->zs_proc_start = gethrtime();
5497 zs->zs_proc_stop = zs->zs_proc_start + zopt_time * NANOSEC;
5499 for (f = 0; f < ZTEST_FUNCS; f++) {
5500 zi = &zs->zs_info[f];
5501 *zi = ztest_info[f];
5502 if (zs->zs_proc_start + zi->zi_interval[0] > zs->zs_proc_stop)
5503 zi->zi_call_next = UINT64_MAX;
5505 zi->zi_call_next = zs->zs_proc_start +
5506 ztest_random(2 * zi->zi_interval[0] + 1);
5510 * Run the tests in a loop. These tests include fault injection
5511 * to verify that self-healing data works, and forced crashes
5512 * to verify that we never lose on-disk consistency.
5514 while (gethrtime() < zs->zs_proc_stop) {
5519 * Initialize the workload counters for each function.
5521 for (f = 0; f < ZTEST_FUNCS; f++) {
5522 zi = &zs->zs_info[f];
5523 zi->zi_call_count = 0;
5524 zi->zi_call_time = 0;
5527 /* Set the allocation switch size */
5528 metaslab_df_alloc_threshold = ztest_random(metaslab_sz / 4) + 1;
5533 fatal(1, "fork failed");
5535 if (pid == 0) { /* child */
5536 struct rlimit rl = { 1024, 1024 };
5537 (void) setrlimit(RLIMIT_NOFILE, &rl);
5538 (void) enable_extended_FILE_stdio(-1, -1);
5543 while (waitpid(pid, &status, 0) != pid)
5546 if (WIFEXITED(status)) {
5547 if (WEXITSTATUS(status) != 0) {
5548 (void) fprintf(stderr,
5549 "child exited with code %d\n",
5550 WEXITSTATUS(status));
5553 } else if (WIFSIGNALED(status)) {
5554 if (WTERMSIG(status) != SIGKILL) {
5555 (void) fprintf(stderr,
5556 "child died with signal %d\n",
5562 (void) fprintf(stderr, "something strange happened "
5569 if (zopt_verbose >= 1) {
5570 hrtime_t now = gethrtime();
5572 now = MIN(now, zs->zs_proc_stop);
5573 print_time(zs->zs_proc_stop - now, timebuf);
5574 nicenum(zs->zs_space, numbuf);
5576 (void) printf("Pass %3d, %8s, %3llu ENOSPC, "
5577 "%4.1f%% of %5s used, %3.0f%% done, %8s to go\n",
5579 WIFEXITED(status) ? "Complete" : "SIGKILL",
5580 (u_longlong_t)zs->zs_enospc_count,
5581 100.0 * zs->zs_alloc / zs->zs_space,
5583 100.0 * (now - zs->zs_proc_start) /
5584 (zopt_time * NANOSEC), timebuf);
5587 if (zopt_verbose >= 2) {
5588 (void) printf("\nWorkload summary:\n\n");
5589 (void) printf("%7s %9s %s\n",
5590 "Calls", "Time", "Function");
5591 (void) printf("%7s %9s %s\n",
5592 "-----", "----", "--------");
5593 for (f = 0; f < ZTEST_FUNCS; f++) {
5596 zi = &zs->zs_info[f];
5597 print_time(zi->zi_call_time, timebuf);
5598 (void) dladdr((void *)zi->zi_func, &dli);
5599 (void) printf("%7llu %9s %s\n",
5600 (u_longlong_t)zi->zi_call_count, timebuf,
5603 (void) printf("\n");
5607 * It's possible that we killed a child during a rename test,
5608 * in which case we'll have a 'ztest_tmp' pool lying around
5609 * instead of 'ztest'. Do a blind rename in case this happened.
5612 if (spa_open(zopt_pool, &spa, FTAG) == 0) {
5613 spa_close(spa, FTAG);
5615 char tmpname[MAXNAMELEN];
5617 kernel_init(FREAD | FWRITE);
5618 (void) snprintf(tmpname, sizeof (tmpname), "%s_tmp",
5620 (void) spa_rename(tmpname, zopt_pool);
5624 ztest_run_zdb(zopt_pool);
5627 if (zopt_verbose >= 1) {
5628 (void) printf("%d killed, %d completed, %.0f%% kill rate\n",
5629 kills, iters - kills, (100.0 * kills) / MAX(1, iters));