4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2011, 2015 by Delphix. All rights reserved.
24 * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
25 * Copyright (c) 2013 Steven Hartland. All rights reserved.
29 * The objective of this program is to provide a DMU/ZAP/SPA stress test
30 * that runs entirely in userland, is easy to use, and easy to extend.
32 * The overall design of the ztest program is as follows:
34 * (1) For each major functional area (e.g. adding vdevs to a pool,
35 * creating and destroying datasets, reading and writing objects, etc)
36 * we have a simple routine to test that functionality. These
37 * individual routines do not have to do anything "stressful".
39 * (2) We turn these simple functionality tests into a stress test by
40 * running them all in parallel, with as many threads as desired,
41 * and spread across as many datasets, objects, and vdevs as desired.
43 * (3) While all this is happening, we inject faults into the pool to
44 * verify that self-healing data really works.
46 * (4) Every time we open a dataset, we change its checksum and compression
47 * functions. Thus even individual objects vary from block to block
48 * in which checksum they use and whether they're compressed.
50 * (5) To verify that we never lose on-disk consistency after a crash,
51 * we run the entire test in a child of the main process.
52 * At random times, the child self-immolates with a SIGKILL.
53 * This is the software equivalent of pulling the power cord.
54 * The parent then runs the test again, using the existing
55 * storage pool, as many times as desired. If backwards compatibility
56 * testing is enabled ztest will sometimes run the "older" version
57 * of ztest after a SIGKILL.
59 * (6) To verify that we don't have future leaks or temporal incursions,
60 * many of the functional tests record the transaction group number
61 * as part of their data. When reading old data, they verify that
62 * the transaction group number is less than the current, open txg.
63 * If you add a new test, please do this if applicable.
65 * (7) Threads are created with a reduced stack size, for sanity checking.
66 * Therefore, it's important not to allocate huge buffers on the stack.
68 * When run with no arguments, ztest runs for about five minutes and
69 * produces no output if successful. To get a little bit of information,
70 * specify -V. To get more information, specify -VV, and so on.
72 * To turn this into an overnight stress test, use -T to specify run time.
74 * You can ask more more vdevs [-v], datasets [-d], or threads [-t]
75 * to increase the pool capacity, fanout, and overall stress level.
77 * Use the -k option to set the desired frequency of kills.
79 * When ztest invokes itself it passes all relevant information through a
80 * temporary file which is mmap-ed in the child process. This allows shared
81 * memory to survive the exec syscall. The ztest_shared_hdr_t struct is always
82 * stored at offset 0 of this file and contains information on the size and
83 * number of shared structures in the file. The information stored in this file
84 * must remain backwards compatible with older versions of ztest so that
85 * ztest can invoke them during backwards compatibility testing (-B).
88 #include <sys/zfs_context.h>
94 #include <sys/dmu_objset.h>
100 #include <sys/resource.h>
103 #include <sys/zil_impl.h>
104 #include <sys/vdev_impl.h>
105 #include <sys/vdev_file.h>
106 #include <sys/spa_impl.h>
107 #include <sys/metaslab_impl.h>
108 #include <sys/dsl_prop.h>
109 #include <sys/dsl_dataset.h>
110 #include <sys/dsl_destroy.h>
111 #include <sys/dsl_scan.h>
112 #include <sys/zio_checksum.h>
113 #include <sys/refcount.h>
114 #include <sys/zfeature.h>
115 #include <sys/dsl_userhold.h>
117 #include <stdio_ext.h>
124 #include <sys/fs/zfs.h>
125 #include <libnvpair.h>
127 #include <execinfo.h> /* for backtrace() */
130 static int ztest_fd_data = -1;
131 static int ztest_fd_rand = -1;
133 typedef struct ztest_shared_hdr {
134 uint64_t zh_hdr_size;
135 uint64_t zh_opts_size;
137 uint64_t zh_stats_size;
138 uint64_t zh_stats_count;
140 uint64_t zh_ds_count;
141 } ztest_shared_hdr_t;
143 static ztest_shared_hdr_t *ztest_shared_hdr;
145 typedef struct ztest_shared_opts {
146 char zo_pool[MAXNAMELEN];
147 char zo_dir[MAXNAMELEN];
148 char zo_alt_ztest[MAXNAMELEN];
149 char zo_alt_libpath[MAXNAMELEN];
151 uint64_t zo_vdevtime;
159 uint64_t zo_passtime;
160 uint64_t zo_killrate;
164 uint64_t zo_maxloops;
165 uint64_t zo_metaslab_gang_bang;
166 } ztest_shared_opts_t;
168 static const ztest_shared_opts_t ztest_opts_defaults = {
169 .zo_pool = { 'z', 't', 'e', 's', 't', '\0' },
170 .zo_dir = { '/', 't', 'm', 'p', '\0' },
171 .zo_alt_ztest = { '\0' },
172 .zo_alt_libpath = { '\0' },
174 .zo_ashift = SPA_MINBLOCKSHIFT,
177 .zo_raidz_parity = 1,
178 .zo_vdev_size = SPA_MINDEVSIZE * 2,
181 .zo_passtime = 60, /* 60 seconds */
182 .zo_killrate = 70, /* 70% kill rate */
185 .zo_time = 300, /* 5 minutes */
186 .zo_maxloops = 50, /* max loops during spa_freeze() */
187 .zo_metaslab_gang_bang = 32 << 10
190 extern uint64_t metaslab_gang_bang;
191 extern uint64_t metaslab_df_alloc_threshold;
192 extern int metaslab_preload_limit;
194 static ztest_shared_opts_t *ztest_shared_opts;
195 static ztest_shared_opts_t ztest_opts;
197 typedef struct ztest_shared_ds {
201 static ztest_shared_ds_t *ztest_shared_ds;
202 #define ZTEST_GET_SHARED_DS(d) (&ztest_shared_ds[d])
204 #define BT_MAGIC 0x123456789abcdefULL
205 #define MAXFAULTS() \
206 (MAX(zs->zs_mirrors, 1) * (ztest_opts.zo_raidz_parity + 1) - 1)
210 ZTEST_IO_WRITE_PATTERN,
211 ZTEST_IO_WRITE_ZEROES,
218 typedef struct ztest_block_tag {
228 typedef struct bufwad {
235 * XXX -- fix zfs range locks to be generic so we can use them here.
257 #define ZTEST_RANGE_LOCKS 64
258 #define ZTEST_OBJECT_LOCKS 64
261 * Object descriptor. Used as a template for object lookup/create/remove.
263 typedef struct ztest_od {
266 dmu_object_type_t od_type;
267 dmu_object_type_t od_crtype;
268 uint64_t od_blocksize;
269 uint64_t od_crblocksize;
272 char od_name[MAXNAMELEN];
278 typedef struct ztest_ds {
279 ztest_shared_ds_t *zd_shared;
281 rwlock_t zd_zilog_lock;
283 ztest_od_t *zd_od; /* debugging aid */
284 char zd_name[MAXNAMELEN];
285 kmutex_t zd_dirobj_lock;
286 rll_t zd_object_lock[ZTEST_OBJECT_LOCKS];
287 rll_t zd_range_lock[ZTEST_RANGE_LOCKS];
291 * Per-iteration state.
293 typedef void ztest_func_t(ztest_ds_t *zd, uint64_t id);
295 typedef struct ztest_info {
296 ztest_func_t *zi_func; /* test function */
297 uint64_t zi_iters; /* iterations per execution */
298 uint64_t *zi_interval; /* execute every <interval> seconds */
299 const char *zi_funcname; /* name of test function */
302 typedef struct ztest_shared_callstate {
303 uint64_t zc_count; /* per-pass count */
304 uint64_t zc_time; /* per-pass time */
305 uint64_t zc_next; /* next time to call this function */
306 } ztest_shared_callstate_t;
308 static ztest_shared_callstate_t *ztest_shared_callstate;
309 #define ZTEST_GET_SHARED_CALLSTATE(c) (&ztest_shared_callstate[c])
311 ztest_func_t ztest_dmu_read_write;
312 ztest_func_t ztest_dmu_write_parallel;
313 ztest_func_t ztest_dmu_object_alloc_free;
314 ztest_func_t ztest_dmu_commit_callbacks;
315 ztest_func_t ztest_zap;
316 ztest_func_t ztest_zap_parallel;
317 ztest_func_t ztest_zil_commit;
318 ztest_func_t ztest_zil_remount;
319 ztest_func_t ztest_dmu_read_write_zcopy;
320 ztest_func_t ztest_dmu_objset_create_destroy;
321 ztest_func_t ztest_dmu_prealloc;
322 ztest_func_t ztest_fzap;
323 ztest_func_t ztest_dmu_snapshot_create_destroy;
324 ztest_func_t ztest_dsl_prop_get_set;
325 ztest_func_t ztest_spa_prop_get_set;
326 ztest_func_t ztest_spa_create_destroy;
327 ztest_func_t ztest_fault_inject;
328 ztest_func_t ztest_ddt_repair;
329 ztest_func_t ztest_dmu_snapshot_hold;
330 ztest_func_t ztest_spa_rename;
331 ztest_func_t ztest_scrub;
332 ztest_func_t ztest_dsl_dataset_promote_busy;
333 ztest_func_t ztest_vdev_attach_detach;
334 ztest_func_t ztest_vdev_LUN_growth;
335 ztest_func_t ztest_vdev_add_remove;
336 ztest_func_t ztest_vdev_aux_add_remove;
337 ztest_func_t ztest_split_pool;
338 ztest_func_t ztest_reguid;
339 ztest_func_t ztest_spa_upgrade;
341 uint64_t zopt_always = 0ULL * NANOSEC; /* all the time */
342 uint64_t zopt_incessant = 1ULL * NANOSEC / 10; /* every 1/10 second */
343 uint64_t zopt_often = 1ULL * NANOSEC; /* every second */
344 uint64_t zopt_sometimes = 10ULL * NANOSEC; /* every 10 seconds */
345 uint64_t zopt_rarely = 60ULL * NANOSEC; /* every 60 seconds */
347 #define ZTI_INIT(func, iters, interval) \
348 { .zi_func = (func), \
349 .zi_iters = (iters), \
350 .zi_interval = (interval), \
351 .zi_funcname = # func }
353 ztest_info_t ztest_info[] = {
354 ZTI_INIT(ztest_dmu_read_write, 1, &zopt_always),
355 ZTI_INIT(ztest_dmu_write_parallel, 10, &zopt_always),
356 ZTI_INIT(ztest_dmu_object_alloc_free, 1, &zopt_always),
357 ZTI_INIT(ztest_dmu_commit_callbacks, 1, &zopt_always),
358 ZTI_INIT(ztest_zap, 30, &zopt_always),
359 ZTI_INIT(ztest_zap_parallel, 100, &zopt_always),
360 ZTI_INIT(ztest_split_pool, 1, &zopt_always),
361 ZTI_INIT(ztest_zil_commit, 1, &zopt_incessant),
362 ZTI_INIT(ztest_zil_remount, 1, &zopt_sometimes),
363 ZTI_INIT(ztest_dmu_read_write_zcopy, 1, &zopt_often),
364 ZTI_INIT(ztest_dmu_objset_create_destroy, 1, &zopt_often),
365 ZTI_INIT(ztest_dsl_prop_get_set, 1, &zopt_often),
366 ZTI_INIT(ztest_spa_prop_get_set, 1, &zopt_sometimes),
368 ZTI_INIT(ztest_dmu_prealloc, 1, &zopt_sometimes),
370 ZTI_INIT(ztest_fzap, 1, &zopt_sometimes),
371 ZTI_INIT(ztest_dmu_snapshot_create_destroy, 1, &zopt_sometimes),
372 ZTI_INIT(ztest_spa_create_destroy, 1, &zopt_sometimes),
373 ZTI_INIT(ztest_fault_inject, 1, &zopt_sometimes),
374 ZTI_INIT(ztest_ddt_repair, 1, &zopt_sometimes),
375 ZTI_INIT(ztest_dmu_snapshot_hold, 1, &zopt_sometimes),
376 ZTI_INIT(ztest_reguid, 1, &zopt_rarely),
377 ZTI_INIT(ztest_spa_rename, 1, &zopt_rarely),
378 ZTI_INIT(ztest_scrub, 1, &zopt_rarely),
379 ZTI_INIT(ztest_spa_upgrade, 1, &zopt_rarely),
380 ZTI_INIT(ztest_dsl_dataset_promote_busy, 1, &zopt_rarely),
381 ZTI_INIT(ztest_vdev_attach_detach, 1, &zopt_sometimes),
382 ZTI_INIT(ztest_vdev_LUN_growth, 1, &zopt_rarely),
383 ZTI_INIT(ztest_vdev_add_remove, 1, &ztest_opts.zo_vdevtime),
384 ZTI_INIT(ztest_vdev_aux_add_remove, 1, &ztest_opts.zo_vdevtime),
387 #define ZTEST_FUNCS (sizeof (ztest_info) / sizeof (ztest_info_t))
390 * The following struct is used to hold a list of uncalled commit callbacks.
391 * The callbacks are ordered by txg number.
393 typedef struct ztest_cb_list {
394 kmutex_t zcl_callbacks_lock;
395 list_t zcl_callbacks;
399 * Stuff we need to share writably between parent and child.
401 typedef struct ztest_shared {
402 boolean_t zs_do_init;
403 hrtime_t zs_proc_start;
404 hrtime_t zs_proc_stop;
405 hrtime_t zs_thread_start;
406 hrtime_t zs_thread_stop;
407 hrtime_t zs_thread_kill;
408 uint64_t zs_enospc_count;
409 uint64_t zs_vdev_next_leaf;
410 uint64_t zs_vdev_aux;
415 uint64_t zs_metaslab_sz;
416 uint64_t zs_metaslab_df_alloc_threshold;
420 #define ID_PARALLEL -1ULL
422 static char ztest_dev_template[] = "%s/%s.%llua";
423 static char ztest_aux_template[] = "%s/%s.%s.%llu";
424 ztest_shared_t *ztest_shared;
426 static spa_t *ztest_spa = NULL;
427 static ztest_ds_t *ztest_ds;
429 static kmutex_t ztest_vdev_lock;
432 * The ztest_name_lock protects the pool and dataset namespace used by
433 * the individual tests. To modify the namespace, consumers must grab
434 * this lock as writer. Grabbing the lock as reader will ensure that the
435 * namespace does not change while the lock is held.
437 static rwlock_t ztest_name_lock;
439 static boolean_t ztest_dump_core = B_TRUE;
440 static boolean_t ztest_exiting;
442 /* Global commit callback list */
443 static ztest_cb_list_t zcl;
444 /* Commit cb delay */
445 static uint64_t zc_min_txg_delay = UINT64_MAX;
446 static int zc_cb_counter = 0;
449 * Minimum number of commit callbacks that need to be registered for us to check
450 * whether the minimum txg delay is acceptable.
452 #define ZTEST_COMMIT_CB_MIN_REG 100
455 * If a number of txgs equal to this threshold have been created after a commit
456 * callback has been registered but not called, then we assume there is an
457 * implementation bug.
459 #define ZTEST_COMMIT_CB_THRESH (TXG_CONCURRENT_STATES + 1000)
461 extern uint64_t metaslab_gang_bang;
462 extern uint64_t metaslab_df_alloc_threshold;
465 ZTEST_META_DNODE = 0,
470 static void usage(boolean_t) __NORETURN;
473 * These libumem hooks provide a reasonable set of defaults for the allocator's
474 * debugging facilities.
477 _umem_debug_init(void)
479 return ("default,verbose"); /* $UMEM_DEBUG setting */
483 _umem_logging_init(void)
485 return ("fail,contents"); /* $UMEM_LOGGING setting */
488 #define BACKTRACE_SZ 100
490 static void sig_handler(int signo)
492 struct sigaction action;
493 #ifdef __GNUC__ /* backtrace() is a GNU extension */
495 void *buffer[BACKTRACE_SZ];
497 nptrs = backtrace(buffer, BACKTRACE_SZ);
498 backtrace_symbols_fd(buffer, nptrs, STDERR_FILENO);
502 * Restore default action and re-raise signal so SIGSEGV and
503 * SIGABRT can trigger a core dump.
505 action.sa_handler = SIG_DFL;
506 sigemptyset(&action.sa_mask);
508 (void) sigaction(signo, &action, NULL);
512 #define FATAL_MSG_SZ 1024
517 fatal(int do_perror, char *message, ...)
520 int save_errno = errno;
523 (void) fflush(stdout);
524 buf = umem_alloc(FATAL_MSG_SZ, UMEM_NOFAIL);
526 va_start(args, message);
527 (void) sprintf(buf, "ztest: ");
529 (void) vsprintf(buf + strlen(buf), message, args);
532 (void) snprintf(buf + strlen(buf), FATAL_MSG_SZ - strlen(buf),
533 ": %s", strerror(save_errno));
535 (void) fprintf(stderr, "%s\n", buf);
536 fatal_msg = buf; /* to ease debugging */
543 str2shift(const char *buf)
545 const char *ends = "BKMGTPEZ";
550 for (i = 0; i < strlen(ends); i++) {
551 if (toupper(buf[0]) == ends[i])
554 if (i == strlen(ends)) {
555 (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n",
559 if (buf[1] == '\0' || (toupper(buf[1]) == 'B' && buf[2] == '\0')) {
562 (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n", buf);
568 nicenumtoull(const char *buf)
573 val = strtoull(buf, &end, 0);
575 (void) fprintf(stderr, "ztest: bad numeric value: %s\n", buf);
577 } else if (end[0] == '.') {
578 double fval = strtod(buf, &end);
579 fval *= pow(2, str2shift(end));
580 if (fval > UINT64_MAX) {
581 (void) fprintf(stderr, "ztest: value too large: %s\n",
585 val = (uint64_t)fval;
587 int shift = str2shift(end);
588 if (shift >= 64 || (val << shift) >> shift != val) {
589 (void) fprintf(stderr, "ztest: value too large: %s\n",
599 usage(boolean_t requested)
601 const ztest_shared_opts_t *zo = &ztest_opts_defaults;
603 char nice_vdev_size[10];
604 char nice_gang_bang[10];
605 FILE *fp = requested ? stdout : stderr;
607 nicenum(zo->zo_vdev_size, nice_vdev_size);
608 nicenum(zo->zo_metaslab_gang_bang, nice_gang_bang);
610 (void) fprintf(fp, "Usage: %s\n"
611 "\t[-v vdevs (default: %llu)]\n"
612 "\t[-s size_of_each_vdev (default: %s)]\n"
613 "\t[-a alignment_shift (default: %d)] use 0 for random\n"
614 "\t[-m mirror_copies (default: %d)]\n"
615 "\t[-r raidz_disks (default: %d)]\n"
616 "\t[-R raidz_parity (default: %d)]\n"
617 "\t[-d datasets (default: %d)]\n"
618 "\t[-t threads (default: %d)]\n"
619 "\t[-g gang_block_threshold (default: %s)]\n"
620 "\t[-i init_count (default: %d)] initialize pool i times\n"
621 "\t[-k kill_percentage (default: %llu%%)]\n"
622 "\t[-p pool_name (default: %s)]\n"
623 "\t[-f dir (default: %s)] file directory for vdev files\n"
624 "\t[-V] verbose (use multiple times for ever more blather)\n"
625 "\t[-E] use existing pool instead of creating new one\n"
626 "\t[-T time (default: %llu sec)] total run time\n"
627 "\t[-F freezeloops (default: %llu)] max loops in spa_freeze()\n"
628 "\t[-P passtime (default: %llu sec)] time per pass\n"
629 "\t[-B alt_ztest (default: <none>)] alternate ztest path\n"
630 "\t[-h] (print help)\n"
633 (u_longlong_t)zo->zo_vdevs, /* -v */
634 nice_vdev_size, /* -s */
635 zo->zo_ashift, /* -a */
636 zo->zo_mirrors, /* -m */
637 zo->zo_raidz, /* -r */
638 zo->zo_raidz_parity, /* -R */
639 zo->zo_datasets, /* -d */
640 zo->zo_threads, /* -t */
641 nice_gang_bang, /* -g */
642 zo->zo_init, /* -i */
643 (u_longlong_t)zo->zo_killrate, /* -k */
644 zo->zo_pool, /* -p */
646 (u_longlong_t)zo->zo_time, /* -T */
647 (u_longlong_t)zo->zo_maxloops, /* -F */
648 (u_longlong_t)zo->zo_passtime);
649 exit(requested ? 0 : 1);
653 process_options(int argc, char **argv)
656 ztest_shared_opts_t *zo = &ztest_opts;
660 char altdir[MAXNAMELEN] = { 0 };
662 bcopy(&ztest_opts_defaults, zo, sizeof (*zo));
664 while ((opt = getopt(argc, argv,
665 "v:s:a:m:r:R:d:t:g:i:k:p:f:VET:P:hF:B:")) != EOF) {
682 value = nicenumtoull(optarg);
686 zo->zo_vdevs = value;
689 zo->zo_vdev_size = MAX(SPA_MINDEVSIZE, value);
692 zo->zo_ashift = value;
695 zo->zo_mirrors = value;
698 zo->zo_raidz = MAX(1, value);
701 zo->zo_raidz_parity = MIN(MAX(value, 1), 3);
704 zo->zo_datasets = MAX(1, value);
707 zo->zo_threads = MAX(1, value);
710 zo->zo_metaslab_gang_bang = MAX(SPA_MINBLOCKSIZE << 1,
717 zo->zo_killrate = value;
720 (void) strlcpy(zo->zo_pool, optarg,
721 sizeof (zo->zo_pool));
724 path = realpath(optarg, NULL);
726 (void) fprintf(stderr, "error: %s: %s\n",
727 optarg, strerror(errno));
730 (void) strlcpy(zo->zo_dir, path,
731 sizeof (zo->zo_dir));
744 zo->zo_passtime = MAX(1, value);
747 zo->zo_maxloops = MAX(1, value);
750 (void) strlcpy(altdir, optarg, sizeof (altdir));
762 zo->zo_raidz_parity = MIN(zo->zo_raidz_parity, zo->zo_raidz - 1);
765 (zo->zo_vdevs > 0 ? zo->zo_time * NANOSEC / zo->zo_vdevs :
768 if (strlen(altdir) > 0) {
776 cmd = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
777 realaltdir = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
779 VERIFY(NULL != realpath(getexecname(), cmd));
780 if (0 != access(altdir, F_OK)) {
781 ztest_dump_core = B_FALSE;
782 fatal(B_TRUE, "invalid alternate ztest path: %s",
785 VERIFY(NULL != realpath(altdir, realaltdir));
788 * 'cmd' should be of the form "<anything>/usr/bin/<isa>/ztest".
789 * We want to extract <isa> to determine if we should use
790 * 32 or 64 bit binaries.
792 bin = strstr(cmd, "/usr/bin/");
793 ztest = strstr(bin, "/ztest");
795 isalen = ztest - isa;
796 (void) snprintf(zo->zo_alt_ztest, sizeof (zo->zo_alt_ztest),
797 "%s/usr/bin/%.*s/ztest", realaltdir, isalen, isa);
798 (void) snprintf(zo->zo_alt_libpath, sizeof (zo->zo_alt_libpath),
799 "%s/usr/lib/%.*s", realaltdir, isalen, isa);
801 if (0 != access(zo->zo_alt_ztest, X_OK)) {
802 ztest_dump_core = B_FALSE;
803 fatal(B_TRUE, "invalid alternate ztest: %s",
805 } else if (0 != access(zo->zo_alt_libpath, X_OK)) {
806 ztest_dump_core = B_FALSE;
807 fatal(B_TRUE, "invalid alternate lib directory %s",
811 umem_free(cmd, MAXPATHLEN);
812 umem_free(realaltdir, MAXPATHLEN);
817 ztest_kill(ztest_shared_t *zs)
819 zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(ztest_spa));
820 zs->zs_space = metaslab_class_get_space(spa_normal_class(ztest_spa));
823 * Before we kill off ztest, make sure that the config is updated.
824 * See comment above spa_config_sync().
826 mutex_enter(&spa_namespace_lock);
827 spa_config_sync(ztest_spa, B_FALSE, B_FALSE);
828 mutex_exit(&spa_namespace_lock);
830 (void) kill(getpid(), SIGKILL);
834 ztest_random(uint64_t range)
838 ASSERT3S(ztest_fd_rand, >=, 0);
843 if (read(ztest_fd_rand, &r, sizeof (r)) != sizeof (r))
844 fatal(1, "short read from /dev/urandom");
851 ztest_record_enospc(const char *s)
853 ztest_shared->zs_enospc_count++;
857 ztest_get_ashift(void)
859 if (ztest_opts.zo_ashift == 0)
860 return (SPA_MINBLOCKSHIFT + ztest_random(5));
861 return (ztest_opts.zo_ashift);
865 make_vdev_file(char *path, char *aux, char *pool, size_t size, uint64_t ashift)
871 pathbuf = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
874 ashift = ztest_get_ashift();
880 vdev = ztest_shared->zs_vdev_aux;
881 (void) snprintf(path, MAXPATHLEN,
882 ztest_aux_template, ztest_opts.zo_dir,
883 pool == NULL ? ztest_opts.zo_pool : pool,
886 vdev = ztest_shared->zs_vdev_next_leaf++;
887 (void) snprintf(path, MAXPATHLEN,
888 ztest_dev_template, ztest_opts.zo_dir,
889 pool == NULL ? ztest_opts.zo_pool : pool, vdev);
894 int fd = open(path, O_RDWR | O_CREAT | O_TRUNC, 0666);
896 fatal(1, "can't open %s", path);
897 if (ftruncate(fd, size) != 0)
898 fatal(1, "can't ftruncate %s", path);
902 VERIFY(nvlist_alloc(&file, NV_UNIQUE_NAME, 0) == 0);
903 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_TYPE, VDEV_TYPE_FILE) == 0);
904 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_PATH, path) == 0);
905 VERIFY(nvlist_add_uint64(file, ZPOOL_CONFIG_ASHIFT, ashift) == 0);
906 umem_free(pathbuf, MAXPATHLEN);
912 make_vdev_raidz(char *path, char *aux, char *pool, size_t size,
913 uint64_t ashift, int r)
915 nvlist_t *raidz, **child;
919 return (make_vdev_file(path, aux, pool, size, ashift));
920 child = umem_alloc(r * sizeof (nvlist_t *), UMEM_NOFAIL);
922 for (c = 0; c < r; c++)
923 child[c] = make_vdev_file(path, aux, pool, size, ashift);
925 VERIFY(nvlist_alloc(&raidz, NV_UNIQUE_NAME, 0) == 0);
926 VERIFY(nvlist_add_string(raidz, ZPOOL_CONFIG_TYPE,
927 VDEV_TYPE_RAIDZ) == 0);
928 VERIFY(nvlist_add_uint64(raidz, ZPOOL_CONFIG_NPARITY,
929 ztest_opts.zo_raidz_parity) == 0);
930 VERIFY(nvlist_add_nvlist_array(raidz, ZPOOL_CONFIG_CHILDREN,
933 for (c = 0; c < r; c++)
934 nvlist_free(child[c]);
936 umem_free(child, r * sizeof (nvlist_t *));
942 make_vdev_mirror(char *path, char *aux, char *pool, size_t size,
943 uint64_t ashift, int r, int m)
945 nvlist_t *mirror, **child;
949 return (make_vdev_raidz(path, aux, pool, size, ashift, r));
951 child = umem_alloc(m * sizeof (nvlist_t *), UMEM_NOFAIL);
953 for (c = 0; c < m; c++)
954 child[c] = make_vdev_raidz(path, aux, pool, size, ashift, r);
956 VERIFY(nvlist_alloc(&mirror, NV_UNIQUE_NAME, 0) == 0);
957 VERIFY(nvlist_add_string(mirror, ZPOOL_CONFIG_TYPE,
958 VDEV_TYPE_MIRROR) == 0);
959 VERIFY(nvlist_add_nvlist_array(mirror, ZPOOL_CONFIG_CHILDREN,
962 for (c = 0; c < m; c++)
963 nvlist_free(child[c]);
965 umem_free(child, m * sizeof (nvlist_t *));
971 make_vdev_root(char *path, char *aux, char *pool, size_t size, uint64_t ashift,
972 int log, int r, int m, int t)
974 nvlist_t *root, **child;
979 child = umem_alloc(t * sizeof (nvlist_t *), UMEM_NOFAIL);
981 for (c = 0; c < t; c++) {
982 child[c] = make_vdev_mirror(path, aux, pool, size, ashift,
984 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_IS_LOG,
988 VERIFY(nvlist_alloc(&root, NV_UNIQUE_NAME, 0) == 0);
989 VERIFY(nvlist_add_string(root, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) == 0);
990 VERIFY(nvlist_add_nvlist_array(root, aux ? aux : ZPOOL_CONFIG_CHILDREN,
993 for (c = 0; c < t; c++)
994 nvlist_free(child[c]);
996 umem_free(child, t * sizeof (nvlist_t *));
1002 * Find a random spa version. Returns back a random spa version in the
1003 * range [initial_version, SPA_VERSION_FEATURES].
1006 ztest_random_spa_version(uint64_t initial_version)
1008 uint64_t version = initial_version;
1010 if (version <= SPA_VERSION_BEFORE_FEATURES) {
1012 ztest_random(SPA_VERSION_BEFORE_FEATURES - version + 1);
1015 if (version > SPA_VERSION_BEFORE_FEATURES)
1016 version = SPA_VERSION_FEATURES;
1018 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
1023 ztest_random_blocksize(void)
1026 * Choose a block size >= the ashift.
1027 * If the SPA supports new MAXBLOCKSIZE, test up to 1MB blocks.
1029 int maxbs = SPA_OLD_MAXBLOCKSHIFT;
1030 if (spa_maxblocksize(ztest_spa) == SPA_MAXBLOCKSIZE)
1032 uint64_t block_shift =
1033 ztest_random(maxbs - ztest_spa->spa_max_ashift + 1);
1034 return (1 << (SPA_MINBLOCKSHIFT + block_shift));
1038 ztest_random_ibshift(void)
1040 return (DN_MIN_INDBLKSHIFT +
1041 ztest_random(DN_MAX_INDBLKSHIFT - DN_MIN_INDBLKSHIFT + 1));
1045 ztest_random_vdev_top(spa_t *spa, boolean_t log_ok)
1048 vdev_t *rvd = spa->spa_root_vdev;
1051 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1054 top = ztest_random(rvd->vdev_children);
1055 tvd = rvd->vdev_child[top];
1056 } while (tvd->vdev_ishole || (tvd->vdev_islog && !log_ok) ||
1057 tvd->vdev_mg == NULL || tvd->vdev_mg->mg_class == NULL);
1063 ztest_random_dsl_prop(zfs_prop_t prop)
1068 value = zfs_prop_random_value(prop, ztest_random(-1ULL));
1069 } while (prop == ZFS_PROP_CHECKSUM && value == ZIO_CHECKSUM_OFF);
1075 ztest_dsl_prop_set_uint64(char *osname, zfs_prop_t prop, uint64_t value,
1078 const char *propname = zfs_prop_to_name(prop);
1079 const char *valname;
1084 error = dsl_prop_set_int(osname, propname,
1085 (inherit ? ZPROP_SRC_NONE : ZPROP_SRC_LOCAL), value);
1087 if (error == ENOSPC) {
1088 ztest_record_enospc(FTAG);
1093 setpoint = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
1094 VERIFY0(dsl_prop_get_integer(osname, propname, &curval, setpoint));
1096 if (ztest_opts.zo_verbose >= 6) {
1099 err = zfs_prop_index_to_string(prop, curval, &valname);
1101 (void) printf("%s %s = %llu at '%s'\n",
1102 osname, propname, (unsigned long long)curval,
1105 (void) printf("%s %s = %s at '%s'\n",
1106 osname, propname, valname, setpoint);
1108 umem_free(setpoint, MAXPATHLEN);
1114 ztest_spa_prop_set_uint64(zpool_prop_t prop, uint64_t value)
1116 spa_t *spa = ztest_spa;
1117 nvlist_t *props = NULL;
1120 VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
1121 VERIFY(nvlist_add_uint64(props, zpool_prop_to_name(prop), value) == 0);
1123 error = spa_prop_set(spa, props);
1127 if (error == ENOSPC) {
1128 ztest_record_enospc(FTAG);
1137 ztest_rll_init(rll_t *rll)
1139 rll->rll_writer = NULL;
1140 rll->rll_readers = 0;
1141 mutex_init(&rll->rll_lock, NULL, MUTEX_DEFAULT, NULL);
1142 cv_init(&rll->rll_cv, NULL, CV_DEFAULT, NULL);
1146 ztest_rll_destroy(rll_t *rll)
1148 ASSERT(rll->rll_writer == NULL);
1149 ASSERT(rll->rll_readers == 0);
1150 mutex_destroy(&rll->rll_lock);
1151 cv_destroy(&rll->rll_cv);
1155 ztest_rll_lock(rll_t *rll, rl_type_t type)
1157 mutex_enter(&rll->rll_lock);
1159 if (type == RL_READER) {
1160 while (rll->rll_writer != NULL)
1161 (void) cv_wait(&rll->rll_cv, &rll->rll_lock);
1164 while (rll->rll_writer != NULL || rll->rll_readers)
1165 (void) cv_wait(&rll->rll_cv, &rll->rll_lock);
1166 rll->rll_writer = curthread;
1169 mutex_exit(&rll->rll_lock);
1173 ztest_rll_unlock(rll_t *rll)
1175 mutex_enter(&rll->rll_lock);
1177 if (rll->rll_writer) {
1178 ASSERT(rll->rll_readers == 0);
1179 rll->rll_writer = NULL;
1181 ASSERT(rll->rll_readers != 0);
1182 ASSERT(rll->rll_writer == NULL);
1186 if (rll->rll_writer == NULL && rll->rll_readers == 0)
1187 cv_broadcast(&rll->rll_cv);
1189 mutex_exit(&rll->rll_lock);
1193 ztest_object_lock(ztest_ds_t *zd, uint64_t object, rl_type_t type)
1195 rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1197 ztest_rll_lock(rll, type);
1201 ztest_object_unlock(ztest_ds_t *zd, uint64_t object)
1203 rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1205 ztest_rll_unlock(rll);
1209 ztest_range_lock(ztest_ds_t *zd, uint64_t object, uint64_t offset,
1210 uint64_t size, rl_type_t type)
1212 uint64_t hash = object ^ (offset % (ZTEST_RANGE_LOCKS + 1));
1213 rll_t *rll = &zd->zd_range_lock[hash & (ZTEST_RANGE_LOCKS - 1)];
1216 rl = umem_alloc(sizeof (*rl), UMEM_NOFAIL);
1217 rl->rl_object = object;
1218 rl->rl_offset = offset;
1222 ztest_rll_lock(rll, type);
1228 ztest_range_unlock(rl_t *rl)
1230 rll_t *rll = rl->rl_lock;
1232 ztest_rll_unlock(rll);
1234 umem_free(rl, sizeof (*rl));
1238 ztest_zd_init(ztest_ds_t *zd, ztest_shared_ds_t *szd, objset_t *os)
1241 zd->zd_zilog = dmu_objset_zil(os);
1242 zd->zd_shared = szd;
1243 dmu_objset_name(os, zd->zd_name);
1246 if (zd->zd_shared != NULL)
1247 zd->zd_shared->zd_seq = 0;
1249 VERIFY(rwlock_init(&zd->zd_zilog_lock, USYNC_THREAD, NULL) == 0);
1250 mutex_init(&zd->zd_dirobj_lock, NULL, MUTEX_DEFAULT, NULL);
1252 for (l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1253 ztest_rll_init(&zd->zd_object_lock[l]);
1255 for (l = 0; l < ZTEST_RANGE_LOCKS; l++)
1256 ztest_rll_init(&zd->zd_range_lock[l]);
1260 ztest_zd_fini(ztest_ds_t *zd)
1264 mutex_destroy(&zd->zd_dirobj_lock);
1265 (void) rwlock_destroy(&zd->zd_zilog_lock);
1267 for (l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1268 ztest_rll_destroy(&zd->zd_object_lock[l]);
1270 for (l = 0; l < ZTEST_RANGE_LOCKS; l++)
1271 ztest_rll_destroy(&zd->zd_range_lock[l]);
1274 #define TXG_MIGHTWAIT (ztest_random(10) == 0 ? TXG_NOWAIT : TXG_WAIT)
1277 ztest_tx_assign(dmu_tx_t *tx, uint64_t txg_how, const char *tag)
1283 * Attempt to assign tx to some transaction group.
1285 error = dmu_tx_assign(tx, txg_how);
1287 if (error == ERESTART) {
1288 ASSERT(txg_how == TXG_NOWAIT);
1291 ASSERT3U(error, ==, ENOSPC);
1292 ztest_record_enospc(tag);
1297 txg = dmu_tx_get_txg(tx);
1303 ztest_pattern_set(void *buf, uint64_t size, uint64_t value)
1306 uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1314 ztest_pattern_match(void *buf, uint64_t size, uint64_t value)
1317 uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1321 diff |= (value - *ip++);
1328 ztest_bt_generate(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1329 uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg)
1331 bt->bt_magic = BT_MAGIC;
1332 bt->bt_objset = dmu_objset_id(os);
1333 bt->bt_object = object;
1334 bt->bt_offset = offset;
1337 bt->bt_crtxg = crtxg;
1341 ztest_bt_verify(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1342 uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg)
1344 ASSERT3U(bt->bt_magic, ==, BT_MAGIC);
1345 ASSERT3U(bt->bt_objset, ==, dmu_objset_id(os));
1346 ASSERT3U(bt->bt_object, ==, object);
1347 ASSERT3U(bt->bt_offset, ==, offset);
1348 ASSERT3U(bt->bt_gen, <=, gen);
1349 ASSERT3U(bt->bt_txg, <=, txg);
1350 ASSERT3U(bt->bt_crtxg, ==, crtxg);
1353 static ztest_block_tag_t *
1354 ztest_bt_bonus(dmu_buf_t *db)
1356 dmu_object_info_t doi;
1357 ztest_block_tag_t *bt;
1359 dmu_object_info_from_db(db, &doi);
1360 ASSERT3U(doi.doi_bonus_size, <=, db->db_size);
1361 ASSERT3U(doi.doi_bonus_size, >=, sizeof (*bt));
1362 bt = (void *)((char *)db->db_data + doi.doi_bonus_size - sizeof (*bt));
1371 #define lrz_type lr_mode
1372 #define lrz_blocksize lr_uid
1373 #define lrz_ibshift lr_gid
1374 #define lrz_bonustype lr_rdev
1375 #define lrz_bonuslen lr_crtime[1]
1378 ztest_log_create(ztest_ds_t *zd, dmu_tx_t *tx, lr_create_t *lr)
1380 char *name = (void *)(lr + 1); /* name follows lr */
1381 size_t namesize = strlen(name) + 1;
1384 if (zil_replaying(zd->zd_zilog, tx))
1387 itx = zil_itx_create(TX_CREATE, sizeof (*lr) + namesize);
1388 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1389 sizeof (*lr) + namesize - sizeof (lr_t));
1391 zil_itx_assign(zd->zd_zilog, itx, tx);
1395 ztest_log_remove(ztest_ds_t *zd, dmu_tx_t *tx, lr_remove_t *lr, uint64_t object)
1397 char *name = (void *)(lr + 1); /* name follows lr */
1398 size_t namesize = strlen(name) + 1;
1401 if (zil_replaying(zd->zd_zilog, tx))
1404 itx = zil_itx_create(TX_REMOVE, sizeof (*lr) + namesize);
1405 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1406 sizeof (*lr) + namesize - sizeof (lr_t));
1408 itx->itx_oid = object;
1409 zil_itx_assign(zd->zd_zilog, itx, tx);
1413 ztest_log_write(ztest_ds_t *zd, dmu_tx_t *tx, lr_write_t *lr)
1416 itx_wr_state_t write_state = ztest_random(WR_NUM_STATES);
1418 if (zil_replaying(zd->zd_zilog, tx))
1421 if (lr->lr_length > ZIL_MAX_LOG_DATA)
1422 write_state = WR_INDIRECT;
1424 itx = zil_itx_create(TX_WRITE,
1425 sizeof (*lr) + (write_state == WR_COPIED ? lr->lr_length : 0));
1427 if (write_state == WR_COPIED &&
1428 dmu_read(zd->zd_os, lr->lr_foid, lr->lr_offset, lr->lr_length,
1429 ((lr_write_t *)&itx->itx_lr) + 1, DMU_READ_NO_PREFETCH) != 0) {
1430 zil_itx_destroy(itx);
1431 itx = zil_itx_create(TX_WRITE, sizeof (*lr));
1432 write_state = WR_NEED_COPY;
1434 itx->itx_private = zd;
1435 itx->itx_wr_state = write_state;
1436 itx->itx_sync = (ztest_random(8) == 0);
1437 itx->itx_sod += (write_state == WR_NEED_COPY ? lr->lr_length : 0);
1439 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1440 sizeof (*lr) - sizeof (lr_t));
1442 zil_itx_assign(zd->zd_zilog, itx, tx);
1446 ztest_log_truncate(ztest_ds_t *zd, dmu_tx_t *tx, lr_truncate_t *lr)
1450 if (zil_replaying(zd->zd_zilog, tx))
1453 itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr));
1454 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1455 sizeof (*lr) - sizeof (lr_t));
1457 itx->itx_sync = B_FALSE;
1458 zil_itx_assign(zd->zd_zilog, itx, tx);
1462 ztest_log_setattr(ztest_ds_t *zd, dmu_tx_t *tx, lr_setattr_t *lr)
1466 if (zil_replaying(zd->zd_zilog, tx))
1469 itx = zil_itx_create(TX_SETATTR, sizeof (*lr));
1470 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1471 sizeof (*lr) - sizeof (lr_t));
1473 itx->itx_sync = B_FALSE;
1474 zil_itx_assign(zd->zd_zilog, itx, tx);
1481 ztest_replay_create(ztest_ds_t *zd, lr_create_t *lr, boolean_t byteswap)
1483 char *name = (void *)(lr + 1); /* name follows lr */
1484 objset_t *os = zd->zd_os;
1485 ztest_block_tag_t *bbt;
1492 byteswap_uint64_array(lr, sizeof (*lr));
1494 ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1495 ASSERT(name[0] != '\0');
1497 tx = dmu_tx_create(os);
1499 dmu_tx_hold_zap(tx, lr->lr_doid, B_TRUE, name);
1501 if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1502 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, B_TRUE, NULL);
1504 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1507 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1511 ASSERT(dmu_objset_zil(os)->zl_replay == !!lr->lr_foid);
1513 if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1514 if (lr->lr_foid == 0) {
1515 lr->lr_foid = zap_create(os,
1516 lr->lrz_type, lr->lrz_bonustype,
1517 lr->lrz_bonuslen, tx);
1519 error = zap_create_claim(os, lr->lr_foid,
1520 lr->lrz_type, lr->lrz_bonustype,
1521 lr->lrz_bonuslen, tx);
1524 if (lr->lr_foid == 0) {
1525 lr->lr_foid = dmu_object_alloc(os,
1526 lr->lrz_type, 0, lr->lrz_bonustype,
1527 lr->lrz_bonuslen, tx);
1529 error = dmu_object_claim(os, lr->lr_foid,
1530 lr->lrz_type, 0, lr->lrz_bonustype,
1531 lr->lrz_bonuslen, tx);
1536 ASSERT3U(error, ==, EEXIST);
1537 ASSERT(zd->zd_zilog->zl_replay);
1542 ASSERT(lr->lr_foid != 0);
1544 if (lr->lrz_type != DMU_OT_ZAP_OTHER)
1545 VERIFY3U(0, ==, dmu_object_set_blocksize(os, lr->lr_foid,
1546 lr->lrz_blocksize, lr->lrz_ibshift, tx));
1548 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1549 bbt = ztest_bt_bonus(db);
1550 dmu_buf_will_dirty(db, tx);
1551 ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_gen, txg, txg);
1552 dmu_buf_rele(db, FTAG);
1554 VERIFY3U(0, ==, zap_add(os, lr->lr_doid, name, sizeof (uint64_t), 1,
1557 (void) ztest_log_create(zd, tx, lr);
1565 ztest_replay_remove(ztest_ds_t *zd, lr_remove_t *lr, boolean_t byteswap)
1567 char *name = (void *)(lr + 1); /* name follows lr */
1568 objset_t *os = zd->zd_os;
1569 dmu_object_info_t doi;
1571 uint64_t object, txg;
1574 byteswap_uint64_array(lr, sizeof (*lr));
1576 ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1577 ASSERT(name[0] != '\0');
1580 zap_lookup(os, lr->lr_doid, name, sizeof (object), 1, &object));
1581 ASSERT(object != 0);
1583 ztest_object_lock(zd, object, RL_WRITER);
1585 VERIFY3U(0, ==, dmu_object_info(os, object, &doi));
1587 tx = dmu_tx_create(os);
1589 dmu_tx_hold_zap(tx, lr->lr_doid, B_FALSE, name);
1590 dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END);
1592 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1594 ztest_object_unlock(zd, object);
1598 if (doi.doi_type == DMU_OT_ZAP_OTHER) {
1599 VERIFY3U(0, ==, zap_destroy(os, object, tx));
1601 VERIFY3U(0, ==, dmu_object_free(os, object, tx));
1604 VERIFY3U(0, ==, zap_remove(os, lr->lr_doid, name, tx));
1606 (void) ztest_log_remove(zd, tx, lr, object);
1610 ztest_object_unlock(zd, object);
1616 ztest_replay_write(ztest_ds_t *zd, lr_write_t *lr, boolean_t byteswap)
1618 objset_t *os = zd->zd_os;
1619 void *data = lr + 1; /* data follows lr */
1620 uint64_t offset, length;
1621 ztest_block_tag_t *bt = data;
1622 ztest_block_tag_t *bbt;
1623 uint64_t gen, txg, lrtxg, crtxg;
1624 dmu_object_info_t doi;
1627 arc_buf_t *abuf = NULL;
1631 byteswap_uint64_array(lr, sizeof (*lr));
1633 offset = lr->lr_offset;
1634 length = lr->lr_length;
1636 /* If it's a dmu_sync() block, write the whole block */
1637 if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
1638 uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
1639 if (length < blocksize) {
1640 offset -= offset % blocksize;
1645 if (bt->bt_magic == BSWAP_64(BT_MAGIC))
1646 byteswap_uint64_array(bt, sizeof (*bt));
1648 if (bt->bt_magic != BT_MAGIC)
1651 ztest_object_lock(zd, lr->lr_foid, RL_READER);
1652 rl = ztest_range_lock(zd, lr->lr_foid, offset, length, RL_WRITER);
1654 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1656 dmu_object_info_from_db(db, &doi);
1658 bbt = ztest_bt_bonus(db);
1659 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1661 crtxg = bbt->bt_crtxg;
1662 lrtxg = lr->lr_common.lrc_txg;
1664 tx = dmu_tx_create(os);
1666 dmu_tx_hold_write(tx, lr->lr_foid, offset, length);
1668 if (ztest_random(8) == 0 && length == doi.doi_data_block_size &&
1669 P2PHASE(offset, length) == 0)
1670 abuf = dmu_request_arcbuf(db, length);
1672 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1675 dmu_return_arcbuf(abuf);
1676 dmu_buf_rele(db, FTAG);
1677 ztest_range_unlock(rl);
1678 ztest_object_unlock(zd, lr->lr_foid);
1684 * Usually, verify the old data before writing new data --
1685 * but not always, because we also want to verify correct
1686 * behavior when the data was not recently read into cache.
1688 ASSERT(offset % doi.doi_data_block_size == 0);
1689 if (ztest_random(4) != 0) {
1690 int prefetch = ztest_random(2) ?
1691 DMU_READ_PREFETCH : DMU_READ_NO_PREFETCH;
1692 ztest_block_tag_t rbt;
1694 VERIFY(dmu_read(os, lr->lr_foid, offset,
1695 sizeof (rbt), &rbt, prefetch) == 0);
1696 if (rbt.bt_magic == BT_MAGIC) {
1697 ztest_bt_verify(&rbt, os, lr->lr_foid,
1698 offset, gen, txg, crtxg);
1703 * Writes can appear to be newer than the bonus buffer because
1704 * the ztest_get_data() callback does a dmu_read() of the
1705 * open-context data, which may be different than the data
1706 * as it was when the write was generated.
1708 if (zd->zd_zilog->zl_replay) {
1709 ztest_bt_verify(bt, os, lr->lr_foid, offset,
1710 MAX(gen, bt->bt_gen), MAX(txg, lrtxg),
1715 * Set the bt's gen/txg to the bonus buffer's gen/txg
1716 * so that all of the usual ASSERTs will work.
1718 ztest_bt_generate(bt, os, lr->lr_foid, offset, gen, txg, crtxg);
1722 dmu_write(os, lr->lr_foid, offset, length, data, tx);
1724 bcopy(data, abuf->b_data, length);
1725 dmu_assign_arcbuf(db, offset, abuf, tx);
1728 (void) ztest_log_write(zd, tx, lr);
1730 dmu_buf_rele(db, FTAG);
1734 ztest_range_unlock(rl);
1735 ztest_object_unlock(zd, lr->lr_foid);
1741 ztest_replay_truncate(ztest_ds_t *zd, lr_truncate_t *lr, boolean_t byteswap)
1743 objset_t *os = zd->zd_os;
1749 byteswap_uint64_array(lr, sizeof (*lr));
1751 ztest_object_lock(zd, lr->lr_foid, RL_READER);
1752 rl = ztest_range_lock(zd, lr->lr_foid, lr->lr_offset, lr->lr_length,
1755 tx = dmu_tx_create(os);
1757 dmu_tx_hold_free(tx, lr->lr_foid, lr->lr_offset, lr->lr_length);
1759 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1761 ztest_range_unlock(rl);
1762 ztest_object_unlock(zd, lr->lr_foid);
1766 VERIFY(dmu_free_range(os, lr->lr_foid, lr->lr_offset,
1767 lr->lr_length, tx) == 0);
1769 (void) ztest_log_truncate(zd, tx, lr);
1773 ztest_range_unlock(rl);
1774 ztest_object_unlock(zd, lr->lr_foid);
1780 ztest_replay_setattr(ztest_ds_t *zd, lr_setattr_t *lr, boolean_t byteswap)
1782 objset_t *os = zd->zd_os;
1785 ztest_block_tag_t *bbt;
1786 uint64_t txg, lrtxg, crtxg;
1789 byteswap_uint64_array(lr, sizeof (*lr));
1791 ztest_object_lock(zd, lr->lr_foid, RL_WRITER);
1793 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1795 tx = dmu_tx_create(os);
1796 dmu_tx_hold_bonus(tx, lr->lr_foid);
1798 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1800 dmu_buf_rele(db, FTAG);
1801 ztest_object_unlock(zd, lr->lr_foid);
1805 bbt = ztest_bt_bonus(db);
1806 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1807 crtxg = bbt->bt_crtxg;
1808 lrtxg = lr->lr_common.lrc_txg;
1810 if (zd->zd_zilog->zl_replay) {
1811 ASSERT(lr->lr_size != 0);
1812 ASSERT(lr->lr_mode != 0);
1816 * Randomly change the size and increment the generation.
1818 lr->lr_size = (ztest_random(db->db_size / sizeof (*bbt)) + 1) *
1820 lr->lr_mode = bbt->bt_gen + 1;
1825 * Verify that the current bonus buffer is not newer than our txg.
1827 ztest_bt_verify(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode,
1828 MAX(txg, lrtxg), crtxg);
1830 dmu_buf_will_dirty(db, tx);
1832 ASSERT3U(lr->lr_size, >=, sizeof (*bbt));
1833 ASSERT3U(lr->lr_size, <=, db->db_size);
1834 VERIFY0(dmu_set_bonus(db, lr->lr_size, tx));
1835 bbt = ztest_bt_bonus(db);
1837 ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode, txg, crtxg);
1839 dmu_buf_rele(db, FTAG);
1841 (void) ztest_log_setattr(zd, tx, lr);
1845 ztest_object_unlock(zd, lr->lr_foid);
1850 zil_replay_func_t ztest_replay_vector[TX_MAX_TYPE] = {
1851 NULL, /* 0 no such transaction type */
1852 (zil_replay_func_t)ztest_replay_create, /* TX_CREATE */
1853 NULL, /* TX_MKDIR */
1854 NULL, /* TX_MKXATTR */
1855 NULL, /* TX_SYMLINK */
1856 (zil_replay_func_t)ztest_replay_remove, /* TX_REMOVE */
1857 NULL, /* TX_RMDIR */
1859 NULL, /* TX_RENAME */
1860 (zil_replay_func_t)ztest_replay_write, /* TX_WRITE */
1861 (zil_replay_func_t)ztest_replay_truncate, /* TX_TRUNCATE */
1862 (zil_replay_func_t)ztest_replay_setattr, /* TX_SETATTR */
1864 NULL, /* TX_CREATE_ACL */
1865 NULL, /* TX_CREATE_ATTR */
1866 NULL, /* TX_CREATE_ACL_ATTR */
1867 NULL, /* TX_MKDIR_ACL */
1868 NULL, /* TX_MKDIR_ATTR */
1869 NULL, /* TX_MKDIR_ACL_ATTR */
1870 NULL, /* TX_WRITE2 */
1874 * ZIL get_data callbacks
1878 ztest_get_done(zgd_t *zgd, int error)
1880 ztest_ds_t *zd = zgd->zgd_private;
1881 uint64_t object = zgd->zgd_rl->rl_object;
1884 dmu_buf_rele(zgd->zgd_db, zgd);
1886 ztest_range_unlock(zgd->zgd_rl);
1887 ztest_object_unlock(zd, object);
1889 if (error == 0 && zgd->zgd_bp)
1890 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
1892 umem_free(zgd, sizeof (*zgd));
1896 ztest_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
1898 ztest_ds_t *zd = arg;
1899 objset_t *os = zd->zd_os;
1900 uint64_t object = lr->lr_foid;
1901 uint64_t offset = lr->lr_offset;
1902 uint64_t size = lr->lr_length;
1903 blkptr_t *bp = &lr->lr_blkptr;
1904 uint64_t txg = lr->lr_common.lrc_txg;
1906 dmu_object_info_t doi;
1911 ztest_object_lock(zd, object, RL_READER);
1912 error = dmu_bonus_hold(os, object, FTAG, &db);
1914 ztest_object_unlock(zd, object);
1918 crtxg = ztest_bt_bonus(db)->bt_crtxg;
1920 if (crtxg == 0 || crtxg > txg) {
1921 dmu_buf_rele(db, FTAG);
1922 ztest_object_unlock(zd, object);
1926 dmu_object_info_from_db(db, &doi);
1927 dmu_buf_rele(db, FTAG);
1930 zgd = umem_zalloc(sizeof (*zgd), UMEM_NOFAIL);
1931 zgd->zgd_zilog = zd->zd_zilog;
1932 zgd->zgd_private = zd;
1934 if (buf != NULL) { /* immediate write */
1935 zgd->zgd_rl = ztest_range_lock(zd, object, offset, size,
1938 error = dmu_read(os, object, offset, size, buf,
1939 DMU_READ_NO_PREFETCH);
1942 size = doi.doi_data_block_size;
1944 offset = P2ALIGN(offset, size);
1946 ASSERT(offset < size);
1950 zgd->zgd_rl = ztest_range_lock(zd, object, offset, size,
1953 error = dmu_buf_hold(os, object, offset, zgd, &db,
1954 DMU_READ_NO_PREFETCH);
1957 blkptr_t *obp = dmu_buf_get_blkptr(db);
1959 ASSERT(BP_IS_HOLE(bp));
1966 ASSERT(db->db_offset == offset);
1967 ASSERT(db->db_size == size);
1969 error = dmu_sync(zio, lr->lr_common.lrc_txg,
1970 ztest_get_done, zgd);
1977 ztest_get_done(zgd, error);
1983 ztest_lr_alloc(size_t lrsize, char *name)
1986 size_t namesize = name ? strlen(name) + 1 : 0;
1988 lr = umem_zalloc(lrsize + namesize, UMEM_NOFAIL);
1991 bcopy(name, lr + lrsize, namesize);
1997 ztest_lr_free(void *lr, size_t lrsize, char *name)
1999 size_t namesize = name ? strlen(name) + 1 : 0;
2001 umem_free(lr, lrsize + namesize);
2005 * Lookup a bunch of objects. Returns the number of objects not found.
2008 ztest_lookup(ztest_ds_t *zd, ztest_od_t *od, int count)
2014 ASSERT(mutex_held(&zd->zd_dirobj_lock));
2016 for (i = 0; i < count; i++, od++) {
2018 error = zap_lookup(zd->zd_os, od->od_dir, od->od_name,
2019 sizeof (uint64_t), 1, &od->od_object);
2021 ASSERT(error == ENOENT);
2022 ASSERT(od->od_object == 0);
2026 ztest_block_tag_t *bbt;
2027 dmu_object_info_t doi;
2029 ASSERT(od->od_object != 0);
2030 ASSERT(missing == 0); /* there should be no gaps */
2032 ztest_object_lock(zd, od->od_object, RL_READER);
2033 VERIFY3U(0, ==, dmu_bonus_hold(zd->zd_os,
2034 od->od_object, FTAG, &db));
2035 dmu_object_info_from_db(db, &doi);
2036 bbt = ztest_bt_bonus(db);
2037 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
2038 od->od_type = doi.doi_type;
2039 od->od_blocksize = doi.doi_data_block_size;
2040 od->od_gen = bbt->bt_gen;
2041 dmu_buf_rele(db, FTAG);
2042 ztest_object_unlock(zd, od->od_object);
2050 ztest_create(ztest_ds_t *zd, ztest_od_t *od, int count)
2055 ASSERT(mutex_held(&zd->zd_dirobj_lock));
2057 for (i = 0; i < count; i++, od++) {
2064 lr_create_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
2066 lr->lr_doid = od->od_dir;
2067 lr->lr_foid = 0; /* 0 to allocate, > 0 to claim */
2068 lr->lrz_type = od->od_crtype;
2069 lr->lrz_blocksize = od->od_crblocksize;
2070 lr->lrz_ibshift = ztest_random_ibshift();
2071 lr->lrz_bonustype = DMU_OT_UINT64_OTHER;
2072 lr->lrz_bonuslen = dmu_bonus_max();
2073 lr->lr_gen = od->od_crgen;
2074 lr->lr_crtime[0] = time(NULL);
2076 if (ztest_replay_create(zd, lr, B_FALSE) != 0) {
2077 ASSERT(missing == 0);
2081 od->od_object = lr->lr_foid;
2082 od->od_type = od->od_crtype;
2083 od->od_blocksize = od->od_crblocksize;
2084 od->od_gen = od->od_crgen;
2085 ASSERT(od->od_object != 0);
2088 ztest_lr_free(lr, sizeof (*lr), od->od_name);
2095 ztest_remove(ztest_ds_t *zd, ztest_od_t *od, int count)
2101 ASSERT(mutex_held(&zd->zd_dirobj_lock));
2105 for (i = count - 1; i >= 0; i--, od--) {
2112 * No object was found.
2114 if (od->od_object == 0)
2117 lr_remove_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
2119 lr->lr_doid = od->od_dir;
2121 if ((error = ztest_replay_remove(zd, lr, B_FALSE)) != 0) {
2122 ASSERT3U(error, ==, ENOSPC);
2127 ztest_lr_free(lr, sizeof (*lr), od->od_name);
2134 ztest_write(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size,
2140 lr = ztest_lr_alloc(sizeof (*lr) + size, NULL);
2142 lr->lr_foid = object;
2143 lr->lr_offset = offset;
2144 lr->lr_length = size;
2146 BP_ZERO(&lr->lr_blkptr);
2148 bcopy(data, lr + 1, size);
2150 error = ztest_replay_write(zd, lr, B_FALSE);
2152 ztest_lr_free(lr, sizeof (*lr) + size, NULL);
2158 ztest_truncate(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2163 lr = ztest_lr_alloc(sizeof (*lr), NULL);
2165 lr->lr_foid = object;
2166 lr->lr_offset = offset;
2167 lr->lr_length = size;
2169 error = ztest_replay_truncate(zd, lr, B_FALSE);
2171 ztest_lr_free(lr, sizeof (*lr), NULL);
2177 ztest_setattr(ztest_ds_t *zd, uint64_t object)
2182 lr = ztest_lr_alloc(sizeof (*lr), NULL);
2184 lr->lr_foid = object;
2188 error = ztest_replay_setattr(zd, lr, B_FALSE);
2190 ztest_lr_free(lr, sizeof (*lr), NULL);
2196 ztest_prealloc(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2198 objset_t *os = zd->zd_os;
2203 txg_wait_synced(dmu_objset_pool(os), 0);
2205 ztest_object_lock(zd, object, RL_READER);
2206 rl = ztest_range_lock(zd, object, offset, size, RL_WRITER);
2208 tx = dmu_tx_create(os);
2210 dmu_tx_hold_write(tx, object, offset, size);
2212 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
2215 dmu_prealloc(os, object, offset, size, tx);
2217 txg_wait_synced(dmu_objset_pool(os), txg);
2219 (void) dmu_free_long_range(os, object, offset, size);
2222 ztest_range_unlock(rl);
2223 ztest_object_unlock(zd, object);
2227 ztest_io(ztest_ds_t *zd, uint64_t object, uint64_t offset)
2230 ztest_block_tag_t wbt;
2231 dmu_object_info_t doi;
2232 enum ztest_io_type io_type;
2236 VERIFY(dmu_object_info(zd->zd_os, object, &doi) == 0);
2237 blocksize = doi.doi_data_block_size;
2238 data = umem_alloc(blocksize, UMEM_NOFAIL);
2241 * Pick an i/o type at random, biased toward writing block tags.
2243 io_type = ztest_random(ZTEST_IO_TYPES);
2244 if (ztest_random(2) == 0)
2245 io_type = ZTEST_IO_WRITE_TAG;
2247 (void) rw_rdlock(&zd->zd_zilog_lock);
2251 case ZTEST_IO_WRITE_TAG:
2252 ztest_bt_generate(&wbt, zd->zd_os, object, offset, 0, 0, 0);
2253 (void) ztest_write(zd, object, offset, sizeof (wbt), &wbt);
2256 case ZTEST_IO_WRITE_PATTERN:
2257 (void) memset(data, 'a' + (object + offset) % 5, blocksize);
2258 if (ztest_random(2) == 0) {
2260 * Induce fletcher2 collisions to ensure that
2261 * zio_ddt_collision() detects and resolves them
2262 * when using fletcher2-verify for deduplication.
2264 ((uint64_t *)data)[0] ^= 1ULL << 63;
2265 ((uint64_t *)data)[4] ^= 1ULL << 63;
2267 (void) ztest_write(zd, object, offset, blocksize, data);
2270 case ZTEST_IO_WRITE_ZEROES:
2271 bzero(data, blocksize);
2272 (void) ztest_write(zd, object, offset, blocksize, data);
2275 case ZTEST_IO_TRUNCATE:
2276 (void) ztest_truncate(zd, object, offset, blocksize);
2279 case ZTEST_IO_SETATTR:
2280 (void) ztest_setattr(zd, object);
2285 case ZTEST_IO_REWRITE:
2286 (void) rw_rdlock(&ztest_name_lock);
2287 err = ztest_dsl_prop_set_uint64(zd->zd_name,
2288 ZFS_PROP_CHECKSUM, spa_dedup_checksum(ztest_spa),
2290 VERIFY(err == 0 || err == ENOSPC);
2291 err = ztest_dsl_prop_set_uint64(zd->zd_name,
2292 ZFS_PROP_COMPRESSION,
2293 ztest_random_dsl_prop(ZFS_PROP_COMPRESSION),
2295 VERIFY(err == 0 || err == ENOSPC);
2296 (void) rw_unlock(&ztest_name_lock);
2298 VERIFY0(dmu_read(zd->zd_os, object, offset, blocksize, data,
2299 DMU_READ_NO_PREFETCH));
2301 (void) ztest_write(zd, object, offset, blocksize, data);
2305 (void) rw_unlock(&zd->zd_zilog_lock);
2307 umem_free(data, blocksize);
2311 * Initialize an object description template.
2314 ztest_od_init(ztest_od_t *od, uint64_t id, char *tag, uint64_t index,
2315 dmu_object_type_t type, uint64_t blocksize, uint64_t gen)
2317 od->od_dir = ZTEST_DIROBJ;
2320 od->od_crtype = type;
2321 od->od_crblocksize = blocksize ? blocksize : ztest_random_blocksize();
2324 od->od_type = DMU_OT_NONE;
2325 od->od_blocksize = 0;
2328 (void) snprintf(od->od_name, sizeof (od->od_name), "%s(%lld)[%llu]",
2329 tag, (longlong_t)id, (u_longlong_t)index);
2333 * Lookup or create the objects for a test using the od template.
2334 * If the objects do not all exist, or if 'remove' is specified,
2335 * remove any existing objects and create new ones. Otherwise,
2336 * use the existing objects.
2339 ztest_object_init(ztest_ds_t *zd, ztest_od_t *od, size_t size, boolean_t remove)
2341 int count = size / sizeof (*od);
2344 mutex_enter(&zd->zd_dirobj_lock);
2345 if ((ztest_lookup(zd, od, count) != 0 || remove) &&
2346 (ztest_remove(zd, od, count) != 0 ||
2347 ztest_create(zd, od, count) != 0))
2350 mutex_exit(&zd->zd_dirobj_lock);
2357 ztest_zil_commit(ztest_ds_t *zd, uint64_t id)
2359 zilog_t *zilog = zd->zd_zilog;
2361 (void) rw_rdlock(&zd->zd_zilog_lock);
2363 zil_commit(zilog, ztest_random(ZTEST_OBJECTS));
2366 * Remember the committed values in zd, which is in parent/child
2367 * shared memory. If we die, the next iteration of ztest_run()
2368 * will verify that the log really does contain this record.
2370 mutex_enter(&zilog->zl_lock);
2371 ASSERT(zd->zd_shared != NULL);
2372 ASSERT3U(zd->zd_shared->zd_seq, <=, zilog->zl_commit_lr_seq);
2373 zd->zd_shared->zd_seq = zilog->zl_commit_lr_seq;
2374 mutex_exit(&zilog->zl_lock);
2376 (void) rw_unlock(&zd->zd_zilog_lock);
2380 * This function is designed to simulate the operations that occur during a
2381 * mount/unmount operation. We hold the dataset across these operations in an
2382 * attempt to expose any implicit assumptions about ZIL management.
2386 ztest_zil_remount(ztest_ds_t *zd, uint64_t id)
2388 objset_t *os = zd->zd_os;
2391 * We grab the zd_dirobj_lock to ensure that no other thread is
2392 * updating the zil (i.e. adding in-memory log records) and the
2393 * zd_zilog_lock to block any I/O.
2395 mutex_enter(&zd->zd_dirobj_lock);
2396 (void) rw_wrlock(&zd->zd_zilog_lock);
2398 /* zfs_sb_teardown() */
2399 zil_close(zd->zd_zilog);
2401 /* zfsvfs_setup() */
2402 VERIFY(zil_open(os, ztest_get_data) == zd->zd_zilog);
2403 zil_replay(os, zd, ztest_replay_vector);
2405 (void) rw_unlock(&zd->zd_zilog_lock);
2406 mutex_exit(&zd->zd_dirobj_lock);
2410 * Verify that we can't destroy an active pool, create an existing pool,
2411 * or create a pool with a bad vdev spec.
2415 ztest_spa_create_destroy(ztest_ds_t *zd, uint64_t id)
2417 ztest_shared_opts_t *zo = &ztest_opts;
2422 * Attempt to create using a bad file.
2424 nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1);
2425 VERIFY3U(ENOENT, ==,
2426 spa_create("ztest_bad_file", nvroot, NULL, NULL));
2427 nvlist_free(nvroot);
2430 * Attempt to create using a bad mirror.
2432 nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 2, 1);
2433 VERIFY3U(ENOENT, ==,
2434 spa_create("ztest_bad_mirror", nvroot, NULL, NULL));
2435 nvlist_free(nvroot);
2438 * Attempt to create an existing pool. It shouldn't matter
2439 * what's in the nvroot; we should fail with EEXIST.
2441 (void) rw_rdlock(&ztest_name_lock);
2442 nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1);
2443 VERIFY3U(EEXIST, ==, spa_create(zo->zo_pool, nvroot, NULL, NULL));
2444 nvlist_free(nvroot);
2445 VERIFY3U(0, ==, spa_open(zo->zo_pool, &spa, FTAG));
2446 VERIFY3U(EBUSY, ==, spa_destroy(zo->zo_pool));
2447 spa_close(spa, FTAG);
2449 (void) rw_unlock(&ztest_name_lock);
2454 ztest_spa_upgrade(ztest_ds_t *zd, uint64_t id)
2457 uint64_t initial_version = SPA_VERSION_INITIAL;
2458 uint64_t version, newversion;
2459 nvlist_t *nvroot, *props;
2462 mutex_enter(&ztest_vdev_lock);
2463 name = kmem_asprintf("%s_upgrade", ztest_opts.zo_pool);
2466 * Clean up from previous runs.
2468 (void) spa_destroy(name);
2470 nvroot = make_vdev_root(NULL, NULL, name, ztest_opts.zo_vdev_size, 0,
2471 0, ztest_opts.zo_raidz, ztest_opts.zo_mirrors, 1);
2474 * If we're configuring a RAIDZ device then make sure that the
2475 * the initial version is capable of supporting that feature.
2477 switch (ztest_opts.zo_raidz_parity) {
2480 initial_version = SPA_VERSION_INITIAL;
2483 initial_version = SPA_VERSION_RAIDZ2;
2486 initial_version = SPA_VERSION_RAIDZ3;
2491 * Create a pool with a spa version that can be upgraded. Pick
2492 * a value between initial_version and SPA_VERSION_BEFORE_FEATURES.
2495 version = ztest_random_spa_version(initial_version);
2496 } while (version > SPA_VERSION_BEFORE_FEATURES);
2498 props = fnvlist_alloc();
2499 fnvlist_add_uint64(props,
2500 zpool_prop_to_name(ZPOOL_PROP_VERSION), version);
2501 VERIFY3S(spa_create(name, nvroot, props, NULL), ==, 0);
2502 fnvlist_free(nvroot);
2503 fnvlist_free(props);
2505 VERIFY3S(spa_open(name, &spa, FTAG), ==, 0);
2506 VERIFY3U(spa_version(spa), ==, version);
2507 newversion = ztest_random_spa_version(version + 1);
2509 if (ztest_opts.zo_verbose >= 4) {
2510 (void) printf("upgrading spa version from %llu to %llu\n",
2511 (u_longlong_t)version, (u_longlong_t)newversion);
2514 spa_upgrade(spa, newversion);
2515 VERIFY3U(spa_version(spa), >, version);
2516 VERIFY3U(spa_version(spa), ==, fnvlist_lookup_uint64(spa->spa_config,
2517 zpool_prop_to_name(ZPOOL_PROP_VERSION)));
2518 spa_close(spa, FTAG);
2521 mutex_exit(&ztest_vdev_lock);
2525 vdev_lookup_by_path(vdev_t *vd, const char *path)
2530 if (vd->vdev_path != NULL && strcmp(path, vd->vdev_path) == 0)
2533 for (c = 0; c < vd->vdev_children; c++)
2534 if ((mvd = vdev_lookup_by_path(vd->vdev_child[c], path)) !=
2542 * Find the first available hole which can be used as a top-level.
2545 find_vdev_hole(spa_t *spa)
2547 vdev_t *rvd = spa->spa_root_vdev;
2550 ASSERT(spa_config_held(spa, SCL_VDEV, RW_READER) == SCL_VDEV);
2552 for (c = 0; c < rvd->vdev_children; c++) {
2553 vdev_t *cvd = rvd->vdev_child[c];
2555 if (cvd->vdev_ishole)
2562 * Verify that vdev_add() works as expected.
2566 ztest_vdev_add_remove(ztest_ds_t *zd, uint64_t id)
2568 ztest_shared_t *zs = ztest_shared;
2569 spa_t *spa = ztest_spa;
2575 mutex_enter(&ztest_vdev_lock);
2576 leaves = MAX(zs->zs_mirrors + zs->zs_splits, 1) * ztest_opts.zo_raidz;
2578 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2580 ztest_shared->zs_vdev_next_leaf = find_vdev_hole(spa) * leaves;
2583 * If we have slogs then remove them 1/4 of the time.
2585 if (spa_has_slogs(spa) && ztest_random(4) == 0) {
2587 * Grab the guid from the head of the log class rotor.
2589 guid = spa_log_class(spa)->mc_rotor->mg_vd->vdev_guid;
2591 spa_config_exit(spa, SCL_VDEV, FTAG);
2594 * We have to grab the zs_name_lock as writer to
2595 * prevent a race between removing a slog (dmu_objset_find)
2596 * and destroying a dataset. Removing the slog will
2597 * grab a reference on the dataset which may cause
2598 * dsl_destroy_head() to fail with EBUSY thus
2599 * leaving the dataset in an inconsistent state.
2601 rw_wrlock(&ztest_name_lock);
2602 error = spa_vdev_remove(spa, guid, B_FALSE);
2603 rw_unlock(&ztest_name_lock);
2605 if (error && error != EEXIST)
2606 fatal(0, "spa_vdev_remove() = %d", error);
2608 spa_config_exit(spa, SCL_VDEV, FTAG);
2611 * Make 1/4 of the devices be log devices.
2613 nvroot = make_vdev_root(NULL, NULL, NULL,
2614 ztest_opts.zo_vdev_size, 0,
2615 ztest_random(4) == 0, ztest_opts.zo_raidz,
2618 error = spa_vdev_add(spa, nvroot);
2619 nvlist_free(nvroot);
2621 if (error == ENOSPC)
2622 ztest_record_enospc("spa_vdev_add");
2623 else if (error != 0)
2624 fatal(0, "spa_vdev_add() = %d", error);
2627 mutex_exit(&ztest_vdev_lock);
2631 * Verify that adding/removing aux devices (l2arc, hot spare) works as expected.
2635 ztest_vdev_aux_add_remove(ztest_ds_t *zd, uint64_t id)
2637 ztest_shared_t *zs = ztest_shared;
2638 spa_t *spa = ztest_spa;
2639 vdev_t *rvd = spa->spa_root_vdev;
2640 spa_aux_vdev_t *sav;
2646 path = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
2648 if (ztest_random(2) == 0) {
2649 sav = &spa->spa_spares;
2650 aux = ZPOOL_CONFIG_SPARES;
2652 sav = &spa->spa_l2cache;
2653 aux = ZPOOL_CONFIG_L2CACHE;
2656 mutex_enter(&ztest_vdev_lock);
2658 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2660 if (sav->sav_count != 0 && ztest_random(4) == 0) {
2662 * Pick a random device to remove.
2664 guid = sav->sav_vdevs[ztest_random(sav->sav_count)]->vdev_guid;
2667 * Find an unused device we can add.
2669 zs->zs_vdev_aux = 0;
2672 (void) snprintf(path, MAXPATHLEN, ztest_aux_template,
2673 ztest_opts.zo_dir, ztest_opts.zo_pool, aux,
2675 for (c = 0; c < sav->sav_count; c++)
2676 if (strcmp(sav->sav_vdevs[c]->vdev_path,
2679 if (c == sav->sav_count &&
2680 vdev_lookup_by_path(rvd, path) == NULL)
2686 spa_config_exit(spa, SCL_VDEV, FTAG);
2692 nvlist_t *nvroot = make_vdev_root(NULL, aux, NULL,
2693 (ztest_opts.zo_vdev_size * 5) / 4, 0, 0, 0, 0, 1);
2694 error = spa_vdev_add(spa, nvroot);
2696 fatal(0, "spa_vdev_add(%p) = %d", nvroot, error);
2697 nvlist_free(nvroot);
2700 * Remove an existing device. Sometimes, dirty its
2701 * vdev state first to make sure we handle removal
2702 * of devices that have pending state changes.
2704 if (ztest_random(2) == 0)
2705 (void) vdev_online(spa, guid, 0, NULL);
2707 error = spa_vdev_remove(spa, guid, B_FALSE);
2708 if (error != 0 && error != EBUSY)
2709 fatal(0, "spa_vdev_remove(%llu) = %d", guid, error);
2712 mutex_exit(&ztest_vdev_lock);
2714 umem_free(path, MAXPATHLEN);
2718 * split a pool if it has mirror tlvdevs
2722 ztest_split_pool(ztest_ds_t *zd, uint64_t id)
2724 ztest_shared_t *zs = ztest_shared;
2725 spa_t *spa = ztest_spa;
2726 vdev_t *rvd = spa->spa_root_vdev;
2727 nvlist_t *tree, **child, *config, *split, **schild;
2728 uint_t c, children, schildren = 0, lastlogid = 0;
2731 mutex_enter(&ztest_vdev_lock);
2733 /* ensure we have a useable config; mirrors of raidz aren't supported */
2734 if (zs->zs_mirrors < 3 || ztest_opts.zo_raidz > 1) {
2735 mutex_exit(&ztest_vdev_lock);
2739 /* clean up the old pool, if any */
2740 (void) spa_destroy("splitp");
2742 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2744 /* generate a config from the existing config */
2745 mutex_enter(&spa->spa_props_lock);
2746 VERIFY(nvlist_lookup_nvlist(spa->spa_config, ZPOOL_CONFIG_VDEV_TREE,
2748 mutex_exit(&spa->spa_props_lock);
2750 VERIFY(nvlist_lookup_nvlist_array(tree, ZPOOL_CONFIG_CHILDREN, &child,
2753 schild = malloc(rvd->vdev_children * sizeof (nvlist_t *));
2754 for (c = 0; c < children; c++) {
2755 vdev_t *tvd = rvd->vdev_child[c];
2759 if (tvd->vdev_islog || tvd->vdev_ops == &vdev_hole_ops) {
2760 VERIFY(nvlist_alloc(&schild[schildren], NV_UNIQUE_NAME,
2762 VERIFY(nvlist_add_string(schild[schildren],
2763 ZPOOL_CONFIG_TYPE, VDEV_TYPE_HOLE) == 0);
2764 VERIFY(nvlist_add_uint64(schild[schildren],
2765 ZPOOL_CONFIG_IS_HOLE, 1) == 0);
2767 lastlogid = schildren;
2772 VERIFY(nvlist_lookup_nvlist_array(child[c],
2773 ZPOOL_CONFIG_CHILDREN, &mchild, &mchildren) == 0);
2774 VERIFY(nvlist_dup(mchild[0], &schild[schildren++], 0) == 0);
2777 /* OK, create a config that can be used to split */
2778 VERIFY(nvlist_alloc(&split, NV_UNIQUE_NAME, 0) == 0);
2779 VERIFY(nvlist_add_string(split, ZPOOL_CONFIG_TYPE,
2780 VDEV_TYPE_ROOT) == 0);
2781 VERIFY(nvlist_add_nvlist_array(split, ZPOOL_CONFIG_CHILDREN, schild,
2782 lastlogid != 0 ? lastlogid : schildren) == 0);
2784 VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, 0) == 0);
2785 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, split) == 0);
2787 for (c = 0; c < schildren; c++)
2788 nvlist_free(schild[c]);
2792 spa_config_exit(spa, SCL_VDEV, FTAG);
2794 (void) rw_wrlock(&ztest_name_lock);
2795 error = spa_vdev_split_mirror(spa, "splitp", config, NULL, B_FALSE);
2796 (void) rw_unlock(&ztest_name_lock);
2798 nvlist_free(config);
2801 (void) printf("successful split - results:\n");
2802 mutex_enter(&spa_namespace_lock);
2803 show_pool_stats(spa);
2804 show_pool_stats(spa_lookup("splitp"));
2805 mutex_exit(&spa_namespace_lock);
2809 mutex_exit(&ztest_vdev_lock);
2814 * Verify that we can attach and detach devices.
2818 ztest_vdev_attach_detach(ztest_ds_t *zd, uint64_t id)
2820 ztest_shared_t *zs = ztest_shared;
2821 spa_t *spa = ztest_spa;
2822 spa_aux_vdev_t *sav = &spa->spa_spares;
2823 vdev_t *rvd = spa->spa_root_vdev;
2824 vdev_t *oldvd, *newvd, *pvd;
2828 uint64_t ashift = ztest_get_ashift();
2829 uint64_t oldguid, pguid;
2830 uint64_t oldsize, newsize;
2831 char *oldpath, *newpath;
2833 int oldvd_has_siblings = B_FALSE;
2834 int newvd_is_spare = B_FALSE;
2836 int error, expected_error;
2838 oldpath = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
2839 newpath = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
2841 mutex_enter(&ztest_vdev_lock);
2842 leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz;
2844 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2847 * Decide whether to do an attach or a replace.
2849 replacing = ztest_random(2);
2852 * Pick a random top-level vdev.
2854 top = ztest_random_vdev_top(spa, B_TRUE);
2857 * Pick a random leaf within it.
2859 leaf = ztest_random(leaves);
2864 oldvd = rvd->vdev_child[top];
2865 if (zs->zs_mirrors >= 1) {
2866 ASSERT(oldvd->vdev_ops == &vdev_mirror_ops);
2867 ASSERT(oldvd->vdev_children >= zs->zs_mirrors);
2868 oldvd = oldvd->vdev_child[leaf / ztest_opts.zo_raidz];
2870 if (ztest_opts.zo_raidz > 1) {
2871 ASSERT(oldvd->vdev_ops == &vdev_raidz_ops);
2872 ASSERT(oldvd->vdev_children == ztest_opts.zo_raidz);
2873 oldvd = oldvd->vdev_child[leaf % ztest_opts.zo_raidz];
2877 * If we're already doing an attach or replace, oldvd may be a
2878 * mirror vdev -- in which case, pick a random child.
2880 while (oldvd->vdev_children != 0) {
2881 oldvd_has_siblings = B_TRUE;
2882 ASSERT(oldvd->vdev_children >= 2);
2883 oldvd = oldvd->vdev_child[ztest_random(oldvd->vdev_children)];
2886 oldguid = oldvd->vdev_guid;
2887 oldsize = vdev_get_min_asize(oldvd);
2888 oldvd_is_log = oldvd->vdev_top->vdev_islog;
2889 (void) strcpy(oldpath, oldvd->vdev_path);
2890 pvd = oldvd->vdev_parent;
2891 pguid = pvd->vdev_guid;
2894 * If oldvd has siblings, then half of the time, detach it.
2896 if (oldvd_has_siblings && ztest_random(2) == 0) {
2897 spa_config_exit(spa, SCL_VDEV, FTAG);
2898 error = spa_vdev_detach(spa, oldguid, pguid, B_FALSE);
2899 if (error != 0 && error != ENODEV && error != EBUSY &&
2901 fatal(0, "detach (%s) returned %d", oldpath, error);
2906 * For the new vdev, choose with equal probability between the two
2907 * standard paths (ending in either 'a' or 'b') or a random hot spare.
2909 if (sav->sav_count != 0 && ztest_random(3) == 0) {
2910 newvd = sav->sav_vdevs[ztest_random(sav->sav_count)];
2911 newvd_is_spare = B_TRUE;
2912 (void) strcpy(newpath, newvd->vdev_path);
2914 (void) snprintf(newpath, MAXPATHLEN, ztest_dev_template,
2915 ztest_opts.zo_dir, ztest_opts.zo_pool,
2916 top * leaves + leaf);
2917 if (ztest_random(2) == 0)
2918 newpath[strlen(newpath) - 1] = 'b';
2919 newvd = vdev_lookup_by_path(rvd, newpath);
2923 newsize = vdev_get_min_asize(newvd);
2926 * Make newsize a little bigger or smaller than oldsize.
2927 * If it's smaller, the attach should fail.
2928 * If it's larger, and we're doing a replace,
2929 * we should get dynamic LUN growth when we're done.
2931 newsize = 10 * oldsize / (9 + ztest_random(3));
2935 * If pvd is not a mirror or root, the attach should fail with ENOTSUP,
2936 * unless it's a replace; in that case any non-replacing parent is OK.
2938 * If newvd is already part of the pool, it should fail with EBUSY.
2940 * If newvd is too small, it should fail with EOVERFLOW.
2942 if (pvd->vdev_ops != &vdev_mirror_ops &&
2943 pvd->vdev_ops != &vdev_root_ops && (!replacing ||
2944 pvd->vdev_ops == &vdev_replacing_ops ||
2945 pvd->vdev_ops == &vdev_spare_ops))
2946 expected_error = ENOTSUP;
2947 else if (newvd_is_spare && (!replacing || oldvd_is_log))
2948 expected_error = ENOTSUP;
2949 else if (newvd == oldvd)
2950 expected_error = replacing ? 0 : EBUSY;
2951 else if (vdev_lookup_by_path(rvd, newpath) != NULL)
2952 expected_error = EBUSY;
2953 else if (newsize < oldsize)
2954 expected_error = EOVERFLOW;
2955 else if (ashift > oldvd->vdev_top->vdev_ashift)
2956 expected_error = EDOM;
2960 spa_config_exit(spa, SCL_VDEV, FTAG);
2963 * Build the nvlist describing newpath.
2965 root = make_vdev_root(newpath, NULL, NULL, newvd == NULL ? newsize : 0,
2966 ashift, 0, 0, 0, 1);
2968 error = spa_vdev_attach(spa, oldguid, root, replacing);
2973 * If our parent was the replacing vdev, but the replace completed,
2974 * then instead of failing with ENOTSUP we may either succeed,
2975 * fail with ENODEV, or fail with EOVERFLOW.
2977 if (expected_error == ENOTSUP &&
2978 (error == 0 || error == ENODEV || error == EOVERFLOW))
2979 expected_error = error;
2982 * If someone grew the LUN, the replacement may be too small.
2984 if (error == EOVERFLOW || error == EBUSY)
2985 expected_error = error;
2987 /* XXX workaround 6690467 */
2988 if (error != expected_error && expected_error != EBUSY) {
2989 fatal(0, "attach (%s %llu, %s %llu, %d) "
2990 "returned %d, expected %d",
2991 oldpath, oldsize, newpath,
2992 newsize, replacing, error, expected_error);
2995 mutex_exit(&ztest_vdev_lock);
2997 umem_free(oldpath, MAXPATHLEN);
2998 umem_free(newpath, MAXPATHLEN);
3002 * Callback function which expands the physical size of the vdev.
3005 grow_vdev(vdev_t *vd, void *arg)
3007 ASSERTV(spa_t *spa = vd->vdev_spa);
3008 size_t *newsize = arg;
3012 ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
3013 ASSERT(vd->vdev_ops->vdev_op_leaf);
3015 if ((fd = open(vd->vdev_path, O_RDWR)) == -1)
3018 fsize = lseek(fd, 0, SEEK_END);
3019 VERIFY(ftruncate(fd, *newsize) == 0);
3021 if (ztest_opts.zo_verbose >= 6) {
3022 (void) printf("%s grew from %lu to %lu bytes\n",
3023 vd->vdev_path, (ulong_t)fsize, (ulong_t)*newsize);
3030 * Callback function which expands a given vdev by calling vdev_online().
3034 online_vdev(vdev_t *vd, void *arg)
3036 spa_t *spa = vd->vdev_spa;
3037 vdev_t *tvd = vd->vdev_top;
3038 uint64_t guid = vd->vdev_guid;
3039 uint64_t generation = spa->spa_config_generation + 1;
3040 vdev_state_t newstate = VDEV_STATE_UNKNOWN;
3043 ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
3044 ASSERT(vd->vdev_ops->vdev_op_leaf);
3046 /* Calling vdev_online will initialize the new metaslabs */
3047 spa_config_exit(spa, SCL_STATE, spa);
3048 error = vdev_online(spa, guid, ZFS_ONLINE_EXPAND, &newstate);
3049 spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3052 * If vdev_online returned an error or the underlying vdev_open
3053 * failed then we abort the expand. The only way to know that
3054 * vdev_open fails is by checking the returned newstate.
3056 if (error || newstate != VDEV_STATE_HEALTHY) {
3057 if (ztest_opts.zo_verbose >= 5) {
3058 (void) printf("Unable to expand vdev, state %llu, "
3059 "error %d\n", (u_longlong_t)newstate, error);
3063 ASSERT3U(newstate, ==, VDEV_STATE_HEALTHY);
3066 * Since we dropped the lock we need to ensure that we're
3067 * still talking to the original vdev. It's possible this
3068 * vdev may have been detached/replaced while we were
3069 * trying to online it.
3071 if (generation != spa->spa_config_generation) {
3072 if (ztest_opts.zo_verbose >= 5) {
3073 (void) printf("vdev configuration has changed, "
3074 "guid %llu, state %llu, expected gen %llu, "
3077 (u_longlong_t)tvd->vdev_state,
3078 (u_longlong_t)generation,
3079 (u_longlong_t)spa->spa_config_generation);
3087 * Traverse the vdev tree calling the supplied function.
3088 * We continue to walk the tree until we either have walked all
3089 * children or we receive a non-NULL return from the callback.
3090 * If a NULL callback is passed, then we just return back the first
3091 * leaf vdev we encounter.
3094 vdev_walk_tree(vdev_t *vd, vdev_t *(*func)(vdev_t *, void *), void *arg)
3098 if (vd->vdev_ops->vdev_op_leaf) {
3102 return (func(vd, arg));
3105 for (c = 0; c < vd->vdev_children; c++) {
3106 vdev_t *cvd = vd->vdev_child[c];
3107 if ((cvd = vdev_walk_tree(cvd, func, arg)) != NULL)
3114 * Verify that dynamic LUN growth works as expected.
3118 ztest_vdev_LUN_growth(ztest_ds_t *zd, uint64_t id)
3120 spa_t *spa = ztest_spa;
3122 metaslab_class_t *mc;
3123 metaslab_group_t *mg;
3124 size_t psize, newsize;
3126 uint64_t old_class_space, new_class_space, old_ms_count, new_ms_count;
3128 mutex_enter(&ztest_vdev_lock);
3129 spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3131 top = ztest_random_vdev_top(spa, B_TRUE);
3133 tvd = spa->spa_root_vdev->vdev_child[top];
3136 old_ms_count = tvd->vdev_ms_count;
3137 old_class_space = metaslab_class_get_space(mc);
3140 * Determine the size of the first leaf vdev associated with
3141 * our top-level device.
3143 vd = vdev_walk_tree(tvd, NULL, NULL);
3144 ASSERT3P(vd, !=, NULL);
3145 ASSERT(vd->vdev_ops->vdev_op_leaf);
3147 psize = vd->vdev_psize;
3150 * We only try to expand the vdev if it's healthy, less than 4x its
3151 * original size, and it has a valid psize.
3153 if (tvd->vdev_state != VDEV_STATE_HEALTHY ||
3154 psize == 0 || psize >= 4 * ztest_opts.zo_vdev_size) {
3155 spa_config_exit(spa, SCL_STATE, spa);
3156 mutex_exit(&ztest_vdev_lock);
3160 newsize = psize + psize / 8;
3161 ASSERT3U(newsize, >, psize);
3163 if (ztest_opts.zo_verbose >= 6) {
3164 (void) printf("Expanding LUN %s from %lu to %lu\n",
3165 vd->vdev_path, (ulong_t)psize, (ulong_t)newsize);
3169 * Growing the vdev is a two step process:
3170 * 1). expand the physical size (i.e. relabel)
3171 * 2). online the vdev to create the new metaslabs
3173 if (vdev_walk_tree(tvd, grow_vdev, &newsize) != NULL ||
3174 vdev_walk_tree(tvd, online_vdev, NULL) != NULL ||
3175 tvd->vdev_state != VDEV_STATE_HEALTHY) {
3176 if (ztest_opts.zo_verbose >= 5) {
3177 (void) printf("Could not expand LUN because "
3178 "the vdev configuration changed.\n");
3180 spa_config_exit(spa, SCL_STATE, spa);
3181 mutex_exit(&ztest_vdev_lock);
3185 spa_config_exit(spa, SCL_STATE, spa);
3188 * Expanding the LUN will update the config asynchronously,
3189 * thus we must wait for the async thread to complete any
3190 * pending tasks before proceeding.
3194 mutex_enter(&spa->spa_async_lock);
3195 done = (spa->spa_async_thread == NULL && !spa->spa_async_tasks);
3196 mutex_exit(&spa->spa_async_lock);
3199 txg_wait_synced(spa_get_dsl(spa), 0);
3200 (void) poll(NULL, 0, 100);
3203 spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3205 tvd = spa->spa_root_vdev->vdev_child[top];
3206 new_ms_count = tvd->vdev_ms_count;
3207 new_class_space = metaslab_class_get_space(mc);
3209 if (tvd->vdev_mg != mg || mg->mg_class != mc) {
3210 if (ztest_opts.zo_verbose >= 5) {
3211 (void) printf("Could not verify LUN expansion due to "
3212 "intervening vdev offline or remove.\n");
3214 spa_config_exit(spa, SCL_STATE, spa);
3215 mutex_exit(&ztest_vdev_lock);
3220 * Make sure we were able to grow the vdev.
3222 if (new_ms_count <= old_ms_count)
3223 fatal(0, "LUN expansion failed: ms_count %llu <= %llu\n",
3224 old_ms_count, new_ms_count);
3227 * Make sure we were able to grow the pool.
3229 if (new_class_space <= old_class_space)
3230 fatal(0, "LUN expansion failed: class_space %llu <= %llu\n",
3231 old_class_space, new_class_space);
3233 if (ztest_opts.zo_verbose >= 5) {
3234 char oldnumbuf[6], newnumbuf[6];
3236 nicenum(old_class_space, oldnumbuf);
3237 nicenum(new_class_space, newnumbuf);
3238 (void) printf("%s grew from %s to %s\n",
3239 spa->spa_name, oldnumbuf, newnumbuf);
3242 spa_config_exit(spa, SCL_STATE, spa);
3243 mutex_exit(&ztest_vdev_lock);
3247 * Verify that dmu_objset_{create,destroy,open,close} work as expected.
3251 ztest_objset_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
3254 * Create the objects common to all ztest datasets.
3256 VERIFY(zap_create_claim(os, ZTEST_DIROBJ,
3257 DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx) == 0);
3261 ztest_dataset_create(char *dsname)
3263 uint64_t zilset = ztest_random(100);
3264 int err = dmu_objset_create(dsname, DMU_OST_OTHER, 0,
3265 ztest_objset_create_cb, NULL);
3267 if (err || zilset < 80)
3270 if (ztest_opts.zo_verbose >= 5)
3271 (void) printf("Setting dataset %s to sync always\n", dsname);
3272 return (ztest_dsl_prop_set_uint64(dsname, ZFS_PROP_SYNC,
3273 ZFS_SYNC_ALWAYS, B_FALSE));
3278 ztest_objset_destroy_cb(const char *name, void *arg)
3281 dmu_object_info_t doi;
3285 * Verify that the dataset contains a directory object.
3287 VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_TRUE, FTAG, &os));
3288 error = dmu_object_info(os, ZTEST_DIROBJ, &doi);
3289 if (error != ENOENT) {
3290 /* We could have crashed in the middle of destroying it */
3292 ASSERT3U(doi.doi_type, ==, DMU_OT_ZAP_OTHER);
3293 ASSERT3S(doi.doi_physical_blocks_512, >=, 0);
3295 dmu_objset_disown(os, FTAG);
3298 * Destroy the dataset.
3300 if (strchr(name, '@') != NULL) {
3301 VERIFY0(dsl_destroy_snapshot(name, B_FALSE));
3303 VERIFY0(dsl_destroy_head(name));
3309 ztest_snapshot_create(char *osname, uint64_t id)
3311 char snapname[MAXNAMELEN];
3314 (void) snprintf(snapname, sizeof (snapname), "%llu", (u_longlong_t)id);
3316 error = dmu_objset_snapshot_one(osname, snapname);
3317 if (error == ENOSPC) {
3318 ztest_record_enospc(FTAG);
3321 if (error != 0 && error != EEXIST) {
3322 fatal(0, "ztest_snapshot_create(%s@%s) = %d", osname,
3329 ztest_snapshot_destroy(char *osname, uint64_t id)
3331 char snapname[MAXNAMELEN];
3334 (void) snprintf(snapname, MAXNAMELEN, "%s@%llu", osname,
3337 error = dsl_destroy_snapshot(snapname, B_FALSE);
3338 if (error != 0 && error != ENOENT)
3339 fatal(0, "ztest_snapshot_destroy(%s) = %d", snapname, error);
3345 ztest_dmu_objset_create_destroy(ztest_ds_t *zd, uint64_t id)
3355 zdtmp = umem_alloc(sizeof (ztest_ds_t), UMEM_NOFAIL);
3356 name = umem_alloc(MAXNAMELEN, UMEM_NOFAIL);
3358 (void) rw_rdlock(&ztest_name_lock);
3360 (void) snprintf(name, MAXNAMELEN, "%s/temp_%llu",
3361 ztest_opts.zo_pool, (u_longlong_t)id);
3364 * If this dataset exists from a previous run, process its replay log
3365 * half of the time. If we don't replay it, then dsl_destroy_head()
3366 * (invoked from ztest_objset_destroy_cb()) should just throw it away.
3368 if (ztest_random(2) == 0 &&
3369 dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os) == 0) {
3370 ztest_zd_init(zdtmp, NULL, os);
3371 zil_replay(os, zdtmp, ztest_replay_vector);
3372 ztest_zd_fini(zdtmp);
3373 dmu_objset_disown(os, FTAG);
3377 * There may be an old instance of the dataset we're about to
3378 * create lying around from a previous run. If so, destroy it
3379 * and all of its snapshots.
3381 (void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
3382 DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
3385 * Verify that the destroyed dataset is no longer in the namespace.
3387 VERIFY3U(ENOENT, ==, dmu_objset_own(name, DMU_OST_OTHER, B_TRUE,
3391 * Verify that we can create a new dataset.
3393 error = ztest_dataset_create(name);
3395 if (error == ENOSPC) {
3396 ztest_record_enospc(FTAG);
3399 fatal(0, "dmu_objset_create(%s) = %d", name, error);
3402 VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os));
3404 ztest_zd_init(zdtmp, NULL, os);
3407 * Open the intent log for it.
3409 zilog = zil_open(os, ztest_get_data);
3412 * Put some objects in there, do a little I/O to them,
3413 * and randomly take a couple of snapshots along the way.
3415 iters = ztest_random(5);
3416 for (i = 0; i < iters; i++) {
3417 ztest_dmu_object_alloc_free(zdtmp, id);
3418 if (ztest_random(iters) == 0)
3419 (void) ztest_snapshot_create(name, i);
3423 * Verify that we cannot create an existing dataset.
3425 VERIFY3U(EEXIST, ==,
3426 dmu_objset_create(name, DMU_OST_OTHER, 0, NULL, NULL));
3429 * Verify that we can hold an objset that is also owned.
3431 VERIFY3U(0, ==, dmu_objset_hold(name, FTAG, &os2));
3432 dmu_objset_rele(os2, FTAG);
3435 * Verify that we cannot own an objset that is already owned.
3438 dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os2));
3441 dmu_objset_disown(os, FTAG);
3442 ztest_zd_fini(zdtmp);
3444 (void) rw_unlock(&ztest_name_lock);
3446 umem_free(name, MAXNAMELEN);
3447 umem_free(zdtmp, sizeof (ztest_ds_t));
3451 * Verify that dmu_snapshot_{create,destroy,open,close} work as expected.
3454 ztest_dmu_snapshot_create_destroy(ztest_ds_t *zd, uint64_t id)
3456 (void) rw_rdlock(&ztest_name_lock);
3457 (void) ztest_snapshot_destroy(zd->zd_name, id);
3458 (void) ztest_snapshot_create(zd->zd_name, id);
3459 (void) rw_unlock(&ztest_name_lock);
3463 * Cleanup non-standard snapshots and clones.
3466 ztest_dsl_dataset_cleanup(char *osname, uint64_t id)
3475 snap1name = umem_alloc(MAXNAMELEN, UMEM_NOFAIL);
3476 clone1name = umem_alloc(MAXNAMELEN, UMEM_NOFAIL);
3477 snap2name = umem_alloc(MAXNAMELEN, UMEM_NOFAIL);
3478 clone2name = umem_alloc(MAXNAMELEN, UMEM_NOFAIL);
3479 snap3name = umem_alloc(MAXNAMELEN, UMEM_NOFAIL);
3481 (void) snprintf(snap1name, MAXNAMELEN, "%s@s1_%llu",
3482 osname, (u_longlong_t)id);
3483 (void) snprintf(clone1name, MAXNAMELEN, "%s/c1_%llu",
3484 osname, (u_longlong_t)id);
3485 (void) snprintf(snap2name, MAXNAMELEN, "%s@s2_%llu",
3486 clone1name, (u_longlong_t)id);
3487 (void) snprintf(clone2name, MAXNAMELEN, "%s/c2_%llu",
3488 osname, (u_longlong_t)id);
3489 (void) snprintf(snap3name, MAXNAMELEN, "%s@s3_%llu",
3490 clone1name, (u_longlong_t)id);
3492 error = dsl_destroy_head(clone2name);
3493 if (error && error != ENOENT)
3494 fatal(0, "dsl_destroy_head(%s) = %d", clone2name, error);
3495 error = dsl_destroy_snapshot(snap3name, B_FALSE);
3496 if (error && error != ENOENT)
3497 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap3name, error);
3498 error = dsl_destroy_snapshot(snap2name, B_FALSE);
3499 if (error && error != ENOENT)
3500 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap2name, error);
3501 error = dsl_destroy_head(clone1name);
3502 if (error && error != ENOENT)
3503 fatal(0, "dsl_destroy_head(%s) = %d", clone1name, error);
3504 error = dsl_destroy_snapshot(snap1name, B_FALSE);
3505 if (error && error != ENOENT)
3506 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap1name, error);
3508 umem_free(snap1name, MAXNAMELEN);
3509 umem_free(clone1name, MAXNAMELEN);
3510 umem_free(snap2name, MAXNAMELEN);
3511 umem_free(clone2name, MAXNAMELEN);
3512 umem_free(snap3name, MAXNAMELEN);
3516 * Verify dsl_dataset_promote handles EBUSY
3519 ztest_dsl_dataset_promote_busy(ztest_ds_t *zd, uint64_t id)
3527 char *osname = zd->zd_name;
3530 snap1name = umem_alloc(MAXNAMELEN, UMEM_NOFAIL);
3531 clone1name = umem_alloc(MAXNAMELEN, UMEM_NOFAIL);
3532 snap2name = umem_alloc(MAXNAMELEN, UMEM_NOFAIL);
3533 clone2name = umem_alloc(MAXNAMELEN, UMEM_NOFAIL);
3534 snap3name = umem_alloc(MAXNAMELEN, UMEM_NOFAIL);
3536 (void) rw_rdlock(&ztest_name_lock);
3538 ztest_dsl_dataset_cleanup(osname, id);
3540 (void) snprintf(snap1name, MAXNAMELEN, "%s@s1_%llu",
3541 osname, (u_longlong_t)id);
3542 (void) snprintf(clone1name, MAXNAMELEN, "%s/c1_%llu",
3543 osname, (u_longlong_t)id);
3544 (void) snprintf(snap2name, MAXNAMELEN, "%s@s2_%llu",
3545 clone1name, (u_longlong_t)id);
3546 (void) snprintf(clone2name, MAXNAMELEN, "%s/c2_%llu",
3547 osname, (u_longlong_t)id);
3548 (void) snprintf(snap3name, MAXNAMELEN, "%s@s3_%llu",
3549 clone1name, (u_longlong_t)id);
3551 error = dmu_objset_snapshot_one(osname, strchr(snap1name, '@') + 1);
3552 if (error && error != EEXIST) {
3553 if (error == ENOSPC) {
3554 ztest_record_enospc(FTAG);
3557 fatal(0, "dmu_take_snapshot(%s) = %d", snap1name, error);
3560 error = dmu_objset_clone(clone1name, snap1name);
3562 if (error == ENOSPC) {
3563 ztest_record_enospc(FTAG);
3566 fatal(0, "dmu_objset_create(%s) = %d", clone1name, error);
3569 error = dmu_objset_snapshot_one(clone1name, strchr(snap2name, '@') + 1);
3570 if (error && error != EEXIST) {
3571 if (error == ENOSPC) {
3572 ztest_record_enospc(FTAG);
3575 fatal(0, "dmu_open_snapshot(%s) = %d", snap2name, error);
3578 error = dmu_objset_snapshot_one(clone1name, strchr(snap3name, '@') + 1);
3579 if (error && error != EEXIST) {
3580 if (error == ENOSPC) {
3581 ztest_record_enospc(FTAG);
3584 fatal(0, "dmu_open_snapshot(%s) = %d", snap3name, error);
3587 error = dmu_objset_clone(clone2name, snap3name);
3589 if (error == ENOSPC) {
3590 ztest_record_enospc(FTAG);
3593 fatal(0, "dmu_objset_create(%s) = %d", clone2name, error);
3596 error = dmu_objset_own(snap2name, DMU_OST_ANY, B_TRUE, FTAG, &os);
3598 fatal(0, "dmu_objset_own(%s) = %d", snap2name, error);
3599 error = dsl_dataset_promote(clone2name, NULL);
3600 if (error == ENOSPC) {
3601 dmu_objset_disown(os, FTAG);
3602 ztest_record_enospc(FTAG);
3606 fatal(0, "dsl_dataset_promote(%s), %d, not EBUSY", clone2name,
3608 dmu_objset_disown(os, FTAG);
3611 ztest_dsl_dataset_cleanup(osname, id);
3613 (void) rw_unlock(&ztest_name_lock);
3615 umem_free(snap1name, MAXNAMELEN);
3616 umem_free(clone1name, MAXNAMELEN);
3617 umem_free(snap2name, MAXNAMELEN);
3618 umem_free(clone2name, MAXNAMELEN);
3619 umem_free(snap3name, MAXNAMELEN);
3622 #undef OD_ARRAY_SIZE
3623 #define OD_ARRAY_SIZE 4
3626 * Verify that dmu_object_{alloc,free} work as expected.
3629 ztest_dmu_object_alloc_free(ztest_ds_t *zd, uint64_t id)
3636 size = sizeof (ztest_od_t) * OD_ARRAY_SIZE;
3637 od = umem_alloc(size, UMEM_NOFAIL);
3638 batchsize = OD_ARRAY_SIZE;
3640 for (b = 0; b < batchsize; b++)
3641 ztest_od_init(od + b, id, FTAG, b, DMU_OT_UINT64_OTHER, 0, 0);
3644 * Destroy the previous batch of objects, create a new batch,
3645 * and do some I/O on the new objects.
3647 if (ztest_object_init(zd, od, size, B_TRUE) != 0)
3650 while (ztest_random(4 * batchsize) != 0)
3651 ztest_io(zd, od[ztest_random(batchsize)].od_object,
3652 ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
3654 umem_free(od, size);
3657 #undef OD_ARRAY_SIZE
3658 #define OD_ARRAY_SIZE 2
3661 * Verify that dmu_{read,write} work as expected.
3664 ztest_dmu_read_write(ztest_ds_t *zd, uint64_t id)
3669 objset_t *os = zd->zd_os;
3670 size = sizeof (ztest_od_t) * OD_ARRAY_SIZE;
3671 od = umem_alloc(size, UMEM_NOFAIL);
3673 int i, freeit, error;
3675 bufwad_t *packbuf, *bigbuf, *pack, *bigH, *bigT;
3676 uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
3677 uint64_t chunksize = (1000 + ztest_random(1000)) * sizeof (uint64_t);
3678 uint64_t regions = 997;
3679 uint64_t stride = 123456789ULL;
3680 uint64_t width = 40;
3681 int free_percent = 5;
3684 * This test uses two objects, packobj and bigobj, that are always
3685 * updated together (i.e. in the same tx) so that their contents are
3686 * in sync and can be compared. Their contents relate to each other
3687 * in a simple way: packobj is a dense array of 'bufwad' structures,
3688 * while bigobj is a sparse array of the same bufwads. Specifically,
3689 * for any index n, there are three bufwads that should be identical:
3691 * packobj, at offset n * sizeof (bufwad_t)
3692 * bigobj, at the head of the nth chunk
3693 * bigobj, at the tail of the nth chunk
3695 * The chunk size is arbitrary. It doesn't have to be a power of two,
3696 * and it doesn't have any relation to the object blocksize.
3697 * The only requirement is that it can hold at least two bufwads.
3699 * Normally, we write the bufwad to each of these locations.
3700 * However, free_percent of the time we instead write zeroes to
3701 * packobj and perform a dmu_free_range() on bigobj. By comparing
3702 * bigobj to packobj, we can verify that the DMU is correctly
3703 * tracking which parts of an object are allocated and free,
3704 * and that the contents of the allocated blocks are correct.
3708 * Read the directory info. If it's the first time, set things up.
3710 ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, chunksize);
3711 ztest_od_init(od + 1, id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize);
3713 if (ztest_object_init(zd, od, size, B_FALSE) != 0) {
3714 umem_free(od, size);
3718 bigobj = od[0].od_object;
3719 packobj = od[1].od_object;
3720 chunksize = od[0].od_gen;
3721 ASSERT(chunksize == od[1].od_gen);
3724 * Prefetch a random chunk of the big object.
3725 * Our aim here is to get some async reads in flight
3726 * for blocks that we may free below; the DMU should
3727 * handle this race correctly.
3729 n = ztest_random(regions) * stride + ztest_random(width);
3730 s = 1 + ztest_random(2 * width - 1);
3731 dmu_prefetch(os, bigobj, 0, n * chunksize, s * chunksize,
3732 ZIO_PRIORITY_SYNC_READ);
3735 * Pick a random index and compute the offsets into packobj and bigobj.
3737 n = ztest_random(regions) * stride + ztest_random(width);
3738 s = 1 + ztest_random(width - 1);
3740 packoff = n * sizeof (bufwad_t);
3741 packsize = s * sizeof (bufwad_t);
3743 bigoff = n * chunksize;
3744 bigsize = s * chunksize;
3746 packbuf = umem_alloc(packsize, UMEM_NOFAIL);
3747 bigbuf = umem_alloc(bigsize, UMEM_NOFAIL);
3750 * free_percent of the time, free a range of bigobj rather than
3753 freeit = (ztest_random(100) < free_percent);
3756 * Read the current contents of our objects.
3758 error = dmu_read(os, packobj, packoff, packsize, packbuf,
3761 error = dmu_read(os, bigobj, bigoff, bigsize, bigbuf,
3766 * Get a tx for the mods to both packobj and bigobj.
3768 tx = dmu_tx_create(os);
3770 dmu_tx_hold_write(tx, packobj, packoff, packsize);
3773 dmu_tx_hold_free(tx, bigobj, bigoff, bigsize);
3775 dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
3777 /* This accounts for setting the checksum/compression. */
3778 dmu_tx_hold_bonus(tx, bigobj);
3780 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3782 umem_free(packbuf, packsize);
3783 umem_free(bigbuf, bigsize);
3784 umem_free(od, size);
3788 enum zio_checksum cksum;
3790 cksum = (enum zio_checksum)
3791 ztest_random_dsl_prop(ZFS_PROP_CHECKSUM);
3792 } while (cksum >= ZIO_CHECKSUM_LEGACY_FUNCTIONS);
3793 dmu_object_set_checksum(os, bigobj, cksum, tx);
3795 enum zio_compress comp;
3797 comp = (enum zio_compress)
3798 ztest_random_dsl_prop(ZFS_PROP_COMPRESSION);
3799 } while (comp >= ZIO_COMPRESS_LEGACY_FUNCTIONS);
3800 dmu_object_set_compress(os, bigobj, comp, tx);
3803 * For each index from n to n + s, verify that the existing bufwad
3804 * in packobj matches the bufwads at the head and tail of the
3805 * corresponding chunk in bigobj. Then update all three bufwads
3806 * with the new values we want to write out.
3808 for (i = 0; i < s; i++) {
3810 pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
3812 bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
3814 bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
3816 ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
3817 ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
3819 if (pack->bw_txg > txg)
3820 fatal(0, "future leak: got %llx, open txg is %llx",
3823 if (pack->bw_data != 0 && pack->bw_index != n + i)
3824 fatal(0, "wrong index: got %llx, wanted %llx+%llx",
3825 pack->bw_index, n, i);
3827 if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
3828 fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
3830 if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
3831 fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
3834 bzero(pack, sizeof (bufwad_t));
3836 pack->bw_index = n + i;
3838 pack->bw_data = 1 + ztest_random(-2ULL);
3845 * We've verified all the old bufwads, and made new ones.
3846 * Now write them out.
3848 dmu_write(os, packobj, packoff, packsize, packbuf, tx);
3851 if (ztest_opts.zo_verbose >= 7) {
3852 (void) printf("freeing offset %llx size %llx"
3854 (u_longlong_t)bigoff,
3855 (u_longlong_t)bigsize,
3858 VERIFY(0 == dmu_free_range(os, bigobj, bigoff, bigsize, tx));
3860 if (ztest_opts.zo_verbose >= 7) {
3861 (void) printf("writing offset %llx size %llx"
3863 (u_longlong_t)bigoff,
3864 (u_longlong_t)bigsize,
3867 dmu_write(os, bigobj, bigoff, bigsize, bigbuf, tx);
3873 * Sanity check the stuff we just wrote.
3876 void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
3877 void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
3879 VERIFY(0 == dmu_read(os, packobj, packoff,
3880 packsize, packcheck, DMU_READ_PREFETCH));
3881 VERIFY(0 == dmu_read(os, bigobj, bigoff,
3882 bigsize, bigcheck, DMU_READ_PREFETCH));
3884 ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
3885 ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
3887 umem_free(packcheck, packsize);
3888 umem_free(bigcheck, bigsize);
3891 umem_free(packbuf, packsize);
3892 umem_free(bigbuf, bigsize);
3893 umem_free(od, size);
3897 compare_and_update_pbbufs(uint64_t s, bufwad_t *packbuf, bufwad_t *bigbuf,
3898 uint64_t bigsize, uint64_t n, uint64_t chunksize, uint64_t txg)
3906 * For each index from n to n + s, verify that the existing bufwad
3907 * in packobj matches the bufwads at the head and tail of the
3908 * corresponding chunk in bigobj. Then update all three bufwads
3909 * with the new values we want to write out.
3911 for (i = 0; i < s; i++) {
3913 pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
3915 bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
3917 bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
3919 ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
3920 ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
3922 if (pack->bw_txg > txg)
3923 fatal(0, "future leak: got %llx, open txg is %llx",
3926 if (pack->bw_data != 0 && pack->bw_index != n + i)
3927 fatal(0, "wrong index: got %llx, wanted %llx+%llx",
3928 pack->bw_index, n, i);
3930 if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
3931 fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
3933 if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
3934 fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
3936 pack->bw_index = n + i;
3938 pack->bw_data = 1 + ztest_random(-2ULL);
3945 #undef OD_ARRAY_SIZE
3946 #define OD_ARRAY_SIZE 2
3949 ztest_dmu_read_write_zcopy(ztest_ds_t *zd, uint64_t id)
3951 objset_t *os = zd->zd_os;
3958 bufwad_t *packbuf, *bigbuf;
3959 uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
3960 uint64_t blocksize = ztest_random_blocksize();
3961 uint64_t chunksize = blocksize;
3962 uint64_t regions = 997;
3963 uint64_t stride = 123456789ULL;
3965 dmu_buf_t *bonus_db;
3966 arc_buf_t **bigbuf_arcbufs;
3967 dmu_object_info_t doi;
3969 size = sizeof (ztest_od_t) * OD_ARRAY_SIZE;
3970 od = umem_alloc(size, UMEM_NOFAIL);
3973 * This test uses two objects, packobj and bigobj, that are always
3974 * updated together (i.e. in the same tx) so that their contents are
3975 * in sync and can be compared. Their contents relate to each other
3976 * in a simple way: packobj is a dense array of 'bufwad' structures,
3977 * while bigobj is a sparse array of the same bufwads. Specifically,
3978 * for any index n, there are three bufwads that should be identical:
3980 * packobj, at offset n * sizeof (bufwad_t)
3981 * bigobj, at the head of the nth chunk
3982 * bigobj, at the tail of the nth chunk
3984 * The chunk size is set equal to bigobj block size so that
3985 * dmu_assign_arcbuf() can be tested for object updates.
3989 * Read the directory info. If it's the first time, set things up.
3991 ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
3992 ztest_od_init(od + 1, id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize);
3995 if (ztest_object_init(zd, od, size, B_FALSE) != 0) {
3996 umem_free(od, size);
4000 bigobj = od[0].od_object;
4001 packobj = od[1].od_object;
4002 blocksize = od[0].od_blocksize;
4003 chunksize = blocksize;
4004 ASSERT(chunksize == od[1].od_gen);
4006 VERIFY(dmu_object_info(os, bigobj, &doi) == 0);
4007 VERIFY(ISP2(doi.doi_data_block_size));
4008 VERIFY(chunksize == doi.doi_data_block_size);
4009 VERIFY(chunksize >= 2 * sizeof (bufwad_t));
4012 * Pick a random index and compute the offsets into packobj and bigobj.
4014 n = ztest_random(regions) * stride + ztest_random(width);
4015 s = 1 + ztest_random(width - 1);
4017 packoff = n * sizeof (bufwad_t);
4018 packsize = s * sizeof (bufwad_t);
4020 bigoff = n * chunksize;
4021 bigsize = s * chunksize;
4023 packbuf = umem_zalloc(packsize, UMEM_NOFAIL);
4024 bigbuf = umem_zalloc(bigsize, UMEM_NOFAIL);
4026 VERIFY3U(0, ==, dmu_bonus_hold(os, bigobj, FTAG, &bonus_db));
4028 bigbuf_arcbufs = umem_zalloc(2 * s * sizeof (arc_buf_t *), UMEM_NOFAIL);
4031 * Iteration 0 test zcopy for DB_UNCACHED dbufs.
4032 * Iteration 1 test zcopy to already referenced dbufs.
4033 * Iteration 2 test zcopy to dirty dbuf in the same txg.
4034 * Iteration 3 test zcopy to dbuf dirty in previous txg.
4035 * Iteration 4 test zcopy when dbuf is no longer dirty.
4036 * Iteration 5 test zcopy when it can't be done.
4037 * Iteration 6 one more zcopy write.
4039 for (i = 0; i < 7; i++) {
4044 * In iteration 5 (i == 5) use arcbufs
4045 * that don't match bigobj blksz to test
4046 * dmu_assign_arcbuf() when it can't directly
4047 * assign an arcbuf to a dbuf.
4049 for (j = 0; j < s; j++) {
4050 if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) {
4052 dmu_request_arcbuf(bonus_db, chunksize);
4054 bigbuf_arcbufs[2 * j] =
4055 dmu_request_arcbuf(bonus_db, chunksize / 2);
4056 bigbuf_arcbufs[2 * j + 1] =
4057 dmu_request_arcbuf(bonus_db, chunksize / 2);
4062 * Get a tx for the mods to both packobj and bigobj.
4064 tx = dmu_tx_create(os);
4066 dmu_tx_hold_write(tx, packobj, packoff, packsize);
4067 dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
4069 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4071 umem_free(packbuf, packsize);
4072 umem_free(bigbuf, bigsize);
4073 for (j = 0; j < s; j++) {
4075 chunksize < (SPA_MINBLOCKSIZE * 2)) {
4076 dmu_return_arcbuf(bigbuf_arcbufs[j]);
4079 bigbuf_arcbufs[2 * j]);
4081 bigbuf_arcbufs[2 * j + 1]);
4084 umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
4085 umem_free(od, size);
4086 dmu_buf_rele(bonus_db, FTAG);
4091 * 50% of the time don't read objects in the 1st iteration to
4092 * test dmu_assign_arcbuf() for the case when there're no
4093 * existing dbufs for the specified offsets.
4095 if (i != 0 || ztest_random(2) != 0) {
4096 error = dmu_read(os, packobj, packoff,
4097 packsize, packbuf, DMU_READ_PREFETCH);
4099 error = dmu_read(os, bigobj, bigoff, bigsize,
4100 bigbuf, DMU_READ_PREFETCH);
4103 compare_and_update_pbbufs(s, packbuf, bigbuf, bigsize,
4107 * We've verified all the old bufwads, and made new ones.
4108 * Now write them out.
4110 dmu_write(os, packobj, packoff, packsize, packbuf, tx);
4111 if (ztest_opts.zo_verbose >= 7) {
4112 (void) printf("writing offset %llx size %llx"
4114 (u_longlong_t)bigoff,
4115 (u_longlong_t)bigsize,
4118 for (off = bigoff, j = 0; j < s; j++, off += chunksize) {
4120 if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) {
4121 bcopy((caddr_t)bigbuf + (off - bigoff),
4122 bigbuf_arcbufs[j]->b_data, chunksize);
4124 bcopy((caddr_t)bigbuf + (off - bigoff),
4125 bigbuf_arcbufs[2 * j]->b_data,
4127 bcopy((caddr_t)bigbuf + (off - bigoff) +
4129 bigbuf_arcbufs[2 * j + 1]->b_data,
4134 VERIFY(dmu_buf_hold(os, bigobj, off,
4135 FTAG, &dbt, DMU_READ_NO_PREFETCH) == 0);
4137 if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) {
4138 dmu_assign_arcbuf(bonus_db, off,
4139 bigbuf_arcbufs[j], tx);
4141 dmu_assign_arcbuf(bonus_db, off,
4142 bigbuf_arcbufs[2 * j], tx);
4143 dmu_assign_arcbuf(bonus_db,
4144 off + chunksize / 2,
4145 bigbuf_arcbufs[2 * j + 1], tx);
4148 dmu_buf_rele(dbt, FTAG);
4154 * Sanity check the stuff we just wrote.
4157 void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
4158 void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
4160 VERIFY(0 == dmu_read(os, packobj, packoff,
4161 packsize, packcheck, DMU_READ_PREFETCH));
4162 VERIFY(0 == dmu_read(os, bigobj, bigoff,
4163 bigsize, bigcheck, DMU_READ_PREFETCH));
4165 ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
4166 ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
4168 umem_free(packcheck, packsize);
4169 umem_free(bigcheck, bigsize);
4172 txg_wait_open(dmu_objset_pool(os), 0);
4173 } else if (i == 3) {
4174 txg_wait_synced(dmu_objset_pool(os), 0);
4178 dmu_buf_rele(bonus_db, FTAG);
4179 umem_free(packbuf, packsize);
4180 umem_free(bigbuf, bigsize);
4181 umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
4182 umem_free(od, size);
4187 ztest_dmu_write_parallel(ztest_ds_t *zd, uint64_t id)
4191 od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
4192 uint64_t offset = (1ULL << (ztest_random(20) + 43)) +
4193 (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
4196 * Have multiple threads write to large offsets in an object
4197 * to verify that parallel writes to an object -- even to the
4198 * same blocks within the object -- doesn't cause any trouble.
4200 ztest_od_init(od, ID_PARALLEL, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
4202 if (ztest_object_init(zd, od, sizeof (ztest_od_t), B_FALSE) != 0)
4205 while (ztest_random(10) != 0)
4206 ztest_io(zd, od->od_object, offset);
4208 umem_free(od, sizeof (ztest_od_t));
4212 ztest_dmu_prealloc(ztest_ds_t *zd, uint64_t id)
4215 uint64_t offset = (1ULL << (ztest_random(4) + SPA_MAXBLOCKSHIFT)) +
4216 (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
4217 uint64_t count = ztest_random(20) + 1;
4218 uint64_t blocksize = ztest_random_blocksize();
4221 od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
4223 ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
4225 if (ztest_object_init(zd, od, sizeof (ztest_od_t),
4226 !ztest_random(2)) != 0) {
4227 umem_free(od, sizeof (ztest_od_t));
4231 if (ztest_truncate(zd, od->od_object, offset, count * blocksize) != 0) {
4232 umem_free(od, sizeof (ztest_od_t));
4236 ztest_prealloc(zd, od->od_object, offset, count * blocksize);
4238 data = umem_zalloc(blocksize, UMEM_NOFAIL);
4240 while (ztest_random(count) != 0) {
4241 uint64_t randoff = offset + (ztest_random(count) * blocksize);
4242 if (ztest_write(zd, od->od_object, randoff, blocksize,
4245 while (ztest_random(4) != 0)
4246 ztest_io(zd, od->od_object, randoff);
4249 umem_free(data, blocksize);
4250 umem_free(od, sizeof (ztest_od_t));
4254 * Verify that zap_{create,destroy,add,remove,update} work as expected.
4256 #define ZTEST_ZAP_MIN_INTS 1
4257 #define ZTEST_ZAP_MAX_INTS 4
4258 #define ZTEST_ZAP_MAX_PROPS 1000
4261 ztest_zap(ztest_ds_t *zd, uint64_t id)
4263 objset_t *os = zd->zd_os;
4266 uint64_t txg, last_txg;
4267 uint64_t value[ZTEST_ZAP_MAX_INTS];
4268 uint64_t zl_ints, zl_intsize, prop;
4271 char propname[100], txgname[100];
4273 char *hc[2] = { "s.acl.h", ".s.open.h.hyLZlg" };
4275 od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
4276 ztest_od_init(od, id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0);
4278 if (ztest_object_init(zd, od, sizeof (ztest_od_t),
4279 !ztest_random(2)) != 0)
4282 object = od->od_object;
4285 * Generate a known hash collision, and verify that
4286 * we can lookup and remove both entries.
4288 tx = dmu_tx_create(os);
4289 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4290 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4293 for (i = 0; i < 2; i++) {
4295 VERIFY3U(0, ==, zap_add(os, object, hc[i], sizeof (uint64_t),
4298 for (i = 0; i < 2; i++) {
4299 VERIFY3U(EEXIST, ==, zap_add(os, object, hc[i],
4300 sizeof (uint64_t), 1, &value[i], tx));
4302 zap_length(os, object, hc[i], &zl_intsize, &zl_ints));
4303 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4304 ASSERT3U(zl_ints, ==, 1);
4306 for (i = 0; i < 2; i++) {
4307 VERIFY3U(0, ==, zap_remove(os, object, hc[i], tx));
4312 * Generate a buch of random entries.
4314 ints = MAX(ZTEST_ZAP_MIN_INTS, object % ZTEST_ZAP_MAX_INTS);
4316 prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
4317 (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
4318 (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
4319 bzero(value, sizeof (value));
4323 * If these zap entries already exist, validate their contents.
4325 error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
4327 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4328 ASSERT3U(zl_ints, ==, 1);
4330 VERIFY(zap_lookup(os, object, txgname, zl_intsize,
4331 zl_ints, &last_txg) == 0);
4333 VERIFY(zap_length(os, object, propname, &zl_intsize,
4336 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4337 ASSERT3U(zl_ints, ==, ints);
4339 VERIFY(zap_lookup(os, object, propname, zl_intsize,
4340 zl_ints, value) == 0);
4342 for (i = 0; i < ints; i++) {
4343 ASSERT3U(value[i], ==, last_txg + object + i);
4346 ASSERT3U(error, ==, ENOENT);
4350 * Atomically update two entries in our zap object.
4351 * The first is named txg_%llu, and contains the txg
4352 * in which the property was last updated. The second
4353 * is named prop_%llu, and the nth element of its value
4354 * should be txg + object + n.
4356 tx = dmu_tx_create(os);
4357 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4358 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4363 fatal(0, "zap future leak: old %llu new %llu", last_txg, txg);
4365 for (i = 0; i < ints; i++)
4366 value[i] = txg + object + i;
4368 VERIFY3U(0, ==, zap_update(os, object, txgname, sizeof (uint64_t),
4370 VERIFY3U(0, ==, zap_update(os, object, propname, sizeof (uint64_t),
4376 * Remove a random pair of entries.
4378 prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
4379 (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
4380 (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
4382 error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
4384 if (error == ENOENT)
4389 tx = dmu_tx_create(os);
4390 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4391 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4394 VERIFY3U(0, ==, zap_remove(os, object, txgname, tx));
4395 VERIFY3U(0, ==, zap_remove(os, object, propname, tx));
4398 umem_free(od, sizeof (ztest_od_t));
4402 * Testcase to test the upgrading of a microzap to fatzap.
4405 ztest_fzap(ztest_ds_t *zd, uint64_t id)
4407 objset_t *os = zd->zd_os;
4409 uint64_t object, txg;
4412 od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
4413 ztest_od_init(od, id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0);
4415 if (ztest_object_init(zd, od, sizeof (ztest_od_t),
4416 !ztest_random(2)) != 0)
4418 object = od->od_object;
4421 * Add entries to this ZAP and make sure it spills over
4422 * and gets upgraded to a fatzap. Also, since we are adding
4423 * 2050 entries we should see ptrtbl growth and leaf-block split.
4425 for (i = 0; i < 2050; i++) {
4426 char name[MAXNAMELEN];
4431 (void) snprintf(name, sizeof (name), "fzap-%llu-%llu",
4432 (u_longlong_t)id, (u_longlong_t)value);
4434 tx = dmu_tx_create(os);
4435 dmu_tx_hold_zap(tx, object, B_TRUE, name);
4436 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4439 error = zap_add(os, object, name, sizeof (uint64_t), 1,
4441 ASSERT(error == 0 || error == EEXIST);
4445 umem_free(od, sizeof (ztest_od_t));
4450 ztest_zap_parallel(ztest_ds_t *zd, uint64_t id)
4452 objset_t *os = zd->zd_os;
4454 uint64_t txg, object, count, wsize, wc, zl_wsize, zl_wc;
4456 int i, namelen, error;
4457 int micro = ztest_random(2);
4458 char name[20], string_value[20];
4461 od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
4462 ztest_od_init(od, ID_PARALLEL, FTAG, micro, DMU_OT_ZAP_OTHER, 0, 0);
4464 if (ztest_object_init(zd, od, sizeof (ztest_od_t), B_FALSE) != 0) {
4465 umem_free(od, sizeof (ztest_od_t));
4469 object = od->od_object;
4472 * Generate a random name of the form 'xxx.....' where each
4473 * x is a random printable character and the dots are dots.
4474 * There are 94 such characters, and the name length goes from
4475 * 6 to 20, so there are 94^3 * 15 = 12,458,760 possible names.
4477 namelen = ztest_random(sizeof (name) - 5) + 5 + 1;
4479 for (i = 0; i < 3; i++)
4480 name[i] = '!' + ztest_random('~' - '!' + 1);
4481 for (; i < namelen - 1; i++)
4485 if ((namelen & 1) || micro) {
4486 wsize = sizeof (txg);
4492 data = string_value;
4496 VERIFY0(zap_count(os, object, &count));
4497 ASSERT(count != -1ULL);
4500 * Select an operation: length, lookup, add, update, remove.
4502 i = ztest_random(5);
4505 tx = dmu_tx_create(os);
4506 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4507 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4510 bcopy(name, string_value, namelen);
4514 bzero(string_value, namelen);
4520 error = zap_length(os, object, name, &zl_wsize, &zl_wc);
4522 ASSERT3U(wsize, ==, zl_wsize);
4523 ASSERT3U(wc, ==, zl_wc);
4525 ASSERT3U(error, ==, ENOENT);
4530 error = zap_lookup(os, object, name, wsize, wc, data);
4532 if (data == string_value &&
4533 bcmp(name, data, namelen) != 0)
4534 fatal(0, "name '%s' != val '%s' len %d",
4535 name, data, namelen);
4537 ASSERT3U(error, ==, ENOENT);
4542 error = zap_add(os, object, name, wsize, wc, data, tx);
4543 ASSERT(error == 0 || error == EEXIST);
4547 VERIFY(zap_update(os, object, name, wsize, wc, data, tx) == 0);
4551 error = zap_remove(os, object, name, tx);
4552 ASSERT(error == 0 || error == ENOENT);
4559 umem_free(od, sizeof (ztest_od_t));
4563 * Commit callback data.
4565 typedef struct ztest_cb_data {
4566 list_node_t zcd_node;
4568 int zcd_expected_err;
4569 boolean_t zcd_added;
4570 boolean_t zcd_called;
4574 /* This is the actual commit callback function */
4576 ztest_commit_callback(void *arg, int error)
4578 ztest_cb_data_t *data = arg;
4579 uint64_t synced_txg;
4581 VERIFY(data != NULL);
4582 VERIFY3S(data->zcd_expected_err, ==, error);
4583 VERIFY(!data->zcd_called);
4585 synced_txg = spa_last_synced_txg(data->zcd_spa);
4586 if (data->zcd_txg > synced_txg)
4587 fatal(0, "commit callback of txg %" PRIu64 " called prematurely"
4588 ", last synced txg = %" PRIu64 "\n", data->zcd_txg,
4591 data->zcd_called = B_TRUE;
4593 if (error == ECANCELED) {
4594 ASSERT0(data->zcd_txg);
4595 ASSERT(!data->zcd_added);
4598 * The private callback data should be destroyed here, but
4599 * since we are going to check the zcd_called field after
4600 * dmu_tx_abort(), we will destroy it there.
4605 ASSERT(data->zcd_added);
4606 ASSERT3U(data->zcd_txg, !=, 0);
4608 (void) mutex_enter(&zcl.zcl_callbacks_lock);
4610 /* See if this cb was called more quickly */
4611 if ((synced_txg - data->zcd_txg) < zc_min_txg_delay)
4612 zc_min_txg_delay = synced_txg - data->zcd_txg;
4614 /* Remove our callback from the list */
4615 list_remove(&zcl.zcl_callbacks, data);
4617 (void) mutex_exit(&zcl.zcl_callbacks_lock);
4619 umem_free(data, sizeof (ztest_cb_data_t));
4622 /* Allocate and initialize callback data structure */
4623 static ztest_cb_data_t *
4624 ztest_create_cb_data(objset_t *os, uint64_t txg)
4626 ztest_cb_data_t *cb_data;
4628 cb_data = umem_zalloc(sizeof (ztest_cb_data_t), UMEM_NOFAIL);
4630 cb_data->zcd_txg = txg;
4631 cb_data->zcd_spa = dmu_objset_spa(os);
4632 list_link_init(&cb_data->zcd_node);
4638 * Commit callback test.
4641 ztest_dmu_commit_callbacks(ztest_ds_t *zd, uint64_t id)
4643 objset_t *os = zd->zd_os;
4646 ztest_cb_data_t *cb_data[3], *tmp_cb;
4647 uint64_t old_txg, txg;
4650 od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
4651 ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
4653 if (ztest_object_init(zd, od, sizeof (ztest_od_t), B_FALSE) != 0) {
4654 umem_free(od, sizeof (ztest_od_t));
4658 tx = dmu_tx_create(os);
4660 cb_data[0] = ztest_create_cb_data(os, 0);
4661 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[0]);
4663 dmu_tx_hold_write(tx, od->od_object, 0, sizeof (uint64_t));
4665 /* Every once in a while, abort the transaction on purpose */
4666 if (ztest_random(100) == 0)
4670 error = dmu_tx_assign(tx, TXG_NOWAIT);
4672 txg = error ? 0 : dmu_tx_get_txg(tx);
4674 cb_data[0]->zcd_txg = txg;
4675 cb_data[1] = ztest_create_cb_data(os, txg);
4676 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[1]);
4680 * It's not a strict requirement to call the registered
4681 * callbacks from inside dmu_tx_abort(), but that's what
4682 * it's supposed to happen in the current implementation
4683 * so we will check for that.
4685 for (i = 0; i < 2; i++) {
4686 cb_data[i]->zcd_expected_err = ECANCELED;
4687 VERIFY(!cb_data[i]->zcd_called);
4692 for (i = 0; i < 2; i++) {
4693 VERIFY(cb_data[i]->zcd_called);
4694 umem_free(cb_data[i], sizeof (ztest_cb_data_t));
4697 umem_free(od, sizeof (ztest_od_t));
4701 cb_data[2] = ztest_create_cb_data(os, txg);
4702 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[2]);
4705 * Read existing data to make sure there isn't a future leak.
4707 VERIFY(0 == dmu_read(os, od->od_object, 0, sizeof (uint64_t),
4708 &old_txg, DMU_READ_PREFETCH));
4711 fatal(0, "future leak: got %" PRIu64 ", open txg is %" PRIu64,
4714 dmu_write(os, od->od_object, 0, sizeof (uint64_t), &txg, tx);
4716 (void) mutex_enter(&zcl.zcl_callbacks_lock);
4719 * Since commit callbacks don't have any ordering requirement and since
4720 * it is theoretically possible for a commit callback to be called
4721 * after an arbitrary amount of time has elapsed since its txg has been
4722 * synced, it is difficult to reliably determine whether a commit
4723 * callback hasn't been called due to high load or due to a flawed
4726 * In practice, we will assume that if after a certain number of txgs a
4727 * commit callback hasn't been called, then most likely there's an
4728 * implementation bug..
4730 tmp_cb = list_head(&zcl.zcl_callbacks);
4731 if (tmp_cb != NULL &&
4732 tmp_cb->zcd_txg + ZTEST_COMMIT_CB_THRESH < txg) {
4733 fatal(0, "Commit callback threshold exceeded, oldest txg: %"
4734 PRIu64 ", open txg: %" PRIu64 "\n", tmp_cb->zcd_txg, txg);
4738 * Let's find the place to insert our callbacks.
4740 * Even though the list is ordered by txg, it is possible for the
4741 * insertion point to not be the end because our txg may already be
4742 * quiescing at this point and other callbacks in the open txg
4743 * (from other objsets) may have sneaked in.
4745 tmp_cb = list_tail(&zcl.zcl_callbacks);
4746 while (tmp_cb != NULL && tmp_cb->zcd_txg > txg)
4747 tmp_cb = list_prev(&zcl.zcl_callbacks, tmp_cb);
4749 /* Add the 3 callbacks to the list */
4750 for (i = 0; i < 3; i++) {
4752 list_insert_head(&zcl.zcl_callbacks, cb_data[i]);
4754 list_insert_after(&zcl.zcl_callbacks, tmp_cb,
4757 cb_data[i]->zcd_added = B_TRUE;
4758 VERIFY(!cb_data[i]->zcd_called);
4760 tmp_cb = cb_data[i];
4765 (void) mutex_exit(&zcl.zcl_callbacks_lock);
4769 umem_free(od, sizeof (ztest_od_t));
4774 ztest_dsl_prop_get_set(ztest_ds_t *zd, uint64_t id)
4776 zfs_prop_t proplist[] = {
4778 ZFS_PROP_COMPRESSION,
4784 (void) rw_rdlock(&ztest_name_lock);
4786 for (p = 0; p < sizeof (proplist) / sizeof (proplist[0]); p++)
4787 (void) ztest_dsl_prop_set_uint64(zd->zd_name, proplist[p],
4788 ztest_random_dsl_prop(proplist[p]), (int)ztest_random(2));
4790 VERIFY0(ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_RECORDSIZE,
4791 ztest_random_blocksize(), (int)ztest_random(2)));
4793 (void) rw_unlock(&ztest_name_lock);
4798 ztest_spa_prop_get_set(ztest_ds_t *zd, uint64_t id)
4800 nvlist_t *props = NULL;
4802 (void) rw_rdlock(&ztest_name_lock);
4804 (void) ztest_spa_prop_set_uint64(ZPOOL_PROP_DEDUPDITTO,
4805 ZIO_DEDUPDITTO_MIN + ztest_random(ZIO_DEDUPDITTO_MIN));
4807 VERIFY0(spa_prop_get(ztest_spa, &props));
4809 if (ztest_opts.zo_verbose >= 6)
4810 dump_nvlist(props, 4);
4814 (void) rw_unlock(&ztest_name_lock);
4818 user_release_one(const char *snapname, const char *holdname)
4820 nvlist_t *snaps, *holds;
4823 snaps = fnvlist_alloc();
4824 holds = fnvlist_alloc();
4825 fnvlist_add_boolean(holds, holdname);
4826 fnvlist_add_nvlist(snaps, snapname, holds);
4827 fnvlist_free(holds);
4828 error = dsl_dataset_user_release(snaps, NULL);
4829 fnvlist_free(snaps);
4834 * Test snapshot hold/release and deferred destroy.
4837 ztest_dmu_snapshot_hold(ztest_ds_t *zd, uint64_t id)
4840 objset_t *os = zd->zd_os;
4844 char clonename[100];
4846 char osname[MAXNAMELEN];
4849 (void) rw_rdlock(&ztest_name_lock);
4851 dmu_objset_name(os, osname);
4853 (void) snprintf(snapname, sizeof (snapname), "sh1_%llu",
4855 (void) snprintf(fullname, sizeof (fullname), "%s@%s", osname, snapname);
4856 (void) snprintf(clonename, sizeof (clonename),
4857 "%s/ch1_%llu", osname, (u_longlong_t)id);
4858 (void) snprintf(tag, sizeof (tag), "tag_%llu", (u_longlong_t)id);
4861 * Clean up from any previous run.
4863 error = dsl_destroy_head(clonename);
4864 if (error != ENOENT)
4866 error = user_release_one(fullname, tag);
4867 if (error != ESRCH && error != ENOENT)
4869 error = dsl_destroy_snapshot(fullname, B_FALSE);
4870 if (error != ENOENT)
4874 * Create snapshot, clone it, mark snap for deferred destroy,
4875 * destroy clone, verify snap was also destroyed.
4877 error = dmu_objset_snapshot_one(osname, snapname);
4879 if (error == ENOSPC) {
4880 ztest_record_enospc("dmu_objset_snapshot");
4883 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
4886 error = dmu_objset_clone(clonename, fullname);
4888 if (error == ENOSPC) {
4889 ztest_record_enospc("dmu_objset_clone");
4892 fatal(0, "dmu_objset_clone(%s) = %d", clonename, error);
4895 error = dsl_destroy_snapshot(fullname, B_TRUE);
4897 fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
4901 error = dsl_destroy_head(clonename);
4903 fatal(0, "dsl_destroy_head(%s) = %d", clonename, error);
4905 error = dmu_objset_hold(fullname, FTAG, &origin);
4906 if (error != ENOENT)
4907 fatal(0, "dmu_objset_hold(%s) = %d", fullname, error);
4910 * Create snapshot, add temporary hold, verify that we can't
4911 * destroy a held snapshot, mark for deferred destroy,
4912 * release hold, verify snapshot was destroyed.
4914 error = dmu_objset_snapshot_one(osname, snapname);
4916 if (error == ENOSPC) {
4917 ztest_record_enospc("dmu_objset_snapshot");
4920 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
4923 holds = fnvlist_alloc();
4924 fnvlist_add_string(holds, fullname, tag);
4925 error = dsl_dataset_user_hold(holds, 0, NULL);
4926 fnvlist_free(holds);
4928 if (error == ENOSPC) {
4929 ztest_record_enospc("dsl_dataset_user_hold");
4932 fatal(0, "dsl_dataset_user_hold(%s, %s) = %u",
4933 fullname, tag, error);
4936 error = dsl_destroy_snapshot(fullname, B_FALSE);
4937 if (error != EBUSY) {
4938 fatal(0, "dsl_destroy_snapshot(%s, B_FALSE) = %d",
4942 error = dsl_destroy_snapshot(fullname, B_TRUE);
4944 fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
4948 error = user_release_one(fullname, tag);
4950 fatal(0, "user_release_one(%s, %s) = %d", fullname, tag, error);
4952 VERIFY3U(dmu_objset_hold(fullname, FTAG, &origin), ==, ENOENT);
4955 (void) rw_unlock(&ztest_name_lock);
4959 * Inject random faults into the on-disk data.
4963 ztest_fault_inject(ztest_ds_t *zd, uint64_t id)
4965 ztest_shared_t *zs = ztest_shared;
4966 spa_t *spa = ztest_spa;
4970 uint64_t bad = 0x1990c0ffeedecadeull;
4975 int bshift = SPA_OLD_MAXBLOCKSHIFT + 2; /* don't scrog all labels */
4981 boolean_t islog = B_FALSE;
4983 path0 = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
4984 pathrand = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
4986 mutex_enter(&ztest_vdev_lock);
4987 maxfaults = MAXFAULTS();
4988 leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz;
4989 mirror_save = zs->zs_mirrors;
4990 mutex_exit(&ztest_vdev_lock);
4992 ASSERT(leaves >= 1);
4995 * Grab the name lock as reader. There are some operations
4996 * which don't like to have their vdevs changed while
4997 * they are in progress (i.e. spa_change_guid). Those
4998 * operations will have grabbed the name lock as writer.
5000 (void) rw_rdlock(&ztest_name_lock);
5003 * We need SCL_STATE here because we're going to look at vd0->vdev_tsd.
5005 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5007 if (ztest_random(2) == 0) {
5009 * Inject errors on a normal data device or slog device.
5011 top = ztest_random_vdev_top(spa, B_TRUE);
5012 leaf = ztest_random(leaves) + zs->zs_splits;
5015 * Generate paths to the first leaf in this top-level vdev,
5016 * and to the random leaf we selected. We'll induce transient
5017 * write failures and random online/offline activity on leaf 0,
5018 * and we'll write random garbage to the randomly chosen leaf.
5020 (void) snprintf(path0, MAXPATHLEN, ztest_dev_template,
5021 ztest_opts.zo_dir, ztest_opts.zo_pool,
5022 top * leaves + zs->zs_splits);
5023 (void) snprintf(pathrand, MAXPATHLEN, ztest_dev_template,
5024 ztest_opts.zo_dir, ztest_opts.zo_pool,
5025 top * leaves + leaf);
5027 vd0 = vdev_lookup_by_path(spa->spa_root_vdev, path0);
5028 if (vd0 != NULL && vd0->vdev_top->vdev_islog)
5032 * If the top-level vdev needs to be resilvered
5033 * then we only allow faults on the device that is
5036 if (vd0 != NULL && maxfaults != 1 &&
5037 (!vdev_resilver_needed(vd0->vdev_top, NULL, NULL) ||
5038 vd0->vdev_resilver_txg != 0)) {
5040 * Make vd0 explicitly claim to be unreadable,
5041 * or unwriteable, or reach behind its back
5042 * and close the underlying fd. We can do this if
5043 * maxfaults == 0 because we'll fail and reexecute,
5044 * and we can do it if maxfaults >= 2 because we'll
5045 * have enough redundancy. If maxfaults == 1, the
5046 * combination of this with injection of random data
5047 * corruption below exceeds the pool's fault tolerance.
5049 vdev_file_t *vf = vd0->vdev_tsd;
5051 if (vf != NULL && ztest_random(3) == 0) {
5052 (void) close(vf->vf_vnode->v_fd);
5053 vf->vf_vnode->v_fd = -1;
5054 } else if (ztest_random(2) == 0) {
5055 vd0->vdev_cant_read = B_TRUE;
5057 vd0->vdev_cant_write = B_TRUE;
5059 guid0 = vd0->vdev_guid;
5063 * Inject errors on an l2cache device.
5065 spa_aux_vdev_t *sav = &spa->spa_l2cache;
5067 if (sav->sav_count == 0) {
5068 spa_config_exit(spa, SCL_STATE, FTAG);
5069 (void) rw_unlock(&ztest_name_lock);
5072 vd0 = sav->sav_vdevs[ztest_random(sav->sav_count)];
5073 guid0 = vd0->vdev_guid;
5074 (void) strcpy(path0, vd0->vdev_path);
5075 (void) strcpy(pathrand, vd0->vdev_path);
5079 maxfaults = INT_MAX; /* no limit on cache devices */
5082 spa_config_exit(spa, SCL_STATE, FTAG);
5083 (void) rw_unlock(&ztest_name_lock);
5086 * If we can tolerate two or more faults, or we're dealing
5087 * with a slog, randomly online/offline vd0.
5089 if ((maxfaults >= 2 || islog) && guid0 != 0) {
5090 if (ztest_random(10) < 6) {
5091 int flags = (ztest_random(2) == 0 ?
5092 ZFS_OFFLINE_TEMPORARY : 0);
5095 * We have to grab the zs_name_lock as writer to
5096 * prevent a race between offlining a slog and
5097 * destroying a dataset. Offlining the slog will
5098 * grab a reference on the dataset which may cause
5099 * dsl_destroy_head() to fail with EBUSY thus
5100 * leaving the dataset in an inconsistent state.
5103 (void) rw_wrlock(&ztest_name_lock);
5105 VERIFY(vdev_offline(spa, guid0, flags) != EBUSY);
5108 (void) rw_unlock(&ztest_name_lock);
5111 * Ideally we would like to be able to randomly
5112 * call vdev_[on|off]line without holding locks
5113 * to force unpredictable failures but the side
5114 * effects of vdev_[on|off]line prevent us from
5115 * doing so. We grab the ztest_vdev_lock here to
5116 * prevent a race between injection testing and
5119 mutex_enter(&ztest_vdev_lock);
5120 (void) vdev_online(spa, guid0, 0, NULL);
5121 mutex_exit(&ztest_vdev_lock);
5129 * We have at least single-fault tolerance, so inject data corruption.
5131 fd = open(pathrand, O_RDWR);
5133 if (fd == -1) /* we hit a gap in the device namespace */
5136 fsize = lseek(fd, 0, SEEK_END);
5138 while (--iters != 0) {
5139 offset = ztest_random(fsize / (leaves << bshift)) *
5140 (leaves << bshift) + (leaf << bshift) +
5141 (ztest_random(1ULL << (bshift - 1)) & -8ULL);
5143 if (offset >= fsize)
5146 mutex_enter(&ztest_vdev_lock);
5147 if (mirror_save != zs->zs_mirrors) {
5148 mutex_exit(&ztest_vdev_lock);
5153 if (pwrite(fd, &bad, sizeof (bad), offset) != sizeof (bad))
5154 fatal(1, "can't inject bad word at 0x%llx in %s",
5157 mutex_exit(&ztest_vdev_lock);
5159 if (ztest_opts.zo_verbose >= 7)
5160 (void) printf("injected bad word into %s,"
5161 " offset 0x%llx\n", pathrand, (u_longlong_t)offset);
5166 umem_free(path0, MAXPATHLEN);
5167 umem_free(pathrand, MAXPATHLEN);
5171 * Verify that DDT repair works as expected.
5174 ztest_ddt_repair(ztest_ds_t *zd, uint64_t id)
5176 ztest_shared_t *zs = ztest_shared;
5177 spa_t *spa = ztest_spa;
5178 objset_t *os = zd->zd_os;
5180 uint64_t object, blocksize, txg, pattern, psize;
5181 enum zio_checksum checksum = spa_dedup_checksum(spa);
5186 int copies = 2 * ZIO_DEDUPDITTO_MIN;
5189 blocksize = ztest_random_blocksize();
5190 blocksize = MIN(blocksize, 2048); /* because we write so many */
5192 od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
5193 ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
5195 if (ztest_object_init(zd, od, sizeof (ztest_od_t), B_FALSE) != 0) {
5196 umem_free(od, sizeof (ztest_od_t));
5201 * Take the name lock as writer to prevent anyone else from changing
5202 * the pool and dataset properies we need to maintain during this test.
5204 (void) rw_wrlock(&ztest_name_lock);
5206 if (ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_DEDUP, checksum,
5208 ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_COPIES, 1,
5210 (void) rw_unlock(&ztest_name_lock);
5211 umem_free(od, sizeof (ztest_od_t));
5215 object = od[0].od_object;
5216 blocksize = od[0].od_blocksize;
5217 pattern = zs->zs_guid ^ dmu_objset_fsid_guid(os);
5219 ASSERT(object != 0);
5221 tx = dmu_tx_create(os);
5222 dmu_tx_hold_write(tx, object, 0, copies * blocksize);
5223 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
5225 (void) rw_unlock(&ztest_name_lock);
5226 umem_free(od, sizeof (ztest_od_t));
5231 * Write all the copies of our block.
5233 for (i = 0; i < copies; i++) {
5234 uint64_t offset = i * blocksize;
5235 int error = dmu_buf_hold(os, object, offset, FTAG, &db,
5236 DMU_READ_NO_PREFETCH);
5238 fatal(B_FALSE, "dmu_buf_hold(%p, %llu, %llu) = %u",
5239 os, (long long)object, (long long) offset, error);
5241 ASSERT(db->db_offset == offset);
5242 ASSERT(db->db_size == blocksize);
5243 ASSERT(ztest_pattern_match(db->db_data, db->db_size, pattern) ||
5244 ztest_pattern_match(db->db_data, db->db_size, 0ULL));
5245 dmu_buf_will_fill(db, tx);
5246 ztest_pattern_set(db->db_data, db->db_size, pattern);
5247 dmu_buf_rele(db, FTAG);
5251 txg_wait_synced(spa_get_dsl(spa), txg);
5254 * Find out what block we got.
5256 VERIFY0(dmu_buf_hold(os, object, 0, FTAG, &db,
5257 DMU_READ_NO_PREFETCH));
5258 blk = *((dmu_buf_impl_t *)db)->db_blkptr;
5259 dmu_buf_rele(db, FTAG);
5262 * Damage the block. Dedup-ditto will save us when we read it later.
5264 psize = BP_GET_PSIZE(&blk);
5265 buf = zio_buf_alloc(psize);
5266 ztest_pattern_set(buf, psize, ~pattern);
5268 (void) zio_wait(zio_rewrite(NULL, spa, 0, &blk,
5269 buf, psize, NULL, NULL, ZIO_PRIORITY_SYNC_WRITE,
5270 ZIO_FLAG_CANFAIL | ZIO_FLAG_INDUCE_DAMAGE, NULL));
5272 zio_buf_free(buf, psize);
5274 (void) rw_unlock(&ztest_name_lock);
5275 umem_free(od, sizeof (ztest_od_t));
5283 ztest_scrub(ztest_ds_t *zd, uint64_t id)
5285 spa_t *spa = ztest_spa;
5287 (void) spa_scan(spa, POOL_SCAN_SCRUB);
5288 (void) poll(NULL, 0, 100); /* wait a moment, then force a restart */
5289 (void) spa_scan(spa, POOL_SCAN_SCRUB);
5293 * Change the guid for the pool.
5297 ztest_reguid(ztest_ds_t *zd, uint64_t id)
5299 spa_t *spa = ztest_spa;
5300 uint64_t orig, load;
5303 orig = spa_guid(spa);
5304 load = spa_load_guid(spa);
5306 (void) rw_wrlock(&ztest_name_lock);
5307 error = spa_change_guid(spa);
5308 (void) rw_unlock(&ztest_name_lock);
5313 if (ztest_opts.zo_verbose >= 4) {
5314 (void) printf("Changed guid old %llu -> %llu\n",
5315 (u_longlong_t)orig, (u_longlong_t)spa_guid(spa));
5318 VERIFY3U(orig, !=, spa_guid(spa));
5319 VERIFY3U(load, ==, spa_load_guid(spa));
5323 * Rename the pool to a different name and then rename it back.
5327 ztest_spa_rename(ztest_ds_t *zd, uint64_t id)
5329 char *oldname, *newname;
5332 (void) rw_wrlock(&ztest_name_lock);
5334 oldname = ztest_opts.zo_pool;
5335 newname = umem_alloc(strlen(oldname) + 5, UMEM_NOFAIL);
5336 (void) strcpy(newname, oldname);
5337 (void) strcat(newname, "_tmp");
5342 VERIFY3U(0, ==, spa_rename(oldname, newname));
5345 * Try to open it under the old name, which shouldn't exist
5347 VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
5350 * Open it under the new name and make sure it's still the same spa_t.
5352 VERIFY3U(0, ==, spa_open(newname, &spa, FTAG));
5354 ASSERT(spa == ztest_spa);
5355 spa_close(spa, FTAG);
5358 * Rename it back to the original
5360 VERIFY3U(0, ==, spa_rename(newname, oldname));
5363 * Make sure it can still be opened
5365 VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG));
5367 ASSERT(spa == ztest_spa);
5368 spa_close(spa, FTAG);
5370 umem_free(newname, strlen(newname) + 1);
5372 (void) rw_unlock(&ztest_name_lock);
5376 ztest_check_path(char *path)
5379 /* return true on success */
5380 return (!stat(path, &s));
5384 ztest_get_zdb_bin(char *bin, int len)
5388 * Try to use ZDB_PATH and in-tree zdb path. If not successful, just
5389 * let popen to search through PATH.
5391 if ((zdb_path = getenv("ZDB_PATH"))) {
5392 strlcpy(bin, zdb_path, len); /* In env */
5393 if (!ztest_check_path(bin)) {
5394 ztest_dump_core = 0;
5395 fatal(1, "invalid ZDB_PATH '%s'", bin);
5400 VERIFY(realpath(getexecname(), bin) != NULL);
5401 if (strstr(bin, "/ztest/")) {
5402 strstr(bin, "/ztest/")[0] = '\0'; /* In-tree */
5403 strcat(bin, "/zdb/zdb");
5404 if (ztest_check_path(bin))
5411 * Verify pool integrity by running zdb.
5414 ztest_run_zdb(char *pool)
5420 const int len = MAXPATHLEN + MAXNAMELEN + 20;
5423 bin = umem_alloc(len, UMEM_NOFAIL);
5424 zdb = umem_alloc(len, UMEM_NOFAIL);
5425 zbuf = umem_alloc(1024, UMEM_NOFAIL);
5427 ztest_get_zdb_bin(bin, len);
5430 "%s -bcc%s%s -d -U %s %s",
5432 ztest_opts.zo_verbose >= 3 ? "s" : "",
5433 ztest_opts.zo_verbose >= 4 ? "v" : "",
5437 if (ztest_opts.zo_verbose >= 5)
5438 (void) printf("Executing %s\n", strstr(zdb, "zdb "));
5440 fp = popen(zdb, "r");
5442 while (fgets(zbuf, 1024, fp) != NULL)
5443 if (ztest_opts.zo_verbose >= 3)
5444 (void) printf("%s", zbuf);
5446 status = pclose(fp);
5451 ztest_dump_core = 0;
5452 if (WIFEXITED(status))
5453 fatal(0, "'%s' exit code %d", zdb, WEXITSTATUS(status));
5455 fatal(0, "'%s' died with signal %d", zdb, WTERMSIG(status));
5457 umem_free(bin, len);
5458 umem_free(zdb, len);
5459 umem_free(zbuf, 1024);
5463 ztest_walk_pool_directory(char *header)
5467 if (ztest_opts.zo_verbose >= 6)
5468 (void) printf("%s\n", header);
5470 mutex_enter(&spa_namespace_lock);
5471 while ((spa = spa_next(spa)) != NULL)
5472 if (ztest_opts.zo_verbose >= 6)
5473 (void) printf("\t%s\n", spa_name(spa));
5474 mutex_exit(&spa_namespace_lock);
5478 ztest_spa_import_export(char *oldname, char *newname)
5480 nvlist_t *config, *newconfig;
5485 if (ztest_opts.zo_verbose >= 4) {
5486 (void) printf("import/export: old = %s, new = %s\n",
5491 * Clean up from previous runs.
5493 (void) spa_destroy(newname);
5496 * Get the pool's configuration and guid.
5498 VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG));
5501 * Kick off a scrub to tickle scrub/export races.
5503 if (ztest_random(2) == 0)
5504 (void) spa_scan(spa, POOL_SCAN_SCRUB);
5506 pool_guid = spa_guid(spa);
5507 spa_close(spa, FTAG);
5509 ztest_walk_pool_directory("pools before export");
5514 VERIFY3U(0, ==, spa_export(oldname, &config, B_FALSE, B_FALSE));
5516 ztest_walk_pool_directory("pools after export");
5521 newconfig = spa_tryimport(config);
5522 ASSERT(newconfig != NULL);
5523 nvlist_free(newconfig);
5526 * Import it under the new name.
5528 error = spa_import(newname, config, NULL, 0);
5530 dump_nvlist(config, 0);
5531 fatal(B_FALSE, "couldn't import pool %s as %s: error %u",
5532 oldname, newname, error);
5535 ztest_walk_pool_directory("pools after import");
5538 * Try to import it again -- should fail with EEXIST.
5540 VERIFY3U(EEXIST, ==, spa_import(newname, config, NULL, 0));
5543 * Try to import it under a different name -- should fail with EEXIST.
5545 VERIFY3U(EEXIST, ==, spa_import(oldname, config, NULL, 0));
5548 * Verify that the pool is no longer visible under the old name.
5550 VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
5553 * Verify that we can open and close the pool using the new name.
5555 VERIFY3U(0, ==, spa_open(newname, &spa, FTAG));
5556 ASSERT(pool_guid == spa_guid(spa));
5557 spa_close(spa, FTAG);
5559 nvlist_free(config);
5563 ztest_resume(spa_t *spa)
5565 if (spa_suspended(spa) && ztest_opts.zo_verbose >= 6)
5566 (void) printf("resuming from suspended state\n");
5567 spa_vdev_state_enter(spa, SCL_NONE);
5568 vdev_clear(spa, NULL);
5569 (void) spa_vdev_state_exit(spa, NULL, 0);
5570 (void) zio_resume(spa);
5574 ztest_resume_thread(void *arg)
5578 while (!ztest_exiting) {
5579 if (spa_suspended(spa))
5581 (void) poll(NULL, 0, 100);
5593 ztest_deadman_alarm(int sig)
5595 fatal(0, "failed to complete within %d seconds of deadline", GRACE);
5600 ztest_execute(int test, ztest_info_t *zi, uint64_t id)
5602 ztest_ds_t *zd = &ztest_ds[id % ztest_opts.zo_datasets];
5603 ztest_shared_callstate_t *zc = ZTEST_GET_SHARED_CALLSTATE(test);
5604 hrtime_t functime = gethrtime();
5607 for (i = 0; i < zi->zi_iters; i++)
5608 zi->zi_func(zd, id);
5610 functime = gethrtime() - functime;
5612 atomic_add_64(&zc->zc_count, 1);
5613 atomic_add_64(&zc->zc_time, functime);
5615 if (ztest_opts.zo_verbose >= 4)
5616 (void) printf("%6.2f sec in %s\n",
5617 (double)functime / NANOSEC, zi->zi_funcname);
5621 ztest_thread(void *arg)
5624 uint64_t id = (uintptr_t)arg;
5625 ztest_shared_t *zs = ztest_shared;
5629 ztest_shared_callstate_t *zc;
5631 while ((now = gethrtime()) < zs->zs_thread_stop) {
5633 * See if it's time to force a crash.
5635 if (now > zs->zs_thread_kill)
5639 * If we're getting ENOSPC with some regularity, stop.
5641 if (zs->zs_enospc_count > 10)
5645 * Pick a random function to execute.
5647 rand = ztest_random(ZTEST_FUNCS);
5648 zi = &ztest_info[rand];
5649 zc = ZTEST_GET_SHARED_CALLSTATE(rand);
5650 call_next = zc->zc_next;
5652 if (now >= call_next &&
5653 atomic_cas_64(&zc->zc_next, call_next, call_next +
5654 ztest_random(2 * zi->zi_interval[0] + 1)) == call_next) {
5655 ztest_execute(rand, zi, id);
5665 ztest_dataset_name(char *dsname, char *pool, int d)
5667 (void) snprintf(dsname, MAXNAMELEN, "%s/ds_%d", pool, d);
5671 ztest_dataset_destroy(int d)
5673 char name[MAXNAMELEN];
5676 ztest_dataset_name(name, ztest_opts.zo_pool, d);
5678 if (ztest_opts.zo_verbose >= 3)
5679 (void) printf("Destroying %s to free up space\n", name);
5682 * Cleanup any non-standard clones and snapshots. In general,
5683 * ztest thread t operates on dataset (t % zopt_datasets),
5684 * so there may be more than one thing to clean up.
5686 for (t = d; t < ztest_opts.zo_threads;
5687 t += ztest_opts.zo_datasets)
5688 ztest_dsl_dataset_cleanup(name, t);
5690 (void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
5691 DS_FIND_SNAPSHOTS | DS_FIND_CHILDREN);
5695 ztest_dataset_dirobj_verify(ztest_ds_t *zd)
5697 uint64_t usedobjs, dirobjs, scratch;
5700 * ZTEST_DIROBJ is the object directory for the entire dataset.
5701 * Therefore, the number of objects in use should equal the
5702 * number of ZTEST_DIROBJ entries, +1 for ZTEST_DIROBJ itself.
5703 * If not, we have an object leak.
5705 * Note that we can only check this in ztest_dataset_open(),
5706 * when the open-context and syncing-context values agree.
5707 * That's because zap_count() returns the open-context value,
5708 * while dmu_objset_space() returns the rootbp fill count.
5710 VERIFY3U(0, ==, zap_count(zd->zd_os, ZTEST_DIROBJ, &dirobjs));
5711 dmu_objset_space(zd->zd_os, &scratch, &scratch, &usedobjs, &scratch);
5712 ASSERT3U(dirobjs + 1, ==, usedobjs);
5716 ztest_dataset_open(int d)
5718 ztest_ds_t *zd = &ztest_ds[d];
5719 uint64_t committed_seq = ZTEST_GET_SHARED_DS(d)->zd_seq;
5722 char name[MAXNAMELEN];
5725 ztest_dataset_name(name, ztest_opts.zo_pool, d);
5727 (void) rw_rdlock(&ztest_name_lock);
5729 error = ztest_dataset_create(name);
5730 if (error == ENOSPC) {
5731 (void) rw_unlock(&ztest_name_lock);
5732 ztest_record_enospc(FTAG);
5735 ASSERT(error == 0 || error == EEXIST);
5737 VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, zd, &os));
5738 (void) rw_unlock(&ztest_name_lock);
5740 ztest_zd_init(zd, ZTEST_GET_SHARED_DS(d), os);
5742 zilog = zd->zd_zilog;
5744 if (zilog->zl_header->zh_claim_lr_seq != 0 &&
5745 zilog->zl_header->zh_claim_lr_seq < committed_seq)
5746 fatal(0, "missing log records: claimed %llu < committed %llu",
5747 zilog->zl_header->zh_claim_lr_seq, committed_seq);
5749 ztest_dataset_dirobj_verify(zd);
5751 zil_replay(os, zd, ztest_replay_vector);
5753 ztest_dataset_dirobj_verify(zd);
5755 if (ztest_opts.zo_verbose >= 6)
5756 (void) printf("%s replay %llu blocks, %llu records, seq %llu\n",
5758 (u_longlong_t)zilog->zl_parse_blk_count,
5759 (u_longlong_t)zilog->zl_parse_lr_count,
5760 (u_longlong_t)zilog->zl_replaying_seq);
5762 zilog = zil_open(os, ztest_get_data);
5764 if (zilog->zl_replaying_seq != 0 &&
5765 zilog->zl_replaying_seq < committed_seq)
5766 fatal(0, "missing log records: replayed %llu < committed %llu",
5767 zilog->zl_replaying_seq, committed_seq);
5773 ztest_dataset_close(int d)
5775 ztest_ds_t *zd = &ztest_ds[d];
5777 zil_close(zd->zd_zilog);
5778 dmu_objset_disown(zd->zd_os, zd);
5784 * Kick off threads to run tests on all datasets in parallel.
5787 ztest_run(ztest_shared_t *zs)
5792 kthread_t *resume_thread;
5797 ztest_exiting = B_FALSE;
5800 * Initialize parent/child shared state.
5802 mutex_init(&ztest_vdev_lock, NULL, MUTEX_DEFAULT, NULL);
5803 VERIFY(rwlock_init(&ztest_name_lock, USYNC_THREAD, NULL) == 0);
5805 zs->zs_thread_start = gethrtime();
5806 zs->zs_thread_stop =
5807 zs->zs_thread_start + ztest_opts.zo_passtime * NANOSEC;
5808 zs->zs_thread_stop = MIN(zs->zs_thread_stop, zs->zs_proc_stop);
5809 zs->zs_thread_kill = zs->zs_thread_stop;
5810 if (ztest_random(100) < ztest_opts.zo_killrate) {
5811 zs->zs_thread_kill -=
5812 ztest_random(ztest_opts.zo_passtime * NANOSEC);
5815 mutex_init(&zcl.zcl_callbacks_lock, NULL, MUTEX_DEFAULT, NULL);
5817 list_create(&zcl.zcl_callbacks, sizeof (ztest_cb_data_t),
5818 offsetof(ztest_cb_data_t, zcd_node));
5823 kernel_init(FREAD | FWRITE);
5824 VERIFY0(spa_open(ztest_opts.zo_pool, &spa, FTAG));
5825 spa->spa_debug = B_TRUE;
5826 metaslab_preload_limit = ztest_random(20) + 1;
5829 VERIFY0(dmu_objset_own(ztest_opts.zo_pool,
5830 DMU_OST_ANY, B_TRUE, FTAG, &os));
5831 zs->zs_guid = dmu_objset_fsid_guid(os);
5832 dmu_objset_disown(os, FTAG);
5834 spa->spa_dedup_ditto = 2 * ZIO_DEDUPDITTO_MIN;
5837 * We don't expect the pool to suspend unless maxfaults == 0,
5838 * in which case ztest_fault_inject() temporarily takes away
5839 * the only valid replica.
5841 if (MAXFAULTS() == 0)
5842 spa->spa_failmode = ZIO_FAILURE_MODE_WAIT;
5844 spa->spa_failmode = ZIO_FAILURE_MODE_PANIC;
5847 * Create a thread to periodically resume suspended I/O.
5849 VERIFY3P((resume_thread = zk_thread_create(NULL, 0,
5850 (thread_func_t)ztest_resume_thread, spa, TS_RUN, NULL, 0, 0,
5851 PTHREAD_CREATE_JOINABLE)), !=, NULL);
5855 * Set a deadman alarm to abort() if we hang.
5857 signal(SIGALRM, ztest_deadman_alarm);
5858 alarm((zs->zs_thread_stop - zs->zs_thread_start) / NANOSEC + GRACE);
5862 * Verify that we can safely inquire about about any object,
5863 * whether it's allocated or not. To make it interesting,
5864 * we probe a 5-wide window around each power of two.
5865 * This hits all edge cases, including zero and the max.
5867 for (t = 0; t < 64; t++) {
5868 for (d = -5; d <= 5; d++) {
5869 error = dmu_object_info(spa->spa_meta_objset,
5870 (1ULL << t) + d, NULL);
5871 ASSERT(error == 0 || error == ENOENT ||
5877 * If we got any ENOSPC errors on the previous run, destroy something.
5879 if (zs->zs_enospc_count != 0) {
5880 int d = ztest_random(ztest_opts.zo_datasets);
5881 ztest_dataset_destroy(d);
5883 zs->zs_enospc_count = 0;
5885 tid = umem_zalloc(ztest_opts.zo_threads * sizeof (kt_did_t),
5888 if (ztest_opts.zo_verbose >= 4)
5889 (void) printf("starting main threads...\n");
5892 * Kick off all the tests that run in parallel.
5894 for (t = 0; t < ztest_opts.zo_threads; t++) {
5897 if (t < ztest_opts.zo_datasets &&
5898 ztest_dataset_open(t) != 0)
5901 VERIFY3P(thread = zk_thread_create(NULL, 0,
5902 (thread_func_t)ztest_thread,
5903 (void *)(uintptr_t)t, TS_RUN, NULL, 0, 0,
5904 PTHREAD_CREATE_JOINABLE), !=, NULL);
5905 tid[t] = thread->t_tid;
5909 * Wait for all of the tests to complete. We go in reverse order
5910 * so we don't close datasets while threads are still using them.
5912 for (t = ztest_opts.zo_threads - 1; t >= 0; t--) {
5913 thread_join(tid[t]);
5914 if (t < ztest_opts.zo_datasets)
5915 ztest_dataset_close(t);
5918 txg_wait_synced(spa_get_dsl(spa), 0);
5920 zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(spa));
5921 zs->zs_space = metaslab_class_get_space(spa_normal_class(spa));
5923 umem_free(tid, ztest_opts.zo_threads * sizeof (kt_did_t));
5925 /* Kill the resume thread */
5926 ztest_exiting = B_TRUE;
5927 thread_join(resume_thread->t_tid);
5931 * Right before closing the pool, kick off a bunch of async I/O;
5932 * spa_close() should wait for it to complete.
5934 for (object = 1; object < 50; object++) {
5935 dmu_prefetch(spa->spa_meta_objset, object, 0, 0, 1ULL << 20,
5936 ZIO_PRIORITY_SYNC_READ);
5939 /* Verify that at least one commit cb was called in a timely fashion */
5940 if (zc_cb_counter >= ZTEST_COMMIT_CB_MIN_REG)
5941 VERIFY0(zc_min_txg_delay);
5943 spa_close(spa, FTAG);
5946 * Verify that we can loop over all pools.
5948 mutex_enter(&spa_namespace_lock);
5949 for (spa = spa_next(NULL); spa != NULL; spa = spa_next(spa))
5950 if (ztest_opts.zo_verbose > 3)
5951 (void) printf("spa_next: found %s\n", spa_name(spa));
5952 mutex_exit(&spa_namespace_lock);
5955 * Verify that we can export the pool and reimport it under a
5958 if (ztest_random(2) == 0) {
5959 char name[MAXNAMELEN];
5960 (void) snprintf(name, MAXNAMELEN, "%s_import",
5961 ztest_opts.zo_pool);
5962 ztest_spa_import_export(ztest_opts.zo_pool, name);
5963 ztest_spa_import_export(name, ztest_opts.zo_pool);
5968 list_destroy(&zcl.zcl_callbacks);
5969 mutex_destroy(&zcl.zcl_callbacks_lock);
5970 (void) rwlock_destroy(&ztest_name_lock);
5971 mutex_destroy(&ztest_vdev_lock);
5977 ztest_ds_t *zd = &ztest_ds[0];
5981 if (ztest_opts.zo_verbose >= 3)
5982 (void) printf("testing spa_freeze()...\n");
5984 kernel_init(FREAD | FWRITE);
5985 VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
5986 VERIFY3U(0, ==, ztest_dataset_open(0));
5987 spa->spa_debug = B_TRUE;
5991 * Force the first log block to be transactionally allocated.
5992 * We have to do this before we freeze the pool -- otherwise
5993 * the log chain won't be anchored.
5995 while (BP_IS_HOLE(&zd->zd_zilog->zl_header->zh_log)) {
5996 ztest_dmu_object_alloc_free(zd, 0);
5997 zil_commit(zd->zd_zilog, 0);
6000 txg_wait_synced(spa_get_dsl(spa), 0);
6003 * Freeze the pool. This stops spa_sync() from doing anything,
6004 * so that the only way to record changes from now on is the ZIL.
6009 * Because it is hard to predict how much space a write will actually
6010 * require beforehand, we leave ourselves some fudge space to write over
6013 uint64_t capacity = metaslab_class_get_space(spa_normal_class(spa)) / 2;
6016 * Run tests that generate log records but don't alter the pool config
6017 * or depend on DSL sync tasks (snapshots, objset create/destroy, etc).
6018 * We do a txg_wait_synced() after each iteration to force the txg
6019 * to increase well beyond the last synced value in the uberblock.
6020 * The ZIL should be OK with that.
6022 * Run a random number of times less than zo_maxloops and ensure we do
6023 * not run out of space on the pool.
6025 while (ztest_random(10) != 0 &&
6026 numloops++ < ztest_opts.zo_maxloops &&
6027 metaslab_class_get_alloc(spa_normal_class(spa)) < capacity) {
6029 ztest_od_init(&od, 0, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
6030 VERIFY0(ztest_object_init(zd, &od, sizeof (od), B_FALSE));
6031 ztest_io(zd, od.od_object,
6032 ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
6033 txg_wait_synced(spa_get_dsl(spa), 0);
6037 * Commit all of the changes we just generated.
6039 zil_commit(zd->zd_zilog, 0);
6040 txg_wait_synced(spa_get_dsl(spa), 0);
6043 * Close our dataset and close the pool.
6045 ztest_dataset_close(0);
6046 spa_close(spa, FTAG);
6050 * Open and close the pool and dataset to induce log replay.
6052 kernel_init(FREAD | FWRITE);
6053 VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
6054 ASSERT(spa_freeze_txg(spa) == UINT64_MAX);
6055 VERIFY3U(0, ==, ztest_dataset_open(0));
6056 ztest_dataset_close(0);
6058 spa->spa_debug = B_TRUE;
6060 txg_wait_synced(spa_get_dsl(spa), 0);
6061 ztest_reguid(NULL, 0);
6063 spa_close(spa, FTAG);
6068 print_time(hrtime_t t, char *timebuf)
6070 hrtime_t s = t / NANOSEC;
6071 hrtime_t m = s / 60;
6072 hrtime_t h = m / 60;
6073 hrtime_t d = h / 24;
6082 (void) sprintf(timebuf,
6083 "%llud%02lluh%02llum%02llus", d, h, m, s);
6085 (void) sprintf(timebuf, "%lluh%02llum%02llus", h, m, s);
6087 (void) sprintf(timebuf, "%llum%02llus", m, s);
6089 (void) sprintf(timebuf, "%llus", s);
6093 make_random_props(void)
6097 VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
6098 if (ztest_random(2) == 0)
6100 VERIFY(nvlist_add_uint64(props, "autoreplace", 1) == 0);
6106 * Create a storage pool with the given name and initial vdev size.
6107 * Then test spa_freeze() functionality.
6110 ztest_init(ztest_shared_t *zs)
6113 nvlist_t *nvroot, *props;
6116 mutex_init(&ztest_vdev_lock, NULL, MUTEX_DEFAULT, NULL);
6117 VERIFY(rwlock_init(&ztest_name_lock, USYNC_THREAD, NULL) == 0);
6119 kernel_init(FREAD | FWRITE);
6122 * Create the storage pool.
6124 (void) spa_destroy(ztest_opts.zo_pool);
6125 ztest_shared->zs_vdev_next_leaf = 0;
6127 zs->zs_mirrors = ztest_opts.zo_mirrors;
6128 nvroot = make_vdev_root(NULL, NULL, NULL, ztest_opts.zo_vdev_size, 0,
6129 0, ztest_opts.zo_raidz, zs->zs_mirrors, 1);
6130 props = make_random_props();
6131 for (i = 0; i < SPA_FEATURES; i++) {
6133 VERIFY3S(-1, !=, asprintf(&buf, "feature@%s",
6134 spa_feature_table[i].fi_uname));
6135 VERIFY3U(0, ==, nvlist_add_uint64(props, buf, 0));
6138 VERIFY3U(0, ==, spa_create(ztest_opts.zo_pool, nvroot, props, NULL));
6139 nvlist_free(nvroot);
6142 VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
6143 zs->zs_metaslab_sz =
6144 1ULL << spa->spa_root_vdev->vdev_child[0]->vdev_ms_shift;
6145 spa_close(spa, FTAG);
6149 ztest_run_zdb(ztest_opts.zo_pool);
6153 ztest_run_zdb(ztest_opts.zo_pool);
6155 (void) rwlock_destroy(&ztest_name_lock);
6156 mutex_destroy(&ztest_vdev_lock);
6162 static char ztest_name_data[] = "/tmp/ztest.data.XXXXXX";
6164 ztest_fd_data = mkstemp(ztest_name_data);
6165 ASSERT3S(ztest_fd_data, >=, 0);
6166 (void) unlink(ztest_name_data);
6170 shared_data_size(ztest_shared_hdr_t *hdr)
6174 size = hdr->zh_hdr_size;
6175 size += hdr->zh_opts_size;
6176 size += hdr->zh_size;
6177 size += hdr->zh_stats_size * hdr->zh_stats_count;
6178 size += hdr->zh_ds_size * hdr->zh_ds_count;
6187 ztest_shared_hdr_t *hdr;
6189 hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
6190 PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
6191 ASSERT(hdr != MAP_FAILED);
6193 VERIFY3U(0, ==, ftruncate(ztest_fd_data, sizeof (ztest_shared_hdr_t)));
6195 hdr->zh_hdr_size = sizeof (ztest_shared_hdr_t);
6196 hdr->zh_opts_size = sizeof (ztest_shared_opts_t);
6197 hdr->zh_size = sizeof (ztest_shared_t);
6198 hdr->zh_stats_size = sizeof (ztest_shared_callstate_t);
6199 hdr->zh_stats_count = ZTEST_FUNCS;
6200 hdr->zh_ds_size = sizeof (ztest_shared_ds_t);
6201 hdr->zh_ds_count = ztest_opts.zo_datasets;
6203 size = shared_data_size(hdr);
6204 VERIFY3U(0, ==, ftruncate(ztest_fd_data, size));
6206 (void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
6213 ztest_shared_hdr_t *hdr;
6216 hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
6217 PROT_READ, MAP_SHARED, ztest_fd_data, 0);
6218 ASSERT(hdr != MAP_FAILED);
6220 size = shared_data_size(hdr);
6222 (void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
6223 hdr = ztest_shared_hdr = (void *)mmap(0, P2ROUNDUP(size, getpagesize()),
6224 PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
6225 ASSERT(hdr != MAP_FAILED);
6226 buf = (uint8_t *)hdr;
6228 offset = hdr->zh_hdr_size;
6229 ztest_shared_opts = (void *)&buf[offset];
6230 offset += hdr->zh_opts_size;
6231 ztest_shared = (void *)&buf[offset];
6232 offset += hdr->zh_size;
6233 ztest_shared_callstate = (void *)&buf[offset];
6234 offset += hdr->zh_stats_size * hdr->zh_stats_count;
6235 ztest_shared_ds = (void *)&buf[offset];
6239 exec_child(char *cmd, char *libpath, boolean_t ignorekill, int *statusp)
6243 char *cmdbuf = NULL;
6248 cmdbuf = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
6249 (void) strlcpy(cmdbuf, getexecname(), MAXPATHLEN);
6254 fatal(1, "fork failed");
6256 if (pid == 0) { /* child */
6257 char *emptyargv[2] = { cmd, NULL };
6258 char fd_data_str[12];
6260 struct rlimit rl = { 1024, 1024 };
6261 (void) setrlimit(RLIMIT_NOFILE, &rl);
6263 (void) close(ztest_fd_rand);
6264 VERIFY(11 >= snprintf(fd_data_str, 12, "%d", ztest_fd_data));
6265 VERIFY(0 == setenv("ZTEST_FD_DATA", fd_data_str, 1));
6267 (void) enable_extended_FILE_stdio(-1, -1);
6268 if (libpath != NULL)
6269 VERIFY(0 == setenv("LD_LIBRARY_PATH", libpath, 1));
6270 (void) execv(cmd, emptyargv);
6271 ztest_dump_core = B_FALSE;
6272 fatal(B_TRUE, "exec failed: %s", cmd);
6275 if (cmdbuf != NULL) {
6276 umem_free(cmdbuf, MAXPATHLEN);
6280 while (waitpid(pid, &status, 0) != pid)
6282 if (statusp != NULL)
6285 if (WIFEXITED(status)) {
6286 if (WEXITSTATUS(status) != 0) {
6287 (void) fprintf(stderr, "child exited with code %d\n",
6288 WEXITSTATUS(status));
6292 } else if (WIFSIGNALED(status)) {
6293 if (!ignorekill || WTERMSIG(status) != SIGKILL) {
6294 (void) fprintf(stderr, "child died with signal %d\n",
6300 (void) fprintf(stderr, "something strange happened to child\n");
6307 ztest_run_init(void)
6311 ztest_shared_t *zs = ztest_shared;
6313 ASSERT(ztest_opts.zo_init != 0);
6316 * Blow away any existing copy of zpool.cache
6318 (void) remove(spa_config_path);
6321 * Create and initialize our storage pool.
6323 for (i = 1; i <= ztest_opts.zo_init; i++) {
6324 bzero(zs, sizeof (ztest_shared_t));
6325 if (ztest_opts.zo_verbose >= 3 &&
6326 ztest_opts.zo_init != 1) {
6327 (void) printf("ztest_init(), pass %d\n", i);
6334 main(int argc, char **argv)
6342 ztest_shared_callstate_t *zc;
6349 char *fd_data_str = getenv("ZTEST_FD_DATA");
6350 struct sigaction action;
6352 (void) setvbuf(stdout, NULL, _IOLBF, 0);
6354 dprintf_setup(&argc, argv);
6356 action.sa_handler = sig_handler;
6357 sigemptyset(&action.sa_mask);
6358 action.sa_flags = 0;
6360 if (sigaction(SIGSEGV, &action, NULL) < 0) {
6361 (void) fprintf(stderr, "ztest: cannot catch SIGSEGV: %s.\n",
6366 if (sigaction(SIGABRT, &action, NULL) < 0) {
6367 (void) fprintf(stderr, "ztest: cannot catch SIGABRT: %s.\n",
6372 ztest_fd_rand = open("/dev/urandom", O_RDONLY);
6373 ASSERT3S(ztest_fd_rand, >=, 0);
6376 process_options(argc, argv);
6381 bcopy(&ztest_opts, ztest_shared_opts,
6382 sizeof (*ztest_shared_opts));
6384 ztest_fd_data = atoi(fd_data_str);
6386 bcopy(ztest_shared_opts, &ztest_opts, sizeof (ztest_opts));
6388 ASSERT3U(ztest_opts.zo_datasets, ==, ztest_shared_hdr->zh_ds_count);
6390 /* Override location of zpool.cache */
6391 VERIFY(asprintf((char **)&spa_config_path, "%s/zpool.cache",
6392 ztest_opts.zo_dir) != -1);
6394 ztest_ds = umem_alloc(ztest_opts.zo_datasets * sizeof (ztest_ds_t),
6399 metaslab_gang_bang = ztest_opts.zo_metaslab_gang_bang;
6400 metaslab_df_alloc_threshold =
6401 zs->zs_metaslab_df_alloc_threshold;
6410 hasalt = (strlen(ztest_opts.zo_alt_ztest) != 0);
6412 if (ztest_opts.zo_verbose >= 1) {
6413 (void) printf("%llu vdevs, %d datasets, %d threads,"
6414 " %llu seconds...\n",
6415 (u_longlong_t)ztest_opts.zo_vdevs,
6416 ztest_opts.zo_datasets,
6417 ztest_opts.zo_threads,
6418 (u_longlong_t)ztest_opts.zo_time);
6421 cmd = umem_alloc(MAXNAMELEN, UMEM_NOFAIL);
6422 (void) strlcpy(cmd, getexecname(), MAXNAMELEN);
6424 zs->zs_do_init = B_TRUE;
6425 if (strlen(ztest_opts.zo_alt_ztest) != 0) {
6426 if (ztest_opts.zo_verbose >= 1) {
6427 (void) printf("Executing older ztest for "
6428 "initialization: %s\n", ztest_opts.zo_alt_ztest);
6430 VERIFY(!exec_child(ztest_opts.zo_alt_ztest,
6431 ztest_opts.zo_alt_libpath, B_FALSE, NULL));
6433 VERIFY(!exec_child(NULL, NULL, B_FALSE, NULL));
6435 zs->zs_do_init = B_FALSE;
6437 zs->zs_proc_start = gethrtime();
6438 zs->zs_proc_stop = zs->zs_proc_start + ztest_opts.zo_time * NANOSEC;
6440 for (f = 0; f < ZTEST_FUNCS; f++) {
6441 zi = &ztest_info[f];
6442 zc = ZTEST_GET_SHARED_CALLSTATE(f);
6443 if (zs->zs_proc_start + zi->zi_interval[0] > zs->zs_proc_stop)
6444 zc->zc_next = UINT64_MAX;
6446 zc->zc_next = zs->zs_proc_start +
6447 ztest_random(2 * zi->zi_interval[0] + 1);
6451 * Run the tests in a loop. These tests include fault injection
6452 * to verify that self-healing data works, and forced crashes
6453 * to verify that we never lose on-disk consistency.
6455 while (gethrtime() < zs->zs_proc_stop) {
6460 * Initialize the workload counters for each function.
6462 for (f = 0; f < ZTEST_FUNCS; f++) {
6463 zc = ZTEST_GET_SHARED_CALLSTATE(f);
6468 /* Set the allocation switch size */
6469 zs->zs_metaslab_df_alloc_threshold =
6470 ztest_random(zs->zs_metaslab_sz / 4) + 1;
6472 if (!hasalt || ztest_random(2) == 0) {
6473 if (hasalt && ztest_opts.zo_verbose >= 1) {
6474 (void) printf("Executing newer ztest: %s\n",
6478 killed = exec_child(cmd, NULL, B_TRUE, &status);
6480 if (hasalt && ztest_opts.zo_verbose >= 1) {
6481 (void) printf("Executing older ztest: %s\n",
6482 ztest_opts.zo_alt_ztest);
6485 killed = exec_child(ztest_opts.zo_alt_ztest,
6486 ztest_opts.zo_alt_libpath, B_TRUE, &status);
6493 if (ztest_opts.zo_verbose >= 1) {
6494 hrtime_t now = gethrtime();
6496 now = MIN(now, zs->zs_proc_stop);
6497 print_time(zs->zs_proc_stop - now, timebuf);
6498 nicenum(zs->zs_space, numbuf);
6500 (void) printf("Pass %3d, %8s, %3llu ENOSPC, "
6501 "%4.1f%% of %5s used, %3.0f%% done, %8s to go\n",
6503 WIFEXITED(status) ? "Complete" : "SIGKILL",
6504 (u_longlong_t)zs->zs_enospc_count,
6505 100.0 * zs->zs_alloc / zs->zs_space,
6507 100.0 * (now - zs->zs_proc_start) /
6508 (ztest_opts.zo_time * NANOSEC), timebuf);
6511 if (ztest_opts.zo_verbose >= 2) {
6512 (void) printf("\nWorkload summary:\n\n");
6513 (void) printf("%7s %9s %s\n",
6514 "Calls", "Time", "Function");
6515 (void) printf("%7s %9s %s\n",
6516 "-----", "----", "--------");
6517 for (f = 0; f < ZTEST_FUNCS; f++) {
6518 zi = &ztest_info[f];
6519 zc = ZTEST_GET_SHARED_CALLSTATE(f);
6520 print_time(zc->zc_time, timebuf);
6521 (void) printf("%7llu %9s %s\n",
6522 (u_longlong_t)zc->zc_count, timebuf,
6525 (void) printf("\n");
6529 * It's possible that we killed a child during a rename test,
6530 * in which case we'll have a 'ztest_tmp' pool lying around
6531 * instead of 'ztest'. Do a blind rename in case this happened.
6534 if (spa_open(ztest_opts.zo_pool, &spa, FTAG) == 0) {
6535 spa_close(spa, FTAG);
6537 char tmpname[MAXNAMELEN];
6539 kernel_init(FREAD | FWRITE);
6540 (void) snprintf(tmpname, sizeof (tmpname), "%s_tmp",
6541 ztest_opts.zo_pool);
6542 (void) spa_rename(tmpname, ztest_opts.zo_pool);
6546 ztest_run_zdb(ztest_opts.zo_pool);
6549 if (ztest_opts.zo_verbose >= 1) {
6551 (void) printf("%d runs of older ztest: %s\n", older,
6552 ztest_opts.zo_alt_ztest);
6553 (void) printf("%d runs of newer ztest: %s\n", newer,
6556 (void) printf("%d killed, %d completed, %.0f%% kill rate\n",
6557 kills, iters - kills, (100.0 * kills) / MAX(1, iters));
6560 umem_free(cmd, MAXNAMELEN);