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,
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 if (ztest_opts.zo_verbose >= 3)
831 zfs_dbgmsg_print(FTAG);
833 (void) kill(getpid(), SIGKILL);
837 ztest_random(uint64_t range)
841 ASSERT3S(ztest_fd_rand, >=, 0);
846 if (read(ztest_fd_rand, &r, sizeof (r)) != sizeof (r))
847 fatal(1, "short read from /dev/urandom");
854 ztest_record_enospc(const char *s)
856 ztest_shared->zs_enospc_count++;
860 ztest_get_ashift(void)
862 if (ztest_opts.zo_ashift == 0)
863 return (SPA_MINBLOCKSHIFT + ztest_random(5));
864 return (ztest_opts.zo_ashift);
868 make_vdev_file(char *path, char *aux, char *pool, size_t size, uint64_t ashift)
874 pathbuf = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
877 ashift = ztest_get_ashift();
883 vdev = ztest_shared->zs_vdev_aux;
884 (void) snprintf(path, MAXPATHLEN,
885 ztest_aux_template, ztest_opts.zo_dir,
886 pool == NULL ? ztest_opts.zo_pool : pool,
889 vdev = ztest_shared->zs_vdev_next_leaf++;
890 (void) snprintf(path, MAXPATHLEN,
891 ztest_dev_template, ztest_opts.zo_dir,
892 pool == NULL ? ztest_opts.zo_pool : pool, vdev);
897 int fd = open(path, O_RDWR | O_CREAT | O_TRUNC, 0666);
899 fatal(1, "can't open %s", path);
900 if (ftruncate(fd, size) != 0)
901 fatal(1, "can't ftruncate %s", path);
905 VERIFY(nvlist_alloc(&file, NV_UNIQUE_NAME, 0) == 0);
906 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_TYPE, VDEV_TYPE_FILE) == 0);
907 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_PATH, path) == 0);
908 VERIFY(nvlist_add_uint64(file, ZPOOL_CONFIG_ASHIFT, ashift) == 0);
909 umem_free(pathbuf, MAXPATHLEN);
915 make_vdev_raidz(char *path, char *aux, char *pool, size_t size,
916 uint64_t ashift, int r)
918 nvlist_t *raidz, **child;
922 return (make_vdev_file(path, aux, pool, size, ashift));
923 child = umem_alloc(r * sizeof (nvlist_t *), UMEM_NOFAIL);
925 for (c = 0; c < r; c++)
926 child[c] = make_vdev_file(path, aux, pool, size, ashift);
928 VERIFY(nvlist_alloc(&raidz, NV_UNIQUE_NAME, 0) == 0);
929 VERIFY(nvlist_add_string(raidz, ZPOOL_CONFIG_TYPE,
930 VDEV_TYPE_RAIDZ) == 0);
931 VERIFY(nvlist_add_uint64(raidz, ZPOOL_CONFIG_NPARITY,
932 ztest_opts.zo_raidz_parity) == 0);
933 VERIFY(nvlist_add_nvlist_array(raidz, ZPOOL_CONFIG_CHILDREN,
936 for (c = 0; c < r; c++)
937 nvlist_free(child[c]);
939 umem_free(child, r * sizeof (nvlist_t *));
945 make_vdev_mirror(char *path, char *aux, char *pool, size_t size,
946 uint64_t ashift, int r, int m)
948 nvlist_t *mirror, **child;
952 return (make_vdev_raidz(path, aux, pool, size, ashift, r));
954 child = umem_alloc(m * sizeof (nvlist_t *), UMEM_NOFAIL);
956 for (c = 0; c < m; c++)
957 child[c] = make_vdev_raidz(path, aux, pool, size, ashift, r);
959 VERIFY(nvlist_alloc(&mirror, NV_UNIQUE_NAME, 0) == 0);
960 VERIFY(nvlist_add_string(mirror, ZPOOL_CONFIG_TYPE,
961 VDEV_TYPE_MIRROR) == 0);
962 VERIFY(nvlist_add_nvlist_array(mirror, ZPOOL_CONFIG_CHILDREN,
965 for (c = 0; c < m; c++)
966 nvlist_free(child[c]);
968 umem_free(child, m * sizeof (nvlist_t *));
974 make_vdev_root(char *path, char *aux, char *pool, size_t size, uint64_t ashift,
975 int log, int r, int m, int t)
977 nvlist_t *root, **child;
982 child = umem_alloc(t * sizeof (nvlist_t *), UMEM_NOFAIL);
984 for (c = 0; c < t; c++) {
985 child[c] = make_vdev_mirror(path, aux, pool, size, ashift,
987 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_IS_LOG,
991 VERIFY(nvlist_alloc(&root, NV_UNIQUE_NAME, 0) == 0);
992 VERIFY(nvlist_add_string(root, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) == 0);
993 VERIFY(nvlist_add_nvlist_array(root, aux ? aux : ZPOOL_CONFIG_CHILDREN,
996 for (c = 0; c < t; c++)
997 nvlist_free(child[c]);
999 umem_free(child, t * sizeof (nvlist_t *));
1005 * Find a random spa version. Returns back a random spa version in the
1006 * range [initial_version, SPA_VERSION_FEATURES].
1009 ztest_random_spa_version(uint64_t initial_version)
1011 uint64_t version = initial_version;
1013 if (version <= SPA_VERSION_BEFORE_FEATURES) {
1015 ztest_random(SPA_VERSION_BEFORE_FEATURES - version + 1);
1018 if (version > SPA_VERSION_BEFORE_FEATURES)
1019 version = SPA_VERSION_FEATURES;
1021 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
1026 ztest_random_blocksize(void)
1029 * Choose a block size >= the ashift.
1030 * If the SPA supports new MAXBLOCKSIZE, test up to 1MB blocks.
1032 int maxbs = SPA_OLD_MAXBLOCKSHIFT;
1033 if (spa_maxblocksize(ztest_spa) == SPA_MAXBLOCKSIZE)
1035 uint64_t block_shift =
1036 ztest_random(maxbs - ztest_spa->spa_max_ashift + 1);
1037 return (1 << (SPA_MINBLOCKSHIFT + block_shift));
1041 ztest_random_ibshift(void)
1043 return (DN_MIN_INDBLKSHIFT +
1044 ztest_random(DN_MAX_INDBLKSHIFT - DN_MIN_INDBLKSHIFT + 1));
1048 ztest_random_vdev_top(spa_t *spa, boolean_t log_ok)
1051 vdev_t *rvd = spa->spa_root_vdev;
1054 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1057 top = ztest_random(rvd->vdev_children);
1058 tvd = rvd->vdev_child[top];
1059 } while (tvd->vdev_ishole || (tvd->vdev_islog && !log_ok) ||
1060 tvd->vdev_mg == NULL || tvd->vdev_mg->mg_class == NULL);
1066 ztest_random_dsl_prop(zfs_prop_t prop)
1071 value = zfs_prop_random_value(prop, ztest_random(-1ULL));
1072 } while (prop == ZFS_PROP_CHECKSUM && value == ZIO_CHECKSUM_OFF);
1078 ztest_dsl_prop_set_uint64(char *osname, zfs_prop_t prop, uint64_t value,
1081 const char *propname = zfs_prop_to_name(prop);
1082 const char *valname;
1087 error = dsl_prop_set_int(osname, propname,
1088 (inherit ? ZPROP_SRC_NONE : ZPROP_SRC_LOCAL), value);
1090 if (error == ENOSPC) {
1091 ztest_record_enospc(FTAG);
1096 setpoint = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
1097 VERIFY0(dsl_prop_get_integer(osname, propname, &curval, setpoint));
1099 if (ztest_opts.zo_verbose >= 6) {
1102 err = zfs_prop_index_to_string(prop, curval, &valname);
1104 (void) printf("%s %s = %llu at '%s'\n",
1105 osname, propname, (unsigned long long)curval,
1108 (void) printf("%s %s = %s at '%s'\n",
1109 osname, propname, valname, setpoint);
1111 umem_free(setpoint, MAXPATHLEN);
1117 ztest_spa_prop_set_uint64(zpool_prop_t prop, uint64_t value)
1119 spa_t *spa = ztest_spa;
1120 nvlist_t *props = NULL;
1123 VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
1124 VERIFY(nvlist_add_uint64(props, zpool_prop_to_name(prop), value) == 0);
1126 error = spa_prop_set(spa, props);
1130 if (error == ENOSPC) {
1131 ztest_record_enospc(FTAG);
1140 ztest_rll_init(rll_t *rll)
1142 rll->rll_writer = NULL;
1143 rll->rll_readers = 0;
1144 mutex_init(&rll->rll_lock, NULL, MUTEX_DEFAULT, NULL);
1145 cv_init(&rll->rll_cv, NULL, CV_DEFAULT, NULL);
1149 ztest_rll_destroy(rll_t *rll)
1151 ASSERT(rll->rll_writer == NULL);
1152 ASSERT(rll->rll_readers == 0);
1153 mutex_destroy(&rll->rll_lock);
1154 cv_destroy(&rll->rll_cv);
1158 ztest_rll_lock(rll_t *rll, rl_type_t type)
1160 mutex_enter(&rll->rll_lock);
1162 if (type == RL_READER) {
1163 while (rll->rll_writer != NULL)
1164 (void) cv_wait(&rll->rll_cv, &rll->rll_lock);
1167 while (rll->rll_writer != NULL || rll->rll_readers)
1168 (void) cv_wait(&rll->rll_cv, &rll->rll_lock);
1169 rll->rll_writer = curthread;
1172 mutex_exit(&rll->rll_lock);
1176 ztest_rll_unlock(rll_t *rll)
1178 mutex_enter(&rll->rll_lock);
1180 if (rll->rll_writer) {
1181 ASSERT(rll->rll_readers == 0);
1182 rll->rll_writer = NULL;
1184 ASSERT(rll->rll_readers != 0);
1185 ASSERT(rll->rll_writer == NULL);
1189 if (rll->rll_writer == NULL && rll->rll_readers == 0)
1190 cv_broadcast(&rll->rll_cv);
1192 mutex_exit(&rll->rll_lock);
1196 ztest_object_lock(ztest_ds_t *zd, uint64_t object, rl_type_t type)
1198 rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1200 ztest_rll_lock(rll, type);
1204 ztest_object_unlock(ztest_ds_t *zd, uint64_t object)
1206 rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1208 ztest_rll_unlock(rll);
1212 ztest_range_lock(ztest_ds_t *zd, uint64_t object, uint64_t offset,
1213 uint64_t size, rl_type_t type)
1215 uint64_t hash = object ^ (offset % (ZTEST_RANGE_LOCKS + 1));
1216 rll_t *rll = &zd->zd_range_lock[hash & (ZTEST_RANGE_LOCKS - 1)];
1219 rl = umem_alloc(sizeof (*rl), UMEM_NOFAIL);
1220 rl->rl_object = object;
1221 rl->rl_offset = offset;
1225 ztest_rll_lock(rll, type);
1231 ztest_range_unlock(rl_t *rl)
1233 rll_t *rll = rl->rl_lock;
1235 ztest_rll_unlock(rll);
1237 umem_free(rl, sizeof (*rl));
1241 ztest_zd_init(ztest_ds_t *zd, ztest_shared_ds_t *szd, objset_t *os)
1244 zd->zd_zilog = dmu_objset_zil(os);
1245 zd->zd_shared = szd;
1246 dmu_objset_name(os, zd->zd_name);
1249 if (zd->zd_shared != NULL)
1250 zd->zd_shared->zd_seq = 0;
1252 VERIFY(rwlock_init(&zd->zd_zilog_lock, USYNC_THREAD, NULL) == 0);
1253 mutex_init(&zd->zd_dirobj_lock, NULL, MUTEX_DEFAULT, NULL);
1255 for (l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1256 ztest_rll_init(&zd->zd_object_lock[l]);
1258 for (l = 0; l < ZTEST_RANGE_LOCKS; l++)
1259 ztest_rll_init(&zd->zd_range_lock[l]);
1263 ztest_zd_fini(ztest_ds_t *zd)
1267 mutex_destroy(&zd->zd_dirobj_lock);
1268 (void) rwlock_destroy(&zd->zd_zilog_lock);
1270 for (l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1271 ztest_rll_destroy(&zd->zd_object_lock[l]);
1273 for (l = 0; l < ZTEST_RANGE_LOCKS; l++)
1274 ztest_rll_destroy(&zd->zd_range_lock[l]);
1277 #define TXG_MIGHTWAIT (ztest_random(10) == 0 ? TXG_NOWAIT : TXG_WAIT)
1280 ztest_tx_assign(dmu_tx_t *tx, uint64_t txg_how, const char *tag)
1286 * Attempt to assign tx to some transaction group.
1288 error = dmu_tx_assign(tx, txg_how);
1290 if (error == ERESTART) {
1291 ASSERT(txg_how == TXG_NOWAIT);
1294 ASSERT3U(error, ==, ENOSPC);
1295 ztest_record_enospc(tag);
1300 txg = dmu_tx_get_txg(tx);
1306 ztest_pattern_set(void *buf, uint64_t size, uint64_t value)
1309 uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1317 ztest_pattern_match(void *buf, uint64_t size, uint64_t value)
1320 uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1324 diff |= (value - *ip++);
1331 ztest_bt_generate(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1332 uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg)
1334 bt->bt_magic = BT_MAGIC;
1335 bt->bt_objset = dmu_objset_id(os);
1336 bt->bt_object = object;
1337 bt->bt_offset = offset;
1340 bt->bt_crtxg = crtxg;
1344 ztest_bt_verify(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1345 uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg)
1347 ASSERT3U(bt->bt_magic, ==, BT_MAGIC);
1348 ASSERT3U(bt->bt_objset, ==, dmu_objset_id(os));
1349 ASSERT3U(bt->bt_object, ==, object);
1350 ASSERT3U(bt->bt_offset, ==, offset);
1351 ASSERT3U(bt->bt_gen, <=, gen);
1352 ASSERT3U(bt->bt_txg, <=, txg);
1353 ASSERT3U(bt->bt_crtxg, ==, crtxg);
1356 static ztest_block_tag_t *
1357 ztest_bt_bonus(dmu_buf_t *db)
1359 dmu_object_info_t doi;
1360 ztest_block_tag_t *bt;
1362 dmu_object_info_from_db(db, &doi);
1363 ASSERT3U(doi.doi_bonus_size, <=, db->db_size);
1364 ASSERT3U(doi.doi_bonus_size, >=, sizeof (*bt));
1365 bt = (void *)((char *)db->db_data + doi.doi_bonus_size - sizeof (*bt));
1374 #define lrz_type lr_mode
1375 #define lrz_blocksize lr_uid
1376 #define lrz_ibshift lr_gid
1377 #define lrz_bonustype lr_rdev
1378 #define lrz_bonuslen lr_crtime[1]
1381 ztest_log_create(ztest_ds_t *zd, dmu_tx_t *tx, lr_create_t *lr)
1383 char *name = (void *)(lr + 1); /* name follows lr */
1384 size_t namesize = strlen(name) + 1;
1387 if (zil_replaying(zd->zd_zilog, tx))
1390 itx = zil_itx_create(TX_CREATE, sizeof (*lr) + namesize);
1391 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1392 sizeof (*lr) + namesize - sizeof (lr_t));
1394 zil_itx_assign(zd->zd_zilog, itx, tx);
1398 ztest_log_remove(ztest_ds_t *zd, dmu_tx_t *tx, lr_remove_t *lr, uint64_t object)
1400 char *name = (void *)(lr + 1); /* name follows lr */
1401 size_t namesize = strlen(name) + 1;
1404 if (zil_replaying(zd->zd_zilog, tx))
1407 itx = zil_itx_create(TX_REMOVE, sizeof (*lr) + namesize);
1408 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1409 sizeof (*lr) + namesize - sizeof (lr_t));
1411 itx->itx_oid = object;
1412 zil_itx_assign(zd->zd_zilog, itx, tx);
1416 ztest_log_write(ztest_ds_t *zd, dmu_tx_t *tx, lr_write_t *lr)
1419 itx_wr_state_t write_state = ztest_random(WR_NUM_STATES);
1421 if (zil_replaying(zd->zd_zilog, tx))
1424 if (lr->lr_length > ZIL_MAX_LOG_DATA)
1425 write_state = WR_INDIRECT;
1427 itx = zil_itx_create(TX_WRITE,
1428 sizeof (*lr) + (write_state == WR_COPIED ? lr->lr_length : 0));
1430 if (write_state == WR_COPIED &&
1431 dmu_read(zd->zd_os, lr->lr_foid, lr->lr_offset, lr->lr_length,
1432 ((lr_write_t *)&itx->itx_lr) + 1, DMU_READ_NO_PREFETCH) != 0) {
1433 zil_itx_destroy(itx);
1434 itx = zil_itx_create(TX_WRITE, sizeof (*lr));
1435 write_state = WR_NEED_COPY;
1437 itx->itx_private = zd;
1438 itx->itx_wr_state = write_state;
1439 itx->itx_sync = (ztest_random(8) == 0);
1440 itx->itx_sod += (write_state == WR_NEED_COPY ? lr->lr_length : 0);
1442 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1443 sizeof (*lr) - sizeof (lr_t));
1445 zil_itx_assign(zd->zd_zilog, itx, tx);
1449 ztest_log_truncate(ztest_ds_t *zd, dmu_tx_t *tx, lr_truncate_t *lr)
1453 if (zil_replaying(zd->zd_zilog, tx))
1456 itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr));
1457 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1458 sizeof (*lr) - sizeof (lr_t));
1460 itx->itx_sync = B_FALSE;
1461 zil_itx_assign(zd->zd_zilog, itx, tx);
1465 ztest_log_setattr(ztest_ds_t *zd, dmu_tx_t *tx, lr_setattr_t *lr)
1469 if (zil_replaying(zd->zd_zilog, tx))
1472 itx = zil_itx_create(TX_SETATTR, sizeof (*lr));
1473 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1474 sizeof (*lr) - sizeof (lr_t));
1476 itx->itx_sync = B_FALSE;
1477 zil_itx_assign(zd->zd_zilog, itx, tx);
1484 ztest_replay_create(ztest_ds_t *zd, lr_create_t *lr, boolean_t byteswap)
1486 char *name = (void *)(lr + 1); /* name follows lr */
1487 objset_t *os = zd->zd_os;
1488 ztest_block_tag_t *bbt;
1495 byteswap_uint64_array(lr, sizeof (*lr));
1497 ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1498 ASSERT(name[0] != '\0');
1500 tx = dmu_tx_create(os);
1502 dmu_tx_hold_zap(tx, lr->lr_doid, B_TRUE, name);
1504 if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1505 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, B_TRUE, NULL);
1507 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1510 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1514 ASSERT(dmu_objset_zil(os)->zl_replay == !!lr->lr_foid);
1516 if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1517 if (lr->lr_foid == 0) {
1518 lr->lr_foid = zap_create(os,
1519 lr->lrz_type, lr->lrz_bonustype,
1520 lr->lrz_bonuslen, tx);
1522 error = zap_create_claim(os, lr->lr_foid,
1523 lr->lrz_type, lr->lrz_bonustype,
1524 lr->lrz_bonuslen, tx);
1527 if (lr->lr_foid == 0) {
1528 lr->lr_foid = dmu_object_alloc(os,
1529 lr->lrz_type, 0, lr->lrz_bonustype,
1530 lr->lrz_bonuslen, tx);
1532 error = dmu_object_claim(os, lr->lr_foid,
1533 lr->lrz_type, 0, lr->lrz_bonustype,
1534 lr->lrz_bonuslen, tx);
1539 ASSERT3U(error, ==, EEXIST);
1540 ASSERT(zd->zd_zilog->zl_replay);
1545 ASSERT(lr->lr_foid != 0);
1547 if (lr->lrz_type != DMU_OT_ZAP_OTHER)
1548 VERIFY3U(0, ==, dmu_object_set_blocksize(os, lr->lr_foid,
1549 lr->lrz_blocksize, lr->lrz_ibshift, tx));
1551 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1552 bbt = ztest_bt_bonus(db);
1553 dmu_buf_will_dirty(db, tx);
1554 ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_gen, txg, txg);
1555 dmu_buf_rele(db, FTAG);
1557 VERIFY3U(0, ==, zap_add(os, lr->lr_doid, name, sizeof (uint64_t), 1,
1560 (void) ztest_log_create(zd, tx, lr);
1568 ztest_replay_remove(ztest_ds_t *zd, lr_remove_t *lr, boolean_t byteswap)
1570 char *name = (void *)(lr + 1); /* name follows lr */
1571 objset_t *os = zd->zd_os;
1572 dmu_object_info_t doi;
1574 uint64_t object, txg;
1577 byteswap_uint64_array(lr, sizeof (*lr));
1579 ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1580 ASSERT(name[0] != '\0');
1583 zap_lookup(os, lr->lr_doid, name, sizeof (object), 1, &object));
1584 ASSERT(object != 0);
1586 ztest_object_lock(zd, object, RL_WRITER);
1588 VERIFY3U(0, ==, dmu_object_info(os, object, &doi));
1590 tx = dmu_tx_create(os);
1592 dmu_tx_hold_zap(tx, lr->lr_doid, B_FALSE, name);
1593 dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END);
1595 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1597 ztest_object_unlock(zd, object);
1601 if (doi.doi_type == DMU_OT_ZAP_OTHER) {
1602 VERIFY3U(0, ==, zap_destroy(os, object, tx));
1604 VERIFY3U(0, ==, dmu_object_free(os, object, tx));
1607 VERIFY3U(0, ==, zap_remove(os, lr->lr_doid, name, tx));
1609 (void) ztest_log_remove(zd, tx, lr, object);
1613 ztest_object_unlock(zd, object);
1619 ztest_replay_write(ztest_ds_t *zd, lr_write_t *lr, boolean_t byteswap)
1621 objset_t *os = zd->zd_os;
1622 void *data = lr + 1; /* data follows lr */
1623 uint64_t offset, length;
1624 ztest_block_tag_t *bt = data;
1625 ztest_block_tag_t *bbt;
1626 uint64_t gen, txg, lrtxg, crtxg;
1627 dmu_object_info_t doi;
1630 arc_buf_t *abuf = NULL;
1634 byteswap_uint64_array(lr, sizeof (*lr));
1636 offset = lr->lr_offset;
1637 length = lr->lr_length;
1639 /* If it's a dmu_sync() block, write the whole block */
1640 if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
1641 uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
1642 if (length < blocksize) {
1643 offset -= offset % blocksize;
1648 if (bt->bt_magic == BSWAP_64(BT_MAGIC))
1649 byteswap_uint64_array(bt, sizeof (*bt));
1651 if (bt->bt_magic != BT_MAGIC)
1654 ztest_object_lock(zd, lr->lr_foid, RL_READER);
1655 rl = ztest_range_lock(zd, lr->lr_foid, offset, length, RL_WRITER);
1657 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1659 dmu_object_info_from_db(db, &doi);
1661 bbt = ztest_bt_bonus(db);
1662 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1664 crtxg = bbt->bt_crtxg;
1665 lrtxg = lr->lr_common.lrc_txg;
1667 tx = dmu_tx_create(os);
1669 dmu_tx_hold_write(tx, lr->lr_foid, offset, length);
1671 if (ztest_random(8) == 0 && length == doi.doi_data_block_size &&
1672 P2PHASE(offset, length) == 0)
1673 abuf = dmu_request_arcbuf(db, length);
1675 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1678 dmu_return_arcbuf(abuf);
1679 dmu_buf_rele(db, FTAG);
1680 ztest_range_unlock(rl);
1681 ztest_object_unlock(zd, lr->lr_foid);
1687 * Usually, verify the old data before writing new data --
1688 * but not always, because we also want to verify correct
1689 * behavior when the data was not recently read into cache.
1691 ASSERT(offset % doi.doi_data_block_size == 0);
1692 if (ztest_random(4) != 0) {
1693 int prefetch = ztest_random(2) ?
1694 DMU_READ_PREFETCH : DMU_READ_NO_PREFETCH;
1695 ztest_block_tag_t rbt;
1697 VERIFY(dmu_read(os, lr->lr_foid, offset,
1698 sizeof (rbt), &rbt, prefetch) == 0);
1699 if (rbt.bt_magic == BT_MAGIC) {
1700 ztest_bt_verify(&rbt, os, lr->lr_foid,
1701 offset, gen, txg, crtxg);
1706 * Writes can appear to be newer than the bonus buffer because
1707 * the ztest_get_data() callback does a dmu_read() of the
1708 * open-context data, which may be different than the data
1709 * as it was when the write was generated.
1711 if (zd->zd_zilog->zl_replay) {
1712 ztest_bt_verify(bt, os, lr->lr_foid, offset,
1713 MAX(gen, bt->bt_gen), MAX(txg, lrtxg),
1718 * Set the bt's gen/txg to the bonus buffer's gen/txg
1719 * so that all of the usual ASSERTs will work.
1721 ztest_bt_generate(bt, os, lr->lr_foid, offset, gen, txg, crtxg);
1725 dmu_write(os, lr->lr_foid, offset, length, data, tx);
1727 bcopy(data, abuf->b_data, length);
1728 dmu_assign_arcbuf(db, offset, abuf, tx);
1731 (void) ztest_log_write(zd, tx, lr);
1733 dmu_buf_rele(db, FTAG);
1737 ztest_range_unlock(rl);
1738 ztest_object_unlock(zd, lr->lr_foid);
1744 ztest_replay_truncate(ztest_ds_t *zd, lr_truncate_t *lr, boolean_t byteswap)
1746 objset_t *os = zd->zd_os;
1752 byteswap_uint64_array(lr, sizeof (*lr));
1754 ztest_object_lock(zd, lr->lr_foid, RL_READER);
1755 rl = ztest_range_lock(zd, lr->lr_foid, lr->lr_offset, lr->lr_length,
1758 tx = dmu_tx_create(os);
1760 dmu_tx_hold_free(tx, lr->lr_foid, lr->lr_offset, lr->lr_length);
1762 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1764 ztest_range_unlock(rl);
1765 ztest_object_unlock(zd, lr->lr_foid);
1769 VERIFY(dmu_free_range(os, lr->lr_foid, lr->lr_offset,
1770 lr->lr_length, tx) == 0);
1772 (void) ztest_log_truncate(zd, tx, lr);
1776 ztest_range_unlock(rl);
1777 ztest_object_unlock(zd, lr->lr_foid);
1783 ztest_replay_setattr(ztest_ds_t *zd, lr_setattr_t *lr, boolean_t byteswap)
1785 objset_t *os = zd->zd_os;
1788 ztest_block_tag_t *bbt;
1789 uint64_t txg, lrtxg, crtxg;
1792 byteswap_uint64_array(lr, sizeof (*lr));
1794 ztest_object_lock(zd, lr->lr_foid, RL_WRITER);
1796 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1798 tx = dmu_tx_create(os);
1799 dmu_tx_hold_bonus(tx, lr->lr_foid);
1801 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1803 dmu_buf_rele(db, FTAG);
1804 ztest_object_unlock(zd, lr->lr_foid);
1808 bbt = ztest_bt_bonus(db);
1809 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1810 crtxg = bbt->bt_crtxg;
1811 lrtxg = lr->lr_common.lrc_txg;
1813 if (zd->zd_zilog->zl_replay) {
1814 ASSERT(lr->lr_size != 0);
1815 ASSERT(lr->lr_mode != 0);
1819 * Randomly change the size and increment the generation.
1821 lr->lr_size = (ztest_random(db->db_size / sizeof (*bbt)) + 1) *
1823 lr->lr_mode = bbt->bt_gen + 1;
1828 * Verify that the current bonus buffer is not newer than our txg.
1830 ztest_bt_verify(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode,
1831 MAX(txg, lrtxg), crtxg);
1833 dmu_buf_will_dirty(db, tx);
1835 ASSERT3U(lr->lr_size, >=, sizeof (*bbt));
1836 ASSERT3U(lr->lr_size, <=, db->db_size);
1837 VERIFY0(dmu_set_bonus(db, lr->lr_size, tx));
1838 bbt = ztest_bt_bonus(db);
1840 ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode, txg, crtxg);
1842 dmu_buf_rele(db, FTAG);
1844 (void) ztest_log_setattr(zd, tx, lr);
1848 ztest_object_unlock(zd, lr->lr_foid);
1853 zil_replay_func_t ztest_replay_vector[TX_MAX_TYPE] = {
1854 NULL, /* 0 no such transaction type */
1855 (zil_replay_func_t)ztest_replay_create, /* TX_CREATE */
1856 NULL, /* TX_MKDIR */
1857 NULL, /* TX_MKXATTR */
1858 NULL, /* TX_SYMLINK */
1859 (zil_replay_func_t)ztest_replay_remove, /* TX_REMOVE */
1860 NULL, /* TX_RMDIR */
1862 NULL, /* TX_RENAME */
1863 (zil_replay_func_t)ztest_replay_write, /* TX_WRITE */
1864 (zil_replay_func_t)ztest_replay_truncate, /* TX_TRUNCATE */
1865 (zil_replay_func_t)ztest_replay_setattr, /* TX_SETATTR */
1867 NULL, /* TX_CREATE_ACL */
1868 NULL, /* TX_CREATE_ATTR */
1869 NULL, /* TX_CREATE_ACL_ATTR */
1870 NULL, /* TX_MKDIR_ACL */
1871 NULL, /* TX_MKDIR_ATTR */
1872 NULL, /* TX_MKDIR_ACL_ATTR */
1873 NULL, /* TX_WRITE2 */
1877 * ZIL get_data callbacks
1881 ztest_get_done(zgd_t *zgd, int error)
1883 ztest_ds_t *zd = zgd->zgd_private;
1884 uint64_t object = zgd->zgd_rl->rl_object;
1887 dmu_buf_rele(zgd->zgd_db, zgd);
1889 ztest_range_unlock(zgd->zgd_rl);
1890 ztest_object_unlock(zd, object);
1892 if (error == 0 && zgd->zgd_bp)
1893 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
1895 umem_free(zgd, sizeof (*zgd));
1899 ztest_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
1901 ztest_ds_t *zd = arg;
1902 objset_t *os = zd->zd_os;
1903 uint64_t object = lr->lr_foid;
1904 uint64_t offset = lr->lr_offset;
1905 uint64_t size = lr->lr_length;
1906 blkptr_t *bp = &lr->lr_blkptr;
1907 uint64_t txg = lr->lr_common.lrc_txg;
1909 dmu_object_info_t doi;
1914 ztest_object_lock(zd, object, RL_READER);
1915 error = dmu_bonus_hold(os, object, FTAG, &db);
1917 ztest_object_unlock(zd, object);
1921 crtxg = ztest_bt_bonus(db)->bt_crtxg;
1923 if (crtxg == 0 || crtxg > txg) {
1924 dmu_buf_rele(db, FTAG);
1925 ztest_object_unlock(zd, object);
1929 dmu_object_info_from_db(db, &doi);
1930 dmu_buf_rele(db, FTAG);
1933 zgd = umem_zalloc(sizeof (*zgd), UMEM_NOFAIL);
1934 zgd->zgd_zilog = zd->zd_zilog;
1935 zgd->zgd_private = zd;
1937 if (buf != NULL) { /* immediate write */
1938 zgd->zgd_rl = ztest_range_lock(zd, object, offset, size,
1941 error = dmu_read(os, object, offset, size, buf,
1942 DMU_READ_NO_PREFETCH);
1945 size = doi.doi_data_block_size;
1947 offset = P2ALIGN(offset, size);
1949 ASSERT(offset < size);
1953 zgd->zgd_rl = ztest_range_lock(zd, object, offset, size,
1956 error = dmu_buf_hold(os, object, offset, zgd, &db,
1957 DMU_READ_NO_PREFETCH);
1960 blkptr_t *obp = dmu_buf_get_blkptr(db);
1962 ASSERT(BP_IS_HOLE(bp));
1969 ASSERT(db->db_offset == offset);
1970 ASSERT(db->db_size == size);
1972 error = dmu_sync(zio, lr->lr_common.lrc_txg,
1973 ztest_get_done, zgd);
1980 ztest_get_done(zgd, error);
1986 ztest_lr_alloc(size_t lrsize, char *name)
1989 size_t namesize = name ? strlen(name) + 1 : 0;
1991 lr = umem_zalloc(lrsize + namesize, UMEM_NOFAIL);
1994 bcopy(name, lr + lrsize, namesize);
2000 ztest_lr_free(void *lr, size_t lrsize, char *name)
2002 size_t namesize = name ? strlen(name) + 1 : 0;
2004 umem_free(lr, lrsize + namesize);
2008 * Lookup a bunch of objects. Returns the number of objects not found.
2011 ztest_lookup(ztest_ds_t *zd, ztest_od_t *od, int count)
2017 ASSERT(mutex_held(&zd->zd_dirobj_lock));
2019 for (i = 0; i < count; i++, od++) {
2021 error = zap_lookup(zd->zd_os, od->od_dir, od->od_name,
2022 sizeof (uint64_t), 1, &od->od_object);
2024 ASSERT(error == ENOENT);
2025 ASSERT(od->od_object == 0);
2029 ztest_block_tag_t *bbt;
2030 dmu_object_info_t doi;
2032 ASSERT(od->od_object != 0);
2033 ASSERT(missing == 0); /* there should be no gaps */
2035 ztest_object_lock(zd, od->od_object, RL_READER);
2036 VERIFY3U(0, ==, dmu_bonus_hold(zd->zd_os,
2037 od->od_object, FTAG, &db));
2038 dmu_object_info_from_db(db, &doi);
2039 bbt = ztest_bt_bonus(db);
2040 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
2041 od->od_type = doi.doi_type;
2042 od->od_blocksize = doi.doi_data_block_size;
2043 od->od_gen = bbt->bt_gen;
2044 dmu_buf_rele(db, FTAG);
2045 ztest_object_unlock(zd, od->od_object);
2053 ztest_create(ztest_ds_t *zd, ztest_od_t *od, int count)
2058 ASSERT(mutex_held(&zd->zd_dirobj_lock));
2060 for (i = 0; i < count; i++, od++) {
2067 lr_create_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
2069 lr->lr_doid = od->od_dir;
2070 lr->lr_foid = 0; /* 0 to allocate, > 0 to claim */
2071 lr->lrz_type = od->od_crtype;
2072 lr->lrz_blocksize = od->od_crblocksize;
2073 lr->lrz_ibshift = ztest_random_ibshift();
2074 lr->lrz_bonustype = DMU_OT_UINT64_OTHER;
2075 lr->lrz_bonuslen = dmu_bonus_max();
2076 lr->lr_gen = od->od_crgen;
2077 lr->lr_crtime[0] = time(NULL);
2079 if (ztest_replay_create(zd, lr, B_FALSE) != 0) {
2080 ASSERT(missing == 0);
2084 od->od_object = lr->lr_foid;
2085 od->od_type = od->od_crtype;
2086 od->od_blocksize = od->od_crblocksize;
2087 od->od_gen = od->od_crgen;
2088 ASSERT(od->od_object != 0);
2091 ztest_lr_free(lr, sizeof (*lr), od->od_name);
2098 ztest_remove(ztest_ds_t *zd, ztest_od_t *od, int count)
2104 ASSERT(mutex_held(&zd->zd_dirobj_lock));
2108 for (i = count - 1; i >= 0; i--, od--) {
2115 * No object was found.
2117 if (od->od_object == 0)
2120 lr_remove_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
2122 lr->lr_doid = od->od_dir;
2124 if ((error = ztest_replay_remove(zd, lr, B_FALSE)) != 0) {
2125 ASSERT3U(error, ==, ENOSPC);
2130 ztest_lr_free(lr, sizeof (*lr), od->od_name);
2137 ztest_write(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size,
2143 lr = ztest_lr_alloc(sizeof (*lr) + size, NULL);
2145 lr->lr_foid = object;
2146 lr->lr_offset = offset;
2147 lr->lr_length = size;
2149 BP_ZERO(&lr->lr_blkptr);
2151 bcopy(data, lr + 1, size);
2153 error = ztest_replay_write(zd, lr, B_FALSE);
2155 ztest_lr_free(lr, sizeof (*lr) + size, NULL);
2161 ztest_truncate(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2166 lr = ztest_lr_alloc(sizeof (*lr), NULL);
2168 lr->lr_foid = object;
2169 lr->lr_offset = offset;
2170 lr->lr_length = size;
2172 error = ztest_replay_truncate(zd, lr, B_FALSE);
2174 ztest_lr_free(lr, sizeof (*lr), NULL);
2180 ztest_setattr(ztest_ds_t *zd, uint64_t object)
2185 lr = ztest_lr_alloc(sizeof (*lr), NULL);
2187 lr->lr_foid = object;
2191 error = ztest_replay_setattr(zd, lr, B_FALSE);
2193 ztest_lr_free(lr, sizeof (*lr), NULL);
2199 ztest_prealloc(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2201 objset_t *os = zd->zd_os;
2206 txg_wait_synced(dmu_objset_pool(os), 0);
2208 ztest_object_lock(zd, object, RL_READER);
2209 rl = ztest_range_lock(zd, object, offset, size, RL_WRITER);
2211 tx = dmu_tx_create(os);
2213 dmu_tx_hold_write(tx, object, offset, size);
2215 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
2218 dmu_prealloc(os, object, offset, size, tx);
2220 txg_wait_synced(dmu_objset_pool(os), txg);
2222 (void) dmu_free_long_range(os, object, offset, size);
2225 ztest_range_unlock(rl);
2226 ztest_object_unlock(zd, object);
2230 ztest_io(ztest_ds_t *zd, uint64_t object, uint64_t offset)
2233 ztest_block_tag_t wbt;
2234 dmu_object_info_t doi;
2235 enum ztest_io_type io_type;
2239 VERIFY(dmu_object_info(zd->zd_os, object, &doi) == 0);
2240 blocksize = doi.doi_data_block_size;
2241 data = umem_alloc(blocksize, UMEM_NOFAIL);
2244 * Pick an i/o type at random, biased toward writing block tags.
2246 io_type = ztest_random(ZTEST_IO_TYPES);
2247 if (ztest_random(2) == 0)
2248 io_type = ZTEST_IO_WRITE_TAG;
2250 (void) rw_rdlock(&zd->zd_zilog_lock);
2254 case ZTEST_IO_WRITE_TAG:
2255 ztest_bt_generate(&wbt, zd->zd_os, object, offset, 0, 0, 0);
2256 (void) ztest_write(zd, object, offset, sizeof (wbt), &wbt);
2259 case ZTEST_IO_WRITE_PATTERN:
2260 (void) memset(data, 'a' + (object + offset) % 5, blocksize);
2261 if (ztest_random(2) == 0) {
2263 * Induce fletcher2 collisions to ensure that
2264 * zio_ddt_collision() detects and resolves them
2265 * when using fletcher2-verify for deduplication.
2267 ((uint64_t *)data)[0] ^= 1ULL << 63;
2268 ((uint64_t *)data)[4] ^= 1ULL << 63;
2270 (void) ztest_write(zd, object, offset, blocksize, data);
2273 case ZTEST_IO_WRITE_ZEROES:
2274 bzero(data, blocksize);
2275 (void) ztest_write(zd, object, offset, blocksize, data);
2278 case ZTEST_IO_TRUNCATE:
2279 (void) ztest_truncate(zd, object, offset, blocksize);
2282 case ZTEST_IO_SETATTR:
2283 (void) ztest_setattr(zd, object);
2288 case ZTEST_IO_REWRITE:
2289 (void) rw_rdlock(&ztest_name_lock);
2290 err = ztest_dsl_prop_set_uint64(zd->zd_name,
2291 ZFS_PROP_CHECKSUM, spa_dedup_checksum(ztest_spa),
2293 VERIFY(err == 0 || err == ENOSPC);
2294 err = ztest_dsl_prop_set_uint64(zd->zd_name,
2295 ZFS_PROP_COMPRESSION,
2296 ztest_random_dsl_prop(ZFS_PROP_COMPRESSION),
2298 VERIFY(err == 0 || err == ENOSPC);
2299 (void) rw_unlock(&ztest_name_lock);
2301 VERIFY0(dmu_read(zd->zd_os, object, offset, blocksize, data,
2302 DMU_READ_NO_PREFETCH));
2304 (void) ztest_write(zd, object, offset, blocksize, data);
2308 (void) rw_unlock(&zd->zd_zilog_lock);
2310 umem_free(data, blocksize);
2314 * Initialize an object description template.
2317 ztest_od_init(ztest_od_t *od, uint64_t id, char *tag, uint64_t index,
2318 dmu_object_type_t type, uint64_t blocksize, uint64_t gen)
2320 od->od_dir = ZTEST_DIROBJ;
2323 od->od_crtype = type;
2324 od->od_crblocksize = blocksize ? blocksize : ztest_random_blocksize();
2327 od->od_type = DMU_OT_NONE;
2328 od->od_blocksize = 0;
2331 (void) snprintf(od->od_name, sizeof (od->od_name), "%s(%lld)[%llu]",
2332 tag, (longlong_t)id, (u_longlong_t)index);
2336 * Lookup or create the objects for a test using the od template.
2337 * If the objects do not all exist, or if 'remove' is specified,
2338 * remove any existing objects and create new ones. Otherwise,
2339 * use the existing objects.
2342 ztest_object_init(ztest_ds_t *zd, ztest_od_t *od, size_t size, boolean_t remove)
2344 int count = size / sizeof (*od);
2347 mutex_enter(&zd->zd_dirobj_lock);
2348 if ((ztest_lookup(zd, od, count) != 0 || remove) &&
2349 (ztest_remove(zd, od, count) != 0 ||
2350 ztest_create(zd, od, count) != 0))
2353 mutex_exit(&zd->zd_dirobj_lock);
2360 ztest_zil_commit(ztest_ds_t *zd, uint64_t id)
2362 zilog_t *zilog = zd->zd_zilog;
2364 (void) rw_rdlock(&zd->zd_zilog_lock);
2366 zil_commit(zilog, ztest_random(ZTEST_OBJECTS));
2369 * Remember the committed values in zd, which is in parent/child
2370 * shared memory. If we die, the next iteration of ztest_run()
2371 * will verify that the log really does contain this record.
2373 mutex_enter(&zilog->zl_lock);
2374 ASSERT(zd->zd_shared != NULL);
2375 ASSERT3U(zd->zd_shared->zd_seq, <=, zilog->zl_commit_lr_seq);
2376 zd->zd_shared->zd_seq = zilog->zl_commit_lr_seq;
2377 mutex_exit(&zilog->zl_lock);
2379 (void) rw_unlock(&zd->zd_zilog_lock);
2383 * This function is designed to simulate the operations that occur during a
2384 * mount/unmount operation. We hold the dataset across these operations in an
2385 * attempt to expose any implicit assumptions about ZIL management.
2389 ztest_zil_remount(ztest_ds_t *zd, uint64_t id)
2391 objset_t *os = zd->zd_os;
2394 * We grab the zd_dirobj_lock to ensure that no other thread is
2395 * updating the zil (i.e. adding in-memory log records) and the
2396 * zd_zilog_lock to block any I/O.
2398 mutex_enter(&zd->zd_dirobj_lock);
2399 (void) rw_wrlock(&zd->zd_zilog_lock);
2401 /* zfs_sb_teardown() */
2402 zil_close(zd->zd_zilog);
2404 /* zfsvfs_setup() */
2405 VERIFY(zil_open(os, ztest_get_data) == zd->zd_zilog);
2406 zil_replay(os, zd, ztest_replay_vector);
2408 (void) rw_unlock(&zd->zd_zilog_lock);
2409 mutex_exit(&zd->zd_dirobj_lock);
2413 * Verify that we can't destroy an active pool, create an existing pool,
2414 * or create a pool with a bad vdev spec.
2418 ztest_spa_create_destroy(ztest_ds_t *zd, uint64_t id)
2420 ztest_shared_opts_t *zo = &ztest_opts;
2425 * Attempt to create using a bad file.
2427 nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1);
2428 VERIFY3U(ENOENT, ==,
2429 spa_create("ztest_bad_file", nvroot, NULL, NULL));
2430 nvlist_free(nvroot);
2433 * Attempt to create using a bad mirror.
2435 nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 2, 1);
2436 VERIFY3U(ENOENT, ==,
2437 spa_create("ztest_bad_mirror", nvroot, NULL, NULL));
2438 nvlist_free(nvroot);
2441 * Attempt to create an existing pool. It shouldn't matter
2442 * what's in the nvroot; we should fail with EEXIST.
2444 (void) rw_rdlock(&ztest_name_lock);
2445 nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1);
2446 VERIFY3U(EEXIST, ==, spa_create(zo->zo_pool, nvroot, NULL, NULL));
2447 nvlist_free(nvroot);
2448 VERIFY3U(0, ==, spa_open(zo->zo_pool, &spa, FTAG));
2449 VERIFY3U(EBUSY, ==, spa_destroy(zo->zo_pool));
2450 spa_close(spa, FTAG);
2452 (void) rw_unlock(&ztest_name_lock);
2457 ztest_spa_upgrade(ztest_ds_t *zd, uint64_t id)
2460 uint64_t initial_version = SPA_VERSION_INITIAL;
2461 uint64_t version, newversion;
2462 nvlist_t *nvroot, *props;
2465 mutex_enter(&ztest_vdev_lock);
2466 name = kmem_asprintf("%s_upgrade", ztest_opts.zo_pool);
2469 * Clean up from previous runs.
2471 (void) spa_destroy(name);
2473 nvroot = make_vdev_root(NULL, NULL, name, ztest_opts.zo_vdev_size, 0,
2474 0, ztest_opts.zo_raidz, ztest_opts.zo_mirrors, 1);
2477 * If we're configuring a RAIDZ device then make sure that the
2478 * the initial version is capable of supporting that feature.
2480 switch (ztest_opts.zo_raidz_parity) {
2483 initial_version = SPA_VERSION_INITIAL;
2486 initial_version = SPA_VERSION_RAIDZ2;
2489 initial_version = SPA_VERSION_RAIDZ3;
2494 * Create a pool with a spa version that can be upgraded. Pick
2495 * a value between initial_version and SPA_VERSION_BEFORE_FEATURES.
2498 version = ztest_random_spa_version(initial_version);
2499 } while (version > SPA_VERSION_BEFORE_FEATURES);
2501 props = fnvlist_alloc();
2502 fnvlist_add_uint64(props,
2503 zpool_prop_to_name(ZPOOL_PROP_VERSION), version);
2504 VERIFY3S(spa_create(name, nvroot, props, NULL), ==, 0);
2505 fnvlist_free(nvroot);
2506 fnvlist_free(props);
2508 VERIFY3S(spa_open(name, &spa, FTAG), ==, 0);
2509 VERIFY3U(spa_version(spa), ==, version);
2510 newversion = ztest_random_spa_version(version + 1);
2512 if (ztest_opts.zo_verbose >= 4) {
2513 (void) printf("upgrading spa version from %llu to %llu\n",
2514 (u_longlong_t)version, (u_longlong_t)newversion);
2517 spa_upgrade(spa, newversion);
2518 VERIFY3U(spa_version(spa), >, version);
2519 VERIFY3U(spa_version(spa), ==, fnvlist_lookup_uint64(spa->spa_config,
2520 zpool_prop_to_name(ZPOOL_PROP_VERSION)));
2521 spa_close(spa, FTAG);
2524 mutex_exit(&ztest_vdev_lock);
2528 vdev_lookup_by_path(vdev_t *vd, const char *path)
2533 if (vd->vdev_path != NULL && strcmp(path, vd->vdev_path) == 0)
2536 for (c = 0; c < vd->vdev_children; c++)
2537 if ((mvd = vdev_lookup_by_path(vd->vdev_child[c], path)) !=
2545 * Find the first available hole which can be used as a top-level.
2548 find_vdev_hole(spa_t *spa)
2550 vdev_t *rvd = spa->spa_root_vdev;
2553 ASSERT(spa_config_held(spa, SCL_VDEV, RW_READER) == SCL_VDEV);
2555 for (c = 0; c < rvd->vdev_children; c++) {
2556 vdev_t *cvd = rvd->vdev_child[c];
2558 if (cvd->vdev_ishole)
2565 * Verify that vdev_add() works as expected.
2569 ztest_vdev_add_remove(ztest_ds_t *zd, uint64_t id)
2571 ztest_shared_t *zs = ztest_shared;
2572 spa_t *spa = ztest_spa;
2578 mutex_enter(&ztest_vdev_lock);
2579 leaves = MAX(zs->zs_mirrors + zs->zs_splits, 1) * ztest_opts.zo_raidz;
2581 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2583 ztest_shared->zs_vdev_next_leaf = find_vdev_hole(spa) * leaves;
2586 * If we have slogs then remove them 1/4 of the time.
2588 if (spa_has_slogs(spa) && ztest_random(4) == 0) {
2590 * Grab the guid from the head of the log class rotor.
2592 guid = spa_log_class(spa)->mc_rotor->mg_vd->vdev_guid;
2594 spa_config_exit(spa, SCL_VDEV, FTAG);
2597 * We have to grab the zs_name_lock as writer to
2598 * prevent a race between removing a slog (dmu_objset_find)
2599 * and destroying a dataset. Removing the slog will
2600 * grab a reference on the dataset which may cause
2601 * dsl_destroy_head() to fail with EBUSY thus
2602 * leaving the dataset in an inconsistent state.
2604 rw_wrlock(&ztest_name_lock);
2605 error = spa_vdev_remove(spa, guid, B_FALSE);
2606 rw_unlock(&ztest_name_lock);
2608 if (error && error != EEXIST)
2609 fatal(0, "spa_vdev_remove() = %d", error);
2611 spa_config_exit(spa, SCL_VDEV, FTAG);
2614 * Make 1/4 of the devices be log devices.
2616 nvroot = make_vdev_root(NULL, NULL, NULL,
2617 ztest_opts.zo_vdev_size, 0,
2618 ztest_random(4) == 0, ztest_opts.zo_raidz,
2621 error = spa_vdev_add(spa, nvroot);
2622 nvlist_free(nvroot);
2624 if (error == ENOSPC)
2625 ztest_record_enospc("spa_vdev_add");
2626 else if (error != 0)
2627 fatal(0, "spa_vdev_add() = %d", error);
2630 mutex_exit(&ztest_vdev_lock);
2634 * Verify that adding/removing aux devices (l2arc, hot spare) works as expected.
2638 ztest_vdev_aux_add_remove(ztest_ds_t *zd, uint64_t id)
2640 ztest_shared_t *zs = ztest_shared;
2641 spa_t *spa = ztest_spa;
2642 vdev_t *rvd = spa->spa_root_vdev;
2643 spa_aux_vdev_t *sav;
2649 path = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
2651 if (ztest_random(2) == 0) {
2652 sav = &spa->spa_spares;
2653 aux = ZPOOL_CONFIG_SPARES;
2655 sav = &spa->spa_l2cache;
2656 aux = ZPOOL_CONFIG_L2CACHE;
2659 mutex_enter(&ztest_vdev_lock);
2661 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2663 if (sav->sav_count != 0 && ztest_random(4) == 0) {
2665 * Pick a random device to remove.
2667 guid = sav->sav_vdevs[ztest_random(sav->sav_count)]->vdev_guid;
2670 * Find an unused device we can add.
2672 zs->zs_vdev_aux = 0;
2675 (void) snprintf(path, MAXPATHLEN, ztest_aux_template,
2676 ztest_opts.zo_dir, ztest_opts.zo_pool, aux,
2678 for (c = 0; c < sav->sav_count; c++)
2679 if (strcmp(sav->sav_vdevs[c]->vdev_path,
2682 if (c == sav->sav_count &&
2683 vdev_lookup_by_path(rvd, path) == NULL)
2689 spa_config_exit(spa, SCL_VDEV, FTAG);
2695 nvlist_t *nvroot = make_vdev_root(NULL, aux, NULL,
2696 (ztest_opts.zo_vdev_size * 5) / 4, 0, 0, 0, 0, 1);
2697 error = spa_vdev_add(spa, nvroot);
2699 fatal(0, "spa_vdev_add(%p) = %d", nvroot, error);
2700 nvlist_free(nvroot);
2703 * Remove an existing device. Sometimes, dirty its
2704 * vdev state first to make sure we handle removal
2705 * of devices that have pending state changes.
2707 if (ztest_random(2) == 0)
2708 (void) vdev_online(spa, guid, 0, NULL);
2710 error = spa_vdev_remove(spa, guid, B_FALSE);
2711 if (error != 0 && error != EBUSY)
2712 fatal(0, "spa_vdev_remove(%llu) = %d", guid, error);
2715 mutex_exit(&ztest_vdev_lock);
2717 umem_free(path, MAXPATHLEN);
2721 * split a pool if it has mirror tlvdevs
2725 ztest_split_pool(ztest_ds_t *zd, uint64_t id)
2727 ztest_shared_t *zs = ztest_shared;
2728 spa_t *spa = ztest_spa;
2729 vdev_t *rvd = spa->spa_root_vdev;
2730 nvlist_t *tree, **child, *config, *split, **schild;
2731 uint_t c, children, schildren = 0, lastlogid = 0;
2734 mutex_enter(&ztest_vdev_lock);
2736 /* ensure we have a useable config; mirrors of raidz aren't supported */
2737 if (zs->zs_mirrors < 3 || ztest_opts.zo_raidz > 1) {
2738 mutex_exit(&ztest_vdev_lock);
2742 /* clean up the old pool, if any */
2743 (void) spa_destroy("splitp");
2745 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2747 /* generate a config from the existing config */
2748 mutex_enter(&spa->spa_props_lock);
2749 VERIFY(nvlist_lookup_nvlist(spa->spa_config, ZPOOL_CONFIG_VDEV_TREE,
2751 mutex_exit(&spa->spa_props_lock);
2753 VERIFY(nvlist_lookup_nvlist_array(tree, ZPOOL_CONFIG_CHILDREN, &child,
2756 schild = malloc(rvd->vdev_children * sizeof (nvlist_t *));
2757 for (c = 0; c < children; c++) {
2758 vdev_t *tvd = rvd->vdev_child[c];
2762 if (tvd->vdev_islog || tvd->vdev_ops == &vdev_hole_ops) {
2763 VERIFY(nvlist_alloc(&schild[schildren], NV_UNIQUE_NAME,
2765 VERIFY(nvlist_add_string(schild[schildren],
2766 ZPOOL_CONFIG_TYPE, VDEV_TYPE_HOLE) == 0);
2767 VERIFY(nvlist_add_uint64(schild[schildren],
2768 ZPOOL_CONFIG_IS_HOLE, 1) == 0);
2770 lastlogid = schildren;
2775 VERIFY(nvlist_lookup_nvlist_array(child[c],
2776 ZPOOL_CONFIG_CHILDREN, &mchild, &mchildren) == 0);
2777 VERIFY(nvlist_dup(mchild[0], &schild[schildren++], 0) == 0);
2780 /* OK, create a config that can be used to split */
2781 VERIFY(nvlist_alloc(&split, NV_UNIQUE_NAME, 0) == 0);
2782 VERIFY(nvlist_add_string(split, ZPOOL_CONFIG_TYPE,
2783 VDEV_TYPE_ROOT) == 0);
2784 VERIFY(nvlist_add_nvlist_array(split, ZPOOL_CONFIG_CHILDREN, schild,
2785 lastlogid != 0 ? lastlogid : schildren) == 0);
2787 VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, 0) == 0);
2788 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, split) == 0);
2790 for (c = 0; c < schildren; c++)
2791 nvlist_free(schild[c]);
2795 spa_config_exit(spa, SCL_VDEV, FTAG);
2797 (void) rw_wrlock(&ztest_name_lock);
2798 error = spa_vdev_split_mirror(spa, "splitp", config, NULL, B_FALSE);
2799 (void) rw_unlock(&ztest_name_lock);
2801 nvlist_free(config);
2804 (void) printf("successful split - results:\n");
2805 mutex_enter(&spa_namespace_lock);
2806 show_pool_stats(spa);
2807 show_pool_stats(spa_lookup("splitp"));
2808 mutex_exit(&spa_namespace_lock);
2812 mutex_exit(&ztest_vdev_lock);
2817 * Verify that we can attach and detach devices.
2821 ztest_vdev_attach_detach(ztest_ds_t *zd, uint64_t id)
2823 ztest_shared_t *zs = ztest_shared;
2824 spa_t *spa = ztest_spa;
2825 spa_aux_vdev_t *sav = &spa->spa_spares;
2826 vdev_t *rvd = spa->spa_root_vdev;
2827 vdev_t *oldvd, *newvd, *pvd;
2831 uint64_t ashift = ztest_get_ashift();
2832 uint64_t oldguid, pguid;
2833 uint64_t oldsize, newsize;
2834 char *oldpath, *newpath;
2836 int oldvd_has_siblings = B_FALSE;
2837 int newvd_is_spare = B_FALSE;
2839 int error, expected_error;
2841 oldpath = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
2842 newpath = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
2844 mutex_enter(&ztest_vdev_lock);
2845 leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz;
2847 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2850 * Decide whether to do an attach or a replace.
2852 replacing = ztest_random(2);
2855 * Pick a random top-level vdev.
2857 top = ztest_random_vdev_top(spa, B_TRUE);
2860 * Pick a random leaf within it.
2862 leaf = ztest_random(leaves);
2867 oldvd = rvd->vdev_child[top];
2868 if (zs->zs_mirrors >= 1) {
2869 ASSERT(oldvd->vdev_ops == &vdev_mirror_ops);
2870 ASSERT(oldvd->vdev_children >= zs->zs_mirrors);
2871 oldvd = oldvd->vdev_child[leaf / ztest_opts.zo_raidz];
2873 if (ztest_opts.zo_raidz > 1) {
2874 ASSERT(oldvd->vdev_ops == &vdev_raidz_ops);
2875 ASSERT(oldvd->vdev_children == ztest_opts.zo_raidz);
2876 oldvd = oldvd->vdev_child[leaf % ztest_opts.zo_raidz];
2880 * If we're already doing an attach or replace, oldvd may be a
2881 * mirror vdev -- in which case, pick a random child.
2883 while (oldvd->vdev_children != 0) {
2884 oldvd_has_siblings = B_TRUE;
2885 ASSERT(oldvd->vdev_children >= 2);
2886 oldvd = oldvd->vdev_child[ztest_random(oldvd->vdev_children)];
2889 oldguid = oldvd->vdev_guid;
2890 oldsize = vdev_get_min_asize(oldvd);
2891 oldvd_is_log = oldvd->vdev_top->vdev_islog;
2892 (void) strcpy(oldpath, oldvd->vdev_path);
2893 pvd = oldvd->vdev_parent;
2894 pguid = pvd->vdev_guid;
2897 * If oldvd has siblings, then half of the time, detach it.
2899 if (oldvd_has_siblings && ztest_random(2) == 0) {
2900 spa_config_exit(spa, SCL_VDEV, FTAG);
2901 error = spa_vdev_detach(spa, oldguid, pguid, B_FALSE);
2902 if (error != 0 && error != ENODEV && error != EBUSY &&
2904 fatal(0, "detach (%s) returned %d", oldpath, error);
2909 * For the new vdev, choose with equal probability between the two
2910 * standard paths (ending in either 'a' or 'b') or a random hot spare.
2912 if (sav->sav_count != 0 && ztest_random(3) == 0) {
2913 newvd = sav->sav_vdevs[ztest_random(sav->sav_count)];
2914 newvd_is_spare = B_TRUE;
2915 (void) strcpy(newpath, newvd->vdev_path);
2917 (void) snprintf(newpath, MAXPATHLEN, ztest_dev_template,
2918 ztest_opts.zo_dir, ztest_opts.zo_pool,
2919 top * leaves + leaf);
2920 if (ztest_random(2) == 0)
2921 newpath[strlen(newpath) - 1] = 'b';
2922 newvd = vdev_lookup_by_path(rvd, newpath);
2926 newsize = vdev_get_min_asize(newvd);
2929 * Make newsize a little bigger or smaller than oldsize.
2930 * If it's smaller, the attach should fail.
2931 * If it's larger, and we're doing a replace,
2932 * we should get dynamic LUN growth when we're done.
2934 newsize = 10 * oldsize / (9 + ztest_random(3));
2938 * If pvd is not a mirror or root, the attach should fail with ENOTSUP,
2939 * unless it's a replace; in that case any non-replacing parent is OK.
2941 * If newvd is already part of the pool, it should fail with EBUSY.
2943 * If newvd is too small, it should fail with EOVERFLOW.
2945 if (pvd->vdev_ops != &vdev_mirror_ops &&
2946 pvd->vdev_ops != &vdev_root_ops && (!replacing ||
2947 pvd->vdev_ops == &vdev_replacing_ops ||
2948 pvd->vdev_ops == &vdev_spare_ops))
2949 expected_error = ENOTSUP;
2950 else if (newvd_is_spare && (!replacing || oldvd_is_log))
2951 expected_error = ENOTSUP;
2952 else if (newvd == oldvd)
2953 expected_error = replacing ? 0 : EBUSY;
2954 else if (vdev_lookup_by_path(rvd, newpath) != NULL)
2955 expected_error = EBUSY;
2956 else if (newsize < oldsize)
2957 expected_error = EOVERFLOW;
2958 else if (ashift > oldvd->vdev_top->vdev_ashift)
2959 expected_error = EDOM;
2963 spa_config_exit(spa, SCL_VDEV, FTAG);
2966 * Build the nvlist describing newpath.
2968 root = make_vdev_root(newpath, NULL, NULL, newvd == NULL ? newsize : 0,
2969 ashift, 0, 0, 0, 1);
2971 error = spa_vdev_attach(spa, oldguid, root, replacing);
2976 * If our parent was the replacing vdev, but the replace completed,
2977 * then instead of failing with ENOTSUP we may either succeed,
2978 * fail with ENODEV, or fail with EOVERFLOW.
2980 if (expected_error == ENOTSUP &&
2981 (error == 0 || error == ENODEV || error == EOVERFLOW))
2982 expected_error = error;
2985 * If someone grew the LUN, the replacement may be too small.
2987 if (error == EOVERFLOW || error == EBUSY)
2988 expected_error = error;
2990 /* XXX workaround 6690467 */
2991 if (error != expected_error && expected_error != EBUSY) {
2992 fatal(0, "attach (%s %llu, %s %llu, %d) "
2993 "returned %d, expected %d",
2994 oldpath, oldsize, newpath,
2995 newsize, replacing, error, expected_error);
2998 mutex_exit(&ztest_vdev_lock);
3000 umem_free(oldpath, MAXPATHLEN);
3001 umem_free(newpath, MAXPATHLEN);
3005 * Callback function which expands the physical size of the vdev.
3008 grow_vdev(vdev_t *vd, void *arg)
3010 ASSERTV(spa_t *spa = vd->vdev_spa);
3011 size_t *newsize = arg;
3015 ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
3016 ASSERT(vd->vdev_ops->vdev_op_leaf);
3018 if ((fd = open(vd->vdev_path, O_RDWR)) == -1)
3021 fsize = lseek(fd, 0, SEEK_END);
3022 VERIFY(ftruncate(fd, *newsize) == 0);
3024 if (ztest_opts.zo_verbose >= 6) {
3025 (void) printf("%s grew from %lu to %lu bytes\n",
3026 vd->vdev_path, (ulong_t)fsize, (ulong_t)*newsize);
3033 * Callback function which expands a given vdev by calling vdev_online().
3037 online_vdev(vdev_t *vd, void *arg)
3039 spa_t *spa = vd->vdev_spa;
3040 vdev_t *tvd = vd->vdev_top;
3041 uint64_t guid = vd->vdev_guid;
3042 uint64_t generation = spa->spa_config_generation + 1;
3043 vdev_state_t newstate = VDEV_STATE_UNKNOWN;
3046 ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
3047 ASSERT(vd->vdev_ops->vdev_op_leaf);
3049 /* Calling vdev_online will initialize the new metaslabs */
3050 spa_config_exit(spa, SCL_STATE, spa);
3051 error = vdev_online(spa, guid, ZFS_ONLINE_EXPAND, &newstate);
3052 spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3055 * If vdev_online returned an error or the underlying vdev_open
3056 * failed then we abort the expand. The only way to know that
3057 * vdev_open fails is by checking the returned newstate.
3059 if (error || newstate != VDEV_STATE_HEALTHY) {
3060 if (ztest_opts.zo_verbose >= 5) {
3061 (void) printf("Unable to expand vdev, state %llu, "
3062 "error %d\n", (u_longlong_t)newstate, error);
3066 ASSERT3U(newstate, ==, VDEV_STATE_HEALTHY);
3069 * Since we dropped the lock we need to ensure that we're
3070 * still talking to the original vdev. It's possible this
3071 * vdev may have been detached/replaced while we were
3072 * trying to online it.
3074 if (generation != spa->spa_config_generation) {
3075 if (ztest_opts.zo_verbose >= 5) {
3076 (void) printf("vdev configuration has changed, "
3077 "guid %llu, state %llu, expected gen %llu, "
3080 (u_longlong_t)tvd->vdev_state,
3081 (u_longlong_t)generation,
3082 (u_longlong_t)spa->spa_config_generation);
3090 * Traverse the vdev tree calling the supplied function.
3091 * We continue to walk the tree until we either have walked all
3092 * children or we receive a non-NULL return from the callback.
3093 * If a NULL callback is passed, then we just return back the first
3094 * leaf vdev we encounter.
3097 vdev_walk_tree(vdev_t *vd, vdev_t *(*func)(vdev_t *, void *), void *arg)
3101 if (vd->vdev_ops->vdev_op_leaf) {
3105 return (func(vd, arg));
3108 for (c = 0; c < vd->vdev_children; c++) {
3109 vdev_t *cvd = vd->vdev_child[c];
3110 if ((cvd = vdev_walk_tree(cvd, func, arg)) != NULL)
3117 * Verify that dynamic LUN growth works as expected.
3121 ztest_vdev_LUN_growth(ztest_ds_t *zd, uint64_t id)
3123 spa_t *spa = ztest_spa;
3125 metaslab_class_t *mc;
3126 metaslab_group_t *mg;
3127 size_t psize, newsize;
3129 uint64_t old_class_space, new_class_space, old_ms_count, new_ms_count;
3131 mutex_enter(&ztest_vdev_lock);
3132 spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3134 top = ztest_random_vdev_top(spa, B_TRUE);
3136 tvd = spa->spa_root_vdev->vdev_child[top];
3139 old_ms_count = tvd->vdev_ms_count;
3140 old_class_space = metaslab_class_get_space(mc);
3143 * Determine the size of the first leaf vdev associated with
3144 * our top-level device.
3146 vd = vdev_walk_tree(tvd, NULL, NULL);
3147 ASSERT3P(vd, !=, NULL);
3148 ASSERT(vd->vdev_ops->vdev_op_leaf);
3150 psize = vd->vdev_psize;
3153 * We only try to expand the vdev if it's healthy, less than 4x its
3154 * original size, and it has a valid psize.
3156 if (tvd->vdev_state != VDEV_STATE_HEALTHY ||
3157 psize == 0 || psize >= 4 * ztest_opts.zo_vdev_size) {
3158 spa_config_exit(spa, SCL_STATE, spa);
3159 mutex_exit(&ztest_vdev_lock);
3163 newsize = psize + psize / 8;
3164 ASSERT3U(newsize, >, psize);
3166 if (ztest_opts.zo_verbose >= 6) {
3167 (void) printf("Expanding LUN %s from %lu to %lu\n",
3168 vd->vdev_path, (ulong_t)psize, (ulong_t)newsize);
3172 * Growing the vdev is a two step process:
3173 * 1). expand the physical size (i.e. relabel)
3174 * 2). online the vdev to create the new metaslabs
3176 if (vdev_walk_tree(tvd, grow_vdev, &newsize) != NULL ||
3177 vdev_walk_tree(tvd, online_vdev, NULL) != NULL ||
3178 tvd->vdev_state != VDEV_STATE_HEALTHY) {
3179 if (ztest_opts.zo_verbose >= 5) {
3180 (void) printf("Could not expand LUN because "
3181 "the vdev configuration changed.\n");
3183 spa_config_exit(spa, SCL_STATE, spa);
3184 mutex_exit(&ztest_vdev_lock);
3188 spa_config_exit(spa, SCL_STATE, spa);
3191 * Expanding the LUN will update the config asynchronously,
3192 * thus we must wait for the async thread to complete any
3193 * pending tasks before proceeding.
3197 mutex_enter(&spa->spa_async_lock);
3198 done = (spa->spa_async_thread == NULL && !spa->spa_async_tasks);
3199 mutex_exit(&spa->spa_async_lock);
3202 txg_wait_synced(spa_get_dsl(spa), 0);
3203 (void) poll(NULL, 0, 100);
3206 spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3208 tvd = spa->spa_root_vdev->vdev_child[top];
3209 new_ms_count = tvd->vdev_ms_count;
3210 new_class_space = metaslab_class_get_space(mc);
3212 if (tvd->vdev_mg != mg || mg->mg_class != mc) {
3213 if (ztest_opts.zo_verbose >= 5) {
3214 (void) printf("Could not verify LUN expansion due to "
3215 "intervening vdev offline or remove.\n");
3217 spa_config_exit(spa, SCL_STATE, spa);
3218 mutex_exit(&ztest_vdev_lock);
3223 * Make sure we were able to grow the vdev.
3225 if (new_ms_count <= old_ms_count)
3226 fatal(0, "LUN expansion failed: ms_count %llu <= %llu\n",
3227 old_ms_count, new_ms_count);
3230 * Make sure we were able to grow the pool.
3232 if (new_class_space <= old_class_space)
3233 fatal(0, "LUN expansion failed: class_space %llu <= %llu\n",
3234 old_class_space, new_class_space);
3236 if (ztest_opts.zo_verbose >= 5) {
3237 char oldnumbuf[6], newnumbuf[6];
3239 nicenum(old_class_space, oldnumbuf);
3240 nicenum(new_class_space, newnumbuf);
3241 (void) printf("%s grew from %s to %s\n",
3242 spa->spa_name, oldnumbuf, newnumbuf);
3245 spa_config_exit(spa, SCL_STATE, spa);
3246 mutex_exit(&ztest_vdev_lock);
3250 * Verify that dmu_objset_{create,destroy,open,close} work as expected.
3254 ztest_objset_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
3257 * Create the objects common to all ztest datasets.
3259 VERIFY(zap_create_claim(os, ZTEST_DIROBJ,
3260 DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx) == 0);
3264 ztest_dataset_create(char *dsname)
3266 uint64_t zilset = ztest_random(100);
3267 int err = dmu_objset_create(dsname, DMU_OST_OTHER, 0,
3268 ztest_objset_create_cb, NULL);
3270 if (err || zilset < 80)
3273 if (ztest_opts.zo_verbose >= 5)
3274 (void) printf("Setting dataset %s to sync always\n", dsname);
3275 return (ztest_dsl_prop_set_uint64(dsname, ZFS_PROP_SYNC,
3276 ZFS_SYNC_ALWAYS, B_FALSE));
3281 ztest_objset_destroy_cb(const char *name, void *arg)
3284 dmu_object_info_t doi;
3288 * Verify that the dataset contains a directory object.
3290 VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_TRUE, FTAG, &os));
3291 error = dmu_object_info(os, ZTEST_DIROBJ, &doi);
3292 if (error != ENOENT) {
3293 /* We could have crashed in the middle of destroying it */
3295 ASSERT3U(doi.doi_type, ==, DMU_OT_ZAP_OTHER);
3296 ASSERT3S(doi.doi_physical_blocks_512, >=, 0);
3298 dmu_objset_disown(os, FTAG);
3301 * Destroy the dataset.
3303 if (strchr(name, '@') != NULL) {
3304 VERIFY0(dsl_destroy_snapshot(name, B_FALSE));
3306 VERIFY0(dsl_destroy_head(name));
3312 ztest_snapshot_create(char *osname, uint64_t id)
3314 char snapname[MAXNAMELEN];
3317 (void) snprintf(snapname, sizeof (snapname), "%llu", (u_longlong_t)id);
3319 error = dmu_objset_snapshot_one(osname, snapname);
3320 if (error == ENOSPC) {
3321 ztest_record_enospc(FTAG);
3324 if (error != 0 && error != EEXIST) {
3325 fatal(0, "ztest_snapshot_create(%s@%s) = %d", osname,
3332 ztest_snapshot_destroy(char *osname, uint64_t id)
3334 char snapname[MAXNAMELEN];
3337 (void) snprintf(snapname, MAXNAMELEN, "%s@%llu", osname,
3340 error = dsl_destroy_snapshot(snapname, B_FALSE);
3341 if (error != 0 && error != ENOENT)
3342 fatal(0, "ztest_snapshot_destroy(%s) = %d", snapname, error);
3348 ztest_dmu_objset_create_destroy(ztest_ds_t *zd, uint64_t id)
3358 zdtmp = umem_alloc(sizeof (ztest_ds_t), UMEM_NOFAIL);
3359 name = umem_alloc(MAXNAMELEN, UMEM_NOFAIL);
3361 (void) rw_rdlock(&ztest_name_lock);
3363 (void) snprintf(name, MAXNAMELEN, "%s/temp_%llu",
3364 ztest_opts.zo_pool, (u_longlong_t)id);
3367 * If this dataset exists from a previous run, process its replay log
3368 * half of the time. If we don't replay it, then dsl_destroy_head()
3369 * (invoked from ztest_objset_destroy_cb()) should just throw it away.
3371 if (ztest_random(2) == 0 &&
3372 dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os) == 0) {
3373 ztest_zd_init(zdtmp, NULL, os);
3374 zil_replay(os, zdtmp, ztest_replay_vector);
3375 ztest_zd_fini(zdtmp);
3376 dmu_objset_disown(os, FTAG);
3380 * There may be an old instance of the dataset we're about to
3381 * create lying around from a previous run. If so, destroy it
3382 * and all of its snapshots.
3384 (void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
3385 DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
3388 * Verify that the destroyed dataset is no longer in the namespace.
3390 VERIFY3U(ENOENT, ==, dmu_objset_own(name, DMU_OST_OTHER, B_TRUE,
3394 * Verify that we can create a new dataset.
3396 error = ztest_dataset_create(name);
3398 if (error == ENOSPC) {
3399 ztest_record_enospc(FTAG);
3402 fatal(0, "dmu_objset_create(%s) = %d", name, error);
3405 VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os));
3407 ztest_zd_init(zdtmp, NULL, os);
3410 * Open the intent log for it.
3412 zilog = zil_open(os, ztest_get_data);
3415 * Put some objects in there, do a little I/O to them,
3416 * and randomly take a couple of snapshots along the way.
3418 iters = ztest_random(5);
3419 for (i = 0; i < iters; i++) {
3420 ztest_dmu_object_alloc_free(zdtmp, id);
3421 if (ztest_random(iters) == 0)
3422 (void) ztest_snapshot_create(name, i);
3426 * Verify that we cannot create an existing dataset.
3428 VERIFY3U(EEXIST, ==,
3429 dmu_objset_create(name, DMU_OST_OTHER, 0, NULL, NULL));
3432 * Verify that we can hold an objset that is also owned.
3434 VERIFY3U(0, ==, dmu_objset_hold(name, FTAG, &os2));
3435 dmu_objset_rele(os2, FTAG);
3438 * Verify that we cannot own an objset that is already owned.
3441 dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os2));
3444 dmu_objset_disown(os, FTAG);
3445 ztest_zd_fini(zdtmp);
3447 (void) rw_unlock(&ztest_name_lock);
3449 umem_free(name, MAXNAMELEN);
3450 umem_free(zdtmp, sizeof (ztest_ds_t));
3454 * Verify that dmu_snapshot_{create,destroy,open,close} work as expected.
3457 ztest_dmu_snapshot_create_destroy(ztest_ds_t *zd, uint64_t id)
3459 (void) rw_rdlock(&ztest_name_lock);
3460 (void) ztest_snapshot_destroy(zd->zd_name, id);
3461 (void) ztest_snapshot_create(zd->zd_name, id);
3462 (void) rw_unlock(&ztest_name_lock);
3466 * Cleanup non-standard snapshots and clones.
3469 ztest_dsl_dataset_cleanup(char *osname, uint64_t id)
3478 snap1name = umem_alloc(MAXNAMELEN, UMEM_NOFAIL);
3479 clone1name = umem_alloc(MAXNAMELEN, UMEM_NOFAIL);
3480 snap2name = umem_alloc(MAXNAMELEN, UMEM_NOFAIL);
3481 clone2name = umem_alloc(MAXNAMELEN, UMEM_NOFAIL);
3482 snap3name = umem_alloc(MAXNAMELEN, UMEM_NOFAIL);
3484 (void) snprintf(snap1name, MAXNAMELEN, "%s@s1_%llu",
3485 osname, (u_longlong_t)id);
3486 (void) snprintf(clone1name, MAXNAMELEN, "%s/c1_%llu",
3487 osname, (u_longlong_t)id);
3488 (void) snprintf(snap2name, MAXNAMELEN, "%s@s2_%llu",
3489 clone1name, (u_longlong_t)id);
3490 (void) snprintf(clone2name, MAXNAMELEN, "%s/c2_%llu",
3491 osname, (u_longlong_t)id);
3492 (void) snprintf(snap3name, MAXNAMELEN, "%s@s3_%llu",
3493 clone1name, (u_longlong_t)id);
3495 error = dsl_destroy_head(clone2name);
3496 if (error && error != ENOENT)
3497 fatal(0, "dsl_destroy_head(%s) = %d", clone2name, error);
3498 error = dsl_destroy_snapshot(snap3name, B_FALSE);
3499 if (error && error != ENOENT)
3500 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap3name, error);
3501 error = dsl_destroy_snapshot(snap2name, B_FALSE);
3502 if (error && error != ENOENT)
3503 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap2name, error);
3504 error = dsl_destroy_head(clone1name);
3505 if (error && error != ENOENT)
3506 fatal(0, "dsl_destroy_head(%s) = %d", clone1name, error);
3507 error = dsl_destroy_snapshot(snap1name, B_FALSE);
3508 if (error && error != ENOENT)
3509 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap1name, error);
3511 umem_free(snap1name, MAXNAMELEN);
3512 umem_free(clone1name, MAXNAMELEN);
3513 umem_free(snap2name, MAXNAMELEN);
3514 umem_free(clone2name, MAXNAMELEN);
3515 umem_free(snap3name, MAXNAMELEN);
3519 * Verify dsl_dataset_promote handles EBUSY
3522 ztest_dsl_dataset_promote_busy(ztest_ds_t *zd, uint64_t id)
3530 char *osname = zd->zd_name;
3533 snap1name = umem_alloc(MAXNAMELEN, UMEM_NOFAIL);
3534 clone1name = umem_alloc(MAXNAMELEN, UMEM_NOFAIL);
3535 snap2name = umem_alloc(MAXNAMELEN, UMEM_NOFAIL);
3536 clone2name = umem_alloc(MAXNAMELEN, UMEM_NOFAIL);
3537 snap3name = umem_alloc(MAXNAMELEN, UMEM_NOFAIL);
3539 (void) rw_rdlock(&ztest_name_lock);
3541 ztest_dsl_dataset_cleanup(osname, id);
3543 (void) snprintf(snap1name, MAXNAMELEN, "%s@s1_%llu",
3544 osname, (u_longlong_t)id);
3545 (void) snprintf(clone1name, MAXNAMELEN, "%s/c1_%llu",
3546 osname, (u_longlong_t)id);
3547 (void) snprintf(snap2name, MAXNAMELEN, "%s@s2_%llu",
3548 clone1name, (u_longlong_t)id);
3549 (void) snprintf(clone2name, MAXNAMELEN, "%s/c2_%llu",
3550 osname, (u_longlong_t)id);
3551 (void) snprintf(snap3name, MAXNAMELEN, "%s@s3_%llu",
3552 clone1name, (u_longlong_t)id);
3554 error = dmu_objset_snapshot_one(osname, strchr(snap1name, '@') + 1);
3555 if (error && error != EEXIST) {
3556 if (error == ENOSPC) {
3557 ztest_record_enospc(FTAG);
3560 fatal(0, "dmu_take_snapshot(%s) = %d", snap1name, error);
3563 error = dmu_objset_clone(clone1name, snap1name);
3565 if (error == ENOSPC) {
3566 ztest_record_enospc(FTAG);
3569 fatal(0, "dmu_objset_create(%s) = %d", clone1name, error);
3572 error = dmu_objset_snapshot_one(clone1name, strchr(snap2name, '@') + 1);
3573 if (error && error != EEXIST) {
3574 if (error == ENOSPC) {
3575 ztest_record_enospc(FTAG);
3578 fatal(0, "dmu_open_snapshot(%s) = %d", snap2name, error);
3581 error = dmu_objset_snapshot_one(clone1name, strchr(snap3name, '@') + 1);
3582 if (error && error != EEXIST) {
3583 if (error == ENOSPC) {
3584 ztest_record_enospc(FTAG);
3587 fatal(0, "dmu_open_snapshot(%s) = %d", snap3name, error);
3590 error = dmu_objset_clone(clone2name, snap3name);
3592 if (error == ENOSPC) {
3593 ztest_record_enospc(FTAG);
3596 fatal(0, "dmu_objset_create(%s) = %d", clone2name, error);
3599 error = dmu_objset_own(snap2name, DMU_OST_ANY, B_TRUE, FTAG, &os);
3601 fatal(0, "dmu_objset_own(%s) = %d", snap2name, error);
3602 error = dsl_dataset_promote(clone2name, NULL);
3603 if (error == ENOSPC) {
3604 dmu_objset_disown(os, FTAG);
3605 ztest_record_enospc(FTAG);
3609 fatal(0, "dsl_dataset_promote(%s), %d, not EBUSY", clone2name,
3611 dmu_objset_disown(os, FTAG);
3614 ztest_dsl_dataset_cleanup(osname, id);
3616 (void) rw_unlock(&ztest_name_lock);
3618 umem_free(snap1name, MAXNAMELEN);
3619 umem_free(clone1name, MAXNAMELEN);
3620 umem_free(snap2name, MAXNAMELEN);
3621 umem_free(clone2name, MAXNAMELEN);
3622 umem_free(snap3name, MAXNAMELEN);
3625 #undef OD_ARRAY_SIZE
3626 #define OD_ARRAY_SIZE 4
3629 * Verify that dmu_object_{alloc,free} work as expected.
3632 ztest_dmu_object_alloc_free(ztest_ds_t *zd, uint64_t id)
3639 size = sizeof (ztest_od_t) * OD_ARRAY_SIZE;
3640 od = umem_alloc(size, UMEM_NOFAIL);
3641 batchsize = OD_ARRAY_SIZE;
3643 for (b = 0; b < batchsize; b++)
3644 ztest_od_init(od + b, id, FTAG, b, DMU_OT_UINT64_OTHER, 0, 0);
3647 * Destroy the previous batch of objects, create a new batch,
3648 * and do some I/O on the new objects.
3650 if (ztest_object_init(zd, od, size, B_TRUE) != 0)
3653 while (ztest_random(4 * batchsize) != 0)
3654 ztest_io(zd, od[ztest_random(batchsize)].od_object,
3655 ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
3657 umem_free(od, size);
3660 #undef OD_ARRAY_SIZE
3661 #define OD_ARRAY_SIZE 2
3664 * Verify that dmu_{read,write} work as expected.
3667 ztest_dmu_read_write(ztest_ds_t *zd, uint64_t id)
3672 objset_t *os = zd->zd_os;
3673 size = sizeof (ztest_od_t) * OD_ARRAY_SIZE;
3674 od = umem_alloc(size, UMEM_NOFAIL);
3676 int i, freeit, error;
3678 bufwad_t *packbuf, *bigbuf, *pack, *bigH, *bigT;
3679 uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
3680 uint64_t chunksize = (1000 + ztest_random(1000)) * sizeof (uint64_t);
3681 uint64_t regions = 997;
3682 uint64_t stride = 123456789ULL;
3683 uint64_t width = 40;
3684 int free_percent = 5;
3687 * This test uses two objects, packobj and bigobj, that are always
3688 * updated together (i.e. in the same tx) so that their contents are
3689 * in sync and can be compared. Their contents relate to each other
3690 * in a simple way: packobj is a dense array of 'bufwad' structures,
3691 * while bigobj is a sparse array of the same bufwads. Specifically,
3692 * for any index n, there are three bufwads that should be identical:
3694 * packobj, at offset n * sizeof (bufwad_t)
3695 * bigobj, at the head of the nth chunk
3696 * bigobj, at the tail of the nth chunk
3698 * The chunk size is arbitrary. It doesn't have to be a power of two,
3699 * and it doesn't have any relation to the object blocksize.
3700 * The only requirement is that it can hold at least two bufwads.
3702 * Normally, we write the bufwad to each of these locations.
3703 * However, free_percent of the time we instead write zeroes to
3704 * packobj and perform a dmu_free_range() on bigobj. By comparing
3705 * bigobj to packobj, we can verify that the DMU is correctly
3706 * tracking which parts of an object are allocated and free,
3707 * and that the contents of the allocated blocks are correct.
3711 * Read the directory info. If it's the first time, set things up.
3713 ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, chunksize);
3714 ztest_od_init(od + 1, id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize);
3716 if (ztest_object_init(zd, od, size, B_FALSE) != 0) {
3717 umem_free(od, size);
3721 bigobj = od[0].od_object;
3722 packobj = od[1].od_object;
3723 chunksize = od[0].od_gen;
3724 ASSERT(chunksize == od[1].od_gen);
3727 * Prefetch a random chunk of the big object.
3728 * Our aim here is to get some async reads in flight
3729 * for blocks that we may free below; the DMU should
3730 * handle this race correctly.
3732 n = ztest_random(regions) * stride + ztest_random(width);
3733 s = 1 + ztest_random(2 * width - 1);
3734 dmu_prefetch(os, bigobj, n * chunksize, s * chunksize);
3737 * Pick a random index and compute the offsets into packobj and bigobj.
3739 n = ztest_random(regions) * stride + ztest_random(width);
3740 s = 1 + ztest_random(width - 1);
3742 packoff = n * sizeof (bufwad_t);
3743 packsize = s * sizeof (bufwad_t);
3745 bigoff = n * chunksize;
3746 bigsize = s * chunksize;
3748 packbuf = umem_alloc(packsize, UMEM_NOFAIL);
3749 bigbuf = umem_alloc(bigsize, UMEM_NOFAIL);
3752 * free_percent of the time, free a range of bigobj rather than
3755 freeit = (ztest_random(100) < free_percent);
3758 * Read the current contents of our objects.
3760 error = dmu_read(os, packobj, packoff, packsize, packbuf,
3763 error = dmu_read(os, bigobj, bigoff, bigsize, bigbuf,
3768 * Get a tx for the mods to both packobj and bigobj.
3770 tx = dmu_tx_create(os);
3772 dmu_tx_hold_write(tx, packobj, packoff, packsize);
3775 dmu_tx_hold_free(tx, bigobj, bigoff, bigsize);
3777 dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
3779 /* This accounts for setting the checksum/compression. */
3780 dmu_tx_hold_bonus(tx, bigobj);
3782 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3784 umem_free(packbuf, packsize);
3785 umem_free(bigbuf, bigsize);
3786 umem_free(od, size);
3790 enum zio_checksum cksum;
3792 cksum = (enum zio_checksum)
3793 ztest_random_dsl_prop(ZFS_PROP_CHECKSUM);
3794 } while (cksum >= ZIO_CHECKSUM_LEGACY_FUNCTIONS);
3795 dmu_object_set_checksum(os, bigobj, cksum, tx);
3797 enum zio_compress comp;
3799 comp = (enum zio_compress)
3800 ztest_random_dsl_prop(ZFS_PROP_COMPRESSION);
3801 } while (comp >= ZIO_COMPRESS_LEGACY_FUNCTIONS);
3802 dmu_object_set_compress(os, bigobj, comp, tx);
3805 * For each index from n to n + s, verify that the existing bufwad
3806 * in packobj matches the bufwads at the head and tail of the
3807 * corresponding chunk in bigobj. Then update all three bufwads
3808 * with the new values we want to write out.
3810 for (i = 0; i < s; i++) {
3812 pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
3814 bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
3816 bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
3818 ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
3819 ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
3821 if (pack->bw_txg > txg)
3822 fatal(0, "future leak: got %llx, open txg is %llx",
3825 if (pack->bw_data != 0 && pack->bw_index != n + i)
3826 fatal(0, "wrong index: got %llx, wanted %llx+%llx",
3827 pack->bw_index, n, i);
3829 if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
3830 fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
3832 if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
3833 fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
3836 bzero(pack, sizeof (bufwad_t));
3838 pack->bw_index = n + i;
3840 pack->bw_data = 1 + ztest_random(-2ULL);
3847 * We've verified all the old bufwads, and made new ones.
3848 * Now write them out.
3850 dmu_write(os, packobj, packoff, packsize, packbuf, tx);
3853 if (ztest_opts.zo_verbose >= 7) {
3854 (void) printf("freeing offset %llx size %llx"
3856 (u_longlong_t)bigoff,
3857 (u_longlong_t)bigsize,
3860 VERIFY(0 == dmu_free_range(os, bigobj, bigoff, bigsize, tx));
3862 if (ztest_opts.zo_verbose >= 7) {
3863 (void) printf("writing offset %llx size %llx"
3865 (u_longlong_t)bigoff,
3866 (u_longlong_t)bigsize,
3869 dmu_write(os, bigobj, bigoff, bigsize, bigbuf, tx);
3875 * Sanity check the stuff we just wrote.
3878 void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
3879 void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
3881 VERIFY(0 == dmu_read(os, packobj, packoff,
3882 packsize, packcheck, DMU_READ_PREFETCH));
3883 VERIFY(0 == dmu_read(os, bigobj, bigoff,
3884 bigsize, bigcheck, DMU_READ_PREFETCH));
3886 ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
3887 ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
3889 umem_free(packcheck, packsize);
3890 umem_free(bigcheck, bigsize);
3893 umem_free(packbuf, packsize);
3894 umem_free(bigbuf, bigsize);
3895 umem_free(od, size);
3899 compare_and_update_pbbufs(uint64_t s, bufwad_t *packbuf, bufwad_t *bigbuf,
3900 uint64_t bigsize, uint64_t n, uint64_t chunksize, uint64_t txg)
3908 * For each index from n to n + s, verify that the existing bufwad
3909 * in packobj matches the bufwads at the head and tail of the
3910 * corresponding chunk in bigobj. Then update all three bufwads
3911 * with the new values we want to write out.
3913 for (i = 0; i < s; i++) {
3915 pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
3917 bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
3919 bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
3921 ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
3922 ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
3924 if (pack->bw_txg > txg)
3925 fatal(0, "future leak: got %llx, open txg is %llx",
3928 if (pack->bw_data != 0 && pack->bw_index != n + i)
3929 fatal(0, "wrong index: got %llx, wanted %llx+%llx",
3930 pack->bw_index, n, i);
3932 if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
3933 fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
3935 if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
3936 fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
3938 pack->bw_index = n + i;
3940 pack->bw_data = 1 + ztest_random(-2ULL);
3947 #undef OD_ARRAY_SIZE
3948 #define OD_ARRAY_SIZE 2
3951 ztest_dmu_read_write_zcopy(ztest_ds_t *zd, uint64_t id)
3953 objset_t *os = zd->zd_os;
3960 bufwad_t *packbuf, *bigbuf;
3961 uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
3962 uint64_t blocksize = ztest_random_blocksize();
3963 uint64_t chunksize = blocksize;
3964 uint64_t regions = 997;
3965 uint64_t stride = 123456789ULL;
3967 dmu_buf_t *bonus_db;
3968 arc_buf_t **bigbuf_arcbufs;
3969 dmu_object_info_t doi;
3971 size = sizeof (ztest_od_t) * OD_ARRAY_SIZE;
3972 od = umem_alloc(size, UMEM_NOFAIL);
3975 * This test uses two objects, packobj and bigobj, that are always
3976 * updated together (i.e. in the same tx) so that their contents are
3977 * in sync and can be compared. Their contents relate to each other
3978 * in a simple way: packobj is a dense array of 'bufwad' structures,
3979 * while bigobj is a sparse array of the same bufwads. Specifically,
3980 * for any index n, there are three bufwads that should be identical:
3982 * packobj, at offset n * sizeof (bufwad_t)
3983 * bigobj, at the head of the nth chunk
3984 * bigobj, at the tail of the nth chunk
3986 * The chunk size is set equal to bigobj block size so that
3987 * dmu_assign_arcbuf() can be tested for object updates.
3991 * Read the directory info. If it's the first time, set things up.
3993 ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
3994 ztest_od_init(od + 1, id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize);
3997 if (ztest_object_init(zd, od, size, B_FALSE) != 0) {
3998 umem_free(od, size);
4002 bigobj = od[0].od_object;
4003 packobj = od[1].od_object;
4004 blocksize = od[0].od_blocksize;
4005 chunksize = blocksize;
4006 ASSERT(chunksize == od[1].od_gen);
4008 VERIFY(dmu_object_info(os, bigobj, &doi) == 0);
4009 VERIFY(ISP2(doi.doi_data_block_size));
4010 VERIFY(chunksize == doi.doi_data_block_size);
4011 VERIFY(chunksize >= 2 * sizeof (bufwad_t));
4014 * Pick a random index and compute the offsets into packobj and bigobj.
4016 n = ztest_random(regions) * stride + ztest_random(width);
4017 s = 1 + ztest_random(width - 1);
4019 packoff = n * sizeof (bufwad_t);
4020 packsize = s * sizeof (bufwad_t);
4022 bigoff = n * chunksize;
4023 bigsize = s * chunksize;
4025 packbuf = umem_zalloc(packsize, UMEM_NOFAIL);
4026 bigbuf = umem_zalloc(bigsize, UMEM_NOFAIL);
4028 VERIFY3U(0, ==, dmu_bonus_hold(os, bigobj, FTAG, &bonus_db));
4030 bigbuf_arcbufs = umem_zalloc(2 * s * sizeof (arc_buf_t *), UMEM_NOFAIL);
4033 * Iteration 0 test zcopy for DB_UNCACHED dbufs.
4034 * Iteration 1 test zcopy to already referenced dbufs.
4035 * Iteration 2 test zcopy to dirty dbuf in the same txg.
4036 * Iteration 3 test zcopy to dbuf dirty in previous txg.
4037 * Iteration 4 test zcopy when dbuf is no longer dirty.
4038 * Iteration 5 test zcopy when it can't be done.
4039 * Iteration 6 one more zcopy write.
4041 for (i = 0; i < 7; i++) {
4046 * In iteration 5 (i == 5) use arcbufs
4047 * that don't match bigobj blksz to test
4048 * dmu_assign_arcbuf() when it can't directly
4049 * assign an arcbuf to a dbuf.
4051 for (j = 0; j < s; j++) {
4052 if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) {
4054 dmu_request_arcbuf(bonus_db, chunksize);
4056 bigbuf_arcbufs[2 * j] =
4057 dmu_request_arcbuf(bonus_db, chunksize / 2);
4058 bigbuf_arcbufs[2 * j + 1] =
4059 dmu_request_arcbuf(bonus_db, chunksize / 2);
4064 * Get a tx for the mods to both packobj and bigobj.
4066 tx = dmu_tx_create(os);
4068 dmu_tx_hold_write(tx, packobj, packoff, packsize);
4069 dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
4071 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4073 umem_free(packbuf, packsize);
4074 umem_free(bigbuf, bigsize);
4075 for (j = 0; j < s; j++) {
4077 chunksize < (SPA_MINBLOCKSIZE * 2)) {
4078 dmu_return_arcbuf(bigbuf_arcbufs[j]);
4081 bigbuf_arcbufs[2 * j]);
4083 bigbuf_arcbufs[2 * j + 1]);
4086 umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
4087 umem_free(od, size);
4088 dmu_buf_rele(bonus_db, FTAG);
4093 * 50% of the time don't read objects in the 1st iteration to
4094 * test dmu_assign_arcbuf() for the case when there're no
4095 * existing dbufs for the specified offsets.
4097 if (i != 0 || ztest_random(2) != 0) {
4098 error = dmu_read(os, packobj, packoff,
4099 packsize, packbuf, DMU_READ_PREFETCH);
4101 error = dmu_read(os, bigobj, bigoff, bigsize,
4102 bigbuf, DMU_READ_PREFETCH);
4105 compare_and_update_pbbufs(s, packbuf, bigbuf, bigsize,
4109 * We've verified all the old bufwads, and made new ones.
4110 * Now write them out.
4112 dmu_write(os, packobj, packoff, packsize, packbuf, tx);
4113 if (ztest_opts.zo_verbose >= 7) {
4114 (void) printf("writing offset %llx size %llx"
4116 (u_longlong_t)bigoff,
4117 (u_longlong_t)bigsize,
4120 for (off = bigoff, j = 0; j < s; j++, off += chunksize) {
4122 if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) {
4123 bcopy((caddr_t)bigbuf + (off - bigoff),
4124 bigbuf_arcbufs[j]->b_data, chunksize);
4126 bcopy((caddr_t)bigbuf + (off - bigoff),
4127 bigbuf_arcbufs[2 * j]->b_data,
4129 bcopy((caddr_t)bigbuf + (off - bigoff) +
4131 bigbuf_arcbufs[2 * j + 1]->b_data,
4136 VERIFY(dmu_buf_hold(os, bigobj, off,
4137 FTAG, &dbt, DMU_READ_NO_PREFETCH) == 0);
4139 if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) {
4140 dmu_assign_arcbuf(bonus_db, off,
4141 bigbuf_arcbufs[j], tx);
4143 dmu_assign_arcbuf(bonus_db, off,
4144 bigbuf_arcbufs[2 * j], tx);
4145 dmu_assign_arcbuf(bonus_db,
4146 off + chunksize / 2,
4147 bigbuf_arcbufs[2 * j + 1], tx);
4150 dmu_buf_rele(dbt, FTAG);
4156 * Sanity check the stuff we just wrote.
4159 void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
4160 void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
4162 VERIFY(0 == dmu_read(os, packobj, packoff,
4163 packsize, packcheck, DMU_READ_PREFETCH));
4164 VERIFY(0 == dmu_read(os, bigobj, bigoff,
4165 bigsize, bigcheck, DMU_READ_PREFETCH));
4167 ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
4168 ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
4170 umem_free(packcheck, packsize);
4171 umem_free(bigcheck, bigsize);
4174 txg_wait_open(dmu_objset_pool(os), 0);
4175 } else if (i == 3) {
4176 txg_wait_synced(dmu_objset_pool(os), 0);
4180 dmu_buf_rele(bonus_db, FTAG);
4181 umem_free(packbuf, packsize);
4182 umem_free(bigbuf, bigsize);
4183 umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
4184 umem_free(od, size);
4189 ztest_dmu_write_parallel(ztest_ds_t *zd, uint64_t id)
4193 od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
4194 uint64_t offset = (1ULL << (ztest_random(20) + 43)) +
4195 (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
4198 * Have multiple threads write to large offsets in an object
4199 * to verify that parallel writes to an object -- even to the
4200 * same blocks within the object -- doesn't cause any trouble.
4202 ztest_od_init(od, ID_PARALLEL, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
4204 if (ztest_object_init(zd, od, sizeof (ztest_od_t), B_FALSE) != 0)
4207 while (ztest_random(10) != 0)
4208 ztest_io(zd, od->od_object, offset);
4210 umem_free(od, sizeof (ztest_od_t));
4214 ztest_dmu_prealloc(ztest_ds_t *zd, uint64_t id)
4217 uint64_t offset = (1ULL << (ztest_random(4) + SPA_MAXBLOCKSHIFT)) +
4218 (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
4219 uint64_t count = ztest_random(20) + 1;
4220 uint64_t blocksize = ztest_random_blocksize();
4223 od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
4225 ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
4227 if (ztest_object_init(zd, od, sizeof (ztest_od_t),
4228 !ztest_random(2)) != 0) {
4229 umem_free(od, sizeof (ztest_od_t));
4233 if (ztest_truncate(zd, od->od_object, offset, count * blocksize) != 0) {
4234 umem_free(od, sizeof (ztest_od_t));
4238 ztest_prealloc(zd, od->od_object, offset, count * blocksize);
4240 data = umem_zalloc(blocksize, UMEM_NOFAIL);
4242 while (ztest_random(count) != 0) {
4243 uint64_t randoff = offset + (ztest_random(count) * blocksize);
4244 if (ztest_write(zd, od->od_object, randoff, blocksize,
4247 while (ztest_random(4) != 0)
4248 ztest_io(zd, od->od_object, randoff);
4251 umem_free(data, blocksize);
4252 umem_free(od, sizeof (ztest_od_t));
4256 * Verify that zap_{create,destroy,add,remove,update} work as expected.
4258 #define ZTEST_ZAP_MIN_INTS 1
4259 #define ZTEST_ZAP_MAX_INTS 4
4260 #define ZTEST_ZAP_MAX_PROPS 1000
4263 ztest_zap(ztest_ds_t *zd, uint64_t id)
4265 objset_t *os = zd->zd_os;
4268 uint64_t txg, last_txg;
4269 uint64_t value[ZTEST_ZAP_MAX_INTS];
4270 uint64_t zl_ints, zl_intsize, prop;
4273 char propname[100], txgname[100];
4275 char *hc[2] = { "s.acl.h", ".s.open.h.hyLZlg" };
4277 od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
4278 ztest_od_init(od, id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0);
4280 if (ztest_object_init(zd, od, sizeof (ztest_od_t),
4281 !ztest_random(2)) != 0)
4284 object = od->od_object;
4287 * Generate a known hash collision, and verify that
4288 * we can lookup and remove both entries.
4290 tx = dmu_tx_create(os);
4291 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4292 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4295 for (i = 0; i < 2; i++) {
4297 VERIFY3U(0, ==, zap_add(os, object, hc[i], sizeof (uint64_t),
4300 for (i = 0; i < 2; i++) {
4301 VERIFY3U(EEXIST, ==, zap_add(os, object, hc[i],
4302 sizeof (uint64_t), 1, &value[i], tx));
4304 zap_length(os, object, hc[i], &zl_intsize, &zl_ints));
4305 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4306 ASSERT3U(zl_ints, ==, 1);
4308 for (i = 0; i < 2; i++) {
4309 VERIFY3U(0, ==, zap_remove(os, object, hc[i], tx));
4314 * Generate a buch of random entries.
4316 ints = MAX(ZTEST_ZAP_MIN_INTS, object % ZTEST_ZAP_MAX_INTS);
4318 prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
4319 (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
4320 (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
4321 bzero(value, sizeof (value));
4325 * If these zap entries already exist, validate their contents.
4327 error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
4329 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4330 ASSERT3U(zl_ints, ==, 1);
4332 VERIFY(zap_lookup(os, object, txgname, zl_intsize,
4333 zl_ints, &last_txg) == 0);
4335 VERIFY(zap_length(os, object, propname, &zl_intsize,
4338 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4339 ASSERT3U(zl_ints, ==, ints);
4341 VERIFY(zap_lookup(os, object, propname, zl_intsize,
4342 zl_ints, value) == 0);
4344 for (i = 0; i < ints; i++) {
4345 ASSERT3U(value[i], ==, last_txg + object + i);
4348 ASSERT3U(error, ==, ENOENT);
4352 * Atomically update two entries in our zap object.
4353 * The first is named txg_%llu, and contains the txg
4354 * in which the property was last updated. The second
4355 * is named prop_%llu, and the nth element of its value
4356 * should be txg + object + n.
4358 tx = dmu_tx_create(os);
4359 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4360 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4365 fatal(0, "zap future leak: old %llu new %llu", last_txg, txg);
4367 for (i = 0; i < ints; i++)
4368 value[i] = txg + object + i;
4370 VERIFY3U(0, ==, zap_update(os, object, txgname, sizeof (uint64_t),
4372 VERIFY3U(0, ==, zap_update(os, object, propname, sizeof (uint64_t),
4378 * Remove a random pair of entries.
4380 prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
4381 (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
4382 (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
4384 error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
4386 if (error == ENOENT)
4391 tx = dmu_tx_create(os);
4392 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4393 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4396 VERIFY3U(0, ==, zap_remove(os, object, txgname, tx));
4397 VERIFY3U(0, ==, zap_remove(os, object, propname, tx));
4400 umem_free(od, sizeof (ztest_od_t));
4404 * Testcase to test the upgrading of a microzap to fatzap.
4407 ztest_fzap(ztest_ds_t *zd, uint64_t id)
4409 objset_t *os = zd->zd_os;
4411 uint64_t object, txg;
4414 od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
4415 ztest_od_init(od, id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0);
4417 if (ztest_object_init(zd, od, sizeof (ztest_od_t),
4418 !ztest_random(2)) != 0)
4420 object = od->od_object;
4423 * Add entries to this ZAP and make sure it spills over
4424 * and gets upgraded to a fatzap. Also, since we are adding
4425 * 2050 entries we should see ptrtbl growth and leaf-block split.
4427 for (i = 0; i < 2050; i++) {
4428 char name[MAXNAMELEN];
4433 (void) snprintf(name, sizeof (name), "fzap-%llu-%llu",
4434 (u_longlong_t)id, (u_longlong_t)value);
4436 tx = dmu_tx_create(os);
4437 dmu_tx_hold_zap(tx, object, B_TRUE, name);
4438 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4441 error = zap_add(os, object, name, sizeof (uint64_t), 1,
4443 ASSERT(error == 0 || error == EEXIST);
4447 umem_free(od, sizeof (ztest_od_t));
4452 ztest_zap_parallel(ztest_ds_t *zd, uint64_t id)
4454 objset_t *os = zd->zd_os;
4456 uint64_t txg, object, count, wsize, wc, zl_wsize, zl_wc;
4458 int i, namelen, error;
4459 int micro = ztest_random(2);
4460 char name[20], string_value[20];
4463 od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
4464 ztest_od_init(od, ID_PARALLEL, FTAG, micro, DMU_OT_ZAP_OTHER, 0, 0);
4466 if (ztest_object_init(zd, od, sizeof (ztest_od_t), B_FALSE) != 0) {
4467 umem_free(od, sizeof (ztest_od_t));
4471 object = od->od_object;
4474 * Generate a random name of the form 'xxx.....' where each
4475 * x is a random printable character and the dots are dots.
4476 * There are 94 such characters, and the name length goes from
4477 * 6 to 20, so there are 94^3 * 15 = 12,458,760 possible names.
4479 namelen = ztest_random(sizeof (name) - 5) + 5 + 1;
4481 for (i = 0; i < 3; i++)
4482 name[i] = '!' + ztest_random('~' - '!' + 1);
4483 for (; i < namelen - 1; i++)
4487 if ((namelen & 1) || micro) {
4488 wsize = sizeof (txg);
4494 data = string_value;
4498 VERIFY0(zap_count(os, object, &count));
4499 ASSERT(count != -1ULL);
4502 * Select an operation: length, lookup, add, update, remove.
4504 i = ztest_random(5);
4507 tx = dmu_tx_create(os);
4508 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4509 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4512 bcopy(name, string_value, namelen);
4516 bzero(string_value, namelen);
4522 error = zap_length(os, object, name, &zl_wsize, &zl_wc);
4524 ASSERT3U(wsize, ==, zl_wsize);
4525 ASSERT3U(wc, ==, zl_wc);
4527 ASSERT3U(error, ==, ENOENT);
4532 error = zap_lookup(os, object, name, wsize, wc, data);
4534 if (data == string_value &&
4535 bcmp(name, data, namelen) != 0)
4536 fatal(0, "name '%s' != val '%s' len %d",
4537 name, data, namelen);
4539 ASSERT3U(error, ==, ENOENT);
4544 error = zap_add(os, object, name, wsize, wc, data, tx);
4545 ASSERT(error == 0 || error == EEXIST);
4549 VERIFY(zap_update(os, object, name, wsize, wc, data, tx) == 0);
4553 error = zap_remove(os, object, name, tx);
4554 ASSERT(error == 0 || error == ENOENT);
4561 umem_free(od, sizeof (ztest_od_t));
4565 * Commit callback data.
4567 typedef struct ztest_cb_data {
4568 list_node_t zcd_node;
4570 int zcd_expected_err;
4571 boolean_t zcd_added;
4572 boolean_t zcd_called;
4576 /* This is the actual commit callback function */
4578 ztest_commit_callback(void *arg, int error)
4580 ztest_cb_data_t *data = arg;
4581 uint64_t synced_txg;
4583 VERIFY(data != NULL);
4584 VERIFY3S(data->zcd_expected_err, ==, error);
4585 VERIFY(!data->zcd_called);
4587 synced_txg = spa_last_synced_txg(data->zcd_spa);
4588 if (data->zcd_txg > synced_txg)
4589 fatal(0, "commit callback of txg %" PRIu64 " called prematurely"
4590 ", last synced txg = %" PRIu64 "\n", data->zcd_txg,
4593 data->zcd_called = B_TRUE;
4595 if (error == ECANCELED) {
4596 ASSERT0(data->zcd_txg);
4597 ASSERT(!data->zcd_added);
4600 * The private callback data should be destroyed here, but
4601 * since we are going to check the zcd_called field after
4602 * dmu_tx_abort(), we will destroy it there.
4607 ASSERT(data->zcd_added);
4608 ASSERT3U(data->zcd_txg, !=, 0);
4610 (void) mutex_enter(&zcl.zcl_callbacks_lock);
4612 /* See if this cb was called more quickly */
4613 if ((synced_txg - data->zcd_txg) < zc_min_txg_delay)
4614 zc_min_txg_delay = synced_txg - data->zcd_txg;
4616 /* Remove our callback from the list */
4617 list_remove(&zcl.zcl_callbacks, data);
4619 (void) mutex_exit(&zcl.zcl_callbacks_lock);
4621 umem_free(data, sizeof (ztest_cb_data_t));
4624 /* Allocate and initialize callback data structure */
4625 static ztest_cb_data_t *
4626 ztest_create_cb_data(objset_t *os, uint64_t txg)
4628 ztest_cb_data_t *cb_data;
4630 cb_data = umem_zalloc(sizeof (ztest_cb_data_t), UMEM_NOFAIL);
4632 cb_data->zcd_txg = txg;
4633 cb_data->zcd_spa = dmu_objset_spa(os);
4634 list_link_init(&cb_data->zcd_node);
4640 * Commit callback test.
4643 ztest_dmu_commit_callbacks(ztest_ds_t *zd, uint64_t id)
4645 objset_t *os = zd->zd_os;
4648 ztest_cb_data_t *cb_data[3], *tmp_cb;
4649 uint64_t old_txg, txg;
4652 od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
4653 ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
4655 if (ztest_object_init(zd, od, sizeof (ztest_od_t), B_FALSE) != 0) {
4656 umem_free(od, sizeof (ztest_od_t));
4660 tx = dmu_tx_create(os);
4662 cb_data[0] = ztest_create_cb_data(os, 0);
4663 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[0]);
4665 dmu_tx_hold_write(tx, od->od_object, 0, sizeof (uint64_t));
4667 /* Every once in a while, abort the transaction on purpose */
4668 if (ztest_random(100) == 0)
4672 error = dmu_tx_assign(tx, TXG_NOWAIT);
4674 txg = error ? 0 : dmu_tx_get_txg(tx);
4676 cb_data[0]->zcd_txg = txg;
4677 cb_data[1] = ztest_create_cb_data(os, txg);
4678 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[1]);
4682 * It's not a strict requirement to call the registered
4683 * callbacks from inside dmu_tx_abort(), but that's what
4684 * it's supposed to happen in the current implementation
4685 * so we will check for that.
4687 for (i = 0; i < 2; i++) {
4688 cb_data[i]->zcd_expected_err = ECANCELED;
4689 VERIFY(!cb_data[i]->zcd_called);
4694 for (i = 0; i < 2; i++) {
4695 VERIFY(cb_data[i]->zcd_called);
4696 umem_free(cb_data[i], sizeof (ztest_cb_data_t));
4699 umem_free(od, sizeof (ztest_od_t));
4703 cb_data[2] = ztest_create_cb_data(os, txg);
4704 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[2]);
4707 * Read existing data to make sure there isn't a future leak.
4709 VERIFY(0 == dmu_read(os, od->od_object, 0, sizeof (uint64_t),
4710 &old_txg, DMU_READ_PREFETCH));
4713 fatal(0, "future leak: got %" PRIu64 ", open txg is %" PRIu64,
4716 dmu_write(os, od->od_object, 0, sizeof (uint64_t), &txg, tx);
4718 (void) mutex_enter(&zcl.zcl_callbacks_lock);
4721 * Since commit callbacks don't have any ordering requirement and since
4722 * it is theoretically possible for a commit callback to be called
4723 * after an arbitrary amount of time has elapsed since its txg has been
4724 * synced, it is difficult to reliably determine whether a commit
4725 * callback hasn't been called due to high load or due to a flawed
4728 * In practice, we will assume that if after a certain number of txgs a
4729 * commit callback hasn't been called, then most likely there's an
4730 * implementation bug..
4732 tmp_cb = list_head(&zcl.zcl_callbacks);
4733 if (tmp_cb != NULL &&
4734 tmp_cb->zcd_txg + ZTEST_COMMIT_CB_THRESH < txg) {
4735 fatal(0, "Commit callback threshold exceeded, oldest txg: %"
4736 PRIu64 ", open txg: %" PRIu64 "\n", tmp_cb->zcd_txg, txg);
4740 * Let's find the place to insert our callbacks.
4742 * Even though the list is ordered by txg, it is possible for the
4743 * insertion point to not be the end because our txg may already be
4744 * quiescing at this point and other callbacks in the open txg
4745 * (from other objsets) may have sneaked in.
4747 tmp_cb = list_tail(&zcl.zcl_callbacks);
4748 while (tmp_cb != NULL && tmp_cb->zcd_txg > txg)
4749 tmp_cb = list_prev(&zcl.zcl_callbacks, tmp_cb);
4751 /* Add the 3 callbacks to the list */
4752 for (i = 0; i < 3; i++) {
4754 list_insert_head(&zcl.zcl_callbacks, cb_data[i]);
4756 list_insert_after(&zcl.zcl_callbacks, tmp_cb,
4759 cb_data[i]->zcd_added = B_TRUE;
4760 VERIFY(!cb_data[i]->zcd_called);
4762 tmp_cb = cb_data[i];
4767 (void) mutex_exit(&zcl.zcl_callbacks_lock);
4771 umem_free(od, sizeof (ztest_od_t));
4776 ztest_dsl_prop_get_set(ztest_ds_t *zd, uint64_t id)
4778 zfs_prop_t proplist[] = {
4780 ZFS_PROP_COMPRESSION,
4786 (void) rw_rdlock(&ztest_name_lock);
4788 for (p = 0; p < sizeof (proplist) / sizeof (proplist[0]); p++)
4789 (void) ztest_dsl_prop_set_uint64(zd->zd_name, proplist[p],
4790 ztest_random_dsl_prop(proplist[p]), (int)ztest_random(2));
4792 VERIFY0(ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_RECORDSIZE,
4793 ztest_random_blocksize(), (int)ztest_random(2)));
4795 (void) rw_unlock(&ztest_name_lock);
4800 ztest_spa_prop_get_set(ztest_ds_t *zd, uint64_t id)
4802 nvlist_t *props = NULL;
4804 (void) rw_rdlock(&ztest_name_lock);
4806 (void) ztest_spa_prop_set_uint64(ZPOOL_PROP_DEDUPDITTO,
4807 ZIO_DEDUPDITTO_MIN + ztest_random(ZIO_DEDUPDITTO_MIN));
4809 VERIFY0(spa_prop_get(ztest_spa, &props));
4811 if (ztest_opts.zo_verbose >= 6)
4812 dump_nvlist(props, 4);
4816 (void) rw_unlock(&ztest_name_lock);
4820 user_release_one(const char *snapname, const char *holdname)
4822 nvlist_t *snaps, *holds;
4825 snaps = fnvlist_alloc();
4826 holds = fnvlist_alloc();
4827 fnvlist_add_boolean(holds, holdname);
4828 fnvlist_add_nvlist(snaps, snapname, holds);
4829 fnvlist_free(holds);
4830 error = dsl_dataset_user_release(snaps, NULL);
4831 fnvlist_free(snaps);
4836 * Test snapshot hold/release and deferred destroy.
4839 ztest_dmu_snapshot_hold(ztest_ds_t *zd, uint64_t id)
4842 objset_t *os = zd->zd_os;
4846 char clonename[100];
4848 char osname[MAXNAMELEN];
4851 (void) rw_rdlock(&ztest_name_lock);
4853 dmu_objset_name(os, osname);
4855 (void) snprintf(snapname, sizeof (snapname), "sh1_%llu",
4857 (void) snprintf(fullname, sizeof (fullname), "%s@%s", osname, snapname);
4858 (void) snprintf(clonename, sizeof (clonename),
4859 "%s/ch1_%llu", osname, (u_longlong_t)id);
4860 (void) snprintf(tag, sizeof (tag), "tag_%llu", (u_longlong_t)id);
4863 * Clean up from any previous run.
4865 error = dsl_destroy_head(clonename);
4866 if (error != ENOENT)
4868 error = user_release_one(fullname, tag);
4869 if (error != ESRCH && error != ENOENT)
4871 error = dsl_destroy_snapshot(fullname, B_FALSE);
4872 if (error != ENOENT)
4876 * Create snapshot, clone it, mark snap for deferred destroy,
4877 * destroy clone, verify snap was also destroyed.
4879 error = dmu_objset_snapshot_one(osname, snapname);
4881 if (error == ENOSPC) {
4882 ztest_record_enospc("dmu_objset_snapshot");
4885 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
4888 error = dmu_objset_clone(clonename, fullname);
4890 if (error == ENOSPC) {
4891 ztest_record_enospc("dmu_objset_clone");
4894 fatal(0, "dmu_objset_clone(%s) = %d", clonename, error);
4897 error = dsl_destroy_snapshot(fullname, B_TRUE);
4899 fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
4903 error = dsl_destroy_head(clonename);
4905 fatal(0, "dsl_destroy_head(%s) = %d", clonename, error);
4907 error = dmu_objset_hold(fullname, FTAG, &origin);
4908 if (error != ENOENT)
4909 fatal(0, "dmu_objset_hold(%s) = %d", fullname, error);
4912 * Create snapshot, add temporary hold, verify that we can't
4913 * destroy a held snapshot, mark for deferred destroy,
4914 * release hold, verify snapshot was destroyed.
4916 error = dmu_objset_snapshot_one(osname, snapname);
4918 if (error == ENOSPC) {
4919 ztest_record_enospc("dmu_objset_snapshot");
4922 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
4925 holds = fnvlist_alloc();
4926 fnvlist_add_string(holds, fullname, tag);
4927 error = dsl_dataset_user_hold(holds, 0, NULL);
4928 fnvlist_free(holds);
4930 if (error == ENOSPC) {
4931 ztest_record_enospc("dsl_dataset_user_hold");
4934 fatal(0, "dsl_dataset_user_hold(%s, %s) = %u",
4935 fullname, tag, error);
4938 error = dsl_destroy_snapshot(fullname, B_FALSE);
4939 if (error != EBUSY) {
4940 fatal(0, "dsl_destroy_snapshot(%s, B_FALSE) = %d",
4944 error = dsl_destroy_snapshot(fullname, B_TRUE);
4946 fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
4950 error = user_release_one(fullname, tag);
4952 fatal(0, "user_release_one(%s, %s) = %d", fullname, tag, error);
4954 VERIFY3U(dmu_objset_hold(fullname, FTAG, &origin), ==, ENOENT);
4957 (void) rw_unlock(&ztest_name_lock);
4961 * Inject random faults into the on-disk data.
4965 ztest_fault_inject(ztest_ds_t *zd, uint64_t id)
4967 ztest_shared_t *zs = ztest_shared;
4968 spa_t *spa = ztest_spa;
4972 uint64_t bad = 0x1990c0ffeedecadeull;
4977 int bshift = SPA_OLD_MAXBLOCKSHIFT + 2; /* don't scrog all labels */
4983 boolean_t islog = B_FALSE;
4985 path0 = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
4986 pathrand = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
4988 mutex_enter(&ztest_vdev_lock);
4989 maxfaults = MAXFAULTS();
4990 leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz;
4991 mirror_save = zs->zs_mirrors;
4992 mutex_exit(&ztest_vdev_lock);
4994 ASSERT(leaves >= 1);
4997 * Grab the name lock as reader. There are some operations
4998 * which don't like to have their vdevs changed while
4999 * they are in progress (i.e. spa_change_guid). Those
5000 * operations will have grabbed the name lock as writer.
5002 (void) rw_rdlock(&ztest_name_lock);
5005 * We need SCL_STATE here because we're going to look at vd0->vdev_tsd.
5007 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5009 if (ztest_random(2) == 0) {
5011 * Inject errors on a normal data device or slog device.
5013 top = ztest_random_vdev_top(spa, B_TRUE);
5014 leaf = ztest_random(leaves) + zs->zs_splits;
5017 * Generate paths to the first leaf in this top-level vdev,
5018 * and to the random leaf we selected. We'll induce transient
5019 * write failures and random online/offline activity on leaf 0,
5020 * and we'll write random garbage to the randomly chosen leaf.
5022 (void) snprintf(path0, MAXPATHLEN, ztest_dev_template,
5023 ztest_opts.zo_dir, ztest_opts.zo_pool,
5024 top * leaves + zs->zs_splits);
5025 (void) snprintf(pathrand, MAXPATHLEN, ztest_dev_template,
5026 ztest_opts.zo_dir, ztest_opts.zo_pool,
5027 top * leaves + leaf);
5029 vd0 = vdev_lookup_by_path(spa->spa_root_vdev, path0);
5030 if (vd0 != NULL && vd0->vdev_top->vdev_islog)
5034 * If the top-level vdev needs to be resilvered
5035 * then we only allow faults on the device that is
5038 if (vd0 != NULL && maxfaults != 1 &&
5039 (!vdev_resilver_needed(vd0->vdev_top, NULL, NULL) ||
5040 vd0->vdev_resilver_txg != 0)) {
5042 * Make vd0 explicitly claim to be unreadable,
5043 * or unwriteable, or reach behind its back
5044 * and close the underlying fd. We can do this if
5045 * maxfaults == 0 because we'll fail and reexecute,
5046 * and we can do it if maxfaults >= 2 because we'll
5047 * have enough redundancy. If maxfaults == 1, the
5048 * combination of this with injection of random data
5049 * corruption below exceeds the pool's fault tolerance.
5051 vdev_file_t *vf = vd0->vdev_tsd;
5053 if (vf != NULL && ztest_random(3) == 0) {
5054 (void) close(vf->vf_vnode->v_fd);
5055 vf->vf_vnode->v_fd = -1;
5056 } else if (ztest_random(2) == 0) {
5057 vd0->vdev_cant_read = B_TRUE;
5059 vd0->vdev_cant_write = B_TRUE;
5061 guid0 = vd0->vdev_guid;
5065 * Inject errors on an l2cache device.
5067 spa_aux_vdev_t *sav = &spa->spa_l2cache;
5069 if (sav->sav_count == 0) {
5070 spa_config_exit(spa, SCL_STATE, FTAG);
5071 (void) rw_unlock(&ztest_name_lock);
5074 vd0 = sav->sav_vdevs[ztest_random(sav->sav_count)];
5075 guid0 = vd0->vdev_guid;
5076 (void) strcpy(path0, vd0->vdev_path);
5077 (void) strcpy(pathrand, vd0->vdev_path);
5081 maxfaults = INT_MAX; /* no limit on cache devices */
5084 spa_config_exit(spa, SCL_STATE, FTAG);
5085 (void) rw_unlock(&ztest_name_lock);
5088 * If we can tolerate two or more faults, or we're dealing
5089 * with a slog, randomly online/offline vd0.
5091 if ((maxfaults >= 2 || islog) && guid0 != 0) {
5092 if (ztest_random(10) < 6) {
5093 int flags = (ztest_random(2) == 0 ?
5094 ZFS_OFFLINE_TEMPORARY : 0);
5097 * We have to grab the zs_name_lock as writer to
5098 * prevent a race between offlining a slog and
5099 * destroying a dataset. Offlining the slog will
5100 * grab a reference on the dataset which may cause
5101 * dsl_destroy_head() to fail with EBUSY thus
5102 * leaving the dataset in an inconsistent state.
5105 (void) rw_wrlock(&ztest_name_lock);
5107 VERIFY(vdev_offline(spa, guid0, flags) != EBUSY);
5110 (void) rw_unlock(&ztest_name_lock);
5113 * Ideally we would like to be able to randomly
5114 * call vdev_[on|off]line without holding locks
5115 * to force unpredictable failures but the side
5116 * effects of vdev_[on|off]line prevent us from
5117 * doing so. We grab the ztest_vdev_lock here to
5118 * prevent a race between injection testing and
5121 mutex_enter(&ztest_vdev_lock);
5122 (void) vdev_online(spa, guid0, 0, NULL);
5123 mutex_exit(&ztest_vdev_lock);
5131 * We have at least single-fault tolerance, so inject data corruption.
5133 fd = open(pathrand, O_RDWR);
5135 if (fd == -1) /* we hit a gap in the device namespace */
5138 fsize = lseek(fd, 0, SEEK_END);
5140 while (--iters != 0) {
5141 offset = ztest_random(fsize / (leaves << bshift)) *
5142 (leaves << bshift) + (leaf << bshift) +
5143 (ztest_random(1ULL << (bshift - 1)) & -8ULL);
5145 if (offset >= fsize)
5148 mutex_enter(&ztest_vdev_lock);
5149 if (mirror_save != zs->zs_mirrors) {
5150 mutex_exit(&ztest_vdev_lock);
5155 if (pwrite(fd, &bad, sizeof (bad), offset) != sizeof (bad))
5156 fatal(1, "can't inject bad word at 0x%llx in %s",
5159 mutex_exit(&ztest_vdev_lock);
5161 if (ztest_opts.zo_verbose >= 7)
5162 (void) printf("injected bad word into %s,"
5163 " offset 0x%llx\n", pathrand, (u_longlong_t)offset);
5168 umem_free(path0, MAXPATHLEN);
5169 umem_free(pathrand, MAXPATHLEN);
5173 * Verify that DDT repair works as expected.
5176 ztest_ddt_repair(ztest_ds_t *zd, uint64_t id)
5178 ztest_shared_t *zs = ztest_shared;
5179 spa_t *spa = ztest_spa;
5180 objset_t *os = zd->zd_os;
5182 uint64_t object, blocksize, txg, pattern, psize;
5183 enum zio_checksum checksum = spa_dedup_checksum(spa);
5188 int copies = 2 * ZIO_DEDUPDITTO_MIN;
5191 blocksize = ztest_random_blocksize();
5192 blocksize = MIN(blocksize, 2048); /* because we write so many */
5194 od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
5195 ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
5197 if (ztest_object_init(zd, od, sizeof (ztest_od_t), B_FALSE) != 0) {
5198 umem_free(od, sizeof (ztest_od_t));
5203 * Take the name lock as writer to prevent anyone else from changing
5204 * the pool and dataset properies we need to maintain during this test.
5206 (void) rw_wrlock(&ztest_name_lock);
5208 if (ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_DEDUP, checksum,
5210 ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_COPIES, 1,
5212 (void) rw_unlock(&ztest_name_lock);
5213 umem_free(od, sizeof (ztest_od_t));
5217 object = od[0].od_object;
5218 blocksize = od[0].od_blocksize;
5219 pattern = zs->zs_guid ^ dmu_objset_fsid_guid(os);
5221 ASSERT(object != 0);
5223 tx = dmu_tx_create(os);
5224 dmu_tx_hold_write(tx, object, 0, copies * blocksize);
5225 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
5227 (void) rw_unlock(&ztest_name_lock);
5228 umem_free(od, sizeof (ztest_od_t));
5233 * Write all the copies of our block.
5235 for (i = 0; i < copies; i++) {
5236 uint64_t offset = i * blocksize;
5237 int error = dmu_buf_hold(os, object, offset, FTAG, &db,
5238 DMU_READ_NO_PREFETCH);
5240 fatal(B_FALSE, "dmu_buf_hold(%p, %llu, %llu) = %u",
5241 os, (long long)object, (long long) offset, error);
5243 ASSERT(db->db_offset == offset);
5244 ASSERT(db->db_size == blocksize);
5245 ASSERT(ztest_pattern_match(db->db_data, db->db_size, pattern) ||
5246 ztest_pattern_match(db->db_data, db->db_size, 0ULL));
5247 dmu_buf_will_fill(db, tx);
5248 ztest_pattern_set(db->db_data, db->db_size, pattern);
5249 dmu_buf_rele(db, FTAG);
5253 txg_wait_synced(spa_get_dsl(spa), txg);
5256 * Find out what block we got.
5258 VERIFY0(dmu_buf_hold(os, object, 0, FTAG, &db,
5259 DMU_READ_NO_PREFETCH));
5260 blk = *((dmu_buf_impl_t *)db)->db_blkptr;
5261 dmu_buf_rele(db, FTAG);
5264 * Damage the block. Dedup-ditto will save us when we read it later.
5266 psize = BP_GET_PSIZE(&blk);
5267 buf = zio_buf_alloc(psize);
5268 ztest_pattern_set(buf, psize, ~pattern);
5270 (void) zio_wait(zio_rewrite(NULL, spa, 0, &blk,
5271 buf, psize, NULL, NULL, ZIO_PRIORITY_SYNC_WRITE,
5272 ZIO_FLAG_CANFAIL | ZIO_FLAG_INDUCE_DAMAGE, NULL));
5274 zio_buf_free(buf, psize);
5276 (void) rw_unlock(&ztest_name_lock);
5277 umem_free(od, sizeof (ztest_od_t));
5285 ztest_scrub(ztest_ds_t *zd, uint64_t id)
5287 spa_t *spa = ztest_spa;
5289 (void) spa_scan(spa, POOL_SCAN_SCRUB);
5290 (void) poll(NULL, 0, 100); /* wait a moment, then force a restart */
5291 (void) spa_scan(spa, POOL_SCAN_SCRUB);
5295 * Change the guid for the pool.
5299 ztest_reguid(ztest_ds_t *zd, uint64_t id)
5301 spa_t *spa = ztest_spa;
5302 uint64_t orig, load;
5305 orig = spa_guid(spa);
5306 load = spa_load_guid(spa);
5308 (void) rw_wrlock(&ztest_name_lock);
5309 error = spa_change_guid(spa);
5310 (void) rw_unlock(&ztest_name_lock);
5315 if (ztest_opts.zo_verbose >= 4) {
5316 (void) printf("Changed guid old %llu -> %llu\n",
5317 (u_longlong_t)orig, (u_longlong_t)spa_guid(spa));
5320 VERIFY3U(orig, !=, spa_guid(spa));
5321 VERIFY3U(load, ==, spa_load_guid(spa));
5325 * Rename the pool to a different name and then rename it back.
5329 ztest_spa_rename(ztest_ds_t *zd, uint64_t id)
5331 char *oldname, *newname;
5334 (void) rw_wrlock(&ztest_name_lock);
5336 oldname = ztest_opts.zo_pool;
5337 newname = umem_alloc(strlen(oldname) + 5, UMEM_NOFAIL);
5338 (void) strcpy(newname, oldname);
5339 (void) strcat(newname, "_tmp");
5344 VERIFY3U(0, ==, spa_rename(oldname, newname));
5347 * Try to open it under the old name, which shouldn't exist
5349 VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
5352 * Open it under the new name and make sure it's still the same spa_t.
5354 VERIFY3U(0, ==, spa_open(newname, &spa, FTAG));
5356 ASSERT(spa == ztest_spa);
5357 spa_close(spa, FTAG);
5360 * Rename it back to the original
5362 VERIFY3U(0, ==, spa_rename(newname, oldname));
5365 * Make sure it can still be opened
5367 VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG));
5369 ASSERT(spa == ztest_spa);
5370 spa_close(spa, FTAG);
5372 umem_free(newname, strlen(newname) + 1);
5374 (void) rw_unlock(&ztest_name_lock);
5378 * Verify pool integrity by running zdb.
5381 ztest_run_zdb(char *pool)
5389 bin = umem_alloc(MAXPATHLEN + MAXNAMELEN + 20, UMEM_NOFAIL);
5390 zdb = umem_alloc(MAXPATHLEN + MAXNAMELEN + 20, UMEM_NOFAIL);
5391 zbuf = umem_alloc(1024, UMEM_NOFAIL);
5393 VERIFY(realpath(getexecname(), bin) != NULL);
5394 if (strncmp(bin, "/usr/sbin/ztest", 15) == 0) {
5395 strcpy(bin, "/usr/sbin/zdb"); /* Installed */
5396 } else if (strncmp(bin, "/sbin/ztest", 11) == 0) {
5397 strcpy(bin, "/sbin/zdb"); /* Installed */
5399 strstr(bin, "/ztest/")[0] = '\0'; /* In-tree */
5400 strcat(bin, "/zdb/zdb");
5404 "%s -bcc%s%s -d -U %s %s",
5406 ztest_opts.zo_verbose >= 3 ? "s" : "",
5407 ztest_opts.zo_verbose >= 4 ? "v" : "",
5411 if (ztest_opts.zo_verbose >= 5)
5412 (void) printf("Executing %s\n", strstr(zdb, "zdb "));
5414 fp = popen(zdb, "r");
5416 while (fgets(zbuf, 1024, fp) != NULL)
5417 if (ztest_opts.zo_verbose >= 3)
5418 (void) printf("%s", zbuf);
5420 status = pclose(fp);
5425 ztest_dump_core = 0;
5426 if (WIFEXITED(status))
5427 fatal(0, "'%s' exit code %d", zdb, WEXITSTATUS(status));
5429 fatal(0, "'%s' died with signal %d", zdb, WTERMSIG(status));
5431 umem_free(bin, MAXPATHLEN + MAXNAMELEN + 20);
5432 umem_free(zdb, MAXPATHLEN + MAXNAMELEN + 20);
5433 umem_free(zbuf, 1024);
5437 ztest_walk_pool_directory(char *header)
5441 if (ztest_opts.zo_verbose >= 6)
5442 (void) printf("%s\n", header);
5444 mutex_enter(&spa_namespace_lock);
5445 while ((spa = spa_next(spa)) != NULL)
5446 if (ztest_opts.zo_verbose >= 6)
5447 (void) printf("\t%s\n", spa_name(spa));
5448 mutex_exit(&spa_namespace_lock);
5452 ztest_spa_import_export(char *oldname, char *newname)
5454 nvlist_t *config, *newconfig;
5459 if (ztest_opts.zo_verbose >= 4) {
5460 (void) printf("import/export: old = %s, new = %s\n",
5465 * Clean up from previous runs.
5467 (void) spa_destroy(newname);
5470 * Get the pool's configuration and guid.
5472 VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG));
5475 * Kick off a scrub to tickle scrub/export races.
5477 if (ztest_random(2) == 0)
5478 (void) spa_scan(spa, POOL_SCAN_SCRUB);
5480 pool_guid = spa_guid(spa);
5481 spa_close(spa, FTAG);
5483 ztest_walk_pool_directory("pools before export");
5488 VERIFY3U(0, ==, spa_export(oldname, &config, B_FALSE, B_FALSE));
5490 ztest_walk_pool_directory("pools after export");
5495 newconfig = spa_tryimport(config);
5496 ASSERT(newconfig != NULL);
5497 nvlist_free(newconfig);
5500 * Import it under the new name.
5502 error = spa_import(newname, config, NULL, 0);
5504 dump_nvlist(config, 0);
5505 fatal(B_FALSE, "couldn't import pool %s as %s: error %u",
5506 oldname, newname, error);
5509 ztest_walk_pool_directory("pools after import");
5512 * Try to import it again -- should fail with EEXIST.
5514 VERIFY3U(EEXIST, ==, spa_import(newname, config, NULL, 0));
5517 * Try to import it under a different name -- should fail with EEXIST.
5519 VERIFY3U(EEXIST, ==, spa_import(oldname, config, NULL, 0));
5522 * Verify that the pool is no longer visible under the old name.
5524 VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
5527 * Verify that we can open and close the pool using the new name.
5529 VERIFY3U(0, ==, spa_open(newname, &spa, FTAG));
5530 ASSERT(pool_guid == spa_guid(spa));
5531 spa_close(spa, FTAG);
5533 nvlist_free(config);
5537 ztest_resume(spa_t *spa)
5539 if (spa_suspended(spa) && ztest_opts.zo_verbose >= 6)
5540 (void) printf("resuming from suspended state\n");
5541 spa_vdev_state_enter(spa, SCL_NONE);
5542 vdev_clear(spa, NULL);
5543 (void) spa_vdev_state_exit(spa, NULL, 0);
5544 (void) zio_resume(spa);
5548 ztest_resume_thread(void *arg)
5552 while (!ztest_exiting) {
5553 if (spa_suspended(spa))
5555 (void) poll(NULL, 0, 100);
5567 ztest_deadman_alarm(int sig)
5569 fatal(0, "failed to complete within %d seconds of deadline", GRACE);
5574 ztest_execute(int test, ztest_info_t *zi, uint64_t id)
5576 ztest_ds_t *zd = &ztest_ds[id % ztest_opts.zo_datasets];
5577 ztest_shared_callstate_t *zc = ZTEST_GET_SHARED_CALLSTATE(test);
5578 hrtime_t functime = gethrtime();
5581 for (i = 0; i < zi->zi_iters; i++)
5582 zi->zi_func(zd, id);
5584 functime = gethrtime() - functime;
5586 atomic_add_64(&zc->zc_count, 1);
5587 atomic_add_64(&zc->zc_time, functime);
5589 if (ztest_opts.zo_verbose >= 4)
5590 (void) printf("%6.2f sec in %s\n",
5591 (double)functime / NANOSEC, zi->zi_funcname);
5595 ztest_thread(void *arg)
5598 uint64_t id = (uintptr_t)arg;
5599 ztest_shared_t *zs = ztest_shared;
5603 ztest_shared_callstate_t *zc;
5605 while ((now = gethrtime()) < zs->zs_thread_stop) {
5607 * See if it's time to force a crash.
5609 if (now > zs->zs_thread_kill)
5613 * If we're getting ENOSPC with some regularity, stop.
5615 if (zs->zs_enospc_count > 10)
5619 * Pick a random function to execute.
5621 rand = ztest_random(ZTEST_FUNCS);
5622 zi = &ztest_info[rand];
5623 zc = ZTEST_GET_SHARED_CALLSTATE(rand);
5624 call_next = zc->zc_next;
5626 if (now >= call_next &&
5627 atomic_cas_64(&zc->zc_next, call_next, call_next +
5628 ztest_random(2 * zi->zi_interval[0] + 1)) == call_next) {
5629 ztest_execute(rand, zi, id);
5639 ztest_dataset_name(char *dsname, char *pool, int d)
5641 (void) snprintf(dsname, MAXNAMELEN, "%s/ds_%d", pool, d);
5645 ztest_dataset_destroy(int d)
5647 char name[MAXNAMELEN];
5650 ztest_dataset_name(name, ztest_opts.zo_pool, d);
5652 if (ztest_opts.zo_verbose >= 3)
5653 (void) printf("Destroying %s to free up space\n", name);
5656 * Cleanup any non-standard clones and snapshots. In general,
5657 * ztest thread t operates on dataset (t % zopt_datasets),
5658 * so there may be more than one thing to clean up.
5660 for (t = d; t < ztest_opts.zo_threads;
5661 t += ztest_opts.zo_datasets)
5662 ztest_dsl_dataset_cleanup(name, t);
5664 (void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
5665 DS_FIND_SNAPSHOTS | DS_FIND_CHILDREN);
5669 ztest_dataset_dirobj_verify(ztest_ds_t *zd)
5671 uint64_t usedobjs, dirobjs, scratch;
5674 * ZTEST_DIROBJ is the object directory for the entire dataset.
5675 * Therefore, the number of objects in use should equal the
5676 * number of ZTEST_DIROBJ entries, +1 for ZTEST_DIROBJ itself.
5677 * If not, we have an object leak.
5679 * Note that we can only check this in ztest_dataset_open(),
5680 * when the open-context and syncing-context values agree.
5681 * That's because zap_count() returns the open-context value,
5682 * while dmu_objset_space() returns the rootbp fill count.
5684 VERIFY3U(0, ==, zap_count(zd->zd_os, ZTEST_DIROBJ, &dirobjs));
5685 dmu_objset_space(zd->zd_os, &scratch, &scratch, &usedobjs, &scratch);
5686 ASSERT3U(dirobjs + 1, ==, usedobjs);
5690 ztest_dataset_open(int d)
5692 ztest_ds_t *zd = &ztest_ds[d];
5693 uint64_t committed_seq = ZTEST_GET_SHARED_DS(d)->zd_seq;
5696 char name[MAXNAMELEN];
5699 ztest_dataset_name(name, ztest_opts.zo_pool, d);
5701 (void) rw_rdlock(&ztest_name_lock);
5703 error = ztest_dataset_create(name);
5704 if (error == ENOSPC) {
5705 (void) rw_unlock(&ztest_name_lock);
5706 ztest_record_enospc(FTAG);
5709 ASSERT(error == 0 || error == EEXIST);
5711 VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, zd, &os));
5712 (void) rw_unlock(&ztest_name_lock);
5714 ztest_zd_init(zd, ZTEST_GET_SHARED_DS(d), os);
5716 zilog = zd->zd_zilog;
5718 if (zilog->zl_header->zh_claim_lr_seq != 0 &&
5719 zilog->zl_header->zh_claim_lr_seq < committed_seq)
5720 fatal(0, "missing log records: claimed %llu < committed %llu",
5721 zilog->zl_header->zh_claim_lr_seq, committed_seq);
5723 ztest_dataset_dirobj_verify(zd);
5725 zil_replay(os, zd, ztest_replay_vector);
5727 ztest_dataset_dirobj_verify(zd);
5729 if (ztest_opts.zo_verbose >= 6)
5730 (void) printf("%s replay %llu blocks, %llu records, seq %llu\n",
5732 (u_longlong_t)zilog->zl_parse_blk_count,
5733 (u_longlong_t)zilog->zl_parse_lr_count,
5734 (u_longlong_t)zilog->zl_replaying_seq);
5736 zilog = zil_open(os, ztest_get_data);
5738 if (zilog->zl_replaying_seq != 0 &&
5739 zilog->zl_replaying_seq < committed_seq)
5740 fatal(0, "missing log records: replayed %llu < committed %llu",
5741 zilog->zl_replaying_seq, committed_seq);
5747 ztest_dataset_close(int d)
5749 ztest_ds_t *zd = &ztest_ds[d];
5751 zil_close(zd->zd_zilog);
5752 dmu_objset_disown(zd->zd_os, zd);
5758 * Kick off threads to run tests on all datasets in parallel.
5761 ztest_run(ztest_shared_t *zs)
5766 kthread_t *resume_thread;
5771 ztest_exiting = B_FALSE;
5774 * Initialize parent/child shared state.
5776 mutex_init(&ztest_vdev_lock, NULL, MUTEX_DEFAULT, NULL);
5777 VERIFY(rwlock_init(&ztest_name_lock, USYNC_THREAD, NULL) == 0);
5779 zs->zs_thread_start = gethrtime();
5780 zs->zs_thread_stop =
5781 zs->zs_thread_start + ztest_opts.zo_passtime * NANOSEC;
5782 zs->zs_thread_stop = MIN(zs->zs_thread_stop, zs->zs_proc_stop);
5783 zs->zs_thread_kill = zs->zs_thread_stop;
5784 if (ztest_random(100) < ztest_opts.zo_killrate) {
5785 zs->zs_thread_kill -=
5786 ztest_random(ztest_opts.zo_passtime * NANOSEC);
5789 mutex_init(&zcl.zcl_callbacks_lock, NULL, MUTEX_DEFAULT, NULL);
5791 list_create(&zcl.zcl_callbacks, sizeof (ztest_cb_data_t),
5792 offsetof(ztest_cb_data_t, zcd_node));
5797 kernel_init(FREAD | FWRITE);
5798 VERIFY0(spa_open(ztest_opts.zo_pool, &spa, FTAG));
5799 spa->spa_debug = B_TRUE;
5800 metaslab_preload_limit = ztest_random(20) + 1;
5803 VERIFY0(dmu_objset_own(ztest_opts.zo_pool,
5804 DMU_OST_ANY, B_TRUE, FTAG, &os));
5805 zs->zs_guid = dmu_objset_fsid_guid(os);
5806 dmu_objset_disown(os, FTAG);
5808 spa->spa_dedup_ditto = 2 * ZIO_DEDUPDITTO_MIN;
5811 * We don't expect the pool to suspend unless maxfaults == 0,
5812 * in which case ztest_fault_inject() temporarily takes away
5813 * the only valid replica.
5815 if (MAXFAULTS() == 0)
5816 spa->spa_failmode = ZIO_FAILURE_MODE_WAIT;
5818 spa->spa_failmode = ZIO_FAILURE_MODE_PANIC;
5821 * Create a thread to periodically resume suspended I/O.
5823 VERIFY3P((resume_thread = zk_thread_create(NULL, 0,
5824 (thread_func_t)ztest_resume_thread, spa, TS_RUN, NULL, 0, 0,
5825 PTHREAD_CREATE_JOINABLE)), !=, NULL);
5829 * Set a deadman alarm to abort() if we hang.
5831 signal(SIGALRM, ztest_deadman_alarm);
5832 alarm((zs->zs_thread_stop - zs->zs_thread_start) / NANOSEC + GRACE);
5836 * Verify that we can safely inquire about about any object,
5837 * whether it's allocated or not. To make it interesting,
5838 * we probe a 5-wide window around each power of two.
5839 * This hits all edge cases, including zero and the max.
5841 for (t = 0; t < 64; t++) {
5842 for (d = -5; d <= 5; d++) {
5843 error = dmu_object_info(spa->spa_meta_objset,
5844 (1ULL << t) + d, NULL);
5845 ASSERT(error == 0 || error == ENOENT ||
5851 * If we got any ENOSPC errors on the previous run, destroy something.
5853 if (zs->zs_enospc_count != 0) {
5854 int d = ztest_random(ztest_opts.zo_datasets);
5855 ztest_dataset_destroy(d);
5857 zs->zs_enospc_count = 0;
5859 tid = umem_zalloc(ztest_opts.zo_threads * sizeof (kt_did_t),
5862 if (ztest_opts.zo_verbose >= 4)
5863 (void) printf("starting main threads...\n");
5866 * Kick off all the tests that run in parallel.
5868 for (t = 0; t < ztest_opts.zo_threads; t++) {
5871 if (t < ztest_opts.zo_datasets &&
5872 ztest_dataset_open(t) != 0)
5875 VERIFY3P(thread = zk_thread_create(NULL, 0,
5876 (thread_func_t)ztest_thread,
5877 (void *)(uintptr_t)t, TS_RUN, NULL, 0, 0,
5878 PTHREAD_CREATE_JOINABLE), !=, NULL);
5879 tid[t] = thread->t_tid;
5883 * Wait for all of the tests to complete. We go in reverse order
5884 * so we don't close datasets while threads are still using them.
5886 for (t = ztest_opts.zo_threads - 1; t >= 0; t--) {
5887 thread_join(tid[t]);
5888 if (t < ztest_opts.zo_datasets)
5889 ztest_dataset_close(t);
5892 txg_wait_synced(spa_get_dsl(spa), 0);
5894 zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(spa));
5895 zs->zs_space = metaslab_class_get_space(spa_normal_class(spa));
5897 if (ztest_opts.zo_verbose >= 3)
5898 zfs_dbgmsg_print(FTAG);
5900 umem_free(tid, ztest_opts.zo_threads * sizeof (kt_did_t));
5902 /* Kill the resume thread */
5903 ztest_exiting = B_TRUE;
5904 thread_join(resume_thread->t_tid);
5908 * Right before closing the pool, kick off a bunch of async I/O;
5909 * spa_close() should wait for it to complete.
5911 for (object = 1; object < 50; object++)
5912 dmu_prefetch(spa->spa_meta_objset, object, 0, 1ULL << 20);
5914 /* Verify that at least one commit cb was called in a timely fashion */
5915 if (zc_cb_counter >= ZTEST_COMMIT_CB_MIN_REG)
5916 VERIFY0(zc_min_txg_delay);
5918 spa_close(spa, FTAG);
5921 * Verify that we can loop over all pools.
5923 mutex_enter(&spa_namespace_lock);
5924 for (spa = spa_next(NULL); spa != NULL; spa = spa_next(spa))
5925 if (ztest_opts.zo_verbose > 3)
5926 (void) printf("spa_next: found %s\n", spa_name(spa));
5927 mutex_exit(&spa_namespace_lock);
5930 * Verify that we can export the pool and reimport it under a
5933 if (ztest_random(2) == 0) {
5934 char name[MAXNAMELEN];
5935 (void) snprintf(name, MAXNAMELEN, "%s_import",
5936 ztest_opts.zo_pool);
5937 ztest_spa_import_export(ztest_opts.zo_pool, name);
5938 ztest_spa_import_export(name, ztest_opts.zo_pool);
5943 list_destroy(&zcl.zcl_callbacks);
5944 mutex_destroy(&zcl.zcl_callbacks_lock);
5945 (void) rwlock_destroy(&ztest_name_lock);
5946 mutex_destroy(&ztest_vdev_lock);
5952 ztest_ds_t *zd = &ztest_ds[0];
5956 if (ztest_opts.zo_verbose >= 3)
5957 (void) printf("testing spa_freeze()...\n");
5959 kernel_init(FREAD | FWRITE);
5960 VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
5961 VERIFY3U(0, ==, ztest_dataset_open(0));
5962 spa->spa_debug = B_TRUE;
5966 * Force the first log block to be transactionally allocated.
5967 * We have to do this before we freeze the pool -- otherwise
5968 * the log chain won't be anchored.
5970 while (BP_IS_HOLE(&zd->zd_zilog->zl_header->zh_log)) {
5971 ztest_dmu_object_alloc_free(zd, 0);
5972 zil_commit(zd->zd_zilog, 0);
5975 txg_wait_synced(spa_get_dsl(spa), 0);
5978 * Freeze the pool. This stops spa_sync() from doing anything,
5979 * so that the only way to record changes from now on is the ZIL.
5984 * Because it is hard to predict how much space a write will actually
5985 * require beforehand, we leave ourselves some fudge space to write over
5988 uint64_t capacity = metaslab_class_get_space(spa_normal_class(spa)) / 2;
5991 * Run tests that generate log records but don't alter the pool config
5992 * or depend on DSL sync tasks (snapshots, objset create/destroy, etc).
5993 * We do a txg_wait_synced() after each iteration to force the txg
5994 * to increase well beyond the last synced value in the uberblock.
5995 * The ZIL should be OK with that.
5997 * Run a random number of times less than zo_maxloops and ensure we do
5998 * not run out of space on the pool.
6000 while (ztest_random(10) != 0 &&
6001 numloops++ < ztest_opts.zo_maxloops &&
6002 metaslab_class_get_alloc(spa_normal_class(spa)) < capacity) {
6004 ztest_od_init(&od, 0, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
6005 VERIFY0(ztest_object_init(zd, &od, sizeof (od), B_FALSE));
6006 ztest_io(zd, od.od_object,
6007 ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
6008 txg_wait_synced(spa_get_dsl(spa), 0);
6012 * Commit all of the changes we just generated.
6014 zil_commit(zd->zd_zilog, 0);
6015 txg_wait_synced(spa_get_dsl(spa), 0);
6018 * Close our dataset and close the pool.
6020 ztest_dataset_close(0);
6021 spa_close(spa, FTAG);
6025 * Open and close the pool and dataset to induce log replay.
6027 kernel_init(FREAD | FWRITE);
6028 VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
6029 ASSERT(spa_freeze_txg(spa) == UINT64_MAX);
6030 VERIFY3U(0, ==, ztest_dataset_open(0));
6031 ztest_dataset_close(0);
6033 spa->spa_debug = B_TRUE;
6035 txg_wait_synced(spa_get_dsl(spa), 0);
6036 ztest_reguid(NULL, 0);
6038 spa_close(spa, FTAG);
6043 print_time(hrtime_t t, char *timebuf)
6045 hrtime_t s = t / NANOSEC;
6046 hrtime_t m = s / 60;
6047 hrtime_t h = m / 60;
6048 hrtime_t d = h / 24;
6057 (void) sprintf(timebuf,
6058 "%llud%02lluh%02llum%02llus", d, h, m, s);
6060 (void) sprintf(timebuf, "%lluh%02llum%02llus", h, m, s);
6062 (void) sprintf(timebuf, "%llum%02llus", m, s);
6064 (void) sprintf(timebuf, "%llus", s);
6068 make_random_props(void)
6072 VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
6073 if (ztest_random(2) == 0)
6075 VERIFY(nvlist_add_uint64(props, "autoreplace", 1) == 0);
6081 * Create a storage pool with the given name and initial vdev size.
6082 * Then test spa_freeze() functionality.
6085 ztest_init(ztest_shared_t *zs)
6088 nvlist_t *nvroot, *props;
6091 mutex_init(&ztest_vdev_lock, NULL, MUTEX_DEFAULT, NULL);
6092 VERIFY(rwlock_init(&ztest_name_lock, USYNC_THREAD, NULL) == 0);
6094 kernel_init(FREAD | FWRITE);
6097 * Create the storage pool.
6099 (void) spa_destroy(ztest_opts.zo_pool);
6100 ztest_shared->zs_vdev_next_leaf = 0;
6102 zs->zs_mirrors = ztest_opts.zo_mirrors;
6103 nvroot = make_vdev_root(NULL, NULL, NULL, ztest_opts.zo_vdev_size, 0,
6104 0, ztest_opts.zo_raidz, zs->zs_mirrors, 1);
6105 props = make_random_props();
6106 for (i = 0; i < SPA_FEATURES; i++) {
6108 VERIFY3S(-1, !=, asprintf(&buf, "feature@%s",
6109 spa_feature_table[i].fi_uname));
6110 VERIFY3U(0, ==, nvlist_add_uint64(props, buf, 0));
6113 VERIFY3U(0, ==, spa_create(ztest_opts.zo_pool, nvroot, props, NULL));
6114 nvlist_free(nvroot);
6117 VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
6118 zs->zs_metaslab_sz =
6119 1ULL << spa->spa_root_vdev->vdev_child[0]->vdev_ms_shift;
6120 spa_close(spa, FTAG);
6124 ztest_run_zdb(ztest_opts.zo_pool);
6128 ztest_run_zdb(ztest_opts.zo_pool);
6130 (void) rwlock_destroy(&ztest_name_lock);
6131 mutex_destroy(&ztest_vdev_lock);
6137 static char ztest_name_data[] = "/tmp/ztest.data.XXXXXX";
6139 ztest_fd_data = mkstemp(ztest_name_data);
6140 ASSERT3S(ztest_fd_data, >=, 0);
6141 (void) unlink(ztest_name_data);
6145 shared_data_size(ztest_shared_hdr_t *hdr)
6149 size = hdr->zh_hdr_size;
6150 size += hdr->zh_opts_size;
6151 size += hdr->zh_size;
6152 size += hdr->zh_stats_size * hdr->zh_stats_count;
6153 size += hdr->zh_ds_size * hdr->zh_ds_count;
6162 ztest_shared_hdr_t *hdr;
6164 hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
6165 PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
6166 VERIFY3P(hdr, !=, MAP_FAILED);
6168 VERIFY3U(0, ==, ftruncate(ztest_fd_data, sizeof (ztest_shared_hdr_t)));
6170 hdr->zh_hdr_size = sizeof (ztest_shared_hdr_t);
6171 hdr->zh_opts_size = sizeof (ztest_shared_opts_t);
6172 hdr->zh_size = sizeof (ztest_shared_t);
6173 hdr->zh_stats_size = sizeof (ztest_shared_callstate_t);
6174 hdr->zh_stats_count = ZTEST_FUNCS;
6175 hdr->zh_ds_size = sizeof (ztest_shared_ds_t);
6176 hdr->zh_ds_count = ztest_opts.zo_datasets;
6178 size = shared_data_size(hdr);
6179 VERIFY3U(0, ==, ftruncate(ztest_fd_data, size));
6181 (void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
6188 ztest_shared_hdr_t *hdr;
6191 hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
6192 PROT_READ, MAP_SHARED, ztest_fd_data, 0);
6193 VERIFY3P(hdr, !=, MAP_FAILED);
6195 size = shared_data_size(hdr);
6197 (void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
6198 hdr = ztest_shared_hdr = (void *)mmap(0, P2ROUNDUP(size, getpagesize()),
6199 PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
6200 VERIFY3P(hdr, !=, MAP_FAILED);
6201 buf = (uint8_t *)hdr;
6203 offset = hdr->zh_hdr_size;
6204 ztest_shared_opts = (void *)&buf[offset];
6205 offset += hdr->zh_opts_size;
6206 ztest_shared = (void *)&buf[offset];
6207 offset += hdr->zh_size;
6208 ztest_shared_callstate = (void *)&buf[offset];
6209 offset += hdr->zh_stats_size * hdr->zh_stats_count;
6210 ztest_shared_ds = (void *)&buf[offset];
6214 exec_child(char *cmd, char *libpath, boolean_t ignorekill, int *statusp)
6218 char *cmdbuf = NULL;
6223 cmdbuf = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
6224 (void) strlcpy(cmdbuf, getexecname(), MAXPATHLEN);
6229 fatal(1, "fork failed");
6231 if (pid == 0) { /* child */
6232 char *emptyargv[2] = { cmd, NULL };
6233 char fd_data_str[12];
6235 struct rlimit rl = { 1024, 1024 };
6236 (void) setrlimit(RLIMIT_NOFILE, &rl);
6238 (void) close(ztest_fd_rand);
6239 VERIFY(11 >= snprintf(fd_data_str, 12, "%d", ztest_fd_data));
6240 VERIFY(0 == setenv("ZTEST_FD_DATA", fd_data_str, 1));
6242 (void) enable_extended_FILE_stdio(-1, -1);
6243 if (libpath != NULL)
6244 VERIFY(0 == setenv("LD_LIBRARY_PATH", libpath, 1));
6245 (void) execv(cmd, emptyargv);
6246 ztest_dump_core = B_FALSE;
6247 fatal(B_TRUE, "exec failed: %s", cmd);
6250 if (cmdbuf != NULL) {
6251 umem_free(cmdbuf, MAXPATHLEN);
6255 while (waitpid(pid, &status, 0) != pid)
6257 if (statusp != NULL)
6260 if (WIFEXITED(status)) {
6261 if (WEXITSTATUS(status) != 0) {
6262 (void) fprintf(stderr, "child exited with code %d\n",
6263 WEXITSTATUS(status));
6267 } else if (WIFSIGNALED(status)) {
6268 if (!ignorekill || WTERMSIG(status) != SIGKILL) {
6269 (void) fprintf(stderr, "child died with signal %d\n",
6275 (void) fprintf(stderr, "something strange happened to child\n");
6282 ztest_run_init(void)
6286 ztest_shared_t *zs = ztest_shared;
6288 ASSERT(ztest_opts.zo_init != 0);
6291 * Blow away any existing copy of zpool.cache
6293 (void) remove(spa_config_path);
6296 * Create and initialize our storage pool.
6298 for (i = 1; i <= ztest_opts.zo_init; i++) {
6299 bzero(zs, sizeof (ztest_shared_t));
6300 if (ztest_opts.zo_verbose >= 3 &&
6301 ztest_opts.zo_init != 1) {
6302 (void) printf("ztest_init(), pass %d\n", i);
6309 main(int argc, char **argv)
6317 ztest_shared_callstate_t *zc;
6324 char *fd_data_str = getenv("ZTEST_FD_DATA");
6325 struct sigaction action;
6327 (void) setvbuf(stdout, NULL, _IOLBF, 0);
6329 dprintf_setup(&argc, argv);
6331 action.sa_handler = sig_handler;
6332 sigemptyset(&action.sa_mask);
6333 action.sa_flags = 0;
6335 if (sigaction(SIGSEGV, &action, NULL) < 0) {
6336 (void) fprintf(stderr, "ztest: cannot catch SIGSEGV: %s.\n",
6341 if (sigaction(SIGABRT, &action, NULL) < 0) {
6342 (void) fprintf(stderr, "ztest: cannot catch SIGABRT: %s.\n",
6347 ztest_fd_rand = open("/dev/urandom", O_RDONLY);
6348 ASSERT3S(ztest_fd_rand, >=, 0);
6351 process_options(argc, argv);
6356 bcopy(&ztest_opts, ztest_shared_opts,
6357 sizeof (*ztest_shared_opts));
6359 ztest_fd_data = atoi(fd_data_str);
6361 bcopy(ztest_shared_opts, &ztest_opts, sizeof (ztest_opts));
6363 ASSERT3U(ztest_opts.zo_datasets, ==, ztest_shared_hdr->zh_ds_count);
6365 /* Override location of zpool.cache */
6366 VERIFY(asprintf((char **)&spa_config_path, "%s/zpool.cache",
6367 ztest_opts.zo_dir) != -1);
6369 ztest_ds = umem_alloc(ztest_opts.zo_datasets * sizeof (ztest_ds_t),
6374 metaslab_gang_bang = ztest_opts.zo_metaslab_gang_bang;
6375 metaslab_df_alloc_threshold =
6376 zs->zs_metaslab_df_alloc_threshold;
6385 hasalt = (strlen(ztest_opts.zo_alt_ztest) != 0);
6387 if (ztest_opts.zo_verbose >= 1) {
6388 (void) printf("%llu vdevs, %d datasets, %d threads,"
6389 " %llu seconds...\n",
6390 (u_longlong_t)ztest_opts.zo_vdevs,
6391 ztest_opts.zo_datasets,
6392 ztest_opts.zo_threads,
6393 (u_longlong_t)ztest_opts.zo_time);
6396 cmd = umem_alloc(MAXNAMELEN, UMEM_NOFAIL);
6397 (void) strlcpy(cmd, getexecname(), MAXNAMELEN);
6399 zs->zs_do_init = B_TRUE;
6400 if (strlen(ztest_opts.zo_alt_ztest) != 0) {
6401 if (ztest_opts.zo_verbose >= 1) {
6402 (void) printf("Executing older ztest for "
6403 "initialization: %s\n", ztest_opts.zo_alt_ztest);
6405 VERIFY(!exec_child(ztest_opts.zo_alt_ztest,
6406 ztest_opts.zo_alt_libpath, B_FALSE, NULL));
6408 VERIFY(!exec_child(NULL, NULL, B_FALSE, NULL));
6410 zs->zs_do_init = B_FALSE;
6412 zs->zs_proc_start = gethrtime();
6413 zs->zs_proc_stop = zs->zs_proc_start + ztest_opts.zo_time * NANOSEC;
6415 for (f = 0; f < ZTEST_FUNCS; f++) {
6416 zi = &ztest_info[f];
6417 zc = ZTEST_GET_SHARED_CALLSTATE(f);
6418 if (zs->zs_proc_start + zi->zi_interval[0] > zs->zs_proc_stop)
6419 zc->zc_next = UINT64_MAX;
6421 zc->zc_next = zs->zs_proc_start +
6422 ztest_random(2 * zi->zi_interval[0] + 1);
6426 * Run the tests in a loop. These tests include fault injection
6427 * to verify that self-healing data works, and forced crashes
6428 * to verify that we never lose on-disk consistency.
6430 while (gethrtime() < zs->zs_proc_stop) {
6435 * Initialize the workload counters for each function.
6437 for (f = 0; f < ZTEST_FUNCS; f++) {
6438 zc = ZTEST_GET_SHARED_CALLSTATE(f);
6443 /* Set the allocation switch size */
6444 zs->zs_metaslab_df_alloc_threshold =
6445 ztest_random(zs->zs_metaslab_sz / 4) + 1;
6447 if (!hasalt || ztest_random(2) == 0) {
6448 if (hasalt && ztest_opts.zo_verbose >= 1) {
6449 (void) printf("Executing newer ztest: %s\n",
6453 killed = exec_child(cmd, NULL, B_TRUE, &status);
6455 if (hasalt && ztest_opts.zo_verbose >= 1) {
6456 (void) printf("Executing older ztest: %s\n",
6457 ztest_opts.zo_alt_ztest);
6460 killed = exec_child(ztest_opts.zo_alt_ztest,
6461 ztest_opts.zo_alt_libpath, B_TRUE, &status);
6468 if (ztest_opts.zo_verbose >= 1) {
6469 hrtime_t now = gethrtime();
6471 now = MIN(now, zs->zs_proc_stop);
6472 print_time(zs->zs_proc_stop - now, timebuf);
6473 nicenum(zs->zs_space, numbuf);
6475 (void) printf("Pass %3d, %8s, %3llu ENOSPC, "
6476 "%4.1f%% of %5s used, %3.0f%% done, %8s to go\n",
6478 WIFEXITED(status) ? "Complete" : "SIGKILL",
6479 (u_longlong_t)zs->zs_enospc_count,
6480 100.0 * zs->zs_alloc / zs->zs_space,
6482 100.0 * (now - zs->zs_proc_start) /
6483 (ztest_opts.zo_time * NANOSEC), timebuf);
6486 if (ztest_opts.zo_verbose >= 2) {
6487 (void) printf("\nWorkload summary:\n\n");
6488 (void) printf("%7s %9s %s\n",
6489 "Calls", "Time", "Function");
6490 (void) printf("%7s %9s %s\n",
6491 "-----", "----", "--------");
6492 for (f = 0; f < ZTEST_FUNCS; f++) {
6493 zi = &ztest_info[f];
6494 zc = ZTEST_GET_SHARED_CALLSTATE(f);
6495 print_time(zc->zc_time, timebuf);
6496 (void) printf("%7llu %9s %s\n",
6497 (u_longlong_t)zc->zc_count, timebuf,
6500 (void) printf("\n");
6504 * It's possible that we killed a child during a rename test,
6505 * in which case we'll have a 'ztest_tmp' pool lying around
6506 * instead of 'ztest'. Do a blind rename in case this happened.
6509 if (spa_open(ztest_opts.zo_pool, &spa, FTAG) == 0) {
6510 spa_close(spa, FTAG);
6512 char tmpname[MAXNAMELEN];
6514 kernel_init(FREAD | FWRITE);
6515 (void) snprintf(tmpname, sizeof (tmpname), "%s_tmp",
6516 ztest_opts.zo_pool);
6517 (void) spa_rename(tmpname, ztest_opts.zo_pool);
6521 ztest_run_zdb(ztest_opts.zo_pool);
6524 if (ztest_opts.zo_verbose >= 1) {
6526 (void) printf("%d runs of older ztest: %s\n", older,
6527 ztest_opts.zo_alt_ztest);
6528 (void) printf("%d runs of newer ztest: %s\n", newer,
6531 (void) printf("%d killed, %d completed, %.0f%% kill rate\n",
6532 kills, iters - kills, (100.0 * kills) / MAX(1, iters));
6535 umem_free(cmd, MAXNAMELEN);