4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2011, 2014 by Delphix. All rights reserved.
24 * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
25 * Copyright (c) 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 * Find the largest ashift used
1029 ztest_spa_get_ashift(void) {
1031 uint64_t ashift = SPA_MINBLOCKSHIFT;
1032 vdev_t *rvd = ztest_spa->spa_root_vdev;
1034 for (i = 0; i < rvd->vdev_children; i++) {
1035 ashift = MAX(ashift, rvd->vdev_child[i]->vdev_ashift);
1041 ztest_random_blocksize(void)
1044 * Choose a block size >= the ashift.
1045 * If the SPA supports new MAXBLOCKSIZE, test up to 1MB blocks.
1047 int maxbs = SPA_OLD_MAXBLOCKSHIFT;
1048 if (spa_maxblocksize(ztest_spa) == SPA_MAXBLOCKSIZE)
1050 uint64_t block_shift = ztest_random(maxbs - ztest_spa_get_ashift() + 1);
1051 return (1 << (SPA_MINBLOCKSHIFT + block_shift));
1055 ztest_random_ibshift(void)
1057 return (DN_MIN_INDBLKSHIFT +
1058 ztest_random(DN_MAX_INDBLKSHIFT - DN_MIN_INDBLKSHIFT + 1));
1062 ztest_random_vdev_top(spa_t *spa, boolean_t log_ok)
1065 vdev_t *rvd = spa->spa_root_vdev;
1068 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1071 top = ztest_random(rvd->vdev_children);
1072 tvd = rvd->vdev_child[top];
1073 } while (tvd->vdev_ishole || (tvd->vdev_islog && !log_ok) ||
1074 tvd->vdev_mg == NULL || tvd->vdev_mg->mg_class == NULL);
1080 ztest_random_dsl_prop(zfs_prop_t prop)
1085 value = zfs_prop_random_value(prop, ztest_random(-1ULL));
1086 } while (prop == ZFS_PROP_CHECKSUM && value == ZIO_CHECKSUM_OFF);
1092 ztest_dsl_prop_set_uint64(char *osname, zfs_prop_t prop, uint64_t value,
1095 const char *propname = zfs_prop_to_name(prop);
1096 const char *valname;
1101 error = dsl_prop_set_int(osname, propname,
1102 (inherit ? ZPROP_SRC_NONE : ZPROP_SRC_LOCAL), value);
1104 if (error == ENOSPC) {
1105 ztest_record_enospc(FTAG);
1110 setpoint = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
1111 VERIFY0(dsl_prop_get_integer(osname, propname, &curval, setpoint));
1113 if (ztest_opts.zo_verbose >= 6) {
1114 VERIFY(zfs_prop_index_to_string(prop, curval, &valname) == 0);
1115 (void) printf("%s %s = %s at '%s'\n",
1116 osname, propname, valname, setpoint);
1118 umem_free(setpoint, MAXPATHLEN);
1124 ztest_spa_prop_set_uint64(zpool_prop_t prop, uint64_t value)
1126 spa_t *spa = ztest_spa;
1127 nvlist_t *props = NULL;
1130 VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
1131 VERIFY(nvlist_add_uint64(props, zpool_prop_to_name(prop), value) == 0);
1133 error = spa_prop_set(spa, props);
1137 if (error == ENOSPC) {
1138 ztest_record_enospc(FTAG);
1147 ztest_rll_init(rll_t *rll)
1149 rll->rll_writer = NULL;
1150 rll->rll_readers = 0;
1151 mutex_init(&rll->rll_lock, NULL, MUTEX_DEFAULT, NULL);
1152 cv_init(&rll->rll_cv, NULL, CV_DEFAULT, NULL);
1156 ztest_rll_destroy(rll_t *rll)
1158 ASSERT(rll->rll_writer == NULL);
1159 ASSERT(rll->rll_readers == 0);
1160 mutex_destroy(&rll->rll_lock);
1161 cv_destroy(&rll->rll_cv);
1165 ztest_rll_lock(rll_t *rll, rl_type_t type)
1167 mutex_enter(&rll->rll_lock);
1169 if (type == RL_READER) {
1170 while (rll->rll_writer != NULL)
1171 (void) cv_wait(&rll->rll_cv, &rll->rll_lock);
1174 while (rll->rll_writer != NULL || rll->rll_readers)
1175 (void) cv_wait(&rll->rll_cv, &rll->rll_lock);
1176 rll->rll_writer = curthread;
1179 mutex_exit(&rll->rll_lock);
1183 ztest_rll_unlock(rll_t *rll)
1185 mutex_enter(&rll->rll_lock);
1187 if (rll->rll_writer) {
1188 ASSERT(rll->rll_readers == 0);
1189 rll->rll_writer = NULL;
1191 ASSERT(rll->rll_readers != 0);
1192 ASSERT(rll->rll_writer == NULL);
1196 if (rll->rll_writer == NULL && rll->rll_readers == 0)
1197 cv_broadcast(&rll->rll_cv);
1199 mutex_exit(&rll->rll_lock);
1203 ztest_object_lock(ztest_ds_t *zd, uint64_t object, rl_type_t type)
1205 rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1207 ztest_rll_lock(rll, type);
1211 ztest_object_unlock(ztest_ds_t *zd, uint64_t object)
1213 rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1215 ztest_rll_unlock(rll);
1219 ztest_range_lock(ztest_ds_t *zd, uint64_t object, uint64_t offset,
1220 uint64_t size, rl_type_t type)
1222 uint64_t hash = object ^ (offset % (ZTEST_RANGE_LOCKS + 1));
1223 rll_t *rll = &zd->zd_range_lock[hash & (ZTEST_RANGE_LOCKS - 1)];
1226 rl = umem_alloc(sizeof (*rl), UMEM_NOFAIL);
1227 rl->rl_object = object;
1228 rl->rl_offset = offset;
1232 ztest_rll_lock(rll, type);
1238 ztest_range_unlock(rl_t *rl)
1240 rll_t *rll = rl->rl_lock;
1242 ztest_rll_unlock(rll);
1244 umem_free(rl, sizeof (*rl));
1248 ztest_zd_init(ztest_ds_t *zd, ztest_shared_ds_t *szd, objset_t *os)
1251 zd->zd_zilog = dmu_objset_zil(os);
1252 zd->zd_shared = szd;
1253 dmu_objset_name(os, zd->zd_name);
1256 if (zd->zd_shared != NULL)
1257 zd->zd_shared->zd_seq = 0;
1259 VERIFY(rwlock_init(&zd->zd_zilog_lock, USYNC_THREAD, NULL) == 0);
1260 mutex_init(&zd->zd_dirobj_lock, NULL, MUTEX_DEFAULT, NULL);
1262 for (l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1263 ztest_rll_init(&zd->zd_object_lock[l]);
1265 for (l = 0; l < ZTEST_RANGE_LOCKS; l++)
1266 ztest_rll_init(&zd->zd_range_lock[l]);
1270 ztest_zd_fini(ztest_ds_t *zd)
1274 mutex_destroy(&zd->zd_dirobj_lock);
1275 (void) rwlock_destroy(&zd->zd_zilog_lock);
1277 for (l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1278 ztest_rll_destroy(&zd->zd_object_lock[l]);
1280 for (l = 0; l < ZTEST_RANGE_LOCKS; l++)
1281 ztest_rll_destroy(&zd->zd_range_lock[l]);
1284 #define TXG_MIGHTWAIT (ztest_random(10) == 0 ? TXG_NOWAIT : TXG_WAIT)
1287 ztest_tx_assign(dmu_tx_t *tx, uint64_t txg_how, const char *tag)
1293 * Attempt to assign tx to some transaction group.
1295 error = dmu_tx_assign(tx, txg_how);
1297 if (error == ERESTART) {
1298 ASSERT(txg_how == TXG_NOWAIT);
1301 ASSERT3U(error, ==, ENOSPC);
1302 ztest_record_enospc(tag);
1307 txg = dmu_tx_get_txg(tx);
1313 ztest_pattern_set(void *buf, uint64_t size, uint64_t value)
1316 uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1324 ztest_pattern_match(void *buf, uint64_t size, uint64_t value)
1327 uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1331 diff |= (value - *ip++);
1338 ztest_bt_generate(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1339 uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg)
1341 bt->bt_magic = BT_MAGIC;
1342 bt->bt_objset = dmu_objset_id(os);
1343 bt->bt_object = object;
1344 bt->bt_offset = offset;
1347 bt->bt_crtxg = crtxg;
1351 ztest_bt_verify(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1352 uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg)
1354 ASSERT3U(bt->bt_magic, ==, BT_MAGIC);
1355 ASSERT3U(bt->bt_objset, ==, dmu_objset_id(os));
1356 ASSERT3U(bt->bt_object, ==, object);
1357 ASSERT3U(bt->bt_offset, ==, offset);
1358 ASSERT3U(bt->bt_gen, <=, gen);
1359 ASSERT3U(bt->bt_txg, <=, txg);
1360 ASSERT3U(bt->bt_crtxg, ==, crtxg);
1363 static ztest_block_tag_t *
1364 ztest_bt_bonus(dmu_buf_t *db)
1366 dmu_object_info_t doi;
1367 ztest_block_tag_t *bt;
1369 dmu_object_info_from_db(db, &doi);
1370 ASSERT3U(doi.doi_bonus_size, <=, db->db_size);
1371 ASSERT3U(doi.doi_bonus_size, >=, sizeof (*bt));
1372 bt = (void *)((char *)db->db_data + doi.doi_bonus_size - sizeof (*bt));
1381 #define lrz_type lr_mode
1382 #define lrz_blocksize lr_uid
1383 #define lrz_ibshift lr_gid
1384 #define lrz_bonustype lr_rdev
1385 #define lrz_bonuslen lr_crtime[1]
1388 ztest_log_create(ztest_ds_t *zd, dmu_tx_t *tx, lr_create_t *lr)
1390 char *name = (void *)(lr + 1); /* name follows lr */
1391 size_t namesize = strlen(name) + 1;
1394 if (zil_replaying(zd->zd_zilog, tx))
1397 itx = zil_itx_create(TX_CREATE, sizeof (*lr) + namesize);
1398 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1399 sizeof (*lr) + namesize - sizeof (lr_t));
1401 zil_itx_assign(zd->zd_zilog, itx, tx);
1405 ztest_log_remove(ztest_ds_t *zd, dmu_tx_t *tx, lr_remove_t *lr, uint64_t object)
1407 char *name = (void *)(lr + 1); /* name follows lr */
1408 size_t namesize = strlen(name) + 1;
1411 if (zil_replaying(zd->zd_zilog, tx))
1414 itx = zil_itx_create(TX_REMOVE, sizeof (*lr) + namesize);
1415 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1416 sizeof (*lr) + namesize - sizeof (lr_t));
1418 itx->itx_oid = object;
1419 zil_itx_assign(zd->zd_zilog, itx, tx);
1423 ztest_log_write(ztest_ds_t *zd, dmu_tx_t *tx, lr_write_t *lr)
1426 itx_wr_state_t write_state = ztest_random(WR_NUM_STATES);
1428 if (zil_replaying(zd->zd_zilog, tx))
1431 if (lr->lr_length > ZIL_MAX_LOG_DATA)
1432 write_state = WR_INDIRECT;
1434 itx = zil_itx_create(TX_WRITE,
1435 sizeof (*lr) + (write_state == WR_COPIED ? lr->lr_length : 0));
1437 if (write_state == WR_COPIED &&
1438 dmu_read(zd->zd_os, lr->lr_foid, lr->lr_offset, lr->lr_length,
1439 ((lr_write_t *)&itx->itx_lr) + 1, DMU_READ_NO_PREFETCH) != 0) {
1440 zil_itx_destroy(itx);
1441 itx = zil_itx_create(TX_WRITE, sizeof (*lr));
1442 write_state = WR_NEED_COPY;
1444 itx->itx_private = zd;
1445 itx->itx_wr_state = write_state;
1446 itx->itx_sync = (ztest_random(8) == 0);
1447 itx->itx_sod += (write_state == WR_NEED_COPY ? lr->lr_length : 0);
1449 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1450 sizeof (*lr) - sizeof (lr_t));
1452 zil_itx_assign(zd->zd_zilog, itx, tx);
1456 ztest_log_truncate(ztest_ds_t *zd, dmu_tx_t *tx, lr_truncate_t *lr)
1460 if (zil_replaying(zd->zd_zilog, tx))
1463 itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr));
1464 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1465 sizeof (*lr) - sizeof (lr_t));
1467 itx->itx_sync = B_FALSE;
1468 zil_itx_assign(zd->zd_zilog, itx, tx);
1472 ztest_log_setattr(ztest_ds_t *zd, dmu_tx_t *tx, lr_setattr_t *lr)
1476 if (zil_replaying(zd->zd_zilog, tx))
1479 itx = zil_itx_create(TX_SETATTR, sizeof (*lr));
1480 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1481 sizeof (*lr) - sizeof (lr_t));
1483 itx->itx_sync = B_FALSE;
1484 zil_itx_assign(zd->zd_zilog, itx, tx);
1491 ztest_replay_create(ztest_ds_t *zd, lr_create_t *lr, boolean_t byteswap)
1493 char *name = (void *)(lr + 1); /* name follows lr */
1494 objset_t *os = zd->zd_os;
1495 ztest_block_tag_t *bbt;
1502 byteswap_uint64_array(lr, sizeof (*lr));
1504 ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1505 ASSERT(name[0] != '\0');
1507 tx = dmu_tx_create(os);
1509 dmu_tx_hold_zap(tx, lr->lr_doid, B_TRUE, name);
1511 if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1512 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, B_TRUE, NULL);
1514 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1517 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1521 ASSERT(dmu_objset_zil(os)->zl_replay == !!lr->lr_foid);
1523 if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1524 if (lr->lr_foid == 0) {
1525 lr->lr_foid = zap_create(os,
1526 lr->lrz_type, lr->lrz_bonustype,
1527 lr->lrz_bonuslen, tx);
1529 error = zap_create_claim(os, lr->lr_foid,
1530 lr->lrz_type, lr->lrz_bonustype,
1531 lr->lrz_bonuslen, tx);
1534 if (lr->lr_foid == 0) {
1535 lr->lr_foid = dmu_object_alloc(os,
1536 lr->lrz_type, 0, lr->lrz_bonustype,
1537 lr->lrz_bonuslen, tx);
1539 error = dmu_object_claim(os, lr->lr_foid,
1540 lr->lrz_type, 0, lr->lrz_bonustype,
1541 lr->lrz_bonuslen, tx);
1546 ASSERT3U(error, ==, EEXIST);
1547 ASSERT(zd->zd_zilog->zl_replay);
1552 ASSERT(lr->lr_foid != 0);
1554 if (lr->lrz_type != DMU_OT_ZAP_OTHER)
1555 VERIFY3U(0, ==, dmu_object_set_blocksize(os, lr->lr_foid,
1556 lr->lrz_blocksize, lr->lrz_ibshift, tx));
1558 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1559 bbt = ztest_bt_bonus(db);
1560 dmu_buf_will_dirty(db, tx);
1561 ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_gen, txg, txg);
1562 dmu_buf_rele(db, FTAG);
1564 VERIFY3U(0, ==, zap_add(os, lr->lr_doid, name, sizeof (uint64_t), 1,
1567 (void) ztest_log_create(zd, tx, lr);
1575 ztest_replay_remove(ztest_ds_t *zd, lr_remove_t *lr, boolean_t byteswap)
1577 char *name = (void *)(lr + 1); /* name follows lr */
1578 objset_t *os = zd->zd_os;
1579 dmu_object_info_t doi;
1581 uint64_t object, txg;
1584 byteswap_uint64_array(lr, sizeof (*lr));
1586 ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1587 ASSERT(name[0] != '\0');
1590 zap_lookup(os, lr->lr_doid, name, sizeof (object), 1, &object));
1591 ASSERT(object != 0);
1593 ztest_object_lock(zd, object, RL_WRITER);
1595 VERIFY3U(0, ==, dmu_object_info(os, object, &doi));
1597 tx = dmu_tx_create(os);
1599 dmu_tx_hold_zap(tx, lr->lr_doid, B_FALSE, name);
1600 dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END);
1602 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1604 ztest_object_unlock(zd, object);
1608 if (doi.doi_type == DMU_OT_ZAP_OTHER) {
1609 VERIFY3U(0, ==, zap_destroy(os, object, tx));
1611 VERIFY3U(0, ==, dmu_object_free(os, object, tx));
1614 VERIFY3U(0, ==, zap_remove(os, lr->lr_doid, name, tx));
1616 (void) ztest_log_remove(zd, tx, lr, object);
1620 ztest_object_unlock(zd, object);
1626 ztest_replay_write(ztest_ds_t *zd, lr_write_t *lr, boolean_t byteswap)
1628 objset_t *os = zd->zd_os;
1629 void *data = lr + 1; /* data follows lr */
1630 uint64_t offset, length;
1631 ztest_block_tag_t *bt = data;
1632 ztest_block_tag_t *bbt;
1633 uint64_t gen, txg, lrtxg, crtxg;
1634 dmu_object_info_t doi;
1637 arc_buf_t *abuf = NULL;
1641 byteswap_uint64_array(lr, sizeof (*lr));
1643 offset = lr->lr_offset;
1644 length = lr->lr_length;
1646 /* If it's a dmu_sync() block, write the whole block */
1647 if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
1648 uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
1649 if (length < blocksize) {
1650 offset -= offset % blocksize;
1655 if (bt->bt_magic == BSWAP_64(BT_MAGIC))
1656 byteswap_uint64_array(bt, sizeof (*bt));
1658 if (bt->bt_magic != BT_MAGIC)
1661 ztest_object_lock(zd, lr->lr_foid, RL_READER);
1662 rl = ztest_range_lock(zd, lr->lr_foid, offset, length, RL_WRITER);
1664 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1666 dmu_object_info_from_db(db, &doi);
1668 bbt = ztest_bt_bonus(db);
1669 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1671 crtxg = bbt->bt_crtxg;
1672 lrtxg = lr->lr_common.lrc_txg;
1674 tx = dmu_tx_create(os);
1676 dmu_tx_hold_write(tx, lr->lr_foid, offset, length);
1678 if (ztest_random(8) == 0 && length == doi.doi_data_block_size &&
1679 P2PHASE(offset, length) == 0)
1680 abuf = dmu_request_arcbuf(db, length);
1682 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1685 dmu_return_arcbuf(abuf);
1686 dmu_buf_rele(db, FTAG);
1687 ztest_range_unlock(rl);
1688 ztest_object_unlock(zd, lr->lr_foid);
1694 * Usually, verify the old data before writing new data --
1695 * but not always, because we also want to verify correct
1696 * behavior when the data was not recently read into cache.
1698 ASSERT(offset % doi.doi_data_block_size == 0);
1699 if (ztest_random(4) != 0) {
1700 int prefetch = ztest_random(2) ?
1701 DMU_READ_PREFETCH : DMU_READ_NO_PREFETCH;
1702 ztest_block_tag_t rbt;
1704 VERIFY(dmu_read(os, lr->lr_foid, offset,
1705 sizeof (rbt), &rbt, prefetch) == 0);
1706 if (rbt.bt_magic == BT_MAGIC) {
1707 ztest_bt_verify(&rbt, os, lr->lr_foid,
1708 offset, gen, txg, crtxg);
1713 * Writes can appear to be newer than the bonus buffer because
1714 * the ztest_get_data() callback does a dmu_read() of the
1715 * open-context data, which may be different than the data
1716 * as it was when the write was generated.
1718 if (zd->zd_zilog->zl_replay) {
1719 ztest_bt_verify(bt, os, lr->lr_foid, offset,
1720 MAX(gen, bt->bt_gen), MAX(txg, lrtxg),
1725 * Set the bt's gen/txg to the bonus buffer's gen/txg
1726 * so that all of the usual ASSERTs will work.
1728 ztest_bt_generate(bt, os, lr->lr_foid, offset, gen, txg, crtxg);
1732 dmu_write(os, lr->lr_foid, offset, length, data, tx);
1734 bcopy(data, abuf->b_data, length);
1735 dmu_assign_arcbuf(db, offset, abuf, tx);
1738 (void) ztest_log_write(zd, tx, lr);
1740 dmu_buf_rele(db, FTAG);
1744 ztest_range_unlock(rl);
1745 ztest_object_unlock(zd, lr->lr_foid);
1751 ztest_replay_truncate(ztest_ds_t *zd, lr_truncate_t *lr, boolean_t byteswap)
1753 objset_t *os = zd->zd_os;
1759 byteswap_uint64_array(lr, sizeof (*lr));
1761 ztest_object_lock(zd, lr->lr_foid, RL_READER);
1762 rl = ztest_range_lock(zd, lr->lr_foid, lr->lr_offset, lr->lr_length,
1765 tx = dmu_tx_create(os);
1767 dmu_tx_hold_free(tx, lr->lr_foid, lr->lr_offset, lr->lr_length);
1769 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1771 ztest_range_unlock(rl);
1772 ztest_object_unlock(zd, lr->lr_foid);
1776 VERIFY(dmu_free_range(os, lr->lr_foid, lr->lr_offset,
1777 lr->lr_length, tx) == 0);
1779 (void) ztest_log_truncate(zd, tx, lr);
1783 ztest_range_unlock(rl);
1784 ztest_object_unlock(zd, lr->lr_foid);
1790 ztest_replay_setattr(ztest_ds_t *zd, lr_setattr_t *lr, boolean_t byteswap)
1792 objset_t *os = zd->zd_os;
1795 ztest_block_tag_t *bbt;
1796 uint64_t txg, lrtxg, crtxg;
1799 byteswap_uint64_array(lr, sizeof (*lr));
1801 ztest_object_lock(zd, lr->lr_foid, RL_WRITER);
1803 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1805 tx = dmu_tx_create(os);
1806 dmu_tx_hold_bonus(tx, lr->lr_foid);
1808 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1810 dmu_buf_rele(db, FTAG);
1811 ztest_object_unlock(zd, lr->lr_foid);
1815 bbt = ztest_bt_bonus(db);
1816 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1817 crtxg = bbt->bt_crtxg;
1818 lrtxg = lr->lr_common.lrc_txg;
1820 if (zd->zd_zilog->zl_replay) {
1821 ASSERT(lr->lr_size != 0);
1822 ASSERT(lr->lr_mode != 0);
1826 * Randomly change the size and increment the generation.
1828 lr->lr_size = (ztest_random(db->db_size / sizeof (*bbt)) + 1) *
1830 lr->lr_mode = bbt->bt_gen + 1;
1835 * Verify that the current bonus buffer is not newer than our txg.
1837 ztest_bt_verify(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode,
1838 MAX(txg, lrtxg), crtxg);
1840 dmu_buf_will_dirty(db, tx);
1842 ASSERT3U(lr->lr_size, >=, sizeof (*bbt));
1843 ASSERT3U(lr->lr_size, <=, db->db_size);
1844 VERIFY0(dmu_set_bonus(db, lr->lr_size, tx));
1845 bbt = ztest_bt_bonus(db);
1847 ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode, txg, crtxg);
1849 dmu_buf_rele(db, FTAG);
1851 (void) ztest_log_setattr(zd, tx, lr);
1855 ztest_object_unlock(zd, lr->lr_foid);
1860 zil_replay_func_t ztest_replay_vector[TX_MAX_TYPE] = {
1861 NULL, /* 0 no such transaction type */
1862 (zil_replay_func_t)ztest_replay_create, /* TX_CREATE */
1863 NULL, /* TX_MKDIR */
1864 NULL, /* TX_MKXATTR */
1865 NULL, /* TX_SYMLINK */
1866 (zil_replay_func_t)ztest_replay_remove, /* TX_REMOVE */
1867 NULL, /* TX_RMDIR */
1869 NULL, /* TX_RENAME */
1870 (zil_replay_func_t)ztest_replay_write, /* TX_WRITE */
1871 (zil_replay_func_t)ztest_replay_truncate, /* TX_TRUNCATE */
1872 (zil_replay_func_t)ztest_replay_setattr, /* TX_SETATTR */
1874 NULL, /* TX_CREATE_ACL */
1875 NULL, /* TX_CREATE_ATTR */
1876 NULL, /* TX_CREATE_ACL_ATTR */
1877 NULL, /* TX_MKDIR_ACL */
1878 NULL, /* TX_MKDIR_ATTR */
1879 NULL, /* TX_MKDIR_ACL_ATTR */
1880 NULL, /* TX_WRITE2 */
1884 * ZIL get_data callbacks
1888 ztest_get_done(zgd_t *zgd, int error)
1890 ztest_ds_t *zd = zgd->zgd_private;
1891 uint64_t object = zgd->zgd_rl->rl_object;
1894 dmu_buf_rele(zgd->zgd_db, zgd);
1896 ztest_range_unlock(zgd->zgd_rl);
1897 ztest_object_unlock(zd, object);
1899 if (error == 0 && zgd->zgd_bp)
1900 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
1902 umem_free(zgd, sizeof (*zgd));
1906 ztest_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
1908 ztest_ds_t *zd = arg;
1909 objset_t *os = zd->zd_os;
1910 uint64_t object = lr->lr_foid;
1911 uint64_t offset = lr->lr_offset;
1912 uint64_t size = lr->lr_length;
1913 blkptr_t *bp = &lr->lr_blkptr;
1914 uint64_t txg = lr->lr_common.lrc_txg;
1916 dmu_object_info_t doi;
1921 ztest_object_lock(zd, object, RL_READER);
1922 error = dmu_bonus_hold(os, object, FTAG, &db);
1924 ztest_object_unlock(zd, object);
1928 crtxg = ztest_bt_bonus(db)->bt_crtxg;
1930 if (crtxg == 0 || crtxg > txg) {
1931 dmu_buf_rele(db, FTAG);
1932 ztest_object_unlock(zd, object);
1936 dmu_object_info_from_db(db, &doi);
1937 dmu_buf_rele(db, FTAG);
1940 zgd = umem_zalloc(sizeof (*zgd), UMEM_NOFAIL);
1941 zgd->zgd_zilog = zd->zd_zilog;
1942 zgd->zgd_private = zd;
1944 if (buf != NULL) { /* immediate write */
1945 zgd->zgd_rl = ztest_range_lock(zd, object, offset, size,
1948 error = dmu_read(os, object, offset, size, buf,
1949 DMU_READ_NO_PREFETCH);
1952 size = doi.doi_data_block_size;
1954 offset = P2ALIGN(offset, size);
1956 ASSERT(offset < size);
1960 zgd->zgd_rl = ztest_range_lock(zd, object, offset, size,
1963 error = dmu_buf_hold(os, object, offset, zgd, &db,
1964 DMU_READ_NO_PREFETCH);
1967 blkptr_t *obp = dmu_buf_get_blkptr(db);
1969 ASSERT(BP_IS_HOLE(bp));
1976 ASSERT(db->db_offset == offset);
1977 ASSERT(db->db_size == size);
1979 error = dmu_sync(zio, lr->lr_common.lrc_txg,
1980 ztest_get_done, zgd);
1987 ztest_get_done(zgd, error);
1993 ztest_lr_alloc(size_t lrsize, char *name)
1996 size_t namesize = name ? strlen(name) + 1 : 0;
1998 lr = umem_zalloc(lrsize + namesize, UMEM_NOFAIL);
2001 bcopy(name, lr + lrsize, namesize);
2007 ztest_lr_free(void *lr, size_t lrsize, char *name)
2009 size_t namesize = name ? strlen(name) + 1 : 0;
2011 umem_free(lr, lrsize + namesize);
2015 * Lookup a bunch of objects. Returns the number of objects not found.
2018 ztest_lookup(ztest_ds_t *zd, ztest_od_t *od, int count)
2024 ASSERT(mutex_held(&zd->zd_dirobj_lock));
2026 for (i = 0; i < count; i++, od++) {
2028 error = zap_lookup(zd->zd_os, od->od_dir, od->od_name,
2029 sizeof (uint64_t), 1, &od->od_object);
2031 ASSERT(error == ENOENT);
2032 ASSERT(od->od_object == 0);
2036 ztest_block_tag_t *bbt;
2037 dmu_object_info_t doi;
2039 ASSERT(od->od_object != 0);
2040 ASSERT(missing == 0); /* there should be no gaps */
2042 ztest_object_lock(zd, od->od_object, RL_READER);
2043 VERIFY3U(0, ==, dmu_bonus_hold(zd->zd_os,
2044 od->od_object, FTAG, &db));
2045 dmu_object_info_from_db(db, &doi);
2046 bbt = ztest_bt_bonus(db);
2047 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
2048 od->od_type = doi.doi_type;
2049 od->od_blocksize = doi.doi_data_block_size;
2050 od->od_gen = bbt->bt_gen;
2051 dmu_buf_rele(db, FTAG);
2052 ztest_object_unlock(zd, od->od_object);
2060 ztest_create(ztest_ds_t *zd, ztest_od_t *od, int count)
2065 ASSERT(mutex_held(&zd->zd_dirobj_lock));
2067 for (i = 0; i < count; i++, od++) {
2074 lr_create_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
2076 lr->lr_doid = od->od_dir;
2077 lr->lr_foid = 0; /* 0 to allocate, > 0 to claim */
2078 lr->lrz_type = od->od_crtype;
2079 lr->lrz_blocksize = od->od_crblocksize;
2080 lr->lrz_ibshift = ztest_random_ibshift();
2081 lr->lrz_bonustype = DMU_OT_UINT64_OTHER;
2082 lr->lrz_bonuslen = dmu_bonus_max();
2083 lr->lr_gen = od->od_crgen;
2084 lr->lr_crtime[0] = time(NULL);
2086 if (ztest_replay_create(zd, lr, B_FALSE) != 0) {
2087 ASSERT(missing == 0);
2091 od->od_object = lr->lr_foid;
2092 od->od_type = od->od_crtype;
2093 od->od_blocksize = od->od_crblocksize;
2094 od->od_gen = od->od_crgen;
2095 ASSERT(od->od_object != 0);
2098 ztest_lr_free(lr, sizeof (*lr), od->od_name);
2105 ztest_remove(ztest_ds_t *zd, ztest_od_t *od, int count)
2111 ASSERT(mutex_held(&zd->zd_dirobj_lock));
2115 for (i = count - 1; i >= 0; i--, od--) {
2122 * No object was found.
2124 if (od->od_object == 0)
2127 lr_remove_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
2129 lr->lr_doid = od->od_dir;
2131 if ((error = ztest_replay_remove(zd, lr, B_FALSE)) != 0) {
2132 ASSERT3U(error, ==, ENOSPC);
2137 ztest_lr_free(lr, sizeof (*lr), od->od_name);
2144 ztest_write(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size,
2150 lr = ztest_lr_alloc(sizeof (*lr) + size, NULL);
2152 lr->lr_foid = object;
2153 lr->lr_offset = offset;
2154 lr->lr_length = size;
2156 BP_ZERO(&lr->lr_blkptr);
2158 bcopy(data, lr + 1, size);
2160 error = ztest_replay_write(zd, lr, B_FALSE);
2162 ztest_lr_free(lr, sizeof (*lr) + size, NULL);
2168 ztest_truncate(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2173 lr = ztest_lr_alloc(sizeof (*lr), NULL);
2175 lr->lr_foid = object;
2176 lr->lr_offset = offset;
2177 lr->lr_length = size;
2179 error = ztest_replay_truncate(zd, lr, B_FALSE);
2181 ztest_lr_free(lr, sizeof (*lr), NULL);
2187 ztest_setattr(ztest_ds_t *zd, uint64_t object)
2192 lr = ztest_lr_alloc(sizeof (*lr), NULL);
2194 lr->lr_foid = object;
2198 error = ztest_replay_setattr(zd, lr, B_FALSE);
2200 ztest_lr_free(lr, sizeof (*lr), NULL);
2206 ztest_prealloc(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2208 objset_t *os = zd->zd_os;
2213 txg_wait_synced(dmu_objset_pool(os), 0);
2215 ztest_object_lock(zd, object, RL_READER);
2216 rl = ztest_range_lock(zd, object, offset, size, RL_WRITER);
2218 tx = dmu_tx_create(os);
2220 dmu_tx_hold_write(tx, object, offset, size);
2222 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
2225 dmu_prealloc(os, object, offset, size, tx);
2227 txg_wait_synced(dmu_objset_pool(os), txg);
2229 (void) dmu_free_long_range(os, object, offset, size);
2232 ztest_range_unlock(rl);
2233 ztest_object_unlock(zd, object);
2237 ztest_io(ztest_ds_t *zd, uint64_t object, uint64_t offset)
2240 ztest_block_tag_t wbt;
2241 dmu_object_info_t doi;
2242 enum ztest_io_type io_type;
2246 VERIFY(dmu_object_info(zd->zd_os, object, &doi) == 0);
2247 blocksize = doi.doi_data_block_size;
2248 data = umem_alloc(blocksize, UMEM_NOFAIL);
2251 * Pick an i/o type at random, biased toward writing block tags.
2253 io_type = ztest_random(ZTEST_IO_TYPES);
2254 if (ztest_random(2) == 0)
2255 io_type = ZTEST_IO_WRITE_TAG;
2257 (void) rw_rdlock(&zd->zd_zilog_lock);
2261 case ZTEST_IO_WRITE_TAG:
2262 ztest_bt_generate(&wbt, zd->zd_os, object, offset, 0, 0, 0);
2263 (void) ztest_write(zd, object, offset, sizeof (wbt), &wbt);
2266 case ZTEST_IO_WRITE_PATTERN:
2267 (void) memset(data, 'a' + (object + offset) % 5, blocksize);
2268 if (ztest_random(2) == 0) {
2270 * Induce fletcher2 collisions to ensure that
2271 * zio_ddt_collision() detects and resolves them
2272 * when using fletcher2-verify for deduplication.
2274 ((uint64_t *)data)[0] ^= 1ULL << 63;
2275 ((uint64_t *)data)[4] ^= 1ULL << 63;
2277 (void) ztest_write(zd, object, offset, blocksize, data);
2280 case ZTEST_IO_WRITE_ZEROES:
2281 bzero(data, blocksize);
2282 (void) ztest_write(zd, object, offset, blocksize, data);
2285 case ZTEST_IO_TRUNCATE:
2286 (void) ztest_truncate(zd, object, offset, blocksize);
2289 case ZTEST_IO_SETATTR:
2290 (void) ztest_setattr(zd, object);
2295 case ZTEST_IO_REWRITE:
2296 (void) rw_rdlock(&ztest_name_lock);
2297 err = ztest_dsl_prop_set_uint64(zd->zd_name,
2298 ZFS_PROP_CHECKSUM, spa_dedup_checksum(ztest_spa),
2300 VERIFY(err == 0 || err == ENOSPC);
2301 err = ztest_dsl_prop_set_uint64(zd->zd_name,
2302 ZFS_PROP_COMPRESSION,
2303 ztest_random_dsl_prop(ZFS_PROP_COMPRESSION),
2305 VERIFY(err == 0 || err == ENOSPC);
2306 (void) rw_unlock(&ztest_name_lock);
2308 VERIFY0(dmu_read(zd->zd_os, object, offset, blocksize, data,
2309 DMU_READ_NO_PREFETCH));
2311 (void) ztest_write(zd, object, offset, blocksize, data);
2315 (void) rw_unlock(&zd->zd_zilog_lock);
2317 umem_free(data, blocksize);
2321 * Initialize an object description template.
2324 ztest_od_init(ztest_od_t *od, uint64_t id, char *tag, uint64_t index,
2325 dmu_object_type_t type, uint64_t blocksize, uint64_t gen)
2327 od->od_dir = ZTEST_DIROBJ;
2330 od->od_crtype = type;
2331 od->od_crblocksize = blocksize ? blocksize : ztest_random_blocksize();
2334 od->od_type = DMU_OT_NONE;
2335 od->od_blocksize = 0;
2338 (void) snprintf(od->od_name, sizeof (od->od_name), "%s(%lld)[%llu]",
2339 tag, (longlong_t)id, (u_longlong_t)index);
2343 * Lookup or create the objects for a test using the od template.
2344 * If the objects do not all exist, or if 'remove' is specified,
2345 * remove any existing objects and create new ones. Otherwise,
2346 * use the existing objects.
2349 ztest_object_init(ztest_ds_t *zd, ztest_od_t *od, size_t size, boolean_t remove)
2351 int count = size / sizeof (*od);
2354 mutex_enter(&zd->zd_dirobj_lock);
2355 if ((ztest_lookup(zd, od, count) != 0 || remove) &&
2356 (ztest_remove(zd, od, count) != 0 ||
2357 ztest_create(zd, od, count) != 0))
2360 mutex_exit(&zd->zd_dirobj_lock);
2367 ztest_zil_commit(ztest_ds_t *zd, uint64_t id)
2369 zilog_t *zilog = zd->zd_zilog;
2371 (void) rw_rdlock(&zd->zd_zilog_lock);
2373 zil_commit(zilog, ztest_random(ZTEST_OBJECTS));
2376 * Remember the committed values in zd, which is in parent/child
2377 * shared memory. If we die, the next iteration of ztest_run()
2378 * will verify that the log really does contain this record.
2380 mutex_enter(&zilog->zl_lock);
2381 ASSERT(zd->zd_shared != NULL);
2382 ASSERT3U(zd->zd_shared->zd_seq, <=, zilog->zl_commit_lr_seq);
2383 zd->zd_shared->zd_seq = zilog->zl_commit_lr_seq;
2384 mutex_exit(&zilog->zl_lock);
2386 (void) rw_unlock(&zd->zd_zilog_lock);
2390 * This function is designed to simulate the operations that occur during a
2391 * mount/unmount operation. We hold the dataset across these operations in an
2392 * attempt to expose any implicit assumptions about ZIL management.
2396 ztest_zil_remount(ztest_ds_t *zd, uint64_t id)
2398 objset_t *os = zd->zd_os;
2401 * We grab the zd_dirobj_lock to ensure that no other thread is
2402 * updating the zil (i.e. adding in-memory log records) and the
2403 * zd_zilog_lock to block any I/O.
2405 mutex_enter(&zd->zd_dirobj_lock);
2406 (void) rw_wrlock(&zd->zd_zilog_lock);
2408 /* zfs_sb_teardown() */
2409 zil_close(zd->zd_zilog);
2411 /* zfsvfs_setup() */
2412 VERIFY(zil_open(os, ztest_get_data) == zd->zd_zilog);
2413 zil_replay(os, zd, ztest_replay_vector);
2415 (void) rw_unlock(&zd->zd_zilog_lock);
2416 mutex_exit(&zd->zd_dirobj_lock);
2420 * Verify that we can't destroy an active pool, create an existing pool,
2421 * or create a pool with a bad vdev spec.
2425 ztest_spa_create_destroy(ztest_ds_t *zd, uint64_t id)
2427 ztest_shared_opts_t *zo = &ztest_opts;
2432 * Attempt to create using a bad file.
2434 nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1);
2435 VERIFY3U(ENOENT, ==,
2436 spa_create("ztest_bad_file", nvroot, NULL, NULL));
2437 nvlist_free(nvroot);
2440 * Attempt to create using a bad mirror.
2442 nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 2, 1);
2443 VERIFY3U(ENOENT, ==,
2444 spa_create("ztest_bad_mirror", nvroot, NULL, NULL));
2445 nvlist_free(nvroot);
2448 * Attempt to create an existing pool. It shouldn't matter
2449 * what's in the nvroot; we should fail with EEXIST.
2451 (void) rw_rdlock(&ztest_name_lock);
2452 nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1);
2453 VERIFY3U(EEXIST, ==, spa_create(zo->zo_pool, nvroot, NULL, NULL));
2454 nvlist_free(nvroot);
2455 VERIFY3U(0, ==, spa_open(zo->zo_pool, &spa, FTAG));
2456 VERIFY3U(EBUSY, ==, spa_destroy(zo->zo_pool));
2457 spa_close(spa, FTAG);
2459 (void) rw_unlock(&ztest_name_lock);
2464 ztest_spa_upgrade(ztest_ds_t *zd, uint64_t id)
2467 uint64_t initial_version = SPA_VERSION_INITIAL;
2468 uint64_t version, newversion;
2469 nvlist_t *nvroot, *props;
2472 mutex_enter(&ztest_vdev_lock);
2473 name = kmem_asprintf("%s_upgrade", ztest_opts.zo_pool);
2476 * Clean up from previous runs.
2478 (void) spa_destroy(name);
2480 nvroot = make_vdev_root(NULL, NULL, name, ztest_opts.zo_vdev_size, 0,
2481 0, ztest_opts.zo_raidz, ztest_opts.zo_mirrors, 1);
2484 * If we're configuring a RAIDZ device then make sure that the
2485 * the initial version is capable of supporting that feature.
2487 switch (ztest_opts.zo_raidz_parity) {
2490 initial_version = SPA_VERSION_INITIAL;
2493 initial_version = SPA_VERSION_RAIDZ2;
2496 initial_version = SPA_VERSION_RAIDZ3;
2501 * Create a pool with a spa version that can be upgraded. Pick
2502 * a value between initial_version and SPA_VERSION_BEFORE_FEATURES.
2505 version = ztest_random_spa_version(initial_version);
2506 } while (version > SPA_VERSION_BEFORE_FEATURES);
2508 props = fnvlist_alloc();
2509 fnvlist_add_uint64(props,
2510 zpool_prop_to_name(ZPOOL_PROP_VERSION), version);
2511 VERIFY3S(spa_create(name, nvroot, props, NULL), ==, 0);
2512 fnvlist_free(nvroot);
2513 fnvlist_free(props);
2515 VERIFY3S(spa_open(name, &spa, FTAG), ==, 0);
2516 VERIFY3U(spa_version(spa), ==, version);
2517 newversion = ztest_random_spa_version(version + 1);
2519 if (ztest_opts.zo_verbose >= 4) {
2520 (void) printf("upgrading spa version from %llu to %llu\n",
2521 (u_longlong_t)version, (u_longlong_t)newversion);
2524 spa_upgrade(spa, newversion);
2525 VERIFY3U(spa_version(spa), >, version);
2526 VERIFY3U(spa_version(spa), ==, fnvlist_lookup_uint64(spa->spa_config,
2527 zpool_prop_to_name(ZPOOL_PROP_VERSION)));
2528 spa_close(spa, FTAG);
2531 mutex_exit(&ztest_vdev_lock);
2535 vdev_lookup_by_path(vdev_t *vd, const char *path)
2540 if (vd->vdev_path != NULL && strcmp(path, vd->vdev_path) == 0)
2543 for (c = 0; c < vd->vdev_children; c++)
2544 if ((mvd = vdev_lookup_by_path(vd->vdev_child[c], path)) !=
2552 * Find the first available hole which can be used as a top-level.
2555 find_vdev_hole(spa_t *spa)
2557 vdev_t *rvd = spa->spa_root_vdev;
2560 ASSERT(spa_config_held(spa, SCL_VDEV, RW_READER) == SCL_VDEV);
2562 for (c = 0; c < rvd->vdev_children; c++) {
2563 vdev_t *cvd = rvd->vdev_child[c];
2565 if (cvd->vdev_ishole)
2572 * Verify that vdev_add() works as expected.
2576 ztest_vdev_add_remove(ztest_ds_t *zd, uint64_t id)
2578 ztest_shared_t *zs = ztest_shared;
2579 spa_t *spa = ztest_spa;
2585 mutex_enter(&ztest_vdev_lock);
2586 leaves = MAX(zs->zs_mirrors + zs->zs_splits, 1) * ztest_opts.zo_raidz;
2588 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2590 ztest_shared->zs_vdev_next_leaf = find_vdev_hole(spa) * leaves;
2593 * If we have slogs then remove them 1/4 of the time.
2595 if (spa_has_slogs(spa) && ztest_random(4) == 0) {
2597 * Grab the guid from the head of the log class rotor.
2599 guid = spa_log_class(spa)->mc_rotor->mg_vd->vdev_guid;
2601 spa_config_exit(spa, SCL_VDEV, FTAG);
2604 * We have to grab the zs_name_lock as writer to
2605 * prevent a race between removing a slog (dmu_objset_find)
2606 * and destroying a dataset. Removing the slog will
2607 * grab a reference on the dataset which may cause
2608 * dsl_destroy_head() to fail with EBUSY thus
2609 * leaving the dataset in an inconsistent state.
2611 rw_wrlock(&ztest_name_lock);
2612 error = spa_vdev_remove(spa, guid, B_FALSE);
2613 rw_unlock(&ztest_name_lock);
2615 if (error && error != EEXIST)
2616 fatal(0, "spa_vdev_remove() = %d", error);
2618 spa_config_exit(spa, SCL_VDEV, FTAG);
2621 * Make 1/4 of the devices be log devices.
2623 nvroot = make_vdev_root(NULL, NULL, NULL,
2624 ztest_opts.zo_vdev_size, 0,
2625 ztest_random(4) == 0, ztest_opts.zo_raidz,
2628 error = spa_vdev_add(spa, nvroot);
2629 nvlist_free(nvroot);
2631 if (error == ENOSPC)
2632 ztest_record_enospc("spa_vdev_add");
2633 else if (error != 0)
2634 fatal(0, "spa_vdev_add() = %d", error);
2637 mutex_exit(&ztest_vdev_lock);
2641 * Verify that adding/removing aux devices (l2arc, hot spare) works as expected.
2645 ztest_vdev_aux_add_remove(ztest_ds_t *zd, uint64_t id)
2647 ztest_shared_t *zs = ztest_shared;
2648 spa_t *spa = ztest_spa;
2649 vdev_t *rvd = spa->spa_root_vdev;
2650 spa_aux_vdev_t *sav;
2656 path = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
2658 if (ztest_random(2) == 0) {
2659 sav = &spa->spa_spares;
2660 aux = ZPOOL_CONFIG_SPARES;
2662 sav = &spa->spa_l2cache;
2663 aux = ZPOOL_CONFIG_L2CACHE;
2666 mutex_enter(&ztest_vdev_lock);
2668 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2670 if (sav->sav_count != 0 && ztest_random(4) == 0) {
2672 * Pick a random device to remove.
2674 guid = sav->sav_vdevs[ztest_random(sav->sav_count)]->vdev_guid;
2677 * Find an unused device we can add.
2679 zs->zs_vdev_aux = 0;
2682 (void) snprintf(path, MAXPATHLEN, ztest_aux_template,
2683 ztest_opts.zo_dir, ztest_opts.zo_pool, aux,
2685 for (c = 0; c < sav->sav_count; c++)
2686 if (strcmp(sav->sav_vdevs[c]->vdev_path,
2689 if (c == sav->sav_count &&
2690 vdev_lookup_by_path(rvd, path) == NULL)
2696 spa_config_exit(spa, SCL_VDEV, FTAG);
2702 nvlist_t *nvroot = make_vdev_root(NULL, aux, NULL,
2703 (ztest_opts.zo_vdev_size * 5) / 4, 0, 0, 0, 0, 1);
2704 error = spa_vdev_add(spa, nvroot);
2706 fatal(0, "spa_vdev_add(%p) = %d", nvroot, error);
2707 nvlist_free(nvroot);
2710 * Remove an existing device. Sometimes, dirty its
2711 * vdev state first to make sure we handle removal
2712 * of devices that have pending state changes.
2714 if (ztest_random(2) == 0)
2715 (void) vdev_online(spa, guid, 0, NULL);
2717 error = spa_vdev_remove(spa, guid, B_FALSE);
2718 if (error != 0 && error != EBUSY)
2719 fatal(0, "spa_vdev_remove(%llu) = %d", guid, error);
2722 mutex_exit(&ztest_vdev_lock);
2724 umem_free(path, MAXPATHLEN);
2728 * split a pool if it has mirror tlvdevs
2732 ztest_split_pool(ztest_ds_t *zd, uint64_t id)
2734 ztest_shared_t *zs = ztest_shared;
2735 spa_t *spa = ztest_spa;
2736 vdev_t *rvd = spa->spa_root_vdev;
2737 nvlist_t *tree, **child, *config, *split, **schild;
2738 uint_t c, children, schildren = 0, lastlogid = 0;
2741 mutex_enter(&ztest_vdev_lock);
2743 /* ensure we have a useable config; mirrors of raidz aren't supported */
2744 if (zs->zs_mirrors < 3 || ztest_opts.zo_raidz > 1) {
2745 mutex_exit(&ztest_vdev_lock);
2749 /* clean up the old pool, if any */
2750 (void) spa_destroy("splitp");
2752 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2754 /* generate a config from the existing config */
2755 mutex_enter(&spa->spa_props_lock);
2756 VERIFY(nvlist_lookup_nvlist(spa->spa_config, ZPOOL_CONFIG_VDEV_TREE,
2758 mutex_exit(&spa->spa_props_lock);
2760 VERIFY(nvlist_lookup_nvlist_array(tree, ZPOOL_CONFIG_CHILDREN, &child,
2763 schild = malloc(rvd->vdev_children * sizeof (nvlist_t *));
2764 for (c = 0; c < children; c++) {
2765 vdev_t *tvd = rvd->vdev_child[c];
2769 if (tvd->vdev_islog || tvd->vdev_ops == &vdev_hole_ops) {
2770 VERIFY(nvlist_alloc(&schild[schildren], NV_UNIQUE_NAME,
2772 VERIFY(nvlist_add_string(schild[schildren],
2773 ZPOOL_CONFIG_TYPE, VDEV_TYPE_HOLE) == 0);
2774 VERIFY(nvlist_add_uint64(schild[schildren],
2775 ZPOOL_CONFIG_IS_HOLE, 1) == 0);
2777 lastlogid = schildren;
2782 VERIFY(nvlist_lookup_nvlist_array(child[c],
2783 ZPOOL_CONFIG_CHILDREN, &mchild, &mchildren) == 0);
2784 VERIFY(nvlist_dup(mchild[0], &schild[schildren++], 0) == 0);
2787 /* OK, create a config that can be used to split */
2788 VERIFY(nvlist_alloc(&split, NV_UNIQUE_NAME, 0) == 0);
2789 VERIFY(nvlist_add_string(split, ZPOOL_CONFIG_TYPE,
2790 VDEV_TYPE_ROOT) == 0);
2791 VERIFY(nvlist_add_nvlist_array(split, ZPOOL_CONFIG_CHILDREN, schild,
2792 lastlogid != 0 ? lastlogid : schildren) == 0);
2794 VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, 0) == 0);
2795 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, split) == 0);
2797 for (c = 0; c < schildren; c++)
2798 nvlist_free(schild[c]);
2802 spa_config_exit(spa, SCL_VDEV, FTAG);
2804 (void) rw_wrlock(&ztest_name_lock);
2805 error = spa_vdev_split_mirror(spa, "splitp", config, NULL, B_FALSE);
2806 (void) rw_unlock(&ztest_name_lock);
2808 nvlist_free(config);
2811 (void) printf("successful split - results:\n");
2812 mutex_enter(&spa_namespace_lock);
2813 show_pool_stats(spa);
2814 show_pool_stats(spa_lookup("splitp"));
2815 mutex_exit(&spa_namespace_lock);
2819 mutex_exit(&ztest_vdev_lock);
2824 * Verify that we can attach and detach devices.
2828 ztest_vdev_attach_detach(ztest_ds_t *zd, uint64_t id)
2830 ztest_shared_t *zs = ztest_shared;
2831 spa_t *spa = ztest_spa;
2832 spa_aux_vdev_t *sav = &spa->spa_spares;
2833 vdev_t *rvd = spa->spa_root_vdev;
2834 vdev_t *oldvd, *newvd, *pvd;
2838 uint64_t ashift = ztest_get_ashift();
2839 uint64_t oldguid, pguid;
2840 uint64_t oldsize, newsize;
2841 char *oldpath, *newpath;
2843 int oldvd_has_siblings = B_FALSE;
2844 int newvd_is_spare = B_FALSE;
2846 int error, expected_error;
2848 oldpath = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
2849 newpath = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
2851 mutex_enter(&ztest_vdev_lock);
2852 leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz;
2854 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2857 * Decide whether to do an attach or a replace.
2859 replacing = ztest_random(2);
2862 * Pick a random top-level vdev.
2864 top = ztest_random_vdev_top(spa, B_TRUE);
2867 * Pick a random leaf within it.
2869 leaf = ztest_random(leaves);
2874 oldvd = rvd->vdev_child[top];
2875 if (zs->zs_mirrors >= 1) {
2876 ASSERT(oldvd->vdev_ops == &vdev_mirror_ops);
2877 ASSERT(oldvd->vdev_children >= zs->zs_mirrors);
2878 oldvd = oldvd->vdev_child[leaf / ztest_opts.zo_raidz];
2880 if (ztest_opts.zo_raidz > 1) {
2881 ASSERT(oldvd->vdev_ops == &vdev_raidz_ops);
2882 ASSERT(oldvd->vdev_children == ztest_opts.zo_raidz);
2883 oldvd = oldvd->vdev_child[leaf % ztest_opts.zo_raidz];
2887 * If we're already doing an attach or replace, oldvd may be a
2888 * mirror vdev -- in which case, pick a random child.
2890 while (oldvd->vdev_children != 0) {
2891 oldvd_has_siblings = B_TRUE;
2892 ASSERT(oldvd->vdev_children >= 2);
2893 oldvd = oldvd->vdev_child[ztest_random(oldvd->vdev_children)];
2896 oldguid = oldvd->vdev_guid;
2897 oldsize = vdev_get_min_asize(oldvd);
2898 oldvd_is_log = oldvd->vdev_top->vdev_islog;
2899 (void) strcpy(oldpath, oldvd->vdev_path);
2900 pvd = oldvd->vdev_parent;
2901 pguid = pvd->vdev_guid;
2904 * If oldvd has siblings, then half of the time, detach it.
2906 if (oldvd_has_siblings && ztest_random(2) == 0) {
2907 spa_config_exit(spa, SCL_VDEV, FTAG);
2908 error = spa_vdev_detach(spa, oldguid, pguid, B_FALSE);
2909 if (error != 0 && error != ENODEV && error != EBUSY &&
2911 fatal(0, "detach (%s) returned %d", oldpath, error);
2916 * For the new vdev, choose with equal probability between the two
2917 * standard paths (ending in either 'a' or 'b') or a random hot spare.
2919 if (sav->sav_count != 0 && ztest_random(3) == 0) {
2920 newvd = sav->sav_vdevs[ztest_random(sav->sav_count)];
2921 newvd_is_spare = B_TRUE;
2922 (void) strcpy(newpath, newvd->vdev_path);
2924 (void) snprintf(newpath, MAXPATHLEN, ztest_dev_template,
2925 ztest_opts.zo_dir, ztest_opts.zo_pool,
2926 top * leaves + leaf);
2927 if (ztest_random(2) == 0)
2928 newpath[strlen(newpath) - 1] = 'b';
2929 newvd = vdev_lookup_by_path(rvd, newpath);
2933 newsize = vdev_get_min_asize(newvd);
2936 * Make newsize a little bigger or smaller than oldsize.
2937 * If it's smaller, the attach should fail.
2938 * If it's larger, and we're doing a replace,
2939 * we should get dynamic LUN growth when we're done.
2941 newsize = 10 * oldsize / (9 + ztest_random(3));
2945 * If pvd is not a mirror or root, the attach should fail with ENOTSUP,
2946 * unless it's a replace; in that case any non-replacing parent is OK.
2948 * If newvd is already part of the pool, it should fail with EBUSY.
2950 * If newvd is too small, it should fail with EOVERFLOW.
2952 if (pvd->vdev_ops != &vdev_mirror_ops &&
2953 pvd->vdev_ops != &vdev_root_ops && (!replacing ||
2954 pvd->vdev_ops == &vdev_replacing_ops ||
2955 pvd->vdev_ops == &vdev_spare_ops))
2956 expected_error = ENOTSUP;
2957 else if (newvd_is_spare && (!replacing || oldvd_is_log))
2958 expected_error = ENOTSUP;
2959 else if (newvd == oldvd)
2960 expected_error = replacing ? 0 : EBUSY;
2961 else if (vdev_lookup_by_path(rvd, newpath) != NULL)
2962 expected_error = EBUSY;
2963 else if (newsize < oldsize)
2964 expected_error = EOVERFLOW;
2965 else if (ashift > oldvd->vdev_top->vdev_ashift)
2966 expected_error = EDOM;
2970 spa_config_exit(spa, SCL_VDEV, FTAG);
2973 * Build the nvlist describing newpath.
2975 root = make_vdev_root(newpath, NULL, NULL, newvd == NULL ? newsize : 0,
2976 ashift, 0, 0, 0, 1);
2978 error = spa_vdev_attach(spa, oldguid, root, replacing);
2983 * If our parent was the replacing vdev, but the replace completed,
2984 * then instead of failing with ENOTSUP we may either succeed,
2985 * fail with ENODEV, or fail with EOVERFLOW.
2987 if (expected_error == ENOTSUP &&
2988 (error == 0 || error == ENODEV || error == EOVERFLOW))
2989 expected_error = error;
2992 * If someone grew the LUN, the replacement may be too small.
2994 if (error == EOVERFLOW || error == EBUSY)
2995 expected_error = error;
2997 /* XXX workaround 6690467 */
2998 if (error != expected_error && expected_error != EBUSY) {
2999 fatal(0, "attach (%s %llu, %s %llu, %d) "
3000 "returned %d, expected %d",
3001 oldpath, oldsize, newpath,
3002 newsize, replacing, error, expected_error);
3005 mutex_exit(&ztest_vdev_lock);
3007 umem_free(oldpath, MAXPATHLEN);
3008 umem_free(newpath, MAXPATHLEN);
3012 * Callback function which expands the physical size of the vdev.
3015 grow_vdev(vdev_t *vd, void *arg)
3017 ASSERTV(spa_t *spa = vd->vdev_spa);
3018 size_t *newsize = arg;
3022 ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
3023 ASSERT(vd->vdev_ops->vdev_op_leaf);
3025 if ((fd = open(vd->vdev_path, O_RDWR)) == -1)
3028 fsize = lseek(fd, 0, SEEK_END);
3029 VERIFY(ftruncate(fd, *newsize) == 0);
3031 if (ztest_opts.zo_verbose >= 6) {
3032 (void) printf("%s grew from %lu to %lu bytes\n",
3033 vd->vdev_path, (ulong_t)fsize, (ulong_t)*newsize);
3040 * Callback function which expands a given vdev by calling vdev_online().
3044 online_vdev(vdev_t *vd, void *arg)
3046 spa_t *spa = vd->vdev_spa;
3047 vdev_t *tvd = vd->vdev_top;
3048 uint64_t guid = vd->vdev_guid;
3049 uint64_t generation = spa->spa_config_generation + 1;
3050 vdev_state_t newstate = VDEV_STATE_UNKNOWN;
3053 ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
3054 ASSERT(vd->vdev_ops->vdev_op_leaf);
3056 /* Calling vdev_online will initialize the new metaslabs */
3057 spa_config_exit(spa, SCL_STATE, spa);
3058 error = vdev_online(spa, guid, ZFS_ONLINE_EXPAND, &newstate);
3059 spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3062 * If vdev_online returned an error or the underlying vdev_open
3063 * failed then we abort the expand. The only way to know that
3064 * vdev_open fails is by checking the returned newstate.
3066 if (error || newstate != VDEV_STATE_HEALTHY) {
3067 if (ztest_opts.zo_verbose >= 5) {
3068 (void) printf("Unable to expand vdev, state %llu, "
3069 "error %d\n", (u_longlong_t)newstate, error);
3073 ASSERT3U(newstate, ==, VDEV_STATE_HEALTHY);
3076 * Since we dropped the lock we need to ensure that we're
3077 * still talking to the original vdev. It's possible this
3078 * vdev may have been detached/replaced while we were
3079 * trying to online it.
3081 if (generation != spa->spa_config_generation) {
3082 if (ztest_opts.zo_verbose >= 5) {
3083 (void) printf("vdev configuration has changed, "
3084 "guid %llu, state %llu, expected gen %llu, "
3087 (u_longlong_t)tvd->vdev_state,
3088 (u_longlong_t)generation,
3089 (u_longlong_t)spa->spa_config_generation);
3097 * Traverse the vdev tree calling the supplied function.
3098 * We continue to walk the tree until we either have walked all
3099 * children or we receive a non-NULL return from the callback.
3100 * If a NULL callback is passed, then we just return back the first
3101 * leaf vdev we encounter.
3104 vdev_walk_tree(vdev_t *vd, vdev_t *(*func)(vdev_t *, void *), void *arg)
3108 if (vd->vdev_ops->vdev_op_leaf) {
3112 return (func(vd, arg));
3115 for (c = 0; c < vd->vdev_children; c++) {
3116 vdev_t *cvd = vd->vdev_child[c];
3117 if ((cvd = vdev_walk_tree(cvd, func, arg)) != NULL)
3124 * Verify that dynamic LUN growth works as expected.
3128 ztest_vdev_LUN_growth(ztest_ds_t *zd, uint64_t id)
3130 spa_t *spa = ztest_spa;
3132 metaslab_class_t *mc;
3133 metaslab_group_t *mg;
3134 size_t psize, newsize;
3136 uint64_t old_class_space, new_class_space, old_ms_count, new_ms_count;
3138 mutex_enter(&ztest_vdev_lock);
3139 spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3141 top = ztest_random_vdev_top(spa, B_TRUE);
3143 tvd = spa->spa_root_vdev->vdev_child[top];
3146 old_ms_count = tvd->vdev_ms_count;
3147 old_class_space = metaslab_class_get_space(mc);
3150 * Determine the size of the first leaf vdev associated with
3151 * our top-level device.
3153 vd = vdev_walk_tree(tvd, NULL, NULL);
3154 ASSERT3P(vd, !=, NULL);
3155 ASSERT(vd->vdev_ops->vdev_op_leaf);
3157 psize = vd->vdev_psize;
3160 * We only try to expand the vdev if it's healthy, less than 4x its
3161 * original size, and it has a valid psize.
3163 if (tvd->vdev_state != VDEV_STATE_HEALTHY ||
3164 psize == 0 || psize >= 4 * ztest_opts.zo_vdev_size) {
3165 spa_config_exit(spa, SCL_STATE, spa);
3166 mutex_exit(&ztest_vdev_lock);
3170 newsize = psize + psize / 8;
3171 ASSERT3U(newsize, >, psize);
3173 if (ztest_opts.zo_verbose >= 6) {
3174 (void) printf("Expanding LUN %s from %lu to %lu\n",
3175 vd->vdev_path, (ulong_t)psize, (ulong_t)newsize);
3179 * Growing the vdev is a two step process:
3180 * 1). expand the physical size (i.e. relabel)
3181 * 2). online the vdev to create the new metaslabs
3183 if (vdev_walk_tree(tvd, grow_vdev, &newsize) != NULL ||
3184 vdev_walk_tree(tvd, online_vdev, NULL) != NULL ||
3185 tvd->vdev_state != VDEV_STATE_HEALTHY) {
3186 if (ztest_opts.zo_verbose >= 5) {
3187 (void) printf("Could not expand LUN because "
3188 "the vdev configuration changed.\n");
3190 spa_config_exit(spa, SCL_STATE, spa);
3191 mutex_exit(&ztest_vdev_lock);
3195 spa_config_exit(spa, SCL_STATE, spa);
3198 * Expanding the LUN will update the config asynchronously,
3199 * thus we must wait for the async thread to complete any
3200 * pending tasks before proceeding.
3204 mutex_enter(&spa->spa_async_lock);
3205 done = (spa->spa_async_thread == NULL && !spa->spa_async_tasks);
3206 mutex_exit(&spa->spa_async_lock);
3209 txg_wait_synced(spa_get_dsl(spa), 0);
3210 (void) poll(NULL, 0, 100);
3213 spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3215 tvd = spa->spa_root_vdev->vdev_child[top];
3216 new_ms_count = tvd->vdev_ms_count;
3217 new_class_space = metaslab_class_get_space(mc);
3219 if (tvd->vdev_mg != mg || mg->mg_class != mc) {
3220 if (ztest_opts.zo_verbose >= 5) {
3221 (void) printf("Could not verify LUN expansion due to "
3222 "intervening vdev offline or remove.\n");
3224 spa_config_exit(spa, SCL_STATE, spa);
3225 mutex_exit(&ztest_vdev_lock);
3230 * Make sure we were able to grow the vdev.
3232 if (new_ms_count <= old_ms_count)
3233 fatal(0, "LUN expansion failed: ms_count %llu <= %llu\n",
3234 old_ms_count, new_ms_count);
3237 * Make sure we were able to grow the pool.
3239 if (new_class_space <= old_class_space)
3240 fatal(0, "LUN expansion failed: class_space %llu <= %llu\n",
3241 old_class_space, new_class_space);
3243 if (ztest_opts.zo_verbose >= 5) {
3244 char oldnumbuf[6], newnumbuf[6];
3246 nicenum(old_class_space, oldnumbuf);
3247 nicenum(new_class_space, newnumbuf);
3248 (void) printf("%s grew from %s to %s\n",
3249 spa->spa_name, oldnumbuf, newnumbuf);
3252 spa_config_exit(spa, SCL_STATE, spa);
3253 mutex_exit(&ztest_vdev_lock);
3257 * Verify that dmu_objset_{create,destroy,open,close} work as expected.
3261 ztest_objset_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
3264 * Create the objects common to all ztest datasets.
3266 VERIFY(zap_create_claim(os, ZTEST_DIROBJ,
3267 DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx) == 0);
3271 ztest_dataset_create(char *dsname)
3273 uint64_t zilset = ztest_random(100);
3274 int err = dmu_objset_create(dsname, DMU_OST_OTHER, 0,
3275 ztest_objset_create_cb, NULL);
3277 if (err || zilset < 80)
3280 if (ztest_opts.zo_verbose >= 5)
3281 (void) printf("Setting dataset %s to sync always\n", dsname);
3282 return (ztest_dsl_prop_set_uint64(dsname, ZFS_PROP_SYNC,
3283 ZFS_SYNC_ALWAYS, B_FALSE));
3288 ztest_objset_destroy_cb(const char *name, void *arg)
3291 dmu_object_info_t doi;
3295 * Verify that the dataset contains a directory object.
3297 VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_TRUE, FTAG, &os));
3298 error = dmu_object_info(os, ZTEST_DIROBJ, &doi);
3299 if (error != ENOENT) {
3300 /* We could have crashed in the middle of destroying it */
3302 ASSERT3U(doi.doi_type, ==, DMU_OT_ZAP_OTHER);
3303 ASSERT3S(doi.doi_physical_blocks_512, >=, 0);
3305 dmu_objset_disown(os, FTAG);
3308 * Destroy the dataset.
3310 if (strchr(name, '@') != NULL) {
3311 VERIFY0(dsl_destroy_snapshot(name, B_FALSE));
3313 VERIFY0(dsl_destroy_head(name));
3319 ztest_snapshot_create(char *osname, uint64_t id)
3321 char snapname[MAXNAMELEN];
3324 (void) snprintf(snapname, sizeof (snapname), "%llu", (u_longlong_t)id);
3326 error = dmu_objset_snapshot_one(osname, snapname);
3327 if (error == ENOSPC) {
3328 ztest_record_enospc(FTAG);
3331 if (error != 0 && error != EEXIST) {
3332 fatal(0, "ztest_snapshot_create(%s@%s) = %d", osname,
3339 ztest_snapshot_destroy(char *osname, uint64_t id)
3341 char snapname[MAXNAMELEN];
3344 (void) snprintf(snapname, MAXNAMELEN, "%s@%llu", osname,
3347 error = dsl_destroy_snapshot(snapname, B_FALSE);
3348 if (error != 0 && error != ENOENT)
3349 fatal(0, "ztest_snapshot_destroy(%s) = %d", snapname, error);
3355 ztest_dmu_objset_create_destroy(ztest_ds_t *zd, uint64_t id)
3365 zdtmp = umem_alloc(sizeof (ztest_ds_t), UMEM_NOFAIL);
3366 name = umem_alloc(MAXNAMELEN, UMEM_NOFAIL);
3368 (void) rw_rdlock(&ztest_name_lock);
3370 (void) snprintf(name, MAXNAMELEN, "%s/temp_%llu",
3371 ztest_opts.zo_pool, (u_longlong_t)id);
3374 * If this dataset exists from a previous run, process its replay log
3375 * half of the time. If we don't replay it, then dsl_destroy_head()
3376 * (invoked from ztest_objset_destroy_cb()) should just throw it away.
3378 if (ztest_random(2) == 0 &&
3379 dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os) == 0) {
3380 ztest_zd_init(zdtmp, NULL, os);
3381 zil_replay(os, zdtmp, ztest_replay_vector);
3382 ztest_zd_fini(zdtmp);
3383 dmu_objset_disown(os, FTAG);
3387 * There may be an old instance of the dataset we're about to
3388 * create lying around from a previous run. If so, destroy it
3389 * and all of its snapshots.
3391 (void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
3392 DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
3395 * Verify that the destroyed dataset is no longer in the namespace.
3397 VERIFY3U(ENOENT, ==, dmu_objset_own(name, DMU_OST_OTHER, B_TRUE,
3401 * Verify that we can create a new dataset.
3403 error = ztest_dataset_create(name);
3405 if (error == ENOSPC) {
3406 ztest_record_enospc(FTAG);
3409 fatal(0, "dmu_objset_create(%s) = %d", name, error);
3412 VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os));
3414 ztest_zd_init(zdtmp, NULL, os);
3417 * Open the intent log for it.
3419 zilog = zil_open(os, ztest_get_data);
3422 * Put some objects in there, do a little I/O to them,
3423 * and randomly take a couple of snapshots along the way.
3425 iters = ztest_random(5);
3426 for (i = 0; i < iters; i++) {
3427 ztest_dmu_object_alloc_free(zdtmp, id);
3428 if (ztest_random(iters) == 0)
3429 (void) ztest_snapshot_create(name, i);
3433 * Verify that we cannot create an existing dataset.
3435 VERIFY3U(EEXIST, ==,
3436 dmu_objset_create(name, DMU_OST_OTHER, 0, NULL, NULL));
3439 * Verify that we can hold an objset that is also owned.
3441 VERIFY3U(0, ==, dmu_objset_hold(name, FTAG, &os2));
3442 dmu_objset_rele(os2, FTAG);
3445 * Verify that we cannot own an objset that is already owned.
3448 dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os2));
3451 dmu_objset_disown(os, FTAG);
3452 ztest_zd_fini(zdtmp);
3454 (void) rw_unlock(&ztest_name_lock);
3456 umem_free(name, MAXNAMELEN);
3457 umem_free(zdtmp, sizeof (ztest_ds_t));
3461 * Verify that dmu_snapshot_{create,destroy,open,close} work as expected.
3464 ztest_dmu_snapshot_create_destroy(ztest_ds_t *zd, uint64_t id)
3466 (void) rw_rdlock(&ztest_name_lock);
3467 (void) ztest_snapshot_destroy(zd->zd_name, id);
3468 (void) ztest_snapshot_create(zd->zd_name, id);
3469 (void) rw_unlock(&ztest_name_lock);
3473 * Cleanup non-standard snapshots and clones.
3476 ztest_dsl_dataset_cleanup(char *osname, uint64_t id)
3485 snap1name = umem_alloc(MAXNAMELEN, UMEM_NOFAIL);
3486 clone1name = umem_alloc(MAXNAMELEN, UMEM_NOFAIL);
3487 snap2name = umem_alloc(MAXNAMELEN, UMEM_NOFAIL);
3488 clone2name = umem_alloc(MAXNAMELEN, UMEM_NOFAIL);
3489 snap3name = umem_alloc(MAXNAMELEN, UMEM_NOFAIL);
3491 (void) snprintf(snap1name, MAXNAMELEN, "%s@s1_%llu",
3492 osname, (u_longlong_t)id);
3493 (void) snprintf(clone1name, MAXNAMELEN, "%s/c1_%llu",
3494 osname, (u_longlong_t)id);
3495 (void) snprintf(snap2name, MAXNAMELEN, "%s@s2_%llu",
3496 clone1name, (u_longlong_t)id);
3497 (void) snprintf(clone2name, MAXNAMELEN, "%s/c2_%llu",
3498 osname, (u_longlong_t)id);
3499 (void) snprintf(snap3name, MAXNAMELEN, "%s@s3_%llu",
3500 clone1name, (u_longlong_t)id);
3502 error = dsl_destroy_head(clone2name);
3503 if (error && error != ENOENT)
3504 fatal(0, "dsl_destroy_head(%s) = %d", clone2name, error);
3505 error = dsl_destroy_snapshot(snap3name, B_FALSE);
3506 if (error && error != ENOENT)
3507 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap3name, error);
3508 error = dsl_destroy_snapshot(snap2name, B_FALSE);
3509 if (error && error != ENOENT)
3510 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap2name, error);
3511 error = dsl_destroy_head(clone1name);
3512 if (error && error != ENOENT)
3513 fatal(0, "dsl_destroy_head(%s) = %d", clone1name, error);
3514 error = dsl_destroy_snapshot(snap1name, B_FALSE);
3515 if (error && error != ENOENT)
3516 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap1name, error);
3518 umem_free(snap1name, MAXNAMELEN);
3519 umem_free(clone1name, MAXNAMELEN);
3520 umem_free(snap2name, MAXNAMELEN);
3521 umem_free(clone2name, MAXNAMELEN);
3522 umem_free(snap3name, MAXNAMELEN);
3526 * Verify dsl_dataset_promote handles EBUSY
3529 ztest_dsl_dataset_promote_busy(ztest_ds_t *zd, uint64_t id)
3537 char *osname = zd->zd_name;
3540 snap1name = umem_alloc(MAXNAMELEN, UMEM_NOFAIL);
3541 clone1name = umem_alloc(MAXNAMELEN, UMEM_NOFAIL);
3542 snap2name = umem_alloc(MAXNAMELEN, UMEM_NOFAIL);
3543 clone2name = umem_alloc(MAXNAMELEN, UMEM_NOFAIL);
3544 snap3name = umem_alloc(MAXNAMELEN, UMEM_NOFAIL);
3546 (void) rw_rdlock(&ztest_name_lock);
3548 ztest_dsl_dataset_cleanup(osname, id);
3550 (void) snprintf(snap1name, MAXNAMELEN, "%s@s1_%llu",
3551 osname, (u_longlong_t)id);
3552 (void) snprintf(clone1name, MAXNAMELEN, "%s/c1_%llu",
3553 osname, (u_longlong_t)id);
3554 (void) snprintf(snap2name, MAXNAMELEN, "%s@s2_%llu",
3555 clone1name, (u_longlong_t)id);
3556 (void) snprintf(clone2name, MAXNAMELEN, "%s/c2_%llu",
3557 osname, (u_longlong_t)id);
3558 (void) snprintf(snap3name, MAXNAMELEN, "%s@s3_%llu",
3559 clone1name, (u_longlong_t)id);
3561 error = dmu_objset_snapshot_one(osname, strchr(snap1name, '@') + 1);
3562 if (error && error != EEXIST) {
3563 if (error == ENOSPC) {
3564 ztest_record_enospc(FTAG);
3567 fatal(0, "dmu_take_snapshot(%s) = %d", snap1name, error);
3570 error = dmu_objset_clone(clone1name, snap1name);
3572 if (error == ENOSPC) {
3573 ztest_record_enospc(FTAG);
3576 fatal(0, "dmu_objset_create(%s) = %d", clone1name, error);
3579 error = dmu_objset_snapshot_one(clone1name, strchr(snap2name, '@') + 1);
3580 if (error && error != EEXIST) {
3581 if (error == ENOSPC) {
3582 ztest_record_enospc(FTAG);
3585 fatal(0, "dmu_open_snapshot(%s) = %d", snap2name, error);
3588 error = dmu_objset_snapshot_one(clone1name, strchr(snap3name, '@') + 1);
3589 if (error && error != EEXIST) {
3590 if (error == ENOSPC) {
3591 ztest_record_enospc(FTAG);
3594 fatal(0, "dmu_open_snapshot(%s) = %d", snap3name, error);
3597 error = dmu_objset_clone(clone2name, snap3name);
3599 if (error == ENOSPC) {
3600 ztest_record_enospc(FTAG);
3603 fatal(0, "dmu_objset_create(%s) = %d", clone2name, error);
3606 error = dmu_objset_own(snap2name, DMU_OST_ANY, B_TRUE, FTAG, &os);
3608 fatal(0, "dmu_objset_own(%s) = %d", snap2name, error);
3609 error = dsl_dataset_promote(clone2name, NULL);
3610 if (error == ENOSPC) {
3611 dmu_objset_disown(os, FTAG);
3612 ztest_record_enospc(FTAG);
3616 fatal(0, "dsl_dataset_promote(%s), %d, not EBUSY", clone2name,
3618 dmu_objset_disown(os, FTAG);
3621 ztest_dsl_dataset_cleanup(osname, id);
3623 (void) rw_unlock(&ztest_name_lock);
3625 umem_free(snap1name, MAXNAMELEN);
3626 umem_free(clone1name, MAXNAMELEN);
3627 umem_free(snap2name, MAXNAMELEN);
3628 umem_free(clone2name, MAXNAMELEN);
3629 umem_free(snap3name, MAXNAMELEN);
3632 #undef OD_ARRAY_SIZE
3633 #define OD_ARRAY_SIZE 4
3636 * Verify that dmu_object_{alloc,free} work as expected.
3639 ztest_dmu_object_alloc_free(ztest_ds_t *zd, uint64_t id)
3646 size = sizeof (ztest_od_t) * OD_ARRAY_SIZE;
3647 od = umem_alloc(size, UMEM_NOFAIL);
3648 batchsize = OD_ARRAY_SIZE;
3650 for (b = 0; b < batchsize; b++)
3651 ztest_od_init(od + b, id, FTAG, b, DMU_OT_UINT64_OTHER, 0, 0);
3654 * Destroy the previous batch of objects, create a new batch,
3655 * and do some I/O on the new objects.
3657 if (ztest_object_init(zd, od, size, B_TRUE) != 0)
3660 while (ztest_random(4 * batchsize) != 0)
3661 ztest_io(zd, od[ztest_random(batchsize)].od_object,
3662 ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
3664 umem_free(od, size);
3667 #undef OD_ARRAY_SIZE
3668 #define OD_ARRAY_SIZE 2
3671 * Verify that dmu_{read,write} work as expected.
3674 ztest_dmu_read_write(ztest_ds_t *zd, uint64_t id)
3679 objset_t *os = zd->zd_os;
3680 size = sizeof (ztest_od_t) * OD_ARRAY_SIZE;
3681 od = umem_alloc(size, UMEM_NOFAIL);
3683 int i, freeit, error;
3685 bufwad_t *packbuf, *bigbuf, *pack, *bigH, *bigT;
3686 uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
3687 uint64_t chunksize = (1000 + ztest_random(1000)) * sizeof (uint64_t);
3688 uint64_t regions = 997;
3689 uint64_t stride = 123456789ULL;
3690 uint64_t width = 40;
3691 int free_percent = 5;
3694 * This test uses two objects, packobj and bigobj, that are always
3695 * updated together (i.e. in the same tx) so that their contents are
3696 * in sync and can be compared. Their contents relate to each other
3697 * in a simple way: packobj is a dense array of 'bufwad' structures,
3698 * while bigobj is a sparse array of the same bufwads. Specifically,
3699 * for any index n, there are three bufwads that should be identical:
3701 * packobj, at offset n * sizeof (bufwad_t)
3702 * bigobj, at the head of the nth chunk
3703 * bigobj, at the tail of the nth chunk
3705 * The chunk size is arbitrary. It doesn't have to be a power of two,
3706 * and it doesn't have any relation to the object blocksize.
3707 * The only requirement is that it can hold at least two bufwads.
3709 * Normally, we write the bufwad to each of these locations.
3710 * However, free_percent of the time we instead write zeroes to
3711 * packobj and perform a dmu_free_range() on bigobj. By comparing
3712 * bigobj to packobj, we can verify that the DMU is correctly
3713 * tracking which parts of an object are allocated and free,
3714 * and that the contents of the allocated blocks are correct.
3718 * Read the directory info. If it's the first time, set things up.
3720 ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, chunksize);
3721 ztest_od_init(od + 1, id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize);
3723 if (ztest_object_init(zd, od, size, B_FALSE) != 0) {
3724 umem_free(od, size);
3728 bigobj = od[0].od_object;
3729 packobj = od[1].od_object;
3730 chunksize = od[0].od_gen;
3731 ASSERT(chunksize == od[1].od_gen);
3734 * Prefetch a random chunk of the big object.
3735 * Our aim here is to get some async reads in flight
3736 * for blocks that we may free below; the DMU should
3737 * handle this race correctly.
3739 n = ztest_random(regions) * stride + ztest_random(width);
3740 s = 1 + ztest_random(2 * width - 1);
3741 dmu_prefetch(os, bigobj, n * chunksize, s * chunksize);
3744 * Pick a random index and compute the offsets into packobj and bigobj.
3746 n = ztest_random(regions) * stride + ztest_random(width);
3747 s = 1 + ztest_random(width - 1);
3749 packoff = n * sizeof (bufwad_t);
3750 packsize = s * sizeof (bufwad_t);
3752 bigoff = n * chunksize;
3753 bigsize = s * chunksize;
3755 packbuf = umem_alloc(packsize, UMEM_NOFAIL);
3756 bigbuf = umem_alloc(bigsize, UMEM_NOFAIL);
3759 * free_percent of the time, free a range of bigobj rather than
3762 freeit = (ztest_random(100) < free_percent);
3765 * Read the current contents of our objects.
3767 error = dmu_read(os, packobj, packoff, packsize, packbuf,
3770 error = dmu_read(os, bigobj, bigoff, bigsize, bigbuf,
3775 * Get a tx for the mods to both packobj and bigobj.
3777 tx = dmu_tx_create(os);
3779 dmu_tx_hold_write(tx, packobj, packoff, packsize);
3782 dmu_tx_hold_free(tx, bigobj, bigoff, bigsize);
3784 dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
3786 /* This accounts for setting the checksum/compression. */
3787 dmu_tx_hold_bonus(tx, bigobj);
3789 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3791 umem_free(packbuf, packsize);
3792 umem_free(bigbuf, bigsize);
3793 umem_free(od, size);
3797 enum zio_checksum cksum;
3799 cksum = (enum zio_checksum)
3800 ztest_random_dsl_prop(ZFS_PROP_CHECKSUM);
3801 } while (cksum >= ZIO_CHECKSUM_LEGACY_FUNCTIONS);
3802 dmu_object_set_checksum(os, bigobj, cksum, tx);
3804 enum zio_compress comp;
3806 comp = (enum zio_compress)
3807 ztest_random_dsl_prop(ZFS_PROP_COMPRESSION);
3808 } while (comp >= ZIO_COMPRESS_LEGACY_FUNCTIONS);
3809 dmu_object_set_compress(os, bigobj, comp, tx);
3812 * For each index from n to n + s, verify that the existing bufwad
3813 * in packobj matches the bufwads at the head and tail of the
3814 * corresponding chunk in bigobj. Then update all three bufwads
3815 * with the new values we want to write out.
3817 for (i = 0; i < s; i++) {
3819 pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
3821 bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
3823 bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
3825 ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
3826 ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
3828 if (pack->bw_txg > txg)
3829 fatal(0, "future leak: got %llx, open txg is %llx",
3832 if (pack->bw_data != 0 && pack->bw_index != n + i)
3833 fatal(0, "wrong index: got %llx, wanted %llx+%llx",
3834 pack->bw_index, n, i);
3836 if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
3837 fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
3839 if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
3840 fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
3843 bzero(pack, sizeof (bufwad_t));
3845 pack->bw_index = n + i;
3847 pack->bw_data = 1 + ztest_random(-2ULL);
3854 * We've verified all the old bufwads, and made new ones.
3855 * Now write them out.
3857 dmu_write(os, packobj, packoff, packsize, packbuf, tx);
3860 if (ztest_opts.zo_verbose >= 7) {
3861 (void) printf("freeing offset %llx size %llx"
3863 (u_longlong_t)bigoff,
3864 (u_longlong_t)bigsize,
3867 VERIFY(0 == dmu_free_range(os, bigobj, bigoff, bigsize, tx));
3869 if (ztest_opts.zo_verbose >= 7) {
3870 (void) printf("writing offset %llx size %llx"
3872 (u_longlong_t)bigoff,
3873 (u_longlong_t)bigsize,
3876 dmu_write(os, bigobj, bigoff, bigsize, bigbuf, tx);
3882 * Sanity check the stuff we just wrote.
3885 void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
3886 void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
3888 VERIFY(0 == dmu_read(os, packobj, packoff,
3889 packsize, packcheck, DMU_READ_PREFETCH));
3890 VERIFY(0 == dmu_read(os, bigobj, bigoff,
3891 bigsize, bigcheck, DMU_READ_PREFETCH));
3893 ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
3894 ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
3896 umem_free(packcheck, packsize);
3897 umem_free(bigcheck, bigsize);
3900 umem_free(packbuf, packsize);
3901 umem_free(bigbuf, bigsize);
3902 umem_free(od, size);
3906 compare_and_update_pbbufs(uint64_t s, bufwad_t *packbuf, bufwad_t *bigbuf,
3907 uint64_t bigsize, uint64_t n, uint64_t chunksize, uint64_t txg)
3915 * For each index from n to n + s, verify that the existing bufwad
3916 * in packobj matches the bufwads at the head and tail of the
3917 * corresponding chunk in bigobj. Then update all three bufwads
3918 * with the new values we want to write out.
3920 for (i = 0; i < s; i++) {
3922 pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
3924 bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
3926 bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
3928 ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
3929 ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
3931 if (pack->bw_txg > txg)
3932 fatal(0, "future leak: got %llx, open txg is %llx",
3935 if (pack->bw_data != 0 && pack->bw_index != n + i)
3936 fatal(0, "wrong index: got %llx, wanted %llx+%llx",
3937 pack->bw_index, n, i);
3939 if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
3940 fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
3942 if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
3943 fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
3945 pack->bw_index = n + i;
3947 pack->bw_data = 1 + ztest_random(-2ULL);
3954 #undef OD_ARRAY_SIZE
3955 #define OD_ARRAY_SIZE 2
3958 ztest_dmu_read_write_zcopy(ztest_ds_t *zd, uint64_t id)
3960 objset_t *os = zd->zd_os;
3967 bufwad_t *packbuf, *bigbuf;
3968 uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
3969 uint64_t blocksize = ztest_random_blocksize();
3970 uint64_t chunksize = blocksize;
3971 uint64_t regions = 997;
3972 uint64_t stride = 123456789ULL;
3974 dmu_buf_t *bonus_db;
3975 arc_buf_t **bigbuf_arcbufs;
3976 dmu_object_info_t doi;
3978 size = sizeof (ztest_od_t) * OD_ARRAY_SIZE;
3979 od = umem_alloc(size, UMEM_NOFAIL);
3982 * This test uses two objects, packobj and bigobj, that are always
3983 * updated together (i.e. in the same tx) so that their contents are
3984 * in sync and can be compared. Their contents relate to each other
3985 * in a simple way: packobj is a dense array of 'bufwad' structures,
3986 * while bigobj is a sparse array of the same bufwads. Specifically,
3987 * for any index n, there are three bufwads that should be identical:
3989 * packobj, at offset n * sizeof (bufwad_t)
3990 * bigobj, at the head of the nth chunk
3991 * bigobj, at the tail of the nth chunk
3993 * The chunk size is set equal to bigobj block size so that
3994 * dmu_assign_arcbuf() can be tested for object updates.
3998 * Read the directory info. If it's the first time, set things up.
4000 ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
4001 ztest_od_init(od + 1, id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize);
4004 if (ztest_object_init(zd, od, size, B_FALSE) != 0) {
4005 umem_free(od, size);
4009 bigobj = od[0].od_object;
4010 packobj = od[1].od_object;
4011 blocksize = od[0].od_blocksize;
4012 chunksize = blocksize;
4013 ASSERT(chunksize == od[1].od_gen);
4015 VERIFY(dmu_object_info(os, bigobj, &doi) == 0);
4016 VERIFY(ISP2(doi.doi_data_block_size));
4017 VERIFY(chunksize == doi.doi_data_block_size);
4018 VERIFY(chunksize >= 2 * sizeof (bufwad_t));
4021 * Pick a random index and compute the offsets into packobj and bigobj.
4023 n = ztest_random(regions) * stride + ztest_random(width);
4024 s = 1 + ztest_random(width - 1);
4026 packoff = n * sizeof (bufwad_t);
4027 packsize = s * sizeof (bufwad_t);
4029 bigoff = n * chunksize;
4030 bigsize = s * chunksize;
4032 packbuf = umem_zalloc(packsize, UMEM_NOFAIL);
4033 bigbuf = umem_zalloc(bigsize, UMEM_NOFAIL);
4035 VERIFY3U(0, ==, dmu_bonus_hold(os, bigobj, FTAG, &bonus_db));
4037 bigbuf_arcbufs = umem_zalloc(2 * s * sizeof (arc_buf_t *), UMEM_NOFAIL);
4040 * Iteration 0 test zcopy for DB_UNCACHED dbufs.
4041 * Iteration 1 test zcopy to already referenced dbufs.
4042 * Iteration 2 test zcopy to dirty dbuf in the same txg.
4043 * Iteration 3 test zcopy to dbuf dirty in previous txg.
4044 * Iteration 4 test zcopy when dbuf is no longer dirty.
4045 * Iteration 5 test zcopy when it can't be done.
4046 * Iteration 6 one more zcopy write.
4048 for (i = 0; i < 7; i++) {
4053 * In iteration 5 (i == 5) use arcbufs
4054 * that don't match bigobj blksz to test
4055 * dmu_assign_arcbuf() when it can't directly
4056 * assign an arcbuf to a dbuf.
4058 for (j = 0; j < s; j++) {
4061 dmu_request_arcbuf(bonus_db, chunksize);
4063 bigbuf_arcbufs[2 * j] =
4064 dmu_request_arcbuf(bonus_db, chunksize / 2);
4065 bigbuf_arcbufs[2 * j + 1] =
4066 dmu_request_arcbuf(bonus_db, chunksize / 2);
4071 * Get a tx for the mods to both packobj and bigobj.
4073 tx = dmu_tx_create(os);
4075 dmu_tx_hold_write(tx, packobj, packoff, packsize);
4076 dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
4078 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4080 umem_free(packbuf, packsize);
4081 umem_free(bigbuf, bigsize);
4082 for (j = 0; j < s; j++) {
4084 dmu_return_arcbuf(bigbuf_arcbufs[j]);
4087 bigbuf_arcbufs[2 * j]);
4089 bigbuf_arcbufs[2 * j + 1]);
4092 umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
4093 umem_free(od, size);
4094 dmu_buf_rele(bonus_db, FTAG);
4099 * 50% of the time don't read objects in the 1st iteration to
4100 * test dmu_assign_arcbuf() for the case when there're no
4101 * existing dbufs for the specified offsets.
4103 if (i != 0 || ztest_random(2) != 0) {
4104 error = dmu_read(os, packobj, packoff,
4105 packsize, packbuf, DMU_READ_PREFETCH);
4107 error = dmu_read(os, bigobj, bigoff, bigsize,
4108 bigbuf, DMU_READ_PREFETCH);
4111 compare_and_update_pbbufs(s, packbuf, bigbuf, bigsize,
4115 * We've verified all the old bufwads, and made new ones.
4116 * Now write them out.
4118 dmu_write(os, packobj, packoff, packsize, packbuf, tx);
4119 if (ztest_opts.zo_verbose >= 7) {
4120 (void) printf("writing offset %llx size %llx"
4122 (u_longlong_t)bigoff,
4123 (u_longlong_t)bigsize,
4126 for (off = bigoff, j = 0; j < s; j++, off += chunksize) {
4129 bcopy((caddr_t)bigbuf + (off - bigoff),
4130 bigbuf_arcbufs[j]->b_data, chunksize);
4132 bcopy((caddr_t)bigbuf + (off - bigoff),
4133 bigbuf_arcbufs[2 * j]->b_data,
4135 bcopy((caddr_t)bigbuf + (off - bigoff) +
4137 bigbuf_arcbufs[2 * j + 1]->b_data,
4142 VERIFY(dmu_buf_hold(os, bigobj, off,
4143 FTAG, &dbt, DMU_READ_NO_PREFETCH) == 0);
4146 dmu_assign_arcbuf(bonus_db, off,
4147 bigbuf_arcbufs[j], tx);
4149 dmu_assign_arcbuf(bonus_db, off,
4150 bigbuf_arcbufs[2 * j], tx);
4151 dmu_assign_arcbuf(bonus_db,
4152 off + chunksize / 2,
4153 bigbuf_arcbufs[2 * j + 1], tx);
4156 dmu_buf_rele(dbt, FTAG);
4162 * Sanity check the stuff we just wrote.
4165 void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
4166 void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
4168 VERIFY(0 == dmu_read(os, packobj, packoff,
4169 packsize, packcheck, DMU_READ_PREFETCH));
4170 VERIFY(0 == dmu_read(os, bigobj, bigoff,
4171 bigsize, bigcheck, DMU_READ_PREFETCH));
4173 ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
4174 ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
4176 umem_free(packcheck, packsize);
4177 umem_free(bigcheck, bigsize);
4180 txg_wait_open(dmu_objset_pool(os), 0);
4181 } else if (i == 3) {
4182 txg_wait_synced(dmu_objset_pool(os), 0);
4186 dmu_buf_rele(bonus_db, FTAG);
4187 umem_free(packbuf, packsize);
4188 umem_free(bigbuf, bigsize);
4189 umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
4190 umem_free(od, size);
4195 ztest_dmu_write_parallel(ztest_ds_t *zd, uint64_t id)
4199 od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
4200 uint64_t offset = (1ULL << (ztest_random(20) + 43)) +
4201 (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
4204 * Have multiple threads write to large offsets in an object
4205 * to verify that parallel writes to an object -- even to the
4206 * same blocks within the object -- doesn't cause any trouble.
4208 ztest_od_init(od, ID_PARALLEL, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
4210 if (ztest_object_init(zd, od, sizeof (ztest_od_t), B_FALSE) != 0)
4213 while (ztest_random(10) != 0)
4214 ztest_io(zd, od->od_object, offset);
4216 umem_free(od, sizeof (ztest_od_t));
4220 ztest_dmu_prealloc(ztest_ds_t *zd, uint64_t id)
4223 uint64_t offset = (1ULL << (ztest_random(4) + SPA_MAXBLOCKSHIFT)) +
4224 (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
4225 uint64_t count = ztest_random(20) + 1;
4226 uint64_t blocksize = ztest_random_blocksize();
4229 od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
4231 ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
4233 if (ztest_object_init(zd, od, sizeof (ztest_od_t),
4234 !ztest_random(2)) != 0) {
4235 umem_free(od, sizeof (ztest_od_t));
4239 if (ztest_truncate(zd, od->od_object, offset, count * blocksize) != 0) {
4240 umem_free(od, sizeof (ztest_od_t));
4244 ztest_prealloc(zd, od->od_object, offset, count * blocksize);
4246 data = umem_zalloc(blocksize, UMEM_NOFAIL);
4248 while (ztest_random(count) != 0) {
4249 uint64_t randoff = offset + (ztest_random(count) * blocksize);
4250 if (ztest_write(zd, od->od_object, randoff, blocksize,
4253 while (ztest_random(4) != 0)
4254 ztest_io(zd, od->od_object, randoff);
4257 umem_free(data, blocksize);
4258 umem_free(od, sizeof (ztest_od_t));
4262 * Verify that zap_{create,destroy,add,remove,update} work as expected.
4264 #define ZTEST_ZAP_MIN_INTS 1
4265 #define ZTEST_ZAP_MAX_INTS 4
4266 #define ZTEST_ZAP_MAX_PROPS 1000
4269 ztest_zap(ztest_ds_t *zd, uint64_t id)
4271 objset_t *os = zd->zd_os;
4274 uint64_t txg, last_txg;
4275 uint64_t value[ZTEST_ZAP_MAX_INTS];
4276 uint64_t zl_ints, zl_intsize, prop;
4279 char propname[100], txgname[100];
4281 char *hc[2] = { "s.acl.h", ".s.open.h.hyLZlg" };
4283 od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
4284 ztest_od_init(od, id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0);
4286 if (ztest_object_init(zd, od, sizeof (ztest_od_t),
4287 !ztest_random(2)) != 0)
4290 object = od->od_object;
4293 * Generate a known hash collision, and verify that
4294 * we can lookup and remove both entries.
4296 tx = dmu_tx_create(os);
4297 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4298 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4301 for (i = 0; i < 2; i++) {
4303 VERIFY3U(0, ==, zap_add(os, object, hc[i], sizeof (uint64_t),
4306 for (i = 0; i < 2; i++) {
4307 VERIFY3U(EEXIST, ==, zap_add(os, object, hc[i],
4308 sizeof (uint64_t), 1, &value[i], tx));
4310 zap_length(os, object, hc[i], &zl_intsize, &zl_ints));
4311 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4312 ASSERT3U(zl_ints, ==, 1);
4314 for (i = 0; i < 2; i++) {
4315 VERIFY3U(0, ==, zap_remove(os, object, hc[i], tx));
4320 * Generate a buch of random entries.
4322 ints = MAX(ZTEST_ZAP_MIN_INTS, object % ZTEST_ZAP_MAX_INTS);
4324 prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
4325 (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
4326 (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
4327 bzero(value, sizeof (value));
4331 * If these zap entries already exist, validate their contents.
4333 error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
4335 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4336 ASSERT3U(zl_ints, ==, 1);
4338 VERIFY(zap_lookup(os, object, txgname, zl_intsize,
4339 zl_ints, &last_txg) == 0);
4341 VERIFY(zap_length(os, object, propname, &zl_intsize,
4344 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4345 ASSERT3U(zl_ints, ==, ints);
4347 VERIFY(zap_lookup(os, object, propname, zl_intsize,
4348 zl_ints, value) == 0);
4350 for (i = 0; i < ints; i++) {
4351 ASSERT3U(value[i], ==, last_txg + object + i);
4354 ASSERT3U(error, ==, ENOENT);
4358 * Atomically update two entries in our zap object.
4359 * The first is named txg_%llu, and contains the txg
4360 * in which the property was last updated. The second
4361 * is named prop_%llu, and the nth element of its value
4362 * should be txg + object + n.
4364 tx = dmu_tx_create(os);
4365 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4366 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4371 fatal(0, "zap future leak: old %llu new %llu", last_txg, txg);
4373 for (i = 0; i < ints; i++)
4374 value[i] = txg + object + i;
4376 VERIFY3U(0, ==, zap_update(os, object, txgname, sizeof (uint64_t),
4378 VERIFY3U(0, ==, zap_update(os, object, propname, sizeof (uint64_t),
4384 * Remove a random pair of entries.
4386 prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
4387 (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
4388 (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
4390 error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
4392 if (error == ENOENT)
4397 tx = dmu_tx_create(os);
4398 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4399 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4402 VERIFY3U(0, ==, zap_remove(os, object, txgname, tx));
4403 VERIFY3U(0, ==, zap_remove(os, object, propname, tx));
4406 umem_free(od, sizeof (ztest_od_t));
4410 * Testcase to test the upgrading of a microzap to fatzap.
4413 ztest_fzap(ztest_ds_t *zd, uint64_t id)
4415 objset_t *os = zd->zd_os;
4417 uint64_t object, txg;
4420 od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
4421 ztest_od_init(od, id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0);
4423 if (ztest_object_init(zd, od, sizeof (ztest_od_t),
4424 !ztest_random(2)) != 0)
4426 object = od->od_object;
4429 * Add entries to this ZAP and make sure it spills over
4430 * and gets upgraded to a fatzap. Also, since we are adding
4431 * 2050 entries we should see ptrtbl growth and leaf-block split.
4433 for (i = 0; i < 2050; i++) {
4434 char name[MAXNAMELEN];
4439 (void) snprintf(name, sizeof (name), "fzap-%llu-%llu",
4440 (u_longlong_t)id, (u_longlong_t)value);
4442 tx = dmu_tx_create(os);
4443 dmu_tx_hold_zap(tx, object, B_TRUE, name);
4444 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4447 error = zap_add(os, object, name, sizeof (uint64_t), 1,
4449 ASSERT(error == 0 || error == EEXIST);
4453 umem_free(od, sizeof (ztest_od_t));
4458 ztest_zap_parallel(ztest_ds_t *zd, uint64_t id)
4460 objset_t *os = zd->zd_os;
4462 uint64_t txg, object, count, wsize, wc, zl_wsize, zl_wc;
4464 int i, namelen, error;
4465 int micro = ztest_random(2);
4466 char name[20], string_value[20];
4469 od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
4470 ztest_od_init(od, ID_PARALLEL, FTAG, micro, DMU_OT_ZAP_OTHER, 0, 0);
4472 if (ztest_object_init(zd, od, sizeof (ztest_od_t), B_FALSE) != 0) {
4473 umem_free(od, sizeof (ztest_od_t));
4477 object = od->od_object;
4480 * Generate a random name of the form 'xxx.....' where each
4481 * x is a random printable character and the dots are dots.
4482 * There are 94 such characters, and the name length goes from
4483 * 6 to 20, so there are 94^3 * 15 = 12,458,760 possible names.
4485 namelen = ztest_random(sizeof (name) - 5) + 5 + 1;
4487 for (i = 0; i < 3; i++)
4488 name[i] = '!' + ztest_random('~' - '!' + 1);
4489 for (; i < namelen - 1; i++)
4493 if ((namelen & 1) || micro) {
4494 wsize = sizeof (txg);
4500 data = string_value;
4504 VERIFY0(zap_count(os, object, &count));
4505 ASSERT(count != -1ULL);
4508 * Select an operation: length, lookup, add, update, remove.
4510 i = ztest_random(5);
4513 tx = dmu_tx_create(os);
4514 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4515 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4518 bcopy(name, string_value, namelen);
4522 bzero(string_value, namelen);
4528 error = zap_length(os, object, name, &zl_wsize, &zl_wc);
4530 ASSERT3U(wsize, ==, zl_wsize);
4531 ASSERT3U(wc, ==, zl_wc);
4533 ASSERT3U(error, ==, ENOENT);
4538 error = zap_lookup(os, object, name, wsize, wc, data);
4540 if (data == string_value &&
4541 bcmp(name, data, namelen) != 0)
4542 fatal(0, "name '%s' != val '%s' len %d",
4543 name, data, namelen);
4545 ASSERT3U(error, ==, ENOENT);
4550 error = zap_add(os, object, name, wsize, wc, data, tx);
4551 ASSERT(error == 0 || error == EEXIST);
4555 VERIFY(zap_update(os, object, name, wsize, wc, data, tx) == 0);
4559 error = zap_remove(os, object, name, tx);
4560 ASSERT(error == 0 || error == ENOENT);
4567 umem_free(od, sizeof (ztest_od_t));
4571 * Commit callback data.
4573 typedef struct ztest_cb_data {
4574 list_node_t zcd_node;
4576 int zcd_expected_err;
4577 boolean_t zcd_added;
4578 boolean_t zcd_called;
4582 /* This is the actual commit callback function */
4584 ztest_commit_callback(void *arg, int error)
4586 ztest_cb_data_t *data = arg;
4587 uint64_t synced_txg;
4589 VERIFY(data != NULL);
4590 VERIFY3S(data->zcd_expected_err, ==, error);
4591 VERIFY(!data->zcd_called);
4593 synced_txg = spa_last_synced_txg(data->zcd_spa);
4594 if (data->zcd_txg > synced_txg)
4595 fatal(0, "commit callback of txg %" PRIu64 " called prematurely"
4596 ", last synced txg = %" PRIu64 "\n", data->zcd_txg,
4599 data->zcd_called = B_TRUE;
4601 if (error == ECANCELED) {
4602 ASSERT0(data->zcd_txg);
4603 ASSERT(!data->zcd_added);
4606 * The private callback data should be destroyed here, but
4607 * since we are going to check the zcd_called field after
4608 * dmu_tx_abort(), we will destroy it there.
4613 ASSERT(data->zcd_added);
4614 ASSERT3U(data->zcd_txg, !=, 0);
4616 (void) mutex_enter(&zcl.zcl_callbacks_lock);
4618 /* See if this cb was called more quickly */
4619 if ((synced_txg - data->zcd_txg) < zc_min_txg_delay)
4620 zc_min_txg_delay = synced_txg - data->zcd_txg;
4622 /* Remove our callback from the list */
4623 list_remove(&zcl.zcl_callbacks, data);
4625 (void) mutex_exit(&zcl.zcl_callbacks_lock);
4627 umem_free(data, sizeof (ztest_cb_data_t));
4630 /* Allocate and initialize callback data structure */
4631 static ztest_cb_data_t *
4632 ztest_create_cb_data(objset_t *os, uint64_t txg)
4634 ztest_cb_data_t *cb_data;
4636 cb_data = umem_zalloc(sizeof (ztest_cb_data_t), UMEM_NOFAIL);
4638 cb_data->zcd_txg = txg;
4639 cb_data->zcd_spa = dmu_objset_spa(os);
4640 list_link_init(&cb_data->zcd_node);
4646 * Commit callback test.
4649 ztest_dmu_commit_callbacks(ztest_ds_t *zd, uint64_t id)
4651 objset_t *os = zd->zd_os;
4654 ztest_cb_data_t *cb_data[3], *tmp_cb;
4655 uint64_t old_txg, txg;
4658 od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
4659 ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
4661 if (ztest_object_init(zd, od, sizeof (ztest_od_t), B_FALSE) != 0) {
4662 umem_free(od, sizeof (ztest_od_t));
4666 tx = dmu_tx_create(os);
4668 cb_data[0] = ztest_create_cb_data(os, 0);
4669 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[0]);
4671 dmu_tx_hold_write(tx, od->od_object, 0, sizeof (uint64_t));
4673 /* Every once in a while, abort the transaction on purpose */
4674 if (ztest_random(100) == 0)
4678 error = dmu_tx_assign(tx, TXG_NOWAIT);
4680 txg = error ? 0 : dmu_tx_get_txg(tx);
4682 cb_data[0]->zcd_txg = txg;
4683 cb_data[1] = ztest_create_cb_data(os, txg);
4684 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[1]);
4688 * It's not a strict requirement to call the registered
4689 * callbacks from inside dmu_tx_abort(), but that's what
4690 * it's supposed to happen in the current implementation
4691 * so we will check for that.
4693 for (i = 0; i < 2; i++) {
4694 cb_data[i]->zcd_expected_err = ECANCELED;
4695 VERIFY(!cb_data[i]->zcd_called);
4700 for (i = 0; i < 2; i++) {
4701 VERIFY(cb_data[i]->zcd_called);
4702 umem_free(cb_data[i], sizeof (ztest_cb_data_t));
4705 umem_free(od, sizeof (ztest_od_t));
4709 cb_data[2] = ztest_create_cb_data(os, txg);
4710 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[2]);
4713 * Read existing data to make sure there isn't a future leak.
4715 VERIFY(0 == dmu_read(os, od->od_object, 0, sizeof (uint64_t),
4716 &old_txg, DMU_READ_PREFETCH));
4719 fatal(0, "future leak: got %" PRIu64 ", open txg is %" PRIu64,
4722 dmu_write(os, od->od_object, 0, sizeof (uint64_t), &txg, tx);
4724 (void) mutex_enter(&zcl.zcl_callbacks_lock);
4727 * Since commit callbacks don't have any ordering requirement and since
4728 * it is theoretically possible for a commit callback to be called
4729 * after an arbitrary amount of time has elapsed since its txg has been
4730 * synced, it is difficult to reliably determine whether a commit
4731 * callback hasn't been called due to high load or due to a flawed
4734 * In practice, we will assume that if after a certain number of txgs a
4735 * commit callback hasn't been called, then most likely there's an
4736 * implementation bug..
4738 tmp_cb = list_head(&zcl.zcl_callbacks);
4739 if (tmp_cb != NULL &&
4740 tmp_cb->zcd_txg + ZTEST_COMMIT_CB_THRESH < txg) {
4741 fatal(0, "Commit callback threshold exceeded, oldest txg: %"
4742 PRIu64 ", open txg: %" PRIu64 "\n", tmp_cb->zcd_txg, txg);
4746 * Let's find the place to insert our callbacks.
4748 * Even though the list is ordered by txg, it is possible for the
4749 * insertion point to not be the end because our txg may already be
4750 * quiescing at this point and other callbacks in the open txg
4751 * (from other objsets) may have sneaked in.
4753 tmp_cb = list_tail(&zcl.zcl_callbacks);
4754 while (tmp_cb != NULL && tmp_cb->zcd_txg > txg)
4755 tmp_cb = list_prev(&zcl.zcl_callbacks, tmp_cb);
4757 /* Add the 3 callbacks to the list */
4758 for (i = 0; i < 3; i++) {
4760 list_insert_head(&zcl.zcl_callbacks, cb_data[i]);
4762 list_insert_after(&zcl.zcl_callbacks, tmp_cb,
4765 cb_data[i]->zcd_added = B_TRUE;
4766 VERIFY(!cb_data[i]->zcd_called);
4768 tmp_cb = cb_data[i];
4773 (void) mutex_exit(&zcl.zcl_callbacks_lock);
4777 umem_free(od, sizeof (ztest_od_t));
4782 ztest_dsl_prop_get_set(ztest_ds_t *zd, uint64_t id)
4784 zfs_prop_t proplist[] = {
4786 ZFS_PROP_COMPRESSION,
4792 (void) rw_rdlock(&ztest_name_lock);
4794 for (p = 0; p < sizeof (proplist) / sizeof (proplist[0]); p++)
4795 (void) ztest_dsl_prop_set_uint64(zd->zd_name, proplist[p],
4796 ztest_random_dsl_prop(proplist[p]), (int)ztest_random(2));
4798 (void) rw_unlock(&ztest_name_lock);
4803 ztest_spa_prop_get_set(ztest_ds_t *zd, uint64_t id)
4805 nvlist_t *props = NULL;
4807 (void) rw_rdlock(&ztest_name_lock);
4809 (void) ztest_spa_prop_set_uint64(ZPOOL_PROP_DEDUPDITTO,
4810 ZIO_DEDUPDITTO_MIN + ztest_random(ZIO_DEDUPDITTO_MIN));
4812 VERIFY0(spa_prop_get(ztest_spa, &props));
4814 if (ztest_opts.zo_verbose >= 6)
4815 dump_nvlist(props, 4);
4819 (void) rw_unlock(&ztest_name_lock);
4823 user_release_one(const char *snapname, const char *holdname)
4825 nvlist_t *snaps, *holds;
4828 snaps = fnvlist_alloc();
4829 holds = fnvlist_alloc();
4830 fnvlist_add_boolean(holds, holdname);
4831 fnvlist_add_nvlist(snaps, snapname, holds);
4832 fnvlist_free(holds);
4833 error = dsl_dataset_user_release(snaps, NULL);
4834 fnvlist_free(snaps);
4839 * Test snapshot hold/release and deferred destroy.
4842 ztest_dmu_snapshot_hold(ztest_ds_t *zd, uint64_t id)
4845 objset_t *os = zd->zd_os;
4849 char clonename[100];
4851 char osname[MAXNAMELEN];
4854 (void) rw_rdlock(&ztest_name_lock);
4856 dmu_objset_name(os, osname);
4858 (void) snprintf(snapname, sizeof (snapname), "sh1_%llu",
4860 (void) snprintf(fullname, sizeof (fullname), "%s@%s", osname, snapname);
4861 (void) snprintf(clonename, sizeof (clonename),
4862 "%s/ch1_%llu", osname, (u_longlong_t)id);
4863 (void) snprintf(tag, sizeof (tag), "tag_%llu", (u_longlong_t)id);
4866 * Clean up from any previous run.
4868 error = dsl_destroy_head(clonename);
4869 if (error != ENOENT)
4871 error = user_release_one(fullname, tag);
4872 if (error != ESRCH && error != ENOENT)
4874 error = dsl_destroy_snapshot(fullname, B_FALSE);
4875 if (error != ENOENT)
4879 * Create snapshot, clone it, mark snap for deferred destroy,
4880 * destroy clone, verify snap was also destroyed.
4882 error = dmu_objset_snapshot_one(osname, snapname);
4884 if (error == ENOSPC) {
4885 ztest_record_enospc("dmu_objset_snapshot");
4888 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
4891 error = dmu_objset_clone(clonename, fullname);
4893 if (error == ENOSPC) {
4894 ztest_record_enospc("dmu_objset_clone");
4897 fatal(0, "dmu_objset_clone(%s) = %d", clonename, error);
4900 error = dsl_destroy_snapshot(fullname, B_TRUE);
4902 fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
4906 error = dsl_destroy_head(clonename);
4908 fatal(0, "dsl_destroy_head(%s) = %d", clonename, error);
4910 error = dmu_objset_hold(fullname, FTAG, &origin);
4911 if (error != ENOENT)
4912 fatal(0, "dmu_objset_hold(%s) = %d", fullname, error);
4915 * Create snapshot, add temporary hold, verify that we can't
4916 * destroy a held snapshot, mark for deferred destroy,
4917 * release hold, verify snapshot was destroyed.
4919 error = dmu_objset_snapshot_one(osname, snapname);
4921 if (error == ENOSPC) {
4922 ztest_record_enospc("dmu_objset_snapshot");
4925 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
4928 holds = fnvlist_alloc();
4929 fnvlist_add_string(holds, fullname, tag);
4930 error = dsl_dataset_user_hold(holds, 0, NULL);
4931 fnvlist_free(holds);
4933 if (error == ENOSPC) {
4934 ztest_record_enospc("dsl_dataset_user_hold");
4937 fatal(0, "dsl_dataset_user_hold(%s, %s) = %u",
4938 fullname, tag, error);
4941 error = dsl_destroy_snapshot(fullname, B_FALSE);
4942 if (error != EBUSY) {
4943 fatal(0, "dsl_destroy_snapshot(%s, B_FALSE) = %d",
4947 error = dsl_destroy_snapshot(fullname, B_TRUE);
4949 fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
4953 error = user_release_one(fullname, tag);
4955 fatal(0, "user_release_one(%s, %s) = %d", fullname, tag, error);
4957 VERIFY3U(dmu_objset_hold(fullname, FTAG, &origin), ==, ENOENT);
4960 (void) rw_unlock(&ztest_name_lock);
4964 * Inject random faults into the on-disk data.
4968 ztest_fault_inject(ztest_ds_t *zd, uint64_t id)
4970 ztest_shared_t *zs = ztest_shared;
4971 spa_t *spa = ztest_spa;
4975 uint64_t bad = 0x1990c0ffeedecadeull;
4980 int bshift = SPA_OLD_MAXBLOCKSHIFT + 2; /* don't scrog all labels */
4986 boolean_t islog = B_FALSE;
4988 path0 = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
4989 pathrand = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
4991 mutex_enter(&ztest_vdev_lock);
4992 maxfaults = MAXFAULTS();
4993 leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz;
4994 mirror_save = zs->zs_mirrors;
4995 mutex_exit(&ztest_vdev_lock);
4997 ASSERT(leaves >= 1);
5000 * Grab the name lock as reader. There are some operations
5001 * which don't like to have their vdevs changed while
5002 * they are in progress (i.e. spa_change_guid). Those
5003 * operations will have grabbed the name lock as writer.
5005 (void) rw_rdlock(&ztest_name_lock);
5008 * We need SCL_STATE here because we're going to look at vd0->vdev_tsd.
5010 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5012 if (ztest_random(2) == 0) {
5014 * Inject errors on a normal data device or slog device.
5016 top = ztest_random_vdev_top(spa, B_TRUE);
5017 leaf = ztest_random(leaves) + zs->zs_splits;
5020 * Generate paths to the first leaf in this top-level vdev,
5021 * and to the random leaf we selected. We'll induce transient
5022 * write failures and random online/offline activity on leaf 0,
5023 * and we'll write random garbage to the randomly chosen leaf.
5025 (void) snprintf(path0, MAXPATHLEN, ztest_dev_template,
5026 ztest_opts.zo_dir, ztest_opts.zo_pool,
5027 top * leaves + zs->zs_splits);
5028 (void) snprintf(pathrand, MAXPATHLEN, ztest_dev_template,
5029 ztest_opts.zo_dir, ztest_opts.zo_pool,
5030 top * leaves + leaf);
5032 vd0 = vdev_lookup_by_path(spa->spa_root_vdev, path0);
5033 if (vd0 != NULL && vd0->vdev_top->vdev_islog)
5037 * If the top-level vdev needs to be resilvered
5038 * then we only allow faults on the device that is
5041 if (vd0 != NULL && maxfaults != 1 &&
5042 (!vdev_resilver_needed(vd0->vdev_top, NULL, NULL) ||
5043 vd0->vdev_resilver_txg != 0)) {
5045 * Make vd0 explicitly claim to be unreadable,
5046 * or unwriteable, or reach behind its back
5047 * and close the underlying fd. We can do this if
5048 * maxfaults == 0 because we'll fail and reexecute,
5049 * and we can do it if maxfaults >= 2 because we'll
5050 * have enough redundancy. If maxfaults == 1, the
5051 * combination of this with injection of random data
5052 * corruption below exceeds the pool's fault tolerance.
5054 vdev_file_t *vf = vd0->vdev_tsd;
5056 if (vf != NULL && ztest_random(3) == 0) {
5057 (void) close(vf->vf_vnode->v_fd);
5058 vf->vf_vnode->v_fd = -1;
5059 } else if (ztest_random(2) == 0) {
5060 vd0->vdev_cant_read = B_TRUE;
5062 vd0->vdev_cant_write = B_TRUE;
5064 guid0 = vd0->vdev_guid;
5068 * Inject errors on an l2cache device.
5070 spa_aux_vdev_t *sav = &spa->spa_l2cache;
5072 if (sav->sav_count == 0) {
5073 spa_config_exit(spa, SCL_STATE, FTAG);
5074 (void) rw_unlock(&ztest_name_lock);
5077 vd0 = sav->sav_vdevs[ztest_random(sav->sav_count)];
5078 guid0 = vd0->vdev_guid;
5079 (void) strcpy(path0, vd0->vdev_path);
5080 (void) strcpy(pathrand, vd0->vdev_path);
5084 maxfaults = INT_MAX; /* no limit on cache devices */
5087 spa_config_exit(spa, SCL_STATE, FTAG);
5088 (void) rw_unlock(&ztest_name_lock);
5091 * If we can tolerate two or more faults, or we're dealing
5092 * with a slog, randomly online/offline vd0.
5094 if ((maxfaults >= 2 || islog) && guid0 != 0) {
5095 if (ztest_random(10) < 6) {
5096 int flags = (ztest_random(2) == 0 ?
5097 ZFS_OFFLINE_TEMPORARY : 0);
5100 * We have to grab the zs_name_lock as writer to
5101 * prevent a race between offlining a slog and
5102 * destroying a dataset. Offlining the slog will
5103 * grab a reference on the dataset which may cause
5104 * dsl_destroy_head() to fail with EBUSY thus
5105 * leaving the dataset in an inconsistent state.
5108 (void) rw_wrlock(&ztest_name_lock);
5110 VERIFY(vdev_offline(spa, guid0, flags) != EBUSY);
5113 (void) rw_unlock(&ztest_name_lock);
5116 * Ideally we would like to be able to randomly
5117 * call vdev_[on|off]line without holding locks
5118 * to force unpredictable failures but the side
5119 * effects of vdev_[on|off]line prevent us from
5120 * doing so. We grab the ztest_vdev_lock here to
5121 * prevent a race between injection testing and
5124 mutex_enter(&ztest_vdev_lock);
5125 (void) vdev_online(spa, guid0, 0, NULL);
5126 mutex_exit(&ztest_vdev_lock);
5134 * We have at least single-fault tolerance, so inject data corruption.
5136 fd = open(pathrand, O_RDWR);
5138 if (fd == -1) /* we hit a gap in the device namespace */
5141 fsize = lseek(fd, 0, SEEK_END);
5143 while (--iters != 0) {
5144 offset = ztest_random(fsize / (leaves << bshift)) *
5145 (leaves << bshift) + (leaf << bshift) +
5146 (ztest_random(1ULL << (bshift - 1)) & -8ULL);
5148 if (offset >= fsize)
5151 mutex_enter(&ztest_vdev_lock);
5152 if (mirror_save != zs->zs_mirrors) {
5153 mutex_exit(&ztest_vdev_lock);
5158 if (pwrite(fd, &bad, sizeof (bad), offset) != sizeof (bad))
5159 fatal(1, "can't inject bad word at 0x%llx in %s",
5162 mutex_exit(&ztest_vdev_lock);
5164 if (ztest_opts.zo_verbose >= 7)
5165 (void) printf("injected bad word into %s,"
5166 " offset 0x%llx\n", pathrand, (u_longlong_t)offset);
5171 umem_free(path0, MAXPATHLEN);
5172 umem_free(pathrand, MAXPATHLEN);
5176 * Verify that DDT repair works as expected.
5179 ztest_ddt_repair(ztest_ds_t *zd, uint64_t id)
5181 ztest_shared_t *zs = ztest_shared;
5182 spa_t *spa = ztest_spa;
5183 objset_t *os = zd->zd_os;
5185 uint64_t object, blocksize, txg, pattern, psize;
5186 enum zio_checksum checksum = spa_dedup_checksum(spa);
5191 int copies = 2 * ZIO_DEDUPDITTO_MIN;
5194 blocksize = ztest_random_blocksize();
5195 blocksize = MIN(blocksize, 2048); /* because we write so many */
5197 od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
5198 ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
5200 if (ztest_object_init(zd, od, sizeof (ztest_od_t), B_FALSE) != 0) {
5201 umem_free(od, sizeof (ztest_od_t));
5206 * Take the name lock as writer to prevent anyone else from changing
5207 * the pool and dataset properies we need to maintain during this test.
5209 (void) rw_wrlock(&ztest_name_lock);
5211 if (ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_DEDUP, checksum,
5213 ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_COPIES, 1,
5215 (void) rw_unlock(&ztest_name_lock);
5216 umem_free(od, sizeof (ztest_od_t));
5220 object = od[0].od_object;
5221 blocksize = od[0].od_blocksize;
5222 pattern = zs->zs_guid ^ dmu_objset_fsid_guid(os);
5224 ASSERT(object != 0);
5226 tx = dmu_tx_create(os);
5227 dmu_tx_hold_write(tx, object, 0, copies * blocksize);
5228 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
5230 (void) rw_unlock(&ztest_name_lock);
5231 umem_free(od, sizeof (ztest_od_t));
5236 * Write all the copies of our block.
5238 for (i = 0; i < copies; i++) {
5239 uint64_t offset = i * blocksize;
5240 int error = dmu_buf_hold(os, object, offset, FTAG, &db,
5241 DMU_READ_NO_PREFETCH);
5243 fatal(B_FALSE, "dmu_buf_hold(%p, %llu, %llu) = %u",
5244 os, (long long)object, (long long) offset, error);
5246 ASSERT(db->db_offset == offset);
5247 ASSERT(db->db_size == blocksize);
5248 ASSERT(ztest_pattern_match(db->db_data, db->db_size, pattern) ||
5249 ztest_pattern_match(db->db_data, db->db_size, 0ULL));
5250 dmu_buf_will_fill(db, tx);
5251 ztest_pattern_set(db->db_data, db->db_size, pattern);
5252 dmu_buf_rele(db, FTAG);
5256 txg_wait_synced(spa_get_dsl(spa), txg);
5259 * Find out what block we got.
5261 VERIFY0(dmu_buf_hold(os, object, 0, FTAG, &db,
5262 DMU_READ_NO_PREFETCH));
5263 blk = *((dmu_buf_impl_t *)db)->db_blkptr;
5264 dmu_buf_rele(db, FTAG);
5267 * Damage the block. Dedup-ditto will save us when we read it later.
5269 psize = BP_GET_PSIZE(&blk);
5270 buf = zio_buf_alloc(psize);
5271 ztest_pattern_set(buf, psize, ~pattern);
5273 (void) zio_wait(zio_rewrite(NULL, spa, 0, &blk,
5274 buf, psize, NULL, NULL, ZIO_PRIORITY_SYNC_WRITE,
5275 ZIO_FLAG_CANFAIL | ZIO_FLAG_INDUCE_DAMAGE, NULL));
5277 zio_buf_free(buf, psize);
5279 (void) rw_unlock(&ztest_name_lock);
5280 umem_free(od, sizeof (ztest_od_t));
5288 ztest_scrub(ztest_ds_t *zd, uint64_t id)
5290 spa_t *spa = ztest_spa;
5292 (void) spa_scan(spa, POOL_SCAN_SCRUB);
5293 (void) poll(NULL, 0, 100); /* wait a moment, then force a restart */
5294 (void) spa_scan(spa, POOL_SCAN_SCRUB);
5298 * Change the guid for the pool.
5302 ztest_reguid(ztest_ds_t *zd, uint64_t id)
5304 spa_t *spa = ztest_spa;
5305 uint64_t orig, load;
5308 orig = spa_guid(spa);
5309 load = spa_load_guid(spa);
5311 (void) rw_wrlock(&ztest_name_lock);
5312 error = spa_change_guid(spa);
5313 (void) rw_unlock(&ztest_name_lock);
5318 if (ztest_opts.zo_verbose >= 4) {
5319 (void) printf("Changed guid old %llu -> %llu\n",
5320 (u_longlong_t)orig, (u_longlong_t)spa_guid(spa));
5323 VERIFY3U(orig, !=, spa_guid(spa));
5324 VERIFY3U(load, ==, spa_load_guid(spa));
5328 * Rename the pool to a different name and then rename it back.
5332 ztest_spa_rename(ztest_ds_t *zd, uint64_t id)
5334 char *oldname, *newname;
5337 (void) rw_wrlock(&ztest_name_lock);
5339 oldname = ztest_opts.zo_pool;
5340 newname = umem_alloc(strlen(oldname) + 5, UMEM_NOFAIL);
5341 (void) strcpy(newname, oldname);
5342 (void) strcat(newname, "_tmp");
5347 VERIFY3U(0, ==, spa_rename(oldname, newname));
5350 * Try to open it under the old name, which shouldn't exist
5352 VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
5355 * Open it under the new name and make sure it's still the same spa_t.
5357 VERIFY3U(0, ==, spa_open(newname, &spa, FTAG));
5359 ASSERT(spa == ztest_spa);
5360 spa_close(spa, FTAG);
5363 * Rename it back to the original
5365 VERIFY3U(0, ==, spa_rename(newname, oldname));
5368 * Make sure it can still be opened
5370 VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG));
5372 ASSERT(spa == ztest_spa);
5373 spa_close(spa, FTAG);
5375 umem_free(newname, strlen(newname) + 1);
5377 (void) rw_unlock(&ztest_name_lock);
5381 * Verify pool integrity by running zdb.
5384 ztest_run_zdb(char *pool)
5392 bin = umem_alloc(MAXPATHLEN + MAXNAMELEN + 20, UMEM_NOFAIL);
5393 zdb = umem_alloc(MAXPATHLEN + MAXNAMELEN + 20, UMEM_NOFAIL);
5394 zbuf = umem_alloc(1024, UMEM_NOFAIL);
5396 VERIFY(realpath(getexecname(), bin) != NULL);
5397 if (strncmp(bin, "/usr/sbin/ztest", 15) == 0) {
5398 strcpy(bin, "/usr/sbin/zdb"); /* Installed */
5399 } else if (strncmp(bin, "/sbin/ztest", 11) == 0) {
5400 strcpy(bin, "/sbin/zdb"); /* Installed */
5402 strstr(bin, "/ztest/")[0] = '\0'; /* In-tree */
5403 strcat(bin, "/zdb/zdb");
5407 "%s -bcc%s%s -d -U %s %s",
5409 ztest_opts.zo_verbose >= 3 ? "s" : "",
5410 ztest_opts.zo_verbose >= 4 ? "v" : "",
5414 if (ztest_opts.zo_verbose >= 5)
5415 (void) printf("Executing %s\n", strstr(zdb, "zdb "));
5417 fp = popen(zdb, "r");
5419 while (fgets(zbuf, 1024, fp) != NULL)
5420 if (ztest_opts.zo_verbose >= 3)
5421 (void) printf("%s", zbuf);
5423 status = pclose(fp);
5428 ztest_dump_core = 0;
5429 if (WIFEXITED(status))
5430 fatal(0, "'%s' exit code %d", zdb, WEXITSTATUS(status));
5432 fatal(0, "'%s' died with signal %d", zdb, WTERMSIG(status));
5434 umem_free(bin, MAXPATHLEN + MAXNAMELEN + 20);
5435 umem_free(zdb, MAXPATHLEN + MAXNAMELEN + 20);
5436 umem_free(zbuf, 1024);
5440 ztest_walk_pool_directory(char *header)
5444 if (ztest_opts.zo_verbose >= 6)
5445 (void) printf("%s\n", header);
5447 mutex_enter(&spa_namespace_lock);
5448 while ((spa = spa_next(spa)) != NULL)
5449 if (ztest_opts.zo_verbose >= 6)
5450 (void) printf("\t%s\n", spa_name(spa));
5451 mutex_exit(&spa_namespace_lock);
5455 ztest_spa_import_export(char *oldname, char *newname)
5457 nvlist_t *config, *newconfig;
5462 if (ztest_opts.zo_verbose >= 4) {
5463 (void) printf("import/export: old = %s, new = %s\n",
5468 * Clean up from previous runs.
5470 (void) spa_destroy(newname);
5473 * Get the pool's configuration and guid.
5475 VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG));
5478 * Kick off a scrub to tickle scrub/export races.
5480 if (ztest_random(2) == 0)
5481 (void) spa_scan(spa, POOL_SCAN_SCRUB);
5483 pool_guid = spa_guid(spa);
5484 spa_close(spa, FTAG);
5486 ztest_walk_pool_directory("pools before export");
5491 VERIFY3U(0, ==, spa_export(oldname, &config, B_FALSE, B_FALSE));
5493 ztest_walk_pool_directory("pools after export");
5498 newconfig = spa_tryimport(config);
5499 ASSERT(newconfig != NULL);
5500 nvlist_free(newconfig);
5503 * Import it under the new name.
5505 error = spa_import(newname, config, NULL, 0);
5507 dump_nvlist(config, 0);
5508 fatal(B_FALSE, "couldn't import pool %s as %s: error %u",
5509 oldname, newname, error);
5512 ztest_walk_pool_directory("pools after import");
5515 * Try to import it again -- should fail with EEXIST.
5517 VERIFY3U(EEXIST, ==, spa_import(newname, config, NULL, 0));
5520 * Try to import it under a different name -- should fail with EEXIST.
5522 VERIFY3U(EEXIST, ==, spa_import(oldname, config, NULL, 0));
5525 * Verify that the pool is no longer visible under the old name.
5527 VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
5530 * Verify that we can open and close the pool using the new name.
5532 VERIFY3U(0, ==, spa_open(newname, &spa, FTAG));
5533 ASSERT(pool_guid == spa_guid(spa));
5534 spa_close(spa, FTAG);
5536 nvlist_free(config);
5540 ztest_resume(spa_t *spa)
5542 if (spa_suspended(spa) && ztest_opts.zo_verbose >= 6)
5543 (void) printf("resuming from suspended state\n");
5544 spa_vdev_state_enter(spa, SCL_NONE);
5545 vdev_clear(spa, NULL);
5546 (void) spa_vdev_state_exit(spa, NULL, 0);
5547 (void) zio_resume(spa);
5551 ztest_resume_thread(void *arg)
5555 while (!ztest_exiting) {
5556 if (spa_suspended(spa))
5558 (void) poll(NULL, 0, 100);
5570 ztest_deadman_alarm(int sig)
5572 fatal(0, "failed to complete within %d seconds of deadline", GRACE);
5577 ztest_execute(int test, ztest_info_t *zi, uint64_t id)
5579 ztest_ds_t *zd = &ztest_ds[id % ztest_opts.zo_datasets];
5580 ztest_shared_callstate_t *zc = ZTEST_GET_SHARED_CALLSTATE(test);
5581 hrtime_t functime = gethrtime();
5584 for (i = 0; i < zi->zi_iters; i++)
5585 zi->zi_func(zd, id);
5587 functime = gethrtime() - functime;
5589 atomic_add_64(&zc->zc_count, 1);
5590 atomic_add_64(&zc->zc_time, functime);
5592 if (ztest_opts.zo_verbose >= 4)
5593 (void) printf("%6.2f sec in %s\n",
5594 (double)functime / NANOSEC, zi->zi_funcname);
5598 ztest_thread(void *arg)
5601 uint64_t id = (uintptr_t)arg;
5602 ztest_shared_t *zs = ztest_shared;
5606 ztest_shared_callstate_t *zc;
5608 while ((now = gethrtime()) < zs->zs_thread_stop) {
5610 * See if it's time to force a crash.
5612 if (now > zs->zs_thread_kill)
5616 * If we're getting ENOSPC with some regularity, stop.
5618 if (zs->zs_enospc_count > 10)
5622 * Pick a random function to execute.
5624 rand = ztest_random(ZTEST_FUNCS);
5625 zi = &ztest_info[rand];
5626 zc = ZTEST_GET_SHARED_CALLSTATE(rand);
5627 call_next = zc->zc_next;
5629 if (now >= call_next &&
5630 atomic_cas_64(&zc->zc_next, call_next, call_next +
5631 ztest_random(2 * zi->zi_interval[0] + 1)) == call_next) {
5632 ztest_execute(rand, zi, id);
5642 ztest_dataset_name(char *dsname, char *pool, int d)
5644 (void) snprintf(dsname, MAXNAMELEN, "%s/ds_%d", pool, d);
5648 ztest_dataset_destroy(int d)
5650 char name[MAXNAMELEN];
5653 ztest_dataset_name(name, ztest_opts.zo_pool, d);
5655 if (ztest_opts.zo_verbose >= 3)
5656 (void) printf("Destroying %s to free up space\n", name);
5659 * Cleanup any non-standard clones and snapshots. In general,
5660 * ztest thread t operates on dataset (t % zopt_datasets),
5661 * so there may be more than one thing to clean up.
5663 for (t = d; t < ztest_opts.zo_threads;
5664 t += ztest_opts.zo_datasets)
5665 ztest_dsl_dataset_cleanup(name, t);
5667 (void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
5668 DS_FIND_SNAPSHOTS | DS_FIND_CHILDREN);
5672 ztest_dataset_dirobj_verify(ztest_ds_t *zd)
5674 uint64_t usedobjs, dirobjs, scratch;
5677 * ZTEST_DIROBJ is the object directory for the entire dataset.
5678 * Therefore, the number of objects in use should equal the
5679 * number of ZTEST_DIROBJ entries, +1 for ZTEST_DIROBJ itself.
5680 * If not, we have an object leak.
5682 * Note that we can only check this in ztest_dataset_open(),
5683 * when the open-context and syncing-context values agree.
5684 * That's because zap_count() returns the open-context value,
5685 * while dmu_objset_space() returns the rootbp fill count.
5687 VERIFY3U(0, ==, zap_count(zd->zd_os, ZTEST_DIROBJ, &dirobjs));
5688 dmu_objset_space(zd->zd_os, &scratch, &scratch, &usedobjs, &scratch);
5689 ASSERT3U(dirobjs + 1, ==, usedobjs);
5693 ztest_dataset_open(int d)
5695 ztest_ds_t *zd = &ztest_ds[d];
5696 uint64_t committed_seq = ZTEST_GET_SHARED_DS(d)->zd_seq;
5699 char name[MAXNAMELEN];
5702 ztest_dataset_name(name, ztest_opts.zo_pool, d);
5704 (void) rw_rdlock(&ztest_name_lock);
5706 error = ztest_dataset_create(name);
5707 if (error == ENOSPC) {
5708 (void) rw_unlock(&ztest_name_lock);
5709 ztest_record_enospc(FTAG);
5712 ASSERT(error == 0 || error == EEXIST);
5714 VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, zd, &os));
5715 (void) rw_unlock(&ztest_name_lock);
5717 ztest_zd_init(zd, ZTEST_GET_SHARED_DS(d), os);
5719 zilog = zd->zd_zilog;
5721 if (zilog->zl_header->zh_claim_lr_seq != 0 &&
5722 zilog->zl_header->zh_claim_lr_seq < committed_seq)
5723 fatal(0, "missing log records: claimed %llu < committed %llu",
5724 zilog->zl_header->zh_claim_lr_seq, committed_seq);
5726 ztest_dataset_dirobj_verify(zd);
5728 zil_replay(os, zd, ztest_replay_vector);
5730 ztest_dataset_dirobj_verify(zd);
5732 if (ztest_opts.zo_verbose >= 6)
5733 (void) printf("%s replay %llu blocks, %llu records, seq %llu\n",
5735 (u_longlong_t)zilog->zl_parse_blk_count,
5736 (u_longlong_t)zilog->zl_parse_lr_count,
5737 (u_longlong_t)zilog->zl_replaying_seq);
5739 zilog = zil_open(os, ztest_get_data);
5741 if (zilog->zl_replaying_seq != 0 &&
5742 zilog->zl_replaying_seq < committed_seq)
5743 fatal(0, "missing log records: replayed %llu < committed %llu",
5744 zilog->zl_replaying_seq, committed_seq);
5750 ztest_dataset_close(int d)
5752 ztest_ds_t *zd = &ztest_ds[d];
5754 zil_close(zd->zd_zilog);
5755 dmu_objset_disown(zd->zd_os, zd);
5761 * Kick off threads to run tests on all datasets in parallel.
5764 ztest_run(ztest_shared_t *zs)
5769 kthread_t *resume_thread;
5774 ztest_exiting = B_FALSE;
5777 * Initialize parent/child shared state.
5779 mutex_init(&ztest_vdev_lock, NULL, MUTEX_DEFAULT, NULL);
5780 VERIFY(rwlock_init(&ztest_name_lock, USYNC_THREAD, NULL) == 0);
5782 zs->zs_thread_start = gethrtime();
5783 zs->zs_thread_stop =
5784 zs->zs_thread_start + ztest_opts.zo_passtime * NANOSEC;
5785 zs->zs_thread_stop = MIN(zs->zs_thread_stop, zs->zs_proc_stop);
5786 zs->zs_thread_kill = zs->zs_thread_stop;
5787 if (ztest_random(100) < ztest_opts.zo_killrate) {
5788 zs->zs_thread_kill -=
5789 ztest_random(ztest_opts.zo_passtime * NANOSEC);
5792 mutex_init(&zcl.zcl_callbacks_lock, NULL, MUTEX_DEFAULT, NULL);
5794 list_create(&zcl.zcl_callbacks, sizeof (ztest_cb_data_t),
5795 offsetof(ztest_cb_data_t, zcd_node));
5800 kernel_init(FREAD | FWRITE);
5801 VERIFY0(spa_open(ztest_opts.zo_pool, &spa, FTAG));
5802 spa->spa_debug = B_TRUE;
5803 metaslab_preload_limit = ztest_random(20) + 1;
5806 VERIFY0(dmu_objset_own(ztest_opts.zo_pool,
5807 DMU_OST_ANY, B_TRUE, FTAG, &os));
5808 zs->zs_guid = dmu_objset_fsid_guid(os);
5809 dmu_objset_disown(os, FTAG);
5811 spa->spa_dedup_ditto = 2 * ZIO_DEDUPDITTO_MIN;
5814 * We don't expect the pool to suspend unless maxfaults == 0,
5815 * in which case ztest_fault_inject() temporarily takes away
5816 * the only valid replica.
5818 if (MAXFAULTS() == 0)
5819 spa->spa_failmode = ZIO_FAILURE_MODE_WAIT;
5821 spa->spa_failmode = ZIO_FAILURE_MODE_PANIC;
5824 * Create a thread to periodically resume suspended I/O.
5826 VERIFY3P((resume_thread = zk_thread_create(NULL, 0,
5827 (thread_func_t)ztest_resume_thread, spa, TS_RUN, NULL, 0, 0,
5828 PTHREAD_CREATE_JOINABLE)), !=, NULL);
5832 * Set a deadman alarm to abort() if we hang.
5834 signal(SIGALRM, ztest_deadman_alarm);
5835 alarm((zs->zs_thread_stop - zs->zs_thread_start) / NANOSEC + GRACE);
5839 * Verify that we can safely inquire about about any object,
5840 * whether it's allocated or not. To make it interesting,
5841 * we probe a 5-wide window around each power of two.
5842 * This hits all edge cases, including zero and the max.
5844 for (t = 0; t < 64; t++) {
5845 for (d = -5; d <= 5; d++) {
5846 error = dmu_object_info(spa->spa_meta_objset,
5847 (1ULL << t) + d, NULL);
5848 ASSERT(error == 0 || error == ENOENT ||
5854 * If we got any ENOSPC errors on the previous run, destroy something.
5856 if (zs->zs_enospc_count != 0) {
5857 int d = ztest_random(ztest_opts.zo_datasets);
5858 ztest_dataset_destroy(d);
5860 zs->zs_enospc_count = 0;
5862 tid = umem_zalloc(ztest_opts.zo_threads * sizeof (kt_did_t),
5865 if (ztest_opts.zo_verbose >= 4)
5866 (void) printf("starting main threads...\n");
5869 * Kick off all the tests that run in parallel.
5871 for (t = 0; t < ztest_opts.zo_threads; t++) {
5874 if (t < ztest_opts.zo_datasets &&
5875 ztest_dataset_open(t) != 0)
5878 VERIFY3P(thread = zk_thread_create(NULL, 0,
5879 (thread_func_t)ztest_thread,
5880 (void *)(uintptr_t)t, TS_RUN, NULL, 0, 0,
5881 PTHREAD_CREATE_JOINABLE), !=, NULL);
5882 tid[t] = thread->t_tid;
5886 * Wait for all of the tests to complete. We go in reverse order
5887 * so we don't close datasets while threads are still using them.
5889 for (t = ztest_opts.zo_threads - 1; t >= 0; t--) {
5890 thread_join(tid[t]);
5891 if (t < ztest_opts.zo_datasets)
5892 ztest_dataset_close(t);
5895 txg_wait_synced(spa_get_dsl(spa), 0);
5897 zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(spa));
5898 zs->zs_space = metaslab_class_get_space(spa_normal_class(spa));
5900 if (ztest_opts.zo_verbose >= 3)
5901 zfs_dbgmsg_print(FTAG);
5903 umem_free(tid, ztest_opts.zo_threads * sizeof (kt_did_t));
5905 /* Kill the resume thread */
5906 ztest_exiting = B_TRUE;
5907 thread_join(resume_thread->t_tid);
5911 * Right before closing the pool, kick off a bunch of async I/O;
5912 * spa_close() should wait for it to complete.
5914 for (object = 1; object < 50; object++)
5915 dmu_prefetch(spa->spa_meta_objset, object, 0, 1ULL << 20);
5917 /* Verify that at least one commit cb was called in a timely fashion */
5918 if (zc_cb_counter >= ZTEST_COMMIT_CB_MIN_REG)
5919 VERIFY0(zc_min_txg_delay);
5921 spa_close(spa, FTAG);
5924 * Verify that we can loop over all pools.
5926 mutex_enter(&spa_namespace_lock);
5927 for (spa = spa_next(NULL); spa != NULL; spa = spa_next(spa))
5928 if (ztest_opts.zo_verbose > 3)
5929 (void) printf("spa_next: found %s\n", spa_name(spa));
5930 mutex_exit(&spa_namespace_lock);
5933 * Verify that we can export the pool and reimport it under a
5936 if (ztest_random(2) == 0) {
5937 char name[MAXNAMELEN];
5938 (void) snprintf(name, MAXNAMELEN, "%s_import",
5939 ztest_opts.zo_pool);
5940 ztest_spa_import_export(ztest_opts.zo_pool, name);
5941 ztest_spa_import_export(name, ztest_opts.zo_pool);
5946 list_destroy(&zcl.zcl_callbacks);
5947 mutex_destroy(&zcl.zcl_callbacks_lock);
5948 (void) rwlock_destroy(&ztest_name_lock);
5949 mutex_destroy(&ztest_vdev_lock);
5955 ztest_ds_t *zd = &ztest_ds[0];
5959 if (ztest_opts.zo_verbose >= 3)
5960 (void) printf("testing spa_freeze()...\n");
5962 kernel_init(FREAD | FWRITE);
5963 VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
5964 VERIFY3U(0, ==, ztest_dataset_open(0));
5965 spa->spa_debug = B_TRUE;
5969 * Force the first log block to be transactionally allocated.
5970 * We have to do this before we freeze the pool -- otherwise
5971 * the log chain won't be anchored.
5973 while (BP_IS_HOLE(&zd->zd_zilog->zl_header->zh_log)) {
5974 ztest_dmu_object_alloc_free(zd, 0);
5975 zil_commit(zd->zd_zilog, 0);
5978 txg_wait_synced(spa_get_dsl(spa), 0);
5981 * Freeze the pool. This stops spa_sync() from doing anything,
5982 * so that the only way to record changes from now on is the ZIL.
5987 * Because it is hard to predict how much space a write will actually
5988 * require beforehand, we leave ourselves some fudge space to write over
5991 uint64_t capacity = metaslab_class_get_space(spa_normal_class(spa)) / 2;
5994 * Run tests that generate log records but don't alter the pool config
5995 * or depend on DSL sync tasks (snapshots, objset create/destroy, etc).
5996 * We do a txg_wait_synced() after each iteration to force the txg
5997 * to increase well beyond the last synced value in the uberblock.
5998 * The ZIL should be OK with that.
6000 * Run a random number of times less than zo_maxloops and ensure we do
6001 * not run out of space on the pool.
6003 while (ztest_random(10) != 0 &&
6004 numloops++ < ztest_opts.zo_maxloops &&
6005 metaslab_class_get_alloc(spa_normal_class(spa)) < capacity) {
6007 ztest_od_init(&od, 0, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
6008 VERIFY0(ztest_object_init(zd, &od, sizeof (od), B_FALSE));
6009 ztest_io(zd, od.od_object,
6010 ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
6011 txg_wait_synced(spa_get_dsl(spa), 0);
6015 * Commit all of the changes we just generated.
6017 zil_commit(zd->zd_zilog, 0);
6018 txg_wait_synced(spa_get_dsl(spa), 0);
6021 * Close our dataset and close the pool.
6023 ztest_dataset_close(0);
6024 spa_close(spa, FTAG);
6028 * Open and close the pool and dataset to induce log replay.
6030 kernel_init(FREAD | FWRITE);
6031 VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
6032 ASSERT(spa_freeze_txg(spa) == UINT64_MAX);
6033 VERIFY3U(0, ==, ztest_dataset_open(0));
6034 ztest_dataset_close(0);
6036 spa->spa_debug = B_TRUE;
6038 txg_wait_synced(spa_get_dsl(spa), 0);
6039 ztest_reguid(NULL, 0);
6041 spa_close(spa, FTAG);
6046 print_time(hrtime_t t, char *timebuf)
6048 hrtime_t s = t / NANOSEC;
6049 hrtime_t m = s / 60;
6050 hrtime_t h = m / 60;
6051 hrtime_t d = h / 24;
6060 (void) sprintf(timebuf,
6061 "%llud%02lluh%02llum%02llus", d, h, m, s);
6063 (void) sprintf(timebuf, "%lluh%02llum%02llus", h, m, s);
6065 (void) sprintf(timebuf, "%llum%02llus", m, s);
6067 (void) sprintf(timebuf, "%llus", s);
6071 make_random_props(void)
6075 VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
6076 if (ztest_random(2) == 0)
6078 VERIFY(nvlist_add_uint64(props, "autoreplace", 1) == 0);
6084 * Create a storage pool with the given name and initial vdev size.
6085 * Then test spa_freeze() functionality.
6088 ztest_init(ztest_shared_t *zs)
6091 nvlist_t *nvroot, *props;
6094 mutex_init(&ztest_vdev_lock, NULL, MUTEX_DEFAULT, NULL);
6095 VERIFY(rwlock_init(&ztest_name_lock, USYNC_THREAD, NULL) == 0);
6097 kernel_init(FREAD | FWRITE);
6100 * Create the storage pool.
6102 (void) spa_destroy(ztest_opts.zo_pool);
6103 ztest_shared->zs_vdev_next_leaf = 0;
6105 zs->zs_mirrors = ztest_opts.zo_mirrors;
6106 nvroot = make_vdev_root(NULL, NULL, NULL, ztest_opts.zo_vdev_size, 0,
6107 0, ztest_opts.zo_raidz, zs->zs_mirrors, 1);
6108 props = make_random_props();
6109 for (i = 0; i < SPA_FEATURES; i++) {
6111 VERIFY3S(-1, !=, asprintf(&buf, "feature@%s",
6112 spa_feature_table[i].fi_uname));
6113 VERIFY3U(0, ==, nvlist_add_uint64(props, buf, 0));
6116 VERIFY3U(0, ==, spa_create(ztest_opts.zo_pool, nvroot, props, NULL));
6117 nvlist_free(nvroot);
6120 VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
6121 zs->zs_metaslab_sz =
6122 1ULL << spa->spa_root_vdev->vdev_child[0]->vdev_ms_shift;
6123 spa_close(spa, FTAG);
6127 ztest_run_zdb(ztest_opts.zo_pool);
6131 ztest_run_zdb(ztest_opts.zo_pool);
6133 (void) rwlock_destroy(&ztest_name_lock);
6134 mutex_destroy(&ztest_vdev_lock);
6140 static char ztest_name_data[] = "/tmp/ztest.data.XXXXXX";
6142 ztest_fd_data = mkstemp(ztest_name_data);
6143 ASSERT3S(ztest_fd_data, >=, 0);
6144 (void) unlink(ztest_name_data);
6148 shared_data_size(ztest_shared_hdr_t *hdr)
6152 size = hdr->zh_hdr_size;
6153 size += hdr->zh_opts_size;
6154 size += hdr->zh_size;
6155 size += hdr->zh_stats_size * hdr->zh_stats_count;
6156 size += hdr->zh_ds_size * hdr->zh_ds_count;
6165 ztest_shared_hdr_t *hdr;
6167 hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
6168 PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
6169 VERIFY3P(hdr, !=, MAP_FAILED);
6171 VERIFY3U(0, ==, ftruncate(ztest_fd_data, sizeof (ztest_shared_hdr_t)));
6173 hdr->zh_hdr_size = sizeof (ztest_shared_hdr_t);
6174 hdr->zh_opts_size = sizeof (ztest_shared_opts_t);
6175 hdr->zh_size = sizeof (ztest_shared_t);
6176 hdr->zh_stats_size = sizeof (ztest_shared_callstate_t);
6177 hdr->zh_stats_count = ZTEST_FUNCS;
6178 hdr->zh_ds_size = sizeof (ztest_shared_ds_t);
6179 hdr->zh_ds_count = ztest_opts.zo_datasets;
6181 size = shared_data_size(hdr);
6182 VERIFY3U(0, ==, ftruncate(ztest_fd_data, size));
6184 (void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
6191 ztest_shared_hdr_t *hdr;
6194 hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
6195 PROT_READ, MAP_SHARED, ztest_fd_data, 0);
6196 VERIFY3P(hdr, !=, MAP_FAILED);
6198 size = shared_data_size(hdr);
6200 (void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
6201 hdr = ztest_shared_hdr = (void *)mmap(0, P2ROUNDUP(size, getpagesize()),
6202 PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
6203 VERIFY3P(hdr, !=, MAP_FAILED);
6204 buf = (uint8_t *)hdr;
6206 offset = hdr->zh_hdr_size;
6207 ztest_shared_opts = (void *)&buf[offset];
6208 offset += hdr->zh_opts_size;
6209 ztest_shared = (void *)&buf[offset];
6210 offset += hdr->zh_size;
6211 ztest_shared_callstate = (void *)&buf[offset];
6212 offset += hdr->zh_stats_size * hdr->zh_stats_count;
6213 ztest_shared_ds = (void *)&buf[offset];
6217 exec_child(char *cmd, char *libpath, boolean_t ignorekill, int *statusp)
6221 char *cmdbuf = NULL;
6226 cmdbuf = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
6227 (void) strlcpy(cmdbuf, getexecname(), MAXPATHLEN);
6232 fatal(1, "fork failed");
6234 if (pid == 0) { /* child */
6235 char *emptyargv[2] = { cmd, NULL };
6236 char fd_data_str[12];
6238 struct rlimit rl = { 1024, 1024 };
6239 (void) setrlimit(RLIMIT_NOFILE, &rl);
6241 (void) close(ztest_fd_rand);
6242 VERIFY(11 >= snprintf(fd_data_str, 12, "%d", ztest_fd_data));
6243 VERIFY(0 == setenv("ZTEST_FD_DATA", fd_data_str, 1));
6245 (void) enable_extended_FILE_stdio(-1, -1);
6246 if (libpath != NULL)
6247 VERIFY(0 == setenv("LD_LIBRARY_PATH", libpath, 1));
6248 (void) execv(cmd, emptyargv);
6249 ztest_dump_core = B_FALSE;
6250 fatal(B_TRUE, "exec failed: %s", cmd);
6253 if (cmdbuf != NULL) {
6254 umem_free(cmdbuf, MAXPATHLEN);
6258 while (waitpid(pid, &status, 0) != pid)
6260 if (statusp != NULL)
6263 if (WIFEXITED(status)) {
6264 if (WEXITSTATUS(status) != 0) {
6265 (void) fprintf(stderr, "child exited with code %d\n",
6266 WEXITSTATUS(status));
6270 } else if (WIFSIGNALED(status)) {
6271 if (!ignorekill || WTERMSIG(status) != SIGKILL) {
6272 (void) fprintf(stderr, "child died with signal %d\n",
6278 (void) fprintf(stderr, "something strange happened to child\n");
6285 ztest_run_init(void)
6289 ztest_shared_t *zs = ztest_shared;
6291 ASSERT(ztest_opts.zo_init != 0);
6294 * Blow away any existing copy of zpool.cache
6296 (void) remove(spa_config_path);
6299 * Create and initialize our storage pool.
6301 for (i = 1; i <= ztest_opts.zo_init; i++) {
6302 bzero(zs, sizeof (ztest_shared_t));
6303 if (ztest_opts.zo_verbose >= 3 &&
6304 ztest_opts.zo_init != 1) {
6305 (void) printf("ztest_init(), pass %d\n", i);
6312 main(int argc, char **argv)
6320 ztest_shared_callstate_t *zc;
6327 char *fd_data_str = getenv("ZTEST_FD_DATA");
6328 struct sigaction action;
6330 (void) setvbuf(stdout, NULL, _IOLBF, 0);
6332 dprintf_setup(&argc, argv);
6334 action.sa_handler = sig_handler;
6335 sigemptyset(&action.sa_mask);
6336 action.sa_flags = 0;
6338 if (sigaction(SIGSEGV, &action, NULL) < 0) {
6339 (void) fprintf(stderr, "ztest: cannot catch SIGSEGV: %s.\n",
6344 if (sigaction(SIGABRT, &action, NULL) < 0) {
6345 (void) fprintf(stderr, "ztest: cannot catch SIGABRT: %s.\n",
6350 ztest_fd_rand = open("/dev/urandom", O_RDONLY);
6351 ASSERT3S(ztest_fd_rand, >=, 0);
6354 process_options(argc, argv);
6359 bcopy(&ztest_opts, ztest_shared_opts,
6360 sizeof (*ztest_shared_opts));
6362 ztest_fd_data = atoi(fd_data_str);
6364 bcopy(ztest_shared_opts, &ztest_opts, sizeof (ztest_opts));
6366 ASSERT3U(ztest_opts.zo_datasets, ==, ztest_shared_hdr->zh_ds_count);
6368 /* Override location of zpool.cache */
6369 VERIFY(asprintf((char **)&spa_config_path, "%s/zpool.cache",
6370 ztest_opts.zo_dir) != -1);
6372 ztest_ds = umem_alloc(ztest_opts.zo_datasets * sizeof (ztest_ds_t),
6377 metaslab_gang_bang = ztest_opts.zo_metaslab_gang_bang;
6378 metaslab_df_alloc_threshold =
6379 zs->zs_metaslab_df_alloc_threshold;
6388 hasalt = (strlen(ztest_opts.zo_alt_ztest) != 0);
6390 if (ztest_opts.zo_verbose >= 1) {
6391 (void) printf("%llu vdevs, %d datasets, %d threads,"
6392 " %llu seconds...\n",
6393 (u_longlong_t)ztest_opts.zo_vdevs,
6394 ztest_opts.zo_datasets,
6395 ztest_opts.zo_threads,
6396 (u_longlong_t)ztest_opts.zo_time);
6399 cmd = umem_alloc(MAXNAMELEN, UMEM_NOFAIL);
6400 (void) strlcpy(cmd, getexecname(), MAXNAMELEN);
6402 zs->zs_do_init = B_TRUE;
6403 if (strlen(ztest_opts.zo_alt_ztest) != 0) {
6404 if (ztest_opts.zo_verbose >= 1) {
6405 (void) printf("Executing older ztest for "
6406 "initialization: %s\n", ztest_opts.zo_alt_ztest);
6408 VERIFY(!exec_child(ztest_opts.zo_alt_ztest,
6409 ztest_opts.zo_alt_libpath, B_FALSE, NULL));
6411 VERIFY(!exec_child(NULL, NULL, B_FALSE, NULL));
6413 zs->zs_do_init = B_FALSE;
6415 zs->zs_proc_start = gethrtime();
6416 zs->zs_proc_stop = zs->zs_proc_start + ztest_opts.zo_time * NANOSEC;
6418 for (f = 0; f < ZTEST_FUNCS; f++) {
6419 zi = &ztest_info[f];
6420 zc = ZTEST_GET_SHARED_CALLSTATE(f);
6421 if (zs->zs_proc_start + zi->zi_interval[0] > zs->zs_proc_stop)
6422 zc->zc_next = UINT64_MAX;
6424 zc->zc_next = zs->zs_proc_start +
6425 ztest_random(2 * zi->zi_interval[0] + 1);
6429 * Run the tests in a loop. These tests include fault injection
6430 * to verify that self-healing data works, and forced crashes
6431 * to verify that we never lose on-disk consistency.
6433 while (gethrtime() < zs->zs_proc_stop) {
6438 * Initialize the workload counters for each function.
6440 for (f = 0; f < ZTEST_FUNCS; f++) {
6441 zc = ZTEST_GET_SHARED_CALLSTATE(f);
6446 /* Set the allocation switch size */
6447 zs->zs_metaslab_df_alloc_threshold =
6448 ztest_random(zs->zs_metaslab_sz / 4) + 1;
6450 if (!hasalt || ztest_random(2) == 0) {
6451 if (hasalt && ztest_opts.zo_verbose >= 1) {
6452 (void) printf("Executing newer ztest: %s\n",
6456 killed = exec_child(cmd, NULL, B_TRUE, &status);
6458 if (hasalt && ztest_opts.zo_verbose >= 1) {
6459 (void) printf("Executing older ztest: %s\n",
6460 ztest_opts.zo_alt_ztest);
6463 killed = exec_child(ztest_opts.zo_alt_ztest,
6464 ztest_opts.zo_alt_libpath, B_TRUE, &status);
6471 if (ztest_opts.zo_verbose >= 1) {
6472 hrtime_t now = gethrtime();
6474 now = MIN(now, zs->zs_proc_stop);
6475 print_time(zs->zs_proc_stop - now, timebuf);
6476 nicenum(zs->zs_space, numbuf);
6478 (void) printf("Pass %3d, %8s, %3llu ENOSPC, "
6479 "%4.1f%% of %5s used, %3.0f%% done, %8s to go\n",
6481 WIFEXITED(status) ? "Complete" : "SIGKILL",
6482 (u_longlong_t)zs->zs_enospc_count,
6483 100.0 * zs->zs_alloc / zs->zs_space,
6485 100.0 * (now - zs->zs_proc_start) /
6486 (ztest_opts.zo_time * NANOSEC), timebuf);
6489 if (ztest_opts.zo_verbose >= 2) {
6490 (void) printf("\nWorkload summary:\n\n");
6491 (void) printf("%7s %9s %s\n",
6492 "Calls", "Time", "Function");
6493 (void) printf("%7s %9s %s\n",
6494 "-----", "----", "--------");
6495 for (f = 0; f < ZTEST_FUNCS; f++) {
6496 zi = &ztest_info[f];
6497 zc = ZTEST_GET_SHARED_CALLSTATE(f);
6498 print_time(zc->zc_time, timebuf);
6499 (void) printf("%7llu %9s %s\n",
6500 (u_longlong_t)zc->zc_count, timebuf,
6503 (void) printf("\n");
6507 * It's possible that we killed a child during a rename test,
6508 * in which case we'll have a 'ztest_tmp' pool lying around
6509 * instead of 'ztest'. Do a blind rename in case this happened.
6512 if (spa_open(ztest_opts.zo_pool, &spa, FTAG) == 0) {
6513 spa_close(spa, FTAG);
6515 char tmpname[MAXNAMELEN];
6517 kernel_init(FREAD | FWRITE);
6518 (void) snprintf(tmpname, sizeof (tmpname), "%s_tmp",
6519 ztest_opts.zo_pool);
6520 (void) spa_rename(tmpname, ztest_opts.zo_pool);
6524 ztest_run_zdb(ztest_opts.zo_pool);
6527 if (ztest_opts.zo_verbose >= 1) {
6529 (void) printf("%d runs of older ztest: %s\n", older,
6530 ztest_opts.zo_alt_ztest);
6531 (void) printf("%d runs of newer ztest: %s\n", newer,
6534 (void) printf("%d killed, %d completed, %.0f%% kill rate\n",
6535 kills, iters - kills, (100.0 * kills) / MAX(1, iters));
6538 umem_free(cmd, MAXNAMELEN);