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, 2016 by Delphix. All rights reserved.
24 * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
25 * Copyright (c) 2013 Steven Hartland. All rights reserved.
26 * Copyright (c) 2014 Integros [integros.com]
27 * Copyright 2017 Joyent, Inc.
31 * The objective of this program is to provide a DMU/ZAP/SPA stress test
32 * that runs entirely in userland, is easy to use, and easy to extend.
34 * The overall design of the ztest program is as follows:
36 * (1) For each major functional area (e.g. adding vdevs to a pool,
37 * creating and destroying datasets, reading and writing objects, etc)
38 * we have a simple routine to test that functionality. These
39 * individual routines do not have to do anything "stressful".
41 * (2) We turn these simple functionality tests into a stress test by
42 * running them all in parallel, with as many threads as desired,
43 * and spread across as many datasets, objects, and vdevs as desired.
45 * (3) While all this is happening, we inject faults into the pool to
46 * verify that self-healing data really works.
48 * (4) Every time we open a dataset, we change its checksum and compression
49 * functions. Thus even individual objects vary from block to block
50 * in which checksum they use and whether they're compressed.
52 * (5) To verify that we never lose on-disk consistency after a crash,
53 * we run the entire test in a child of the main process.
54 * At random times, the child self-immolates with a SIGKILL.
55 * This is the software equivalent of pulling the power cord.
56 * The parent then runs the test again, using the existing
57 * storage pool, as many times as desired. If backwards compatibility
58 * testing is enabled ztest will sometimes run the "older" version
59 * of ztest after a SIGKILL.
61 * (6) To verify that we don't have future leaks or temporal incursions,
62 * many of the functional tests record the transaction group number
63 * as part of their data. When reading old data, they verify that
64 * the transaction group number is less than the current, open txg.
65 * If you add a new test, please do this if applicable.
67 * (7) Threads are created with a reduced stack size, for sanity checking.
68 * Therefore, it's important not to allocate huge buffers on the stack.
70 * When run with no arguments, ztest runs for about five minutes and
71 * produces no output if successful. To get a little bit of information,
72 * specify -V. To get more information, specify -VV, and so on.
74 * To turn this into an overnight stress test, use -T to specify run time.
76 * You can ask more more vdevs [-v], datasets [-d], or threads [-t]
77 * to increase the pool capacity, fanout, and overall stress level.
79 * Use the -k option to set the desired frequency of kills.
81 * When ztest invokes itself it passes all relevant information through a
82 * temporary file which is mmap-ed in the child process. This allows shared
83 * memory to survive the exec syscall. The ztest_shared_hdr_t struct is always
84 * stored at offset 0 of this file and contains information on the size and
85 * number of shared structures in the file. The information stored in this file
86 * must remain backwards compatible with older versions of ztest so that
87 * ztest can invoke them during backwards compatibility testing (-B).
90 #include <sys/zfs_context.h>
96 #include <sys/dmu_objset.h>
100 #include <sys/wait.h>
101 #include <sys/mman.h>
102 #include <sys/resource.h>
105 #include <sys/zil_impl.h>
106 #include <sys/zfs_rlock.h>
107 #include <sys/vdev_impl.h>
108 #include <sys/vdev_file.h>
109 #include <sys/spa_impl.h>
110 #include <sys/metaslab_impl.h>
111 #include <sys/dsl_prop.h>
112 #include <sys/dsl_dataset.h>
113 #include <sys/dsl_destroy.h>
114 #include <sys/dsl_scan.h>
115 #include <sys/zio_checksum.h>
116 #include <sys/refcount.h>
117 #include <sys/zfeature.h>
118 #include <sys/dsl_userhold.h>
121 #include <stdio_ext.h>
128 #include <sys/fs/zfs.h>
129 #include <zfs_fletcher.h>
130 #include <libnvpair.h>
133 #include <execinfo.h> /* for backtrace() */
136 static int ztest_fd_data = -1;
137 static int ztest_fd_rand = -1;
139 typedef struct ztest_shared_hdr {
140 uint64_t zh_hdr_size;
141 uint64_t zh_opts_size;
143 uint64_t zh_stats_size;
144 uint64_t zh_stats_count;
146 uint64_t zh_ds_count;
147 } ztest_shared_hdr_t;
149 static ztest_shared_hdr_t *ztest_shared_hdr;
151 typedef struct ztest_shared_opts {
152 char zo_pool[ZFS_MAX_DATASET_NAME_LEN];
153 char zo_dir[ZFS_MAX_DATASET_NAME_LEN];
154 char zo_alt_ztest[MAXNAMELEN];
155 char zo_alt_libpath[MAXNAMELEN];
157 uint64_t zo_vdevtime;
165 uint64_t zo_passtime;
166 uint64_t zo_killrate;
170 uint64_t zo_maxloops;
171 uint64_t zo_metaslab_gang_bang;
173 } ztest_shared_opts_t;
175 static const ztest_shared_opts_t ztest_opts_defaults = {
176 .zo_pool = { 'z', 't', 'e', 's', 't', '\0' },
177 .zo_dir = { '/', 't', 'm', 'p', '\0' },
178 .zo_alt_ztest = { '\0' },
179 .zo_alt_libpath = { '\0' },
181 .zo_ashift = SPA_MINBLOCKSHIFT,
184 .zo_raidz_parity = 1,
185 .zo_vdev_size = SPA_MINDEVSIZE * 4, /* 256m default size */
188 .zo_passtime = 60, /* 60 seconds */
189 .zo_killrate = 70, /* 70% kill rate */
193 .zo_time = 300, /* 5 minutes */
194 .zo_maxloops = 50, /* max loops during spa_freeze() */
195 .zo_metaslab_gang_bang = 32 << 10
198 extern uint64_t metaslab_gang_bang;
199 extern uint64_t metaslab_df_alloc_threshold;
200 extern int metaslab_preload_limit;
201 extern boolean_t zfs_compressed_arc_enabled;
202 extern int zfs_abd_scatter_enabled;
204 static ztest_shared_opts_t *ztest_shared_opts;
205 static ztest_shared_opts_t ztest_opts;
206 static char *ztest_wkeydata = "abcdefghijklmnopqrstuvwxyz012345";
208 typedef struct ztest_shared_ds {
212 static ztest_shared_ds_t *ztest_shared_ds;
213 #define ZTEST_GET_SHARED_DS(d) (&ztest_shared_ds[d])
215 #define BT_MAGIC 0x123456789abcdefULL
216 #define MAXFAULTS() \
217 (MAX(zs->zs_mirrors, 1) * (ztest_opts.zo_raidz_parity + 1) - 1)
221 ZTEST_IO_WRITE_PATTERN,
222 ZTEST_IO_WRITE_ZEROES,
229 typedef struct ztest_block_tag {
233 uint64_t bt_dnodesize;
240 typedef struct bufwad {
258 #define ZTEST_RANGE_LOCKS 64
259 #define ZTEST_OBJECT_LOCKS 64
262 * Object descriptor. Used as a template for object lookup/create/remove.
264 typedef struct ztest_od {
267 dmu_object_type_t od_type;
268 dmu_object_type_t od_crtype;
269 uint64_t od_blocksize;
270 uint64_t od_crblocksize;
271 uint64_t od_crdnodesize;
274 char od_name[ZFS_MAX_DATASET_NAME_LEN];
280 typedef struct ztest_ds {
281 ztest_shared_ds_t *zd_shared;
283 rwlock_t zd_zilog_lock;
285 ztest_od_t *zd_od; /* debugging aid */
286 char zd_name[ZFS_MAX_DATASET_NAME_LEN];
287 kmutex_t zd_dirobj_lock;
288 rll_t zd_object_lock[ZTEST_OBJECT_LOCKS];
289 zll_t zd_range_lock[ZTEST_RANGE_LOCKS];
293 * Per-iteration state.
295 typedef void ztest_func_t(ztest_ds_t *zd, uint64_t id);
297 typedef struct ztest_info {
298 ztest_func_t *zi_func; /* test function */
299 uint64_t zi_iters; /* iterations per execution */
300 uint64_t *zi_interval; /* execute every <interval> seconds */
301 const char *zi_funcname; /* name of test function */
304 typedef struct ztest_shared_callstate {
305 uint64_t zc_count; /* per-pass count */
306 uint64_t zc_time; /* per-pass time */
307 uint64_t zc_next; /* next time to call this function */
308 } ztest_shared_callstate_t;
310 static ztest_shared_callstate_t *ztest_shared_callstate;
311 #define ZTEST_GET_SHARED_CALLSTATE(c) (&ztest_shared_callstate[c])
313 ztest_func_t ztest_dmu_read_write;
314 ztest_func_t ztest_dmu_write_parallel;
315 ztest_func_t ztest_dmu_object_alloc_free;
316 ztest_func_t ztest_dmu_commit_callbacks;
317 ztest_func_t ztest_zap;
318 ztest_func_t ztest_zap_parallel;
319 ztest_func_t ztest_zil_commit;
320 ztest_func_t ztest_zil_remount;
321 ztest_func_t ztest_dmu_read_write_zcopy;
322 ztest_func_t ztest_dmu_objset_create_destroy;
323 ztest_func_t ztest_dmu_prealloc;
324 ztest_func_t ztest_fzap;
325 ztest_func_t ztest_dmu_snapshot_create_destroy;
326 ztest_func_t ztest_dsl_prop_get_set;
327 ztest_func_t ztest_spa_prop_get_set;
328 ztest_func_t ztest_spa_create_destroy;
329 ztest_func_t ztest_fault_inject;
330 ztest_func_t ztest_ddt_repair;
331 ztest_func_t ztest_dmu_snapshot_hold;
332 ztest_func_t ztest_mmp_enable_disable;
333 ztest_func_t ztest_spa_rename;
334 ztest_func_t ztest_scrub;
335 ztest_func_t ztest_dsl_dataset_promote_busy;
336 ztest_func_t ztest_vdev_attach_detach;
337 ztest_func_t ztest_vdev_LUN_growth;
338 ztest_func_t ztest_vdev_add_remove;
339 ztest_func_t ztest_vdev_aux_add_remove;
340 ztest_func_t ztest_split_pool;
341 ztest_func_t ztest_reguid;
342 ztest_func_t ztest_spa_upgrade;
343 ztest_func_t ztest_fletcher;
344 ztest_func_t ztest_fletcher_incr;
345 ztest_func_t ztest_verify_dnode_bt;
347 uint64_t zopt_always = 0ULL * NANOSEC; /* all the time */
348 uint64_t zopt_incessant = 1ULL * NANOSEC / 10; /* every 1/10 second */
349 uint64_t zopt_often = 1ULL * NANOSEC; /* every second */
350 uint64_t zopt_sometimes = 10ULL * NANOSEC; /* every 10 seconds */
351 uint64_t zopt_rarely = 60ULL * NANOSEC; /* every 60 seconds */
353 #define ZTI_INIT(func, iters, interval) \
354 { .zi_func = (func), \
355 .zi_iters = (iters), \
356 .zi_interval = (interval), \
357 .zi_funcname = # func }
359 ztest_info_t ztest_info[] = {
360 ZTI_INIT(ztest_dmu_read_write, 1, &zopt_always),
361 ZTI_INIT(ztest_dmu_write_parallel, 10, &zopt_always),
362 ZTI_INIT(ztest_dmu_object_alloc_free, 1, &zopt_always),
363 ZTI_INIT(ztest_dmu_commit_callbacks, 1, &zopt_always),
364 ZTI_INIT(ztest_zap, 30, &zopt_always),
365 ZTI_INIT(ztest_zap_parallel, 100, &zopt_always),
366 ZTI_INIT(ztest_split_pool, 1, &zopt_always),
367 ZTI_INIT(ztest_zil_commit, 1, &zopt_incessant),
368 ZTI_INIT(ztest_zil_remount, 1, &zopt_sometimes),
369 ZTI_INIT(ztest_dmu_read_write_zcopy, 1, &zopt_often),
370 ZTI_INIT(ztest_dmu_objset_create_destroy, 1, &zopt_often),
371 ZTI_INIT(ztest_dsl_prop_get_set, 1, &zopt_often),
372 ZTI_INIT(ztest_spa_prop_get_set, 1, &zopt_sometimes),
374 ZTI_INIT(ztest_dmu_prealloc, 1, &zopt_sometimes),
376 ZTI_INIT(ztest_fzap, 1, &zopt_sometimes),
377 ZTI_INIT(ztest_dmu_snapshot_create_destroy, 1, &zopt_sometimes),
378 ZTI_INIT(ztest_spa_create_destroy, 1, &zopt_sometimes),
379 ZTI_INIT(ztest_fault_inject, 1, &zopt_sometimes),
380 ZTI_INIT(ztest_ddt_repair, 1, &zopt_sometimes),
381 ZTI_INIT(ztest_dmu_snapshot_hold, 1, &zopt_sometimes),
382 ZTI_INIT(ztest_mmp_enable_disable, 1, &zopt_sometimes),
383 ZTI_INIT(ztest_reguid, 1, &zopt_rarely),
384 ZTI_INIT(ztest_spa_rename, 1, &zopt_rarely),
385 ZTI_INIT(ztest_scrub, 1, &zopt_rarely),
386 ZTI_INIT(ztest_spa_upgrade, 1, &zopt_rarely),
387 ZTI_INIT(ztest_dsl_dataset_promote_busy, 1, &zopt_rarely),
388 ZTI_INIT(ztest_vdev_attach_detach, 1, &zopt_sometimes),
389 ZTI_INIT(ztest_vdev_LUN_growth, 1, &zopt_rarely),
390 ZTI_INIT(ztest_vdev_add_remove, 1, &ztest_opts.zo_vdevtime),
391 ZTI_INIT(ztest_vdev_aux_add_remove, 1, &ztest_opts.zo_vdevtime),
392 ZTI_INIT(ztest_fletcher, 1, &zopt_rarely),
393 ZTI_INIT(ztest_fletcher_incr, 1, &zopt_rarely),
394 ZTI_INIT(ztest_verify_dnode_bt, 1, &zopt_sometimes),
397 #define ZTEST_FUNCS (sizeof (ztest_info) / sizeof (ztest_info_t))
400 * The following struct is used to hold a list of uncalled commit callbacks.
401 * The callbacks are ordered by txg number.
403 typedef struct ztest_cb_list {
404 kmutex_t zcl_callbacks_lock;
405 list_t zcl_callbacks;
409 * Stuff we need to share writably between parent and child.
411 typedef struct ztest_shared {
412 boolean_t zs_do_init;
413 hrtime_t zs_proc_start;
414 hrtime_t zs_proc_stop;
415 hrtime_t zs_thread_start;
416 hrtime_t zs_thread_stop;
417 hrtime_t zs_thread_kill;
418 uint64_t zs_enospc_count;
419 uint64_t zs_vdev_next_leaf;
420 uint64_t zs_vdev_aux;
425 uint64_t zs_metaslab_sz;
426 uint64_t zs_metaslab_df_alloc_threshold;
430 #define ID_PARALLEL -1ULL
432 static char ztest_dev_template[] = "%s/%s.%llua";
433 static char ztest_aux_template[] = "%s/%s.%s.%llu";
434 ztest_shared_t *ztest_shared;
436 static spa_t *ztest_spa = NULL;
437 static ztest_ds_t *ztest_ds;
439 static kmutex_t ztest_vdev_lock;
442 * The ztest_name_lock protects the pool and dataset namespace used by
443 * the individual tests. To modify the namespace, consumers must grab
444 * this lock as writer. Grabbing the lock as reader will ensure that the
445 * namespace does not change while the lock is held.
447 static rwlock_t ztest_name_lock;
449 static boolean_t ztest_dump_core = B_TRUE;
450 static boolean_t ztest_exiting;
452 /* Global commit callback list */
453 static ztest_cb_list_t zcl;
454 /* Commit cb delay */
455 static uint64_t zc_min_txg_delay = UINT64_MAX;
456 static int zc_cb_counter = 0;
459 * Minimum number of commit callbacks that need to be registered for us to check
460 * whether the minimum txg delay is acceptable.
462 #define ZTEST_COMMIT_CB_MIN_REG 100
465 * If a number of txgs equal to this threshold have been created after a commit
466 * callback has been registered but not called, then we assume there is an
467 * implementation bug.
469 #define ZTEST_COMMIT_CB_THRESH (TXG_CONCURRENT_STATES + 1000)
471 extern uint64_t metaslab_gang_bang;
472 extern uint64_t metaslab_df_alloc_threshold;
475 ZTEST_META_DNODE = 0,
480 static void usage(boolean_t) __NORETURN;
483 * These libumem hooks provide a reasonable set of defaults for the allocator's
484 * debugging facilities.
487 _umem_debug_init(void)
489 return ("default,verbose"); /* $UMEM_DEBUG setting */
493 _umem_logging_init(void)
495 return ("fail,contents"); /* $UMEM_LOGGING setting */
498 #define BACKTRACE_SZ 100
500 static void sig_handler(int signo)
502 struct sigaction action;
503 #ifdef __GLIBC__ /* backtrace() is a GNU extension */
505 void *buffer[BACKTRACE_SZ];
507 nptrs = backtrace(buffer, BACKTRACE_SZ);
508 backtrace_symbols_fd(buffer, nptrs, STDERR_FILENO);
512 * Restore default action and re-raise signal so SIGSEGV and
513 * SIGABRT can trigger a core dump.
515 action.sa_handler = SIG_DFL;
516 sigemptyset(&action.sa_mask);
518 (void) sigaction(signo, &action, NULL);
522 #define FATAL_MSG_SZ 1024
527 fatal(int do_perror, char *message, ...)
530 int save_errno = errno;
533 (void) fflush(stdout);
534 buf = umem_alloc(FATAL_MSG_SZ, UMEM_NOFAIL);
536 va_start(args, message);
537 (void) sprintf(buf, "ztest: ");
539 (void) vsprintf(buf + strlen(buf), message, args);
542 (void) snprintf(buf + strlen(buf), FATAL_MSG_SZ - strlen(buf),
543 ": %s", strerror(save_errno));
545 (void) fprintf(stderr, "%s\n", buf);
546 fatal_msg = buf; /* to ease debugging */
553 str2shift(const char *buf)
555 const char *ends = "BKMGTPEZ";
560 for (i = 0; i < strlen(ends); i++) {
561 if (toupper(buf[0]) == ends[i])
564 if (i == strlen(ends)) {
565 (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n",
569 if (buf[1] == '\0' || (toupper(buf[1]) == 'B' && buf[2] == '\0')) {
572 (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n", buf);
578 nicenumtoull(const char *buf)
583 val = strtoull(buf, &end, 0);
585 (void) fprintf(stderr, "ztest: bad numeric value: %s\n", buf);
587 } else if (end[0] == '.') {
588 double fval = strtod(buf, &end);
589 fval *= pow(2, str2shift(end));
590 if (fval > UINT64_MAX) {
591 (void) fprintf(stderr, "ztest: value too large: %s\n",
595 val = (uint64_t)fval;
597 int shift = str2shift(end);
598 if (shift >= 64 || (val << shift) >> shift != val) {
599 (void) fprintf(stderr, "ztest: value too large: %s\n",
609 usage(boolean_t requested)
611 const ztest_shared_opts_t *zo = &ztest_opts_defaults;
613 char nice_vdev_size[NN_NUMBUF_SZ];
614 char nice_gang_bang[NN_NUMBUF_SZ];
615 FILE *fp = requested ? stdout : stderr;
617 nicenum(zo->zo_vdev_size, nice_vdev_size, sizeof (nice_vdev_size));
618 nicenum(zo->zo_metaslab_gang_bang, nice_gang_bang,
619 sizeof (nice_gang_bang));
621 (void) fprintf(fp, "Usage: %s\n"
622 "\t[-v vdevs (default: %llu)]\n"
623 "\t[-s size_of_each_vdev (default: %s)]\n"
624 "\t[-a alignment_shift (default: %d)] use 0 for random\n"
625 "\t[-m mirror_copies (default: %d)]\n"
626 "\t[-r raidz_disks (default: %d)]\n"
627 "\t[-R raidz_parity (default: %d)]\n"
628 "\t[-d datasets (default: %d)]\n"
629 "\t[-t threads (default: %d)]\n"
630 "\t[-g gang_block_threshold (default: %s)]\n"
631 "\t[-i init_count (default: %d)] initialize pool i times\n"
632 "\t[-k kill_percentage (default: %llu%%)]\n"
633 "\t[-p pool_name (default: %s)]\n"
634 "\t[-f dir (default: %s)] file directory for vdev files\n"
635 "\t[-M] Multi-host simulate pool imported on remote host\n"
636 "\t[-V] verbose (use multiple times for ever more blather)\n"
637 "\t[-E] use existing pool instead of creating new one\n"
638 "\t[-T time (default: %llu sec)] total run time\n"
639 "\t[-F freezeloops (default: %llu)] max loops in spa_freeze()\n"
640 "\t[-P passtime (default: %llu sec)] time per pass\n"
641 "\t[-B alt_ztest (default: <none>)] alternate ztest path\n"
642 "\t[-o variable=value] ... set global variable to an unsigned\n"
643 "\t 32-bit integer value\n"
644 "\t[-h] (print help)\n"
647 (u_longlong_t)zo->zo_vdevs, /* -v */
648 nice_vdev_size, /* -s */
649 zo->zo_ashift, /* -a */
650 zo->zo_mirrors, /* -m */
651 zo->zo_raidz, /* -r */
652 zo->zo_raidz_parity, /* -R */
653 zo->zo_datasets, /* -d */
654 zo->zo_threads, /* -t */
655 nice_gang_bang, /* -g */
656 zo->zo_init, /* -i */
657 (u_longlong_t)zo->zo_killrate, /* -k */
658 zo->zo_pool, /* -p */
660 (u_longlong_t)zo->zo_time, /* -T */
661 (u_longlong_t)zo->zo_maxloops, /* -F */
662 (u_longlong_t)zo->zo_passtime);
663 exit(requested ? 0 : 1);
667 process_options(int argc, char **argv)
670 ztest_shared_opts_t *zo = &ztest_opts;
674 char altdir[MAXNAMELEN] = { 0 };
676 bcopy(&ztest_opts_defaults, zo, sizeof (*zo));
678 while ((opt = getopt(argc, argv,
679 "v:s:a:m:r:R:d:t:g:i:k:p:f:MVET:P:hF:B:o:")) != EOF) {
696 value = nicenumtoull(optarg);
700 zo->zo_vdevs = value;
703 zo->zo_vdev_size = MAX(SPA_MINDEVSIZE, value);
706 zo->zo_ashift = value;
709 zo->zo_mirrors = value;
712 zo->zo_raidz = MAX(1, value);
715 zo->zo_raidz_parity = MIN(MAX(value, 1), 3);
718 zo->zo_datasets = MAX(1, value);
721 zo->zo_threads = MAX(1, value);
724 zo->zo_metaslab_gang_bang = MAX(SPA_MINBLOCKSIZE << 1,
731 zo->zo_killrate = value;
734 (void) strlcpy(zo->zo_pool, optarg,
735 sizeof (zo->zo_pool));
738 path = realpath(optarg, NULL);
740 (void) fprintf(stderr, "error: %s: %s\n",
741 optarg, strerror(errno));
744 (void) strlcpy(zo->zo_dir, path,
745 sizeof (zo->zo_dir));
762 zo->zo_passtime = MAX(1, value);
765 zo->zo_maxloops = MAX(1, value);
768 (void) strlcpy(altdir, optarg, sizeof (altdir));
771 if (set_global_var(optarg) != 0)
784 zo->zo_raidz_parity = MIN(zo->zo_raidz_parity, zo->zo_raidz - 1);
787 (zo->zo_vdevs > 0 ? zo->zo_time * NANOSEC / zo->zo_vdevs :
790 if (strlen(altdir) > 0) {
798 cmd = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
799 realaltdir = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
801 VERIFY(NULL != realpath(getexecname(), cmd));
802 if (0 != access(altdir, F_OK)) {
803 ztest_dump_core = B_FALSE;
804 fatal(B_TRUE, "invalid alternate ztest path: %s",
807 VERIFY(NULL != realpath(altdir, realaltdir));
810 * 'cmd' should be of the form "<anything>/usr/bin/<isa>/ztest".
811 * We want to extract <isa> to determine if we should use
812 * 32 or 64 bit binaries.
814 bin = strstr(cmd, "/usr/bin/");
815 ztest = strstr(bin, "/ztest");
817 isalen = ztest - isa;
818 (void) snprintf(zo->zo_alt_ztest, sizeof (zo->zo_alt_ztest),
819 "%s/usr/bin/%.*s/ztest", realaltdir, isalen, isa);
820 (void) snprintf(zo->zo_alt_libpath, sizeof (zo->zo_alt_libpath),
821 "%s/usr/lib/%.*s", realaltdir, isalen, isa);
823 if (0 != access(zo->zo_alt_ztest, X_OK)) {
824 ztest_dump_core = B_FALSE;
825 fatal(B_TRUE, "invalid alternate ztest: %s",
827 } else if (0 != access(zo->zo_alt_libpath, X_OK)) {
828 ztest_dump_core = B_FALSE;
829 fatal(B_TRUE, "invalid alternate lib directory %s",
833 umem_free(cmd, MAXPATHLEN);
834 umem_free(realaltdir, MAXPATHLEN);
839 ztest_kill(ztest_shared_t *zs)
841 zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(ztest_spa));
842 zs->zs_space = metaslab_class_get_space(spa_normal_class(ztest_spa));
845 * Before we kill off ztest, make sure that the config is updated.
846 * See comment above spa_config_sync().
848 mutex_enter(&spa_namespace_lock);
849 spa_config_sync(ztest_spa, B_FALSE, B_FALSE);
850 mutex_exit(&spa_namespace_lock);
852 (void) kill(getpid(), SIGKILL);
856 ztest_random(uint64_t range)
860 ASSERT3S(ztest_fd_rand, >=, 0);
865 if (read(ztest_fd_rand, &r, sizeof (r)) != sizeof (r))
866 fatal(1, "short read from /dev/urandom");
873 ztest_record_enospc(const char *s)
875 ztest_shared->zs_enospc_count++;
879 ztest_get_ashift(void)
881 if (ztest_opts.zo_ashift == 0)
882 return (SPA_MINBLOCKSHIFT + ztest_random(5));
883 return (ztest_opts.zo_ashift);
887 make_vdev_file(char *path, char *aux, char *pool, size_t size, uint64_t ashift)
893 pathbuf = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
896 ashift = ztest_get_ashift();
902 vdev = ztest_shared->zs_vdev_aux;
903 (void) snprintf(path, MAXPATHLEN,
904 ztest_aux_template, ztest_opts.zo_dir,
905 pool == NULL ? ztest_opts.zo_pool : pool,
908 vdev = ztest_shared->zs_vdev_next_leaf++;
909 (void) snprintf(path, MAXPATHLEN,
910 ztest_dev_template, ztest_opts.zo_dir,
911 pool == NULL ? ztest_opts.zo_pool : pool, vdev);
916 int fd = open(path, O_RDWR | O_CREAT | O_TRUNC, 0666);
918 fatal(1, "can't open %s", path);
919 if (ftruncate(fd, size) != 0)
920 fatal(1, "can't ftruncate %s", path);
924 VERIFY(nvlist_alloc(&file, NV_UNIQUE_NAME, 0) == 0);
925 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_TYPE, VDEV_TYPE_FILE) == 0);
926 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_PATH, path) == 0);
927 VERIFY(nvlist_add_uint64(file, ZPOOL_CONFIG_ASHIFT, ashift) == 0);
928 umem_free(pathbuf, MAXPATHLEN);
934 make_vdev_raidz(char *path, char *aux, char *pool, size_t size,
935 uint64_t ashift, int r)
937 nvlist_t *raidz, **child;
941 return (make_vdev_file(path, aux, pool, size, ashift));
942 child = umem_alloc(r * sizeof (nvlist_t *), UMEM_NOFAIL);
944 for (c = 0; c < r; c++)
945 child[c] = make_vdev_file(path, aux, pool, size, ashift);
947 VERIFY(nvlist_alloc(&raidz, NV_UNIQUE_NAME, 0) == 0);
948 VERIFY(nvlist_add_string(raidz, ZPOOL_CONFIG_TYPE,
949 VDEV_TYPE_RAIDZ) == 0);
950 VERIFY(nvlist_add_uint64(raidz, ZPOOL_CONFIG_NPARITY,
951 ztest_opts.zo_raidz_parity) == 0);
952 VERIFY(nvlist_add_nvlist_array(raidz, ZPOOL_CONFIG_CHILDREN,
955 for (c = 0; c < r; c++)
956 nvlist_free(child[c]);
958 umem_free(child, r * sizeof (nvlist_t *));
964 make_vdev_mirror(char *path, char *aux, char *pool, size_t size,
965 uint64_t ashift, int r, int m)
967 nvlist_t *mirror, **child;
971 return (make_vdev_raidz(path, aux, pool, size, ashift, r));
973 child = umem_alloc(m * sizeof (nvlist_t *), UMEM_NOFAIL);
975 for (c = 0; c < m; c++)
976 child[c] = make_vdev_raidz(path, aux, pool, size, ashift, r);
978 VERIFY(nvlist_alloc(&mirror, NV_UNIQUE_NAME, 0) == 0);
979 VERIFY(nvlist_add_string(mirror, ZPOOL_CONFIG_TYPE,
980 VDEV_TYPE_MIRROR) == 0);
981 VERIFY(nvlist_add_nvlist_array(mirror, ZPOOL_CONFIG_CHILDREN,
984 for (c = 0; c < m; c++)
985 nvlist_free(child[c]);
987 umem_free(child, m * sizeof (nvlist_t *));
993 make_vdev_root(char *path, char *aux, char *pool, size_t size, uint64_t ashift,
994 int log, int r, int m, int t)
996 nvlist_t *root, **child;
1001 child = umem_alloc(t * sizeof (nvlist_t *), UMEM_NOFAIL);
1003 for (c = 0; c < t; c++) {
1004 child[c] = make_vdev_mirror(path, aux, pool, size, ashift,
1006 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_IS_LOG,
1010 VERIFY(nvlist_alloc(&root, NV_UNIQUE_NAME, 0) == 0);
1011 VERIFY(nvlist_add_string(root, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) == 0);
1012 VERIFY(nvlist_add_nvlist_array(root, aux ? aux : ZPOOL_CONFIG_CHILDREN,
1015 for (c = 0; c < t; c++)
1016 nvlist_free(child[c]);
1018 umem_free(child, t * sizeof (nvlist_t *));
1024 * Find a random spa version. Returns back a random spa version in the
1025 * range [initial_version, SPA_VERSION_FEATURES].
1028 ztest_random_spa_version(uint64_t initial_version)
1030 uint64_t version = initial_version;
1032 if (version <= SPA_VERSION_BEFORE_FEATURES) {
1034 ztest_random(SPA_VERSION_BEFORE_FEATURES - version + 1);
1037 if (version > SPA_VERSION_BEFORE_FEATURES)
1038 version = SPA_VERSION_FEATURES;
1040 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
1045 ztest_random_blocksize(void)
1048 * Choose a block size >= the ashift.
1049 * If the SPA supports new MAXBLOCKSIZE, test up to 1MB blocks.
1051 int maxbs = SPA_OLD_MAXBLOCKSHIFT;
1052 if (spa_maxblocksize(ztest_spa) == SPA_MAXBLOCKSIZE)
1054 uint64_t block_shift =
1055 ztest_random(maxbs - ztest_spa->spa_max_ashift + 1);
1056 return (1 << (SPA_MINBLOCKSHIFT + block_shift));
1060 ztest_random_dnodesize(void)
1063 int max_slots = spa_maxdnodesize(ztest_spa) >> DNODE_SHIFT;
1065 if (max_slots == DNODE_MIN_SLOTS)
1066 return (DNODE_MIN_SIZE);
1069 * Weight the random distribution more heavily toward smaller
1070 * dnode sizes since that is more likely to reflect real-world
1073 ASSERT3U(max_slots, >, 4);
1074 switch (ztest_random(10)) {
1076 slots = 5 + ztest_random(max_slots - 4);
1079 slots = 2 + ztest_random(3);
1086 return (slots << DNODE_SHIFT);
1090 ztest_random_ibshift(void)
1092 return (DN_MIN_INDBLKSHIFT +
1093 ztest_random(DN_MAX_INDBLKSHIFT - DN_MIN_INDBLKSHIFT + 1));
1097 ztest_random_vdev_top(spa_t *spa, boolean_t log_ok)
1100 vdev_t *rvd = spa->spa_root_vdev;
1103 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1106 top = ztest_random(rvd->vdev_children);
1107 tvd = rvd->vdev_child[top];
1108 } while (tvd->vdev_ishole || (tvd->vdev_islog && !log_ok) ||
1109 tvd->vdev_mg == NULL || tvd->vdev_mg->mg_class == NULL);
1115 ztest_random_dsl_prop(zfs_prop_t prop)
1120 value = zfs_prop_random_value(prop, ztest_random(-1ULL));
1121 } while (prop == ZFS_PROP_CHECKSUM && value == ZIO_CHECKSUM_OFF);
1127 ztest_dsl_prop_set_uint64(char *osname, zfs_prop_t prop, uint64_t value,
1130 const char *propname = zfs_prop_to_name(prop);
1131 const char *valname;
1136 error = dsl_prop_set_int(osname, propname,
1137 (inherit ? ZPROP_SRC_NONE : ZPROP_SRC_LOCAL), value);
1139 if (error == ENOSPC) {
1140 ztest_record_enospc(FTAG);
1145 setpoint = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
1146 VERIFY0(dsl_prop_get_integer(osname, propname, &curval, setpoint));
1148 if (ztest_opts.zo_verbose >= 6) {
1151 err = zfs_prop_index_to_string(prop, curval, &valname);
1153 (void) printf("%s %s = %llu at '%s'\n", osname,
1154 propname, (unsigned long long)curval, setpoint);
1156 (void) printf("%s %s = %s at '%s'\n",
1157 osname, propname, valname, setpoint);
1159 umem_free(setpoint, MAXPATHLEN);
1165 ztest_spa_prop_set_uint64(zpool_prop_t prop, uint64_t value)
1167 spa_t *spa = ztest_spa;
1168 nvlist_t *props = NULL;
1171 VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
1172 VERIFY(nvlist_add_uint64(props, zpool_prop_to_name(prop), value) == 0);
1174 error = spa_prop_set(spa, props);
1178 if (error == ENOSPC) {
1179 ztest_record_enospc(FTAG);
1188 ztest_dmu_objset_own(const char *name, dmu_objset_type_t type,
1189 boolean_t readonly, boolean_t decrypt, void *tag, objset_t **osp)
1193 err = dmu_objset_own(name, type, readonly, decrypt, tag, osp);
1194 if (decrypt && err == EACCES) {
1195 char ddname[ZFS_MAX_DATASET_NAME_LEN];
1196 dsl_crypto_params_t *dcp;
1197 nvlist_t *crypto_args = fnvlist_alloc();
1200 /* spa_keystore_load_wkey() expects a dsl dir name */
1201 strcpy(ddname, name);
1202 cp = strchr(ddname, '@');
1206 fnvlist_add_uint8_array(crypto_args, "wkeydata",
1207 (uint8_t *)ztest_wkeydata, WRAPPING_KEY_LEN);
1208 VERIFY0(dsl_crypto_params_create_nvlist(DCP_CMD_NONE, NULL,
1209 crypto_args, &dcp));
1210 err = spa_keystore_load_wkey(ddname, dcp, B_FALSE);
1211 dsl_crypto_params_free(dcp, B_FALSE);
1212 fnvlist_free(crypto_args);
1217 err = dmu_objset_own(name, type, readonly, decrypt, tag, osp);
1225 * Object and range lock mechanics
1228 list_node_t z_lnode;
1229 refcount_t z_refcnt;
1231 zfs_rlock_t z_range_lock;
1236 ztest_znode_t *z_ztznode;
1239 static ztest_znode_t *
1240 ztest_znode_init(uint64_t object)
1242 ztest_znode_t *zp = umem_alloc(sizeof (*zp), UMEM_NOFAIL);
1244 list_link_init(&zp->z_lnode);
1245 refcount_create(&zp->z_refcnt);
1246 zp->z_object = object;
1247 zfs_rlock_init(&zp->z_range_lock);
1253 ztest_znode_fini(ztest_znode_t *zp)
1255 ASSERT(refcount_is_zero(&zp->z_refcnt));
1256 zfs_rlock_destroy(&zp->z_range_lock);
1258 refcount_destroy(&zp->z_refcnt);
1259 list_link_init(&zp->z_lnode);
1260 umem_free(zp, sizeof (*zp));
1264 ztest_zll_init(zll_t *zll)
1266 mutex_init(&zll->z_lock, NULL, MUTEX_DEFAULT, NULL);
1267 list_create(&zll->z_list, sizeof (ztest_znode_t),
1268 offsetof(ztest_znode_t, z_lnode));
1272 ztest_zll_destroy(zll_t *zll)
1274 list_destroy(&zll->z_list);
1275 mutex_destroy(&zll->z_lock);
1278 #define RL_TAG "range_lock"
1279 static ztest_znode_t *
1280 ztest_znode_get(ztest_ds_t *zd, uint64_t object)
1282 zll_t *zll = &zd->zd_range_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1283 ztest_znode_t *zp = NULL;
1284 mutex_enter(&zll->z_lock);
1285 for (zp = list_head(&zll->z_list); (zp);
1286 zp = list_next(&zll->z_list, zp)) {
1287 if (zp->z_object == object) {
1288 refcount_add(&zp->z_refcnt, RL_TAG);
1293 zp = ztest_znode_init(object);
1294 refcount_add(&zp->z_refcnt, RL_TAG);
1295 list_insert_head(&zll->z_list, zp);
1297 mutex_exit(&zll->z_lock);
1302 ztest_znode_put(ztest_ds_t *zd, ztest_znode_t *zp)
1305 ASSERT3U(zp->z_object, !=, 0);
1306 zll = &zd->zd_range_lock[zp->z_object & (ZTEST_OBJECT_LOCKS - 1)];
1307 mutex_enter(&zll->z_lock);
1308 refcount_remove(&zp->z_refcnt, RL_TAG);
1309 if (refcount_is_zero(&zp->z_refcnt)) {
1310 list_remove(&zll->z_list, zp);
1311 ztest_znode_fini(zp);
1313 mutex_exit(&zll->z_lock);
1318 ztest_rll_init(rll_t *rll)
1320 rll->rll_writer = NULL;
1321 rll->rll_readers = 0;
1322 mutex_init(&rll->rll_lock, NULL, MUTEX_DEFAULT, NULL);
1323 cv_init(&rll->rll_cv, NULL, CV_DEFAULT, NULL);
1327 ztest_rll_destroy(rll_t *rll)
1329 ASSERT(rll->rll_writer == NULL);
1330 ASSERT(rll->rll_readers == 0);
1331 mutex_destroy(&rll->rll_lock);
1332 cv_destroy(&rll->rll_cv);
1336 ztest_rll_lock(rll_t *rll, rl_type_t type)
1338 mutex_enter(&rll->rll_lock);
1340 if (type == RL_READER) {
1341 while (rll->rll_writer != NULL)
1342 (void) cv_wait(&rll->rll_cv, &rll->rll_lock);
1345 while (rll->rll_writer != NULL || rll->rll_readers)
1346 (void) cv_wait(&rll->rll_cv, &rll->rll_lock);
1347 rll->rll_writer = curthread;
1350 mutex_exit(&rll->rll_lock);
1354 ztest_rll_unlock(rll_t *rll)
1356 mutex_enter(&rll->rll_lock);
1358 if (rll->rll_writer) {
1359 ASSERT(rll->rll_readers == 0);
1360 rll->rll_writer = NULL;
1362 ASSERT(rll->rll_readers != 0);
1363 ASSERT(rll->rll_writer == NULL);
1367 if (rll->rll_writer == NULL && rll->rll_readers == 0)
1368 cv_broadcast(&rll->rll_cv);
1370 mutex_exit(&rll->rll_lock);
1374 ztest_object_lock(ztest_ds_t *zd, uint64_t object, rl_type_t type)
1376 rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1378 ztest_rll_lock(rll, type);
1382 ztest_object_unlock(ztest_ds_t *zd, uint64_t object)
1384 rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1386 ztest_rll_unlock(rll);
1389 static ztest_zrl_t *
1390 ztest_zrl_init(rl_t *rl, ztest_znode_t *zp)
1392 ztest_zrl_t *zrl = umem_alloc(sizeof (*zrl), UMEM_NOFAIL);
1394 zrl->z_ztznode = zp;
1399 ztest_zrl_fini(ztest_zrl_t *zrl)
1401 umem_free(zrl, sizeof (*zrl));
1404 static ztest_zrl_t *
1405 ztest_range_lock(ztest_ds_t *zd, uint64_t object, uint64_t offset,
1406 uint64_t size, rl_type_t type)
1408 ztest_znode_t *zp = ztest_znode_get(zd, object);
1409 rl_t *rl = zfs_range_lock(&zp->z_range_lock, offset,
1411 return (ztest_zrl_init(rl, zp));
1415 ztest_range_unlock(ztest_ds_t *zd, ztest_zrl_t *zrl)
1417 zfs_range_unlock(zrl->z_rl);
1418 ztest_znode_put(zd, zrl->z_ztznode);
1419 ztest_zrl_fini(zrl);
1423 ztest_zd_init(ztest_ds_t *zd, ztest_shared_ds_t *szd, objset_t *os)
1426 zd->zd_zilog = dmu_objset_zil(os);
1427 zd->zd_shared = szd;
1428 dmu_objset_name(os, zd->zd_name);
1431 if (zd->zd_shared != NULL)
1432 zd->zd_shared->zd_seq = 0;
1434 VERIFY(rwlock_init(&zd->zd_zilog_lock, USYNC_THREAD, NULL) == 0);
1435 mutex_init(&zd->zd_dirobj_lock, NULL, MUTEX_DEFAULT, NULL);
1437 for (l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1438 ztest_rll_init(&zd->zd_object_lock[l]);
1440 for (l = 0; l < ZTEST_RANGE_LOCKS; l++)
1441 ztest_zll_init(&zd->zd_range_lock[l]);
1445 ztest_zd_fini(ztest_ds_t *zd)
1449 mutex_destroy(&zd->zd_dirobj_lock);
1450 (void) rwlock_destroy(&zd->zd_zilog_lock);
1452 for (l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1453 ztest_rll_destroy(&zd->zd_object_lock[l]);
1455 for (l = 0; l < ZTEST_RANGE_LOCKS; l++)
1456 ztest_zll_destroy(&zd->zd_range_lock[l]);
1459 #define TXG_MIGHTWAIT (ztest_random(10) == 0 ? TXG_NOWAIT : TXG_WAIT)
1462 ztest_tx_assign(dmu_tx_t *tx, uint64_t txg_how, const char *tag)
1468 * Attempt to assign tx to some transaction group.
1470 error = dmu_tx_assign(tx, txg_how);
1472 if (error == ERESTART) {
1473 ASSERT(txg_how == TXG_NOWAIT);
1476 ASSERT3U(error, ==, ENOSPC);
1477 ztest_record_enospc(tag);
1482 txg = dmu_tx_get_txg(tx);
1488 ztest_pattern_set(void *buf, uint64_t size, uint64_t value)
1491 uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1499 ztest_pattern_match(void *buf, uint64_t size, uint64_t value)
1502 uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1506 diff |= (value - *ip++);
1513 ztest_bt_generate(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1514 uint64_t dnodesize, uint64_t offset, uint64_t gen, uint64_t txg,
1517 bt->bt_magic = BT_MAGIC;
1518 bt->bt_objset = dmu_objset_id(os);
1519 bt->bt_object = object;
1520 bt->bt_dnodesize = dnodesize;
1521 bt->bt_offset = offset;
1524 bt->bt_crtxg = crtxg;
1528 ztest_bt_verify(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1529 uint64_t dnodesize, uint64_t offset, uint64_t gen, uint64_t txg,
1532 ASSERT3U(bt->bt_magic, ==, BT_MAGIC);
1533 ASSERT3U(bt->bt_objset, ==, dmu_objset_id(os));
1534 ASSERT3U(bt->bt_object, ==, object);
1535 ASSERT3U(bt->bt_dnodesize, ==, dnodesize);
1536 ASSERT3U(bt->bt_offset, ==, offset);
1537 ASSERT3U(bt->bt_gen, <=, gen);
1538 ASSERT3U(bt->bt_txg, <=, txg);
1539 ASSERT3U(bt->bt_crtxg, ==, crtxg);
1542 static ztest_block_tag_t *
1543 ztest_bt_bonus(dmu_buf_t *db)
1545 dmu_object_info_t doi;
1546 ztest_block_tag_t *bt;
1548 dmu_object_info_from_db(db, &doi);
1549 ASSERT3U(doi.doi_bonus_size, <=, db->db_size);
1550 ASSERT3U(doi.doi_bonus_size, >=, sizeof (*bt));
1551 bt = (void *)((char *)db->db_data + doi.doi_bonus_size - sizeof (*bt));
1557 * Generate a token to fill up unused bonus buffer space. Try to make
1558 * it unique to the object, generation, and offset to verify that data
1559 * is not getting overwritten by data from other dnodes.
1561 #define ZTEST_BONUS_FILL_TOKEN(obj, ds, gen, offset) \
1562 (((ds) << 48) | ((gen) << 32) | ((obj) << 8) | (offset))
1565 * Fill up the unused bonus buffer region before the block tag with a
1566 * verifiable pattern. Filling the whole bonus area with non-zero data
1567 * helps ensure that all dnode traversal code properly skips the
1568 * interior regions of large dnodes.
1571 ztest_fill_unused_bonus(dmu_buf_t *db, void *end, uint64_t obj,
1572 objset_t *os, uint64_t gen)
1576 ASSERT(IS_P2ALIGNED((char *)end - (char *)db->db_data, 8));
1578 for (bonusp = db->db_data; bonusp < (uint64_t *)end; bonusp++) {
1579 uint64_t token = ZTEST_BONUS_FILL_TOKEN(obj, dmu_objset_id(os),
1580 gen, bonusp - (uint64_t *)db->db_data);
1586 * Verify that the unused area of a bonus buffer is filled with the
1590 ztest_verify_unused_bonus(dmu_buf_t *db, void *end, uint64_t obj,
1591 objset_t *os, uint64_t gen)
1595 for (bonusp = db->db_data; bonusp < (uint64_t *)end; bonusp++) {
1596 uint64_t token = ZTEST_BONUS_FILL_TOKEN(obj, dmu_objset_id(os),
1597 gen, bonusp - (uint64_t *)db->db_data);
1598 VERIFY3U(*bonusp, ==, token);
1606 #define lrz_type lr_mode
1607 #define lrz_blocksize lr_uid
1608 #define lrz_ibshift lr_gid
1609 #define lrz_bonustype lr_rdev
1610 #define lrz_dnodesize lr_crtime[1]
1613 ztest_log_create(ztest_ds_t *zd, dmu_tx_t *tx, lr_create_t *lr)
1615 char *name = (void *)(lr + 1); /* name follows lr */
1616 size_t namesize = strlen(name) + 1;
1619 if (zil_replaying(zd->zd_zilog, tx))
1622 itx = zil_itx_create(TX_CREATE, sizeof (*lr) + namesize);
1623 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1624 sizeof (*lr) + namesize - sizeof (lr_t));
1626 zil_itx_assign(zd->zd_zilog, itx, tx);
1630 ztest_log_remove(ztest_ds_t *zd, dmu_tx_t *tx, lr_remove_t *lr, uint64_t object)
1632 char *name = (void *)(lr + 1); /* name follows lr */
1633 size_t namesize = strlen(name) + 1;
1636 if (zil_replaying(zd->zd_zilog, tx))
1639 itx = zil_itx_create(TX_REMOVE, sizeof (*lr) + namesize);
1640 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1641 sizeof (*lr) + namesize - sizeof (lr_t));
1643 itx->itx_oid = object;
1644 zil_itx_assign(zd->zd_zilog, itx, tx);
1648 ztest_log_write(ztest_ds_t *zd, dmu_tx_t *tx, lr_write_t *lr)
1651 itx_wr_state_t write_state = ztest_random(WR_NUM_STATES);
1653 if (zil_replaying(zd->zd_zilog, tx))
1656 if (lr->lr_length > ZIL_MAX_LOG_DATA)
1657 write_state = WR_INDIRECT;
1659 itx = zil_itx_create(TX_WRITE,
1660 sizeof (*lr) + (write_state == WR_COPIED ? lr->lr_length : 0));
1662 if (write_state == WR_COPIED &&
1663 dmu_read(zd->zd_os, lr->lr_foid, lr->lr_offset, lr->lr_length,
1664 ((lr_write_t *)&itx->itx_lr) + 1, DMU_READ_NO_PREFETCH) != 0) {
1665 zil_itx_destroy(itx);
1666 itx = zil_itx_create(TX_WRITE, sizeof (*lr));
1667 write_state = WR_NEED_COPY;
1669 itx->itx_private = zd;
1670 itx->itx_wr_state = write_state;
1671 itx->itx_sync = (ztest_random(8) == 0);
1673 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1674 sizeof (*lr) - sizeof (lr_t));
1676 zil_itx_assign(zd->zd_zilog, itx, tx);
1680 ztest_log_truncate(ztest_ds_t *zd, dmu_tx_t *tx, lr_truncate_t *lr)
1684 if (zil_replaying(zd->zd_zilog, tx))
1687 itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr));
1688 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1689 sizeof (*lr) - sizeof (lr_t));
1691 itx->itx_sync = B_FALSE;
1692 zil_itx_assign(zd->zd_zilog, itx, tx);
1696 ztest_log_setattr(ztest_ds_t *zd, dmu_tx_t *tx, lr_setattr_t *lr)
1700 if (zil_replaying(zd->zd_zilog, tx))
1703 itx = zil_itx_create(TX_SETATTR, sizeof (*lr));
1704 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1705 sizeof (*lr) - sizeof (lr_t));
1707 itx->itx_sync = B_FALSE;
1708 zil_itx_assign(zd->zd_zilog, itx, tx);
1715 ztest_replay_create(void *arg1, void *arg2, boolean_t byteswap)
1717 ztest_ds_t *zd = arg1;
1718 lr_create_t *lr = arg2;
1719 char *name = (void *)(lr + 1); /* name follows lr */
1720 objset_t *os = zd->zd_os;
1721 ztest_block_tag_t *bbt;
1729 byteswap_uint64_array(lr, sizeof (*lr));
1731 ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1732 ASSERT(name[0] != '\0');
1734 tx = dmu_tx_create(os);
1736 dmu_tx_hold_zap(tx, lr->lr_doid, B_TRUE, name);
1738 if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1739 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, B_TRUE, NULL);
1741 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1744 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1748 ASSERT(dmu_objset_zil(os)->zl_replay == !!lr->lr_foid);
1749 bonuslen = DN_BONUS_SIZE(lr->lrz_dnodesize);
1751 if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1752 if (lr->lr_foid == 0) {
1753 lr->lr_foid = zap_create_dnsize(os,
1754 lr->lrz_type, lr->lrz_bonustype,
1755 bonuslen, lr->lrz_dnodesize, tx);
1757 error = zap_create_claim_dnsize(os, lr->lr_foid,
1758 lr->lrz_type, lr->lrz_bonustype,
1759 bonuslen, lr->lrz_dnodesize, tx);
1762 if (lr->lr_foid == 0) {
1763 lr->lr_foid = dmu_object_alloc_dnsize(os,
1764 lr->lrz_type, 0, lr->lrz_bonustype,
1765 bonuslen, lr->lrz_dnodesize, tx);
1767 error = dmu_object_claim_dnsize(os, lr->lr_foid,
1768 lr->lrz_type, 0, lr->lrz_bonustype,
1769 bonuslen, lr->lrz_dnodesize, tx);
1774 ASSERT3U(error, ==, EEXIST);
1775 ASSERT(zd->zd_zilog->zl_replay);
1780 ASSERT(lr->lr_foid != 0);
1782 if (lr->lrz_type != DMU_OT_ZAP_OTHER)
1783 VERIFY3U(0, ==, dmu_object_set_blocksize(os, lr->lr_foid,
1784 lr->lrz_blocksize, lr->lrz_ibshift, tx));
1786 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1787 bbt = ztest_bt_bonus(db);
1788 dmu_buf_will_dirty(db, tx);
1789 ztest_bt_generate(bbt, os, lr->lr_foid, lr->lrz_dnodesize, -1ULL,
1790 lr->lr_gen, txg, txg);
1791 ztest_fill_unused_bonus(db, bbt, lr->lr_foid, os, lr->lr_gen);
1792 dmu_buf_rele(db, FTAG);
1794 VERIFY3U(0, ==, zap_add(os, lr->lr_doid, name, sizeof (uint64_t), 1,
1797 (void) ztest_log_create(zd, tx, lr);
1805 ztest_replay_remove(void *arg1, void *arg2, boolean_t byteswap)
1807 ztest_ds_t *zd = arg1;
1808 lr_remove_t *lr = arg2;
1809 char *name = (void *)(lr + 1); /* name follows lr */
1810 objset_t *os = zd->zd_os;
1811 dmu_object_info_t doi;
1813 uint64_t object, txg;
1816 byteswap_uint64_array(lr, sizeof (*lr));
1818 ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1819 ASSERT(name[0] != '\0');
1822 zap_lookup(os, lr->lr_doid, name, sizeof (object), 1, &object));
1823 ASSERT(object != 0);
1825 ztest_object_lock(zd, object, RL_WRITER);
1827 VERIFY3U(0, ==, dmu_object_info(os, object, &doi));
1829 tx = dmu_tx_create(os);
1831 dmu_tx_hold_zap(tx, lr->lr_doid, B_FALSE, name);
1832 dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END);
1834 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1836 ztest_object_unlock(zd, object);
1840 if (doi.doi_type == DMU_OT_ZAP_OTHER) {
1841 VERIFY3U(0, ==, zap_destroy(os, object, tx));
1843 VERIFY3U(0, ==, dmu_object_free(os, object, tx));
1846 VERIFY3U(0, ==, zap_remove(os, lr->lr_doid, name, tx));
1848 (void) ztest_log_remove(zd, tx, lr, object);
1852 ztest_object_unlock(zd, object);
1858 ztest_replay_write(void *arg1, void *arg2, boolean_t byteswap)
1860 ztest_ds_t *zd = arg1;
1861 lr_write_t *lr = arg2;
1862 objset_t *os = zd->zd_os;
1863 void *data = lr + 1; /* data follows lr */
1864 uint64_t offset, length;
1865 ztest_block_tag_t *bt = data;
1866 ztest_block_tag_t *bbt;
1867 uint64_t gen, txg, lrtxg, crtxg;
1868 dmu_object_info_t doi;
1871 arc_buf_t *abuf = NULL;
1875 byteswap_uint64_array(lr, sizeof (*lr));
1877 offset = lr->lr_offset;
1878 length = lr->lr_length;
1880 /* If it's a dmu_sync() block, write the whole block */
1881 if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
1882 uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
1883 if (length < blocksize) {
1884 offset -= offset % blocksize;
1889 if (bt->bt_magic == BSWAP_64(BT_MAGIC))
1890 byteswap_uint64_array(bt, sizeof (*bt));
1892 if (bt->bt_magic != BT_MAGIC)
1895 ztest_object_lock(zd, lr->lr_foid, RL_READER);
1896 rl = ztest_range_lock(zd, lr->lr_foid, offset, length, RL_WRITER);
1898 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1900 dmu_object_info_from_db(db, &doi);
1902 bbt = ztest_bt_bonus(db);
1903 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1905 crtxg = bbt->bt_crtxg;
1906 lrtxg = lr->lr_common.lrc_txg;
1908 tx = dmu_tx_create(os);
1910 dmu_tx_hold_write(tx, lr->lr_foid, offset, length);
1912 if (ztest_random(8) == 0 && length == doi.doi_data_block_size &&
1913 P2PHASE(offset, length) == 0)
1914 abuf = dmu_request_arcbuf(db, length);
1916 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1919 dmu_return_arcbuf(abuf);
1920 dmu_buf_rele(db, FTAG);
1921 ztest_range_unlock(zd, rl);
1922 ztest_object_unlock(zd, lr->lr_foid);
1928 * Usually, verify the old data before writing new data --
1929 * but not always, because we also want to verify correct
1930 * behavior when the data was not recently read into cache.
1932 ASSERT(offset % doi.doi_data_block_size == 0);
1933 if (ztest_random(4) != 0) {
1934 int prefetch = ztest_random(2) ?
1935 DMU_READ_PREFETCH : DMU_READ_NO_PREFETCH;
1936 ztest_block_tag_t rbt;
1938 VERIFY(dmu_read(os, lr->lr_foid, offset,
1939 sizeof (rbt), &rbt, prefetch) == 0);
1940 if (rbt.bt_magic == BT_MAGIC) {
1941 ztest_bt_verify(&rbt, os, lr->lr_foid, 0,
1942 offset, gen, txg, crtxg);
1947 * Writes can appear to be newer than the bonus buffer because
1948 * the ztest_get_data() callback does a dmu_read() of the
1949 * open-context data, which may be different than the data
1950 * as it was when the write was generated.
1952 if (zd->zd_zilog->zl_replay) {
1953 ztest_bt_verify(bt, os, lr->lr_foid, 0, offset,
1954 MAX(gen, bt->bt_gen), MAX(txg, lrtxg),
1959 * Set the bt's gen/txg to the bonus buffer's gen/txg
1960 * so that all of the usual ASSERTs will work.
1962 ztest_bt_generate(bt, os, lr->lr_foid, 0, offset, gen, txg,
1967 dmu_write(os, lr->lr_foid, offset, length, data, tx);
1969 bcopy(data, abuf->b_data, length);
1970 dmu_assign_arcbuf_by_dbuf(db, offset, abuf, tx);
1973 (void) ztest_log_write(zd, tx, lr);
1975 dmu_buf_rele(db, FTAG);
1979 ztest_range_unlock(zd, rl);
1980 ztest_object_unlock(zd, lr->lr_foid);
1986 ztest_replay_truncate(void *arg1, void *arg2, boolean_t byteswap)
1988 ztest_ds_t *zd = arg1;
1989 lr_truncate_t *lr = arg2;
1990 objset_t *os = zd->zd_os;
1996 byteswap_uint64_array(lr, sizeof (*lr));
1998 ztest_object_lock(zd, lr->lr_foid, RL_READER);
1999 rl = ztest_range_lock(zd, lr->lr_foid, lr->lr_offset, lr->lr_length,
2002 tx = dmu_tx_create(os);
2004 dmu_tx_hold_free(tx, lr->lr_foid, lr->lr_offset, lr->lr_length);
2006 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
2008 ztest_range_unlock(zd, rl);
2009 ztest_object_unlock(zd, lr->lr_foid);
2013 VERIFY(dmu_free_range(os, lr->lr_foid, lr->lr_offset,
2014 lr->lr_length, tx) == 0);
2016 (void) ztest_log_truncate(zd, tx, lr);
2020 ztest_range_unlock(zd, rl);
2021 ztest_object_unlock(zd, lr->lr_foid);
2027 ztest_replay_setattr(void *arg1, void *arg2, boolean_t byteswap)
2029 ztest_ds_t *zd = arg1;
2030 lr_setattr_t *lr = arg2;
2031 objset_t *os = zd->zd_os;
2034 ztest_block_tag_t *bbt;
2035 uint64_t txg, lrtxg, crtxg, dnodesize;
2038 byteswap_uint64_array(lr, sizeof (*lr));
2040 ztest_object_lock(zd, lr->lr_foid, RL_WRITER);
2042 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
2044 tx = dmu_tx_create(os);
2045 dmu_tx_hold_bonus(tx, lr->lr_foid);
2047 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
2049 dmu_buf_rele(db, FTAG);
2050 ztest_object_unlock(zd, lr->lr_foid);
2054 bbt = ztest_bt_bonus(db);
2055 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
2056 crtxg = bbt->bt_crtxg;
2057 lrtxg = lr->lr_common.lrc_txg;
2058 dnodesize = bbt->bt_dnodesize;
2060 if (zd->zd_zilog->zl_replay) {
2061 ASSERT(lr->lr_size != 0);
2062 ASSERT(lr->lr_mode != 0);
2066 * Randomly change the size and increment the generation.
2068 lr->lr_size = (ztest_random(db->db_size / sizeof (*bbt)) + 1) *
2070 lr->lr_mode = bbt->bt_gen + 1;
2075 * Verify that the current bonus buffer is not newer than our txg.
2077 ztest_bt_verify(bbt, os, lr->lr_foid, dnodesize, -1ULL, lr->lr_mode,
2078 MAX(txg, lrtxg), crtxg);
2080 dmu_buf_will_dirty(db, tx);
2082 ASSERT3U(lr->lr_size, >=, sizeof (*bbt));
2083 ASSERT3U(lr->lr_size, <=, db->db_size);
2084 VERIFY0(dmu_set_bonus(db, lr->lr_size, tx));
2085 bbt = ztest_bt_bonus(db);
2087 ztest_bt_generate(bbt, os, lr->lr_foid, dnodesize, -1ULL, lr->lr_mode,
2089 ztest_fill_unused_bonus(db, bbt, lr->lr_foid, os, bbt->bt_gen);
2090 dmu_buf_rele(db, FTAG);
2092 (void) ztest_log_setattr(zd, tx, lr);
2096 ztest_object_unlock(zd, lr->lr_foid);
2101 zil_replay_func_t *ztest_replay_vector[TX_MAX_TYPE] = {
2102 NULL, /* 0 no such transaction type */
2103 ztest_replay_create, /* TX_CREATE */
2104 NULL, /* TX_MKDIR */
2105 NULL, /* TX_MKXATTR */
2106 NULL, /* TX_SYMLINK */
2107 ztest_replay_remove, /* TX_REMOVE */
2108 NULL, /* TX_RMDIR */
2110 NULL, /* TX_RENAME */
2111 ztest_replay_write, /* TX_WRITE */
2112 ztest_replay_truncate, /* TX_TRUNCATE */
2113 ztest_replay_setattr, /* TX_SETATTR */
2115 NULL, /* TX_CREATE_ACL */
2116 NULL, /* TX_CREATE_ATTR */
2117 NULL, /* TX_CREATE_ACL_ATTR */
2118 NULL, /* TX_MKDIR_ACL */
2119 NULL, /* TX_MKDIR_ATTR */
2120 NULL, /* TX_MKDIR_ACL_ATTR */
2121 NULL, /* TX_WRITE2 */
2125 * ZIL get_data callbacks
2127 typedef struct ztest_zgd_private {
2131 } ztest_zgd_private_t;
2134 ztest_get_done(zgd_t *zgd, int error)
2136 ztest_zgd_private_t *zzp = zgd->zgd_private;
2137 ztest_ds_t *zd = zzp->z_zd;
2138 uint64_t object = zzp->z_object;
2141 dmu_buf_rele(zgd->zgd_db, zgd);
2143 ztest_range_unlock(zd, zzp->z_rl);
2144 ztest_object_unlock(zd, object);
2146 if (error == 0 && zgd->zgd_bp)
2147 zil_lwb_add_block(zgd->zgd_lwb, zgd->zgd_bp);
2149 umem_free(zgd, sizeof (*zgd));
2150 umem_free(zzp, sizeof (*zzp));
2154 ztest_get_data(void *arg, lr_write_t *lr, char *buf, struct lwb *lwb,
2157 ztest_ds_t *zd = arg;
2158 objset_t *os = zd->zd_os;
2159 uint64_t object = lr->lr_foid;
2160 uint64_t offset = lr->lr_offset;
2161 uint64_t size = lr->lr_length;
2162 uint64_t txg = lr->lr_common.lrc_txg;
2164 dmu_object_info_t doi;
2168 ztest_zgd_private_t *zgd_private;
2170 ASSERT3P(lwb, !=, NULL);
2171 ASSERT3P(zio, !=, NULL);
2172 ASSERT3U(size, !=, 0);
2174 ztest_object_lock(zd, object, RL_READER);
2175 error = dmu_bonus_hold(os, object, FTAG, &db);
2177 ztest_object_unlock(zd, object);
2181 crtxg = ztest_bt_bonus(db)->bt_crtxg;
2183 if (crtxg == 0 || crtxg > txg) {
2184 dmu_buf_rele(db, FTAG);
2185 ztest_object_unlock(zd, object);
2189 dmu_object_info_from_db(db, &doi);
2190 dmu_buf_rele(db, FTAG);
2193 zgd = umem_zalloc(sizeof (*zgd), UMEM_NOFAIL);
2195 zgd_private = umem_zalloc(sizeof (ztest_zgd_private_t), UMEM_NOFAIL);
2196 zgd_private->z_zd = zd;
2197 zgd_private->z_object = object;
2198 zgd->zgd_private = zgd_private;
2200 if (buf != NULL) { /* immediate write */
2201 zgd_private->z_rl = ztest_range_lock(zd, object, offset, size,
2203 zgd->zgd_rl = zgd_private->z_rl->z_rl;
2205 error = dmu_read(os, object, offset, size, buf,
2206 DMU_READ_NO_PREFETCH);
2209 size = doi.doi_data_block_size;
2211 offset = P2ALIGN(offset, size);
2213 ASSERT(offset < size);
2217 zgd_private->z_rl = ztest_range_lock(zd, object, offset, size,
2219 zgd->zgd_rl = zgd_private->z_rl->z_rl;
2221 error = dmu_buf_hold(os, object, offset, zgd, &db,
2222 DMU_READ_NO_PREFETCH);
2225 blkptr_t *bp = &lr->lr_blkptr;
2230 ASSERT(db->db_offset == offset);
2231 ASSERT(db->db_size == size);
2233 error = dmu_sync(zio, lr->lr_common.lrc_txg,
2234 ztest_get_done, zgd);
2241 ztest_get_done(zgd, error);
2247 ztest_lr_alloc(size_t lrsize, char *name)
2250 size_t namesize = name ? strlen(name) + 1 : 0;
2252 lr = umem_zalloc(lrsize + namesize, UMEM_NOFAIL);
2255 bcopy(name, lr + lrsize, namesize);
2261 ztest_lr_free(void *lr, size_t lrsize, char *name)
2263 size_t namesize = name ? strlen(name) + 1 : 0;
2265 umem_free(lr, lrsize + namesize);
2269 * Lookup a bunch of objects. Returns the number of objects not found.
2272 ztest_lookup(ztest_ds_t *zd, ztest_od_t *od, int count)
2278 ASSERT(MUTEX_HELD(&zd->zd_dirobj_lock));
2280 for (i = 0; i < count; i++, od++) {
2282 error = zap_lookup(zd->zd_os, od->od_dir, od->od_name,
2283 sizeof (uint64_t), 1, &od->od_object);
2285 ASSERT(error == ENOENT);
2286 ASSERT(od->od_object == 0);
2290 ztest_block_tag_t *bbt;
2291 dmu_object_info_t doi;
2293 ASSERT(od->od_object != 0);
2294 ASSERT(missing == 0); /* there should be no gaps */
2296 ztest_object_lock(zd, od->od_object, RL_READER);
2297 VERIFY3U(0, ==, dmu_bonus_hold(zd->zd_os,
2298 od->od_object, FTAG, &db));
2299 dmu_object_info_from_db(db, &doi);
2300 bbt = ztest_bt_bonus(db);
2301 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
2302 od->od_type = doi.doi_type;
2303 od->od_blocksize = doi.doi_data_block_size;
2304 od->od_gen = bbt->bt_gen;
2305 dmu_buf_rele(db, FTAG);
2306 ztest_object_unlock(zd, od->od_object);
2314 ztest_create(ztest_ds_t *zd, ztest_od_t *od, int count)
2319 ASSERT(MUTEX_HELD(&zd->zd_dirobj_lock));
2321 for (i = 0; i < count; i++, od++) {
2328 lr_create_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
2330 lr->lr_doid = od->od_dir;
2331 lr->lr_foid = 0; /* 0 to allocate, > 0 to claim */
2332 lr->lrz_type = od->od_crtype;
2333 lr->lrz_blocksize = od->od_crblocksize;
2334 lr->lrz_ibshift = ztest_random_ibshift();
2335 lr->lrz_bonustype = DMU_OT_UINT64_OTHER;
2336 lr->lrz_dnodesize = od->od_crdnodesize;
2337 lr->lr_gen = od->od_crgen;
2338 lr->lr_crtime[0] = time(NULL);
2340 if (ztest_replay_create(zd, lr, B_FALSE) != 0) {
2341 ASSERT(missing == 0);
2345 od->od_object = lr->lr_foid;
2346 od->od_type = od->od_crtype;
2347 od->od_blocksize = od->od_crblocksize;
2348 od->od_gen = od->od_crgen;
2349 ASSERT(od->od_object != 0);
2352 ztest_lr_free(lr, sizeof (*lr), od->od_name);
2359 ztest_remove(ztest_ds_t *zd, ztest_od_t *od, int count)
2365 ASSERT(MUTEX_HELD(&zd->zd_dirobj_lock));
2369 for (i = count - 1; i >= 0; i--, od--) {
2376 * No object was found.
2378 if (od->od_object == 0)
2381 lr_remove_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
2383 lr->lr_doid = od->od_dir;
2385 if ((error = ztest_replay_remove(zd, lr, B_FALSE)) != 0) {
2386 ASSERT3U(error, ==, ENOSPC);
2391 ztest_lr_free(lr, sizeof (*lr), od->od_name);
2398 ztest_write(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size,
2404 lr = ztest_lr_alloc(sizeof (*lr) + size, NULL);
2406 lr->lr_foid = object;
2407 lr->lr_offset = offset;
2408 lr->lr_length = size;
2410 BP_ZERO(&lr->lr_blkptr);
2412 bcopy(data, lr + 1, size);
2414 error = ztest_replay_write(zd, lr, B_FALSE);
2416 ztest_lr_free(lr, sizeof (*lr) + size, NULL);
2422 ztest_truncate(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2427 lr = ztest_lr_alloc(sizeof (*lr), NULL);
2429 lr->lr_foid = object;
2430 lr->lr_offset = offset;
2431 lr->lr_length = size;
2433 error = ztest_replay_truncate(zd, lr, B_FALSE);
2435 ztest_lr_free(lr, sizeof (*lr), NULL);
2441 ztest_setattr(ztest_ds_t *zd, uint64_t object)
2446 lr = ztest_lr_alloc(sizeof (*lr), NULL);
2448 lr->lr_foid = object;
2452 error = ztest_replay_setattr(zd, lr, B_FALSE);
2454 ztest_lr_free(lr, sizeof (*lr), NULL);
2460 ztest_prealloc(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2462 objset_t *os = zd->zd_os;
2467 txg_wait_synced(dmu_objset_pool(os), 0);
2469 ztest_object_lock(zd, object, RL_READER);
2470 rl = ztest_range_lock(zd, object, offset, size, RL_WRITER);
2472 tx = dmu_tx_create(os);
2474 dmu_tx_hold_write(tx, object, offset, size);
2476 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
2479 dmu_prealloc(os, object, offset, size, tx);
2481 txg_wait_synced(dmu_objset_pool(os), txg);
2483 (void) dmu_free_long_range(os, object, offset, size);
2486 ztest_range_unlock(zd, rl);
2487 ztest_object_unlock(zd, object);
2491 ztest_io(ztest_ds_t *zd, uint64_t object, uint64_t offset)
2494 ztest_block_tag_t wbt;
2495 dmu_object_info_t doi;
2496 enum ztest_io_type io_type;
2500 VERIFY(dmu_object_info(zd->zd_os, object, &doi) == 0);
2501 blocksize = doi.doi_data_block_size;
2502 data = umem_alloc(blocksize, UMEM_NOFAIL);
2505 * Pick an i/o type at random, biased toward writing block tags.
2507 io_type = ztest_random(ZTEST_IO_TYPES);
2508 if (ztest_random(2) == 0)
2509 io_type = ZTEST_IO_WRITE_TAG;
2511 (void) rw_rdlock(&zd->zd_zilog_lock);
2515 case ZTEST_IO_WRITE_TAG:
2516 ztest_bt_generate(&wbt, zd->zd_os, object, doi.doi_dnodesize,
2518 (void) ztest_write(zd, object, offset, sizeof (wbt), &wbt);
2521 case ZTEST_IO_WRITE_PATTERN:
2522 (void) memset(data, 'a' + (object + offset) % 5, blocksize);
2523 if (ztest_random(2) == 0) {
2525 * Induce fletcher2 collisions to ensure that
2526 * zio_ddt_collision() detects and resolves them
2527 * when using fletcher2-verify for deduplication.
2529 ((uint64_t *)data)[0] ^= 1ULL << 63;
2530 ((uint64_t *)data)[4] ^= 1ULL << 63;
2532 (void) ztest_write(zd, object, offset, blocksize, data);
2535 case ZTEST_IO_WRITE_ZEROES:
2536 bzero(data, blocksize);
2537 (void) ztest_write(zd, object, offset, blocksize, data);
2540 case ZTEST_IO_TRUNCATE:
2541 (void) ztest_truncate(zd, object, offset, blocksize);
2544 case ZTEST_IO_SETATTR:
2545 (void) ztest_setattr(zd, object);
2550 case ZTEST_IO_REWRITE:
2551 (void) rw_rdlock(&ztest_name_lock);
2552 err = ztest_dsl_prop_set_uint64(zd->zd_name,
2553 ZFS_PROP_CHECKSUM, spa_dedup_checksum(ztest_spa),
2555 VERIFY(err == 0 || err == ENOSPC);
2556 err = ztest_dsl_prop_set_uint64(zd->zd_name,
2557 ZFS_PROP_COMPRESSION,
2558 ztest_random_dsl_prop(ZFS_PROP_COMPRESSION),
2560 VERIFY(err == 0 || err == ENOSPC);
2561 (void) rw_unlock(&ztest_name_lock);
2563 VERIFY0(dmu_read(zd->zd_os, object, offset, blocksize, data,
2564 DMU_READ_NO_PREFETCH));
2566 (void) ztest_write(zd, object, offset, blocksize, data);
2570 (void) rw_unlock(&zd->zd_zilog_lock);
2572 umem_free(data, blocksize);
2576 * Initialize an object description template.
2579 ztest_od_init(ztest_od_t *od, uint64_t id, char *tag, uint64_t index,
2580 dmu_object_type_t type, uint64_t blocksize, uint64_t dnodesize,
2583 od->od_dir = ZTEST_DIROBJ;
2586 od->od_crtype = type;
2587 od->od_crblocksize = blocksize ? blocksize : ztest_random_blocksize();
2588 od->od_crdnodesize = dnodesize ? dnodesize : ztest_random_dnodesize();
2591 od->od_type = DMU_OT_NONE;
2592 od->od_blocksize = 0;
2595 (void) snprintf(od->od_name, sizeof (od->od_name), "%s(%lld)[%llu]",
2596 tag, (longlong_t)id, (u_longlong_t)index);
2600 * Lookup or create the objects for a test using the od template.
2601 * If the objects do not all exist, or if 'remove' is specified,
2602 * remove any existing objects and create new ones. Otherwise,
2603 * use the existing objects.
2606 ztest_object_init(ztest_ds_t *zd, ztest_od_t *od, size_t size, boolean_t remove)
2608 int count = size / sizeof (*od);
2611 mutex_enter(&zd->zd_dirobj_lock);
2612 if ((ztest_lookup(zd, od, count) != 0 || remove) &&
2613 (ztest_remove(zd, od, count) != 0 ||
2614 ztest_create(zd, od, count) != 0))
2617 mutex_exit(&zd->zd_dirobj_lock);
2624 ztest_zil_commit(ztest_ds_t *zd, uint64_t id)
2626 zilog_t *zilog = zd->zd_zilog;
2628 (void) rw_rdlock(&zd->zd_zilog_lock);
2630 zil_commit(zilog, ztest_random(ZTEST_OBJECTS));
2633 * Remember the committed values in zd, which is in parent/child
2634 * shared memory. If we die, the next iteration of ztest_run()
2635 * will verify that the log really does contain this record.
2637 mutex_enter(&zilog->zl_lock);
2638 ASSERT(zd->zd_shared != NULL);
2639 ASSERT3U(zd->zd_shared->zd_seq, <=, zilog->zl_commit_lr_seq);
2640 zd->zd_shared->zd_seq = zilog->zl_commit_lr_seq;
2641 mutex_exit(&zilog->zl_lock);
2643 (void) rw_unlock(&zd->zd_zilog_lock);
2647 * This function is designed to simulate the operations that occur during a
2648 * mount/unmount operation. We hold the dataset across these operations in an
2649 * attempt to expose any implicit assumptions about ZIL management.
2653 ztest_zil_remount(ztest_ds_t *zd, uint64_t id)
2655 objset_t *os = zd->zd_os;
2658 * We grab the zd_dirobj_lock to ensure that no other thread is
2659 * updating the zil (i.e. adding in-memory log records) and the
2660 * zd_zilog_lock to block any I/O.
2662 mutex_enter(&zd->zd_dirobj_lock);
2663 (void) rw_wrlock(&zd->zd_zilog_lock);
2665 /* zfsvfs_teardown() */
2666 zil_close(zd->zd_zilog);
2668 /* zfsvfs_setup() */
2669 VERIFY(zil_open(os, ztest_get_data) == zd->zd_zilog);
2670 zil_replay(os, zd, ztest_replay_vector);
2672 (void) rw_unlock(&zd->zd_zilog_lock);
2673 mutex_exit(&zd->zd_dirobj_lock);
2677 * Verify that we can't destroy an active pool, create an existing pool,
2678 * or create a pool with a bad vdev spec.
2682 ztest_spa_create_destroy(ztest_ds_t *zd, uint64_t id)
2684 ztest_shared_opts_t *zo = &ztest_opts;
2688 if (zo->zo_mmp_test)
2692 * Attempt to create using a bad file.
2694 nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1);
2695 VERIFY3U(ENOENT, ==,
2696 spa_create("ztest_bad_file", nvroot, NULL, NULL, NULL));
2697 nvlist_free(nvroot);
2700 * Attempt to create using a bad mirror.
2702 nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 2, 1);
2703 VERIFY3U(ENOENT, ==,
2704 spa_create("ztest_bad_mirror", nvroot, NULL, NULL, NULL));
2705 nvlist_free(nvroot);
2708 * Attempt to create an existing pool. It shouldn't matter
2709 * what's in the nvroot; we should fail with EEXIST.
2711 (void) rw_rdlock(&ztest_name_lock);
2712 nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1);
2713 VERIFY3U(EEXIST, ==,
2714 spa_create(zo->zo_pool, nvroot, NULL, NULL, NULL));
2715 nvlist_free(nvroot);
2716 VERIFY3U(0, ==, spa_open(zo->zo_pool, &spa, FTAG));
2717 VERIFY3U(EBUSY, ==, spa_destroy(zo->zo_pool));
2718 spa_close(spa, FTAG);
2720 (void) rw_unlock(&ztest_name_lock);
2724 * Start and then stop the MMP threads to ensure the startup and shutdown code
2725 * works properly. Actual protection and property-related code tested via ZTS.
2729 ztest_mmp_enable_disable(ztest_ds_t *zd, uint64_t id)
2731 ztest_shared_opts_t *zo = &ztest_opts;
2732 spa_t *spa = ztest_spa;
2734 if (zo->zo_mmp_test)
2737 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
2738 mutex_enter(&spa->spa_props_lock);
2740 if (!spa_multihost(spa)) {
2741 spa->spa_multihost = B_TRUE;
2742 mmp_thread_start(spa);
2745 mutex_exit(&spa->spa_props_lock);
2746 spa_config_exit(spa, SCL_CONFIG, FTAG);
2748 txg_wait_synced(spa_get_dsl(spa), 0);
2749 mmp_signal_all_threads();
2750 txg_wait_synced(spa_get_dsl(spa), 0);
2752 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
2753 mutex_enter(&spa->spa_props_lock);
2755 if (spa_multihost(spa)) {
2756 mmp_thread_stop(spa);
2757 spa->spa_multihost = B_FALSE;
2760 mutex_exit(&spa->spa_props_lock);
2761 spa_config_exit(spa, SCL_CONFIG, FTAG);
2766 ztest_spa_upgrade(ztest_ds_t *zd, uint64_t id)
2769 uint64_t initial_version = SPA_VERSION_INITIAL;
2770 uint64_t version, newversion;
2771 nvlist_t *nvroot, *props;
2774 if (ztest_opts.zo_mmp_test)
2777 mutex_enter(&ztest_vdev_lock);
2778 name = kmem_asprintf("%s_upgrade", ztest_opts.zo_pool);
2781 * Clean up from previous runs.
2783 (void) spa_destroy(name);
2785 nvroot = make_vdev_root(NULL, NULL, name, ztest_opts.zo_vdev_size, 0,
2786 0, ztest_opts.zo_raidz, ztest_opts.zo_mirrors, 1);
2789 * If we're configuring a RAIDZ device then make sure that the
2790 * the initial version is capable of supporting that feature.
2792 switch (ztest_opts.zo_raidz_parity) {
2795 initial_version = SPA_VERSION_INITIAL;
2798 initial_version = SPA_VERSION_RAIDZ2;
2801 initial_version = SPA_VERSION_RAIDZ3;
2806 * Create a pool with a spa version that can be upgraded. Pick
2807 * a value between initial_version and SPA_VERSION_BEFORE_FEATURES.
2810 version = ztest_random_spa_version(initial_version);
2811 } while (version > SPA_VERSION_BEFORE_FEATURES);
2813 props = fnvlist_alloc();
2814 fnvlist_add_uint64(props,
2815 zpool_prop_to_name(ZPOOL_PROP_VERSION), version);
2816 VERIFY3S(spa_create(name, nvroot, props, NULL, NULL), ==, 0);
2817 fnvlist_free(nvroot);
2818 fnvlist_free(props);
2820 VERIFY3S(spa_open(name, &spa, FTAG), ==, 0);
2821 VERIFY3U(spa_version(spa), ==, version);
2822 newversion = ztest_random_spa_version(version + 1);
2824 if (ztest_opts.zo_verbose >= 4) {
2825 (void) printf("upgrading spa version from %llu to %llu\n",
2826 (u_longlong_t)version, (u_longlong_t)newversion);
2829 spa_upgrade(spa, newversion);
2830 VERIFY3U(spa_version(spa), >, version);
2831 VERIFY3U(spa_version(spa), ==, fnvlist_lookup_uint64(spa->spa_config,
2832 zpool_prop_to_name(ZPOOL_PROP_VERSION)));
2833 spa_close(spa, FTAG);
2836 mutex_exit(&ztest_vdev_lock);
2840 vdev_lookup_by_path(vdev_t *vd, const char *path)
2845 if (vd->vdev_path != NULL && strcmp(path, vd->vdev_path) == 0)
2848 for (c = 0; c < vd->vdev_children; c++)
2849 if ((mvd = vdev_lookup_by_path(vd->vdev_child[c], path)) !=
2857 * Find the first available hole which can be used as a top-level.
2860 find_vdev_hole(spa_t *spa)
2862 vdev_t *rvd = spa->spa_root_vdev;
2865 ASSERT(spa_config_held(spa, SCL_VDEV, RW_READER) == SCL_VDEV);
2867 for (c = 0; c < rvd->vdev_children; c++) {
2868 vdev_t *cvd = rvd->vdev_child[c];
2870 if (cvd->vdev_ishole)
2877 * Verify that vdev_add() works as expected.
2881 ztest_vdev_add_remove(ztest_ds_t *zd, uint64_t id)
2883 ztest_shared_t *zs = ztest_shared;
2884 spa_t *spa = ztest_spa;
2890 if (ztest_opts.zo_mmp_test)
2893 mutex_enter(&ztest_vdev_lock);
2894 leaves = MAX(zs->zs_mirrors + zs->zs_splits, 1) * ztest_opts.zo_raidz;
2896 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2898 ztest_shared->zs_vdev_next_leaf = find_vdev_hole(spa) * leaves;
2901 * If we have slogs then remove them 1/4 of the time.
2903 if (spa_has_slogs(spa) && ztest_random(4) == 0) {
2905 * Grab the guid from the head of the log class rotor.
2907 guid = spa_log_class(spa)->mc_rotor->mg_vd->vdev_guid;
2909 spa_config_exit(spa, SCL_VDEV, FTAG);
2912 * We have to grab the zs_name_lock as writer to
2913 * prevent a race between removing a slog (dmu_objset_find)
2914 * and destroying a dataset. Removing the slog will
2915 * grab a reference on the dataset which may cause
2916 * dsl_destroy_head() to fail with EBUSY thus
2917 * leaving the dataset in an inconsistent state.
2919 rw_wrlock(&ztest_name_lock);
2920 error = spa_vdev_remove(spa, guid, B_FALSE);
2921 rw_unlock(&ztest_name_lock);
2923 if (error && error != EEXIST)
2924 fatal(0, "spa_vdev_remove() = %d", error);
2926 spa_config_exit(spa, SCL_VDEV, FTAG);
2929 * Make 1/4 of the devices be log devices.
2931 nvroot = make_vdev_root(NULL, NULL, NULL,
2932 ztest_opts.zo_vdev_size, 0,
2933 ztest_random(4) == 0, ztest_opts.zo_raidz,
2936 error = spa_vdev_add(spa, nvroot);
2937 nvlist_free(nvroot);
2939 if (error == ENOSPC)
2940 ztest_record_enospc("spa_vdev_add");
2941 else if (error != 0)
2942 fatal(0, "spa_vdev_add() = %d", error);
2945 mutex_exit(&ztest_vdev_lock);
2949 * Verify that adding/removing aux devices (l2arc, hot spare) works as expected.
2953 ztest_vdev_aux_add_remove(ztest_ds_t *zd, uint64_t id)
2955 ztest_shared_t *zs = ztest_shared;
2956 spa_t *spa = ztest_spa;
2957 vdev_t *rvd = spa->spa_root_vdev;
2958 spa_aux_vdev_t *sav;
2964 if (ztest_opts.zo_mmp_test)
2967 path = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
2969 if (ztest_random(2) == 0) {
2970 sav = &spa->spa_spares;
2971 aux = ZPOOL_CONFIG_SPARES;
2973 sav = &spa->spa_l2cache;
2974 aux = ZPOOL_CONFIG_L2CACHE;
2977 mutex_enter(&ztest_vdev_lock);
2979 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2981 if (sav->sav_count != 0 && ztest_random(4) == 0) {
2983 * Pick a random device to remove.
2985 guid = sav->sav_vdevs[ztest_random(sav->sav_count)]->vdev_guid;
2988 * Find an unused device we can add.
2990 zs->zs_vdev_aux = 0;
2993 (void) snprintf(path, MAXPATHLEN, ztest_aux_template,
2994 ztest_opts.zo_dir, ztest_opts.zo_pool, aux,
2996 for (c = 0; c < sav->sav_count; c++)
2997 if (strcmp(sav->sav_vdevs[c]->vdev_path,
3000 if (c == sav->sav_count &&
3001 vdev_lookup_by_path(rvd, path) == NULL)
3007 spa_config_exit(spa, SCL_VDEV, FTAG);
3013 nvlist_t *nvroot = make_vdev_root(NULL, aux, NULL,
3014 (ztest_opts.zo_vdev_size * 5) / 4, 0, 0, 0, 0, 1);
3015 error = spa_vdev_add(spa, nvroot);
3017 fatal(0, "spa_vdev_add(%p) = %d", nvroot, error);
3018 nvlist_free(nvroot);
3021 * Remove an existing device. Sometimes, dirty its
3022 * vdev state first to make sure we handle removal
3023 * of devices that have pending state changes.
3025 if (ztest_random(2) == 0)
3026 (void) vdev_online(spa, guid, 0, NULL);
3028 error = spa_vdev_remove(spa, guid, B_FALSE);
3029 if (error != 0 && error != EBUSY)
3030 fatal(0, "spa_vdev_remove(%llu) = %d", guid, error);
3033 mutex_exit(&ztest_vdev_lock);
3035 umem_free(path, MAXPATHLEN);
3039 * split a pool if it has mirror tlvdevs
3043 ztest_split_pool(ztest_ds_t *zd, uint64_t id)
3045 ztest_shared_t *zs = ztest_shared;
3046 spa_t *spa = ztest_spa;
3047 vdev_t *rvd = spa->spa_root_vdev;
3048 nvlist_t *tree, **child, *config, *split, **schild;
3049 uint_t c, children, schildren = 0, lastlogid = 0;
3052 if (ztest_opts.zo_mmp_test)
3055 mutex_enter(&ztest_vdev_lock);
3057 /* ensure we have a useable config; mirrors of raidz aren't supported */
3058 if (zs->zs_mirrors < 3 || ztest_opts.zo_raidz > 1) {
3059 mutex_exit(&ztest_vdev_lock);
3063 /* clean up the old pool, if any */
3064 (void) spa_destroy("splitp");
3066 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
3068 /* generate a config from the existing config */
3069 mutex_enter(&spa->spa_props_lock);
3070 VERIFY(nvlist_lookup_nvlist(spa->spa_config, ZPOOL_CONFIG_VDEV_TREE,
3072 mutex_exit(&spa->spa_props_lock);
3074 VERIFY(nvlist_lookup_nvlist_array(tree, ZPOOL_CONFIG_CHILDREN, &child,
3077 schild = malloc(rvd->vdev_children * sizeof (nvlist_t *));
3078 for (c = 0; c < children; c++) {
3079 vdev_t *tvd = rvd->vdev_child[c];
3083 if (tvd->vdev_islog || tvd->vdev_ops == &vdev_hole_ops) {
3084 VERIFY(nvlist_alloc(&schild[schildren], NV_UNIQUE_NAME,
3086 VERIFY(nvlist_add_string(schild[schildren],
3087 ZPOOL_CONFIG_TYPE, VDEV_TYPE_HOLE) == 0);
3088 VERIFY(nvlist_add_uint64(schild[schildren],
3089 ZPOOL_CONFIG_IS_HOLE, 1) == 0);
3091 lastlogid = schildren;
3096 VERIFY(nvlist_lookup_nvlist_array(child[c],
3097 ZPOOL_CONFIG_CHILDREN, &mchild, &mchildren) == 0);
3098 VERIFY(nvlist_dup(mchild[0], &schild[schildren++], 0) == 0);
3101 /* OK, create a config that can be used to split */
3102 VERIFY(nvlist_alloc(&split, NV_UNIQUE_NAME, 0) == 0);
3103 VERIFY(nvlist_add_string(split, ZPOOL_CONFIG_TYPE,
3104 VDEV_TYPE_ROOT) == 0);
3105 VERIFY(nvlist_add_nvlist_array(split, ZPOOL_CONFIG_CHILDREN, schild,
3106 lastlogid != 0 ? lastlogid : schildren) == 0);
3108 VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, 0) == 0);
3109 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, split) == 0);
3111 for (c = 0; c < schildren; c++)
3112 nvlist_free(schild[c]);
3116 spa_config_exit(spa, SCL_VDEV, FTAG);
3118 (void) rw_wrlock(&ztest_name_lock);
3119 error = spa_vdev_split_mirror(spa, "splitp", config, NULL, B_FALSE);
3120 (void) rw_unlock(&ztest_name_lock);
3122 nvlist_free(config);
3125 (void) printf("successful split - results:\n");
3126 mutex_enter(&spa_namespace_lock);
3127 show_pool_stats(spa);
3128 show_pool_stats(spa_lookup("splitp"));
3129 mutex_exit(&spa_namespace_lock);
3133 mutex_exit(&ztest_vdev_lock);
3138 * Verify that we can attach and detach devices.
3142 ztest_vdev_attach_detach(ztest_ds_t *zd, uint64_t id)
3144 ztest_shared_t *zs = ztest_shared;
3145 spa_t *spa = ztest_spa;
3146 spa_aux_vdev_t *sav = &spa->spa_spares;
3147 vdev_t *rvd = spa->spa_root_vdev;
3148 vdev_t *oldvd, *newvd, *pvd;
3152 uint64_t ashift = ztest_get_ashift();
3153 uint64_t oldguid, pguid;
3154 uint64_t oldsize, newsize;
3155 char *oldpath, *newpath;
3157 int oldvd_has_siblings = B_FALSE;
3158 int newvd_is_spare = B_FALSE;
3160 int error, expected_error;
3162 if (ztest_opts.zo_mmp_test)
3165 oldpath = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
3166 newpath = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
3168 mutex_enter(&ztest_vdev_lock);
3169 leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz;
3171 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
3174 * Decide whether to do an attach or a replace.
3176 replacing = ztest_random(2);
3179 * Pick a random top-level vdev.
3181 top = ztest_random_vdev_top(spa, B_TRUE);
3184 * Pick a random leaf within it.
3186 leaf = ztest_random(leaves);
3191 oldvd = rvd->vdev_child[top];
3192 if (zs->zs_mirrors >= 1) {
3193 ASSERT(oldvd->vdev_ops == &vdev_mirror_ops);
3194 ASSERT(oldvd->vdev_children >= zs->zs_mirrors);
3195 oldvd = oldvd->vdev_child[leaf / ztest_opts.zo_raidz];
3197 if (ztest_opts.zo_raidz > 1) {
3198 ASSERT(oldvd->vdev_ops == &vdev_raidz_ops);
3199 ASSERT(oldvd->vdev_children == ztest_opts.zo_raidz);
3200 oldvd = oldvd->vdev_child[leaf % ztest_opts.zo_raidz];
3204 * If we're already doing an attach or replace, oldvd may be a
3205 * mirror vdev -- in which case, pick a random child.
3207 while (oldvd->vdev_children != 0) {
3208 oldvd_has_siblings = B_TRUE;
3209 ASSERT(oldvd->vdev_children >= 2);
3210 oldvd = oldvd->vdev_child[ztest_random(oldvd->vdev_children)];
3213 oldguid = oldvd->vdev_guid;
3214 oldsize = vdev_get_min_asize(oldvd);
3215 oldvd_is_log = oldvd->vdev_top->vdev_islog;
3216 (void) strcpy(oldpath, oldvd->vdev_path);
3217 pvd = oldvd->vdev_parent;
3218 pguid = pvd->vdev_guid;
3221 * If oldvd has siblings, then half of the time, detach it.
3223 if (oldvd_has_siblings && ztest_random(2) == 0) {
3224 spa_config_exit(spa, SCL_VDEV, FTAG);
3225 error = spa_vdev_detach(spa, oldguid, pguid, B_FALSE);
3226 if (error != 0 && error != ENODEV && error != EBUSY &&
3228 fatal(0, "detach (%s) returned %d", oldpath, error);
3233 * For the new vdev, choose with equal probability between the two
3234 * standard paths (ending in either 'a' or 'b') or a random hot spare.
3236 if (sav->sav_count != 0 && ztest_random(3) == 0) {
3237 newvd = sav->sav_vdevs[ztest_random(sav->sav_count)];
3238 newvd_is_spare = B_TRUE;
3239 (void) strcpy(newpath, newvd->vdev_path);
3241 (void) snprintf(newpath, MAXPATHLEN, ztest_dev_template,
3242 ztest_opts.zo_dir, ztest_opts.zo_pool,
3243 top * leaves + leaf);
3244 if (ztest_random(2) == 0)
3245 newpath[strlen(newpath) - 1] = 'b';
3246 newvd = vdev_lookup_by_path(rvd, newpath);
3250 newsize = vdev_get_min_asize(newvd);
3253 * Make newsize a little bigger or smaller than oldsize.
3254 * If it's smaller, the attach should fail.
3255 * If it's larger, and we're doing a replace,
3256 * we should get dynamic LUN growth when we're done.
3258 newsize = 10 * oldsize / (9 + ztest_random(3));
3262 * If pvd is not a mirror or root, the attach should fail with ENOTSUP,
3263 * unless it's a replace; in that case any non-replacing parent is OK.
3265 * If newvd is already part of the pool, it should fail with EBUSY.
3267 * If newvd is too small, it should fail with EOVERFLOW.
3269 if (pvd->vdev_ops != &vdev_mirror_ops &&
3270 pvd->vdev_ops != &vdev_root_ops && (!replacing ||
3271 pvd->vdev_ops == &vdev_replacing_ops ||
3272 pvd->vdev_ops == &vdev_spare_ops))
3273 expected_error = ENOTSUP;
3274 else if (newvd_is_spare && (!replacing || oldvd_is_log))
3275 expected_error = ENOTSUP;
3276 else if (newvd == oldvd)
3277 expected_error = replacing ? 0 : EBUSY;
3278 else if (vdev_lookup_by_path(rvd, newpath) != NULL)
3279 expected_error = EBUSY;
3280 else if (newsize < oldsize)
3281 expected_error = EOVERFLOW;
3282 else if (ashift > oldvd->vdev_top->vdev_ashift)
3283 expected_error = EDOM;
3287 spa_config_exit(spa, SCL_VDEV, FTAG);
3290 * Build the nvlist describing newpath.
3292 root = make_vdev_root(newpath, NULL, NULL, newvd == NULL ? newsize : 0,
3293 ashift, 0, 0, 0, 1);
3295 error = spa_vdev_attach(spa, oldguid, root, replacing);
3300 * If our parent was the replacing vdev, but the replace completed,
3301 * then instead of failing with ENOTSUP we may either succeed,
3302 * fail with ENODEV, or fail with EOVERFLOW.
3304 if (expected_error == ENOTSUP &&
3305 (error == 0 || error == ENODEV || error == EOVERFLOW))
3306 expected_error = error;
3309 * If someone grew the LUN, the replacement may be too small.
3311 if (error == EOVERFLOW || error == EBUSY)
3312 expected_error = error;
3314 /* XXX workaround 6690467 */
3315 if (error != expected_error && expected_error != EBUSY) {
3316 fatal(0, "attach (%s %llu, %s %llu, %d) "
3317 "returned %d, expected %d",
3318 oldpath, oldsize, newpath,
3319 newsize, replacing, error, expected_error);
3322 mutex_exit(&ztest_vdev_lock);
3324 umem_free(oldpath, MAXPATHLEN);
3325 umem_free(newpath, MAXPATHLEN);
3329 * Callback function which expands the physical size of the vdev.
3332 grow_vdev(vdev_t *vd, void *arg)
3334 ASSERTV(spa_t *spa = vd->vdev_spa);
3335 size_t *newsize = arg;
3339 ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
3340 ASSERT(vd->vdev_ops->vdev_op_leaf);
3342 if ((fd = open(vd->vdev_path, O_RDWR)) == -1)
3345 fsize = lseek(fd, 0, SEEK_END);
3346 VERIFY(ftruncate(fd, *newsize) == 0);
3348 if (ztest_opts.zo_verbose >= 6) {
3349 (void) printf("%s grew from %lu to %lu bytes\n",
3350 vd->vdev_path, (ulong_t)fsize, (ulong_t)*newsize);
3357 * Callback function which expands a given vdev by calling vdev_online().
3361 online_vdev(vdev_t *vd, void *arg)
3363 spa_t *spa = vd->vdev_spa;
3364 vdev_t *tvd = vd->vdev_top;
3365 uint64_t guid = vd->vdev_guid;
3366 uint64_t generation = spa->spa_config_generation + 1;
3367 vdev_state_t newstate = VDEV_STATE_UNKNOWN;
3370 ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
3371 ASSERT(vd->vdev_ops->vdev_op_leaf);
3373 /* Calling vdev_online will initialize the new metaslabs */
3374 spa_config_exit(spa, SCL_STATE, spa);
3375 error = vdev_online(spa, guid, ZFS_ONLINE_EXPAND, &newstate);
3376 spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3379 * If vdev_online returned an error or the underlying vdev_open
3380 * failed then we abort the expand. The only way to know that
3381 * vdev_open fails is by checking the returned newstate.
3383 if (error || newstate != VDEV_STATE_HEALTHY) {
3384 if (ztest_opts.zo_verbose >= 5) {
3385 (void) printf("Unable to expand vdev, state %llu, "
3386 "error %d\n", (u_longlong_t)newstate, error);
3390 ASSERT3U(newstate, ==, VDEV_STATE_HEALTHY);
3393 * Since we dropped the lock we need to ensure that we're
3394 * still talking to the original vdev. It's possible this
3395 * vdev may have been detached/replaced while we were
3396 * trying to online it.
3398 if (generation != spa->spa_config_generation) {
3399 if (ztest_opts.zo_verbose >= 5) {
3400 (void) printf("vdev configuration has changed, "
3401 "guid %llu, state %llu, expected gen %llu, "
3404 (u_longlong_t)tvd->vdev_state,
3405 (u_longlong_t)generation,
3406 (u_longlong_t)spa->spa_config_generation);
3414 * Traverse the vdev tree calling the supplied function.
3415 * We continue to walk the tree until we either have walked all
3416 * children or we receive a non-NULL return from the callback.
3417 * If a NULL callback is passed, then we just return back the first
3418 * leaf vdev we encounter.
3421 vdev_walk_tree(vdev_t *vd, vdev_t *(*func)(vdev_t *, void *), void *arg)
3425 if (vd->vdev_ops->vdev_op_leaf) {
3429 return (func(vd, arg));
3432 for (c = 0; c < vd->vdev_children; c++) {
3433 vdev_t *cvd = vd->vdev_child[c];
3434 if ((cvd = vdev_walk_tree(cvd, func, arg)) != NULL)
3441 * Verify that dynamic LUN growth works as expected.
3445 ztest_vdev_LUN_growth(ztest_ds_t *zd, uint64_t id)
3447 spa_t *spa = ztest_spa;
3449 metaslab_class_t *mc;
3450 metaslab_group_t *mg;
3451 size_t psize, newsize;
3453 uint64_t old_class_space, new_class_space, old_ms_count, new_ms_count;
3455 mutex_enter(&ztest_vdev_lock);
3456 spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3458 top = ztest_random_vdev_top(spa, B_TRUE);
3460 tvd = spa->spa_root_vdev->vdev_child[top];
3463 old_ms_count = tvd->vdev_ms_count;
3464 old_class_space = metaslab_class_get_space(mc);
3467 * Determine the size of the first leaf vdev associated with
3468 * our top-level device.
3470 vd = vdev_walk_tree(tvd, NULL, NULL);
3471 ASSERT3P(vd, !=, NULL);
3472 ASSERT(vd->vdev_ops->vdev_op_leaf);
3474 psize = vd->vdev_psize;
3477 * We only try to expand the vdev if it's healthy, less than 4x its
3478 * original size, and it has a valid psize.
3480 if (tvd->vdev_state != VDEV_STATE_HEALTHY ||
3481 psize == 0 || psize >= 4 * ztest_opts.zo_vdev_size) {
3482 spa_config_exit(spa, SCL_STATE, spa);
3483 mutex_exit(&ztest_vdev_lock);
3487 newsize = psize + psize / 8;
3488 ASSERT3U(newsize, >, psize);
3490 if (ztest_opts.zo_verbose >= 6) {
3491 (void) printf("Expanding LUN %s from %lu to %lu\n",
3492 vd->vdev_path, (ulong_t)psize, (ulong_t)newsize);
3496 * Growing the vdev is a two step process:
3497 * 1). expand the physical size (i.e. relabel)
3498 * 2). online the vdev to create the new metaslabs
3500 if (vdev_walk_tree(tvd, grow_vdev, &newsize) != NULL ||
3501 vdev_walk_tree(tvd, online_vdev, NULL) != NULL ||
3502 tvd->vdev_state != VDEV_STATE_HEALTHY) {
3503 if (ztest_opts.zo_verbose >= 5) {
3504 (void) printf("Could not expand LUN because "
3505 "the vdev configuration changed.\n");
3507 spa_config_exit(spa, SCL_STATE, spa);
3508 mutex_exit(&ztest_vdev_lock);
3512 spa_config_exit(spa, SCL_STATE, spa);
3515 * Expanding the LUN will update the config asynchronously,
3516 * thus we must wait for the async thread to complete any
3517 * pending tasks before proceeding.
3521 mutex_enter(&spa->spa_async_lock);
3522 done = (spa->spa_async_thread == NULL && !spa->spa_async_tasks);
3523 mutex_exit(&spa->spa_async_lock);
3526 txg_wait_synced(spa_get_dsl(spa), 0);
3527 (void) poll(NULL, 0, 100);
3530 spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3532 tvd = spa->spa_root_vdev->vdev_child[top];
3533 new_ms_count = tvd->vdev_ms_count;
3534 new_class_space = metaslab_class_get_space(mc);
3536 if (tvd->vdev_mg != mg || mg->mg_class != mc) {
3537 if (ztest_opts.zo_verbose >= 5) {
3538 (void) printf("Could not verify LUN expansion due to "
3539 "intervening vdev offline or remove.\n");
3541 spa_config_exit(spa, SCL_STATE, spa);
3542 mutex_exit(&ztest_vdev_lock);
3547 * Make sure we were able to grow the vdev.
3549 if (new_ms_count <= old_ms_count)
3550 fatal(0, "LUN expansion failed: ms_count %llu <= %llu\n",
3551 old_ms_count, new_ms_count);
3554 * Make sure we were able to grow the pool.
3556 if (new_class_space <= old_class_space)
3557 fatal(0, "LUN expansion failed: class_space %llu <= %llu\n",
3558 old_class_space, new_class_space);
3560 if (ztest_opts.zo_verbose >= 5) {
3561 char oldnumbuf[NN_NUMBUF_SZ], newnumbuf[NN_NUMBUF_SZ];
3563 nicenum(old_class_space, oldnumbuf, sizeof (oldnumbuf));
3564 nicenum(new_class_space, newnumbuf, sizeof (newnumbuf));
3565 (void) printf("%s grew from %s to %s\n",
3566 spa->spa_name, oldnumbuf, newnumbuf);
3569 spa_config_exit(spa, SCL_STATE, spa);
3570 mutex_exit(&ztest_vdev_lock);
3574 * Verify that dmu_objset_{create,destroy,open,close} work as expected.
3578 ztest_objset_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
3581 * Create the objects common to all ztest datasets.
3583 VERIFY(zap_create_claim(os, ZTEST_DIROBJ,
3584 DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx) == 0);
3588 ztest_dataset_create(char *dsname)
3592 dsl_crypto_params_t *dcp = NULL;
3595 * 50% of the time, we create encrypted datasets
3596 * using a random cipher suite and a hard-coded
3599 rand = ztest_random(2);
3601 nvlist_t *crypto_args = fnvlist_alloc();
3602 nvlist_t *props = fnvlist_alloc();
3604 /* slight bias towards the default cipher suite */
3605 rand = ztest_random(ZIO_CRYPT_FUNCTIONS);
3606 if (rand < ZIO_CRYPT_AES_128_CCM)
3607 rand = ZIO_CRYPT_ON;
3609 fnvlist_add_uint64(props,
3610 zfs_prop_to_name(ZFS_PROP_ENCRYPTION), rand);
3611 fnvlist_add_uint8_array(crypto_args, "wkeydata",
3612 (uint8_t *)ztest_wkeydata, WRAPPING_KEY_LEN);
3615 * These parameters aren't really used by the kernel. They
3616 * are simply stored so that userspace knows how to load
3619 fnvlist_add_uint64(props,
3620 zfs_prop_to_name(ZFS_PROP_KEYFORMAT), ZFS_KEYFORMAT_RAW);
3621 fnvlist_add_string(props,
3622 zfs_prop_to_name(ZFS_PROP_KEYLOCATION), "prompt");
3623 fnvlist_add_uint64(props,
3624 zfs_prop_to_name(ZFS_PROP_PBKDF2_SALT), 0ULL);
3625 fnvlist_add_uint64(props,
3626 zfs_prop_to_name(ZFS_PROP_PBKDF2_ITERS), 0ULL);
3628 VERIFY0(dsl_crypto_params_create_nvlist(DCP_CMD_NONE, props,
3629 crypto_args, &dcp));
3631 fnvlist_free(crypto_args);
3632 fnvlist_free(props);
3635 err = dmu_objset_create(dsname, DMU_OST_OTHER, 0, dcp,
3636 ztest_objset_create_cb, NULL);
3637 dsl_crypto_params_free(dcp, !!err);
3639 rand = ztest_random(100);
3640 if (err || rand < 80)
3643 if (ztest_opts.zo_verbose >= 5)
3644 (void) printf("Setting dataset %s to sync always\n", dsname);
3645 return (ztest_dsl_prop_set_uint64(dsname, ZFS_PROP_SYNC,
3646 ZFS_SYNC_ALWAYS, B_FALSE));
3651 ztest_objset_destroy_cb(const char *name, void *arg)
3654 dmu_object_info_t doi;
3658 * Verify that the dataset contains a directory object.
3660 VERIFY0(ztest_dmu_objset_own(name, DMU_OST_OTHER, B_TRUE,
3661 B_TRUE, FTAG, &os));
3662 error = dmu_object_info(os, ZTEST_DIROBJ, &doi);
3663 if (error != ENOENT) {
3664 /* We could have crashed in the middle of destroying it */
3666 ASSERT3U(doi.doi_type, ==, DMU_OT_ZAP_OTHER);
3667 ASSERT3S(doi.doi_physical_blocks_512, >=, 0);
3669 dmu_objset_disown(os, B_TRUE, FTAG);
3672 * Destroy the dataset.
3674 if (strchr(name, '@') != NULL) {
3675 VERIFY0(dsl_destroy_snapshot(name, B_TRUE));
3677 error = dsl_destroy_head(name);
3678 /* There could be a hold on this dataset */
3686 ztest_snapshot_create(char *osname, uint64_t id)
3688 char snapname[ZFS_MAX_DATASET_NAME_LEN];
3691 (void) snprintf(snapname, sizeof (snapname), "%llu", (u_longlong_t)id);
3693 error = dmu_objset_snapshot_one(osname, snapname);
3694 if (error == ENOSPC) {
3695 ztest_record_enospc(FTAG);
3698 if (error != 0 && error != EEXIST) {
3699 fatal(0, "ztest_snapshot_create(%s@%s) = %d", osname,
3706 ztest_snapshot_destroy(char *osname, uint64_t id)
3708 char snapname[ZFS_MAX_DATASET_NAME_LEN];
3711 (void) snprintf(snapname, sizeof (snapname), "%s@%llu", osname,
3714 error = dsl_destroy_snapshot(snapname, B_FALSE);
3715 if (error != 0 && error != ENOENT)
3716 fatal(0, "ztest_snapshot_destroy(%s) = %d", snapname, error);
3722 ztest_dmu_objset_create_destroy(ztest_ds_t *zd, uint64_t id)
3728 char name[ZFS_MAX_DATASET_NAME_LEN];
3732 zdtmp = umem_alloc(sizeof (ztest_ds_t), UMEM_NOFAIL);
3734 (void) rw_rdlock(&ztest_name_lock);
3736 (void) snprintf(name, sizeof (name), "%s/temp_%llu",
3737 ztest_opts.zo_pool, (u_longlong_t)id);
3740 * If this dataset exists from a previous run, process its replay log
3741 * half of the time. If we don't replay it, then dsl_destroy_head()
3742 * (invoked from ztest_objset_destroy_cb()) should just throw it away.
3744 if (ztest_random(2) == 0 &&
3745 ztest_dmu_objset_own(name, DMU_OST_OTHER, B_FALSE,
3746 B_TRUE, FTAG, &os) == 0) {
3747 ztest_zd_init(zdtmp, NULL, os);
3748 zil_replay(os, zdtmp, ztest_replay_vector);
3749 ztest_zd_fini(zdtmp);
3750 txg_wait_synced(dmu_objset_pool(os), 0);
3751 dmu_objset_disown(os, B_TRUE, FTAG);
3755 * There may be an old instance of the dataset we're about to
3756 * create lying around from a previous run. If so, destroy it
3757 * and all of its snapshots.
3759 (void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
3760 DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
3763 * Verify that the destroyed dataset is no longer in the namespace.
3765 VERIFY3U(ENOENT, ==, ztest_dmu_objset_own(name, DMU_OST_OTHER, B_TRUE,
3766 B_TRUE, FTAG, &os));
3769 * Verify that we can create a new dataset.
3771 error = ztest_dataset_create(name);
3773 if (error == ENOSPC) {
3774 ztest_record_enospc(FTAG);
3777 fatal(0, "dmu_objset_create(%s) = %d", name, error);
3780 VERIFY0(ztest_dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, B_TRUE,
3783 ztest_zd_init(zdtmp, NULL, os);
3786 * Open the intent log for it.
3788 zilog = zil_open(os, ztest_get_data);
3791 * Put some objects in there, do a little I/O to them,
3792 * and randomly take a couple of snapshots along the way.
3794 iters = ztest_random(5);
3795 for (i = 0; i < iters; i++) {
3796 ztest_dmu_object_alloc_free(zdtmp, id);
3797 if (ztest_random(iters) == 0)
3798 (void) ztest_snapshot_create(name, i);
3802 * Verify that we cannot create an existing dataset.
3804 VERIFY3U(EEXIST, ==,
3805 dmu_objset_create(name, DMU_OST_OTHER, 0, NULL, NULL, NULL));
3808 * Verify that we can hold an objset that is also owned.
3810 VERIFY3U(0, ==, dmu_objset_hold(name, FTAG, &os2));
3811 dmu_objset_rele(os2, FTAG);
3814 * Verify that we cannot own an objset that is already owned.
3816 VERIFY3U(EBUSY, ==, ztest_dmu_objset_own(name, DMU_OST_OTHER,
3817 B_FALSE, B_TRUE, FTAG, &os2));
3820 txg_wait_synced(spa_get_dsl(os->os_spa), 0);
3821 dmu_objset_disown(os, B_TRUE, FTAG);
3822 ztest_zd_fini(zdtmp);
3824 (void) rw_unlock(&ztest_name_lock);
3826 umem_free(zdtmp, sizeof (ztest_ds_t));
3830 * Verify that dmu_snapshot_{create,destroy,open,close} work as expected.
3833 ztest_dmu_snapshot_create_destroy(ztest_ds_t *zd, uint64_t id)
3835 (void) rw_rdlock(&ztest_name_lock);
3836 (void) ztest_snapshot_destroy(zd->zd_name, id);
3837 (void) ztest_snapshot_create(zd->zd_name, id);
3838 (void) rw_unlock(&ztest_name_lock);
3842 * Cleanup non-standard snapshots and clones.
3845 ztest_dsl_dataset_cleanup(char *osname, uint64_t id)
3854 snap1name = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
3855 clone1name = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
3856 snap2name = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
3857 clone2name = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
3858 snap3name = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
3860 (void) snprintf(snap1name, ZFS_MAX_DATASET_NAME_LEN,
3861 "%s@s1_%llu", osname, (u_longlong_t)id);
3862 (void) snprintf(clone1name, ZFS_MAX_DATASET_NAME_LEN,
3863 "%s/c1_%llu", osname, (u_longlong_t)id);
3864 (void) snprintf(snap2name, ZFS_MAX_DATASET_NAME_LEN,
3865 "%s@s2_%llu", clone1name, (u_longlong_t)id);
3866 (void) snprintf(clone2name, ZFS_MAX_DATASET_NAME_LEN,
3867 "%s/c2_%llu", osname, (u_longlong_t)id);
3868 (void) snprintf(snap3name, ZFS_MAX_DATASET_NAME_LEN,
3869 "%s@s3_%llu", clone1name, (u_longlong_t)id);
3871 error = dsl_destroy_head(clone2name);
3872 if (error && error != ENOENT)
3873 fatal(0, "dsl_destroy_head(%s) = %d", clone2name, error);
3874 error = dsl_destroy_snapshot(snap3name, B_FALSE);
3875 if (error && error != ENOENT)
3876 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap3name, error);
3877 error = dsl_destroy_snapshot(snap2name, B_FALSE);
3878 if (error && error != ENOENT)
3879 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap2name, error);
3880 error = dsl_destroy_head(clone1name);
3881 if (error && error != ENOENT)
3882 fatal(0, "dsl_destroy_head(%s) = %d", clone1name, error);
3883 error = dsl_destroy_snapshot(snap1name, B_FALSE);
3884 if (error && error != ENOENT)
3885 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap1name, error);
3887 umem_free(snap1name, ZFS_MAX_DATASET_NAME_LEN);
3888 umem_free(clone1name, ZFS_MAX_DATASET_NAME_LEN);
3889 umem_free(snap2name, ZFS_MAX_DATASET_NAME_LEN);
3890 umem_free(clone2name, ZFS_MAX_DATASET_NAME_LEN);
3891 umem_free(snap3name, ZFS_MAX_DATASET_NAME_LEN);
3895 * Verify dsl_dataset_promote handles EBUSY
3898 ztest_dsl_dataset_promote_busy(ztest_ds_t *zd, uint64_t id)
3906 char *osname = zd->zd_name;
3909 snap1name = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
3910 clone1name = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
3911 snap2name = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
3912 clone2name = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
3913 snap3name = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
3915 (void) rw_rdlock(&ztest_name_lock);
3917 ztest_dsl_dataset_cleanup(osname, id);
3919 (void) snprintf(snap1name, ZFS_MAX_DATASET_NAME_LEN,
3920 "%s@s1_%llu", osname, (u_longlong_t)id);
3921 (void) snprintf(clone1name, ZFS_MAX_DATASET_NAME_LEN,
3922 "%s/c1_%llu", osname, (u_longlong_t)id);
3923 (void) snprintf(snap2name, ZFS_MAX_DATASET_NAME_LEN,
3924 "%s@s2_%llu", clone1name, (u_longlong_t)id);
3925 (void) snprintf(clone2name, ZFS_MAX_DATASET_NAME_LEN,
3926 "%s/c2_%llu", osname, (u_longlong_t)id);
3927 (void) snprintf(snap3name, ZFS_MAX_DATASET_NAME_LEN,
3928 "%s@s3_%llu", clone1name, (u_longlong_t)id);
3930 error = dmu_objset_snapshot_one(osname, strchr(snap1name, '@') + 1);
3931 if (error && error != EEXIST) {
3932 if (error == ENOSPC) {
3933 ztest_record_enospc(FTAG);
3936 fatal(0, "dmu_take_snapshot(%s) = %d", snap1name, error);
3939 error = dmu_objset_clone(clone1name, snap1name);
3941 if (error == ENOSPC) {
3942 ztest_record_enospc(FTAG);
3945 fatal(0, "dmu_objset_create(%s) = %d", clone1name, error);
3948 error = dmu_objset_snapshot_one(clone1name, strchr(snap2name, '@') + 1);
3949 if (error && error != EEXIST) {
3950 if (error == ENOSPC) {
3951 ztest_record_enospc(FTAG);
3954 fatal(0, "dmu_open_snapshot(%s) = %d", snap2name, error);
3957 error = dmu_objset_snapshot_one(clone1name, strchr(snap3name, '@') + 1);
3958 if (error && error != EEXIST) {
3959 if (error == ENOSPC) {
3960 ztest_record_enospc(FTAG);
3963 fatal(0, "dmu_open_snapshot(%s) = %d", snap3name, error);
3966 error = dmu_objset_clone(clone2name, snap3name);
3968 if (error == ENOSPC) {
3969 ztest_record_enospc(FTAG);
3972 fatal(0, "dmu_objset_create(%s) = %d", clone2name, error);
3975 error = ztest_dmu_objset_own(snap2name, DMU_OST_ANY, B_TRUE, B_TRUE,
3978 fatal(0, "dmu_objset_own(%s) = %d", snap2name, error);
3979 error = dsl_dataset_promote(clone2name, NULL);
3980 if (error == ENOSPC) {
3981 dmu_objset_disown(os, B_TRUE, FTAG);
3982 ztest_record_enospc(FTAG);
3986 fatal(0, "dsl_dataset_promote(%s), %d, not EBUSY", clone2name,
3988 dmu_objset_disown(os, B_TRUE, FTAG);
3991 ztest_dsl_dataset_cleanup(osname, id);
3993 (void) rw_unlock(&ztest_name_lock);
3995 umem_free(snap1name, ZFS_MAX_DATASET_NAME_LEN);
3996 umem_free(clone1name, ZFS_MAX_DATASET_NAME_LEN);
3997 umem_free(snap2name, ZFS_MAX_DATASET_NAME_LEN);
3998 umem_free(clone2name, ZFS_MAX_DATASET_NAME_LEN);
3999 umem_free(snap3name, ZFS_MAX_DATASET_NAME_LEN);
4002 #undef OD_ARRAY_SIZE
4003 #define OD_ARRAY_SIZE 4
4006 * Verify that dmu_object_{alloc,free} work as expected.
4009 ztest_dmu_object_alloc_free(ztest_ds_t *zd, uint64_t id)
4016 size = sizeof (ztest_od_t) * OD_ARRAY_SIZE;
4017 od = umem_alloc(size, UMEM_NOFAIL);
4018 batchsize = OD_ARRAY_SIZE;
4020 for (b = 0; b < batchsize; b++)
4021 ztest_od_init(od + b, id, FTAG, b, DMU_OT_UINT64_OTHER,
4025 * Destroy the previous batch of objects, create a new batch,
4026 * and do some I/O on the new objects.
4028 if (ztest_object_init(zd, od, size, B_TRUE) != 0)
4031 while (ztest_random(4 * batchsize) != 0)
4032 ztest_io(zd, od[ztest_random(batchsize)].od_object,
4033 ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
4035 umem_free(od, size);
4038 #undef OD_ARRAY_SIZE
4039 #define OD_ARRAY_SIZE 2
4042 * Verify that dmu_{read,write} work as expected.
4045 ztest_dmu_read_write(ztest_ds_t *zd, uint64_t id)
4050 objset_t *os = zd->zd_os;
4051 size = sizeof (ztest_od_t) * OD_ARRAY_SIZE;
4052 od = umem_alloc(size, UMEM_NOFAIL);
4054 int i, freeit, error;
4056 bufwad_t *packbuf, *bigbuf, *pack, *bigH, *bigT;
4057 uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
4058 uint64_t chunksize = (1000 + ztest_random(1000)) * sizeof (uint64_t);
4059 uint64_t regions = 997;
4060 uint64_t stride = 123456789ULL;
4061 uint64_t width = 40;
4062 int free_percent = 5;
4065 * This test uses two objects, packobj and bigobj, that are always
4066 * updated together (i.e. in the same tx) so that their contents are
4067 * in sync and can be compared. Their contents relate to each other
4068 * in a simple way: packobj is a dense array of 'bufwad' structures,
4069 * while bigobj is a sparse array of the same bufwads. Specifically,
4070 * for any index n, there are three bufwads that should be identical:
4072 * packobj, at offset n * sizeof (bufwad_t)
4073 * bigobj, at the head of the nth chunk
4074 * bigobj, at the tail of the nth chunk
4076 * The chunk size is arbitrary. It doesn't have to be a power of two,
4077 * and it doesn't have any relation to the object blocksize.
4078 * The only requirement is that it can hold at least two bufwads.
4080 * Normally, we write the bufwad to each of these locations.
4081 * However, free_percent of the time we instead write zeroes to
4082 * packobj and perform a dmu_free_range() on bigobj. By comparing
4083 * bigobj to packobj, we can verify that the DMU is correctly
4084 * tracking which parts of an object are allocated and free,
4085 * and that the contents of the allocated blocks are correct.
4089 * Read the directory info. If it's the first time, set things up.
4091 ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0, chunksize);
4092 ztest_od_init(od + 1, id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, 0,
4095 if (ztest_object_init(zd, od, size, B_FALSE) != 0) {
4096 umem_free(od, size);
4100 bigobj = od[0].od_object;
4101 packobj = od[1].od_object;
4102 chunksize = od[0].od_gen;
4103 ASSERT(chunksize == od[1].od_gen);
4106 * Prefetch a random chunk of the big object.
4107 * Our aim here is to get some async reads in flight
4108 * for blocks that we may free below; the DMU should
4109 * handle this race correctly.
4111 n = ztest_random(regions) * stride + ztest_random(width);
4112 s = 1 + ztest_random(2 * width - 1);
4113 dmu_prefetch(os, bigobj, 0, n * chunksize, s * chunksize,
4114 ZIO_PRIORITY_SYNC_READ);
4117 * Pick a random index and compute the offsets into packobj and bigobj.
4119 n = ztest_random(regions) * stride + ztest_random(width);
4120 s = 1 + ztest_random(width - 1);
4122 packoff = n * sizeof (bufwad_t);
4123 packsize = s * sizeof (bufwad_t);
4125 bigoff = n * chunksize;
4126 bigsize = s * chunksize;
4128 packbuf = umem_alloc(packsize, UMEM_NOFAIL);
4129 bigbuf = umem_alloc(bigsize, UMEM_NOFAIL);
4132 * free_percent of the time, free a range of bigobj rather than
4135 freeit = (ztest_random(100) < free_percent);
4138 * Read the current contents of our objects.
4140 error = dmu_read(os, packobj, packoff, packsize, packbuf,
4143 error = dmu_read(os, bigobj, bigoff, bigsize, bigbuf,
4148 * Get a tx for the mods to both packobj and bigobj.
4150 tx = dmu_tx_create(os);
4152 dmu_tx_hold_write(tx, packobj, packoff, packsize);
4155 dmu_tx_hold_free(tx, bigobj, bigoff, bigsize);
4157 dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
4159 /* This accounts for setting the checksum/compression. */
4160 dmu_tx_hold_bonus(tx, bigobj);
4162 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4164 umem_free(packbuf, packsize);
4165 umem_free(bigbuf, bigsize);
4166 umem_free(od, size);
4170 enum zio_checksum cksum;
4172 cksum = (enum zio_checksum)
4173 ztest_random_dsl_prop(ZFS_PROP_CHECKSUM);
4174 } while (cksum >= ZIO_CHECKSUM_LEGACY_FUNCTIONS);
4175 dmu_object_set_checksum(os, bigobj, cksum, tx);
4177 enum zio_compress comp;
4179 comp = (enum zio_compress)
4180 ztest_random_dsl_prop(ZFS_PROP_COMPRESSION);
4181 } while (comp >= ZIO_COMPRESS_LEGACY_FUNCTIONS);
4182 dmu_object_set_compress(os, bigobj, comp, tx);
4185 * For each index from n to n + s, verify that the existing bufwad
4186 * in packobj matches the bufwads at the head and tail of the
4187 * corresponding chunk in bigobj. Then update all three bufwads
4188 * with the new values we want to write out.
4190 for (i = 0; i < s; i++) {
4192 pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
4194 bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
4196 bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
4198 ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
4199 ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
4201 if (pack->bw_txg > txg)
4202 fatal(0, "future leak: got %llx, open txg is %llx",
4205 if (pack->bw_data != 0 && pack->bw_index != n + i)
4206 fatal(0, "wrong index: got %llx, wanted %llx+%llx",
4207 pack->bw_index, n, i);
4209 if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
4210 fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
4212 if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
4213 fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
4216 bzero(pack, sizeof (bufwad_t));
4218 pack->bw_index = n + i;
4220 pack->bw_data = 1 + ztest_random(-2ULL);
4227 * We've verified all the old bufwads, and made new ones.
4228 * Now write them out.
4230 dmu_write(os, packobj, packoff, packsize, packbuf, tx);
4233 if (ztest_opts.zo_verbose >= 7) {
4234 (void) printf("freeing offset %llx size %llx"
4236 (u_longlong_t)bigoff,
4237 (u_longlong_t)bigsize,
4240 VERIFY(0 == dmu_free_range(os, bigobj, bigoff, bigsize, tx));
4242 if (ztest_opts.zo_verbose >= 7) {
4243 (void) printf("writing offset %llx size %llx"
4245 (u_longlong_t)bigoff,
4246 (u_longlong_t)bigsize,
4249 dmu_write(os, bigobj, bigoff, bigsize, bigbuf, tx);
4255 * Sanity check the stuff we just wrote.
4258 void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
4259 void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
4261 VERIFY(0 == dmu_read(os, packobj, packoff,
4262 packsize, packcheck, DMU_READ_PREFETCH));
4263 VERIFY(0 == dmu_read(os, bigobj, bigoff,
4264 bigsize, bigcheck, DMU_READ_PREFETCH));
4266 ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
4267 ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
4269 umem_free(packcheck, packsize);
4270 umem_free(bigcheck, bigsize);
4273 umem_free(packbuf, packsize);
4274 umem_free(bigbuf, bigsize);
4275 umem_free(od, size);
4279 compare_and_update_pbbufs(uint64_t s, bufwad_t *packbuf, bufwad_t *bigbuf,
4280 uint64_t bigsize, uint64_t n, uint64_t chunksize, uint64_t txg)
4288 * For each index from n to n + s, verify that the existing bufwad
4289 * in packobj matches the bufwads at the head and tail of the
4290 * corresponding chunk in bigobj. Then update all three bufwads
4291 * with the new values we want to write out.
4293 for (i = 0; i < s; i++) {
4295 pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
4297 bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
4299 bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
4301 ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
4302 ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
4304 if (pack->bw_txg > txg)
4305 fatal(0, "future leak: got %llx, open txg is %llx",
4308 if (pack->bw_data != 0 && pack->bw_index != n + i)
4309 fatal(0, "wrong index: got %llx, wanted %llx+%llx",
4310 pack->bw_index, n, i);
4312 if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
4313 fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
4315 if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
4316 fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
4318 pack->bw_index = n + i;
4320 pack->bw_data = 1 + ztest_random(-2ULL);
4327 #undef OD_ARRAY_SIZE
4328 #define OD_ARRAY_SIZE 2
4331 ztest_dmu_read_write_zcopy(ztest_ds_t *zd, uint64_t id)
4333 objset_t *os = zd->zd_os;
4340 bufwad_t *packbuf, *bigbuf;
4341 uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
4342 uint64_t blocksize = ztest_random_blocksize();
4343 uint64_t chunksize = blocksize;
4344 uint64_t regions = 997;
4345 uint64_t stride = 123456789ULL;
4347 dmu_buf_t *bonus_db;
4348 arc_buf_t **bigbuf_arcbufs;
4349 dmu_object_info_t doi;
4351 size = sizeof (ztest_od_t) * OD_ARRAY_SIZE;
4352 od = umem_alloc(size, UMEM_NOFAIL);
4355 * This test uses two objects, packobj and bigobj, that are always
4356 * updated together (i.e. in the same tx) so that their contents are
4357 * in sync and can be compared. Their contents relate to each other
4358 * in a simple way: packobj is a dense array of 'bufwad' structures,
4359 * while bigobj is a sparse array of the same bufwads. Specifically,
4360 * for any index n, there are three bufwads that should be identical:
4362 * packobj, at offset n * sizeof (bufwad_t)
4363 * bigobj, at the head of the nth chunk
4364 * bigobj, at the tail of the nth chunk
4366 * The chunk size is set equal to bigobj block size so that
4367 * dmu_assign_arcbuf_by_dbuf() can be tested for object updates.
4371 * Read the directory info. If it's the first time, set things up.
4373 ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0, 0);
4374 ztest_od_init(od + 1, id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, 0,
4378 if (ztest_object_init(zd, od, size, B_FALSE) != 0) {
4379 umem_free(od, size);
4383 bigobj = od[0].od_object;
4384 packobj = od[1].od_object;
4385 blocksize = od[0].od_blocksize;
4386 chunksize = blocksize;
4387 ASSERT(chunksize == od[1].od_gen);
4389 VERIFY(dmu_object_info(os, bigobj, &doi) == 0);
4390 VERIFY(ISP2(doi.doi_data_block_size));
4391 VERIFY(chunksize == doi.doi_data_block_size);
4392 VERIFY(chunksize >= 2 * sizeof (bufwad_t));
4395 * Pick a random index and compute the offsets into packobj and bigobj.
4397 n = ztest_random(regions) * stride + ztest_random(width);
4398 s = 1 + ztest_random(width - 1);
4400 packoff = n * sizeof (bufwad_t);
4401 packsize = s * sizeof (bufwad_t);
4403 bigoff = n * chunksize;
4404 bigsize = s * chunksize;
4406 packbuf = umem_zalloc(packsize, UMEM_NOFAIL);
4407 bigbuf = umem_zalloc(bigsize, UMEM_NOFAIL);
4409 VERIFY3U(0, ==, dmu_bonus_hold(os, bigobj, FTAG, &bonus_db));
4411 bigbuf_arcbufs = umem_zalloc(2 * s * sizeof (arc_buf_t *), UMEM_NOFAIL);
4414 * Iteration 0 test zcopy for DB_UNCACHED dbufs.
4415 * Iteration 1 test zcopy to already referenced dbufs.
4416 * Iteration 2 test zcopy to dirty dbuf in the same txg.
4417 * Iteration 3 test zcopy to dbuf dirty in previous txg.
4418 * Iteration 4 test zcopy when dbuf is no longer dirty.
4419 * Iteration 5 test zcopy when it can't be done.
4420 * Iteration 6 one more zcopy write.
4422 for (i = 0; i < 7; i++) {
4427 * In iteration 5 (i == 5) use arcbufs
4428 * that don't match bigobj blksz to test
4429 * dmu_assign_arcbuf_by_dbuf() when it can't directly
4430 * assign an arcbuf to a dbuf.
4432 for (j = 0; j < s; j++) {
4433 if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) {
4435 dmu_request_arcbuf(bonus_db, chunksize);
4437 bigbuf_arcbufs[2 * j] =
4438 dmu_request_arcbuf(bonus_db, chunksize / 2);
4439 bigbuf_arcbufs[2 * j + 1] =
4440 dmu_request_arcbuf(bonus_db, chunksize / 2);
4445 * Get a tx for the mods to both packobj and bigobj.
4447 tx = dmu_tx_create(os);
4449 dmu_tx_hold_write(tx, packobj, packoff, packsize);
4450 dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
4452 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4454 umem_free(packbuf, packsize);
4455 umem_free(bigbuf, bigsize);
4456 for (j = 0; j < s; j++) {
4458 chunksize < (SPA_MINBLOCKSIZE * 2)) {
4459 dmu_return_arcbuf(bigbuf_arcbufs[j]);
4462 bigbuf_arcbufs[2 * j]);
4464 bigbuf_arcbufs[2 * j + 1]);
4467 umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
4468 umem_free(od, size);
4469 dmu_buf_rele(bonus_db, FTAG);
4474 * 50% of the time don't read objects in the 1st iteration to
4475 * test dmu_assign_arcbuf_by_dbuf() for the case when there are
4476 * no existing dbufs for the specified offsets.
4478 if (i != 0 || ztest_random(2) != 0) {
4479 error = dmu_read(os, packobj, packoff,
4480 packsize, packbuf, DMU_READ_PREFETCH);
4482 error = dmu_read(os, bigobj, bigoff, bigsize,
4483 bigbuf, DMU_READ_PREFETCH);
4486 compare_and_update_pbbufs(s, packbuf, bigbuf, bigsize,
4490 * We've verified all the old bufwads, and made new ones.
4491 * Now write them out.
4493 dmu_write(os, packobj, packoff, packsize, packbuf, tx);
4494 if (ztest_opts.zo_verbose >= 7) {
4495 (void) printf("writing offset %llx size %llx"
4497 (u_longlong_t)bigoff,
4498 (u_longlong_t)bigsize,
4501 for (off = bigoff, j = 0; j < s; j++, off += chunksize) {
4503 if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) {
4504 bcopy((caddr_t)bigbuf + (off - bigoff),
4505 bigbuf_arcbufs[j]->b_data, chunksize);
4507 bcopy((caddr_t)bigbuf + (off - bigoff),
4508 bigbuf_arcbufs[2 * j]->b_data,
4510 bcopy((caddr_t)bigbuf + (off - bigoff) +
4512 bigbuf_arcbufs[2 * j + 1]->b_data,
4517 VERIFY(dmu_buf_hold(os, bigobj, off,
4518 FTAG, &dbt, DMU_READ_NO_PREFETCH) == 0);
4520 if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) {
4521 dmu_assign_arcbuf_by_dbuf(bonus_db, off,
4522 bigbuf_arcbufs[j], tx);
4524 dmu_assign_arcbuf_by_dbuf(bonus_db, off,
4525 bigbuf_arcbufs[2 * j], tx);
4526 dmu_assign_arcbuf_by_dbuf(bonus_db,
4527 off + chunksize / 2,
4528 bigbuf_arcbufs[2 * j + 1], tx);
4531 dmu_buf_rele(dbt, FTAG);
4537 * Sanity check the stuff we just wrote.
4540 void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
4541 void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
4543 VERIFY(0 == dmu_read(os, packobj, packoff,
4544 packsize, packcheck, DMU_READ_PREFETCH));
4545 VERIFY(0 == dmu_read(os, bigobj, bigoff,
4546 bigsize, bigcheck, DMU_READ_PREFETCH));
4548 ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
4549 ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
4551 umem_free(packcheck, packsize);
4552 umem_free(bigcheck, bigsize);
4555 txg_wait_open(dmu_objset_pool(os), 0);
4556 } else if (i == 3) {
4557 txg_wait_synced(dmu_objset_pool(os), 0);
4561 dmu_buf_rele(bonus_db, FTAG);
4562 umem_free(packbuf, packsize);
4563 umem_free(bigbuf, bigsize);
4564 umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
4565 umem_free(od, size);
4570 ztest_dmu_write_parallel(ztest_ds_t *zd, uint64_t id)
4574 od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
4575 uint64_t offset = (1ULL << (ztest_random(20) + 43)) +
4576 (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
4579 * Have multiple threads write to large offsets in an object
4580 * to verify that parallel writes to an object -- even to the
4581 * same blocks within the object -- doesn't cause any trouble.
4583 ztest_od_init(od, ID_PARALLEL, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0, 0);
4585 if (ztest_object_init(zd, od, sizeof (ztest_od_t), B_FALSE) != 0)
4588 while (ztest_random(10) != 0)
4589 ztest_io(zd, od->od_object, offset);
4591 umem_free(od, sizeof (ztest_od_t));
4595 ztest_dmu_prealloc(ztest_ds_t *zd, uint64_t id)
4598 uint64_t offset = (1ULL << (ztest_random(4) + SPA_MAXBLOCKSHIFT)) +
4599 (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
4600 uint64_t count = ztest_random(20) + 1;
4601 uint64_t blocksize = ztest_random_blocksize();
4604 od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
4606 ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0, 0);
4608 if (ztest_object_init(zd, od, sizeof (ztest_od_t),
4609 !ztest_random(2)) != 0) {
4610 umem_free(od, sizeof (ztest_od_t));
4614 if (ztest_truncate(zd, od->od_object, offset, count * blocksize) != 0) {
4615 umem_free(od, sizeof (ztest_od_t));
4619 ztest_prealloc(zd, od->od_object, offset, count * blocksize);
4621 data = umem_zalloc(blocksize, UMEM_NOFAIL);
4623 while (ztest_random(count) != 0) {
4624 uint64_t randoff = offset + (ztest_random(count) * blocksize);
4625 if (ztest_write(zd, od->od_object, randoff, blocksize,
4628 while (ztest_random(4) != 0)
4629 ztest_io(zd, od->od_object, randoff);
4632 umem_free(data, blocksize);
4633 umem_free(od, sizeof (ztest_od_t));
4637 * Verify that zap_{create,destroy,add,remove,update} work as expected.
4639 #define ZTEST_ZAP_MIN_INTS 1
4640 #define ZTEST_ZAP_MAX_INTS 4
4641 #define ZTEST_ZAP_MAX_PROPS 1000
4644 ztest_zap(ztest_ds_t *zd, uint64_t id)
4646 objset_t *os = zd->zd_os;
4649 uint64_t txg, last_txg;
4650 uint64_t value[ZTEST_ZAP_MAX_INTS];
4651 uint64_t zl_ints, zl_intsize, prop;
4654 char propname[100], txgname[100];
4656 char *hc[2] = { "s.acl.h", ".s.open.h.hyLZlg" };
4658 od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
4659 ztest_od_init(od, id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0, 0);
4661 if (ztest_object_init(zd, od, sizeof (ztest_od_t),
4662 !ztest_random(2)) != 0)
4665 object = od->od_object;
4668 * Generate a known hash collision, and verify that
4669 * we can lookup and remove both entries.
4671 tx = dmu_tx_create(os);
4672 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4673 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4676 for (i = 0; i < 2; i++) {
4678 VERIFY3U(0, ==, zap_add(os, object, hc[i], sizeof (uint64_t),
4681 for (i = 0; i < 2; i++) {
4682 VERIFY3U(EEXIST, ==, zap_add(os, object, hc[i],
4683 sizeof (uint64_t), 1, &value[i], tx));
4685 zap_length(os, object, hc[i], &zl_intsize, &zl_ints));
4686 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4687 ASSERT3U(zl_ints, ==, 1);
4689 for (i = 0; i < 2; i++) {
4690 VERIFY3U(0, ==, zap_remove(os, object, hc[i], tx));
4695 * Generate a buch of random entries.
4697 ints = MAX(ZTEST_ZAP_MIN_INTS, object % ZTEST_ZAP_MAX_INTS);
4699 prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
4700 (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
4701 (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
4702 bzero(value, sizeof (value));
4706 * If these zap entries already exist, validate their contents.
4708 error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
4710 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4711 ASSERT3U(zl_ints, ==, 1);
4713 VERIFY(zap_lookup(os, object, txgname, zl_intsize,
4714 zl_ints, &last_txg) == 0);
4716 VERIFY(zap_length(os, object, propname, &zl_intsize,
4719 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4720 ASSERT3U(zl_ints, ==, ints);
4722 VERIFY(zap_lookup(os, object, propname, zl_intsize,
4723 zl_ints, value) == 0);
4725 for (i = 0; i < ints; i++) {
4726 ASSERT3U(value[i], ==, last_txg + object + i);
4729 ASSERT3U(error, ==, ENOENT);
4733 * Atomically update two entries in our zap object.
4734 * The first is named txg_%llu, and contains the txg
4735 * in which the property was last updated. The second
4736 * is named prop_%llu, and the nth element of its value
4737 * should be txg + object + n.
4739 tx = dmu_tx_create(os);
4740 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4741 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4746 fatal(0, "zap future leak: old %llu new %llu", last_txg, txg);
4748 for (i = 0; i < ints; i++)
4749 value[i] = txg + object + i;
4751 VERIFY3U(0, ==, zap_update(os, object, txgname, sizeof (uint64_t),
4753 VERIFY3U(0, ==, zap_update(os, object, propname, sizeof (uint64_t),
4759 * Remove a random pair of entries.
4761 prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
4762 (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
4763 (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
4765 error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
4767 if (error == ENOENT)
4772 tx = dmu_tx_create(os);
4773 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4774 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4777 VERIFY3U(0, ==, zap_remove(os, object, txgname, tx));
4778 VERIFY3U(0, ==, zap_remove(os, object, propname, tx));
4781 umem_free(od, sizeof (ztest_od_t));
4785 * Testcase to test the upgrading of a microzap to fatzap.
4788 ztest_fzap(ztest_ds_t *zd, uint64_t id)
4790 objset_t *os = zd->zd_os;
4792 uint64_t object, txg;
4795 od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
4796 ztest_od_init(od, id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0, 0);
4798 if (ztest_object_init(zd, od, sizeof (ztest_od_t),
4799 !ztest_random(2)) != 0)
4801 object = od->od_object;
4804 * Add entries to this ZAP and make sure it spills over
4805 * and gets upgraded to a fatzap. Also, since we are adding
4806 * 2050 entries we should see ptrtbl growth and leaf-block split.
4808 for (i = 0; i < 2050; i++) {
4809 char name[ZFS_MAX_DATASET_NAME_LEN];
4814 (void) snprintf(name, sizeof (name), "fzap-%llu-%llu",
4815 (u_longlong_t)id, (u_longlong_t)value);
4817 tx = dmu_tx_create(os);
4818 dmu_tx_hold_zap(tx, object, B_TRUE, name);
4819 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4822 error = zap_add(os, object, name, sizeof (uint64_t), 1,
4824 ASSERT(error == 0 || error == EEXIST);
4828 umem_free(od, sizeof (ztest_od_t));
4833 ztest_zap_parallel(ztest_ds_t *zd, uint64_t id)
4835 objset_t *os = zd->zd_os;
4837 uint64_t txg, object, count, wsize, wc, zl_wsize, zl_wc;
4839 int i, namelen, error;
4840 int micro = ztest_random(2);
4841 char name[20], string_value[20];
4844 od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
4845 ztest_od_init(od, ID_PARALLEL, FTAG, micro, DMU_OT_ZAP_OTHER, 0, 0, 0);
4847 if (ztest_object_init(zd, od, sizeof (ztest_od_t), B_FALSE) != 0) {
4848 umem_free(od, sizeof (ztest_od_t));
4852 object = od->od_object;
4855 * Generate a random name of the form 'xxx.....' where each
4856 * x is a random printable character and the dots are dots.
4857 * There are 94 such characters, and the name length goes from
4858 * 6 to 20, so there are 94^3 * 15 = 12,458,760 possible names.
4860 namelen = ztest_random(sizeof (name) - 5) + 5 + 1;
4862 for (i = 0; i < 3; i++)
4863 name[i] = '!' + ztest_random('~' - '!' + 1);
4864 for (; i < namelen - 1; i++)
4868 if ((namelen & 1) || micro) {
4869 wsize = sizeof (txg);
4875 data = string_value;
4879 VERIFY0(zap_count(os, object, &count));
4880 ASSERT(count != -1ULL);
4883 * Select an operation: length, lookup, add, update, remove.
4885 i = ztest_random(5);
4888 tx = dmu_tx_create(os);
4889 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4890 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4892 umem_free(od, sizeof (ztest_od_t));
4895 bcopy(name, string_value, namelen);
4899 bzero(string_value, namelen);
4905 error = zap_length(os, object, name, &zl_wsize, &zl_wc);
4907 ASSERT3U(wsize, ==, zl_wsize);
4908 ASSERT3U(wc, ==, zl_wc);
4910 ASSERT3U(error, ==, ENOENT);
4915 error = zap_lookup(os, object, name, wsize, wc, data);
4917 if (data == string_value &&
4918 bcmp(name, data, namelen) != 0)
4919 fatal(0, "name '%s' != val '%s' len %d",
4920 name, data, namelen);
4922 ASSERT3U(error, ==, ENOENT);
4927 error = zap_add(os, object, name, wsize, wc, data, tx);
4928 ASSERT(error == 0 || error == EEXIST);
4932 VERIFY(zap_update(os, object, name, wsize, wc, data, tx) == 0);
4936 error = zap_remove(os, object, name, tx);
4937 ASSERT(error == 0 || error == ENOENT);
4944 umem_free(od, sizeof (ztest_od_t));
4948 * Commit callback data.
4950 typedef struct ztest_cb_data {
4951 list_node_t zcd_node;
4953 int zcd_expected_err;
4954 boolean_t zcd_added;
4955 boolean_t zcd_called;
4959 /* This is the actual commit callback function */
4961 ztest_commit_callback(void *arg, int error)
4963 ztest_cb_data_t *data = arg;
4964 uint64_t synced_txg;
4966 VERIFY(data != NULL);
4967 VERIFY3S(data->zcd_expected_err, ==, error);
4968 VERIFY(!data->zcd_called);
4970 synced_txg = spa_last_synced_txg(data->zcd_spa);
4971 if (data->zcd_txg > synced_txg)
4972 fatal(0, "commit callback of txg %" PRIu64 " called prematurely"
4973 ", last synced txg = %" PRIu64 "\n", data->zcd_txg,
4976 data->zcd_called = B_TRUE;
4978 if (error == ECANCELED) {
4979 ASSERT0(data->zcd_txg);
4980 ASSERT(!data->zcd_added);
4983 * The private callback data should be destroyed here, but
4984 * since we are going to check the zcd_called field after
4985 * dmu_tx_abort(), we will destroy it there.
4990 ASSERT(data->zcd_added);
4991 ASSERT3U(data->zcd_txg, !=, 0);
4993 (void) mutex_enter(&zcl.zcl_callbacks_lock);
4995 /* See if this cb was called more quickly */
4996 if ((synced_txg - data->zcd_txg) < zc_min_txg_delay)
4997 zc_min_txg_delay = synced_txg - data->zcd_txg;
4999 /* Remove our callback from the list */
5000 list_remove(&zcl.zcl_callbacks, data);
5002 (void) mutex_exit(&zcl.zcl_callbacks_lock);
5004 umem_free(data, sizeof (ztest_cb_data_t));
5007 /* Allocate and initialize callback data structure */
5008 static ztest_cb_data_t *
5009 ztest_create_cb_data(objset_t *os, uint64_t txg)
5011 ztest_cb_data_t *cb_data;
5013 cb_data = umem_zalloc(sizeof (ztest_cb_data_t), UMEM_NOFAIL);
5015 cb_data->zcd_txg = txg;
5016 cb_data->zcd_spa = dmu_objset_spa(os);
5017 list_link_init(&cb_data->zcd_node);
5023 * Commit callback test.
5026 ztest_dmu_commit_callbacks(ztest_ds_t *zd, uint64_t id)
5028 objset_t *os = zd->zd_os;
5031 ztest_cb_data_t *cb_data[3], *tmp_cb;
5032 uint64_t old_txg, txg;
5035 od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
5036 ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0, 0);
5038 if (ztest_object_init(zd, od, sizeof (ztest_od_t), B_FALSE) != 0) {
5039 umem_free(od, sizeof (ztest_od_t));
5043 tx = dmu_tx_create(os);
5045 cb_data[0] = ztest_create_cb_data(os, 0);
5046 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[0]);
5048 dmu_tx_hold_write(tx, od->od_object, 0, sizeof (uint64_t));
5050 /* Every once in a while, abort the transaction on purpose */
5051 if (ztest_random(100) == 0)
5055 error = dmu_tx_assign(tx, TXG_NOWAIT);
5057 txg = error ? 0 : dmu_tx_get_txg(tx);
5059 cb_data[0]->zcd_txg = txg;
5060 cb_data[1] = ztest_create_cb_data(os, txg);
5061 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[1]);
5065 * It's not a strict requirement to call the registered
5066 * callbacks from inside dmu_tx_abort(), but that's what
5067 * it's supposed to happen in the current implementation
5068 * so we will check for that.
5070 for (i = 0; i < 2; i++) {
5071 cb_data[i]->zcd_expected_err = ECANCELED;
5072 VERIFY(!cb_data[i]->zcd_called);
5077 for (i = 0; i < 2; i++) {
5078 VERIFY(cb_data[i]->zcd_called);
5079 umem_free(cb_data[i], sizeof (ztest_cb_data_t));
5082 umem_free(od, sizeof (ztest_od_t));
5086 cb_data[2] = ztest_create_cb_data(os, txg);
5087 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[2]);
5090 * Read existing data to make sure there isn't a future leak.
5092 VERIFY(0 == dmu_read(os, od->od_object, 0, sizeof (uint64_t),
5093 &old_txg, DMU_READ_PREFETCH));
5096 fatal(0, "future leak: got %" PRIu64 ", open txg is %" PRIu64,
5099 dmu_write(os, od->od_object, 0, sizeof (uint64_t), &txg, tx);
5101 (void) mutex_enter(&zcl.zcl_callbacks_lock);
5104 * Since commit callbacks don't have any ordering requirement and since
5105 * it is theoretically possible for a commit callback to be called
5106 * after an arbitrary amount of time has elapsed since its txg has been
5107 * synced, it is difficult to reliably determine whether a commit
5108 * callback hasn't been called due to high load or due to a flawed
5111 * In practice, we will assume that if after a certain number of txgs a
5112 * commit callback hasn't been called, then most likely there's an
5113 * implementation bug..
5115 tmp_cb = list_head(&zcl.zcl_callbacks);
5116 if (tmp_cb != NULL &&
5117 tmp_cb->zcd_txg + ZTEST_COMMIT_CB_THRESH < txg) {
5118 fatal(0, "Commit callback threshold exceeded, oldest txg: %"
5119 PRIu64 ", open txg: %" PRIu64 "\n", tmp_cb->zcd_txg, txg);
5123 * Let's find the place to insert our callbacks.
5125 * Even though the list is ordered by txg, it is possible for the
5126 * insertion point to not be the end because our txg may already be
5127 * quiescing at this point and other callbacks in the open txg
5128 * (from other objsets) may have sneaked in.
5130 tmp_cb = list_tail(&zcl.zcl_callbacks);
5131 while (tmp_cb != NULL && tmp_cb->zcd_txg > txg)
5132 tmp_cb = list_prev(&zcl.zcl_callbacks, tmp_cb);
5134 /* Add the 3 callbacks to the list */
5135 for (i = 0; i < 3; i++) {
5137 list_insert_head(&zcl.zcl_callbacks, cb_data[i]);
5139 list_insert_after(&zcl.zcl_callbacks, tmp_cb,
5142 cb_data[i]->zcd_added = B_TRUE;
5143 VERIFY(!cb_data[i]->zcd_called);
5145 tmp_cb = cb_data[i];
5150 (void) mutex_exit(&zcl.zcl_callbacks_lock);
5154 umem_free(od, sizeof (ztest_od_t));
5158 * Visit each object in the dataset. Verify that its properties
5159 * are consistent what was stored in the block tag when it was created,
5160 * and that its unused bonus buffer space has not been overwritten.
5164 ztest_verify_dnode_bt(ztest_ds_t *zd, uint64_t id)
5166 objset_t *os = zd->zd_os;
5170 for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, 0)) {
5171 ztest_block_tag_t *bt = NULL;
5172 dmu_object_info_t doi;
5175 ztest_object_lock(zd, obj, RL_READER);
5176 if (dmu_bonus_hold(os, obj, FTAG, &db) != 0) {
5177 ztest_object_unlock(zd, obj);
5181 dmu_object_info_from_db(db, &doi);
5182 if (doi.doi_bonus_size >= sizeof (*bt))
5183 bt = ztest_bt_bonus(db);
5185 if (bt && bt->bt_magic == BT_MAGIC) {
5186 ztest_bt_verify(bt, os, obj, doi.doi_dnodesize,
5187 bt->bt_offset, bt->bt_gen, bt->bt_txg,
5189 ztest_verify_unused_bonus(db, bt, obj, os, bt->bt_gen);
5192 dmu_buf_rele(db, FTAG);
5193 ztest_object_unlock(zd, obj);
5199 ztest_dsl_prop_get_set(ztest_ds_t *zd, uint64_t id)
5201 zfs_prop_t proplist[] = {
5203 ZFS_PROP_COMPRESSION,
5209 (void) rw_rdlock(&ztest_name_lock);
5211 for (p = 0; p < sizeof (proplist) / sizeof (proplist[0]); p++)
5212 (void) ztest_dsl_prop_set_uint64(zd->zd_name, proplist[p],
5213 ztest_random_dsl_prop(proplist[p]), (int)ztest_random(2));
5215 VERIFY0(ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_RECORDSIZE,
5216 ztest_random_blocksize(), (int)ztest_random(2)));
5218 (void) rw_unlock(&ztest_name_lock);
5223 ztest_spa_prop_get_set(ztest_ds_t *zd, uint64_t id)
5225 nvlist_t *props = NULL;
5227 (void) rw_rdlock(&ztest_name_lock);
5229 (void) ztest_spa_prop_set_uint64(ZPOOL_PROP_DEDUPDITTO,
5230 ZIO_DEDUPDITTO_MIN + ztest_random(ZIO_DEDUPDITTO_MIN));
5232 VERIFY0(spa_prop_get(ztest_spa, &props));
5234 if (ztest_opts.zo_verbose >= 6)
5235 dump_nvlist(props, 4);
5239 (void) rw_unlock(&ztest_name_lock);
5243 user_release_one(const char *snapname, const char *holdname)
5245 nvlist_t *snaps, *holds;
5248 snaps = fnvlist_alloc();
5249 holds = fnvlist_alloc();
5250 fnvlist_add_boolean(holds, holdname);
5251 fnvlist_add_nvlist(snaps, snapname, holds);
5252 fnvlist_free(holds);
5253 error = dsl_dataset_user_release(snaps, NULL);
5254 fnvlist_free(snaps);
5259 * Test snapshot hold/release and deferred destroy.
5262 ztest_dmu_snapshot_hold(ztest_ds_t *zd, uint64_t id)
5265 objset_t *os = zd->zd_os;
5269 char clonename[100];
5271 char osname[ZFS_MAX_DATASET_NAME_LEN];
5274 (void) rw_rdlock(&ztest_name_lock);
5276 dmu_objset_name(os, osname);
5278 (void) snprintf(snapname, sizeof (snapname), "sh1_%llu",
5280 (void) snprintf(fullname, sizeof (fullname), "%s@%s", osname, snapname);
5281 (void) snprintf(clonename, sizeof (clonename),
5282 "%s/ch1_%llu", osname, (u_longlong_t)id);
5283 (void) snprintf(tag, sizeof (tag), "tag_%llu", (u_longlong_t)id);
5286 * Clean up from any previous run.
5288 error = dsl_destroy_head(clonename);
5289 if (error != ENOENT)
5291 error = user_release_one(fullname, tag);
5292 if (error != ESRCH && error != ENOENT)
5294 error = dsl_destroy_snapshot(fullname, B_FALSE);
5295 if (error != ENOENT)
5299 * Create snapshot, clone it, mark snap for deferred destroy,
5300 * destroy clone, verify snap was also destroyed.
5302 error = dmu_objset_snapshot_one(osname, snapname);
5304 if (error == ENOSPC) {
5305 ztest_record_enospc("dmu_objset_snapshot");
5308 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
5311 error = dmu_objset_clone(clonename, fullname);
5313 if (error == ENOSPC) {
5314 ztest_record_enospc("dmu_objset_clone");
5317 fatal(0, "dmu_objset_clone(%s) = %d", clonename, error);
5320 error = dsl_destroy_snapshot(fullname, B_TRUE);
5322 fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
5326 error = dsl_destroy_head(clonename);
5328 fatal(0, "dsl_destroy_head(%s) = %d", clonename, error);
5330 error = dmu_objset_hold(fullname, FTAG, &origin);
5331 if (error != ENOENT)
5332 fatal(0, "dmu_objset_hold(%s) = %d", fullname, error);
5335 * Create snapshot, add temporary hold, verify that we can't
5336 * destroy a held snapshot, mark for deferred destroy,
5337 * release hold, verify snapshot was destroyed.
5339 error = dmu_objset_snapshot_one(osname, snapname);
5341 if (error == ENOSPC) {
5342 ztest_record_enospc("dmu_objset_snapshot");
5345 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
5348 holds = fnvlist_alloc();
5349 fnvlist_add_string(holds, fullname, tag);
5350 error = dsl_dataset_user_hold(holds, 0, NULL);
5351 fnvlist_free(holds);
5353 if (error == ENOSPC) {
5354 ztest_record_enospc("dsl_dataset_user_hold");
5357 fatal(0, "dsl_dataset_user_hold(%s, %s) = %u",
5358 fullname, tag, error);
5361 error = dsl_destroy_snapshot(fullname, B_FALSE);
5362 if (error != EBUSY) {
5363 fatal(0, "dsl_destroy_snapshot(%s, B_FALSE) = %d",
5367 error = dsl_destroy_snapshot(fullname, B_TRUE);
5369 fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
5373 error = user_release_one(fullname, tag);
5375 fatal(0, "user_release_one(%s, %s) = %d", fullname, tag, error);
5377 VERIFY3U(dmu_objset_hold(fullname, FTAG, &origin), ==, ENOENT);
5380 (void) rw_unlock(&ztest_name_lock);
5384 * Inject random faults into the on-disk data.
5388 ztest_fault_inject(ztest_ds_t *zd, uint64_t id)
5390 ztest_shared_t *zs = ztest_shared;
5391 spa_t *spa = ztest_spa;
5395 uint64_t bad = 0x1990c0ffeedecadeull;
5400 int bshift = SPA_MAXBLOCKSHIFT + 2;
5406 boolean_t islog = B_FALSE;
5408 path0 = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
5409 pathrand = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
5411 mutex_enter(&ztest_vdev_lock);
5412 maxfaults = MAXFAULTS();
5413 leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz;
5414 mirror_save = zs->zs_mirrors;
5415 mutex_exit(&ztest_vdev_lock);
5417 ASSERT(leaves >= 1);
5420 * Grab the name lock as reader. There are some operations
5421 * which don't like to have their vdevs changed while
5422 * they are in progress (i.e. spa_change_guid). Those
5423 * operations will have grabbed the name lock as writer.
5425 (void) rw_rdlock(&ztest_name_lock);
5428 * We need SCL_STATE here because we're going to look at vd0->vdev_tsd.
5430 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5432 if (ztest_random(2) == 0) {
5434 * Inject errors on a normal data device or slog device.
5436 top = ztest_random_vdev_top(spa, B_TRUE);
5437 leaf = ztest_random(leaves) + zs->zs_splits;
5440 * Generate paths to the first leaf in this top-level vdev,
5441 * and to the random leaf we selected. We'll induce transient
5442 * write failures and random online/offline activity on leaf 0,
5443 * and we'll write random garbage to the randomly chosen leaf.
5445 (void) snprintf(path0, MAXPATHLEN, ztest_dev_template,
5446 ztest_opts.zo_dir, ztest_opts.zo_pool,
5447 top * leaves + zs->zs_splits);
5448 (void) snprintf(pathrand, MAXPATHLEN, ztest_dev_template,
5449 ztest_opts.zo_dir, ztest_opts.zo_pool,
5450 top * leaves + leaf);
5452 vd0 = vdev_lookup_by_path(spa->spa_root_vdev, path0);
5453 if (vd0 != NULL && vd0->vdev_top->vdev_islog)
5457 * If the top-level vdev needs to be resilvered
5458 * then we only allow faults on the device that is
5461 if (vd0 != NULL && maxfaults != 1 &&
5462 (!vdev_resilver_needed(vd0->vdev_top, NULL, NULL) ||
5463 vd0->vdev_resilver_txg != 0)) {
5465 * Make vd0 explicitly claim to be unreadable,
5466 * or unwriteable, or reach behind its back
5467 * and close the underlying fd. We can do this if
5468 * maxfaults == 0 because we'll fail and reexecute,
5469 * and we can do it if maxfaults >= 2 because we'll
5470 * have enough redundancy. If maxfaults == 1, the
5471 * combination of this with injection of random data
5472 * corruption below exceeds the pool's fault tolerance.
5474 vdev_file_t *vf = vd0->vdev_tsd;
5476 if (vf != NULL && ztest_random(3) == 0) {
5477 (void) close(vf->vf_vnode->v_fd);
5478 vf->vf_vnode->v_fd = -1;
5479 } else if (ztest_random(2) == 0) {
5480 vd0->vdev_cant_read = B_TRUE;
5482 vd0->vdev_cant_write = B_TRUE;
5484 guid0 = vd0->vdev_guid;
5488 * Inject errors on an l2cache device.
5490 spa_aux_vdev_t *sav = &spa->spa_l2cache;
5492 if (sav->sav_count == 0) {
5493 spa_config_exit(spa, SCL_STATE, FTAG);
5494 (void) rw_unlock(&ztest_name_lock);
5497 vd0 = sav->sav_vdevs[ztest_random(sav->sav_count)];
5498 guid0 = vd0->vdev_guid;
5499 (void) strcpy(path0, vd0->vdev_path);
5500 (void) strcpy(pathrand, vd0->vdev_path);
5504 maxfaults = INT_MAX; /* no limit on cache devices */
5507 spa_config_exit(spa, SCL_STATE, FTAG);
5508 (void) rw_unlock(&ztest_name_lock);
5511 * If we can tolerate two or more faults, or we're dealing
5512 * with a slog, randomly online/offline vd0.
5514 if ((maxfaults >= 2 || islog) && guid0 != 0) {
5515 if (ztest_random(10) < 6) {
5516 int flags = (ztest_random(2) == 0 ?
5517 ZFS_OFFLINE_TEMPORARY : 0);
5520 * We have to grab the zs_name_lock as writer to
5521 * prevent a race between offlining a slog and
5522 * destroying a dataset. Offlining the slog will
5523 * grab a reference on the dataset which may cause
5524 * dsl_destroy_head() to fail with EBUSY thus
5525 * leaving the dataset in an inconsistent state.
5528 (void) rw_wrlock(&ztest_name_lock);
5530 VERIFY(vdev_offline(spa, guid0, flags) != EBUSY);
5533 (void) rw_unlock(&ztest_name_lock);
5536 * Ideally we would like to be able to randomly
5537 * call vdev_[on|off]line without holding locks
5538 * to force unpredictable failures but the side
5539 * effects of vdev_[on|off]line prevent us from
5540 * doing so. We grab the ztest_vdev_lock here to
5541 * prevent a race between injection testing and
5544 mutex_enter(&ztest_vdev_lock);
5545 (void) vdev_online(spa, guid0, 0, NULL);
5546 mutex_exit(&ztest_vdev_lock);
5554 * We have at least single-fault tolerance, so inject data corruption.
5556 fd = open(pathrand, O_RDWR);
5558 if (fd == -1) /* we hit a gap in the device namespace */
5561 fsize = lseek(fd, 0, SEEK_END);
5563 while (--iters != 0) {
5565 * The offset must be chosen carefully to ensure that
5566 * we do not inject a given logical block with errors
5567 * on two different leaf devices, because ZFS can not
5568 * tolerate that (if maxfaults==1).
5570 * We divide each leaf into chunks of size
5571 * (# leaves * SPA_MAXBLOCKSIZE * 4). Within each chunk
5572 * there is a series of ranges to which we can inject errors.
5573 * Each range can accept errors on only a single leaf vdev.
5574 * The error injection ranges are separated by ranges
5575 * which we will not inject errors on any device (DMZs).
5576 * Each DMZ must be large enough such that a single block
5577 * can not straddle it, so that a single block can not be
5578 * a target in two different injection ranges (on different
5581 * For example, with 3 leaves, each chunk looks like:
5582 * 0 to 32M: injection range for leaf 0
5583 * 32M to 64M: DMZ - no injection allowed
5584 * 64M to 96M: injection range for leaf 1
5585 * 96M to 128M: DMZ - no injection allowed
5586 * 128M to 160M: injection range for leaf 2
5587 * 160M to 192M: DMZ - no injection allowed
5589 offset = ztest_random(fsize / (leaves << bshift)) *
5590 (leaves << bshift) + (leaf << bshift) +
5591 (ztest_random(1ULL << (bshift - 1)) & -8ULL);
5594 * Only allow damage to the labels at one end of the vdev.
5596 * If all labels are damaged, the device will be totally
5597 * inaccessible, which will result in loss of data,
5598 * because we also damage (parts of) the other side of
5601 * Additionally, we will always have both an even and an
5602 * odd label, so that we can handle crashes in the
5603 * middle of vdev_config_sync().
5605 if ((leaf & 1) == 0 && offset < VDEV_LABEL_START_SIZE)
5609 * The two end labels are stored at the "end" of the disk, but
5610 * the end of the disk (vdev_psize) is aligned to
5611 * sizeof (vdev_label_t).
5613 uint64_t psize = P2ALIGN(fsize, sizeof (vdev_label_t));
5614 if ((leaf & 1) == 1 &&
5615 offset + sizeof (bad) > psize - VDEV_LABEL_END_SIZE)
5618 mutex_enter(&ztest_vdev_lock);
5619 if (mirror_save != zs->zs_mirrors) {
5620 mutex_exit(&ztest_vdev_lock);
5625 if (pwrite(fd, &bad, sizeof (bad), offset) != sizeof (bad))
5626 fatal(1, "can't inject bad word at 0x%llx in %s",
5629 mutex_exit(&ztest_vdev_lock);
5631 if (ztest_opts.zo_verbose >= 7)
5632 (void) printf("injected bad word into %s,"
5633 " offset 0x%llx\n", pathrand, (u_longlong_t)offset);
5638 umem_free(path0, MAXPATHLEN);
5639 umem_free(pathrand, MAXPATHLEN);
5643 * Verify that DDT repair works as expected.
5646 ztest_ddt_repair(ztest_ds_t *zd, uint64_t id)
5648 ztest_shared_t *zs = ztest_shared;
5649 spa_t *spa = ztest_spa;
5650 objset_t *os = zd->zd_os;
5652 uint64_t object, blocksize, txg, pattern, psize;
5653 enum zio_checksum checksum = spa_dedup_checksum(spa);
5658 int copies = 2 * ZIO_DEDUPDITTO_MIN;
5661 blocksize = ztest_random_blocksize();
5662 blocksize = MIN(blocksize, 2048); /* because we write so many */
5664 od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
5665 ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0, 0);
5667 if (ztest_object_init(zd, od, sizeof (ztest_od_t), B_FALSE) != 0) {
5668 umem_free(od, sizeof (ztest_od_t));
5673 * Take the name lock as writer to prevent anyone else from changing
5674 * the pool and dataset properies we need to maintain during this test.
5676 (void) rw_wrlock(&ztest_name_lock);
5678 if (ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_DEDUP, checksum,
5680 ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_COPIES, 1,
5682 (void) rw_unlock(&ztest_name_lock);
5683 umem_free(od, sizeof (ztest_od_t));
5687 dmu_objset_stats_t dds;
5688 dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
5689 dmu_objset_fast_stat(os, &dds);
5690 dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
5692 object = od[0].od_object;
5693 blocksize = od[0].od_blocksize;
5694 pattern = zs->zs_guid ^ dds.dds_guid;
5696 ASSERT(object != 0);
5698 tx = dmu_tx_create(os);
5699 dmu_tx_hold_write(tx, object, 0, copies * blocksize);
5700 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
5702 (void) rw_unlock(&ztest_name_lock);
5703 umem_free(od, sizeof (ztest_od_t));
5708 * Write all the copies of our block.
5710 for (i = 0; i < copies; i++) {
5711 uint64_t offset = i * blocksize;
5712 int error = dmu_buf_hold(os, object, offset, FTAG, &db,
5713 DMU_READ_NO_PREFETCH);
5715 fatal(B_FALSE, "dmu_buf_hold(%p, %llu, %llu) = %u",
5716 os, (long long)object, (long long) offset, error);
5718 ASSERT(db->db_offset == offset);
5719 ASSERT(db->db_size == blocksize);
5720 ASSERT(ztest_pattern_match(db->db_data, db->db_size, pattern) ||
5721 ztest_pattern_match(db->db_data, db->db_size, 0ULL));
5722 dmu_buf_will_fill(db, tx);
5723 ztest_pattern_set(db->db_data, db->db_size, pattern);
5724 dmu_buf_rele(db, FTAG);
5728 txg_wait_synced(spa_get_dsl(spa), txg);
5731 * Find out what block we got.
5733 VERIFY0(dmu_buf_hold(os, object, 0, FTAG, &db,
5734 DMU_READ_NO_PREFETCH));
5735 blk = *((dmu_buf_impl_t *)db)->db_blkptr;
5736 dmu_buf_rele(db, FTAG);
5739 * Damage the block. Dedup-ditto will save us when we read it later.
5741 psize = BP_GET_PSIZE(&blk);
5742 abd = abd_alloc_linear(psize, B_TRUE);
5743 ztest_pattern_set(abd_to_buf(abd), psize, ~pattern);
5745 (void) zio_wait(zio_rewrite(NULL, spa, 0, &blk,
5746 abd, psize, NULL, NULL, ZIO_PRIORITY_SYNC_WRITE,
5747 ZIO_FLAG_CANFAIL | ZIO_FLAG_INDUCE_DAMAGE, NULL));
5751 (void) rw_unlock(&ztest_name_lock);
5752 umem_free(od, sizeof (ztest_od_t));
5760 ztest_scrub(ztest_ds_t *zd, uint64_t id)
5762 spa_t *spa = ztest_spa;
5764 (void) spa_scan(spa, POOL_SCAN_SCRUB);
5765 (void) poll(NULL, 0, 100); /* wait a moment, then force a restart */
5766 (void) spa_scan(spa, POOL_SCAN_SCRUB);
5770 * Change the guid for the pool.
5774 ztest_reguid(ztest_ds_t *zd, uint64_t id)
5776 spa_t *spa = ztest_spa;
5777 uint64_t orig, load;
5780 if (ztest_opts.zo_mmp_test)
5783 orig = spa_guid(spa);
5784 load = spa_load_guid(spa);
5786 (void) rw_wrlock(&ztest_name_lock);
5787 error = spa_change_guid(spa);
5788 (void) rw_unlock(&ztest_name_lock);
5793 if (ztest_opts.zo_verbose >= 4) {
5794 (void) printf("Changed guid old %llu -> %llu\n",
5795 (u_longlong_t)orig, (u_longlong_t)spa_guid(spa));
5798 VERIFY3U(orig, !=, spa_guid(spa));
5799 VERIFY3U(load, ==, spa_load_guid(spa));
5803 * Rename the pool to a different name and then rename it back.
5807 ztest_spa_rename(ztest_ds_t *zd, uint64_t id)
5809 char *oldname, *newname;
5812 if (ztest_opts.zo_mmp_test)
5815 (void) rw_wrlock(&ztest_name_lock);
5817 oldname = ztest_opts.zo_pool;
5818 newname = umem_alloc(strlen(oldname) + 5, UMEM_NOFAIL);
5819 (void) strcpy(newname, oldname);
5820 (void) strcat(newname, "_tmp");
5825 VERIFY3U(0, ==, spa_rename(oldname, newname));
5828 * Try to open it under the old name, which shouldn't exist
5830 VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
5833 * Open it under the new name and make sure it's still the same spa_t.
5835 VERIFY3U(0, ==, spa_open(newname, &spa, FTAG));
5837 ASSERT(spa == ztest_spa);
5838 spa_close(spa, FTAG);
5841 * Rename it back to the original
5843 VERIFY3U(0, ==, spa_rename(newname, oldname));
5846 * Make sure it can still be opened
5848 VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG));
5850 ASSERT(spa == ztest_spa);
5851 spa_close(spa, FTAG);
5853 umem_free(newname, strlen(newname) + 1);
5855 (void) rw_unlock(&ztest_name_lock);
5859 ztest_fletcher(ztest_ds_t *zd, uint64_t id)
5861 hrtime_t end = gethrtime() + NANOSEC;
5863 while (gethrtime() <= end) {
5864 int run_count = 100;
5866 struct abd *abd_data, *abd_meta;
5871 zio_cksum_t zc_ref_byteswap;
5873 size = ztest_random_blocksize();
5875 buf = umem_alloc(size, UMEM_NOFAIL);
5876 abd_data = abd_alloc(size, B_FALSE);
5877 abd_meta = abd_alloc(size, B_TRUE);
5879 for (i = 0, ptr = buf; i < size / sizeof (*ptr); i++, ptr++)
5880 *ptr = ztest_random(UINT_MAX);
5882 abd_copy_from_buf_off(abd_data, buf, 0, size);
5883 abd_copy_from_buf_off(abd_meta, buf, 0, size);
5885 VERIFY0(fletcher_4_impl_set("scalar"));
5886 fletcher_4_native(buf, size, NULL, &zc_ref);
5887 fletcher_4_byteswap(buf, size, NULL, &zc_ref_byteswap);
5889 VERIFY0(fletcher_4_impl_set("cycle"));
5890 while (run_count-- > 0) {
5892 zio_cksum_t zc_byteswap;
5894 fletcher_4_byteswap(buf, size, NULL, &zc_byteswap);
5895 fletcher_4_native(buf, size, NULL, &zc);
5897 VERIFY0(bcmp(&zc, &zc_ref, sizeof (zc)));
5898 VERIFY0(bcmp(&zc_byteswap, &zc_ref_byteswap,
5899 sizeof (zc_byteswap)));
5901 /* Test ABD - data */
5902 abd_fletcher_4_byteswap(abd_data, size, NULL,
5904 abd_fletcher_4_native(abd_data, size, NULL, &zc);
5906 VERIFY0(bcmp(&zc, &zc_ref, sizeof (zc)));
5907 VERIFY0(bcmp(&zc_byteswap, &zc_ref_byteswap,
5908 sizeof (zc_byteswap)));
5910 /* Test ABD - metadata */
5911 abd_fletcher_4_byteswap(abd_meta, size, NULL,
5913 abd_fletcher_4_native(abd_meta, size, NULL, &zc);
5915 VERIFY0(bcmp(&zc, &zc_ref, sizeof (zc)));
5916 VERIFY0(bcmp(&zc_byteswap, &zc_ref_byteswap,
5917 sizeof (zc_byteswap)));
5921 umem_free(buf, size);
5928 ztest_fletcher_incr(ztest_ds_t *zd, uint64_t id)
5935 zio_cksum_t zc_ref_bswap;
5937 hrtime_t end = gethrtime() + NANOSEC;
5939 while (gethrtime() <= end) {
5940 int run_count = 100;
5942 size = ztest_random_blocksize();
5943 buf = umem_alloc(size, UMEM_NOFAIL);
5945 for (i = 0, ptr = buf; i < size / sizeof (*ptr); i++, ptr++)
5946 *ptr = ztest_random(UINT_MAX);
5948 VERIFY0(fletcher_4_impl_set("scalar"));
5949 fletcher_4_native(buf, size, NULL, &zc_ref);
5950 fletcher_4_byteswap(buf, size, NULL, &zc_ref_bswap);
5952 VERIFY0(fletcher_4_impl_set("cycle"));
5954 while (run_count-- > 0) {
5956 zio_cksum_t zc_bswap;
5959 ZIO_SET_CHECKSUM(&zc, 0, 0, 0, 0);
5960 ZIO_SET_CHECKSUM(&zc_bswap, 0, 0, 0, 0);
5962 while (pos < size) {
5963 size_t inc = 64 * ztest_random(size / 67);
5964 /* sometimes add few bytes to test non-simd */
5965 if (ztest_random(100) < 10)
5966 inc += P2ALIGN(ztest_random(64),
5969 if (inc > (size - pos))
5972 fletcher_4_incremental_native(buf + pos, inc,
5974 fletcher_4_incremental_byteswap(buf + pos, inc,
5980 VERIFY3U(pos, ==, size);
5982 VERIFY(ZIO_CHECKSUM_EQUAL(zc, zc_ref));
5983 VERIFY(ZIO_CHECKSUM_EQUAL(zc_bswap, zc_ref_bswap));
5986 * verify if incremental on the whole buffer is
5987 * equivalent to non-incremental version
5989 ZIO_SET_CHECKSUM(&zc, 0, 0, 0, 0);
5990 ZIO_SET_CHECKSUM(&zc_bswap, 0, 0, 0, 0);
5992 fletcher_4_incremental_native(buf, size, &zc);
5993 fletcher_4_incremental_byteswap(buf, size, &zc_bswap);
5995 VERIFY(ZIO_CHECKSUM_EQUAL(zc, zc_ref));
5996 VERIFY(ZIO_CHECKSUM_EQUAL(zc_bswap, zc_ref_bswap));
5999 umem_free(buf, size);
6004 ztest_check_path(char *path)
6007 /* return true on success */
6008 return (!stat(path, &s));
6012 ztest_get_zdb_bin(char *bin, int len)
6016 * Try to use ZDB_PATH and in-tree zdb path. If not successful, just
6017 * let popen to search through PATH.
6019 if ((zdb_path = getenv("ZDB_PATH"))) {
6020 strlcpy(bin, zdb_path, len); /* In env */
6021 if (!ztest_check_path(bin)) {
6022 ztest_dump_core = 0;
6023 fatal(1, "invalid ZDB_PATH '%s'", bin);
6028 VERIFY(realpath(getexecname(), bin) != NULL);
6029 if (strstr(bin, "/ztest/")) {
6030 strstr(bin, "/ztest/")[0] = '\0'; /* In-tree */
6031 strcat(bin, "/zdb/zdb");
6032 if (ztest_check_path(bin))
6039 * Verify pool integrity by running zdb.
6042 ztest_run_zdb(char *pool)
6048 const int len = MAXPATHLEN + MAXNAMELEN + 20;
6051 bin = umem_alloc(len, UMEM_NOFAIL);
6052 zdb = umem_alloc(len, UMEM_NOFAIL);
6053 zbuf = umem_alloc(1024, UMEM_NOFAIL);
6055 ztest_get_zdb_bin(bin, len);
6058 "%s -bcc%s%s -G -d -U %s %s",
6060 ztest_opts.zo_verbose >= 3 ? "s" : "",
6061 ztest_opts.zo_verbose >= 4 ? "v" : "",
6065 if (ztest_opts.zo_verbose >= 5)
6066 (void) printf("Executing %s\n", strstr(zdb, "zdb "));
6068 fp = popen(zdb, "r");
6070 while (fgets(zbuf, 1024, fp) != NULL)
6071 if (ztest_opts.zo_verbose >= 3)
6072 (void) printf("%s", zbuf);
6074 status = pclose(fp);
6079 ztest_dump_core = 0;
6080 if (WIFEXITED(status))
6081 fatal(0, "'%s' exit code %d", zdb, WEXITSTATUS(status));
6083 fatal(0, "'%s' died with signal %d", zdb, WTERMSIG(status));
6085 umem_free(bin, len);
6086 umem_free(zdb, len);
6087 umem_free(zbuf, 1024);
6091 ztest_walk_pool_directory(char *header)
6095 if (ztest_opts.zo_verbose >= 6)
6096 (void) printf("%s\n", header);
6098 mutex_enter(&spa_namespace_lock);
6099 while ((spa = spa_next(spa)) != NULL)
6100 if (ztest_opts.zo_verbose >= 6)
6101 (void) printf("\t%s\n", spa_name(spa));
6102 mutex_exit(&spa_namespace_lock);
6106 ztest_spa_import_export(char *oldname, char *newname)
6108 nvlist_t *config, *newconfig;
6113 if (ztest_opts.zo_verbose >= 4) {
6114 (void) printf("import/export: old = %s, new = %s\n",
6119 * Clean up from previous runs.
6121 (void) spa_destroy(newname);
6124 * Get the pool's configuration and guid.
6126 VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG));
6129 * Kick off a scrub to tickle scrub/export races.
6131 if (ztest_random(2) == 0)
6132 (void) spa_scan(spa, POOL_SCAN_SCRUB);
6134 pool_guid = spa_guid(spa);
6135 spa_close(spa, FTAG);
6137 ztest_walk_pool_directory("pools before export");
6142 VERIFY3U(0, ==, spa_export(oldname, &config, B_FALSE, B_FALSE));
6144 ztest_walk_pool_directory("pools after export");
6149 newconfig = spa_tryimport(config);
6150 ASSERT(newconfig != NULL);
6151 nvlist_free(newconfig);
6154 * Import it under the new name.
6156 error = spa_import(newname, config, NULL, 0);
6158 dump_nvlist(config, 0);
6159 fatal(B_FALSE, "couldn't import pool %s as %s: error %u",
6160 oldname, newname, error);
6163 ztest_walk_pool_directory("pools after import");
6166 * Try to import it again -- should fail with EEXIST.
6168 VERIFY3U(EEXIST, ==, spa_import(newname, config, NULL, 0));
6171 * Try to import it under a different name -- should fail with EEXIST.
6173 VERIFY3U(EEXIST, ==, spa_import(oldname, config, NULL, 0));
6176 * Verify that the pool is no longer visible under the old name.
6178 VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
6181 * Verify that we can open and close the pool using the new name.
6183 VERIFY3U(0, ==, spa_open(newname, &spa, FTAG));
6184 ASSERT(pool_guid == spa_guid(spa));
6185 spa_close(spa, FTAG);
6187 nvlist_free(config);
6191 ztest_resume(spa_t *spa)
6193 if (spa_suspended(spa) && ztest_opts.zo_verbose >= 6)
6194 (void) printf("resuming from suspended state\n");
6195 spa_vdev_state_enter(spa, SCL_NONE);
6196 vdev_clear(spa, NULL);
6197 (void) spa_vdev_state_exit(spa, NULL, 0);
6198 (void) zio_resume(spa);
6202 ztest_resume_thread(void *arg)
6206 while (!ztest_exiting) {
6207 if (spa_suspended(spa))
6209 (void) poll(NULL, 0, 100);
6212 * Periodically change the zfs_compressed_arc_enabled setting.
6214 if (ztest_random(10) == 0)
6215 zfs_compressed_arc_enabled = ztest_random(2);
6218 * Periodically change the zfs_abd_scatter_enabled setting.
6220 if (ztest_random(10) == 0)
6221 zfs_abd_scatter_enabled = ztest_random(2);
6231 ztest_deadman_alarm(int sig)
6233 fatal(0, "failed to complete within %d seconds of deadline", GRACE);
6238 ztest_execute(int test, ztest_info_t *zi, uint64_t id)
6240 ztest_ds_t *zd = &ztest_ds[id % ztest_opts.zo_datasets];
6241 ztest_shared_callstate_t *zc = ZTEST_GET_SHARED_CALLSTATE(test);
6242 hrtime_t functime = gethrtime();
6245 for (i = 0; i < zi->zi_iters; i++)
6246 zi->zi_func(zd, id);
6248 functime = gethrtime() - functime;
6250 atomic_add_64(&zc->zc_count, 1);
6251 atomic_add_64(&zc->zc_time, functime);
6253 if (ztest_opts.zo_verbose >= 4)
6254 (void) printf("%6.2f sec in %s\n",
6255 (double)functime / NANOSEC, zi->zi_funcname);
6259 ztest_thread(void *arg)
6262 uint64_t id = (uintptr_t)arg;
6263 ztest_shared_t *zs = ztest_shared;
6267 ztest_shared_callstate_t *zc;
6269 while ((now = gethrtime()) < zs->zs_thread_stop) {
6271 * See if it's time to force a crash.
6273 if (now > zs->zs_thread_kill)
6277 * If we're getting ENOSPC with some regularity, stop.
6279 if (zs->zs_enospc_count > 10)
6283 * Pick a random function to execute.
6285 rand = ztest_random(ZTEST_FUNCS);
6286 zi = &ztest_info[rand];
6287 zc = ZTEST_GET_SHARED_CALLSTATE(rand);
6288 call_next = zc->zc_next;
6290 if (now >= call_next &&
6291 atomic_cas_64(&zc->zc_next, call_next, call_next +
6292 ztest_random(2 * zi->zi_interval[0] + 1)) == call_next) {
6293 ztest_execute(rand, zi, id);
6301 ztest_dataset_name(char *dsname, char *pool, int d)
6303 (void) snprintf(dsname, ZFS_MAX_DATASET_NAME_LEN, "%s/ds_%d", pool, d);
6307 ztest_dataset_destroy(int d)
6309 char name[ZFS_MAX_DATASET_NAME_LEN];
6312 ztest_dataset_name(name, ztest_opts.zo_pool, d);
6314 if (ztest_opts.zo_verbose >= 3)
6315 (void) printf("Destroying %s to free up space\n", name);
6318 * Cleanup any non-standard clones and snapshots. In general,
6319 * ztest thread t operates on dataset (t % zopt_datasets),
6320 * so there may be more than one thing to clean up.
6322 for (t = d; t < ztest_opts.zo_threads;
6323 t += ztest_opts.zo_datasets)
6324 ztest_dsl_dataset_cleanup(name, t);
6326 (void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
6327 DS_FIND_SNAPSHOTS | DS_FIND_CHILDREN);
6331 ztest_dataset_dirobj_verify(ztest_ds_t *zd)
6333 uint64_t usedobjs, dirobjs, scratch;
6336 * ZTEST_DIROBJ is the object directory for the entire dataset.
6337 * Therefore, the number of objects in use should equal the
6338 * number of ZTEST_DIROBJ entries, +1 for ZTEST_DIROBJ itself.
6339 * If not, we have an object leak.
6341 * Note that we can only check this in ztest_dataset_open(),
6342 * when the open-context and syncing-context values agree.
6343 * That's because zap_count() returns the open-context value,
6344 * while dmu_objset_space() returns the rootbp fill count.
6346 VERIFY3U(0, ==, zap_count(zd->zd_os, ZTEST_DIROBJ, &dirobjs));
6347 dmu_objset_space(zd->zd_os, &scratch, &scratch, &usedobjs, &scratch);
6348 ASSERT3U(dirobjs + 1, ==, usedobjs);
6352 ztest_dataset_open(int d)
6354 ztest_ds_t *zd = &ztest_ds[d];
6355 uint64_t committed_seq = ZTEST_GET_SHARED_DS(d)->zd_seq;
6358 char name[ZFS_MAX_DATASET_NAME_LEN];
6361 ztest_dataset_name(name, ztest_opts.zo_pool, d);
6363 (void) rw_rdlock(&ztest_name_lock);
6365 error = ztest_dataset_create(name);
6366 if (error == ENOSPC) {
6367 (void) rw_unlock(&ztest_name_lock);
6368 ztest_record_enospc(FTAG);
6371 ASSERT(error == 0 || error == EEXIST);
6373 VERIFY0(ztest_dmu_objset_own(name, DMU_OST_OTHER, B_FALSE,
6375 (void) rw_unlock(&ztest_name_lock);
6377 ztest_zd_init(zd, ZTEST_GET_SHARED_DS(d), os);
6379 zilog = zd->zd_zilog;
6381 if (zilog->zl_header->zh_claim_lr_seq != 0 &&
6382 zilog->zl_header->zh_claim_lr_seq < committed_seq)
6383 fatal(0, "missing log records: claimed %llu < committed %llu",
6384 zilog->zl_header->zh_claim_lr_seq, committed_seq);
6386 ztest_dataset_dirobj_verify(zd);
6388 zil_replay(os, zd, ztest_replay_vector);
6390 ztest_dataset_dirobj_verify(zd);
6392 if (ztest_opts.zo_verbose >= 6)
6393 (void) printf("%s replay %llu blocks, %llu records, seq %llu\n",
6395 (u_longlong_t)zilog->zl_parse_blk_count,
6396 (u_longlong_t)zilog->zl_parse_lr_count,
6397 (u_longlong_t)zilog->zl_replaying_seq);
6399 zilog = zil_open(os, ztest_get_data);
6401 if (zilog->zl_replaying_seq != 0 &&
6402 zilog->zl_replaying_seq < committed_seq)
6403 fatal(0, "missing log records: replayed %llu < committed %llu",
6404 zilog->zl_replaying_seq, committed_seq);
6410 ztest_dataset_close(int d)
6412 ztest_ds_t *zd = &ztest_ds[d];
6414 zil_close(zd->zd_zilog);
6415 txg_wait_synced(spa_get_dsl(zd->zd_os->os_spa), 0);
6416 dmu_objset_disown(zd->zd_os, B_TRUE, zd);
6422 * Kick off threads to run tests on all datasets in parallel.
6425 ztest_run(ztest_shared_t *zs)
6429 kthread_t *resume_thread;
6430 kthread_t **run_threads;
6435 ztest_exiting = B_FALSE;
6438 * Initialize parent/child shared state.
6440 mutex_init(&ztest_vdev_lock, NULL, MUTEX_DEFAULT, NULL);
6441 VERIFY(rwlock_init(&ztest_name_lock, USYNC_THREAD, NULL) == 0);
6443 zs->zs_thread_start = gethrtime();
6444 zs->zs_thread_stop =
6445 zs->zs_thread_start + ztest_opts.zo_passtime * NANOSEC;
6446 zs->zs_thread_stop = MIN(zs->zs_thread_stop, zs->zs_proc_stop);
6447 zs->zs_thread_kill = zs->zs_thread_stop;
6448 if (ztest_random(100) < ztest_opts.zo_killrate) {
6449 zs->zs_thread_kill -=
6450 ztest_random(ztest_opts.zo_passtime * NANOSEC);
6453 mutex_init(&zcl.zcl_callbacks_lock, NULL, MUTEX_DEFAULT, NULL);
6455 list_create(&zcl.zcl_callbacks, sizeof (ztest_cb_data_t),
6456 offsetof(ztest_cb_data_t, zcd_node));
6461 kernel_init(FREAD | FWRITE);
6462 VERIFY0(spa_open(ztest_opts.zo_pool, &spa, FTAG));
6463 spa->spa_debug = B_TRUE;
6464 metaslab_preload_limit = ztest_random(20) + 1;
6467 dmu_objset_stats_t dds;
6468 VERIFY0(ztest_dmu_objset_own(ztest_opts.zo_pool,
6469 DMU_OST_ANY, B_TRUE, B_TRUE, FTAG, &os));
6470 dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
6471 dmu_objset_fast_stat(os, &dds);
6472 dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
6473 zs->zs_guid = dds.dds_guid;
6474 dmu_objset_disown(os, B_TRUE, FTAG);
6476 spa->spa_dedup_ditto = 2 * ZIO_DEDUPDITTO_MIN;
6479 * We don't expect the pool to suspend unless maxfaults == 0,
6480 * in which case ztest_fault_inject() temporarily takes away
6481 * the only valid replica.
6483 if (MAXFAULTS() == 0)
6484 spa->spa_failmode = ZIO_FAILURE_MODE_WAIT;
6486 spa->spa_failmode = ZIO_FAILURE_MODE_PANIC;
6489 * Create a thread to periodically resume suspended I/O.
6491 resume_thread = thread_create(NULL, 0, ztest_resume_thread,
6492 spa, 0, NULL, TS_RUN | TS_JOINABLE, defclsyspri);
6496 * Set a deadman alarm to abort() if we hang.
6498 signal(SIGALRM, ztest_deadman_alarm);
6499 alarm((zs->zs_thread_stop - zs->zs_thread_start) / NANOSEC + GRACE);
6503 * Verify that we can safely inquire about about any object,
6504 * whether it's allocated or not. To make it interesting,
6505 * we probe a 5-wide window around each power of two.
6506 * This hits all edge cases, including zero and the max.
6508 for (t = 0; t < 64; t++) {
6509 for (d = -5; d <= 5; d++) {
6510 error = dmu_object_info(spa->spa_meta_objset,
6511 (1ULL << t) + d, NULL);
6512 ASSERT(error == 0 || error == ENOENT ||
6518 * If we got any ENOSPC errors on the previous run, destroy something.
6520 if (zs->zs_enospc_count != 0) {
6521 int d = ztest_random(ztest_opts.zo_datasets);
6522 ztest_dataset_destroy(d);
6524 zs->zs_enospc_count = 0;
6526 run_threads = umem_zalloc(ztest_opts.zo_threads * sizeof (kthread_t *),
6529 if (ztest_opts.zo_verbose >= 4)
6530 (void) printf("starting main threads...\n");
6533 * Kick off all the tests that run in parallel.
6535 for (t = 0; t < ztest_opts.zo_threads; t++) {
6536 if (t < ztest_opts.zo_datasets && ztest_dataset_open(t) != 0) {
6537 umem_free(run_threads, ztest_opts.zo_threads *
6538 sizeof (kthread_t *));
6542 run_threads[t] = thread_create(NULL, 0, ztest_thread,
6543 (void *)(uintptr_t)t, 0, NULL, TS_RUN | TS_JOINABLE,
6548 * Wait for all of the tests to complete. We go in reverse order
6549 * so we don't close datasets while threads are still using them.
6551 for (t = ztest_opts.zo_threads - 1; t >= 0; t--) {
6552 VERIFY0(thread_join(run_threads[t]));
6553 if (t < ztest_opts.zo_datasets)
6554 ztest_dataset_close(t);
6557 txg_wait_synced(spa_get_dsl(spa), 0);
6559 zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(spa));
6560 zs->zs_space = metaslab_class_get_space(spa_normal_class(spa));
6562 umem_free(run_threads, ztest_opts.zo_threads * sizeof (kthread_t *));
6564 /* Kill the resume thread */
6565 ztest_exiting = B_TRUE;
6566 VERIFY0(thread_join(resume_thread));
6570 * Right before closing the pool, kick off a bunch of async I/O;
6571 * spa_close() should wait for it to complete.
6573 for (object = 1; object < 50; object++) {
6574 dmu_prefetch(spa->spa_meta_objset, object, 0, 0, 1ULL << 20,
6575 ZIO_PRIORITY_SYNC_READ);
6578 /* Verify that at least one commit cb was called in a timely fashion */
6579 if (zc_cb_counter >= ZTEST_COMMIT_CB_MIN_REG)
6580 VERIFY0(zc_min_txg_delay);
6582 spa_close(spa, FTAG);
6585 * Verify that we can loop over all pools.
6587 mutex_enter(&spa_namespace_lock);
6588 for (spa = spa_next(NULL); spa != NULL; spa = spa_next(spa))
6589 if (ztest_opts.zo_verbose > 3)
6590 (void) printf("spa_next: found %s\n", spa_name(spa));
6591 mutex_exit(&spa_namespace_lock);
6594 * Verify that we can export the pool and reimport it under a
6597 if ((ztest_random(2) == 0) && !ztest_opts.zo_mmp_test) {
6598 char name[ZFS_MAX_DATASET_NAME_LEN];
6599 (void) snprintf(name, sizeof (name), "%s_import",
6600 ztest_opts.zo_pool);
6601 ztest_spa_import_export(ztest_opts.zo_pool, name);
6602 ztest_spa_import_export(name, ztest_opts.zo_pool);
6607 list_destroy(&zcl.zcl_callbacks);
6608 mutex_destroy(&zcl.zcl_callbacks_lock);
6609 (void) rwlock_destroy(&ztest_name_lock);
6610 mutex_destroy(&ztest_vdev_lock);
6616 ztest_ds_t *zd = &ztest_ds[0];
6620 if (ztest_opts.zo_verbose >= 3)
6621 (void) printf("testing spa_freeze()...\n");
6623 kernel_init(FREAD | FWRITE);
6624 VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
6625 VERIFY3U(0, ==, ztest_dataset_open(0));
6626 spa->spa_debug = B_TRUE;
6630 * Force the first log block to be transactionally allocated.
6631 * We have to do this before we freeze the pool -- otherwise
6632 * the log chain won't be anchored.
6634 while (BP_IS_HOLE(&zd->zd_zilog->zl_header->zh_log)) {
6635 ztest_dmu_object_alloc_free(zd, 0);
6636 zil_commit(zd->zd_zilog, 0);
6639 txg_wait_synced(spa_get_dsl(spa), 0);
6642 * Freeze the pool. This stops spa_sync() from doing anything,
6643 * so that the only way to record changes from now on is the ZIL.
6648 * Because it is hard to predict how much space a write will actually
6649 * require beforehand, we leave ourselves some fudge space to write over
6652 uint64_t capacity = metaslab_class_get_space(spa_normal_class(spa)) / 2;
6655 * Run tests that generate log records but don't alter the pool config
6656 * or depend on DSL sync tasks (snapshots, objset create/destroy, etc).
6657 * We do a txg_wait_synced() after each iteration to force the txg
6658 * to increase well beyond the last synced value in the uberblock.
6659 * The ZIL should be OK with that.
6661 * Run a random number of times less than zo_maxloops and ensure we do
6662 * not run out of space on the pool.
6664 while (ztest_random(10) != 0 &&
6665 numloops++ < ztest_opts.zo_maxloops &&
6666 metaslab_class_get_alloc(spa_normal_class(spa)) < capacity) {
6668 ztest_od_init(&od, 0, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0, 0);
6669 VERIFY0(ztest_object_init(zd, &od, sizeof (od), B_FALSE));
6670 ztest_io(zd, od.od_object,
6671 ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
6672 txg_wait_synced(spa_get_dsl(spa), 0);
6676 * Commit all of the changes we just generated.
6678 zil_commit(zd->zd_zilog, 0);
6679 txg_wait_synced(spa_get_dsl(spa), 0);
6682 * Close our dataset and close the pool.
6684 ztest_dataset_close(0);
6685 spa_close(spa, FTAG);
6689 * Open and close the pool and dataset to induce log replay.
6691 kernel_init(FREAD | FWRITE);
6692 VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
6693 ASSERT(spa_freeze_txg(spa) == UINT64_MAX);
6694 VERIFY3U(0, ==, ztest_dataset_open(0));
6695 spa->spa_debug = B_TRUE;
6697 txg_wait_synced(spa_get_dsl(spa), 0);
6698 ztest_dataset_close(0);
6699 ztest_reguid(NULL, 0);
6701 spa_close(spa, FTAG);
6706 print_time(hrtime_t t, char *timebuf)
6708 hrtime_t s = t / NANOSEC;
6709 hrtime_t m = s / 60;
6710 hrtime_t h = m / 60;
6711 hrtime_t d = h / 24;
6720 (void) sprintf(timebuf,
6721 "%llud%02lluh%02llum%02llus", d, h, m, s);
6723 (void) sprintf(timebuf, "%lluh%02llum%02llus", h, m, s);
6725 (void) sprintf(timebuf, "%llum%02llus", m, s);
6727 (void) sprintf(timebuf, "%llus", s);
6731 make_random_props(void)
6735 VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
6736 if (ztest_random(2) == 0)
6738 VERIFY(nvlist_add_uint64(props, "autoreplace", 1) == 0);
6744 * Import a storage pool with the given name.
6747 ztest_import(ztest_shared_t *zs)
6749 libzfs_handle_t *hdl;
6750 importargs_t args = { 0 };
6752 nvlist_t *cfg = NULL;
6754 char *searchdirs[nsearch];
6755 char *name = ztest_opts.zo_pool;
6756 int flags = ZFS_IMPORT_MISSING_LOG;
6759 mutex_init(&ztest_vdev_lock, NULL, MUTEX_DEFAULT, NULL);
6760 VERIFY(rwlock_init(&ztest_name_lock, USYNC_THREAD, NULL) == 0);
6762 kernel_init(FREAD | FWRITE);
6763 hdl = libzfs_init();
6765 searchdirs[0] = ztest_opts.zo_dir;
6766 args.paths = nsearch;
6767 args.path = searchdirs;
6768 args.can_be_active = B_FALSE;
6770 error = zpool_tryimport(hdl, name, &cfg, &args);
6772 (void) fatal(0, "No pools found\n");
6774 VERIFY0(spa_import(name, cfg, NULL, flags));
6775 VERIFY0(spa_open(name, &spa, FTAG));
6776 zs->zs_metaslab_sz =
6777 1ULL << spa->spa_root_vdev->vdev_child[0]->vdev_ms_shift;
6778 spa_close(spa, FTAG);
6783 if (!ztest_opts.zo_mmp_test) {
6784 ztest_run_zdb(ztest_opts.zo_pool);
6786 ztest_run_zdb(ztest_opts.zo_pool);
6789 (void) rwlock_destroy(&ztest_name_lock);
6790 mutex_destroy(&ztest_vdev_lock);
6794 * Create a storage pool with the given name and initial vdev size.
6795 * Then test spa_freeze() functionality.
6798 ztest_init(ztest_shared_t *zs)
6801 nvlist_t *nvroot, *props;
6804 mutex_init(&ztest_vdev_lock, NULL, MUTEX_DEFAULT, NULL);
6805 VERIFY(rwlock_init(&ztest_name_lock, USYNC_THREAD, NULL) == 0);
6807 kernel_init(FREAD | FWRITE);
6810 * Create the storage pool.
6812 (void) spa_destroy(ztest_opts.zo_pool);
6813 ztest_shared->zs_vdev_next_leaf = 0;
6815 zs->zs_mirrors = ztest_opts.zo_mirrors;
6816 nvroot = make_vdev_root(NULL, NULL, NULL, ztest_opts.zo_vdev_size, 0,
6817 0, ztest_opts.zo_raidz, zs->zs_mirrors, 1);
6818 props = make_random_props();
6819 for (i = 0; i < SPA_FEATURES; i++) {
6821 VERIFY3S(-1, !=, asprintf(&buf, "feature@%s",
6822 spa_feature_table[i].fi_uname));
6823 VERIFY3U(0, ==, nvlist_add_uint64(props, buf, 0));
6827 spa_create(ztest_opts.zo_pool, nvroot, props, NULL, NULL));
6828 nvlist_free(nvroot);
6831 VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
6832 zs->zs_metaslab_sz =
6833 1ULL << spa->spa_root_vdev->vdev_child[0]->vdev_ms_shift;
6834 spa_close(spa, FTAG);
6838 if (!ztest_opts.zo_mmp_test) {
6839 ztest_run_zdb(ztest_opts.zo_pool);
6841 ztest_run_zdb(ztest_opts.zo_pool);
6844 (void) rwlock_destroy(&ztest_name_lock);
6845 mutex_destroy(&ztest_vdev_lock);
6851 static char ztest_name_data[] = "/tmp/ztest.data.XXXXXX";
6853 ztest_fd_data = mkstemp(ztest_name_data);
6854 ASSERT3S(ztest_fd_data, >=, 0);
6855 (void) unlink(ztest_name_data);
6859 shared_data_size(ztest_shared_hdr_t *hdr)
6863 size = hdr->zh_hdr_size;
6864 size += hdr->zh_opts_size;
6865 size += hdr->zh_size;
6866 size += hdr->zh_stats_size * hdr->zh_stats_count;
6867 size += hdr->zh_ds_size * hdr->zh_ds_count;
6876 ztest_shared_hdr_t *hdr;
6878 hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
6879 PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
6880 ASSERT(hdr != MAP_FAILED);
6882 VERIFY3U(0, ==, ftruncate(ztest_fd_data, sizeof (ztest_shared_hdr_t)));
6884 hdr->zh_hdr_size = sizeof (ztest_shared_hdr_t);
6885 hdr->zh_opts_size = sizeof (ztest_shared_opts_t);
6886 hdr->zh_size = sizeof (ztest_shared_t);
6887 hdr->zh_stats_size = sizeof (ztest_shared_callstate_t);
6888 hdr->zh_stats_count = ZTEST_FUNCS;
6889 hdr->zh_ds_size = sizeof (ztest_shared_ds_t);
6890 hdr->zh_ds_count = ztest_opts.zo_datasets;
6892 size = shared_data_size(hdr);
6893 VERIFY3U(0, ==, ftruncate(ztest_fd_data, size));
6895 (void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
6902 ztest_shared_hdr_t *hdr;
6905 hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
6906 PROT_READ, MAP_SHARED, ztest_fd_data, 0);
6907 ASSERT(hdr != MAP_FAILED);
6909 size = shared_data_size(hdr);
6911 (void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
6912 hdr = ztest_shared_hdr = (void *)mmap(0, P2ROUNDUP(size, getpagesize()),
6913 PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
6914 ASSERT(hdr != MAP_FAILED);
6915 buf = (uint8_t *)hdr;
6917 offset = hdr->zh_hdr_size;
6918 ztest_shared_opts = (void *)&buf[offset];
6919 offset += hdr->zh_opts_size;
6920 ztest_shared = (void *)&buf[offset];
6921 offset += hdr->zh_size;
6922 ztest_shared_callstate = (void *)&buf[offset];
6923 offset += hdr->zh_stats_size * hdr->zh_stats_count;
6924 ztest_shared_ds = (void *)&buf[offset];
6928 exec_child(char *cmd, char *libpath, boolean_t ignorekill, int *statusp)
6932 char *cmdbuf = NULL;
6937 cmdbuf = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
6938 (void) strlcpy(cmdbuf, getexecname(), MAXPATHLEN);
6943 fatal(1, "fork failed");
6945 if (pid == 0) { /* child */
6946 char *emptyargv[2] = { cmd, NULL };
6947 char fd_data_str[12];
6949 struct rlimit rl = { 1024, 1024 };
6950 (void) setrlimit(RLIMIT_NOFILE, &rl);
6952 (void) close(ztest_fd_rand);
6953 VERIFY(11 >= snprintf(fd_data_str, 12, "%d", ztest_fd_data));
6954 VERIFY(0 == setenv("ZTEST_FD_DATA", fd_data_str, 1));
6956 (void) enable_extended_FILE_stdio(-1, -1);
6957 if (libpath != NULL)
6958 VERIFY(0 == setenv("LD_LIBRARY_PATH", libpath, 1));
6959 (void) execv(cmd, emptyargv);
6960 ztest_dump_core = B_FALSE;
6961 fatal(B_TRUE, "exec failed: %s", cmd);
6964 if (cmdbuf != NULL) {
6965 umem_free(cmdbuf, MAXPATHLEN);
6969 while (waitpid(pid, &status, 0) != pid)
6971 if (statusp != NULL)
6974 if (WIFEXITED(status)) {
6975 if (WEXITSTATUS(status) != 0) {
6976 (void) fprintf(stderr, "child exited with code %d\n",
6977 WEXITSTATUS(status));
6981 } else if (WIFSIGNALED(status)) {
6982 if (!ignorekill || WTERMSIG(status) != SIGKILL) {
6983 (void) fprintf(stderr, "child died with signal %d\n",
6989 (void) fprintf(stderr, "something strange happened to child\n");
6996 ztest_run_init(void)
7000 ztest_shared_t *zs = ztest_shared;
7003 * Blow away any existing copy of zpool.cache
7005 (void) remove(spa_config_path);
7007 if (ztest_opts.zo_init == 0) {
7008 if (ztest_opts.zo_verbose >= 1)
7009 (void) printf("Importing pool %s\n",
7010 ztest_opts.zo_pool);
7016 * Create and initialize our storage pool.
7018 for (i = 1; i <= ztest_opts.zo_init; i++) {
7019 bzero(zs, sizeof (ztest_shared_t));
7020 if (ztest_opts.zo_verbose >= 3 &&
7021 ztest_opts.zo_init != 1) {
7022 (void) printf("ztest_init(), pass %d\n", i);
7029 main(int argc, char **argv)
7037 ztest_shared_callstate_t *zc;
7039 char numbuf[NN_NUMBUF_SZ];
7044 char *fd_data_str = getenv("ZTEST_FD_DATA");
7045 struct sigaction action;
7047 (void) setvbuf(stdout, NULL, _IOLBF, 0);
7049 dprintf_setup(&argc, argv);
7051 action.sa_handler = sig_handler;
7052 sigemptyset(&action.sa_mask);
7053 action.sa_flags = 0;
7055 if (sigaction(SIGSEGV, &action, NULL) < 0) {
7056 (void) fprintf(stderr, "ztest: cannot catch SIGSEGV: %s.\n",
7061 if (sigaction(SIGABRT, &action, NULL) < 0) {
7062 (void) fprintf(stderr, "ztest: cannot catch SIGABRT: %s.\n",
7067 ztest_fd_rand = open("/dev/urandom", O_RDONLY);
7068 ASSERT3S(ztest_fd_rand, >=, 0);
7071 process_options(argc, argv);
7076 bcopy(&ztest_opts, ztest_shared_opts,
7077 sizeof (*ztest_shared_opts));
7079 ztest_fd_data = atoi(fd_data_str);
7081 bcopy(ztest_shared_opts, &ztest_opts, sizeof (ztest_opts));
7083 ASSERT3U(ztest_opts.zo_datasets, ==, ztest_shared_hdr->zh_ds_count);
7085 /* Override location of zpool.cache */
7086 VERIFY(asprintf((char **)&spa_config_path, "%s/zpool.cache",
7087 ztest_opts.zo_dir) != -1);
7089 ztest_ds = umem_alloc(ztest_opts.zo_datasets * sizeof (ztest_ds_t),
7094 metaslab_gang_bang = ztest_opts.zo_metaslab_gang_bang;
7095 metaslab_df_alloc_threshold =
7096 zs->zs_metaslab_df_alloc_threshold;
7105 hasalt = (strlen(ztest_opts.zo_alt_ztest) != 0);
7107 if (ztest_opts.zo_verbose >= 1) {
7108 (void) printf("%llu vdevs, %d datasets, %d threads,"
7109 " %llu seconds...\n",
7110 (u_longlong_t)ztest_opts.zo_vdevs,
7111 ztest_opts.zo_datasets,
7112 ztest_opts.zo_threads,
7113 (u_longlong_t)ztest_opts.zo_time);
7116 cmd = umem_alloc(MAXNAMELEN, UMEM_NOFAIL);
7117 (void) strlcpy(cmd, getexecname(), MAXNAMELEN);
7119 zs->zs_do_init = B_TRUE;
7120 if (strlen(ztest_opts.zo_alt_ztest) != 0) {
7121 if (ztest_opts.zo_verbose >= 1) {
7122 (void) printf("Executing older ztest for "
7123 "initialization: %s\n", ztest_opts.zo_alt_ztest);
7125 VERIFY(!exec_child(ztest_opts.zo_alt_ztest,
7126 ztest_opts.zo_alt_libpath, B_FALSE, NULL));
7128 VERIFY(!exec_child(NULL, NULL, B_FALSE, NULL));
7130 zs->zs_do_init = B_FALSE;
7132 zs->zs_proc_start = gethrtime();
7133 zs->zs_proc_stop = zs->zs_proc_start + ztest_opts.zo_time * NANOSEC;
7135 for (f = 0; f < ZTEST_FUNCS; f++) {
7136 zi = &ztest_info[f];
7137 zc = ZTEST_GET_SHARED_CALLSTATE(f);
7138 if (zs->zs_proc_start + zi->zi_interval[0] > zs->zs_proc_stop)
7139 zc->zc_next = UINT64_MAX;
7141 zc->zc_next = zs->zs_proc_start +
7142 ztest_random(2 * zi->zi_interval[0] + 1);
7146 * Run the tests in a loop. These tests include fault injection
7147 * to verify that self-healing data works, and forced crashes
7148 * to verify that we never lose on-disk consistency.
7150 while (gethrtime() < zs->zs_proc_stop) {
7155 * Initialize the workload counters for each function.
7157 for (f = 0; f < ZTEST_FUNCS; f++) {
7158 zc = ZTEST_GET_SHARED_CALLSTATE(f);
7163 /* Set the allocation switch size */
7164 zs->zs_metaslab_df_alloc_threshold =
7165 ztest_random(zs->zs_metaslab_sz / 4) + 1;
7167 if (!hasalt || ztest_random(2) == 0) {
7168 if (hasalt && ztest_opts.zo_verbose >= 1) {
7169 (void) printf("Executing newer ztest: %s\n",
7173 killed = exec_child(cmd, NULL, B_TRUE, &status);
7175 if (hasalt && ztest_opts.zo_verbose >= 1) {
7176 (void) printf("Executing older ztest: %s\n",
7177 ztest_opts.zo_alt_ztest);
7180 killed = exec_child(ztest_opts.zo_alt_ztest,
7181 ztest_opts.zo_alt_libpath, B_TRUE, &status);
7188 if (ztest_opts.zo_verbose >= 1) {
7189 hrtime_t now = gethrtime();
7191 now = MIN(now, zs->zs_proc_stop);
7192 print_time(zs->zs_proc_stop - now, timebuf);
7193 nicenum(zs->zs_space, numbuf, sizeof (numbuf));
7195 (void) printf("Pass %3d, %8s, %3llu ENOSPC, "
7196 "%4.1f%% of %5s used, %3.0f%% done, %8s to go\n",
7198 WIFEXITED(status) ? "Complete" : "SIGKILL",
7199 (u_longlong_t)zs->zs_enospc_count,
7200 100.0 * zs->zs_alloc / zs->zs_space,
7202 100.0 * (now - zs->zs_proc_start) /
7203 (ztest_opts.zo_time * NANOSEC), timebuf);
7206 if (ztest_opts.zo_verbose >= 2) {
7207 (void) printf("\nWorkload summary:\n\n");
7208 (void) printf("%7s %9s %s\n",
7209 "Calls", "Time", "Function");
7210 (void) printf("%7s %9s %s\n",
7211 "-----", "----", "--------");
7212 for (f = 0; f < ZTEST_FUNCS; f++) {
7213 zi = &ztest_info[f];
7214 zc = ZTEST_GET_SHARED_CALLSTATE(f);
7215 print_time(zc->zc_time, timebuf);
7216 (void) printf("%7llu %9s %s\n",
7217 (u_longlong_t)zc->zc_count, timebuf,
7220 (void) printf("\n");
7224 * It's possible that we killed a child during a rename test,
7225 * in which case we'll have a 'ztest_tmp' pool lying around
7226 * instead of 'ztest'. Do a blind rename in case this happened.
7229 if (spa_open(ztest_opts.zo_pool, &spa, FTAG) == 0) {
7230 spa_close(spa, FTAG);
7232 char tmpname[ZFS_MAX_DATASET_NAME_LEN];
7234 kernel_init(FREAD | FWRITE);
7235 (void) snprintf(tmpname, sizeof (tmpname), "%s_tmp",
7236 ztest_opts.zo_pool);
7237 (void) spa_rename(tmpname, ztest_opts.zo_pool);
7241 if (!ztest_opts.zo_mmp_test)
7242 ztest_run_zdb(ztest_opts.zo_pool);
7245 if (ztest_opts.zo_verbose >= 1) {
7247 (void) printf("%d runs of older ztest: %s\n", older,
7248 ztest_opts.zo_alt_ztest);
7249 (void) printf("%d runs of newer ztest: %s\n", newer,
7252 (void) printf("%d killed, %d completed, %.0f%% kill rate\n",
7253 kills, iters - kills, (100.0 * kills) / MAX(1, iters));
7256 umem_free(cmd, MAXNAMELEN);