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, 2017 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_force_ganging;
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_force_ganging = 32 << 10
198 extern uint64_t metaslab_force_ganging;
199 extern uint64_t metaslab_df_alloc_threshold;
200 extern unsigned long zfs_deadman_synctime_ms;
201 extern int metaslab_preload_limit;
202 extern boolean_t zfs_compressed_arc_enabled;
203 extern int zfs_abd_scatter_enabled;
204 extern int dmu_object_alloc_chunk_shift;
206 static ztest_shared_opts_t *ztest_shared_opts;
207 static ztest_shared_opts_t ztest_opts;
208 static char *ztest_wkeydata = "abcdefghijklmnopqrstuvwxyz012345";
210 typedef struct ztest_shared_ds {
214 static ztest_shared_ds_t *ztest_shared_ds;
215 #define ZTEST_GET_SHARED_DS(d) (&ztest_shared_ds[d])
217 #define BT_MAGIC 0x123456789abcdefULL
218 #define MAXFAULTS(zs) \
219 (MAX((zs)->zs_mirrors, 1) * (ztest_opts.zo_raidz_parity + 1) - 1)
223 ZTEST_IO_WRITE_PATTERN,
224 ZTEST_IO_WRITE_ZEROES,
231 typedef struct ztest_block_tag {
235 uint64_t bt_dnodesize;
242 typedef struct bufwad {
260 #define ZTEST_RANGE_LOCKS 64
261 #define ZTEST_OBJECT_LOCKS 64
264 * Object descriptor. Used as a template for object lookup/create/remove.
266 typedef struct ztest_od {
269 dmu_object_type_t od_type;
270 dmu_object_type_t od_crtype;
271 uint64_t od_blocksize;
272 uint64_t od_crblocksize;
273 uint64_t od_crdnodesize;
276 char od_name[ZFS_MAX_DATASET_NAME_LEN];
282 typedef struct ztest_ds {
283 ztest_shared_ds_t *zd_shared;
285 pthread_rwlock_t zd_zilog_lock;
287 ztest_od_t *zd_od; /* debugging aid */
288 char zd_name[ZFS_MAX_DATASET_NAME_LEN];
289 kmutex_t zd_dirobj_lock;
290 rll_t zd_object_lock[ZTEST_OBJECT_LOCKS];
291 zll_t zd_range_lock[ZTEST_RANGE_LOCKS];
295 * Per-iteration state.
297 typedef void ztest_func_t(ztest_ds_t *zd, uint64_t id);
299 typedef struct ztest_info {
300 ztest_func_t *zi_func; /* test function */
301 uint64_t zi_iters; /* iterations per execution */
302 uint64_t *zi_interval; /* execute every <interval> seconds */
303 const char *zi_funcname; /* name of test function */
306 typedef struct ztest_shared_callstate {
307 uint64_t zc_count; /* per-pass count */
308 uint64_t zc_time; /* per-pass time */
309 uint64_t zc_next; /* next time to call this function */
310 } ztest_shared_callstate_t;
312 static ztest_shared_callstate_t *ztest_shared_callstate;
313 #define ZTEST_GET_SHARED_CALLSTATE(c) (&ztest_shared_callstate[c])
315 ztest_func_t ztest_dmu_read_write;
316 ztest_func_t ztest_dmu_write_parallel;
317 ztest_func_t ztest_dmu_object_alloc_free;
318 ztest_func_t ztest_dmu_object_next_chunk;
319 ztest_func_t ztest_dmu_commit_callbacks;
320 ztest_func_t ztest_zap;
321 ztest_func_t ztest_zap_parallel;
322 ztest_func_t ztest_zil_commit;
323 ztest_func_t ztest_zil_remount;
324 ztest_func_t ztest_dmu_read_write_zcopy;
325 ztest_func_t ztest_dmu_objset_create_destroy;
326 ztest_func_t ztest_dmu_prealloc;
327 ztest_func_t ztest_fzap;
328 ztest_func_t ztest_dmu_snapshot_create_destroy;
329 ztest_func_t ztest_dsl_prop_get_set;
330 ztest_func_t ztest_spa_prop_get_set;
331 ztest_func_t ztest_spa_create_destroy;
332 ztest_func_t ztest_fault_inject;
333 ztest_func_t ztest_ddt_repair;
334 ztest_func_t ztest_dmu_snapshot_hold;
335 ztest_func_t ztest_mmp_enable_disable;
336 ztest_func_t ztest_spa_rename;
337 ztest_func_t ztest_scrub;
338 ztest_func_t ztest_dsl_dataset_promote_busy;
339 ztest_func_t ztest_vdev_attach_detach;
340 ztest_func_t ztest_vdev_LUN_growth;
341 ztest_func_t ztest_vdev_add_remove;
342 ztest_func_t ztest_vdev_aux_add_remove;
343 ztest_func_t ztest_split_pool;
344 ztest_func_t ztest_reguid;
345 ztest_func_t ztest_spa_upgrade;
346 ztest_func_t ztest_device_removal;
347 ztest_func_t ztest_remap_blocks;
348 ztest_func_t ztest_spa_checkpoint_create_discard;
349 ztest_func_t ztest_fletcher;
350 ztest_func_t ztest_fletcher_incr;
351 ztest_func_t ztest_verify_dnode_bt;
353 uint64_t zopt_always = 0ULL * NANOSEC; /* all the time */
354 uint64_t zopt_incessant = 1ULL * NANOSEC / 10; /* every 1/10 second */
355 uint64_t zopt_often = 1ULL * NANOSEC; /* every second */
356 uint64_t zopt_sometimes = 10ULL * NANOSEC; /* every 10 seconds */
357 uint64_t zopt_rarely = 60ULL * NANOSEC; /* every 60 seconds */
359 #define ZTI_INIT(func, iters, interval) \
360 { .zi_func = (func), \
361 .zi_iters = (iters), \
362 .zi_interval = (interval), \
363 .zi_funcname = # func }
365 ztest_info_t ztest_info[] = {
366 ZTI_INIT(ztest_dmu_read_write, 1, &zopt_always),
367 ZTI_INIT(ztest_dmu_write_parallel, 10, &zopt_always),
368 ZTI_INIT(ztest_dmu_object_alloc_free, 1, &zopt_always),
369 ZTI_INIT(ztest_dmu_object_next_chunk, 1, &zopt_sometimes),
370 ZTI_INIT(ztest_dmu_commit_callbacks, 1, &zopt_always),
371 ZTI_INIT(ztest_zap, 30, &zopt_always),
372 ZTI_INIT(ztest_zap_parallel, 100, &zopt_always),
373 ZTI_INIT(ztest_split_pool, 1, &zopt_always),
374 ZTI_INIT(ztest_zil_commit, 1, &zopt_incessant),
375 ZTI_INIT(ztest_zil_remount, 1, &zopt_sometimes),
376 ZTI_INIT(ztest_dmu_read_write_zcopy, 1, &zopt_often),
377 ZTI_INIT(ztest_dmu_objset_create_destroy, 1, &zopt_often),
378 ZTI_INIT(ztest_dsl_prop_get_set, 1, &zopt_often),
379 ZTI_INIT(ztest_spa_prop_get_set, 1, &zopt_sometimes),
381 ZTI_INIT(ztest_dmu_prealloc, 1, &zopt_sometimes),
383 ZTI_INIT(ztest_fzap, 1, &zopt_sometimes),
384 ZTI_INIT(ztest_dmu_snapshot_create_destroy, 1, &zopt_sometimes),
385 ZTI_INIT(ztest_spa_create_destroy, 1, &zopt_sometimes),
386 ZTI_INIT(ztest_fault_inject, 1, &zopt_sometimes),
387 ZTI_INIT(ztest_ddt_repair, 1, &zopt_sometimes),
388 ZTI_INIT(ztest_dmu_snapshot_hold, 1, &zopt_sometimes),
389 ZTI_INIT(ztest_mmp_enable_disable, 1, &zopt_sometimes),
390 ZTI_INIT(ztest_reguid, 1, &zopt_rarely),
391 ZTI_INIT(ztest_spa_rename, 1, &zopt_rarely),
392 ZTI_INIT(ztest_scrub, 1, &zopt_rarely),
393 ZTI_INIT(ztest_spa_upgrade, 1, &zopt_rarely),
394 ZTI_INIT(ztest_dsl_dataset_promote_busy, 1, &zopt_rarely),
395 ZTI_INIT(ztest_vdev_attach_detach, 1, &zopt_sometimes),
396 ZTI_INIT(ztest_vdev_LUN_growth, 1, &zopt_rarely),
397 ZTI_INIT(ztest_vdev_add_remove, 1, &ztest_opts.zo_vdevtime),
398 ZTI_INIT(ztest_vdev_aux_add_remove, 1, &ztest_opts.zo_vdevtime),
399 ZTI_INIT(ztest_device_removal, 1, &zopt_sometimes),
400 ZTI_INIT(ztest_remap_blocks, 1, &zopt_sometimes),
401 ZTI_INIT(ztest_spa_checkpoint_create_discard, 1, &zopt_rarely),
402 ZTI_INIT(ztest_fletcher, 1, &zopt_rarely),
403 ZTI_INIT(ztest_fletcher_incr, 1, &zopt_rarely),
404 ZTI_INIT(ztest_verify_dnode_bt, 1, &zopt_sometimes),
407 #define ZTEST_FUNCS (sizeof (ztest_info) / sizeof (ztest_info_t))
410 * The following struct is used to hold a list of uncalled commit callbacks.
411 * The callbacks are ordered by txg number.
413 typedef struct ztest_cb_list {
414 kmutex_t zcl_callbacks_lock;
415 list_t zcl_callbacks;
419 * Stuff we need to share writably between parent and child.
421 typedef struct ztest_shared {
422 boolean_t zs_do_init;
423 hrtime_t zs_proc_start;
424 hrtime_t zs_proc_stop;
425 hrtime_t zs_thread_start;
426 hrtime_t zs_thread_stop;
427 hrtime_t zs_thread_kill;
428 uint64_t zs_enospc_count;
429 uint64_t zs_vdev_next_leaf;
430 uint64_t zs_vdev_aux;
435 uint64_t zs_metaslab_sz;
436 uint64_t zs_metaslab_df_alloc_threshold;
440 #define ID_PARALLEL -1ULL
442 static char ztest_dev_template[] = "%s/%s.%llua";
443 static char ztest_aux_template[] = "%s/%s.%s.%llu";
444 ztest_shared_t *ztest_shared;
446 static spa_t *ztest_spa = NULL;
447 static ztest_ds_t *ztest_ds;
449 static kmutex_t ztest_vdev_lock;
450 static boolean_t ztest_device_removal_active = B_FALSE;
451 static kmutex_t ztest_checkpoint_lock;
454 * The ztest_name_lock protects the pool and dataset namespace used by
455 * the individual tests. To modify the namespace, consumers must grab
456 * this lock as writer. Grabbing the lock as reader will ensure that the
457 * namespace does not change while the lock is held.
459 static pthread_rwlock_t ztest_name_lock;
461 static boolean_t ztest_dump_core = B_TRUE;
462 static boolean_t ztest_dump_debug_buffer = B_FALSE;
463 static boolean_t ztest_exiting;
465 /* Global commit callback list */
466 static ztest_cb_list_t zcl;
467 /* Commit cb delay */
468 static uint64_t zc_min_txg_delay = UINT64_MAX;
469 static int zc_cb_counter = 0;
472 * Minimum number of commit callbacks that need to be registered for us to check
473 * whether the minimum txg delay is acceptable.
475 #define ZTEST_COMMIT_CB_MIN_REG 100
478 * If a number of txgs equal to this threshold have been created after a commit
479 * callback has been registered but not called, then we assume there is an
480 * implementation bug.
482 #define ZTEST_COMMIT_CB_THRESH (TXG_CONCURRENT_STATES + 1000)
484 extern uint64_t metaslab_gang_bang;
485 extern uint64_t metaslab_df_alloc_threshold;
488 ZTEST_META_DNODE = 0,
493 static void usage(boolean_t) __NORETURN;
496 * These libumem hooks provide a reasonable set of defaults for the allocator's
497 * debugging facilities.
500 _umem_debug_init(void)
502 return ("default,verbose"); /* $UMEM_DEBUG setting */
506 _umem_logging_init(void)
508 return ("fail,contents"); /* $UMEM_LOGGING setting */
512 dump_debug_buffer(void)
514 if (!ztest_dump_debug_buffer)
518 zfs_dbgmsg_print("ztest");
521 #define BACKTRACE_SZ 100
523 static void sig_handler(int signo)
525 struct sigaction action;
526 #ifdef __GLIBC__ /* backtrace() is a GNU extension */
528 void *buffer[BACKTRACE_SZ];
530 nptrs = backtrace(buffer, BACKTRACE_SZ);
531 backtrace_symbols_fd(buffer, nptrs, STDERR_FILENO);
536 * Restore default action and re-raise signal so SIGSEGV and
537 * SIGABRT can trigger a core dump.
539 action.sa_handler = SIG_DFL;
540 sigemptyset(&action.sa_mask);
542 (void) sigaction(signo, &action, NULL);
546 #define FATAL_MSG_SZ 1024
551 fatal(int do_perror, char *message, ...)
554 int save_errno = errno;
557 (void) fflush(stdout);
558 buf = umem_alloc(FATAL_MSG_SZ, UMEM_NOFAIL);
560 va_start(args, message);
561 (void) sprintf(buf, "ztest: ");
563 (void) vsprintf(buf + strlen(buf), message, args);
566 (void) snprintf(buf + strlen(buf), FATAL_MSG_SZ - strlen(buf),
567 ": %s", strerror(save_errno));
569 (void) fprintf(stderr, "%s\n", buf);
570 fatal_msg = buf; /* to ease debugging */
580 str2shift(const char *buf)
582 const char *ends = "BKMGTPEZ";
587 for (i = 0; i < strlen(ends); i++) {
588 if (toupper(buf[0]) == ends[i])
591 if (i == strlen(ends)) {
592 (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n",
596 if (buf[1] == '\0' || (toupper(buf[1]) == 'B' && buf[2] == '\0')) {
599 (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n", buf);
605 nicenumtoull(const char *buf)
610 val = strtoull(buf, &end, 0);
612 (void) fprintf(stderr, "ztest: bad numeric value: %s\n", buf);
614 } else if (end[0] == '.') {
615 double fval = strtod(buf, &end);
616 fval *= pow(2, str2shift(end));
617 if (fval > UINT64_MAX) {
618 (void) fprintf(stderr, "ztest: value too large: %s\n",
622 val = (uint64_t)fval;
624 int shift = str2shift(end);
625 if (shift >= 64 || (val << shift) >> shift != val) {
626 (void) fprintf(stderr, "ztest: value too large: %s\n",
636 usage(boolean_t requested)
638 const ztest_shared_opts_t *zo = &ztest_opts_defaults;
640 char nice_vdev_size[NN_NUMBUF_SZ];
641 char nice_force_ganging[NN_NUMBUF_SZ];
642 FILE *fp = requested ? stdout : stderr;
644 nicenum(zo->zo_vdev_size, nice_vdev_size, sizeof (nice_vdev_size));
645 nicenum(zo->zo_metaslab_force_ganging, nice_force_ganging,
646 sizeof (nice_force_ganging));
648 (void) fprintf(fp, "Usage: %s\n"
649 "\t[-v vdevs (default: %llu)]\n"
650 "\t[-s size_of_each_vdev (default: %s)]\n"
651 "\t[-a alignment_shift (default: %d)] use 0 for random\n"
652 "\t[-m mirror_copies (default: %d)]\n"
653 "\t[-r raidz_disks (default: %d)]\n"
654 "\t[-R raidz_parity (default: %d)]\n"
655 "\t[-d datasets (default: %d)]\n"
656 "\t[-t threads (default: %d)]\n"
657 "\t[-g gang_block_threshold (default: %s)]\n"
658 "\t[-i init_count (default: %d)] initialize pool i times\n"
659 "\t[-k kill_percentage (default: %llu%%)]\n"
660 "\t[-p pool_name (default: %s)]\n"
661 "\t[-f dir (default: %s)] file directory for vdev files\n"
662 "\t[-M] Multi-host simulate pool imported on remote host\n"
663 "\t[-V] verbose (use multiple times for ever more blather)\n"
664 "\t[-E] use existing pool instead of creating new one\n"
665 "\t[-T time (default: %llu sec)] total run time\n"
666 "\t[-F freezeloops (default: %llu)] max loops in spa_freeze()\n"
667 "\t[-P passtime (default: %llu sec)] time per pass\n"
668 "\t[-B alt_ztest (default: <none>)] alternate ztest path\n"
669 "\t[-o variable=value] ... set global variable to an unsigned\n"
670 "\t 32-bit integer value\n"
671 "\t[-G dump zfs_dbgmsg buffer before exiting due to an error\n"
672 "\t[-h] (print help)\n"
675 (u_longlong_t)zo->zo_vdevs, /* -v */
676 nice_vdev_size, /* -s */
677 zo->zo_ashift, /* -a */
678 zo->zo_mirrors, /* -m */
679 zo->zo_raidz, /* -r */
680 zo->zo_raidz_parity, /* -R */
681 zo->zo_datasets, /* -d */
682 zo->zo_threads, /* -t */
683 nice_force_ganging, /* -g */
684 zo->zo_init, /* -i */
685 (u_longlong_t)zo->zo_killrate, /* -k */
686 zo->zo_pool, /* -p */
688 (u_longlong_t)zo->zo_time, /* -T */
689 (u_longlong_t)zo->zo_maxloops, /* -F */
690 (u_longlong_t)zo->zo_passtime);
691 exit(requested ? 0 : 1);
695 process_options(int argc, char **argv)
698 ztest_shared_opts_t *zo = &ztest_opts;
702 char altdir[MAXNAMELEN] = { 0 };
704 bcopy(&ztest_opts_defaults, zo, sizeof (*zo));
706 while ((opt = getopt(argc, argv,
707 "v:s:a:m:r:R:d:t:g:i:k:p:f:MVET:P:hF:B:o:G")) != EOF) {
724 value = nicenumtoull(optarg);
728 zo->zo_vdevs = value;
731 zo->zo_vdev_size = MAX(SPA_MINDEVSIZE, value);
734 zo->zo_ashift = value;
737 zo->zo_mirrors = value;
740 zo->zo_raidz = MAX(1, value);
743 zo->zo_raidz_parity = MIN(MAX(value, 1), 3);
746 zo->zo_datasets = MAX(1, value);
749 zo->zo_threads = MAX(1, value);
752 zo->zo_metaslab_force_ganging =
753 MAX(SPA_MINBLOCKSIZE << 1, value);
759 zo->zo_killrate = value;
762 (void) strlcpy(zo->zo_pool, optarg,
763 sizeof (zo->zo_pool));
766 path = realpath(optarg, NULL);
768 (void) fprintf(stderr, "error: %s: %s\n",
769 optarg, strerror(errno));
772 (void) strlcpy(zo->zo_dir, path,
773 sizeof (zo->zo_dir));
790 zo->zo_passtime = MAX(1, value);
793 zo->zo_maxloops = MAX(1, value);
796 (void) strlcpy(altdir, optarg, sizeof (altdir));
799 if (set_global_var(optarg) != 0)
803 ztest_dump_debug_buffer = B_TRUE;
815 zo->zo_raidz_parity = MIN(zo->zo_raidz_parity, zo->zo_raidz - 1);
818 (zo->zo_vdevs > 0 ? zo->zo_time * NANOSEC / zo->zo_vdevs :
821 if (strlen(altdir) > 0) {
829 cmd = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
830 realaltdir = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
832 VERIFY(NULL != realpath(getexecname(), cmd));
833 if (0 != access(altdir, F_OK)) {
834 ztest_dump_core = B_FALSE;
835 fatal(B_TRUE, "invalid alternate ztest path: %s",
838 VERIFY(NULL != realpath(altdir, realaltdir));
841 * 'cmd' should be of the form "<anything>/usr/bin/<isa>/ztest".
842 * We want to extract <isa> to determine if we should use
843 * 32 or 64 bit binaries.
845 bin = strstr(cmd, "/usr/bin/");
846 ztest = strstr(bin, "/ztest");
848 isalen = ztest - isa;
849 (void) snprintf(zo->zo_alt_ztest, sizeof (zo->zo_alt_ztest),
850 "%s/usr/bin/%.*s/ztest", realaltdir, isalen, isa);
851 (void) snprintf(zo->zo_alt_libpath, sizeof (zo->zo_alt_libpath),
852 "%s/usr/lib/%.*s", realaltdir, isalen, isa);
854 if (0 != access(zo->zo_alt_ztest, X_OK)) {
855 ztest_dump_core = B_FALSE;
856 fatal(B_TRUE, "invalid alternate ztest: %s",
858 } else if (0 != access(zo->zo_alt_libpath, X_OK)) {
859 ztest_dump_core = B_FALSE;
860 fatal(B_TRUE, "invalid alternate lib directory %s",
864 umem_free(cmd, MAXPATHLEN);
865 umem_free(realaltdir, MAXPATHLEN);
870 ztest_kill(ztest_shared_t *zs)
872 zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(ztest_spa));
873 zs->zs_space = metaslab_class_get_space(spa_normal_class(ztest_spa));
876 * Before we kill off ztest, make sure that the config is updated.
877 * See comment above spa_write_cachefile().
879 mutex_enter(&spa_namespace_lock);
880 spa_write_cachefile(ztest_spa, B_FALSE, B_FALSE);
881 mutex_exit(&spa_namespace_lock);
883 (void) kill(getpid(), SIGKILL);
887 ztest_random(uint64_t range)
891 ASSERT3S(ztest_fd_rand, >=, 0);
896 if (read(ztest_fd_rand, &r, sizeof (r)) != sizeof (r))
897 fatal(1, "short read from /dev/urandom");
904 ztest_record_enospc(const char *s)
906 ztest_shared->zs_enospc_count++;
910 ztest_get_ashift(void)
912 if (ztest_opts.zo_ashift == 0)
913 return (SPA_MINBLOCKSHIFT + ztest_random(5));
914 return (ztest_opts.zo_ashift);
918 make_vdev_file(char *path, char *aux, char *pool, size_t size, uint64_t ashift)
924 pathbuf = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
927 ashift = ztest_get_ashift();
933 vdev = ztest_shared->zs_vdev_aux;
934 (void) snprintf(path, MAXPATHLEN,
935 ztest_aux_template, ztest_opts.zo_dir,
936 pool == NULL ? ztest_opts.zo_pool : pool,
939 vdev = ztest_shared->zs_vdev_next_leaf++;
940 (void) snprintf(path, MAXPATHLEN,
941 ztest_dev_template, ztest_opts.zo_dir,
942 pool == NULL ? ztest_opts.zo_pool : pool, vdev);
947 int fd = open(path, O_RDWR | O_CREAT | O_TRUNC, 0666);
949 fatal(1, "can't open %s", path);
950 if (ftruncate(fd, size) != 0)
951 fatal(1, "can't ftruncate %s", path);
955 VERIFY(nvlist_alloc(&file, NV_UNIQUE_NAME, 0) == 0);
956 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_TYPE, VDEV_TYPE_FILE) == 0);
957 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_PATH, path) == 0);
958 VERIFY(nvlist_add_uint64(file, ZPOOL_CONFIG_ASHIFT, ashift) == 0);
959 umem_free(pathbuf, MAXPATHLEN);
965 make_vdev_raidz(char *path, char *aux, char *pool, size_t size,
966 uint64_t ashift, int r)
968 nvlist_t *raidz, **child;
972 return (make_vdev_file(path, aux, pool, size, ashift));
973 child = umem_alloc(r * sizeof (nvlist_t *), UMEM_NOFAIL);
975 for (c = 0; c < r; c++)
976 child[c] = make_vdev_file(path, aux, pool, size, ashift);
978 VERIFY(nvlist_alloc(&raidz, NV_UNIQUE_NAME, 0) == 0);
979 VERIFY(nvlist_add_string(raidz, ZPOOL_CONFIG_TYPE,
980 VDEV_TYPE_RAIDZ) == 0);
981 VERIFY(nvlist_add_uint64(raidz, ZPOOL_CONFIG_NPARITY,
982 ztest_opts.zo_raidz_parity) == 0);
983 VERIFY(nvlist_add_nvlist_array(raidz, ZPOOL_CONFIG_CHILDREN,
986 for (c = 0; c < r; c++)
987 nvlist_free(child[c]);
989 umem_free(child, r * sizeof (nvlist_t *));
995 make_vdev_mirror(char *path, char *aux, char *pool, size_t size,
996 uint64_t ashift, int r, int m)
998 nvlist_t *mirror, **child;
1002 return (make_vdev_raidz(path, aux, pool, size, ashift, r));
1004 child = umem_alloc(m * sizeof (nvlist_t *), UMEM_NOFAIL);
1006 for (c = 0; c < m; c++)
1007 child[c] = make_vdev_raidz(path, aux, pool, size, ashift, r);
1009 VERIFY(nvlist_alloc(&mirror, NV_UNIQUE_NAME, 0) == 0);
1010 VERIFY(nvlist_add_string(mirror, ZPOOL_CONFIG_TYPE,
1011 VDEV_TYPE_MIRROR) == 0);
1012 VERIFY(nvlist_add_nvlist_array(mirror, ZPOOL_CONFIG_CHILDREN,
1015 for (c = 0; c < m; c++)
1016 nvlist_free(child[c]);
1018 umem_free(child, m * sizeof (nvlist_t *));
1024 make_vdev_root(char *path, char *aux, char *pool, size_t size, uint64_t ashift,
1025 int log, int r, int m, int t)
1027 nvlist_t *root, **child;
1032 child = umem_alloc(t * sizeof (nvlist_t *), UMEM_NOFAIL);
1034 for (c = 0; c < t; c++) {
1035 child[c] = make_vdev_mirror(path, aux, pool, size, ashift,
1037 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_IS_LOG,
1041 VERIFY(nvlist_alloc(&root, NV_UNIQUE_NAME, 0) == 0);
1042 VERIFY(nvlist_add_string(root, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) == 0);
1043 VERIFY(nvlist_add_nvlist_array(root, aux ? aux : ZPOOL_CONFIG_CHILDREN,
1046 for (c = 0; c < t; c++)
1047 nvlist_free(child[c]);
1049 umem_free(child, t * sizeof (nvlist_t *));
1055 * Find a random spa version. Returns back a random spa version in the
1056 * range [initial_version, SPA_VERSION_FEATURES].
1059 ztest_random_spa_version(uint64_t initial_version)
1061 uint64_t version = initial_version;
1063 if (version <= SPA_VERSION_BEFORE_FEATURES) {
1065 ztest_random(SPA_VERSION_BEFORE_FEATURES - version + 1);
1068 if (version > SPA_VERSION_BEFORE_FEATURES)
1069 version = SPA_VERSION_FEATURES;
1071 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
1076 ztest_random_blocksize(void)
1079 * Choose a block size >= the ashift.
1080 * If the SPA supports new MAXBLOCKSIZE, test up to 1MB blocks.
1082 int maxbs = SPA_OLD_MAXBLOCKSHIFT;
1083 if (spa_maxblocksize(ztest_spa) == SPA_MAXBLOCKSIZE)
1085 uint64_t block_shift =
1086 ztest_random(maxbs - ztest_spa->spa_max_ashift + 1);
1087 return (1 << (SPA_MINBLOCKSHIFT + block_shift));
1091 ztest_random_dnodesize(void)
1094 int max_slots = spa_maxdnodesize(ztest_spa) >> DNODE_SHIFT;
1096 if (max_slots == DNODE_MIN_SLOTS)
1097 return (DNODE_MIN_SIZE);
1100 * Weight the random distribution more heavily toward smaller
1101 * dnode sizes since that is more likely to reflect real-world
1104 ASSERT3U(max_slots, >, 4);
1105 switch (ztest_random(10)) {
1107 slots = 5 + ztest_random(max_slots - 4);
1110 slots = 2 + ztest_random(3);
1117 return (slots << DNODE_SHIFT);
1121 ztest_random_ibshift(void)
1123 return (DN_MIN_INDBLKSHIFT +
1124 ztest_random(DN_MAX_INDBLKSHIFT - DN_MIN_INDBLKSHIFT + 1));
1128 ztest_random_vdev_top(spa_t *spa, boolean_t log_ok)
1131 vdev_t *rvd = spa->spa_root_vdev;
1134 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1137 top = ztest_random(rvd->vdev_children);
1138 tvd = rvd->vdev_child[top];
1139 } while (!vdev_is_concrete(tvd) || (tvd->vdev_islog && !log_ok) ||
1140 tvd->vdev_mg == NULL || tvd->vdev_mg->mg_class == NULL);
1146 ztest_random_dsl_prop(zfs_prop_t prop)
1151 value = zfs_prop_random_value(prop, ztest_random(-1ULL));
1152 } while (prop == ZFS_PROP_CHECKSUM && value == ZIO_CHECKSUM_OFF);
1158 ztest_dsl_prop_set_uint64(char *osname, zfs_prop_t prop, uint64_t value,
1161 const char *propname = zfs_prop_to_name(prop);
1162 const char *valname;
1167 error = dsl_prop_set_int(osname, propname,
1168 (inherit ? ZPROP_SRC_NONE : ZPROP_SRC_LOCAL), value);
1170 if (error == ENOSPC) {
1171 ztest_record_enospc(FTAG);
1176 setpoint = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
1177 VERIFY0(dsl_prop_get_integer(osname, propname, &curval, setpoint));
1179 if (ztest_opts.zo_verbose >= 6) {
1182 err = zfs_prop_index_to_string(prop, curval, &valname);
1184 (void) printf("%s %s = %llu at '%s'\n", osname,
1185 propname, (unsigned long long)curval, setpoint);
1187 (void) printf("%s %s = %s at '%s'\n",
1188 osname, propname, valname, setpoint);
1190 umem_free(setpoint, MAXPATHLEN);
1196 ztest_spa_prop_set_uint64(zpool_prop_t prop, uint64_t value)
1198 spa_t *spa = ztest_spa;
1199 nvlist_t *props = NULL;
1202 VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
1203 VERIFY(nvlist_add_uint64(props, zpool_prop_to_name(prop), value) == 0);
1205 error = spa_prop_set(spa, props);
1209 if (error == ENOSPC) {
1210 ztest_record_enospc(FTAG);
1219 ztest_dmu_objset_own(const char *name, dmu_objset_type_t type,
1220 boolean_t readonly, boolean_t decrypt, void *tag, objset_t **osp)
1224 err = dmu_objset_own(name, type, readonly, decrypt, tag, osp);
1225 if (decrypt && err == EACCES) {
1226 char ddname[ZFS_MAX_DATASET_NAME_LEN];
1227 dsl_crypto_params_t *dcp;
1228 nvlist_t *crypto_args = fnvlist_alloc();
1231 /* spa_keystore_load_wkey() expects a dsl dir name */
1232 strcpy(ddname, name);
1233 cp = strchr(ddname, '@');
1237 fnvlist_add_uint8_array(crypto_args, "wkeydata",
1238 (uint8_t *)ztest_wkeydata, WRAPPING_KEY_LEN);
1239 VERIFY0(dsl_crypto_params_create_nvlist(DCP_CMD_NONE, NULL,
1240 crypto_args, &dcp));
1241 err = spa_keystore_load_wkey(ddname, dcp, B_FALSE);
1242 dsl_crypto_params_free(dcp, B_FALSE);
1243 fnvlist_free(crypto_args);
1248 err = dmu_objset_own(name, type, readonly, decrypt, tag, osp);
1256 * Object and range lock mechanics
1259 list_node_t z_lnode;
1260 refcount_t z_refcnt;
1262 zfs_rlock_t z_range_lock;
1267 ztest_znode_t *z_ztznode;
1270 static ztest_znode_t *
1271 ztest_znode_init(uint64_t object)
1273 ztest_znode_t *zp = umem_alloc(sizeof (*zp), UMEM_NOFAIL);
1275 list_link_init(&zp->z_lnode);
1276 refcount_create(&zp->z_refcnt);
1277 zp->z_object = object;
1278 zfs_rlock_init(&zp->z_range_lock);
1284 ztest_znode_fini(ztest_znode_t *zp)
1286 ASSERT(refcount_is_zero(&zp->z_refcnt));
1287 zfs_rlock_destroy(&zp->z_range_lock);
1289 refcount_destroy(&zp->z_refcnt);
1290 list_link_init(&zp->z_lnode);
1291 umem_free(zp, sizeof (*zp));
1295 ztest_zll_init(zll_t *zll)
1297 mutex_init(&zll->z_lock, NULL, MUTEX_DEFAULT, NULL);
1298 list_create(&zll->z_list, sizeof (ztest_znode_t),
1299 offsetof(ztest_znode_t, z_lnode));
1303 ztest_zll_destroy(zll_t *zll)
1305 list_destroy(&zll->z_list);
1306 mutex_destroy(&zll->z_lock);
1309 #define RL_TAG "range_lock"
1310 static ztest_znode_t *
1311 ztest_znode_get(ztest_ds_t *zd, uint64_t object)
1313 zll_t *zll = &zd->zd_range_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1314 ztest_znode_t *zp = NULL;
1315 mutex_enter(&zll->z_lock);
1316 for (zp = list_head(&zll->z_list); (zp);
1317 zp = list_next(&zll->z_list, zp)) {
1318 if (zp->z_object == object) {
1319 refcount_add(&zp->z_refcnt, RL_TAG);
1324 zp = ztest_znode_init(object);
1325 refcount_add(&zp->z_refcnt, RL_TAG);
1326 list_insert_head(&zll->z_list, zp);
1328 mutex_exit(&zll->z_lock);
1333 ztest_znode_put(ztest_ds_t *zd, ztest_znode_t *zp)
1336 ASSERT3U(zp->z_object, !=, 0);
1337 zll = &zd->zd_range_lock[zp->z_object & (ZTEST_OBJECT_LOCKS - 1)];
1338 mutex_enter(&zll->z_lock);
1339 refcount_remove(&zp->z_refcnt, RL_TAG);
1340 if (refcount_is_zero(&zp->z_refcnt)) {
1341 list_remove(&zll->z_list, zp);
1342 ztest_znode_fini(zp);
1344 mutex_exit(&zll->z_lock);
1349 ztest_rll_init(rll_t *rll)
1351 rll->rll_writer = NULL;
1352 rll->rll_readers = 0;
1353 mutex_init(&rll->rll_lock, NULL, MUTEX_DEFAULT, NULL);
1354 cv_init(&rll->rll_cv, NULL, CV_DEFAULT, NULL);
1358 ztest_rll_destroy(rll_t *rll)
1360 ASSERT(rll->rll_writer == NULL);
1361 ASSERT(rll->rll_readers == 0);
1362 mutex_destroy(&rll->rll_lock);
1363 cv_destroy(&rll->rll_cv);
1367 ztest_rll_lock(rll_t *rll, rl_type_t type)
1369 mutex_enter(&rll->rll_lock);
1371 if (type == RL_READER) {
1372 while (rll->rll_writer != NULL)
1373 (void) cv_wait(&rll->rll_cv, &rll->rll_lock);
1376 while (rll->rll_writer != NULL || rll->rll_readers)
1377 (void) cv_wait(&rll->rll_cv, &rll->rll_lock);
1378 rll->rll_writer = curthread;
1381 mutex_exit(&rll->rll_lock);
1385 ztest_rll_unlock(rll_t *rll)
1387 mutex_enter(&rll->rll_lock);
1389 if (rll->rll_writer) {
1390 ASSERT(rll->rll_readers == 0);
1391 rll->rll_writer = NULL;
1393 ASSERT(rll->rll_readers != 0);
1394 ASSERT(rll->rll_writer == NULL);
1398 if (rll->rll_writer == NULL && rll->rll_readers == 0)
1399 cv_broadcast(&rll->rll_cv);
1401 mutex_exit(&rll->rll_lock);
1405 ztest_object_lock(ztest_ds_t *zd, uint64_t object, rl_type_t type)
1407 rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1409 ztest_rll_lock(rll, type);
1413 ztest_object_unlock(ztest_ds_t *zd, uint64_t object)
1415 rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1417 ztest_rll_unlock(rll);
1420 static ztest_zrl_t *
1421 ztest_zrl_init(rl_t *rl, ztest_znode_t *zp)
1423 ztest_zrl_t *zrl = umem_alloc(sizeof (*zrl), UMEM_NOFAIL);
1425 zrl->z_ztznode = zp;
1430 ztest_zrl_fini(ztest_zrl_t *zrl)
1432 umem_free(zrl, sizeof (*zrl));
1435 static ztest_zrl_t *
1436 ztest_range_lock(ztest_ds_t *zd, uint64_t object, uint64_t offset,
1437 uint64_t size, rl_type_t type)
1439 ztest_znode_t *zp = ztest_znode_get(zd, object);
1440 rl_t *rl = zfs_range_lock(&zp->z_range_lock, offset,
1442 return (ztest_zrl_init(rl, zp));
1446 ztest_range_unlock(ztest_ds_t *zd, ztest_zrl_t *zrl)
1448 zfs_range_unlock(zrl->z_rl);
1449 ztest_znode_put(zd, zrl->z_ztznode);
1450 ztest_zrl_fini(zrl);
1454 ztest_zd_init(ztest_ds_t *zd, ztest_shared_ds_t *szd, objset_t *os)
1457 zd->zd_zilog = dmu_objset_zil(os);
1458 zd->zd_shared = szd;
1459 dmu_objset_name(os, zd->zd_name);
1462 if (zd->zd_shared != NULL)
1463 zd->zd_shared->zd_seq = 0;
1465 VERIFY0(pthread_rwlock_init(&zd->zd_zilog_lock, NULL));
1466 mutex_init(&zd->zd_dirobj_lock, NULL, MUTEX_DEFAULT, NULL);
1468 for (l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1469 ztest_rll_init(&zd->zd_object_lock[l]);
1471 for (l = 0; l < ZTEST_RANGE_LOCKS; l++)
1472 ztest_zll_init(&zd->zd_range_lock[l]);
1476 ztest_zd_fini(ztest_ds_t *zd)
1480 mutex_destroy(&zd->zd_dirobj_lock);
1481 (void) pthread_rwlock_destroy(&zd->zd_zilog_lock);
1483 for (l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1484 ztest_rll_destroy(&zd->zd_object_lock[l]);
1486 for (l = 0; l < ZTEST_RANGE_LOCKS; l++)
1487 ztest_zll_destroy(&zd->zd_range_lock[l]);
1490 #define TXG_MIGHTWAIT (ztest_random(10) == 0 ? TXG_NOWAIT : TXG_WAIT)
1493 ztest_tx_assign(dmu_tx_t *tx, uint64_t txg_how, const char *tag)
1499 * Attempt to assign tx to some transaction group.
1501 error = dmu_tx_assign(tx, txg_how);
1503 if (error == ERESTART) {
1504 ASSERT(txg_how == TXG_NOWAIT);
1507 ASSERT3U(error, ==, ENOSPC);
1508 ztest_record_enospc(tag);
1513 txg = dmu_tx_get_txg(tx);
1519 ztest_pattern_set(void *buf, uint64_t size, uint64_t value)
1522 uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1530 ztest_pattern_match(void *buf, uint64_t size, uint64_t value)
1533 uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1537 diff |= (value - *ip++);
1544 ztest_bt_generate(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1545 uint64_t dnodesize, uint64_t offset, uint64_t gen, uint64_t txg,
1548 bt->bt_magic = BT_MAGIC;
1549 bt->bt_objset = dmu_objset_id(os);
1550 bt->bt_object = object;
1551 bt->bt_dnodesize = dnodesize;
1552 bt->bt_offset = offset;
1555 bt->bt_crtxg = crtxg;
1559 ztest_bt_verify(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1560 uint64_t dnodesize, uint64_t offset, uint64_t gen, uint64_t txg,
1563 ASSERT3U(bt->bt_magic, ==, BT_MAGIC);
1564 ASSERT3U(bt->bt_objset, ==, dmu_objset_id(os));
1565 ASSERT3U(bt->bt_object, ==, object);
1566 ASSERT3U(bt->bt_dnodesize, ==, dnodesize);
1567 ASSERT3U(bt->bt_offset, ==, offset);
1568 ASSERT3U(bt->bt_gen, <=, gen);
1569 ASSERT3U(bt->bt_txg, <=, txg);
1570 ASSERT3U(bt->bt_crtxg, ==, crtxg);
1573 static ztest_block_tag_t *
1574 ztest_bt_bonus(dmu_buf_t *db)
1576 dmu_object_info_t doi;
1577 ztest_block_tag_t *bt;
1579 dmu_object_info_from_db(db, &doi);
1580 ASSERT3U(doi.doi_bonus_size, <=, db->db_size);
1581 ASSERT3U(doi.doi_bonus_size, >=, sizeof (*bt));
1582 bt = (void *)((char *)db->db_data + doi.doi_bonus_size - sizeof (*bt));
1588 * Generate a token to fill up unused bonus buffer space. Try to make
1589 * it unique to the object, generation, and offset to verify that data
1590 * is not getting overwritten by data from other dnodes.
1592 #define ZTEST_BONUS_FILL_TOKEN(obj, ds, gen, offset) \
1593 (((ds) << 48) | ((gen) << 32) | ((obj) << 8) | (offset))
1596 * Fill up the unused bonus buffer region before the block tag with a
1597 * verifiable pattern. Filling the whole bonus area with non-zero data
1598 * helps ensure that all dnode traversal code properly skips the
1599 * interior regions of large dnodes.
1602 ztest_fill_unused_bonus(dmu_buf_t *db, void *end, uint64_t obj,
1603 objset_t *os, uint64_t gen)
1607 ASSERT(IS_P2ALIGNED((char *)end - (char *)db->db_data, 8));
1609 for (bonusp = db->db_data; bonusp < (uint64_t *)end; bonusp++) {
1610 uint64_t token = ZTEST_BONUS_FILL_TOKEN(obj, dmu_objset_id(os),
1611 gen, bonusp - (uint64_t *)db->db_data);
1617 * Verify that the unused area of a bonus buffer is filled with the
1621 ztest_verify_unused_bonus(dmu_buf_t *db, void *end, uint64_t obj,
1622 objset_t *os, uint64_t gen)
1626 for (bonusp = db->db_data; bonusp < (uint64_t *)end; bonusp++) {
1627 uint64_t token = ZTEST_BONUS_FILL_TOKEN(obj, dmu_objset_id(os),
1628 gen, bonusp - (uint64_t *)db->db_data);
1629 VERIFY3U(*bonusp, ==, token);
1637 #define lrz_type lr_mode
1638 #define lrz_blocksize lr_uid
1639 #define lrz_ibshift lr_gid
1640 #define lrz_bonustype lr_rdev
1641 #define lrz_dnodesize lr_crtime[1]
1644 ztest_log_create(ztest_ds_t *zd, dmu_tx_t *tx, lr_create_t *lr)
1646 char *name = (void *)(lr + 1); /* name follows lr */
1647 size_t namesize = strlen(name) + 1;
1650 if (zil_replaying(zd->zd_zilog, tx))
1653 itx = zil_itx_create(TX_CREATE, sizeof (*lr) + namesize);
1654 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1655 sizeof (*lr) + namesize - sizeof (lr_t));
1657 zil_itx_assign(zd->zd_zilog, itx, tx);
1661 ztest_log_remove(ztest_ds_t *zd, dmu_tx_t *tx, lr_remove_t *lr, uint64_t object)
1663 char *name = (void *)(lr + 1); /* name follows lr */
1664 size_t namesize = strlen(name) + 1;
1667 if (zil_replaying(zd->zd_zilog, tx))
1670 itx = zil_itx_create(TX_REMOVE, sizeof (*lr) + namesize);
1671 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1672 sizeof (*lr) + namesize - sizeof (lr_t));
1674 itx->itx_oid = object;
1675 zil_itx_assign(zd->zd_zilog, itx, tx);
1679 ztest_log_write(ztest_ds_t *zd, dmu_tx_t *tx, lr_write_t *lr)
1682 itx_wr_state_t write_state = ztest_random(WR_NUM_STATES);
1684 if (zil_replaying(zd->zd_zilog, tx))
1687 if (lr->lr_length > ZIL_MAX_LOG_DATA)
1688 write_state = WR_INDIRECT;
1690 itx = zil_itx_create(TX_WRITE,
1691 sizeof (*lr) + (write_state == WR_COPIED ? lr->lr_length : 0));
1693 if (write_state == WR_COPIED &&
1694 dmu_read(zd->zd_os, lr->lr_foid, lr->lr_offset, lr->lr_length,
1695 ((lr_write_t *)&itx->itx_lr) + 1, DMU_READ_NO_PREFETCH) != 0) {
1696 zil_itx_destroy(itx);
1697 itx = zil_itx_create(TX_WRITE, sizeof (*lr));
1698 write_state = WR_NEED_COPY;
1700 itx->itx_private = zd;
1701 itx->itx_wr_state = write_state;
1702 itx->itx_sync = (ztest_random(8) == 0);
1704 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1705 sizeof (*lr) - sizeof (lr_t));
1707 zil_itx_assign(zd->zd_zilog, itx, tx);
1711 ztest_log_truncate(ztest_ds_t *zd, dmu_tx_t *tx, lr_truncate_t *lr)
1715 if (zil_replaying(zd->zd_zilog, tx))
1718 itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr));
1719 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1720 sizeof (*lr) - sizeof (lr_t));
1722 itx->itx_sync = B_FALSE;
1723 zil_itx_assign(zd->zd_zilog, itx, tx);
1727 ztest_log_setattr(ztest_ds_t *zd, dmu_tx_t *tx, lr_setattr_t *lr)
1731 if (zil_replaying(zd->zd_zilog, tx))
1734 itx = zil_itx_create(TX_SETATTR, sizeof (*lr));
1735 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1736 sizeof (*lr) - sizeof (lr_t));
1738 itx->itx_sync = B_FALSE;
1739 zil_itx_assign(zd->zd_zilog, itx, tx);
1746 ztest_replay_create(void *arg1, void *arg2, boolean_t byteswap)
1748 ztest_ds_t *zd = arg1;
1749 lr_create_t *lr = arg2;
1750 char *name = (void *)(lr + 1); /* name follows lr */
1751 objset_t *os = zd->zd_os;
1752 ztest_block_tag_t *bbt;
1760 byteswap_uint64_array(lr, sizeof (*lr));
1762 ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1763 ASSERT(name[0] != '\0');
1765 tx = dmu_tx_create(os);
1767 dmu_tx_hold_zap(tx, lr->lr_doid, B_TRUE, name);
1769 if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1770 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, B_TRUE, NULL);
1772 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1775 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1779 ASSERT(dmu_objset_zil(os)->zl_replay == !!lr->lr_foid);
1780 bonuslen = DN_BONUS_SIZE(lr->lrz_dnodesize);
1782 if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1783 if (lr->lr_foid == 0) {
1784 lr->lr_foid = zap_create_dnsize(os,
1785 lr->lrz_type, lr->lrz_bonustype,
1786 bonuslen, lr->lrz_dnodesize, tx);
1788 error = zap_create_claim_dnsize(os, lr->lr_foid,
1789 lr->lrz_type, lr->lrz_bonustype,
1790 bonuslen, lr->lrz_dnodesize, tx);
1793 if (lr->lr_foid == 0) {
1794 lr->lr_foid = dmu_object_alloc_dnsize(os,
1795 lr->lrz_type, 0, lr->lrz_bonustype,
1796 bonuslen, lr->lrz_dnodesize, tx);
1798 error = dmu_object_claim_dnsize(os, lr->lr_foid,
1799 lr->lrz_type, 0, lr->lrz_bonustype,
1800 bonuslen, lr->lrz_dnodesize, tx);
1805 ASSERT3U(error, ==, EEXIST);
1806 ASSERT(zd->zd_zilog->zl_replay);
1811 ASSERT(lr->lr_foid != 0);
1813 if (lr->lrz_type != DMU_OT_ZAP_OTHER)
1814 VERIFY3U(0, ==, dmu_object_set_blocksize(os, lr->lr_foid,
1815 lr->lrz_blocksize, lr->lrz_ibshift, tx));
1817 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1818 bbt = ztest_bt_bonus(db);
1819 dmu_buf_will_dirty(db, tx);
1820 ztest_bt_generate(bbt, os, lr->lr_foid, lr->lrz_dnodesize, -1ULL,
1821 lr->lr_gen, txg, txg);
1822 ztest_fill_unused_bonus(db, bbt, lr->lr_foid, os, lr->lr_gen);
1823 dmu_buf_rele(db, FTAG);
1825 VERIFY3U(0, ==, zap_add(os, lr->lr_doid, name, sizeof (uint64_t), 1,
1828 (void) ztest_log_create(zd, tx, lr);
1836 ztest_replay_remove(void *arg1, void *arg2, boolean_t byteswap)
1838 ztest_ds_t *zd = arg1;
1839 lr_remove_t *lr = arg2;
1840 char *name = (void *)(lr + 1); /* name follows lr */
1841 objset_t *os = zd->zd_os;
1842 dmu_object_info_t doi;
1844 uint64_t object, txg;
1847 byteswap_uint64_array(lr, sizeof (*lr));
1849 ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1850 ASSERT(name[0] != '\0');
1853 zap_lookup(os, lr->lr_doid, name, sizeof (object), 1, &object));
1854 ASSERT(object != 0);
1856 ztest_object_lock(zd, object, RL_WRITER);
1858 VERIFY3U(0, ==, dmu_object_info(os, object, &doi));
1860 tx = dmu_tx_create(os);
1862 dmu_tx_hold_zap(tx, lr->lr_doid, B_FALSE, name);
1863 dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END);
1865 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1867 ztest_object_unlock(zd, object);
1871 if (doi.doi_type == DMU_OT_ZAP_OTHER) {
1872 VERIFY3U(0, ==, zap_destroy(os, object, tx));
1874 VERIFY3U(0, ==, dmu_object_free(os, object, tx));
1877 VERIFY3U(0, ==, zap_remove(os, lr->lr_doid, name, tx));
1879 (void) ztest_log_remove(zd, tx, lr, object);
1883 ztest_object_unlock(zd, object);
1889 ztest_replay_write(void *arg1, void *arg2, boolean_t byteswap)
1891 ztest_ds_t *zd = arg1;
1892 lr_write_t *lr = arg2;
1893 objset_t *os = zd->zd_os;
1894 void *data = lr + 1; /* data follows lr */
1895 uint64_t offset, length;
1896 ztest_block_tag_t *bt = data;
1897 ztest_block_tag_t *bbt;
1898 uint64_t gen, txg, lrtxg, crtxg;
1899 dmu_object_info_t doi;
1902 arc_buf_t *abuf = NULL;
1906 byteswap_uint64_array(lr, sizeof (*lr));
1908 offset = lr->lr_offset;
1909 length = lr->lr_length;
1911 /* If it's a dmu_sync() block, write the whole block */
1912 if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
1913 uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
1914 if (length < blocksize) {
1915 offset -= offset % blocksize;
1920 if (bt->bt_magic == BSWAP_64(BT_MAGIC))
1921 byteswap_uint64_array(bt, sizeof (*bt));
1923 if (bt->bt_magic != BT_MAGIC)
1926 ztest_object_lock(zd, lr->lr_foid, RL_READER);
1927 rl = ztest_range_lock(zd, lr->lr_foid, offset, length, RL_WRITER);
1929 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1931 dmu_object_info_from_db(db, &doi);
1933 bbt = ztest_bt_bonus(db);
1934 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1936 crtxg = bbt->bt_crtxg;
1937 lrtxg = lr->lr_common.lrc_txg;
1939 tx = dmu_tx_create(os);
1941 dmu_tx_hold_write(tx, lr->lr_foid, offset, length);
1943 if (ztest_random(8) == 0 && length == doi.doi_data_block_size &&
1944 P2PHASE(offset, length) == 0)
1945 abuf = dmu_request_arcbuf(db, length);
1947 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1950 dmu_return_arcbuf(abuf);
1951 dmu_buf_rele(db, FTAG);
1952 ztest_range_unlock(zd, rl);
1953 ztest_object_unlock(zd, lr->lr_foid);
1959 * Usually, verify the old data before writing new data --
1960 * but not always, because we also want to verify correct
1961 * behavior when the data was not recently read into cache.
1963 ASSERT(offset % doi.doi_data_block_size == 0);
1964 if (ztest_random(4) != 0) {
1965 int prefetch = ztest_random(2) ?
1966 DMU_READ_PREFETCH : DMU_READ_NO_PREFETCH;
1967 ztest_block_tag_t rbt;
1969 VERIFY(dmu_read(os, lr->lr_foid, offset,
1970 sizeof (rbt), &rbt, prefetch) == 0);
1971 if (rbt.bt_magic == BT_MAGIC) {
1972 ztest_bt_verify(&rbt, os, lr->lr_foid, 0,
1973 offset, gen, txg, crtxg);
1978 * Writes can appear to be newer than the bonus buffer because
1979 * the ztest_get_data() callback does a dmu_read() of the
1980 * open-context data, which may be different than the data
1981 * as it was when the write was generated.
1983 if (zd->zd_zilog->zl_replay) {
1984 ztest_bt_verify(bt, os, lr->lr_foid, 0, offset,
1985 MAX(gen, bt->bt_gen), MAX(txg, lrtxg),
1990 * Set the bt's gen/txg to the bonus buffer's gen/txg
1991 * so that all of the usual ASSERTs will work.
1993 ztest_bt_generate(bt, os, lr->lr_foid, 0, offset, gen, txg,
1998 dmu_write(os, lr->lr_foid, offset, length, data, tx);
2000 bcopy(data, abuf->b_data, length);
2001 dmu_assign_arcbuf_by_dbuf(db, offset, abuf, tx);
2004 (void) ztest_log_write(zd, tx, lr);
2006 dmu_buf_rele(db, FTAG);
2010 ztest_range_unlock(zd, rl);
2011 ztest_object_unlock(zd, lr->lr_foid);
2017 ztest_replay_truncate(void *arg1, void *arg2, boolean_t byteswap)
2019 ztest_ds_t *zd = arg1;
2020 lr_truncate_t *lr = arg2;
2021 objset_t *os = zd->zd_os;
2027 byteswap_uint64_array(lr, sizeof (*lr));
2029 ztest_object_lock(zd, lr->lr_foid, RL_READER);
2030 rl = ztest_range_lock(zd, lr->lr_foid, lr->lr_offset, lr->lr_length,
2033 tx = dmu_tx_create(os);
2035 dmu_tx_hold_free(tx, lr->lr_foid, lr->lr_offset, lr->lr_length);
2037 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
2039 ztest_range_unlock(zd, rl);
2040 ztest_object_unlock(zd, lr->lr_foid);
2044 VERIFY(dmu_free_range(os, lr->lr_foid, lr->lr_offset,
2045 lr->lr_length, tx) == 0);
2047 (void) ztest_log_truncate(zd, tx, lr);
2051 ztest_range_unlock(zd, rl);
2052 ztest_object_unlock(zd, lr->lr_foid);
2058 ztest_replay_setattr(void *arg1, void *arg2, boolean_t byteswap)
2060 ztest_ds_t *zd = arg1;
2061 lr_setattr_t *lr = arg2;
2062 objset_t *os = zd->zd_os;
2065 ztest_block_tag_t *bbt;
2066 uint64_t txg, lrtxg, crtxg, dnodesize;
2069 byteswap_uint64_array(lr, sizeof (*lr));
2071 ztest_object_lock(zd, lr->lr_foid, RL_WRITER);
2073 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
2075 tx = dmu_tx_create(os);
2076 dmu_tx_hold_bonus(tx, lr->lr_foid);
2078 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
2080 dmu_buf_rele(db, FTAG);
2081 ztest_object_unlock(zd, lr->lr_foid);
2085 bbt = ztest_bt_bonus(db);
2086 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
2087 crtxg = bbt->bt_crtxg;
2088 lrtxg = lr->lr_common.lrc_txg;
2089 dnodesize = bbt->bt_dnodesize;
2091 if (zd->zd_zilog->zl_replay) {
2092 ASSERT(lr->lr_size != 0);
2093 ASSERT(lr->lr_mode != 0);
2097 * Randomly change the size and increment the generation.
2099 lr->lr_size = (ztest_random(db->db_size / sizeof (*bbt)) + 1) *
2101 lr->lr_mode = bbt->bt_gen + 1;
2106 * Verify that the current bonus buffer is not newer than our txg.
2108 ztest_bt_verify(bbt, os, lr->lr_foid, dnodesize, -1ULL, lr->lr_mode,
2109 MAX(txg, lrtxg), crtxg);
2111 dmu_buf_will_dirty(db, tx);
2113 ASSERT3U(lr->lr_size, >=, sizeof (*bbt));
2114 ASSERT3U(lr->lr_size, <=, db->db_size);
2115 VERIFY0(dmu_set_bonus(db, lr->lr_size, tx));
2116 bbt = ztest_bt_bonus(db);
2118 ztest_bt_generate(bbt, os, lr->lr_foid, dnodesize, -1ULL, lr->lr_mode,
2120 ztest_fill_unused_bonus(db, bbt, lr->lr_foid, os, bbt->bt_gen);
2121 dmu_buf_rele(db, FTAG);
2123 (void) ztest_log_setattr(zd, tx, lr);
2127 ztest_object_unlock(zd, lr->lr_foid);
2132 zil_replay_func_t *ztest_replay_vector[TX_MAX_TYPE] = {
2133 NULL, /* 0 no such transaction type */
2134 ztest_replay_create, /* TX_CREATE */
2135 NULL, /* TX_MKDIR */
2136 NULL, /* TX_MKXATTR */
2137 NULL, /* TX_SYMLINK */
2138 ztest_replay_remove, /* TX_REMOVE */
2139 NULL, /* TX_RMDIR */
2141 NULL, /* TX_RENAME */
2142 ztest_replay_write, /* TX_WRITE */
2143 ztest_replay_truncate, /* TX_TRUNCATE */
2144 ztest_replay_setattr, /* TX_SETATTR */
2146 NULL, /* TX_CREATE_ACL */
2147 NULL, /* TX_CREATE_ATTR */
2148 NULL, /* TX_CREATE_ACL_ATTR */
2149 NULL, /* TX_MKDIR_ACL */
2150 NULL, /* TX_MKDIR_ATTR */
2151 NULL, /* TX_MKDIR_ACL_ATTR */
2152 NULL, /* TX_WRITE2 */
2156 * ZIL get_data callbacks
2158 typedef struct ztest_zgd_private {
2162 } ztest_zgd_private_t;
2165 ztest_get_done(zgd_t *zgd, int error)
2167 ztest_zgd_private_t *zzp = zgd->zgd_private;
2168 ztest_ds_t *zd = zzp->z_zd;
2169 uint64_t object = zzp->z_object;
2172 dmu_buf_rele(zgd->zgd_db, zgd);
2174 ztest_range_unlock(zd, zzp->z_rl);
2175 ztest_object_unlock(zd, object);
2177 if (error == 0 && zgd->zgd_bp)
2178 zil_lwb_add_block(zgd->zgd_lwb, zgd->zgd_bp);
2180 umem_free(zgd, sizeof (*zgd));
2181 umem_free(zzp, sizeof (*zzp));
2185 ztest_get_data(void *arg, lr_write_t *lr, char *buf, struct lwb *lwb,
2188 ztest_ds_t *zd = arg;
2189 objset_t *os = zd->zd_os;
2190 uint64_t object = lr->lr_foid;
2191 uint64_t offset = lr->lr_offset;
2192 uint64_t size = lr->lr_length;
2193 uint64_t txg = lr->lr_common.lrc_txg;
2195 dmu_object_info_t doi;
2199 ztest_zgd_private_t *zgd_private;
2201 ASSERT3P(lwb, !=, NULL);
2202 ASSERT3P(zio, !=, NULL);
2203 ASSERT3U(size, !=, 0);
2205 ztest_object_lock(zd, object, RL_READER);
2206 error = dmu_bonus_hold(os, object, FTAG, &db);
2208 ztest_object_unlock(zd, object);
2212 crtxg = ztest_bt_bonus(db)->bt_crtxg;
2214 if (crtxg == 0 || crtxg > txg) {
2215 dmu_buf_rele(db, FTAG);
2216 ztest_object_unlock(zd, object);
2220 dmu_object_info_from_db(db, &doi);
2221 dmu_buf_rele(db, FTAG);
2224 zgd = umem_zalloc(sizeof (*zgd), UMEM_NOFAIL);
2226 zgd_private = umem_zalloc(sizeof (ztest_zgd_private_t), UMEM_NOFAIL);
2227 zgd_private->z_zd = zd;
2228 zgd_private->z_object = object;
2229 zgd->zgd_private = zgd_private;
2231 if (buf != NULL) { /* immediate write */
2232 zgd_private->z_rl = ztest_range_lock(zd, object, offset, size,
2234 zgd->zgd_rl = zgd_private->z_rl->z_rl;
2236 error = dmu_read(os, object, offset, size, buf,
2237 DMU_READ_NO_PREFETCH);
2240 size = doi.doi_data_block_size;
2242 offset = P2ALIGN(offset, size);
2244 ASSERT(offset < size);
2248 zgd_private->z_rl = ztest_range_lock(zd, object, offset, size,
2250 zgd->zgd_rl = zgd_private->z_rl->z_rl;
2252 error = dmu_buf_hold(os, object, offset, zgd, &db,
2253 DMU_READ_NO_PREFETCH);
2256 blkptr_t *bp = &lr->lr_blkptr;
2261 ASSERT(db->db_offset == offset);
2262 ASSERT(db->db_size == size);
2264 error = dmu_sync(zio, lr->lr_common.lrc_txg,
2265 ztest_get_done, zgd);
2272 ztest_get_done(zgd, error);
2278 ztest_lr_alloc(size_t lrsize, char *name)
2281 size_t namesize = name ? strlen(name) + 1 : 0;
2283 lr = umem_zalloc(lrsize + namesize, UMEM_NOFAIL);
2286 bcopy(name, lr + lrsize, namesize);
2292 ztest_lr_free(void *lr, size_t lrsize, char *name)
2294 size_t namesize = name ? strlen(name) + 1 : 0;
2296 umem_free(lr, lrsize + namesize);
2300 * Lookup a bunch of objects. Returns the number of objects not found.
2303 ztest_lookup(ztest_ds_t *zd, ztest_od_t *od, int count)
2309 ASSERT(MUTEX_HELD(&zd->zd_dirobj_lock));
2311 for (i = 0; i < count; i++, od++) {
2313 error = zap_lookup(zd->zd_os, od->od_dir, od->od_name,
2314 sizeof (uint64_t), 1, &od->od_object);
2316 ASSERT(error == ENOENT);
2317 ASSERT(od->od_object == 0);
2321 ztest_block_tag_t *bbt;
2322 dmu_object_info_t doi;
2324 ASSERT(od->od_object != 0);
2325 ASSERT(missing == 0); /* there should be no gaps */
2327 ztest_object_lock(zd, od->od_object, RL_READER);
2328 VERIFY3U(0, ==, dmu_bonus_hold(zd->zd_os,
2329 od->od_object, FTAG, &db));
2330 dmu_object_info_from_db(db, &doi);
2331 bbt = ztest_bt_bonus(db);
2332 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
2333 od->od_type = doi.doi_type;
2334 od->od_blocksize = doi.doi_data_block_size;
2335 od->od_gen = bbt->bt_gen;
2336 dmu_buf_rele(db, FTAG);
2337 ztest_object_unlock(zd, od->od_object);
2345 ztest_create(ztest_ds_t *zd, ztest_od_t *od, int count)
2350 ASSERT(MUTEX_HELD(&zd->zd_dirobj_lock));
2352 for (i = 0; i < count; i++, od++) {
2359 lr_create_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
2361 lr->lr_doid = od->od_dir;
2362 lr->lr_foid = 0; /* 0 to allocate, > 0 to claim */
2363 lr->lrz_type = od->od_crtype;
2364 lr->lrz_blocksize = od->od_crblocksize;
2365 lr->lrz_ibshift = ztest_random_ibshift();
2366 lr->lrz_bonustype = DMU_OT_UINT64_OTHER;
2367 lr->lrz_dnodesize = od->od_crdnodesize;
2368 lr->lr_gen = od->od_crgen;
2369 lr->lr_crtime[0] = time(NULL);
2371 if (ztest_replay_create(zd, lr, B_FALSE) != 0) {
2372 ASSERT(missing == 0);
2376 od->od_object = lr->lr_foid;
2377 od->od_type = od->od_crtype;
2378 od->od_blocksize = od->od_crblocksize;
2379 od->od_gen = od->od_crgen;
2380 ASSERT(od->od_object != 0);
2383 ztest_lr_free(lr, sizeof (*lr), od->od_name);
2390 ztest_remove(ztest_ds_t *zd, ztest_od_t *od, int count)
2396 ASSERT(MUTEX_HELD(&zd->zd_dirobj_lock));
2400 for (i = count - 1; i >= 0; i--, od--) {
2407 * No object was found.
2409 if (od->od_object == 0)
2412 lr_remove_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
2414 lr->lr_doid = od->od_dir;
2416 if ((error = ztest_replay_remove(zd, lr, B_FALSE)) != 0) {
2417 ASSERT3U(error, ==, ENOSPC);
2422 ztest_lr_free(lr, sizeof (*lr), od->od_name);
2429 ztest_write(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size,
2435 lr = ztest_lr_alloc(sizeof (*lr) + size, NULL);
2437 lr->lr_foid = object;
2438 lr->lr_offset = offset;
2439 lr->lr_length = size;
2441 BP_ZERO(&lr->lr_blkptr);
2443 bcopy(data, lr + 1, size);
2445 error = ztest_replay_write(zd, lr, B_FALSE);
2447 ztest_lr_free(lr, sizeof (*lr) + size, NULL);
2453 ztest_truncate(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2458 lr = ztest_lr_alloc(sizeof (*lr), NULL);
2460 lr->lr_foid = object;
2461 lr->lr_offset = offset;
2462 lr->lr_length = size;
2464 error = ztest_replay_truncate(zd, lr, B_FALSE);
2466 ztest_lr_free(lr, sizeof (*lr), NULL);
2472 ztest_setattr(ztest_ds_t *zd, uint64_t object)
2477 lr = ztest_lr_alloc(sizeof (*lr), NULL);
2479 lr->lr_foid = object;
2483 error = ztest_replay_setattr(zd, lr, B_FALSE);
2485 ztest_lr_free(lr, sizeof (*lr), NULL);
2491 ztest_prealloc(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2493 objset_t *os = zd->zd_os;
2498 txg_wait_synced(dmu_objset_pool(os), 0);
2500 ztest_object_lock(zd, object, RL_READER);
2501 rl = ztest_range_lock(zd, object, offset, size, RL_WRITER);
2503 tx = dmu_tx_create(os);
2505 dmu_tx_hold_write(tx, object, offset, size);
2507 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
2510 dmu_prealloc(os, object, offset, size, tx);
2512 txg_wait_synced(dmu_objset_pool(os), txg);
2514 (void) dmu_free_long_range(os, object, offset, size);
2517 ztest_range_unlock(zd, rl);
2518 ztest_object_unlock(zd, object);
2522 ztest_io(ztest_ds_t *zd, uint64_t object, uint64_t offset)
2525 ztest_block_tag_t wbt;
2526 dmu_object_info_t doi;
2527 enum ztest_io_type io_type;
2531 VERIFY(dmu_object_info(zd->zd_os, object, &doi) == 0);
2532 blocksize = doi.doi_data_block_size;
2533 data = umem_alloc(blocksize, UMEM_NOFAIL);
2536 * Pick an i/o type at random, biased toward writing block tags.
2538 io_type = ztest_random(ZTEST_IO_TYPES);
2539 if (ztest_random(2) == 0)
2540 io_type = ZTEST_IO_WRITE_TAG;
2542 (void) pthread_rwlock_rdlock(&zd->zd_zilog_lock);
2546 case ZTEST_IO_WRITE_TAG:
2547 ztest_bt_generate(&wbt, zd->zd_os, object, doi.doi_dnodesize,
2549 (void) ztest_write(zd, object, offset, sizeof (wbt), &wbt);
2552 case ZTEST_IO_WRITE_PATTERN:
2553 (void) memset(data, 'a' + (object + offset) % 5, blocksize);
2554 if (ztest_random(2) == 0) {
2556 * Induce fletcher2 collisions to ensure that
2557 * zio_ddt_collision() detects and resolves them
2558 * when using fletcher2-verify for deduplication.
2560 ((uint64_t *)data)[0] ^= 1ULL << 63;
2561 ((uint64_t *)data)[4] ^= 1ULL << 63;
2563 (void) ztest_write(zd, object, offset, blocksize, data);
2566 case ZTEST_IO_WRITE_ZEROES:
2567 bzero(data, blocksize);
2568 (void) ztest_write(zd, object, offset, blocksize, data);
2571 case ZTEST_IO_TRUNCATE:
2572 (void) ztest_truncate(zd, object, offset, blocksize);
2575 case ZTEST_IO_SETATTR:
2576 (void) ztest_setattr(zd, object);
2581 case ZTEST_IO_REWRITE:
2582 (void) pthread_rwlock_rdlock(&ztest_name_lock);
2583 err = ztest_dsl_prop_set_uint64(zd->zd_name,
2584 ZFS_PROP_CHECKSUM, spa_dedup_checksum(ztest_spa),
2586 VERIFY(err == 0 || err == ENOSPC);
2587 err = ztest_dsl_prop_set_uint64(zd->zd_name,
2588 ZFS_PROP_COMPRESSION,
2589 ztest_random_dsl_prop(ZFS_PROP_COMPRESSION),
2591 VERIFY(err == 0 || err == ENOSPC);
2592 (void) pthread_rwlock_unlock(&ztest_name_lock);
2594 VERIFY0(dmu_read(zd->zd_os, object, offset, blocksize, data,
2595 DMU_READ_NO_PREFETCH));
2597 (void) ztest_write(zd, object, offset, blocksize, data);
2601 (void) pthread_rwlock_unlock(&zd->zd_zilog_lock);
2603 umem_free(data, blocksize);
2607 * Initialize an object description template.
2610 ztest_od_init(ztest_od_t *od, uint64_t id, char *tag, uint64_t index,
2611 dmu_object_type_t type, uint64_t blocksize, uint64_t dnodesize,
2614 od->od_dir = ZTEST_DIROBJ;
2617 od->od_crtype = type;
2618 od->od_crblocksize = blocksize ? blocksize : ztest_random_blocksize();
2619 od->od_crdnodesize = dnodesize ? dnodesize : ztest_random_dnodesize();
2622 od->od_type = DMU_OT_NONE;
2623 od->od_blocksize = 0;
2626 (void) snprintf(od->od_name, sizeof (od->od_name), "%s(%lld)[%llu]",
2627 tag, (longlong_t)id, (u_longlong_t)index);
2631 * Lookup or create the objects for a test using the od template.
2632 * If the objects do not all exist, or if 'remove' is specified,
2633 * remove any existing objects and create new ones. Otherwise,
2634 * use the existing objects.
2637 ztest_object_init(ztest_ds_t *zd, ztest_od_t *od, size_t size, boolean_t remove)
2639 int count = size / sizeof (*od);
2642 mutex_enter(&zd->zd_dirobj_lock);
2643 if ((ztest_lookup(zd, od, count) != 0 || remove) &&
2644 (ztest_remove(zd, od, count) != 0 ||
2645 ztest_create(zd, od, count) != 0))
2648 mutex_exit(&zd->zd_dirobj_lock);
2655 ztest_zil_commit(ztest_ds_t *zd, uint64_t id)
2657 zilog_t *zilog = zd->zd_zilog;
2659 (void) pthread_rwlock_rdlock(&zd->zd_zilog_lock);
2661 zil_commit(zilog, ztest_random(ZTEST_OBJECTS));
2664 * Remember the committed values in zd, which is in parent/child
2665 * shared memory. If we die, the next iteration of ztest_run()
2666 * will verify that the log really does contain this record.
2668 mutex_enter(&zilog->zl_lock);
2669 ASSERT(zd->zd_shared != NULL);
2670 ASSERT3U(zd->zd_shared->zd_seq, <=, zilog->zl_commit_lr_seq);
2671 zd->zd_shared->zd_seq = zilog->zl_commit_lr_seq;
2672 mutex_exit(&zilog->zl_lock);
2674 (void) pthread_rwlock_unlock(&zd->zd_zilog_lock);
2678 * This function is designed to simulate the operations that occur during a
2679 * mount/unmount operation. We hold the dataset across these operations in an
2680 * attempt to expose any implicit assumptions about ZIL management.
2684 ztest_zil_remount(ztest_ds_t *zd, uint64_t id)
2686 objset_t *os = zd->zd_os;
2689 * We grab the zd_dirobj_lock to ensure that no other thread is
2690 * updating the zil (i.e. adding in-memory log records) and the
2691 * zd_zilog_lock to block any I/O.
2693 mutex_enter(&zd->zd_dirobj_lock);
2694 (void) pthread_rwlock_wrlock(&zd->zd_zilog_lock);
2696 /* zfsvfs_teardown() */
2697 zil_close(zd->zd_zilog);
2699 /* zfsvfs_setup() */
2700 VERIFY(zil_open(os, ztest_get_data) == zd->zd_zilog);
2701 zil_replay(os, zd, ztest_replay_vector);
2703 (void) pthread_rwlock_unlock(&zd->zd_zilog_lock);
2704 mutex_exit(&zd->zd_dirobj_lock);
2708 * Verify that we can't destroy an active pool, create an existing pool,
2709 * or create a pool with a bad vdev spec.
2713 ztest_spa_create_destroy(ztest_ds_t *zd, uint64_t id)
2715 ztest_shared_opts_t *zo = &ztest_opts;
2719 if (zo->zo_mmp_test)
2723 * Attempt to create using a bad file.
2725 nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1);
2726 VERIFY3U(ENOENT, ==,
2727 spa_create("ztest_bad_file", nvroot, NULL, NULL, NULL));
2728 nvlist_free(nvroot);
2731 * Attempt to create using a bad mirror.
2733 nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 2, 1);
2734 VERIFY3U(ENOENT, ==,
2735 spa_create("ztest_bad_mirror", nvroot, NULL, NULL, NULL));
2736 nvlist_free(nvroot);
2739 * Attempt to create an existing pool. It shouldn't matter
2740 * what's in the nvroot; we should fail with EEXIST.
2742 (void) pthread_rwlock_rdlock(&ztest_name_lock);
2743 nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1);
2744 VERIFY3U(EEXIST, ==,
2745 spa_create(zo->zo_pool, nvroot, NULL, NULL, NULL));
2746 nvlist_free(nvroot);
2747 VERIFY3U(0, ==, spa_open(zo->zo_pool, &spa, FTAG));
2748 VERIFY3U(EBUSY, ==, spa_destroy(zo->zo_pool));
2749 spa_close(spa, FTAG);
2751 (void) pthread_rwlock_unlock(&ztest_name_lock);
2755 * Start and then stop the MMP threads to ensure the startup and shutdown code
2756 * works properly. Actual protection and property-related code tested via ZTS.
2760 ztest_mmp_enable_disable(ztest_ds_t *zd, uint64_t id)
2762 ztest_shared_opts_t *zo = &ztest_opts;
2763 spa_t *spa = ztest_spa;
2765 if (zo->zo_mmp_test)
2768 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
2769 mutex_enter(&spa->spa_props_lock);
2771 if (!spa_multihost(spa)) {
2772 spa->spa_multihost = B_TRUE;
2773 mmp_thread_start(spa);
2776 mutex_exit(&spa->spa_props_lock);
2777 spa_config_exit(spa, SCL_CONFIG, FTAG);
2779 txg_wait_synced(spa_get_dsl(spa), 0);
2780 mmp_signal_all_threads();
2781 txg_wait_synced(spa_get_dsl(spa), 0);
2783 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
2784 mutex_enter(&spa->spa_props_lock);
2786 if (spa_multihost(spa)) {
2787 mmp_thread_stop(spa);
2788 spa->spa_multihost = B_FALSE;
2791 mutex_exit(&spa->spa_props_lock);
2792 spa_config_exit(spa, SCL_CONFIG, FTAG);
2797 ztest_spa_upgrade(ztest_ds_t *zd, uint64_t id)
2800 uint64_t initial_version = SPA_VERSION_INITIAL;
2801 uint64_t version, newversion;
2802 nvlist_t *nvroot, *props;
2805 if (ztest_opts.zo_mmp_test)
2808 mutex_enter(&ztest_vdev_lock);
2809 name = kmem_asprintf("%s_upgrade", ztest_opts.zo_pool);
2812 * Clean up from previous runs.
2814 (void) spa_destroy(name);
2816 nvroot = make_vdev_root(NULL, NULL, name, ztest_opts.zo_vdev_size, 0,
2817 0, ztest_opts.zo_raidz, ztest_opts.zo_mirrors, 1);
2820 * If we're configuring a RAIDZ device then make sure that the
2821 * the initial version is capable of supporting that feature.
2823 switch (ztest_opts.zo_raidz_parity) {
2826 initial_version = SPA_VERSION_INITIAL;
2829 initial_version = SPA_VERSION_RAIDZ2;
2832 initial_version = SPA_VERSION_RAIDZ3;
2837 * Create a pool with a spa version that can be upgraded. Pick
2838 * a value between initial_version and SPA_VERSION_BEFORE_FEATURES.
2841 version = ztest_random_spa_version(initial_version);
2842 } while (version > SPA_VERSION_BEFORE_FEATURES);
2844 props = fnvlist_alloc();
2845 fnvlist_add_uint64(props,
2846 zpool_prop_to_name(ZPOOL_PROP_VERSION), version);
2847 VERIFY3S(spa_create(name, nvroot, props, NULL, NULL), ==, 0);
2848 fnvlist_free(nvroot);
2849 fnvlist_free(props);
2851 VERIFY3S(spa_open(name, &spa, FTAG), ==, 0);
2852 VERIFY3U(spa_version(spa), ==, version);
2853 newversion = ztest_random_spa_version(version + 1);
2855 if (ztest_opts.zo_verbose >= 4) {
2856 (void) printf("upgrading spa version from %llu to %llu\n",
2857 (u_longlong_t)version, (u_longlong_t)newversion);
2860 spa_upgrade(spa, newversion);
2861 VERIFY3U(spa_version(spa), >, version);
2862 VERIFY3U(spa_version(spa), ==, fnvlist_lookup_uint64(spa->spa_config,
2863 zpool_prop_to_name(ZPOOL_PROP_VERSION)));
2864 spa_close(spa, FTAG);
2867 mutex_exit(&ztest_vdev_lock);
2871 ztest_spa_checkpoint(spa_t *spa)
2873 ASSERT(MUTEX_HELD(&ztest_checkpoint_lock));
2875 int error = spa_checkpoint(spa->spa_name);
2879 case ZFS_ERR_DEVRM_IN_PROGRESS:
2880 case ZFS_ERR_DISCARDING_CHECKPOINT:
2881 case ZFS_ERR_CHECKPOINT_EXISTS:
2884 ztest_record_enospc(FTAG);
2887 fatal(0, "spa_checkpoint(%s) = %d", spa->spa_name, error);
2892 ztest_spa_discard_checkpoint(spa_t *spa)
2894 ASSERT(MUTEX_HELD(&ztest_checkpoint_lock));
2896 int error = spa_checkpoint_discard(spa->spa_name);
2900 case ZFS_ERR_DISCARDING_CHECKPOINT:
2901 case ZFS_ERR_NO_CHECKPOINT:
2904 fatal(0, "spa_discard_checkpoint(%s) = %d",
2905 spa->spa_name, error);
2912 ztest_spa_checkpoint_create_discard(ztest_ds_t *zd, uint64_t id)
2914 spa_t *spa = ztest_spa;
2916 mutex_enter(&ztest_checkpoint_lock);
2917 if (ztest_random(2) == 0) {
2918 ztest_spa_checkpoint(spa);
2920 ztest_spa_discard_checkpoint(spa);
2922 mutex_exit(&ztest_checkpoint_lock);
2927 vdev_lookup_by_path(vdev_t *vd, const char *path)
2932 if (vd->vdev_path != NULL && strcmp(path, vd->vdev_path) == 0)
2935 for (c = 0; c < vd->vdev_children; c++)
2936 if ((mvd = vdev_lookup_by_path(vd->vdev_child[c], path)) !=
2944 * Find the first available hole which can be used as a top-level.
2947 find_vdev_hole(spa_t *spa)
2949 vdev_t *rvd = spa->spa_root_vdev;
2952 ASSERT(spa_config_held(spa, SCL_VDEV, RW_READER) == SCL_VDEV);
2954 for (c = 0; c < rvd->vdev_children; c++) {
2955 vdev_t *cvd = rvd->vdev_child[c];
2957 if (cvd->vdev_ishole)
2964 * Verify that vdev_add() works as expected.
2968 ztest_vdev_add_remove(ztest_ds_t *zd, uint64_t id)
2970 ztest_shared_t *zs = ztest_shared;
2971 spa_t *spa = ztest_spa;
2977 if (ztest_opts.zo_mmp_test)
2980 mutex_enter(&ztest_vdev_lock);
2981 leaves = MAX(zs->zs_mirrors + zs->zs_splits, 1) * ztest_opts.zo_raidz;
2983 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2985 ztest_shared->zs_vdev_next_leaf = find_vdev_hole(spa) * leaves;
2988 * If we have slogs then remove them 1/4 of the time.
2990 if (spa_has_slogs(spa) && ztest_random(4) == 0) {
2992 * Grab the guid from the head of the log class rotor.
2994 guid = spa_log_class(spa)->mc_rotor->mg_vd->vdev_guid;
2996 spa_config_exit(spa, SCL_VDEV, FTAG);
2999 * We have to grab the zs_name_lock as writer to
3000 * prevent a race between removing a slog (dmu_objset_find)
3001 * and destroying a dataset. Removing the slog will
3002 * grab a reference on the dataset which may cause
3003 * dsl_destroy_head() to fail with EBUSY thus
3004 * leaving the dataset in an inconsistent state.
3006 pthread_rwlock_wrlock(&ztest_name_lock);
3007 error = spa_vdev_remove(spa, guid, B_FALSE);
3008 pthread_rwlock_unlock(&ztest_name_lock);
3012 case EEXIST: /* Generic zil_reset() error */
3013 case EBUSY: /* Replay required */
3014 case EACCES: /* Crypto key not loaded */
3015 case ZFS_ERR_CHECKPOINT_EXISTS:
3016 case ZFS_ERR_DISCARDING_CHECKPOINT:
3019 fatal(0, "spa_vdev_remove() = %d", error);
3022 spa_config_exit(spa, SCL_VDEV, FTAG);
3025 * Make 1/4 of the devices be log devices.
3027 nvroot = make_vdev_root(NULL, NULL, NULL,
3028 ztest_opts.zo_vdev_size, 0,
3029 ztest_random(4) == 0, ztest_opts.zo_raidz,
3032 error = spa_vdev_add(spa, nvroot);
3033 nvlist_free(nvroot);
3039 ztest_record_enospc("spa_vdev_add");
3042 fatal(0, "spa_vdev_add() = %d", error);
3046 mutex_exit(&ztest_vdev_lock);
3050 * Verify that adding/removing aux devices (l2arc, hot spare) works as expected.
3054 ztest_vdev_aux_add_remove(ztest_ds_t *zd, uint64_t id)
3056 ztest_shared_t *zs = ztest_shared;
3057 spa_t *spa = ztest_spa;
3058 vdev_t *rvd = spa->spa_root_vdev;
3059 spa_aux_vdev_t *sav;
3065 if (ztest_opts.zo_mmp_test)
3068 path = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
3070 if (ztest_random(2) == 0) {
3071 sav = &spa->spa_spares;
3072 aux = ZPOOL_CONFIG_SPARES;
3074 sav = &spa->spa_l2cache;
3075 aux = ZPOOL_CONFIG_L2CACHE;
3078 mutex_enter(&ztest_vdev_lock);
3080 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
3082 if (sav->sav_count != 0 && ztest_random(4) == 0) {
3084 * Pick a random device to remove.
3086 guid = sav->sav_vdevs[ztest_random(sav->sav_count)]->vdev_guid;
3089 * Find an unused device we can add.
3091 zs->zs_vdev_aux = 0;
3094 (void) snprintf(path, MAXPATHLEN, ztest_aux_template,
3095 ztest_opts.zo_dir, ztest_opts.zo_pool, aux,
3097 for (c = 0; c < sav->sav_count; c++)
3098 if (strcmp(sav->sav_vdevs[c]->vdev_path,
3101 if (c == sav->sav_count &&
3102 vdev_lookup_by_path(rvd, path) == NULL)
3108 spa_config_exit(spa, SCL_VDEV, FTAG);
3114 nvlist_t *nvroot = make_vdev_root(NULL, aux, NULL,
3115 (ztest_opts.zo_vdev_size * 5) / 4, 0, 0, 0, 0, 1);
3116 error = spa_vdev_add(spa, nvroot);
3122 fatal(0, "spa_vdev_add(%p) = %d", nvroot, error);
3124 nvlist_free(nvroot);
3127 * Remove an existing device. Sometimes, dirty its
3128 * vdev state first to make sure we handle removal
3129 * of devices that have pending state changes.
3131 if (ztest_random(2) == 0)
3132 (void) vdev_online(spa, guid, 0, NULL);
3134 error = spa_vdev_remove(spa, guid, B_FALSE);
3139 case ZFS_ERR_CHECKPOINT_EXISTS:
3140 case ZFS_ERR_DISCARDING_CHECKPOINT:
3143 fatal(0, "spa_vdev_remove(%llu) = %d", guid, error);
3147 mutex_exit(&ztest_vdev_lock);
3149 umem_free(path, MAXPATHLEN);
3153 * split a pool if it has mirror tlvdevs
3157 ztest_split_pool(ztest_ds_t *zd, uint64_t id)
3159 ztest_shared_t *zs = ztest_shared;
3160 spa_t *spa = ztest_spa;
3161 vdev_t *rvd = spa->spa_root_vdev;
3162 nvlist_t *tree, **child, *config, *split, **schild;
3163 uint_t c, children, schildren = 0, lastlogid = 0;
3166 if (ztest_opts.zo_mmp_test)
3169 mutex_enter(&ztest_vdev_lock);
3171 /* ensure we have a useable config; mirrors of raidz aren't supported */
3172 if (zs->zs_mirrors < 3 || ztest_opts.zo_raidz > 1) {
3173 mutex_exit(&ztest_vdev_lock);
3177 /* clean up the old pool, if any */
3178 (void) spa_destroy("splitp");
3180 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
3182 /* generate a config from the existing config */
3183 mutex_enter(&spa->spa_props_lock);
3184 VERIFY(nvlist_lookup_nvlist(spa->spa_config, ZPOOL_CONFIG_VDEV_TREE,
3186 mutex_exit(&spa->spa_props_lock);
3188 VERIFY(nvlist_lookup_nvlist_array(tree, ZPOOL_CONFIG_CHILDREN, &child,
3191 schild = malloc(rvd->vdev_children * sizeof (nvlist_t *));
3192 for (c = 0; c < children; c++) {
3193 vdev_t *tvd = rvd->vdev_child[c];
3197 if (tvd->vdev_islog || tvd->vdev_ops == &vdev_hole_ops) {
3198 VERIFY(nvlist_alloc(&schild[schildren], NV_UNIQUE_NAME,
3200 VERIFY(nvlist_add_string(schild[schildren],
3201 ZPOOL_CONFIG_TYPE, VDEV_TYPE_HOLE) == 0);
3202 VERIFY(nvlist_add_uint64(schild[schildren],
3203 ZPOOL_CONFIG_IS_HOLE, 1) == 0);
3205 lastlogid = schildren;
3210 VERIFY(nvlist_lookup_nvlist_array(child[c],
3211 ZPOOL_CONFIG_CHILDREN, &mchild, &mchildren) == 0);
3212 VERIFY(nvlist_dup(mchild[0], &schild[schildren++], 0) == 0);
3215 /* OK, create a config that can be used to split */
3216 VERIFY(nvlist_alloc(&split, NV_UNIQUE_NAME, 0) == 0);
3217 VERIFY(nvlist_add_string(split, ZPOOL_CONFIG_TYPE,
3218 VDEV_TYPE_ROOT) == 0);
3219 VERIFY(nvlist_add_nvlist_array(split, ZPOOL_CONFIG_CHILDREN, schild,
3220 lastlogid != 0 ? lastlogid : schildren) == 0);
3222 VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, 0) == 0);
3223 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, split) == 0);
3225 for (c = 0; c < schildren; c++)
3226 nvlist_free(schild[c]);
3230 spa_config_exit(spa, SCL_VDEV, FTAG);
3232 (void) pthread_rwlock_wrlock(&ztest_name_lock);
3233 error = spa_vdev_split_mirror(spa, "splitp", config, NULL, B_FALSE);
3234 (void) pthread_rwlock_unlock(&ztest_name_lock);
3236 nvlist_free(config);
3239 (void) printf("successful split - results:\n");
3240 mutex_enter(&spa_namespace_lock);
3241 show_pool_stats(spa);
3242 show_pool_stats(spa_lookup("splitp"));
3243 mutex_exit(&spa_namespace_lock);
3247 mutex_exit(&ztest_vdev_lock);
3251 * Verify that we can attach and detach devices.
3255 ztest_vdev_attach_detach(ztest_ds_t *zd, uint64_t id)
3257 ztest_shared_t *zs = ztest_shared;
3258 spa_t *spa = ztest_spa;
3259 spa_aux_vdev_t *sav = &spa->spa_spares;
3260 vdev_t *rvd = spa->spa_root_vdev;
3261 vdev_t *oldvd, *newvd, *pvd;
3265 uint64_t ashift = ztest_get_ashift();
3266 uint64_t oldguid, pguid;
3267 uint64_t oldsize, newsize;
3268 char *oldpath, *newpath;
3270 int oldvd_has_siblings = B_FALSE;
3271 int newvd_is_spare = B_FALSE;
3273 int error, expected_error;
3275 if (ztest_opts.zo_mmp_test)
3278 oldpath = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
3279 newpath = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
3281 mutex_enter(&ztest_vdev_lock);
3282 leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz;
3284 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3287 * If a vdev is in the process of being removed, its removal may
3288 * finish while we are in progress, leading to an unexpected error
3289 * value. Don't bother trying to attach while we are in the middle
3292 if (ztest_device_removal_active) {
3293 spa_config_exit(spa, SCL_ALL, FTAG);
3294 mutex_exit(&ztest_vdev_lock);
3299 * Decide whether to do an attach or a replace.
3301 replacing = ztest_random(2);
3304 * Pick a random top-level vdev.
3306 top = ztest_random_vdev_top(spa, B_TRUE);
3309 * Pick a random leaf within it.
3311 leaf = ztest_random(leaves);
3316 oldvd = rvd->vdev_child[top];
3317 if (zs->zs_mirrors >= 1) {
3318 ASSERT(oldvd->vdev_ops == &vdev_mirror_ops);
3319 ASSERT(oldvd->vdev_children >= zs->zs_mirrors);
3320 oldvd = oldvd->vdev_child[leaf / ztest_opts.zo_raidz];
3322 if (ztest_opts.zo_raidz > 1) {
3323 ASSERT(oldvd->vdev_ops == &vdev_raidz_ops);
3324 ASSERT(oldvd->vdev_children == ztest_opts.zo_raidz);
3325 oldvd = oldvd->vdev_child[leaf % ztest_opts.zo_raidz];
3329 * If we're already doing an attach or replace, oldvd may be a
3330 * mirror vdev -- in which case, pick a random child.
3332 while (oldvd->vdev_children != 0) {
3333 oldvd_has_siblings = B_TRUE;
3334 ASSERT(oldvd->vdev_children >= 2);
3335 oldvd = oldvd->vdev_child[ztest_random(oldvd->vdev_children)];
3338 oldguid = oldvd->vdev_guid;
3339 oldsize = vdev_get_min_asize(oldvd);
3340 oldvd_is_log = oldvd->vdev_top->vdev_islog;
3341 (void) strcpy(oldpath, oldvd->vdev_path);
3342 pvd = oldvd->vdev_parent;
3343 pguid = pvd->vdev_guid;
3346 * If oldvd has siblings, then half of the time, detach it.
3348 if (oldvd_has_siblings && ztest_random(2) == 0) {
3349 spa_config_exit(spa, SCL_ALL, FTAG);
3350 error = spa_vdev_detach(spa, oldguid, pguid, B_FALSE);
3351 if (error != 0 && error != ENODEV && error != EBUSY &&
3352 error != ENOTSUP && error != ZFS_ERR_CHECKPOINT_EXISTS &&
3353 error != ZFS_ERR_DISCARDING_CHECKPOINT)
3354 fatal(0, "detach (%s) returned %d", oldpath, error);
3359 * For the new vdev, choose with equal probability between the two
3360 * standard paths (ending in either 'a' or 'b') or a random hot spare.
3362 if (sav->sav_count != 0 && ztest_random(3) == 0) {
3363 newvd = sav->sav_vdevs[ztest_random(sav->sav_count)];
3364 newvd_is_spare = B_TRUE;
3365 (void) strcpy(newpath, newvd->vdev_path);
3367 (void) snprintf(newpath, MAXPATHLEN, ztest_dev_template,
3368 ztest_opts.zo_dir, ztest_opts.zo_pool,
3369 top * leaves + leaf);
3370 if (ztest_random(2) == 0)
3371 newpath[strlen(newpath) - 1] = 'b';
3372 newvd = vdev_lookup_by_path(rvd, newpath);
3377 * Reopen to ensure the vdev's asize field isn't stale.
3380 newsize = vdev_get_min_asize(newvd);
3383 * Make newsize a little bigger or smaller than oldsize.
3384 * If it's smaller, the attach should fail.
3385 * If it's larger, and we're doing a replace,
3386 * we should get dynamic LUN growth when we're done.
3388 newsize = 10 * oldsize / (9 + ztest_random(3));
3392 * If pvd is not a mirror or root, the attach should fail with ENOTSUP,
3393 * unless it's a replace; in that case any non-replacing parent is OK.
3395 * If newvd is already part of the pool, it should fail with EBUSY.
3397 * If newvd is too small, it should fail with EOVERFLOW.
3399 if (pvd->vdev_ops != &vdev_mirror_ops &&
3400 pvd->vdev_ops != &vdev_root_ops && (!replacing ||
3401 pvd->vdev_ops == &vdev_replacing_ops ||
3402 pvd->vdev_ops == &vdev_spare_ops))
3403 expected_error = ENOTSUP;
3404 else if (newvd_is_spare && (!replacing || oldvd_is_log))
3405 expected_error = ENOTSUP;
3406 else if (newvd == oldvd)
3407 expected_error = replacing ? 0 : EBUSY;
3408 else if (vdev_lookup_by_path(rvd, newpath) != NULL)
3409 expected_error = EBUSY;
3410 else if (newsize < oldsize)
3411 expected_error = EOVERFLOW;
3412 else if (ashift > oldvd->vdev_top->vdev_ashift)
3413 expected_error = EDOM;
3417 spa_config_exit(spa, SCL_ALL, FTAG);
3420 * Build the nvlist describing newpath.
3422 root = make_vdev_root(newpath, NULL, NULL, newvd == NULL ? newsize : 0,
3423 ashift, 0, 0, 0, 1);
3425 error = spa_vdev_attach(spa, oldguid, root, replacing);
3430 * If our parent was the replacing vdev, but the replace completed,
3431 * then instead of failing with ENOTSUP we may either succeed,
3432 * fail with ENODEV, or fail with EOVERFLOW.
3434 if (expected_error == ENOTSUP &&
3435 (error == 0 || error == ENODEV || error == EOVERFLOW))
3436 expected_error = error;
3439 * If someone grew the LUN, the replacement may be too small.
3441 if (error == EOVERFLOW || error == EBUSY)
3442 expected_error = error;
3444 if (error == ZFS_ERR_CHECKPOINT_EXISTS ||
3445 error == ZFS_ERR_DISCARDING_CHECKPOINT)
3446 expected_error = error;
3448 /* XXX workaround 6690467 */
3449 if (error != expected_error && expected_error != EBUSY) {
3450 fatal(0, "attach (%s %llu, %s %llu, %d) "
3451 "returned %d, expected %d",
3452 oldpath, oldsize, newpath,
3453 newsize, replacing, error, expected_error);
3456 mutex_exit(&ztest_vdev_lock);
3458 umem_free(oldpath, MAXPATHLEN);
3459 umem_free(newpath, MAXPATHLEN);
3464 ztest_device_removal(ztest_ds_t *zd, uint64_t id)
3466 spa_t *spa = ztest_spa;
3471 mutex_enter(&ztest_vdev_lock);
3473 if (ztest_device_removal_active) {
3474 mutex_exit(&ztest_vdev_lock);
3479 * Remove a random top-level vdev and wait for removal to finish.
3481 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
3482 vd = vdev_lookup_top(spa, ztest_random_vdev_top(spa, B_FALSE));
3483 guid = vd->vdev_guid;
3484 spa_config_exit(spa, SCL_VDEV, FTAG);
3486 error = spa_vdev_remove(spa, guid, B_FALSE);
3488 ztest_device_removal_active = B_TRUE;
3489 mutex_exit(&ztest_vdev_lock);
3491 while (spa->spa_vdev_removal != NULL)
3492 txg_wait_synced(spa_get_dsl(spa), 0);
3494 mutex_exit(&ztest_vdev_lock);
3499 * The pool needs to be scrubbed after completing device removal.
3500 * Failure to do so may result in checksum errors due to the
3501 * strategy employed by ztest_fault_inject() when selecting which
3502 * offset are redundant and can be damaged.
3504 error = spa_scan(spa, POOL_SCAN_SCRUB);
3506 while (dsl_scan_scrubbing(spa_get_dsl(spa)))
3507 txg_wait_synced(spa_get_dsl(spa), 0);
3510 mutex_enter(&ztest_vdev_lock);
3511 ztest_device_removal_active = B_FALSE;
3512 mutex_exit(&ztest_vdev_lock);
3516 * Callback function which expands the physical size of the vdev.
3519 grow_vdev(vdev_t *vd, void *arg)
3521 ASSERTV(spa_t *spa = vd->vdev_spa);
3522 size_t *newsize = arg;
3526 ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
3527 ASSERT(vd->vdev_ops->vdev_op_leaf);
3529 if ((fd = open(vd->vdev_path, O_RDWR)) == -1)
3532 fsize = lseek(fd, 0, SEEK_END);
3533 VERIFY(ftruncate(fd, *newsize) == 0);
3535 if (ztest_opts.zo_verbose >= 6) {
3536 (void) printf("%s grew from %lu to %lu bytes\n",
3537 vd->vdev_path, (ulong_t)fsize, (ulong_t)*newsize);
3544 * Callback function which expands a given vdev by calling vdev_online().
3548 online_vdev(vdev_t *vd, void *arg)
3550 spa_t *spa = vd->vdev_spa;
3551 vdev_t *tvd = vd->vdev_top;
3552 uint64_t guid = vd->vdev_guid;
3553 uint64_t generation = spa->spa_config_generation + 1;
3554 vdev_state_t newstate = VDEV_STATE_UNKNOWN;
3557 ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
3558 ASSERT(vd->vdev_ops->vdev_op_leaf);
3560 /* Calling vdev_online will initialize the new metaslabs */
3561 spa_config_exit(spa, SCL_STATE, spa);
3562 error = vdev_online(spa, guid, ZFS_ONLINE_EXPAND, &newstate);
3563 spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3566 * If vdev_online returned an error or the underlying vdev_open
3567 * failed then we abort the expand. The only way to know that
3568 * vdev_open fails is by checking the returned newstate.
3570 if (error || newstate != VDEV_STATE_HEALTHY) {
3571 if (ztest_opts.zo_verbose >= 5) {
3572 (void) printf("Unable to expand vdev, state %llu, "
3573 "error %d\n", (u_longlong_t)newstate, error);
3577 ASSERT3U(newstate, ==, VDEV_STATE_HEALTHY);
3580 * Since we dropped the lock we need to ensure that we're
3581 * still talking to the original vdev. It's possible this
3582 * vdev may have been detached/replaced while we were
3583 * trying to online it.
3585 if (generation != spa->spa_config_generation) {
3586 if (ztest_opts.zo_verbose >= 5) {
3587 (void) printf("vdev configuration has changed, "
3588 "guid %llu, state %llu, expected gen %llu, "
3591 (u_longlong_t)tvd->vdev_state,
3592 (u_longlong_t)generation,
3593 (u_longlong_t)spa->spa_config_generation);
3601 * Traverse the vdev tree calling the supplied function.
3602 * We continue to walk the tree until we either have walked all
3603 * children or we receive a non-NULL return from the callback.
3604 * If a NULL callback is passed, then we just return back the first
3605 * leaf vdev we encounter.
3608 vdev_walk_tree(vdev_t *vd, vdev_t *(*func)(vdev_t *, void *), void *arg)
3612 if (vd->vdev_ops->vdev_op_leaf) {
3616 return (func(vd, arg));
3619 for (c = 0; c < vd->vdev_children; c++) {
3620 vdev_t *cvd = vd->vdev_child[c];
3621 if ((cvd = vdev_walk_tree(cvd, func, arg)) != NULL)
3628 * Verify that dynamic LUN growth works as expected.
3632 ztest_vdev_LUN_growth(ztest_ds_t *zd, uint64_t id)
3634 spa_t *spa = ztest_spa;
3636 metaslab_class_t *mc;
3637 metaslab_group_t *mg;
3638 size_t psize, newsize;
3640 uint64_t old_class_space, new_class_space, old_ms_count, new_ms_count;
3642 mutex_enter(&ztest_checkpoint_lock);
3643 mutex_enter(&ztest_vdev_lock);
3644 spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3647 * If there is a vdev removal in progress, it could complete while
3648 * we are running, in which case we would not be able to verify
3649 * that the metaslab_class space increased (because it decreases
3650 * when the device removal completes).
3652 if (ztest_device_removal_active) {
3653 spa_config_exit(spa, SCL_STATE, spa);
3654 mutex_exit(&ztest_vdev_lock);
3655 mutex_exit(&ztest_checkpoint_lock);
3659 top = ztest_random_vdev_top(spa, B_TRUE);
3661 tvd = spa->spa_root_vdev->vdev_child[top];
3664 old_ms_count = tvd->vdev_ms_count;
3665 old_class_space = metaslab_class_get_space(mc);
3668 * Determine the size of the first leaf vdev associated with
3669 * our top-level device.
3671 vd = vdev_walk_tree(tvd, NULL, NULL);
3672 ASSERT3P(vd, !=, NULL);
3673 ASSERT(vd->vdev_ops->vdev_op_leaf);
3675 psize = vd->vdev_psize;
3678 * We only try to expand the vdev if it's healthy, less than 4x its
3679 * original size, and it has a valid psize.
3681 if (tvd->vdev_state != VDEV_STATE_HEALTHY ||
3682 psize == 0 || psize >= 4 * ztest_opts.zo_vdev_size) {
3683 spa_config_exit(spa, SCL_STATE, spa);
3684 mutex_exit(&ztest_vdev_lock);
3685 mutex_exit(&ztest_checkpoint_lock);
3689 newsize = psize + psize / 8;
3690 ASSERT3U(newsize, >, psize);
3692 if (ztest_opts.zo_verbose >= 6) {
3693 (void) printf("Expanding LUN %s from %lu to %lu\n",
3694 vd->vdev_path, (ulong_t)psize, (ulong_t)newsize);
3698 * Growing the vdev is a two step process:
3699 * 1). expand the physical size (i.e. relabel)
3700 * 2). online the vdev to create the new metaslabs
3702 if (vdev_walk_tree(tvd, grow_vdev, &newsize) != NULL ||
3703 vdev_walk_tree(tvd, online_vdev, NULL) != NULL ||
3704 tvd->vdev_state != VDEV_STATE_HEALTHY) {
3705 if (ztest_opts.zo_verbose >= 5) {
3706 (void) printf("Could not expand LUN because "
3707 "the vdev configuration changed.\n");
3709 spa_config_exit(spa, SCL_STATE, spa);
3710 mutex_exit(&ztest_vdev_lock);
3711 mutex_exit(&ztest_checkpoint_lock);
3715 spa_config_exit(spa, SCL_STATE, spa);
3718 * Expanding the LUN will update the config asynchronously,
3719 * thus we must wait for the async thread to complete any
3720 * pending tasks before proceeding.
3724 mutex_enter(&spa->spa_async_lock);
3725 done = (spa->spa_async_thread == NULL && !spa->spa_async_tasks);
3726 mutex_exit(&spa->spa_async_lock);
3729 txg_wait_synced(spa_get_dsl(spa), 0);
3730 (void) poll(NULL, 0, 100);
3733 spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3735 tvd = spa->spa_root_vdev->vdev_child[top];
3736 new_ms_count = tvd->vdev_ms_count;
3737 new_class_space = metaslab_class_get_space(mc);
3739 if (tvd->vdev_mg != mg || mg->mg_class != mc) {
3740 if (ztest_opts.zo_verbose >= 5) {
3741 (void) printf("Could not verify LUN expansion due to "
3742 "intervening vdev offline or remove.\n");
3744 spa_config_exit(spa, SCL_STATE, spa);
3745 mutex_exit(&ztest_vdev_lock);
3746 mutex_exit(&ztest_checkpoint_lock);
3751 * Make sure we were able to grow the vdev.
3753 if (new_ms_count <= old_ms_count) {
3754 fatal(0, "LUN expansion failed: ms_count %llu < %llu\n",
3755 old_ms_count, new_ms_count);
3759 * Make sure we were able to grow the pool.
3761 if (new_class_space <= old_class_space) {
3762 fatal(0, "LUN expansion failed: class_space %llu < %llu\n",
3763 old_class_space, new_class_space);
3766 if (ztest_opts.zo_verbose >= 5) {
3767 char oldnumbuf[NN_NUMBUF_SZ], newnumbuf[NN_NUMBUF_SZ];
3769 nicenum(old_class_space, oldnumbuf, sizeof (oldnumbuf));
3770 nicenum(new_class_space, newnumbuf, sizeof (newnumbuf));
3771 (void) printf("%s grew from %s to %s\n",
3772 spa->spa_name, oldnumbuf, newnumbuf);
3775 spa_config_exit(spa, SCL_STATE, spa);
3776 mutex_exit(&ztest_vdev_lock);
3777 mutex_exit(&ztest_checkpoint_lock);
3781 * Verify that dmu_objset_{create,destroy,open,close} work as expected.
3785 ztest_objset_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
3788 * Create the objects common to all ztest datasets.
3790 VERIFY(zap_create_claim(os, ZTEST_DIROBJ,
3791 DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx) == 0);
3795 ztest_dataset_create(char *dsname)
3799 dsl_crypto_params_t *dcp = NULL;
3802 * 50% of the time, we create encrypted datasets
3803 * using a random cipher suite and a hard-coded
3806 rand = ztest_random(2);
3808 nvlist_t *crypto_args = fnvlist_alloc();
3809 nvlist_t *props = fnvlist_alloc();
3811 /* slight bias towards the default cipher suite */
3812 rand = ztest_random(ZIO_CRYPT_FUNCTIONS);
3813 if (rand < ZIO_CRYPT_AES_128_CCM)
3814 rand = ZIO_CRYPT_ON;
3816 fnvlist_add_uint64(props,
3817 zfs_prop_to_name(ZFS_PROP_ENCRYPTION), rand);
3818 fnvlist_add_uint8_array(crypto_args, "wkeydata",
3819 (uint8_t *)ztest_wkeydata, WRAPPING_KEY_LEN);
3822 * These parameters aren't really used by the kernel. They
3823 * are simply stored so that userspace knows how to load
3826 fnvlist_add_uint64(props,
3827 zfs_prop_to_name(ZFS_PROP_KEYFORMAT), ZFS_KEYFORMAT_RAW);
3828 fnvlist_add_string(props,
3829 zfs_prop_to_name(ZFS_PROP_KEYLOCATION), "prompt");
3830 fnvlist_add_uint64(props,
3831 zfs_prop_to_name(ZFS_PROP_PBKDF2_SALT), 0ULL);
3832 fnvlist_add_uint64(props,
3833 zfs_prop_to_name(ZFS_PROP_PBKDF2_ITERS), 0ULL);
3835 VERIFY0(dsl_crypto_params_create_nvlist(DCP_CMD_NONE, props,
3836 crypto_args, &dcp));
3838 fnvlist_free(crypto_args);
3839 fnvlist_free(props);
3842 err = dmu_objset_create(dsname, DMU_OST_OTHER, 0, dcp,
3843 ztest_objset_create_cb, NULL);
3844 dsl_crypto_params_free(dcp, !!err);
3846 rand = ztest_random(100);
3847 if (err || rand < 80)
3850 if (ztest_opts.zo_verbose >= 5)
3851 (void) printf("Setting dataset %s to sync always\n", dsname);
3852 return (ztest_dsl_prop_set_uint64(dsname, ZFS_PROP_SYNC,
3853 ZFS_SYNC_ALWAYS, B_FALSE));
3858 ztest_objset_destroy_cb(const char *name, void *arg)
3861 dmu_object_info_t doi;
3865 * Verify that the dataset contains a directory object.
3867 VERIFY0(ztest_dmu_objset_own(name, DMU_OST_OTHER, B_TRUE,
3868 B_TRUE, FTAG, &os));
3869 error = dmu_object_info(os, ZTEST_DIROBJ, &doi);
3870 if (error != ENOENT) {
3871 /* We could have crashed in the middle of destroying it */
3873 ASSERT3U(doi.doi_type, ==, DMU_OT_ZAP_OTHER);
3874 ASSERT3S(doi.doi_physical_blocks_512, >=, 0);
3876 dmu_objset_disown(os, B_TRUE, FTAG);
3879 * Destroy the dataset.
3881 if (strchr(name, '@') != NULL) {
3882 VERIFY0(dsl_destroy_snapshot(name, B_TRUE));
3884 error = dsl_destroy_head(name);
3885 /* There could be a hold on this dataset */
3893 ztest_snapshot_create(char *osname, uint64_t id)
3895 char snapname[ZFS_MAX_DATASET_NAME_LEN];
3898 (void) snprintf(snapname, sizeof (snapname), "%llu", (u_longlong_t)id);
3900 error = dmu_objset_snapshot_one(osname, snapname);
3901 if (error == ENOSPC) {
3902 ztest_record_enospc(FTAG);
3905 if (error != 0 && error != EEXIST) {
3906 fatal(0, "ztest_snapshot_create(%s@%s) = %d", osname,
3913 ztest_snapshot_destroy(char *osname, uint64_t id)
3915 char snapname[ZFS_MAX_DATASET_NAME_LEN];
3918 (void) snprintf(snapname, sizeof (snapname), "%s@%llu", osname,
3921 error = dsl_destroy_snapshot(snapname, B_FALSE);
3922 if (error != 0 && error != ENOENT)
3923 fatal(0, "ztest_snapshot_destroy(%s) = %d", snapname, error);
3929 ztest_dmu_objset_create_destroy(ztest_ds_t *zd, uint64_t id)
3935 char name[ZFS_MAX_DATASET_NAME_LEN];
3939 zdtmp = umem_alloc(sizeof (ztest_ds_t), UMEM_NOFAIL);
3941 (void) pthread_rwlock_rdlock(&ztest_name_lock);
3943 (void) snprintf(name, sizeof (name), "%s/temp_%llu",
3944 ztest_opts.zo_pool, (u_longlong_t)id);
3947 * If this dataset exists from a previous run, process its replay log
3948 * half of the time. If we don't replay it, then dsl_destroy_head()
3949 * (invoked from ztest_objset_destroy_cb()) should just throw it away.
3951 if (ztest_random(2) == 0 &&
3952 ztest_dmu_objset_own(name, DMU_OST_OTHER, B_FALSE,
3953 B_TRUE, FTAG, &os) == 0) {
3954 ztest_zd_init(zdtmp, NULL, os);
3955 zil_replay(os, zdtmp, ztest_replay_vector);
3956 ztest_zd_fini(zdtmp);
3957 txg_wait_synced(dmu_objset_pool(os), 0);
3958 dmu_objset_disown(os, B_TRUE, FTAG);
3962 * There may be an old instance of the dataset we're about to
3963 * create lying around from a previous run. If so, destroy it
3964 * and all of its snapshots.
3966 (void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
3967 DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
3970 * Verify that the destroyed dataset is no longer in the namespace.
3972 VERIFY3U(ENOENT, ==, ztest_dmu_objset_own(name, DMU_OST_OTHER, B_TRUE,
3973 B_TRUE, FTAG, &os));
3976 * Verify that we can create a new dataset.
3978 error = ztest_dataset_create(name);
3980 if (error == ENOSPC) {
3981 ztest_record_enospc(FTAG);
3984 fatal(0, "dmu_objset_create(%s) = %d", name, error);
3987 VERIFY0(ztest_dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, B_TRUE,
3990 ztest_zd_init(zdtmp, NULL, os);
3993 * Open the intent log for it.
3995 zilog = zil_open(os, ztest_get_data);
3998 * Put some objects in there, do a little I/O to them,
3999 * and randomly take a couple of snapshots along the way.
4001 iters = ztest_random(5);
4002 for (i = 0; i < iters; i++) {
4003 ztest_dmu_object_alloc_free(zdtmp, id);
4004 if (ztest_random(iters) == 0)
4005 (void) ztest_snapshot_create(name, i);
4009 * Verify that we cannot create an existing dataset.
4011 VERIFY3U(EEXIST, ==,
4012 dmu_objset_create(name, DMU_OST_OTHER, 0, NULL, NULL, NULL));
4015 * Verify that we can hold an objset that is also owned.
4017 VERIFY3U(0, ==, dmu_objset_hold(name, FTAG, &os2));
4018 dmu_objset_rele(os2, FTAG);
4021 * Verify that we cannot own an objset that is already owned.
4023 VERIFY3U(EBUSY, ==, ztest_dmu_objset_own(name, DMU_OST_OTHER,
4024 B_FALSE, B_TRUE, FTAG, &os2));
4027 txg_wait_synced(spa_get_dsl(os->os_spa), 0);
4028 dmu_objset_disown(os, B_TRUE, FTAG);
4029 ztest_zd_fini(zdtmp);
4031 (void) pthread_rwlock_unlock(&ztest_name_lock);
4033 umem_free(zdtmp, sizeof (ztest_ds_t));
4037 * Verify that dmu_snapshot_{create,destroy,open,close} work as expected.
4040 ztest_dmu_snapshot_create_destroy(ztest_ds_t *zd, uint64_t id)
4042 (void) pthread_rwlock_rdlock(&ztest_name_lock);
4043 (void) ztest_snapshot_destroy(zd->zd_name, id);
4044 (void) ztest_snapshot_create(zd->zd_name, id);
4045 (void) pthread_rwlock_unlock(&ztest_name_lock);
4049 * Cleanup non-standard snapshots and clones.
4052 ztest_dsl_dataset_cleanup(char *osname, uint64_t id)
4061 snap1name = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
4062 clone1name = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
4063 snap2name = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
4064 clone2name = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
4065 snap3name = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
4067 (void) snprintf(snap1name, ZFS_MAX_DATASET_NAME_LEN,
4068 "%s@s1_%llu", osname, (u_longlong_t)id);
4069 (void) snprintf(clone1name, ZFS_MAX_DATASET_NAME_LEN,
4070 "%s/c1_%llu", osname, (u_longlong_t)id);
4071 (void) snprintf(snap2name, ZFS_MAX_DATASET_NAME_LEN,
4072 "%s@s2_%llu", clone1name, (u_longlong_t)id);
4073 (void) snprintf(clone2name, ZFS_MAX_DATASET_NAME_LEN,
4074 "%s/c2_%llu", osname, (u_longlong_t)id);
4075 (void) snprintf(snap3name, ZFS_MAX_DATASET_NAME_LEN,
4076 "%s@s3_%llu", clone1name, (u_longlong_t)id);
4078 error = dsl_destroy_head(clone2name);
4079 if (error && error != ENOENT)
4080 fatal(0, "dsl_destroy_head(%s) = %d", clone2name, error);
4081 error = dsl_destroy_snapshot(snap3name, B_FALSE);
4082 if (error && error != ENOENT)
4083 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap3name, error);
4084 error = dsl_destroy_snapshot(snap2name, B_FALSE);
4085 if (error && error != ENOENT)
4086 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap2name, error);
4087 error = dsl_destroy_head(clone1name);
4088 if (error && error != ENOENT)
4089 fatal(0, "dsl_destroy_head(%s) = %d", clone1name, error);
4090 error = dsl_destroy_snapshot(snap1name, B_FALSE);
4091 if (error && error != ENOENT)
4092 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap1name, error);
4094 umem_free(snap1name, ZFS_MAX_DATASET_NAME_LEN);
4095 umem_free(clone1name, ZFS_MAX_DATASET_NAME_LEN);
4096 umem_free(snap2name, ZFS_MAX_DATASET_NAME_LEN);
4097 umem_free(clone2name, ZFS_MAX_DATASET_NAME_LEN);
4098 umem_free(snap3name, ZFS_MAX_DATASET_NAME_LEN);
4102 * Verify dsl_dataset_promote handles EBUSY
4105 ztest_dsl_dataset_promote_busy(ztest_ds_t *zd, uint64_t id)
4113 char *osname = zd->zd_name;
4116 snap1name = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
4117 clone1name = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
4118 snap2name = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
4119 clone2name = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
4120 snap3name = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
4122 (void) pthread_rwlock_rdlock(&ztest_name_lock);
4124 ztest_dsl_dataset_cleanup(osname, id);
4126 (void) snprintf(snap1name, ZFS_MAX_DATASET_NAME_LEN,
4127 "%s@s1_%llu", osname, (u_longlong_t)id);
4128 (void) snprintf(clone1name, ZFS_MAX_DATASET_NAME_LEN,
4129 "%s/c1_%llu", osname, (u_longlong_t)id);
4130 (void) snprintf(snap2name, ZFS_MAX_DATASET_NAME_LEN,
4131 "%s@s2_%llu", clone1name, (u_longlong_t)id);
4132 (void) snprintf(clone2name, ZFS_MAX_DATASET_NAME_LEN,
4133 "%s/c2_%llu", osname, (u_longlong_t)id);
4134 (void) snprintf(snap3name, ZFS_MAX_DATASET_NAME_LEN,
4135 "%s@s3_%llu", clone1name, (u_longlong_t)id);
4137 error = dmu_objset_snapshot_one(osname, strchr(snap1name, '@') + 1);
4138 if (error && error != EEXIST) {
4139 if (error == ENOSPC) {
4140 ztest_record_enospc(FTAG);
4143 fatal(0, "dmu_take_snapshot(%s) = %d", snap1name, error);
4146 error = dmu_objset_clone(clone1name, snap1name);
4148 if (error == ENOSPC) {
4149 ztest_record_enospc(FTAG);
4152 fatal(0, "dmu_objset_create(%s) = %d", clone1name, error);
4155 error = dmu_objset_snapshot_one(clone1name, strchr(snap2name, '@') + 1);
4156 if (error && error != EEXIST) {
4157 if (error == ENOSPC) {
4158 ztest_record_enospc(FTAG);
4161 fatal(0, "dmu_open_snapshot(%s) = %d", snap2name, error);
4164 error = dmu_objset_snapshot_one(clone1name, strchr(snap3name, '@') + 1);
4165 if (error && error != EEXIST) {
4166 if (error == ENOSPC) {
4167 ztest_record_enospc(FTAG);
4170 fatal(0, "dmu_open_snapshot(%s) = %d", snap3name, error);
4173 error = dmu_objset_clone(clone2name, snap3name);
4175 if (error == ENOSPC) {
4176 ztest_record_enospc(FTAG);
4179 fatal(0, "dmu_objset_create(%s) = %d", clone2name, error);
4182 error = ztest_dmu_objset_own(snap2name, DMU_OST_ANY, B_TRUE, B_TRUE,
4185 fatal(0, "dmu_objset_own(%s) = %d", snap2name, error);
4186 error = dsl_dataset_promote(clone2name, NULL);
4187 if (error == ENOSPC) {
4188 dmu_objset_disown(os, B_TRUE, FTAG);
4189 ztest_record_enospc(FTAG);
4193 fatal(0, "dsl_dataset_promote(%s), %d, not EBUSY", clone2name,
4195 dmu_objset_disown(os, B_TRUE, FTAG);
4198 ztest_dsl_dataset_cleanup(osname, id);
4200 (void) pthread_rwlock_unlock(&ztest_name_lock);
4202 umem_free(snap1name, ZFS_MAX_DATASET_NAME_LEN);
4203 umem_free(clone1name, ZFS_MAX_DATASET_NAME_LEN);
4204 umem_free(snap2name, ZFS_MAX_DATASET_NAME_LEN);
4205 umem_free(clone2name, ZFS_MAX_DATASET_NAME_LEN);
4206 umem_free(snap3name, ZFS_MAX_DATASET_NAME_LEN);
4209 #undef OD_ARRAY_SIZE
4210 #define OD_ARRAY_SIZE 4
4213 * Verify that dmu_object_{alloc,free} work as expected.
4216 ztest_dmu_object_alloc_free(ztest_ds_t *zd, uint64_t id)
4223 size = sizeof (ztest_od_t) * OD_ARRAY_SIZE;
4224 od = umem_alloc(size, UMEM_NOFAIL);
4225 batchsize = OD_ARRAY_SIZE;
4227 for (b = 0; b < batchsize; b++)
4228 ztest_od_init(od + b, id, FTAG, b, DMU_OT_UINT64_OTHER,
4232 * Destroy the previous batch of objects, create a new batch,
4233 * and do some I/O on the new objects.
4235 if (ztest_object_init(zd, od, size, B_TRUE) != 0)
4238 while (ztest_random(4 * batchsize) != 0)
4239 ztest_io(zd, od[ztest_random(batchsize)].od_object,
4240 ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
4242 umem_free(od, size);
4246 * Rewind the global allocator to verify object allocation backfilling.
4249 ztest_dmu_object_next_chunk(ztest_ds_t *zd, uint64_t id)
4251 objset_t *os = zd->zd_os;
4252 int dnodes_per_chunk = 1 << dmu_object_alloc_chunk_shift;
4256 * Rewind the global allocator randomly back to a lower object number
4257 * to force backfilling and reclamation of recently freed dnodes.
4259 mutex_enter(&os->os_obj_lock);
4260 object = ztest_random(os->os_obj_next_chunk);
4261 os->os_obj_next_chunk = P2ALIGN(object, dnodes_per_chunk);
4262 mutex_exit(&os->os_obj_lock);
4265 #undef OD_ARRAY_SIZE
4266 #define OD_ARRAY_SIZE 2
4269 * Verify that dmu_{read,write} work as expected.
4272 ztest_dmu_read_write(ztest_ds_t *zd, uint64_t id)
4277 objset_t *os = zd->zd_os;
4278 size = sizeof (ztest_od_t) * OD_ARRAY_SIZE;
4279 od = umem_alloc(size, UMEM_NOFAIL);
4281 int i, freeit, error;
4283 bufwad_t *packbuf, *bigbuf, *pack, *bigH, *bigT;
4284 uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
4285 uint64_t chunksize = (1000 + ztest_random(1000)) * sizeof (uint64_t);
4286 uint64_t regions = 997;
4287 uint64_t stride = 123456789ULL;
4288 uint64_t width = 40;
4289 int free_percent = 5;
4292 * This test uses two objects, packobj and bigobj, that are always
4293 * updated together (i.e. in the same tx) so that their contents are
4294 * in sync and can be compared. Their contents relate to each other
4295 * in a simple way: packobj is a dense array of 'bufwad' structures,
4296 * while bigobj is a sparse array of the same bufwads. Specifically,
4297 * for any index n, there are three bufwads that should be identical:
4299 * packobj, at offset n * sizeof (bufwad_t)
4300 * bigobj, at the head of the nth chunk
4301 * bigobj, at the tail of the nth chunk
4303 * The chunk size is arbitrary. It doesn't have to be a power of two,
4304 * and it doesn't have any relation to the object blocksize.
4305 * The only requirement is that it can hold at least two bufwads.
4307 * Normally, we write the bufwad to each of these locations.
4308 * However, free_percent of the time we instead write zeroes to
4309 * packobj and perform a dmu_free_range() on bigobj. By comparing
4310 * bigobj to packobj, we can verify that the DMU is correctly
4311 * tracking which parts of an object are allocated and free,
4312 * and that the contents of the allocated blocks are correct.
4316 * Read the directory info. If it's the first time, set things up.
4318 ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0, chunksize);
4319 ztest_od_init(od + 1, id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, 0,
4322 if (ztest_object_init(zd, od, size, B_FALSE) != 0) {
4323 umem_free(od, size);
4327 bigobj = od[0].od_object;
4328 packobj = od[1].od_object;
4329 chunksize = od[0].od_gen;
4330 ASSERT(chunksize == od[1].od_gen);
4333 * Prefetch a random chunk of the big object.
4334 * Our aim here is to get some async reads in flight
4335 * for blocks that we may free below; the DMU should
4336 * handle this race correctly.
4338 n = ztest_random(regions) * stride + ztest_random(width);
4339 s = 1 + ztest_random(2 * width - 1);
4340 dmu_prefetch(os, bigobj, 0, n * chunksize, s * chunksize,
4341 ZIO_PRIORITY_SYNC_READ);
4344 * Pick a random index and compute the offsets into packobj and bigobj.
4346 n = ztest_random(regions) * stride + ztest_random(width);
4347 s = 1 + ztest_random(width - 1);
4349 packoff = n * sizeof (bufwad_t);
4350 packsize = s * sizeof (bufwad_t);
4352 bigoff = n * chunksize;
4353 bigsize = s * chunksize;
4355 packbuf = umem_alloc(packsize, UMEM_NOFAIL);
4356 bigbuf = umem_alloc(bigsize, UMEM_NOFAIL);
4359 * free_percent of the time, free a range of bigobj rather than
4362 freeit = (ztest_random(100) < free_percent);
4365 * Read the current contents of our objects.
4367 error = dmu_read(os, packobj, packoff, packsize, packbuf,
4370 error = dmu_read(os, bigobj, bigoff, bigsize, bigbuf,
4375 * Get a tx for the mods to both packobj and bigobj.
4377 tx = dmu_tx_create(os);
4379 dmu_tx_hold_write(tx, packobj, packoff, packsize);
4382 dmu_tx_hold_free(tx, bigobj, bigoff, bigsize);
4384 dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
4386 /* This accounts for setting the checksum/compression. */
4387 dmu_tx_hold_bonus(tx, bigobj);
4389 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4391 umem_free(packbuf, packsize);
4392 umem_free(bigbuf, bigsize);
4393 umem_free(od, size);
4397 enum zio_checksum cksum;
4399 cksum = (enum zio_checksum)
4400 ztest_random_dsl_prop(ZFS_PROP_CHECKSUM);
4401 } while (cksum >= ZIO_CHECKSUM_LEGACY_FUNCTIONS);
4402 dmu_object_set_checksum(os, bigobj, cksum, tx);
4404 enum zio_compress comp;
4406 comp = (enum zio_compress)
4407 ztest_random_dsl_prop(ZFS_PROP_COMPRESSION);
4408 } while (comp >= ZIO_COMPRESS_LEGACY_FUNCTIONS);
4409 dmu_object_set_compress(os, bigobj, comp, tx);
4412 * For each index from n to n + s, verify that the existing bufwad
4413 * in packobj matches the bufwads at the head and tail of the
4414 * corresponding chunk in bigobj. Then update all three bufwads
4415 * with the new values we want to write out.
4417 for (i = 0; i < s; i++) {
4419 pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
4421 bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
4423 bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
4425 ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
4426 ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
4428 if (pack->bw_txg > txg)
4429 fatal(0, "future leak: got %llx, open txg is %llx",
4432 if (pack->bw_data != 0 && pack->bw_index != n + i)
4433 fatal(0, "wrong index: got %llx, wanted %llx+%llx",
4434 pack->bw_index, n, i);
4436 if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
4437 fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
4439 if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
4440 fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
4443 bzero(pack, sizeof (bufwad_t));
4445 pack->bw_index = n + i;
4447 pack->bw_data = 1 + ztest_random(-2ULL);
4454 * We've verified all the old bufwads, and made new ones.
4455 * Now write them out.
4457 dmu_write(os, packobj, packoff, packsize, packbuf, tx);
4460 if (ztest_opts.zo_verbose >= 7) {
4461 (void) printf("freeing offset %llx size %llx"
4463 (u_longlong_t)bigoff,
4464 (u_longlong_t)bigsize,
4467 VERIFY(0 == dmu_free_range(os, bigobj, bigoff, bigsize, tx));
4469 if (ztest_opts.zo_verbose >= 7) {
4470 (void) printf("writing offset %llx size %llx"
4472 (u_longlong_t)bigoff,
4473 (u_longlong_t)bigsize,
4476 dmu_write(os, bigobj, bigoff, bigsize, bigbuf, tx);
4482 * Sanity check the stuff we just wrote.
4485 void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
4486 void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
4488 VERIFY(0 == dmu_read(os, packobj, packoff,
4489 packsize, packcheck, DMU_READ_PREFETCH));
4490 VERIFY(0 == dmu_read(os, bigobj, bigoff,
4491 bigsize, bigcheck, DMU_READ_PREFETCH));
4493 ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
4494 ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
4496 umem_free(packcheck, packsize);
4497 umem_free(bigcheck, bigsize);
4500 umem_free(packbuf, packsize);
4501 umem_free(bigbuf, bigsize);
4502 umem_free(od, size);
4506 compare_and_update_pbbufs(uint64_t s, bufwad_t *packbuf, bufwad_t *bigbuf,
4507 uint64_t bigsize, uint64_t n, uint64_t chunksize, uint64_t txg)
4515 * For each index from n to n + s, verify that the existing bufwad
4516 * in packobj matches the bufwads at the head and tail of the
4517 * corresponding chunk in bigobj. Then update all three bufwads
4518 * with the new values we want to write out.
4520 for (i = 0; i < s; i++) {
4522 pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
4524 bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
4526 bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
4528 ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
4529 ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
4531 if (pack->bw_txg > txg)
4532 fatal(0, "future leak: got %llx, open txg is %llx",
4535 if (pack->bw_data != 0 && pack->bw_index != n + i)
4536 fatal(0, "wrong index: got %llx, wanted %llx+%llx",
4537 pack->bw_index, n, i);
4539 if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
4540 fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
4542 if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
4543 fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
4545 pack->bw_index = n + i;
4547 pack->bw_data = 1 + ztest_random(-2ULL);
4554 #undef OD_ARRAY_SIZE
4555 #define OD_ARRAY_SIZE 2
4558 ztest_dmu_read_write_zcopy(ztest_ds_t *zd, uint64_t id)
4560 objset_t *os = zd->zd_os;
4567 bufwad_t *packbuf, *bigbuf;
4568 uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
4569 uint64_t blocksize = ztest_random_blocksize();
4570 uint64_t chunksize = blocksize;
4571 uint64_t regions = 997;
4572 uint64_t stride = 123456789ULL;
4574 dmu_buf_t *bonus_db;
4575 arc_buf_t **bigbuf_arcbufs;
4576 dmu_object_info_t doi;
4578 size = sizeof (ztest_od_t) * OD_ARRAY_SIZE;
4579 od = umem_alloc(size, UMEM_NOFAIL);
4582 * This test uses two objects, packobj and bigobj, that are always
4583 * updated together (i.e. in the same tx) so that their contents are
4584 * in sync and can be compared. Their contents relate to each other
4585 * in a simple way: packobj is a dense array of 'bufwad' structures,
4586 * while bigobj is a sparse array of the same bufwads. Specifically,
4587 * for any index n, there are three bufwads that should be identical:
4589 * packobj, at offset n * sizeof (bufwad_t)
4590 * bigobj, at the head of the nth chunk
4591 * bigobj, at the tail of the nth chunk
4593 * The chunk size is set equal to bigobj block size so that
4594 * dmu_assign_arcbuf_by_dbuf() can be tested for object updates.
4598 * Read the directory info. If it's the first time, set things up.
4600 ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0, 0);
4601 ztest_od_init(od + 1, id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, 0,
4605 if (ztest_object_init(zd, od, size, B_FALSE) != 0) {
4606 umem_free(od, size);
4610 bigobj = od[0].od_object;
4611 packobj = od[1].od_object;
4612 blocksize = od[0].od_blocksize;
4613 chunksize = blocksize;
4614 ASSERT(chunksize == od[1].od_gen);
4616 VERIFY(dmu_object_info(os, bigobj, &doi) == 0);
4617 VERIFY(ISP2(doi.doi_data_block_size));
4618 VERIFY(chunksize == doi.doi_data_block_size);
4619 VERIFY(chunksize >= 2 * sizeof (bufwad_t));
4622 * Pick a random index and compute the offsets into packobj and bigobj.
4624 n = ztest_random(regions) * stride + ztest_random(width);
4625 s = 1 + ztest_random(width - 1);
4627 packoff = n * sizeof (bufwad_t);
4628 packsize = s * sizeof (bufwad_t);
4630 bigoff = n * chunksize;
4631 bigsize = s * chunksize;
4633 packbuf = umem_zalloc(packsize, UMEM_NOFAIL);
4634 bigbuf = umem_zalloc(bigsize, UMEM_NOFAIL);
4636 VERIFY3U(0, ==, dmu_bonus_hold(os, bigobj, FTAG, &bonus_db));
4638 bigbuf_arcbufs = umem_zalloc(2 * s * sizeof (arc_buf_t *), UMEM_NOFAIL);
4641 * Iteration 0 test zcopy for DB_UNCACHED dbufs.
4642 * Iteration 1 test zcopy to already referenced dbufs.
4643 * Iteration 2 test zcopy to dirty dbuf in the same txg.
4644 * Iteration 3 test zcopy to dbuf dirty in previous txg.
4645 * Iteration 4 test zcopy when dbuf is no longer dirty.
4646 * Iteration 5 test zcopy when it can't be done.
4647 * Iteration 6 one more zcopy write.
4649 for (i = 0; i < 7; i++) {
4654 * In iteration 5 (i == 5) use arcbufs
4655 * that don't match bigobj blksz to test
4656 * dmu_assign_arcbuf_by_dbuf() when it can't directly
4657 * assign an arcbuf to a dbuf.
4659 for (j = 0; j < s; j++) {
4660 if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) {
4662 dmu_request_arcbuf(bonus_db, chunksize);
4664 bigbuf_arcbufs[2 * j] =
4665 dmu_request_arcbuf(bonus_db, chunksize / 2);
4666 bigbuf_arcbufs[2 * j + 1] =
4667 dmu_request_arcbuf(bonus_db, chunksize / 2);
4672 * Get a tx for the mods to both packobj and bigobj.
4674 tx = dmu_tx_create(os);
4676 dmu_tx_hold_write(tx, packobj, packoff, packsize);
4677 dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
4679 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4681 umem_free(packbuf, packsize);
4682 umem_free(bigbuf, bigsize);
4683 for (j = 0; j < s; j++) {
4685 chunksize < (SPA_MINBLOCKSIZE * 2)) {
4686 dmu_return_arcbuf(bigbuf_arcbufs[j]);
4689 bigbuf_arcbufs[2 * j]);
4691 bigbuf_arcbufs[2 * j + 1]);
4694 umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
4695 umem_free(od, size);
4696 dmu_buf_rele(bonus_db, FTAG);
4701 * 50% of the time don't read objects in the 1st iteration to
4702 * test dmu_assign_arcbuf_by_dbuf() for the case when there are
4703 * no existing dbufs for the specified offsets.
4705 if (i != 0 || ztest_random(2) != 0) {
4706 error = dmu_read(os, packobj, packoff,
4707 packsize, packbuf, DMU_READ_PREFETCH);
4709 error = dmu_read(os, bigobj, bigoff, bigsize,
4710 bigbuf, DMU_READ_PREFETCH);
4713 compare_and_update_pbbufs(s, packbuf, bigbuf, bigsize,
4717 * We've verified all the old bufwads, and made new ones.
4718 * Now write them out.
4720 dmu_write(os, packobj, packoff, packsize, packbuf, tx);
4721 if (ztest_opts.zo_verbose >= 7) {
4722 (void) printf("writing offset %llx size %llx"
4724 (u_longlong_t)bigoff,
4725 (u_longlong_t)bigsize,
4728 for (off = bigoff, j = 0; j < s; j++, off += chunksize) {
4730 if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) {
4731 bcopy((caddr_t)bigbuf + (off - bigoff),
4732 bigbuf_arcbufs[j]->b_data, chunksize);
4734 bcopy((caddr_t)bigbuf + (off - bigoff),
4735 bigbuf_arcbufs[2 * j]->b_data,
4737 bcopy((caddr_t)bigbuf + (off - bigoff) +
4739 bigbuf_arcbufs[2 * j + 1]->b_data,
4744 VERIFY(dmu_buf_hold(os, bigobj, off,
4745 FTAG, &dbt, DMU_READ_NO_PREFETCH) == 0);
4747 if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) {
4748 dmu_assign_arcbuf_by_dbuf(bonus_db, off,
4749 bigbuf_arcbufs[j], tx);
4751 dmu_assign_arcbuf_by_dbuf(bonus_db, off,
4752 bigbuf_arcbufs[2 * j], tx);
4753 dmu_assign_arcbuf_by_dbuf(bonus_db,
4754 off + chunksize / 2,
4755 bigbuf_arcbufs[2 * j + 1], tx);
4758 dmu_buf_rele(dbt, FTAG);
4764 * Sanity check the stuff we just wrote.
4767 void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
4768 void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
4770 VERIFY(0 == dmu_read(os, packobj, packoff,
4771 packsize, packcheck, DMU_READ_PREFETCH));
4772 VERIFY(0 == dmu_read(os, bigobj, bigoff,
4773 bigsize, bigcheck, DMU_READ_PREFETCH));
4775 ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
4776 ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
4778 umem_free(packcheck, packsize);
4779 umem_free(bigcheck, bigsize);
4782 txg_wait_open(dmu_objset_pool(os), 0);
4783 } else if (i == 3) {
4784 txg_wait_synced(dmu_objset_pool(os), 0);
4788 dmu_buf_rele(bonus_db, FTAG);
4789 umem_free(packbuf, packsize);
4790 umem_free(bigbuf, bigsize);
4791 umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
4792 umem_free(od, size);
4797 ztest_dmu_write_parallel(ztest_ds_t *zd, uint64_t id)
4801 od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
4802 uint64_t offset = (1ULL << (ztest_random(20) + 43)) +
4803 (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
4806 * Have multiple threads write to large offsets in an object
4807 * to verify that parallel writes to an object -- even to the
4808 * same blocks within the object -- doesn't cause any trouble.
4810 ztest_od_init(od, ID_PARALLEL, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0, 0);
4812 if (ztest_object_init(zd, od, sizeof (ztest_od_t), B_FALSE) != 0)
4815 while (ztest_random(10) != 0)
4816 ztest_io(zd, od->od_object, offset);
4818 umem_free(od, sizeof (ztest_od_t));
4822 ztest_dmu_prealloc(ztest_ds_t *zd, uint64_t id)
4825 uint64_t offset = (1ULL << (ztest_random(4) + SPA_MAXBLOCKSHIFT)) +
4826 (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
4827 uint64_t count = ztest_random(20) + 1;
4828 uint64_t blocksize = ztest_random_blocksize();
4831 od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
4833 ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0, 0);
4835 if (ztest_object_init(zd, od, sizeof (ztest_od_t),
4836 !ztest_random(2)) != 0) {
4837 umem_free(od, sizeof (ztest_od_t));
4841 if (ztest_truncate(zd, od->od_object, offset, count * blocksize) != 0) {
4842 umem_free(od, sizeof (ztest_od_t));
4846 ztest_prealloc(zd, od->od_object, offset, count * blocksize);
4848 data = umem_zalloc(blocksize, UMEM_NOFAIL);
4850 while (ztest_random(count) != 0) {
4851 uint64_t randoff = offset + (ztest_random(count) * blocksize);
4852 if (ztest_write(zd, od->od_object, randoff, blocksize,
4855 while (ztest_random(4) != 0)
4856 ztest_io(zd, od->od_object, randoff);
4859 umem_free(data, blocksize);
4860 umem_free(od, sizeof (ztest_od_t));
4864 * Verify that zap_{create,destroy,add,remove,update} work as expected.
4866 #define ZTEST_ZAP_MIN_INTS 1
4867 #define ZTEST_ZAP_MAX_INTS 4
4868 #define ZTEST_ZAP_MAX_PROPS 1000
4871 ztest_zap(ztest_ds_t *zd, uint64_t id)
4873 objset_t *os = zd->zd_os;
4876 uint64_t txg, last_txg;
4877 uint64_t value[ZTEST_ZAP_MAX_INTS];
4878 uint64_t zl_ints, zl_intsize, prop;
4881 char propname[100], txgname[100];
4883 char *hc[2] = { "s.acl.h", ".s.open.h.hyLZlg" };
4885 od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
4886 ztest_od_init(od, id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0, 0);
4888 if (ztest_object_init(zd, od, sizeof (ztest_od_t),
4889 !ztest_random(2)) != 0)
4892 object = od->od_object;
4895 * Generate a known hash collision, and verify that
4896 * we can lookup and remove both entries.
4898 tx = dmu_tx_create(os);
4899 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4900 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4903 for (i = 0; i < 2; i++) {
4905 VERIFY3U(0, ==, zap_add(os, object, hc[i], sizeof (uint64_t),
4908 for (i = 0; i < 2; i++) {
4909 VERIFY3U(EEXIST, ==, zap_add(os, object, hc[i],
4910 sizeof (uint64_t), 1, &value[i], tx));
4912 zap_length(os, object, hc[i], &zl_intsize, &zl_ints));
4913 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4914 ASSERT3U(zl_ints, ==, 1);
4916 for (i = 0; i < 2; i++) {
4917 VERIFY3U(0, ==, zap_remove(os, object, hc[i], tx));
4922 * Generate a buch of random entries.
4924 ints = MAX(ZTEST_ZAP_MIN_INTS, object % ZTEST_ZAP_MAX_INTS);
4926 prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
4927 (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
4928 (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
4929 bzero(value, sizeof (value));
4933 * If these zap entries already exist, validate their contents.
4935 error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
4937 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4938 ASSERT3U(zl_ints, ==, 1);
4940 VERIFY(zap_lookup(os, object, txgname, zl_intsize,
4941 zl_ints, &last_txg) == 0);
4943 VERIFY(zap_length(os, object, propname, &zl_intsize,
4946 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4947 ASSERT3U(zl_ints, ==, ints);
4949 VERIFY(zap_lookup(os, object, propname, zl_intsize,
4950 zl_ints, value) == 0);
4952 for (i = 0; i < ints; i++) {
4953 ASSERT3U(value[i], ==, last_txg + object + i);
4956 ASSERT3U(error, ==, ENOENT);
4960 * Atomically update two entries in our zap object.
4961 * The first is named txg_%llu, and contains the txg
4962 * in which the property was last updated. The second
4963 * is named prop_%llu, and the nth element of its value
4964 * should be txg + object + n.
4966 tx = dmu_tx_create(os);
4967 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4968 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4973 fatal(0, "zap future leak: old %llu new %llu", last_txg, txg);
4975 for (i = 0; i < ints; i++)
4976 value[i] = txg + object + i;
4978 VERIFY3U(0, ==, zap_update(os, object, txgname, sizeof (uint64_t),
4980 VERIFY3U(0, ==, zap_update(os, object, propname, sizeof (uint64_t),
4986 * Remove a random pair of entries.
4988 prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
4989 (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
4990 (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
4992 error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
4994 if (error == ENOENT)
4999 tx = dmu_tx_create(os);
5000 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
5001 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
5004 VERIFY3U(0, ==, zap_remove(os, object, txgname, tx));
5005 VERIFY3U(0, ==, zap_remove(os, object, propname, tx));
5008 umem_free(od, sizeof (ztest_od_t));
5012 * Testcase to test the upgrading of a microzap to fatzap.
5015 ztest_fzap(ztest_ds_t *zd, uint64_t id)
5017 objset_t *os = zd->zd_os;
5019 uint64_t object, txg;
5022 od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
5023 ztest_od_init(od, id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0, 0);
5025 if (ztest_object_init(zd, od, sizeof (ztest_od_t),
5026 !ztest_random(2)) != 0)
5028 object = od->od_object;
5031 * Add entries to this ZAP and make sure it spills over
5032 * and gets upgraded to a fatzap. Also, since we are adding
5033 * 2050 entries we should see ptrtbl growth and leaf-block split.
5035 for (i = 0; i < 2050; i++) {
5036 char name[ZFS_MAX_DATASET_NAME_LEN];
5041 (void) snprintf(name, sizeof (name), "fzap-%llu-%llu",
5042 (u_longlong_t)id, (u_longlong_t)value);
5044 tx = dmu_tx_create(os);
5045 dmu_tx_hold_zap(tx, object, B_TRUE, name);
5046 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
5049 error = zap_add(os, object, name, sizeof (uint64_t), 1,
5051 ASSERT(error == 0 || error == EEXIST);
5055 umem_free(od, sizeof (ztest_od_t));
5060 ztest_zap_parallel(ztest_ds_t *zd, uint64_t id)
5062 objset_t *os = zd->zd_os;
5064 uint64_t txg, object, count, wsize, wc, zl_wsize, zl_wc;
5066 int i, namelen, error;
5067 int micro = ztest_random(2);
5068 char name[20], string_value[20];
5071 od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
5072 ztest_od_init(od, ID_PARALLEL, FTAG, micro, DMU_OT_ZAP_OTHER, 0, 0, 0);
5074 if (ztest_object_init(zd, od, sizeof (ztest_od_t), B_FALSE) != 0) {
5075 umem_free(od, sizeof (ztest_od_t));
5079 object = od->od_object;
5082 * Generate a random name of the form 'xxx.....' where each
5083 * x is a random printable character and the dots are dots.
5084 * There are 94 such characters, and the name length goes from
5085 * 6 to 20, so there are 94^3 * 15 = 12,458,760 possible names.
5087 namelen = ztest_random(sizeof (name) - 5) + 5 + 1;
5089 for (i = 0; i < 3; i++)
5090 name[i] = '!' + ztest_random('~' - '!' + 1);
5091 for (; i < namelen - 1; i++)
5095 if ((namelen & 1) || micro) {
5096 wsize = sizeof (txg);
5102 data = string_value;
5106 VERIFY0(zap_count(os, object, &count));
5107 ASSERT(count != -1ULL);
5110 * Select an operation: length, lookup, add, update, remove.
5112 i = ztest_random(5);
5115 tx = dmu_tx_create(os);
5116 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
5117 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
5119 umem_free(od, sizeof (ztest_od_t));
5122 bcopy(name, string_value, namelen);
5126 bzero(string_value, namelen);
5132 error = zap_length(os, object, name, &zl_wsize, &zl_wc);
5134 ASSERT3U(wsize, ==, zl_wsize);
5135 ASSERT3U(wc, ==, zl_wc);
5137 ASSERT3U(error, ==, ENOENT);
5142 error = zap_lookup(os, object, name, wsize, wc, data);
5144 if (data == string_value &&
5145 bcmp(name, data, namelen) != 0)
5146 fatal(0, "name '%s' != val '%s' len %d",
5147 name, data, namelen);
5149 ASSERT3U(error, ==, ENOENT);
5154 error = zap_add(os, object, name, wsize, wc, data, tx);
5155 ASSERT(error == 0 || error == EEXIST);
5159 VERIFY(zap_update(os, object, name, wsize, wc, data, tx) == 0);
5163 error = zap_remove(os, object, name, tx);
5164 ASSERT(error == 0 || error == ENOENT);
5171 umem_free(od, sizeof (ztest_od_t));
5175 * Commit callback data.
5177 typedef struct ztest_cb_data {
5178 list_node_t zcd_node;
5180 int zcd_expected_err;
5181 boolean_t zcd_added;
5182 boolean_t zcd_called;
5186 /* This is the actual commit callback function */
5188 ztest_commit_callback(void *arg, int error)
5190 ztest_cb_data_t *data = arg;
5191 uint64_t synced_txg;
5193 VERIFY(data != NULL);
5194 VERIFY3S(data->zcd_expected_err, ==, error);
5195 VERIFY(!data->zcd_called);
5197 synced_txg = spa_last_synced_txg(data->zcd_spa);
5198 if (data->zcd_txg > synced_txg)
5199 fatal(0, "commit callback of txg %" PRIu64 " called prematurely"
5200 ", last synced txg = %" PRIu64 "\n", data->zcd_txg,
5203 data->zcd_called = B_TRUE;
5205 if (error == ECANCELED) {
5206 ASSERT0(data->zcd_txg);
5207 ASSERT(!data->zcd_added);
5210 * The private callback data should be destroyed here, but
5211 * since we are going to check the zcd_called field after
5212 * dmu_tx_abort(), we will destroy it there.
5217 ASSERT(data->zcd_added);
5218 ASSERT3U(data->zcd_txg, !=, 0);
5220 (void) mutex_enter(&zcl.zcl_callbacks_lock);
5222 /* See if this cb was called more quickly */
5223 if ((synced_txg - data->zcd_txg) < zc_min_txg_delay)
5224 zc_min_txg_delay = synced_txg - data->zcd_txg;
5226 /* Remove our callback from the list */
5227 list_remove(&zcl.zcl_callbacks, data);
5229 (void) mutex_exit(&zcl.zcl_callbacks_lock);
5231 umem_free(data, sizeof (ztest_cb_data_t));
5234 /* Allocate and initialize callback data structure */
5235 static ztest_cb_data_t *
5236 ztest_create_cb_data(objset_t *os, uint64_t txg)
5238 ztest_cb_data_t *cb_data;
5240 cb_data = umem_zalloc(sizeof (ztest_cb_data_t), UMEM_NOFAIL);
5242 cb_data->zcd_txg = txg;
5243 cb_data->zcd_spa = dmu_objset_spa(os);
5244 list_link_init(&cb_data->zcd_node);
5250 * Commit callback test.
5253 ztest_dmu_commit_callbacks(ztest_ds_t *zd, uint64_t id)
5255 objset_t *os = zd->zd_os;
5258 ztest_cb_data_t *cb_data[3], *tmp_cb;
5259 uint64_t old_txg, txg;
5262 od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
5263 ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0, 0);
5265 if (ztest_object_init(zd, od, sizeof (ztest_od_t), B_FALSE) != 0) {
5266 umem_free(od, sizeof (ztest_od_t));
5270 tx = dmu_tx_create(os);
5272 cb_data[0] = ztest_create_cb_data(os, 0);
5273 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[0]);
5275 dmu_tx_hold_write(tx, od->od_object, 0, sizeof (uint64_t));
5277 /* Every once in a while, abort the transaction on purpose */
5278 if (ztest_random(100) == 0)
5282 error = dmu_tx_assign(tx, TXG_NOWAIT);
5284 txg = error ? 0 : dmu_tx_get_txg(tx);
5286 cb_data[0]->zcd_txg = txg;
5287 cb_data[1] = ztest_create_cb_data(os, txg);
5288 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[1]);
5292 * It's not a strict requirement to call the registered
5293 * callbacks from inside dmu_tx_abort(), but that's what
5294 * it's supposed to happen in the current implementation
5295 * so we will check for that.
5297 for (i = 0; i < 2; i++) {
5298 cb_data[i]->zcd_expected_err = ECANCELED;
5299 VERIFY(!cb_data[i]->zcd_called);
5304 for (i = 0; i < 2; i++) {
5305 VERIFY(cb_data[i]->zcd_called);
5306 umem_free(cb_data[i], sizeof (ztest_cb_data_t));
5309 umem_free(od, sizeof (ztest_od_t));
5313 cb_data[2] = ztest_create_cb_data(os, txg);
5314 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[2]);
5317 * Read existing data to make sure there isn't a future leak.
5319 VERIFY(0 == dmu_read(os, od->od_object, 0, sizeof (uint64_t),
5320 &old_txg, DMU_READ_PREFETCH));
5323 fatal(0, "future leak: got %" PRIu64 ", open txg is %" PRIu64,
5326 dmu_write(os, od->od_object, 0, sizeof (uint64_t), &txg, tx);
5328 (void) mutex_enter(&zcl.zcl_callbacks_lock);
5331 * Since commit callbacks don't have any ordering requirement and since
5332 * it is theoretically possible for a commit callback to be called
5333 * after an arbitrary amount of time has elapsed since its txg has been
5334 * synced, it is difficult to reliably determine whether a commit
5335 * callback hasn't been called due to high load or due to a flawed
5338 * In practice, we will assume that if after a certain number of txgs a
5339 * commit callback hasn't been called, then most likely there's an
5340 * implementation bug..
5342 tmp_cb = list_head(&zcl.zcl_callbacks);
5343 if (tmp_cb != NULL &&
5344 tmp_cb->zcd_txg + ZTEST_COMMIT_CB_THRESH < txg) {
5345 fatal(0, "Commit callback threshold exceeded, oldest txg: %"
5346 PRIu64 ", open txg: %" PRIu64 "\n", tmp_cb->zcd_txg, txg);
5350 * Let's find the place to insert our callbacks.
5352 * Even though the list is ordered by txg, it is possible for the
5353 * insertion point to not be the end because our txg may already be
5354 * quiescing at this point and other callbacks in the open txg
5355 * (from other objsets) may have sneaked in.
5357 tmp_cb = list_tail(&zcl.zcl_callbacks);
5358 while (tmp_cb != NULL && tmp_cb->zcd_txg > txg)
5359 tmp_cb = list_prev(&zcl.zcl_callbacks, tmp_cb);
5361 /* Add the 3 callbacks to the list */
5362 for (i = 0; i < 3; i++) {
5364 list_insert_head(&zcl.zcl_callbacks, cb_data[i]);
5366 list_insert_after(&zcl.zcl_callbacks, tmp_cb,
5369 cb_data[i]->zcd_added = B_TRUE;
5370 VERIFY(!cb_data[i]->zcd_called);
5372 tmp_cb = cb_data[i];
5377 (void) mutex_exit(&zcl.zcl_callbacks_lock);
5381 umem_free(od, sizeof (ztest_od_t));
5385 * Visit each object in the dataset. Verify that its properties
5386 * are consistent what was stored in the block tag when it was created,
5387 * and that its unused bonus buffer space has not been overwritten.
5391 ztest_verify_dnode_bt(ztest_ds_t *zd, uint64_t id)
5393 objset_t *os = zd->zd_os;
5397 for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, 0)) {
5398 ztest_block_tag_t *bt = NULL;
5399 dmu_object_info_t doi;
5402 ztest_object_lock(zd, obj, RL_READER);
5403 if (dmu_bonus_hold(os, obj, FTAG, &db) != 0) {
5404 ztest_object_unlock(zd, obj);
5408 dmu_object_info_from_db(db, &doi);
5409 if (doi.doi_bonus_size >= sizeof (*bt))
5410 bt = ztest_bt_bonus(db);
5412 if (bt && bt->bt_magic == BT_MAGIC) {
5413 ztest_bt_verify(bt, os, obj, doi.doi_dnodesize,
5414 bt->bt_offset, bt->bt_gen, bt->bt_txg,
5416 ztest_verify_unused_bonus(db, bt, obj, os, bt->bt_gen);
5419 dmu_buf_rele(db, FTAG);
5420 ztest_object_unlock(zd, obj);
5426 ztest_dsl_prop_get_set(ztest_ds_t *zd, uint64_t id)
5428 zfs_prop_t proplist[] = {
5430 ZFS_PROP_COMPRESSION,
5436 (void) pthread_rwlock_rdlock(&ztest_name_lock);
5438 for (p = 0; p < sizeof (proplist) / sizeof (proplist[0]); p++)
5439 (void) ztest_dsl_prop_set_uint64(zd->zd_name, proplist[p],
5440 ztest_random_dsl_prop(proplist[p]), (int)ztest_random(2));
5442 VERIFY0(ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_RECORDSIZE,
5443 ztest_random_blocksize(), (int)ztest_random(2)));
5445 (void) pthread_rwlock_unlock(&ztest_name_lock);
5450 ztest_remap_blocks(ztest_ds_t *zd, uint64_t id)
5452 (void) pthread_rwlock_rdlock(&ztest_name_lock);
5454 int error = dmu_objset_remap_indirects(zd->zd_name);
5455 if (error == ENOSPC)
5459 (void) pthread_rwlock_unlock(&ztest_name_lock);
5464 ztest_spa_prop_get_set(ztest_ds_t *zd, uint64_t id)
5466 nvlist_t *props = NULL;
5468 (void) pthread_rwlock_rdlock(&ztest_name_lock);
5470 (void) ztest_spa_prop_set_uint64(ZPOOL_PROP_DEDUPDITTO,
5471 ZIO_DEDUPDITTO_MIN + ztest_random(ZIO_DEDUPDITTO_MIN));
5473 VERIFY0(spa_prop_get(ztest_spa, &props));
5475 if (ztest_opts.zo_verbose >= 6)
5476 dump_nvlist(props, 4);
5480 (void) pthread_rwlock_unlock(&ztest_name_lock);
5484 user_release_one(const char *snapname, const char *holdname)
5486 nvlist_t *snaps, *holds;
5489 snaps = fnvlist_alloc();
5490 holds = fnvlist_alloc();
5491 fnvlist_add_boolean(holds, holdname);
5492 fnvlist_add_nvlist(snaps, snapname, holds);
5493 fnvlist_free(holds);
5494 error = dsl_dataset_user_release(snaps, NULL);
5495 fnvlist_free(snaps);
5500 * Test snapshot hold/release and deferred destroy.
5503 ztest_dmu_snapshot_hold(ztest_ds_t *zd, uint64_t id)
5506 objset_t *os = zd->zd_os;
5510 char clonename[100];
5512 char osname[ZFS_MAX_DATASET_NAME_LEN];
5515 (void) pthread_rwlock_rdlock(&ztest_name_lock);
5517 dmu_objset_name(os, osname);
5519 (void) snprintf(snapname, sizeof (snapname), "sh1_%llu",
5521 (void) snprintf(fullname, sizeof (fullname), "%s@%s", osname, snapname);
5522 (void) snprintf(clonename, sizeof (clonename),
5523 "%s/ch1_%llu", osname, (u_longlong_t)id);
5524 (void) snprintf(tag, sizeof (tag), "tag_%llu", (u_longlong_t)id);
5527 * Clean up from any previous run.
5529 error = dsl_destroy_head(clonename);
5530 if (error != ENOENT)
5532 error = user_release_one(fullname, tag);
5533 if (error != ESRCH && error != ENOENT)
5535 error = dsl_destroy_snapshot(fullname, B_FALSE);
5536 if (error != ENOENT)
5540 * Create snapshot, clone it, mark snap for deferred destroy,
5541 * destroy clone, verify snap was also destroyed.
5543 error = dmu_objset_snapshot_one(osname, snapname);
5545 if (error == ENOSPC) {
5546 ztest_record_enospc("dmu_objset_snapshot");
5549 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
5552 error = dmu_objset_clone(clonename, fullname);
5554 if (error == ENOSPC) {
5555 ztest_record_enospc("dmu_objset_clone");
5558 fatal(0, "dmu_objset_clone(%s) = %d", clonename, error);
5561 error = dsl_destroy_snapshot(fullname, B_TRUE);
5563 fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
5567 error = dsl_destroy_head(clonename);
5569 fatal(0, "dsl_destroy_head(%s) = %d", clonename, error);
5571 error = dmu_objset_hold(fullname, FTAG, &origin);
5572 if (error != ENOENT)
5573 fatal(0, "dmu_objset_hold(%s) = %d", fullname, error);
5576 * Create snapshot, add temporary hold, verify that we can't
5577 * destroy a held snapshot, mark for deferred destroy,
5578 * release hold, verify snapshot was destroyed.
5580 error = dmu_objset_snapshot_one(osname, snapname);
5582 if (error == ENOSPC) {
5583 ztest_record_enospc("dmu_objset_snapshot");
5586 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
5589 holds = fnvlist_alloc();
5590 fnvlist_add_string(holds, fullname, tag);
5591 error = dsl_dataset_user_hold(holds, 0, NULL);
5592 fnvlist_free(holds);
5594 if (error == ENOSPC) {
5595 ztest_record_enospc("dsl_dataset_user_hold");
5598 fatal(0, "dsl_dataset_user_hold(%s, %s) = %u",
5599 fullname, tag, error);
5602 error = dsl_destroy_snapshot(fullname, B_FALSE);
5603 if (error != EBUSY) {
5604 fatal(0, "dsl_destroy_snapshot(%s, B_FALSE) = %d",
5608 error = dsl_destroy_snapshot(fullname, B_TRUE);
5610 fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
5614 error = user_release_one(fullname, tag);
5616 fatal(0, "user_release_one(%s, %s) = %d", fullname, tag, error);
5618 VERIFY3U(dmu_objset_hold(fullname, FTAG, &origin), ==, ENOENT);
5621 (void) pthread_rwlock_unlock(&ztest_name_lock);
5625 * Inject random faults into the on-disk data.
5629 ztest_fault_inject(ztest_ds_t *zd, uint64_t id)
5631 ztest_shared_t *zs = ztest_shared;
5632 spa_t *spa = ztest_spa;
5636 uint64_t bad = 0x1990c0ffeedecadeull;
5641 int bshift = SPA_MAXBLOCKSHIFT + 2;
5647 boolean_t islog = B_FALSE;
5649 path0 = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
5650 pathrand = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
5652 mutex_enter(&ztest_vdev_lock);
5655 * Device removal is in progress, fault injection must be disabled
5656 * until it completes and the pool is scrubbed. The fault injection
5657 * strategy for damaging blocks does not take in to account evacuated
5658 * blocks which may have already been damaged.
5660 if (ztest_device_removal_active) {
5661 mutex_exit(&ztest_vdev_lock);
5665 maxfaults = MAXFAULTS(zs);
5666 leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz;
5667 mirror_save = zs->zs_mirrors;
5668 mutex_exit(&ztest_vdev_lock);
5670 ASSERT(leaves >= 1);
5673 * Grab the name lock as reader. There are some operations
5674 * which don't like to have their vdevs changed while
5675 * they are in progress (i.e. spa_change_guid). Those
5676 * operations will have grabbed the name lock as writer.
5678 (void) pthread_rwlock_rdlock(&ztest_name_lock);
5681 * We need SCL_STATE here because we're going to look at vd0->vdev_tsd.
5683 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5685 if (ztest_random(2) == 0) {
5687 * Inject errors on a normal data device or slog device.
5689 top = ztest_random_vdev_top(spa, B_TRUE);
5690 leaf = ztest_random(leaves) + zs->zs_splits;
5693 * Generate paths to the first leaf in this top-level vdev,
5694 * and to the random leaf we selected. We'll induce transient
5695 * write failures and random online/offline activity on leaf 0,
5696 * and we'll write random garbage to the randomly chosen leaf.
5698 (void) snprintf(path0, MAXPATHLEN, ztest_dev_template,
5699 ztest_opts.zo_dir, ztest_opts.zo_pool,
5700 top * leaves + zs->zs_splits);
5701 (void) snprintf(pathrand, MAXPATHLEN, ztest_dev_template,
5702 ztest_opts.zo_dir, ztest_opts.zo_pool,
5703 top * leaves + leaf);
5705 vd0 = vdev_lookup_by_path(spa->spa_root_vdev, path0);
5706 if (vd0 != NULL && vd0->vdev_top->vdev_islog)
5710 * If the top-level vdev needs to be resilvered
5711 * then we only allow faults on the device that is
5714 if (vd0 != NULL && maxfaults != 1 &&
5715 (!vdev_resilver_needed(vd0->vdev_top, NULL, NULL) ||
5716 vd0->vdev_resilver_txg != 0)) {
5718 * Make vd0 explicitly claim to be unreadable,
5719 * or unwriteable, or reach behind its back
5720 * and close the underlying fd. We can do this if
5721 * maxfaults == 0 because we'll fail and reexecute,
5722 * and we can do it if maxfaults >= 2 because we'll
5723 * have enough redundancy. If maxfaults == 1, the
5724 * combination of this with injection of random data
5725 * corruption below exceeds the pool's fault tolerance.
5727 vdev_file_t *vf = vd0->vdev_tsd;
5729 zfs_dbgmsg("injecting fault to vdev %llu; maxfaults=%d",
5730 (long long)vd0->vdev_id, (int)maxfaults);
5732 if (vf != NULL && ztest_random(3) == 0) {
5733 (void) close(vf->vf_vnode->v_fd);
5734 vf->vf_vnode->v_fd = -1;
5735 } else if (ztest_random(2) == 0) {
5736 vd0->vdev_cant_read = B_TRUE;
5738 vd0->vdev_cant_write = B_TRUE;
5740 guid0 = vd0->vdev_guid;
5744 * Inject errors on an l2cache device.
5746 spa_aux_vdev_t *sav = &spa->spa_l2cache;
5748 if (sav->sav_count == 0) {
5749 spa_config_exit(spa, SCL_STATE, FTAG);
5750 (void) pthread_rwlock_unlock(&ztest_name_lock);
5753 vd0 = sav->sav_vdevs[ztest_random(sav->sav_count)];
5754 guid0 = vd0->vdev_guid;
5755 (void) strcpy(path0, vd0->vdev_path);
5756 (void) strcpy(pathrand, vd0->vdev_path);
5760 maxfaults = INT_MAX; /* no limit on cache devices */
5763 spa_config_exit(spa, SCL_STATE, FTAG);
5764 (void) pthread_rwlock_unlock(&ztest_name_lock);
5767 * If we can tolerate two or more faults, or we're dealing
5768 * with a slog, randomly online/offline vd0.
5770 if ((maxfaults >= 2 || islog) && guid0 != 0) {
5771 if (ztest_random(10) < 6) {
5772 int flags = (ztest_random(2) == 0 ?
5773 ZFS_OFFLINE_TEMPORARY : 0);
5776 * We have to grab the zs_name_lock as writer to
5777 * prevent a race between offlining a slog and
5778 * destroying a dataset. Offlining the slog will
5779 * grab a reference on the dataset which may cause
5780 * dsl_destroy_head() to fail with EBUSY thus
5781 * leaving the dataset in an inconsistent state.
5784 (void) pthread_rwlock_wrlock(&ztest_name_lock);
5786 VERIFY(vdev_offline(spa, guid0, flags) != EBUSY);
5789 (void) pthread_rwlock_unlock(&ztest_name_lock);
5792 * Ideally we would like to be able to randomly
5793 * call vdev_[on|off]line without holding locks
5794 * to force unpredictable failures but the side
5795 * effects of vdev_[on|off]line prevent us from
5796 * doing so. We grab the ztest_vdev_lock here to
5797 * prevent a race between injection testing and
5800 mutex_enter(&ztest_vdev_lock);
5801 (void) vdev_online(spa, guid0, 0, NULL);
5802 mutex_exit(&ztest_vdev_lock);
5810 * We have at least single-fault tolerance, so inject data corruption.
5812 fd = open(pathrand, O_RDWR);
5814 if (fd == -1) /* we hit a gap in the device namespace */
5817 fsize = lseek(fd, 0, SEEK_END);
5819 while (--iters != 0) {
5821 * The offset must be chosen carefully to ensure that
5822 * we do not inject a given logical block with errors
5823 * on two different leaf devices, because ZFS can not
5824 * tolerate that (if maxfaults==1).
5826 * We divide each leaf into chunks of size
5827 * (# leaves * SPA_MAXBLOCKSIZE * 4). Within each chunk
5828 * there is a series of ranges to which we can inject errors.
5829 * Each range can accept errors on only a single leaf vdev.
5830 * The error injection ranges are separated by ranges
5831 * which we will not inject errors on any device (DMZs).
5832 * Each DMZ must be large enough such that a single block
5833 * can not straddle it, so that a single block can not be
5834 * a target in two different injection ranges (on different
5837 * For example, with 3 leaves, each chunk looks like:
5838 * 0 to 32M: injection range for leaf 0
5839 * 32M to 64M: DMZ - no injection allowed
5840 * 64M to 96M: injection range for leaf 1
5841 * 96M to 128M: DMZ - no injection allowed
5842 * 128M to 160M: injection range for leaf 2
5843 * 160M to 192M: DMZ - no injection allowed
5845 offset = ztest_random(fsize / (leaves << bshift)) *
5846 (leaves << bshift) + (leaf << bshift) +
5847 (ztest_random(1ULL << (bshift - 1)) & -8ULL);
5850 * Only allow damage to the labels at one end of the vdev.
5852 * If all labels are damaged, the device will be totally
5853 * inaccessible, which will result in loss of data,
5854 * because we also damage (parts of) the other side of
5857 * Additionally, we will always have both an even and an
5858 * odd label, so that we can handle crashes in the
5859 * middle of vdev_config_sync().
5861 if ((leaf & 1) == 0 && offset < VDEV_LABEL_START_SIZE)
5865 * The two end labels are stored at the "end" of the disk, but
5866 * the end of the disk (vdev_psize) is aligned to
5867 * sizeof (vdev_label_t).
5869 uint64_t psize = P2ALIGN(fsize, sizeof (vdev_label_t));
5870 if ((leaf & 1) == 1 &&
5871 offset + sizeof (bad) > psize - VDEV_LABEL_END_SIZE)
5874 mutex_enter(&ztest_vdev_lock);
5875 if (mirror_save != zs->zs_mirrors) {
5876 mutex_exit(&ztest_vdev_lock);
5881 if (pwrite(fd, &bad, sizeof (bad), offset) != sizeof (bad))
5882 fatal(1, "can't inject bad word at 0x%llx in %s",
5885 mutex_exit(&ztest_vdev_lock);
5887 if (ztest_opts.zo_verbose >= 7)
5888 (void) printf("injected bad word into %s,"
5889 " offset 0x%llx\n", pathrand, (u_longlong_t)offset);
5894 umem_free(path0, MAXPATHLEN);
5895 umem_free(pathrand, MAXPATHLEN);
5899 * Verify that DDT repair works as expected.
5902 ztest_ddt_repair(ztest_ds_t *zd, uint64_t id)
5904 ztest_shared_t *zs = ztest_shared;
5905 spa_t *spa = ztest_spa;
5906 objset_t *os = zd->zd_os;
5908 uint64_t object, blocksize, txg, pattern, psize;
5909 enum zio_checksum checksum = spa_dedup_checksum(spa);
5914 int copies = 2 * ZIO_DEDUPDITTO_MIN;
5917 blocksize = ztest_random_blocksize();
5918 blocksize = MIN(blocksize, 2048); /* because we write so many */
5920 od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
5921 ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0, 0);
5923 if (ztest_object_init(zd, od, sizeof (ztest_od_t), B_FALSE) != 0) {
5924 umem_free(od, sizeof (ztest_od_t));
5929 * Take the name lock as writer to prevent anyone else from changing
5930 * the pool and dataset properies we need to maintain during this test.
5932 (void) pthread_rwlock_wrlock(&ztest_name_lock);
5934 if (ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_DEDUP, checksum,
5936 ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_COPIES, 1,
5938 (void) pthread_rwlock_unlock(&ztest_name_lock);
5939 umem_free(od, sizeof (ztest_od_t));
5943 dmu_objset_stats_t dds;
5944 dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
5945 dmu_objset_fast_stat(os, &dds);
5946 dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
5948 object = od[0].od_object;
5949 blocksize = od[0].od_blocksize;
5950 pattern = zs->zs_guid ^ dds.dds_guid;
5952 ASSERT(object != 0);
5954 tx = dmu_tx_create(os);
5955 dmu_tx_hold_write(tx, object, 0, copies * blocksize);
5956 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
5958 (void) pthread_rwlock_unlock(&ztest_name_lock);
5959 umem_free(od, sizeof (ztest_od_t));
5964 * Write all the copies of our block.
5966 for (i = 0; i < copies; i++) {
5967 uint64_t offset = i * blocksize;
5968 int error = dmu_buf_hold(os, object, offset, FTAG, &db,
5969 DMU_READ_NO_PREFETCH);
5971 fatal(B_FALSE, "dmu_buf_hold(%p, %llu, %llu) = %u",
5972 os, (long long)object, (long long) offset, error);
5974 ASSERT(db->db_offset == offset);
5975 ASSERT(db->db_size == blocksize);
5976 ASSERT(ztest_pattern_match(db->db_data, db->db_size, pattern) ||
5977 ztest_pattern_match(db->db_data, db->db_size, 0ULL));
5978 dmu_buf_will_fill(db, tx);
5979 ztest_pattern_set(db->db_data, db->db_size, pattern);
5980 dmu_buf_rele(db, FTAG);
5984 txg_wait_synced(spa_get_dsl(spa), txg);
5987 * Find out what block we got.
5989 VERIFY0(dmu_buf_hold(os, object, 0, FTAG, &db,
5990 DMU_READ_NO_PREFETCH));
5991 blk = *((dmu_buf_impl_t *)db)->db_blkptr;
5992 dmu_buf_rele(db, FTAG);
5995 * Damage the block. Dedup-ditto will save us when we read it later.
5997 psize = BP_GET_PSIZE(&blk);
5998 abd = abd_alloc_linear(psize, B_TRUE);
5999 ztest_pattern_set(abd_to_buf(abd), psize, ~pattern);
6001 (void) zio_wait(zio_rewrite(NULL, spa, 0, &blk,
6002 abd, psize, NULL, NULL, ZIO_PRIORITY_SYNC_WRITE,
6003 ZIO_FLAG_CANFAIL | ZIO_FLAG_INDUCE_DAMAGE, NULL));
6007 (void) pthread_rwlock_unlock(&ztest_name_lock);
6008 umem_free(od, sizeof (ztest_od_t));
6016 ztest_scrub(ztest_ds_t *zd, uint64_t id)
6018 spa_t *spa = ztest_spa;
6021 * Scrub in progress by device removal.
6023 if (ztest_device_removal_active)
6026 (void) spa_scan(spa, POOL_SCAN_SCRUB);
6027 (void) poll(NULL, 0, 100); /* wait a moment, then force a restart */
6028 (void) spa_scan(spa, POOL_SCAN_SCRUB);
6032 * Change the guid for the pool.
6036 ztest_reguid(ztest_ds_t *zd, uint64_t id)
6038 spa_t *spa = ztest_spa;
6039 uint64_t orig, load;
6042 if (ztest_opts.zo_mmp_test)
6045 orig = spa_guid(spa);
6046 load = spa_load_guid(spa);
6048 (void) pthread_rwlock_wrlock(&ztest_name_lock);
6049 error = spa_change_guid(spa);
6050 (void) pthread_rwlock_unlock(&ztest_name_lock);
6055 if (ztest_opts.zo_verbose >= 4) {
6056 (void) printf("Changed guid old %llu -> %llu\n",
6057 (u_longlong_t)orig, (u_longlong_t)spa_guid(spa));
6060 VERIFY3U(orig, !=, spa_guid(spa));
6061 VERIFY3U(load, ==, spa_load_guid(spa));
6065 * Rename the pool to a different name and then rename it back.
6069 ztest_spa_rename(ztest_ds_t *zd, uint64_t id)
6071 char *oldname, *newname;
6074 if (ztest_opts.zo_mmp_test)
6077 (void) pthread_rwlock_wrlock(&ztest_name_lock);
6079 oldname = ztest_opts.zo_pool;
6080 newname = umem_alloc(strlen(oldname) + 5, UMEM_NOFAIL);
6081 (void) strcpy(newname, oldname);
6082 (void) strcat(newname, "_tmp");
6087 VERIFY3U(0, ==, spa_rename(oldname, newname));
6090 * Try to open it under the old name, which shouldn't exist
6092 VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
6095 * Open it under the new name and make sure it's still the same spa_t.
6097 VERIFY3U(0, ==, spa_open(newname, &spa, FTAG));
6099 ASSERT(spa == ztest_spa);
6100 spa_close(spa, FTAG);
6103 * Rename it back to the original
6105 VERIFY3U(0, ==, spa_rename(newname, oldname));
6108 * Make sure it can still be opened
6110 VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG));
6112 ASSERT(spa == ztest_spa);
6113 spa_close(spa, FTAG);
6115 umem_free(newname, strlen(newname) + 1);
6117 (void) pthread_rwlock_unlock(&ztest_name_lock);
6121 ztest_fletcher(ztest_ds_t *zd, uint64_t id)
6123 hrtime_t end = gethrtime() + NANOSEC;
6125 while (gethrtime() <= end) {
6126 int run_count = 100;
6128 struct abd *abd_data, *abd_meta;
6133 zio_cksum_t zc_ref_byteswap;
6135 size = ztest_random_blocksize();
6137 buf = umem_alloc(size, UMEM_NOFAIL);
6138 abd_data = abd_alloc(size, B_FALSE);
6139 abd_meta = abd_alloc(size, B_TRUE);
6141 for (i = 0, ptr = buf; i < size / sizeof (*ptr); i++, ptr++)
6142 *ptr = ztest_random(UINT_MAX);
6144 abd_copy_from_buf_off(abd_data, buf, 0, size);
6145 abd_copy_from_buf_off(abd_meta, buf, 0, size);
6147 VERIFY0(fletcher_4_impl_set("scalar"));
6148 fletcher_4_native(buf, size, NULL, &zc_ref);
6149 fletcher_4_byteswap(buf, size, NULL, &zc_ref_byteswap);
6151 VERIFY0(fletcher_4_impl_set("cycle"));
6152 while (run_count-- > 0) {
6154 zio_cksum_t zc_byteswap;
6156 fletcher_4_byteswap(buf, size, NULL, &zc_byteswap);
6157 fletcher_4_native(buf, size, NULL, &zc);
6159 VERIFY0(bcmp(&zc, &zc_ref, sizeof (zc)));
6160 VERIFY0(bcmp(&zc_byteswap, &zc_ref_byteswap,
6161 sizeof (zc_byteswap)));
6163 /* Test ABD - data */
6164 abd_fletcher_4_byteswap(abd_data, size, NULL,
6166 abd_fletcher_4_native(abd_data, size, NULL, &zc);
6168 VERIFY0(bcmp(&zc, &zc_ref, sizeof (zc)));
6169 VERIFY0(bcmp(&zc_byteswap, &zc_ref_byteswap,
6170 sizeof (zc_byteswap)));
6172 /* Test ABD - metadata */
6173 abd_fletcher_4_byteswap(abd_meta, size, NULL,
6175 abd_fletcher_4_native(abd_meta, size, NULL, &zc);
6177 VERIFY0(bcmp(&zc, &zc_ref, sizeof (zc)));
6178 VERIFY0(bcmp(&zc_byteswap, &zc_ref_byteswap,
6179 sizeof (zc_byteswap)));
6183 umem_free(buf, size);
6190 ztest_fletcher_incr(ztest_ds_t *zd, uint64_t id)
6197 zio_cksum_t zc_ref_bswap;
6199 hrtime_t end = gethrtime() + NANOSEC;
6201 while (gethrtime() <= end) {
6202 int run_count = 100;
6204 size = ztest_random_blocksize();
6205 buf = umem_alloc(size, UMEM_NOFAIL);
6207 for (i = 0, ptr = buf; i < size / sizeof (*ptr); i++, ptr++)
6208 *ptr = ztest_random(UINT_MAX);
6210 VERIFY0(fletcher_4_impl_set("scalar"));
6211 fletcher_4_native(buf, size, NULL, &zc_ref);
6212 fletcher_4_byteswap(buf, size, NULL, &zc_ref_bswap);
6214 VERIFY0(fletcher_4_impl_set("cycle"));
6216 while (run_count-- > 0) {
6218 zio_cksum_t zc_bswap;
6221 ZIO_SET_CHECKSUM(&zc, 0, 0, 0, 0);
6222 ZIO_SET_CHECKSUM(&zc_bswap, 0, 0, 0, 0);
6224 while (pos < size) {
6225 size_t inc = 64 * ztest_random(size / 67);
6226 /* sometimes add few bytes to test non-simd */
6227 if (ztest_random(100) < 10)
6228 inc += P2ALIGN(ztest_random(64),
6231 if (inc > (size - pos))
6234 fletcher_4_incremental_native(buf + pos, inc,
6236 fletcher_4_incremental_byteswap(buf + pos, inc,
6242 VERIFY3U(pos, ==, size);
6244 VERIFY(ZIO_CHECKSUM_EQUAL(zc, zc_ref));
6245 VERIFY(ZIO_CHECKSUM_EQUAL(zc_bswap, zc_ref_bswap));
6248 * verify if incremental on the whole buffer is
6249 * equivalent to non-incremental version
6251 ZIO_SET_CHECKSUM(&zc, 0, 0, 0, 0);
6252 ZIO_SET_CHECKSUM(&zc_bswap, 0, 0, 0, 0);
6254 fletcher_4_incremental_native(buf, size, &zc);
6255 fletcher_4_incremental_byteswap(buf, size, &zc_bswap);
6257 VERIFY(ZIO_CHECKSUM_EQUAL(zc, zc_ref));
6258 VERIFY(ZIO_CHECKSUM_EQUAL(zc_bswap, zc_ref_bswap));
6261 umem_free(buf, size);
6266 ztest_check_path(char *path)
6269 /* return true on success */
6270 return (!stat(path, &s));
6274 ztest_get_zdb_bin(char *bin, int len)
6278 * Try to use ZDB_PATH and in-tree zdb path. If not successful, just
6279 * let popen to search through PATH.
6281 if ((zdb_path = getenv("ZDB_PATH"))) {
6282 strlcpy(bin, zdb_path, len); /* In env */
6283 if (!ztest_check_path(bin)) {
6284 ztest_dump_core = 0;
6285 fatal(1, "invalid ZDB_PATH '%s'", bin);
6290 VERIFY(realpath(getexecname(), bin) != NULL);
6291 if (strstr(bin, "/ztest/")) {
6292 strstr(bin, "/ztest/")[0] = '\0'; /* In-tree */
6293 strcat(bin, "/zdb/zdb");
6294 if (ztest_check_path(bin))
6301 * Verify pool integrity by running zdb.
6304 ztest_run_zdb(char *pool)
6310 const int len = MAXPATHLEN + MAXNAMELEN + 20;
6313 bin = umem_alloc(len, UMEM_NOFAIL);
6314 zdb = umem_alloc(len, UMEM_NOFAIL);
6315 zbuf = umem_alloc(1024, UMEM_NOFAIL);
6317 ztest_get_zdb_bin(bin, len);
6320 "%s -bcc%s%s -G -d -U %s "
6321 "-o zfs_reconstruct_indirect_combinations_max=1000000 %s",
6323 ztest_opts.zo_verbose >= 3 ? "s" : "",
6324 ztest_opts.zo_verbose >= 4 ? "v" : "",
6328 if (ztest_opts.zo_verbose >= 5)
6329 (void) printf("Executing %s\n", strstr(zdb, "zdb "));
6331 fp = popen(zdb, "r");
6333 while (fgets(zbuf, 1024, fp) != NULL)
6334 if (ztest_opts.zo_verbose >= 3)
6335 (void) printf("%s", zbuf);
6337 status = pclose(fp);
6342 ztest_dump_core = 0;
6343 if (WIFEXITED(status))
6344 fatal(0, "'%s' exit code %d", zdb, WEXITSTATUS(status));
6346 fatal(0, "'%s' died with signal %d", zdb, WTERMSIG(status));
6348 umem_free(bin, len);
6349 umem_free(zdb, len);
6350 umem_free(zbuf, 1024);
6354 ztest_walk_pool_directory(char *header)
6358 if (ztest_opts.zo_verbose >= 6)
6359 (void) printf("%s\n", header);
6361 mutex_enter(&spa_namespace_lock);
6362 while ((spa = spa_next(spa)) != NULL)
6363 if (ztest_opts.zo_verbose >= 6)
6364 (void) printf("\t%s\n", spa_name(spa));
6365 mutex_exit(&spa_namespace_lock);
6369 ztest_spa_import_export(char *oldname, char *newname)
6371 nvlist_t *config, *newconfig;
6376 if (ztest_opts.zo_verbose >= 4) {
6377 (void) printf("import/export: old = %s, new = %s\n",
6382 * Clean up from previous runs.
6384 (void) spa_destroy(newname);
6387 * Get the pool's configuration and guid.
6389 VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG));
6392 * Kick off a scrub to tickle scrub/export races.
6394 if (ztest_random(2) == 0)
6395 (void) spa_scan(spa, POOL_SCAN_SCRUB);
6397 pool_guid = spa_guid(spa);
6398 spa_close(spa, FTAG);
6400 ztest_walk_pool_directory("pools before export");
6405 VERIFY3U(0, ==, spa_export(oldname, &config, B_FALSE, B_FALSE));
6407 ztest_walk_pool_directory("pools after export");
6412 newconfig = spa_tryimport(config);
6413 ASSERT(newconfig != NULL);
6414 nvlist_free(newconfig);
6417 * Import it under the new name.
6419 error = spa_import(newname, config, NULL, 0);
6421 dump_nvlist(config, 0);
6422 fatal(B_FALSE, "couldn't import pool %s as %s: error %u",
6423 oldname, newname, error);
6426 ztest_walk_pool_directory("pools after import");
6429 * Try to import it again -- should fail with EEXIST.
6431 VERIFY3U(EEXIST, ==, spa_import(newname, config, NULL, 0));
6434 * Try to import it under a different name -- should fail with EEXIST.
6436 VERIFY3U(EEXIST, ==, spa_import(oldname, config, NULL, 0));
6439 * Verify that the pool is no longer visible under the old name.
6441 VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
6444 * Verify that we can open and close the pool using the new name.
6446 VERIFY3U(0, ==, spa_open(newname, &spa, FTAG));
6447 ASSERT(pool_guid == spa_guid(spa));
6448 spa_close(spa, FTAG);
6450 nvlist_free(config);
6454 ztest_resume(spa_t *spa)
6456 if (spa_suspended(spa) && ztest_opts.zo_verbose >= 6)
6457 (void) printf("resuming from suspended state\n");
6458 spa_vdev_state_enter(spa, SCL_NONE);
6459 vdev_clear(spa, NULL);
6460 (void) spa_vdev_state_exit(spa, NULL, 0);
6461 (void) zio_resume(spa);
6465 ztest_resume_thread(void *arg)
6469 while (!ztest_exiting) {
6470 if (spa_suspended(spa))
6472 (void) poll(NULL, 0, 100);
6475 * Periodically change the zfs_compressed_arc_enabled setting.
6477 if (ztest_random(10) == 0)
6478 zfs_compressed_arc_enabled = ztest_random(2);
6481 * Periodically change the zfs_abd_scatter_enabled setting.
6483 if (ztest_random(10) == 0)
6484 zfs_abd_scatter_enabled = ztest_random(2);
6491 ztest_deadman_thread(void *arg)
6493 ztest_shared_t *zs = arg;
6494 spa_t *spa = ztest_spa;
6495 hrtime_t delta, overdue, total = 0;
6498 delta = zs->zs_thread_stop - zs->zs_thread_start +
6499 MSEC2NSEC(zfs_deadman_synctime_ms);
6501 (void) poll(NULL, 0, (int)NSEC2MSEC(delta));
6504 * If the pool is suspended then fail immediately. Otherwise,
6505 * check to see if the pool is making any progress. If
6506 * vdev_deadman() discovers that there hasn't been any recent
6507 * I/Os then it will end up aborting the tests.
6509 if (spa_suspended(spa) || spa->spa_root_vdev == NULL) {
6510 fatal(0, "aborting test after %llu seconds because "
6511 "pool has transitioned to a suspended state.",
6512 zfs_deadman_synctime_ms / 1000);
6514 vdev_deadman(spa->spa_root_vdev, FTAG);
6517 * If the process doesn't complete within a grace period of
6518 * zfs_deadman_synctime_ms over the expected finish time,
6519 * then it may be hung and is terminated.
6521 overdue = zs->zs_proc_stop + MSEC2NSEC(zfs_deadman_synctime_ms);
6522 total += zfs_deadman_synctime_ms / 1000;
6523 if (gethrtime() > overdue) {
6524 fatal(0, "aborting test after %llu seconds because "
6525 "the process is overdue for termination.", total);
6528 (void) printf("ztest has been running for %lld seconds\n",
6534 ztest_execute(int test, ztest_info_t *zi, uint64_t id)
6536 ztest_ds_t *zd = &ztest_ds[id % ztest_opts.zo_datasets];
6537 ztest_shared_callstate_t *zc = ZTEST_GET_SHARED_CALLSTATE(test);
6538 hrtime_t functime = gethrtime();
6541 for (i = 0; i < zi->zi_iters; i++)
6542 zi->zi_func(zd, id);
6544 functime = gethrtime() - functime;
6546 atomic_add_64(&zc->zc_count, 1);
6547 atomic_add_64(&zc->zc_time, functime);
6549 if (ztest_opts.zo_verbose >= 4)
6550 (void) printf("%6.2f sec in %s\n",
6551 (double)functime / NANOSEC, zi->zi_funcname);
6555 ztest_thread(void *arg)
6558 uint64_t id = (uintptr_t)arg;
6559 ztest_shared_t *zs = ztest_shared;
6563 ztest_shared_callstate_t *zc;
6565 while ((now = gethrtime()) < zs->zs_thread_stop) {
6567 * See if it's time to force a crash.
6569 if (now > zs->zs_thread_kill)
6573 * If we're getting ENOSPC with some regularity, stop.
6575 if (zs->zs_enospc_count > 10)
6579 * Pick a random function to execute.
6581 rand = ztest_random(ZTEST_FUNCS);
6582 zi = &ztest_info[rand];
6583 zc = ZTEST_GET_SHARED_CALLSTATE(rand);
6584 call_next = zc->zc_next;
6586 if (now >= call_next &&
6587 atomic_cas_64(&zc->zc_next, call_next, call_next +
6588 ztest_random(2 * zi->zi_interval[0] + 1)) == call_next) {
6589 ztest_execute(rand, zi, id);
6597 ztest_dataset_name(char *dsname, char *pool, int d)
6599 (void) snprintf(dsname, ZFS_MAX_DATASET_NAME_LEN, "%s/ds_%d", pool, d);
6603 ztest_dataset_destroy(int d)
6605 char name[ZFS_MAX_DATASET_NAME_LEN];
6608 ztest_dataset_name(name, ztest_opts.zo_pool, d);
6610 if (ztest_opts.zo_verbose >= 3)
6611 (void) printf("Destroying %s to free up space\n", name);
6614 * Cleanup any non-standard clones and snapshots. In general,
6615 * ztest thread t operates on dataset (t % zopt_datasets),
6616 * so there may be more than one thing to clean up.
6618 for (t = d; t < ztest_opts.zo_threads;
6619 t += ztest_opts.zo_datasets)
6620 ztest_dsl_dataset_cleanup(name, t);
6622 (void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
6623 DS_FIND_SNAPSHOTS | DS_FIND_CHILDREN);
6627 ztest_dataset_dirobj_verify(ztest_ds_t *zd)
6629 uint64_t usedobjs, dirobjs, scratch;
6632 * ZTEST_DIROBJ is the object directory for the entire dataset.
6633 * Therefore, the number of objects in use should equal the
6634 * number of ZTEST_DIROBJ entries, +1 for ZTEST_DIROBJ itself.
6635 * If not, we have an object leak.
6637 * Note that we can only check this in ztest_dataset_open(),
6638 * when the open-context and syncing-context values agree.
6639 * That's because zap_count() returns the open-context value,
6640 * while dmu_objset_space() returns the rootbp fill count.
6642 VERIFY3U(0, ==, zap_count(zd->zd_os, ZTEST_DIROBJ, &dirobjs));
6643 dmu_objset_space(zd->zd_os, &scratch, &scratch, &usedobjs, &scratch);
6644 ASSERT3U(dirobjs + 1, ==, usedobjs);
6648 ztest_dataset_open(int d)
6650 ztest_ds_t *zd = &ztest_ds[d];
6651 uint64_t committed_seq = ZTEST_GET_SHARED_DS(d)->zd_seq;
6654 char name[ZFS_MAX_DATASET_NAME_LEN];
6657 ztest_dataset_name(name, ztest_opts.zo_pool, d);
6659 (void) pthread_rwlock_rdlock(&ztest_name_lock);
6661 error = ztest_dataset_create(name);
6662 if (error == ENOSPC) {
6663 (void) pthread_rwlock_unlock(&ztest_name_lock);
6664 ztest_record_enospc(FTAG);
6667 ASSERT(error == 0 || error == EEXIST);
6669 VERIFY0(ztest_dmu_objset_own(name, DMU_OST_OTHER, B_FALSE,
6671 (void) pthread_rwlock_unlock(&ztest_name_lock);
6673 ztest_zd_init(zd, ZTEST_GET_SHARED_DS(d), os);
6675 zilog = zd->zd_zilog;
6677 if (zilog->zl_header->zh_claim_lr_seq != 0 &&
6678 zilog->zl_header->zh_claim_lr_seq < committed_seq)
6679 fatal(0, "missing log records: claimed %llu < committed %llu",
6680 zilog->zl_header->zh_claim_lr_seq, committed_seq);
6682 ztest_dataset_dirobj_verify(zd);
6684 zil_replay(os, zd, ztest_replay_vector);
6686 ztest_dataset_dirobj_verify(zd);
6688 if (ztest_opts.zo_verbose >= 6)
6689 (void) printf("%s replay %llu blocks, %llu records, seq %llu\n",
6691 (u_longlong_t)zilog->zl_parse_blk_count,
6692 (u_longlong_t)zilog->zl_parse_lr_count,
6693 (u_longlong_t)zilog->zl_replaying_seq);
6695 zilog = zil_open(os, ztest_get_data);
6697 if (zilog->zl_replaying_seq != 0 &&
6698 zilog->zl_replaying_seq < committed_seq)
6699 fatal(0, "missing log records: replayed %llu < committed %llu",
6700 zilog->zl_replaying_seq, committed_seq);
6706 ztest_dataset_close(int d)
6708 ztest_ds_t *zd = &ztest_ds[d];
6710 zil_close(zd->zd_zilog);
6711 txg_wait_synced(spa_get_dsl(zd->zd_os->os_spa), 0);
6712 dmu_objset_disown(zd->zd_os, B_TRUE, zd);
6718 * Kick off threads to run tests on all datasets in parallel.
6721 ztest_run(ztest_shared_t *zs)
6725 kthread_t *resume_thread;
6726 kthread_t **run_threads;
6731 ztest_exiting = B_FALSE;
6734 * Initialize parent/child shared state.
6736 mutex_init(&ztest_vdev_lock, NULL, MUTEX_DEFAULT, NULL);
6737 mutex_init(&ztest_checkpoint_lock, NULL, MUTEX_DEFAULT, NULL);
6738 VERIFY0(pthread_rwlock_init(&ztest_name_lock, NULL));
6740 zs->zs_thread_start = gethrtime();
6741 zs->zs_thread_stop =
6742 zs->zs_thread_start + ztest_opts.zo_passtime * NANOSEC;
6743 zs->zs_thread_stop = MIN(zs->zs_thread_stop, zs->zs_proc_stop);
6744 zs->zs_thread_kill = zs->zs_thread_stop;
6745 if (ztest_random(100) < ztest_opts.zo_killrate) {
6746 zs->zs_thread_kill -=
6747 ztest_random(ztest_opts.zo_passtime * NANOSEC);
6750 mutex_init(&zcl.zcl_callbacks_lock, NULL, MUTEX_DEFAULT, NULL);
6752 list_create(&zcl.zcl_callbacks, sizeof (ztest_cb_data_t),
6753 offsetof(ztest_cb_data_t, zcd_node));
6758 kernel_init(FREAD | FWRITE);
6759 VERIFY0(spa_open(ztest_opts.zo_pool, &spa, FTAG));
6760 metaslab_preload_limit = ztest_random(20) + 1;
6763 dmu_objset_stats_t dds;
6764 VERIFY0(ztest_dmu_objset_own(ztest_opts.zo_pool,
6765 DMU_OST_ANY, B_TRUE, B_TRUE, FTAG, &os));
6766 dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
6767 dmu_objset_fast_stat(os, &dds);
6768 dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
6769 zs->zs_guid = dds.dds_guid;
6770 dmu_objset_disown(os, B_TRUE, FTAG);
6772 spa->spa_dedup_ditto = 2 * ZIO_DEDUPDITTO_MIN;
6775 * Create a thread to periodically resume suspended I/O.
6777 resume_thread = thread_create(NULL, 0, ztest_resume_thread,
6778 spa, 0, NULL, TS_RUN | TS_JOINABLE, defclsyspri);
6781 * Create a deadman thread and set to panic if we hang.
6783 (void) thread_create(NULL, 0, ztest_deadman_thread,
6784 zs, 0, NULL, TS_RUN | TS_JOINABLE, defclsyspri);
6786 spa->spa_deadman_failmode = ZIO_FAILURE_MODE_PANIC;
6789 * Verify that we can safely inquire about any object,
6790 * whether it's allocated or not. To make it interesting,
6791 * we probe a 5-wide window around each power of two.
6792 * This hits all edge cases, including zero and the max.
6794 for (t = 0; t < 64; t++) {
6795 for (d = -5; d <= 5; d++) {
6796 error = dmu_object_info(spa->spa_meta_objset,
6797 (1ULL << t) + d, NULL);
6798 ASSERT(error == 0 || error == ENOENT ||
6804 * If we got any ENOSPC errors on the previous run, destroy something.
6806 if (zs->zs_enospc_count != 0) {
6807 int d = ztest_random(ztest_opts.zo_datasets);
6808 ztest_dataset_destroy(d);
6810 zs->zs_enospc_count = 0;
6812 run_threads = umem_zalloc(ztest_opts.zo_threads * sizeof (kthread_t *),
6815 if (ztest_opts.zo_verbose >= 4)
6816 (void) printf("starting main threads...\n");
6819 * Kick off all the tests that run in parallel.
6821 for (t = 0; t < ztest_opts.zo_threads; t++) {
6822 if (t < ztest_opts.zo_datasets && ztest_dataset_open(t) != 0) {
6823 umem_free(run_threads, ztest_opts.zo_threads *
6824 sizeof (kthread_t *));
6828 run_threads[t] = thread_create(NULL, 0, ztest_thread,
6829 (void *)(uintptr_t)t, 0, NULL, TS_RUN | TS_JOINABLE,
6834 * Wait for all of the tests to complete. We go in reverse order
6835 * so we don't close datasets while threads are still using them.
6837 for (t = ztest_opts.zo_threads - 1; t >= 0; t--) {
6838 VERIFY0(thread_join(run_threads[t]));
6839 if (t < ztest_opts.zo_datasets)
6840 ztest_dataset_close(t);
6843 txg_wait_synced(spa_get_dsl(spa), 0);
6845 zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(spa));
6846 zs->zs_space = metaslab_class_get_space(spa_normal_class(spa));
6848 umem_free(run_threads, ztest_opts.zo_threads * sizeof (kthread_t *));
6850 /* Kill the resume thread */
6851 ztest_exiting = B_TRUE;
6852 VERIFY0(thread_join(resume_thread));
6856 * Right before closing the pool, kick off a bunch of async I/O;
6857 * spa_close() should wait for it to complete.
6859 for (object = 1; object < 50; object++) {
6860 dmu_prefetch(spa->spa_meta_objset, object, 0, 0, 1ULL << 20,
6861 ZIO_PRIORITY_SYNC_READ);
6864 /* Verify that at least one commit cb was called in a timely fashion */
6865 if (zc_cb_counter >= ZTEST_COMMIT_CB_MIN_REG)
6866 VERIFY0(zc_min_txg_delay);
6868 spa_close(spa, FTAG);
6871 * Verify that we can loop over all pools.
6873 mutex_enter(&spa_namespace_lock);
6874 for (spa = spa_next(NULL); spa != NULL; spa = spa_next(spa))
6875 if (ztest_opts.zo_verbose > 3)
6876 (void) printf("spa_next: found %s\n", spa_name(spa));
6877 mutex_exit(&spa_namespace_lock);
6880 * Verify that we can export the pool and reimport it under a
6883 if ((ztest_random(2) == 0) && !ztest_opts.zo_mmp_test) {
6884 char name[ZFS_MAX_DATASET_NAME_LEN];
6885 (void) snprintf(name, sizeof (name), "%s_import",
6886 ztest_opts.zo_pool);
6887 ztest_spa_import_export(ztest_opts.zo_pool, name);
6888 ztest_spa_import_export(name, ztest_opts.zo_pool);
6893 list_destroy(&zcl.zcl_callbacks);
6894 mutex_destroy(&zcl.zcl_callbacks_lock);
6895 (void) pthread_rwlock_destroy(&ztest_name_lock);
6896 mutex_destroy(&ztest_vdev_lock);
6897 mutex_destroy(&ztest_checkpoint_lock);
6903 ztest_ds_t *zd = &ztest_ds[0];
6907 if (ztest_opts.zo_verbose >= 3)
6908 (void) printf("testing spa_freeze()...\n");
6910 kernel_init(FREAD | FWRITE);
6911 VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
6912 VERIFY3U(0, ==, ztest_dataset_open(0));
6916 * Force the first log block to be transactionally allocated.
6917 * We have to do this before we freeze the pool -- otherwise
6918 * the log chain won't be anchored.
6920 while (BP_IS_HOLE(&zd->zd_zilog->zl_header->zh_log)) {
6921 ztest_dmu_object_alloc_free(zd, 0);
6922 zil_commit(zd->zd_zilog, 0);
6925 txg_wait_synced(spa_get_dsl(spa), 0);
6928 * Freeze the pool. This stops spa_sync() from doing anything,
6929 * so that the only way to record changes from now on is the ZIL.
6934 * Because it is hard to predict how much space a write will actually
6935 * require beforehand, we leave ourselves some fudge space to write over
6938 uint64_t capacity = metaslab_class_get_space(spa_normal_class(spa)) / 2;
6941 * Run tests that generate log records but don't alter the pool config
6942 * or depend on DSL sync tasks (snapshots, objset create/destroy, etc).
6943 * We do a txg_wait_synced() after each iteration to force the txg
6944 * to increase well beyond the last synced value in the uberblock.
6945 * The ZIL should be OK with that.
6947 * Run a random number of times less than zo_maxloops and ensure we do
6948 * not run out of space on the pool.
6950 while (ztest_random(10) != 0 &&
6951 numloops++ < ztest_opts.zo_maxloops &&
6952 metaslab_class_get_alloc(spa_normal_class(spa)) < capacity) {
6954 ztest_od_init(&od, 0, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0, 0);
6955 VERIFY0(ztest_object_init(zd, &od, sizeof (od), B_FALSE));
6956 ztest_io(zd, od.od_object,
6957 ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
6958 txg_wait_synced(spa_get_dsl(spa), 0);
6962 * Commit all of the changes we just generated.
6964 zil_commit(zd->zd_zilog, 0);
6965 txg_wait_synced(spa_get_dsl(spa), 0);
6968 * Close our dataset and close the pool.
6970 ztest_dataset_close(0);
6971 spa_close(spa, FTAG);
6975 * Open and close the pool and dataset to induce log replay.
6977 kernel_init(FREAD | FWRITE);
6978 VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
6979 ASSERT(spa_freeze_txg(spa) == UINT64_MAX);
6980 VERIFY3U(0, ==, ztest_dataset_open(0));
6982 txg_wait_synced(spa_get_dsl(spa), 0);
6983 ztest_dataset_close(0);
6984 ztest_reguid(NULL, 0);
6986 spa_close(spa, FTAG);
6991 print_time(hrtime_t t, char *timebuf)
6993 hrtime_t s = t / NANOSEC;
6994 hrtime_t m = s / 60;
6995 hrtime_t h = m / 60;
6996 hrtime_t d = h / 24;
7005 (void) sprintf(timebuf,
7006 "%llud%02lluh%02llum%02llus", d, h, m, s);
7008 (void) sprintf(timebuf, "%lluh%02llum%02llus", h, m, s);
7010 (void) sprintf(timebuf, "%llum%02llus", m, s);
7012 (void) sprintf(timebuf, "%llus", s);
7016 make_random_props(void)
7020 VERIFY0(nvlist_alloc(&props, NV_UNIQUE_NAME, 0));
7021 if (ztest_random(2) == 0)
7024 VERIFY0(nvlist_add_uint64(props,
7025 zpool_prop_to_name(ZPOOL_PROP_AUTOREPLACE), 1));
7031 * Import a storage pool with the given name.
7034 ztest_import(ztest_shared_t *zs)
7036 libzfs_handle_t *hdl;
7037 importargs_t args = { 0 };
7039 nvlist_t *cfg = NULL;
7041 char *searchdirs[nsearch];
7042 char *name = ztest_opts.zo_pool;
7043 int flags = ZFS_IMPORT_MISSING_LOG;
7046 mutex_init(&ztest_vdev_lock, NULL, MUTEX_DEFAULT, NULL);
7047 mutex_init(&ztest_checkpoint_lock, NULL, MUTEX_DEFAULT, NULL);
7048 VERIFY0(pthread_rwlock_init(&ztest_name_lock, NULL));
7050 kernel_init(FREAD | FWRITE);
7051 hdl = libzfs_init();
7053 searchdirs[0] = ztest_opts.zo_dir;
7054 args.paths = nsearch;
7055 args.path = searchdirs;
7056 args.can_be_active = B_FALSE;
7058 error = zpool_tryimport(hdl, name, &cfg, &args);
7060 (void) fatal(0, "No pools found\n");
7062 VERIFY0(spa_import(name, cfg, NULL, flags));
7063 VERIFY0(spa_open(name, &spa, FTAG));
7064 zs->zs_metaslab_sz =
7065 1ULL << spa->spa_root_vdev->vdev_child[0]->vdev_ms_shift;
7066 spa_close(spa, FTAG);
7071 if (!ztest_opts.zo_mmp_test) {
7072 ztest_run_zdb(ztest_opts.zo_pool);
7074 ztest_run_zdb(ztest_opts.zo_pool);
7077 (void) pthread_rwlock_destroy(&ztest_name_lock);
7078 mutex_destroy(&ztest_vdev_lock);
7079 mutex_destroy(&ztest_checkpoint_lock);
7083 * Create a storage pool with the given name and initial vdev size.
7084 * Then test spa_freeze() functionality.
7087 ztest_init(ztest_shared_t *zs)
7090 nvlist_t *nvroot, *props;
7093 mutex_init(&ztest_vdev_lock, NULL, MUTEX_DEFAULT, NULL);
7094 mutex_init(&ztest_checkpoint_lock, NULL, MUTEX_DEFAULT, NULL);
7095 VERIFY0(pthread_rwlock_init(&ztest_name_lock, NULL));
7097 kernel_init(FREAD | FWRITE);
7100 * Create the storage pool.
7102 (void) spa_destroy(ztest_opts.zo_pool);
7103 ztest_shared->zs_vdev_next_leaf = 0;
7105 zs->zs_mirrors = ztest_opts.zo_mirrors;
7106 nvroot = make_vdev_root(NULL, NULL, NULL, ztest_opts.zo_vdev_size, 0,
7107 0, ztest_opts.zo_raidz, zs->zs_mirrors, 1);
7108 props = make_random_props();
7111 * We don't expect the pool to suspend unless maxfaults == 0,
7112 * in which case ztest_fault_inject() temporarily takes away
7113 * the only valid replica.
7115 VERIFY0(nvlist_add_uint64(props,
7116 zpool_prop_to_name(ZPOOL_PROP_FAILUREMODE),
7117 MAXFAULTS(zs) ? ZIO_FAILURE_MODE_PANIC : ZIO_FAILURE_MODE_WAIT));
7119 for (i = 0; i < SPA_FEATURES; i++) {
7121 VERIFY3S(-1, !=, asprintf(&buf, "feature@%s",
7122 spa_feature_table[i].fi_uname));
7123 VERIFY3U(0, ==, nvlist_add_uint64(props, buf, 0));
7127 VERIFY0(spa_create(ztest_opts.zo_pool, nvroot, props, NULL, NULL));
7128 nvlist_free(nvroot);
7131 VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
7132 zs->zs_metaslab_sz =
7133 1ULL << spa->spa_root_vdev->vdev_child[0]->vdev_ms_shift;
7134 spa_close(spa, FTAG);
7138 if (!ztest_opts.zo_mmp_test) {
7139 ztest_run_zdb(ztest_opts.zo_pool);
7141 ztest_run_zdb(ztest_opts.zo_pool);
7144 (void) pthread_rwlock_destroy(&ztest_name_lock);
7145 mutex_destroy(&ztest_vdev_lock);
7146 mutex_destroy(&ztest_checkpoint_lock);
7152 static char ztest_name_data[] = "/tmp/ztest.data.XXXXXX";
7154 ztest_fd_data = mkstemp(ztest_name_data);
7155 ASSERT3S(ztest_fd_data, >=, 0);
7156 (void) unlink(ztest_name_data);
7160 shared_data_size(ztest_shared_hdr_t *hdr)
7164 size = hdr->zh_hdr_size;
7165 size += hdr->zh_opts_size;
7166 size += hdr->zh_size;
7167 size += hdr->zh_stats_size * hdr->zh_stats_count;
7168 size += hdr->zh_ds_size * hdr->zh_ds_count;
7177 ztest_shared_hdr_t *hdr;
7179 hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
7180 PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
7181 ASSERT(hdr != MAP_FAILED);
7183 VERIFY3U(0, ==, ftruncate(ztest_fd_data, sizeof (ztest_shared_hdr_t)));
7185 hdr->zh_hdr_size = sizeof (ztest_shared_hdr_t);
7186 hdr->zh_opts_size = sizeof (ztest_shared_opts_t);
7187 hdr->zh_size = sizeof (ztest_shared_t);
7188 hdr->zh_stats_size = sizeof (ztest_shared_callstate_t);
7189 hdr->zh_stats_count = ZTEST_FUNCS;
7190 hdr->zh_ds_size = sizeof (ztest_shared_ds_t);
7191 hdr->zh_ds_count = ztest_opts.zo_datasets;
7193 size = shared_data_size(hdr);
7194 VERIFY3U(0, ==, ftruncate(ztest_fd_data, size));
7196 (void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
7203 ztest_shared_hdr_t *hdr;
7206 hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
7207 PROT_READ, MAP_SHARED, ztest_fd_data, 0);
7208 ASSERT(hdr != MAP_FAILED);
7210 size = shared_data_size(hdr);
7212 (void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
7213 hdr = ztest_shared_hdr = (void *)mmap(0, P2ROUNDUP(size, getpagesize()),
7214 PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
7215 ASSERT(hdr != MAP_FAILED);
7216 buf = (uint8_t *)hdr;
7218 offset = hdr->zh_hdr_size;
7219 ztest_shared_opts = (void *)&buf[offset];
7220 offset += hdr->zh_opts_size;
7221 ztest_shared = (void *)&buf[offset];
7222 offset += hdr->zh_size;
7223 ztest_shared_callstate = (void *)&buf[offset];
7224 offset += hdr->zh_stats_size * hdr->zh_stats_count;
7225 ztest_shared_ds = (void *)&buf[offset];
7229 exec_child(char *cmd, char *libpath, boolean_t ignorekill, int *statusp)
7233 char *cmdbuf = NULL;
7238 cmdbuf = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
7239 (void) strlcpy(cmdbuf, getexecname(), MAXPATHLEN);
7244 fatal(1, "fork failed");
7246 if (pid == 0) { /* child */
7247 char *emptyargv[2] = { cmd, NULL };
7248 char fd_data_str[12];
7250 struct rlimit rl = { 1024, 1024 };
7251 (void) setrlimit(RLIMIT_NOFILE, &rl);
7253 (void) close(ztest_fd_rand);
7254 VERIFY(11 >= snprintf(fd_data_str, 12, "%d", ztest_fd_data));
7255 VERIFY(0 == setenv("ZTEST_FD_DATA", fd_data_str, 1));
7257 (void) enable_extended_FILE_stdio(-1, -1);
7258 if (libpath != NULL)
7259 VERIFY(0 == setenv("LD_LIBRARY_PATH", libpath, 1));
7260 (void) execv(cmd, emptyargv);
7261 ztest_dump_core = B_FALSE;
7262 fatal(B_TRUE, "exec failed: %s", cmd);
7265 if (cmdbuf != NULL) {
7266 umem_free(cmdbuf, MAXPATHLEN);
7270 while (waitpid(pid, &status, 0) != pid)
7272 if (statusp != NULL)
7275 if (WIFEXITED(status)) {
7276 if (WEXITSTATUS(status) != 0) {
7277 (void) fprintf(stderr, "child exited with code %d\n",
7278 WEXITSTATUS(status));
7282 } else if (WIFSIGNALED(status)) {
7283 if (!ignorekill || WTERMSIG(status) != SIGKILL) {
7284 (void) fprintf(stderr, "child died with signal %d\n",
7290 (void) fprintf(stderr, "something strange happened to child\n");
7297 ztest_run_init(void)
7301 ztest_shared_t *zs = ztest_shared;
7304 * Blow away any existing copy of zpool.cache
7306 (void) remove(spa_config_path);
7308 if (ztest_opts.zo_init == 0) {
7309 if (ztest_opts.zo_verbose >= 1)
7310 (void) printf("Importing pool %s\n",
7311 ztest_opts.zo_pool);
7317 * Create and initialize our storage pool.
7319 for (i = 1; i <= ztest_opts.zo_init; i++) {
7320 bzero(zs, sizeof (ztest_shared_t));
7321 if (ztest_opts.zo_verbose >= 3 &&
7322 ztest_opts.zo_init != 1) {
7323 (void) printf("ztest_init(), pass %d\n", i);
7330 main(int argc, char **argv)
7338 ztest_shared_callstate_t *zc;
7340 char numbuf[NN_NUMBUF_SZ];
7345 char *fd_data_str = getenv("ZTEST_FD_DATA");
7346 struct sigaction action;
7348 (void) setvbuf(stdout, NULL, _IOLBF, 0);
7350 dprintf_setup(&argc, argv);
7351 zfs_deadman_synctime_ms = 300000;
7353 action.sa_handler = sig_handler;
7354 sigemptyset(&action.sa_mask);
7355 action.sa_flags = 0;
7357 if (sigaction(SIGSEGV, &action, NULL) < 0) {
7358 (void) fprintf(stderr, "ztest: cannot catch SIGSEGV: %s.\n",
7363 if (sigaction(SIGABRT, &action, NULL) < 0) {
7364 (void) fprintf(stderr, "ztest: cannot catch SIGABRT: %s.\n",
7370 * Force random_get_bytes() to use /dev/urandom in order to prevent
7371 * ztest from needlessly depleting the system entropy pool.
7373 random_path = "/dev/urandom";
7374 ztest_fd_rand = open(random_path, O_RDONLY);
7375 ASSERT3S(ztest_fd_rand, >=, 0);
7378 process_options(argc, argv);
7383 bcopy(&ztest_opts, ztest_shared_opts,
7384 sizeof (*ztest_shared_opts));
7386 ztest_fd_data = atoi(fd_data_str);
7388 bcopy(ztest_shared_opts, &ztest_opts, sizeof (ztest_opts));
7390 ASSERT3U(ztest_opts.zo_datasets, ==, ztest_shared_hdr->zh_ds_count);
7392 /* Override location of zpool.cache */
7393 VERIFY(asprintf((char **)&spa_config_path, "%s/zpool.cache",
7394 ztest_opts.zo_dir) != -1);
7396 ztest_ds = umem_alloc(ztest_opts.zo_datasets * sizeof (ztest_ds_t),
7401 metaslab_force_ganging = ztest_opts.zo_metaslab_force_ganging;
7402 metaslab_df_alloc_threshold =
7403 zs->zs_metaslab_df_alloc_threshold;
7412 hasalt = (strlen(ztest_opts.zo_alt_ztest) != 0);
7414 if (ztest_opts.zo_verbose >= 1) {
7415 (void) printf("%llu vdevs, %d datasets, %d threads,"
7416 " %llu seconds...\n",
7417 (u_longlong_t)ztest_opts.zo_vdevs,
7418 ztest_opts.zo_datasets,
7419 ztest_opts.zo_threads,
7420 (u_longlong_t)ztest_opts.zo_time);
7423 cmd = umem_alloc(MAXNAMELEN, UMEM_NOFAIL);
7424 (void) strlcpy(cmd, getexecname(), MAXNAMELEN);
7426 zs->zs_do_init = B_TRUE;
7427 if (strlen(ztest_opts.zo_alt_ztest) != 0) {
7428 if (ztest_opts.zo_verbose >= 1) {
7429 (void) printf("Executing older ztest for "
7430 "initialization: %s\n", ztest_opts.zo_alt_ztest);
7432 VERIFY(!exec_child(ztest_opts.zo_alt_ztest,
7433 ztest_opts.zo_alt_libpath, B_FALSE, NULL));
7435 VERIFY(!exec_child(NULL, NULL, B_FALSE, NULL));
7437 zs->zs_do_init = B_FALSE;
7439 zs->zs_proc_start = gethrtime();
7440 zs->zs_proc_stop = zs->zs_proc_start + ztest_opts.zo_time * NANOSEC;
7442 for (f = 0; f < ZTEST_FUNCS; f++) {
7443 zi = &ztest_info[f];
7444 zc = ZTEST_GET_SHARED_CALLSTATE(f);
7445 if (zs->zs_proc_start + zi->zi_interval[0] > zs->zs_proc_stop)
7446 zc->zc_next = UINT64_MAX;
7448 zc->zc_next = zs->zs_proc_start +
7449 ztest_random(2 * zi->zi_interval[0] + 1);
7453 * Run the tests in a loop. These tests include fault injection
7454 * to verify that self-healing data works, and forced crashes
7455 * to verify that we never lose on-disk consistency.
7457 while (gethrtime() < zs->zs_proc_stop) {
7462 * Initialize the workload counters for each function.
7464 for (f = 0; f < ZTEST_FUNCS; f++) {
7465 zc = ZTEST_GET_SHARED_CALLSTATE(f);
7470 /* Set the allocation switch size */
7471 zs->zs_metaslab_df_alloc_threshold =
7472 ztest_random(zs->zs_metaslab_sz / 4) + 1;
7474 if (!hasalt || ztest_random(2) == 0) {
7475 if (hasalt && ztest_opts.zo_verbose >= 1) {
7476 (void) printf("Executing newer ztest: %s\n",
7480 killed = exec_child(cmd, NULL, B_TRUE, &status);
7482 if (hasalt && ztest_opts.zo_verbose >= 1) {
7483 (void) printf("Executing older ztest: %s\n",
7484 ztest_opts.zo_alt_ztest);
7487 killed = exec_child(ztest_opts.zo_alt_ztest,
7488 ztest_opts.zo_alt_libpath, B_TRUE, &status);
7495 if (ztest_opts.zo_verbose >= 1) {
7496 hrtime_t now = gethrtime();
7498 now = MIN(now, zs->zs_proc_stop);
7499 print_time(zs->zs_proc_stop - now, timebuf);
7500 nicenum(zs->zs_space, numbuf, sizeof (numbuf));
7502 (void) printf("Pass %3d, %8s, %3llu ENOSPC, "
7503 "%4.1f%% of %5s used, %3.0f%% done, %8s to go\n",
7505 WIFEXITED(status) ? "Complete" : "SIGKILL",
7506 (u_longlong_t)zs->zs_enospc_count,
7507 100.0 * zs->zs_alloc / zs->zs_space,
7509 100.0 * (now - zs->zs_proc_start) /
7510 (ztest_opts.zo_time * NANOSEC), timebuf);
7513 if (ztest_opts.zo_verbose >= 2) {
7514 (void) printf("\nWorkload summary:\n\n");
7515 (void) printf("%7s %9s %s\n",
7516 "Calls", "Time", "Function");
7517 (void) printf("%7s %9s %s\n",
7518 "-----", "----", "--------");
7519 for (f = 0; f < ZTEST_FUNCS; f++) {
7520 zi = &ztest_info[f];
7521 zc = ZTEST_GET_SHARED_CALLSTATE(f);
7522 print_time(zc->zc_time, timebuf);
7523 (void) printf("%7llu %9s %s\n",
7524 (u_longlong_t)zc->zc_count, timebuf,
7527 (void) printf("\n");
7531 * It's possible that we killed a child during a rename test,
7532 * in which case we'll have a 'ztest_tmp' pool lying around
7533 * instead of 'ztest'. Do a blind rename in case this happened.
7536 if (spa_open(ztest_opts.zo_pool, &spa, FTAG) == 0) {
7537 spa_close(spa, FTAG);
7539 char tmpname[ZFS_MAX_DATASET_NAME_LEN];
7541 kernel_init(FREAD | FWRITE);
7542 (void) snprintf(tmpname, sizeof (tmpname), "%s_tmp",
7543 ztest_opts.zo_pool);
7544 (void) spa_rename(tmpname, ztest_opts.zo_pool);
7548 if (!ztest_opts.zo_mmp_test)
7549 ztest_run_zdb(ztest_opts.zo_pool);
7552 if (ztest_opts.zo_verbose >= 1) {
7554 (void) printf("%d runs of older ztest: %s\n", older,
7555 ztest_opts.zo_alt_ztest);
7556 (void) printf("%d runs of newer ztest: %s\n", newer,
7559 (void) printf("%d killed, %d completed, %.0f%% kill rate\n",
7560 kills, iters - kills, (100.0 * kills) / MAX(1, iters));
7563 umem_free(cmd, MAXNAMELEN);