4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2011, 2016 by Delphix. All rights reserved.
24 * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
25 * Copyright (c) 2013 Steven Hartland. All rights reserved.
29 * The objective of this program is to provide a DMU/ZAP/SPA stress test
30 * that runs entirely in userland, is easy to use, and easy to extend.
32 * The overall design of the ztest program is as follows:
34 * (1) For each major functional area (e.g. adding vdevs to a pool,
35 * creating and destroying datasets, reading and writing objects, etc)
36 * we have a simple routine to test that functionality. These
37 * individual routines do not have to do anything "stressful".
39 * (2) We turn these simple functionality tests into a stress test by
40 * running them all in parallel, with as many threads as desired,
41 * and spread across as many datasets, objects, and vdevs as desired.
43 * (3) While all this is happening, we inject faults into the pool to
44 * verify that self-healing data really works.
46 * (4) Every time we open a dataset, we change its checksum and compression
47 * functions. Thus even individual objects vary from block to block
48 * in which checksum they use and whether they're compressed.
50 * (5) To verify that we never lose on-disk consistency after a crash,
51 * we run the entire test in a child of the main process.
52 * At random times, the child self-immolates with a SIGKILL.
53 * This is the software equivalent of pulling the power cord.
54 * The parent then runs the test again, using the existing
55 * storage pool, as many times as desired. If backwards compatibility
56 * testing is enabled ztest will sometimes run the "older" version
57 * of ztest after a SIGKILL.
59 * (6) To verify that we don't have future leaks or temporal incursions,
60 * many of the functional tests record the transaction group number
61 * as part of their data. When reading old data, they verify that
62 * the transaction group number is less than the current, open txg.
63 * If you add a new test, please do this if applicable.
65 * (7) Threads are created with a reduced stack size, for sanity checking.
66 * Therefore, it's important not to allocate huge buffers on the stack.
68 * When run with no arguments, ztest runs for about five minutes and
69 * produces no output if successful. To get a little bit of information,
70 * specify -V. To get more information, specify -VV, and so on.
72 * To turn this into an overnight stress test, use -T to specify run time.
74 * You can ask more more vdevs [-v], datasets [-d], or threads [-t]
75 * to increase the pool capacity, fanout, and overall stress level.
77 * Use the -k option to set the desired frequency of kills.
79 * When ztest invokes itself it passes all relevant information through a
80 * temporary file which is mmap-ed in the child process. This allows shared
81 * memory to survive the exec syscall. The ztest_shared_hdr_t struct is always
82 * stored at offset 0 of this file and contains information on the size and
83 * number of shared structures in the file. The information stored in this file
84 * must remain backwards compatible with older versions of ztest so that
85 * ztest can invoke them during backwards compatibility testing (-B).
88 #include <sys/zfs_context.h>
94 #include <sys/dmu_objset.h>
100 #include <sys/resource.h>
103 #include <sys/zil_impl.h>
104 #include <sys/zfs_rlock.h>
105 #include <sys/vdev_impl.h>
106 #include <sys/vdev_file.h>
107 #include <sys/spa_impl.h>
108 #include <sys/metaslab_impl.h>
109 #include <sys/dsl_prop.h>
110 #include <sys/dsl_dataset.h>
111 #include <sys/dsl_destroy.h>
112 #include <sys/dsl_scan.h>
113 #include <sys/zio_checksum.h>
114 #include <sys/refcount.h>
115 #include <sys/zfeature.h>
116 #include <sys/dsl_userhold.h>
119 #include <stdio_ext.h>
126 #include <sys/fs/zfs.h>
127 #include <zfs_fletcher.h>
128 #include <libnvpair.h>
131 #include <execinfo.h> /* for backtrace() */
134 static int ztest_fd_data = -1;
135 static int ztest_fd_rand = -1;
137 typedef struct ztest_shared_hdr {
138 uint64_t zh_hdr_size;
139 uint64_t zh_opts_size;
141 uint64_t zh_stats_size;
142 uint64_t zh_stats_count;
144 uint64_t zh_ds_count;
145 } ztest_shared_hdr_t;
147 static ztest_shared_hdr_t *ztest_shared_hdr;
149 typedef struct ztest_shared_opts {
150 char zo_pool[ZFS_MAX_DATASET_NAME_LEN];
151 char zo_dir[ZFS_MAX_DATASET_NAME_LEN];
152 char zo_alt_ztest[MAXNAMELEN];
153 char zo_alt_libpath[MAXNAMELEN];
155 uint64_t zo_vdevtime;
163 uint64_t zo_passtime;
164 uint64_t zo_killrate;
168 uint64_t zo_maxloops;
169 uint64_t zo_metaslab_gang_bang;
171 } ztest_shared_opts_t;
173 static const ztest_shared_opts_t ztest_opts_defaults = {
174 .zo_pool = { 'z', 't', 'e', 's', 't', '\0' },
175 .zo_dir = { '/', 't', 'm', 'p', '\0' },
176 .zo_alt_ztest = { '\0' },
177 .zo_alt_libpath = { '\0' },
179 .zo_ashift = SPA_MINBLOCKSHIFT,
182 .zo_raidz_parity = 1,
183 .zo_vdev_size = SPA_MINDEVSIZE * 4, /* 256m default size */
186 .zo_passtime = 60, /* 60 seconds */
187 .zo_killrate = 70, /* 70% kill rate */
191 .zo_time = 300, /* 5 minutes */
192 .zo_maxloops = 50, /* max loops during spa_freeze() */
193 .zo_metaslab_gang_bang = 32 << 10
196 extern uint64_t metaslab_gang_bang;
197 extern uint64_t metaslab_df_alloc_threshold;
198 extern int metaslab_preload_limit;
199 extern boolean_t zfs_compressed_arc_enabled;
200 extern int zfs_abd_scatter_enabled;
202 static ztest_shared_opts_t *ztest_shared_opts;
203 static ztest_shared_opts_t ztest_opts;
204 static char *ztest_wkeydata = "abcdefghijklmnopqrstuvwxyz012345";
206 typedef struct ztest_shared_ds {
210 static ztest_shared_ds_t *ztest_shared_ds;
211 #define ZTEST_GET_SHARED_DS(d) (&ztest_shared_ds[d])
213 #define BT_MAGIC 0x123456789abcdefULL
214 #define MAXFAULTS() \
215 (MAX(zs->zs_mirrors, 1) * (ztest_opts.zo_raidz_parity + 1) - 1)
219 ZTEST_IO_WRITE_PATTERN,
220 ZTEST_IO_WRITE_ZEROES,
227 typedef struct ztest_block_tag {
231 uint64_t bt_dnodesize;
238 typedef struct bufwad {
256 #define ZTEST_RANGE_LOCKS 64
257 #define ZTEST_OBJECT_LOCKS 64
260 * Object descriptor. Used as a template for object lookup/create/remove.
262 typedef struct ztest_od {
265 dmu_object_type_t od_type;
266 dmu_object_type_t od_crtype;
267 uint64_t od_blocksize;
268 uint64_t od_crblocksize;
269 uint64_t od_crdnodesize;
272 char od_name[ZFS_MAX_DATASET_NAME_LEN];
278 typedef struct ztest_ds {
279 ztest_shared_ds_t *zd_shared;
281 rwlock_t zd_zilog_lock;
283 ztest_od_t *zd_od; /* debugging aid */
284 char zd_name[ZFS_MAX_DATASET_NAME_LEN];
285 kmutex_t zd_dirobj_lock;
286 rll_t zd_object_lock[ZTEST_OBJECT_LOCKS];
287 zll_t zd_range_lock[ZTEST_RANGE_LOCKS];
291 * Per-iteration state.
293 typedef void ztest_func_t(ztest_ds_t *zd, uint64_t id);
295 typedef struct ztest_info {
296 ztest_func_t *zi_func; /* test function */
297 uint64_t zi_iters; /* iterations per execution */
298 uint64_t *zi_interval; /* execute every <interval> seconds */
299 const char *zi_funcname; /* name of test function */
302 typedef struct ztest_shared_callstate {
303 uint64_t zc_count; /* per-pass count */
304 uint64_t zc_time; /* per-pass time */
305 uint64_t zc_next; /* next time to call this function */
306 } ztest_shared_callstate_t;
308 static ztest_shared_callstate_t *ztest_shared_callstate;
309 #define ZTEST_GET_SHARED_CALLSTATE(c) (&ztest_shared_callstate[c])
311 ztest_func_t ztest_dmu_read_write;
312 ztest_func_t ztest_dmu_write_parallel;
313 ztest_func_t ztest_dmu_object_alloc_free;
314 ztest_func_t ztest_dmu_commit_callbacks;
315 ztest_func_t ztest_zap;
316 ztest_func_t ztest_zap_parallel;
317 ztest_func_t ztest_zil_commit;
318 ztest_func_t ztest_zil_remount;
319 ztest_func_t ztest_dmu_read_write_zcopy;
320 ztest_func_t ztest_dmu_objset_create_destroy;
321 ztest_func_t ztest_dmu_prealloc;
322 ztest_func_t ztest_fzap;
323 ztest_func_t ztest_dmu_snapshot_create_destroy;
324 ztest_func_t ztest_dsl_prop_get_set;
325 ztest_func_t ztest_spa_prop_get_set;
326 ztest_func_t ztest_spa_create_destroy;
327 ztest_func_t ztest_fault_inject;
328 ztest_func_t ztest_ddt_repair;
329 ztest_func_t ztest_dmu_snapshot_hold;
330 ztest_func_t ztest_mmp_enable_disable;
331 ztest_func_t ztest_spa_rename;
332 ztest_func_t ztest_scrub;
333 ztest_func_t ztest_dsl_dataset_promote_busy;
334 ztest_func_t ztest_vdev_attach_detach;
335 ztest_func_t ztest_vdev_LUN_growth;
336 ztest_func_t ztest_vdev_add_remove;
337 ztest_func_t ztest_vdev_aux_add_remove;
338 ztest_func_t ztest_split_pool;
339 ztest_func_t ztest_reguid;
340 ztest_func_t ztest_spa_upgrade;
341 ztest_func_t ztest_fletcher;
342 ztest_func_t ztest_fletcher_incr;
343 ztest_func_t ztest_verify_dnode_bt;
345 uint64_t zopt_always = 0ULL * NANOSEC; /* all the time */
346 uint64_t zopt_incessant = 1ULL * NANOSEC / 10; /* every 1/10 second */
347 uint64_t zopt_often = 1ULL * NANOSEC; /* every second */
348 uint64_t zopt_sometimes = 10ULL * NANOSEC; /* every 10 seconds */
349 uint64_t zopt_rarely = 60ULL * NANOSEC; /* every 60 seconds */
351 #define ZTI_INIT(func, iters, interval) \
352 { .zi_func = (func), \
353 .zi_iters = (iters), \
354 .zi_interval = (interval), \
355 .zi_funcname = # func }
357 ztest_info_t ztest_info[] = {
358 ZTI_INIT(ztest_dmu_read_write, 1, &zopt_always),
359 ZTI_INIT(ztest_dmu_write_parallel, 10, &zopt_always),
360 ZTI_INIT(ztest_dmu_object_alloc_free, 1, &zopt_always),
361 ZTI_INIT(ztest_dmu_commit_callbacks, 1, &zopt_always),
362 ZTI_INIT(ztest_zap, 30, &zopt_always),
363 ZTI_INIT(ztest_zap_parallel, 100, &zopt_always),
364 ZTI_INIT(ztest_split_pool, 1, &zopt_always),
365 ZTI_INIT(ztest_zil_commit, 1, &zopt_incessant),
366 ZTI_INIT(ztest_zil_remount, 1, &zopt_sometimes),
367 ZTI_INIT(ztest_dmu_read_write_zcopy, 1, &zopt_often),
368 ZTI_INIT(ztest_dmu_objset_create_destroy, 1, &zopt_often),
369 ZTI_INIT(ztest_dsl_prop_get_set, 1, &zopt_often),
370 ZTI_INIT(ztest_spa_prop_get_set, 1, &zopt_sometimes),
372 ZTI_INIT(ztest_dmu_prealloc, 1, &zopt_sometimes),
374 ZTI_INIT(ztest_fzap, 1, &zopt_sometimes),
375 ZTI_INIT(ztest_dmu_snapshot_create_destroy, 1, &zopt_sometimes),
376 ZTI_INIT(ztest_spa_create_destroy, 1, &zopt_sometimes),
377 ZTI_INIT(ztest_fault_inject, 1, &zopt_sometimes),
378 ZTI_INIT(ztest_ddt_repair, 1, &zopt_sometimes),
379 ZTI_INIT(ztest_dmu_snapshot_hold, 1, &zopt_sometimes),
380 ZTI_INIT(ztest_mmp_enable_disable, 1, &zopt_sometimes),
381 ZTI_INIT(ztest_reguid, 1, &zopt_rarely),
382 ZTI_INIT(ztest_spa_rename, 1, &zopt_rarely),
383 ZTI_INIT(ztest_scrub, 1, &zopt_rarely),
384 ZTI_INIT(ztest_spa_upgrade, 1, &zopt_rarely),
385 ZTI_INIT(ztest_dsl_dataset_promote_busy, 1, &zopt_rarely),
386 ZTI_INIT(ztest_vdev_attach_detach, 1, &zopt_sometimes),
387 ZTI_INIT(ztest_vdev_LUN_growth, 1, &zopt_rarely),
388 ZTI_INIT(ztest_vdev_add_remove, 1, &ztest_opts.zo_vdevtime),
389 ZTI_INIT(ztest_vdev_aux_add_remove, 1, &ztest_opts.zo_vdevtime),
390 ZTI_INIT(ztest_fletcher, 1, &zopt_rarely),
391 ZTI_INIT(ztest_fletcher_incr, 1, &zopt_rarely),
392 ZTI_INIT(ztest_verify_dnode_bt, 1, &zopt_sometimes),
395 #define ZTEST_FUNCS (sizeof (ztest_info) / sizeof (ztest_info_t))
398 * The following struct is used to hold a list of uncalled commit callbacks.
399 * The callbacks are ordered by txg number.
401 typedef struct ztest_cb_list {
402 kmutex_t zcl_callbacks_lock;
403 list_t zcl_callbacks;
407 * Stuff we need to share writably between parent and child.
409 typedef struct ztest_shared {
410 boolean_t zs_do_init;
411 hrtime_t zs_proc_start;
412 hrtime_t zs_proc_stop;
413 hrtime_t zs_thread_start;
414 hrtime_t zs_thread_stop;
415 hrtime_t zs_thread_kill;
416 uint64_t zs_enospc_count;
417 uint64_t zs_vdev_next_leaf;
418 uint64_t zs_vdev_aux;
423 uint64_t zs_metaslab_sz;
424 uint64_t zs_metaslab_df_alloc_threshold;
428 #define ID_PARALLEL -1ULL
430 static char ztest_dev_template[] = "%s/%s.%llua";
431 static char ztest_aux_template[] = "%s/%s.%s.%llu";
432 ztest_shared_t *ztest_shared;
434 static spa_t *ztest_spa = NULL;
435 static ztest_ds_t *ztest_ds;
437 static kmutex_t ztest_vdev_lock;
440 * The ztest_name_lock protects the pool and dataset namespace used by
441 * the individual tests. To modify the namespace, consumers must grab
442 * this lock as writer. Grabbing the lock as reader will ensure that the
443 * namespace does not change while the lock is held.
445 static rwlock_t ztest_name_lock;
447 static boolean_t ztest_dump_core = B_TRUE;
448 static boolean_t ztest_exiting;
450 /* Global commit callback list */
451 static ztest_cb_list_t zcl;
452 /* Commit cb delay */
453 static uint64_t zc_min_txg_delay = UINT64_MAX;
454 static int zc_cb_counter = 0;
457 * Minimum number of commit callbacks that need to be registered for us to check
458 * whether the minimum txg delay is acceptable.
460 #define ZTEST_COMMIT_CB_MIN_REG 100
463 * If a number of txgs equal to this threshold have been created after a commit
464 * callback has been registered but not called, then we assume there is an
465 * implementation bug.
467 #define ZTEST_COMMIT_CB_THRESH (TXG_CONCURRENT_STATES + 1000)
469 extern uint64_t metaslab_gang_bang;
470 extern uint64_t metaslab_df_alloc_threshold;
473 ZTEST_META_DNODE = 0,
478 static void usage(boolean_t) __NORETURN;
481 * These libumem hooks provide a reasonable set of defaults for the allocator's
482 * debugging facilities.
485 _umem_debug_init(void)
487 return ("default,verbose"); /* $UMEM_DEBUG setting */
491 _umem_logging_init(void)
493 return ("fail,contents"); /* $UMEM_LOGGING setting */
496 #define BACKTRACE_SZ 100
498 static void sig_handler(int signo)
500 struct sigaction action;
501 #ifdef __GLIBC__ /* backtrace() is a GNU extension */
503 void *buffer[BACKTRACE_SZ];
505 nptrs = backtrace(buffer, BACKTRACE_SZ);
506 backtrace_symbols_fd(buffer, nptrs, STDERR_FILENO);
510 * Restore default action and re-raise signal so SIGSEGV and
511 * SIGABRT can trigger a core dump.
513 action.sa_handler = SIG_DFL;
514 sigemptyset(&action.sa_mask);
516 (void) sigaction(signo, &action, NULL);
520 #define FATAL_MSG_SZ 1024
525 fatal(int do_perror, char *message, ...)
528 int save_errno = errno;
531 (void) fflush(stdout);
532 buf = umem_alloc(FATAL_MSG_SZ, UMEM_NOFAIL);
534 va_start(args, message);
535 (void) sprintf(buf, "ztest: ");
537 (void) vsprintf(buf + strlen(buf), message, args);
540 (void) snprintf(buf + strlen(buf), FATAL_MSG_SZ - strlen(buf),
541 ": %s", strerror(save_errno));
543 (void) fprintf(stderr, "%s\n", buf);
544 fatal_msg = buf; /* to ease debugging */
551 str2shift(const char *buf)
553 const char *ends = "BKMGTPEZ";
558 for (i = 0; i < strlen(ends); i++) {
559 if (toupper(buf[0]) == ends[i])
562 if (i == strlen(ends)) {
563 (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n",
567 if (buf[1] == '\0' || (toupper(buf[1]) == 'B' && buf[2] == '\0')) {
570 (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n", buf);
576 nicenumtoull(const char *buf)
581 val = strtoull(buf, &end, 0);
583 (void) fprintf(stderr, "ztest: bad numeric value: %s\n", buf);
585 } else if (end[0] == '.') {
586 double fval = strtod(buf, &end);
587 fval *= pow(2, str2shift(end));
588 if (fval > UINT64_MAX) {
589 (void) fprintf(stderr, "ztest: value too large: %s\n",
593 val = (uint64_t)fval;
595 int shift = str2shift(end);
596 if (shift >= 64 || (val << shift) >> shift != val) {
597 (void) fprintf(stderr, "ztest: value too large: %s\n",
607 usage(boolean_t requested)
609 const ztest_shared_opts_t *zo = &ztest_opts_defaults;
611 char nice_vdev_size[10];
612 char nice_gang_bang[10];
613 FILE *fp = requested ? stdout : stderr;
615 nicenum(zo->zo_vdev_size, nice_vdev_size);
616 nicenum(zo->zo_metaslab_gang_bang, nice_gang_bang);
618 (void) fprintf(fp, "Usage: %s\n"
619 "\t[-v vdevs (default: %llu)]\n"
620 "\t[-s size_of_each_vdev (default: %s)]\n"
621 "\t[-a alignment_shift (default: %d)] use 0 for random\n"
622 "\t[-m mirror_copies (default: %d)]\n"
623 "\t[-r raidz_disks (default: %d)]\n"
624 "\t[-R raidz_parity (default: %d)]\n"
625 "\t[-d datasets (default: %d)]\n"
626 "\t[-t threads (default: %d)]\n"
627 "\t[-g gang_block_threshold (default: %s)]\n"
628 "\t[-i init_count (default: %d)] initialize pool i times\n"
629 "\t[-k kill_percentage (default: %llu%%)]\n"
630 "\t[-p pool_name (default: %s)]\n"
631 "\t[-f dir (default: %s)] file directory for vdev files\n"
632 "\t[-M] Multi-host simulate pool imported on remote host\n"
633 "\t[-V] verbose (use multiple times for ever more blather)\n"
634 "\t[-E] use existing pool instead of creating new one\n"
635 "\t[-T time (default: %llu sec)] total run time\n"
636 "\t[-F freezeloops (default: %llu)] max loops in spa_freeze()\n"
637 "\t[-P passtime (default: %llu sec)] time per pass\n"
638 "\t[-B alt_ztest (default: <none>)] alternate ztest path\n"
639 "\t[-o variable=value] ... set global variable to an unsigned\n"
640 "\t 32-bit integer value\n"
641 "\t[-h] (print help)\n"
644 (u_longlong_t)zo->zo_vdevs, /* -v */
645 nice_vdev_size, /* -s */
646 zo->zo_ashift, /* -a */
647 zo->zo_mirrors, /* -m */
648 zo->zo_raidz, /* -r */
649 zo->zo_raidz_parity, /* -R */
650 zo->zo_datasets, /* -d */
651 zo->zo_threads, /* -t */
652 nice_gang_bang, /* -g */
653 zo->zo_init, /* -i */
654 (u_longlong_t)zo->zo_killrate, /* -k */
655 zo->zo_pool, /* -p */
657 (u_longlong_t)zo->zo_time, /* -T */
658 (u_longlong_t)zo->zo_maxloops, /* -F */
659 (u_longlong_t)zo->zo_passtime);
660 exit(requested ? 0 : 1);
664 process_options(int argc, char **argv)
667 ztest_shared_opts_t *zo = &ztest_opts;
671 char altdir[MAXNAMELEN] = { 0 };
673 bcopy(&ztest_opts_defaults, zo, sizeof (*zo));
675 while ((opt = getopt(argc, argv,
676 "v:s:a:m:r:R:d:t:g:i:k:p:f:MVET:P:hF:B:o:")) != EOF) {
693 value = nicenumtoull(optarg);
697 zo->zo_vdevs = value;
700 zo->zo_vdev_size = MAX(SPA_MINDEVSIZE, value);
703 zo->zo_ashift = value;
706 zo->zo_mirrors = value;
709 zo->zo_raidz = MAX(1, value);
712 zo->zo_raidz_parity = MIN(MAX(value, 1), 3);
715 zo->zo_datasets = MAX(1, value);
718 zo->zo_threads = MAX(1, value);
721 zo->zo_metaslab_gang_bang = MAX(SPA_MINBLOCKSIZE << 1,
728 zo->zo_killrate = value;
731 (void) strlcpy(zo->zo_pool, optarg,
732 sizeof (zo->zo_pool));
735 path = realpath(optarg, NULL);
737 (void) fprintf(stderr, "error: %s: %s\n",
738 optarg, strerror(errno));
741 (void) strlcpy(zo->zo_dir, path,
742 sizeof (zo->zo_dir));
759 zo->zo_passtime = MAX(1, value);
762 zo->zo_maxloops = MAX(1, value);
765 (void) strlcpy(altdir, optarg, sizeof (altdir));
768 if (set_global_var(optarg) != 0)
781 zo->zo_raidz_parity = MIN(zo->zo_raidz_parity, zo->zo_raidz - 1);
784 (zo->zo_vdevs > 0 ? zo->zo_time * NANOSEC / zo->zo_vdevs :
787 if (strlen(altdir) > 0) {
795 cmd = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
796 realaltdir = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
798 VERIFY(NULL != realpath(getexecname(), cmd));
799 if (0 != access(altdir, F_OK)) {
800 ztest_dump_core = B_FALSE;
801 fatal(B_TRUE, "invalid alternate ztest path: %s",
804 VERIFY(NULL != realpath(altdir, realaltdir));
807 * 'cmd' should be of the form "<anything>/usr/bin/<isa>/ztest".
808 * We want to extract <isa> to determine if we should use
809 * 32 or 64 bit binaries.
811 bin = strstr(cmd, "/usr/bin/");
812 ztest = strstr(bin, "/ztest");
814 isalen = ztest - isa;
815 (void) snprintf(zo->zo_alt_ztest, sizeof (zo->zo_alt_ztest),
816 "%s/usr/bin/%.*s/ztest", realaltdir, isalen, isa);
817 (void) snprintf(zo->zo_alt_libpath, sizeof (zo->zo_alt_libpath),
818 "%s/usr/lib/%.*s", realaltdir, isalen, isa);
820 if (0 != access(zo->zo_alt_ztest, X_OK)) {
821 ztest_dump_core = B_FALSE;
822 fatal(B_TRUE, "invalid alternate ztest: %s",
824 } else if (0 != access(zo->zo_alt_libpath, X_OK)) {
825 ztest_dump_core = B_FALSE;
826 fatal(B_TRUE, "invalid alternate lib directory %s",
830 umem_free(cmd, MAXPATHLEN);
831 umem_free(realaltdir, MAXPATHLEN);
836 ztest_kill(ztest_shared_t *zs)
838 zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(ztest_spa));
839 zs->zs_space = metaslab_class_get_space(spa_normal_class(ztest_spa));
842 * Before we kill off ztest, make sure that the config is updated.
843 * See comment above spa_config_sync().
845 mutex_enter(&spa_namespace_lock);
846 spa_config_sync(ztest_spa, B_FALSE, B_FALSE);
847 mutex_exit(&spa_namespace_lock);
849 (void) kill(getpid(), SIGKILL);
853 ztest_random(uint64_t range)
857 ASSERT3S(ztest_fd_rand, >=, 0);
862 if (read(ztest_fd_rand, &r, sizeof (r)) != sizeof (r))
863 fatal(1, "short read from /dev/urandom");
870 ztest_record_enospc(const char *s)
872 ztest_shared->zs_enospc_count++;
876 ztest_get_ashift(void)
878 if (ztest_opts.zo_ashift == 0)
879 return (SPA_MINBLOCKSHIFT + ztest_random(5));
880 return (ztest_opts.zo_ashift);
884 make_vdev_file(char *path, char *aux, char *pool, size_t size, uint64_t ashift)
890 pathbuf = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
893 ashift = ztest_get_ashift();
899 vdev = ztest_shared->zs_vdev_aux;
900 (void) snprintf(path, MAXPATHLEN,
901 ztest_aux_template, ztest_opts.zo_dir,
902 pool == NULL ? ztest_opts.zo_pool : pool,
905 vdev = ztest_shared->zs_vdev_next_leaf++;
906 (void) snprintf(path, MAXPATHLEN,
907 ztest_dev_template, ztest_opts.zo_dir,
908 pool == NULL ? ztest_opts.zo_pool : pool, vdev);
913 int fd = open(path, O_RDWR | O_CREAT | O_TRUNC, 0666);
915 fatal(1, "can't open %s", path);
916 if (ftruncate(fd, size) != 0)
917 fatal(1, "can't ftruncate %s", path);
921 VERIFY(nvlist_alloc(&file, NV_UNIQUE_NAME, 0) == 0);
922 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_TYPE, VDEV_TYPE_FILE) == 0);
923 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_PATH, path) == 0);
924 VERIFY(nvlist_add_uint64(file, ZPOOL_CONFIG_ASHIFT, ashift) == 0);
925 umem_free(pathbuf, MAXPATHLEN);
931 make_vdev_raidz(char *path, char *aux, char *pool, size_t size,
932 uint64_t ashift, int r)
934 nvlist_t *raidz, **child;
938 return (make_vdev_file(path, aux, pool, size, ashift));
939 child = umem_alloc(r * sizeof (nvlist_t *), UMEM_NOFAIL);
941 for (c = 0; c < r; c++)
942 child[c] = make_vdev_file(path, aux, pool, size, ashift);
944 VERIFY(nvlist_alloc(&raidz, NV_UNIQUE_NAME, 0) == 0);
945 VERIFY(nvlist_add_string(raidz, ZPOOL_CONFIG_TYPE,
946 VDEV_TYPE_RAIDZ) == 0);
947 VERIFY(nvlist_add_uint64(raidz, ZPOOL_CONFIG_NPARITY,
948 ztest_opts.zo_raidz_parity) == 0);
949 VERIFY(nvlist_add_nvlist_array(raidz, ZPOOL_CONFIG_CHILDREN,
952 for (c = 0; c < r; c++)
953 nvlist_free(child[c]);
955 umem_free(child, r * sizeof (nvlist_t *));
961 make_vdev_mirror(char *path, char *aux, char *pool, size_t size,
962 uint64_t ashift, int r, int m)
964 nvlist_t *mirror, **child;
968 return (make_vdev_raidz(path, aux, pool, size, ashift, r));
970 child = umem_alloc(m * sizeof (nvlist_t *), UMEM_NOFAIL);
972 for (c = 0; c < m; c++)
973 child[c] = make_vdev_raidz(path, aux, pool, size, ashift, r);
975 VERIFY(nvlist_alloc(&mirror, NV_UNIQUE_NAME, 0) == 0);
976 VERIFY(nvlist_add_string(mirror, ZPOOL_CONFIG_TYPE,
977 VDEV_TYPE_MIRROR) == 0);
978 VERIFY(nvlist_add_nvlist_array(mirror, ZPOOL_CONFIG_CHILDREN,
981 for (c = 0; c < m; c++)
982 nvlist_free(child[c]);
984 umem_free(child, m * sizeof (nvlist_t *));
990 make_vdev_root(char *path, char *aux, char *pool, size_t size, uint64_t ashift,
991 int log, int r, int m, int t)
993 nvlist_t *root, **child;
998 child = umem_alloc(t * sizeof (nvlist_t *), UMEM_NOFAIL);
1000 for (c = 0; c < t; c++) {
1001 child[c] = make_vdev_mirror(path, aux, pool, size, ashift,
1003 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_IS_LOG,
1007 VERIFY(nvlist_alloc(&root, NV_UNIQUE_NAME, 0) == 0);
1008 VERIFY(nvlist_add_string(root, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) == 0);
1009 VERIFY(nvlist_add_nvlist_array(root, aux ? aux : ZPOOL_CONFIG_CHILDREN,
1012 for (c = 0; c < t; c++)
1013 nvlist_free(child[c]);
1015 umem_free(child, t * sizeof (nvlist_t *));
1021 * Find a random spa version. Returns back a random spa version in the
1022 * range [initial_version, SPA_VERSION_FEATURES].
1025 ztest_random_spa_version(uint64_t initial_version)
1027 uint64_t version = initial_version;
1029 if (version <= SPA_VERSION_BEFORE_FEATURES) {
1031 ztest_random(SPA_VERSION_BEFORE_FEATURES - version + 1);
1034 if (version > SPA_VERSION_BEFORE_FEATURES)
1035 version = SPA_VERSION_FEATURES;
1037 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
1042 ztest_random_blocksize(void)
1045 * Choose a block size >= the ashift.
1046 * If the SPA supports new MAXBLOCKSIZE, test up to 1MB blocks.
1048 int maxbs = SPA_OLD_MAXBLOCKSHIFT;
1049 if (spa_maxblocksize(ztest_spa) == SPA_MAXBLOCKSIZE)
1051 uint64_t block_shift =
1052 ztest_random(maxbs - ztest_spa->spa_max_ashift + 1);
1053 return (1 << (SPA_MINBLOCKSHIFT + block_shift));
1057 ztest_random_dnodesize(void)
1060 int max_slots = spa_maxdnodesize(ztest_spa) >> DNODE_SHIFT;
1062 if (max_slots == DNODE_MIN_SLOTS)
1063 return (DNODE_MIN_SIZE);
1066 * Weight the random distribution more heavily toward smaller
1067 * dnode sizes since that is more likely to reflect real-world
1070 ASSERT3U(max_slots, >, 4);
1071 switch (ztest_random(10)) {
1073 slots = 5 + ztest_random(max_slots - 4);
1076 slots = 2 + ztest_random(3);
1083 return (slots << DNODE_SHIFT);
1087 ztest_random_ibshift(void)
1089 return (DN_MIN_INDBLKSHIFT +
1090 ztest_random(DN_MAX_INDBLKSHIFT - DN_MIN_INDBLKSHIFT + 1));
1094 ztest_random_vdev_top(spa_t *spa, boolean_t log_ok)
1097 vdev_t *rvd = spa->spa_root_vdev;
1100 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1103 top = ztest_random(rvd->vdev_children);
1104 tvd = rvd->vdev_child[top];
1105 } while (tvd->vdev_ishole || (tvd->vdev_islog && !log_ok) ||
1106 tvd->vdev_mg == NULL || tvd->vdev_mg->mg_class == NULL);
1112 ztest_random_dsl_prop(zfs_prop_t prop)
1117 value = zfs_prop_random_value(prop, ztest_random(-1ULL));
1118 } while (prop == ZFS_PROP_CHECKSUM && value == ZIO_CHECKSUM_OFF);
1124 ztest_dsl_prop_set_uint64(char *osname, zfs_prop_t prop, uint64_t value,
1127 const char *propname = zfs_prop_to_name(prop);
1128 const char *valname;
1133 error = dsl_prop_set_int(osname, propname,
1134 (inherit ? ZPROP_SRC_NONE : ZPROP_SRC_LOCAL), value);
1136 if (error == ENOSPC) {
1137 ztest_record_enospc(FTAG);
1142 setpoint = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
1143 VERIFY0(dsl_prop_get_integer(osname, propname, &curval, setpoint));
1145 if (ztest_opts.zo_verbose >= 6) {
1148 err = zfs_prop_index_to_string(prop, curval, &valname);
1150 (void) printf("%s %s = %llu at '%s'\n", osname,
1151 propname, (unsigned long long)curval, setpoint);
1153 (void) printf("%s %s = %s at '%s'\n",
1154 osname, propname, valname, setpoint);
1156 umem_free(setpoint, MAXPATHLEN);
1162 ztest_spa_prop_set_uint64(zpool_prop_t prop, uint64_t value)
1164 spa_t *spa = ztest_spa;
1165 nvlist_t *props = NULL;
1168 VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
1169 VERIFY(nvlist_add_uint64(props, zpool_prop_to_name(prop), value) == 0);
1171 error = spa_prop_set(spa, props);
1175 if (error == ENOSPC) {
1176 ztest_record_enospc(FTAG);
1185 ztest_dmu_objset_own(const char *name, dmu_objset_type_t type,
1186 boolean_t readonly, boolean_t decrypt, void *tag, objset_t **osp)
1190 err = dmu_objset_own(name, type, readonly, decrypt, tag, osp);
1191 if (decrypt && err == EACCES) {
1192 char ddname[ZFS_MAX_DATASET_NAME_LEN];
1193 dsl_crypto_params_t *dcp;
1194 nvlist_t *crypto_args = fnvlist_alloc();
1197 /* spa_keystore_load_wkey() expects a dsl dir name */
1198 strcpy(ddname, name);
1199 cp = strchr(ddname, '@');
1203 fnvlist_add_uint8_array(crypto_args, "wkeydata",
1204 (uint8_t *)ztest_wkeydata, WRAPPING_KEY_LEN);
1205 VERIFY0(dsl_crypto_params_create_nvlist(DCP_CMD_NONE, NULL,
1206 crypto_args, &dcp));
1207 err = spa_keystore_load_wkey(ddname, dcp, B_FALSE);
1208 dsl_crypto_params_free(dcp, B_FALSE);
1209 fnvlist_free(crypto_args);
1214 err = dmu_objset_own(name, type, readonly, decrypt, tag, osp);
1222 * Object and range lock mechanics
1225 list_node_t z_lnode;
1226 refcount_t z_refcnt;
1228 zfs_rlock_t z_range_lock;
1233 ztest_znode_t *z_ztznode;
1236 static ztest_znode_t *
1237 ztest_znode_init(uint64_t object)
1239 ztest_znode_t *zp = umem_alloc(sizeof (*zp), UMEM_NOFAIL);
1241 list_link_init(&zp->z_lnode);
1242 refcount_create(&zp->z_refcnt);
1243 zp->z_object = object;
1244 zfs_rlock_init(&zp->z_range_lock);
1250 ztest_znode_fini(ztest_znode_t *zp)
1252 ASSERT(refcount_is_zero(&zp->z_refcnt));
1253 zfs_rlock_destroy(&zp->z_range_lock);
1255 refcount_destroy(&zp->z_refcnt);
1256 list_link_init(&zp->z_lnode);
1257 umem_free(zp, sizeof (*zp));
1261 ztest_zll_init(zll_t *zll)
1263 mutex_init(&zll->z_lock, NULL, MUTEX_DEFAULT, NULL);
1264 list_create(&zll->z_list, sizeof (ztest_znode_t),
1265 offsetof(ztest_znode_t, z_lnode));
1269 ztest_zll_destroy(zll_t *zll)
1271 list_destroy(&zll->z_list);
1272 mutex_destroy(&zll->z_lock);
1275 #define RL_TAG "range_lock"
1276 static ztest_znode_t *
1277 ztest_znode_get(ztest_ds_t *zd, uint64_t object)
1279 zll_t *zll = &zd->zd_range_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1280 ztest_znode_t *zp = NULL;
1281 mutex_enter(&zll->z_lock);
1282 for (zp = list_head(&zll->z_list); (zp);
1283 zp = list_next(&zll->z_list, zp)) {
1284 if (zp->z_object == object) {
1285 refcount_add(&zp->z_refcnt, RL_TAG);
1290 zp = ztest_znode_init(object);
1291 refcount_add(&zp->z_refcnt, RL_TAG);
1292 list_insert_head(&zll->z_list, zp);
1294 mutex_exit(&zll->z_lock);
1299 ztest_znode_put(ztest_ds_t *zd, ztest_znode_t *zp)
1302 ASSERT3U(zp->z_object, !=, 0);
1303 zll = &zd->zd_range_lock[zp->z_object & (ZTEST_OBJECT_LOCKS - 1)];
1304 mutex_enter(&zll->z_lock);
1305 refcount_remove(&zp->z_refcnt, RL_TAG);
1306 if (refcount_is_zero(&zp->z_refcnt)) {
1307 list_remove(&zll->z_list, zp);
1308 ztest_znode_fini(zp);
1310 mutex_exit(&zll->z_lock);
1315 ztest_rll_init(rll_t *rll)
1317 rll->rll_writer = NULL;
1318 rll->rll_readers = 0;
1319 mutex_init(&rll->rll_lock, NULL, MUTEX_DEFAULT, NULL);
1320 cv_init(&rll->rll_cv, NULL, CV_DEFAULT, NULL);
1324 ztest_rll_destroy(rll_t *rll)
1326 ASSERT(rll->rll_writer == NULL);
1327 ASSERT(rll->rll_readers == 0);
1328 mutex_destroy(&rll->rll_lock);
1329 cv_destroy(&rll->rll_cv);
1333 ztest_rll_lock(rll_t *rll, rl_type_t type)
1335 mutex_enter(&rll->rll_lock);
1337 if (type == RL_READER) {
1338 while (rll->rll_writer != NULL)
1339 (void) cv_wait(&rll->rll_cv, &rll->rll_lock);
1342 while (rll->rll_writer != NULL || rll->rll_readers)
1343 (void) cv_wait(&rll->rll_cv, &rll->rll_lock);
1344 rll->rll_writer = curthread;
1347 mutex_exit(&rll->rll_lock);
1351 ztest_rll_unlock(rll_t *rll)
1353 mutex_enter(&rll->rll_lock);
1355 if (rll->rll_writer) {
1356 ASSERT(rll->rll_readers == 0);
1357 rll->rll_writer = NULL;
1359 ASSERT(rll->rll_readers != 0);
1360 ASSERT(rll->rll_writer == NULL);
1364 if (rll->rll_writer == NULL && rll->rll_readers == 0)
1365 cv_broadcast(&rll->rll_cv);
1367 mutex_exit(&rll->rll_lock);
1371 ztest_object_lock(ztest_ds_t *zd, uint64_t object, rl_type_t type)
1373 rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1375 ztest_rll_lock(rll, type);
1379 ztest_object_unlock(ztest_ds_t *zd, uint64_t object)
1381 rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1383 ztest_rll_unlock(rll);
1386 static ztest_zrl_t *
1387 ztest_zrl_init(rl_t *rl, ztest_znode_t *zp)
1389 ztest_zrl_t *zrl = umem_alloc(sizeof (*zrl), UMEM_NOFAIL);
1391 zrl->z_ztznode = zp;
1396 ztest_zrl_fini(ztest_zrl_t *zrl)
1398 umem_free(zrl, sizeof (*zrl));
1401 static ztest_zrl_t *
1402 ztest_range_lock(ztest_ds_t *zd, uint64_t object, uint64_t offset,
1403 uint64_t size, rl_type_t type)
1405 ztest_znode_t *zp = ztest_znode_get(zd, object);
1406 rl_t *rl = zfs_range_lock(&zp->z_range_lock, offset,
1408 return (ztest_zrl_init(rl, zp));
1412 ztest_range_unlock(ztest_ds_t *zd, ztest_zrl_t *zrl)
1414 zfs_range_unlock(zrl->z_rl);
1415 ztest_znode_put(zd, zrl->z_ztznode);
1416 ztest_zrl_fini(zrl);
1420 ztest_zd_init(ztest_ds_t *zd, ztest_shared_ds_t *szd, objset_t *os)
1423 zd->zd_zilog = dmu_objset_zil(os);
1424 zd->zd_shared = szd;
1425 dmu_objset_name(os, zd->zd_name);
1428 if (zd->zd_shared != NULL)
1429 zd->zd_shared->zd_seq = 0;
1431 VERIFY(rwlock_init(&zd->zd_zilog_lock, USYNC_THREAD, NULL) == 0);
1432 mutex_init(&zd->zd_dirobj_lock, NULL, MUTEX_DEFAULT, NULL);
1434 for (l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1435 ztest_rll_init(&zd->zd_object_lock[l]);
1437 for (l = 0; l < ZTEST_RANGE_LOCKS; l++)
1438 ztest_zll_init(&zd->zd_range_lock[l]);
1442 ztest_zd_fini(ztest_ds_t *zd)
1446 mutex_destroy(&zd->zd_dirobj_lock);
1447 (void) rwlock_destroy(&zd->zd_zilog_lock);
1449 for (l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1450 ztest_rll_destroy(&zd->zd_object_lock[l]);
1452 for (l = 0; l < ZTEST_RANGE_LOCKS; l++)
1453 ztest_zll_destroy(&zd->zd_range_lock[l]);
1456 #define TXG_MIGHTWAIT (ztest_random(10) == 0 ? TXG_NOWAIT : TXG_WAIT)
1459 ztest_tx_assign(dmu_tx_t *tx, uint64_t txg_how, const char *tag)
1465 * Attempt to assign tx to some transaction group.
1467 error = dmu_tx_assign(tx, txg_how);
1469 if (error == ERESTART) {
1470 ASSERT(txg_how == TXG_NOWAIT);
1473 ASSERT3U(error, ==, ENOSPC);
1474 ztest_record_enospc(tag);
1479 txg = dmu_tx_get_txg(tx);
1485 ztest_pattern_set(void *buf, uint64_t size, uint64_t value)
1488 uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1496 ztest_pattern_match(void *buf, uint64_t size, uint64_t value)
1499 uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1503 diff |= (value - *ip++);
1510 ztest_bt_generate(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1511 uint64_t dnodesize, uint64_t offset, uint64_t gen, uint64_t txg,
1514 bt->bt_magic = BT_MAGIC;
1515 bt->bt_objset = dmu_objset_id(os);
1516 bt->bt_object = object;
1517 bt->bt_dnodesize = dnodesize;
1518 bt->bt_offset = offset;
1521 bt->bt_crtxg = crtxg;
1525 ztest_bt_verify(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1526 uint64_t dnodesize, uint64_t offset, uint64_t gen, uint64_t txg,
1529 ASSERT3U(bt->bt_magic, ==, BT_MAGIC);
1530 ASSERT3U(bt->bt_objset, ==, dmu_objset_id(os));
1531 ASSERT3U(bt->bt_object, ==, object);
1532 ASSERT3U(bt->bt_dnodesize, ==, dnodesize);
1533 ASSERT3U(bt->bt_offset, ==, offset);
1534 ASSERT3U(bt->bt_gen, <=, gen);
1535 ASSERT3U(bt->bt_txg, <=, txg);
1536 ASSERT3U(bt->bt_crtxg, ==, crtxg);
1539 static ztest_block_tag_t *
1540 ztest_bt_bonus(dmu_buf_t *db)
1542 dmu_object_info_t doi;
1543 ztest_block_tag_t *bt;
1545 dmu_object_info_from_db(db, &doi);
1546 ASSERT3U(doi.doi_bonus_size, <=, db->db_size);
1547 ASSERT3U(doi.doi_bonus_size, >=, sizeof (*bt));
1548 bt = (void *)((char *)db->db_data + doi.doi_bonus_size - sizeof (*bt));
1554 * Generate a token to fill up unused bonus buffer space. Try to make
1555 * it unique to the object, generation, and offset to verify that data
1556 * is not getting overwritten by data from other dnodes.
1558 #define ZTEST_BONUS_FILL_TOKEN(obj, ds, gen, offset) \
1559 (((ds) << 48) | ((gen) << 32) | ((obj) << 8) | (offset))
1562 * Fill up the unused bonus buffer region before the block tag with a
1563 * verifiable pattern. Filling the whole bonus area with non-zero data
1564 * helps ensure that all dnode traversal code properly skips the
1565 * interior regions of large dnodes.
1568 ztest_fill_unused_bonus(dmu_buf_t *db, void *end, uint64_t obj,
1569 objset_t *os, uint64_t gen)
1573 ASSERT(IS_P2ALIGNED((char *)end - (char *)db->db_data, 8));
1575 for (bonusp = db->db_data; bonusp < (uint64_t *)end; bonusp++) {
1576 uint64_t token = ZTEST_BONUS_FILL_TOKEN(obj, dmu_objset_id(os),
1577 gen, bonusp - (uint64_t *)db->db_data);
1583 * Verify that the unused area of a bonus buffer is filled with the
1587 ztest_verify_unused_bonus(dmu_buf_t *db, void *end, uint64_t obj,
1588 objset_t *os, uint64_t gen)
1592 for (bonusp = db->db_data; bonusp < (uint64_t *)end; bonusp++) {
1593 uint64_t token = ZTEST_BONUS_FILL_TOKEN(obj, dmu_objset_id(os),
1594 gen, bonusp - (uint64_t *)db->db_data);
1595 VERIFY3U(*bonusp, ==, token);
1603 #define lrz_type lr_mode
1604 #define lrz_blocksize lr_uid
1605 #define lrz_ibshift lr_gid
1606 #define lrz_bonustype lr_rdev
1607 #define lrz_dnodesize lr_crtime[1]
1610 ztest_log_create(ztest_ds_t *zd, dmu_tx_t *tx, lr_create_t *lr)
1612 char *name = (void *)(lr + 1); /* name follows lr */
1613 size_t namesize = strlen(name) + 1;
1616 if (zil_replaying(zd->zd_zilog, tx))
1619 itx = zil_itx_create(TX_CREATE, sizeof (*lr) + namesize);
1620 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1621 sizeof (*lr) + namesize - sizeof (lr_t));
1623 zil_itx_assign(zd->zd_zilog, itx, tx);
1627 ztest_log_remove(ztest_ds_t *zd, dmu_tx_t *tx, lr_remove_t *lr, uint64_t object)
1629 char *name = (void *)(lr + 1); /* name follows lr */
1630 size_t namesize = strlen(name) + 1;
1633 if (zil_replaying(zd->zd_zilog, tx))
1636 itx = zil_itx_create(TX_REMOVE, sizeof (*lr) + namesize);
1637 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1638 sizeof (*lr) + namesize - sizeof (lr_t));
1640 itx->itx_oid = object;
1641 zil_itx_assign(zd->zd_zilog, itx, tx);
1645 ztest_log_write(ztest_ds_t *zd, dmu_tx_t *tx, lr_write_t *lr)
1648 itx_wr_state_t write_state = ztest_random(WR_NUM_STATES);
1650 if (zil_replaying(zd->zd_zilog, tx))
1653 if (lr->lr_length > ZIL_MAX_LOG_DATA)
1654 write_state = WR_INDIRECT;
1656 itx = zil_itx_create(TX_WRITE,
1657 sizeof (*lr) + (write_state == WR_COPIED ? lr->lr_length : 0));
1659 if (write_state == WR_COPIED &&
1660 dmu_read(zd->zd_os, lr->lr_foid, lr->lr_offset, lr->lr_length,
1661 ((lr_write_t *)&itx->itx_lr) + 1, DMU_READ_NO_PREFETCH) != 0) {
1662 zil_itx_destroy(itx);
1663 itx = zil_itx_create(TX_WRITE, sizeof (*lr));
1664 write_state = WR_NEED_COPY;
1666 itx->itx_private = zd;
1667 itx->itx_wr_state = write_state;
1668 itx->itx_sync = (ztest_random(8) == 0);
1670 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1671 sizeof (*lr) - sizeof (lr_t));
1673 zil_itx_assign(zd->zd_zilog, itx, tx);
1677 ztest_log_truncate(ztest_ds_t *zd, dmu_tx_t *tx, lr_truncate_t *lr)
1681 if (zil_replaying(zd->zd_zilog, tx))
1684 itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr));
1685 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1686 sizeof (*lr) - sizeof (lr_t));
1688 itx->itx_sync = B_FALSE;
1689 zil_itx_assign(zd->zd_zilog, itx, tx);
1693 ztest_log_setattr(ztest_ds_t *zd, dmu_tx_t *tx, lr_setattr_t *lr)
1697 if (zil_replaying(zd->zd_zilog, tx))
1700 itx = zil_itx_create(TX_SETATTR, sizeof (*lr));
1701 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1702 sizeof (*lr) - sizeof (lr_t));
1704 itx->itx_sync = B_FALSE;
1705 zil_itx_assign(zd->zd_zilog, itx, tx);
1712 ztest_replay_create(ztest_ds_t *zd, lr_create_t *lr, boolean_t byteswap)
1714 char *name = (void *)(lr + 1); /* name follows lr */
1715 objset_t *os = zd->zd_os;
1716 ztest_block_tag_t *bbt;
1724 byteswap_uint64_array(lr, sizeof (*lr));
1726 ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1727 ASSERT(name[0] != '\0');
1729 tx = dmu_tx_create(os);
1731 dmu_tx_hold_zap(tx, lr->lr_doid, B_TRUE, name);
1733 if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1734 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, B_TRUE, NULL);
1736 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1739 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1743 ASSERT(dmu_objset_zil(os)->zl_replay == !!lr->lr_foid);
1744 bonuslen = DN_BONUS_SIZE(lr->lrz_dnodesize);
1746 if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1747 if (lr->lr_foid == 0) {
1748 lr->lr_foid = zap_create_dnsize(os,
1749 lr->lrz_type, lr->lrz_bonustype,
1750 bonuslen, lr->lrz_dnodesize, tx);
1752 error = zap_create_claim_dnsize(os, lr->lr_foid,
1753 lr->lrz_type, lr->lrz_bonustype,
1754 bonuslen, lr->lrz_dnodesize, tx);
1757 if (lr->lr_foid == 0) {
1758 lr->lr_foid = dmu_object_alloc_dnsize(os,
1759 lr->lrz_type, 0, lr->lrz_bonustype,
1760 bonuslen, lr->lrz_dnodesize, tx);
1762 error = dmu_object_claim_dnsize(os, lr->lr_foid,
1763 lr->lrz_type, 0, lr->lrz_bonustype,
1764 bonuslen, lr->lrz_dnodesize, tx);
1769 ASSERT3U(error, ==, EEXIST);
1770 ASSERT(zd->zd_zilog->zl_replay);
1775 ASSERT(lr->lr_foid != 0);
1777 if (lr->lrz_type != DMU_OT_ZAP_OTHER)
1778 VERIFY3U(0, ==, dmu_object_set_blocksize(os, lr->lr_foid,
1779 lr->lrz_blocksize, lr->lrz_ibshift, tx));
1781 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1782 bbt = ztest_bt_bonus(db);
1783 dmu_buf_will_dirty(db, tx);
1784 ztest_bt_generate(bbt, os, lr->lr_foid, lr->lrz_dnodesize, -1ULL,
1785 lr->lr_gen, txg, txg);
1786 ztest_fill_unused_bonus(db, bbt, lr->lr_foid, os, lr->lr_gen);
1787 dmu_buf_rele(db, FTAG);
1789 VERIFY3U(0, ==, zap_add(os, lr->lr_doid, name, sizeof (uint64_t), 1,
1792 (void) ztest_log_create(zd, tx, lr);
1800 ztest_replay_remove(ztest_ds_t *zd, lr_remove_t *lr, boolean_t byteswap)
1802 char *name = (void *)(lr + 1); /* name follows lr */
1803 objset_t *os = zd->zd_os;
1804 dmu_object_info_t doi;
1806 uint64_t object, txg;
1809 byteswap_uint64_array(lr, sizeof (*lr));
1811 ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1812 ASSERT(name[0] != '\0');
1815 zap_lookup(os, lr->lr_doid, name, sizeof (object), 1, &object));
1816 ASSERT(object != 0);
1818 ztest_object_lock(zd, object, RL_WRITER);
1820 VERIFY3U(0, ==, dmu_object_info(os, object, &doi));
1822 tx = dmu_tx_create(os);
1824 dmu_tx_hold_zap(tx, lr->lr_doid, B_FALSE, name);
1825 dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END);
1827 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1829 ztest_object_unlock(zd, object);
1833 if (doi.doi_type == DMU_OT_ZAP_OTHER) {
1834 VERIFY3U(0, ==, zap_destroy(os, object, tx));
1836 VERIFY3U(0, ==, dmu_object_free(os, object, tx));
1839 VERIFY3U(0, ==, zap_remove(os, lr->lr_doid, name, tx));
1841 (void) ztest_log_remove(zd, tx, lr, object);
1845 ztest_object_unlock(zd, object);
1851 ztest_replay_write(ztest_ds_t *zd, lr_write_t *lr, boolean_t byteswap)
1853 objset_t *os = zd->zd_os;
1854 void *data = lr + 1; /* data follows lr */
1855 uint64_t offset, length;
1856 ztest_block_tag_t *bt = data;
1857 ztest_block_tag_t *bbt;
1858 uint64_t gen, txg, lrtxg, crtxg;
1859 dmu_object_info_t doi;
1862 arc_buf_t *abuf = NULL;
1866 byteswap_uint64_array(lr, sizeof (*lr));
1868 offset = lr->lr_offset;
1869 length = lr->lr_length;
1871 /* If it's a dmu_sync() block, write the whole block */
1872 if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
1873 uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
1874 if (length < blocksize) {
1875 offset -= offset % blocksize;
1880 if (bt->bt_magic == BSWAP_64(BT_MAGIC))
1881 byteswap_uint64_array(bt, sizeof (*bt));
1883 if (bt->bt_magic != BT_MAGIC)
1886 ztest_object_lock(zd, lr->lr_foid, RL_READER);
1887 rl = ztest_range_lock(zd, lr->lr_foid, offset, length, RL_WRITER);
1889 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1891 dmu_object_info_from_db(db, &doi);
1893 bbt = ztest_bt_bonus(db);
1894 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1896 crtxg = bbt->bt_crtxg;
1897 lrtxg = lr->lr_common.lrc_txg;
1899 tx = dmu_tx_create(os);
1901 dmu_tx_hold_write(tx, lr->lr_foid, offset, length);
1903 if (ztest_random(8) == 0 && length == doi.doi_data_block_size &&
1904 P2PHASE(offset, length) == 0)
1905 abuf = dmu_request_arcbuf(db, length);
1907 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1910 dmu_return_arcbuf(abuf);
1911 dmu_buf_rele(db, FTAG);
1912 ztest_range_unlock(zd, rl);
1913 ztest_object_unlock(zd, lr->lr_foid);
1919 * Usually, verify the old data before writing new data --
1920 * but not always, because we also want to verify correct
1921 * behavior when the data was not recently read into cache.
1923 ASSERT(offset % doi.doi_data_block_size == 0);
1924 if (ztest_random(4) != 0) {
1925 int prefetch = ztest_random(2) ?
1926 DMU_READ_PREFETCH : DMU_READ_NO_PREFETCH;
1927 ztest_block_tag_t rbt;
1929 VERIFY(dmu_read(os, lr->lr_foid, offset,
1930 sizeof (rbt), &rbt, prefetch) == 0);
1931 if (rbt.bt_magic == BT_MAGIC) {
1932 ztest_bt_verify(&rbt, os, lr->lr_foid, 0,
1933 offset, gen, txg, crtxg);
1938 * Writes can appear to be newer than the bonus buffer because
1939 * the ztest_get_data() callback does a dmu_read() of the
1940 * open-context data, which may be different than the data
1941 * as it was when the write was generated.
1943 if (zd->zd_zilog->zl_replay) {
1944 ztest_bt_verify(bt, os, lr->lr_foid, 0, offset,
1945 MAX(gen, bt->bt_gen), MAX(txg, lrtxg),
1950 * Set the bt's gen/txg to the bonus buffer's gen/txg
1951 * so that all of the usual ASSERTs will work.
1953 ztest_bt_generate(bt, os, lr->lr_foid, 0, offset, gen, txg,
1958 dmu_write(os, lr->lr_foid, offset, length, data, tx);
1960 bcopy(data, abuf->b_data, length);
1961 dmu_assign_arcbuf_by_dbuf(db, offset, abuf, tx);
1964 (void) ztest_log_write(zd, tx, lr);
1966 dmu_buf_rele(db, FTAG);
1970 ztest_range_unlock(zd, rl);
1971 ztest_object_unlock(zd, lr->lr_foid);
1977 ztest_replay_truncate(ztest_ds_t *zd, lr_truncate_t *lr, boolean_t byteswap)
1979 objset_t *os = zd->zd_os;
1985 byteswap_uint64_array(lr, sizeof (*lr));
1987 ztest_object_lock(zd, lr->lr_foid, RL_READER);
1988 rl = ztest_range_lock(zd, lr->lr_foid, lr->lr_offset, lr->lr_length,
1991 tx = dmu_tx_create(os);
1993 dmu_tx_hold_free(tx, lr->lr_foid, lr->lr_offset, lr->lr_length);
1995 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1997 ztest_range_unlock(zd, rl);
1998 ztest_object_unlock(zd, lr->lr_foid);
2002 VERIFY(dmu_free_range(os, lr->lr_foid, lr->lr_offset,
2003 lr->lr_length, tx) == 0);
2005 (void) ztest_log_truncate(zd, tx, lr);
2009 ztest_range_unlock(zd, rl);
2010 ztest_object_unlock(zd, lr->lr_foid);
2016 ztest_replay_setattr(ztest_ds_t *zd, lr_setattr_t *lr, boolean_t byteswap)
2018 objset_t *os = zd->zd_os;
2021 ztest_block_tag_t *bbt;
2022 uint64_t txg, lrtxg, crtxg, dnodesize;
2025 byteswap_uint64_array(lr, sizeof (*lr));
2027 ztest_object_lock(zd, lr->lr_foid, RL_WRITER);
2029 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
2031 tx = dmu_tx_create(os);
2032 dmu_tx_hold_bonus(tx, lr->lr_foid);
2034 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
2036 dmu_buf_rele(db, FTAG);
2037 ztest_object_unlock(zd, lr->lr_foid);
2041 bbt = ztest_bt_bonus(db);
2042 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
2043 crtxg = bbt->bt_crtxg;
2044 lrtxg = lr->lr_common.lrc_txg;
2045 dnodesize = bbt->bt_dnodesize;
2047 if (zd->zd_zilog->zl_replay) {
2048 ASSERT(lr->lr_size != 0);
2049 ASSERT(lr->lr_mode != 0);
2053 * Randomly change the size and increment the generation.
2055 lr->lr_size = (ztest_random(db->db_size / sizeof (*bbt)) + 1) *
2057 lr->lr_mode = bbt->bt_gen + 1;
2062 * Verify that the current bonus buffer is not newer than our txg.
2064 ztest_bt_verify(bbt, os, lr->lr_foid, dnodesize, -1ULL, lr->lr_mode,
2065 MAX(txg, lrtxg), crtxg);
2067 dmu_buf_will_dirty(db, tx);
2069 ASSERT3U(lr->lr_size, >=, sizeof (*bbt));
2070 ASSERT3U(lr->lr_size, <=, db->db_size);
2071 VERIFY0(dmu_set_bonus(db, lr->lr_size, tx));
2072 bbt = ztest_bt_bonus(db);
2074 ztest_bt_generate(bbt, os, lr->lr_foid, dnodesize, -1ULL, lr->lr_mode,
2076 ztest_fill_unused_bonus(db, bbt, lr->lr_foid, os, bbt->bt_gen);
2077 dmu_buf_rele(db, FTAG);
2079 (void) ztest_log_setattr(zd, tx, lr);
2083 ztest_object_unlock(zd, lr->lr_foid);
2088 zil_replay_func_t ztest_replay_vector[TX_MAX_TYPE] = {
2089 NULL, /* 0 no such transaction type */
2090 (zil_replay_func_t)ztest_replay_create, /* TX_CREATE */
2091 NULL, /* TX_MKDIR */
2092 NULL, /* TX_MKXATTR */
2093 NULL, /* TX_SYMLINK */
2094 (zil_replay_func_t)ztest_replay_remove, /* TX_REMOVE */
2095 NULL, /* TX_RMDIR */
2097 NULL, /* TX_RENAME */
2098 (zil_replay_func_t)ztest_replay_write, /* TX_WRITE */
2099 (zil_replay_func_t)ztest_replay_truncate, /* TX_TRUNCATE */
2100 (zil_replay_func_t)ztest_replay_setattr, /* TX_SETATTR */
2102 NULL, /* TX_CREATE_ACL */
2103 NULL, /* TX_CREATE_ATTR */
2104 NULL, /* TX_CREATE_ACL_ATTR */
2105 NULL, /* TX_MKDIR_ACL */
2106 NULL, /* TX_MKDIR_ATTR */
2107 NULL, /* TX_MKDIR_ACL_ATTR */
2108 NULL, /* TX_WRITE2 */
2112 * ZIL get_data callbacks
2114 typedef struct ztest_zgd_private {
2118 } ztest_zgd_private_t;
2121 ztest_get_done(zgd_t *zgd, int error)
2123 ztest_zgd_private_t *zzp = zgd->zgd_private;
2124 ztest_ds_t *zd = zzp->z_zd;
2125 uint64_t object = zzp->z_object;
2128 dmu_buf_rele(zgd->zgd_db, zgd);
2130 ztest_range_unlock(zd, zzp->z_rl);
2131 ztest_object_unlock(zd, object);
2133 if (error == 0 && zgd->zgd_bp)
2134 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
2136 umem_free(zgd, sizeof (*zgd));
2137 umem_free(zzp, sizeof (*zzp));
2141 ztest_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
2143 ztest_ds_t *zd = arg;
2144 objset_t *os = zd->zd_os;
2145 uint64_t object = lr->lr_foid;
2146 uint64_t offset = lr->lr_offset;
2147 uint64_t size = lr->lr_length;
2148 uint64_t txg = lr->lr_common.lrc_txg;
2150 dmu_object_info_t doi;
2154 ztest_zgd_private_t *zgd_private;
2156 ztest_object_lock(zd, object, RL_READER);
2157 error = dmu_bonus_hold(os, object, FTAG, &db);
2159 ztest_object_unlock(zd, object);
2163 crtxg = ztest_bt_bonus(db)->bt_crtxg;
2165 if (crtxg == 0 || crtxg > txg) {
2166 dmu_buf_rele(db, FTAG);
2167 ztest_object_unlock(zd, object);
2171 dmu_object_info_from_db(db, &doi);
2172 dmu_buf_rele(db, FTAG);
2175 zgd = umem_zalloc(sizeof (*zgd), UMEM_NOFAIL);
2176 zgd->zgd_zilog = zd->zd_zilog;
2177 zgd_private = umem_zalloc(sizeof (ztest_zgd_private_t), UMEM_NOFAIL);
2178 zgd_private->z_zd = zd;
2179 zgd_private->z_object = object;
2180 zgd->zgd_private = zgd_private;
2182 if (buf != NULL) { /* immediate write */
2183 zgd_private->z_rl = ztest_range_lock(zd, object, offset, size,
2185 zgd->zgd_rl = zgd_private->z_rl->z_rl;
2187 error = dmu_read(os, object, offset, size, buf,
2188 DMU_READ_NO_PREFETCH);
2191 size = doi.doi_data_block_size;
2193 offset = P2ALIGN(offset, size);
2195 ASSERT(offset < size);
2199 zgd_private->z_rl = ztest_range_lock(zd, object, offset, size,
2201 zgd->zgd_rl = zgd_private->z_rl->z_rl;
2203 error = dmu_buf_hold(os, object, offset, zgd, &db,
2204 DMU_READ_NO_PREFETCH);
2207 blkptr_t *bp = &lr->lr_blkptr;
2212 ASSERT(db->db_offset == offset);
2213 ASSERT(db->db_size == size);
2215 error = dmu_sync(zio, lr->lr_common.lrc_txg,
2216 ztest_get_done, zgd);
2223 ztest_get_done(zgd, error);
2229 ztest_lr_alloc(size_t lrsize, char *name)
2232 size_t namesize = name ? strlen(name) + 1 : 0;
2234 lr = umem_zalloc(lrsize + namesize, UMEM_NOFAIL);
2237 bcopy(name, lr + lrsize, namesize);
2243 ztest_lr_free(void *lr, size_t lrsize, char *name)
2245 size_t namesize = name ? strlen(name) + 1 : 0;
2247 umem_free(lr, lrsize + namesize);
2251 * Lookup a bunch of objects. Returns the number of objects not found.
2254 ztest_lookup(ztest_ds_t *zd, ztest_od_t *od, int count)
2260 ASSERT(MUTEX_HELD(&zd->zd_dirobj_lock));
2262 for (i = 0; i < count; i++, od++) {
2264 error = zap_lookup(zd->zd_os, od->od_dir, od->od_name,
2265 sizeof (uint64_t), 1, &od->od_object);
2267 ASSERT(error == ENOENT);
2268 ASSERT(od->od_object == 0);
2272 ztest_block_tag_t *bbt;
2273 dmu_object_info_t doi;
2275 ASSERT(od->od_object != 0);
2276 ASSERT(missing == 0); /* there should be no gaps */
2278 ztest_object_lock(zd, od->od_object, RL_READER);
2279 VERIFY3U(0, ==, dmu_bonus_hold(zd->zd_os,
2280 od->od_object, FTAG, &db));
2281 dmu_object_info_from_db(db, &doi);
2282 bbt = ztest_bt_bonus(db);
2283 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
2284 od->od_type = doi.doi_type;
2285 od->od_blocksize = doi.doi_data_block_size;
2286 od->od_gen = bbt->bt_gen;
2287 dmu_buf_rele(db, FTAG);
2288 ztest_object_unlock(zd, od->od_object);
2296 ztest_create(ztest_ds_t *zd, ztest_od_t *od, int count)
2301 ASSERT(MUTEX_HELD(&zd->zd_dirobj_lock));
2303 for (i = 0; i < count; i++, od++) {
2310 lr_create_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
2312 lr->lr_doid = od->od_dir;
2313 lr->lr_foid = 0; /* 0 to allocate, > 0 to claim */
2314 lr->lrz_type = od->od_crtype;
2315 lr->lrz_blocksize = od->od_crblocksize;
2316 lr->lrz_ibshift = ztest_random_ibshift();
2317 lr->lrz_bonustype = DMU_OT_UINT64_OTHER;
2318 lr->lrz_dnodesize = od->od_crdnodesize;
2319 lr->lr_gen = od->od_crgen;
2320 lr->lr_crtime[0] = time(NULL);
2322 if (ztest_replay_create(zd, lr, B_FALSE) != 0) {
2323 ASSERT(missing == 0);
2327 od->od_object = lr->lr_foid;
2328 od->od_type = od->od_crtype;
2329 od->od_blocksize = od->od_crblocksize;
2330 od->od_gen = od->od_crgen;
2331 ASSERT(od->od_object != 0);
2334 ztest_lr_free(lr, sizeof (*lr), od->od_name);
2341 ztest_remove(ztest_ds_t *zd, ztest_od_t *od, int count)
2347 ASSERT(MUTEX_HELD(&zd->zd_dirobj_lock));
2351 for (i = count - 1; i >= 0; i--, od--) {
2358 * No object was found.
2360 if (od->od_object == 0)
2363 lr_remove_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
2365 lr->lr_doid = od->od_dir;
2367 if ((error = ztest_replay_remove(zd, lr, B_FALSE)) != 0) {
2368 ASSERT3U(error, ==, ENOSPC);
2373 ztest_lr_free(lr, sizeof (*lr), od->od_name);
2380 ztest_write(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size,
2386 lr = ztest_lr_alloc(sizeof (*lr) + size, NULL);
2388 lr->lr_foid = object;
2389 lr->lr_offset = offset;
2390 lr->lr_length = size;
2392 BP_ZERO(&lr->lr_blkptr);
2394 bcopy(data, lr + 1, size);
2396 error = ztest_replay_write(zd, lr, B_FALSE);
2398 ztest_lr_free(lr, sizeof (*lr) + size, NULL);
2404 ztest_truncate(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2409 lr = ztest_lr_alloc(sizeof (*lr), NULL);
2411 lr->lr_foid = object;
2412 lr->lr_offset = offset;
2413 lr->lr_length = size;
2415 error = ztest_replay_truncate(zd, lr, B_FALSE);
2417 ztest_lr_free(lr, sizeof (*lr), NULL);
2423 ztest_setattr(ztest_ds_t *zd, uint64_t object)
2428 lr = ztest_lr_alloc(sizeof (*lr), NULL);
2430 lr->lr_foid = object;
2434 error = ztest_replay_setattr(zd, lr, B_FALSE);
2436 ztest_lr_free(lr, sizeof (*lr), NULL);
2442 ztest_prealloc(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2444 objset_t *os = zd->zd_os;
2449 txg_wait_synced(dmu_objset_pool(os), 0);
2451 ztest_object_lock(zd, object, RL_READER);
2452 rl = ztest_range_lock(zd, object, offset, size, RL_WRITER);
2454 tx = dmu_tx_create(os);
2456 dmu_tx_hold_write(tx, object, offset, size);
2458 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
2461 dmu_prealloc(os, object, offset, size, tx);
2463 txg_wait_synced(dmu_objset_pool(os), txg);
2465 (void) dmu_free_long_range(os, object, offset, size);
2468 ztest_range_unlock(zd, rl);
2469 ztest_object_unlock(zd, object);
2473 ztest_io(ztest_ds_t *zd, uint64_t object, uint64_t offset)
2476 ztest_block_tag_t wbt;
2477 dmu_object_info_t doi;
2478 enum ztest_io_type io_type;
2482 VERIFY(dmu_object_info(zd->zd_os, object, &doi) == 0);
2483 blocksize = doi.doi_data_block_size;
2484 data = umem_alloc(blocksize, UMEM_NOFAIL);
2487 * Pick an i/o type at random, biased toward writing block tags.
2489 io_type = ztest_random(ZTEST_IO_TYPES);
2490 if (ztest_random(2) == 0)
2491 io_type = ZTEST_IO_WRITE_TAG;
2493 (void) rw_rdlock(&zd->zd_zilog_lock);
2497 case ZTEST_IO_WRITE_TAG:
2498 ztest_bt_generate(&wbt, zd->zd_os, object, doi.doi_dnodesize,
2500 (void) ztest_write(zd, object, offset, sizeof (wbt), &wbt);
2503 case ZTEST_IO_WRITE_PATTERN:
2504 (void) memset(data, 'a' + (object + offset) % 5, blocksize);
2505 if (ztest_random(2) == 0) {
2507 * Induce fletcher2 collisions to ensure that
2508 * zio_ddt_collision() detects and resolves them
2509 * when using fletcher2-verify for deduplication.
2511 ((uint64_t *)data)[0] ^= 1ULL << 63;
2512 ((uint64_t *)data)[4] ^= 1ULL << 63;
2514 (void) ztest_write(zd, object, offset, blocksize, data);
2517 case ZTEST_IO_WRITE_ZEROES:
2518 bzero(data, blocksize);
2519 (void) ztest_write(zd, object, offset, blocksize, data);
2522 case ZTEST_IO_TRUNCATE:
2523 (void) ztest_truncate(zd, object, offset, blocksize);
2526 case ZTEST_IO_SETATTR:
2527 (void) ztest_setattr(zd, object);
2532 case ZTEST_IO_REWRITE:
2533 (void) rw_rdlock(&ztest_name_lock);
2534 err = ztest_dsl_prop_set_uint64(zd->zd_name,
2535 ZFS_PROP_CHECKSUM, spa_dedup_checksum(ztest_spa),
2537 VERIFY(err == 0 || err == ENOSPC);
2538 err = ztest_dsl_prop_set_uint64(zd->zd_name,
2539 ZFS_PROP_COMPRESSION,
2540 ztest_random_dsl_prop(ZFS_PROP_COMPRESSION),
2542 VERIFY(err == 0 || err == ENOSPC);
2543 (void) rw_unlock(&ztest_name_lock);
2545 VERIFY0(dmu_read(zd->zd_os, object, offset, blocksize, data,
2546 DMU_READ_NO_PREFETCH));
2548 (void) ztest_write(zd, object, offset, blocksize, data);
2552 (void) rw_unlock(&zd->zd_zilog_lock);
2554 umem_free(data, blocksize);
2558 * Initialize an object description template.
2561 ztest_od_init(ztest_od_t *od, uint64_t id, char *tag, uint64_t index,
2562 dmu_object_type_t type, uint64_t blocksize, uint64_t dnodesize,
2565 od->od_dir = ZTEST_DIROBJ;
2568 od->od_crtype = type;
2569 od->od_crblocksize = blocksize ? blocksize : ztest_random_blocksize();
2570 od->od_crdnodesize = dnodesize ? dnodesize : ztest_random_dnodesize();
2573 od->od_type = DMU_OT_NONE;
2574 od->od_blocksize = 0;
2577 (void) snprintf(od->od_name, sizeof (od->od_name), "%s(%lld)[%llu]",
2578 tag, (longlong_t)id, (u_longlong_t)index);
2582 * Lookup or create the objects for a test using the od template.
2583 * If the objects do not all exist, or if 'remove' is specified,
2584 * remove any existing objects and create new ones. Otherwise,
2585 * use the existing objects.
2588 ztest_object_init(ztest_ds_t *zd, ztest_od_t *od, size_t size, boolean_t remove)
2590 int count = size / sizeof (*od);
2593 mutex_enter(&zd->zd_dirobj_lock);
2594 if ((ztest_lookup(zd, od, count) != 0 || remove) &&
2595 (ztest_remove(zd, od, count) != 0 ||
2596 ztest_create(zd, od, count) != 0))
2599 mutex_exit(&zd->zd_dirobj_lock);
2606 ztest_zil_commit(ztest_ds_t *zd, uint64_t id)
2608 zilog_t *zilog = zd->zd_zilog;
2610 (void) rw_rdlock(&zd->zd_zilog_lock);
2612 zil_commit(zilog, ztest_random(ZTEST_OBJECTS));
2615 * Remember the committed values in zd, which is in parent/child
2616 * shared memory. If we die, the next iteration of ztest_run()
2617 * will verify that the log really does contain this record.
2619 mutex_enter(&zilog->zl_lock);
2620 ASSERT(zd->zd_shared != NULL);
2621 ASSERT3U(zd->zd_shared->zd_seq, <=, zilog->zl_commit_lr_seq);
2622 zd->zd_shared->zd_seq = zilog->zl_commit_lr_seq;
2623 mutex_exit(&zilog->zl_lock);
2625 (void) rw_unlock(&zd->zd_zilog_lock);
2629 * This function is designed to simulate the operations that occur during a
2630 * mount/unmount operation. We hold the dataset across these operations in an
2631 * attempt to expose any implicit assumptions about ZIL management.
2635 ztest_zil_remount(ztest_ds_t *zd, uint64_t id)
2637 objset_t *os = zd->zd_os;
2640 * We grab the zd_dirobj_lock to ensure that no other thread is
2641 * updating the zil (i.e. adding in-memory log records) and the
2642 * zd_zilog_lock to block any I/O.
2644 mutex_enter(&zd->zd_dirobj_lock);
2645 (void) rw_wrlock(&zd->zd_zilog_lock);
2647 /* zfsvfs_teardown() */
2648 zil_close(zd->zd_zilog);
2650 /* zfsvfs_setup() */
2651 VERIFY(zil_open(os, ztest_get_data) == zd->zd_zilog);
2652 zil_replay(os, zd, ztest_replay_vector);
2654 (void) rw_unlock(&zd->zd_zilog_lock);
2655 mutex_exit(&zd->zd_dirobj_lock);
2659 * Verify that we can't destroy an active pool, create an existing pool,
2660 * or create a pool with a bad vdev spec.
2664 ztest_spa_create_destroy(ztest_ds_t *zd, uint64_t id)
2666 ztest_shared_opts_t *zo = &ztest_opts;
2670 if (zo->zo_mmp_test)
2674 * Attempt to create using a bad file.
2676 nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1);
2677 VERIFY3U(ENOENT, ==,
2678 spa_create("ztest_bad_file", nvroot, NULL, NULL, NULL));
2679 nvlist_free(nvroot);
2682 * Attempt to create using a bad mirror.
2684 nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 2, 1);
2685 VERIFY3U(ENOENT, ==,
2686 spa_create("ztest_bad_mirror", nvroot, NULL, NULL, NULL));
2687 nvlist_free(nvroot);
2690 * Attempt to create an existing pool. It shouldn't matter
2691 * what's in the nvroot; we should fail with EEXIST.
2693 (void) rw_rdlock(&ztest_name_lock);
2694 nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1);
2695 VERIFY3U(EEXIST, ==,
2696 spa_create(zo->zo_pool, nvroot, NULL, NULL, NULL));
2697 nvlist_free(nvroot);
2698 VERIFY3U(0, ==, spa_open(zo->zo_pool, &spa, FTAG));
2699 VERIFY3U(EBUSY, ==, spa_destroy(zo->zo_pool));
2700 spa_close(spa, FTAG);
2702 (void) rw_unlock(&ztest_name_lock);
2706 * Start and then stop the MMP threads to ensure the startup and shutdown code
2707 * works properly. Actual protection and property-related code tested via ZTS.
2711 ztest_mmp_enable_disable(ztest_ds_t *zd, uint64_t id)
2713 ztest_shared_opts_t *zo = &ztest_opts;
2714 spa_t *spa = ztest_spa;
2716 if (zo->zo_mmp_test)
2719 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
2720 mutex_enter(&spa->spa_props_lock);
2722 if (!spa_multihost(spa)) {
2723 spa->spa_multihost = B_TRUE;
2724 mmp_thread_start(spa);
2727 mutex_exit(&spa->spa_props_lock);
2728 spa_config_exit(spa, SCL_CONFIG, FTAG);
2730 txg_wait_synced(spa_get_dsl(spa), 0);
2731 mmp_signal_all_threads();
2732 txg_wait_synced(spa_get_dsl(spa), 0);
2734 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
2735 mutex_enter(&spa->spa_props_lock);
2737 if (spa_multihost(spa)) {
2738 mmp_thread_stop(spa);
2739 spa->spa_multihost = B_FALSE;
2742 mutex_exit(&spa->spa_props_lock);
2743 spa_config_exit(spa, SCL_CONFIG, FTAG);
2748 ztest_spa_upgrade(ztest_ds_t *zd, uint64_t id)
2751 uint64_t initial_version = SPA_VERSION_INITIAL;
2752 uint64_t version, newversion;
2753 nvlist_t *nvroot, *props;
2756 if (ztest_opts.zo_mmp_test)
2759 mutex_enter(&ztest_vdev_lock);
2760 name = kmem_asprintf("%s_upgrade", ztest_opts.zo_pool);
2763 * Clean up from previous runs.
2765 (void) spa_destroy(name);
2767 nvroot = make_vdev_root(NULL, NULL, name, ztest_opts.zo_vdev_size, 0,
2768 0, ztest_opts.zo_raidz, ztest_opts.zo_mirrors, 1);
2771 * If we're configuring a RAIDZ device then make sure that the
2772 * the initial version is capable of supporting that feature.
2774 switch (ztest_opts.zo_raidz_parity) {
2777 initial_version = SPA_VERSION_INITIAL;
2780 initial_version = SPA_VERSION_RAIDZ2;
2783 initial_version = SPA_VERSION_RAIDZ3;
2788 * Create a pool with a spa version that can be upgraded. Pick
2789 * a value between initial_version and SPA_VERSION_BEFORE_FEATURES.
2792 version = ztest_random_spa_version(initial_version);
2793 } while (version > SPA_VERSION_BEFORE_FEATURES);
2795 props = fnvlist_alloc();
2796 fnvlist_add_uint64(props,
2797 zpool_prop_to_name(ZPOOL_PROP_VERSION), version);
2798 VERIFY3S(spa_create(name, nvroot, props, NULL, NULL), ==, 0);
2799 fnvlist_free(nvroot);
2800 fnvlist_free(props);
2802 VERIFY3S(spa_open(name, &spa, FTAG), ==, 0);
2803 VERIFY3U(spa_version(spa), ==, version);
2804 newversion = ztest_random_spa_version(version + 1);
2806 if (ztest_opts.zo_verbose >= 4) {
2807 (void) printf("upgrading spa version from %llu to %llu\n",
2808 (u_longlong_t)version, (u_longlong_t)newversion);
2811 spa_upgrade(spa, newversion);
2812 VERIFY3U(spa_version(spa), >, version);
2813 VERIFY3U(spa_version(spa), ==, fnvlist_lookup_uint64(spa->spa_config,
2814 zpool_prop_to_name(ZPOOL_PROP_VERSION)));
2815 spa_close(spa, FTAG);
2818 mutex_exit(&ztest_vdev_lock);
2822 vdev_lookup_by_path(vdev_t *vd, const char *path)
2827 if (vd->vdev_path != NULL && strcmp(path, vd->vdev_path) == 0)
2830 for (c = 0; c < vd->vdev_children; c++)
2831 if ((mvd = vdev_lookup_by_path(vd->vdev_child[c], path)) !=
2839 * Find the first available hole which can be used as a top-level.
2842 find_vdev_hole(spa_t *spa)
2844 vdev_t *rvd = spa->spa_root_vdev;
2847 ASSERT(spa_config_held(spa, SCL_VDEV, RW_READER) == SCL_VDEV);
2849 for (c = 0; c < rvd->vdev_children; c++) {
2850 vdev_t *cvd = rvd->vdev_child[c];
2852 if (cvd->vdev_ishole)
2859 * Verify that vdev_add() works as expected.
2863 ztest_vdev_add_remove(ztest_ds_t *zd, uint64_t id)
2865 ztest_shared_t *zs = ztest_shared;
2866 spa_t *spa = ztest_spa;
2872 if (ztest_opts.zo_mmp_test)
2875 mutex_enter(&ztest_vdev_lock);
2876 leaves = MAX(zs->zs_mirrors + zs->zs_splits, 1) * ztest_opts.zo_raidz;
2878 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2880 ztest_shared->zs_vdev_next_leaf = find_vdev_hole(spa) * leaves;
2883 * If we have slogs then remove them 1/4 of the time.
2885 if (spa_has_slogs(spa) && ztest_random(4) == 0) {
2887 * Grab the guid from the head of the log class rotor.
2889 guid = spa_log_class(spa)->mc_rotor->mg_vd->vdev_guid;
2891 spa_config_exit(spa, SCL_VDEV, FTAG);
2894 * We have to grab the zs_name_lock as writer to
2895 * prevent a race between removing a slog (dmu_objset_find)
2896 * and destroying a dataset. Removing the slog will
2897 * grab a reference on the dataset which may cause
2898 * dsl_destroy_head() to fail with EBUSY thus
2899 * leaving the dataset in an inconsistent state.
2901 rw_wrlock(&ztest_name_lock);
2902 error = spa_vdev_remove(spa, guid, B_FALSE);
2903 rw_unlock(&ztest_name_lock);
2905 if (error && error != EEXIST)
2906 fatal(0, "spa_vdev_remove() = %d", error);
2908 spa_config_exit(spa, SCL_VDEV, FTAG);
2911 * Make 1/4 of the devices be log devices.
2913 nvroot = make_vdev_root(NULL, NULL, NULL,
2914 ztest_opts.zo_vdev_size, 0,
2915 ztest_random(4) == 0, ztest_opts.zo_raidz,
2918 error = spa_vdev_add(spa, nvroot);
2919 nvlist_free(nvroot);
2921 if (error == ENOSPC)
2922 ztest_record_enospc("spa_vdev_add");
2923 else if (error != 0)
2924 fatal(0, "spa_vdev_add() = %d", error);
2927 mutex_exit(&ztest_vdev_lock);
2931 * Verify that adding/removing aux devices (l2arc, hot spare) works as expected.
2935 ztest_vdev_aux_add_remove(ztest_ds_t *zd, uint64_t id)
2937 ztest_shared_t *zs = ztest_shared;
2938 spa_t *spa = ztest_spa;
2939 vdev_t *rvd = spa->spa_root_vdev;
2940 spa_aux_vdev_t *sav;
2946 if (ztest_opts.zo_mmp_test)
2949 path = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
2951 if (ztest_random(2) == 0) {
2952 sav = &spa->spa_spares;
2953 aux = ZPOOL_CONFIG_SPARES;
2955 sav = &spa->spa_l2cache;
2956 aux = ZPOOL_CONFIG_L2CACHE;
2959 mutex_enter(&ztest_vdev_lock);
2961 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2963 if (sav->sav_count != 0 && ztest_random(4) == 0) {
2965 * Pick a random device to remove.
2967 guid = sav->sav_vdevs[ztest_random(sav->sav_count)]->vdev_guid;
2970 * Find an unused device we can add.
2972 zs->zs_vdev_aux = 0;
2975 (void) snprintf(path, MAXPATHLEN, ztest_aux_template,
2976 ztest_opts.zo_dir, ztest_opts.zo_pool, aux,
2978 for (c = 0; c < sav->sav_count; c++)
2979 if (strcmp(sav->sav_vdevs[c]->vdev_path,
2982 if (c == sav->sav_count &&
2983 vdev_lookup_by_path(rvd, path) == NULL)
2989 spa_config_exit(spa, SCL_VDEV, FTAG);
2995 nvlist_t *nvroot = make_vdev_root(NULL, aux, NULL,
2996 (ztest_opts.zo_vdev_size * 5) / 4, 0, 0, 0, 0, 1);
2997 error = spa_vdev_add(spa, nvroot);
2999 fatal(0, "spa_vdev_add(%p) = %d", nvroot, error);
3000 nvlist_free(nvroot);
3003 * Remove an existing device. Sometimes, dirty its
3004 * vdev state first to make sure we handle removal
3005 * of devices that have pending state changes.
3007 if (ztest_random(2) == 0)
3008 (void) vdev_online(spa, guid, 0, NULL);
3010 error = spa_vdev_remove(spa, guid, B_FALSE);
3011 if (error != 0 && error != EBUSY)
3012 fatal(0, "spa_vdev_remove(%llu) = %d", guid, error);
3015 mutex_exit(&ztest_vdev_lock);
3017 umem_free(path, MAXPATHLEN);
3021 * split a pool if it has mirror tlvdevs
3025 ztest_split_pool(ztest_ds_t *zd, uint64_t id)
3027 ztest_shared_t *zs = ztest_shared;
3028 spa_t *spa = ztest_spa;
3029 vdev_t *rvd = spa->spa_root_vdev;
3030 nvlist_t *tree, **child, *config, *split, **schild;
3031 uint_t c, children, schildren = 0, lastlogid = 0;
3034 if (ztest_opts.zo_mmp_test)
3037 mutex_enter(&ztest_vdev_lock);
3039 /* ensure we have a useable config; mirrors of raidz aren't supported */
3040 if (zs->zs_mirrors < 3 || ztest_opts.zo_raidz > 1) {
3041 mutex_exit(&ztest_vdev_lock);
3045 /* clean up the old pool, if any */
3046 (void) spa_destroy("splitp");
3048 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
3050 /* generate a config from the existing config */
3051 mutex_enter(&spa->spa_props_lock);
3052 VERIFY(nvlist_lookup_nvlist(spa->spa_config, ZPOOL_CONFIG_VDEV_TREE,
3054 mutex_exit(&spa->spa_props_lock);
3056 VERIFY(nvlist_lookup_nvlist_array(tree, ZPOOL_CONFIG_CHILDREN, &child,
3059 schild = malloc(rvd->vdev_children * sizeof (nvlist_t *));
3060 for (c = 0; c < children; c++) {
3061 vdev_t *tvd = rvd->vdev_child[c];
3065 if (tvd->vdev_islog || tvd->vdev_ops == &vdev_hole_ops) {
3066 VERIFY(nvlist_alloc(&schild[schildren], NV_UNIQUE_NAME,
3068 VERIFY(nvlist_add_string(schild[schildren],
3069 ZPOOL_CONFIG_TYPE, VDEV_TYPE_HOLE) == 0);
3070 VERIFY(nvlist_add_uint64(schild[schildren],
3071 ZPOOL_CONFIG_IS_HOLE, 1) == 0);
3073 lastlogid = schildren;
3078 VERIFY(nvlist_lookup_nvlist_array(child[c],
3079 ZPOOL_CONFIG_CHILDREN, &mchild, &mchildren) == 0);
3080 VERIFY(nvlist_dup(mchild[0], &schild[schildren++], 0) == 0);
3083 /* OK, create a config that can be used to split */
3084 VERIFY(nvlist_alloc(&split, NV_UNIQUE_NAME, 0) == 0);
3085 VERIFY(nvlist_add_string(split, ZPOOL_CONFIG_TYPE,
3086 VDEV_TYPE_ROOT) == 0);
3087 VERIFY(nvlist_add_nvlist_array(split, ZPOOL_CONFIG_CHILDREN, schild,
3088 lastlogid != 0 ? lastlogid : schildren) == 0);
3090 VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, 0) == 0);
3091 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, split) == 0);
3093 for (c = 0; c < schildren; c++)
3094 nvlist_free(schild[c]);
3098 spa_config_exit(spa, SCL_VDEV, FTAG);
3100 (void) rw_wrlock(&ztest_name_lock);
3101 error = spa_vdev_split_mirror(spa, "splitp", config, NULL, B_FALSE);
3102 (void) rw_unlock(&ztest_name_lock);
3104 nvlist_free(config);
3107 (void) printf("successful split - results:\n");
3108 mutex_enter(&spa_namespace_lock);
3109 show_pool_stats(spa);
3110 show_pool_stats(spa_lookup("splitp"));
3111 mutex_exit(&spa_namespace_lock);
3115 mutex_exit(&ztest_vdev_lock);
3120 * Verify that we can attach and detach devices.
3124 ztest_vdev_attach_detach(ztest_ds_t *zd, uint64_t id)
3126 ztest_shared_t *zs = ztest_shared;
3127 spa_t *spa = ztest_spa;
3128 spa_aux_vdev_t *sav = &spa->spa_spares;
3129 vdev_t *rvd = spa->spa_root_vdev;
3130 vdev_t *oldvd, *newvd, *pvd;
3134 uint64_t ashift = ztest_get_ashift();
3135 uint64_t oldguid, pguid;
3136 uint64_t oldsize, newsize;
3137 char *oldpath, *newpath;
3139 int oldvd_has_siblings = B_FALSE;
3140 int newvd_is_spare = B_FALSE;
3142 int error, expected_error;
3144 if (ztest_opts.zo_mmp_test)
3147 oldpath = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
3148 newpath = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
3150 mutex_enter(&ztest_vdev_lock);
3151 leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz;
3153 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
3156 * Decide whether to do an attach or a replace.
3158 replacing = ztest_random(2);
3161 * Pick a random top-level vdev.
3163 top = ztest_random_vdev_top(spa, B_TRUE);
3166 * Pick a random leaf within it.
3168 leaf = ztest_random(leaves);
3173 oldvd = rvd->vdev_child[top];
3174 if (zs->zs_mirrors >= 1) {
3175 ASSERT(oldvd->vdev_ops == &vdev_mirror_ops);
3176 ASSERT(oldvd->vdev_children >= zs->zs_mirrors);
3177 oldvd = oldvd->vdev_child[leaf / ztest_opts.zo_raidz];
3179 if (ztest_opts.zo_raidz > 1) {
3180 ASSERT(oldvd->vdev_ops == &vdev_raidz_ops);
3181 ASSERT(oldvd->vdev_children == ztest_opts.zo_raidz);
3182 oldvd = oldvd->vdev_child[leaf % ztest_opts.zo_raidz];
3186 * If we're already doing an attach or replace, oldvd may be a
3187 * mirror vdev -- in which case, pick a random child.
3189 while (oldvd->vdev_children != 0) {
3190 oldvd_has_siblings = B_TRUE;
3191 ASSERT(oldvd->vdev_children >= 2);
3192 oldvd = oldvd->vdev_child[ztest_random(oldvd->vdev_children)];
3195 oldguid = oldvd->vdev_guid;
3196 oldsize = vdev_get_min_asize(oldvd);
3197 oldvd_is_log = oldvd->vdev_top->vdev_islog;
3198 (void) strcpy(oldpath, oldvd->vdev_path);
3199 pvd = oldvd->vdev_parent;
3200 pguid = pvd->vdev_guid;
3203 * If oldvd has siblings, then half of the time, detach it.
3205 if (oldvd_has_siblings && ztest_random(2) == 0) {
3206 spa_config_exit(spa, SCL_VDEV, FTAG);
3207 error = spa_vdev_detach(spa, oldguid, pguid, B_FALSE);
3208 if (error != 0 && error != ENODEV && error != EBUSY &&
3210 fatal(0, "detach (%s) returned %d", oldpath, error);
3215 * For the new vdev, choose with equal probability between the two
3216 * standard paths (ending in either 'a' or 'b') or a random hot spare.
3218 if (sav->sav_count != 0 && ztest_random(3) == 0) {
3219 newvd = sav->sav_vdevs[ztest_random(sav->sav_count)];
3220 newvd_is_spare = B_TRUE;
3221 (void) strcpy(newpath, newvd->vdev_path);
3223 (void) snprintf(newpath, MAXPATHLEN, ztest_dev_template,
3224 ztest_opts.zo_dir, ztest_opts.zo_pool,
3225 top * leaves + leaf);
3226 if (ztest_random(2) == 0)
3227 newpath[strlen(newpath) - 1] = 'b';
3228 newvd = vdev_lookup_by_path(rvd, newpath);
3232 newsize = vdev_get_min_asize(newvd);
3235 * Make newsize a little bigger or smaller than oldsize.
3236 * If it's smaller, the attach should fail.
3237 * If it's larger, and we're doing a replace,
3238 * we should get dynamic LUN growth when we're done.
3240 newsize = 10 * oldsize / (9 + ztest_random(3));
3244 * If pvd is not a mirror or root, the attach should fail with ENOTSUP,
3245 * unless it's a replace; in that case any non-replacing parent is OK.
3247 * If newvd is already part of the pool, it should fail with EBUSY.
3249 * If newvd is too small, it should fail with EOVERFLOW.
3251 if (pvd->vdev_ops != &vdev_mirror_ops &&
3252 pvd->vdev_ops != &vdev_root_ops && (!replacing ||
3253 pvd->vdev_ops == &vdev_replacing_ops ||
3254 pvd->vdev_ops == &vdev_spare_ops))
3255 expected_error = ENOTSUP;
3256 else if (newvd_is_spare && (!replacing || oldvd_is_log))
3257 expected_error = ENOTSUP;
3258 else if (newvd == oldvd)
3259 expected_error = replacing ? 0 : EBUSY;
3260 else if (vdev_lookup_by_path(rvd, newpath) != NULL)
3261 expected_error = EBUSY;
3262 else if (newsize < oldsize)
3263 expected_error = EOVERFLOW;
3264 else if (ashift > oldvd->vdev_top->vdev_ashift)
3265 expected_error = EDOM;
3269 spa_config_exit(spa, SCL_VDEV, FTAG);
3272 * Build the nvlist describing newpath.
3274 root = make_vdev_root(newpath, NULL, NULL, newvd == NULL ? newsize : 0,
3275 ashift, 0, 0, 0, 1);
3277 error = spa_vdev_attach(spa, oldguid, root, replacing);
3282 * If our parent was the replacing vdev, but the replace completed,
3283 * then instead of failing with ENOTSUP we may either succeed,
3284 * fail with ENODEV, or fail with EOVERFLOW.
3286 if (expected_error == ENOTSUP &&
3287 (error == 0 || error == ENODEV || error == EOVERFLOW))
3288 expected_error = error;
3291 * If someone grew the LUN, the replacement may be too small.
3293 if (error == EOVERFLOW || error == EBUSY)
3294 expected_error = error;
3296 /* XXX workaround 6690467 */
3297 if (error != expected_error && expected_error != EBUSY) {
3298 fatal(0, "attach (%s %llu, %s %llu, %d) "
3299 "returned %d, expected %d",
3300 oldpath, oldsize, newpath,
3301 newsize, replacing, error, expected_error);
3304 mutex_exit(&ztest_vdev_lock);
3306 umem_free(oldpath, MAXPATHLEN);
3307 umem_free(newpath, MAXPATHLEN);
3311 * Callback function which expands the physical size of the vdev.
3314 grow_vdev(vdev_t *vd, void *arg)
3316 ASSERTV(spa_t *spa = vd->vdev_spa);
3317 size_t *newsize = arg;
3321 ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
3322 ASSERT(vd->vdev_ops->vdev_op_leaf);
3324 if ((fd = open(vd->vdev_path, O_RDWR)) == -1)
3327 fsize = lseek(fd, 0, SEEK_END);
3328 VERIFY(ftruncate(fd, *newsize) == 0);
3330 if (ztest_opts.zo_verbose >= 6) {
3331 (void) printf("%s grew from %lu to %lu bytes\n",
3332 vd->vdev_path, (ulong_t)fsize, (ulong_t)*newsize);
3339 * Callback function which expands a given vdev by calling vdev_online().
3343 online_vdev(vdev_t *vd, void *arg)
3345 spa_t *spa = vd->vdev_spa;
3346 vdev_t *tvd = vd->vdev_top;
3347 uint64_t guid = vd->vdev_guid;
3348 uint64_t generation = spa->spa_config_generation + 1;
3349 vdev_state_t newstate = VDEV_STATE_UNKNOWN;
3352 ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
3353 ASSERT(vd->vdev_ops->vdev_op_leaf);
3355 /* Calling vdev_online will initialize the new metaslabs */
3356 spa_config_exit(spa, SCL_STATE, spa);
3357 error = vdev_online(spa, guid, ZFS_ONLINE_EXPAND, &newstate);
3358 spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3361 * If vdev_online returned an error or the underlying vdev_open
3362 * failed then we abort the expand. The only way to know that
3363 * vdev_open fails is by checking the returned newstate.
3365 if (error || newstate != VDEV_STATE_HEALTHY) {
3366 if (ztest_opts.zo_verbose >= 5) {
3367 (void) printf("Unable to expand vdev, state %llu, "
3368 "error %d\n", (u_longlong_t)newstate, error);
3372 ASSERT3U(newstate, ==, VDEV_STATE_HEALTHY);
3375 * Since we dropped the lock we need to ensure that we're
3376 * still talking to the original vdev. It's possible this
3377 * vdev may have been detached/replaced while we were
3378 * trying to online it.
3380 if (generation != spa->spa_config_generation) {
3381 if (ztest_opts.zo_verbose >= 5) {
3382 (void) printf("vdev configuration has changed, "
3383 "guid %llu, state %llu, expected gen %llu, "
3386 (u_longlong_t)tvd->vdev_state,
3387 (u_longlong_t)generation,
3388 (u_longlong_t)spa->spa_config_generation);
3396 * Traverse the vdev tree calling the supplied function.
3397 * We continue to walk the tree until we either have walked all
3398 * children or we receive a non-NULL return from the callback.
3399 * If a NULL callback is passed, then we just return back the first
3400 * leaf vdev we encounter.
3403 vdev_walk_tree(vdev_t *vd, vdev_t *(*func)(vdev_t *, void *), void *arg)
3407 if (vd->vdev_ops->vdev_op_leaf) {
3411 return (func(vd, arg));
3414 for (c = 0; c < vd->vdev_children; c++) {
3415 vdev_t *cvd = vd->vdev_child[c];
3416 if ((cvd = vdev_walk_tree(cvd, func, arg)) != NULL)
3423 * Verify that dynamic LUN growth works as expected.
3427 ztest_vdev_LUN_growth(ztest_ds_t *zd, uint64_t id)
3429 spa_t *spa = ztest_spa;
3431 metaslab_class_t *mc;
3432 metaslab_group_t *mg;
3433 size_t psize, newsize;
3435 uint64_t old_class_space, new_class_space, old_ms_count, new_ms_count;
3437 mutex_enter(&ztest_vdev_lock);
3438 spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3440 top = ztest_random_vdev_top(spa, B_TRUE);
3442 tvd = spa->spa_root_vdev->vdev_child[top];
3445 old_ms_count = tvd->vdev_ms_count;
3446 old_class_space = metaslab_class_get_space(mc);
3449 * Determine the size of the first leaf vdev associated with
3450 * our top-level device.
3452 vd = vdev_walk_tree(tvd, NULL, NULL);
3453 ASSERT3P(vd, !=, NULL);
3454 ASSERT(vd->vdev_ops->vdev_op_leaf);
3456 psize = vd->vdev_psize;
3459 * We only try to expand the vdev if it's healthy, less than 4x its
3460 * original size, and it has a valid psize.
3462 if (tvd->vdev_state != VDEV_STATE_HEALTHY ||
3463 psize == 0 || psize >= 4 * ztest_opts.zo_vdev_size) {
3464 spa_config_exit(spa, SCL_STATE, spa);
3465 mutex_exit(&ztest_vdev_lock);
3469 newsize = psize + psize / 8;
3470 ASSERT3U(newsize, >, psize);
3472 if (ztest_opts.zo_verbose >= 6) {
3473 (void) printf("Expanding LUN %s from %lu to %lu\n",
3474 vd->vdev_path, (ulong_t)psize, (ulong_t)newsize);
3478 * Growing the vdev is a two step process:
3479 * 1). expand the physical size (i.e. relabel)
3480 * 2). online the vdev to create the new metaslabs
3482 if (vdev_walk_tree(tvd, grow_vdev, &newsize) != NULL ||
3483 vdev_walk_tree(tvd, online_vdev, NULL) != NULL ||
3484 tvd->vdev_state != VDEV_STATE_HEALTHY) {
3485 if (ztest_opts.zo_verbose >= 5) {
3486 (void) printf("Could not expand LUN because "
3487 "the vdev configuration changed.\n");
3489 spa_config_exit(spa, SCL_STATE, spa);
3490 mutex_exit(&ztest_vdev_lock);
3494 spa_config_exit(spa, SCL_STATE, spa);
3497 * Expanding the LUN will update the config asynchronously,
3498 * thus we must wait for the async thread to complete any
3499 * pending tasks before proceeding.
3503 mutex_enter(&spa->spa_async_lock);
3504 done = (spa->spa_async_thread == NULL && !spa->spa_async_tasks);
3505 mutex_exit(&spa->spa_async_lock);
3508 txg_wait_synced(spa_get_dsl(spa), 0);
3509 (void) poll(NULL, 0, 100);
3512 spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3514 tvd = spa->spa_root_vdev->vdev_child[top];
3515 new_ms_count = tvd->vdev_ms_count;
3516 new_class_space = metaslab_class_get_space(mc);
3518 if (tvd->vdev_mg != mg || mg->mg_class != mc) {
3519 if (ztest_opts.zo_verbose >= 5) {
3520 (void) printf("Could not verify LUN expansion due to "
3521 "intervening vdev offline or remove.\n");
3523 spa_config_exit(spa, SCL_STATE, spa);
3524 mutex_exit(&ztest_vdev_lock);
3529 * Make sure we were able to grow the vdev.
3531 if (new_ms_count <= old_ms_count)
3532 fatal(0, "LUN expansion failed: ms_count %llu <= %llu\n",
3533 old_ms_count, new_ms_count);
3536 * Make sure we were able to grow the pool.
3538 if (new_class_space <= old_class_space)
3539 fatal(0, "LUN expansion failed: class_space %llu <= %llu\n",
3540 old_class_space, new_class_space);
3542 if (ztest_opts.zo_verbose >= 5) {
3543 char oldnumbuf[6], newnumbuf[6];
3545 nicenum(old_class_space, oldnumbuf);
3546 nicenum(new_class_space, newnumbuf);
3547 (void) printf("%s grew from %s to %s\n",
3548 spa->spa_name, oldnumbuf, newnumbuf);
3551 spa_config_exit(spa, SCL_STATE, spa);
3552 mutex_exit(&ztest_vdev_lock);
3556 * Verify that dmu_objset_{create,destroy,open,close} work as expected.
3560 ztest_objset_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
3563 * Create the objects common to all ztest datasets.
3565 VERIFY(zap_create_claim(os, ZTEST_DIROBJ,
3566 DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx) == 0);
3570 ztest_dataset_create(char *dsname)
3574 dsl_crypto_params_t *dcp = NULL;
3577 * 50% of the time, we create encrypted datasets
3578 * using a random cipher suite and a hard-coded
3581 rand = ztest_random(2);
3583 nvlist_t *crypto_args = fnvlist_alloc();
3584 nvlist_t *props = fnvlist_alloc();
3586 /* slight bias towards the default cipher suite */
3587 rand = ztest_random(ZIO_CRYPT_FUNCTIONS);
3588 if (rand < ZIO_CRYPT_AES_128_CCM)
3589 rand = ZIO_CRYPT_ON;
3591 fnvlist_add_uint64(props,
3592 zfs_prop_to_name(ZFS_PROP_ENCRYPTION), rand);
3593 fnvlist_add_uint8_array(crypto_args, "wkeydata",
3594 (uint8_t *)ztest_wkeydata, WRAPPING_KEY_LEN);
3597 * These parameters aren't really used by the kernel. They
3598 * are simply stored so that userspace knows how to load
3601 fnvlist_add_uint64(props,
3602 zfs_prop_to_name(ZFS_PROP_KEYFORMAT), ZFS_KEYFORMAT_RAW);
3603 fnvlist_add_string(props,
3604 zfs_prop_to_name(ZFS_PROP_KEYLOCATION), "prompt");
3605 fnvlist_add_uint64(props,
3606 zfs_prop_to_name(ZFS_PROP_PBKDF2_SALT), 0ULL);
3607 fnvlist_add_uint64(props,
3608 zfs_prop_to_name(ZFS_PROP_PBKDF2_ITERS), 0ULL);
3610 VERIFY0(dsl_crypto_params_create_nvlist(DCP_CMD_NONE, props,
3611 crypto_args, &dcp));
3613 fnvlist_free(crypto_args);
3614 fnvlist_free(props);
3617 err = dmu_objset_create(dsname, DMU_OST_OTHER, 0, dcp,
3618 ztest_objset_create_cb, NULL);
3619 dsl_crypto_params_free(dcp, !!err);
3621 rand = ztest_random(100);
3622 if (err || rand < 80)
3625 if (ztest_opts.zo_verbose >= 5)
3626 (void) printf("Setting dataset %s to sync always\n", dsname);
3627 return (ztest_dsl_prop_set_uint64(dsname, ZFS_PROP_SYNC,
3628 ZFS_SYNC_ALWAYS, B_FALSE));
3633 ztest_objset_destroy_cb(const char *name, void *arg)
3636 dmu_object_info_t doi;
3640 * Verify that the dataset contains a directory object.
3642 VERIFY0(ztest_dmu_objset_own(name, DMU_OST_OTHER, B_TRUE,
3643 B_TRUE, FTAG, &os));
3644 error = dmu_object_info(os, ZTEST_DIROBJ, &doi);
3645 if (error != ENOENT) {
3646 /* We could have crashed in the middle of destroying it */
3648 ASSERT3U(doi.doi_type, ==, DMU_OT_ZAP_OTHER);
3649 ASSERT3S(doi.doi_physical_blocks_512, >=, 0);
3651 dmu_objset_disown(os, B_TRUE, FTAG);
3654 * Destroy the dataset.
3656 if (strchr(name, '@') != NULL) {
3657 VERIFY0(dsl_destroy_snapshot(name, B_TRUE));
3659 error = dsl_destroy_head(name);
3660 /* There could be a hold on this dataset */
3668 ztest_snapshot_create(char *osname, uint64_t id)
3670 char snapname[ZFS_MAX_DATASET_NAME_LEN];
3673 (void) snprintf(snapname, sizeof (snapname), "%llu", (u_longlong_t)id);
3675 error = dmu_objset_snapshot_one(osname, snapname);
3676 if (error == ENOSPC) {
3677 ztest_record_enospc(FTAG);
3680 if (error != 0 && error != EEXIST) {
3681 fatal(0, "ztest_snapshot_create(%s@%s) = %d", osname,
3688 ztest_snapshot_destroy(char *osname, uint64_t id)
3690 char snapname[ZFS_MAX_DATASET_NAME_LEN];
3693 (void) snprintf(snapname, sizeof (snapname), "%s@%llu", osname,
3696 error = dsl_destroy_snapshot(snapname, B_FALSE);
3697 if (error != 0 && error != ENOENT)
3698 fatal(0, "ztest_snapshot_destroy(%s) = %d", snapname, error);
3704 ztest_dmu_objset_create_destroy(ztest_ds_t *zd, uint64_t id)
3710 char name[ZFS_MAX_DATASET_NAME_LEN];
3714 zdtmp = umem_alloc(sizeof (ztest_ds_t), UMEM_NOFAIL);
3716 (void) rw_rdlock(&ztest_name_lock);
3718 (void) snprintf(name, sizeof (name), "%s/temp_%llu",
3719 ztest_opts.zo_pool, (u_longlong_t)id);
3722 * If this dataset exists from a previous run, process its replay log
3723 * half of the time. If we don't replay it, then dsl_destroy_head()
3724 * (invoked from ztest_objset_destroy_cb()) should just throw it away.
3726 if (ztest_random(2) == 0 &&
3727 ztest_dmu_objset_own(name, DMU_OST_OTHER, B_FALSE,
3728 B_TRUE, FTAG, &os) == 0) {
3729 ztest_zd_init(zdtmp, NULL, os);
3730 zil_replay(os, zdtmp, ztest_replay_vector);
3731 ztest_zd_fini(zdtmp);
3732 txg_wait_synced(dmu_objset_pool(os), 0);
3733 dmu_objset_disown(os, B_TRUE, FTAG);
3737 * There may be an old instance of the dataset we're about to
3738 * create lying around from a previous run. If so, destroy it
3739 * and all of its snapshots.
3741 (void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
3742 DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
3745 * Verify that the destroyed dataset is no longer in the namespace.
3747 VERIFY3U(ENOENT, ==, ztest_dmu_objset_own(name, DMU_OST_OTHER, B_TRUE,
3748 B_TRUE, FTAG, &os));
3751 * Verify that we can create a new dataset.
3753 error = ztest_dataset_create(name);
3755 if (error == ENOSPC) {
3756 ztest_record_enospc(FTAG);
3759 fatal(0, "dmu_objset_create(%s) = %d", name, error);
3762 VERIFY0(ztest_dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, B_TRUE,
3765 ztest_zd_init(zdtmp, NULL, os);
3768 * Open the intent log for it.
3770 zilog = zil_open(os, ztest_get_data);
3773 * Put some objects in there, do a little I/O to them,
3774 * and randomly take a couple of snapshots along the way.
3776 iters = ztest_random(5);
3777 for (i = 0; i < iters; i++) {
3778 ztest_dmu_object_alloc_free(zdtmp, id);
3779 if (ztest_random(iters) == 0)
3780 (void) ztest_snapshot_create(name, i);
3784 * Verify that we cannot create an existing dataset.
3786 VERIFY3U(EEXIST, ==,
3787 dmu_objset_create(name, DMU_OST_OTHER, 0, NULL, NULL, NULL));
3790 * Verify that we can hold an objset that is also owned.
3792 VERIFY3U(0, ==, dmu_objset_hold(name, FTAG, &os2));
3793 dmu_objset_rele(os2, FTAG);
3796 * Verify that we cannot own an objset that is already owned.
3798 VERIFY3U(EBUSY, ==, ztest_dmu_objset_own(name, DMU_OST_OTHER,
3799 B_FALSE, B_TRUE, FTAG, &os2));
3802 txg_wait_synced(spa_get_dsl(os->os_spa), 0);
3803 dmu_objset_disown(os, B_TRUE, FTAG);
3804 ztest_zd_fini(zdtmp);
3806 (void) rw_unlock(&ztest_name_lock);
3808 umem_free(zdtmp, sizeof (ztest_ds_t));
3812 * Verify that dmu_snapshot_{create,destroy,open,close} work as expected.
3815 ztest_dmu_snapshot_create_destroy(ztest_ds_t *zd, uint64_t id)
3817 (void) rw_rdlock(&ztest_name_lock);
3818 (void) ztest_snapshot_destroy(zd->zd_name, id);
3819 (void) ztest_snapshot_create(zd->zd_name, id);
3820 (void) rw_unlock(&ztest_name_lock);
3824 * Cleanup non-standard snapshots and clones.
3827 ztest_dsl_dataset_cleanup(char *osname, uint64_t id)
3836 snap1name = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
3837 clone1name = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
3838 snap2name = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
3839 clone2name = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
3840 snap3name = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
3842 (void) snprintf(snap1name, ZFS_MAX_DATASET_NAME_LEN,
3843 "%s@s1_%llu", osname, (u_longlong_t)id);
3844 (void) snprintf(clone1name, ZFS_MAX_DATASET_NAME_LEN,
3845 "%s/c1_%llu", osname, (u_longlong_t)id);
3846 (void) snprintf(snap2name, ZFS_MAX_DATASET_NAME_LEN,
3847 "%s@s2_%llu", clone1name, (u_longlong_t)id);
3848 (void) snprintf(clone2name, ZFS_MAX_DATASET_NAME_LEN,
3849 "%s/c2_%llu", osname, (u_longlong_t)id);
3850 (void) snprintf(snap3name, ZFS_MAX_DATASET_NAME_LEN,
3851 "%s@s3_%llu", clone1name, (u_longlong_t)id);
3853 error = dsl_destroy_head(clone2name);
3854 if (error && error != ENOENT)
3855 fatal(0, "dsl_destroy_head(%s) = %d", clone2name, error);
3856 error = dsl_destroy_snapshot(snap3name, B_FALSE);
3857 if (error && error != ENOENT)
3858 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap3name, error);
3859 error = dsl_destroy_snapshot(snap2name, B_FALSE);
3860 if (error && error != ENOENT)
3861 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap2name, error);
3862 error = dsl_destroy_head(clone1name);
3863 if (error && error != ENOENT)
3864 fatal(0, "dsl_destroy_head(%s) = %d", clone1name, error);
3865 error = dsl_destroy_snapshot(snap1name, B_FALSE);
3866 if (error && error != ENOENT)
3867 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap1name, error);
3869 umem_free(snap1name, ZFS_MAX_DATASET_NAME_LEN);
3870 umem_free(clone1name, ZFS_MAX_DATASET_NAME_LEN);
3871 umem_free(snap2name, ZFS_MAX_DATASET_NAME_LEN);
3872 umem_free(clone2name, ZFS_MAX_DATASET_NAME_LEN);
3873 umem_free(snap3name, ZFS_MAX_DATASET_NAME_LEN);
3877 * Verify dsl_dataset_promote handles EBUSY
3880 ztest_dsl_dataset_promote_busy(ztest_ds_t *zd, uint64_t id)
3888 char *osname = zd->zd_name;
3891 snap1name = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
3892 clone1name = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
3893 snap2name = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
3894 clone2name = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
3895 snap3name = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
3897 (void) rw_rdlock(&ztest_name_lock);
3899 ztest_dsl_dataset_cleanup(osname, id);
3901 (void) snprintf(snap1name, ZFS_MAX_DATASET_NAME_LEN,
3902 "%s@s1_%llu", osname, (u_longlong_t)id);
3903 (void) snprintf(clone1name, ZFS_MAX_DATASET_NAME_LEN,
3904 "%s/c1_%llu", osname, (u_longlong_t)id);
3905 (void) snprintf(snap2name, ZFS_MAX_DATASET_NAME_LEN,
3906 "%s@s2_%llu", clone1name, (u_longlong_t)id);
3907 (void) snprintf(clone2name, ZFS_MAX_DATASET_NAME_LEN,
3908 "%s/c2_%llu", osname, (u_longlong_t)id);
3909 (void) snprintf(snap3name, ZFS_MAX_DATASET_NAME_LEN,
3910 "%s@s3_%llu", clone1name, (u_longlong_t)id);
3912 error = dmu_objset_snapshot_one(osname, strchr(snap1name, '@') + 1);
3913 if (error && error != EEXIST) {
3914 if (error == ENOSPC) {
3915 ztest_record_enospc(FTAG);
3918 fatal(0, "dmu_take_snapshot(%s) = %d", snap1name, error);
3921 error = dmu_objset_clone(clone1name, snap1name);
3923 if (error == ENOSPC) {
3924 ztest_record_enospc(FTAG);
3927 fatal(0, "dmu_objset_create(%s) = %d", clone1name, error);
3930 error = dmu_objset_snapshot_one(clone1name, strchr(snap2name, '@') + 1);
3931 if (error && error != EEXIST) {
3932 if (error == ENOSPC) {
3933 ztest_record_enospc(FTAG);
3936 fatal(0, "dmu_open_snapshot(%s) = %d", snap2name, error);
3939 error = dmu_objset_snapshot_one(clone1name, strchr(snap3name, '@') + 1);
3940 if (error && error != EEXIST) {
3941 if (error == ENOSPC) {
3942 ztest_record_enospc(FTAG);
3945 fatal(0, "dmu_open_snapshot(%s) = %d", snap3name, error);
3948 error = dmu_objset_clone(clone2name, snap3name);
3950 if (error == ENOSPC) {
3951 ztest_record_enospc(FTAG);
3954 fatal(0, "dmu_objset_create(%s) = %d", clone2name, error);
3957 error = ztest_dmu_objset_own(snap2name, DMU_OST_ANY, B_TRUE, B_TRUE,
3960 fatal(0, "dmu_objset_own(%s) = %d", snap2name, error);
3961 error = dsl_dataset_promote(clone2name, NULL);
3962 if (error == ENOSPC) {
3963 dmu_objset_disown(os, B_TRUE, FTAG);
3964 ztest_record_enospc(FTAG);
3968 fatal(0, "dsl_dataset_promote(%s), %d, not EBUSY", clone2name,
3970 dmu_objset_disown(os, B_TRUE, FTAG);
3973 ztest_dsl_dataset_cleanup(osname, id);
3975 (void) rw_unlock(&ztest_name_lock);
3977 umem_free(snap1name, ZFS_MAX_DATASET_NAME_LEN);
3978 umem_free(clone1name, ZFS_MAX_DATASET_NAME_LEN);
3979 umem_free(snap2name, ZFS_MAX_DATASET_NAME_LEN);
3980 umem_free(clone2name, ZFS_MAX_DATASET_NAME_LEN);
3981 umem_free(snap3name, ZFS_MAX_DATASET_NAME_LEN);
3984 #undef OD_ARRAY_SIZE
3985 #define OD_ARRAY_SIZE 4
3988 * Verify that dmu_object_{alloc,free} work as expected.
3991 ztest_dmu_object_alloc_free(ztest_ds_t *zd, uint64_t id)
3998 size = sizeof (ztest_od_t) * OD_ARRAY_SIZE;
3999 od = umem_alloc(size, UMEM_NOFAIL);
4000 batchsize = OD_ARRAY_SIZE;
4002 for (b = 0; b < batchsize; b++)
4003 ztest_od_init(od + b, id, FTAG, b, DMU_OT_UINT64_OTHER,
4007 * Destroy the previous batch of objects, create a new batch,
4008 * and do some I/O on the new objects.
4010 if (ztest_object_init(zd, od, size, B_TRUE) != 0)
4013 while (ztest_random(4 * batchsize) != 0)
4014 ztest_io(zd, od[ztest_random(batchsize)].od_object,
4015 ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
4017 umem_free(od, size);
4020 #undef OD_ARRAY_SIZE
4021 #define OD_ARRAY_SIZE 2
4024 * Verify that dmu_{read,write} work as expected.
4027 ztest_dmu_read_write(ztest_ds_t *zd, uint64_t id)
4032 objset_t *os = zd->zd_os;
4033 size = sizeof (ztest_od_t) * OD_ARRAY_SIZE;
4034 od = umem_alloc(size, UMEM_NOFAIL);
4036 int i, freeit, error;
4038 bufwad_t *packbuf, *bigbuf, *pack, *bigH, *bigT;
4039 uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
4040 uint64_t chunksize = (1000 + ztest_random(1000)) * sizeof (uint64_t);
4041 uint64_t regions = 997;
4042 uint64_t stride = 123456789ULL;
4043 uint64_t width = 40;
4044 int free_percent = 5;
4047 * This test uses two objects, packobj and bigobj, that are always
4048 * updated together (i.e. in the same tx) so that their contents are
4049 * in sync and can be compared. Their contents relate to each other
4050 * in a simple way: packobj is a dense array of 'bufwad' structures,
4051 * while bigobj is a sparse array of the same bufwads. Specifically,
4052 * for any index n, there are three bufwads that should be identical:
4054 * packobj, at offset n * sizeof (bufwad_t)
4055 * bigobj, at the head of the nth chunk
4056 * bigobj, at the tail of the nth chunk
4058 * The chunk size is arbitrary. It doesn't have to be a power of two,
4059 * and it doesn't have any relation to the object blocksize.
4060 * The only requirement is that it can hold at least two bufwads.
4062 * Normally, we write the bufwad to each of these locations.
4063 * However, free_percent of the time we instead write zeroes to
4064 * packobj and perform a dmu_free_range() on bigobj. By comparing
4065 * bigobj to packobj, we can verify that the DMU is correctly
4066 * tracking which parts of an object are allocated and free,
4067 * and that the contents of the allocated blocks are correct.
4071 * Read the directory info. If it's the first time, set things up.
4073 ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0, chunksize);
4074 ztest_od_init(od + 1, id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, 0,
4077 if (ztest_object_init(zd, od, size, B_FALSE) != 0) {
4078 umem_free(od, size);
4082 bigobj = od[0].od_object;
4083 packobj = od[1].od_object;
4084 chunksize = od[0].od_gen;
4085 ASSERT(chunksize == od[1].od_gen);
4088 * Prefetch a random chunk of the big object.
4089 * Our aim here is to get some async reads in flight
4090 * for blocks that we may free below; the DMU should
4091 * handle this race correctly.
4093 n = ztest_random(regions) * stride + ztest_random(width);
4094 s = 1 + ztest_random(2 * width - 1);
4095 dmu_prefetch(os, bigobj, 0, n * chunksize, s * chunksize,
4096 ZIO_PRIORITY_SYNC_READ);
4099 * Pick a random index and compute the offsets into packobj and bigobj.
4101 n = ztest_random(regions) * stride + ztest_random(width);
4102 s = 1 + ztest_random(width - 1);
4104 packoff = n * sizeof (bufwad_t);
4105 packsize = s * sizeof (bufwad_t);
4107 bigoff = n * chunksize;
4108 bigsize = s * chunksize;
4110 packbuf = umem_alloc(packsize, UMEM_NOFAIL);
4111 bigbuf = umem_alloc(bigsize, UMEM_NOFAIL);
4114 * free_percent of the time, free a range of bigobj rather than
4117 freeit = (ztest_random(100) < free_percent);
4120 * Read the current contents of our objects.
4122 error = dmu_read(os, packobj, packoff, packsize, packbuf,
4125 error = dmu_read(os, bigobj, bigoff, bigsize, bigbuf,
4130 * Get a tx for the mods to both packobj and bigobj.
4132 tx = dmu_tx_create(os);
4134 dmu_tx_hold_write(tx, packobj, packoff, packsize);
4137 dmu_tx_hold_free(tx, bigobj, bigoff, bigsize);
4139 dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
4141 /* This accounts for setting the checksum/compression. */
4142 dmu_tx_hold_bonus(tx, bigobj);
4144 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4146 umem_free(packbuf, packsize);
4147 umem_free(bigbuf, bigsize);
4148 umem_free(od, size);
4152 enum zio_checksum cksum;
4154 cksum = (enum zio_checksum)
4155 ztest_random_dsl_prop(ZFS_PROP_CHECKSUM);
4156 } while (cksum >= ZIO_CHECKSUM_LEGACY_FUNCTIONS);
4157 dmu_object_set_checksum(os, bigobj, cksum, tx);
4159 enum zio_compress comp;
4161 comp = (enum zio_compress)
4162 ztest_random_dsl_prop(ZFS_PROP_COMPRESSION);
4163 } while (comp >= ZIO_COMPRESS_LEGACY_FUNCTIONS);
4164 dmu_object_set_compress(os, bigobj, comp, tx);
4167 * For each index from n to n + s, verify that the existing bufwad
4168 * in packobj matches the bufwads at the head and tail of the
4169 * corresponding chunk in bigobj. Then update all three bufwads
4170 * with the new values we want to write out.
4172 for (i = 0; i < s; i++) {
4174 pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
4176 bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
4178 bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
4180 ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
4181 ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
4183 if (pack->bw_txg > txg)
4184 fatal(0, "future leak: got %llx, open txg is %llx",
4187 if (pack->bw_data != 0 && pack->bw_index != n + i)
4188 fatal(0, "wrong index: got %llx, wanted %llx+%llx",
4189 pack->bw_index, n, i);
4191 if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
4192 fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
4194 if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
4195 fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
4198 bzero(pack, sizeof (bufwad_t));
4200 pack->bw_index = n + i;
4202 pack->bw_data = 1 + ztest_random(-2ULL);
4209 * We've verified all the old bufwads, and made new ones.
4210 * Now write them out.
4212 dmu_write(os, packobj, packoff, packsize, packbuf, tx);
4215 if (ztest_opts.zo_verbose >= 7) {
4216 (void) printf("freeing offset %llx size %llx"
4218 (u_longlong_t)bigoff,
4219 (u_longlong_t)bigsize,
4222 VERIFY(0 == dmu_free_range(os, bigobj, bigoff, bigsize, tx));
4224 if (ztest_opts.zo_verbose >= 7) {
4225 (void) printf("writing offset %llx size %llx"
4227 (u_longlong_t)bigoff,
4228 (u_longlong_t)bigsize,
4231 dmu_write(os, bigobj, bigoff, bigsize, bigbuf, tx);
4237 * Sanity check the stuff we just wrote.
4240 void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
4241 void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
4243 VERIFY(0 == dmu_read(os, packobj, packoff,
4244 packsize, packcheck, DMU_READ_PREFETCH));
4245 VERIFY(0 == dmu_read(os, bigobj, bigoff,
4246 bigsize, bigcheck, DMU_READ_PREFETCH));
4248 ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
4249 ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
4251 umem_free(packcheck, packsize);
4252 umem_free(bigcheck, bigsize);
4255 umem_free(packbuf, packsize);
4256 umem_free(bigbuf, bigsize);
4257 umem_free(od, size);
4261 compare_and_update_pbbufs(uint64_t s, bufwad_t *packbuf, bufwad_t *bigbuf,
4262 uint64_t bigsize, uint64_t n, uint64_t chunksize, uint64_t txg)
4270 * For each index from n to n + s, verify that the existing bufwad
4271 * in packobj matches the bufwads at the head and tail of the
4272 * corresponding chunk in bigobj. Then update all three bufwads
4273 * with the new values we want to write out.
4275 for (i = 0; i < s; i++) {
4277 pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
4279 bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
4281 bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
4283 ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
4284 ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
4286 if (pack->bw_txg > txg)
4287 fatal(0, "future leak: got %llx, open txg is %llx",
4290 if (pack->bw_data != 0 && pack->bw_index != n + i)
4291 fatal(0, "wrong index: got %llx, wanted %llx+%llx",
4292 pack->bw_index, n, i);
4294 if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
4295 fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
4297 if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
4298 fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
4300 pack->bw_index = n + i;
4302 pack->bw_data = 1 + ztest_random(-2ULL);
4309 #undef OD_ARRAY_SIZE
4310 #define OD_ARRAY_SIZE 2
4313 ztest_dmu_read_write_zcopy(ztest_ds_t *zd, uint64_t id)
4315 objset_t *os = zd->zd_os;
4322 bufwad_t *packbuf, *bigbuf;
4323 uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
4324 uint64_t blocksize = ztest_random_blocksize();
4325 uint64_t chunksize = blocksize;
4326 uint64_t regions = 997;
4327 uint64_t stride = 123456789ULL;
4329 dmu_buf_t *bonus_db;
4330 arc_buf_t **bigbuf_arcbufs;
4331 dmu_object_info_t doi;
4333 size = sizeof (ztest_od_t) * OD_ARRAY_SIZE;
4334 od = umem_alloc(size, UMEM_NOFAIL);
4337 * This test uses two objects, packobj and bigobj, that are always
4338 * updated together (i.e. in the same tx) so that their contents are
4339 * in sync and can be compared. Their contents relate to each other
4340 * in a simple way: packobj is a dense array of 'bufwad' structures,
4341 * while bigobj is a sparse array of the same bufwads. Specifically,
4342 * for any index n, there are three bufwads that should be identical:
4344 * packobj, at offset n * sizeof (bufwad_t)
4345 * bigobj, at the head of the nth chunk
4346 * bigobj, at the tail of the nth chunk
4348 * The chunk size is set equal to bigobj block size so that
4349 * dmu_assign_arcbuf_by_dbuf() can be tested for object updates.
4353 * Read the directory info. If it's the first time, set things up.
4355 ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0, 0);
4356 ztest_od_init(od + 1, id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, 0,
4360 if (ztest_object_init(zd, od, size, B_FALSE) != 0) {
4361 umem_free(od, size);
4365 bigobj = od[0].od_object;
4366 packobj = od[1].od_object;
4367 blocksize = od[0].od_blocksize;
4368 chunksize = blocksize;
4369 ASSERT(chunksize == od[1].od_gen);
4371 VERIFY(dmu_object_info(os, bigobj, &doi) == 0);
4372 VERIFY(ISP2(doi.doi_data_block_size));
4373 VERIFY(chunksize == doi.doi_data_block_size);
4374 VERIFY(chunksize >= 2 * sizeof (bufwad_t));
4377 * Pick a random index and compute the offsets into packobj and bigobj.
4379 n = ztest_random(regions) * stride + ztest_random(width);
4380 s = 1 + ztest_random(width - 1);
4382 packoff = n * sizeof (bufwad_t);
4383 packsize = s * sizeof (bufwad_t);
4385 bigoff = n * chunksize;
4386 bigsize = s * chunksize;
4388 packbuf = umem_zalloc(packsize, UMEM_NOFAIL);
4389 bigbuf = umem_zalloc(bigsize, UMEM_NOFAIL);
4391 VERIFY3U(0, ==, dmu_bonus_hold(os, bigobj, FTAG, &bonus_db));
4393 bigbuf_arcbufs = umem_zalloc(2 * s * sizeof (arc_buf_t *), UMEM_NOFAIL);
4396 * Iteration 0 test zcopy for DB_UNCACHED dbufs.
4397 * Iteration 1 test zcopy to already referenced dbufs.
4398 * Iteration 2 test zcopy to dirty dbuf in the same txg.
4399 * Iteration 3 test zcopy to dbuf dirty in previous txg.
4400 * Iteration 4 test zcopy when dbuf is no longer dirty.
4401 * Iteration 5 test zcopy when it can't be done.
4402 * Iteration 6 one more zcopy write.
4404 for (i = 0; i < 7; i++) {
4409 * In iteration 5 (i == 5) use arcbufs
4410 * that don't match bigobj blksz to test
4411 * dmu_assign_arcbuf_by_dbuf() when it can't directly
4412 * assign an arcbuf to a dbuf.
4414 for (j = 0; j < s; j++) {
4415 if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) {
4417 dmu_request_arcbuf(bonus_db, chunksize);
4419 bigbuf_arcbufs[2 * j] =
4420 dmu_request_arcbuf(bonus_db, chunksize / 2);
4421 bigbuf_arcbufs[2 * j + 1] =
4422 dmu_request_arcbuf(bonus_db, chunksize / 2);
4427 * Get a tx for the mods to both packobj and bigobj.
4429 tx = dmu_tx_create(os);
4431 dmu_tx_hold_write(tx, packobj, packoff, packsize);
4432 dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
4434 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4436 umem_free(packbuf, packsize);
4437 umem_free(bigbuf, bigsize);
4438 for (j = 0; j < s; j++) {
4440 chunksize < (SPA_MINBLOCKSIZE * 2)) {
4441 dmu_return_arcbuf(bigbuf_arcbufs[j]);
4444 bigbuf_arcbufs[2 * j]);
4446 bigbuf_arcbufs[2 * j + 1]);
4449 umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
4450 umem_free(od, size);
4451 dmu_buf_rele(bonus_db, FTAG);
4456 * 50% of the time don't read objects in the 1st iteration to
4457 * test dmu_assign_arcbuf_by_dbuf() for the case when there are
4458 * no existing dbufs for the specified offsets.
4460 if (i != 0 || ztest_random(2) != 0) {
4461 error = dmu_read(os, packobj, packoff,
4462 packsize, packbuf, DMU_READ_PREFETCH);
4464 error = dmu_read(os, bigobj, bigoff, bigsize,
4465 bigbuf, DMU_READ_PREFETCH);
4468 compare_and_update_pbbufs(s, packbuf, bigbuf, bigsize,
4472 * We've verified all the old bufwads, and made new ones.
4473 * Now write them out.
4475 dmu_write(os, packobj, packoff, packsize, packbuf, tx);
4476 if (ztest_opts.zo_verbose >= 7) {
4477 (void) printf("writing offset %llx size %llx"
4479 (u_longlong_t)bigoff,
4480 (u_longlong_t)bigsize,
4483 for (off = bigoff, j = 0; j < s; j++, off += chunksize) {
4485 if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) {
4486 bcopy((caddr_t)bigbuf + (off - bigoff),
4487 bigbuf_arcbufs[j]->b_data, chunksize);
4489 bcopy((caddr_t)bigbuf + (off - bigoff),
4490 bigbuf_arcbufs[2 * j]->b_data,
4492 bcopy((caddr_t)bigbuf + (off - bigoff) +
4494 bigbuf_arcbufs[2 * j + 1]->b_data,
4499 VERIFY(dmu_buf_hold(os, bigobj, off,
4500 FTAG, &dbt, DMU_READ_NO_PREFETCH) == 0);
4502 if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) {
4503 dmu_assign_arcbuf_by_dbuf(bonus_db, off,
4504 bigbuf_arcbufs[j], tx);
4506 dmu_assign_arcbuf_by_dbuf(bonus_db, off,
4507 bigbuf_arcbufs[2 * j], tx);
4508 dmu_assign_arcbuf_by_dbuf(bonus_db,
4509 off + chunksize / 2,
4510 bigbuf_arcbufs[2 * j + 1], tx);
4513 dmu_buf_rele(dbt, FTAG);
4519 * Sanity check the stuff we just wrote.
4522 void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
4523 void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
4525 VERIFY(0 == dmu_read(os, packobj, packoff,
4526 packsize, packcheck, DMU_READ_PREFETCH));
4527 VERIFY(0 == dmu_read(os, bigobj, bigoff,
4528 bigsize, bigcheck, DMU_READ_PREFETCH));
4530 ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
4531 ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
4533 umem_free(packcheck, packsize);
4534 umem_free(bigcheck, bigsize);
4537 txg_wait_open(dmu_objset_pool(os), 0);
4538 } else if (i == 3) {
4539 txg_wait_synced(dmu_objset_pool(os), 0);
4543 dmu_buf_rele(bonus_db, FTAG);
4544 umem_free(packbuf, packsize);
4545 umem_free(bigbuf, bigsize);
4546 umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
4547 umem_free(od, size);
4552 ztest_dmu_write_parallel(ztest_ds_t *zd, uint64_t id)
4556 od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
4557 uint64_t offset = (1ULL << (ztest_random(20) + 43)) +
4558 (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
4561 * Have multiple threads write to large offsets in an object
4562 * to verify that parallel writes to an object -- even to the
4563 * same blocks within the object -- doesn't cause any trouble.
4565 ztest_od_init(od, ID_PARALLEL, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0, 0);
4567 if (ztest_object_init(zd, od, sizeof (ztest_od_t), B_FALSE) != 0)
4570 while (ztest_random(10) != 0)
4571 ztest_io(zd, od->od_object, offset);
4573 umem_free(od, sizeof (ztest_od_t));
4577 ztest_dmu_prealloc(ztest_ds_t *zd, uint64_t id)
4580 uint64_t offset = (1ULL << (ztest_random(4) + SPA_MAXBLOCKSHIFT)) +
4581 (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
4582 uint64_t count = ztest_random(20) + 1;
4583 uint64_t blocksize = ztest_random_blocksize();
4586 od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
4588 ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0, 0);
4590 if (ztest_object_init(zd, od, sizeof (ztest_od_t),
4591 !ztest_random(2)) != 0) {
4592 umem_free(od, sizeof (ztest_od_t));
4596 if (ztest_truncate(zd, od->od_object, offset, count * blocksize) != 0) {
4597 umem_free(od, sizeof (ztest_od_t));
4601 ztest_prealloc(zd, od->od_object, offset, count * blocksize);
4603 data = umem_zalloc(blocksize, UMEM_NOFAIL);
4605 while (ztest_random(count) != 0) {
4606 uint64_t randoff = offset + (ztest_random(count) * blocksize);
4607 if (ztest_write(zd, od->od_object, randoff, blocksize,
4610 while (ztest_random(4) != 0)
4611 ztest_io(zd, od->od_object, randoff);
4614 umem_free(data, blocksize);
4615 umem_free(od, sizeof (ztest_od_t));
4619 * Verify that zap_{create,destroy,add,remove,update} work as expected.
4621 #define ZTEST_ZAP_MIN_INTS 1
4622 #define ZTEST_ZAP_MAX_INTS 4
4623 #define ZTEST_ZAP_MAX_PROPS 1000
4626 ztest_zap(ztest_ds_t *zd, uint64_t id)
4628 objset_t *os = zd->zd_os;
4631 uint64_t txg, last_txg;
4632 uint64_t value[ZTEST_ZAP_MAX_INTS];
4633 uint64_t zl_ints, zl_intsize, prop;
4636 char propname[100], txgname[100];
4638 char *hc[2] = { "s.acl.h", ".s.open.h.hyLZlg" };
4640 od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
4641 ztest_od_init(od, id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0, 0);
4643 if (ztest_object_init(zd, od, sizeof (ztest_od_t),
4644 !ztest_random(2)) != 0)
4647 object = od->od_object;
4650 * Generate a known hash collision, and verify that
4651 * we can lookup and remove both entries.
4653 tx = dmu_tx_create(os);
4654 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4655 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4658 for (i = 0; i < 2; i++) {
4660 VERIFY3U(0, ==, zap_add(os, object, hc[i], sizeof (uint64_t),
4663 for (i = 0; i < 2; i++) {
4664 VERIFY3U(EEXIST, ==, zap_add(os, object, hc[i],
4665 sizeof (uint64_t), 1, &value[i], tx));
4667 zap_length(os, object, hc[i], &zl_intsize, &zl_ints));
4668 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4669 ASSERT3U(zl_ints, ==, 1);
4671 for (i = 0; i < 2; i++) {
4672 VERIFY3U(0, ==, zap_remove(os, object, hc[i], tx));
4677 * Generate a buch of random entries.
4679 ints = MAX(ZTEST_ZAP_MIN_INTS, object % ZTEST_ZAP_MAX_INTS);
4681 prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
4682 (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
4683 (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
4684 bzero(value, sizeof (value));
4688 * If these zap entries already exist, validate their contents.
4690 error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
4692 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4693 ASSERT3U(zl_ints, ==, 1);
4695 VERIFY(zap_lookup(os, object, txgname, zl_intsize,
4696 zl_ints, &last_txg) == 0);
4698 VERIFY(zap_length(os, object, propname, &zl_intsize,
4701 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4702 ASSERT3U(zl_ints, ==, ints);
4704 VERIFY(zap_lookup(os, object, propname, zl_intsize,
4705 zl_ints, value) == 0);
4707 for (i = 0; i < ints; i++) {
4708 ASSERT3U(value[i], ==, last_txg + object + i);
4711 ASSERT3U(error, ==, ENOENT);
4715 * Atomically update two entries in our zap object.
4716 * The first is named txg_%llu, and contains the txg
4717 * in which the property was last updated. The second
4718 * is named prop_%llu, and the nth element of its value
4719 * should be txg + object + n.
4721 tx = dmu_tx_create(os);
4722 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4723 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4728 fatal(0, "zap future leak: old %llu new %llu", last_txg, txg);
4730 for (i = 0; i < ints; i++)
4731 value[i] = txg + object + i;
4733 VERIFY3U(0, ==, zap_update(os, object, txgname, sizeof (uint64_t),
4735 VERIFY3U(0, ==, zap_update(os, object, propname, sizeof (uint64_t),
4741 * Remove a random pair of entries.
4743 prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
4744 (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
4745 (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
4747 error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
4749 if (error == ENOENT)
4754 tx = dmu_tx_create(os);
4755 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4756 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4759 VERIFY3U(0, ==, zap_remove(os, object, txgname, tx));
4760 VERIFY3U(0, ==, zap_remove(os, object, propname, tx));
4763 umem_free(od, sizeof (ztest_od_t));
4767 * Testcase to test the upgrading of a microzap to fatzap.
4770 ztest_fzap(ztest_ds_t *zd, uint64_t id)
4772 objset_t *os = zd->zd_os;
4774 uint64_t object, txg;
4777 od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
4778 ztest_od_init(od, id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0, 0);
4780 if (ztest_object_init(zd, od, sizeof (ztest_od_t),
4781 !ztest_random(2)) != 0)
4783 object = od->od_object;
4786 * Add entries to this ZAP and make sure it spills over
4787 * and gets upgraded to a fatzap. Also, since we are adding
4788 * 2050 entries we should see ptrtbl growth and leaf-block split.
4790 for (i = 0; i < 2050; i++) {
4791 char name[ZFS_MAX_DATASET_NAME_LEN];
4796 (void) snprintf(name, sizeof (name), "fzap-%llu-%llu",
4797 (u_longlong_t)id, (u_longlong_t)value);
4799 tx = dmu_tx_create(os);
4800 dmu_tx_hold_zap(tx, object, B_TRUE, name);
4801 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4804 error = zap_add(os, object, name, sizeof (uint64_t), 1,
4806 ASSERT(error == 0 || error == EEXIST);
4810 umem_free(od, sizeof (ztest_od_t));
4815 ztest_zap_parallel(ztest_ds_t *zd, uint64_t id)
4817 objset_t *os = zd->zd_os;
4819 uint64_t txg, object, count, wsize, wc, zl_wsize, zl_wc;
4821 int i, namelen, error;
4822 int micro = ztest_random(2);
4823 char name[20], string_value[20];
4826 od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
4827 ztest_od_init(od, ID_PARALLEL, FTAG, micro, DMU_OT_ZAP_OTHER, 0, 0, 0);
4829 if (ztest_object_init(zd, od, sizeof (ztest_od_t), B_FALSE) != 0) {
4830 umem_free(od, sizeof (ztest_od_t));
4834 object = od->od_object;
4837 * Generate a random name of the form 'xxx.....' where each
4838 * x is a random printable character and the dots are dots.
4839 * There are 94 such characters, and the name length goes from
4840 * 6 to 20, so there are 94^3 * 15 = 12,458,760 possible names.
4842 namelen = ztest_random(sizeof (name) - 5) + 5 + 1;
4844 for (i = 0; i < 3; i++)
4845 name[i] = '!' + ztest_random('~' - '!' + 1);
4846 for (; i < namelen - 1; i++)
4850 if ((namelen & 1) || micro) {
4851 wsize = sizeof (txg);
4857 data = string_value;
4861 VERIFY0(zap_count(os, object, &count));
4862 ASSERT(count != -1ULL);
4865 * Select an operation: length, lookup, add, update, remove.
4867 i = ztest_random(5);
4870 tx = dmu_tx_create(os);
4871 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4872 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4874 umem_free(od, sizeof (ztest_od_t));
4877 bcopy(name, string_value, namelen);
4881 bzero(string_value, namelen);
4887 error = zap_length(os, object, name, &zl_wsize, &zl_wc);
4889 ASSERT3U(wsize, ==, zl_wsize);
4890 ASSERT3U(wc, ==, zl_wc);
4892 ASSERT3U(error, ==, ENOENT);
4897 error = zap_lookup(os, object, name, wsize, wc, data);
4899 if (data == string_value &&
4900 bcmp(name, data, namelen) != 0)
4901 fatal(0, "name '%s' != val '%s' len %d",
4902 name, data, namelen);
4904 ASSERT3U(error, ==, ENOENT);
4909 error = zap_add(os, object, name, wsize, wc, data, tx);
4910 ASSERT(error == 0 || error == EEXIST);
4914 VERIFY(zap_update(os, object, name, wsize, wc, data, tx) == 0);
4918 error = zap_remove(os, object, name, tx);
4919 ASSERT(error == 0 || error == ENOENT);
4926 umem_free(od, sizeof (ztest_od_t));
4930 * Commit callback data.
4932 typedef struct ztest_cb_data {
4933 list_node_t zcd_node;
4935 int zcd_expected_err;
4936 boolean_t zcd_added;
4937 boolean_t zcd_called;
4941 /* This is the actual commit callback function */
4943 ztest_commit_callback(void *arg, int error)
4945 ztest_cb_data_t *data = arg;
4946 uint64_t synced_txg;
4948 VERIFY(data != NULL);
4949 VERIFY3S(data->zcd_expected_err, ==, error);
4950 VERIFY(!data->zcd_called);
4952 synced_txg = spa_last_synced_txg(data->zcd_spa);
4953 if (data->zcd_txg > synced_txg)
4954 fatal(0, "commit callback of txg %" PRIu64 " called prematurely"
4955 ", last synced txg = %" PRIu64 "\n", data->zcd_txg,
4958 data->zcd_called = B_TRUE;
4960 if (error == ECANCELED) {
4961 ASSERT0(data->zcd_txg);
4962 ASSERT(!data->zcd_added);
4965 * The private callback data should be destroyed here, but
4966 * since we are going to check the zcd_called field after
4967 * dmu_tx_abort(), we will destroy it there.
4972 ASSERT(data->zcd_added);
4973 ASSERT3U(data->zcd_txg, !=, 0);
4975 (void) mutex_enter(&zcl.zcl_callbacks_lock);
4977 /* See if this cb was called more quickly */
4978 if ((synced_txg - data->zcd_txg) < zc_min_txg_delay)
4979 zc_min_txg_delay = synced_txg - data->zcd_txg;
4981 /* Remove our callback from the list */
4982 list_remove(&zcl.zcl_callbacks, data);
4984 (void) mutex_exit(&zcl.zcl_callbacks_lock);
4986 umem_free(data, sizeof (ztest_cb_data_t));
4989 /* Allocate and initialize callback data structure */
4990 static ztest_cb_data_t *
4991 ztest_create_cb_data(objset_t *os, uint64_t txg)
4993 ztest_cb_data_t *cb_data;
4995 cb_data = umem_zalloc(sizeof (ztest_cb_data_t), UMEM_NOFAIL);
4997 cb_data->zcd_txg = txg;
4998 cb_data->zcd_spa = dmu_objset_spa(os);
4999 list_link_init(&cb_data->zcd_node);
5005 * Commit callback test.
5008 ztest_dmu_commit_callbacks(ztest_ds_t *zd, uint64_t id)
5010 objset_t *os = zd->zd_os;
5013 ztest_cb_data_t *cb_data[3], *tmp_cb;
5014 uint64_t old_txg, txg;
5017 od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
5018 ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0, 0);
5020 if (ztest_object_init(zd, od, sizeof (ztest_od_t), B_FALSE) != 0) {
5021 umem_free(od, sizeof (ztest_od_t));
5025 tx = dmu_tx_create(os);
5027 cb_data[0] = ztest_create_cb_data(os, 0);
5028 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[0]);
5030 dmu_tx_hold_write(tx, od->od_object, 0, sizeof (uint64_t));
5032 /* Every once in a while, abort the transaction on purpose */
5033 if (ztest_random(100) == 0)
5037 error = dmu_tx_assign(tx, TXG_NOWAIT);
5039 txg = error ? 0 : dmu_tx_get_txg(tx);
5041 cb_data[0]->zcd_txg = txg;
5042 cb_data[1] = ztest_create_cb_data(os, txg);
5043 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[1]);
5047 * It's not a strict requirement to call the registered
5048 * callbacks from inside dmu_tx_abort(), but that's what
5049 * it's supposed to happen in the current implementation
5050 * so we will check for that.
5052 for (i = 0; i < 2; i++) {
5053 cb_data[i]->zcd_expected_err = ECANCELED;
5054 VERIFY(!cb_data[i]->zcd_called);
5059 for (i = 0; i < 2; i++) {
5060 VERIFY(cb_data[i]->zcd_called);
5061 umem_free(cb_data[i], sizeof (ztest_cb_data_t));
5064 umem_free(od, sizeof (ztest_od_t));
5068 cb_data[2] = ztest_create_cb_data(os, txg);
5069 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[2]);
5072 * Read existing data to make sure there isn't a future leak.
5074 VERIFY(0 == dmu_read(os, od->od_object, 0, sizeof (uint64_t),
5075 &old_txg, DMU_READ_PREFETCH));
5078 fatal(0, "future leak: got %" PRIu64 ", open txg is %" PRIu64,
5081 dmu_write(os, od->od_object, 0, sizeof (uint64_t), &txg, tx);
5083 (void) mutex_enter(&zcl.zcl_callbacks_lock);
5086 * Since commit callbacks don't have any ordering requirement and since
5087 * it is theoretically possible for a commit callback to be called
5088 * after an arbitrary amount of time has elapsed since its txg has been
5089 * synced, it is difficult to reliably determine whether a commit
5090 * callback hasn't been called due to high load or due to a flawed
5093 * In practice, we will assume that if after a certain number of txgs a
5094 * commit callback hasn't been called, then most likely there's an
5095 * implementation bug..
5097 tmp_cb = list_head(&zcl.zcl_callbacks);
5098 if (tmp_cb != NULL &&
5099 tmp_cb->zcd_txg + ZTEST_COMMIT_CB_THRESH < txg) {
5100 fatal(0, "Commit callback threshold exceeded, oldest txg: %"
5101 PRIu64 ", open txg: %" PRIu64 "\n", tmp_cb->zcd_txg, txg);
5105 * Let's find the place to insert our callbacks.
5107 * Even though the list is ordered by txg, it is possible for the
5108 * insertion point to not be the end because our txg may already be
5109 * quiescing at this point and other callbacks in the open txg
5110 * (from other objsets) may have sneaked in.
5112 tmp_cb = list_tail(&zcl.zcl_callbacks);
5113 while (tmp_cb != NULL && tmp_cb->zcd_txg > txg)
5114 tmp_cb = list_prev(&zcl.zcl_callbacks, tmp_cb);
5116 /* Add the 3 callbacks to the list */
5117 for (i = 0; i < 3; i++) {
5119 list_insert_head(&zcl.zcl_callbacks, cb_data[i]);
5121 list_insert_after(&zcl.zcl_callbacks, tmp_cb,
5124 cb_data[i]->zcd_added = B_TRUE;
5125 VERIFY(!cb_data[i]->zcd_called);
5127 tmp_cb = cb_data[i];
5132 (void) mutex_exit(&zcl.zcl_callbacks_lock);
5136 umem_free(od, sizeof (ztest_od_t));
5140 * Visit each object in the dataset. Verify that its properties
5141 * are consistent what was stored in the block tag when it was created,
5142 * and that its unused bonus buffer space has not been overwritten.
5146 ztest_verify_dnode_bt(ztest_ds_t *zd, uint64_t id)
5148 objset_t *os = zd->zd_os;
5152 for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, 0)) {
5153 ztest_block_tag_t *bt = NULL;
5154 dmu_object_info_t doi;
5157 if (dmu_bonus_hold(os, obj, FTAG, &db) != 0)
5160 dmu_object_info_from_db(db, &doi);
5161 if (doi.doi_bonus_size >= sizeof (*bt))
5162 bt = ztest_bt_bonus(db);
5164 if (bt && bt->bt_magic == BT_MAGIC) {
5165 ztest_bt_verify(bt, os, obj, doi.doi_dnodesize,
5166 bt->bt_offset, bt->bt_gen, bt->bt_txg,
5168 ztest_verify_unused_bonus(db, bt, obj, os, bt->bt_gen);
5171 dmu_buf_rele(db, FTAG);
5177 ztest_dsl_prop_get_set(ztest_ds_t *zd, uint64_t id)
5179 zfs_prop_t proplist[] = {
5181 ZFS_PROP_COMPRESSION,
5187 (void) rw_rdlock(&ztest_name_lock);
5189 for (p = 0; p < sizeof (proplist) / sizeof (proplist[0]); p++)
5190 (void) ztest_dsl_prop_set_uint64(zd->zd_name, proplist[p],
5191 ztest_random_dsl_prop(proplist[p]), (int)ztest_random(2));
5193 VERIFY0(ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_RECORDSIZE,
5194 ztest_random_blocksize(), (int)ztest_random(2)));
5196 (void) rw_unlock(&ztest_name_lock);
5201 ztest_spa_prop_get_set(ztest_ds_t *zd, uint64_t id)
5203 nvlist_t *props = NULL;
5205 (void) rw_rdlock(&ztest_name_lock);
5207 (void) ztest_spa_prop_set_uint64(ZPOOL_PROP_DEDUPDITTO,
5208 ZIO_DEDUPDITTO_MIN + ztest_random(ZIO_DEDUPDITTO_MIN));
5210 VERIFY0(spa_prop_get(ztest_spa, &props));
5212 if (ztest_opts.zo_verbose >= 6)
5213 dump_nvlist(props, 4);
5217 (void) rw_unlock(&ztest_name_lock);
5221 user_release_one(const char *snapname, const char *holdname)
5223 nvlist_t *snaps, *holds;
5226 snaps = fnvlist_alloc();
5227 holds = fnvlist_alloc();
5228 fnvlist_add_boolean(holds, holdname);
5229 fnvlist_add_nvlist(snaps, snapname, holds);
5230 fnvlist_free(holds);
5231 error = dsl_dataset_user_release(snaps, NULL);
5232 fnvlist_free(snaps);
5237 * Test snapshot hold/release and deferred destroy.
5240 ztest_dmu_snapshot_hold(ztest_ds_t *zd, uint64_t id)
5243 objset_t *os = zd->zd_os;
5247 char clonename[100];
5249 char osname[ZFS_MAX_DATASET_NAME_LEN];
5252 (void) rw_rdlock(&ztest_name_lock);
5254 dmu_objset_name(os, osname);
5256 (void) snprintf(snapname, sizeof (snapname), "sh1_%llu",
5258 (void) snprintf(fullname, sizeof (fullname), "%s@%s", osname, snapname);
5259 (void) snprintf(clonename, sizeof (clonename),
5260 "%s/ch1_%llu", osname, (u_longlong_t)id);
5261 (void) snprintf(tag, sizeof (tag), "tag_%llu", (u_longlong_t)id);
5264 * Clean up from any previous run.
5266 error = dsl_destroy_head(clonename);
5267 if (error != ENOENT)
5269 error = user_release_one(fullname, tag);
5270 if (error != ESRCH && error != ENOENT)
5272 error = dsl_destroy_snapshot(fullname, B_FALSE);
5273 if (error != ENOENT)
5277 * Create snapshot, clone it, mark snap for deferred destroy,
5278 * destroy clone, verify snap was also destroyed.
5280 error = dmu_objset_snapshot_one(osname, snapname);
5282 if (error == ENOSPC) {
5283 ztest_record_enospc("dmu_objset_snapshot");
5286 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
5289 error = dmu_objset_clone(clonename, fullname);
5291 if (error == ENOSPC) {
5292 ztest_record_enospc("dmu_objset_clone");
5295 fatal(0, "dmu_objset_clone(%s) = %d", clonename, error);
5298 error = dsl_destroy_snapshot(fullname, B_TRUE);
5300 fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
5304 error = dsl_destroy_head(clonename);
5306 fatal(0, "dsl_destroy_head(%s) = %d", clonename, error);
5308 error = dmu_objset_hold(fullname, FTAG, &origin);
5309 if (error != ENOENT)
5310 fatal(0, "dmu_objset_hold(%s) = %d", fullname, error);
5313 * Create snapshot, add temporary hold, verify that we can't
5314 * destroy a held snapshot, mark for deferred destroy,
5315 * release hold, verify snapshot was destroyed.
5317 error = dmu_objset_snapshot_one(osname, snapname);
5319 if (error == ENOSPC) {
5320 ztest_record_enospc("dmu_objset_snapshot");
5323 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
5326 holds = fnvlist_alloc();
5327 fnvlist_add_string(holds, fullname, tag);
5328 error = dsl_dataset_user_hold(holds, 0, NULL);
5329 fnvlist_free(holds);
5331 if (error == ENOSPC) {
5332 ztest_record_enospc("dsl_dataset_user_hold");
5335 fatal(0, "dsl_dataset_user_hold(%s, %s) = %u",
5336 fullname, tag, error);
5339 error = dsl_destroy_snapshot(fullname, B_FALSE);
5340 if (error != EBUSY) {
5341 fatal(0, "dsl_destroy_snapshot(%s, B_FALSE) = %d",
5345 error = dsl_destroy_snapshot(fullname, B_TRUE);
5347 fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
5351 error = user_release_one(fullname, tag);
5353 fatal(0, "user_release_one(%s, %s) = %d", fullname, tag, error);
5355 VERIFY3U(dmu_objset_hold(fullname, FTAG, &origin), ==, ENOENT);
5358 (void) rw_unlock(&ztest_name_lock);
5362 * Inject random faults into the on-disk data.
5366 ztest_fault_inject(ztest_ds_t *zd, uint64_t id)
5368 ztest_shared_t *zs = ztest_shared;
5369 spa_t *spa = ztest_spa;
5373 uint64_t bad = 0x1990c0ffeedecadeull;
5378 int bshift = SPA_MAXBLOCKSHIFT + 2;
5384 boolean_t islog = B_FALSE;
5386 path0 = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
5387 pathrand = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
5389 mutex_enter(&ztest_vdev_lock);
5390 maxfaults = MAXFAULTS();
5391 leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz;
5392 mirror_save = zs->zs_mirrors;
5393 mutex_exit(&ztest_vdev_lock);
5395 ASSERT(leaves >= 1);
5398 * Grab the name lock as reader. There are some operations
5399 * which don't like to have their vdevs changed while
5400 * they are in progress (i.e. spa_change_guid). Those
5401 * operations will have grabbed the name lock as writer.
5403 (void) rw_rdlock(&ztest_name_lock);
5406 * We need SCL_STATE here because we're going to look at vd0->vdev_tsd.
5408 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5410 if (ztest_random(2) == 0) {
5412 * Inject errors on a normal data device or slog device.
5414 top = ztest_random_vdev_top(spa, B_TRUE);
5415 leaf = ztest_random(leaves) + zs->zs_splits;
5418 * Generate paths to the first leaf in this top-level vdev,
5419 * and to the random leaf we selected. We'll induce transient
5420 * write failures and random online/offline activity on leaf 0,
5421 * and we'll write random garbage to the randomly chosen leaf.
5423 (void) snprintf(path0, MAXPATHLEN, ztest_dev_template,
5424 ztest_opts.zo_dir, ztest_opts.zo_pool,
5425 top * leaves + zs->zs_splits);
5426 (void) snprintf(pathrand, MAXPATHLEN, ztest_dev_template,
5427 ztest_opts.zo_dir, ztest_opts.zo_pool,
5428 top * leaves + leaf);
5430 vd0 = vdev_lookup_by_path(spa->spa_root_vdev, path0);
5431 if (vd0 != NULL && vd0->vdev_top->vdev_islog)
5435 * If the top-level vdev needs to be resilvered
5436 * then we only allow faults on the device that is
5439 if (vd0 != NULL && maxfaults != 1 &&
5440 (!vdev_resilver_needed(vd0->vdev_top, NULL, NULL) ||
5441 vd0->vdev_resilver_txg != 0)) {
5443 * Make vd0 explicitly claim to be unreadable,
5444 * or unwriteable, or reach behind its back
5445 * and close the underlying fd. We can do this if
5446 * maxfaults == 0 because we'll fail and reexecute,
5447 * and we can do it if maxfaults >= 2 because we'll
5448 * have enough redundancy. If maxfaults == 1, the
5449 * combination of this with injection of random data
5450 * corruption below exceeds the pool's fault tolerance.
5452 vdev_file_t *vf = vd0->vdev_tsd;
5454 if (vf != NULL && ztest_random(3) == 0) {
5455 (void) close(vf->vf_vnode->v_fd);
5456 vf->vf_vnode->v_fd = -1;
5457 } else if (ztest_random(2) == 0) {
5458 vd0->vdev_cant_read = B_TRUE;
5460 vd0->vdev_cant_write = B_TRUE;
5462 guid0 = vd0->vdev_guid;
5466 * Inject errors on an l2cache device.
5468 spa_aux_vdev_t *sav = &spa->spa_l2cache;
5470 if (sav->sav_count == 0) {
5471 spa_config_exit(spa, SCL_STATE, FTAG);
5472 (void) rw_unlock(&ztest_name_lock);
5475 vd0 = sav->sav_vdevs[ztest_random(sav->sav_count)];
5476 guid0 = vd0->vdev_guid;
5477 (void) strcpy(path0, vd0->vdev_path);
5478 (void) strcpy(pathrand, vd0->vdev_path);
5482 maxfaults = INT_MAX; /* no limit on cache devices */
5485 spa_config_exit(spa, SCL_STATE, FTAG);
5486 (void) rw_unlock(&ztest_name_lock);
5489 * If we can tolerate two or more faults, or we're dealing
5490 * with a slog, randomly online/offline vd0.
5492 if ((maxfaults >= 2 || islog) && guid0 != 0) {
5493 if (ztest_random(10) < 6) {
5494 int flags = (ztest_random(2) == 0 ?
5495 ZFS_OFFLINE_TEMPORARY : 0);
5498 * We have to grab the zs_name_lock as writer to
5499 * prevent a race between offlining a slog and
5500 * destroying a dataset. Offlining the slog will
5501 * grab a reference on the dataset which may cause
5502 * dsl_destroy_head() to fail with EBUSY thus
5503 * leaving the dataset in an inconsistent state.
5506 (void) rw_wrlock(&ztest_name_lock);
5508 VERIFY(vdev_offline(spa, guid0, flags) != EBUSY);
5511 (void) rw_unlock(&ztest_name_lock);
5514 * Ideally we would like to be able to randomly
5515 * call vdev_[on|off]line without holding locks
5516 * to force unpredictable failures but the side
5517 * effects of vdev_[on|off]line prevent us from
5518 * doing so. We grab the ztest_vdev_lock here to
5519 * prevent a race between injection testing and
5522 mutex_enter(&ztest_vdev_lock);
5523 (void) vdev_online(spa, guid0, 0, NULL);
5524 mutex_exit(&ztest_vdev_lock);
5532 * We have at least single-fault tolerance, so inject data corruption.
5534 fd = open(pathrand, O_RDWR);
5536 if (fd == -1) /* we hit a gap in the device namespace */
5539 fsize = lseek(fd, 0, SEEK_END);
5541 while (--iters != 0) {
5543 * The offset must be chosen carefully to ensure that
5544 * we do not inject a given logical block with errors
5545 * on two different leaf devices, because ZFS can not
5546 * tolerate that (if maxfaults==1).
5548 * We divide each leaf into chunks of size
5549 * (# leaves * SPA_MAXBLOCKSIZE * 4). Within each chunk
5550 * there is a series of ranges to which we can inject errors.
5551 * Each range can accept errors on only a single leaf vdev.
5552 * The error injection ranges are separated by ranges
5553 * which we will not inject errors on any device (DMZs).
5554 * Each DMZ must be large enough such that a single block
5555 * can not straddle it, so that a single block can not be
5556 * a target in two different injection ranges (on different
5559 * For example, with 3 leaves, each chunk looks like:
5560 * 0 to 32M: injection range for leaf 0
5561 * 32M to 64M: DMZ - no injection allowed
5562 * 64M to 96M: injection range for leaf 1
5563 * 96M to 128M: DMZ - no injection allowed
5564 * 128M to 160M: injection range for leaf 2
5565 * 160M to 192M: DMZ - no injection allowed
5567 offset = ztest_random(fsize / (leaves << bshift)) *
5568 (leaves << bshift) + (leaf << bshift) +
5569 (ztest_random(1ULL << (bshift - 1)) & -8ULL);
5572 * Only allow damage to the labels at one end of the vdev.
5574 * If all labels are damaged, the device will be totally
5575 * inaccessible, which will result in loss of data,
5576 * because we also damage (parts of) the other side of
5579 * Additionally, we will always have both an even and an
5580 * odd label, so that we can handle crashes in the
5581 * middle of vdev_config_sync().
5583 if ((leaf & 1) == 0 && offset < VDEV_LABEL_START_SIZE)
5587 * The two end labels are stored at the "end" of the disk, but
5588 * the end of the disk (vdev_psize) is aligned to
5589 * sizeof (vdev_label_t).
5591 uint64_t psize = P2ALIGN(fsize, sizeof (vdev_label_t));
5592 if ((leaf & 1) == 1 &&
5593 offset + sizeof (bad) > psize - VDEV_LABEL_END_SIZE)
5596 mutex_enter(&ztest_vdev_lock);
5597 if (mirror_save != zs->zs_mirrors) {
5598 mutex_exit(&ztest_vdev_lock);
5603 if (pwrite(fd, &bad, sizeof (bad), offset) != sizeof (bad))
5604 fatal(1, "can't inject bad word at 0x%llx in %s",
5607 mutex_exit(&ztest_vdev_lock);
5609 if (ztest_opts.zo_verbose >= 7)
5610 (void) printf("injected bad word into %s,"
5611 " offset 0x%llx\n", pathrand, (u_longlong_t)offset);
5616 umem_free(path0, MAXPATHLEN);
5617 umem_free(pathrand, MAXPATHLEN);
5621 * Verify that DDT repair works as expected.
5624 ztest_ddt_repair(ztest_ds_t *zd, uint64_t id)
5626 ztest_shared_t *zs = ztest_shared;
5627 spa_t *spa = ztest_spa;
5628 objset_t *os = zd->zd_os;
5630 uint64_t object, blocksize, txg, pattern, psize;
5631 enum zio_checksum checksum = spa_dedup_checksum(spa);
5636 int copies = 2 * ZIO_DEDUPDITTO_MIN;
5639 blocksize = ztest_random_blocksize();
5640 blocksize = MIN(blocksize, 2048); /* because we write so many */
5642 od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
5643 ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0, 0);
5645 if (ztest_object_init(zd, od, sizeof (ztest_od_t), B_FALSE) != 0) {
5646 umem_free(od, sizeof (ztest_od_t));
5651 * Take the name lock as writer to prevent anyone else from changing
5652 * the pool and dataset properies we need to maintain during this test.
5654 (void) rw_wrlock(&ztest_name_lock);
5656 if (ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_DEDUP, checksum,
5658 ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_COPIES, 1,
5660 (void) rw_unlock(&ztest_name_lock);
5661 umem_free(od, sizeof (ztest_od_t));
5665 dmu_objset_stats_t dds;
5666 dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
5667 dmu_objset_fast_stat(os, &dds);
5668 dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
5670 object = od[0].od_object;
5671 blocksize = od[0].od_blocksize;
5672 pattern = zs->zs_guid ^ dds.dds_guid;
5674 ASSERT(object != 0);
5676 tx = dmu_tx_create(os);
5677 dmu_tx_hold_write(tx, object, 0, copies * blocksize);
5678 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
5680 (void) rw_unlock(&ztest_name_lock);
5681 umem_free(od, sizeof (ztest_od_t));
5686 * Write all the copies of our block.
5688 for (i = 0; i < copies; i++) {
5689 uint64_t offset = i * blocksize;
5690 int error = dmu_buf_hold(os, object, offset, FTAG, &db,
5691 DMU_READ_NO_PREFETCH);
5693 fatal(B_FALSE, "dmu_buf_hold(%p, %llu, %llu) = %u",
5694 os, (long long)object, (long long) offset, error);
5696 ASSERT(db->db_offset == offset);
5697 ASSERT(db->db_size == blocksize);
5698 ASSERT(ztest_pattern_match(db->db_data, db->db_size, pattern) ||
5699 ztest_pattern_match(db->db_data, db->db_size, 0ULL));
5700 dmu_buf_will_fill(db, tx);
5701 ztest_pattern_set(db->db_data, db->db_size, pattern);
5702 dmu_buf_rele(db, FTAG);
5706 txg_wait_synced(spa_get_dsl(spa), txg);
5709 * Find out what block we got.
5711 VERIFY0(dmu_buf_hold(os, object, 0, FTAG, &db,
5712 DMU_READ_NO_PREFETCH));
5713 blk = *((dmu_buf_impl_t *)db)->db_blkptr;
5714 dmu_buf_rele(db, FTAG);
5717 * Damage the block. Dedup-ditto will save us when we read it later.
5719 psize = BP_GET_PSIZE(&blk);
5720 abd = abd_alloc_linear(psize, B_TRUE);
5721 ztest_pattern_set(abd_to_buf(abd), psize, ~pattern);
5723 (void) zio_wait(zio_rewrite(NULL, spa, 0, &blk,
5724 abd, psize, NULL, NULL, ZIO_PRIORITY_SYNC_WRITE,
5725 ZIO_FLAG_CANFAIL | ZIO_FLAG_INDUCE_DAMAGE, NULL));
5729 (void) rw_unlock(&ztest_name_lock);
5730 umem_free(od, sizeof (ztest_od_t));
5738 ztest_scrub(ztest_ds_t *zd, uint64_t id)
5740 spa_t *spa = ztest_spa;
5742 (void) spa_scan(spa, POOL_SCAN_SCRUB);
5743 (void) poll(NULL, 0, 100); /* wait a moment, then force a restart */
5744 (void) spa_scan(spa, POOL_SCAN_SCRUB);
5748 * Change the guid for the pool.
5752 ztest_reguid(ztest_ds_t *zd, uint64_t id)
5754 spa_t *spa = ztest_spa;
5755 uint64_t orig, load;
5758 if (ztest_opts.zo_mmp_test)
5761 orig = spa_guid(spa);
5762 load = spa_load_guid(spa);
5764 (void) rw_wrlock(&ztest_name_lock);
5765 error = spa_change_guid(spa);
5766 (void) rw_unlock(&ztest_name_lock);
5771 if (ztest_opts.zo_verbose >= 4) {
5772 (void) printf("Changed guid old %llu -> %llu\n",
5773 (u_longlong_t)orig, (u_longlong_t)spa_guid(spa));
5776 VERIFY3U(orig, !=, spa_guid(spa));
5777 VERIFY3U(load, ==, spa_load_guid(spa));
5781 * Rename the pool to a different name and then rename it back.
5785 ztest_spa_rename(ztest_ds_t *zd, uint64_t id)
5787 char *oldname, *newname;
5790 if (ztest_opts.zo_mmp_test)
5793 (void) rw_wrlock(&ztest_name_lock);
5795 oldname = ztest_opts.zo_pool;
5796 newname = umem_alloc(strlen(oldname) + 5, UMEM_NOFAIL);
5797 (void) strcpy(newname, oldname);
5798 (void) strcat(newname, "_tmp");
5803 VERIFY3U(0, ==, spa_rename(oldname, newname));
5806 * Try to open it under the old name, which shouldn't exist
5808 VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
5811 * Open it under the new name and make sure it's still the same spa_t.
5813 VERIFY3U(0, ==, spa_open(newname, &spa, FTAG));
5815 ASSERT(spa == ztest_spa);
5816 spa_close(spa, FTAG);
5819 * Rename it back to the original
5821 VERIFY3U(0, ==, spa_rename(newname, oldname));
5824 * Make sure it can still be opened
5826 VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG));
5828 ASSERT(spa == ztest_spa);
5829 spa_close(spa, FTAG);
5831 umem_free(newname, strlen(newname) + 1);
5833 (void) rw_unlock(&ztest_name_lock);
5837 ztest_fletcher(ztest_ds_t *zd, uint64_t id)
5839 hrtime_t end = gethrtime() + NANOSEC;
5841 while (gethrtime() <= end) {
5842 int run_count = 100;
5844 struct abd *abd_data, *abd_meta;
5849 zio_cksum_t zc_ref_byteswap;
5851 size = ztest_random_blocksize();
5853 buf = umem_alloc(size, UMEM_NOFAIL);
5854 abd_data = abd_alloc(size, B_FALSE);
5855 abd_meta = abd_alloc(size, B_TRUE);
5857 for (i = 0, ptr = buf; i < size / sizeof (*ptr); i++, ptr++)
5858 *ptr = ztest_random(UINT_MAX);
5860 abd_copy_from_buf_off(abd_data, buf, 0, size);
5861 abd_copy_from_buf_off(abd_meta, buf, 0, size);
5863 VERIFY0(fletcher_4_impl_set("scalar"));
5864 fletcher_4_native(buf, size, NULL, &zc_ref);
5865 fletcher_4_byteswap(buf, size, NULL, &zc_ref_byteswap);
5867 VERIFY0(fletcher_4_impl_set("cycle"));
5868 while (run_count-- > 0) {
5870 zio_cksum_t zc_byteswap;
5872 fletcher_4_byteswap(buf, size, NULL, &zc_byteswap);
5873 fletcher_4_native(buf, size, NULL, &zc);
5875 VERIFY0(bcmp(&zc, &zc_ref, sizeof (zc)));
5876 VERIFY0(bcmp(&zc_byteswap, &zc_ref_byteswap,
5877 sizeof (zc_byteswap)));
5879 /* Test ABD - data */
5880 abd_fletcher_4_byteswap(abd_data, size, NULL,
5882 abd_fletcher_4_native(abd_data, size, NULL, &zc);
5884 VERIFY0(bcmp(&zc, &zc_ref, sizeof (zc)));
5885 VERIFY0(bcmp(&zc_byteswap, &zc_ref_byteswap,
5886 sizeof (zc_byteswap)));
5888 /* Test ABD - metadata */
5889 abd_fletcher_4_byteswap(abd_meta, size, NULL,
5891 abd_fletcher_4_native(abd_meta, size, NULL, &zc);
5893 VERIFY0(bcmp(&zc, &zc_ref, sizeof (zc)));
5894 VERIFY0(bcmp(&zc_byteswap, &zc_ref_byteswap,
5895 sizeof (zc_byteswap)));
5899 umem_free(buf, size);
5906 ztest_fletcher_incr(ztest_ds_t *zd, uint64_t id)
5913 zio_cksum_t zc_ref_bswap;
5915 hrtime_t end = gethrtime() + NANOSEC;
5917 while (gethrtime() <= end) {
5918 int run_count = 100;
5920 size = ztest_random_blocksize();
5921 buf = umem_alloc(size, UMEM_NOFAIL);
5923 for (i = 0, ptr = buf; i < size / sizeof (*ptr); i++, ptr++)
5924 *ptr = ztest_random(UINT_MAX);
5926 VERIFY0(fletcher_4_impl_set("scalar"));
5927 fletcher_4_native(buf, size, NULL, &zc_ref);
5928 fletcher_4_byteswap(buf, size, NULL, &zc_ref_bswap);
5930 VERIFY0(fletcher_4_impl_set("cycle"));
5932 while (run_count-- > 0) {
5934 zio_cksum_t zc_bswap;
5937 ZIO_SET_CHECKSUM(&zc, 0, 0, 0, 0);
5938 ZIO_SET_CHECKSUM(&zc_bswap, 0, 0, 0, 0);
5940 while (pos < size) {
5941 size_t inc = 64 * ztest_random(size / 67);
5942 /* sometimes add few bytes to test non-simd */
5943 if (ztest_random(100) < 10)
5944 inc += P2ALIGN(ztest_random(64),
5947 if (inc > (size - pos))
5950 fletcher_4_incremental_native(buf + pos, inc,
5952 fletcher_4_incremental_byteswap(buf + pos, inc,
5958 VERIFY3U(pos, ==, size);
5960 VERIFY(ZIO_CHECKSUM_EQUAL(zc, zc_ref));
5961 VERIFY(ZIO_CHECKSUM_EQUAL(zc_bswap, zc_ref_bswap));
5964 * verify if incremental on the whole buffer is
5965 * equivalent to non-incremental version
5967 ZIO_SET_CHECKSUM(&zc, 0, 0, 0, 0);
5968 ZIO_SET_CHECKSUM(&zc_bswap, 0, 0, 0, 0);
5970 fletcher_4_incremental_native(buf, size, &zc);
5971 fletcher_4_incremental_byteswap(buf, size, &zc_bswap);
5973 VERIFY(ZIO_CHECKSUM_EQUAL(zc, zc_ref));
5974 VERIFY(ZIO_CHECKSUM_EQUAL(zc_bswap, zc_ref_bswap));
5977 umem_free(buf, size);
5982 ztest_check_path(char *path)
5985 /* return true on success */
5986 return (!stat(path, &s));
5990 ztest_get_zdb_bin(char *bin, int len)
5994 * Try to use ZDB_PATH and in-tree zdb path. If not successful, just
5995 * let popen to search through PATH.
5997 if ((zdb_path = getenv("ZDB_PATH"))) {
5998 strlcpy(bin, zdb_path, len); /* In env */
5999 if (!ztest_check_path(bin)) {
6000 ztest_dump_core = 0;
6001 fatal(1, "invalid ZDB_PATH '%s'", bin);
6006 VERIFY(realpath(getexecname(), bin) != NULL);
6007 if (strstr(bin, "/ztest/")) {
6008 strstr(bin, "/ztest/")[0] = '\0'; /* In-tree */
6009 strcat(bin, "/zdb/zdb");
6010 if (ztest_check_path(bin))
6017 * Verify pool integrity by running zdb.
6020 ztest_run_zdb(char *pool)
6026 const int len = MAXPATHLEN + MAXNAMELEN + 20;
6029 bin = umem_alloc(len, UMEM_NOFAIL);
6030 zdb = umem_alloc(len, UMEM_NOFAIL);
6031 zbuf = umem_alloc(1024, UMEM_NOFAIL);
6033 ztest_get_zdb_bin(bin, len);
6036 "%s -bcc%s%s -G -d -U %s %s",
6038 ztest_opts.zo_verbose >= 3 ? "s" : "",
6039 ztest_opts.zo_verbose >= 4 ? "v" : "",
6043 if (ztest_opts.zo_verbose >= 5)
6044 (void) printf("Executing %s\n", strstr(zdb, "zdb "));
6046 fp = popen(zdb, "r");
6048 while (fgets(zbuf, 1024, fp) != NULL)
6049 if (ztest_opts.zo_verbose >= 3)
6050 (void) printf("%s", zbuf);
6052 status = pclose(fp);
6057 ztest_dump_core = 0;
6058 if (WIFEXITED(status))
6059 fatal(0, "'%s' exit code %d", zdb, WEXITSTATUS(status));
6061 fatal(0, "'%s' died with signal %d", zdb, WTERMSIG(status));
6063 umem_free(bin, len);
6064 umem_free(zdb, len);
6065 umem_free(zbuf, 1024);
6069 ztest_walk_pool_directory(char *header)
6073 if (ztest_opts.zo_verbose >= 6)
6074 (void) printf("%s\n", header);
6076 mutex_enter(&spa_namespace_lock);
6077 while ((spa = spa_next(spa)) != NULL)
6078 if (ztest_opts.zo_verbose >= 6)
6079 (void) printf("\t%s\n", spa_name(spa));
6080 mutex_exit(&spa_namespace_lock);
6084 ztest_spa_import_export(char *oldname, char *newname)
6086 nvlist_t *config, *newconfig;
6091 if (ztest_opts.zo_verbose >= 4) {
6092 (void) printf("import/export: old = %s, new = %s\n",
6097 * Clean up from previous runs.
6099 (void) spa_destroy(newname);
6102 * Get the pool's configuration and guid.
6104 VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG));
6107 * Kick off a scrub to tickle scrub/export races.
6109 if (ztest_random(2) == 0)
6110 (void) spa_scan(spa, POOL_SCAN_SCRUB);
6112 pool_guid = spa_guid(spa);
6113 spa_close(spa, FTAG);
6115 ztest_walk_pool_directory("pools before export");
6120 VERIFY3U(0, ==, spa_export(oldname, &config, B_FALSE, B_FALSE));
6122 ztest_walk_pool_directory("pools after export");
6127 newconfig = spa_tryimport(config);
6128 ASSERT(newconfig != NULL);
6129 nvlist_free(newconfig);
6132 * Import it under the new name.
6134 error = spa_import(newname, config, NULL, 0);
6136 dump_nvlist(config, 0);
6137 fatal(B_FALSE, "couldn't import pool %s as %s: error %u",
6138 oldname, newname, error);
6141 ztest_walk_pool_directory("pools after import");
6144 * Try to import it again -- should fail with EEXIST.
6146 VERIFY3U(EEXIST, ==, spa_import(newname, config, NULL, 0));
6149 * Try to import it under a different name -- should fail with EEXIST.
6151 VERIFY3U(EEXIST, ==, spa_import(oldname, config, NULL, 0));
6154 * Verify that the pool is no longer visible under the old name.
6156 VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
6159 * Verify that we can open and close the pool using the new name.
6161 VERIFY3U(0, ==, spa_open(newname, &spa, FTAG));
6162 ASSERT(pool_guid == spa_guid(spa));
6163 spa_close(spa, FTAG);
6165 nvlist_free(config);
6169 ztest_resume(spa_t *spa)
6171 if (spa_suspended(spa) && ztest_opts.zo_verbose >= 6)
6172 (void) printf("resuming from suspended state\n");
6173 spa_vdev_state_enter(spa, SCL_NONE);
6174 vdev_clear(spa, NULL);
6175 (void) spa_vdev_state_exit(spa, NULL, 0);
6176 (void) zio_resume(spa);
6180 ztest_resume_thread(void *arg)
6184 while (!ztest_exiting) {
6185 if (spa_suspended(spa))
6187 (void) poll(NULL, 0, 100);
6190 * Periodically change the zfs_compressed_arc_enabled setting.
6192 if (ztest_random(10) == 0)
6193 zfs_compressed_arc_enabled = ztest_random(2);
6196 * Periodically change the zfs_abd_scatter_enabled setting.
6198 if (ztest_random(10) == 0)
6199 zfs_abd_scatter_enabled = ztest_random(2);
6209 ztest_deadman_alarm(int sig)
6211 fatal(0, "failed to complete within %d seconds of deadline", GRACE);
6216 ztest_execute(int test, ztest_info_t *zi, uint64_t id)
6218 ztest_ds_t *zd = &ztest_ds[id % ztest_opts.zo_datasets];
6219 ztest_shared_callstate_t *zc = ZTEST_GET_SHARED_CALLSTATE(test);
6220 hrtime_t functime = gethrtime();
6223 for (i = 0; i < zi->zi_iters; i++)
6224 zi->zi_func(zd, id);
6226 functime = gethrtime() - functime;
6228 atomic_add_64(&zc->zc_count, 1);
6229 atomic_add_64(&zc->zc_time, functime);
6231 if (ztest_opts.zo_verbose >= 4)
6232 (void) printf("%6.2f sec in %s\n",
6233 (double)functime / NANOSEC, zi->zi_funcname);
6237 ztest_thread(void *arg)
6240 uint64_t id = (uintptr_t)arg;
6241 ztest_shared_t *zs = ztest_shared;
6245 ztest_shared_callstate_t *zc;
6247 while ((now = gethrtime()) < zs->zs_thread_stop) {
6249 * See if it's time to force a crash.
6251 if (now > zs->zs_thread_kill)
6255 * If we're getting ENOSPC with some regularity, stop.
6257 if (zs->zs_enospc_count > 10)
6261 * Pick a random function to execute.
6263 rand = ztest_random(ZTEST_FUNCS);
6264 zi = &ztest_info[rand];
6265 zc = ZTEST_GET_SHARED_CALLSTATE(rand);
6266 call_next = zc->zc_next;
6268 if (now >= call_next &&
6269 atomic_cas_64(&zc->zc_next, call_next, call_next +
6270 ztest_random(2 * zi->zi_interval[0] + 1)) == call_next) {
6271 ztest_execute(rand, zi, id);
6279 ztest_dataset_name(char *dsname, char *pool, int d)
6281 (void) snprintf(dsname, ZFS_MAX_DATASET_NAME_LEN, "%s/ds_%d", pool, d);
6285 ztest_dataset_destroy(int d)
6287 char name[ZFS_MAX_DATASET_NAME_LEN];
6290 ztest_dataset_name(name, ztest_opts.zo_pool, d);
6292 if (ztest_opts.zo_verbose >= 3)
6293 (void) printf("Destroying %s to free up space\n", name);
6296 * Cleanup any non-standard clones and snapshots. In general,
6297 * ztest thread t operates on dataset (t % zopt_datasets),
6298 * so there may be more than one thing to clean up.
6300 for (t = d; t < ztest_opts.zo_threads;
6301 t += ztest_opts.zo_datasets)
6302 ztest_dsl_dataset_cleanup(name, t);
6304 (void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
6305 DS_FIND_SNAPSHOTS | DS_FIND_CHILDREN);
6309 ztest_dataset_dirobj_verify(ztest_ds_t *zd)
6311 uint64_t usedobjs, dirobjs, scratch;
6314 * ZTEST_DIROBJ is the object directory for the entire dataset.
6315 * Therefore, the number of objects in use should equal the
6316 * number of ZTEST_DIROBJ entries, +1 for ZTEST_DIROBJ itself.
6317 * If not, we have an object leak.
6319 * Note that we can only check this in ztest_dataset_open(),
6320 * when the open-context and syncing-context values agree.
6321 * That's because zap_count() returns the open-context value,
6322 * while dmu_objset_space() returns the rootbp fill count.
6324 VERIFY3U(0, ==, zap_count(zd->zd_os, ZTEST_DIROBJ, &dirobjs));
6325 dmu_objset_space(zd->zd_os, &scratch, &scratch, &usedobjs, &scratch);
6326 ASSERT3U(dirobjs + 1, ==, usedobjs);
6330 ztest_dataset_open(int d)
6332 ztest_ds_t *zd = &ztest_ds[d];
6333 uint64_t committed_seq = ZTEST_GET_SHARED_DS(d)->zd_seq;
6336 char name[ZFS_MAX_DATASET_NAME_LEN];
6339 ztest_dataset_name(name, ztest_opts.zo_pool, d);
6341 (void) rw_rdlock(&ztest_name_lock);
6343 error = ztest_dataset_create(name);
6344 if (error == ENOSPC) {
6345 (void) rw_unlock(&ztest_name_lock);
6346 ztest_record_enospc(FTAG);
6349 ASSERT(error == 0 || error == EEXIST);
6351 VERIFY0(ztest_dmu_objset_own(name, DMU_OST_OTHER, B_FALSE,
6353 (void) rw_unlock(&ztest_name_lock);
6355 ztest_zd_init(zd, ZTEST_GET_SHARED_DS(d), os);
6357 zilog = zd->zd_zilog;
6359 if (zilog->zl_header->zh_claim_lr_seq != 0 &&
6360 zilog->zl_header->zh_claim_lr_seq < committed_seq)
6361 fatal(0, "missing log records: claimed %llu < committed %llu",
6362 zilog->zl_header->zh_claim_lr_seq, committed_seq);
6364 ztest_dataset_dirobj_verify(zd);
6366 zil_replay(os, zd, ztest_replay_vector);
6368 ztest_dataset_dirobj_verify(zd);
6370 if (ztest_opts.zo_verbose >= 6)
6371 (void) printf("%s replay %llu blocks, %llu records, seq %llu\n",
6373 (u_longlong_t)zilog->zl_parse_blk_count,
6374 (u_longlong_t)zilog->zl_parse_lr_count,
6375 (u_longlong_t)zilog->zl_replaying_seq);
6377 zilog = zil_open(os, ztest_get_data);
6379 if (zilog->zl_replaying_seq != 0 &&
6380 zilog->zl_replaying_seq < committed_seq)
6381 fatal(0, "missing log records: replayed %llu < committed %llu",
6382 zilog->zl_replaying_seq, committed_seq);
6388 ztest_dataset_close(int d)
6390 ztest_ds_t *zd = &ztest_ds[d];
6392 zil_close(zd->zd_zilog);
6393 txg_wait_synced(spa_get_dsl(zd->zd_os->os_spa), 0);
6394 dmu_objset_disown(zd->zd_os, B_TRUE, zd);
6400 * Kick off threads to run tests on all datasets in parallel.
6403 ztest_run(ztest_shared_t *zs)
6407 kthread_t *resume_thread;
6408 kthread_t **run_threads;
6413 ztest_exiting = B_FALSE;
6416 * Initialize parent/child shared state.
6418 mutex_init(&ztest_vdev_lock, NULL, MUTEX_DEFAULT, NULL);
6419 VERIFY(rwlock_init(&ztest_name_lock, USYNC_THREAD, NULL) == 0);
6421 zs->zs_thread_start = gethrtime();
6422 zs->zs_thread_stop =
6423 zs->zs_thread_start + ztest_opts.zo_passtime * NANOSEC;
6424 zs->zs_thread_stop = MIN(zs->zs_thread_stop, zs->zs_proc_stop);
6425 zs->zs_thread_kill = zs->zs_thread_stop;
6426 if (ztest_random(100) < ztest_opts.zo_killrate) {
6427 zs->zs_thread_kill -=
6428 ztest_random(ztest_opts.zo_passtime * NANOSEC);
6431 mutex_init(&zcl.zcl_callbacks_lock, NULL, MUTEX_DEFAULT, NULL);
6433 list_create(&zcl.zcl_callbacks, sizeof (ztest_cb_data_t),
6434 offsetof(ztest_cb_data_t, zcd_node));
6439 kernel_init(FREAD | FWRITE);
6440 VERIFY0(spa_open(ztest_opts.zo_pool, &spa, FTAG));
6441 spa->spa_debug = B_TRUE;
6442 metaslab_preload_limit = ztest_random(20) + 1;
6445 dmu_objset_stats_t dds;
6446 VERIFY0(ztest_dmu_objset_own(ztest_opts.zo_pool,
6447 DMU_OST_ANY, B_TRUE, B_TRUE, FTAG, &os));
6448 dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
6449 dmu_objset_fast_stat(os, &dds);
6450 dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
6451 zs->zs_guid = dds.dds_guid;
6452 dmu_objset_disown(os, B_TRUE, FTAG);
6454 spa->spa_dedup_ditto = 2 * ZIO_DEDUPDITTO_MIN;
6457 * We don't expect the pool to suspend unless maxfaults == 0,
6458 * in which case ztest_fault_inject() temporarily takes away
6459 * the only valid replica.
6461 if (MAXFAULTS() == 0)
6462 spa->spa_failmode = ZIO_FAILURE_MODE_WAIT;
6464 spa->spa_failmode = ZIO_FAILURE_MODE_PANIC;
6467 * Create a thread to periodically resume suspended I/O.
6469 resume_thread = thread_create(NULL, 0, ztest_resume_thread,
6470 spa, 0, NULL, TS_RUN | TS_JOINABLE, defclsyspri);
6474 * Set a deadman alarm to abort() if we hang.
6476 signal(SIGALRM, ztest_deadman_alarm);
6477 alarm((zs->zs_thread_stop - zs->zs_thread_start) / NANOSEC + GRACE);
6481 * Verify that we can safely inquire about about any object,
6482 * whether it's allocated or not. To make it interesting,
6483 * we probe a 5-wide window around each power of two.
6484 * This hits all edge cases, including zero and the max.
6486 for (t = 0; t < 64; t++) {
6487 for (d = -5; d <= 5; d++) {
6488 error = dmu_object_info(spa->spa_meta_objset,
6489 (1ULL << t) + d, NULL);
6490 ASSERT(error == 0 || error == ENOENT ||
6496 * If we got any ENOSPC errors on the previous run, destroy something.
6498 if (zs->zs_enospc_count != 0) {
6499 int d = ztest_random(ztest_opts.zo_datasets);
6500 ztest_dataset_destroy(d);
6502 zs->zs_enospc_count = 0;
6504 run_threads = umem_zalloc(ztest_opts.zo_threads * sizeof (kthread_t *),
6507 if (ztest_opts.zo_verbose >= 4)
6508 (void) printf("starting main threads...\n");
6511 * Kick off all the tests that run in parallel.
6513 for (t = 0; t < ztest_opts.zo_threads; t++) {
6514 if (t < ztest_opts.zo_datasets && ztest_dataset_open(t) != 0) {
6515 umem_free(run_threads, ztest_opts.zo_threads *
6516 sizeof (kthread_t *));
6520 run_threads[t] = thread_create(NULL, 0, ztest_thread,
6521 (void *)(uintptr_t)t, 0, NULL, TS_RUN | TS_JOINABLE,
6526 * Wait for all of the tests to complete. We go in reverse order
6527 * so we don't close datasets while threads are still using them.
6529 for (t = ztest_opts.zo_threads - 1; t >= 0; t--) {
6530 VERIFY0(thread_join(run_threads[t]));
6531 if (t < ztest_opts.zo_datasets)
6532 ztest_dataset_close(t);
6535 txg_wait_synced(spa_get_dsl(spa), 0);
6537 zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(spa));
6538 zs->zs_space = metaslab_class_get_space(spa_normal_class(spa));
6540 umem_free(run_threads, ztest_opts.zo_threads * sizeof (kthread_t *));
6542 /* Kill the resume thread */
6543 ztest_exiting = B_TRUE;
6544 VERIFY0(thread_join(resume_thread));
6548 * Right before closing the pool, kick off a bunch of async I/O;
6549 * spa_close() should wait for it to complete.
6551 for (object = 1; object < 50; object++) {
6552 dmu_prefetch(spa->spa_meta_objset, object, 0, 0, 1ULL << 20,
6553 ZIO_PRIORITY_SYNC_READ);
6556 /* Verify that at least one commit cb was called in a timely fashion */
6557 if (zc_cb_counter >= ZTEST_COMMIT_CB_MIN_REG)
6558 VERIFY0(zc_min_txg_delay);
6560 spa_close(spa, FTAG);
6563 * Verify that we can loop over all pools.
6565 mutex_enter(&spa_namespace_lock);
6566 for (spa = spa_next(NULL); spa != NULL; spa = spa_next(spa))
6567 if (ztest_opts.zo_verbose > 3)
6568 (void) printf("spa_next: found %s\n", spa_name(spa));
6569 mutex_exit(&spa_namespace_lock);
6572 * Verify that we can export the pool and reimport it under a
6575 if ((ztest_random(2) == 0) && !ztest_opts.zo_mmp_test) {
6576 char name[ZFS_MAX_DATASET_NAME_LEN];
6577 (void) snprintf(name, sizeof (name), "%s_import",
6578 ztest_opts.zo_pool);
6579 ztest_spa_import_export(ztest_opts.zo_pool, name);
6580 ztest_spa_import_export(name, ztest_opts.zo_pool);
6585 list_destroy(&zcl.zcl_callbacks);
6586 mutex_destroy(&zcl.zcl_callbacks_lock);
6587 (void) rwlock_destroy(&ztest_name_lock);
6588 mutex_destroy(&ztest_vdev_lock);
6594 ztest_ds_t *zd = &ztest_ds[0];
6598 if (ztest_opts.zo_verbose >= 3)
6599 (void) printf("testing spa_freeze()...\n");
6601 kernel_init(FREAD | FWRITE);
6602 VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
6603 VERIFY3U(0, ==, ztest_dataset_open(0));
6604 spa->spa_debug = B_TRUE;
6608 * Force the first log block to be transactionally allocated.
6609 * We have to do this before we freeze the pool -- otherwise
6610 * the log chain won't be anchored.
6612 while (BP_IS_HOLE(&zd->zd_zilog->zl_header->zh_log)) {
6613 ztest_dmu_object_alloc_free(zd, 0);
6614 zil_commit(zd->zd_zilog, 0);
6617 txg_wait_synced(spa_get_dsl(spa), 0);
6620 * Freeze the pool. This stops spa_sync() from doing anything,
6621 * so that the only way to record changes from now on is the ZIL.
6626 * Because it is hard to predict how much space a write will actually
6627 * require beforehand, we leave ourselves some fudge space to write over
6630 uint64_t capacity = metaslab_class_get_space(spa_normal_class(spa)) / 2;
6633 * Run tests that generate log records but don't alter the pool config
6634 * or depend on DSL sync tasks (snapshots, objset create/destroy, etc).
6635 * We do a txg_wait_synced() after each iteration to force the txg
6636 * to increase well beyond the last synced value in the uberblock.
6637 * The ZIL should be OK with that.
6639 * Run a random number of times less than zo_maxloops and ensure we do
6640 * not run out of space on the pool.
6642 while (ztest_random(10) != 0 &&
6643 numloops++ < ztest_opts.zo_maxloops &&
6644 metaslab_class_get_alloc(spa_normal_class(spa)) < capacity) {
6646 ztest_od_init(&od, 0, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0, 0);
6647 VERIFY0(ztest_object_init(zd, &od, sizeof (od), B_FALSE));
6648 ztest_io(zd, od.od_object,
6649 ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
6650 txg_wait_synced(spa_get_dsl(spa), 0);
6654 * Commit all of the changes we just generated.
6656 zil_commit(zd->zd_zilog, 0);
6657 txg_wait_synced(spa_get_dsl(spa), 0);
6660 * Close our dataset and close the pool.
6662 ztest_dataset_close(0);
6663 spa_close(spa, FTAG);
6667 * Open and close the pool and dataset to induce log replay.
6669 kernel_init(FREAD | FWRITE);
6670 VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
6671 ASSERT(spa_freeze_txg(spa) == UINT64_MAX);
6672 VERIFY3U(0, ==, ztest_dataset_open(0));
6673 spa->spa_debug = B_TRUE;
6675 txg_wait_synced(spa_get_dsl(spa), 0);
6676 ztest_dataset_close(0);
6677 ztest_reguid(NULL, 0);
6679 spa_close(spa, FTAG);
6684 print_time(hrtime_t t, char *timebuf)
6686 hrtime_t s = t / NANOSEC;
6687 hrtime_t m = s / 60;
6688 hrtime_t h = m / 60;
6689 hrtime_t d = h / 24;
6698 (void) sprintf(timebuf,
6699 "%llud%02lluh%02llum%02llus", d, h, m, s);
6701 (void) sprintf(timebuf, "%lluh%02llum%02llus", h, m, s);
6703 (void) sprintf(timebuf, "%llum%02llus", m, s);
6705 (void) sprintf(timebuf, "%llus", s);
6709 make_random_props(void)
6713 VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
6714 if (ztest_random(2) == 0)
6716 VERIFY(nvlist_add_uint64(props, "autoreplace", 1) == 0);
6722 * Import a storage pool with the given name.
6725 ztest_import(ztest_shared_t *zs)
6727 libzfs_handle_t *hdl;
6728 importargs_t args = { 0 };
6730 nvlist_t *cfg = NULL;
6732 char *searchdirs[nsearch];
6733 char *name = ztest_opts.zo_pool;
6734 int flags = ZFS_IMPORT_MISSING_LOG;
6737 mutex_init(&ztest_vdev_lock, NULL, MUTEX_DEFAULT, NULL);
6738 VERIFY(rwlock_init(&ztest_name_lock, USYNC_THREAD, NULL) == 0);
6740 kernel_init(FREAD | FWRITE);
6741 hdl = libzfs_init();
6743 searchdirs[0] = ztest_opts.zo_dir;
6744 args.paths = nsearch;
6745 args.path = searchdirs;
6746 args.can_be_active = B_FALSE;
6748 error = zpool_tryimport(hdl, name, &cfg, &args);
6750 (void) fatal(0, "No pools found\n");
6752 VERIFY0(spa_import(name, cfg, NULL, flags));
6753 VERIFY0(spa_open(name, &spa, FTAG));
6754 zs->zs_metaslab_sz =
6755 1ULL << spa->spa_root_vdev->vdev_child[0]->vdev_ms_shift;
6756 spa_close(spa, FTAG);
6761 if (!ztest_opts.zo_mmp_test) {
6762 ztest_run_zdb(ztest_opts.zo_pool);
6764 ztest_run_zdb(ztest_opts.zo_pool);
6767 (void) rwlock_destroy(&ztest_name_lock);
6768 mutex_destroy(&ztest_vdev_lock);
6772 * Create a storage pool with the given name and initial vdev size.
6773 * Then test spa_freeze() functionality.
6776 ztest_init(ztest_shared_t *zs)
6779 nvlist_t *nvroot, *props;
6782 mutex_init(&ztest_vdev_lock, NULL, MUTEX_DEFAULT, NULL);
6783 VERIFY(rwlock_init(&ztest_name_lock, USYNC_THREAD, NULL) == 0);
6785 kernel_init(FREAD | FWRITE);
6788 * Create the storage pool.
6790 (void) spa_destroy(ztest_opts.zo_pool);
6791 ztest_shared->zs_vdev_next_leaf = 0;
6793 zs->zs_mirrors = ztest_opts.zo_mirrors;
6794 nvroot = make_vdev_root(NULL, NULL, NULL, ztest_opts.zo_vdev_size, 0,
6795 0, ztest_opts.zo_raidz, zs->zs_mirrors, 1);
6796 props = make_random_props();
6797 for (i = 0; i < SPA_FEATURES; i++) {
6799 VERIFY3S(-1, !=, asprintf(&buf, "feature@%s",
6800 spa_feature_table[i].fi_uname));
6801 VERIFY3U(0, ==, nvlist_add_uint64(props, buf, 0));
6805 spa_create(ztest_opts.zo_pool, nvroot, props, NULL, NULL));
6806 nvlist_free(nvroot);
6809 VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
6810 zs->zs_metaslab_sz =
6811 1ULL << spa->spa_root_vdev->vdev_child[0]->vdev_ms_shift;
6812 spa_close(spa, FTAG);
6816 if (!ztest_opts.zo_mmp_test) {
6817 ztest_run_zdb(ztest_opts.zo_pool);
6819 ztest_run_zdb(ztest_opts.zo_pool);
6822 (void) rwlock_destroy(&ztest_name_lock);
6823 mutex_destroy(&ztest_vdev_lock);
6829 static char ztest_name_data[] = "/tmp/ztest.data.XXXXXX";
6831 ztest_fd_data = mkstemp(ztest_name_data);
6832 ASSERT3S(ztest_fd_data, >=, 0);
6833 (void) unlink(ztest_name_data);
6837 shared_data_size(ztest_shared_hdr_t *hdr)
6841 size = hdr->zh_hdr_size;
6842 size += hdr->zh_opts_size;
6843 size += hdr->zh_size;
6844 size += hdr->zh_stats_size * hdr->zh_stats_count;
6845 size += hdr->zh_ds_size * hdr->zh_ds_count;
6854 ztest_shared_hdr_t *hdr;
6856 hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
6857 PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
6858 ASSERT(hdr != MAP_FAILED);
6860 VERIFY3U(0, ==, ftruncate(ztest_fd_data, sizeof (ztest_shared_hdr_t)));
6862 hdr->zh_hdr_size = sizeof (ztest_shared_hdr_t);
6863 hdr->zh_opts_size = sizeof (ztest_shared_opts_t);
6864 hdr->zh_size = sizeof (ztest_shared_t);
6865 hdr->zh_stats_size = sizeof (ztest_shared_callstate_t);
6866 hdr->zh_stats_count = ZTEST_FUNCS;
6867 hdr->zh_ds_size = sizeof (ztest_shared_ds_t);
6868 hdr->zh_ds_count = ztest_opts.zo_datasets;
6870 size = shared_data_size(hdr);
6871 VERIFY3U(0, ==, ftruncate(ztest_fd_data, size));
6873 (void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
6880 ztest_shared_hdr_t *hdr;
6883 hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
6884 PROT_READ, MAP_SHARED, ztest_fd_data, 0);
6885 ASSERT(hdr != MAP_FAILED);
6887 size = shared_data_size(hdr);
6889 (void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
6890 hdr = ztest_shared_hdr = (void *)mmap(0, P2ROUNDUP(size, getpagesize()),
6891 PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
6892 ASSERT(hdr != MAP_FAILED);
6893 buf = (uint8_t *)hdr;
6895 offset = hdr->zh_hdr_size;
6896 ztest_shared_opts = (void *)&buf[offset];
6897 offset += hdr->zh_opts_size;
6898 ztest_shared = (void *)&buf[offset];
6899 offset += hdr->zh_size;
6900 ztest_shared_callstate = (void *)&buf[offset];
6901 offset += hdr->zh_stats_size * hdr->zh_stats_count;
6902 ztest_shared_ds = (void *)&buf[offset];
6906 exec_child(char *cmd, char *libpath, boolean_t ignorekill, int *statusp)
6910 char *cmdbuf = NULL;
6915 cmdbuf = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
6916 (void) strlcpy(cmdbuf, getexecname(), MAXPATHLEN);
6921 fatal(1, "fork failed");
6923 if (pid == 0) { /* child */
6924 char *emptyargv[2] = { cmd, NULL };
6925 char fd_data_str[12];
6927 struct rlimit rl = { 1024, 1024 };
6928 (void) setrlimit(RLIMIT_NOFILE, &rl);
6930 (void) close(ztest_fd_rand);
6931 VERIFY(11 >= snprintf(fd_data_str, 12, "%d", ztest_fd_data));
6932 VERIFY(0 == setenv("ZTEST_FD_DATA", fd_data_str, 1));
6934 (void) enable_extended_FILE_stdio(-1, -1);
6935 if (libpath != NULL)
6936 VERIFY(0 == setenv("LD_LIBRARY_PATH", libpath, 1));
6937 (void) execv(cmd, emptyargv);
6938 ztest_dump_core = B_FALSE;
6939 fatal(B_TRUE, "exec failed: %s", cmd);
6942 if (cmdbuf != NULL) {
6943 umem_free(cmdbuf, MAXPATHLEN);
6947 while (waitpid(pid, &status, 0) != pid)
6949 if (statusp != NULL)
6952 if (WIFEXITED(status)) {
6953 if (WEXITSTATUS(status) != 0) {
6954 (void) fprintf(stderr, "child exited with code %d\n",
6955 WEXITSTATUS(status));
6959 } else if (WIFSIGNALED(status)) {
6960 if (!ignorekill || WTERMSIG(status) != SIGKILL) {
6961 (void) fprintf(stderr, "child died with signal %d\n",
6967 (void) fprintf(stderr, "something strange happened to child\n");
6974 ztest_run_init(void)
6978 ztest_shared_t *zs = ztest_shared;
6981 * Blow away any existing copy of zpool.cache
6983 (void) remove(spa_config_path);
6985 if (ztest_opts.zo_init == 0) {
6986 if (ztest_opts.zo_verbose >= 1)
6987 (void) printf("Importing pool %s\n",
6988 ztest_opts.zo_pool);
6994 * Create and initialize our storage pool.
6996 for (i = 1; i <= ztest_opts.zo_init; i++) {
6997 bzero(zs, sizeof (ztest_shared_t));
6998 if (ztest_opts.zo_verbose >= 3 &&
6999 ztest_opts.zo_init != 1) {
7000 (void) printf("ztest_init(), pass %d\n", i);
7007 main(int argc, char **argv)
7015 ztest_shared_callstate_t *zc;
7022 char *fd_data_str = getenv("ZTEST_FD_DATA");
7023 struct sigaction action;
7025 (void) setvbuf(stdout, NULL, _IOLBF, 0);
7027 dprintf_setup(&argc, argv);
7029 action.sa_handler = sig_handler;
7030 sigemptyset(&action.sa_mask);
7031 action.sa_flags = 0;
7033 if (sigaction(SIGSEGV, &action, NULL) < 0) {
7034 (void) fprintf(stderr, "ztest: cannot catch SIGSEGV: %s.\n",
7039 if (sigaction(SIGABRT, &action, NULL) < 0) {
7040 (void) fprintf(stderr, "ztest: cannot catch SIGABRT: %s.\n",
7045 ztest_fd_rand = open("/dev/urandom", O_RDONLY);
7046 ASSERT3S(ztest_fd_rand, >=, 0);
7049 process_options(argc, argv);
7054 bcopy(&ztest_opts, ztest_shared_opts,
7055 sizeof (*ztest_shared_opts));
7057 ztest_fd_data = atoi(fd_data_str);
7059 bcopy(ztest_shared_opts, &ztest_opts, sizeof (ztest_opts));
7061 ASSERT3U(ztest_opts.zo_datasets, ==, ztest_shared_hdr->zh_ds_count);
7063 /* Override location of zpool.cache */
7064 VERIFY(asprintf((char **)&spa_config_path, "%s/zpool.cache",
7065 ztest_opts.zo_dir) != -1);
7067 ztest_ds = umem_alloc(ztest_opts.zo_datasets * sizeof (ztest_ds_t),
7072 metaslab_gang_bang = ztest_opts.zo_metaslab_gang_bang;
7073 metaslab_df_alloc_threshold =
7074 zs->zs_metaslab_df_alloc_threshold;
7083 hasalt = (strlen(ztest_opts.zo_alt_ztest) != 0);
7085 if (ztest_opts.zo_verbose >= 1) {
7086 (void) printf("%llu vdevs, %d datasets, %d threads,"
7087 " %llu seconds...\n",
7088 (u_longlong_t)ztest_opts.zo_vdevs,
7089 ztest_opts.zo_datasets,
7090 ztest_opts.zo_threads,
7091 (u_longlong_t)ztest_opts.zo_time);
7094 cmd = umem_alloc(MAXNAMELEN, UMEM_NOFAIL);
7095 (void) strlcpy(cmd, getexecname(), MAXNAMELEN);
7097 zs->zs_do_init = B_TRUE;
7098 if (strlen(ztest_opts.zo_alt_ztest) != 0) {
7099 if (ztest_opts.zo_verbose >= 1) {
7100 (void) printf("Executing older ztest for "
7101 "initialization: %s\n", ztest_opts.zo_alt_ztest);
7103 VERIFY(!exec_child(ztest_opts.zo_alt_ztest,
7104 ztest_opts.zo_alt_libpath, B_FALSE, NULL));
7106 VERIFY(!exec_child(NULL, NULL, B_FALSE, NULL));
7108 zs->zs_do_init = B_FALSE;
7110 zs->zs_proc_start = gethrtime();
7111 zs->zs_proc_stop = zs->zs_proc_start + ztest_opts.zo_time * NANOSEC;
7113 for (f = 0; f < ZTEST_FUNCS; f++) {
7114 zi = &ztest_info[f];
7115 zc = ZTEST_GET_SHARED_CALLSTATE(f);
7116 if (zs->zs_proc_start + zi->zi_interval[0] > zs->zs_proc_stop)
7117 zc->zc_next = UINT64_MAX;
7119 zc->zc_next = zs->zs_proc_start +
7120 ztest_random(2 * zi->zi_interval[0] + 1);
7124 * Run the tests in a loop. These tests include fault injection
7125 * to verify that self-healing data works, and forced crashes
7126 * to verify that we never lose on-disk consistency.
7128 while (gethrtime() < zs->zs_proc_stop) {
7133 * Initialize the workload counters for each function.
7135 for (f = 0; f < ZTEST_FUNCS; f++) {
7136 zc = ZTEST_GET_SHARED_CALLSTATE(f);
7141 /* Set the allocation switch size */
7142 zs->zs_metaslab_df_alloc_threshold =
7143 ztest_random(zs->zs_metaslab_sz / 4) + 1;
7145 if (!hasalt || ztest_random(2) == 0) {
7146 if (hasalt && ztest_opts.zo_verbose >= 1) {
7147 (void) printf("Executing newer ztest: %s\n",
7151 killed = exec_child(cmd, NULL, B_TRUE, &status);
7153 if (hasalt && ztest_opts.zo_verbose >= 1) {
7154 (void) printf("Executing older ztest: %s\n",
7155 ztest_opts.zo_alt_ztest);
7158 killed = exec_child(ztest_opts.zo_alt_ztest,
7159 ztest_opts.zo_alt_libpath, B_TRUE, &status);
7166 if (ztest_opts.zo_verbose >= 1) {
7167 hrtime_t now = gethrtime();
7169 now = MIN(now, zs->zs_proc_stop);
7170 print_time(zs->zs_proc_stop - now, timebuf);
7171 nicenum(zs->zs_space, numbuf);
7173 (void) printf("Pass %3d, %8s, %3llu ENOSPC, "
7174 "%4.1f%% of %5s used, %3.0f%% done, %8s to go\n",
7176 WIFEXITED(status) ? "Complete" : "SIGKILL",
7177 (u_longlong_t)zs->zs_enospc_count,
7178 100.0 * zs->zs_alloc / zs->zs_space,
7180 100.0 * (now - zs->zs_proc_start) /
7181 (ztest_opts.zo_time * NANOSEC), timebuf);
7184 if (ztest_opts.zo_verbose >= 2) {
7185 (void) printf("\nWorkload summary:\n\n");
7186 (void) printf("%7s %9s %s\n",
7187 "Calls", "Time", "Function");
7188 (void) printf("%7s %9s %s\n",
7189 "-----", "----", "--------");
7190 for (f = 0; f < ZTEST_FUNCS; f++) {
7191 zi = &ztest_info[f];
7192 zc = ZTEST_GET_SHARED_CALLSTATE(f);
7193 print_time(zc->zc_time, timebuf);
7194 (void) printf("%7llu %9s %s\n",
7195 (u_longlong_t)zc->zc_count, timebuf,
7198 (void) printf("\n");
7202 * It's possible that we killed a child during a rename test,
7203 * in which case we'll have a 'ztest_tmp' pool lying around
7204 * instead of 'ztest'. Do a blind rename in case this happened.
7207 if (spa_open(ztest_opts.zo_pool, &spa, FTAG) == 0) {
7208 spa_close(spa, FTAG);
7210 char tmpname[ZFS_MAX_DATASET_NAME_LEN];
7212 kernel_init(FREAD | FWRITE);
7213 (void) snprintf(tmpname, sizeof (tmpname), "%s_tmp",
7214 ztest_opts.zo_pool);
7215 (void) spa_rename(tmpname, ztest_opts.zo_pool);
7219 if (!ztest_opts.zo_mmp_test)
7220 ztest_run_zdb(ztest_opts.zo_pool);
7223 if (ztest_opts.zo_verbose >= 1) {
7225 (void) printf("%d runs of older ztest: %s\n", older,
7226 ztest_opts.zo_alt_ztest);
7227 (void) printf("%d runs of newer ztest: %s\n", newer,
7230 (void) printf("%d killed, %d completed, %.0f%% kill rate\n",
7231 kills, iters - kills, (100.0 * kills) / MAX(1, iters));
7234 umem_free(cmd, MAXNAMELEN);