4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2011, 2015 by Delphix. All rights reserved.
24 * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
25 * Copyright (c) 2013 Steven Hartland. All rights reserved.
29 * The objective of this program is to provide a DMU/ZAP/SPA stress test
30 * that runs entirely in userland, is easy to use, and easy to extend.
32 * The overall design of the ztest program is as follows:
34 * (1) For each major functional area (e.g. adding vdevs to a pool,
35 * creating and destroying datasets, reading and writing objects, etc)
36 * we have a simple routine to test that functionality. These
37 * individual routines do not have to do anything "stressful".
39 * (2) We turn these simple functionality tests into a stress test by
40 * running them all in parallel, with as many threads as desired,
41 * and spread across as many datasets, objects, and vdevs as desired.
43 * (3) While all this is happening, we inject faults into the pool to
44 * verify that self-healing data really works.
46 * (4) Every time we open a dataset, we change its checksum and compression
47 * functions. Thus even individual objects vary from block to block
48 * in which checksum they use and whether they're compressed.
50 * (5) To verify that we never lose on-disk consistency after a crash,
51 * we run the entire test in a child of the main process.
52 * At random times, the child self-immolates with a SIGKILL.
53 * This is the software equivalent of pulling the power cord.
54 * The parent then runs the test again, using the existing
55 * storage pool, as many times as desired. If backwards compatibility
56 * testing is enabled ztest will sometimes run the "older" version
57 * of ztest after a SIGKILL.
59 * (6) To verify that we don't have future leaks or temporal incursions,
60 * many of the functional tests record the transaction group number
61 * as part of their data. When reading old data, they verify that
62 * the transaction group number is less than the current, open txg.
63 * If you add a new test, please do this if applicable.
65 * (7) Threads are created with a reduced stack size, for sanity checking.
66 * Therefore, it's important not to allocate huge buffers on the stack.
68 * When run with no arguments, ztest runs for about five minutes and
69 * produces no output if successful. To get a little bit of information,
70 * specify -V. To get more information, specify -VV, and so on.
72 * To turn this into an overnight stress test, use -T to specify run time.
74 * You can ask more more vdevs [-v], datasets [-d], or threads [-t]
75 * to increase the pool capacity, fanout, and overall stress level.
77 * Use the -k option to set the desired frequency of kills.
79 * When ztest invokes itself it passes all relevant information through a
80 * temporary file which is mmap-ed in the child process. This allows shared
81 * memory to survive the exec syscall. The ztest_shared_hdr_t struct is always
82 * stored at offset 0 of this file and contains information on the size and
83 * number of shared structures in the file. The information stored in this file
84 * must remain backwards compatible with older versions of ztest so that
85 * ztest can invoke them during backwards compatibility testing (-B).
88 #include <sys/zfs_context.h>
94 #include <sys/dmu_objset.h>
100 #include <sys/resource.h>
103 #include <sys/zil_impl.h>
104 #include <sys/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>
130 #include <execinfo.h> /* for backtrace() */
133 static int ztest_fd_data = -1;
134 static int ztest_fd_rand = -1;
136 typedef struct ztest_shared_hdr {
137 uint64_t zh_hdr_size;
138 uint64_t zh_opts_size;
140 uint64_t zh_stats_size;
141 uint64_t zh_stats_count;
143 uint64_t zh_ds_count;
144 } ztest_shared_hdr_t;
146 static ztest_shared_hdr_t *ztest_shared_hdr;
148 typedef struct ztest_shared_opts {
149 char zo_pool[ZFS_MAX_DATASET_NAME_LEN];
150 char zo_dir[ZFS_MAX_DATASET_NAME_LEN];
151 char zo_alt_ztest[MAXNAMELEN];
152 char zo_alt_libpath[MAXNAMELEN];
154 uint64_t zo_vdevtime;
162 uint64_t zo_passtime;
163 uint64_t zo_killrate;
167 uint64_t zo_maxloops;
168 uint64_t zo_metaslab_gang_bang;
169 } ztest_shared_opts_t;
171 static const ztest_shared_opts_t ztest_opts_defaults = {
172 .zo_pool = { 'z', 't', 'e', 's', 't', '\0' },
173 .zo_dir = { '/', 't', 'm', 'p', '\0' },
174 .zo_alt_ztest = { '\0' },
175 .zo_alt_libpath = { '\0' },
177 .zo_ashift = SPA_MINBLOCKSHIFT,
180 .zo_raidz_parity = 1,
181 .zo_vdev_size = SPA_MINDEVSIZE * 2,
184 .zo_passtime = 60, /* 60 seconds */
185 .zo_killrate = 70, /* 70% kill rate */
188 .zo_time = 300, /* 5 minutes */
189 .zo_maxloops = 50, /* max loops during spa_freeze() */
190 .zo_metaslab_gang_bang = 32 << 10
193 extern uint64_t metaslab_gang_bang;
194 extern uint64_t metaslab_df_alloc_threshold;
195 extern int metaslab_preload_limit;
196 extern boolean_t zfs_compressed_arc_enabled;
197 extern int zfs_abd_scatter_enabled;
199 static ztest_shared_opts_t *ztest_shared_opts;
200 static ztest_shared_opts_t ztest_opts;
202 typedef struct ztest_shared_ds {
206 static ztest_shared_ds_t *ztest_shared_ds;
207 #define ZTEST_GET_SHARED_DS(d) (&ztest_shared_ds[d])
209 #define BT_MAGIC 0x123456789abcdefULL
210 #define MAXFAULTS() \
211 (MAX(zs->zs_mirrors, 1) * (ztest_opts.zo_raidz_parity + 1) - 1)
215 ZTEST_IO_WRITE_PATTERN,
216 ZTEST_IO_WRITE_ZEROES,
223 typedef struct ztest_block_tag {
227 uint64_t bt_dnodesize;
234 typedef struct bufwad {
252 #define ZTEST_RANGE_LOCKS 64
253 #define ZTEST_OBJECT_LOCKS 64
256 * Object descriptor. Used as a template for object lookup/create/remove.
258 typedef struct ztest_od {
261 dmu_object_type_t od_type;
262 dmu_object_type_t od_crtype;
263 uint64_t od_blocksize;
264 uint64_t od_crblocksize;
265 uint64_t od_crdnodesize;
268 char od_name[ZFS_MAX_DATASET_NAME_LEN];
274 typedef struct ztest_ds {
275 ztest_shared_ds_t *zd_shared;
277 rwlock_t zd_zilog_lock;
279 ztest_od_t *zd_od; /* debugging aid */
280 char zd_name[ZFS_MAX_DATASET_NAME_LEN];
281 kmutex_t zd_dirobj_lock;
282 rll_t zd_object_lock[ZTEST_OBJECT_LOCKS];
283 zll_t zd_range_lock[ZTEST_RANGE_LOCKS];
287 * Per-iteration state.
289 typedef void ztest_func_t(ztest_ds_t *zd, uint64_t id);
291 typedef struct ztest_info {
292 ztest_func_t *zi_func; /* test function */
293 uint64_t zi_iters; /* iterations per execution */
294 uint64_t *zi_interval; /* execute every <interval> seconds */
295 const char *zi_funcname; /* name of test function */
298 typedef struct ztest_shared_callstate {
299 uint64_t zc_count; /* per-pass count */
300 uint64_t zc_time; /* per-pass time */
301 uint64_t zc_next; /* next time to call this function */
302 } ztest_shared_callstate_t;
304 static ztest_shared_callstate_t *ztest_shared_callstate;
305 #define ZTEST_GET_SHARED_CALLSTATE(c) (&ztest_shared_callstate[c])
307 ztest_func_t ztest_dmu_read_write;
308 ztest_func_t ztest_dmu_write_parallel;
309 ztest_func_t ztest_dmu_object_alloc_free;
310 ztest_func_t ztest_dmu_commit_callbacks;
311 ztest_func_t ztest_zap;
312 ztest_func_t ztest_zap_parallel;
313 ztest_func_t ztest_zil_commit;
314 ztest_func_t ztest_zil_remount;
315 ztest_func_t ztest_dmu_read_write_zcopy;
316 ztest_func_t ztest_dmu_objset_create_destroy;
317 ztest_func_t ztest_dmu_prealloc;
318 ztest_func_t ztest_fzap;
319 ztest_func_t ztest_dmu_snapshot_create_destroy;
320 ztest_func_t ztest_dsl_prop_get_set;
321 ztest_func_t ztest_spa_prop_get_set;
322 ztest_func_t ztest_spa_create_destroy;
323 ztest_func_t ztest_fault_inject;
324 ztest_func_t ztest_ddt_repair;
325 ztest_func_t ztest_dmu_snapshot_hold;
326 ztest_func_t ztest_spa_rename;
327 ztest_func_t ztest_scrub;
328 ztest_func_t ztest_dsl_dataset_promote_busy;
329 ztest_func_t ztest_vdev_attach_detach;
330 ztest_func_t ztest_vdev_LUN_growth;
331 ztest_func_t ztest_vdev_add_remove;
332 ztest_func_t ztest_vdev_aux_add_remove;
333 ztest_func_t ztest_split_pool;
334 ztest_func_t ztest_reguid;
335 ztest_func_t ztest_spa_upgrade;
336 ztest_func_t ztest_fletcher;
337 ztest_func_t ztest_fletcher_incr;
338 ztest_func_t ztest_verify_dnode_bt;
340 uint64_t zopt_always = 0ULL * NANOSEC; /* all the time */
341 uint64_t zopt_incessant = 1ULL * NANOSEC / 10; /* every 1/10 second */
342 uint64_t zopt_often = 1ULL * NANOSEC; /* every second */
343 uint64_t zopt_sometimes = 10ULL * NANOSEC; /* every 10 seconds */
344 uint64_t zopt_rarely = 60ULL * NANOSEC; /* every 60 seconds */
346 #define ZTI_INIT(func, iters, interval) \
347 { .zi_func = (func), \
348 .zi_iters = (iters), \
349 .zi_interval = (interval), \
350 .zi_funcname = # func }
352 ztest_info_t ztest_info[] = {
353 ZTI_INIT(ztest_dmu_read_write, 1, &zopt_always),
354 ZTI_INIT(ztest_dmu_write_parallel, 10, &zopt_always),
355 ZTI_INIT(ztest_dmu_object_alloc_free, 1, &zopt_always),
356 ZTI_INIT(ztest_dmu_commit_callbacks, 1, &zopt_always),
357 ZTI_INIT(ztest_zap, 30, &zopt_always),
358 ZTI_INIT(ztest_zap_parallel, 100, &zopt_always),
359 ZTI_INIT(ztest_split_pool, 1, &zopt_always),
360 ZTI_INIT(ztest_zil_commit, 1, &zopt_incessant),
361 ZTI_INIT(ztest_zil_remount, 1, &zopt_sometimes),
362 ZTI_INIT(ztest_dmu_read_write_zcopy, 1, &zopt_often),
363 ZTI_INIT(ztest_dmu_objset_create_destroy, 1, &zopt_often),
364 ZTI_INIT(ztest_dsl_prop_get_set, 1, &zopt_often),
365 ZTI_INIT(ztest_spa_prop_get_set, 1, &zopt_sometimes),
367 ZTI_INIT(ztest_dmu_prealloc, 1, &zopt_sometimes),
369 ZTI_INIT(ztest_fzap, 1, &zopt_sometimes),
370 ZTI_INIT(ztest_dmu_snapshot_create_destroy, 1, &zopt_sometimes),
371 ZTI_INIT(ztest_spa_create_destroy, 1, &zopt_sometimes),
372 ZTI_INIT(ztest_fault_inject, 1, &zopt_sometimes),
373 ZTI_INIT(ztest_ddt_repair, 1, &zopt_sometimes),
374 ZTI_INIT(ztest_dmu_snapshot_hold, 1, &zopt_sometimes),
375 ZTI_INIT(ztest_reguid, 1, &zopt_rarely),
376 ZTI_INIT(ztest_spa_rename, 1, &zopt_rarely),
377 ZTI_INIT(ztest_scrub, 1, &zopt_rarely),
378 ZTI_INIT(ztest_spa_upgrade, 1, &zopt_rarely),
379 ZTI_INIT(ztest_dsl_dataset_promote_busy, 1, &zopt_rarely),
380 ZTI_INIT(ztest_vdev_attach_detach, 1, &zopt_sometimes),
381 ZTI_INIT(ztest_vdev_LUN_growth, 1, &zopt_rarely),
382 ZTI_INIT(ztest_vdev_add_remove, 1, &ztest_opts.zo_vdevtime),
383 ZTI_INIT(ztest_vdev_aux_add_remove, 1, &ztest_opts.zo_vdevtime),
384 ZTI_INIT(ztest_fletcher, 1, &zopt_rarely),
385 ZTI_INIT(ztest_fletcher_incr, 1, &zopt_rarely),
386 ZTI_INIT(ztest_verify_dnode_bt, 1, &zopt_sometimes),
389 #define ZTEST_FUNCS (sizeof (ztest_info) / sizeof (ztest_info_t))
392 * The following struct is used to hold a list of uncalled commit callbacks.
393 * The callbacks are ordered by txg number.
395 typedef struct ztest_cb_list {
396 kmutex_t zcl_callbacks_lock;
397 list_t zcl_callbacks;
401 * Stuff we need to share writably between parent and child.
403 typedef struct ztest_shared {
404 boolean_t zs_do_init;
405 hrtime_t zs_proc_start;
406 hrtime_t zs_proc_stop;
407 hrtime_t zs_thread_start;
408 hrtime_t zs_thread_stop;
409 hrtime_t zs_thread_kill;
410 uint64_t zs_enospc_count;
411 uint64_t zs_vdev_next_leaf;
412 uint64_t zs_vdev_aux;
417 uint64_t zs_metaslab_sz;
418 uint64_t zs_metaslab_df_alloc_threshold;
422 #define ID_PARALLEL -1ULL
424 static char ztest_dev_template[] = "%s/%s.%llua";
425 static char ztest_aux_template[] = "%s/%s.%s.%llu";
426 ztest_shared_t *ztest_shared;
428 static spa_t *ztest_spa = NULL;
429 static ztest_ds_t *ztest_ds;
431 static kmutex_t ztest_vdev_lock;
434 * The ztest_name_lock protects the pool and dataset namespace used by
435 * the individual tests. To modify the namespace, consumers must grab
436 * this lock as writer. Grabbing the lock as reader will ensure that the
437 * namespace does not change while the lock is held.
439 static rwlock_t ztest_name_lock;
441 static boolean_t ztest_dump_core = B_TRUE;
442 static boolean_t ztest_exiting;
444 /* Global commit callback list */
445 static ztest_cb_list_t zcl;
446 /* Commit cb delay */
447 static uint64_t zc_min_txg_delay = UINT64_MAX;
448 static int zc_cb_counter = 0;
451 * Minimum number of commit callbacks that need to be registered for us to check
452 * whether the minimum txg delay is acceptable.
454 #define ZTEST_COMMIT_CB_MIN_REG 100
457 * If a number of txgs equal to this threshold have been created after a commit
458 * callback has been registered but not called, then we assume there is an
459 * implementation bug.
461 #define ZTEST_COMMIT_CB_THRESH (TXG_CONCURRENT_STATES + 1000)
463 extern uint64_t metaslab_gang_bang;
464 extern uint64_t metaslab_df_alloc_threshold;
467 ZTEST_META_DNODE = 0,
472 static void usage(boolean_t) __NORETURN;
475 * These libumem hooks provide a reasonable set of defaults for the allocator's
476 * debugging facilities.
479 _umem_debug_init(void)
481 return ("default,verbose"); /* $UMEM_DEBUG setting */
485 _umem_logging_init(void)
487 return ("fail,contents"); /* $UMEM_LOGGING setting */
490 #define BACKTRACE_SZ 100
492 static void sig_handler(int signo)
494 struct sigaction action;
495 #ifdef __GLIBC__ /* backtrace() is a GNU extension */
497 void *buffer[BACKTRACE_SZ];
499 nptrs = backtrace(buffer, BACKTRACE_SZ);
500 backtrace_symbols_fd(buffer, nptrs, STDERR_FILENO);
504 * Restore default action and re-raise signal so SIGSEGV and
505 * SIGABRT can trigger a core dump.
507 action.sa_handler = SIG_DFL;
508 sigemptyset(&action.sa_mask);
510 (void) sigaction(signo, &action, NULL);
514 #define FATAL_MSG_SZ 1024
519 fatal(int do_perror, char *message, ...)
522 int save_errno = errno;
525 (void) fflush(stdout);
526 buf = umem_alloc(FATAL_MSG_SZ, UMEM_NOFAIL);
528 va_start(args, message);
529 (void) sprintf(buf, "ztest: ");
531 (void) vsprintf(buf + strlen(buf), message, args);
534 (void) snprintf(buf + strlen(buf), FATAL_MSG_SZ - strlen(buf),
535 ": %s", strerror(save_errno));
537 (void) fprintf(stderr, "%s\n", buf);
538 fatal_msg = buf; /* to ease debugging */
545 str2shift(const char *buf)
547 const char *ends = "BKMGTPEZ";
552 for (i = 0; i < strlen(ends); i++) {
553 if (toupper(buf[0]) == ends[i])
556 if (i == strlen(ends)) {
557 (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n",
561 if (buf[1] == '\0' || (toupper(buf[1]) == 'B' && buf[2] == '\0')) {
564 (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n", buf);
570 nicenumtoull(const char *buf)
575 val = strtoull(buf, &end, 0);
577 (void) fprintf(stderr, "ztest: bad numeric value: %s\n", buf);
579 } else if (end[0] == '.') {
580 double fval = strtod(buf, &end);
581 fval *= pow(2, str2shift(end));
582 if (fval > UINT64_MAX) {
583 (void) fprintf(stderr, "ztest: value too large: %s\n",
587 val = (uint64_t)fval;
589 int shift = str2shift(end);
590 if (shift >= 64 || (val << shift) >> shift != val) {
591 (void) fprintf(stderr, "ztest: value too large: %s\n",
601 usage(boolean_t requested)
603 const ztest_shared_opts_t *zo = &ztest_opts_defaults;
605 char nice_vdev_size[10];
606 char nice_gang_bang[10];
607 FILE *fp = requested ? stdout : stderr;
609 nicenum(zo->zo_vdev_size, nice_vdev_size);
610 nicenum(zo->zo_metaslab_gang_bang, nice_gang_bang);
612 (void) fprintf(fp, "Usage: %s\n"
613 "\t[-v vdevs (default: %llu)]\n"
614 "\t[-s size_of_each_vdev (default: %s)]\n"
615 "\t[-a alignment_shift (default: %d)] use 0 for random\n"
616 "\t[-m mirror_copies (default: %d)]\n"
617 "\t[-r raidz_disks (default: %d)]\n"
618 "\t[-R raidz_parity (default: %d)]\n"
619 "\t[-d datasets (default: %d)]\n"
620 "\t[-t threads (default: %d)]\n"
621 "\t[-g gang_block_threshold (default: %s)]\n"
622 "\t[-i init_count (default: %d)] initialize pool i times\n"
623 "\t[-k kill_percentage (default: %llu%%)]\n"
624 "\t[-p pool_name (default: %s)]\n"
625 "\t[-f dir (default: %s)] file directory for vdev files\n"
626 "\t[-V] verbose (use multiple times for ever more blather)\n"
627 "\t[-E] use existing pool instead of creating new one\n"
628 "\t[-T time (default: %llu sec)] total run time\n"
629 "\t[-F freezeloops (default: %llu)] max loops in spa_freeze()\n"
630 "\t[-P passtime (default: %llu sec)] time per pass\n"
631 "\t[-B alt_ztest (default: <none>)] alternate ztest path\n"
632 "\t[-h] (print help)\n"
635 (u_longlong_t)zo->zo_vdevs, /* -v */
636 nice_vdev_size, /* -s */
637 zo->zo_ashift, /* -a */
638 zo->zo_mirrors, /* -m */
639 zo->zo_raidz, /* -r */
640 zo->zo_raidz_parity, /* -R */
641 zo->zo_datasets, /* -d */
642 zo->zo_threads, /* -t */
643 nice_gang_bang, /* -g */
644 zo->zo_init, /* -i */
645 (u_longlong_t)zo->zo_killrate, /* -k */
646 zo->zo_pool, /* -p */
648 (u_longlong_t)zo->zo_time, /* -T */
649 (u_longlong_t)zo->zo_maxloops, /* -F */
650 (u_longlong_t)zo->zo_passtime);
651 exit(requested ? 0 : 1);
655 process_options(int argc, char **argv)
658 ztest_shared_opts_t *zo = &ztest_opts;
662 char altdir[MAXNAMELEN] = { 0 };
664 bcopy(&ztest_opts_defaults, zo, sizeof (*zo));
666 while ((opt = getopt(argc, argv,
667 "v:s:a:m:r:R:d:t:g:i:k:p:f:VET:P:hF:B:")) != EOF) {
684 value = nicenumtoull(optarg);
688 zo->zo_vdevs = value;
691 zo->zo_vdev_size = MAX(SPA_MINDEVSIZE, value);
694 zo->zo_ashift = value;
697 zo->zo_mirrors = value;
700 zo->zo_raidz = MAX(1, value);
703 zo->zo_raidz_parity = MIN(MAX(value, 1), 3);
706 zo->zo_datasets = MAX(1, value);
709 zo->zo_threads = MAX(1, value);
712 zo->zo_metaslab_gang_bang = MAX(SPA_MINBLOCKSIZE << 1,
719 zo->zo_killrate = value;
722 (void) strlcpy(zo->zo_pool, optarg,
723 sizeof (zo->zo_pool));
726 path = realpath(optarg, NULL);
728 (void) fprintf(stderr, "error: %s: %s\n",
729 optarg, strerror(errno));
732 (void) strlcpy(zo->zo_dir, path,
733 sizeof (zo->zo_dir));
747 zo->zo_passtime = MAX(1, value);
750 zo->zo_maxloops = MAX(1, value);
753 (void) strlcpy(altdir, optarg, sizeof (altdir));
765 zo->zo_raidz_parity = MIN(zo->zo_raidz_parity, zo->zo_raidz - 1);
768 (zo->zo_vdevs > 0 ? zo->zo_time * NANOSEC / zo->zo_vdevs :
771 if (strlen(altdir) > 0) {
779 cmd = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
780 realaltdir = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
782 VERIFY(NULL != realpath(getexecname(), cmd));
783 if (0 != access(altdir, F_OK)) {
784 ztest_dump_core = B_FALSE;
785 fatal(B_TRUE, "invalid alternate ztest path: %s",
788 VERIFY(NULL != realpath(altdir, realaltdir));
791 * 'cmd' should be of the form "<anything>/usr/bin/<isa>/ztest".
792 * We want to extract <isa> to determine if we should use
793 * 32 or 64 bit binaries.
795 bin = strstr(cmd, "/usr/bin/");
796 ztest = strstr(bin, "/ztest");
798 isalen = ztest - isa;
799 (void) snprintf(zo->zo_alt_ztest, sizeof (zo->zo_alt_ztest),
800 "%s/usr/bin/%.*s/ztest", realaltdir, isalen, isa);
801 (void) snprintf(zo->zo_alt_libpath, sizeof (zo->zo_alt_libpath),
802 "%s/usr/lib/%.*s", realaltdir, isalen, isa);
804 if (0 != access(zo->zo_alt_ztest, X_OK)) {
805 ztest_dump_core = B_FALSE;
806 fatal(B_TRUE, "invalid alternate ztest: %s",
808 } else if (0 != access(zo->zo_alt_libpath, X_OK)) {
809 ztest_dump_core = B_FALSE;
810 fatal(B_TRUE, "invalid alternate lib directory %s",
814 umem_free(cmd, MAXPATHLEN);
815 umem_free(realaltdir, MAXPATHLEN);
820 ztest_kill(ztest_shared_t *zs)
822 zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(ztest_spa));
823 zs->zs_space = metaslab_class_get_space(spa_normal_class(ztest_spa));
826 * Before we kill off ztest, make sure that the config is updated.
827 * See comment above spa_config_sync().
829 mutex_enter(&spa_namespace_lock);
830 spa_config_sync(ztest_spa, B_FALSE, B_FALSE);
831 mutex_exit(&spa_namespace_lock);
833 (void) kill(getpid(), SIGKILL);
837 ztest_random(uint64_t range)
841 ASSERT3S(ztest_fd_rand, >=, 0);
846 if (read(ztest_fd_rand, &r, sizeof (r)) != sizeof (r))
847 fatal(1, "short read from /dev/urandom");
854 ztest_record_enospc(const char *s)
856 ztest_shared->zs_enospc_count++;
860 ztest_get_ashift(void)
862 if (ztest_opts.zo_ashift == 0)
863 return (SPA_MINBLOCKSHIFT + ztest_random(5));
864 return (ztest_opts.zo_ashift);
868 make_vdev_file(char *path, char *aux, char *pool, size_t size, uint64_t ashift)
874 pathbuf = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
877 ashift = ztest_get_ashift();
883 vdev = ztest_shared->zs_vdev_aux;
884 (void) snprintf(path, MAXPATHLEN,
885 ztest_aux_template, ztest_opts.zo_dir,
886 pool == NULL ? ztest_opts.zo_pool : pool,
889 vdev = ztest_shared->zs_vdev_next_leaf++;
890 (void) snprintf(path, MAXPATHLEN,
891 ztest_dev_template, ztest_opts.zo_dir,
892 pool == NULL ? ztest_opts.zo_pool : pool, vdev);
897 int fd = open(path, O_RDWR | O_CREAT | O_TRUNC, 0666);
899 fatal(1, "can't open %s", path);
900 if (ftruncate(fd, size) != 0)
901 fatal(1, "can't ftruncate %s", path);
905 VERIFY(nvlist_alloc(&file, NV_UNIQUE_NAME, 0) == 0);
906 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_TYPE, VDEV_TYPE_FILE) == 0);
907 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_PATH, path) == 0);
908 VERIFY(nvlist_add_uint64(file, ZPOOL_CONFIG_ASHIFT, ashift) == 0);
909 umem_free(pathbuf, MAXPATHLEN);
915 make_vdev_raidz(char *path, char *aux, char *pool, size_t size,
916 uint64_t ashift, int r)
918 nvlist_t *raidz, **child;
922 return (make_vdev_file(path, aux, pool, size, ashift));
923 child = umem_alloc(r * sizeof (nvlist_t *), UMEM_NOFAIL);
925 for (c = 0; c < r; c++)
926 child[c] = make_vdev_file(path, aux, pool, size, ashift);
928 VERIFY(nvlist_alloc(&raidz, NV_UNIQUE_NAME, 0) == 0);
929 VERIFY(nvlist_add_string(raidz, ZPOOL_CONFIG_TYPE,
930 VDEV_TYPE_RAIDZ) == 0);
931 VERIFY(nvlist_add_uint64(raidz, ZPOOL_CONFIG_NPARITY,
932 ztest_opts.zo_raidz_parity) == 0);
933 VERIFY(nvlist_add_nvlist_array(raidz, ZPOOL_CONFIG_CHILDREN,
936 for (c = 0; c < r; c++)
937 nvlist_free(child[c]);
939 umem_free(child, r * sizeof (nvlist_t *));
945 make_vdev_mirror(char *path, char *aux, char *pool, size_t size,
946 uint64_t ashift, int r, int m)
948 nvlist_t *mirror, **child;
952 return (make_vdev_raidz(path, aux, pool, size, ashift, r));
954 child = umem_alloc(m * sizeof (nvlist_t *), UMEM_NOFAIL);
956 for (c = 0; c < m; c++)
957 child[c] = make_vdev_raidz(path, aux, pool, size, ashift, r);
959 VERIFY(nvlist_alloc(&mirror, NV_UNIQUE_NAME, 0) == 0);
960 VERIFY(nvlist_add_string(mirror, ZPOOL_CONFIG_TYPE,
961 VDEV_TYPE_MIRROR) == 0);
962 VERIFY(nvlist_add_nvlist_array(mirror, ZPOOL_CONFIG_CHILDREN,
965 for (c = 0; c < m; c++)
966 nvlist_free(child[c]);
968 umem_free(child, m * sizeof (nvlist_t *));
974 make_vdev_root(char *path, char *aux, char *pool, size_t size, uint64_t ashift,
975 int log, int r, int m, int t)
977 nvlist_t *root, **child;
982 child = umem_alloc(t * sizeof (nvlist_t *), UMEM_NOFAIL);
984 for (c = 0; c < t; c++) {
985 child[c] = make_vdev_mirror(path, aux, pool, size, ashift,
987 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_IS_LOG,
991 VERIFY(nvlist_alloc(&root, NV_UNIQUE_NAME, 0) == 0);
992 VERIFY(nvlist_add_string(root, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) == 0);
993 VERIFY(nvlist_add_nvlist_array(root, aux ? aux : ZPOOL_CONFIG_CHILDREN,
996 for (c = 0; c < t; c++)
997 nvlist_free(child[c]);
999 umem_free(child, t * sizeof (nvlist_t *));
1005 * Find a random spa version. Returns back a random spa version in the
1006 * range [initial_version, SPA_VERSION_FEATURES].
1009 ztest_random_spa_version(uint64_t initial_version)
1011 uint64_t version = initial_version;
1013 if (version <= SPA_VERSION_BEFORE_FEATURES) {
1015 ztest_random(SPA_VERSION_BEFORE_FEATURES - version + 1);
1018 if (version > SPA_VERSION_BEFORE_FEATURES)
1019 version = SPA_VERSION_FEATURES;
1021 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
1026 ztest_random_blocksize(void)
1029 * Choose a block size >= the ashift.
1030 * If the SPA supports new MAXBLOCKSIZE, test up to 1MB blocks.
1032 int maxbs = SPA_OLD_MAXBLOCKSHIFT;
1033 if (spa_maxblocksize(ztest_spa) == SPA_MAXBLOCKSIZE)
1035 uint64_t block_shift =
1036 ztest_random(maxbs - ztest_spa->spa_max_ashift + 1);
1037 return (1 << (SPA_MINBLOCKSHIFT + block_shift));
1041 ztest_random_dnodesize(void)
1044 int max_slots = spa_maxdnodesize(ztest_spa) >> DNODE_SHIFT;
1046 if (max_slots == DNODE_MIN_SLOTS)
1047 return (DNODE_MIN_SIZE);
1050 * Weight the random distribution more heavily toward smaller
1051 * dnode sizes since that is more likely to reflect real-world
1054 ASSERT3U(max_slots, >, 4);
1055 switch (ztest_random(10)) {
1057 slots = 5 + ztest_random(max_slots - 4);
1060 slots = 2 + ztest_random(3);
1067 return (slots << DNODE_SHIFT);
1071 ztest_random_ibshift(void)
1073 return (DN_MIN_INDBLKSHIFT +
1074 ztest_random(DN_MAX_INDBLKSHIFT - DN_MIN_INDBLKSHIFT + 1));
1078 ztest_random_vdev_top(spa_t *spa, boolean_t log_ok)
1081 vdev_t *rvd = spa->spa_root_vdev;
1084 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1087 top = ztest_random(rvd->vdev_children);
1088 tvd = rvd->vdev_child[top];
1089 } while (tvd->vdev_ishole || (tvd->vdev_islog && !log_ok) ||
1090 tvd->vdev_mg == NULL || tvd->vdev_mg->mg_class == NULL);
1096 ztest_random_dsl_prop(zfs_prop_t prop)
1101 value = zfs_prop_random_value(prop, ztest_random(-1ULL));
1102 } while (prop == ZFS_PROP_CHECKSUM && value == ZIO_CHECKSUM_OFF);
1108 ztest_dsl_prop_set_uint64(char *osname, zfs_prop_t prop, uint64_t value,
1111 const char *propname = zfs_prop_to_name(prop);
1112 const char *valname;
1117 error = dsl_prop_set_int(osname, propname,
1118 (inherit ? ZPROP_SRC_NONE : ZPROP_SRC_LOCAL), value);
1120 if (error == ENOSPC) {
1121 ztest_record_enospc(FTAG);
1126 setpoint = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
1127 VERIFY0(dsl_prop_get_integer(osname, propname, &curval, setpoint));
1129 if (ztest_opts.zo_verbose >= 6) {
1132 err = zfs_prop_index_to_string(prop, curval, &valname);
1134 (void) printf("%s %s = %llu at '%s'\n",
1135 osname, propname, (unsigned long long)curval,
1138 (void) printf("%s %s = %s at '%s'\n",
1139 osname, propname, valname, setpoint);
1141 umem_free(setpoint, MAXPATHLEN);
1147 ztest_spa_prop_set_uint64(zpool_prop_t prop, uint64_t value)
1149 spa_t *spa = ztest_spa;
1150 nvlist_t *props = NULL;
1153 VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
1154 VERIFY(nvlist_add_uint64(props, zpool_prop_to_name(prop), value) == 0);
1156 error = spa_prop_set(spa, props);
1160 if (error == ENOSPC) {
1161 ztest_record_enospc(FTAG);
1171 * Object and range lock mechanics
1174 list_node_t z_lnode;
1175 refcount_t z_refcnt;
1177 zfs_rlock_t z_range_lock;
1182 ztest_znode_t *z_ztznode;
1185 static ztest_znode_t *
1186 ztest_znode_init(uint64_t object)
1188 ztest_znode_t *zp = umem_alloc(sizeof (*zp), UMEM_NOFAIL);
1190 list_link_init(&zp->z_lnode);
1191 refcount_create(&zp->z_refcnt);
1192 zp->z_object = object;
1193 zfs_rlock_init(&zp->z_range_lock);
1199 ztest_znode_fini(ztest_znode_t *zp)
1201 ASSERT(refcount_is_zero(&zp->z_refcnt));
1202 zfs_rlock_destroy(&zp->z_range_lock);
1204 refcount_destroy(&zp->z_refcnt);
1205 list_link_init(&zp->z_lnode);
1206 umem_free(zp, sizeof (*zp));
1210 ztest_zll_init(zll_t *zll)
1212 mutex_init(&zll->z_lock, NULL, MUTEX_DEFAULT, NULL);
1213 list_create(&zll->z_list, sizeof (ztest_znode_t),
1214 offsetof(ztest_znode_t, z_lnode));
1218 ztest_zll_destroy(zll_t *zll)
1220 list_destroy(&zll->z_list);
1221 mutex_destroy(&zll->z_lock);
1224 #define RL_TAG "range_lock"
1225 static ztest_znode_t *
1226 ztest_znode_get(ztest_ds_t *zd, uint64_t object)
1228 zll_t *zll = &zd->zd_range_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1229 ztest_znode_t *zp = NULL;
1230 mutex_enter(&zll->z_lock);
1231 for (zp = list_head(&zll->z_list); (zp);
1232 zp = list_next(&zll->z_list, zp)) {
1233 if (zp->z_object == object) {
1234 refcount_add(&zp->z_refcnt, RL_TAG);
1239 zp = ztest_znode_init(object);
1240 refcount_add(&zp->z_refcnt, RL_TAG);
1241 list_insert_head(&zll->z_list, zp);
1243 mutex_exit(&zll->z_lock);
1248 ztest_znode_put(ztest_ds_t *zd, ztest_znode_t *zp)
1251 ASSERT3U(zp->z_object, !=, 0);
1252 zll = &zd->zd_range_lock[zp->z_object & (ZTEST_OBJECT_LOCKS - 1)];
1253 mutex_enter(&zll->z_lock);
1254 refcount_remove(&zp->z_refcnt, RL_TAG);
1255 if (refcount_is_zero(&zp->z_refcnt)) {
1256 list_remove(&zll->z_list, zp);
1257 ztest_znode_fini(zp);
1259 mutex_exit(&zll->z_lock);
1264 ztest_rll_init(rll_t *rll)
1266 rll->rll_writer = NULL;
1267 rll->rll_readers = 0;
1268 mutex_init(&rll->rll_lock, NULL, MUTEX_DEFAULT, NULL);
1269 cv_init(&rll->rll_cv, NULL, CV_DEFAULT, NULL);
1273 ztest_rll_destroy(rll_t *rll)
1275 ASSERT(rll->rll_writer == NULL);
1276 ASSERT(rll->rll_readers == 0);
1277 mutex_destroy(&rll->rll_lock);
1278 cv_destroy(&rll->rll_cv);
1282 ztest_rll_lock(rll_t *rll, rl_type_t type)
1284 mutex_enter(&rll->rll_lock);
1286 if (type == RL_READER) {
1287 while (rll->rll_writer != NULL)
1288 (void) cv_wait(&rll->rll_cv, &rll->rll_lock);
1291 while (rll->rll_writer != NULL || rll->rll_readers)
1292 (void) cv_wait(&rll->rll_cv, &rll->rll_lock);
1293 rll->rll_writer = curthread;
1296 mutex_exit(&rll->rll_lock);
1300 ztest_rll_unlock(rll_t *rll)
1302 mutex_enter(&rll->rll_lock);
1304 if (rll->rll_writer) {
1305 ASSERT(rll->rll_readers == 0);
1306 rll->rll_writer = NULL;
1308 ASSERT(rll->rll_readers != 0);
1309 ASSERT(rll->rll_writer == NULL);
1313 if (rll->rll_writer == NULL && rll->rll_readers == 0)
1314 cv_broadcast(&rll->rll_cv);
1316 mutex_exit(&rll->rll_lock);
1320 ztest_object_lock(ztest_ds_t *zd, uint64_t object, rl_type_t type)
1322 rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1324 ztest_rll_lock(rll, type);
1328 ztest_object_unlock(ztest_ds_t *zd, uint64_t object)
1330 rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1332 ztest_rll_unlock(rll);
1335 static ztest_zrl_t *
1336 ztest_zrl_init(rl_t *rl, ztest_znode_t *zp)
1338 ztest_zrl_t *zrl = umem_alloc(sizeof (*zrl), UMEM_NOFAIL);
1340 zrl->z_ztznode = zp;
1345 ztest_zrl_fini(ztest_zrl_t *zrl)
1347 umem_free(zrl, sizeof (*zrl));
1350 static ztest_zrl_t *
1351 ztest_range_lock(ztest_ds_t *zd, uint64_t object, uint64_t offset,
1352 uint64_t size, rl_type_t type)
1354 ztest_znode_t *zp = ztest_znode_get(zd, object);
1355 rl_t *rl = zfs_range_lock(&zp->z_range_lock, offset,
1357 return (ztest_zrl_init(rl, zp));
1361 ztest_range_unlock(ztest_ds_t *zd, ztest_zrl_t *zrl)
1363 zfs_range_unlock(zrl->z_rl);
1364 ztest_znode_put(zd, zrl->z_ztznode);
1365 ztest_zrl_fini(zrl);
1369 ztest_zd_init(ztest_ds_t *zd, ztest_shared_ds_t *szd, objset_t *os)
1372 zd->zd_zilog = dmu_objset_zil(os);
1373 zd->zd_shared = szd;
1374 dmu_objset_name(os, zd->zd_name);
1377 if (zd->zd_shared != NULL)
1378 zd->zd_shared->zd_seq = 0;
1380 VERIFY(rwlock_init(&zd->zd_zilog_lock, USYNC_THREAD, NULL) == 0);
1381 mutex_init(&zd->zd_dirobj_lock, NULL, MUTEX_DEFAULT, NULL);
1383 for (l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1384 ztest_rll_init(&zd->zd_object_lock[l]);
1386 for (l = 0; l < ZTEST_RANGE_LOCKS; l++)
1387 ztest_zll_init(&zd->zd_range_lock[l]);
1391 ztest_zd_fini(ztest_ds_t *zd)
1395 mutex_destroy(&zd->zd_dirobj_lock);
1396 (void) rwlock_destroy(&zd->zd_zilog_lock);
1398 for (l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1399 ztest_rll_destroy(&zd->zd_object_lock[l]);
1401 for (l = 0; l < ZTEST_RANGE_LOCKS; l++)
1402 ztest_zll_destroy(&zd->zd_range_lock[l]);
1405 #define TXG_MIGHTWAIT (ztest_random(10) == 0 ? TXG_NOWAIT : TXG_WAIT)
1408 ztest_tx_assign(dmu_tx_t *tx, uint64_t txg_how, const char *tag)
1414 * Attempt to assign tx to some transaction group.
1416 error = dmu_tx_assign(tx, txg_how);
1418 if (error == ERESTART) {
1419 ASSERT(txg_how == TXG_NOWAIT);
1422 ASSERT3U(error, ==, ENOSPC);
1423 ztest_record_enospc(tag);
1428 txg = dmu_tx_get_txg(tx);
1434 ztest_pattern_set(void *buf, uint64_t size, uint64_t value)
1437 uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1445 ztest_pattern_match(void *buf, uint64_t size, uint64_t value)
1448 uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1452 diff |= (value - *ip++);
1459 ztest_bt_generate(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1460 uint64_t dnodesize, uint64_t offset, uint64_t gen, uint64_t txg,
1463 bt->bt_magic = BT_MAGIC;
1464 bt->bt_objset = dmu_objset_id(os);
1465 bt->bt_object = object;
1466 bt->bt_dnodesize = dnodesize;
1467 bt->bt_offset = offset;
1470 bt->bt_crtxg = crtxg;
1474 ztest_bt_verify(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1475 uint64_t dnodesize, uint64_t offset, uint64_t gen, uint64_t txg,
1478 ASSERT3U(bt->bt_magic, ==, BT_MAGIC);
1479 ASSERT3U(bt->bt_objset, ==, dmu_objset_id(os));
1480 ASSERT3U(bt->bt_object, ==, object);
1481 ASSERT3U(bt->bt_dnodesize, ==, dnodesize);
1482 ASSERT3U(bt->bt_offset, ==, offset);
1483 ASSERT3U(bt->bt_gen, <=, gen);
1484 ASSERT3U(bt->bt_txg, <=, txg);
1485 ASSERT3U(bt->bt_crtxg, ==, crtxg);
1488 static ztest_block_tag_t *
1489 ztest_bt_bonus(dmu_buf_t *db)
1491 dmu_object_info_t doi;
1492 ztest_block_tag_t *bt;
1494 dmu_object_info_from_db(db, &doi);
1495 ASSERT3U(doi.doi_bonus_size, <=, db->db_size);
1496 ASSERT3U(doi.doi_bonus_size, >=, sizeof (*bt));
1497 bt = (void *)((char *)db->db_data + doi.doi_bonus_size - sizeof (*bt));
1503 * Generate a token to fill up unused bonus buffer space. Try to make
1504 * it unique to the object, generation, and offset to verify that data
1505 * is not getting overwritten by data from other dnodes.
1507 #define ZTEST_BONUS_FILL_TOKEN(obj, ds, gen, offset) \
1508 (((ds) << 48) | ((gen) << 32) | ((obj) << 8) | (offset))
1511 * Fill up the unused bonus buffer region before the block tag with a
1512 * verifiable pattern. Filling the whole bonus area with non-zero data
1513 * helps ensure that all dnode traversal code properly skips the
1514 * interior regions of large dnodes.
1517 ztest_fill_unused_bonus(dmu_buf_t *db, void *end, uint64_t obj,
1518 objset_t *os, uint64_t gen)
1522 ASSERT(IS_P2ALIGNED((char *)end - (char *)db->db_data, 8));
1524 for (bonusp = db->db_data; bonusp < (uint64_t *)end; bonusp++) {
1525 uint64_t token = ZTEST_BONUS_FILL_TOKEN(obj, dmu_objset_id(os),
1526 gen, bonusp - (uint64_t *)db->db_data);
1532 * Verify that the unused area of a bonus buffer is filled with the
1536 ztest_verify_unused_bonus(dmu_buf_t *db, void *end, uint64_t obj,
1537 objset_t *os, uint64_t gen)
1541 for (bonusp = db->db_data; bonusp < (uint64_t *)end; bonusp++) {
1542 uint64_t token = ZTEST_BONUS_FILL_TOKEN(obj, dmu_objset_id(os),
1543 gen, bonusp - (uint64_t *)db->db_data);
1544 VERIFY3U(*bonusp, ==, token);
1552 #define lrz_type lr_mode
1553 #define lrz_blocksize lr_uid
1554 #define lrz_ibshift lr_gid
1555 #define lrz_bonustype lr_rdev
1556 #define lrz_dnodesize lr_crtime[1]
1559 ztest_log_create(ztest_ds_t *zd, dmu_tx_t *tx, lr_create_t *lr)
1561 char *name = (void *)(lr + 1); /* name follows lr */
1562 size_t namesize = strlen(name) + 1;
1565 if (zil_replaying(zd->zd_zilog, tx))
1568 itx = zil_itx_create(TX_CREATE, sizeof (*lr) + namesize);
1569 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1570 sizeof (*lr) + namesize - sizeof (lr_t));
1572 zil_itx_assign(zd->zd_zilog, itx, tx);
1576 ztest_log_remove(ztest_ds_t *zd, dmu_tx_t *tx, lr_remove_t *lr, uint64_t object)
1578 char *name = (void *)(lr + 1); /* name follows lr */
1579 size_t namesize = strlen(name) + 1;
1582 if (zil_replaying(zd->zd_zilog, tx))
1585 itx = zil_itx_create(TX_REMOVE, sizeof (*lr) + namesize);
1586 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1587 sizeof (*lr) + namesize - sizeof (lr_t));
1589 itx->itx_oid = object;
1590 zil_itx_assign(zd->zd_zilog, itx, tx);
1594 ztest_log_write(ztest_ds_t *zd, dmu_tx_t *tx, lr_write_t *lr)
1597 itx_wr_state_t write_state = ztest_random(WR_NUM_STATES);
1599 if (zil_replaying(zd->zd_zilog, tx))
1602 if (lr->lr_length > ZIL_MAX_LOG_DATA)
1603 write_state = WR_INDIRECT;
1605 itx = zil_itx_create(TX_WRITE,
1606 sizeof (*lr) + (write_state == WR_COPIED ? lr->lr_length : 0));
1608 if (write_state == WR_COPIED &&
1609 dmu_read(zd->zd_os, lr->lr_foid, lr->lr_offset, lr->lr_length,
1610 ((lr_write_t *)&itx->itx_lr) + 1, DMU_READ_NO_PREFETCH) != 0) {
1611 zil_itx_destroy(itx);
1612 itx = zil_itx_create(TX_WRITE, sizeof (*lr));
1613 write_state = WR_NEED_COPY;
1615 itx->itx_private = zd;
1616 itx->itx_wr_state = write_state;
1617 itx->itx_sync = (ztest_random(8) == 0);
1618 itx->itx_sod += (write_state == WR_NEED_COPY ? lr->lr_length : 0);
1620 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1621 sizeof (*lr) - sizeof (lr_t));
1623 zil_itx_assign(zd->zd_zilog, itx, tx);
1627 ztest_log_truncate(ztest_ds_t *zd, dmu_tx_t *tx, lr_truncate_t *lr)
1631 if (zil_replaying(zd->zd_zilog, tx))
1634 itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr));
1635 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1636 sizeof (*lr) - sizeof (lr_t));
1638 itx->itx_sync = B_FALSE;
1639 zil_itx_assign(zd->zd_zilog, itx, tx);
1643 ztest_log_setattr(ztest_ds_t *zd, dmu_tx_t *tx, lr_setattr_t *lr)
1647 if (zil_replaying(zd->zd_zilog, tx))
1650 itx = zil_itx_create(TX_SETATTR, sizeof (*lr));
1651 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1652 sizeof (*lr) - sizeof (lr_t));
1654 itx->itx_sync = B_FALSE;
1655 zil_itx_assign(zd->zd_zilog, itx, tx);
1662 ztest_replay_create(ztest_ds_t *zd, lr_create_t *lr, boolean_t byteswap)
1664 char *name = (void *)(lr + 1); /* name follows lr */
1665 objset_t *os = zd->zd_os;
1666 ztest_block_tag_t *bbt;
1674 byteswap_uint64_array(lr, sizeof (*lr));
1676 ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1677 ASSERT(name[0] != '\0');
1679 tx = dmu_tx_create(os);
1681 dmu_tx_hold_zap(tx, lr->lr_doid, B_TRUE, name);
1683 if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1684 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, B_TRUE, NULL);
1686 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1689 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1693 ASSERT(dmu_objset_zil(os)->zl_replay == !!lr->lr_foid);
1694 bonuslen = DN_BONUS_SIZE(lr->lrz_dnodesize);
1696 if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1697 if (lr->lr_foid == 0) {
1698 lr->lr_foid = zap_create_dnsize(os,
1699 lr->lrz_type, lr->lrz_bonustype,
1700 bonuslen, lr->lrz_dnodesize, tx);
1702 error = zap_create_claim_dnsize(os, lr->lr_foid,
1703 lr->lrz_type, lr->lrz_bonustype,
1704 bonuslen, lr->lrz_dnodesize, tx);
1707 if (lr->lr_foid == 0) {
1708 lr->lr_foid = dmu_object_alloc_dnsize(os,
1709 lr->lrz_type, 0, lr->lrz_bonustype,
1710 bonuslen, lr->lrz_dnodesize, tx);
1712 error = dmu_object_claim_dnsize(os, lr->lr_foid,
1713 lr->lrz_type, 0, lr->lrz_bonustype,
1714 bonuslen, lr->lrz_dnodesize, tx);
1719 ASSERT3U(error, ==, EEXIST);
1720 ASSERT(zd->zd_zilog->zl_replay);
1725 ASSERT(lr->lr_foid != 0);
1727 if (lr->lrz_type != DMU_OT_ZAP_OTHER)
1728 VERIFY3U(0, ==, dmu_object_set_blocksize(os, lr->lr_foid,
1729 lr->lrz_blocksize, lr->lrz_ibshift, tx));
1731 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1732 bbt = ztest_bt_bonus(db);
1733 dmu_buf_will_dirty(db, tx);
1734 ztest_bt_generate(bbt, os, lr->lr_foid, lr->lrz_dnodesize, -1ULL,
1735 lr->lr_gen, txg, txg);
1736 ztest_fill_unused_bonus(db, bbt, lr->lr_foid, os, lr->lr_gen);
1737 dmu_buf_rele(db, FTAG);
1739 VERIFY3U(0, ==, zap_add(os, lr->lr_doid, name, sizeof (uint64_t), 1,
1742 (void) ztest_log_create(zd, tx, lr);
1750 ztest_replay_remove(ztest_ds_t *zd, lr_remove_t *lr, boolean_t byteswap)
1752 char *name = (void *)(lr + 1); /* name follows lr */
1753 objset_t *os = zd->zd_os;
1754 dmu_object_info_t doi;
1756 uint64_t object, txg;
1759 byteswap_uint64_array(lr, sizeof (*lr));
1761 ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1762 ASSERT(name[0] != '\0');
1765 zap_lookup(os, lr->lr_doid, name, sizeof (object), 1, &object));
1766 ASSERT(object != 0);
1768 ztest_object_lock(zd, object, RL_WRITER);
1770 VERIFY3U(0, ==, dmu_object_info(os, object, &doi));
1772 tx = dmu_tx_create(os);
1774 dmu_tx_hold_zap(tx, lr->lr_doid, B_FALSE, name);
1775 dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END);
1777 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1779 ztest_object_unlock(zd, object);
1783 if (doi.doi_type == DMU_OT_ZAP_OTHER) {
1784 VERIFY3U(0, ==, zap_destroy(os, object, tx));
1786 VERIFY3U(0, ==, dmu_object_free(os, object, tx));
1789 VERIFY3U(0, ==, zap_remove(os, lr->lr_doid, name, tx));
1791 (void) ztest_log_remove(zd, tx, lr, object);
1795 ztest_object_unlock(zd, object);
1801 ztest_replay_write(ztest_ds_t *zd, lr_write_t *lr, boolean_t byteswap)
1803 objset_t *os = zd->zd_os;
1804 void *data = lr + 1; /* data follows lr */
1805 uint64_t offset, length;
1806 ztest_block_tag_t *bt = data;
1807 ztest_block_tag_t *bbt;
1808 uint64_t gen, txg, lrtxg, crtxg;
1809 dmu_object_info_t doi;
1812 arc_buf_t *abuf = NULL;
1816 byteswap_uint64_array(lr, sizeof (*lr));
1818 offset = lr->lr_offset;
1819 length = lr->lr_length;
1821 /* If it's a dmu_sync() block, write the whole block */
1822 if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
1823 uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
1824 if (length < blocksize) {
1825 offset -= offset % blocksize;
1830 if (bt->bt_magic == BSWAP_64(BT_MAGIC))
1831 byteswap_uint64_array(bt, sizeof (*bt));
1833 if (bt->bt_magic != BT_MAGIC)
1836 ztest_object_lock(zd, lr->lr_foid, RL_READER);
1837 rl = ztest_range_lock(zd, lr->lr_foid, offset, length, RL_WRITER);
1839 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1841 dmu_object_info_from_db(db, &doi);
1843 bbt = ztest_bt_bonus(db);
1844 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1846 crtxg = bbt->bt_crtxg;
1847 lrtxg = lr->lr_common.lrc_txg;
1849 tx = dmu_tx_create(os);
1851 dmu_tx_hold_write(tx, lr->lr_foid, offset, length);
1853 if (ztest_random(8) == 0 && length == doi.doi_data_block_size &&
1854 P2PHASE(offset, length) == 0)
1855 abuf = dmu_request_arcbuf(db, length);
1857 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1860 dmu_return_arcbuf(abuf);
1861 dmu_buf_rele(db, FTAG);
1862 ztest_range_unlock(zd, rl);
1863 ztest_object_unlock(zd, lr->lr_foid);
1869 * Usually, verify the old data before writing new data --
1870 * but not always, because we also want to verify correct
1871 * behavior when the data was not recently read into cache.
1873 ASSERT(offset % doi.doi_data_block_size == 0);
1874 if (ztest_random(4) != 0) {
1875 int prefetch = ztest_random(2) ?
1876 DMU_READ_PREFETCH : DMU_READ_NO_PREFETCH;
1877 ztest_block_tag_t rbt;
1879 VERIFY(dmu_read(os, lr->lr_foid, offset,
1880 sizeof (rbt), &rbt, prefetch) == 0);
1881 if (rbt.bt_magic == BT_MAGIC) {
1882 ztest_bt_verify(&rbt, os, lr->lr_foid, 0,
1883 offset, gen, txg, crtxg);
1888 * Writes can appear to be newer than the bonus buffer because
1889 * the ztest_get_data() callback does a dmu_read() of the
1890 * open-context data, which may be different than the data
1891 * as it was when the write was generated.
1893 if (zd->zd_zilog->zl_replay) {
1894 ztest_bt_verify(bt, os, lr->lr_foid, 0, offset,
1895 MAX(gen, bt->bt_gen), MAX(txg, lrtxg),
1900 * Set the bt's gen/txg to the bonus buffer's gen/txg
1901 * so that all of the usual ASSERTs will work.
1903 ztest_bt_generate(bt, os, lr->lr_foid, 0, offset, gen, txg,
1908 dmu_write(os, lr->lr_foid, offset, length, data, tx);
1910 bcopy(data, abuf->b_data, length);
1911 dmu_assign_arcbuf(db, offset, abuf, tx);
1914 (void) ztest_log_write(zd, tx, lr);
1916 dmu_buf_rele(db, FTAG);
1920 ztest_range_unlock(zd, rl);
1921 ztest_object_unlock(zd, lr->lr_foid);
1927 ztest_replay_truncate(ztest_ds_t *zd, lr_truncate_t *lr, boolean_t byteswap)
1929 objset_t *os = zd->zd_os;
1935 byteswap_uint64_array(lr, sizeof (*lr));
1937 ztest_object_lock(zd, lr->lr_foid, RL_READER);
1938 rl = ztest_range_lock(zd, lr->lr_foid, lr->lr_offset, lr->lr_length,
1941 tx = dmu_tx_create(os);
1943 dmu_tx_hold_free(tx, lr->lr_foid, lr->lr_offset, lr->lr_length);
1945 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1947 ztest_range_unlock(zd, rl);
1948 ztest_object_unlock(zd, lr->lr_foid);
1952 VERIFY(dmu_free_range(os, lr->lr_foid, lr->lr_offset,
1953 lr->lr_length, tx) == 0);
1955 (void) ztest_log_truncate(zd, tx, lr);
1959 ztest_range_unlock(zd, rl);
1960 ztest_object_unlock(zd, lr->lr_foid);
1966 ztest_replay_setattr(ztest_ds_t *zd, lr_setattr_t *lr, boolean_t byteswap)
1968 objset_t *os = zd->zd_os;
1971 ztest_block_tag_t *bbt;
1972 uint64_t txg, lrtxg, crtxg, dnodesize;
1975 byteswap_uint64_array(lr, sizeof (*lr));
1977 ztest_object_lock(zd, lr->lr_foid, RL_WRITER);
1979 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1981 tx = dmu_tx_create(os);
1982 dmu_tx_hold_bonus(tx, lr->lr_foid);
1984 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1986 dmu_buf_rele(db, FTAG);
1987 ztest_object_unlock(zd, lr->lr_foid);
1991 bbt = ztest_bt_bonus(db);
1992 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1993 crtxg = bbt->bt_crtxg;
1994 lrtxg = lr->lr_common.lrc_txg;
1995 dnodesize = bbt->bt_dnodesize;
1997 if (zd->zd_zilog->zl_replay) {
1998 ASSERT(lr->lr_size != 0);
1999 ASSERT(lr->lr_mode != 0);
2003 * Randomly change the size and increment the generation.
2005 lr->lr_size = (ztest_random(db->db_size / sizeof (*bbt)) + 1) *
2007 lr->lr_mode = bbt->bt_gen + 1;
2012 * Verify that the current bonus buffer is not newer than our txg.
2014 ztest_bt_verify(bbt, os, lr->lr_foid, dnodesize, -1ULL, lr->lr_mode,
2015 MAX(txg, lrtxg), crtxg);
2017 dmu_buf_will_dirty(db, tx);
2019 ASSERT3U(lr->lr_size, >=, sizeof (*bbt));
2020 ASSERT3U(lr->lr_size, <=, db->db_size);
2021 VERIFY0(dmu_set_bonus(db, lr->lr_size, tx));
2022 bbt = ztest_bt_bonus(db);
2024 ztest_bt_generate(bbt, os, lr->lr_foid, dnodesize, -1ULL, lr->lr_mode,
2026 ztest_fill_unused_bonus(db, bbt, lr->lr_foid, os, bbt->bt_gen);
2027 dmu_buf_rele(db, FTAG);
2029 (void) ztest_log_setattr(zd, tx, lr);
2033 ztest_object_unlock(zd, lr->lr_foid);
2038 zil_replay_func_t ztest_replay_vector[TX_MAX_TYPE] = {
2039 NULL, /* 0 no such transaction type */
2040 (zil_replay_func_t)ztest_replay_create, /* TX_CREATE */
2041 NULL, /* TX_MKDIR */
2042 NULL, /* TX_MKXATTR */
2043 NULL, /* TX_SYMLINK */
2044 (zil_replay_func_t)ztest_replay_remove, /* TX_REMOVE */
2045 NULL, /* TX_RMDIR */
2047 NULL, /* TX_RENAME */
2048 (zil_replay_func_t)ztest_replay_write, /* TX_WRITE */
2049 (zil_replay_func_t)ztest_replay_truncate, /* TX_TRUNCATE */
2050 (zil_replay_func_t)ztest_replay_setattr, /* TX_SETATTR */
2052 NULL, /* TX_CREATE_ACL */
2053 NULL, /* TX_CREATE_ATTR */
2054 NULL, /* TX_CREATE_ACL_ATTR */
2055 NULL, /* TX_MKDIR_ACL */
2056 NULL, /* TX_MKDIR_ATTR */
2057 NULL, /* TX_MKDIR_ACL_ATTR */
2058 NULL, /* TX_WRITE2 */
2062 * ZIL get_data callbacks
2064 typedef struct ztest_zgd_private {
2068 } ztest_zgd_private_t;
2071 ztest_get_done(zgd_t *zgd, int error)
2073 ztest_zgd_private_t *zzp = zgd->zgd_private;
2074 ztest_ds_t *zd = zzp->z_zd;
2075 uint64_t object = zzp->z_object;
2078 dmu_buf_rele(zgd->zgd_db, zgd);
2080 ztest_range_unlock(zd, zzp->z_rl);
2081 ztest_object_unlock(zd, object);
2083 if (error == 0 && zgd->zgd_bp)
2084 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
2086 umem_free(zgd, sizeof (*zgd));
2087 umem_free(zzp, sizeof (*zzp));
2091 ztest_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
2093 ztest_ds_t *zd = arg;
2094 objset_t *os = zd->zd_os;
2095 uint64_t object = lr->lr_foid;
2096 uint64_t offset = lr->lr_offset;
2097 uint64_t size = lr->lr_length;
2098 blkptr_t *bp = &lr->lr_blkptr;
2099 uint64_t txg = lr->lr_common.lrc_txg;
2101 dmu_object_info_t doi;
2105 ztest_zgd_private_t *zgd_private;
2107 ztest_object_lock(zd, object, RL_READER);
2108 error = dmu_bonus_hold(os, object, FTAG, &db);
2110 ztest_object_unlock(zd, object);
2114 crtxg = ztest_bt_bonus(db)->bt_crtxg;
2116 if (crtxg == 0 || crtxg > txg) {
2117 dmu_buf_rele(db, FTAG);
2118 ztest_object_unlock(zd, object);
2122 dmu_object_info_from_db(db, &doi);
2123 dmu_buf_rele(db, FTAG);
2126 zgd = umem_zalloc(sizeof (*zgd), UMEM_NOFAIL);
2127 zgd->zgd_zilog = zd->zd_zilog;
2128 zgd_private = umem_zalloc(sizeof (ztest_zgd_private_t), UMEM_NOFAIL);
2129 zgd_private->z_zd = zd;
2130 zgd_private->z_object = object;
2131 zgd->zgd_private = zgd_private;
2133 if (buf != NULL) { /* immediate write */
2134 zgd_private->z_rl = ztest_range_lock(zd, object, offset, size,
2137 error = dmu_read(os, object, offset, size, buf,
2138 DMU_READ_NO_PREFETCH);
2141 size = doi.doi_data_block_size;
2143 offset = P2ALIGN(offset, size);
2145 ASSERT(offset < size);
2149 zgd_private->z_rl = ztest_range_lock(zd, object, offset, size,
2152 error = dmu_buf_hold(os, object, offset, zgd, &db,
2153 DMU_READ_NO_PREFETCH);
2156 blkptr_t *obp = dmu_buf_get_blkptr(db);
2158 ASSERT(BP_IS_HOLE(bp));
2165 ASSERT(db->db_offset == offset);
2166 ASSERT(db->db_size == size);
2168 error = dmu_sync(zio, lr->lr_common.lrc_txg,
2169 ztest_get_done, zgd);
2176 ztest_get_done(zgd, error);
2182 ztest_lr_alloc(size_t lrsize, char *name)
2185 size_t namesize = name ? strlen(name) + 1 : 0;
2187 lr = umem_zalloc(lrsize + namesize, UMEM_NOFAIL);
2190 bcopy(name, lr + lrsize, namesize);
2196 ztest_lr_free(void *lr, size_t lrsize, char *name)
2198 size_t namesize = name ? strlen(name) + 1 : 0;
2200 umem_free(lr, lrsize + namesize);
2204 * Lookup a bunch of objects. Returns the number of objects not found.
2207 ztest_lookup(ztest_ds_t *zd, ztest_od_t *od, int count)
2213 ASSERT(mutex_held(&zd->zd_dirobj_lock));
2215 for (i = 0; i < count; i++, od++) {
2217 error = zap_lookup(zd->zd_os, od->od_dir, od->od_name,
2218 sizeof (uint64_t), 1, &od->od_object);
2220 ASSERT(error == ENOENT);
2221 ASSERT(od->od_object == 0);
2225 ztest_block_tag_t *bbt;
2226 dmu_object_info_t doi;
2228 ASSERT(od->od_object != 0);
2229 ASSERT(missing == 0); /* there should be no gaps */
2231 ztest_object_lock(zd, od->od_object, RL_READER);
2232 VERIFY3U(0, ==, dmu_bonus_hold(zd->zd_os,
2233 od->od_object, FTAG, &db));
2234 dmu_object_info_from_db(db, &doi);
2235 bbt = ztest_bt_bonus(db);
2236 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
2237 od->od_type = doi.doi_type;
2238 od->od_blocksize = doi.doi_data_block_size;
2239 od->od_gen = bbt->bt_gen;
2240 dmu_buf_rele(db, FTAG);
2241 ztest_object_unlock(zd, od->od_object);
2249 ztest_create(ztest_ds_t *zd, ztest_od_t *od, int count)
2254 ASSERT(mutex_held(&zd->zd_dirobj_lock));
2256 for (i = 0; i < count; i++, od++) {
2263 lr_create_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
2265 lr->lr_doid = od->od_dir;
2266 lr->lr_foid = 0; /* 0 to allocate, > 0 to claim */
2267 lr->lrz_type = od->od_crtype;
2268 lr->lrz_blocksize = od->od_crblocksize;
2269 lr->lrz_ibshift = ztest_random_ibshift();
2270 lr->lrz_bonustype = DMU_OT_UINT64_OTHER;
2271 lr->lrz_dnodesize = od->od_crdnodesize;
2272 lr->lr_gen = od->od_crgen;
2273 lr->lr_crtime[0] = time(NULL);
2275 if (ztest_replay_create(zd, lr, B_FALSE) != 0) {
2276 ASSERT(missing == 0);
2280 od->od_object = lr->lr_foid;
2281 od->od_type = od->od_crtype;
2282 od->od_blocksize = od->od_crblocksize;
2283 od->od_gen = od->od_crgen;
2284 ASSERT(od->od_object != 0);
2287 ztest_lr_free(lr, sizeof (*lr), od->od_name);
2294 ztest_remove(ztest_ds_t *zd, ztest_od_t *od, int count)
2300 ASSERT(mutex_held(&zd->zd_dirobj_lock));
2304 for (i = count - 1; i >= 0; i--, od--) {
2311 * No object was found.
2313 if (od->od_object == 0)
2316 lr_remove_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
2318 lr->lr_doid = od->od_dir;
2320 if ((error = ztest_replay_remove(zd, lr, B_FALSE)) != 0) {
2321 ASSERT3U(error, ==, ENOSPC);
2326 ztest_lr_free(lr, sizeof (*lr), od->od_name);
2333 ztest_write(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size,
2339 lr = ztest_lr_alloc(sizeof (*lr) + size, NULL);
2341 lr->lr_foid = object;
2342 lr->lr_offset = offset;
2343 lr->lr_length = size;
2345 BP_ZERO(&lr->lr_blkptr);
2347 bcopy(data, lr + 1, size);
2349 error = ztest_replay_write(zd, lr, B_FALSE);
2351 ztest_lr_free(lr, sizeof (*lr) + size, NULL);
2357 ztest_truncate(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2362 lr = ztest_lr_alloc(sizeof (*lr), NULL);
2364 lr->lr_foid = object;
2365 lr->lr_offset = offset;
2366 lr->lr_length = size;
2368 error = ztest_replay_truncate(zd, lr, B_FALSE);
2370 ztest_lr_free(lr, sizeof (*lr), NULL);
2376 ztest_setattr(ztest_ds_t *zd, uint64_t object)
2381 lr = ztest_lr_alloc(sizeof (*lr), NULL);
2383 lr->lr_foid = object;
2387 error = ztest_replay_setattr(zd, lr, B_FALSE);
2389 ztest_lr_free(lr, sizeof (*lr), NULL);
2395 ztest_prealloc(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2397 objset_t *os = zd->zd_os;
2402 txg_wait_synced(dmu_objset_pool(os), 0);
2404 ztest_object_lock(zd, object, RL_READER);
2405 rl = ztest_range_lock(zd, object, offset, size, RL_WRITER);
2407 tx = dmu_tx_create(os);
2409 dmu_tx_hold_write(tx, object, offset, size);
2411 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
2414 dmu_prealloc(os, object, offset, size, tx);
2416 txg_wait_synced(dmu_objset_pool(os), txg);
2418 (void) dmu_free_long_range(os, object, offset, size);
2421 ztest_range_unlock(zd, rl);
2422 ztest_object_unlock(zd, object);
2426 ztest_io(ztest_ds_t *zd, uint64_t object, uint64_t offset)
2429 ztest_block_tag_t wbt;
2430 dmu_object_info_t doi;
2431 enum ztest_io_type io_type;
2435 VERIFY(dmu_object_info(zd->zd_os, object, &doi) == 0);
2436 blocksize = doi.doi_data_block_size;
2437 data = umem_alloc(blocksize, UMEM_NOFAIL);
2440 * Pick an i/o type at random, biased toward writing block tags.
2442 io_type = ztest_random(ZTEST_IO_TYPES);
2443 if (ztest_random(2) == 0)
2444 io_type = ZTEST_IO_WRITE_TAG;
2446 (void) rw_rdlock(&zd->zd_zilog_lock);
2450 case ZTEST_IO_WRITE_TAG:
2451 ztest_bt_generate(&wbt, zd->zd_os, object, doi.doi_dnodesize,
2453 (void) ztest_write(zd, object, offset, sizeof (wbt), &wbt);
2456 case ZTEST_IO_WRITE_PATTERN:
2457 (void) memset(data, 'a' + (object + offset) % 5, blocksize);
2458 if (ztest_random(2) == 0) {
2460 * Induce fletcher2 collisions to ensure that
2461 * zio_ddt_collision() detects and resolves them
2462 * when using fletcher2-verify for deduplication.
2464 ((uint64_t *)data)[0] ^= 1ULL << 63;
2465 ((uint64_t *)data)[4] ^= 1ULL << 63;
2467 (void) ztest_write(zd, object, offset, blocksize, data);
2470 case ZTEST_IO_WRITE_ZEROES:
2471 bzero(data, blocksize);
2472 (void) ztest_write(zd, object, offset, blocksize, data);
2475 case ZTEST_IO_TRUNCATE:
2476 (void) ztest_truncate(zd, object, offset, blocksize);
2479 case ZTEST_IO_SETATTR:
2480 (void) ztest_setattr(zd, object);
2485 case ZTEST_IO_REWRITE:
2486 (void) rw_rdlock(&ztest_name_lock);
2487 err = ztest_dsl_prop_set_uint64(zd->zd_name,
2488 ZFS_PROP_CHECKSUM, spa_dedup_checksum(ztest_spa),
2490 VERIFY(err == 0 || err == ENOSPC);
2491 err = ztest_dsl_prop_set_uint64(zd->zd_name,
2492 ZFS_PROP_COMPRESSION,
2493 ztest_random_dsl_prop(ZFS_PROP_COMPRESSION),
2495 VERIFY(err == 0 || err == ENOSPC);
2496 (void) rw_unlock(&ztest_name_lock);
2498 VERIFY0(dmu_read(zd->zd_os, object, offset, blocksize, data,
2499 DMU_READ_NO_PREFETCH));
2501 (void) ztest_write(zd, object, offset, blocksize, data);
2505 (void) rw_unlock(&zd->zd_zilog_lock);
2507 umem_free(data, blocksize);
2511 * Initialize an object description template.
2514 ztest_od_init(ztest_od_t *od, uint64_t id, char *tag, uint64_t index,
2515 dmu_object_type_t type, uint64_t blocksize, uint64_t dnodesize,
2518 od->od_dir = ZTEST_DIROBJ;
2521 od->od_crtype = type;
2522 od->od_crblocksize = blocksize ? blocksize : ztest_random_blocksize();
2523 od->od_crdnodesize = dnodesize ? dnodesize : ztest_random_dnodesize();
2526 od->od_type = DMU_OT_NONE;
2527 od->od_blocksize = 0;
2530 (void) snprintf(od->od_name, sizeof (od->od_name), "%s(%lld)[%llu]",
2531 tag, (longlong_t)id, (u_longlong_t)index);
2535 * Lookup or create the objects for a test using the od template.
2536 * If the objects do not all exist, or if 'remove' is specified,
2537 * remove any existing objects and create new ones. Otherwise,
2538 * use the existing objects.
2541 ztest_object_init(ztest_ds_t *zd, ztest_od_t *od, size_t size, boolean_t remove)
2543 int count = size / sizeof (*od);
2546 mutex_enter(&zd->zd_dirobj_lock);
2547 if ((ztest_lookup(zd, od, count) != 0 || remove) &&
2548 (ztest_remove(zd, od, count) != 0 ||
2549 ztest_create(zd, od, count) != 0))
2552 mutex_exit(&zd->zd_dirobj_lock);
2559 ztest_zil_commit(ztest_ds_t *zd, uint64_t id)
2561 zilog_t *zilog = zd->zd_zilog;
2563 (void) rw_rdlock(&zd->zd_zilog_lock);
2565 zil_commit(zilog, ztest_random(ZTEST_OBJECTS));
2568 * Remember the committed values in zd, which is in parent/child
2569 * shared memory. If we die, the next iteration of ztest_run()
2570 * will verify that the log really does contain this record.
2572 mutex_enter(&zilog->zl_lock);
2573 ASSERT(zd->zd_shared != NULL);
2574 ASSERT3U(zd->zd_shared->zd_seq, <=, zilog->zl_commit_lr_seq);
2575 zd->zd_shared->zd_seq = zilog->zl_commit_lr_seq;
2576 mutex_exit(&zilog->zl_lock);
2578 (void) rw_unlock(&zd->zd_zilog_lock);
2582 * This function is designed to simulate the operations that occur during a
2583 * mount/unmount operation. We hold the dataset across these operations in an
2584 * attempt to expose any implicit assumptions about ZIL management.
2588 ztest_zil_remount(ztest_ds_t *zd, uint64_t id)
2590 objset_t *os = zd->zd_os;
2593 * We grab the zd_dirobj_lock to ensure that no other thread is
2594 * updating the zil (i.e. adding in-memory log records) and the
2595 * zd_zilog_lock to block any I/O.
2597 mutex_enter(&zd->zd_dirobj_lock);
2598 (void) rw_wrlock(&zd->zd_zilog_lock);
2600 /* zfs_sb_teardown() */
2601 zil_close(zd->zd_zilog);
2603 /* zfsvfs_setup() */
2604 VERIFY(zil_open(os, ztest_get_data) == zd->zd_zilog);
2605 zil_replay(os, zd, ztest_replay_vector);
2607 (void) rw_unlock(&zd->zd_zilog_lock);
2608 mutex_exit(&zd->zd_dirobj_lock);
2612 * Verify that we can't destroy an active pool, create an existing pool,
2613 * or create a pool with a bad vdev spec.
2617 ztest_spa_create_destroy(ztest_ds_t *zd, uint64_t id)
2619 ztest_shared_opts_t *zo = &ztest_opts;
2624 * Attempt to create using a bad file.
2626 nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1);
2627 VERIFY3U(ENOENT, ==,
2628 spa_create("ztest_bad_file", nvroot, NULL, NULL));
2629 nvlist_free(nvroot);
2632 * Attempt to create using a bad mirror.
2634 nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 2, 1);
2635 VERIFY3U(ENOENT, ==,
2636 spa_create("ztest_bad_mirror", nvroot, NULL, NULL));
2637 nvlist_free(nvroot);
2640 * Attempt to create an existing pool. It shouldn't matter
2641 * what's in the nvroot; we should fail with EEXIST.
2643 (void) rw_rdlock(&ztest_name_lock);
2644 nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1);
2645 VERIFY3U(EEXIST, ==, spa_create(zo->zo_pool, nvroot, NULL, NULL));
2646 nvlist_free(nvroot);
2647 VERIFY3U(0, ==, spa_open(zo->zo_pool, &spa, FTAG));
2648 VERIFY3U(EBUSY, ==, spa_destroy(zo->zo_pool));
2649 spa_close(spa, FTAG);
2651 (void) rw_unlock(&ztest_name_lock);
2656 ztest_spa_upgrade(ztest_ds_t *zd, uint64_t id)
2659 uint64_t initial_version = SPA_VERSION_INITIAL;
2660 uint64_t version, newversion;
2661 nvlist_t *nvroot, *props;
2664 mutex_enter(&ztest_vdev_lock);
2665 name = kmem_asprintf("%s_upgrade", ztest_opts.zo_pool);
2668 * Clean up from previous runs.
2670 (void) spa_destroy(name);
2672 nvroot = make_vdev_root(NULL, NULL, name, ztest_opts.zo_vdev_size, 0,
2673 0, ztest_opts.zo_raidz, ztest_opts.zo_mirrors, 1);
2676 * If we're configuring a RAIDZ device then make sure that the
2677 * the initial version is capable of supporting that feature.
2679 switch (ztest_opts.zo_raidz_parity) {
2682 initial_version = SPA_VERSION_INITIAL;
2685 initial_version = SPA_VERSION_RAIDZ2;
2688 initial_version = SPA_VERSION_RAIDZ3;
2693 * Create a pool with a spa version that can be upgraded. Pick
2694 * a value between initial_version and SPA_VERSION_BEFORE_FEATURES.
2697 version = ztest_random_spa_version(initial_version);
2698 } while (version > SPA_VERSION_BEFORE_FEATURES);
2700 props = fnvlist_alloc();
2701 fnvlist_add_uint64(props,
2702 zpool_prop_to_name(ZPOOL_PROP_VERSION), version);
2703 VERIFY3S(spa_create(name, nvroot, props, NULL), ==, 0);
2704 fnvlist_free(nvroot);
2705 fnvlist_free(props);
2707 VERIFY3S(spa_open(name, &spa, FTAG), ==, 0);
2708 VERIFY3U(spa_version(spa), ==, version);
2709 newversion = ztest_random_spa_version(version + 1);
2711 if (ztest_opts.zo_verbose >= 4) {
2712 (void) printf("upgrading spa version from %llu to %llu\n",
2713 (u_longlong_t)version, (u_longlong_t)newversion);
2716 spa_upgrade(spa, newversion);
2717 VERIFY3U(spa_version(spa), >, version);
2718 VERIFY3U(spa_version(spa), ==, fnvlist_lookup_uint64(spa->spa_config,
2719 zpool_prop_to_name(ZPOOL_PROP_VERSION)));
2720 spa_close(spa, FTAG);
2723 mutex_exit(&ztest_vdev_lock);
2727 vdev_lookup_by_path(vdev_t *vd, const char *path)
2732 if (vd->vdev_path != NULL && strcmp(path, vd->vdev_path) == 0)
2735 for (c = 0; c < vd->vdev_children; c++)
2736 if ((mvd = vdev_lookup_by_path(vd->vdev_child[c], path)) !=
2744 * Find the first available hole which can be used as a top-level.
2747 find_vdev_hole(spa_t *spa)
2749 vdev_t *rvd = spa->spa_root_vdev;
2752 ASSERT(spa_config_held(spa, SCL_VDEV, RW_READER) == SCL_VDEV);
2754 for (c = 0; c < rvd->vdev_children; c++) {
2755 vdev_t *cvd = rvd->vdev_child[c];
2757 if (cvd->vdev_ishole)
2764 * Verify that vdev_add() works as expected.
2768 ztest_vdev_add_remove(ztest_ds_t *zd, uint64_t id)
2770 ztest_shared_t *zs = ztest_shared;
2771 spa_t *spa = ztest_spa;
2777 mutex_enter(&ztest_vdev_lock);
2778 leaves = MAX(zs->zs_mirrors + zs->zs_splits, 1) * ztest_opts.zo_raidz;
2780 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2782 ztest_shared->zs_vdev_next_leaf = find_vdev_hole(spa) * leaves;
2785 * If we have slogs then remove them 1/4 of the time.
2787 if (spa_has_slogs(spa) && ztest_random(4) == 0) {
2789 * Grab the guid from the head of the log class rotor.
2791 guid = spa_log_class(spa)->mc_rotor->mg_vd->vdev_guid;
2793 spa_config_exit(spa, SCL_VDEV, FTAG);
2796 * We have to grab the zs_name_lock as writer to
2797 * prevent a race between removing a slog (dmu_objset_find)
2798 * and destroying a dataset. Removing the slog will
2799 * grab a reference on the dataset which may cause
2800 * dsl_destroy_head() to fail with EBUSY thus
2801 * leaving the dataset in an inconsistent state.
2803 rw_wrlock(&ztest_name_lock);
2804 error = spa_vdev_remove(spa, guid, B_FALSE);
2805 rw_unlock(&ztest_name_lock);
2807 if (error && error != EEXIST)
2808 fatal(0, "spa_vdev_remove() = %d", error);
2810 spa_config_exit(spa, SCL_VDEV, FTAG);
2813 * Make 1/4 of the devices be log devices.
2815 nvroot = make_vdev_root(NULL, NULL, NULL,
2816 ztest_opts.zo_vdev_size, 0,
2817 ztest_random(4) == 0, ztest_opts.zo_raidz,
2820 error = spa_vdev_add(spa, nvroot);
2821 nvlist_free(nvroot);
2823 if (error == ENOSPC)
2824 ztest_record_enospc("spa_vdev_add");
2825 else if (error != 0)
2826 fatal(0, "spa_vdev_add() = %d", error);
2829 mutex_exit(&ztest_vdev_lock);
2833 * Verify that adding/removing aux devices (l2arc, hot spare) works as expected.
2837 ztest_vdev_aux_add_remove(ztest_ds_t *zd, uint64_t id)
2839 ztest_shared_t *zs = ztest_shared;
2840 spa_t *spa = ztest_spa;
2841 vdev_t *rvd = spa->spa_root_vdev;
2842 spa_aux_vdev_t *sav;
2848 path = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
2850 if (ztest_random(2) == 0) {
2851 sav = &spa->spa_spares;
2852 aux = ZPOOL_CONFIG_SPARES;
2854 sav = &spa->spa_l2cache;
2855 aux = ZPOOL_CONFIG_L2CACHE;
2858 mutex_enter(&ztest_vdev_lock);
2860 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2862 if (sav->sav_count != 0 && ztest_random(4) == 0) {
2864 * Pick a random device to remove.
2866 guid = sav->sav_vdevs[ztest_random(sav->sav_count)]->vdev_guid;
2869 * Find an unused device we can add.
2871 zs->zs_vdev_aux = 0;
2874 (void) snprintf(path, MAXPATHLEN, ztest_aux_template,
2875 ztest_opts.zo_dir, ztest_opts.zo_pool, aux,
2877 for (c = 0; c < sav->sav_count; c++)
2878 if (strcmp(sav->sav_vdevs[c]->vdev_path,
2881 if (c == sav->sav_count &&
2882 vdev_lookup_by_path(rvd, path) == NULL)
2888 spa_config_exit(spa, SCL_VDEV, FTAG);
2894 nvlist_t *nvroot = make_vdev_root(NULL, aux, NULL,
2895 (ztest_opts.zo_vdev_size * 5) / 4, 0, 0, 0, 0, 1);
2896 error = spa_vdev_add(spa, nvroot);
2898 fatal(0, "spa_vdev_add(%p) = %d", nvroot, error);
2899 nvlist_free(nvroot);
2902 * Remove an existing device. Sometimes, dirty its
2903 * vdev state first to make sure we handle removal
2904 * of devices that have pending state changes.
2906 if (ztest_random(2) == 0)
2907 (void) vdev_online(spa, guid, 0, NULL);
2909 error = spa_vdev_remove(spa, guid, B_FALSE);
2910 if (error != 0 && error != EBUSY)
2911 fatal(0, "spa_vdev_remove(%llu) = %d", guid, error);
2914 mutex_exit(&ztest_vdev_lock);
2916 umem_free(path, MAXPATHLEN);
2920 * split a pool if it has mirror tlvdevs
2924 ztest_split_pool(ztest_ds_t *zd, uint64_t id)
2926 ztest_shared_t *zs = ztest_shared;
2927 spa_t *spa = ztest_spa;
2928 vdev_t *rvd = spa->spa_root_vdev;
2929 nvlist_t *tree, **child, *config, *split, **schild;
2930 uint_t c, children, schildren = 0, lastlogid = 0;
2933 mutex_enter(&ztest_vdev_lock);
2935 /* ensure we have a useable config; mirrors of raidz aren't supported */
2936 if (zs->zs_mirrors < 3 || ztest_opts.zo_raidz > 1) {
2937 mutex_exit(&ztest_vdev_lock);
2941 /* clean up the old pool, if any */
2942 (void) spa_destroy("splitp");
2944 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2946 /* generate a config from the existing config */
2947 mutex_enter(&spa->spa_props_lock);
2948 VERIFY(nvlist_lookup_nvlist(spa->spa_config, ZPOOL_CONFIG_VDEV_TREE,
2950 mutex_exit(&spa->spa_props_lock);
2952 VERIFY(nvlist_lookup_nvlist_array(tree, ZPOOL_CONFIG_CHILDREN, &child,
2955 schild = malloc(rvd->vdev_children * sizeof (nvlist_t *));
2956 for (c = 0; c < children; c++) {
2957 vdev_t *tvd = rvd->vdev_child[c];
2961 if (tvd->vdev_islog || tvd->vdev_ops == &vdev_hole_ops) {
2962 VERIFY(nvlist_alloc(&schild[schildren], NV_UNIQUE_NAME,
2964 VERIFY(nvlist_add_string(schild[schildren],
2965 ZPOOL_CONFIG_TYPE, VDEV_TYPE_HOLE) == 0);
2966 VERIFY(nvlist_add_uint64(schild[schildren],
2967 ZPOOL_CONFIG_IS_HOLE, 1) == 0);
2969 lastlogid = schildren;
2974 VERIFY(nvlist_lookup_nvlist_array(child[c],
2975 ZPOOL_CONFIG_CHILDREN, &mchild, &mchildren) == 0);
2976 VERIFY(nvlist_dup(mchild[0], &schild[schildren++], 0) == 0);
2979 /* OK, create a config that can be used to split */
2980 VERIFY(nvlist_alloc(&split, NV_UNIQUE_NAME, 0) == 0);
2981 VERIFY(nvlist_add_string(split, ZPOOL_CONFIG_TYPE,
2982 VDEV_TYPE_ROOT) == 0);
2983 VERIFY(nvlist_add_nvlist_array(split, ZPOOL_CONFIG_CHILDREN, schild,
2984 lastlogid != 0 ? lastlogid : schildren) == 0);
2986 VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, 0) == 0);
2987 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, split) == 0);
2989 for (c = 0; c < schildren; c++)
2990 nvlist_free(schild[c]);
2994 spa_config_exit(spa, SCL_VDEV, FTAG);
2996 (void) rw_wrlock(&ztest_name_lock);
2997 error = spa_vdev_split_mirror(spa, "splitp", config, NULL, B_FALSE);
2998 (void) rw_unlock(&ztest_name_lock);
3000 nvlist_free(config);
3003 (void) printf("successful split - results:\n");
3004 mutex_enter(&spa_namespace_lock);
3005 show_pool_stats(spa);
3006 show_pool_stats(spa_lookup("splitp"));
3007 mutex_exit(&spa_namespace_lock);
3011 mutex_exit(&ztest_vdev_lock);
3016 * Verify that we can attach and detach devices.
3020 ztest_vdev_attach_detach(ztest_ds_t *zd, uint64_t id)
3022 ztest_shared_t *zs = ztest_shared;
3023 spa_t *spa = ztest_spa;
3024 spa_aux_vdev_t *sav = &spa->spa_spares;
3025 vdev_t *rvd = spa->spa_root_vdev;
3026 vdev_t *oldvd, *newvd, *pvd;
3030 uint64_t ashift = ztest_get_ashift();
3031 uint64_t oldguid, pguid;
3032 uint64_t oldsize, newsize;
3033 char *oldpath, *newpath;
3035 int oldvd_has_siblings = B_FALSE;
3036 int newvd_is_spare = B_FALSE;
3038 int error, expected_error;
3040 oldpath = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
3041 newpath = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
3043 mutex_enter(&ztest_vdev_lock);
3044 leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz;
3046 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
3049 * Decide whether to do an attach or a replace.
3051 replacing = ztest_random(2);
3054 * Pick a random top-level vdev.
3056 top = ztest_random_vdev_top(spa, B_TRUE);
3059 * Pick a random leaf within it.
3061 leaf = ztest_random(leaves);
3066 oldvd = rvd->vdev_child[top];
3067 if (zs->zs_mirrors >= 1) {
3068 ASSERT(oldvd->vdev_ops == &vdev_mirror_ops);
3069 ASSERT(oldvd->vdev_children >= zs->zs_mirrors);
3070 oldvd = oldvd->vdev_child[leaf / ztest_opts.zo_raidz];
3072 if (ztest_opts.zo_raidz > 1) {
3073 ASSERT(oldvd->vdev_ops == &vdev_raidz_ops);
3074 ASSERT(oldvd->vdev_children == ztest_opts.zo_raidz);
3075 oldvd = oldvd->vdev_child[leaf % ztest_opts.zo_raidz];
3079 * If we're already doing an attach or replace, oldvd may be a
3080 * mirror vdev -- in which case, pick a random child.
3082 while (oldvd->vdev_children != 0) {
3083 oldvd_has_siblings = B_TRUE;
3084 ASSERT(oldvd->vdev_children >= 2);
3085 oldvd = oldvd->vdev_child[ztest_random(oldvd->vdev_children)];
3088 oldguid = oldvd->vdev_guid;
3089 oldsize = vdev_get_min_asize(oldvd);
3090 oldvd_is_log = oldvd->vdev_top->vdev_islog;
3091 (void) strcpy(oldpath, oldvd->vdev_path);
3092 pvd = oldvd->vdev_parent;
3093 pguid = pvd->vdev_guid;
3096 * If oldvd has siblings, then half of the time, detach it.
3098 if (oldvd_has_siblings && ztest_random(2) == 0) {
3099 spa_config_exit(spa, SCL_VDEV, FTAG);
3100 error = spa_vdev_detach(spa, oldguid, pguid, B_FALSE);
3101 if (error != 0 && error != ENODEV && error != EBUSY &&
3103 fatal(0, "detach (%s) returned %d", oldpath, error);
3108 * For the new vdev, choose with equal probability between the two
3109 * standard paths (ending in either 'a' or 'b') or a random hot spare.
3111 if (sav->sav_count != 0 && ztest_random(3) == 0) {
3112 newvd = sav->sav_vdevs[ztest_random(sav->sav_count)];
3113 newvd_is_spare = B_TRUE;
3114 (void) strcpy(newpath, newvd->vdev_path);
3116 (void) snprintf(newpath, MAXPATHLEN, ztest_dev_template,
3117 ztest_opts.zo_dir, ztest_opts.zo_pool,
3118 top * leaves + leaf);
3119 if (ztest_random(2) == 0)
3120 newpath[strlen(newpath) - 1] = 'b';
3121 newvd = vdev_lookup_by_path(rvd, newpath);
3125 newsize = vdev_get_min_asize(newvd);
3128 * Make newsize a little bigger or smaller than oldsize.
3129 * If it's smaller, the attach should fail.
3130 * If it's larger, and we're doing a replace,
3131 * we should get dynamic LUN growth when we're done.
3133 newsize = 10 * oldsize / (9 + ztest_random(3));
3137 * If pvd is not a mirror or root, the attach should fail with ENOTSUP,
3138 * unless it's a replace; in that case any non-replacing parent is OK.
3140 * If newvd is already part of the pool, it should fail with EBUSY.
3142 * If newvd is too small, it should fail with EOVERFLOW.
3144 if (pvd->vdev_ops != &vdev_mirror_ops &&
3145 pvd->vdev_ops != &vdev_root_ops && (!replacing ||
3146 pvd->vdev_ops == &vdev_replacing_ops ||
3147 pvd->vdev_ops == &vdev_spare_ops))
3148 expected_error = ENOTSUP;
3149 else if (newvd_is_spare && (!replacing || oldvd_is_log))
3150 expected_error = ENOTSUP;
3151 else if (newvd == oldvd)
3152 expected_error = replacing ? 0 : EBUSY;
3153 else if (vdev_lookup_by_path(rvd, newpath) != NULL)
3154 expected_error = EBUSY;
3155 else if (newsize < oldsize)
3156 expected_error = EOVERFLOW;
3157 else if (ashift > oldvd->vdev_top->vdev_ashift)
3158 expected_error = EDOM;
3162 spa_config_exit(spa, SCL_VDEV, FTAG);
3165 * Build the nvlist describing newpath.
3167 root = make_vdev_root(newpath, NULL, NULL, newvd == NULL ? newsize : 0,
3168 ashift, 0, 0, 0, 1);
3170 error = spa_vdev_attach(spa, oldguid, root, replacing);
3175 * If our parent was the replacing vdev, but the replace completed,
3176 * then instead of failing with ENOTSUP we may either succeed,
3177 * fail with ENODEV, or fail with EOVERFLOW.
3179 if (expected_error == ENOTSUP &&
3180 (error == 0 || error == ENODEV || error == EOVERFLOW))
3181 expected_error = error;
3184 * If someone grew the LUN, the replacement may be too small.
3186 if (error == EOVERFLOW || error == EBUSY)
3187 expected_error = error;
3189 /* XXX workaround 6690467 */
3190 if (error != expected_error && expected_error != EBUSY) {
3191 fatal(0, "attach (%s %llu, %s %llu, %d) "
3192 "returned %d, expected %d",
3193 oldpath, oldsize, newpath,
3194 newsize, replacing, error, expected_error);
3197 mutex_exit(&ztest_vdev_lock);
3199 umem_free(oldpath, MAXPATHLEN);
3200 umem_free(newpath, MAXPATHLEN);
3204 * Callback function which expands the physical size of the vdev.
3207 grow_vdev(vdev_t *vd, void *arg)
3209 ASSERTV(spa_t *spa = vd->vdev_spa);
3210 size_t *newsize = arg;
3214 ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
3215 ASSERT(vd->vdev_ops->vdev_op_leaf);
3217 if ((fd = open(vd->vdev_path, O_RDWR)) == -1)
3220 fsize = lseek(fd, 0, SEEK_END);
3221 VERIFY(ftruncate(fd, *newsize) == 0);
3223 if (ztest_opts.zo_verbose >= 6) {
3224 (void) printf("%s grew from %lu to %lu bytes\n",
3225 vd->vdev_path, (ulong_t)fsize, (ulong_t)*newsize);
3232 * Callback function which expands a given vdev by calling vdev_online().
3236 online_vdev(vdev_t *vd, void *arg)
3238 spa_t *spa = vd->vdev_spa;
3239 vdev_t *tvd = vd->vdev_top;
3240 uint64_t guid = vd->vdev_guid;
3241 uint64_t generation = spa->spa_config_generation + 1;
3242 vdev_state_t newstate = VDEV_STATE_UNKNOWN;
3245 ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
3246 ASSERT(vd->vdev_ops->vdev_op_leaf);
3248 /* Calling vdev_online will initialize the new metaslabs */
3249 spa_config_exit(spa, SCL_STATE, spa);
3250 error = vdev_online(spa, guid, ZFS_ONLINE_EXPAND, &newstate);
3251 spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3254 * If vdev_online returned an error or the underlying vdev_open
3255 * failed then we abort the expand. The only way to know that
3256 * vdev_open fails is by checking the returned newstate.
3258 if (error || newstate != VDEV_STATE_HEALTHY) {
3259 if (ztest_opts.zo_verbose >= 5) {
3260 (void) printf("Unable to expand vdev, state %llu, "
3261 "error %d\n", (u_longlong_t)newstate, error);
3265 ASSERT3U(newstate, ==, VDEV_STATE_HEALTHY);
3268 * Since we dropped the lock we need to ensure that we're
3269 * still talking to the original vdev. It's possible this
3270 * vdev may have been detached/replaced while we were
3271 * trying to online it.
3273 if (generation != spa->spa_config_generation) {
3274 if (ztest_opts.zo_verbose >= 5) {
3275 (void) printf("vdev configuration has changed, "
3276 "guid %llu, state %llu, expected gen %llu, "
3279 (u_longlong_t)tvd->vdev_state,
3280 (u_longlong_t)generation,
3281 (u_longlong_t)spa->spa_config_generation);
3289 * Traverse the vdev tree calling the supplied function.
3290 * We continue to walk the tree until we either have walked all
3291 * children or we receive a non-NULL return from the callback.
3292 * If a NULL callback is passed, then we just return back the first
3293 * leaf vdev we encounter.
3296 vdev_walk_tree(vdev_t *vd, vdev_t *(*func)(vdev_t *, void *), void *arg)
3300 if (vd->vdev_ops->vdev_op_leaf) {
3304 return (func(vd, arg));
3307 for (c = 0; c < vd->vdev_children; c++) {
3308 vdev_t *cvd = vd->vdev_child[c];
3309 if ((cvd = vdev_walk_tree(cvd, func, arg)) != NULL)
3316 * Verify that dynamic LUN growth works as expected.
3320 ztest_vdev_LUN_growth(ztest_ds_t *zd, uint64_t id)
3322 spa_t *spa = ztest_spa;
3324 metaslab_class_t *mc;
3325 metaslab_group_t *mg;
3326 size_t psize, newsize;
3328 uint64_t old_class_space, new_class_space, old_ms_count, new_ms_count;
3330 mutex_enter(&ztest_vdev_lock);
3331 spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3333 top = ztest_random_vdev_top(spa, B_TRUE);
3335 tvd = spa->spa_root_vdev->vdev_child[top];
3338 old_ms_count = tvd->vdev_ms_count;
3339 old_class_space = metaslab_class_get_space(mc);
3342 * Determine the size of the first leaf vdev associated with
3343 * our top-level device.
3345 vd = vdev_walk_tree(tvd, NULL, NULL);
3346 ASSERT3P(vd, !=, NULL);
3347 ASSERT(vd->vdev_ops->vdev_op_leaf);
3349 psize = vd->vdev_psize;
3352 * We only try to expand the vdev if it's healthy, less than 4x its
3353 * original size, and it has a valid psize.
3355 if (tvd->vdev_state != VDEV_STATE_HEALTHY ||
3356 psize == 0 || psize >= 4 * ztest_opts.zo_vdev_size) {
3357 spa_config_exit(spa, SCL_STATE, spa);
3358 mutex_exit(&ztest_vdev_lock);
3362 newsize = psize + psize / 8;
3363 ASSERT3U(newsize, >, psize);
3365 if (ztest_opts.zo_verbose >= 6) {
3366 (void) printf("Expanding LUN %s from %lu to %lu\n",
3367 vd->vdev_path, (ulong_t)psize, (ulong_t)newsize);
3371 * Growing the vdev is a two step process:
3372 * 1). expand the physical size (i.e. relabel)
3373 * 2). online the vdev to create the new metaslabs
3375 if (vdev_walk_tree(tvd, grow_vdev, &newsize) != NULL ||
3376 vdev_walk_tree(tvd, online_vdev, NULL) != NULL ||
3377 tvd->vdev_state != VDEV_STATE_HEALTHY) {
3378 if (ztest_opts.zo_verbose >= 5) {
3379 (void) printf("Could not expand LUN because "
3380 "the vdev configuration changed.\n");
3382 spa_config_exit(spa, SCL_STATE, spa);
3383 mutex_exit(&ztest_vdev_lock);
3387 spa_config_exit(spa, SCL_STATE, spa);
3390 * Expanding the LUN will update the config asynchronously,
3391 * thus we must wait for the async thread to complete any
3392 * pending tasks before proceeding.
3396 mutex_enter(&spa->spa_async_lock);
3397 done = (spa->spa_async_thread == NULL && !spa->spa_async_tasks);
3398 mutex_exit(&spa->spa_async_lock);
3401 txg_wait_synced(spa_get_dsl(spa), 0);
3402 (void) poll(NULL, 0, 100);
3405 spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3407 tvd = spa->spa_root_vdev->vdev_child[top];
3408 new_ms_count = tvd->vdev_ms_count;
3409 new_class_space = metaslab_class_get_space(mc);
3411 if (tvd->vdev_mg != mg || mg->mg_class != mc) {
3412 if (ztest_opts.zo_verbose >= 5) {
3413 (void) printf("Could not verify LUN expansion due to "
3414 "intervening vdev offline or remove.\n");
3416 spa_config_exit(spa, SCL_STATE, spa);
3417 mutex_exit(&ztest_vdev_lock);
3422 * Make sure we were able to grow the vdev.
3424 if (new_ms_count <= old_ms_count)
3425 fatal(0, "LUN expansion failed: ms_count %llu <= %llu\n",
3426 old_ms_count, new_ms_count);
3429 * Make sure we were able to grow the pool.
3431 if (new_class_space <= old_class_space)
3432 fatal(0, "LUN expansion failed: class_space %llu <= %llu\n",
3433 old_class_space, new_class_space);
3435 if (ztest_opts.zo_verbose >= 5) {
3436 char oldnumbuf[6], newnumbuf[6];
3438 nicenum(old_class_space, oldnumbuf);
3439 nicenum(new_class_space, newnumbuf);
3440 (void) printf("%s grew from %s to %s\n",
3441 spa->spa_name, oldnumbuf, newnumbuf);
3444 spa_config_exit(spa, SCL_STATE, spa);
3445 mutex_exit(&ztest_vdev_lock);
3449 * Verify that dmu_objset_{create,destroy,open,close} work as expected.
3453 ztest_objset_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
3456 * Create the objects common to all ztest datasets.
3458 VERIFY(zap_create_claim(os, ZTEST_DIROBJ,
3459 DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx) == 0);
3463 ztest_dataset_create(char *dsname)
3465 uint64_t zilset = ztest_random(100);
3466 int err = dmu_objset_create(dsname, DMU_OST_OTHER, 0,
3467 ztest_objset_create_cb, NULL);
3469 if (err || zilset < 80)
3472 if (ztest_opts.zo_verbose >= 5)
3473 (void) printf("Setting dataset %s to sync always\n", dsname);
3474 return (ztest_dsl_prop_set_uint64(dsname, ZFS_PROP_SYNC,
3475 ZFS_SYNC_ALWAYS, B_FALSE));
3480 ztest_objset_destroy_cb(const char *name, void *arg)
3483 dmu_object_info_t doi;
3487 * Verify that the dataset contains a directory object.
3489 VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_TRUE, FTAG, &os));
3490 error = dmu_object_info(os, ZTEST_DIROBJ, &doi);
3491 if (error != ENOENT) {
3492 /* We could have crashed in the middle of destroying it */
3494 ASSERT3U(doi.doi_type, ==, DMU_OT_ZAP_OTHER);
3495 ASSERT3S(doi.doi_physical_blocks_512, >=, 0);
3497 dmu_objset_disown(os, FTAG);
3500 * Destroy the dataset.
3502 if (strchr(name, '@') != NULL) {
3503 VERIFY0(dsl_destroy_snapshot(name, B_FALSE));
3505 VERIFY0(dsl_destroy_head(name));
3511 ztest_snapshot_create(char *osname, uint64_t id)
3513 char snapname[ZFS_MAX_DATASET_NAME_LEN];
3516 (void) snprintf(snapname, sizeof (snapname), "%llu", (u_longlong_t)id);
3518 error = dmu_objset_snapshot_one(osname, snapname);
3519 if (error == ENOSPC) {
3520 ztest_record_enospc(FTAG);
3523 if (error != 0 && error != EEXIST) {
3524 fatal(0, "ztest_snapshot_create(%s@%s) = %d", osname,
3531 ztest_snapshot_destroy(char *osname, uint64_t id)
3533 char snapname[ZFS_MAX_DATASET_NAME_LEN];
3536 (void) snprintf(snapname, sizeof (snapname), "%s@%llu", osname,
3539 error = dsl_destroy_snapshot(snapname, B_FALSE);
3540 if (error != 0 && error != ENOENT)
3541 fatal(0, "ztest_snapshot_destroy(%s) = %d", snapname, error);
3547 ztest_dmu_objset_create_destroy(ztest_ds_t *zd, uint64_t id)
3553 char name[ZFS_MAX_DATASET_NAME_LEN];
3557 zdtmp = umem_alloc(sizeof (ztest_ds_t), UMEM_NOFAIL);
3559 (void) rw_rdlock(&ztest_name_lock);
3561 (void) snprintf(name, sizeof (name), "%s/temp_%llu",
3562 ztest_opts.zo_pool, (u_longlong_t)id);
3565 * If this dataset exists from a previous run, process its replay log
3566 * half of the time. If we don't replay it, then dsl_destroy_head()
3567 * (invoked from ztest_objset_destroy_cb()) should just throw it away.
3569 if (ztest_random(2) == 0 &&
3570 dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os) == 0) {
3571 ztest_zd_init(zdtmp, NULL, os);
3572 zil_replay(os, zdtmp, ztest_replay_vector);
3573 ztest_zd_fini(zdtmp);
3574 dmu_objset_disown(os, FTAG);
3578 * There may be an old instance of the dataset we're about to
3579 * create lying around from a previous run. If so, destroy it
3580 * and all of its snapshots.
3582 (void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
3583 DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
3586 * Verify that the destroyed dataset is no longer in the namespace.
3588 VERIFY3U(ENOENT, ==, dmu_objset_own(name, DMU_OST_OTHER, B_TRUE,
3592 * Verify that we can create a new dataset.
3594 error = ztest_dataset_create(name);
3596 if (error == ENOSPC) {
3597 ztest_record_enospc(FTAG);
3600 fatal(0, "dmu_objset_create(%s) = %d", name, error);
3603 VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os));
3605 ztest_zd_init(zdtmp, NULL, os);
3608 * Open the intent log for it.
3610 zilog = zil_open(os, ztest_get_data);
3613 * Put some objects in there, do a little I/O to them,
3614 * and randomly take a couple of snapshots along the way.
3616 iters = ztest_random(5);
3617 for (i = 0; i < iters; i++) {
3618 ztest_dmu_object_alloc_free(zdtmp, id);
3619 if (ztest_random(iters) == 0)
3620 (void) ztest_snapshot_create(name, i);
3624 * Verify that we cannot create an existing dataset.
3626 VERIFY3U(EEXIST, ==,
3627 dmu_objset_create(name, DMU_OST_OTHER, 0, NULL, NULL));
3630 * Verify that we can hold an objset that is also owned.
3632 VERIFY3U(0, ==, dmu_objset_hold(name, FTAG, &os2));
3633 dmu_objset_rele(os2, FTAG);
3636 * Verify that we cannot own an objset that is already owned.
3639 dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os2));
3642 dmu_objset_disown(os, FTAG);
3643 ztest_zd_fini(zdtmp);
3645 (void) rw_unlock(&ztest_name_lock);
3647 umem_free(zdtmp, sizeof (ztest_ds_t));
3651 * Verify that dmu_snapshot_{create,destroy,open,close} work as expected.
3654 ztest_dmu_snapshot_create_destroy(ztest_ds_t *zd, uint64_t id)
3656 (void) rw_rdlock(&ztest_name_lock);
3657 (void) ztest_snapshot_destroy(zd->zd_name, id);
3658 (void) ztest_snapshot_create(zd->zd_name, id);
3659 (void) rw_unlock(&ztest_name_lock);
3663 * Cleanup non-standard snapshots and clones.
3666 ztest_dsl_dataset_cleanup(char *osname, uint64_t id)
3675 snap1name = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
3676 clone1name = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
3677 snap2name = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
3678 clone2name = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
3679 snap3name = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
3681 (void) snprintf(snap1name, ZFS_MAX_DATASET_NAME_LEN,
3682 "%s@s1_%llu", osname, (u_longlong_t)id);
3683 (void) snprintf(clone1name, ZFS_MAX_DATASET_NAME_LEN,
3684 "%s/c1_%llu", osname, (u_longlong_t)id);
3685 (void) snprintf(snap2name, ZFS_MAX_DATASET_NAME_LEN,
3686 "%s@s2_%llu", clone1name, (u_longlong_t)id);
3687 (void) snprintf(clone2name, ZFS_MAX_DATASET_NAME_LEN,
3688 "%s/c2_%llu", osname, (u_longlong_t)id);
3689 (void) snprintf(snap3name, ZFS_MAX_DATASET_NAME_LEN,
3690 "%s@s3_%llu", clone1name, (u_longlong_t)id);
3692 error = dsl_destroy_head(clone2name);
3693 if (error && error != ENOENT)
3694 fatal(0, "dsl_destroy_head(%s) = %d", clone2name, error);
3695 error = dsl_destroy_snapshot(snap3name, B_FALSE);
3696 if (error && error != ENOENT)
3697 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap3name, error);
3698 error = dsl_destroy_snapshot(snap2name, B_FALSE);
3699 if (error && error != ENOENT)
3700 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap2name, error);
3701 error = dsl_destroy_head(clone1name);
3702 if (error && error != ENOENT)
3703 fatal(0, "dsl_destroy_head(%s) = %d", clone1name, error);
3704 error = dsl_destroy_snapshot(snap1name, B_FALSE);
3705 if (error && error != ENOENT)
3706 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap1name, error);
3708 umem_free(snap1name, ZFS_MAX_DATASET_NAME_LEN);
3709 umem_free(clone1name, ZFS_MAX_DATASET_NAME_LEN);
3710 umem_free(snap2name, ZFS_MAX_DATASET_NAME_LEN);
3711 umem_free(clone2name, ZFS_MAX_DATASET_NAME_LEN);
3712 umem_free(snap3name, ZFS_MAX_DATASET_NAME_LEN);
3716 * Verify dsl_dataset_promote handles EBUSY
3719 ztest_dsl_dataset_promote_busy(ztest_ds_t *zd, uint64_t id)
3727 char *osname = zd->zd_name;
3730 snap1name = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
3731 clone1name = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
3732 snap2name = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
3733 clone2name = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
3734 snap3name = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
3736 (void) rw_rdlock(&ztest_name_lock);
3738 ztest_dsl_dataset_cleanup(osname, id);
3740 (void) snprintf(snap1name, ZFS_MAX_DATASET_NAME_LEN,
3741 "%s@s1_%llu", osname, (u_longlong_t)id);
3742 (void) snprintf(clone1name, ZFS_MAX_DATASET_NAME_LEN,
3743 "%s/c1_%llu", osname, (u_longlong_t)id);
3744 (void) snprintf(snap2name, ZFS_MAX_DATASET_NAME_LEN,
3745 "%s@s2_%llu", clone1name, (u_longlong_t)id);
3746 (void) snprintf(clone2name, ZFS_MAX_DATASET_NAME_LEN,
3747 "%s/c2_%llu", osname, (u_longlong_t)id);
3748 (void) snprintf(snap3name, ZFS_MAX_DATASET_NAME_LEN,
3749 "%s@s3_%llu", clone1name, (u_longlong_t)id);
3751 error = dmu_objset_snapshot_one(osname, strchr(snap1name, '@') + 1);
3752 if (error && error != EEXIST) {
3753 if (error == ENOSPC) {
3754 ztest_record_enospc(FTAG);
3757 fatal(0, "dmu_take_snapshot(%s) = %d", snap1name, error);
3760 error = dmu_objset_clone(clone1name, snap1name);
3762 if (error == ENOSPC) {
3763 ztest_record_enospc(FTAG);
3766 fatal(0, "dmu_objset_create(%s) = %d", clone1name, error);
3769 error = dmu_objset_snapshot_one(clone1name, strchr(snap2name, '@') + 1);
3770 if (error && error != EEXIST) {
3771 if (error == ENOSPC) {
3772 ztest_record_enospc(FTAG);
3775 fatal(0, "dmu_open_snapshot(%s) = %d", snap2name, error);
3778 error = dmu_objset_snapshot_one(clone1name, strchr(snap3name, '@') + 1);
3779 if (error && error != EEXIST) {
3780 if (error == ENOSPC) {
3781 ztest_record_enospc(FTAG);
3784 fatal(0, "dmu_open_snapshot(%s) = %d", snap3name, error);
3787 error = dmu_objset_clone(clone2name, snap3name);
3789 if (error == ENOSPC) {
3790 ztest_record_enospc(FTAG);
3793 fatal(0, "dmu_objset_create(%s) = %d", clone2name, error);
3796 error = dmu_objset_own(snap2name, DMU_OST_ANY, B_TRUE, FTAG, &os);
3798 fatal(0, "dmu_objset_own(%s) = %d", snap2name, error);
3799 error = dsl_dataset_promote(clone2name, NULL);
3800 if (error == ENOSPC) {
3801 dmu_objset_disown(os, FTAG);
3802 ztest_record_enospc(FTAG);
3806 fatal(0, "dsl_dataset_promote(%s), %d, not EBUSY", clone2name,
3808 dmu_objset_disown(os, FTAG);
3811 ztest_dsl_dataset_cleanup(osname, id);
3813 (void) rw_unlock(&ztest_name_lock);
3815 umem_free(snap1name, ZFS_MAX_DATASET_NAME_LEN);
3816 umem_free(clone1name, ZFS_MAX_DATASET_NAME_LEN);
3817 umem_free(snap2name, ZFS_MAX_DATASET_NAME_LEN);
3818 umem_free(clone2name, ZFS_MAX_DATASET_NAME_LEN);
3819 umem_free(snap3name, ZFS_MAX_DATASET_NAME_LEN);
3822 #undef OD_ARRAY_SIZE
3823 #define OD_ARRAY_SIZE 4
3826 * Verify that dmu_object_{alloc,free} work as expected.
3829 ztest_dmu_object_alloc_free(ztest_ds_t *zd, uint64_t id)
3836 size = sizeof (ztest_od_t) * OD_ARRAY_SIZE;
3837 od = umem_alloc(size, UMEM_NOFAIL);
3838 batchsize = OD_ARRAY_SIZE;
3840 for (b = 0; b < batchsize; b++)
3841 ztest_od_init(od + b, id, FTAG, b, DMU_OT_UINT64_OTHER,
3845 * Destroy the previous batch of objects, create a new batch,
3846 * and do some I/O on the new objects.
3848 if (ztest_object_init(zd, od, size, B_TRUE) != 0)
3851 while (ztest_random(4 * batchsize) != 0)
3852 ztest_io(zd, od[ztest_random(batchsize)].od_object,
3853 ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
3855 umem_free(od, size);
3858 #undef OD_ARRAY_SIZE
3859 #define OD_ARRAY_SIZE 2
3862 * Verify that dmu_{read,write} work as expected.
3865 ztest_dmu_read_write(ztest_ds_t *zd, uint64_t id)
3870 objset_t *os = zd->zd_os;
3871 size = sizeof (ztest_od_t) * OD_ARRAY_SIZE;
3872 od = umem_alloc(size, UMEM_NOFAIL);
3874 int i, freeit, error;
3876 bufwad_t *packbuf, *bigbuf, *pack, *bigH, *bigT;
3877 uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
3878 uint64_t chunksize = (1000 + ztest_random(1000)) * sizeof (uint64_t);
3879 uint64_t regions = 997;
3880 uint64_t stride = 123456789ULL;
3881 uint64_t width = 40;
3882 int free_percent = 5;
3885 * This test uses two objects, packobj and bigobj, that are always
3886 * updated together (i.e. in the same tx) so that their contents are
3887 * in sync and can be compared. Their contents relate to each other
3888 * in a simple way: packobj is a dense array of 'bufwad' structures,
3889 * while bigobj is a sparse array of the same bufwads. Specifically,
3890 * for any index n, there are three bufwads that should be identical:
3892 * packobj, at offset n * sizeof (bufwad_t)
3893 * bigobj, at the head of the nth chunk
3894 * bigobj, at the tail of the nth chunk
3896 * The chunk size is arbitrary. It doesn't have to be a power of two,
3897 * and it doesn't have any relation to the object blocksize.
3898 * The only requirement is that it can hold at least two bufwads.
3900 * Normally, we write the bufwad to each of these locations.
3901 * However, free_percent of the time we instead write zeroes to
3902 * packobj and perform a dmu_free_range() on bigobj. By comparing
3903 * bigobj to packobj, we can verify that the DMU is correctly
3904 * tracking which parts of an object are allocated and free,
3905 * and that the contents of the allocated blocks are correct.
3909 * Read the directory info. If it's the first time, set things up.
3911 ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0, chunksize);
3912 ztest_od_init(od + 1, id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, 0,
3915 if (ztest_object_init(zd, od, size, B_FALSE) != 0) {
3916 umem_free(od, size);
3920 bigobj = od[0].od_object;
3921 packobj = od[1].od_object;
3922 chunksize = od[0].od_gen;
3923 ASSERT(chunksize == od[1].od_gen);
3926 * Prefetch a random chunk of the big object.
3927 * Our aim here is to get some async reads in flight
3928 * for blocks that we may free below; the DMU should
3929 * handle this race correctly.
3931 n = ztest_random(regions) * stride + ztest_random(width);
3932 s = 1 + ztest_random(2 * width - 1);
3933 dmu_prefetch(os, bigobj, 0, n * chunksize, s * chunksize,
3934 ZIO_PRIORITY_SYNC_READ);
3937 * Pick a random index and compute the offsets into packobj and bigobj.
3939 n = ztest_random(regions) * stride + ztest_random(width);
3940 s = 1 + ztest_random(width - 1);
3942 packoff = n * sizeof (bufwad_t);
3943 packsize = s * sizeof (bufwad_t);
3945 bigoff = n * chunksize;
3946 bigsize = s * chunksize;
3948 packbuf = umem_alloc(packsize, UMEM_NOFAIL);
3949 bigbuf = umem_alloc(bigsize, UMEM_NOFAIL);
3952 * free_percent of the time, free a range of bigobj rather than
3955 freeit = (ztest_random(100) < free_percent);
3958 * Read the current contents of our objects.
3960 error = dmu_read(os, packobj, packoff, packsize, packbuf,
3963 error = dmu_read(os, bigobj, bigoff, bigsize, bigbuf,
3968 * Get a tx for the mods to both packobj and bigobj.
3970 tx = dmu_tx_create(os);
3972 dmu_tx_hold_write(tx, packobj, packoff, packsize);
3975 dmu_tx_hold_free(tx, bigobj, bigoff, bigsize);
3977 dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
3979 /* This accounts for setting the checksum/compression. */
3980 dmu_tx_hold_bonus(tx, bigobj);
3982 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3984 umem_free(packbuf, packsize);
3985 umem_free(bigbuf, bigsize);
3986 umem_free(od, size);
3990 enum zio_checksum cksum;
3992 cksum = (enum zio_checksum)
3993 ztest_random_dsl_prop(ZFS_PROP_CHECKSUM);
3994 } while (cksum >= ZIO_CHECKSUM_LEGACY_FUNCTIONS);
3995 dmu_object_set_checksum(os, bigobj, cksum, tx);
3997 enum zio_compress comp;
3999 comp = (enum zio_compress)
4000 ztest_random_dsl_prop(ZFS_PROP_COMPRESSION);
4001 } while (comp >= ZIO_COMPRESS_LEGACY_FUNCTIONS);
4002 dmu_object_set_compress(os, bigobj, comp, tx);
4005 * For each index from n to n + s, verify that the existing bufwad
4006 * in packobj matches the bufwads at the head and tail of the
4007 * corresponding chunk in bigobj. Then update all three bufwads
4008 * with the new values we want to write out.
4010 for (i = 0; i < s; i++) {
4012 pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
4014 bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
4016 bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
4018 ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
4019 ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
4021 if (pack->bw_txg > txg)
4022 fatal(0, "future leak: got %llx, open txg is %llx",
4025 if (pack->bw_data != 0 && pack->bw_index != n + i)
4026 fatal(0, "wrong index: got %llx, wanted %llx+%llx",
4027 pack->bw_index, n, i);
4029 if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
4030 fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
4032 if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
4033 fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
4036 bzero(pack, sizeof (bufwad_t));
4038 pack->bw_index = n + i;
4040 pack->bw_data = 1 + ztest_random(-2ULL);
4047 * We've verified all the old bufwads, and made new ones.
4048 * Now write them out.
4050 dmu_write(os, packobj, packoff, packsize, packbuf, tx);
4053 if (ztest_opts.zo_verbose >= 7) {
4054 (void) printf("freeing offset %llx size %llx"
4056 (u_longlong_t)bigoff,
4057 (u_longlong_t)bigsize,
4060 VERIFY(0 == dmu_free_range(os, bigobj, bigoff, bigsize, tx));
4062 if (ztest_opts.zo_verbose >= 7) {
4063 (void) printf("writing offset %llx size %llx"
4065 (u_longlong_t)bigoff,
4066 (u_longlong_t)bigsize,
4069 dmu_write(os, bigobj, bigoff, bigsize, bigbuf, tx);
4075 * Sanity check the stuff we just wrote.
4078 void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
4079 void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
4081 VERIFY(0 == dmu_read(os, packobj, packoff,
4082 packsize, packcheck, DMU_READ_PREFETCH));
4083 VERIFY(0 == dmu_read(os, bigobj, bigoff,
4084 bigsize, bigcheck, DMU_READ_PREFETCH));
4086 ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
4087 ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
4089 umem_free(packcheck, packsize);
4090 umem_free(bigcheck, bigsize);
4093 umem_free(packbuf, packsize);
4094 umem_free(bigbuf, bigsize);
4095 umem_free(od, size);
4099 compare_and_update_pbbufs(uint64_t s, bufwad_t *packbuf, bufwad_t *bigbuf,
4100 uint64_t bigsize, uint64_t n, uint64_t chunksize, uint64_t txg)
4108 * For each index from n to n + s, verify that the existing bufwad
4109 * in packobj matches the bufwads at the head and tail of the
4110 * corresponding chunk in bigobj. Then update all three bufwads
4111 * with the new values we want to write out.
4113 for (i = 0; i < s; i++) {
4115 pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
4117 bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
4119 bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
4121 ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
4122 ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
4124 if (pack->bw_txg > txg)
4125 fatal(0, "future leak: got %llx, open txg is %llx",
4128 if (pack->bw_data != 0 && pack->bw_index != n + i)
4129 fatal(0, "wrong index: got %llx, wanted %llx+%llx",
4130 pack->bw_index, n, i);
4132 if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
4133 fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
4135 if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
4136 fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
4138 pack->bw_index = n + i;
4140 pack->bw_data = 1 + ztest_random(-2ULL);
4147 #undef OD_ARRAY_SIZE
4148 #define OD_ARRAY_SIZE 2
4151 ztest_dmu_read_write_zcopy(ztest_ds_t *zd, uint64_t id)
4153 objset_t *os = zd->zd_os;
4160 bufwad_t *packbuf, *bigbuf;
4161 uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
4162 uint64_t blocksize = ztest_random_blocksize();
4163 uint64_t chunksize = blocksize;
4164 uint64_t regions = 997;
4165 uint64_t stride = 123456789ULL;
4167 dmu_buf_t *bonus_db;
4168 arc_buf_t **bigbuf_arcbufs;
4169 dmu_object_info_t doi;
4171 size = sizeof (ztest_od_t) * OD_ARRAY_SIZE;
4172 od = umem_alloc(size, UMEM_NOFAIL);
4175 * This test uses two objects, packobj and bigobj, that are always
4176 * updated together (i.e. in the same tx) so that their contents are
4177 * in sync and can be compared. Their contents relate to each other
4178 * in a simple way: packobj is a dense array of 'bufwad' structures,
4179 * while bigobj is a sparse array of the same bufwads. Specifically,
4180 * for any index n, there are three bufwads that should be identical:
4182 * packobj, at offset n * sizeof (bufwad_t)
4183 * bigobj, at the head of the nth chunk
4184 * bigobj, at the tail of the nth chunk
4186 * The chunk size is set equal to bigobj block size so that
4187 * dmu_assign_arcbuf() can be tested for object updates.
4191 * Read the directory info. If it's the first time, set things up.
4193 ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0, 0);
4194 ztest_od_init(od + 1, id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, 0,
4198 if (ztest_object_init(zd, od, size, B_FALSE) != 0) {
4199 umem_free(od, size);
4203 bigobj = od[0].od_object;
4204 packobj = od[1].od_object;
4205 blocksize = od[0].od_blocksize;
4206 chunksize = blocksize;
4207 ASSERT(chunksize == od[1].od_gen);
4209 VERIFY(dmu_object_info(os, bigobj, &doi) == 0);
4210 VERIFY(ISP2(doi.doi_data_block_size));
4211 VERIFY(chunksize == doi.doi_data_block_size);
4212 VERIFY(chunksize >= 2 * sizeof (bufwad_t));
4215 * Pick a random index and compute the offsets into packobj and bigobj.
4217 n = ztest_random(regions) * stride + ztest_random(width);
4218 s = 1 + ztest_random(width - 1);
4220 packoff = n * sizeof (bufwad_t);
4221 packsize = s * sizeof (bufwad_t);
4223 bigoff = n * chunksize;
4224 bigsize = s * chunksize;
4226 packbuf = umem_zalloc(packsize, UMEM_NOFAIL);
4227 bigbuf = umem_zalloc(bigsize, UMEM_NOFAIL);
4229 VERIFY3U(0, ==, dmu_bonus_hold(os, bigobj, FTAG, &bonus_db));
4231 bigbuf_arcbufs = umem_zalloc(2 * s * sizeof (arc_buf_t *), UMEM_NOFAIL);
4234 * Iteration 0 test zcopy for DB_UNCACHED dbufs.
4235 * Iteration 1 test zcopy to already referenced dbufs.
4236 * Iteration 2 test zcopy to dirty dbuf in the same txg.
4237 * Iteration 3 test zcopy to dbuf dirty in previous txg.
4238 * Iteration 4 test zcopy when dbuf is no longer dirty.
4239 * Iteration 5 test zcopy when it can't be done.
4240 * Iteration 6 one more zcopy write.
4242 for (i = 0; i < 7; i++) {
4247 * In iteration 5 (i == 5) use arcbufs
4248 * that don't match bigobj blksz to test
4249 * dmu_assign_arcbuf() when it can't directly
4250 * assign an arcbuf to a dbuf.
4252 for (j = 0; j < s; j++) {
4253 if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) {
4255 dmu_request_arcbuf(bonus_db, chunksize);
4257 bigbuf_arcbufs[2 * j] =
4258 dmu_request_arcbuf(bonus_db, chunksize / 2);
4259 bigbuf_arcbufs[2 * j + 1] =
4260 dmu_request_arcbuf(bonus_db, chunksize / 2);
4265 * Get a tx for the mods to both packobj and bigobj.
4267 tx = dmu_tx_create(os);
4269 dmu_tx_hold_write(tx, packobj, packoff, packsize);
4270 dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
4272 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4274 umem_free(packbuf, packsize);
4275 umem_free(bigbuf, bigsize);
4276 for (j = 0; j < s; j++) {
4278 chunksize < (SPA_MINBLOCKSIZE * 2)) {
4279 dmu_return_arcbuf(bigbuf_arcbufs[j]);
4282 bigbuf_arcbufs[2 * j]);
4284 bigbuf_arcbufs[2 * j + 1]);
4287 umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
4288 umem_free(od, size);
4289 dmu_buf_rele(bonus_db, FTAG);
4294 * 50% of the time don't read objects in the 1st iteration to
4295 * test dmu_assign_arcbuf() for the case when there're no
4296 * existing dbufs for the specified offsets.
4298 if (i != 0 || ztest_random(2) != 0) {
4299 error = dmu_read(os, packobj, packoff,
4300 packsize, packbuf, DMU_READ_PREFETCH);
4302 error = dmu_read(os, bigobj, bigoff, bigsize,
4303 bigbuf, DMU_READ_PREFETCH);
4306 compare_and_update_pbbufs(s, packbuf, bigbuf, bigsize,
4310 * We've verified all the old bufwads, and made new ones.
4311 * Now write them out.
4313 dmu_write(os, packobj, packoff, packsize, packbuf, tx);
4314 if (ztest_opts.zo_verbose >= 7) {
4315 (void) printf("writing offset %llx size %llx"
4317 (u_longlong_t)bigoff,
4318 (u_longlong_t)bigsize,
4321 for (off = bigoff, j = 0; j < s; j++, off += chunksize) {
4323 if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) {
4324 bcopy((caddr_t)bigbuf + (off - bigoff),
4325 bigbuf_arcbufs[j]->b_data, chunksize);
4327 bcopy((caddr_t)bigbuf + (off - bigoff),
4328 bigbuf_arcbufs[2 * j]->b_data,
4330 bcopy((caddr_t)bigbuf + (off - bigoff) +
4332 bigbuf_arcbufs[2 * j + 1]->b_data,
4337 VERIFY(dmu_buf_hold(os, bigobj, off,
4338 FTAG, &dbt, DMU_READ_NO_PREFETCH) == 0);
4340 if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) {
4341 dmu_assign_arcbuf(bonus_db, off,
4342 bigbuf_arcbufs[j], tx);
4344 dmu_assign_arcbuf(bonus_db, off,
4345 bigbuf_arcbufs[2 * j], tx);
4346 dmu_assign_arcbuf(bonus_db,
4347 off + chunksize / 2,
4348 bigbuf_arcbufs[2 * j + 1], tx);
4351 dmu_buf_rele(dbt, FTAG);
4357 * Sanity check the stuff we just wrote.
4360 void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
4361 void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
4363 VERIFY(0 == dmu_read(os, packobj, packoff,
4364 packsize, packcheck, DMU_READ_PREFETCH));
4365 VERIFY(0 == dmu_read(os, bigobj, bigoff,
4366 bigsize, bigcheck, DMU_READ_PREFETCH));
4368 ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
4369 ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
4371 umem_free(packcheck, packsize);
4372 umem_free(bigcheck, bigsize);
4375 txg_wait_open(dmu_objset_pool(os), 0);
4376 } else if (i == 3) {
4377 txg_wait_synced(dmu_objset_pool(os), 0);
4381 dmu_buf_rele(bonus_db, FTAG);
4382 umem_free(packbuf, packsize);
4383 umem_free(bigbuf, bigsize);
4384 umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
4385 umem_free(od, size);
4390 ztest_dmu_write_parallel(ztest_ds_t *zd, uint64_t id)
4394 od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
4395 uint64_t offset = (1ULL << (ztest_random(20) + 43)) +
4396 (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
4399 * Have multiple threads write to large offsets in an object
4400 * to verify that parallel writes to an object -- even to the
4401 * same blocks within the object -- doesn't cause any trouble.
4403 ztest_od_init(od, ID_PARALLEL, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0, 0);
4405 if (ztest_object_init(zd, od, sizeof (ztest_od_t), B_FALSE) != 0)
4408 while (ztest_random(10) != 0)
4409 ztest_io(zd, od->od_object, offset);
4411 umem_free(od, sizeof (ztest_od_t));
4415 ztest_dmu_prealloc(ztest_ds_t *zd, uint64_t id)
4418 uint64_t offset = (1ULL << (ztest_random(4) + SPA_MAXBLOCKSHIFT)) +
4419 (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
4420 uint64_t count = ztest_random(20) + 1;
4421 uint64_t blocksize = ztest_random_blocksize();
4424 od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
4426 ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0, 0);
4428 if (ztest_object_init(zd, od, sizeof (ztest_od_t),
4429 !ztest_random(2)) != 0) {
4430 umem_free(od, sizeof (ztest_od_t));
4434 if (ztest_truncate(zd, od->od_object, offset, count * blocksize) != 0) {
4435 umem_free(od, sizeof (ztest_od_t));
4439 ztest_prealloc(zd, od->od_object, offset, count * blocksize);
4441 data = umem_zalloc(blocksize, UMEM_NOFAIL);
4443 while (ztest_random(count) != 0) {
4444 uint64_t randoff = offset + (ztest_random(count) * blocksize);
4445 if (ztest_write(zd, od->od_object, randoff, blocksize,
4448 while (ztest_random(4) != 0)
4449 ztest_io(zd, od->od_object, randoff);
4452 umem_free(data, blocksize);
4453 umem_free(od, sizeof (ztest_od_t));
4457 * Verify that zap_{create,destroy,add,remove,update} work as expected.
4459 #define ZTEST_ZAP_MIN_INTS 1
4460 #define ZTEST_ZAP_MAX_INTS 4
4461 #define ZTEST_ZAP_MAX_PROPS 1000
4464 ztest_zap(ztest_ds_t *zd, uint64_t id)
4466 objset_t *os = zd->zd_os;
4469 uint64_t txg, last_txg;
4470 uint64_t value[ZTEST_ZAP_MAX_INTS];
4471 uint64_t zl_ints, zl_intsize, prop;
4474 char propname[100], txgname[100];
4476 char *hc[2] = { "s.acl.h", ".s.open.h.hyLZlg" };
4478 od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
4479 ztest_od_init(od, id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0, 0);
4481 if (ztest_object_init(zd, od, sizeof (ztest_od_t),
4482 !ztest_random(2)) != 0)
4485 object = od->od_object;
4488 * Generate a known hash collision, and verify that
4489 * we can lookup and remove both entries.
4491 tx = dmu_tx_create(os);
4492 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4493 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4496 for (i = 0; i < 2; i++) {
4498 VERIFY3U(0, ==, zap_add(os, object, hc[i], sizeof (uint64_t),
4501 for (i = 0; i < 2; i++) {
4502 VERIFY3U(EEXIST, ==, zap_add(os, object, hc[i],
4503 sizeof (uint64_t), 1, &value[i], tx));
4505 zap_length(os, object, hc[i], &zl_intsize, &zl_ints));
4506 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4507 ASSERT3U(zl_ints, ==, 1);
4509 for (i = 0; i < 2; i++) {
4510 VERIFY3U(0, ==, zap_remove(os, object, hc[i], tx));
4515 * Generate a buch of random entries.
4517 ints = MAX(ZTEST_ZAP_MIN_INTS, object % ZTEST_ZAP_MAX_INTS);
4519 prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
4520 (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
4521 (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
4522 bzero(value, sizeof (value));
4526 * If these zap entries already exist, validate their contents.
4528 error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
4530 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4531 ASSERT3U(zl_ints, ==, 1);
4533 VERIFY(zap_lookup(os, object, txgname, zl_intsize,
4534 zl_ints, &last_txg) == 0);
4536 VERIFY(zap_length(os, object, propname, &zl_intsize,
4539 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4540 ASSERT3U(zl_ints, ==, ints);
4542 VERIFY(zap_lookup(os, object, propname, zl_intsize,
4543 zl_ints, value) == 0);
4545 for (i = 0; i < ints; i++) {
4546 ASSERT3U(value[i], ==, last_txg + object + i);
4549 ASSERT3U(error, ==, ENOENT);
4553 * Atomically update two entries in our zap object.
4554 * The first is named txg_%llu, and contains the txg
4555 * in which the property was last updated. The second
4556 * is named prop_%llu, and the nth element of its value
4557 * should be txg + object + n.
4559 tx = dmu_tx_create(os);
4560 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4561 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4566 fatal(0, "zap future leak: old %llu new %llu", last_txg, txg);
4568 for (i = 0; i < ints; i++)
4569 value[i] = txg + object + i;
4571 VERIFY3U(0, ==, zap_update(os, object, txgname, sizeof (uint64_t),
4573 VERIFY3U(0, ==, zap_update(os, object, propname, sizeof (uint64_t),
4579 * Remove a random pair of entries.
4581 prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
4582 (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
4583 (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
4585 error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
4587 if (error == ENOENT)
4592 tx = dmu_tx_create(os);
4593 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4594 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4597 VERIFY3U(0, ==, zap_remove(os, object, txgname, tx));
4598 VERIFY3U(0, ==, zap_remove(os, object, propname, tx));
4601 umem_free(od, sizeof (ztest_od_t));
4605 * Testcase to test the upgrading of a microzap to fatzap.
4608 ztest_fzap(ztest_ds_t *zd, uint64_t id)
4610 objset_t *os = zd->zd_os;
4612 uint64_t object, txg;
4615 od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
4616 ztest_od_init(od, id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0, 0);
4618 if (ztest_object_init(zd, od, sizeof (ztest_od_t),
4619 !ztest_random(2)) != 0)
4621 object = od->od_object;
4624 * Add entries to this ZAP and make sure it spills over
4625 * and gets upgraded to a fatzap. Also, since we are adding
4626 * 2050 entries we should see ptrtbl growth and leaf-block split.
4628 for (i = 0; i < 2050; i++) {
4629 char name[ZFS_MAX_DATASET_NAME_LEN];
4634 (void) snprintf(name, sizeof (name), "fzap-%llu-%llu",
4635 (u_longlong_t)id, (u_longlong_t)value);
4637 tx = dmu_tx_create(os);
4638 dmu_tx_hold_zap(tx, object, B_TRUE, name);
4639 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4642 error = zap_add(os, object, name, sizeof (uint64_t), 1,
4644 ASSERT(error == 0 || error == EEXIST);
4648 umem_free(od, sizeof (ztest_od_t));
4653 ztest_zap_parallel(ztest_ds_t *zd, uint64_t id)
4655 objset_t *os = zd->zd_os;
4657 uint64_t txg, object, count, wsize, wc, zl_wsize, zl_wc;
4659 int i, namelen, error;
4660 int micro = ztest_random(2);
4661 char name[20], string_value[20];
4664 od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
4665 ztest_od_init(od, ID_PARALLEL, FTAG, micro, DMU_OT_ZAP_OTHER, 0, 0, 0);
4667 if (ztest_object_init(zd, od, sizeof (ztest_od_t), B_FALSE) != 0) {
4668 umem_free(od, sizeof (ztest_od_t));
4672 object = od->od_object;
4675 * Generate a random name of the form 'xxx.....' where each
4676 * x is a random printable character and the dots are dots.
4677 * There are 94 such characters, and the name length goes from
4678 * 6 to 20, so there are 94^3 * 15 = 12,458,760 possible names.
4680 namelen = ztest_random(sizeof (name) - 5) + 5 + 1;
4682 for (i = 0; i < 3; i++)
4683 name[i] = '!' + ztest_random('~' - '!' + 1);
4684 for (; i < namelen - 1; i++)
4688 if ((namelen & 1) || micro) {
4689 wsize = sizeof (txg);
4695 data = string_value;
4699 VERIFY0(zap_count(os, object, &count));
4700 ASSERT(count != -1ULL);
4703 * Select an operation: length, lookup, add, update, remove.
4705 i = ztest_random(5);
4708 tx = dmu_tx_create(os);
4709 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4710 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4712 umem_free(od, sizeof (ztest_od_t));
4715 bcopy(name, string_value, namelen);
4719 bzero(string_value, namelen);
4725 error = zap_length(os, object, name, &zl_wsize, &zl_wc);
4727 ASSERT3U(wsize, ==, zl_wsize);
4728 ASSERT3U(wc, ==, zl_wc);
4730 ASSERT3U(error, ==, ENOENT);
4735 error = zap_lookup(os, object, name, wsize, wc, data);
4737 if (data == string_value &&
4738 bcmp(name, data, namelen) != 0)
4739 fatal(0, "name '%s' != val '%s' len %d",
4740 name, data, namelen);
4742 ASSERT3U(error, ==, ENOENT);
4747 error = zap_add(os, object, name, wsize, wc, data, tx);
4748 ASSERT(error == 0 || error == EEXIST);
4752 VERIFY(zap_update(os, object, name, wsize, wc, data, tx) == 0);
4756 error = zap_remove(os, object, name, tx);
4757 ASSERT(error == 0 || error == ENOENT);
4764 umem_free(od, sizeof (ztest_od_t));
4768 * Commit callback data.
4770 typedef struct ztest_cb_data {
4771 list_node_t zcd_node;
4773 int zcd_expected_err;
4774 boolean_t zcd_added;
4775 boolean_t zcd_called;
4779 /* This is the actual commit callback function */
4781 ztest_commit_callback(void *arg, int error)
4783 ztest_cb_data_t *data = arg;
4784 uint64_t synced_txg;
4786 VERIFY(data != NULL);
4787 VERIFY3S(data->zcd_expected_err, ==, error);
4788 VERIFY(!data->zcd_called);
4790 synced_txg = spa_last_synced_txg(data->zcd_spa);
4791 if (data->zcd_txg > synced_txg)
4792 fatal(0, "commit callback of txg %" PRIu64 " called prematurely"
4793 ", last synced txg = %" PRIu64 "\n", data->zcd_txg,
4796 data->zcd_called = B_TRUE;
4798 if (error == ECANCELED) {
4799 ASSERT0(data->zcd_txg);
4800 ASSERT(!data->zcd_added);
4803 * The private callback data should be destroyed here, but
4804 * since we are going to check the zcd_called field after
4805 * dmu_tx_abort(), we will destroy it there.
4810 ASSERT(data->zcd_added);
4811 ASSERT3U(data->zcd_txg, !=, 0);
4813 (void) mutex_enter(&zcl.zcl_callbacks_lock);
4815 /* See if this cb was called more quickly */
4816 if ((synced_txg - data->zcd_txg) < zc_min_txg_delay)
4817 zc_min_txg_delay = synced_txg - data->zcd_txg;
4819 /* Remove our callback from the list */
4820 list_remove(&zcl.zcl_callbacks, data);
4822 (void) mutex_exit(&zcl.zcl_callbacks_lock);
4824 umem_free(data, sizeof (ztest_cb_data_t));
4827 /* Allocate and initialize callback data structure */
4828 static ztest_cb_data_t *
4829 ztest_create_cb_data(objset_t *os, uint64_t txg)
4831 ztest_cb_data_t *cb_data;
4833 cb_data = umem_zalloc(sizeof (ztest_cb_data_t), UMEM_NOFAIL);
4835 cb_data->zcd_txg = txg;
4836 cb_data->zcd_spa = dmu_objset_spa(os);
4837 list_link_init(&cb_data->zcd_node);
4843 * Commit callback test.
4846 ztest_dmu_commit_callbacks(ztest_ds_t *zd, uint64_t id)
4848 objset_t *os = zd->zd_os;
4851 ztest_cb_data_t *cb_data[3], *tmp_cb;
4852 uint64_t old_txg, txg;
4855 od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
4856 ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0, 0);
4858 if (ztest_object_init(zd, od, sizeof (ztest_od_t), B_FALSE) != 0) {
4859 umem_free(od, sizeof (ztest_od_t));
4863 tx = dmu_tx_create(os);
4865 cb_data[0] = ztest_create_cb_data(os, 0);
4866 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[0]);
4868 dmu_tx_hold_write(tx, od->od_object, 0, sizeof (uint64_t));
4870 /* Every once in a while, abort the transaction on purpose */
4871 if (ztest_random(100) == 0)
4875 error = dmu_tx_assign(tx, TXG_NOWAIT);
4877 txg = error ? 0 : dmu_tx_get_txg(tx);
4879 cb_data[0]->zcd_txg = txg;
4880 cb_data[1] = ztest_create_cb_data(os, txg);
4881 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[1]);
4885 * It's not a strict requirement to call the registered
4886 * callbacks from inside dmu_tx_abort(), but that's what
4887 * it's supposed to happen in the current implementation
4888 * so we will check for that.
4890 for (i = 0; i < 2; i++) {
4891 cb_data[i]->zcd_expected_err = ECANCELED;
4892 VERIFY(!cb_data[i]->zcd_called);
4897 for (i = 0; i < 2; i++) {
4898 VERIFY(cb_data[i]->zcd_called);
4899 umem_free(cb_data[i], sizeof (ztest_cb_data_t));
4902 umem_free(od, sizeof (ztest_od_t));
4906 cb_data[2] = ztest_create_cb_data(os, txg);
4907 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[2]);
4910 * Read existing data to make sure there isn't a future leak.
4912 VERIFY(0 == dmu_read(os, od->od_object, 0, sizeof (uint64_t),
4913 &old_txg, DMU_READ_PREFETCH));
4916 fatal(0, "future leak: got %" PRIu64 ", open txg is %" PRIu64,
4919 dmu_write(os, od->od_object, 0, sizeof (uint64_t), &txg, tx);
4921 (void) mutex_enter(&zcl.zcl_callbacks_lock);
4924 * Since commit callbacks don't have any ordering requirement and since
4925 * it is theoretically possible for a commit callback to be called
4926 * after an arbitrary amount of time has elapsed since its txg has been
4927 * synced, it is difficult to reliably determine whether a commit
4928 * callback hasn't been called due to high load or due to a flawed
4931 * In practice, we will assume that if after a certain number of txgs a
4932 * commit callback hasn't been called, then most likely there's an
4933 * implementation bug..
4935 tmp_cb = list_head(&zcl.zcl_callbacks);
4936 if (tmp_cb != NULL &&
4937 tmp_cb->zcd_txg + ZTEST_COMMIT_CB_THRESH < txg) {
4938 fatal(0, "Commit callback threshold exceeded, oldest txg: %"
4939 PRIu64 ", open txg: %" PRIu64 "\n", tmp_cb->zcd_txg, txg);
4943 * Let's find the place to insert our callbacks.
4945 * Even though the list is ordered by txg, it is possible for the
4946 * insertion point to not be the end because our txg may already be
4947 * quiescing at this point and other callbacks in the open txg
4948 * (from other objsets) may have sneaked in.
4950 tmp_cb = list_tail(&zcl.zcl_callbacks);
4951 while (tmp_cb != NULL && tmp_cb->zcd_txg > txg)
4952 tmp_cb = list_prev(&zcl.zcl_callbacks, tmp_cb);
4954 /* Add the 3 callbacks to the list */
4955 for (i = 0; i < 3; i++) {
4957 list_insert_head(&zcl.zcl_callbacks, cb_data[i]);
4959 list_insert_after(&zcl.zcl_callbacks, tmp_cb,
4962 cb_data[i]->zcd_added = B_TRUE;
4963 VERIFY(!cb_data[i]->zcd_called);
4965 tmp_cb = cb_data[i];
4970 (void) mutex_exit(&zcl.zcl_callbacks_lock);
4974 umem_free(od, sizeof (ztest_od_t));
4978 * Visit each object in the dataset. Verify that its properties
4979 * are consistent what was stored in the block tag when it was created,
4980 * and that its unused bonus buffer space has not been overwritten.
4983 ztest_verify_dnode_bt(ztest_ds_t *zd, uint64_t id)
4985 objset_t *os = zd->zd_os;
4989 for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, 0)) {
4990 ztest_block_tag_t *bt = NULL;
4991 dmu_object_info_t doi;
4994 if (dmu_bonus_hold(os, obj, FTAG, &db) != 0)
4997 dmu_object_info_from_db(db, &doi);
4998 if (doi.doi_bonus_size >= sizeof (*bt))
4999 bt = ztest_bt_bonus(db);
5001 if (bt && bt->bt_magic == BT_MAGIC) {
5002 ztest_bt_verify(bt, os, obj, doi.doi_dnodesize,
5003 bt->bt_offset, bt->bt_gen, bt->bt_txg,
5005 ztest_verify_unused_bonus(db, bt, obj, os, bt->bt_gen);
5008 dmu_buf_rele(db, FTAG);
5014 ztest_dsl_prop_get_set(ztest_ds_t *zd, uint64_t id)
5016 zfs_prop_t proplist[] = {
5018 ZFS_PROP_COMPRESSION,
5024 (void) rw_rdlock(&ztest_name_lock);
5026 for (p = 0; p < sizeof (proplist) / sizeof (proplist[0]); p++)
5027 (void) ztest_dsl_prop_set_uint64(zd->zd_name, proplist[p],
5028 ztest_random_dsl_prop(proplist[p]), (int)ztest_random(2));
5030 VERIFY0(ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_RECORDSIZE,
5031 ztest_random_blocksize(), (int)ztest_random(2)));
5033 (void) rw_unlock(&ztest_name_lock);
5038 ztest_spa_prop_get_set(ztest_ds_t *zd, uint64_t id)
5040 nvlist_t *props = NULL;
5042 (void) rw_rdlock(&ztest_name_lock);
5044 (void) ztest_spa_prop_set_uint64(ZPOOL_PROP_DEDUPDITTO,
5045 ZIO_DEDUPDITTO_MIN + ztest_random(ZIO_DEDUPDITTO_MIN));
5047 VERIFY0(spa_prop_get(ztest_spa, &props));
5049 if (ztest_opts.zo_verbose >= 6)
5050 dump_nvlist(props, 4);
5054 (void) rw_unlock(&ztest_name_lock);
5058 user_release_one(const char *snapname, const char *holdname)
5060 nvlist_t *snaps, *holds;
5063 snaps = fnvlist_alloc();
5064 holds = fnvlist_alloc();
5065 fnvlist_add_boolean(holds, holdname);
5066 fnvlist_add_nvlist(snaps, snapname, holds);
5067 fnvlist_free(holds);
5068 error = dsl_dataset_user_release(snaps, NULL);
5069 fnvlist_free(snaps);
5074 * Test snapshot hold/release and deferred destroy.
5077 ztest_dmu_snapshot_hold(ztest_ds_t *zd, uint64_t id)
5080 objset_t *os = zd->zd_os;
5084 char clonename[100];
5086 char osname[ZFS_MAX_DATASET_NAME_LEN];
5089 (void) rw_rdlock(&ztest_name_lock);
5091 dmu_objset_name(os, osname);
5093 (void) snprintf(snapname, sizeof (snapname), "sh1_%llu",
5095 (void) snprintf(fullname, sizeof (fullname), "%s@%s", osname, snapname);
5096 (void) snprintf(clonename, sizeof (clonename),
5097 "%s/ch1_%llu", osname, (u_longlong_t)id);
5098 (void) snprintf(tag, sizeof (tag), "tag_%llu", (u_longlong_t)id);
5101 * Clean up from any previous run.
5103 error = dsl_destroy_head(clonename);
5104 if (error != ENOENT)
5106 error = user_release_one(fullname, tag);
5107 if (error != ESRCH && error != ENOENT)
5109 error = dsl_destroy_snapshot(fullname, B_FALSE);
5110 if (error != ENOENT)
5114 * Create snapshot, clone it, mark snap for deferred destroy,
5115 * destroy clone, verify snap was also destroyed.
5117 error = dmu_objset_snapshot_one(osname, snapname);
5119 if (error == ENOSPC) {
5120 ztest_record_enospc("dmu_objset_snapshot");
5123 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
5126 error = dmu_objset_clone(clonename, fullname);
5128 if (error == ENOSPC) {
5129 ztest_record_enospc("dmu_objset_clone");
5132 fatal(0, "dmu_objset_clone(%s) = %d", clonename, error);
5135 error = dsl_destroy_snapshot(fullname, B_TRUE);
5137 fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
5141 error = dsl_destroy_head(clonename);
5143 fatal(0, "dsl_destroy_head(%s) = %d", clonename, error);
5145 error = dmu_objset_hold(fullname, FTAG, &origin);
5146 if (error != ENOENT)
5147 fatal(0, "dmu_objset_hold(%s) = %d", fullname, error);
5150 * Create snapshot, add temporary hold, verify that we can't
5151 * destroy a held snapshot, mark for deferred destroy,
5152 * release hold, verify snapshot was destroyed.
5154 error = dmu_objset_snapshot_one(osname, snapname);
5156 if (error == ENOSPC) {
5157 ztest_record_enospc("dmu_objset_snapshot");
5160 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
5163 holds = fnvlist_alloc();
5164 fnvlist_add_string(holds, fullname, tag);
5165 error = dsl_dataset_user_hold(holds, 0, NULL);
5166 fnvlist_free(holds);
5168 if (error == ENOSPC) {
5169 ztest_record_enospc("dsl_dataset_user_hold");
5172 fatal(0, "dsl_dataset_user_hold(%s, %s) = %u",
5173 fullname, tag, error);
5176 error = dsl_destroy_snapshot(fullname, B_FALSE);
5177 if (error != EBUSY) {
5178 fatal(0, "dsl_destroy_snapshot(%s, B_FALSE) = %d",
5182 error = dsl_destroy_snapshot(fullname, B_TRUE);
5184 fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
5188 error = user_release_one(fullname, tag);
5190 fatal(0, "user_release_one(%s, %s) = %d", fullname, tag, error);
5192 VERIFY3U(dmu_objset_hold(fullname, FTAG, &origin), ==, ENOENT);
5195 (void) rw_unlock(&ztest_name_lock);
5199 * Inject random faults into the on-disk data.
5203 ztest_fault_inject(ztest_ds_t *zd, uint64_t id)
5205 ztest_shared_t *zs = ztest_shared;
5206 spa_t *spa = ztest_spa;
5210 uint64_t bad = 0x1990c0ffeedecadeull;
5215 int bshift = SPA_MAXBLOCKSHIFT + 2; /* don't scrog all labels */
5221 boolean_t islog = B_FALSE;
5223 path0 = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
5224 pathrand = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
5226 mutex_enter(&ztest_vdev_lock);
5227 maxfaults = MAXFAULTS();
5228 leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz;
5229 mirror_save = zs->zs_mirrors;
5230 mutex_exit(&ztest_vdev_lock);
5232 ASSERT(leaves >= 1);
5235 * Grab the name lock as reader. There are some operations
5236 * which don't like to have their vdevs changed while
5237 * they are in progress (i.e. spa_change_guid). Those
5238 * operations will have grabbed the name lock as writer.
5240 (void) rw_rdlock(&ztest_name_lock);
5243 * We need SCL_STATE here because we're going to look at vd0->vdev_tsd.
5245 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5247 if (ztest_random(2) == 0) {
5249 * Inject errors on a normal data device or slog device.
5251 top = ztest_random_vdev_top(spa, B_TRUE);
5252 leaf = ztest_random(leaves) + zs->zs_splits;
5255 * Generate paths to the first leaf in this top-level vdev,
5256 * and to the random leaf we selected. We'll induce transient
5257 * write failures and random online/offline activity on leaf 0,
5258 * and we'll write random garbage to the randomly chosen leaf.
5260 (void) snprintf(path0, MAXPATHLEN, ztest_dev_template,
5261 ztest_opts.zo_dir, ztest_opts.zo_pool,
5262 top * leaves + zs->zs_splits);
5263 (void) snprintf(pathrand, MAXPATHLEN, ztest_dev_template,
5264 ztest_opts.zo_dir, ztest_opts.zo_pool,
5265 top * leaves + leaf);
5267 vd0 = vdev_lookup_by_path(spa->spa_root_vdev, path0);
5268 if (vd0 != NULL && vd0->vdev_top->vdev_islog)
5272 * If the top-level vdev needs to be resilvered
5273 * then we only allow faults on the device that is
5276 if (vd0 != NULL && maxfaults != 1 &&
5277 (!vdev_resilver_needed(vd0->vdev_top, NULL, NULL) ||
5278 vd0->vdev_resilver_txg != 0)) {
5280 * Make vd0 explicitly claim to be unreadable,
5281 * or unwriteable, or reach behind its back
5282 * and close the underlying fd. We can do this if
5283 * maxfaults == 0 because we'll fail and reexecute,
5284 * and we can do it if maxfaults >= 2 because we'll
5285 * have enough redundancy. If maxfaults == 1, the
5286 * combination of this with injection of random data
5287 * corruption below exceeds the pool's fault tolerance.
5289 vdev_file_t *vf = vd0->vdev_tsd;
5291 if (vf != NULL && ztest_random(3) == 0) {
5292 (void) close(vf->vf_vnode->v_fd);
5293 vf->vf_vnode->v_fd = -1;
5294 } else if (ztest_random(2) == 0) {
5295 vd0->vdev_cant_read = B_TRUE;
5297 vd0->vdev_cant_write = B_TRUE;
5299 guid0 = vd0->vdev_guid;
5303 * Inject errors on an l2cache device.
5305 spa_aux_vdev_t *sav = &spa->spa_l2cache;
5307 if (sav->sav_count == 0) {
5308 spa_config_exit(spa, SCL_STATE, FTAG);
5309 (void) rw_unlock(&ztest_name_lock);
5312 vd0 = sav->sav_vdevs[ztest_random(sav->sav_count)];
5313 guid0 = vd0->vdev_guid;
5314 (void) strcpy(path0, vd0->vdev_path);
5315 (void) strcpy(pathrand, vd0->vdev_path);
5319 maxfaults = INT_MAX; /* no limit on cache devices */
5322 spa_config_exit(spa, SCL_STATE, FTAG);
5323 (void) rw_unlock(&ztest_name_lock);
5326 * If we can tolerate two or more faults, or we're dealing
5327 * with a slog, randomly online/offline vd0.
5329 if ((maxfaults >= 2 || islog) && guid0 != 0) {
5330 if (ztest_random(10) < 6) {
5331 int flags = (ztest_random(2) == 0 ?
5332 ZFS_OFFLINE_TEMPORARY : 0);
5335 * We have to grab the zs_name_lock as writer to
5336 * prevent a race between offlining a slog and
5337 * destroying a dataset. Offlining the slog will
5338 * grab a reference on the dataset which may cause
5339 * dsl_destroy_head() to fail with EBUSY thus
5340 * leaving the dataset in an inconsistent state.
5343 (void) rw_wrlock(&ztest_name_lock);
5345 VERIFY(vdev_offline(spa, guid0, flags) != EBUSY);
5348 (void) rw_unlock(&ztest_name_lock);
5351 * Ideally we would like to be able to randomly
5352 * call vdev_[on|off]line without holding locks
5353 * to force unpredictable failures but the side
5354 * effects of vdev_[on|off]line prevent us from
5355 * doing so. We grab the ztest_vdev_lock here to
5356 * prevent a race between injection testing and
5359 mutex_enter(&ztest_vdev_lock);
5360 (void) vdev_online(spa, guid0, 0, NULL);
5361 mutex_exit(&ztest_vdev_lock);
5369 * We have at least single-fault tolerance, so inject data corruption.
5371 fd = open(pathrand, O_RDWR);
5373 if (fd == -1) /* we hit a gap in the device namespace */
5376 fsize = lseek(fd, 0, SEEK_END);
5378 while (--iters != 0) {
5380 * The offset must be chosen carefully to ensure that
5381 * we do not inject a given logical block with errors
5382 * on two different leaf devices, because ZFS can not
5383 * tolerate that (if maxfaults==1).
5385 * We divide each leaf into chunks of size
5386 * (# leaves * SPA_MAXBLOCKSIZE * 4). Within each chunk
5387 * there is a series of ranges to which we can inject errors.
5388 * Each range can accept errors on only a single leaf vdev.
5389 * The error injection ranges are separated by ranges
5390 * which we will not inject errors on any device (DMZs).
5391 * Each DMZ must be large enough such that a single block
5392 * can not straddle it, so that a single block can not be
5393 * a target in two different injection ranges (on different
5396 * For example, with 3 leaves, each chunk looks like:
5397 * 0 to 32M: injection range for leaf 0
5398 * 32M to 64M: DMZ - no injection allowed
5399 * 64M to 96M: injection range for leaf 1
5400 * 96M to 128M: DMZ - no injection allowed
5401 * 128M to 160M: injection range for leaf 2
5402 * 160M to 192M: DMZ - no injection allowed
5404 offset = ztest_random(fsize / (leaves << bshift)) *
5405 (leaves << bshift) + (leaf << bshift) +
5406 (ztest_random(1ULL << (bshift - 1)) & -8ULL);
5408 if (offset >= fsize)
5411 mutex_enter(&ztest_vdev_lock);
5412 if (mirror_save != zs->zs_mirrors) {
5413 mutex_exit(&ztest_vdev_lock);
5418 if (pwrite(fd, &bad, sizeof (bad), offset) != sizeof (bad))
5419 fatal(1, "can't inject bad word at 0x%llx in %s",
5422 mutex_exit(&ztest_vdev_lock);
5424 if (ztest_opts.zo_verbose >= 7)
5425 (void) printf("injected bad word into %s,"
5426 " offset 0x%llx\n", pathrand, (u_longlong_t)offset);
5431 umem_free(path0, MAXPATHLEN);
5432 umem_free(pathrand, MAXPATHLEN);
5436 * Verify that DDT repair works as expected.
5439 ztest_ddt_repair(ztest_ds_t *zd, uint64_t id)
5441 ztest_shared_t *zs = ztest_shared;
5442 spa_t *spa = ztest_spa;
5443 objset_t *os = zd->zd_os;
5445 uint64_t object, blocksize, txg, pattern, psize;
5446 enum zio_checksum checksum = spa_dedup_checksum(spa);
5451 int copies = 2 * ZIO_DEDUPDITTO_MIN;
5454 blocksize = ztest_random_blocksize();
5455 blocksize = MIN(blocksize, 2048); /* because we write so many */
5457 od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
5458 ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0, 0);
5460 if (ztest_object_init(zd, od, sizeof (ztest_od_t), B_FALSE) != 0) {
5461 umem_free(od, sizeof (ztest_od_t));
5466 * Take the name lock as writer to prevent anyone else from changing
5467 * the pool and dataset properies we need to maintain during this test.
5469 (void) rw_wrlock(&ztest_name_lock);
5471 if (ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_DEDUP, checksum,
5473 ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_COPIES, 1,
5475 (void) rw_unlock(&ztest_name_lock);
5476 umem_free(od, sizeof (ztest_od_t));
5480 object = od[0].od_object;
5481 blocksize = od[0].od_blocksize;
5482 pattern = zs->zs_guid ^ dmu_objset_fsid_guid(os);
5484 ASSERT(object != 0);
5486 tx = dmu_tx_create(os);
5487 dmu_tx_hold_write(tx, object, 0, copies * blocksize);
5488 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
5490 (void) rw_unlock(&ztest_name_lock);
5491 umem_free(od, sizeof (ztest_od_t));
5496 * Write all the copies of our block.
5498 for (i = 0; i < copies; i++) {
5499 uint64_t offset = i * blocksize;
5500 int error = dmu_buf_hold(os, object, offset, FTAG, &db,
5501 DMU_READ_NO_PREFETCH);
5503 fatal(B_FALSE, "dmu_buf_hold(%p, %llu, %llu) = %u",
5504 os, (long long)object, (long long) offset, error);
5506 ASSERT(db->db_offset == offset);
5507 ASSERT(db->db_size == blocksize);
5508 ASSERT(ztest_pattern_match(db->db_data, db->db_size, pattern) ||
5509 ztest_pattern_match(db->db_data, db->db_size, 0ULL));
5510 dmu_buf_will_fill(db, tx);
5511 ztest_pattern_set(db->db_data, db->db_size, pattern);
5512 dmu_buf_rele(db, FTAG);
5516 txg_wait_synced(spa_get_dsl(spa), txg);
5519 * Find out what block we got.
5521 VERIFY0(dmu_buf_hold(os, object, 0, FTAG, &db,
5522 DMU_READ_NO_PREFETCH));
5523 blk = *((dmu_buf_impl_t *)db)->db_blkptr;
5524 dmu_buf_rele(db, FTAG);
5527 * Damage the block. Dedup-ditto will save us when we read it later.
5529 psize = BP_GET_PSIZE(&blk);
5530 abd = abd_alloc_linear(psize, B_TRUE);
5531 ztest_pattern_set(abd_to_buf(abd), psize, ~pattern);
5533 (void) zio_wait(zio_rewrite(NULL, spa, 0, &blk,
5534 abd, psize, NULL, NULL, ZIO_PRIORITY_SYNC_WRITE,
5535 ZIO_FLAG_CANFAIL | ZIO_FLAG_INDUCE_DAMAGE, NULL));
5539 (void) rw_unlock(&ztest_name_lock);
5540 umem_free(od, sizeof (ztest_od_t));
5548 ztest_scrub(ztest_ds_t *zd, uint64_t id)
5550 spa_t *spa = ztest_spa;
5552 (void) spa_scan(spa, POOL_SCAN_SCRUB);
5553 (void) poll(NULL, 0, 100); /* wait a moment, then force a restart */
5554 (void) spa_scan(spa, POOL_SCAN_SCRUB);
5558 * Change the guid for the pool.
5562 ztest_reguid(ztest_ds_t *zd, uint64_t id)
5564 spa_t *spa = ztest_spa;
5565 uint64_t orig, load;
5568 orig = spa_guid(spa);
5569 load = spa_load_guid(spa);
5571 (void) rw_wrlock(&ztest_name_lock);
5572 error = spa_change_guid(spa);
5573 (void) rw_unlock(&ztest_name_lock);
5578 if (ztest_opts.zo_verbose >= 4) {
5579 (void) printf("Changed guid old %llu -> %llu\n",
5580 (u_longlong_t)orig, (u_longlong_t)spa_guid(spa));
5583 VERIFY3U(orig, !=, spa_guid(spa));
5584 VERIFY3U(load, ==, spa_load_guid(spa));
5588 * Rename the pool to a different name and then rename it back.
5592 ztest_spa_rename(ztest_ds_t *zd, uint64_t id)
5594 char *oldname, *newname;
5597 (void) rw_wrlock(&ztest_name_lock);
5599 oldname = ztest_opts.zo_pool;
5600 newname = umem_alloc(strlen(oldname) + 5, UMEM_NOFAIL);
5601 (void) strcpy(newname, oldname);
5602 (void) strcat(newname, "_tmp");
5607 VERIFY3U(0, ==, spa_rename(oldname, newname));
5610 * Try to open it under the old name, which shouldn't exist
5612 VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
5615 * Open it under the new name and make sure it's still the same spa_t.
5617 VERIFY3U(0, ==, spa_open(newname, &spa, FTAG));
5619 ASSERT(spa == ztest_spa);
5620 spa_close(spa, FTAG);
5623 * Rename it back to the original
5625 VERIFY3U(0, ==, spa_rename(newname, oldname));
5628 * Make sure it can still be opened
5630 VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG));
5632 ASSERT(spa == ztest_spa);
5633 spa_close(spa, FTAG);
5635 umem_free(newname, strlen(newname) + 1);
5637 (void) rw_unlock(&ztest_name_lock);
5641 ztest_fletcher(ztest_ds_t *zd, uint64_t id)
5643 hrtime_t end = gethrtime() + NANOSEC;
5645 while (gethrtime() <= end) {
5646 int run_count = 100;
5652 zio_cksum_t zc_ref_byteswap;
5654 size = ztest_random_blocksize();
5655 buf = umem_alloc(size, UMEM_NOFAIL);
5657 for (i = 0, ptr = buf; i < size / sizeof (*ptr); i++, ptr++)
5658 *ptr = ztest_random(UINT_MAX);
5660 VERIFY0(fletcher_4_impl_set("scalar"));
5661 fletcher_4_native(buf, size, NULL, &zc_ref);
5662 fletcher_4_byteswap(buf, size, NULL, &zc_ref_byteswap);
5664 VERIFY0(fletcher_4_impl_set("cycle"));
5665 while (run_count-- > 0) {
5667 zio_cksum_t zc_byteswap;
5669 fletcher_4_byteswap(buf, size, NULL, &zc_byteswap);
5670 fletcher_4_native(buf, size, NULL, &zc);
5672 VERIFY0(bcmp(&zc, &zc_ref, sizeof (zc)));
5673 VERIFY0(bcmp(&zc_byteswap, &zc_ref_byteswap,
5674 sizeof (zc_byteswap)));
5677 umem_free(buf, size);
5682 ztest_fletcher_incr(ztest_ds_t *zd, uint64_t id)
5689 zio_cksum_t zc_ref_bswap;
5691 hrtime_t end = gethrtime() + NANOSEC;
5693 while (gethrtime() <= end) {
5694 int run_count = 100;
5696 size = ztest_random_blocksize();
5697 buf = umem_alloc(size, UMEM_NOFAIL);
5699 for (i = 0, ptr = buf; i < size / sizeof (*ptr); i++, ptr++)
5700 *ptr = ztest_random(UINT_MAX);
5702 VERIFY0(fletcher_4_impl_set("scalar"));
5703 fletcher_4_native(buf, size, NULL, &zc_ref);
5704 fletcher_4_byteswap(buf, size, NULL, &zc_ref_bswap);
5706 VERIFY0(fletcher_4_impl_set("cycle"));
5708 while (run_count-- > 0) {
5710 zio_cksum_t zc_bswap;
5713 ZIO_SET_CHECKSUM(&zc, 0, 0, 0, 0);
5714 ZIO_SET_CHECKSUM(&zc_bswap, 0, 0, 0, 0);
5716 while (pos < size) {
5717 size_t inc = 64 * ztest_random(size / 67);
5718 /* sometimes add few bytes to test non-simd */
5719 if (ztest_random(100) < 10)
5720 inc += P2ALIGN(ztest_random(64),
5723 if (inc > (size - pos))
5726 fletcher_4_incremental_native(buf + pos, inc,
5728 fletcher_4_incremental_byteswap(buf + pos, inc,
5734 VERIFY3U(pos, ==, size);
5736 VERIFY(ZIO_CHECKSUM_EQUAL(zc, zc_ref));
5737 VERIFY(ZIO_CHECKSUM_EQUAL(zc_bswap, zc_ref_bswap));
5740 * verify if incremental on the whole buffer is
5741 * equivalent to non-incremental version
5743 ZIO_SET_CHECKSUM(&zc, 0, 0, 0, 0);
5744 ZIO_SET_CHECKSUM(&zc_bswap, 0, 0, 0, 0);
5746 fletcher_4_incremental_native(buf, size, &zc);
5747 fletcher_4_incremental_byteswap(buf, size, &zc_bswap);
5749 VERIFY(ZIO_CHECKSUM_EQUAL(zc, zc_ref));
5750 VERIFY(ZIO_CHECKSUM_EQUAL(zc_bswap, zc_ref_bswap));
5753 umem_free(buf, size);
5758 ztest_check_path(char *path)
5761 /* return true on success */
5762 return (!stat(path, &s));
5766 ztest_get_zdb_bin(char *bin, int len)
5770 * Try to use ZDB_PATH and in-tree zdb path. If not successful, just
5771 * let popen to search through PATH.
5773 if ((zdb_path = getenv("ZDB_PATH"))) {
5774 strlcpy(bin, zdb_path, len); /* In env */
5775 if (!ztest_check_path(bin)) {
5776 ztest_dump_core = 0;
5777 fatal(1, "invalid ZDB_PATH '%s'", bin);
5782 VERIFY(realpath(getexecname(), bin) != NULL);
5783 if (strstr(bin, "/ztest/")) {
5784 strstr(bin, "/ztest/")[0] = '\0'; /* In-tree */
5785 strcat(bin, "/zdb/zdb");
5786 if (ztest_check_path(bin))
5793 * Verify pool integrity by running zdb.
5796 ztest_run_zdb(char *pool)
5802 const int len = MAXPATHLEN + MAXNAMELEN + 20;
5805 bin = umem_alloc(len, UMEM_NOFAIL);
5806 zdb = umem_alloc(len, UMEM_NOFAIL);
5807 zbuf = umem_alloc(1024, UMEM_NOFAIL);
5809 ztest_get_zdb_bin(bin, len);
5812 "%s -bcc%s%s -d -U %s %s",
5814 ztest_opts.zo_verbose >= 3 ? "s" : "",
5815 ztest_opts.zo_verbose >= 4 ? "v" : "",
5819 if (ztest_opts.zo_verbose >= 5)
5820 (void) printf("Executing %s\n", strstr(zdb, "zdb "));
5822 fp = popen(zdb, "r");
5824 while (fgets(zbuf, 1024, fp) != NULL)
5825 if (ztest_opts.zo_verbose >= 3)
5826 (void) printf("%s", zbuf);
5828 status = pclose(fp);
5833 ztest_dump_core = 0;
5834 if (WIFEXITED(status))
5835 fatal(0, "'%s' exit code %d", zdb, WEXITSTATUS(status));
5837 fatal(0, "'%s' died with signal %d", zdb, WTERMSIG(status));
5839 umem_free(bin, len);
5840 umem_free(zdb, len);
5841 umem_free(zbuf, 1024);
5845 ztest_walk_pool_directory(char *header)
5849 if (ztest_opts.zo_verbose >= 6)
5850 (void) printf("%s\n", header);
5852 mutex_enter(&spa_namespace_lock);
5853 while ((spa = spa_next(spa)) != NULL)
5854 if (ztest_opts.zo_verbose >= 6)
5855 (void) printf("\t%s\n", spa_name(spa));
5856 mutex_exit(&spa_namespace_lock);
5860 ztest_spa_import_export(char *oldname, char *newname)
5862 nvlist_t *config, *newconfig;
5867 if (ztest_opts.zo_verbose >= 4) {
5868 (void) printf("import/export: old = %s, new = %s\n",
5873 * Clean up from previous runs.
5875 (void) spa_destroy(newname);
5878 * Get the pool's configuration and guid.
5880 VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG));
5883 * Kick off a scrub to tickle scrub/export races.
5885 if (ztest_random(2) == 0)
5886 (void) spa_scan(spa, POOL_SCAN_SCRUB);
5888 pool_guid = spa_guid(spa);
5889 spa_close(spa, FTAG);
5891 ztest_walk_pool_directory("pools before export");
5896 VERIFY3U(0, ==, spa_export(oldname, &config, B_FALSE, B_FALSE));
5898 ztest_walk_pool_directory("pools after export");
5903 newconfig = spa_tryimport(config);
5904 ASSERT(newconfig != NULL);
5905 nvlist_free(newconfig);
5908 * Import it under the new name.
5910 error = spa_import(newname, config, NULL, 0);
5912 dump_nvlist(config, 0);
5913 fatal(B_FALSE, "couldn't import pool %s as %s: error %u",
5914 oldname, newname, error);
5917 ztest_walk_pool_directory("pools after import");
5920 * Try to import it again -- should fail with EEXIST.
5922 VERIFY3U(EEXIST, ==, spa_import(newname, config, NULL, 0));
5925 * Try to import it under a different name -- should fail with EEXIST.
5927 VERIFY3U(EEXIST, ==, spa_import(oldname, config, NULL, 0));
5930 * Verify that the pool is no longer visible under the old name.
5932 VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
5935 * Verify that we can open and close the pool using the new name.
5937 VERIFY3U(0, ==, spa_open(newname, &spa, FTAG));
5938 ASSERT(pool_guid == spa_guid(spa));
5939 spa_close(spa, FTAG);
5941 nvlist_free(config);
5945 ztest_resume(spa_t *spa)
5947 if (spa_suspended(spa) && ztest_opts.zo_verbose >= 6)
5948 (void) printf("resuming from suspended state\n");
5949 spa_vdev_state_enter(spa, SCL_NONE);
5950 vdev_clear(spa, NULL);
5951 (void) spa_vdev_state_exit(spa, NULL, 0);
5952 (void) zio_resume(spa);
5956 ztest_resume_thread(void *arg)
5960 while (!ztest_exiting) {
5961 if (spa_suspended(spa))
5963 (void) poll(NULL, 0, 100);
5966 * Periodically change the zfs_compressed_arc_enabled setting.
5968 if (ztest_random(10) == 0)
5969 zfs_compressed_arc_enabled = ztest_random(2);
5972 * Periodically change the zfs_abd_scatter_enabled setting.
5974 if (ztest_random(10) == 0)
5975 zfs_abd_scatter_enabled = ztest_random(2);
5987 ztest_deadman_alarm(int sig)
5989 fatal(0, "failed to complete within %d seconds of deadline", GRACE);
5994 ztest_execute(int test, ztest_info_t *zi, uint64_t id)
5996 ztest_ds_t *zd = &ztest_ds[id % ztest_opts.zo_datasets];
5997 ztest_shared_callstate_t *zc = ZTEST_GET_SHARED_CALLSTATE(test);
5998 hrtime_t functime = gethrtime();
6001 for (i = 0; i < zi->zi_iters; i++)
6002 zi->zi_func(zd, id);
6004 functime = gethrtime() - functime;
6006 atomic_add_64(&zc->zc_count, 1);
6007 atomic_add_64(&zc->zc_time, functime);
6009 if (ztest_opts.zo_verbose >= 4)
6010 (void) printf("%6.2f sec in %s\n",
6011 (double)functime / NANOSEC, zi->zi_funcname);
6015 ztest_thread(void *arg)
6018 uint64_t id = (uintptr_t)arg;
6019 ztest_shared_t *zs = ztest_shared;
6023 ztest_shared_callstate_t *zc;
6025 while ((now = gethrtime()) < zs->zs_thread_stop) {
6027 * See if it's time to force a crash.
6029 if (now > zs->zs_thread_kill)
6033 * If we're getting ENOSPC with some regularity, stop.
6035 if (zs->zs_enospc_count > 10)
6039 * Pick a random function to execute.
6041 rand = ztest_random(ZTEST_FUNCS);
6042 zi = &ztest_info[rand];
6043 zc = ZTEST_GET_SHARED_CALLSTATE(rand);
6044 call_next = zc->zc_next;
6046 if (now >= call_next &&
6047 atomic_cas_64(&zc->zc_next, call_next, call_next +
6048 ztest_random(2 * zi->zi_interval[0] + 1)) == call_next) {
6049 ztest_execute(rand, zi, id);
6059 ztest_dataset_name(char *dsname, char *pool, int d)
6061 (void) snprintf(dsname, ZFS_MAX_DATASET_NAME_LEN, "%s/ds_%d", pool, d);
6065 ztest_dataset_destroy(int d)
6067 char name[ZFS_MAX_DATASET_NAME_LEN];
6070 ztest_dataset_name(name, ztest_opts.zo_pool, d);
6072 if (ztest_opts.zo_verbose >= 3)
6073 (void) printf("Destroying %s to free up space\n", name);
6076 * Cleanup any non-standard clones and snapshots. In general,
6077 * ztest thread t operates on dataset (t % zopt_datasets),
6078 * so there may be more than one thing to clean up.
6080 for (t = d; t < ztest_opts.zo_threads;
6081 t += ztest_opts.zo_datasets)
6082 ztest_dsl_dataset_cleanup(name, t);
6084 (void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
6085 DS_FIND_SNAPSHOTS | DS_FIND_CHILDREN);
6089 ztest_dataset_dirobj_verify(ztest_ds_t *zd)
6091 uint64_t usedobjs, dirobjs, scratch;
6094 * ZTEST_DIROBJ is the object directory for the entire dataset.
6095 * Therefore, the number of objects in use should equal the
6096 * number of ZTEST_DIROBJ entries, +1 for ZTEST_DIROBJ itself.
6097 * If not, we have an object leak.
6099 * Note that we can only check this in ztest_dataset_open(),
6100 * when the open-context and syncing-context values agree.
6101 * That's because zap_count() returns the open-context value,
6102 * while dmu_objset_space() returns the rootbp fill count.
6104 VERIFY3U(0, ==, zap_count(zd->zd_os, ZTEST_DIROBJ, &dirobjs));
6105 dmu_objset_space(zd->zd_os, &scratch, &scratch, &usedobjs, &scratch);
6106 ASSERT3U(dirobjs + 1, ==, usedobjs);
6110 ztest_dataset_open(int d)
6112 ztest_ds_t *zd = &ztest_ds[d];
6113 uint64_t committed_seq = ZTEST_GET_SHARED_DS(d)->zd_seq;
6116 char name[ZFS_MAX_DATASET_NAME_LEN];
6119 ztest_dataset_name(name, ztest_opts.zo_pool, d);
6121 (void) rw_rdlock(&ztest_name_lock);
6123 error = ztest_dataset_create(name);
6124 if (error == ENOSPC) {
6125 (void) rw_unlock(&ztest_name_lock);
6126 ztest_record_enospc(FTAG);
6129 ASSERT(error == 0 || error == EEXIST);
6131 VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, zd, &os));
6132 (void) rw_unlock(&ztest_name_lock);
6134 ztest_zd_init(zd, ZTEST_GET_SHARED_DS(d), os);
6136 zilog = zd->zd_zilog;
6138 if (zilog->zl_header->zh_claim_lr_seq != 0 &&
6139 zilog->zl_header->zh_claim_lr_seq < committed_seq)
6140 fatal(0, "missing log records: claimed %llu < committed %llu",
6141 zilog->zl_header->zh_claim_lr_seq, committed_seq);
6143 ztest_dataset_dirobj_verify(zd);
6145 zil_replay(os, zd, ztest_replay_vector);
6147 ztest_dataset_dirobj_verify(zd);
6149 if (ztest_opts.zo_verbose >= 6)
6150 (void) printf("%s replay %llu blocks, %llu records, seq %llu\n",
6152 (u_longlong_t)zilog->zl_parse_blk_count,
6153 (u_longlong_t)zilog->zl_parse_lr_count,
6154 (u_longlong_t)zilog->zl_replaying_seq);
6156 zilog = zil_open(os, ztest_get_data);
6158 if (zilog->zl_replaying_seq != 0 &&
6159 zilog->zl_replaying_seq < committed_seq)
6160 fatal(0, "missing log records: replayed %llu < committed %llu",
6161 zilog->zl_replaying_seq, committed_seq);
6167 ztest_dataset_close(int d)
6169 ztest_ds_t *zd = &ztest_ds[d];
6171 zil_close(zd->zd_zilog);
6172 dmu_objset_disown(zd->zd_os, zd);
6178 * Kick off threads to run tests on all datasets in parallel.
6181 ztest_run(ztest_shared_t *zs)
6186 kthread_t *resume_thread;
6191 ztest_exiting = B_FALSE;
6194 * Initialize parent/child shared state.
6196 mutex_init(&ztest_vdev_lock, NULL, MUTEX_DEFAULT, NULL);
6197 VERIFY(rwlock_init(&ztest_name_lock, USYNC_THREAD, NULL) == 0);
6199 zs->zs_thread_start = gethrtime();
6200 zs->zs_thread_stop =
6201 zs->zs_thread_start + ztest_opts.zo_passtime * NANOSEC;
6202 zs->zs_thread_stop = MIN(zs->zs_thread_stop, zs->zs_proc_stop);
6203 zs->zs_thread_kill = zs->zs_thread_stop;
6204 if (ztest_random(100) < ztest_opts.zo_killrate) {
6205 zs->zs_thread_kill -=
6206 ztest_random(ztest_opts.zo_passtime * NANOSEC);
6209 mutex_init(&zcl.zcl_callbacks_lock, NULL, MUTEX_DEFAULT, NULL);
6211 list_create(&zcl.zcl_callbacks, sizeof (ztest_cb_data_t),
6212 offsetof(ztest_cb_data_t, zcd_node));
6217 kernel_init(FREAD | FWRITE);
6218 VERIFY0(spa_open(ztest_opts.zo_pool, &spa, FTAG));
6219 spa->spa_debug = B_TRUE;
6220 metaslab_preload_limit = ztest_random(20) + 1;
6223 VERIFY0(dmu_objset_own(ztest_opts.zo_pool,
6224 DMU_OST_ANY, B_TRUE, FTAG, &os));
6225 zs->zs_guid = dmu_objset_fsid_guid(os);
6226 dmu_objset_disown(os, FTAG);
6228 spa->spa_dedup_ditto = 2 * ZIO_DEDUPDITTO_MIN;
6231 * We don't expect the pool to suspend unless maxfaults == 0,
6232 * in which case ztest_fault_inject() temporarily takes away
6233 * the only valid replica.
6235 if (MAXFAULTS() == 0)
6236 spa->spa_failmode = ZIO_FAILURE_MODE_WAIT;
6238 spa->spa_failmode = ZIO_FAILURE_MODE_PANIC;
6241 * Create a thread to periodically resume suspended I/O.
6243 VERIFY3P((resume_thread = zk_thread_create(NULL, 0,
6244 (thread_func_t)ztest_resume_thread, spa, TS_RUN, NULL, 0, 0,
6245 PTHREAD_CREATE_JOINABLE)), !=, NULL);
6249 * Set a deadman alarm to abort() if we hang.
6251 signal(SIGALRM, ztest_deadman_alarm);
6252 alarm((zs->zs_thread_stop - zs->zs_thread_start) / NANOSEC + GRACE);
6256 * Verify that we can safely inquire about about any object,
6257 * whether it's allocated or not. To make it interesting,
6258 * we probe a 5-wide window around each power of two.
6259 * This hits all edge cases, including zero and the max.
6261 for (t = 0; t < 64; t++) {
6262 for (d = -5; d <= 5; d++) {
6263 error = dmu_object_info(spa->spa_meta_objset,
6264 (1ULL << t) + d, NULL);
6265 ASSERT(error == 0 || error == ENOENT ||
6271 * If we got any ENOSPC errors on the previous run, destroy something.
6273 if (zs->zs_enospc_count != 0) {
6274 int d = ztest_random(ztest_opts.zo_datasets);
6275 ztest_dataset_destroy(d);
6277 zs->zs_enospc_count = 0;
6279 tid = umem_zalloc(ztest_opts.zo_threads * sizeof (kt_did_t),
6282 if (ztest_opts.zo_verbose >= 4)
6283 (void) printf("starting main threads...\n");
6286 * Kick off all the tests that run in parallel.
6288 for (t = 0; t < ztest_opts.zo_threads; t++) {
6291 if (t < ztest_opts.zo_datasets &&
6292 ztest_dataset_open(t) != 0) {
6294 ztest_opts.zo_threads * sizeof (kt_did_t));
6298 VERIFY3P(thread = zk_thread_create(NULL, 0,
6299 (thread_func_t)ztest_thread,
6300 (void *)(uintptr_t)t, TS_RUN, NULL, 0, 0,
6301 PTHREAD_CREATE_JOINABLE), !=, NULL);
6302 tid[t] = thread->t_tid;
6306 * Wait for all of the tests to complete. We go in reverse order
6307 * so we don't close datasets while threads are still using them.
6309 for (t = ztest_opts.zo_threads - 1; t >= 0; t--) {
6310 thread_join(tid[t]);
6311 if (t < ztest_opts.zo_datasets)
6312 ztest_dataset_close(t);
6315 txg_wait_synced(spa_get_dsl(spa), 0);
6317 zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(spa));
6318 zs->zs_space = metaslab_class_get_space(spa_normal_class(spa));
6320 umem_free(tid, ztest_opts.zo_threads * sizeof (kt_did_t));
6322 /* Kill the resume thread */
6323 ztest_exiting = B_TRUE;
6324 thread_join(resume_thread->t_tid);
6328 * Right before closing the pool, kick off a bunch of async I/O;
6329 * spa_close() should wait for it to complete.
6331 for (object = 1; object < 50; object++) {
6332 dmu_prefetch(spa->spa_meta_objset, object, 0, 0, 1ULL << 20,
6333 ZIO_PRIORITY_SYNC_READ);
6336 /* Verify that at least one commit cb was called in a timely fashion */
6337 if (zc_cb_counter >= ZTEST_COMMIT_CB_MIN_REG)
6338 VERIFY0(zc_min_txg_delay);
6340 spa_close(spa, FTAG);
6343 * Verify that we can loop over all pools.
6345 mutex_enter(&spa_namespace_lock);
6346 for (spa = spa_next(NULL); spa != NULL; spa = spa_next(spa))
6347 if (ztest_opts.zo_verbose > 3)
6348 (void) printf("spa_next: found %s\n", spa_name(spa));
6349 mutex_exit(&spa_namespace_lock);
6352 * Verify that we can export the pool and reimport it under a
6355 if (ztest_random(2) == 0) {
6356 char name[ZFS_MAX_DATASET_NAME_LEN];
6357 (void) snprintf(name, sizeof (name), "%s_import",
6358 ztest_opts.zo_pool);
6359 ztest_spa_import_export(ztest_opts.zo_pool, name);
6360 ztest_spa_import_export(name, ztest_opts.zo_pool);
6365 list_destroy(&zcl.zcl_callbacks);
6366 mutex_destroy(&zcl.zcl_callbacks_lock);
6367 (void) rwlock_destroy(&ztest_name_lock);
6368 mutex_destroy(&ztest_vdev_lock);
6374 ztest_ds_t *zd = &ztest_ds[0];
6378 if (ztest_opts.zo_verbose >= 3)
6379 (void) printf("testing spa_freeze()...\n");
6381 kernel_init(FREAD | FWRITE);
6382 VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
6383 VERIFY3U(0, ==, ztest_dataset_open(0));
6384 spa->spa_debug = B_TRUE;
6388 * Force the first log block to be transactionally allocated.
6389 * We have to do this before we freeze the pool -- otherwise
6390 * the log chain won't be anchored.
6392 while (BP_IS_HOLE(&zd->zd_zilog->zl_header->zh_log)) {
6393 ztest_dmu_object_alloc_free(zd, 0);
6394 zil_commit(zd->zd_zilog, 0);
6397 txg_wait_synced(spa_get_dsl(spa), 0);
6400 * Freeze the pool. This stops spa_sync() from doing anything,
6401 * so that the only way to record changes from now on is the ZIL.
6406 * Because it is hard to predict how much space a write will actually
6407 * require beforehand, we leave ourselves some fudge space to write over
6410 uint64_t capacity = metaslab_class_get_space(spa_normal_class(spa)) / 2;
6413 * Run tests that generate log records but don't alter the pool config
6414 * or depend on DSL sync tasks (snapshots, objset create/destroy, etc).
6415 * We do a txg_wait_synced() after each iteration to force the txg
6416 * to increase well beyond the last synced value in the uberblock.
6417 * The ZIL should be OK with that.
6419 * Run a random number of times less than zo_maxloops and ensure we do
6420 * not run out of space on the pool.
6422 while (ztest_random(10) != 0 &&
6423 numloops++ < ztest_opts.zo_maxloops &&
6424 metaslab_class_get_alloc(spa_normal_class(spa)) < capacity) {
6426 ztest_od_init(&od, 0, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0, 0);
6427 VERIFY0(ztest_object_init(zd, &od, sizeof (od), B_FALSE));
6428 ztest_io(zd, od.od_object,
6429 ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
6430 txg_wait_synced(spa_get_dsl(spa), 0);
6434 * Commit all of the changes we just generated.
6436 zil_commit(zd->zd_zilog, 0);
6437 txg_wait_synced(spa_get_dsl(spa), 0);
6440 * Close our dataset and close the pool.
6442 ztest_dataset_close(0);
6443 spa_close(spa, FTAG);
6447 * Open and close the pool and dataset to induce log replay.
6449 kernel_init(FREAD | FWRITE);
6450 VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
6451 ASSERT(spa_freeze_txg(spa) == UINT64_MAX);
6452 VERIFY3U(0, ==, ztest_dataset_open(0));
6453 ztest_dataset_close(0);
6455 spa->spa_debug = B_TRUE;
6457 txg_wait_synced(spa_get_dsl(spa), 0);
6458 ztest_reguid(NULL, 0);
6460 spa_close(spa, FTAG);
6465 print_time(hrtime_t t, char *timebuf)
6467 hrtime_t s = t / NANOSEC;
6468 hrtime_t m = s / 60;
6469 hrtime_t h = m / 60;
6470 hrtime_t d = h / 24;
6479 (void) sprintf(timebuf,
6480 "%llud%02lluh%02llum%02llus", d, h, m, s);
6482 (void) sprintf(timebuf, "%lluh%02llum%02llus", h, m, s);
6484 (void) sprintf(timebuf, "%llum%02llus", m, s);
6486 (void) sprintf(timebuf, "%llus", s);
6490 make_random_props(void)
6494 VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
6495 if (ztest_random(2) == 0)
6497 VERIFY(nvlist_add_uint64(props, "autoreplace", 1) == 0);
6503 * Create a storage pool with the given name and initial vdev size.
6504 * Then test spa_freeze() functionality.
6507 ztest_init(ztest_shared_t *zs)
6510 nvlist_t *nvroot, *props;
6513 mutex_init(&ztest_vdev_lock, NULL, MUTEX_DEFAULT, NULL);
6514 VERIFY(rwlock_init(&ztest_name_lock, USYNC_THREAD, NULL) == 0);
6516 kernel_init(FREAD | FWRITE);
6519 * Create the storage pool.
6521 (void) spa_destroy(ztest_opts.zo_pool);
6522 ztest_shared->zs_vdev_next_leaf = 0;
6524 zs->zs_mirrors = ztest_opts.zo_mirrors;
6525 nvroot = make_vdev_root(NULL, NULL, NULL, ztest_opts.zo_vdev_size, 0,
6526 0, ztest_opts.zo_raidz, zs->zs_mirrors, 1);
6527 props = make_random_props();
6528 for (i = 0; i < SPA_FEATURES; i++) {
6530 VERIFY3S(-1, !=, asprintf(&buf, "feature@%s",
6531 spa_feature_table[i].fi_uname));
6532 VERIFY3U(0, ==, nvlist_add_uint64(props, buf, 0));
6535 VERIFY3U(0, ==, spa_create(ztest_opts.zo_pool, nvroot, props, NULL));
6536 nvlist_free(nvroot);
6539 VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
6540 zs->zs_metaslab_sz =
6541 1ULL << spa->spa_root_vdev->vdev_child[0]->vdev_ms_shift;
6542 spa_close(spa, FTAG);
6546 ztest_run_zdb(ztest_opts.zo_pool);
6550 ztest_run_zdb(ztest_opts.zo_pool);
6552 (void) rwlock_destroy(&ztest_name_lock);
6553 mutex_destroy(&ztest_vdev_lock);
6559 static char ztest_name_data[] = "/tmp/ztest.data.XXXXXX";
6561 ztest_fd_data = mkstemp(ztest_name_data);
6562 ASSERT3S(ztest_fd_data, >=, 0);
6563 (void) unlink(ztest_name_data);
6567 shared_data_size(ztest_shared_hdr_t *hdr)
6571 size = hdr->zh_hdr_size;
6572 size += hdr->zh_opts_size;
6573 size += hdr->zh_size;
6574 size += hdr->zh_stats_size * hdr->zh_stats_count;
6575 size += hdr->zh_ds_size * hdr->zh_ds_count;
6584 ztest_shared_hdr_t *hdr;
6586 hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
6587 PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
6588 ASSERT(hdr != MAP_FAILED);
6590 VERIFY3U(0, ==, ftruncate(ztest_fd_data, sizeof (ztest_shared_hdr_t)));
6592 hdr->zh_hdr_size = sizeof (ztest_shared_hdr_t);
6593 hdr->zh_opts_size = sizeof (ztest_shared_opts_t);
6594 hdr->zh_size = sizeof (ztest_shared_t);
6595 hdr->zh_stats_size = sizeof (ztest_shared_callstate_t);
6596 hdr->zh_stats_count = ZTEST_FUNCS;
6597 hdr->zh_ds_size = sizeof (ztest_shared_ds_t);
6598 hdr->zh_ds_count = ztest_opts.zo_datasets;
6600 size = shared_data_size(hdr);
6601 VERIFY3U(0, ==, ftruncate(ztest_fd_data, size));
6603 (void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
6610 ztest_shared_hdr_t *hdr;
6613 hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
6614 PROT_READ, MAP_SHARED, ztest_fd_data, 0);
6615 ASSERT(hdr != MAP_FAILED);
6617 size = shared_data_size(hdr);
6619 (void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
6620 hdr = ztest_shared_hdr = (void *)mmap(0, P2ROUNDUP(size, getpagesize()),
6621 PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
6622 ASSERT(hdr != MAP_FAILED);
6623 buf = (uint8_t *)hdr;
6625 offset = hdr->zh_hdr_size;
6626 ztest_shared_opts = (void *)&buf[offset];
6627 offset += hdr->zh_opts_size;
6628 ztest_shared = (void *)&buf[offset];
6629 offset += hdr->zh_size;
6630 ztest_shared_callstate = (void *)&buf[offset];
6631 offset += hdr->zh_stats_size * hdr->zh_stats_count;
6632 ztest_shared_ds = (void *)&buf[offset];
6636 exec_child(char *cmd, char *libpath, boolean_t ignorekill, int *statusp)
6640 char *cmdbuf = NULL;
6645 cmdbuf = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
6646 (void) strlcpy(cmdbuf, getexecname(), MAXPATHLEN);
6651 fatal(1, "fork failed");
6653 if (pid == 0) { /* child */
6654 char *emptyargv[2] = { cmd, NULL };
6655 char fd_data_str[12];
6657 struct rlimit rl = { 1024, 1024 };
6658 (void) setrlimit(RLIMIT_NOFILE, &rl);
6660 (void) close(ztest_fd_rand);
6661 VERIFY(11 >= snprintf(fd_data_str, 12, "%d", ztest_fd_data));
6662 VERIFY(0 == setenv("ZTEST_FD_DATA", fd_data_str, 1));
6664 (void) enable_extended_FILE_stdio(-1, -1);
6665 if (libpath != NULL)
6666 VERIFY(0 == setenv("LD_LIBRARY_PATH", libpath, 1));
6667 (void) execv(cmd, emptyargv);
6668 ztest_dump_core = B_FALSE;
6669 fatal(B_TRUE, "exec failed: %s", cmd);
6672 if (cmdbuf != NULL) {
6673 umem_free(cmdbuf, MAXPATHLEN);
6677 while (waitpid(pid, &status, 0) != pid)
6679 if (statusp != NULL)
6682 if (WIFEXITED(status)) {
6683 if (WEXITSTATUS(status) != 0) {
6684 (void) fprintf(stderr, "child exited with code %d\n",
6685 WEXITSTATUS(status));
6689 } else if (WIFSIGNALED(status)) {
6690 if (!ignorekill || WTERMSIG(status) != SIGKILL) {
6691 (void) fprintf(stderr, "child died with signal %d\n",
6697 (void) fprintf(stderr, "something strange happened to child\n");
6704 ztest_run_init(void)
6708 ztest_shared_t *zs = ztest_shared;
6710 ASSERT(ztest_opts.zo_init != 0);
6713 * Blow away any existing copy of zpool.cache
6715 (void) remove(spa_config_path);
6718 * Create and initialize our storage pool.
6720 for (i = 1; i <= ztest_opts.zo_init; i++) {
6721 bzero(zs, sizeof (ztest_shared_t));
6722 if (ztest_opts.zo_verbose >= 3 &&
6723 ztest_opts.zo_init != 1) {
6724 (void) printf("ztest_init(), pass %d\n", i);
6731 main(int argc, char **argv)
6739 ztest_shared_callstate_t *zc;
6746 char *fd_data_str = getenv("ZTEST_FD_DATA");
6747 struct sigaction action;
6749 (void) setvbuf(stdout, NULL, _IOLBF, 0);
6751 dprintf_setup(&argc, argv);
6753 action.sa_handler = sig_handler;
6754 sigemptyset(&action.sa_mask);
6755 action.sa_flags = 0;
6757 if (sigaction(SIGSEGV, &action, NULL) < 0) {
6758 (void) fprintf(stderr, "ztest: cannot catch SIGSEGV: %s.\n",
6763 if (sigaction(SIGABRT, &action, NULL) < 0) {
6764 (void) fprintf(stderr, "ztest: cannot catch SIGABRT: %s.\n",
6769 ztest_fd_rand = open("/dev/urandom", O_RDONLY);
6770 ASSERT3S(ztest_fd_rand, >=, 0);
6773 process_options(argc, argv);
6778 bcopy(&ztest_opts, ztest_shared_opts,
6779 sizeof (*ztest_shared_opts));
6781 ztest_fd_data = atoi(fd_data_str);
6783 bcopy(ztest_shared_opts, &ztest_opts, sizeof (ztest_opts));
6785 ASSERT3U(ztest_opts.zo_datasets, ==, ztest_shared_hdr->zh_ds_count);
6787 /* Override location of zpool.cache */
6788 VERIFY(asprintf((char **)&spa_config_path, "%s/zpool.cache",
6789 ztest_opts.zo_dir) != -1);
6791 ztest_ds = umem_alloc(ztest_opts.zo_datasets * sizeof (ztest_ds_t),
6796 metaslab_gang_bang = ztest_opts.zo_metaslab_gang_bang;
6797 metaslab_df_alloc_threshold =
6798 zs->zs_metaslab_df_alloc_threshold;
6807 hasalt = (strlen(ztest_opts.zo_alt_ztest) != 0);
6809 if (ztest_opts.zo_verbose >= 1) {
6810 (void) printf("%llu vdevs, %d datasets, %d threads,"
6811 " %llu seconds...\n",
6812 (u_longlong_t)ztest_opts.zo_vdevs,
6813 ztest_opts.zo_datasets,
6814 ztest_opts.zo_threads,
6815 (u_longlong_t)ztest_opts.zo_time);
6818 cmd = umem_alloc(MAXNAMELEN, UMEM_NOFAIL);
6819 (void) strlcpy(cmd, getexecname(), MAXNAMELEN);
6821 zs->zs_do_init = B_TRUE;
6822 if (strlen(ztest_opts.zo_alt_ztest) != 0) {
6823 if (ztest_opts.zo_verbose >= 1) {
6824 (void) printf("Executing older ztest for "
6825 "initialization: %s\n", ztest_opts.zo_alt_ztest);
6827 VERIFY(!exec_child(ztest_opts.zo_alt_ztest,
6828 ztest_opts.zo_alt_libpath, B_FALSE, NULL));
6830 VERIFY(!exec_child(NULL, NULL, B_FALSE, NULL));
6832 zs->zs_do_init = B_FALSE;
6834 zs->zs_proc_start = gethrtime();
6835 zs->zs_proc_stop = zs->zs_proc_start + ztest_opts.zo_time * NANOSEC;
6837 for (f = 0; f < ZTEST_FUNCS; f++) {
6838 zi = &ztest_info[f];
6839 zc = ZTEST_GET_SHARED_CALLSTATE(f);
6840 if (zs->zs_proc_start + zi->zi_interval[0] > zs->zs_proc_stop)
6841 zc->zc_next = UINT64_MAX;
6843 zc->zc_next = zs->zs_proc_start +
6844 ztest_random(2 * zi->zi_interval[0] + 1);
6848 * Run the tests in a loop. These tests include fault injection
6849 * to verify that self-healing data works, and forced crashes
6850 * to verify that we never lose on-disk consistency.
6852 while (gethrtime() < zs->zs_proc_stop) {
6857 * Initialize the workload counters for each function.
6859 for (f = 0; f < ZTEST_FUNCS; f++) {
6860 zc = ZTEST_GET_SHARED_CALLSTATE(f);
6865 /* Set the allocation switch size */
6866 zs->zs_metaslab_df_alloc_threshold =
6867 ztest_random(zs->zs_metaslab_sz / 4) + 1;
6869 if (!hasalt || ztest_random(2) == 0) {
6870 if (hasalt && ztest_opts.zo_verbose >= 1) {
6871 (void) printf("Executing newer ztest: %s\n",
6875 killed = exec_child(cmd, NULL, B_TRUE, &status);
6877 if (hasalt && ztest_opts.zo_verbose >= 1) {
6878 (void) printf("Executing older ztest: %s\n",
6879 ztest_opts.zo_alt_ztest);
6882 killed = exec_child(ztest_opts.zo_alt_ztest,
6883 ztest_opts.zo_alt_libpath, B_TRUE, &status);
6890 if (ztest_opts.zo_verbose >= 1) {
6891 hrtime_t now = gethrtime();
6893 now = MIN(now, zs->zs_proc_stop);
6894 print_time(zs->zs_proc_stop - now, timebuf);
6895 nicenum(zs->zs_space, numbuf);
6897 (void) printf("Pass %3d, %8s, %3llu ENOSPC, "
6898 "%4.1f%% of %5s used, %3.0f%% done, %8s to go\n",
6900 WIFEXITED(status) ? "Complete" : "SIGKILL",
6901 (u_longlong_t)zs->zs_enospc_count,
6902 100.0 * zs->zs_alloc / zs->zs_space,
6904 100.0 * (now - zs->zs_proc_start) /
6905 (ztest_opts.zo_time * NANOSEC), timebuf);
6908 if (ztest_opts.zo_verbose >= 2) {
6909 (void) printf("\nWorkload summary:\n\n");
6910 (void) printf("%7s %9s %s\n",
6911 "Calls", "Time", "Function");
6912 (void) printf("%7s %9s %s\n",
6913 "-----", "----", "--------");
6914 for (f = 0; f < ZTEST_FUNCS; f++) {
6915 zi = &ztest_info[f];
6916 zc = ZTEST_GET_SHARED_CALLSTATE(f);
6917 print_time(zc->zc_time, timebuf);
6918 (void) printf("%7llu %9s %s\n",
6919 (u_longlong_t)zc->zc_count, timebuf,
6922 (void) printf("\n");
6926 * It's possible that we killed a child during a rename test,
6927 * in which case we'll have a 'ztest_tmp' pool lying around
6928 * instead of 'ztest'. Do a blind rename in case this happened.
6931 if (spa_open(ztest_opts.zo_pool, &spa, FTAG) == 0) {
6932 spa_close(spa, FTAG);
6934 char tmpname[ZFS_MAX_DATASET_NAME_LEN];
6936 kernel_init(FREAD | FWRITE);
6937 (void) snprintf(tmpname, sizeof (tmpname), "%s_tmp",
6938 ztest_opts.zo_pool);
6939 (void) spa_rename(tmpname, ztest_opts.zo_pool);
6943 ztest_run_zdb(ztest_opts.zo_pool);
6946 if (ztest_opts.zo_verbose >= 1) {
6948 (void) printf("%d runs of older ztest: %s\n", older,
6949 ztest_opts.zo_alt_ztest);
6950 (void) printf("%d runs of newer ztest: %s\n", newer,
6953 (void) printf("%d killed, %d completed, %.0f%% kill rate\n",
6954 kills, iters - kills, (100.0 * kills) / MAX(1, iters));
6957 umem_free(cmd, MAXNAMELEN);