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>
118 #include <stdio_ext.h>
125 #include <sys/fs/zfs.h>
126 #include <zfs_fletcher.h>
127 #include <libnvpair.h>
129 #include <execinfo.h> /* for backtrace() */
132 static int ztest_fd_data = -1;
133 static int ztest_fd_rand = -1;
135 typedef struct ztest_shared_hdr {
136 uint64_t zh_hdr_size;
137 uint64_t zh_opts_size;
139 uint64_t zh_stats_size;
140 uint64_t zh_stats_count;
142 uint64_t zh_ds_count;
143 } ztest_shared_hdr_t;
145 static ztest_shared_hdr_t *ztest_shared_hdr;
147 typedef struct ztest_shared_opts {
148 char zo_pool[ZFS_MAX_DATASET_NAME_LEN];
149 char zo_dir[ZFS_MAX_DATASET_NAME_LEN];
150 char zo_alt_ztest[MAXNAMELEN];
151 char zo_alt_libpath[MAXNAMELEN];
153 uint64_t zo_vdevtime;
161 uint64_t zo_passtime;
162 uint64_t zo_killrate;
166 uint64_t zo_maxloops;
167 uint64_t zo_metaslab_gang_bang;
168 } ztest_shared_opts_t;
170 static const ztest_shared_opts_t ztest_opts_defaults = {
171 .zo_pool = { 'z', 't', 'e', 's', 't', '\0' },
172 .zo_dir = { '/', 't', 'm', 'p', '\0' },
173 .zo_alt_ztest = { '\0' },
174 .zo_alt_libpath = { '\0' },
176 .zo_ashift = SPA_MINBLOCKSHIFT,
179 .zo_raidz_parity = 1,
180 .zo_vdev_size = SPA_MINDEVSIZE * 2,
183 .zo_passtime = 60, /* 60 seconds */
184 .zo_killrate = 70, /* 70% kill rate */
187 .zo_time = 300, /* 5 minutes */
188 .zo_maxloops = 50, /* max loops during spa_freeze() */
189 .zo_metaslab_gang_bang = 32 << 10
192 extern uint64_t metaslab_gang_bang;
193 extern uint64_t metaslab_df_alloc_threshold;
194 extern int metaslab_preload_limit;
195 extern boolean_t zfs_compressed_arc_enabled;
197 static ztest_shared_opts_t *ztest_shared_opts;
198 static ztest_shared_opts_t ztest_opts;
200 typedef struct ztest_shared_ds {
204 static ztest_shared_ds_t *ztest_shared_ds;
205 #define ZTEST_GET_SHARED_DS(d) (&ztest_shared_ds[d])
207 #define BT_MAGIC 0x123456789abcdefULL
208 #define MAXFAULTS() \
209 (MAX(zs->zs_mirrors, 1) * (ztest_opts.zo_raidz_parity + 1) - 1)
213 ZTEST_IO_WRITE_PATTERN,
214 ZTEST_IO_WRITE_ZEROES,
221 typedef struct ztest_block_tag {
225 uint64_t bt_dnodesize;
232 typedef struct bufwad {
250 #define ZTEST_RANGE_LOCKS 64
251 #define ZTEST_OBJECT_LOCKS 64
254 * Object descriptor. Used as a template for object lookup/create/remove.
256 typedef struct ztest_od {
259 dmu_object_type_t od_type;
260 dmu_object_type_t od_crtype;
261 uint64_t od_blocksize;
262 uint64_t od_crblocksize;
263 uint64_t od_crdnodesize;
266 char od_name[ZFS_MAX_DATASET_NAME_LEN];
272 typedef struct ztest_ds {
273 ztest_shared_ds_t *zd_shared;
275 rwlock_t zd_zilog_lock;
277 ztest_od_t *zd_od; /* debugging aid */
278 char zd_name[ZFS_MAX_DATASET_NAME_LEN];
279 kmutex_t zd_dirobj_lock;
280 rll_t zd_object_lock[ZTEST_OBJECT_LOCKS];
281 zll_t zd_range_lock[ZTEST_RANGE_LOCKS];
285 * Per-iteration state.
287 typedef void ztest_func_t(ztest_ds_t *zd, uint64_t id);
289 typedef struct ztest_info {
290 ztest_func_t *zi_func; /* test function */
291 uint64_t zi_iters; /* iterations per execution */
292 uint64_t *zi_interval; /* execute every <interval> seconds */
293 const char *zi_funcname; /* name of test function */
296 typedef struct ztest_shared_callstate {
297 uint64_t zc_count; /* per-pass count */
298 uint64_t zc_time; /* per-pass time */
299 uint64_t zc_next; /* next time to call this function */
300 } ztest_shared_callstate_t;
302 static ztest_shared_callstate_t *ztest_shared_callstate;
303 #define ZTEST_GET_SHARED_CALLSTATE(c) (&ztest_shared_callstate[c])
305 ztest_func_t ztest_dmu_read_write;
306 ztest_func_t ztest_dmu_write_parallel;
307 ztest_func_t ztest_dmu_object_alloc_free;
308 ztest_func_t ztest_dmu_commit_callbacks;
309 ztest_func_t ztest_zap;
310 ztest_func_t ztest_zap_parallel;
311 ztest_func_t ztest_zil_commit;
312 ztest_func_t ztest_zil_remount;
313 ztest_func_t ztest_dmu_read_write_zcopy;
314 ztest_func_t ztest_dmu_objset_create_destroy;
315 ztest_func_t ztest_dmu_prealloc;
316 ztest_func_t ztest_fzap;
317 ztest_func_t ztest_dmu_snapshot_create_destroy;
318 ztest_func_t ztest_dsl_prop_get_set;
319 ztest_func_t ztest_spa_prop_get_set;
320 ztest_func_t ztest_spa_create_destroy;
321 ztest_func_t ztest_fault_inject;
322 ztest_func_t ztest_ddt_repair;
323 ztest_func_t ztest_dmu_snapshot_hold;
324 ztest_func_t ztest_spa_rename;
325 ztest_func_t ztest_scrub;
326 ztest_func_t ztest_dsl_dataset_promote_busy;
327 ztest_func_t ztest_vdev_attach_detach;
328 ztest_func_t ztest_vdev_LUN_growth;
329 ztest_func_t ztest_vdev_add_remove;
330 ztest_func_t ztest_vdev_aux_add_remove;
331 ztest_func_t ztest_split_pool;
332 ztest_func_t ztest_reguid;
333 ztest_func_t ztest_spa_upgrade;
334 ztest_func_t ztest_fletcher;
335 ztest_func_t ztest_verify_dnode_bt;
337 uint64_t zopt_always = 0ULL * NANOSEC; /* all the time */
338 uint64_t zopt_incessant = 1ULL * NANOSEC / 10; /* every 1/10 second */
339 uint64_t zopt_often = 1ULL * NANOSEC; /* every second */
340 uint64_t zopt_sometimes = 10ULL * NANOSEC; /* every 10 seconds */
341 uint64_t zopt_rarely = 60ULL * NANOSEC; /* every 60 seconds */
343 #define ZTI_INIT(func, iters, interval) \
344 { .zi_func = (func), \
345 .zi_iters = (iters), \
346 .zi_interval = (interval), \
347 .zi_funcname = # func }
349 ztest_info_t ztest_info[] = {
350 ZTI_INIT(ztest_dmu_read_write, 1, &zopt_always),
351 ZTI_INIT(ztest_dmu_write_parallel, 10, &zopt_always),
352 ZTI_INIT(ztest_dmu_object_alloc_free, 1, &zopt_always),
353 ZTI_INIT(ztest_dmu_commit_callbacks, 1, &zopt_always),
354 ZTI_INIT(ztest_zap, 30, &zopt_always),
355 ZTI_INIT(ztest_zap_parallel, 100, &zopt_always),
356 ZTI_INIT(ztest_split_pool, 1, &zopt_always),
357 ZTI_INIT(ztest_zil_commit, 1, &zopt_incessant),
358 ZTI_INIT(ztest_zil_remount, 1, &zopt_sometimes),
359 ZTI_INIT(ztest_dmu_read_write_zcopy, 1, &zopt_often),
360 ZTI_INIT(ztest_dmu_objset_create_destroy, 1, &zopt_often),
361 ZTI_INIT(ztest_dsl_prop_get_set, 1, &zopt_often),
362 ZTI_INIT(ztest_spa_prop_get_set, 1, &zopt_sometimes),
364 ZTI_INIT(ztest_dmu_prealloc, 1, &zopt_sometimes),
366 ZTI_INIT(ztest_fzap, 1, &zopt_sometimes),
367 ZTI_INIT(ztest_dmu_snapshot_create_destroy, 1, &zopt_sometimes),
368 ZTI_INIT(ztest_spa_create_destroy, 1, &zopt_sometimes),
369 ZTI_INIT(ztest_fault_inject, 1, &zopt_sometimes),
370 ZTI_INIT(ztest_ddt_repair, 1, &zopt_sometimes),
371 ZTI_INIT(ztest_dmu_snapshot_hold, 1, &zopt_sometimes),
372 ZTI_INIT(ztest_reguid, 1, &zopt_rarely),
373 ZTI_INIT(ztest_spa_rename, 1, &zopt_rarely),
374 ZTI_INIT(ztest_scrub, 1, &zopt_rarely),
375 ZTI_INIT(ztest_spa_upgrade, 1, &zopt_rarely),
376 ZTI_INIT(ztest_dsl_dataset_promote_busy, 1, &zopt_rarely),
377 ZTI_INIT(ztest_vdev_attach_detach, 1, &zopt_sometimes),
378 ZTI_INIT(ztest_vdev_LUN_growth, 1, &zopt_rarely),
379 ZTI_INIT(ztest_vdev_add_remove, 1, &ztest_opts.zo_vdevtime),
380 ZTI_INIT(ztest_vdev_aux_add_remove, 1, &ztest_opts.zo_vdevtime),
381 ZTI_INIT(ztest_fletcher, 1, &zopt_rarely),
382 ZTI_INIT(ztest_verify_dnode_bt, 1, &zopt_sometimes),
385 #define ZTEST_FUNCS (sizeof (ztest_info) / sizeof (ztest_info_t))
388 * The following struct is used to hold a list of uncalled commit callbacks.
389 * The callbacks are ordered by txg number.
391 typedef struct ztest_cb_list {
392 kmutex_t zcl_callbacks_lock;
393 list_t zcl_callbacks;
397 * Stuff we need to share writably between parent and child.
399 typedef struct ztest_shared {
400 boolean_t zs_do_init;
401 hrtime_t zs_proc_start;
402 hrtime_t zs_proc_stop;
403 hrtime_t zs_thread_start;
404 hrtime_t zs_thread_stop;
405 hrtime_t zs_thread_kill;
406 uint64_t zs_enospc_count;
407 uint64_t zs_vdev_next_leaf;
408 uint64_t zs_vdev_aux;
413 uint64_t zs_metaslab_sz;
414 uint64_t zs_metaslab_df_alloc_threshold;
418 #define ID_PARALLEL -1ULL
420 static char ztest_dev_template[] = "%s/%s.%llua";
421 static char ztest_aux_template[] = "%s/%s.%s.%llu";
422 ztest_shared_t *ztest_shared;
424 static spa_t *ztest_spa = NULL;
425 static ztest_ds_t *ztest_ds;
427 static kmutex_t ztest_vdev_lock;
430 * The ztest_name_lock protects the pool and dataset namespace used by
431 * the individual tests. To modify the namespace, consumers must grab
432 * this lock as writer. Grabbing the lock as reader will ensure that the
433 * namespace does not change while the lock is held.
435 static rwlock_t ztest_name_lock;
437 static boolean_t ztest_dump_core = B_TRUE;
438 static boolean_t ztest_exiting;
440 /* Global commit callback list */
441 static ztest_cb_list_t zcl;
442 /* Commit cb delay */
443 static uint64_t zc_min_txg_delay = UINT64_MAX;
444 static int zc_cb_counter = 0;
447 * Minimum number of commit callbacks that need to be registered for us to check
448 * whether the minimum txg delay is acceptable.
450 #define ZTEST_COMMIT_CB_MIN_REG 100
453 * If a number of txgs equal to this threshold have been created after a commit
454 * callback has been registered but not called, then we assume there is an
455 * implementation bug.
457 #define ZTEST_COMMIT_CB_THRESH (TXG_CONCURRENT_STATES + 1000)
459 extern uint64_t metaslab_gang_bang;
460 extern uint64_t metaslab_df_alloc_threshold;
463 ZTEST_META_DNODE = 0,
468 static void usage(boolean_t) __NORETURN;
471 * These libumem hooks provide a reasonable set of defaults for the allocator's
472 * debugging facilities.
475 _umem_debug_init(void)
477 return ("default,verbose"); /* $UMEM_DEBUG setting */
481 _umem_logging_init(void)
483 return ("fail,contents"); /* $UMEM_LOGGING setting */
486 #define BACKTRACE_SZ 100
488 static void sig_handler(int signo)
490 struct sigaction action;
491 #ifdef __GLIBC__ /* backtrace() is a GNU extension */
493 void *buffer[BACKTRACE_SZ];
495 nptrs = backtrace(buffer, BACKTRACE_SZ);
496 backtrace_symbols_fd(buffer, nptrs, STDERR_FILENO);
500 * Restore default action and re-raise signal so SIGSEGV and
501 * SIGABRT can trigger a core dump.
503 action.sa_handler = SIG_DFL;
504 sigemptyset(&action.sa_mask);
506 (void) sigaction(signo, &action, NULL);
510 #define FATAL_MSG_SZ 1024
515 fatal(int do_perror, char *message, ...)
518 int save_errno = errno;
521 (void) fflush(stdout);
522 buf = umem_alloc(FATAL_MSG_SZ, UMEM_NOFAIL);
524 va_start(args, message);
525 (void) sprintf(buf, "ztest: ");
527 (void) vsprintf(buf + strlen(buf), message, args);
530 (void) snprintf(buf + strlen(buf), FATAL_MSG_SZ - strlen(buf),
531 ": %s", strerror(save_errno));
533 (void) fprintf(stderr, "%s\n", buf);
534 fatal_msg = buf; /* to ease debugging */
541 str2shift(const char *buf)
543 const char *ends = "BKMGTPEZ";
548 for (i = 0; i < strlen(ends); i++) {
549 if (toupper(buf[0]) == ends[i])
552 if (i == strlen(ends)) {
553 (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n",
557 if (buf[1] == '\0' || (toupper(buf[1]) == 'B' && buf[2] == '\0')) {
560 (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n", buf);
566 nicenumtoull(const char *buf)
571 val = strtoull(buf, &end, 0);
573 (void) fprintf(stderr, "ztest: bad numeric value: %s\n", buf);
575 } else if (end[0] == '.') {
576 double fval = strtod(buf, &end);
577 fval *= pow(2, str2shift(end));
578 if (fval > UINT64_MAX) {
579 (void) fprintf(stderr, "ztest: value too large: %s\n",
583 val = (uint64_t)fval;
585 int shift = str2shift(end);
586 if (shift >= 64 || (val << shift) >> shift != val) {
587 (void) fprintf(stderr, "ztest: value too large: %s\n",
597 usage(boolean_t requested)
599 const ztest_shared_opts_t *zo = &ztest_opts_defaults;
601 char nice_vdev_size[10];
602 char nice_gang_bang[10];
603 FILE *fp = requested ? stdout : stderr;
605 nicenum(zo->zo_vdev_size, nice_vdev_size);
606 nicenum(zo->zo_metaslab_gang_bang, nice_gang_bang);
608 (void) fprintf(fp, "Usage: %s\n"
609 "\t[-v vdevs (default: %llu)]\n"
610 "\t[-s size_of_each_vdev (default: %s)]\n"
611 "\t[-a alignment_shift (default: %d)] use 0 for random\n"
612 "\t[-m mirror_copies (default: %d)]\n"
613 "\t[-r raidz_disks (default: %d)]\n"
614 "\t[-R raidz_parity (default: %d)]\n"
615 "\t[-d datasets (default: %d)]\n"
616 "\t[-t threads (default: %d)]\n"
617 "\t[-g gang_block_threshold (default: %s)]\n"
618 "\t[-i init_count (default: %d)] initialize pool i times\n"
619 "\t[-k kill_percentage (default: %llu%%)]\n"
620 "\t[-p pool_name (default: %s)]\n"
621 "\t[-f dir (default: %s)] file directory for vdev files\n"
622 "\t[-V] verbose (use multiple times for ever more blather)\n"
623 "\t[-E] use existing pool instead of creating new one\n"
624 "\t[-T time (default: %llu sec)] total run time\n"
625 "\t[-F freezeloops (default: %llu)] max loops in spa_freeze()\n"
626 "\t[-P passtime (default: %llu sec)] time per pass\n"
627 "\t[-B alt_ztest (default: <none>)] alternate ztest path\n"
628 "\t[-h] (print help)\n"
631 (u_longlong_t)zo->zo_vdevs, /* -v */
632 nice_vdev_size, /* -s */
633 zo->zo_ashift, /* -a */
634 zo->zo_mirrors, /* -m */
635 zo->zo_raidz, /* -r */
636 zo->zo_raidz_parity, /* -R */
637 zo->zo_datasets, /* -d */
638 zo->zo_threads, /* -t */
639 nice_gang_bang, /* -g */
640 zo->zo_init, /* -i */
641 (u_longlong_t)zo->zo_killrate, /* -k */
642 zo->zo_pool, /* -p */
644 (u_longlong_t)zo->zo_time, /* -T */
645 (u_longlong_t)zo->zo_maxloops, /* -F */
646 (u_longlong_t)zo->zo_passtime);
647 exit(requested ? 0 : 1);
651 process_options(int argc, char **argv)
654 ztest_shared_opts_t *zo = &ztest_opts;
658 char altdir[MAXNAMELEN] = { 0 };
660 bcopy(&ztest_opts_defaults, zo, sizeof (*zo));
662 while ((opt = getopt(argc, argv,
663 "v:s:a:m:r:R:d:t:g:i:k:p:f:VET:P:hF:B:")) != EOF) {
680 value = nicenumtoull(optarg);
684 zo->zo_vdevs = value;
687 zo->zo_vdev_size = MAX(SPA_MINDEVSIZE, value);
690 zo->zo_ashift = value;
693 zo->zo_mirrors = value;
696 zo->zo_raidz = MAX(1, value);
699 zo->zo_raidz_parity = MIN(MAX(value, 1), 3);
702 zo->zo_datasets = MAX(1, value);
705 zo->zo_threads = MAX(1, value);
708 zo->zo_metaslab_gang_bang = MAX(SPA_MINBLOCKSIZE << 1,
715 zo->zo_killrate = value;
718 (void) strlcpy(zo->zo_pool, optarg,
719 sizeof (zo->zo_pool));
722 path = realpath(optarg, NULL);
724 (void) fprintf(stderr, "error: %s: %s\n",
725 optarg, strerror(errno));
728 (void) strlcpy(zo->zo_dir, path,
729 sizeof (zo->zo_dir));
743 zo->zo_passtime = MAX(1, value);
746 zo->zo_maxloops = MAX(1, value);
749 (void) strlcpy(altdir, optarg, sizeof (altdir));
761 zo->zo_raidz_parity = MIN(zo->zo_raidz_parity, zo->zo_raidz - 1);
764 (zo->zo_vdevs > 0 ? zo->zo_time * NANOSEC / zo->zo_vdevs :
767 if (strlen(altdir) > 0) {
775 cmd = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
776 realaltdir = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
778 VERIFY(NULL != realpath(getexecname(), cmd));
779 if (0 != access(altdir, F_OK)) {
780 ztest_dump_core = B_FALSE;
781 fatal(B_TRUE, "invalid alternate ztest path: %s",
784 VERIFY(NULL != realpath(altdir, realaltdir));
787 * 'cmd' should be of the form "<anything>/usr/bin/<isa>/ztest".
788 * We want to extract <isa> to determine if we should use
789 * 32 or 64 bit binaries.
791 bin = strstr(cmd, "/usr/bin/");
792 ztest = strstr(bin, "/ztest");
794 isalen = ztest - isa;
795 (void) snprintf(zo->zo_alt_ztest, sizeof (zo->zo_alt_ztest),
796 "%s/usr/bin/%.*s/ztest", realaltdir, isalen, isa);
797 (void) snprintf(zo->zo_alt_libpath, sizeof (zo->zo_alt_libpath),
798 "%s/usr/lib/%.*s", realaltdir, isalen, isa);
800 if (0 != access(zo->zo_alt_ztest, X_OK)) {
801 ztest_dump_core = B_FALSE;
802 fatal(B_TRUE, "invalid alternate ztest: %s",
804 } else if (0 != access(zo->zo_alt_libpath, X_OK)) {
805 ztest_dump_core = B_FALSE;
806 fatal(B_TRUE, "invalid alternate lib directory %s",
810 umem_free(cmd, MAXPATHLEN);
811 umem_free(realaltdir, MAXPATHLEN);
816 ztest_kill(ztest_shared_t *zs)
818 zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(ztest_spa));
819 zs->zs_space = metaslab_class_get_space(spa_normal_class(ztest_spa));
822 * Before we kill off ztest, make sure that the config is updated.
823 * See comment above spa_config_sync().
825 mutex_enter(&spa_namespace_lock);
826 spa_config_sync(ztest_spa, B_FALSE, B_FALSE);
827 mutex_exit(&spa_namespace_lock);
829 (void) kill(getpid(), SIGKILL);
833 ztest_random(uint64_t range)
837 ASSERT3S(ztest_fd_rand, >=, 0);
842 if (read(ztest_fd_rand, &r, sizeof (r)) != sizeof (r))
843 fatal(1, "short read from /dev/urandom");
850 ztest_record_enospc(const char *s)
852 ztest_shared->zs_enospc_count++;
856 ztest_get_ashift(void)
858 if (ztest_opts.zo_ashift == 0)
859 return (SPA_MINBLOCKSHIFT + ztest_random(5));
860 return (ztest_opts.zo_ashift);
864 make_vdev_file(char *path, char *aux, char *pool, size_t size, uint64_t ashift)
870 pathbuf = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
873 ashift = ztest_get_ashift();
879 vdev = ztest_shared->zs_vdev_aux;
880 (void) snprintf(path, MAXPATHLEN,
881 ztest_aux_template, ztest_opts.zo_dir,
882 pool == NULL ? ztest_opts.zo_pool : pool,
885 vdev = ztest_shared->zs_vdev_next_leaf++;
886 (void) snprintf(path, MAXPATHLEN,
887 ztest_dev_template, ztest_opts.zo_dir,
888 pool == NULL ? ztest_opts.zo_pool : pool, vdev);
893 int fd = open(path, O_RDWR | O_CREAT | O_TRUNC, 0666);
895 fatal(1, "can't open %s", path);
896 if (ftruncate(fd, size) != 0)
897 fatal(1, "can't ftruncate %s", path);
901 VERIFY(nvlist_alloc(&file, NV_UNIQUE_NAME, 0) == 0);
902 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_TYPE, VDEV_TYPE_FILE) == 0);
903 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_PATH, path) == 0);
904 VERIFY(nvlist_add_uint64(file, ZPOOL_CONFIG_ASHIFT, ashift) == 0);
905 umem_free(pathbuf, MAXPATHLEN);
911 make_vdev_raidz(char *path, char *aux, char *pool, size_t size,
912 uint64_t ashift, int r)
914 nvlist_t *raidz, **child;
918 return (make_vdev_file(path, aux, pool, size, ashift));
919 child = umem_alloc(r * sizeof (nvlist_t *), UMEM_NOFAIL);
921 for (c = 0; c < r; c++)
922 child[c] = make_vdev_file(path, aux, pool, size, ashift);
924 VERIFY(nvlist_alloc(&raidz, NV_UNIQUE_NAME, 0) == 0);
925 VERIFY(nvlist_add_string(raidz, ZPOOL_CONFIG_TYPE,
926 VDEV_TYPE_RAIDZ) == 0);
927 VERIFY(nvlist_add_uint64(raidz, ZPOOL_CONFIG_NPARITY,
928 ztest_opts.zo_raidz_parity) == 0);
929 VERIFY(nvlist_add_nvlist_array(raidz, ZPOOL_CONFIG_CHILDREN,
932 for (c = 0; c < r; c++)
933 nvlist_free(child[c]);
935 umem_free(child, r * sizeof (nvlist_t *));
941 make_vdev_mirror(char *path, char *aux, char *pool, size_t size,
942 uint64_t ashift, int r, int m)
944 nvlist_t *mirror, **child;
948 return (make_vdev_raidz(path, aux, pool, size, ashift, r));
950 child = umem_alloc(m * sizeof (nvlist_t *), UMEM_NOFAIL);
952 for (c = 0; c < m; c++)
953 child[c] = make_vdev_raidz(path, aux, pool, size, ashift, r);
955 VERIFY(nvlist_alloc(&mirror, NV_UNIQUE_NAME, 0) == 0);
956 VERIFY(nvlist_add_string(mirror, ZPOOL_CONFIG_TYPE,
957 VDEV_TYPE_MIRROR) == 0);
958 VERIFY(nvlist_add_nvlist_array(mirror, ZPOOL_CONFIG_CHILDREN,
961 for (c = 0; c < m; c++)
962 nvlist_free(child[c]);
964 umem_free(child, m * sizeof (nvlist_t *));
970 make_vdev_root(char *path, char *aux, char *pool, size_t size, uint64_t ashift,
971 int log, int r, int m, int t)
973 nvlist_t *root, **child;
978 child = umem_alloc(t * sizeof (nvlist_t *), UMEM_NOFAIL);
980 for (c = 0; c < t; c++) {
981 child[c] = make_vdev_mirror(path, aux, pool, size, ashift,
983 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_IS_LOG,
987 VERIFY(nvlist_alloc(&root, NV_UNIQUE_NAME, 0) == 0);
988 VERIFY(nvlist_add_string(root, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) == 0);
989 VERIFY(nvlist_add_nvlist_array(root, aux ? aux : ZPOOL_CONFIG_CHILDREN,
992 for (c = 0; c < t; c++)
993 nvlist_free(child[c]);
995 umem_free(child, t * sizeof (nvlist_t *));
1001 * Find a random spa version. Returns back a random spa version in the
1002 * range [initial_version, SPA_VERSION_FEATURES].
1005 ztest_random_spa_version(uint64_t initial_version)
1007 uint64_t version = initial_version;
1009 if (version <= SPA_VERSION_BEFORE_FEATURES) {
1011 ztest_random(SPA_VERSION_BEFORE_FEATURES - version + 1);
1014 if (version > SPA_VERSION_BEFORE_FEATURES)
1015 version = SPA_VERSION_FEATURES;
1017 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
1022 ztest_random_blocksize(void)
1025 * Choose a block size >= the ashift.
1026 * If the SPA supports new MAXBLOCKSIZE, test up to 1MB blocks.
1028 int maxbs = SPA_OLD_MAXBLOCKSHIFT;
1029 if (spa_maxblocksize(ztest_spa) == SPA_MAXBLOCKSIZE)
1031 uint64_t block_shift =
1032 ztest_random(maxbs - ztest_spa->spa_max_ashift + 1);
1033 return (1 << (SPA_MINBLOCKSHIFT + block_shift));
1037 ztest_random_dnodesize(void)
1040 int max_slots = spa_maxdnodesize(ztest_spa) >> DNODE_SHIFT;
1042 if (max_slots == DNODE_MIN_SLOTS)
1043 return (DNODE_MIN_SIZE);
1046 * Weight the random distribution more heavily toward smaller
1047 * dnode sizes since that is more likely to reflect real-world
1050 ASSERT3U(max_slots, >, 4);
1051 switch (ztest_random(10)) {
1053 slots = 5 + ztest_random(max_slots - 4);
1056 slots = 2 + ztest_random(3);
1063 return (slots << DNODE_SHIFT);
1067 ztest_random_ibshift(void)
1069 return (DN_MIN_INDBLKSHIFT +
1070 ztest_random(DN_MAX_INDBLKSHIFT - DN_MIN_INDBLKSHIFT + 1));
1074 ztest_random_vdev_top(spa_t *spa, boolean_t log_ok)
1077 vdev_t *rvd = spa->spa_root_vdev;
1080 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1083 top = ztest_random(rvd->vdev_children);
1084 tvd = rvd->vdev_child[top];
1085 } while (tvd->vdev_ishole || (tvd->vdev_islog && !log_ok) ||
1086 tvd->vdev_mg == NULL || tvd->vdev_mg->mg_class == NULL);
1092 ztest_random_dsl_prop(zfs_prop_t prop)
1097 value = zfs_prop_random_value(prop, ztest_random(-1ULL));
1098 } while (prop == ZFS_PROP_CHECKSUM && value == ZIO_CHECKSUM_OFF);
1104 ztest_dsl_prop_set_uint64(char *osname, zfs_prop_t prop, uint64_t value,
1107 const char *propname = zfs_prop_to_name(prop);
1108 const char *valname;
1113 error = dsl_prop_set_int(osname, propname,
1114 (inherit ? ZPROP_SRC_NONE : ZPROP_SRC_LOCAL), value);
1116 if (error == ENOSPC) {
1117 ztest_record_enospc(FTAG);
1122 setpoint = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
1123 VERIFY0(dsl_prop_get_integer(osname, propname, &curval, setpoint));
1125 if (ztest_opts.zo_verbose >= 6) {
1128 err = zfs_prop_index_to_string(prop, curval, &valname);
1130 (void) printf("%s %s = %llu at '%s'\n",
1131 osname, propname, (unsigned long long)curval,
1134 (void) printf("%s %s = %s at '%s'\n",
1135 osname, propname, valname, setpoint);
1137 umem_free(setpoint, MAXPATHLEN);
1143 ztest_spa_prop_set_uint64(zpool_prop_t prop, uint64_t value)
1145 spa_t *spa = ztest_spa;
1146 nvlist_t *props = NULL;
1149 VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
1150 VERIFY(nvlist_add_uint64(props, zpool_prop_to_name(prop), value) == 0);
1152 error = spa_prop_set(spa, props);
1156 if (error == ENOSPC) {
1157 ztest_record_enospc(FTAG);
1167 * Object and range lock mechanics
1170 list_node_t z_lnode;
1171 refcount_t z_refcnt;
1173 zfs_rlock_t z_range_lock;
1178 ztest_znode_t *z_ztznode;
1181 static ztest_znode_t *
1182 ztest_znode_init(uint64_t object)
1184 ztest_znode_t *zp = umem_alloc(sizeof (*zp), UMEM_NOFAIL);
1186 list_link_init(&zp->z_lnode);
1187 refcount_create(&zp->z_refcnt);
1188 zp->z_object = object;
1189 zfs_rlock_init(&zp->z_range_lock);
1195 ztest_znode_fini(ztest_znode_t *zp)
1197 ASSERT(refcount_is_zero(&zp->z_refcnt));
1198 zfs_rlock_destroy(&zp->z_range_lock);
1200 refcount_destroy(&zp->z_refcnt);
1201 list_link_init(&zp->z_lnode);
1202 umem_free(zp, sizeof (*zp));
1206 ztest_zll_init(zll_t *zll)
1208 mutex_init(&zll->z_lock, NULL, MUTEX_DEFAULT, NULL);
1209 list_create(&zll->z_list, sizeof (ztest_znode_t),
1210 offsetof(ztest_znode_t, z_lnode));
1214 ztest_zll_destroy(zll_t *zll)
1216 list_destroy(&zll->z_list);
1217 mutex_destroy(&zll->z_lock);
1220 #define RL_TAG "range_lock"
1221 static ztest_znode_t *
1222 ztest_znode_get(ztest_ds_t *zd, uint64_t object)
1224 zll_t *zll = &zd->zd_range_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1225 ztest_znode_t *zp = NULL;
1226 mutex_enter(&zll->z_lock);
1227 for (zp = list_head(&zll->z_list); (zp);
1228 zp = list_next(&zll->z_list, zp)) {
1229 if (zp->z_object == object) {
1230 refcount_add(&zp->z_refcnt, RL_TAG);
1235 zp = ztest_znode_init(object);
1236 refcount_add(&zp->z_refcnt, RL_TAG);
1237 list_insert_head(&zll->z_list, zp);
1239 mutex_exit(&zll->z_lock);
1244 ztest_znode_put(ztest_ds_t *zd, ztest_znode_t *zp)
1247 ASSERT3U(zp->z_object, !=, 0);
1248 zll = &zd->zd_range_lock[zp->z_object & (ZTEST_OBJECT_LOCKS - 1)];
1249 mutex_enter(&zll->z_lock);
1250 refcount_remove(&zp->z_refcnt, RL_TAG);
1251 if (refcount_is_zero(&zp->z_refcnt)) {
1252 list_remove(&zll->z_list, zp);
1253 ztest_znode_fini(zp);
1255 mutex_exit(&zll->z_lock);
1260 ztest_rll_init(rll_t *rll)
1262 rll->rll_writer = NULL;
1263 rll->rll_readers = 0;
1264 mutex_init(&rll->rll_lock, NULL, MUTEX_DEFAULT, NULL);
1265 cv_init(&rll->rll_cv, NULL, CV_DEFAULT, NULL);
1269 ztest_rll_destroy(rll_t *rll)
1271 ASSERT(rll->rll_writer == NULL);
1272 ASSERT(rll->rll_readers == 0);
1273 mutex_destroy(&rll->rll_lock);
1274 cv_destroy(&rll->rll_cv);
1278 ztest_rll_lock(rll_t *rll, rl_type_t type)
1280 mutex_enter(&rll->rll_lock);
1282 if (type == RL_READER) {
1283 while (rll->rll_writer != NULL)
1284 (void) cv_wait(&rll->rll_cv, &rll->rll_lock);
1287 while (rll->rll_writer != NULL || rll->rll_readers)
1288 (void) cv_wait(&rll->rll_cv, &rll->rll_lock);
1289 rll->rll_writer = curthread;
1292 mutex_exit(&rll->rll_lock);
1296 ztest_rll_unlock(rll_t *rll)
1298 mutex_enter(&rll->rll_lock);
1300 if (rll->rll_writer) {
1301 ASSERT(rll->rll_readers == 0);
1302 rll->rll_writer = NULL;
1304 ASSERT(rll->rll_readers != 0);
1305 ASSERT(rll->rll_writer == NULL);
1309 if (rll->rll_writer == NULL && rll->rll_readers == 0)
1310 cv_broadcast(&rll->rll_cv);
1312 mutex_exit(&rll->rll_lock);
1316 ztest_object_lock(ztest_ds_t *zd, uint64_t object, rl_type_t type)
1318 rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1320 ztest_rll_lock(rll, type);
1324 ztest_object_unlock(ztest_ds_t *zd, uint64_t object)
1326 rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1328 ztest_rll_unlock(rll);
1331 static ztest_zrl_t *
1332 ztest_zrl_init(rl_t *rl, ztest_znode_t *zp)
1334 ztest_zrl_t *zrl = umem_alloc(sizeof (*zrl), UMEM_NOFAIL);
1336 zrl->z_ztznode = zp;
1341 ztest_zrl_fini(ztest_zrl_t *zrl)
1343 umem_free(zrl, sizeof (*zrl));
1346 static ztest_zrl_t *
1347 ztest_range_lock(ztest_ds_t *zd, uint64_t object, uint64_t offset,
1348 uint64_t size, rl_type_t type)
1350 ztest_znode_t *zp = ztest_znode_get(zd, object);
1351 rl_t *rl = zfs_range_lock(&zp->z_range_lock, offset,
1353 return (ztest_zrl_init(rl, zp));
1357 ztest_range_unlock(ztest_ds_t *zd, ztest_zrl_t *zrl)
1359 zfs_range_unlock(zrl->z_rl);
1360 ztest_znode_put(zd, zrl->z_ztznode);
1361 ztest_zrl_fini(zrl);
1365 ztest_zd_init(ztest_ds_t *zd, ztest_shared_ds_t *szd, objset_t *os)
1368 zd->zd_zilog = dmu_objset_zil(os);
1369 zd->zd_shared = szd;
1370 dmu_objset_name(os, zd->zd_name);
1373 if (zd->zd_shared != NULL)
1374 zd->zd_shared->zd_seq = 0;
1376 VERIFY(rwlock_init(&zd->zd_zilog_lock, USYNC_THREAD, NULL) == 0);
1377 mutex_init(&zd->zd_dirobj_lock, NULL, MUTEX_DEFAULT, NULL);
1379 for (l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1380 ztest_rll_init(&zd->zd_object_lock[l]);
1382 for (l = 0; l < ZTEST_RANGE_LOCKS; l++)
1383 ztest_zll_init(&zd->zd_range_lock[l]);
1387 ztest_zd_fini(ztest_ds_t *zd)
1391 mutex_destroy(&zd->zd_dirobj_lock);
1392 (void) rwlock_destroy(&zd->zd_zilog_lock);
1394 for (l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1395 ztest_rll_destroy(&zd->zd_object_lock[l]);
1397 for (l = 0; l < ZTEST_RANGE_LOCKS; l++)
1398 ztest_zll_destroy(&zd->zd_range_lock[l]);
1401 #define TXG_MIGHTWAIT (ztest_random(10) == 0 ? TXG_NOWAIT : TXG_WAIT)
1404 ztest_tx_assign(dmu_tx_t *tx, uint64_t txg_how, const char *tag)
1410 * Attempt to assign tx to some transaction group.
1412 error = dmu_tx_assign(tx, txg_how);
1414 if (error == ERESTART) {
1415 ASSERT(txg_how == TXG_NOWAIT);
1418 ASSERT3U(error, ==, ENOSPC);
1419 ztest_record_enospc(tag);
1424 txg = dmu_tx_get_txg(tx);
1430 ztest_pattern_set(void *buf, uint64_t size, uint64_t value)
1433 uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1441 ztest_pattern_match(void *buf, uint64_t size, uint64_t value)
1444 uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1448 diff |= (value - *ip++);
1455 ztest_bt_generate(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1456 uint64_t dnodesize, uint64_t offset, uint64_t gen, uint64_t txg,
1459 bt->bt_magic = BT_MAGIC;
1460 bt->bt_objset = dmu_objset_id(os);
1461 bt->bt_object = object;
1462 bt->bt_dnodesize = dnodesize;
1463 bt->bt_offset = offset;
1466 bt->bt_crtxg = crtxg;
1470 ztest_bt_verify(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1471 uint64_t dnodesize, uint64_t offset, uint64_t gen, uint64_t txg,
1474 ASSERT3U(bt->bt_magic, ==, BT_MAGIC);
1475 ASSERT3U(bt->bt_objset, ==, dmu_objset_id(os));
1476 ASSERT3U(bt->bt_object, ==, object);
1477 ASSERT3U(bt->bt_dnodesize, ==, dnodesize);
1478 ASSERT3U(bt->bt_offset, ==, offset);
1479 ASSERT3U(bt->bt_gen, <=, gen);
1480 ASSERT3U(bt->bt_txg, <=, txg);
1481 ASSERT3U(bt->bt_crtxg, ==, crtxg);
1484 static ztest_block_tag_t *
1485 ztest_bt_bonus(dmu_buf_t *db)
1487 dmu_object_info_t doi;
1488 ztest_block_tag_t *bt;
1490 dmu_object_info_from_db(db, &doi);
1491 ASSERT3U(doi.doi_bonus_size, <=, db->db_size);
1492 ASSERT3U(doi.doi_bonus_size, >=, sizeof (*bt));
1493 bt = (void *)((char *)db->db_data + doi.doi_bonus_size - sizeof (*bt));
1499 * Generate a token to fill up unused bonus buffer space. Try to make
1500 * it unique to the object, generation, and offset to verify that data
1501 * is not getting overwritten by data from other dnodes.
1503 #define ZTEST_BONUS_FILL_TOKEN(obj, ds, gen, offset) \
1504 (((ds) << 48) | ((gen) << 32) | ((obj) << 8) | (offset))
1507 * Fill up the unused bonus buffer region before the block tag with a
1508 * verifiable pattern. Filling the whole bonus area with non-zero data
1509 * helps ensure that all dnode traversal code properly skips the
1510 * interior regions of large dnodes.
1513 ztest_fill_unused_bonus(dmu_buf_t *db, void *end, uint64_t obj,
1514 objset_t *os, uint64_t gen)
1518 ASSERT(IS_P2ALIGNED((char *)end - (char *)db->db_data, 8));
1520 for (bonusp = db->db_data; bonusp < (uint64_t *)end; bonusp++) {
1521 uint64_t token = ZTEST_BONUS_FILL_TOKEN(obj, dmu_objset_id(os),
1522 gen, bonusp - (uint64_t *)db->db_data);
1528 * Verify that the unused area of a bonus buffer is filled with the
1532 ztest_verify_unused_bonus(dmu_buf_t *db, void *end, uint64_t obj,
1533 objset_t *os, uint64_t gen)
1537 for (bonusp = db->db_data; bonusp < (uint64_t *)end; bonusp++) {
1538 uint64_t token = ZTEST_BONUS_FILL_TOKEN(obj, dmu_objset_id(os),
1539 gen, bonusp - (uint64_t *)db->db_data);
1540 VERIFY3U(*bonusp, ==, token);
1548 #define lrz_type lr_mode
1549 #define lrz_blocksize lr_uid
1550 #define lrz_ibshift lr_gid
1551 #define lrz_bonustype lr_rdev
1552 #define lrz_dnodesize lr_crtime[1]
1555 ztest_log_create(ztest_ds_t *zd, dmu_tx_t *tx, lr_create_t *lr)
1557 char *name = (void *)(lr + 1); /* name follows lr */
1558 size_t namesize = strlen(name) + 1;
1561 if (zil_replaying(zd->zd_zilog, tx))
1564 itx = zil_itx_create(TX_CREATE, sizeof (*lr) + namesize);
1565 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1566 sizeof (*lr) + namesize - sizeof (lr_t));
1568 zil_itx_assign(zd->zd_zilog, itx, tx);
1572 ztest_log_remove(ztest_ds_t *zd, dmu_tx_t *tx, lr_remove_t *lr, uint64_t object)
1574 char *name = (void *)(lr + 1); /* name follows lr */
1575 size_t namesize = strlen(name) + 1;
1578 if (zil_replaying(zd->zd_zilog, tx))
1581 itx = zil_itx_create(TX_REMOVE, sizeof (*lr) + namesize);
1582 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1583 sizeof (*lr) + namesize - sizeof (lr_t));
1585 itx->itx_oid = object;
1586 zil_itx_assign(zd->zd_zilog, itx, tx);
1590 ztest_log_write(ztest_ds_t *zd, dmu_tx_t *tx, lr_write_t *lr)
1593 itx_wr_state_t write_state = ztest_random(WR_NUM_STATES);
1595 if (zil_replaying(zd->zd_zilog, tx))
1598 if (lr->lr_length > ZIL_MAX_LOG_DATA)
1599 write_state = WR_INDIRECT;
1601 itx = zil_itx_create(TX_WRITE,
1602 sizeof (*lr) + (write_state == WR_COPIED ? lr->lr_length : 0));
1604 if (write_state == WR_COPIED &&
1605 dmu_read(zd->zd_os, lr->lr_foid, lr->lr_offset, lr->lr_length,
1606 ((lr_write_t *)&itx->itx_lr) + 1, DMU_READ_NO_PREFETCH) != 0) {
1607 zil_itx_destroy(itx);
1608 itx = zil_itx_create(TX_WRITE, sizeof (*lr));
1609 write_state = WR_NEED_COPY;
1611 itx->itx_private = zd;
1612 itx->itx_wr_state = write_state;
1613 itx->itx_sync = (ztest_random(8) == 0);
1614 itx->itx_sod += (write_state == WR_NEED_COPY ? lr->lr_length : 0);
1616 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1617 sizeof (*lr) - sizeof (lr_t));
1619 zil_itx_assign(zd->zd_zilog, itx, tx);
1623 ztest_log_truncate(ztest_ds_t *zd, dmu_tx_t *tx, lr_truncate_t *lr)
1627 if (zil_replaying(zd->zd_zilog, tx))
1630 itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr));
1631 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1632 sizeof (*lr) - sizeof (lr_t));
1634 itx->itx_sync = B_FALSE;
1635 zil_itx_assign(zd->zd_zilog, itx, tx);
1639 ztest_log_setattr(ztest_ds_t *zd, dmu_tx_t *tx, lr_setattr_t *lr)
1643 if (zil_replaying(zd->zd_zilog, tx))
1646 itx = zil_itx_create(TX_SETATTR, sizeof (*lr));
1647 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1648 sizeof (*lr) - sizeof (lr_t));
1650 itx->itx_sync = B_FALSE;
1651 zil_itx_assign(zd->zd_zilog, itx, tx);
1658 ztest_replay_create(ztest_ds_t *zd, lr_create_t *lr, boolean_t byteswap)
1660 char *name = (void *)(lr + 1); /* name follows lr */
1661 objset_t *os = zd->zd_os;
1662 ztest_block_tag_t *bbt;
1670 byteswap_uint64_array(lr, sizeof (*lr));
1672 ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1673 ASSERT(name[0] != '\0');
1675 tx = dmu_tx_create(os);
1677 dmu_tx_hold_zap(tx, lr->lr_doid, B_TRUE, name);
1679 if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1680 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, B_TRUE, NULL);
1682 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1685 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1689 ASSERT(dmu_objset_zil(os)->zl_replay == !!lr->lr_foid);
1690 bonuslen = DN_BONUS_SIZE(lr->lrz_dnodesize);
1692 if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1693 if (lr->lr_foid == 0) {
1694 lr->lr_foid = zap_create_dnsize(os,
1695 lr->lrz_type, lr->lrz_bonustype,
1696 bonuslen, lr->lrz_dnodesize, tx);
1698 error = zap_create_claim_dnsize(os, lr->lr_foid,
1699 lr->lrz_type, lr->lrz_bonustype,
1700 bonuslen, lr->lrz_dnodesize, tx);
1703 if (lr->lr_foid == 0) {
1704 lr->lr_foid = dmu_object_alloc_dnsize(os,
1705 lr->lrz_type, 0, lr->lrz_bonustype,
1706 bonuslen, lr->lrz_dnodesize, tx);
1708 error = dmu_object_claim_dnsize(os, lr->lr_foid,
1709 lr->lrz_type, 0, lr->lrz_bonustype,
1710 bonuslen, lr->lrz_dnodesize, tx);
1715 ASSERT3U(error, ==, EEXIST);
1716 ASSERT(zd->zd_zilog->zl_replay);
1721 ASSERT(lr->lr_foid != 0);
1723 if (lr->lrz_type != DMU_OT_ZAP_OTHER)
1724 VERIFY3U(0, ==, dmu_object_set_blocksize(os, lr->lr_foid,
1725 lr->lrz_blocksize, lr->lrz_ibshift, tx));
1727 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1728 bbt = ztest_bt_bonus(db);
1729 dmu_buf_will_dirty(db, tx);
1730 ztest_bt_generate(bbt, os, lr->lr_foid, lr->lrz_dnodesize, -1ULL,
1731 lr->lr_gen, txg, txg);
1732 ztest_fill_unused_bonus(db, bbt, lr->lr_foid, os, lr->lr_gen);
1733 dmu_buf_rele(db, FTAG);
1735 VERIFY3U(0, ==, zap_add(os, lr->lr_doid, name, sizeof (uint64_t), 1,
1738 (void) ztest_log_create(zd, tx, lr);
1746 ztest_replay_remove(ztest_ds_t *zd, lr_remove_t *lr, boolean_t byteswap)
1748 char *name = (void *)(lr + 1); /* name follows lr */
1749 objset_t *os = zd->zd_os;
1750 dmu_object_info_t doi;
1752 uint64_t object, txg;
1755 byteswap_uint64_array(lr, sizeof (*lr));
1757 ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1758 ASSERT(name[0] != '\0');
1761 zap_lookup(os, lr->lr_doid, name, sizeof (object), 1, &object));
1762 ASSERT(object != 0);
1764 ztest_object_lock(zd, object, RL_WRITER);
1766 VERIFY3U(0, ==, dmu_object_info(os, object, &doi));
1768 tx = dmu_tx_create(os);
1770 dmu_tx_hold_zap(tx, lr->lr_doid, B_FALSE, name);
1771 dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END);
1773 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1775 ztest_object_unlock(zd, object);
1779 if (doi.doi_type == DMU_OT_ZAP_OTHER) {
1780 VERIFY3U(0, ==, zap_destroy(os, object, tx));
1782 VERIFY3U(0, ==, dmu_object_free(os, object, tx));
1785 VERIFY3U(0, ==, zap_remove(os, lr->lr_doid, name, tx));
1787 (void) ztest_log_remove(zd, tx, lr, object);
1791 ztest_object_unlock(zd, object);
1797 ztest_replay_write(ztest_ds_t *zd, lr_write_t *lr, boolean_t byteswap)
1799 objset_t *os = zd->zd_os;
1800 void *data = lr + 1; /* data follows lr */
1801 uint64_t offset, length;
1802 ztest_block_tag_t *bt = data;
1803 ztest_block_tag_t *bbt;
1804 uint64_t gen, txg, lrtxg, crtxg;
1805 dmu_object_info_t doi;
1808 arc_buf_t *abuf = NULL;
1812 byteswap_uint64_array(lr, sizeof (*lr));
1814 offset = lr->lr_offset;
1815 length = lr->lr_length;
1817 /* If it's a dmu_sync() block, write the whole block */
1818 if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
1819 uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
1820 if (length < blocksize) {
1821 offset -= offset % blocksize;
1826 if (bt->bt_magic == BSWAP_64(BT_MAGIC))
1827 byteswap_uint64_array(bt, sizeof (*bt));
1829 if (bt->bt_magic != BT_MAGIC)
1832 ztest_object_lock(zd, lr->lr_foid, RL_READER);
1833 rl = ztest_range_lock(zd, lr->lr_foid, offset, length, RL_WRITER);
1835 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1837 dmu_object_info_from_db(db, &doi);
1839 bbt = ztest_bt_bonus(db);
1840 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1842 crtxg = bbt->bt_crtxg;
1843 lrtxg = lr->lr_common.lrc_txg;
1845 tx = dmu_tx_create(os);
1847 dmu_tx_hold_write(tx, lr->lr_foid, offset, length);
1849 if (ztest_random(8) == 0 && length == doi.doi_data_block_size &&
1850 P2PHASE(offset, length) == 0)
1851 abuf = dmu_request_arcbuf(db, length);
1853 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1856 dmu_return_arcbuf(abuf);
1857 dmu_buf_rele(db, FTAG);
1858 ztest_range_unlock(zd, rl);
1859 ztest_object_unlock(zd, lr->lr_foid);
1865 * Usually, verify the old data before writing new data --
1866 * but not always, because we also want to verify correct
1867 * behavior when the data was not recently read into cache.
1869 ASSERT(offset % doi.doi_data_block_size == 0);
1870 if (ztest_random(4) != 0) {
1871 int prefetch = ztest_random(2) ?
1872 DMU_READ_PREFETCH : DMU_READ_NO_PREFETCH;
1873 ztest_block_tag_t rbt;
1875 VERIFY(dmu_read(os, lr->lr_foid, offset,
1876 sizeof (rbt), &rbt, prefetch) == 0);
1877 if (rbt.bt_magic == BT_MAGIC) {
1878 ztest_bt_verify(&rbt, os, lr->lr_foid, 0,
1879 offset, gen, txg, crtxg);
1884 * Writes can appear to be newer than the bonus buffer because
1885 * the ztest_get_data() callback does a dmu_read() of the
1886 * open-context data, which may be different than the data
1887 * as it was when the write was generated.
1889 if (zd->zd_zilog->zl_replay) {
1890 ztest_bt_verify(bt, os, lr->lr_foid, 0, offset,
1891 MAX(gen, bt->bt_gen), MAX(txg, lrtxg),
1896 * Set the bt's gen/txg to the bonus buffer's gen/txg
1897 * so that all of the usual ASSERTs will work.
1899 ztest_bt_generate(bt, os, lr->lr_foid, 0, offset, gen, txg,
1904 dmu_write(os, lr->lr_foid, offset, length, data, tx);
1906 bcopy(data, abuf->b_data, length);
1907 dmu_assign_arcbuf(db, offset, abuf, tx);
1910 (void) ztest_log_write(zd, tx, lr);
1912 dmu_buf_rele(db, FTAG);
1916 ztest_range_unlock(zd, rl);
1917 ztest_object_unlock(zd, lr->lr_foid);
1923 ztest_replay_truncate(ztest_ds_t *zd, lr_truncate_t *lr, boolean_t byteswap)
1925 objset_t *os = zd->zd_os;
1931 byteswap_uint64_array(lr, sizeof (*lr));
1933 ztest_object_lock(zd, lr->lr_foid, RL_READER);
1934 rl = ztest_range_lock(zd, lr->lr_foid, lr->lr_offset, lr->lr_length,
1937 tx = dmu_tx_create(os);
1939 dmu_tx_hold_free(tx, lr->lr_foid, lr->lr_offset, lr->lr_length);
1941 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1943 ztest_range_unlock(zd, rl);
1944 ztest_object_unlock(zd, lr->lr_foid);
1948 VERIFY(dmu_free_range(os, lr->lr_foid, lr->lr_offset,
1949 lr->lr_length, tx) == 0);
1951 (void) ztest_log_truncate(zd, tx, lr);
1955 ztest_range_unlock(zd, rl);
1956 ztest_object_unlock(zd, lr->lr_foid);
1962 ztest_replay_setattr(ztest_ds_t *zd, lr_setattr_t *lr, boolean_t byteswap)
1964 objset_t *os = zd->zd_os;
1967 ztest_block_tag_t *bbt;
1968 uint64_t txg, lrtxg, crtxg, dnodesize;
1971 byteswap_uint64_array(lr, sizeof (*lr));
1973 ztest_object_lock(zd, lr->lr_foid, RL_WRITER);
1975 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1977 tx = dmu_tx_create(os);
1978 dmu_tx_hold_bonus(tx, lr->lr_foid);
1980 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1982 dmu_buf_rele(db, FTAG);
1983 ztest_object_unlock(zd, lr->lr_foid);
1987 bbt = ztest_bt_bonus(db);
1988 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1989 crtxg = bbt->bt_crtxg;
1990 lrtxg = lr->lr_common.lrc_txg;
1991 dnodesize = bbt->bt_dnodesize;
1993 if (zd->zd_zilog->zl_replay) {
1994 ASSERT(lr->lr_size != 0);
1995 ASSERT(lr->lr_mode != 0);
1999 * Randomly change the size and increment the generation.
2001 lr->lr_size = (ztest_random(db->db_size / sizeof (*bbt)) + 1) *
2003 lr->lr_mode = bbt->bt_gen + 1;
2008 * Verify that the current bonus buffer is not newer than our txg.
2010 ztest_bt_verify(bbt, os, lr->lr_foid, dnodesize, -1ULL, lr->lr_mode,
2011 MAX(txg, lrtxg), crtxg);
2013 dmu_buf_will_dirty(db, tx);
2015 ASSERT3U(lr->lr_size, >=, sizeof (*bbt));
2016 ASSERT3U(lr->lr_size, <=, db->db_size);
2017 VERIFY0(dmu_set_bonus(db, lr->lr_size, tx));
2018 bbt = ztest_bt_bonus(db);
2020 ztest_bt_generate(bbt, os, lr->lr_foid, dnodesize, -1ULL, lr->lr_mode,
2022 ztest_fill_unused_bonus(db, bbt, lr->lr_foid, os, bbt->bt_gen);
2023 dmu_buf_rele(db, FTAG);
2025 (void) ztest_log_setattr(zd, tx, lr);
2029 ztest_object_unlock(zd, lr->lr_foid);
2034 zil_replay_func_t ztest_replay_vector[TX_MAX_TYPE] = {
2035 NULL, /* 0 no such transaction type */
2036 (zil_replay_func_t)ztest_replay_create, /* TX_CREATE */
2037 NULL, /* TX_MKDIR */
2038 NULL, /* TX_MKXATTR */
2039 NULL, /* TX_SYMLINK */
2040 (zil_replay_func_t)ztest_replay_remove, /* TX_REMOVE */
2041 NULL, /* TX_RMDIR */
2043 NULL, /* TX_RENAME */
2044 (zil_replay_func_t)ztest_replay_write, /* TX_WRITE */
2045 (zil_replay_func_t)ztest_replay_truncate, /* TX_TRUNCATE */
2046 (zil_replay_func_t)ztest_replay_setattr, /* TX_SETATTR */
2048 NULL, /* TX_CREATE_ACL */
2049 NULL, /* TX_CREATE_ATTR */
2050 NULL, /* TX_CREATE_ACL_ATTR */
2051 NULL, /* TX_MKDIR_ACL */
2052 NULL, /* TX_MKDIR_ATTR */
2053 NULL, /* TX_MKDIR_ACL_ATTR */
2054 NULL, /* TX_WRITE2 */
2058 * ZIL get_data callbacks
2060 typedef struct ztest_zgd_private {
2064 } ztest_zgd_private_t;
2067 ztest_get_done(zgd_t *zgd, int error)
2069 ztest_zgd_private_t *zzp = zgd->zgd_private;
2070 ztest_ds_t *zd = zzp->z_zd;
2071 uint64_t object = zzp->z_object;
2074 dmu_buf_rele(zgd->zgd_db, zgd);
2076 ztest_range_unlock(zd, zzp->z_rl);
2077 ztest_object_unlock(zd, object);
2079 if (error == 0 && zgd->zgd_bp)
2080 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
2082 umem_free(zgd, sizeof (*zgd));
2083 umem_free(zzp, sizeof (*zzp));
2087 ztest_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
2089 ztest_ds_t *zd = arg;
2090 objset_t *os = zd->zd_os;
2091 uint64_t object = lr->lr_foid;
2092 uint64_t offset = lr->lr_offset;
2093 uint64_t size = lr->lr_length;
2094 blkptr_t *bp = &lr->lr_blkptr;
2095 uint64_t txg = lr->lr_common.lrc_txg;
2097 dmu_object_info_t doi;
2101 ztest_zgd_private_t *zgd_private;
2103 ztest_object_lock(zd, object, RL_READER);
2104 error = dmu_bonus_hold(os, object, FTAG, &db);
2106 ztest_object_unlock(zd, object);
2110 crtxg = ztest_bt_bonus(db)->bt_crtxg;
2112 if (crtxg == 0 || crtxg > txg) {
2113 dmu_buf_rele(db, FTAG);
2114 ztest_object_unlock(zd, object);
2118 dmu_object_info_from_db(db, &doi);
2119 dmu_buf_rele(db, FTAG);
2122 zgd = umem_zalloc(sizeof (*zgd), UMEM_NOFAIL);
2123 zgd->zgd_zilog = zd->zd_zilog;
2124 zgd_private = umem_zalloc(sizeof (ztest_zgd_private_t), UMEM_NOFAIL);
2125 zgd_private->z_zd = zd;
2126 zgd_private->z_object = object;
2127 zgd->zgd_private = zgd_private;
2129 if (buf != NULL) { /* immediate write */
2130 zgd_private->z_rl = ztest_range_lock(zd, object, offset, size,
2133 error = dmu_read(os, object, offset, size, buf,
2134 DMU_READ_NO_PREFETCH);
2137 size = doi.doi_data_block_size;
2139 offset = P2ALIGN(offset, size);
2141 ASSERT(offset < size);
2145 zgd_private->z_rl = ztest_range_lock(zd, object, offset, size,
2148 error = dmu_buf_hold(os, object, offset, zgd, &db,
2149 DMU_READ_NO_PREFETCH);
2152 blkptr_t *obp = dmu_buf_get_blkptr(db);
2154 ASSERT(BP_IS_HOLE(bp));
2161 ASSERT(db->db_offset == offset);
2162 ASSERT(db->db_size == size);
2164 error = dmu_sync(zio, lr->lr_common.lrc_txg,
2165 ztest_get_done, zgd);
2172 ztest_get_done(zgd, error);
2178 ztest_lr_alloc(size_t lrsize, char *name)
2181 size_t namesize = name ? strlen(name) + 1 : 0;
2183 lr = umem_zalloc(lrsize + namesize, UMEM_NOFAIL);
2186 bcopy(name, lr + lrsize, namesize);
2192 ztest_lr_free(void *lr, size_t lrsize, char *name)
2194 size_t namesize = name ? strlen(name) + 1 : 0;
2196 umem_free(lr, lrsize + namesize);
2200 * Lookup a bunch of objects. Returns the number of objects not found.
2203 ztest_lookup(ztest_ds_t *zd, ztest_od_t *od, int count)
2209 ASSERT(mutex_held(&zd->zd_dirobj_lock));
2211 for (i = 0; i < count; i++, od++) {
2213 error = zap_lookup(zd->zd_os, od->od_dir, od->od_name,
2214 sizeof (uint64_t), 1, &od->od_object);
2216 ASSERT(error == ENOENT);
2217 ASSERT(od->od_object == 0);
2221 ztest_block_tag_t *bbt;
2222 dmu_object_info_t doi;
2224 ASSERT(od->od_object != 0);
2225 ASSERT(missing == 0); /* there should be no gaps */
2227 ztest_object_lock(zd, od->od_object, RL_READER);
2228 VERIFY3U(0, ==, dmu_bonus_hold(zd->zd_os,
2229 od->od_object, FTAG, &db));
2230 dmu_object_info_from_db(db, &doi);
2231 bbt = ztest_bt_bonus(db);
2232 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
2233 od->od_type = doi.doi_type;
2234 od->od_blocksize = doi.doi_data_block_size;
2235 od->od_gen = bbt->bt_gen;
2236 dmu_buf_rele(db, FTAG);
2237 ztest_object_unlock(zd, od->od_object);
2245 ztest_create(ztest_ds_t *zd, ztest_od_t *od, int count)
2250 ASSERT(mutex_held(&zd->zd_dirobj_lock));
2252 for (i = 0; i < count; i++, od++) {
2259 lr_create_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
2261 lr->lr_doid = od->od_dir;
2262 lr->lr_foid = 0; /* 0 to allocate, > 0 to claim */
2263 lr->lrz_type = od->od_crtype;
2264 lr->lrz_blocksize = od->od_crblocksize;
2265 lr->lrz_ibshift = ztest_random_ibshift();
2266 lr->lrz_bonustype = DMU_OT_UINT64_OTHER;
2267 lr->lrz_dnodesize = od->od_crdnodesize;
2268 lr->lr_gen = od->od_crgen;
2269 lr->lr_crtime[0] = time(NULL);
2271 if (ztest_replay_create(zd, lr, B_FALSE) != 0) {
2272 ASSERT(missing == 0);
2276 od->od_object = lr->lr_foid;
2277 od->od_type = od->od_crtype;
2278 od->od_blocksize = od->od_crblocksize;
2279 od->od_gen = od->od_crgen;
2280 ASSERT(od->od_object != 0);
2283 ztest_lr_free(lr, sizeof (*lr), od->od_name);
2290 ztest_remove(ztest_ds_t *zd, ztest_od_t *od, int count)
2296 ASSERT(mutex_held(&zd->zd_dirobj_lock));
2300 for (i = count - 1; i >= 0; i--, od--) {
2307 * No object was found.
2309 if (od->od_object == 0)
2312 lr_remove_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
2314 lr->lr_doid = od->od_dir;
2316 if ((error = ztest_replay_remove(zd, lr, B_FALSE)) != 0) {
2317 ASSERT3U(error, ==, ENOSPC);
2322 ztest_lr_free(lr, sizeof (*lr), od->od_name);
2329 ztest_write(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size,
2335 lr = ztest_lr_alloc(sizeof (*lr) + size, NULL);
2337 lr->lr_foid = object;
2338 lr->lr_offset = offset;
2339 lr->lr_length = size;
2341 BP_ZERO(&lr->lr_blkptr);
2343 bcopy(data, lr + 1, size);
2345 error = ztest_replay_write(zd, lr, B_FALSE);
2347 ztest_lr_free(lr, sizeof (*lr) + size, NULL);
2353 ztest_truncate(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2358 lr = ztest_lr_alloc(sizeof (*lr), NULL);
2360 lr->lr_foid = object;
2361 lr->lr_offset = offset;
2362 lr->lr_length = size;
2364 error = ztest_replay_truncate(zd, lr, B_FALSE);
2366 ztest_lr_free(lr, sizeof (*lr), NULL);
2372 ztest_setattr(ztest_ds_t *zd, uint64_t object)
2377 lr = ztest_lr_alloc(sizeof (*lr), NULL);
2379 lr->lr_foid = object;
2383 error = ztest_replay_setattr(zd, lr, B_FALSE);
2385 ztest_lr_free(lr, sizeof (*lr), NULL);
2391 ztest_prealloc(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2393 objset_t *os = zd->zd_os;
2398 txg_wait_synced(dmu_objset_pool(os), 0);
2400 ztest_object_lock(zd, object, RL_READER);
2401 rl = ztest_range_lock(zd, object, offset, size, RL_WRITER);
2403 tx = dmu_tx_create(os);
2405 dmu_tx_hold_write(tx, object, offset, size);
2407 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
2410 dmu_prealloc(os, object, offset, size, tx);
2412 txg_wait_synced(dmu_objset_pool(os), txg);
2414 (void) dmu_free_long_range(os, object, offset, size);
2417 ztest_range_unlock(zd, rl);
2418 ztest_object_unlock(zd, object);
2422 ztest_io(ztest_ds_t *zd, uint64_t object, uint64_t offset)
2425 ztest_block_tag_t wbt;
2426 dmu_object_info_t doi;
2427 enum ztest_io_type io_type;
2431 VERIFY(dmu_object_info(zd->zd_os, object, &doi) == 0);
2432 blocksize = doi.doi_data_block_size;
2433 data = umem_alloc(blocksize, UMEM_NOFAIL);
2436 * Pick an i/o type at random, biased toward writing block tags.
2438 io_type = ztest_random(ZTEST_IO_TYPES);
2439 if (ztest_random(2) == 0)
2440 io_type = ZTEST_IO_WRITE_TAG;
2442 (void) rw_rdlock(&zd->zd_zilog_lock);
2446 case ZTEST_IO_WRITE_TAG:
2447 ztest_bt_generate(&wbt, zd->zd_os, object, doi.doi_dnodesize,
2449 (void) ztest_write(zd, object, offset, sizeof (wbt), &wbt);
2452 case ZTEST_IO_WRITE_PATTERN:
2453 (void) memset(data, 'a' + (object + offset) % 5, blocksize);
2454 if (ztest_random(2) == 0) {
2456 * Induce fletcher2 collisions to ensure that
2457 * zio_ddt_collision() detects and resolves them
2458 * when using fletcher2-verify for deduplication.
2460 ((uint64_t *)data)[0] ^= 1ULL << 63;
2461 ((uint64_t *)data)[4] ^= 1ULL << 63;
2463 (void) ztest_write(zd, object, offset, blocksize, data);
2466 case ZTEST_IO_WRITE_ZEROES:
2467 bzero(data, blocksize);
2468 (void) ztest_write(zd, object, offset, blocksize, data);
2471 case ZTEST_IO_TRUNCATE:
2472 (void) ztest_truncate(zd, object, offset, blocksize);
2475 case ZTEST_IO_SETATTR:
2476 (void) ztest_setattr(zd, object);
2481 case ZTEST_IO_REWRITE:
2482 (void) rw_rdlock(&ztest_name_lock);
2483 err = ztest_dsl_prop_set_uint64(zd->zd_name,
2484 ZFS_PROP_CHECKSUM, spa_dedup_checksum(ztest_spa),
2486 VERIFY(err == 0 || err == ENOSPC);
2487 err = ztest_dsl_prop_set_uint64(zd->zd_name,
2488 ZFS_PROP_COMPRESSION,
2489 ztest_random_dsl_prop(ZFS_PROP_COMPRESSION),
2491 VERIFY(err == 0 || err == ENOSPC);
2492 (void) rw_unlock(&ztest_name_lock);
2494 VERIFY0(dmu_read(zd->zd_os, object, offset, blocksize, data,
2495 DMU_READ_NO_PREFETCH));
2497 (void) ztest_write(zd, object, offset, blocksize, data);
2501 (void) rw_unlock(&zd->zd_zilog_lock);
2503 umem_free(data, blocksize);
2507 * Initialize an object description template.
2510 ztest_od_init(ztest_od_t *od, uint64_t id, char *tag, uint64_t index,
2511 dmu_object_type_t type, uint64_t blocksize, uint64_t dnodesize,
2514 od->od_dir = ZTEST_DIROBJ;
2517 od->od_crtype = type;
2518 od->od_crblocksize = blocksize ? blocksize : ztest_random_blocksize();
2519 od->od_crdnodesize = dnodesize ? dnodesize : ztest_random_dnodesize();
2522 od->od_type = DMU_OT_NONE;
2523 od->od_blocksize = 0;
2526 (void) snprintf(od->od_name, sizeof (od->od_name), "%s(%lld)[%llu]",
2527 tag, (longlong_t)id, (u_longlong_t)index);
2531 * Lookup or create the objects for a test using the od template.
2532 * If the objects do not all exist, or if 'remove' is specified,
2533 * remove any existing objects and create new ones. Otherwise,
2534 * use the existing objects.
2537 ztest_object_init(ztest_ds_t *zd, ztest_od_t *od, size_t size, boolean_t remove)
2539 int count = size / sizeof (*od);
2542 mutex_enter(&zd->zd_dirobj_lock);
2543 if ((ztest_lookup(zd, od, count) != 0 || remove) &&
2544 (ztest_remove(zd, od, count) != 0 ||
2545 ztest_create(zd, od, count) != 0))
2548 mutex_exit(&zd->zd_dirobj_lock);
2555 ztest_zil_commit(ztest_ds_t *zd, uint64_t id)
2557 zilog_t *zilog = zd->zd_zilog;
2559 (void) rw_rdlock(&zd->zd_zilog_lock);
2561 zil_commit(zilog, ztest_random(ZTEST_OBJECTS));
2564 * Remember the committed values in zd, which is in parent/child
2565 * shared memory. If we die, the next iteration of ztest_run()
2566 * will verify that the log really does contain this record.
2568 mutex_enter(&zilog->zl_lock);
2569 ASSERT(zd->zd_shared != NULL);
2570 ASSERT3U(zd->zd_shared->zd_seq, <=, zilog->zl_commit_lr_seq);
2571 zd->zd_shared->zd_seq = zilog->zl_commit_lr_seq;
2572 mutex_exit(&zilog->zl_lock);
2574 (void) rw_unlock(&zd->zd_zilog_lock);
2578 * This function is designed to simulate the operations that occur during a
2579 * mount/unmount operation. We hold the dataset across these operations in an
2580 * attempt to expose any implicit assumptions about ZIL management.
2584 ztest_zil_remount(ztest_ds_t *zd, uint64_t id)
2586 objset_t *os = zd->zd_os;
2589 * We grab the zd_dirobj_lock to ensure that no other thread is
2590 * updating the zil (i.e. adding in-memory log records) and the
2591 * zd_zilog_lock to block any I/O.
2593 mutex_enter(&zd->zd_dirobj_lock);
2594 (void) rw_wrlock(&zd->zd_zilog_lock);
2596 /* zfs_sb_teardown() */
2597 zil_close(zd->zd_zilog);
2599 /* zfsvfs_setup() */
2600 VERIFY(zil_open(os, ztest_get_data) == zd->zd_zilog);
2601 zil_replay(os, zd, ztest_replay_vector);
2603 (void) rw_unlock(&zd->zd_zilog_lock);
2604 mutex_exit(&zd->zd_dirobj_lock);
2608 * Verify that we can't destroy an active pool, create an existing pool,
2609 * or create a pool with a bad vdev spec.
2613 ztest_spa_create_destroy(ztest_ds_t *zd, uint64_t id)
2615 ztest_shared_opts_t *zo = &ztest_opts;
2620 * Attempt to create using a bad file.
2622 nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1);
2623 VERIFY3U(ENOENT, ==,
2624 spa_create("ztest_bad_file", nvroot, NULL, NULL));
2625 nvlist_free(nvroot);
2628 * Attempt to create using a bad mirror.
2630 nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 2, 1);
2631 VERIFY3U(ENOENT, ==,
2632 spa_create("ztest_bad_mirror", nvroot, NULL, NULL));
2633 nvlist_free(nvroot);
2636 * Attempt to create an existing pool. It shouldn't matter
2637 * what's in the nvroot; we should fail with EEXIST.
2639 (void) rw_rdlock(&ztest_name_lock);
2640 nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1);
2641 VERIFY3U(EEXIST, ==, spa_create(zo->zo_pool, nvroot, NULL, NULL));
2642 nvlist_free(nvroot);
2643 VERIFY3U(0, ==, spa_open(zo->zo_pool, &spa, FTAG));
2644 VERIFY3U(EBUSY, ==, spa_destroy(zo->zo_pool));
2645 spa_close(spa, FTAG);
2647 (void) rw_unlock(&ztest_name_lock);
2652 ztest_spa_upgrade(ztest_ds_t *zd, uint64_t id)
2655 uint64_t initial_version = SPA_VERSION_INITIAL;
2656 uint64_t version, newversion;
2657 nvlist_t *nvroot, *props;
2660 mutex_enter(&ztest_vdev_lock);
2661 name = kmem_asprintf("%s_upgrade", ztest_opts.zo_pool);
2664 * Clean up from previous runs.
2666 (void) spa_destroy(name);
2668 nvroot = make_vdev_root(NULL, NULL, name, ztest_opts.zo_vdev_size, 0,
2669 0, ztest_opts.zo_raidz, ztest_opts.zo_mirrors, 1);
2672 * If we're configuring a RAIDZ device then make sure that the
2673 * the initial version is capable of supporting that feature.
2675 switch (ztest_opts.zo_raidz_parity) {
2678 initial_version = SPA_VERSION_INITIAL;
2681 initial_version = SPA_VERSION_RAIDZ2;
2684 initial_version = SPA_VERSION_RAIDZ3;
2689 * Create a pool with a spa version that can be upgraded. Pick
2690 * a value between initial_version and SPA_VERSION_BEFORE_FEATURES.
2693 version = ztest_random_spa_version(initial_version);
2694 } while (version > SPA_VERSION_BEFORE_FEATURES);
2696 props = fnvlist_alloc();
2697 fnvlist_add_uint64(props,
2698 zpool_prop_to_name(ZPOOL_PROP_VERSION), version);
2699 VERIFY3S(spa_create(name, nvroot, props, NULL), ==, 0);
2700 fnvlist_free(nvroot);
2701 fnvlist_free(props);
2703 VERIFY3S(spa_open(name, &spa, FTAG), ==, 0);
2704 VERIFY3U(spa_version(spa), ==, version);
2705 newversion = ztest_random_spa_version(version + 1);
2707 if (ztest_opts.zo_verbose >= 4) {
2708 (void) printf("upgrading spa version from %llu to %llu\n",
2709 (u_longlong_t)version, (u_longlong_t)newversion);
2712 spa_upgrade(spa, newversion);
2713 VERIFY3U(spa_version(spa), >, version);
2714 VERIFY3U(spa_version(spa), ==, fnvlist_lookup_uint64(spa->spa_config,
2715 zpool_prop_to_name(ZPOOL_PROP_VERSION)));
2716 spa_close(spa, FTAG);
2719 mutex_exit(&ztest_vdev_lock);
2723 vdev_lookup_by_path(vdev_t *vd, const char *path)
2728 if (vd->vdev_path != NULL && strcmp(path, vd->vdev_path) == 0)
2731 for (c = 0; c < vd->vdev_children; c++)
2732 if ((mvd = vdev_lookup_by_path(vd->vdev_child[c], path)) !=
2740 * Find the first available hole which can be used as a top-level.
2743 find_vdev_hole(spa_t *spa)
2745 vdev_t *rvd = spa->spa_root_vdev;
2748 ASSERT(spa_config_held(spa, SCL_VDEV, RW_READER) == SCL_VDEV);
2750 for (c = 0; c < rvd->vdev_children; c++) {
2751 vdev_t *cvd = rvd->vdev_child[c];
2753 if (cvd->vdev_ishole)
2760 * Verify that vdev_add() works as expected.
2764 ztest_vdev_add_remove(ztest_ds_t *zd, uint64_t id)
2766 ztest_shared_t *zs = ztest_shared;
2767 spa_t *spa = ztest_spa;
2773 mutex_enter(&ztest_vdev_lock);
2774 leaves = MAX(zs->zs_mirrors + zs->zs_splits, 1) * ztest_opts.zo_raidz;
2776 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2778 ztest_shared->zs_vdev_next_leaf = find_vdev_hole(spa) * leaves;
2781 * If we have slogs then remove them 1/4 of the time.
2783 if (spa_has_slogs(spa) && ztest_random(4) == 0) {
2785 * Grab the guid from the head of the log class rotor.
2787 guid = spa_log_class(spa)->mc_rotor->mg_vd->vdev_guid;
2789 spa_config_exit(spa, SCL_VDEV, FTAG);
2792 * We have to grab the zs_name_lock as writer to
2793 * prevent a race between removing a slog (dmu_objset_find)
2794 * and destroying a dataset. Removing the slog will
2795 * grab a reference on the dataset which may cause
2796 * dsl_destroy_head() to fail with EBUSY thus
2797 * leaving the dataset in an inconsistent state.
2799 rw_wrlock(&ztest_name_lock);
2800 error = spa_vdev_remove(spa, guid, B_FALSE);
2801 rw_unlock(&ztest_name_lock);
2803 if (error && error != EEXIST)
2804 fatal(0, "spa_vdev_remove() = %d", error);
2806 spa_config_exit(spa, SCL_VDEV, FTAG);
2809 * Make 1/4 of the devices be log devices.
2811 nvroot = make_vdev_root(NULL, NULL, NULL,
2812 ztest_opts.zo_vdev_size, 0,
2813 ztest_random(4) == 0, ztest_opts.zo_raidz,
2816 error = spa_vdev_add(spa, nvroot);
2817 nvlist_free(nvroot);
2819 if (error == ENOSPC)
2820 ztest_record_enospc("spa_vdev_add");
2821 else if (error != 0)
2822 fatal(0, "spa_vdev_add() = %d", error);
2825 mutex_exit(&ztest_vdev_lock);
2829 * Verify that adding/removing aux devices (l2arc, hot spare) works as expected.
2833 ztest_vdev_aux_add_remove(ztest_ds_t *zd, uint64_t id)
2835 ztest_shared_t *zs = ztest_shared;
2836 spa_t *spa = ztest_spa;
2837 vdev_t *rvd = spa->spa_root_vdev;
2838 spa_aux_vdev_t *sav;
2844 path = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
2846 if (ztest_random(2) == 0) {
2847 sav = &spa->spa_spares;
2848 aux = ZPOOL_CONFIG_SPARES;
2850 sav = &spa->spa_l2cache;
2851 aux = ZPOOL_CONFIG_L2CACHE;
2854 mutex_enter(&ztest_vdev_lock);
2856 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2858 if (sav->sav_count != 0 && ztest_random(4) == 0) {
2860 * Pick a random device to remove.
2862 guid = sav->sav_vdevs[ztest_random(sav->sav_count)]->vdev_guid;
2865 * Find an unused device we can add.
2867 zs->zs_vdev_aux = 0;
2870 (void) snprintf(path, MAXPATHLEN, ztest_aux_template,
2871 ztest_opts.zo_dir, ztest_opts.zo_pool, aux,
2873 for (c = 0; c < sav->sav_count; c++)
2874 if (strcmp(sav->sav_vdevs[c]->vdev_path,
2877 if (c == sav->sav_count &&
2878 vdev_lookup_by_path(rvd, path) == NULL)
2884 spa_config_exit(spa, SCL_VDEV, FTAG);
2890 nvlist_t *nvroot = make_vdev_root(NULL, aux, NULL,
2891 (ztest_opts.zo_vdev_size * 5) / 4, 0, 0, 0, 0, 1);
2892 error = spa_vdev_add(spa, nvroot);
2894 fatal(0, "spa_vdev_add(%p) = %d", nvroot, error);
2895 nvlist_free(nvroot);
2898 * Remove an existing device. Sometimes, dirty its
2899 * vdev state first to make sure we handle removal
2900 * of devices that have pending state changes.
2902 if (ztest_random(2) == 0)
2903 (void) vdev_online(spa, guid, 0, NULL);
2905 error = spa_vdev_remove(spa, guid, B_FALSE);
2906 if (error != 0 && error != EBUSY)
2907 fatal(0, "spa_vdev_remove(%llu) = %d", guid, error);
2910 mutex_exit(&ztest_vdev_lock);
2912 umem_free(path, MAXPATHLEN);
2916 * split a pool if it has mirror tlvdevs
2920 ztest_split_pool(ztest_ds_t *zd, uint64_t id)
2922 ztest_shared_t *zs = ztest_shared;
2923 spa_t *spa = ztest_spa;
2924 vdev_t *rvd = spa->spa_root_vdev;
2925 nvlist_t *tree, **child, *config, *split, **schild;
2926 uint_t c, children, schildren = 0, lastlogid = 0;
2929 mutex_enter(&ztest_vdev_lock);
2931 /* ensure we have a useable config; mirrors of raidz aren't supported */
2932 if (zs->zs_mirrors < 3 || ztest_opts.zo_raidz > 1) {
2933 mutex_exit(&ztest_vdev_lock);
2937 /* clean up the old pool, if any */
2938 (void) spa_destroy("splitp");
2940 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2942 /* generate a config from the existing config */
2943 mutex_enter(&spa->spa_props_lock);
2944 VERIFY(nvlist_lookup_nvlist(spa->spa_config, ZPOOL_CONFIG_VDEV_TREE,
2946 mutex_exit(&spa->spa_props_lock);
2948 VERIFY(nvlist_lookup_nvlist_array(tree, ZPOOL_CONFIG_CHILDREN, &child,
2951 schild = malloc(rvd->vdev_children * sizeof (nvlist_t *));
2952 for (c = 0; c < children; c++) {
2953 vdev_t *tvd = rvd->vdev_child[c];
2957 if (tvd->vdev_islog || tvd->vdev_ops == &vdev_hole_ops) {
2958 VERIFY(nvlist_alloc(&schild[schildren], NV_UNIQUE_NAME,
2960 VERIFY(nvlist_add_string(schild[schildren],
2961 ZPOOL_CONFIG_TYPE, VDEV_TYPE_HOLE) == 0);
2962 VERIFY(nvlist_add_uint64(schild[schildren],
2963 ZPOOL_CONFIG_IS_HOLE, 1) == 0);
2965 lastlogid = schildren;
2970 VERIFY(nvlist_lookup_nvlist_array(child[c],
2971 ZPOOL_CONFIG_CHILDREN, &mchild, &mchildren) == 0);
2972 VERIFY(nvlist_dup(mchild[0], &schild[schildren++], 0) == 0);
2975 /* OK, create a config that can be used to split */
2976 VERIFY(nvlist_alloc(&split, NV_UNIQUE_NAME, 0) == 0);
2977 VERIFY(nvlist_add_string(split, ZPOOL_CONFIG_TYPE,
2978 VDEV_TYPE_ROOT) == 0);
2979 VERIFY(nvlist_add_nvlist_array(split, ZPOOL_CONFIG_CHILDREN, schild,
2980 lastlogid != 0 ? lastlogid : schildren) == 0);
2982 VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, 0) == 0);
2983 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, split) == 0);
2985 for (c = 0; c < schildren; c++)
2986 nvlist_free(schild[c]);
2990 spa_config_exit(spa, SCL_VDEV, FTAG);
2992 (void) rw_wrlock(&ztest_name_lock);
2993 error = spa_vdev_split_mirror(spa, "splitp", config, NULL, B_FALSE);
2994 (void) rw_unlock(&ztest_name_lock);
2996 nvlist_free(config);
2999 (void) printf("successful split - results:\n");
3000 mutex_enter(&spa_namespace_lock);
3001 show_pool_stats(spa);
3002 show_pool_stats(spa_lookup("splitp"));
3003 mutex_exit(&spa_namespace_lock);
3007 mutex_exit(&ztest_vdev_lock);
3012 * Verify that we can attach and detach devices.
3016 ztest_vdev_attach_detach(ztest_ds_t *zd, uint64_t id)
3018 ztest_shared_t *zs = ztest_shared;
3019 spa_t *spa = ztest_spa;
3020 spa_aux_vdev_t *sav = &spa->spa_spares;
3021 vdev_t *rvd = spa->spa_root_vdev;
3022 vdev_t *oldvd, *newvd, *pvd;
3026 uint64_t ashift = ztest_get_ashift();
3027 uint64_t oldguid, pguid;
3028 uint64_t oldsize, newsize;
3029 char *oldpath, *newpath;
3031 int oldvd_has_siblings = B_FALSE;
3032 int newvd_is_spare = B_FALSE;
3034 int error, expected_error;
3036 oldpath = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
3037 newpath = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
3039 mutex_enter(&ztest_vdev_lock);
3040 leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz;
3042 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
3045 * Decide whether to do an attach or a replace.
3047 replacing = ztest_random(2);
3050 * Pick a random top-level vdev.
3052 top = ztest_random_vdev_top(spa, B_TRUE);
3055 * Pick a random leaf within it.
3057 leaf = ztest_random(leaves);
3062 oldvd = rvd->vdev_child[top];
3063 if (zs->zs_mirrors >= 1) {
3064 ASSERT(oldvd->vdev_ops == &vdev_mirror_ops);
3065 ASSERT(oldvd->vdev_children >= zs->zs_mirrors);
3066 oldvd = oldvd->vdev_child[leaf / ztest_opts.zo_raidz];
3068 if (ztest_opts.zo_raidz > 1) {
3069 ASSERT(oldvd->vdev_ops == &vdev_raidz_ops);
3070 ASSERT(oldvd->vdev_children == ztest_opts.zo_raidz);
3071 oldvd = oldvd->vdev_child[leaf % ztest_opts.zo_raidz];
3075 * If we're already doing an attach or replace, oldvd may be a
3076 * mirror vdev -- in which case, pick a random child.
3078 while (oldvd->vdev_children != 0) {
3079 oldvd_has_siblings = B_TRUE;
3080 ASSERT(oldvd->vdev_children >= 2);
3081 oldvd = oldvd->vdev_child[ztest_random(oldvd->vdev_children)];
3084 oldguid = oldvd->vdev_guid;
3085 oldsize = vdev_get_min_asize(oldvd);
3086 oldvd_is_log = oldvd->vdev_top->vdev_islog;
3087 (void) strcpy(oldpath, oldvd->vdev_path);
3088 pvd = oldvd->vdev_parent;
3089 pguid = pvd->vdev_guid;
3092 * If oldvd has siblings, then half of the time, detach it.
3094 if (oldvd_has_siblings && ztest_random(2) == 0) {
3095 spa_config_exit(spa, SCL_VDEV, FTAG);
3096 error = spa_vdev_detach(spa, oldguid, pguid, B_FALSE);
3097 if (error != 0 && error != ENODEV && error != EBUSY &&
3099 fatal(0, "detach (%s) returned %d", oldpath, error);
3104 * For the new vdev, choose with equal probability between the two
3105 * standard paths (ending in either 'a' or 'b') or a random hot spare.
3107 if (sav->sav_count != 0 && ztest_random(3) == 0) {
3108 newvd = sav->sav_vdevs[ztest_random(sav->sav_count)];
3109 newvd_is_spare = B_TRUE;
3110 (void) strcpy(newpath, newvd->vdev_path);
3112 (void) snprintf(newpath, MAXPATHLEN, ztest_dev_template,
3113 ztest_opts.zo_dir, ztest_opts.zo_pool,
3114 top * leaves + leaf);
3115 if (ztest_random(2) == 0)
3116 newpath[strlen(newpath) - 1] = 'b';
3117 newvd = vdev_lookup_by_path(rvd, newpath);
3121 newsize = vdev_get_min_asize(newvd);
3124 * Make newsize a little bigger or smaller than oldsize.
3125 * If it's smaller, the attach should fail.
3126 * If it's larger, and we're doing a replace,
3127 * we should get dynamic LUN growth when we're done.
3129 newsize = 10 * oldsize / (9 + ztest_random(3));
3133 * If pvd is not a mirror or root, the attach should fail with ENOTSUP,
3134 * unless it's a replace; in that case any non-replacing parent is OK.
3136 * If newvd is already part of the pool, it should fail with EBUSY.
3138 * If newvd is too small, it should fail with EOVERFLOW.
3140 if (pvd->vdev_ops != &vdev_mirror_ops &&
3141 pvd->vdev_ops != &vdev_root_ops && (!replacing ||
3142 pvd->vdev_ops == &vdev_replacing_ops ||
3143 pvd->vdev_ops == &vdev_spare_ops))
3144 expected_error = ENOTSUP;
3145 else if (newvd_is_spare && (!replacing || oldvd_is_log))
3146 expected_error = ENOTSUP;
3147 else if (newvd == oldvd)
3148 expected_error = replacing ? 0 : EBUSY;
3149 else if (vdev_lookup_by_path(rvd, newpath) != NULL)
3150 expected_error = EBUSY;
3151 else if (newsize < oldsize)
3152 expected_error = EOVERFLOW;
3153 else if (ashift > oldvd->vdev_top->vdev_ashift)
3154 expected_error = EDOM;
3158 spa_config_exit(spa, SCL_VDEV, FTAG);
3161 * Build the nvlist describing newpath.
3163 root = make_vdev_root(newpath, NULL, NULL, newvd == NULL ? newsize : 0,
3164 ashift, 0, 0, 0, 1);
3166 error = spa_vdev_attach(spa, oldguid, root, replacing);
3171 * If our parent was the replacing vdev, but the replace completed,
3172 * then instead of failing with ENOTSUP we may either succeed,
3173 * fail with ENODEV, or fail with EOVERFLOW.
3175 if (expected_error == ENOTSUP &&
3176 (error == 0 || error == ENODEV || error == EOVERFLOW))
3177 expected_error = error;
3180 * If someone grew the LUN, the replacement may be too small.
3182 if (error == EOVERFLOW || error == EBUSY)
3183 expected_error = error;
3185 /* XXX workaround 6690467 */
3186 if (error != expected_error && expected_error != EBUSY) {
3187 fatal(0, "attach (%s %llu, %s %llu, %d) "
3188 "returned %d, expected %d",
3189 oldpath, oldsize, newpath,
3190 newsize, replacing, error, expected_error);
3193 mutex_exit(&ztest_vdev_lock);
3195 umem_free(oldpath, MAXPATHLEN);
3196 umem_free(newpath, MAXPATHLEN);
3200 * Callback function which expands the physical size of the vdev.
3203 grow_vdev(vdev_t *vd, void *arg)
3205 ASSERTV(spa_t *spa = vd->vdev_spa);
3206 size_t *newsize = arg;
3210 ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
3211 ASSERT(vd->vdev_ops->vdev_op_leaf);
3213 if ((fd = open(vd->vdev_path, O_RDWR)) == -1)
3216 fsize = lseek(fd, 0, SEEK_END);
3217 VERIFY(ftruncate(fd, *newsize) == 0);
3219 if (ztest_opts.zo_verbose >= 6) {
3220 (void) printf("%s grew from %lu to %lu bytes\n",
3221 vd->vdev_path, (ulong_t)fsize, (ulong_t)*newsize);
3228 * Callback function which expands a given vdev by calling vdev_online().
3232 online_vdev(vdev_t *vd, void *arg)
3234 spa_t *spa = vd->vdev_spa;
3235 vdev_t *tvd = vd->vdev_top;
3236 uint64_t guid = vd->vdev_guid;
3237 uint64_t generation = spa->spa_config_generation + 1;
3238 vdev_state_t newstate = VDEV_STATE_UNKNOWN;
3241 ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
3242 ASSERT(vd->vdev_ops->vdev_op_leaf);
3244 /* Calling vdev_online will initialize the new metaslabs */
3245 spa_config_exit(spa, SCL_STATE, spa);
3246 error = vdev_online(spa, guid, ZFS_ONLINE_EXPAND, &newstate);
3247 spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3250 * If vdev_online returned an error or the underlying vdev_open
3251 * failed then we abort the expand. The only way to know that
3252 * vdev_open fails is by checking the returned newstate.
3254 if (error || newstate != VDEV_STATE_HEALTHY) {
3255 if (ztest_opts.zo_verbose >= 5) {
3256 (void) printf("Unable to expand vdev, state %llu, "
3257 "error %d\n", (u_longlong_t)newstate, error);
3261 ASSERT3U(newstate, ==, VDEV_STATE_HEALTHY);
3264 * Since we dropped the lock we need to ensure that we're
3265 * still talking to the original vdev. It's possible this
3266 * vdev may have been detached/replaced while we were
3267 * trying to online it.
3269 if (generation != spa->spa_config_generation) {
3270 if (ztest_opts.zo_verbose >= 5) {
3271 (void) printf("vdev configuration has changed, "
3272 "guid %llu, state %llu, expected gen %llu, "
3275 (u_longlong_t)tvd->vdev_state,
3276 (u_longlong_t)generation,
3277 (u_longlong_t)spa->spa_config_generation);
3285 * Traverse the vdev tree calling the supplied function.
3286 * We continue to walk the tree until we either have walked all
3287 * children or we receive a non-NULL return from the callback.
3288 * If a NULL callback is passed, then we just return back the first
3289 * leaf vdev we encounter.
3292 vdev_walk_tree(vdev_t *vd, vdev_t *(*func)(vdev_t *, void *), void *arg)
3296 if (vd->vdev_ops->vdev_op_leaf) {
3300 return (func(vd, arg));
3303 for (c = 0; c < vd->vdev_children; c++) {
3304 vdev_t *cvd = vd->vdev_child[c];
3305 if ((cvd = vdev_walk_tree(cvd, func, arg)) != NULL)
3312 * Verify that dynamic LUN growth works as expected.
3316 ztest_vdev_LUN_growth(ztest_ds_t *zd, uint64_t id)
3318 spa_t *spa = ztest_spa;
3320 metaslab_class_t *mc;
3321 metaslab_group_t *mg;
3322 size_t psize, newsize;
3324 uint64_t old_class_space, new_class_space, old_ms_count, new_ms_count;
3326 mutex_enter(&ztest_vdev_lock);
3327 spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3329 top = ztest_random_vdev_top(spa, B_TRUE);
3331 tvd = spa->spa_root_vdev->vdev_child[top];
3334 old_ms_count = tvd->vdev_ms_count;
3335 old_class_space = metaslab_class_get_space(mc);
3338 * Determine the size of the first leaf vdev associated with
3339 * our top-level device.
3341 vd = vdev_walk_tree(tvd, NULL, NULL);
3342 ASSERT3P(vd, !=, NULL);
3343 ASSERT(vd->vdev_ops->vdev_op_leaf);
3345 psize = vd->vdev_psize;
3348 * We only try to expand the vdev if it's healthy, less than 4x its
3349 * original size, and it has a valid psize.
3351 if (tvd->vdev_state != VDEV_STATE_HEALTHY ||
3352 psize == 0 || psize >= 4 * ztest_opts.zo_vdev_size) {
3353 spa_config_exit(spa, SCL_STATE, spa);
3354 mutex_exit(&ztest_vdev_lock);
3358 newsize = psize + psize / 8;
3359 ASSERT3U(newsize, >, psize);
3361 if (ztest_opts.zo_verbose >= 6) {
3362 (void) printf("Expanding LUN %s from %lu to %lu\n",
3363 vd->vdev_path, (ulong_t)psize, (ulong_t)newsize);
3367 * Growing the vdev is a two step process:
3368 * 1). expand the physical size (i.e. relabel)
3369 * 2). online the vdev to create the new metaslabs
3371 if (vdev_walk_tree(tvd, grow_vdev, &newsize) != NULL ||
3372 vdev_walk_tree(tvd, online_vdev, NULL) != NULL ||
3373 tvd->vdev_state != VDEV_STATE_HEALTHY) {
3374 if (ztest_opts.zo_verbose >= 5) {
3375 (void) printf("Could not expand LUN because "
3376 "the vdev configuration changed.\n");
3378 spa_config_exit(spa, SCL_STATE, spa);
3379 mutex_exit(&ztest_vdev_lock);
3383 spa_config_exit(spa, SCL_STATE, spa);
3386 * Expanding the LUN will update the config asynchronously,
3387 * thus we must wait for the async thread to complete any
3388 * pending tasks before proceeding.
3392 mutex_enter(&spa->spa_async_lock);
3393 done = (spa->spa_async_thread == NULL && !spa->spa_async_tasks);
3394 mutex_exit(&spa->spa_async_lock);
3397 txg_wait_synced(spa_get_dsl(spa), 0);
3398 (void) poll(NULL, 0, 100);
3401 spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3403 tvd = spa->spa_root_vdev->vdev_child[top];
3404 new_ms_count = tvd->vdev_ms_count;
3405 new_class_space = metaslab_class_get_space(mc);
3407 if (tvd->vdev_mg != mg || mg->mg_class != mc) {
3408 if (ztest_opts.zo_verbose >= 5) {
3409 (void) printf("Could not verify LUN expansion due to "
3410 "intervening vdev offline or remove.\n");
3412 spa_config_exit(spa, SCL_STATE, spa);
3413 mutex_exit(&ztest_vdev_lock);
3418 * Make sure we were able to grow the vdev.
3420 if (new_ms_count <= old_ms_count)
3421 fatal(0, "LUN expansion failed: ms_count %llu <= %llu\n",
3422 old_ms_count, new_ms_count);
3425 * Make sure we were able to grow the pool.
3427 if (new_class_space <= old_class_space)
3428 fatal(0, "LUN expansion failed: class_space %llu <= %llu\n",
3429 old_class_space, new_class_space);
3431 if (ztest_opts.zo_verbose >= 5) {
3432 char oldnumbuf[6], newnumbuf[6];
3434 nicenum(old_class_space, oldnumbuf);
3435 nicenum(new_class_space, newnumbuf);
3436 (void) printf("%s grew from %s to %s\n",
3437 spa->spa_name, oldnumbuf, newnumbuf);
3440 spa_config_exit(spa, SCL_STATE, spa);
3441 mutex_exit(&ztest_vdev_lock);
3445 * Verify that dmu_objset_{create,destroy,open,close} work as expected.
3449 ztest_objset_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
3452 * Create the objects common to all ztest datasets.
3454 VERIFY(zap_create_claim(os, ZTEST_DIROBJ,
3455 DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx) == 0);
3459 ztest_dataset_create(char *dsname)
3461 uint64_t zilset = ztest_random(100);
3462 int err = dmu_objset_create(dsname, DMU_OST_OTHER, 0,
3463 ztest_objset_create_cb, NULL);
3465 if (err || zilset < 80)
3468 if (ztest_opts.zo_verbose >= 5)
3469 (void) printf("Setting dataset %s to sync always\n", dsname);
3470 return (ztest_dsl_prop_set_uint64(dsname, ZFS_PROP_SYNC,
3471 ZFS_SYNC_ALWAYS, B_FALSE));
3476 ztest_objset_destroy_cb(const char *name, void *arg)
3479 dmu_object_info_t doi;
3483 * Verify that the dataset contains a directory object.
3485 VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_TRUE, FTAG, &os));
3486 error = dmu_object_info(os, ZTEST_DIROBJ, &doi);
3487 if (error != ENOENT) {
3488 /* We could have crashed in the middle of destroying it */
3490 ASSERT3U(doi.doi_type, ==, DMU_OT_ZAP_OTHER);
3491 ASSERT3S(doi.doi_physical_blocks_512, >=, 0);
3493 dmu_objset_disown(os, FTAG);
3496 * Destroy the dataset.
3498 if (strchr(name, '@') != NULL) {
3499 VERIFY0(dsl_destroy_snapshot(name, B_FALSE));
3501 VERIFY0(dsl_destroy_head(name));
3507 ztest_snapshot_create(char *osname, uint64_t id)
3509 char snapname[ZFS_MAX_DATASET_NAME_LEN];
3512 (void) snprintf(snapname, sizeof (snapname), "%llu", (u_longlong_t)id);
3514 error = dmu_objset_snapshot_one(osname, snapname);
3515 if (error == ENOSPC) {
3516 ztest_record_enospc(FTAG);
3519 if (error != 0 && error != EEXIST) {
3520 fatal(0, "ztest_snapshot_create(%s@%s) = %d", osname,
3527 ztest_snapshot_destroy(char *osname, uint64_t id)
3529 char snapname[ZFS_MAX_DATASET_NAME_LEN];
3532 (void) snprintf(snapname, sizeof (snapname), "%s@%llu", osname,
3535 error = dsl_destroy_snapshot(snapname, B_FALSE);
3536 if (error != 0 && error != ENOENT)
3537 fatal(0, "ztest_snapshot_destroy(%s) = %d", snapname, error);
3543 ztest_dmu_objset_create_destroy(ztest_ds_t *zd, uint64_t id)
3549 char name[ZFS_MAX_DATASET_NAME_LEN];
3553 zdtmp = umem_alloc(sizeof (ztest_ds_t), UMEM_NOFAIL);
3555 (void) rw_rdlock(&ztest_name_lock);
3557 (void) snprintf(name, sizeof (name), "%s/temp_%llu",
3558 ztest_opts.zo_pool, (u_longlong_t)id);
3561 * If this dataset exists from a previous run, process its replay log
3562 * half of the time. If we don't replay it, then dsl_destroy_head()
3563 * (invoked from ztest_objset_destroy_cb()) should just throw it away.
3565 if (ztest_random(2) == 0 &&
3566 dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os) == 0) {
3567 ztest_zd_init(zdtmp, NULL, os);
3568 zil_replay(os, zdtmp, ztest_replay_vector);
3569 ztest_zd_fini(zdtmp);
3570 dmu_objset_disown(os, FTAG);
3574 * There may be an old instance of the dataset we're about to
3575 * create lying around from a previous run. If so, destroy it
3576 * and all of its snapshots.
3578 (void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
3579 DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
3582 * Verify that the destroyed dataset is no longer in the namespace.
3584 VERIFY3U(ENOENT, ==, dmu_objset_own(name, DMU_OST_OTHER, B_TRUE,
3588 * Verify that we can create a new dataset.
3590 error = ztest_dataset_create(name);
3592 if (error == ENOSPC) {
3593 ztest_record_enospc(FTAG);
3596 fatal(0, "dmu_objset_create(%s) = %d", name, error);
3599 VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os));
3601 ztest_zd_init(zdtmp, NULL, os);
3604 * Open the intent log for it.
3606 zilog = zil_open(os, ztest_get_data);
3609 * Put some objects in there, do a little I/O to them,
3610 * and randomly take a couple of snapshots along the way.
3612 iters = ztest_random(5);
3613 for (i = 0; i < iters; i++) {
3614 ztest_dmu_object_alloc_free(zdtmp, id);
3615 if (ztest_random(iters) == 0)
3616 (void) ztest_snapshot_create(name, i);
3620 * Verify that we cannot create an existing dataset.
3622 VERIFY3U(EEXIST, ==,
3623 dmu_objset_create(name, DMU_OST_OTHER, 0, NULL, NULL));
3626 * Verify that we can hold an objset that is also owned.
3628 VERIFY3U(0, ==, dmu_objset_hold(name, FTAG, &os2));
3629 dmu_objset_rele(os2, FTAG);
3632 * Verify that we cannot own an objset that is already owned.
3635 dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os2));
3638 dmu_objset_disown(os, FTAG);
3639 ztest_zd_fini(zdtmp);
3641 (void) rw_unlock(&ztest_name_lock);
3643 umem_free(zdtmp, sizeof (ztest_ds_t));
3647 * Verify that dmu_snapshot_{create,destroy,open,close} work as expected.
3650 ztest_dmu_snapshot_create_destroy(ztest_ds_t *zd, uint64_t id)
3652 (void) rw_rdlock(&ztest_name_lock);
3653 (void) ztest_snapshot_destroy(zd->zd_name, id);
3654 (void) ztest_snapshot_create(zd->zd_name, id);
3655 (void) rw_unlock(&ztest_name_lock);
3659 * Cleanup non-standard snapshots and clones.
3662 ztest_dsl_dataset_cleanup(char *osname, uint64_t id)
3671 snap1name = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
3672 clone1name = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
3673 snap2name = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
3674 clone2name = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
3675 snap3name = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
3677 (void) snprintf(snap1name, ZFS_MAX_DATASET_NAME_LEN,
3678 "%s@s1_%llu", osname, (u_longlong_t)id);
3679 (void) snprintf(clone1name, ZFS_MAX_DATASET_NAME_LEN,
3680 "%s/c1_%llu", osname, (u_longlong_t)id);
3681 (void) snprintf(snap2name, ZFS_MAX_DATASET_NAME_LEN,
3682 "%s@s2_%llu", clone1name, (u_longlong_t)id);
3683 (void) snprintf(clone2name, ZFS_MAX_DATASET_NAME_LEN,
3684 "%s/c2_%llu", osname, (u_longlong_t)id);
3685 (void) snprintf(snap3name, ZFS_MAX_DATASET_NAME_LEN,
3686 "%s@s3_%llu", clone1name, (u_longlong_t)id);
3688 error = dsl_destroy_head(clone2name);
3689 if (error && error != ENOENT)
3690 fatal(0, "dsl_destroy_head(%s) = %d", clone2name, error);
3691 error = dsl_destroy_snapshot(snap3name, B_FALSE);
3692 if (error && error != ENOENT)
3693 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap3name, error);
3694 error = dsl_destroy_snapshot(snap2name, B_FALSE);
3695 if (error && error != ENOENT)
3696 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap2name, error);
3697 error = dsl_destroy_head(clone1name);
3698 if (error && error != ENOENT)
3699 fatal(0, "dsl_destroy_head(%s) = %d", clone1name, error);
3700 error = dsl_destroy_snapshot(snap1name, B_FALSE);
3701 if (error && error != ENOENT)
3702 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap1name, error);
3704 umem_free(snap1name, ZFS_MAX_DATASET_NAME_LEN);
3705 umem_free(clone1name, ZFS_MAX_DATASET_NAME_LEN);
3706 umem_free(snap2name, ZFS_MAX_DATASET_NAME_LEN);
3707 umem_free(clone2name, ZFS_MAX_DATASET_NAME_LEN);
3708 umem_free(snap3name, ZFS_MAX_DATASET_NAME_LEN);
3712 * Verify dsl_dataset_promote handles EBUSY
3715 ztest_dsl_dataset_promote_busy(ztest_ds_t *zd, uint64_t id)
3723 char *osname = zd->zd_name;
3726 snap1name = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
3727 clone1name = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
3728 snap2name = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
3729 clone2name = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
3730 snap3name = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
3732 (void) rw_rdlock(&ztest_name_lock);
3734 ztest_dsl_dataset_cleanup(osname, id);
3736 (void) snprintf(snap1name, ZFS_MAX_DATASET_NAME_LEN,
3737 "%s@s1_%llu", osname, (u_longlong_t)id);
3738 (void) snprintf(clone1name, ZFS_MAX_DATASET_NAME_LEN,
3739 "%s/c1_%llu", osname, (u_longlong_t)id);
3740 (void) snprintf(snap2name, ZFS_MAX_DATASET_NAME_LEN,
3741 "%s@s2_%llu", clone1name, (u_longlong_t)id);
3742 (void) snprintf(clone2name, ZFS_MAX_DATASET_NAME_LEN,
3743 "%s/c2_%llu", osname, (u_longlong_t)id);
3744 (void) snprintf(snap3name, ZFS_MAX_DATASET_NAME_LEN,
3745 "%s@s3_%llu", clone1name, (u_longlong_t)id);
3747 error = dmu_objset_snapshot_one(osname, strchr(snap1name, '@') + 1);
3748 if (error && error != EEXIST) {
3749 if (error == ENOSPC) {
3750 ztest_record_enospc(FTAG);
3753 fatal(0, "dmu_take_snapshot(%s) = %d", snap1name, error);
3756 error = dmu_objset_clone(clone1name, snap1name);
3758 if (error == ENOSPC) {
3759 ztest_record_enospc(FTAG);
3762 fatal(0, "dmu_objset_create(%s) = %d", clone1name, error);
3765 error = dmu_objset_snapshot_one(clone1name, strchr(snap2name, '@') + 1);
3766 if (error && error != EEXIST) {
3767 if (error == ENOSPC) {
3768 ztest_record_enospc(FTAG);
3771 fatal(0, "dmu_open_snapshot(%s) = %d", snap2name, error);
3774 error = dmu_objset_snapshot_one(clone1name, strchr(snap3name, '@') + 1);
3775 if (error && error != EEXIST) {
3776 if (error == ENOSPC) {
3777 ztest_record_enospc(FTAG);
3780 fatal(0, "dmu_open_snapshot(%s) = %d", snap3name, error);
3783 error = dmu_objset_clone(clone2name, snap3name);
3785 if (error == ENOSPC) {
3786 ztest_record_enospc(FTAG);
3789 fatal(0, "dmu_objset_create(%s) = %d", clone2name, error);
3792 error = dmu_objset_own(snap2name, DMU_OST_ANY, B_TRUE, FTAG, &os);
3794 fatal(0, "dmu_objset_own(%s) = %d", snap2name, error);
3795 error = dsl_dataset_promote(clone2name, NULL);
3796 if (error == ENOSPC) {
3797 dmu_objset_disown(os, FTAG);
3798 ztest_record_enospc(FTAG);
3802 fatal(0, "dsl_dataset_promote(%s), %d, not EBUSY", clone2name,
3804 dmu_objset_disown(os, FTAG);
3807 ztest_dsl_dataset_cleanup(osname, id);
3809 (void) rw_unlock(&ztest_name_lock);
3811 umem_free(snap1name, ZFS_MAX_DATASET_NAME_LEN);
3812 umem_free(clone1name, ZFS_MAX_DATASET_NAME_LEN);
3813 umem_free(snap2name, ZFS_MAX_DATASET_NAME_LEN);
3814 umem_free(clone2name, ZFS_MAX_DATASET_NAME_LEN);
3815 umem_free(snap3name, ZFS_MAX_DATASET_NAME_LEN);
3818 #undef OD_ARRAY_SIZE
3819 #define OD_ARRAY_SIZE 4
3822 * Verify that dmu_object_{alloc,free} work as expected.
3825 ztest_dmu_object_alloc_free(ztest_ds_t *zd, uint64_t id)
3832 size = sizeof (ztest_od_t) * OD_ARRAY_SIZE;
3833 od = umem_alloc(size, UMEM_NOFAIL);
3834 batchsize = OD_ARRAY_SIZE;
3836 for (b = 0; b < batchsize; b++)
3837 ztest_od_init(od + b, id, FTAG, b, DMU_OT_UINT64_OTHER,
3841 * Destroy the previous batch of objects, create a new batch,
3842 * and do some I/O on the new objects.
3844 if (ztest_object_init(zd, od, size, B_TRUE) != 0)
3847 while (ztest_random(4 * batchsize) != 0)
3848 ztest_io(zd, od[ztest_random(batchsize)].od_object,
3849 ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
3851 umem_free(od, size);
3854 #undef OD_ARRAY_SIZE
3855 #define OD_ARRAY_SIZE 2
3858 * Verify that dmu_{read,write} work as expected.
3861 ztest_dmu_read_write(ztest_ds_t *zd, uint64_t id)
3866 objset_t *os = zd->zd_os;
3867 size = sizeof (ztest_od_t) * OD_ARRAY_SIZE;
3868 od = umem_alloc(size, UMEM_NOFAIL);
3870 int i, freeit, error;
3872 bufwad_t *packbuf, *bigbuf, *pack, *bigH, *bigT;
3873 uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
3874 uint64_t chunksize = (1000 + ztest_random(1000)) * sizeof (uint64_t);
3875 uint64_t regions = 997;
3876 uint64_t stride = 123456789ULL;
3877 uint64_t width = 40;
3878 int free_percent = 5;
3881 * This test uses two objects, packobj and bigobj, that are always
3882 * updated together (i.e. in the same tx) so that their contents are
3883 * in sync and can be compared. Their contents relate to each other
3884 * in a simple way: packobj is a dense array of 'bufwad' structures,
3885 * while bigobj is a sparse array of the same bufwads. Specifically,
3886 * for any index n, there are three bufwads that should be identical:
3888 * packobj, at offset n * sizeof (bufwad_t)
3889 * bigobj, at the head of the nth chunk
3890 * bigobj, at the tail of the nth chunk
3892 * The chunk size is arbitrary. It doesn't have to be a power of two,
3893 * and it doesn't have any relation to the object blocksize.
3894 * The only requirement is that it can hold at least two bufwads.
3896 * Normally, we write the bufwad to each of these locations.
3897 * However, free_percent of the time we instead write zeroes to
3898 * packobj and perform a dmu_free_range() on bigobj. By comparing
3899 * bigobj to packobj, we can verify that the DMU is correctly
3900 * tracking which parts of an object are allocated and free,
3901 * and that the contents of the allocated blocks are correct.
3905 * Read the directory info. If it's the first time, set things up.
3907 ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0, chunksize);
3908 ztest_od_init(od + 1, id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, 0,
3911 if (ztest_object_init(zd, od, size, B_FALSE) != 0) {
3912 umem_free(od, size);
3916 bigobj = od[0].od_object;
3917 packobj = od[1].od_object;
3918 chunksize = od[0].od_gen;
3919 ASSERT(chunksize == od[1].od_gen);
3922 * Prefetch a random chunk of the big object.
3923 * Our aim here is to get some async reads in flight
3924 * for blocks that we may free below; the DMU should
3925 * handle this race correctly.
3927 n = ztest_random(regions) * stride + ztest_random(width);
3928 s = 1 + ztest_random(2 * width - 1);
3929 dmu_prefetch(os, bigobj, 0, n * chunksize, s * chunksize,
3930 ZIO_PRIORITY_SYNC_READ);
3933 * Pick a random index and compute the offsets into packobj and bigobj.
3935 n = ztest_random(regions) * stride + ztest_random(width);
3936 s = 1 + ztest_random(width - 1);
3938 packoff = n * sizeof (bufwad_t);
3939 packsize = s * sizeof (bufwad_t);
3941 bigoff = n * chunksize;
3942 bigsize = s * chunksize;
3944 packbuf = umem_alloc(packsize, UMEM_NOFAIL);
3945 bigbuf = umem_alloc(bigsize, UMEM_NOFAIL);
3948 * free_percent of the time, free a range of bigobj rather than
3951 freeit = (ztest_random(100) < free_percent);
3954 * Read the current contents of our objects.
3956 error = dmu_read(os, packobj, packoff, packsize, packbuf,
3959 error = dmu_read(os, bigobj, bigoff, bigsize, bigbuf,
3964 * Get a tx for the mods to both packobj and bigobj.
3966 tx = dmu_tx_create(os);
3968 dmu_tx_hold_write(tx, packobj, packoff, packsize);
3971 dmu_tx_hold_free(tx, bigobj, bigoff, bigsize);
3973 dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
3975 /* This accounts for setting the checksum/compression. */
3976 dmu_tx_hold_bonus(tx, bigobj);
3978 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3980 umem_free(packbuf, packsize);
3981 umem_free(bigbuf, bigsize);
3982 umem_free(od, size);
3986 enum zio_checksum cksum;
3988 cksum = (enum zio_checksum)
3989 ztest_random_dsl_prop(ZFS_PROP_CHECKSUM);
3990 } while (cksum >= ZIO_CHECKSUM_LEGACY_FUNCTIONS);
3991 dmu_object_set_checksum(os, bigobj, cksum, tx);
3993 enum zio_compress comp;
3995 comp = (enum zio_compress)
3996 ztest_random_dsl_prop(ZFS_PROP_COMPRESSION);
3997 } while (comp >= ZIO_COMPRESS_LEGACY_FUNCTIONS);
3998 dmu_object_set_compress(os, bigobj, comp, tx);
4001 * For each index from n to n + s, verify that the existing bufwad
4002 * in packobj matches the bufwads at the head and tail of the
4003 * corresponding chunk in bigobj. Then update all three bufwads
4004 * with the new values we want to write out.
4006 for (i = 0; i < s; i++) {
4008 pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
4010 bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
4012 bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
4014 ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
4015 ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
4017 if (pack->bw_txg > txg)
4018 fatal(0, "future leak: got %llx, open txg is %llx",
4021 if (pack->bw_data != 0 && pack->bw_index != n + i)
4022 fatal(0, "wrong index: got %llx, wanted %llx+%llx",
4023 pack->bw_index, n, i);
4025 if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
4026 fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
4028 if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
4029 fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
4032 bzero(pack, sizeof (bufwad_t));
4034 pack->bw_index = n + i;
4036 pack->bw_data = 1 + ztest_random(-2ULL);
4043 * We've verified all the old bufwads, and made new ones.
4044 * Now write them out.
4046 dmu_write(os, packobj, packoff, packsize, packbuf, tx);
4049 if (ztest_opts.zo_verbose >= 7) {
4050 (void) printf("freeing offset %llx size %llx"
4052 (u_longlong_t)bigoff,
4053 (u_longlong_t)bigsize,
4056 VERIFY(0 == dmu_free_range(os, bigobj, bigoff, bigsize, tx));
4058 if (ztest_opts.zo_verbose >= 7) {
4059 (void) printf("writing offset %llx size %llx"
4061 (u_longlong_t)bigoff,
4062 (u_longlong_t)bigsize,
4065 dmu_write(os, bigobj, bigoff, bigsize, bigbuf, tx);
4071 * Sanity check the stuff we just wrote.
4074 void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
4075 void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
4077 VERIFY(0 == dmu_read(os, packobj, packoff,
4078 packsize, packcheck, DMU_READ_PREFETCH));
4079 VERIFY(0 == dmu_read(os, bigobj, bigoff,
4080 bigsize, bigcheck, DMU_READ_PREFETCH));
4082 ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
4083 ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
4085 umem_free(packcheck, packsize);
4086 umem_free(bigcheck, bigsize);
4089 umem_free(packbuf, packsize);
4090 umem_free(bigbuf, bigsize);
4091 umem_free(od, size);
4095 compare_and_update_pbbufs(uint64_t s, bufwad_t *packbuf, bufwad_t *bigbuf,
4096 uint64_t bigsize, uint64_t n, uint64_t chunksize, uint64_t txg)
4104 * For each index from n to n + s, verify that the existing bufwad
4105 * in packobj matches the bufwads at the head and tail of the
4106 * corresponding chunk in bigobj. Then update all three bufwads
4107 * with the new values we want to write out.
4109 for (i = 0; i < s; i++) {
4111 pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
4113 bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
4115 bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
4117 ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
4118 ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
4120 if (pack->bw_txg > txg)
4121 fatal(0, "future leak: got %llx, open txg is %llx",
4124 if (pack->bw_data != 0 && pack->bw_index != n + i)
4125 fatal(0, "wrong index: got %llx, wanted %llx+%llx",
4126 pack->bw_index, n, i);
4128 if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
4129 fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
4131 if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
4132 fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
4134 pack->bw_index = n + i;
4136 pack->bw_data = 1 + ztest_random(-2ULL);
4143 #undef OD_ARRAY_SIZE
4144 #define OD_ARRAY_SIZE 2
4147 ztest_dmu_read_write_zcopy(ztest_ds_t *zd, uint64_t id)
4149 objset_t *os = zd->zd_os;
4156 bufwad_t *packbuf, *bigbuf;
4157 uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
4158 uint64_t blocksize = ztest_random_blocksize();
4159 uint64_t chunksize = blocksize;
4160 uint64_t regions = 997;
4161 uint64_t stride = 123456789ULL;
4163 dmu_buf_t *bonus_db;
4164 arc_buf_t **bigbuf_arcbufs;
4165 dmu_object_info_t doi;
4167 size = sizeof (ztest_od_t) * OD_ARRAY_SIZE;
4168 od = umem_alloc(size, UMEM_NOFAIL);
4171 * This test uses two objects, packobj and bigobj, that are always
4172 * updated together (i.e. in the same tx) so that their contents are
4173 * in sync and can be compared. Their contents relate to each other
4174 * in a simple way: packobj is a dense array of 'bufwad' structures,
4175 * while bigobj is a sparse array of the same bufwads. Specifically,
4176 * for any index n, there are three bufwads that should be identical:
4178 * packobj, at offset n * sizeof (bufwad_t)
4179 * bigobj, at the head of the nth chunk
4180 * bigobj, at the tail of the nth chunk
4182 * The chunk size is set equal to bigobj block size so that
4183 * dmu_assign_arcbuf() can be tested for object updates.
4187 * Read the directory info. If it's the first time, set things up.
4189 ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0, 0);
4190 ztest_od_init(od + 1, id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, 0,
4194 if (ztest_object_init(zd, od, size, B_FALSE) != 0) {
4195 umem_free(od, size);
4199 bigobj = od[0].od_object;
4200 packobj = od[1].od_object;
4201 blocksize = od[0].od_blocksize;
4202 chunksize = blocksize;
4203 ASSERT(chunksize == od[1].od_gen);
4205 VERIFY(dmu_object_info(os, bigobj, &doi) == 0);
4206 VERIFY(ISP2(doi.doi_data_block_size));
4207 VERIFY(chunksize == doi.doi_data_block_size);
4208 VERIFY(chunksize >= 2 * sizeof (bufwad_t));
4211 * Pick a random index and compute the offsets into packobj and bigobj.
4213 n = ztest_random(regions) * stride + ztest_random(width);
4214 s = 1 + ztest_random(width - 1);
4216 packoff = n * sizeof (bufwad_t);
4217 packsize = s * sizeof (bufwad_t);
4219 bigoff = n * chunksize;
4220 bigsize = s * chunksize;
4222 packbuf = umem_zalloc(packsize, UMEM_NOFAIL);
4223 bigbuf = umem_zalloc(bigsize, UMEM_NOFAIL);
4225 VERIFY3U(0, ==, dmu_bonus_hold(os, bigobj, FTAG, &bonus_db));
4227 bigbuf_arcbufs = umem_zalloc(2 * s * sizeof (arc_buf_t *), UMEM_NOFAIL);
4230 * Iteration 0 test zcopy for DB_UNCACHED dbufs.
4231 * Iteration 1 test zcopy to already referenced dbufs.
4232 * Iteration 2 test zcopy to dirty dbuf in the same txg.
4233 * Iteration 3 test zcopy to dbuf dirty in previous txg.
4234 * Iteration 4 test zcopy when dbuf is no longer dirty.
4235 * Iteration 5 test zcopy when it can't be done.
4236 * Iteration 6 one more zcopy write.
4238 for (i = 0; i < 7; i++) {
4243 * In iteration 5 (i == 5) use arcbufs
4244 * that don't match bigobj blksz to test
4245 * dmu_assign_arcbuf() when it can't directly
4246 * assign an arcbuf to a dbuf.
4248 for (j = 0; j < s; j++) {
4249 if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) {
4251 dmu_request_arcbuf(bonus_db, chunksize);
4253 bigbuf_arcbufs[2 * j] =
4254 dmu_request_arcbuf(bonus_db, chunksize / 2);
4255 bigbuf_arcbufs[2 * j + 1] =
4256 dmu_request_arcbuf(bonus_db, chunksize / 2);
4261 * Get a tx for the mods to both packobj and bigobj.
4263 tx = dmu_tx_create(os);
4265 dmu_tx_hold_write(tx, packobj, packoff, packsize);
4266 dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
4268 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4270 umem_free(packbuf, packsize);
4271 umem_free(bigbuf, bigsize);
4272 for (j = 0; j < s; j++) {
4274 chunksize < (SPA_MINBLOCKSIZE * 2)) {
4275 dmu_return_arcbuf(bigbuf_arcbufs[j]);
4278 bigbuf_arcbufs[2 * j]);
4280 bigbuf_arcbufs[2 * j + 1]);
4283 umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
4284 umem_free(od, size);
4285 dmu_buf_rele(bonus_db, FTAG);
4290 * 50% of the time don't read objects in the 1st iteration to
4291 * test dmu_assign_arcbuf() for the case when there're no
4292 * existing dbufs for the specified offsets.
4294 if (i != 0 || ztest_random(2) != 0) {
4295 error = dmu_read(os, packobj, packoff,
4296 packsize, packbuf, DMU_READ_PREFETCH);
4298 error = dmu_read(os, bigobj, bigoff, bigsize,
4299 bigbuf, DMU_READ_PREFETCH);
4302 compare_and_update_pbbufs(s, packbuf, bigbuf, bigsize,
4306 * We've verified all the old bufwads, and made new ones.
4307 * Now write them out.
4309 dmu_write(os, packobj, packoff, packsize, packbuf, tx);
4310 if (ztest_opts.zo_verbose >= 7) {
4311 (void) printf("writing offset %llx size %llx"
4313 (u_longlong_t)bigoff,
4314 (u_longlong_t)bigsize,
4317 for (off = bigoff, j = 0; j < s; j++, off += chunksize) {
4319 if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) {
4320 bcopy((caddr_t)bigbuf + (off - bigoff),
4321 bigbuf_arcbufs[j]->b_data, chunksize);
4323 bcopy((caddr_t)bigbuf + (off - bigoff),
4324 bigbuf_arcbufs[2 * j]->b_data,
4326 bcopy((caddr_t)bigbuf + (off - bigoff) +
4328 bigbuf_arcbufs[2 * j + 1]->b_data,
4333 VERIFY(dmu_buf_hold(os, bigobj, off,
4334 FTAG, &dbt, DMU_READ_NO_PREFETCH) == 0);
4336 if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) {
4337 dmu_assign_arcbuf(bonus_db, off,
4338 bigbuf_arcbufs[j], tx);
4340 dmu_assign_arcbuf(bonus_db, off,
4341 bigbuf_arcbufs[2 * j], tx);
4342 dmu_assign_arcbuf(bonus_db,
4343 off + chunksize / 2,
4344 bigbuf_arcbufs[2 * j + 1], tx);
4347 dmu_buf_rele(dbt, FTAG);
4353 * Sanity check the stuff we just wrote.
4356 void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
4357 void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
4359 VERIFY(0 == dmu_read(os, packobj, packoff,
4360 packsize, packcheck, DMU_READ_PREFETCH));
4361 VERIFY(0 == dmu_read(os, bigobj, bigoff,
4362 bigsize, bigcheck, DMU_READ_PREFETCH));
4364 ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
4365 ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
4367 umem_free(packcheck, packsize);
4368 umem_free(bigcheck, bigsize);
4371 txg_wait_open(dmu_objset_pool(os), 0);
4372 } else if (i == 3) {
4373 txg_wait_synced(dmu_objset_pool(os), 0);
4377 dmu_buf_rele(bonus_db, FTAG);
4378 umem_free(packbuf, packsize);
4379 umem_free(bigbuf, bigsize);
4380 umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
4381 umem_free(od, size);
4386 ztest_dmu_write_parallel(ztest_ds_t *zd, uint64_t id)
4390 od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
4391 uint64_t offset = (1ULL << (ztest_random(20) + 43)) +
4392 (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
4395 * Have multiple threads write to large offsets in an object
4396 * to verify that parallel writes to an object -- even to the
4397 * same blocks within the object -- doesn't cause any trouble.
4399 ztest_od_init(od, ID_PARALLEL, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0, 0);
4401 if (ztest_object_init(zd, od, sizeof (ztest_od_t), B_FALSE) != 0)
4404 while (ztest_random(10) != 0)
4405 ztest_io(zd, od->od_object, offset);
4407 umem_free(od, sizeof (ztest_od_t));
4411 ztest_dmu_prealloc(ztest_ds_t *zd, uint64_t id)
4414 uint64_t offset = (1ULL << (ztest_random(4) + SPA_MAXBLOCKSHIFT)) +
4415 (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
4416 uint64_t count = ztest_random(20) + 1;
4417 uint64_t blocksize = ztest_random_blocksize();
4420 od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
4422 ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0, 0);
4424 if (ztest_object_init(zd, od, sizeof (ztest_od_t),
4425 !ztest_random(2)) != 0) {
4426 umem_free(od, sizeof (ztest_od_t));
4430 if (ztest_truncate(zd, od->od_object, offset, count * blocksize) != 0) {
4431 umem_free(od, sizeof (ztest_od_t));
4435 ztest_prealloc(zd, od->od_object, offset, count * blocksize);
4437 data = umem_zalloc(blocksize, UMEM_NOFAIL);
4439 while (ztest_random(count) != 0) {
4440 uint64_t randoff = offset + (ztest_random(count) * blocksize);
4441 if (ztest_write(zd, od->od_object, randoff, blocksize,
4444 while (ztest_random(4) != 0)
4445 ztest_io(zd, od->od_object, randoff);
4448 umem_free(data, blocksize);
4449 umem_free(od, sizeof (ztest_od_t));
4453 * Verify that zap_{create,destroy,add,remove,update} work as expected.
4455 #define ZTEST_ZAP_MIN_INTS 1
4456 #define ZTEST_ZAP_MAX_INTS 4
4457 #define ZTEST_ZAP_MAX_PROPS 1000
4460 ztest_zap(ztest_ds_t *zd, uint64_t id)
4462 objset_t *os = zd->zd_os;
4465 uint64_t txg, last_txg;
4466 uint64_t value[ZTEST_ZAP_MAX_INTS];
4467 uint64_t zl_ints, zl_intsize, prop;
4470 char propname[100], txgname[100];
4472 char *hc[2] = { "s.acl.h", ".s.open.h.hyLZlg" };
4474 od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
4475 ztest_od_init(od, id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0, 0);
4477 if (ztest_object_init(zd, od, sizeof (ztest_od_t),
4478 !ztest_random(2)) != 0)
4481 object = od->od_object;
4484 * Generate a known hash collision, and verify that
4485 * we can lookup and remove both entries.
4487 tx = dmu_tx_create(os);
4488 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4489 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4492 for (i = 0; i < 2; i++) {
4494 VERIFY3U(0, ==, zap_add(os, object, hc[i], sizeof (uint64_t),
4497 for (i = 0; i < 2; i++) {
4498 VERIFY3U(EEXIST, ==, zap_add(os, object, hc[i],
4499 sizeof (uint64_t), 1, &value[i], tx));
4501 zap_length(os, object, hc[i], &zl_intsize, &zl_ints));
4502 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4503 ASSERT3U(zl_ints, ==, 1);
4505 for (i = 0; i < 2; i++) {
4506 VERIFY3U(0, ==, zap_remove(os, object, hc[i], tx));
4511 * Generate a buch of random entries.
4513 ints = MAX(ZTEST_ZAP_MIN_INTS, object % ZTEST_ZAP_MAX_INTS);
4515 prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
4516 (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
4517 (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
4518 bzero(value, sizeof (value));
4522 * If these zap entries already exist, validate their contents.
4524 error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
4526 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4527 ASSERT3U(zl_ints, ==, 1);
4529 VERIFY(zap_lookup(os, object, txgname, zl_intsize,
4530 zl_ints, &last_txg) == 0);
4532 VERIFY(zap_length(os, object, propname, &zl_intsize,
4535 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4536 ASSERT3U(zl_ints, ==, ints);
4538 VERIFY(zap_lookup(os, object, propname, zl_intsize,
4539 zl_ints, value) == 0);
4541 for (i = 0; i < ints; i++) {
4542 ASSERT3U(value[i], ==, last_txg + object + i);
4545 ASSERT3U(error, ==, ENOENT);
4549 * Atomically update two entries in our zap object.
4550 * The first is named txg_%llu, and contains the txg
4551 * in which the property was last updated. The second
4552 * is named prop_%llu, and the nth element of its value
4553 * should be txg + object + n.
4555 tx = dmu_tx_create(os);
4556 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4557 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4562 fatal(0, "zap future leak: old %llu new %llu", last_txg, txg);
4564 for (i = 0; i < ints; i++)
4565 value[i] = txg + object + i;
4567 VERIFY3U(0, ==, zap_update(os, object, txgname, sizeof (uint64_t),
4569 VERIFY3U(0, ==, zap_update(os, object, propname, sizeof (uint64_t),
4575 * Remove a random pair of entries.
4577 prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
4578 (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
4579 (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
4581 error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
4583 if (error == ENOENT)
4588 tx = dmu_tx_create(os);
4589 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4590 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4593 VERIFY3U(0, ==, zap_remove(os, object, txgname, tx));
4594 VERIFY3U(0, ==, zap_remove(os, object, propname, tx));
4597 umem_free(od, sizeof (ztest_od_t));
4601 * Testcase to test the upgrading of a microzap to fatzap.
4604 ztest_fzap(ztest_ds_t *zd, uint64_t id)
4606 objset_t *os = zd->zd_os;
4608 uint64_t object, txg;
4611 od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
4612 ztest_od_init(od, id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0, 0);
4614 if (ztest_object_init(zd, od, sizeof (ztest_od_t),
4615 !ztest_random(2)) != 0)
4617 object = od->od_object;
4620 * Add entries to this ZAP and make sure it spills over
4621 * and gets upgraded to a fatzap. Also, since we are adding
4622 * 2050 entries we should see ptrtbl growth and leaf-block split.
4624 for (i = 0; i < 2050; i++) {
4625 char name[ZFS_MAX_DATASET_NAME_LEN];
4630 (void) snprintf(name, sizeof (name), "fzap-%llu-%llu",
4631 (u_longlong_t)id, (u_longlong_t)value);
4633 tx = dmu_tx_create(os);
4634 dmu_tx_hold_zap(tx, object, B_TRUE, name);
4635 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4638 error = zap_add(os, object, name, sizeof (uint64_t), 1,
4640 ASSERT(error == 0 || error == EEXIST);
4644 umem_free(od, sizeof (ztest_od_t));
4649 ztest_zap_parallel(ztest_ds_t *zd, uint64_t id)
4651 objset_t *os = zd->zd_os;
4653 uint64_t txg, object, count, wsize, wc, zl_wsize, zl_wc;
4655 int i, namelen, error;
4656 int micro = ztest_random(2);
4657 char name[20], string_value[20];
4660 od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
4661 ztest_od_init(od, ID_PARALLEL, FTAG, micro, DMU_OT_ZAP_OTHER, 0, 0, 0);
4663 if (ztest_object_init(zd, od, sizeof (ztest_od_t), B_FALSE) != 0) {
4664 umem_free(od, sizeof (ztest_od_t));
4668 object = od->od_object;
4671 * Generate a random name of the form 'xxx.....' where each
4672 * x is a random printable character and the dots are dots.
4673 * There are 94 such characters, and the name length goes from
4674 * 6 to 20, so there are 94^3 * 15 = 12,458,760 possible names.
4676 namelen = ztest_random(sizeof (name) - 5) + 5 + 1;
4678 for (i = 0; i < 3; i++)
4679 name[i] = '!' + ztest_random('~' - '!' + 1);
4680 for (; i < namelen - 1; i++)
4684 if ((namelen & 1) || micro) {
4685 wsize = sizeof (txg);
4691 data = string_value;
4695 VERIFY0(zap_count(os, object, &count));
4696 ASSERT(count != -1ULL);
4699 * Select an operation: length, lookup, add, update, remove.
4701 i = ztest_random(5);
4704 tx = dmu_tx_create(os);
4705 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4706 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4708 umem_free(od, sizeof (ztest_od_t));
4711 bcopy(name, string_value, namelen);
4715 bzero(string_value, namelen);
4721 error = zap_length(os, object, name, &zl_wsize, &zl_wc);
4723 ASSERT3U(wsize, ==, zl_wsize);
4724 ASSERT3U(wc, ==, zl_wc);
4726 ASSERT3U(error, ==, ENOENT);
4731 error = zap_lookup(os, object, name, wsize, wc, data);
4733 if (data == string_value &&
4734 bcmp(name, data, namelen) != 0)
4735 fatal(0, "name '%s' != val '%s' len %d",
4736 name, data, namelen);
4738 ASSERT3U(error, ==, ENOENT);
4743 error = zap_add(os, object, name, wsize, wc, data, tx);
4744 ASSERT(error == 0 || error == EEXIST);
4748 VERIFY(zap_update(os, object, name, wsize, wc, data, tx) == 0);
4752 error = zap_remove(os, object, name, tx);
4753 ASSERT(error == 0 || error == ENOENT);
4760 umem_free(od, sizeof (ztest_od_t));
4764 * Commit callback data.
4766 typedef struct ztest_cb_data {
4767 list_node_t zcd_node;
4769 int zcd_expected_err;
4770 boolean_t zcd_added;
4771 boolean_t zcd_called;
4775 /* This is the actual commit callback function */
4777 ztest_commit_callback(void *arg, int error)
4779 ztest_cb_data_t *data = arg;
4780 uint64_t synced_txg;
4782 VERIFY(data != NULL);
4783 VERIFY3S(data->zcd_expected_err, ==, error);
4784 VERIFY(!data->zcd_called);
4786 synced_txg = spa_last_synced_txg(data->zcd_spa);
4787 if (data->zcd_txg > synced_txg)
4788 fatal(0, "commit callback of txg %" PRIu64 " called prematurely"
4789 ", last synced txg = %" PRIu64 "\n", data->zcd_txg,
4792 data->zcd_called = B_TRUE;
4794 if (error == ECANCELED) {
4795 ASSERT0(data->zcd_txg);
4796 ASSERT(!data->zcd_added);
4799 * The private callback data should be destroyed here, but
4800 * since we are going to check the zcd_called field after
4801 * dmu_tx_abort(), we will destroy it there.
4806 ASSERT(data->zcd_added);
4807 ASSERT3U(data->zcd_txg, !=, 0);
4809 (void) mutex_enter(&zcl.zcl_callbacks_lock);
4811 /* See if this cb was called more quickly */
4812 if ((synced_txg - data->zcd_txg) < zc_min_txg_delay)
4813 zc_min_txg_delay = synced_txg - data->zcd_txg;
4815 /* Remove our callback from the list */
4816 list_remove(&zcl.zcl_callbacks, data);
4818 (void) mutex_exit(&zcl.zcl_callbacks_lock);
4820 umem_free(data, sizeof (ztest_cb_data_t));
4823 /* Allocate and initialize callback data structure */
4824 static ztest_cb_data_t *
4825 ztest_create_cb_data(objset_t *os, uint64_t txg)
4827 ztest_cb_data_t *cb_data;
4829 cb_data = umem_zalloc(sizeof (ztest_cb_data_t), UMEM_NOFAIL);
4831 cb_data->zcd_txg = txg;
4832 cb_data->zcd_spa = dmu_objset_spa(os);
4833 list_link_init(&cb_data->zcd_node);
4839 * Commit callback test.
4842 ztest_dmu_commit_callbacks(ztest_ds_t *zd, uint64_t id)
4844 objset_t *os = zd->zd_os;
4847 ztest_cb_data_t *cb_data[3], *tmp_cb;
4848 uint64_t old_txg, txg;
4851 od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
4852 ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0, 0);
4854 if (ztest_object_init(zd, od, sizeof (ztest_od_t), B_FALSE) != 0) {
4855 umem_free(od, sizeof (ztest_od_t));
4859 tx = dmu_tx_create(os);
4861 cb_data[0] = ztest_create_cb_data(os, 0);
4862 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[0]);
4864 dmu_tx_hold_write(tx, od->od_object, 0, sizeof (uint64_t));
4866 /* Every once in a while, abort the transaction on purpose */
4867 if (ztest_random(100) == 0)
4871 error = dmu_tx_assign(tx, TXG_NOWAIT);
4873 txg = error ? 0 : dmu_tx_get_txg(tx);
4875 cb_data[0]->zcd_txg = txg;
4876 cb_data[1] = ztest_create_cb_data(os, txg);
4877 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[1]);
4881 * It's not a strict requirement to call the registered
4882 * callbacks from inside dmu_tx_abort(), but that's what
4883 * it's supposed to happen in the current implementation
4884 * so we will check for that.
4886 for (i = 0; i < 2; i++) {
4887 cb_data[i]->zcd_expected_err = ECANCELED;
4888 VERIFY(!cb_data[i]->zcd_called);
4893 for (i = 0; i < 2; i++) {
4894 VERIFY(cb_data[i]->zcd_called);
4895 umem_free(cb_data[i], sizeof (ztest_cb_data_t));
4898 umem_free(od, sizeof (ztest_od_t));
4902 cb_data[2] = ztest_create_cb_data(os, txg);
4903 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[2]);
4906 * Read existing data to make sure there isn't a future leak.
4908 VERIFY(0 == dmu_read(os, od->od_object, 0, sizeof (uint64_t),
4909 &old_txg, DMU_READ_PREFETCH));
4912 fatal(0, "future leak: got %" PRIu64 ", open txg is %" PRIu64,
4915 dmu_write(os, od->od_object, 0, sizeof (uint64_t), &txg, tx);
4917 (void) mutex_enter(&zcl.zcl_callbacks_lock);
4920 * Since commit callbacks don't have any ordering requirement and since
4921 * it is theoretically possible for a commit callback to be called
4922 * after an arbitrary amount of time has elapsed since its txg has been
4923 * synced, it is difficult to reliably determine whether a commit
4924 * callback hasn't been called due to high load or due to a flawed
4927 * In practice, we will assume that if after a certain number of txgs a
4928 * commit callback hasn't been called, then most likely there's an
4929 * implementation bug..
4931 tmp_cb = list_head(&zcl.zcl_callbacks);
4932 if (tmp_cb != NULL &&
4933 tmp_cb->zcd_txg + ZTEST_COMMIT_CB_THRESH < txg) {
4934 fatal(0, "Commit callback threshold exceeded, oldest txg: %"
4935 PRIu64 ", open txg: %" PRIu64 "\n", tmp_cb->zcd_txg, txg);
4939 * Let's find the place to insert our callbacks.
4941 * Even though the list is ordered by txg, it is possible for the
4942 * insertion point to not be the end because our txg may already be
4943 * quiescing at this point and other callbacks in the open txg
4944 * (from other objsets) may have sneaked in.
4946 tmp_cb = list_tail(&zcl.zcl_callbacks);
4947 while (tmp_cb != NULL && tmp_cb->zcd_txg > txg)
4948 tmp_cb = list_prev(&zcl.zcl_callbacks, tmp_cb);
4950 /* Add the 3 callbacks to the list */
4951 for (i = 0; i < 3; i++) {
4953 list_insert_head(&zcl.zcl_callbacks, cb_data[i]);
4955 list_insert_after(&zcl.zcl_callbacks, tmp_cb,
4958 cb_data[i]->zcd_added = B_TRUE;
4959 VERIFY(!cb_data[i]->zcd_called);
4961 tmp_cb = cb_data[i];
4966 (void) mutex_exit(&zcl.zcl_callbacks_lock);
4970 umem_free(od, sizeof (ztest_od_t));
4974 * Visit each object in the dataset. Verify that its properties
4975 * are consistent what was stored in the block tag when it was created,
4976 * and that its unused bonus buffer space has not been overwritten.
4979 ztest_verify_dnode_bt(ztest_ds_t *zd, uint64_t id)
4981 objset_t *os = zd->zd_os;
4985 for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, 0)) {
4986 ztest_block_tag_t *bt = NULL;
4987 dmu_object_info_t doi;
4990 if (dmu_bonus_hold(os, obj, FTAG, &db) != 0)
4993 dmu_object_info_from_db(db, &doi);
4994 if (doi.doi_bonus_size >= sizeof (*bt))
4995 bt = ztest_bt_bonus(db);
4997 if (bt && bt->bt_magic == BT_MAGIC) {
4998 ztest_bt_verify(bt, os, obj, doi.doi_dnodesize,
4999 bt->bt_offset, bt->bt_gen, bt->bt_txg,
5001 ztest_verify_unused_bonus(db, bt, obj, os, bt->bt_gen);
5004 dmu_buf_rele(db, FTAG);
5010 ztest_dsl_prop_get_set(ztest_ds_t *zd, uint64_t id)
5012 zfs_prop_t proplist[] = {
5014 ZFS_PROP_COMPRESSION,
5020 (void) rw_rdlock(&ztest_name_lock);
5022 for (p = 0; p < sizeof (proplist) / sizeof (proplist[0]); p++)
5023 (void) ztest_dsl_prop_set_uint64(zd->zd_name, proplist[p],
5024 ztest_random_dsl_prop(proplist[p]), (int)ztest_random(2));
5026 VERIFY0(ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_RECORDSIZE,
5027 ztest_random_blocksize(), (int)ztest_random(2)));
5029 (void) rw_unlock(&ztest_name_lock);
5034 ztest_spa_prop_get_set(ztest_ds_t *zd, uint64_t id)
5036 nvlist_t *props = NULL;
5038 (void) rw_rdlock(&ztest_name_lock);
5040 (void) ztest_spa_prop_set_uint64(ZPOOL_PROP_DEDUPDITTO,
5041 ZIO_DEDUPDITTO_MIN + ztest_random(ZIO_DEDUPDITTO_MIN));
5043 VERIFY0(spa_prop_get(ztest_spa, &props));
5045 if (ztest_opts.zo_verbose >= 6)
5046 dump_nvlist(props, 4);
5050 (void) rw_unlock(&ztest_name_lock);
5054 user_release_one(const char *snapname, const char *holdname)
5056 nvlist_t *snaps, *holds;
5059 snaps = fnvlist_alloc();
5060 holds = fnvlist_alloc();
5061 fnvlist_add_boolean(holds, holdname);
5062 fnvlist_add_nvlist(snaps, snapname, holds);
5063 fnvlist_free(holds);
5064 error = dsl_dataset_user_release(snaps, NULL);
5065 fnvlist_free(snaps);
5070 * Test snapshot hold/release and deferred destroy.
5073 ztest_dmu_snapshot_hold(ztest_ds_t *zd, uint64_t id)
5076 objset_t *os = zd->zd_os;
5080 char clonename[100];
5082 char osname[ZFS_MAX_DATASET_NAME_LEN];
5085 (void) rw_rdlock(&ztest_name_lock);
5087 dmu_objset_name(os, osname);
5089 (void) snprintf(snapname, sizeof (snapname), "sh1_%llu",
5091 (void) snprintf(fullname, sizeof (fullname), "%s@%s", osname, snapname);
5092 (void) snprintf(clonename, sizeof (clonename),
5093 "%s/ch1_%llu", osname, (u_longlong_t)id);
5094 (void) snprintf(tag, sizeof (tag), "tag_%llu", (u_longlong_t)id);
5097 * Clean up from any previous run.
5099 error = dsl_destroy_head(clonename);
5100 if (error != ENOENT)
5102 error = user_release_one(fullname, tag);
5103 if (error != ESRCH && error != ENOENT)
5105 error = dsl_destroy_snapshot(fullname, B_FALSE);
5106 if (error != ENOENT)
5110 * Create snapshot, clone it, mark snap for deferred destroy,
5111 * destroy clone, verify snap was also destroyed.
5113 error = dmu_objset_snapshot_one(osname, snapname);
5115 if (error == ENOSPC) {
5116 ztest_record_enospc("dmu_objset_snapshot");
5119 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
5122 error = dmu_objset_clone(clonename, fullname);
5124 if (error == ENOSPC) {
5125 ztest_record_enospc("dmu_objset_clone");
5128 fatal(0, "dmu_objset_clone(%s) = %d", clonename, error);
5131 error = dsl_destroy_snapshot(fullname, B_TRUE);
5133 fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
5137 error = dsl_destroy_head(clonename);
5139 fatal(0, "dsl_destroy_head(%s) = %d", clonename, error);
5141 error = dmu_objset_hold(fullname, FTAG, &origin);
5142 if (error != ENOENT)
5143 fatal(0, "dmu_objset_hold(%s) = %d", fullname, error);
5146 * Create snapshot, add temporary hold, verify that we can't
5147 * destroy a held snapshot, mark for deferred destroy,
5148 * release hold, verify snapshot was destroyed.
5150 error = dmu_objset_snapshot_one(osname, snapname);
5152 if (error == ENOSPC) {
5153 ztest_record_enospc("dmu_objset_snapshot");
5156 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
5159 holds = fnvlist_alloc();
5160 fnvlist_add_string(holds, fullname, tag);
5161 error = dsl_dataset_user_hold(holds, 0, NULL);
5162 fnvlist_free(holds);
5164 if (error == ENOSPC) {
5165 ztest_record_enospc("dsl_dataset_user_hold");
5168 fatal(0, "dsl_dataset_user_hold(%s, %s) = %u",
5169 fullname, tag, error);
5172 error = dsl_destroy_snapshot(fullname, B_FALSE);
5173 if (error != EBUSY) {
5174 fatal(0, "dsl_destroy_snapshot(%s, B_FALSE) = %d",
5178 error = dsl_destroy_snapshot(fullname, B_TRUE);
5180 fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
5184 error = user_release_one(fullname, tag);
5186 fatal(0, "user_release_one(%s, %s) = %d", fullname, tag, error);
5188 VERIFY3U(dmu_objset_hold(fullname, FTAG, &origin), ==, ENOENT);
5191 (void) rw_unlock(&ztest_name_lock);
5195 * Inject random faults into the on-disk data.
5199 ztest_fault_inject(ztest_ds_t *zd, uint64_t id)
5201 ztest_shared_t *zs = ztest_shared;
5202 spa_t *spa = ztest_spa;
5206 uint64_t bad = 0x1990c0ffeedecadeull;
5211 int bshift = SPA_MAXBLOCKSHIFT + 2; /* don't scrog all labels */
5217 boolean_t islog = B_FALSE;
5219 path0 = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
5220 pathrand = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
5222 mutex_enter(&ztest_vdev_lock);
5223 maxfaults = MAXFAULTS();
5224 leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz;
5225 mirror_save = zs->zs_mirrors;
5226 mutex_exit(&ztest_vdev_lock);
5228 ASSERT(leaves >= 1);
5231 * Grab the name lock as reader. There are some operations
5232 * which don't like to have their vdevs changed while
5233 * they are in progress (i.e. spa_change_guid). Those
5234 * operations will have grabbed the name lock as writer.
5236 (void) rw_rdlock(&ztest_name_lock);
5239 * We need SCL_STATE here because we're going to look at vd0->vdev_tsd.
5241 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5243 if (ztest_random(2) == 0) {
5245 * Inject errors on a normal data device or slog device.
5247 top = ztest_random_vdev_top(spa, B_TRUE);
5248 leaf = ztest_random(leaves) + zs->zs_splits;
5251 * Generate paths to the first leaf in this top-level vdev,
5252 * and to the random leaf we selected. We'll induce transient
5253 * write failures and random online/offline activity on leaf 0,
5254 * and we'll write random garbage to the randomly chosen leaf.
5256 (void) snprintf(path0, MAXPATHLEN, ztest_dev_template,
5257 ztest_opts.zo_dir, ztest_opts.zo_pool,
5258 top * leaves + zs->zs_splits);
5259 (void) snprintf(pathrand, MAXPATHLEN, ztest_dev_template,
5260 ztest_opts.zo_dir, ztest_opts.zo_pool,
5261 top * leaves + leaf);
5263 vd0 = vdev_lookup_by_path(spa->spa_root_vdev, path0);
5264 if (vd0 != NULL && vd0->vdev_top->vdev_islog)
5268 * If the top-level vdev needs to be resilvered
5269 * then we only allow faults on the device that is
5272 if (vd0 != NULL && maxfaults != 1 &&
5273 (!vdev_resilver_needed(vd0->vdev_top, NULL, NULL) ||
5274 vd0->vdev_resilver_txg != 0)) {
5276 * Make vd0 explicitly claim to be unreadable,
5277 * or unwriteable, or reach behind its back
5278 * and close the underlying fd. We can do this if
5279 * maxfaults == 0 because we'll fail and reexecute,
5280 * and we can do it if maxfaults >= 2 because we'll
5281 * have enough redundancy. If maxfaults == 1, the
5282 * combination of this with injection of random data
5283 * corruption below exceeds the pool's fault tolerance.
5285 vdev_file_t *vf = vd0->vdev_tsd;
5287 if (vf != NULL && ztest_random(3) == 0) {
5288 (void) close(vf->vf_vnode->v_fd);
5289 vf->vf_vnode->v_fd = -1;
5290 } else if (ztest_random(2) == 0) {
5291 vd0->vdev_cant_read = B_TRUE;
5293 vd0->vdev_cant_write = B_TRUE;
5295 guid0 = vd0->vdev_guid;
5299 * Inject errors on an l2cache device.
5301 spa_aux_vdev_t *sav = &spa->spa_l2cache;
5303 if (sav->sav_count == 0) {
5304 spa_config_exit(spa, SCL_STATE, FTAG);
5305 (void) rw_unlock(&ztest_name_lock);
5308 vd0 = sav->sav_vdevs[ztest_random(sav->sav_count)];
5309 guid0 = vd0->vdev_guid;
5310 (void) strcpy(path0, vd0->vdev_path);
5311 (void) strcpy(pathrand, vd0->vdev_path);
5315 maxfaults = INT_MAX; /* no limit on cache devices */
5318 spa_config_exit(spa, SCL_STATE, FTAG);
5319 (void) rw_unlock(&ztest_name_lock);
5322 * If we can tolerate two or more faults, or we're dealing
5323 * with a slog, randomly online/offline vd0.
5325 if ((maxfaults >= 2 || islog) && guid0 != 0) {
5326 if (ztest_random(10) < 6) {
5327 int flags = (ztest_random(2) == 0 ?
5328 ZFS_OFFLINE_TEMPORARY : 0);
5331 * We have to grab the zs_name_lock as writer to
5332 * prevent a race between offlining a slog and
5333 * destroying a dataset. Offlining the slog will
5334 * grab a reference on the dataset which may cause
5335 * dsl_destroy_head() to fail with EBUSY thus
5336 * leaving the dataset in an inconsistent state.
5339 (void) rw_wrlock(&ztest_name_lock);
5341 VERIFY(vdev_offline(spa, guid0, flags) != EBUSY);
5344 (void) rw_unlock(&ztest_name_lock);
5347 * Ideally we would like to be able to randomly
5348 * call vdev_[on|off]line without holding locks
5349 * to force unpredictable failures but the side
5350 * effects of vdev_[on|off]line prevent us from
5351 * doing so. We grab the ztest_vdev_lock here to
5352 * prevent a race between injection testing and
5355 mutex_enter(&ztest_vdev_lock);
5356 (void) vdev_online(spa, guid0, 0, NULL);
5357 mutex_exit(&ztest_vdev_lock);
5365 * We have at least single-fault tolerance, so inject data corruption.
5367 fd = open(pathrand, O_RDWR);
5369 if (fd == -1) /* we hit a gap in the device namespace */
5372 fsize = lseek(fd, 0, SEEK_END);
5374 while (--iters != 0) {
5376 * The offset must be chosen carefully to ensure that
5377 * we do not inject a given logical block with errors
5378 * on two different leaf devices, because ZFS can not
5379 * tolerate that (if maxfaults==1).
5381 * We divide each leaf into chunks of size
5382 * (# leaves * SPA_MAXBLOCKSIZE * 4). Within each chunk
5383 * there is a series of ranges to which we can inject errors.
5384 * Each range can accept errors on only a single leaf vdev.
5385 * The error injection ranges are separated by ranges
5386 * which we will not inject errors on any device (DMZs).
5387 * Each DMZ must be large enough such that a single block
5388 * can not straddle it, so that a single block can not be
5389 * a target in two different injection ranges (on different
5392 * For example, with 3 leaves, each chunk looks like:
5393 * 0 to 32M: injection range for leaf 0
5394 * 32M to 64M: DMZ - no injection allowed
5395 * 64M to 96M: injection range for leaf 1
5396 * 96M to 128M: DMZ - no injection allowed
5397 * 128M to 160M: injection range for leaf 2
5398 * 160M to 192M: DMZ - no injection allowed
5400 offset = ztest_random(fsize / (leaves << bshift)) *
5401 (leaves << bshift) + (leaf << bshift) +
5402 (ztest_random(1ULL << (bshift - 1)) & -8ULL);
5404 if (offset >= fsize)
5407 mutex_enter(&ztest_vdev_lock);
5408 if (mirror_save != zs->zs_mirrors) {
5409 mutex_exit(&ztest_vdev_lock);
5414 if (pwrite(fd, &bad, sizeof (bad), offset) != sizeof (bad))
5415 fatal(1, "can't inject bad word at 0x%llx in %s",
5418 mutex_exit(&ztest_vdev_lock);
5420 if (ztest_opts.zo_verbose >= 7)
5421 (void) printf("injected bad word into %s,"
5422 " offset 0x%llx\n", pathrand, (u_longlong_t)offset);
5427 umem_free(path0, MAXPATHLEN);
5428 umem_free(pathrand, MAXPATHLEN);
5432 * Verify that DDT repair works as expected.
5435 ztest_ddt_repair(ztest_ds_t *zd, uint64_t id)
5437 ztest_shared_t *zs = ztest_shared;
5438 spa_t *spa = ztest_spa;
5439 objset_t *os = zd->zd_os;
5441 uint64_t object, blocksize, txg, pattern, psize;
5442 enum zio_checksum checksum = spa_dedup_checksum(spa);
5447 int copies = 2 * ZIO_DEDUPDITTO_MIN;
5450 blocksize = ztest_random_blocksize();
5451 blocksize = MIN(blocksize, 2048); /* because we write so many */
5453 od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
5454 ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0, 0);
5456 if (ztest_object_init(zd, od, sizeof (ztest_od_t), B_FALSE) != 0) {
5457 umem_free(od, sizeof (ztest_od_t));
5462 * Take the name lock as writer to prevent anyone else from changing
5463 * the pool and dataset properies we need to maintain during this test.
5465 (void) rw_wrlock(&ztest_name_lock);
5467 if (ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_DEDUP, checksum,
5469 ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_COPIES, 1,
5471 (void) rw_unlock(&ztest_name_lock);
5472 umem_free(od, sizeof (ztest_od_t));
5476 object = od[0].od_object;
5477 blocksize = od[0].od_blocksize;
5478 pattern = zs->zs_guid ^ dmu_objset_fsid_guid(os);
5480 ASSERT(object != 0);
5482 tx = dmu_tx_create(os);
5483 dmu_tx_hold_write(tx, object, 0, copies * blocksize);
5484 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
5486 (void) rw_unlock(&ztest_name_lock);
5487 umem_free(od, sizeof (ztest_od_t));
5492 * Write all the copies of our block.
5494 for (i = 0; i < copies; i++) {
5495 uint64_t offset = i * blocksize;
5496 int error = dmu_buf_hold(os, object, offset, FTAG, &db,
5497 DMU_READ_NO_PREFETCH);
5499 fatal(B_FALSE, "dmu_buf_hold(%p, %llu, %llu) = %u",
5500 os, (long long)object, (long long) offset, error);
5502 ASSERT(db->db_offset == offset);
5503 ASSERT(db->db_size == blocksize);
5504 ASSERT(ztest_pattern_match(db->db_data, db->db_size, pattern) ||
5505 ztest_pattern_match(db->db_data, db->db_size, 0ULL));
5506 dmu_buf_will_fill(db, tx);
5507 ztest_pattern_set(db->db_data, db->db_size, pattern);
5508 dmu_buf_rele(db, FTAG);
5512 txg_wait_synced(spa_get_dsl(spa), txg);
5515 * Find out what block we got.
5517 VERIFY0(dmu_buf_hold(os, object, 0, FTAG, &db,
5518 DMU_READ_NO_PREFETCH));
5519 blk = *((dmu_buf_impl_t *)db)->db_blkptr;
5520 dmu_buf_rele(db, FTAG);
5523 * Damage the block. Dedup-ditto will save us when we read it later.
5525 psize = BP_GET_PSIZE(&blk);
5526 buf = zio_buf_alloc(psize);
5527 ztest_pattern_set(buf, psize, ~pattern);
5529 (void) zio_wait(zio_rewrite(NULL, spa, 0, &blk,
5530 buf, psize, NULL, NULL, ZIO_PRIORITY_SYNC_WRITE,
5531 ZIO_FLAG_CANFAIL | ZIO_FLAG_INDUCE_DAMAGE, NULL));
5533 zio_buf_free(buf, psize);
5535 (void) rw_unlock(&ztest_name_lock);
5536 umem_free(od, sizeof (ztest_od_t));
5544 ztest_scrub(ztest_ds_t *zd, uint64_t id)
5546 spa_t *spa = ztest_spa;
5548 (void) spa_scan(spa, POOL_SCAN_SCRUB);
5549 (void) poll(NULL, 0, 100); /* wait a moment, then force a restart */
5550 (void) spa_scan(spa, POOL_SCAN_SCRUB);
5554 * Change the guid for the pool.
5558 ztest_reguid(ztest_ds_t *zd, uint64_t id)
5560 spa_t *spa = ztest_spa;
5561 uint64_t orig, load;
5564 orig = spa_guid(spa);
5565 load = spa_load_guid(spa);
5567 (void) rw_wrlock(&ztest_name_lock);
5568 error = spa_change_guid(spa);
5569 (void) rw_unlock(&ztest_name_lock);
5574 if (ztest_opts.zo_verbose >= 4) {
5575 (void) printf("Changed guid old %llu -> %llu\n",
5576 (u_longlong_t)orig, (u_longlong_t)spa_guid(spa));
5579 VERIFY3U(orig, !=, spa_guid(spa));
5580 VERIFY3U(load, ==, spa_load_guid(spa));
5584 * Rename the pool to a different name and then rename it back.
5588 ztest_spa_rename(ztest_ds_t *zd, uint64_t id)
5590 char *oldname, *newname;
5593 (void) rw_wrlock(&ztest_name_lock);
5595 oldname = ztest_opts.zo_pool;
5596 newname = umem_alloc(strlen(oldname) + 5, UMEM_NOFAIL);
5597 (void) strcpy(newname, oldname);
5598 (void) strcat(newname, "_tmp");
5603 VERIFY3U(0, ==, spa_rename(oldname, newname));
5606 * Try to open it under the old name, which shouldn't exist
5608 VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
5611 * Open it under the new name and make sure it's still the same spa_t.
5613 VERIFY3U(0, ==, spa_open(newname, &spa, FTAG));
5615 ASSERT(spa == ztest_spa);
5616 spa_close(spa, FTAG);
5619 * Rename it back to the original
5621 VERIFY3U(0, ==, spa_rename(newname, oldname));
5624 * Make sure it can still be opened
5626 VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG));
5628 ASSERT(spa == ztest_spa);
5629 spa_close(spa, FTAG);
5631 umem_free(newname, strlen(newname) + 1);
5633 (void) rw_unlock(&ztest_name_lock);
5637 ztest_fletcher(ztest_ds_t *zd, uint64_t id)
5639 hrtime_t end = gethrtime() + NANOSEC;
5641 while (gethrtime() <= end) {
5642 int run_count = 100;
5648 zio_cksum_t zc_ref_byteswap;
5650 size = ztest_random_blocksize();
5651 buf = umem_alloc(size, UMEM_NOFAIL);
5653 for (i = 0, ptr = buf; i < size / sizeof (*ptr); i++, ptr++)
5654 *ptr = ztest_random(UINT_MAX);
5656 VERIFY0(fletcher_4_impl_set("scalar"));
5657 fletcher_4_native(buf, size, NULL, &zc_ref);
5658 fletcher_4_byteswap(buf, size, NULL, &zc_ref_byteswap);
5660 VERIFY0(fletcher_4_impl_set("cycle"));
5661 while (run_count-- > 0) {
5663 zio_cksum_t zc_byteswap;
5665 fletcher_4_byteswap(buf, size, NULL, &zc_byteswap);
5666 fletcher_4_native(buf, size, NULL, &zc);
5668 VERIFY0(bcmp(&zc, &zc_ref, sizeof (zc)));
5669 VERIFY0(bcmp(&zc_byteswap, &zc_ref_byteswap,
5670 sizeof (zc_byteswap)));
5673 umem_free(buf, size);
5678 ztest_check_path(char *path)
5681 /* return true on success */
5682 return (!stat(path, &s));
5686 ztest_get_zdb_bin(char *bin, int len)
5690 * Try to use ZDB_PATH and in-tree zdb path. If not successful, just
5691 * let popen to search through PATH.
5693 if ((zdb_path = getenv("ZDB_PATH"))) {
5694 strlcpy(bin, zdb_path, len); /* In env */
5695 if (!ztest_check_path(bin)) {
5696 ztest_dump_core = 0;
5697 fatal(1, "invalid ZDB_PATH '%s'", bin);
5702 VERIFY(realpath(getexecname(), bin) != NULL);
5703 if (strstr(bin, "/ztest/")) {
5704 strstr(bin, "/ztest/")[0] = '\0'; /* In-tree */
5705 strcat(bin, "/zdb/zdb");
5706 if (ztest_check_path(bin))
5713 * Verify pool integrity by running zdb.
5716 ztest_run_zdb(char *pool)
5722 const int len = MAXPATHLEN + MAXNAMELEN + 20;
5725 bin = umem_alloc(len, UMEM_NOFAIL);
5726 zdb = umem_alloc(len, UMEM_NOFAIL);
5727 zbuf = umem_alloc(1024, UMEM_NOFAIL);
5729 ztest_get_zdb_bin(bin, len);
5732 "%s -bcc%s%s -d -U %s %s",
5734 ztest_opts.zo_verbose >= 3 ? "s" : "",
5735 ztest_opts.zo_verbose >= 4 ? "v" : "",
5739 if (ztest_opts.zo_verbose >= 5)
5740 (void) printf("Executing %s\n", strstr(zdb, "zdb "));
5742 fp = popen(zdb, "r");
5744 while (fgets(zbuf, 1024, fp) != NULL)
5745 if (ztest_opts.zo_verbose >= 3)
5746 (void) printf("%s", zbuf);
5748 status = pclose(fp);
5753 ztest_dump_core = 0;
5754 if (WIFEXITED(status))
5755 fatal(0, "'%s' exit code %d", zdb, WEXITSTATUS(status));
5757 fatal(0, "'%s' died with signal %d", zdb, WTERMSIG(status));
5759 umem_free(bin, len);
5760 umem_free(zdb, len);
5761 umem_free(zbuf, 1024);
5765 ztest_walk_pool_directory(char *header)
5769 if (ztest_opts.zo_verbose >= 6)
5770 (void) printf("%s\n", header);
5772 mutex_enter(&spa_namespace_lock);
5773 while ((spa = spa_next(spa)) != NULL)
5774 if (ztest_opts.zo_verbose >= 6)
5775 (void) printf("\t%s\n", spa_name(spa));
5776 mutex_exit(&spa_namespace_lock);
5780 ztest_spa_import_export(char *oldname, char *newname)
5782 nvlist_t *config, *newconfig;
5787 if (ztest_opts.zo_verbose >= 4) {
5788 (void) printf("import/export: old = %s, new = %s\n",
5793 * Clean up from previous runs.
5795 (void) spa_destroy(newname);
5798 * Get the pool's configuration and guid.
5800 VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG));
5803 * Kick off a scrub to tickle scrub/export races.
5805 if (ztest_random(2) == 0)
5806 (void) spa_scan(spa, POOL_SCAN_SCRUB);
5808 pool_guid = spa_guid(spa);
5809 spa_close(spa, FTAG);
5811 ztest_walk_pool_directory("pools before export");
5816 VERIFY3U(0, ==, spa_export(oldname, &config, B_FALSE, B_FALSE));
5818 ztest_walk_pool_directory("pools after export");
5823 newconfig = spa_tryimport(config);
5824 ASSERT(newconfig != NULL);
5825 nvlist_free(newconfig);
5828 * Import it under the new name.
5830 error = spa_import(newname, config, NULL, 0);
5832 dump_nvlist(config, 0);
5833 fatal(B_FALSE, "couldn't import pool %s as %s: error %u",
5834 oldname, newname, error);
5837 ztest_walk_pool_directory("pools after import");
5840 * Try to import it again -- should fail with EEXIST.
5842 VERIFY3U(EEXIST, ==, spa_import(newname, config, NULL, 0));
5845 * Try to import it under a different name -- should fail with EEXIST.
5847 VERIFY3U(EEXIST, ==, spa_import(oldname, config, NULL, 0));
5850 * Verify that the pool is no longer visible under the old name.
5852 VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
5855 * Verify that we can open and close the pool using the new name.
5857 VERIFY3U(0, ==, spa_open(newname, &spa, FTAG));
5858 ASSERT(pool_guid == spa_guid(spa));
5859 spa_close(spa, FTAG);
5861 nvlist_free(config);
5865 ztest_resume(spa_t *spa)
5867 if (spa_suspended(spa) && ztest_opts.zo_verbose >= 6)
5868 (void) printf("resuming from suspended state\n");
5869 spa_vdev_state_enter(spa, SCL_NONE);
5870 vdev_clear(spa, NULL);
5871 (void) spa_vdev_state_exit(spa, NULL, 0);
5872 (void) zio_resume(spa);
5876 ztest_resume_thread(void *arg)
5880 while (!ztest_exiting) {
5881 if (spa_suspended(spa))
5883 (void) poll(NULL, 0, 100);
5886 * Periodically change the zfs_compressed_arc_enabled setting.
5888 if (ztest_random(10) == 0)
5889 zfs_compressed_arc_enabled = ztest_random(2);
5901 ztest_deadman_alarm(int sig)
5903 fatal(0, "failed to complete within %d seconds of deadline", GRACE);
5908 ztest_execute(int test, ztest_info_t *zi, uint64_t id)
5910 ztest_ds_t *zd = &ztest_ds[id % ztest_opts.zo_datasets];
5911 ztest_shared_callstate_t *zc = ZTEST_GET_SHARED_CALLSTATE(test);
5912 hrtime_t functime = gethrtime();
5915 for (i = 0; i < zi->zi_iters; i++)
5916 zi->zi_func(zd, id);
5918 functime = gethrtime() - functime;
5920 atomic_add_64(&zc->zc_count, 1);
5921 atomic_add_64(&zc->zc_time, functime);
5923 if (ztest_opts.zo_verbose >= 4)
5924 (void) printf("%6.2f sec in %s\n",
5925 (double)functime / NANOSEC, zi->zi_funcname);
5929 ztest_thread(void *arg)
5932 uint64_t id = (uintptr_t)arg;
5933 ztest_shared_t *zs = ztest_shared;
5937 ztest_shared_callstate_t *zc;
5939 while ((now = gethrtime()) < zs->zs_thread_stop) {
5941 * See if it's time to force a crash.
5943 if (now > zs->zs_thread_kill)
5947 * If we're getting ENOSPC with some regularity, stop.
5949 if (zs->zs_enospc_count > 10)
5953 * Pick a random function to execute.
5955 rand = ztest_random(ZTEST_FUNCS);
5956 zi = &ztest_info[rand];
5957 zc = ZTEST_GET_SHARED_CALLSTATE(rand);
5958 call_next = zc->zc_next;
5960 if (now >= call_next &&
5961 atomic_cas_64(&zc->zc_next, call_next, call_next +
5962 ztest_random(2 * zi->zi_interval[0] + 1)) == call_next) {
5963 ztest_execute(rand, zi, id);
5973 ztest_dataset_name(char *dsname, char *pool, int d)
5975 (void) snprintf(dsname, ZFS_MAX_DATASET_NAME_LEN, "%s/ds_%d", pool, d);
5979 ztest_dataset_destroy(int d)
5981 char name[ZFS_MAX_DATASET_NAME_LEN];
5984 ztest_dataset_name(name, ztest_opts.zo_pool, d);
5986 if (ztest_opts.zo_verbose >= 3)
5987 (void) printf("Destroying %s to free up space\n", name);
5990 * Cleanup any non-standard clones and snapshots. In general,
5991 * ztest thread t operates on dataset (t % zopt_datasets),
5992 * so there may be more than one thing to clean up.
5994 for (t = d; t < ztest_opts.zo_threads;
5995 t += ztest_opts.zo_datasets)
5996 ztest_dsl_dataset_cleanup(name, t);
5998 (void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
5999 DS_FIND_SNAPSHOTS | DS_FIND_CHILDREN);
6003 ztest_dataset_dirobj_verify(ztest_ds_t *zd)
6005 uint64_t usedobjs, dirobjs, scratch;
6008 * ZTEST_DIROBJ is the object directory for the entire dataset.
6009 * Therefore, the number of objects in use should equal the
6010 * number of ZTEST_DIROBJ entries, +1 for ZTEST_DIROBJ itself.
6011 * If not, we have an object leak.
6013 * Note that we can only check this in ztest_dataset_open(),
6014 * when the open-context and syncing-context values agree.
6015 * That's because zap_count() returns the open-context value,
6016 * while dmu_objset_space() returns the rootbp fill count.
6018 VERIFY3U(0, ==, zap_count(zd->zd_os, ZTEST_DIROBJ, &dirobjs));
6019 dmu_objset_space(zd->zd_os, &scratch, &scratch, &usedobjs, &scratch);
6020 ASSERT3U(dirobjs + 1, ==, usedobjs);
6024 ztest_dataset_open(int d)
6026 ztest_ds_t *zd = &ztest_ds[d];
6027 uint64_t committed_seq = ZTEST_GET_SHARED_DS(d)->zd_seq;
6030 char name[ZFS_MAX_DATASET_NAME_LEN];
6033 ztest_dataset_name(name, ztest_opts.zo_pool, d);
6035 (void) rw_rdlock(&ztest_name_lock);
6037 error = ztest_dataset_create(name);
6038 if (error == ENOSPC) {
6039 (void) rw_unlock(&ztest_name_lock);
6040 ztest_record_enospc(FTAG);
6043 ASSERT(error == 0 || error == EEXIST);
6045 VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, zd, &os));
6046 (void) rw_unlock(&ztest_name_lock);
6048 ztest_zd_init(zd, ZTEST_GET_SHARED_DS(d), os);
6050 zilog = zd->zd_zilog;
6052 if (zilog->zl_header->zh_claim_lr_seq != 0 &&
6053 zilog->zl_header->zh_claim_lr_seq < committed_seq)
6054 fatal(0, "missing log records: claimed %llu < committed %llu",
6055 zilog->zl_header->zh_claim_lr_seq, committed_seq);
6057 ztest_dataset_dirobj_verify(zd);
6059 zil_replay(os, zd, ztest_replay_vector);
6061 ztest_dataset_dirobj_verify(zd);
6063 if (ztest_opts.zo_verbose >= 6)
6064 (void) printf("%s replay %llu blocks, %llu records, seq %llu\n",
6066 (u_longlong_t)zilog->zl_parse_blk_count,
6067 (u_longlong_t)zilog->zl_parse_lr_count,
6068 (u_longlong_t)zilog->zl_replaying_seq);
6070 zilog = zil_open(os, ztest_get_data);
6072 if (zilog->zl_replaying_seq != 0 &&
6073 zilog->zl_replaying_seq < committed_seq)
6074 fatal(0, "missing log records: replayed %llu < committed %llu",
6075 zilog->zl_replaying_seq, committed_seq);
6081 ztest_dataset_close(int d)
6083 ztest_ds_t *zd = &ztest_ds[d];
6085 zil_close(zd->zd_zilog);
6086 dmu_objset_disown(zd->zd_os, zd);
6092 * Kick off threads to run tests on all datasets in parallel.
6095 ztest_run(ztest_shared_t *zs)
6100 kthread_t *resume_thread;
6105 ztest_exiting = B_FALSE;
6108 * Initialize parent/child shared state.
6110 mutex_init(&ztest_vdev_lock, NULL, MUTEX_DEFAULT, NULL);
6111 VERIFY(rwlock_init(&ztest_name_lock, USYNC_THREAD, NULL) == 0);
6113 zs->zs_thread_start = gethrtime();
6114 zs->zs_thread_stop =
6115 zs->zs_thread_start + ztest_opts.zo_passtime * NANOSEC;
6116 zs->zs_thread_stop = MIN(zs->zs_thread_stop, zs->zs_proc_stop);
6117 zs->zs_thread_kill = zs->zs_thread_stop;
6118 if (ztest_random(100) < ztest_opts.zo_killrate) {
6119 zs->zs_thread_kill -=
6120 ztest_random(ztest_opts.zo_passtime * NANOSEC);
6123 mutex_init(&zcl.zcl_callbacks_lock, NULL, MUTEX_DEFAULT, NULL);
6125 list_create(&zcl.zcl_callbacks, sizeof (ztest_cb_data_t),
6126 offsetof(ztest_cb_data_t, zcd_node));
6131 kernel_init(FREAD | FWRITE);
6132 VERIFY0(spa_open(ztest_opts.zo_pool, &spa, FTAG));
6133 spa->spa_debug = B_TRUE;
6134 metaslab_preload_limit = ztest_random(20) + 1;
6137 VERIFY0(dmu_objset_own(ztest_opts.zo_pool,
6138 DMU_OST_ANY, B_TRUE, FTAG, &os));
6139 zs->zs_guid = dmu_objset_fsid_guid(os);
6140 dmu_objset_disown(os, FTAG);
6142 spa->spa_dedup_ditto = 2 * ZIO_DEDUPDITTO_MIN;
6145 * We don't expect the pool to suspend unless maxfaults == 0,
6146 * in which case ztest_fault_inject() temporarily takes away
6147 * the only valid replica.
6149 if (MAXFAULTS() == 0)
6150 spa->spa_failmode = ZIO_FAILURE_MODE_WAIT;
6152 spa->spa_failmode = ZIO_FAILURE_MODE_PANIC;
6155 * Create a thread to periodically resume suspended I/O.
6157 VERIFY3P((resume_thread = zk_thread_create(NULL, 0,
6158 (thread_func_t)ztest_resume_thread, spa, TS_RUN, NULL, 0, 0,
6159 PTHREAD_CREATE_JOINABLE)), !=, NULL);
6163 * Set a deadman alarm to abort() if we hang.
6165 signal(SIGALRM, ztest_deadman_alarm);
6166 alarm((zs->zs_thread_stop - zs->zs_thread_start) / NANOSEC + GRACE);
6170 * Verify that we can safely inquire about about any object,
6171 * whether it's allocated or not. To make it interesting,
6172 * we probe a 5-wide window around each power of two.
6173 * This hits all edge cases, including zero and the max.
6175 for (t = 0; t < 64; t++) {
6176 for (d = -5; d <= 5; d++) {
6177 error = dmu_object_info(spa->spa_meta_objset,
6178 (1ULL << t) + d, NULL);
6179 ASSERT(error == 0 || error == ENOENT ||
6185 * If we got any ENOSPC errors on the previous run, destroy something.
6187 if (zs->zs_enospc_count != 0) {
6188 int d = ztest_random(ztest_opts.zo_datasets);
6189 ztest_dataset_destroy(d);
6191 zs->zs_enospc_count = 0;
6193 tid = umem_zalloc(ztest_opts.zo_threads * sizeof (kt_did_t),
6196 if (ztest_opts.zo_verbose >= 4)
6197 (void) printf("starting main threads...\n");
6200 * Kick off all the tests that run in parallel.
6202 for (t = 0; t < ztest_opts.zo_threads; t++) {
6205 if (t < ztest_opts.zo_datasets &&
6206 ztest_dataset_open(t) != 0)
6209 VERIFY3P(thread = zk_thread_create(NULL, 0,
6210 (thread_func_t)ztest_thread,
6211 (void *)(uintptr_t)t, TS_RUN, NULL, 0, 0,
6212 PTHREAD_CREATE_JOINABLE), !=, NULL);
6213 tid[t] = thread->t_tid;
6217 * Wait for all of the tests to complete. We go in reverse order
6218 * so we don't close datasets while threads are still using them.
6220 for (t = ztest_opts.zo_threads - 1; t >= 0; t--) {
6221 thread_join(tid[t]);
6222 if (t < ztest_opts.zo_datasets)
6223 ztest_dataset_close(t);
6226 txg_wait_synced(spa_get_dsl(spa), 0);
6228 zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(spa));
6229 zs->zs_space = metaslab_class_get_space(spa_normal_class(spa));
6231 umem_free(tid, ztest_opts.zo_threads * sizeof (kt_did_t));
6233 /* Kill the resume thread */
6234 ztest_exiting = B_TRUE;
6235 thread_join(resume_thread->t_tid);
6239 * Right before closing the pool, kick off a bunch of async I/O;
6240 * spa_close() should wait for it to complete.
6242 for (object = 1; object < 50; object++) {
6243 dmu_prefetch(spa->spa_meta_objset, object, 0, 0, 1ULL << 20,
6244 ZIO_PRIORITY_SYNC_READ);
6247 /* Verify that at least one commit cb was called in a timely fashion */
6248 if (zc_cb_counter >= ZTEST_COMMIT_CB_MIN_REG)
6249 VERIFY0(zc_min_txg_delay);
6251 spa_close(spa, FTAG);
6254 * Verify that we can loop over all pools.
6256 mutex_enter(&spa_namespace_lock);
6257 for (spa = spa_next(NULL); spa != NULL; spa = spa_next(spa))
6258 if (ztest_opts.zo_verbose > 3)
6259 (void) printf("spa_next: found %s\n", spa_name(spa));
6260 mutex_exit(&spa_namespace_lock);
6263 * Verify that we can export the pool and reimport it under a
6266 if (ztest_random(2) == 0) {
6267 char name[ZFS_MAX_DATASET_NAME_LEN];
6268 (void) snprintf(name, sizeof (name), "%s_import",
6269 ztest_opts.zo_pool);
6270 ztest_spa_import_export(ztest_opts.zo_pool, name);
6271 ztest_spa_import_export(name, ztest_opts.zo_pool);
6276 list_destroy(&zcl.zcl_callbacks);
6277 mutex_destroy(&zcl.zcl_callbacks_lock);
6278 (void) rwlock_destroy(&ztest_name_lock);
6279 mutex_destroy(&ztest_vdev_lock);
6285 ztest_ds_t *zd = &ztest_ds[0];
6289 if (ztest_opts.zo_verbose >= 3)
6290 (void) printf("testing spa_freeze()...\n");
6292 kernel_init(FREAD | FWRITE);
6293 VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
6294 VERIFY3U(0, ==, ztest_dataset_open(0));
6295 spa->spa_debug = B_TRUE;
6299 * Force the first log block to be transactionally allocated.
6300 * We have to do this before we freeze the pool -- otherwise
6301 * the log chain won't be anchored.
6303 while (BP_IS_HOLE(&zd->zd_zilog->zl_header->zh_log)) {
6304 ztest_dmu_object_alloc_free(zd, 0);
6305 zil_commit(zd->zd_zilog, 0);
6308 txg_wait_synced(spa_get_dsl(spa), 0);
6311 * Freeze the pool. This stops spa_sync() from doing anything,
6312 * so that the only way to record changes from now on is the ZIL.
6317 * Because it is hard to predict how much space a write will actually
6318 * require beforehand, we leave ourselves some fudge space to write over
6321 uint64_t capacity = metaslab_class_get_space(spa_normal_class(spa)) / 2;
6324 * Run tests that generate log records but don't alter the pool config
6325 * or depend on DSL sync tasks (snapshots, objset create/destroy, etc).
6326 * We do a txg_wait_synced() after each iteration to force the txg
6327 * to increase well beyond the last synced value in the uberblock.
6328 * The ZIL should be OK with that.
6330 * Run a random number of times less than zo_maxloops and ensure we do
6331 * not run out of space on the pool.
6333 while (ztest_random(10) != 0 &&
6334 numloops++ < ztest_opts.zo_maxloops &&
6335 metaslab_class_get_alloc(spa_normal_class(spa)) < capacity) {
6337 ztest_od_init(&od, 0, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0, 0);
6338 VERIFY0(ztest_object_init(zd, &od, sizeof (od), B_FALSE));
6339 ztest_io(zd, od.od_object,
6340 ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
6341 txg_wait_synced(spa_get_dsl(spa), 0);
6345 * Commit all of the changes we just generated.
6347 zil_commit(zd->zd_zilog, 0);
6348 txg_wait_synced(spa_get_dsl(spa), 0);
6351 * Close our dataset and close the pool.
6353 ztest_dataset_close(0);
6354 spa_close(spa, FTAG);
6358 * Open and close the pool and dataset to induce log replay.
6360 kernel_init(FREAD | FWRITE);
6361 VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
6362 ASSERT(spa_freeze_txg(spa) == UINT64_MAX);
6363 VERIFY3U(0, ==, ztest_dataset_open(0));
6364 ztest_dataset_close(0);
6366 spa->spa_debug = B_TRUE;
6368 txg_wait_synced(spa_get_dsl(spa), 0);
6369 ztest_reguid(NULL, 0);
6371 spa_close(spa, FTAG);
6376 print_time(hrtime_t t, char *timebuf)
6378 hrtime_t s = t / NANOSEC;
6379 hrtime_t m = s / 60;
6380 hrtime_t h = m / 60;
6381 hrtime_t d = h / 24;
6390 (void) sprintf(timebuf,
6391 "%llud%02lluh%02llum%02llus", d, h, m, s);
6393 (void) sprintf(timebuf, "%lluh%02llum%02llus", h, m, s);
6395 (void) sprintf(timebuf, "%llum%02llus", m, s);
6397 (void) sprintf(timebuf, "%llus", s);
6401 make_random_props(void)
6405 VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
6406 if (ztest_random(2) == 0)
6408 VERIFY(nvlist_add_uint64(props, "autoreplace", 1) == 0);
6414 * Create a storage pool with the given name and initial vdev size.
6415 * Then test spa_freeze() functionality.
6418 ztest_init(ztest_shared_t *zs)
6421 nvlist_t *nvroot, *props;
6424 mutex_init(&ztest_vdev_lock, NULL, MUTEX_DEFAULT, NULL);
6425 VERIFY(rwlock_init(&ztest_name_lock, USYNC_THREAD, NULL) == 0);
6427 kernel_init(FREAD | FWRITE);
6430 * Create the storage pool.
6432 (void) spa_destroy(ztest_opts.zo_pool);
6433 ztest_shared->zs_vdev_next_leaf = 0;
6435 zs->zs_mirrors = ztest_opts.zo_mirrors;
6436 nvroot = make_vdev_root(NULL, NULL, NULL, ztest_opts.zo_vdev_size, 0,
6437 0, ztest_opts.zo_raidz, zs->zs_mirrors, 1);
6438 props = make_random_props();
6439 for (i = 0; i < SPA_FEATURES; i++) {
6441 VERIFY3S(-1, !=, asprintf(&buf, "feature@%s",
6442 spa_feature_table[i].fi_uname));
6443 VERIFY3U(0, ==, nvlist_add_uint64(props, buf, 0));
6446 VERIFY3U(0, ==, spa_create(ztest_opts.zo_pool, nvroot, props, NULL));
6447 nvlist_free(nvroot);
6450 VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
6451 zs->zs_metaslab_sz =
6452 1ULL << spa->spa_root_vdev->vdev_child[0]->vdev_ms_shift;
6453 spa_close(spa, FTAG);
6457 ztest_run_zdb(ztest_opts.zo_pool);
6461 ztest_run_zdb(ztest_opts.zo_pool);
6463 (void) rwlock_destroy(&ztest_name_lock);
6464 mutex_destroy(&ztest_vdev_lock);
6470 static char ztest_name_data[] = "/tmp/ztest.data.XXXXXX";
6472 ztest_fd_data = mkstemp(ztest_name_data);
6473 ASSERT3S(ztest_fd_data, >=, 0);
6474 (void) unlink(ztest_name_data);
6478 shared_data_size(ztest_shared_hdr_t *hdr)
6482 size = hdr->zh_hdr_size;
6483 size += hdr->zh_opts_size;
6484 size += hdr->zh_size;
6485 size += hdr->zh_stats_size * hdr->zh_stats_count;
6486 size += hdr->zh_ds_size * hdr->zh_ds_count;
6495 ztest_shared_hdr_t *hdr;
6497 hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
6498 PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
6499 ASSERT(hdr != MAP_FAILED);
6501 VERIFY3U(0, ==, ftruncate(ztest_fd_data, sizeof (ztest_shared_hdr_t)));
6503 hdr->zh_hdr_size = sizeof (ztest_shared_hdr_t);
6504 hdr->zh_opts_size = sizeof (ztest_shared_opts_t);
6505 hdr->zh_size = sizeof (ztest_shared_t);
6506 hdr->zh_stats_size = sizeof (ztest_shared_callstate_t);
6507 hdr->zh_stats_count = ZTEST_FUNCS;
6508 hdr->zh_ds_size = sizeof (ztest_shared_ds_t);
6509 hdr->zh_ds_count = ztest_opts.zo_datasets;
6511 size = shared_data_size(hdr);
6512 VERIFY3U(0, ==, ftruncate(ztest_fd_data, size));
6514 (void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
6521 ztest_shared_hdr_t *hdr;
6524 hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
6525 PROT_READ, MAP_SHARED, ztest_fd_data, 0);
6526 ASSERT(hdr != MAP_FAILED);
6528 size = shared_data_size(hdr);
6530 (void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
6531 hdr = ztest_shared_hdr = (void *)mmap(0, P2ROUNDUP(size, getpagesize()),
6532 PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
6533 ASSERT(hdr != MAP_FAILED);
6534 buf = (uint8_t *)hdr;
6536 offset = hdr->zh_hdr_size;
6537 ztest_shared_opts = (void *)&buf[offset];
6538 offset += hdr->zh_opts_size;
6539 ztest_shared = (void *)&buf[offset];
6540 offset += hdr->zh_size;
6541 ztest_shared_callstate = (void *)&buf[offset];
6542 offset += hdr->zh_stats_size * hdr->zh_stats_count;
6543 ztest_shared_ds = (void *)&buf[offset];
6547 exec_child(char *cmd, char *libpath, boolean_t ignorekill, int *statusp)
6551 char *cmdbuf = NULL;
6556 cmdbuf = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
6557 (void) strlcpy(cmdbuf, getexecname(), MAXPATHLEN);
6562 fatal(1, "fork failed");
6564 if (pid == 0) { /* child */
6565 char *emptyargv[2] = { cmd, NULL };
6566 char fd_data_str[12];
6568 struct rlimit rl = { 1024, 1024 };
6569 (void) setrlimit(RLIMIT_NOFILE, &rl);
6571 (void) close(ztest_fd_rand);
6572 VERIFY(11 >= snprintf(fd_data_str, 12, "%d", ztest_fd_data));
6573 VERIFY(0 == setenv("ZTEST_FD_DATA", fd_data_str, 1));
6575 (void) enable_extended_FILE_stdio(-1, -1);
6576 if (libpath != NULL)
6577 VERIFY(0 == setenv("LD_LIBRARY_PATH", libpath, 1));
6578 (void) execv(cmd, emptyargv);
6579 ztest_dump_core = B_FALSE;
6580 fatal(B_TRUE, "exec failed: %s", cmd);
6583 if (cmdbuf != NULL) {
6584 umem_free(cmdbuf, MAXPATHLEN);
6588 while (waitpid(pid, &status, 0) != pid)
6590 if (statusp != NULL)
6593 if (WIFEXITED(status)) {
6594 if (WEXITSTATUS(status) != 0) {
6595 (void) fprintf(stderr, "child exited with code %d\n",
6596 WEXITSTATUS(status));
6600 } else if (WIFSIGNALED(status)) {
6601 if (!ignorekill || WTERMSIG(status) != SIGKILL) {
6602 (void) fprintf(stderr, "child died with signal %d\n",
6608 (void) fprintf(stderr, "something strange happened to child\n");
6615 ztest_run_init(void)
6619 ztest_shared_t *zs = ztest_shared;
6621 ASSERT(ztest_opts.zo_init != 0);
6624 * Blow away any existing copy of zpool.cache
6626 (void) remove(spa_config_path);
6629 * Create and initialize our storage pool.
6631 for (i = 1; i <= ztest_opts.zo_init; i++) {
6632 bzero(zs, sizeof (ztest_shared_t));
6633 if (ztest_opts.zo_verbose >= 3 &&
6634 ztest_opts.zo_init != 1) {
6635 (void) printf("ztest_init(), pass %d\n", i);
6642 main(int argc, char **argv)
6650 ztest_shared_callstate_t *zc;
6657 char *fd_data_str = getenv("ZTEST_FD_DATA");
6658 struct sigaction action;
6660 (void) setvbuf(stdout, NULL, _IOLBF, 0);
6662 dprintf_setup(&argc, argv);
6664 action.sa_handler = sig_handler;
6665 sigemptyset(&action.sa_mask);
6666 action.sa_flags = 0;
6668 if (sigaction(SIGSEGV, &action, NULL) < 0) {
6669 (void) fprintf(stderr, "ztest: cannot catch SIGSEGV: %s.\n",
6674 if (sigaction(SIGABRT, &action, NULL) < 0) {
6675 (void) fprintf(stderr, "ztest: cannot catch SIGABRT: %s.\n",
6680 ztest_fd_rand = open("/dev/urandom", O_RDONLY);
6681 ASSERT3S(ztest_fd_rand, >=, 0);
6684 process_options(argc, argv);
6689 bcopy(&ztest_opts, ztest_shared_opts,
6690 sizeof (*ztest_shared_opts));
6692 ztest_fd_data = atoi(fd_data_str);
6694 bcopy(ztest_shared_opts, &ztest_opts, sizeof (ztest_opts));
6696 ASSERT3U(ztest_opts.zo_datasets, ==, ztest_shared_hdr->zh_ds_count);
6698 /* Override location of zpool.cache */
6699 VERIFY(asprintf((char **)&spa_config_path, "%s/zpool.cache",
6700 ztest_opts.zo_dir) != -1);
6702 ztest_ds = umem_alloc(ztest_opts.zo_datasets * sizeof (ztest_ds_t),
6707 metaslab_gang_bang = ztest_opts.zo_metaslab_gang_bang;
6708 metaslab_df_alloc_threshold =
6709 zs->zs_metaslab_df_alloc_threshold;
6718 hasalt = (strlen(ztest_opts.zo_alt_ztest) != 0);
6720 if (ztest_opts.zo_verbose >= 1) {
6721 (void) printf("%llu vdevs, %d datasets, %d threads,"
6722 " %llu seconds...\n",
6723 (u_longlong_t)ztest_opts.zo_vdevs,
6724 ztest_opts.zo_datasets,
6725 ztest_opts.zo_threads,
6726 (u_longlong_t)ztest_opts.zo_time);
6729 cmd = umem_alloc(MAXNAMELEN, UMEM_NOFAIL);
6730 (void) strlcpy(cmd, getexecname(), MAXNAMELEN);
6732 zs->zs_do_init = B_TRUE;
6733 if (strlen(ztest_opts.zo_alt_ztest) != 0) {
6734 if (ztest_opts.zo_verbose >= 1) {
6735 (void) printf("Executing older ztest for "
6736 "initialization: %s\n", ztest_opts.zo_alt_ztest);
6738 VERIFY(!exec_child(ztest_opts.zo_alt_ztest,
6739 ztest_opts.zo_alt_libpath, B_FALSE, NULL));
6741 VERIFY(!exec_child(NULL, NULL, B_FALSE, NULL));
6743 zs->zs_do_init = B_FALSE;
6745 zs->zs_proc_start = gethrtime();
6746 zs->zs_proc_stop = zs->zs_proc_start + ztest_opts.zo_time * NANOSEC;
6748 for (f = 0; f < ZTEST_FUNCS; f++) {
6749 zi = &ztest_info[f];
6750 zc = ZTEST_GET_SHARED_CALLSTATE(f);
6751 if (zs->zs_proc_start + zi->zi_interval[0] > zs->zs_proc_stop)
6752 zc->zc_next = UINT64_MAX;
6754 zc->zc_next = zs->zs_proc_start +
6755 ztest_random(2 * zi->zi_interval[0] + 1);
6759 * Run the tests in a loop. These tests include fault injection
6760 * to verify that self-healing data works, and forced crashes
6761 * to verify that we never lose on-disk consistency.
6763 while (gethrtime() < zs->zs_proc_stop) {
6768 * Initialize the workload counters for each function.
6770 for (f = 0; f < ZTEST_FUNCS; f++) {
6771 zc = ZTEST_GET_SHARED_CALLSTATE(f);
6776 /* Set the allocation switch size */
6777 zs->zs_metaslab_df_alloc_threshold =
6778 ztest_random(zs->zs_metaslab_sz / 4) + 1;
6780 if (!hasalt || ztest_random(2) == 0) {
6781 if (hasalt && ztest_opts.zo_verbose >= 1) {
6782 (void) printf("Executing newer ztest: %s\n",
6786 killed = exec_child(cmd, NULL, B_TRUE, &status);
6788 if (hasalt && ztest_opts.zo_verbose >= 1) {
6789 (void) printf("Executing older ztest: %s\n",
6790 ztest_opts.zo_alt_ztest);
6793 killed = exec_child(ztest_opts.zo_alt_ztest,
6794 ztest_opts.zo_alt_libpath, B_TRUE, &status);
6801 if (ztest_opts.zo_verbose >= 1) {
6802 hrtime_t now = gethrtime();
6804 now = MIN(now, zs->zs_proc_stop);
6805 print_time(zs->zs_proc_stop - now, timebuf);
6806 nicenum(zs->zs_space, numbuf);
6808 (void) printf("Pass %3d, %8s, %3llu ENOSPC, "
6809 "%4.1f%% of %5s used, %3.0f%% done, %8s to go\n",
6811 WIFEXITED(status) ? "Complete" : "SIGKILL",
6812 (u_longlong_t)zs->zs_enospc_count,
6813 100.0 * zs->zs_alloc / zs->zs_space,
6815 100.0 * (now - zs->zs_proc_start) /
6816 (ztest_opts.zo_time * NANOSEC), timebuf);
6819 if (ztest_opts.zo_verbose >= 2) {
6820 (void) printf("\nWorkload summary:\n\n");
6821 (void) printf("%7s %9s %s\n",
6822 "Calls", "Time", "Function");
6823 (void) printf("%7s %9s %s\n",
6824 "-----", "----", "--------");
6825 for (f = 0; f < ZTEST_FUNCS; f++) {
6826 zi = &ztest_info[f];
6827 zc = ZTEST_GET_SHARED_CALLSTATE(f);
6828 print_time(zc->zc_time, timebuf);
6829 (void) printf("%7llu %9s %s\n",
6830 (u_longlong_t)zc->zc_count, timebuf,
6833 (void) printf("\n");
6837 * It's possible that we killed a child during a rename test,
6838 * in which case we'll have a 'ztest_tmp' pool lying around
6839 * instead of 'ztest'. Do a blind rename in case this happened.
6842 if (spa_open(ztest_opts.zo_pool, &spa, FTAG) == 0) {
6843 spa_close(spa, FTAG);
6845 char tmpname[ZFS_MAX_DATASET_NAME_LEN];
6847 kernel_init(FREAD | FWRITE);
6848 (void) snprintf(tmpname, sizeof (tmpname), "%s_tmp",
6849 ztest_opts.zo_pool);
6850 (void) spa_rename(tmpname, ztest_opts.zo_pool);
6854 ztest_run_zdb(ztest_opts.zo_pool);
6857 if (ztest_opts.zo_verbose >= 1) {
6859 (void) printf("%d runs of older ztest: %s\n", older,
6860 ztest_opts.zo_alt_ztest);
6861 (void) printf("%d runs of newer ztest: %s\n", newer,
6864 (void) printf("%d killed, %d completed, %.0f%% kill rate\n",
6865 kills, iters - kills, (100.0 * kills) / MAX(1, iters));
6868 umem_free(cmd, MAXNAMELEN);