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) 2013 by Delphix. All rights reserved.
24 * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
25 * Copyright (c) 2012 Martin Matuska <mm@FreeBSD.org>. All rights reserved.
26 * Copyright (c) 2013 Steven Hartland. All rights reserved.
30 * The objective of this program is to provide a DMU/ZAP/SPA stress test
31 * that runs entirely in userland, is easy to use, and easy to extend.
33 * The overall design of the ztest program is as follows:
35 * (1) For each major functional area (e.g. adding vdevs to a pool,
36 * creating and destroying datasets, reading and writing objects, etc)
37 * we have a simple routine to test that functionality. These
38 * individual routines do not have to do anything "stressful".
40 * (2) We turn these simple functionality tests into a stress test by
41 * running them all in parallel, with as many threads as desired,
42 * and spread across as many datasets, objects, and vdevs as desired.
44 * (3) While all this is happening, we inject faults into the pool to
45 * verify that self-healing data really works.
47 * (4) Every time we open a dataset, we change its checksum and compression
48 * functions. Thus even individual objects vary from block to block
49 * in which checksum they use and whether they're compressed.
51 * (5) To verify that we never lose on-disk consistency after a crash,
52 * we run the entire test in a child of the main process.
53 * At random times, the child self-immolates with a SIGKILL.
54 * This is the software equivalent of pulling the power cord.
55 * The parent then runs the test again, using the existing
56 * storage pool, as many times as desired. If backwards compatability
57 * testing is enabled ztest will sometimes run the "older" version
58 * of ztest after a SIGKILL.
60 * (6) To verify that we don't have future leaks or temporal incursions,
61 * many of the functional tests record the transaction group number
62 * as part of their data. When reading old data, they verify that
63 * the transaction group number is less than the current, open txg.
64 * If you add a new test, please do this if applicable.
66 * When run with no arguments, ztest runs for about five minutes and
67 * produces no output if successful. To get a little bit of information,
68 * specify -V. To get more information, specify -VV, and so on.
70 * To turn this into an overnight stress test, use -T to specify run time.
72 * You can ask more more vdevs [-v], datasets [-d], or threads [-t]
73 * to increase the pool capacity, fanout, and overall stress level.
75 * Use the -k option to set the desired frequency of kills.
77 * When ztest invokes itself it passes all relevant information through a
78 * temporary file which is mmap-ed in the child process. This allows shared
79 * memory to survive the exec syscall. The ztest_shared_hdr_t struct is always
80 * stored at offset 0 of this file and contains information on the size and
81 * number of shared structures in the file. The information stored in this file
82 * must remain backwards compatible with older versions of ztest so that
83 * ztest can invoke them during backwards compatibility testing (-B).
86 #include <sys/zfs_context.h>
92 #include <sys/dmu_objset.h>
98 #include <sys/resource.h>
101 #include <sys/zil_impl.h>
102 #include <sys/vdev_impl.h>
103 #include <sys/vdev_file.h>
104 #include <sys/spa_impl.h>
105 #include <sys/metaslab_impl.h>
106 #include <sys/dsl_prop.h>
107 #include <sys/dsl_dataset.h>
108 #include <sys/dsl_destroy.h>
109 #include <sys/dsl_scan.h>
110 #include <sys/zio_checksum.h>
111 #include <sys/refcount.h>
112 #include <sys/zfeature.h>
113 #include <sys/dsl_userhold.h>
115 #include <stdio_ext.h>
124 #include <sys/fs/zfs.h>
125 #include <libnvpair.h>
127 static int ztest_fd_data = -1;
128 static int ztest_fd_rand = -1;
130 typedef struct ztest_shared_hdr {
131 uint64_t zh_hdr_size;
132 uint64_t zh_opts_size;
134 uint64_t zh_stats_size;
135 uint64_t zh_stats_count;
137 uint64_t zh_ds_count;
138 } ztest_shared_hdr_t;
140 static ztest_shared_hdr_t *ztest_shared_hdr;
142 typedef struct ztest_shared_opts {
143 char zo_pool[MAXNAMELEN];
144 char zo_dir[MAXNAMELEN];
145 char zo_alt_ztest[MAXNAMELEN];
146 char zo_alt_libpath[MAXNAMELEN];
148 uint64_t zo_vdevtime;
156 uint64_t zo_passtime;
157 uint64_t zo_killrate;
161 uint64_t zo_maxloops;
162 uint64_t zo_metaslab_gang_bang;
163 } ztest_shared_opts_t;
165 static const ztest_shared_opts_t ztest_opts_defaults = {
166 .zo_pool = { 'z', 't', 'e', 's', 't', '\0' },
167 .zo_dir = { '/', 't', 'm', 'p', '\0' },
168 .zo_alt_ztest = { '\0' },
169 .zo_alt_libpath = { '\0' },
171 .zo_ashift = SPA_MINBLOCKSHIFT,
174 .zo_raidz_parity = 1,
175 .zo_vdev_size = SPA_MINDEVSIZE,
178 .zo_passtime = 60, /* 60 seconds */
179 .zo_killrate = 70, /* 70% kill rate */
182 .zo_time = 300, /* 5 minutes */
183 .zo_maxloops = 50, /* max loops during spa_freeze() */
184 .zo_metaslab_gang_bang = 32 << 10
187 extern uint64_t metaslab_gang_bang;
188 extern uint64_t metaslab_df_alloc_threshold;
189 extern uint64_t zfs_deadman_synctime_ms;
191 static ztest_shared_opts_t *ztest_shared_opts;
192 static ztest_shared_opts_t ztest_opts;
194 typedef struct ztest_shared_ds {
198 static ztest_shared_ds_t *ztest_shared_ds;
199 #define ZTEST_GET_SHARED_DS(d) (&ztest_shared_ds[d])
201 #define BT_MAGIC 0x123456789abcdefULL
202 #define MAXFAULTS() \
203 (MAX(zs->zs_mirrors, 1) * (ztest_opts.zo_raidz_parity + 1) - 1)
207 ZTEST_IO_WRITE_PATTERN,
208 ZTEST_IO_WRITE_ZEROES,
215 typedef struct ztest_block_tag {
225 typedef struct bufwad {
232 * XXX -- fix zfs range locks to be generic so we can use them here.
254 #define ZTEST_RANGE_LOCKS 64
255 #define ZTEST_OBJECT_LOCKS 64
258 * Object descriptor. Used as a template for object lookup/create/remove.
260 typedef struct ztest_od {
263 dmu_object_type_t od_type;
264 dmu_object_type_t od_crtype;
265 uint64_t od_blocksize;
266 uint64_t od_crblocksize;
269 char od_name[MAXNAMELEN];
275 typedef struct ztest_ds {
276 ztest_shared_ds_t *zd_shared;
278 rwlock_t zd_zilog_lock;
280 ztest_od_t *zd_od; /* debugging aid */
281 char zd_name[MAXNAMELEN];
282 mutex_t zd_dirobj_lock;
283 rll_t zd_object_lock[ZTEST_OBJECT_LOCKS];
284 rll_t zd_range_lock[ZTEST_RANGE_LOCKS];
288 * Per-iteration state.
290 typedef void ztest_func_t(ztest_ds_t *zd, uint64_t id);
292 typedef struct ztest_info {
293 ztest_func_t *zi_func; /* test function */
294 uint64_t zi_iters; /* iterations per execution */
295 uint64_t *zi_interval; /* execute every <interval> seconds */
298 typedef struct ztest_shared_callstate {
299 uint64_t zc_count; /* per-pass count */
300 uint64_t zc_time; /* per-pass time */
301 uint64_t zc_next; /* next time to call this function */
302 } ztest_shared_callstate_t;
304 static ztest_shared_callstate_t *ztest_shared_callstate;
305 #define ZTEST_GET_SHARED_CALLSTATE(c) (&ztest_shared_callstate[c])
308 * Note: these aren't static because we want dladdr() to work.
310 ztest_func_t ztest_dmu_read_write;
311 ztest_func_t ztest_dmu_write_parallel;
312 ztest_func_t ztest_dmu_object_alloc_free;
313 ztest_func_t ztest_dmu_commit_callbacks;
314 ztest_func_t ztest_zap;
315 ztest_func_t ztest_zap_parallel;
316 ztest_func_t ztest_zil_commit;
317 ztest_func_t ztest_zil_remount;
318 ztest_func_t ztest_dmu_read_write_zcopy;
319 ztest_func_t ztest_dmu_objset_create_destroy;
320 ztest_func_t ztest_dmu_prealloc;
321 ztest_func_t ztest_fzap;
322 ztest_func_t ztest_dmu_snapshot_create_destroy;
323 ztest_func_t ztest_dsl_prop_get_set;
324 ztest_func_t ztest_spa_prop_get_set;
325 ztest_func_t ztest_spa_create_destroy;
326 ztest_func_t ztest_fault_inject;
327 ztest_func_t ztest_ddt_repair;
328 ztest_func_t ztest_dmu_snapshot_hold;
329 ztest_func_t ztest_spa_rename;
330 ztest_func_t ztest_scrub;
331 ztest_func_t ztest_dsl_dataset_promote_busy;
332 ztest_func_t ztest_vdev_attach_detach;
333 ztest_func_t ztest_vdev_LUN_growth;
334 ztest_func_t ztest_vdev_add_remove;
335 ztest_func_t ztest_vdev_aux_add_remove;
336 ztest_func_t ztest_split_pool;
337 ztest_func_t ztest_reguid;
338 ztest_func_t ztest_spa_upgrade;
340 uint64_t zopt_always = 0ULL * NANOSEC; /* all the time */
341 uint64_t zopt_incessant = 1ULL * NANOSEC / 10; /* every 1/10 second */
342 uint64_t zopt_often = 1ULL * NANOSEC; /* every second */
343 uint64_t zopt_sometimes = 10ULL * NANOSEC; /* every 10 seconds */
344 uint64_t zopt_rarely = 60ULL * NANOSEC; /* every 60 seconds */
346 ztest_info_t ztest_info[] = {
347 { ztest_dmu_read_write, 1, &zopt_always },
348 { ztest_dmu_write_parallel, 10, &zopt_always },
349 { ztest_dmu_object_alloc_free, 1, &zopt_always },
350 { ztest_dmu_commit_callbacks, 1, &zopt_always },
351 { ztest_zap, 30, &zopt_always },
352 { ztest_zap_parallel, 100, &zopt_always },
353 { ztest_split_pool, 1, &zopt_always },
354 { ztest_zil_commit, 1, &zopt_incessant },
355 { ztest_zil_remount, 1, &zopt_sometimes },
356 { ztest_dmu_read_write_zcopy, 1, &zopt_often },
357 { ztest_dmu_objset_create_destroy, 1, &zopt_often },
358 { ztest_dsl_prop_get_set, 1, &zopt_often },
359 { ztest_spa_prop_get_set, 1, &zopt_sometimes },
361 { ztest_dmu_prealloc, 1, &zopt_sometimes },
363 { ztest_fzap, 1, &zopt_sometimes },
364 { ztest_dmu_snapshot_create_destroy, 1, &zopt_sometimes },
365 { ztest_spa_create_destroy, 1, &zopt_sometimes },
366 { ztest_fault_inject, 1, &zopt_sometimes },
367 { ztest_ddt_repair, 1, &zopt_sometimes },
368 { ztest_dmu_snapshot_hold, 1, &zopt_sometimes },
369 { ztest_reguid, 1, &zopt_rarely },
370 { ztest_spa_rename, 1, &zopt_rarely },
371 { ztest_scrub, 1, &zopt_rarely },
372 { ztest_spa_upgrade, 1, &zopt_rarely },
373 { ztest_dsl_dataset_promote_busy, 1, &zopt_rarely },
374 { ztest_vdev_attach_detach, 1, &zopt_sometimes },
375 { ztest_vdev_LUN_growth, 1, &zopt_rarely },
376 { ztest_vdev_add_remove, 1,
377 &ztest_opts.zo_vdevtime },
378 { ztest_vdev_aux_add_remove, 1,
379 &ztest_opts.zo_vdevtime },
382 #define ZTEST_FUNCS (sizeof (ztest_info) / sizeof (ztest_info_t))
385 * The following struct is used to hold a list of uncalled commit callbacks.
386 * The callbacks are ordered by txg number.
388 typedef struct ztest_cb_list {
389 mutex_t zcl_callbacks_lock;
390 list_t zcl_callbacks;
394 * Stuff we need to share writably between parent and child.
396 typedef struct ztest_shared {
397 boolean_t zs_do_init;
398 hrtime_t zs_proc_start;
399 hrtime_t zs_proc_stop;
400 hrtime_t zs_thread_start;
401 hrtime_t zs_thread_stop;
402 hrtime_t zs_thread_kill;
403 uint64_t zs_enospc_count;
404 uint64_t zs_vdev_next_leaf;
405 uint64_t zs_vdev_aux;
410 uint64_t zs_metaslab_sz;
411 uint64_t zs_metaslab_df_alloc_threshold;
415 #define ID_PARALLEL -1ULL
417 static char ztest_dev_template[] = "%s/%s.%llua";
418 static char ztest_aux_template[] = "%s/%s.%s.%llu";
419 ztest_shared_t *ztest_shared;
421 static spa_t *ztest_spa = NULL;
422 static ztest_ds_t *ztest_ds;
424 static mutex_t ztest_vdev_lock;
427 * The ztest_name_lock protects the pool and dataset namespace used by
428 * the individual tests. To modify the namespace, consumers must grab
429 * this lock as writer. Grabbing the lock as reader will ensure that the
430 * namespace does not change while the lock is held.
432 static rwlock_t ztest_name_lock;
434 static boolean_t ztest_dump_core = B_TRUE;
435 static boolean_t ztest_exiting;
437 /* Global commit callback list */
438 static ztest_cb_list_t zcl;
441 ZTEST_META_DNODE = 0,
446 static void usage(boolean_t) __NORETURN;
449 * These libumem hooks provide a reasonable set of defaults for the allocator's
450 * debugging facilities.
455 return ("default,verbose"); /* $UMEM_DEBUG setting */
459 _umem_logging_init(void)
461 return ("fail,contents"); /* $UMEM_LOGGING setting */
464 #define FATAL_MSG_SZ 1024
469 fatal(int do_perror, char *message, ...)
472 int save_errno = errno;
473 char buf[FATAL_MSG_SZ];
475 (void) fflush(stdout);
477 va_start(args, message);
478 (void) sprintf(buf, "ztest: ");
480 (void) vsprintf(buf + strlen(buf), message, args);
483 (void) snprintf(buf + strlen(buf), FATAL_MSG_SZ - strlen(buf),
484 ": %s", strerror(save_errno));
486 (void) fprintf(stderr, "%s\n", buf);
487 fatal_msg = buf; /* to ease debugging */
494 str2shift(const char *buf)
496 const char *ends = "BKMGTPEZ";
501 for (i = 0; i < strlen(ends); i++) {
502 if (toupper(buf[0]) == ends[i])
505 if (i == strlen(ends)) {
506 (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n",
510 if (buf[1] == '\0' || (toupper(buf[1]) == 'B' && buf[2] == '\0')) {
513 (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n", buf);
519 nicenumtoull(const char *buf)
524 val = strtoull(buf, &end, 0);
526 (void) fprintf(stderr, "ztest: bad numeric value: %s\n", buf);
528 } else if (end[0] == '.') {
529 double fval = strtod(buf, &end);
530 fval *= pow(2, str2shift(end));
531 if (fval > UINT64_MAX) {
532 (void) fprintf(stderr, "ztest: value too large: %s\n",
536 val = (uint64_t)fval;
538 int shift = str2shift(end);
539 if (shift >= 64 || (val << shift) >> shift != val) {
540 (void) fprintf(stderr, "ztest: value too large: %s\n",
550 usage(boolean_t requested)
552 const ztest_shared_opts_t *zo = &ztest_opts_defaults;
554 char nice_vdev_size[10];
555 char nice_gang_bang[10];
556 FILE *fp = requested ? stdout : stderr;
558 nicenum(zo->zo_vdev_size, nice_vdev_size);
559 nicenum(zo->zo_metaslab_gang_bang, nice_gang_bang);
561 (void) fprintf(fp, "Usage: %s\n"
562 "\t[-v vdevs (default: %llu)]\n"
563 "\t[-s size_of_each_vdev (default: %s)]\n"
564 "\t[-a alignment_shift (default: %d)] use 0 for random\n"
565 "\t[-m mirror_copies (default: %d)]\n"
566 "\t[-r raidz_disks (default: %d)]\n"
567 "\t[-R raidz_parity (default: %d)]\n"
568 "\t[-d datasets (default: %d)]\n"
569 "\t[-t threads (default: %d)]\n"
570 "\t[-g gang_block_threshold (default: %s)]\n"
571 "\t[-i init_count (default: %d)] initialize pool i times\n"
572 "\t[-k kill_percentage (default: %llu%%)]\n"
573 "\t[-p pool_name (default: %s)]\n"
574 "\t[-f dir (default: %s)] file directory for vdev files\n"
575 "\t[-V] verbose (use multiple times for ever more blather)\n"
576 "\t[-E] use existing pool instead of creating new one\n"
577 "\t[-T time (default: %llu sec)] total run time\n"
578 "\t[-F freezeloops (default: %llu)] max loops in spa_freeze()\n"
579 "\t[-P passtime (default: %llu sec)] time per pass\n"
580 "\t[-B alt_ztest (default: <none>)] alternate ztest path\n"
581 "\t[-h] (print help)\n"
584 (u_longlong_t)zo->zo_vdevs, /* -v */
585 nice_vdev_size, /* -s */
586 zo->zo_ashift, /* -a */
587 zo->zo_mirrors, /* -m */
588 zo->zo_raidz, /* -r */
589 zo->zo_raidz_parity, /* -R */
590 zo->zo_datasets, /* -d */
591 zo->zo_threads, /* -t */
592 nice_gang_bang, /* -g */
593 zo->zo_init, /* -i */
594 (u_longlong_t)zo->zo_killrate, /* -k */
595 zo->zo_pool, /* -p */
597 (u_longlong_t)zo->zo_time, /* -T */
598 (u_longlong_t)zo->zo_maxloops, /* -F */
599 (u_longlong_t)zo->zo_passtime);
600 exit(requested ? 0 : 1);
604 process_options(int argc, char **argv)
607 ztest_shared_opts_t *zo = &ztest_opts;
611 char altdir[MAXNAMELEN] = { 0 };
613 bcopy(&ztest_opts_defaults, zo, sizeof (*zo));
615 while ((opt = getopt(argc, argv,
616 "v:s:a:m:r:R:d:t:g:i:k:p:f:VET:P:hF:B:")) != EOF) {
633 value = nicenumtoull(optarg);
637 zo->zo_vdevs = value;
640 zo->zo_vdev_size = MAX(SPA_MINDEVSIZE, value);
643 zo->zo_ashift = value;
646 zo->zo_mirrors = value;
649 zo->zo_raidz = MAX(1, value);
652 zo->zo_raidz_parity = MIN(MAX(value, 1), 3);
655 zo->zo_datasets = MAX(1, value);
658 zo->zo_threads = MAX(1, value);
661 zo->zo_metaslab_gang_bang = MAX(SPA_MINBLOCKSIZE << 1,
668 zo->zo_killrate = value;
671 (void) strlcpy(zo->zo_pool, optarg,
672 sizeof (zo->zo_pool));
675 path = realpath(optarg, NULL);
677 (void) fprintf(stderr, "error: %s: %s\n",
678 optarg, strerror(errno));
681 (void) strlcpy(zo->zo_dir, path,
682 sizeof (zo->zo_dir));
695 zo->zo_passtime = MAX(1, value);
698 zo->zo_maxloops = MAX(1, value);
701 (void) strlcpy(altdir, optarg, sizeof (altdir));
713 zo->zo_raidz_parity = MIN(zo->zo_raidz_parity, zo->zo_raidz - 1);
716 (zo->zo_vdevs > 0 ? zo->zo_time * NANOSEC / zo->zo_vdevs :
719 if (strlen(altdir) > 0) {
727 cmd = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
728 realaltdir = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
730 VERIFY(NULL != realpath(getexecname(), cmd));
731 if (0 != access(altdir, F_OK)) {
732 ztest_dump_core = B_FALSE;
733 fatal(B_TRUE, "invalid alternate ztest path: %s",
736 VERIFY(NULL != realpath(altdir, realaltdir));
739 * 'cmd' should be of the form "<anything>/usr/bin/<isa>/ztest".
740 * We want to extract <isa> to determine if we should use
741 * 32 or 64 bit binaries.
743 bin = strstr(cmd, "/usr/bin/");
744 ztest = strstr(bin, "/ztest");
746 isalen = ztest - isa;
747 (void) snprintf(zo->zo_alt_ztest, sizeof (zo->zo_alt_ztest),
748 "%s/usr/bin/%.*s/ztest", realaltdir, isalen, isa);
749 (void) snprintf(zo->zo_alt_libpath, sizeof (zo->zo_alt_libpath),
750 "%s/usr/lib/%.*s", realaltdir, isalen, isa);
752 if (0 != access(zo->zo_alt_ztest, X_OK)) {
753 ztest_dump_core = B_FALSE;
754 fatal(B_TRUE, "invalid alternate ztest: %s",
756 } else if (0 != access(zo->zo_alt_libpath, X_OK)) {
757 ztest_dump_core = B_FALSE;
758 fatal(B_TRUE, "invalid alternate lib directory %s",
762 umem_free(cmd, MAXPATHLEN);
763 umem_free(realaltdir, MAXPATHLEN);
768 ztest_kill(ztest_shared_t *zs)
770 zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(ztest_spa));
771 zs->zs_space = metaslab_class_get_space(spa_normal_class(ztest_spa));
774 * Before we kill off ztest, make sure that the config is updated.
775 * See comment above spa_config_sync().
777 mutex_enter(&spa_namespace_lock);
778 spa_config_sync(ztest_spa, B_FALSE, B_FALSE);
779 mutex_exit(&spa_namespace_lock);
781 zfs_dbgmsg_print(FTAG);
782 (void) kill(getpid(), SIGKILL);
786 ztest_random(uint64_t range)
790 ASSERT3S(ztest_fd_rand, >=, 0);
795 if (read(ztest_fd_rand, &r, sizeof (r)) != sizeof (r))
796 fatal(1, "short read from /dev/urandom");
803 ztest_record_enospc(const char *s)
805 ztest_shared->zs_enospc_count++;
809 ztest_get_ashift(void)
811 if (ztest_opts.zo_ashift == 0)
812 return (SPA_MINBLOCKSHIFT + ztest_random(3));
813 return (ztest_opts.zo_ashift);
817 make_vdev_file(char *path, char *aux, char *pool, size_t size, uint64_t ashift)
819 char pathbuf[MAXPATHLEN];
824 ashift = ztest_get_ashift();
830 vdev = ztest_shared->zs_vdev_aux;
831 (void) snprintf(path, sizeof (pathbuf),
832 ztest_aux_template, ztest_opts.zo_dir,
833 pool == NULL ? ztest_opts.zo_pool : pool,
836 vdev = ztest_shared->zs_vdev_next_leaf++;
837 (void) snprintf(path, sizeof (pathbuf),
838 ztest_dev_template, ztest_opts.zo_dir,
839 pool == NULL ? ztest_opts.zo_pool : pool, vdev);
844 int fd = open(path, O_RDWR | O_CREAT | O_TRUNC, 0666);
846 fatal(1, "can't open %s", path);
847 if (ftruncate(fd, size) != 0)
848 fatal(1, "can't ftruncate %s", path);
852 VERIFY(nvlist_alloc(&file, NV_UNIQUE_NAME, 0) == 0);
853 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_TYPE, VDEV_TYPE_FILE) == 0);
854 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_PATH, path) == 0);
855 VERIFY(nvlist_add_uint64(file, ZPOOL_CONFIG_ASHIFT, ashift) == 0);
861 make_vdev_raidz(char *path, char *aux, char *pool, size_t size,
862 uint64_t ashift, int r)
864 nvlist_t *raidz, **child;
868 return (make_vdev_file(path, aux, pool, size, ashift));
869 child = umem_alloc(r * sizeof (nvlist_t *), UMEM_NOFAIL);
871 for (c = 0; c < r; c++)
872 child[c] = make_vdev_file(path, aux, pool, size, ashift);
874 VERIFY(nvlist_alloc(&raidz, NV_UNIQUE_NAME, 0) == 0);
875 VERIFY(nvlist_add_string(raidz, ZPOOL_CONFIG_TYPE,
876 VDEV_TYPE_RAIDZ) == 0);
877 VERIFY(nvlist_add_uint64(raidz, ZPOOL_CONFIG_NPARITY,
878 ztest_opts.zo_raidz_parity) == 0);
879 VERIFY(nvlist_add_nvlist_array(raidz, ZPOOL_CONFIG_CHILDREN,
882 for (c = 0; c < r; c++)
883 nvlist_free(child[c]);
885 umem_free(child, r * sizeof (nvlist_t *));
891 make_vdev_mirror(char *path, char *aux, char *pool, size_t size,
892 uint64_t ashift, int r, int m)
894 nvlist_t *mirror, **child;
898 return (make_vdev_raidz(path, aux, pool, size, ashift, r));
900 child = umem_alloc(m * sizeof (nvlist_t *), UMEM_NOFAIL);
902 for (c = 0; c < m; c++)
903 child[c] = make_vdev_raidz(path, aux, pool, size, ashift, r);
905 VERIFY(nvlist_alloc(&mirror, NV_UNIQUE_NAME, 0) == 0);
906 VERIFY(nvlist_add_string(mirror, ZPOOL_CONFIG_TYPE,
907 VDEV_TYPE_MIRROR) == 0);
908 VERIFY(nvlist_add_nvlist_array(mirror, ZPOOL_CONFIG_CHILDREN,
911 for (c = 0; c < m; c++)
912 nvlist_free(child[c]);
914 umem_free(child, m * sizeof (nvlist_t *));
920 make_vdev_root(char *path, char *aux, char *pool, size_t size, uint64_t ashift,
921 int log, int r, int m, int t)
923 nvlist_t *root, **child;
928 child = umem_alloc(t * sizeof (nvlist_t *), UMEM_NOFAIL);
930 for (c = 0; c < t; c++) {
931 child[c] = make_vdev_mirror(path, aux, pool, size, ashift,
933 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_IS_LOG,
937 VERIFY(nvlist_alloc(&root, NV_UNIQUE_NAME, 0) == 0);
938 VERIFY(nvlist_add_string(root, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) == 0);
939 VERIFY(nvlist_add_nvlist_array(root, aux ? aux : ZPOOL_CONFIG_CHILDREN,
942 for (c = 0; c < t; c++)
943 nvlist_free(child[c]);
945 umem_free(child, t * sizeof (nvlist_t *));
951 * Find a random spa version. Returns back a random spa version in the
952 * range [initial_version, SPA_VERSION_FEATURES].
955 ztest_random_spa_version(uint64_t initial_version)
957 uint64_t version = initial_version;
959 if (version <= SPA_VERSION_BEFORE_FEATURES) {
961 ztest_random(SPA_VERSION_BEFORE_FEATURES - version + 1);
964 if (version > SPA_VERSION_BEFORE_FEATURES)
965 version = SPA_VERSION_FEATURES;
967 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
972 ztest_random_blocksize(void)
974 return (1 << (SPA_MINBLOCKSHIFT +
975 ztest_random(SPA_MAXBLOCKSHIFT - SPA_MINBLOCKSHIFT + 1)));
979 ztest_random_ibshift(void)
981 return (DN_MIN_INDBLKSHIFT +
982 ztest_random(DN_MAX_INDBLKSHIFT - DN_MIN_INDBLKSHIFT + 1));
986 ztest_random_vdev_top(spa_t *spa, boolean_t log_ok)
989 vdev_t *rvd = spa->spa_root_vdev;
992 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
995 top = ztest_random(rvd->vdev_children);
996 tvd = rvd->vdev_child[top];
997 } while (tvd->vdev_ishole || (tvd->vdev_islog && !log_ok) ||
998 tvd->vdev_mg == NULL || tvd->vdev_mg->mg_class == NULL);
1004 ztest_random_dsl_prop(zfs_prop_t prop)
1009 value = zfs_prop_random_value(prop, ztest_random(-1ULL));
1010 } while (prop == ZFS_PROP_CHECKSUM && value == ZIO_CHECKSUM_OFF);
1016 ztest_dsl_prop_set_uint64(char *osname, zfs_prop_t prop, uint64_t value,
1019 const char *propname = zfs_prop_to_name(prop);
1020 const char *valname;
1021 char setpoint[MAXPATHLEN];
1025 error = dsl_prop_set_int(osname, propname,
1026 (inherit ? ZPROP_SRC_NONE : ZPROP_SRC_LOCAL), value);
1028 if (error == ENOSPC) {
1029 ztest_record_enospc(FTAG);
1034 VERIFY0(dsl_prop_get_integer(osname, propname, &curval, setpoint));
1036 if (ztest_opts.zo_verbose >= 6) {
1037 VERIFY(zfs_prop_index_to_string(prop, curval, &valname) == 0);
1038 (void) printf("%s %s = %s at '%s'\n",
1039 osname, propname, valname, setpoint);
1046 ztest_spa_prop_set_uint64(zpool_prop_t prop, uint64_t value)
1048 spa_t *spa = ztest_spa;
1049 nvlist_t *props = NULL;
1052 VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
1053 VERIFY(nvlist_add_uint64(props, zpool_prop_to_name(prop), value) == 0);
1055 error = spa_prop_set(spa, props);
1059 if (error == ENOSPC) {
1060 ztest_record_enospc(FTAG);
1069 ztest_rll_init(rll_t *rll)
1071 rll->rll_writer = NULL;
1072 rll->rll_readers = 0;
1073 VERIFY(_mutex_init(&rll->rll_lock, USYNC_THREAD, NULL) == 0);
1074 VERIFY(cond_init(&rll->rll_cv, USYNC_THREAD, NULL) == 0);
1078 ztest_rll_destroy(rll_t *rll)
1080 ASSERT(rll->rll_writer == NULL);
1081 ASSERT(rll->rll_readers == 0);
1082 VERIFY(_mutex_destroy(&rll->rll_lock) == 0);
1083 VERIFY(cond_destroy(&rll->rll_cv) == 0);
1087 ztest_rll_lock(rll_t *rll, rl_type_t type)
1089 VERIFY(mutex_lock(&rll->rll_lock) == 0);
1091 if (type == RL_READER) {
1092 while (rll->rll_writer != NULL)
1093 (void) cond_wait(&rll->rll_cv, &rll->rll_lock);
1096 while (rll->rll_writer != NULL || rll->rll_readers)
1097 (void) cond_wait(&rll->rll_cv, &rll->rll_lock);
1098 rll->rll_writer = curthread;
1101 VERIFY(mutex_unlock(&rll->rll_lock) == 0);
1105 ztest_rll_unlock(rll_t *rll)
1107 VERIFY(mutex_lock(&rll->rll_lock) == 0);
1109 if (rll->rll_writer) {
1110 ASSERT(rll->rll_readers == 0);
1111 rll->rll_writer = NULL;
1113 ASSERT(rll->rll_readers != 0);
1114 ASSERT(rll->rll_writer == NULL);
1118 if (rll->rll_writer == NULL && rll->rll_readers == 0)
1119 VERIFY(cond_broadcast(&rll->rll_cv) == 0);
1121 VERIFY(mutex_unlock(&rll->rll_lock) == 0);
1125 ztest_object_lock(ztest_ds_t *zd, uint64_t object, rl_type_t type)
1127 rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1129 ztest_rll_lock(rll, type);
1133 ztest_object_unlock(ztest_ds_t *zd, uint64_t object)
1135 rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1137 ztest_rll_unlock(rll);
1141 ztest_range_lock(ztest_ds_t *zd, uint64_t object, uint64_t offset,
1142 uint64_t size, rl_type_t type)
1144 uint64_t hash = object ^ (offset % (ZTEST_RANGE_LOCKS + 1));
1145 rll_t *rll = &zd->zd_range_lock[hash & (ZTEST_RANGE_LOCKS - 1)];
1148 rl = umem_alloc(sizeof (*rl), UMEM_NOFAIL);
1149 rl->rl_object = object;
1150 rl->rl_offset = offset;
1154 ztest_rll_lock(rll, type);
1160 ztest_range_unlock(rl_t *rl)
1162 rll_t *rll = rl->rl_lock;
1164 ztest_rll_unlock(rll);
1166 umem_free(rl, sizeof (*rl));
1170 ztest_zd_init(ztest_ds_t *zd, ztest_shared_ds_t *szd, objset_t *os)
1173 zd->zd_zilog = dmu_objset_zil(os);
1174 zd->zd_shared = szd;
1175 dmu_objset_name(os, zd->zd_name);
1177 if (zd->zd_shared != NULL)
1178 zd->zd_shared->zd_seq = 0;
1180 VERIFY(rwlock_init(&zd->zd_zilog_lock, USYNC_THREAD, NULL) == 0);
1181 VERIFY(_mutex_init(&zd->zd_dirobj_lock, USYNC_THREAD, NULL) == 0);
1183 for (int l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1184 ztest_rll_init(&zd->zd_object_lock[l]);
1186 for (int l = 0; l < ZTEST_RANGE_LOCKS; l++)
1187 ztest_rll_init(&zd->zd_range_lock[l]);
1191 ztest_zd_fini(ztest_ds_t *zd)
1193 VERIFY(_mutex_destroy(&zd->zd_dirobj_lock) == 0);
1195 for (int l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1196 ztest_rll_destroy(&zd->zd_object_lock[l]);
1198 for (int l = 0; l < ZTEST_RANGE_LOCKS; l++)
1199 ztest_rll_destroy(&zd->zd_range_lock[l]);
1202 #define TXG_MIGHTWAIT (ztest_random(10) == 0 ? TXG_NOWAIT : TXG_WAIT)
1205 ztest_tx_assign(dmu_tx_t *tx, uint64_t txg_how, const char *tag)
1211 * Attempt to assign tx to some transaction group.
1213 error = dmu_tx_assign(tx, txg_how);
1215 if (error == ERESTART) {
1216 ASSERT(txg_how == TXG_NOWAIT);
1219 ASSERT3U(error, ==, ENOSPC);
1220 ztest_record_enospc(tag);
1225 txg = dmu_tx_get_txg(tx);
1231 ztest_pattern_set(void *buf, uint64_t size, uint64_t value)
1234 uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1241 ztest_pattern_match(void *buf, uint64_t size, uint64_t value)
1244 uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1248 diff |= (value - *ip++);
1254 ztest_bt_generate(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1255 uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg)
1257 bt->bt_magic = BT_MAGIC;
1258 bt->bt_objset = dmu_objset_id(os);
1259 bt->bt_object = object;
1260 bt->bt_offset = offset;
1263 bt->bt_crtxg = crtxg;
1267 ztest_bt_verify(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1268 uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg)
1270 ASSERT(bt->bt_magic == BT_MAGIC);
1271 ASSERT(bt->bt_objset == dmu_objset_id(os));
1272 ASSERT(bt->bt_object == object);
1273 ASSERT(bt->bt_offset == offset);
1274 ASSERT(bt->bt_gen <= gen);
1275 ASSERT(bt->bt_txg <= txg);
1276 ASSERT(bt->bt_crtxg == crtxg);
1279 static ztest_block_tag_t *
1280 ztest_bt_bonus(dmu_buf_t *db)
1282 dmu_object_info_t doi;
1283 ztest_block_tag_t *bt;
1285 dmu_object_info_from_db(db, &doi);
1286 ASSERT3U(doi.doi_bonus_size, <=, db->db_size);
1287 ASSERT3U(doi.doi_bonus_size, >=, sizeof (*bt));
1288 bt = (void *)((char *)db->db_data + doi.doi_bonus_size - sizeof (*bt));
1297 #define lrz_type lr_mode
1298 #define lrz_blocksize lr_uid
1299 #define lrz_ibshift lr_gid
1300 #define lrz_bonustype lr_rdev
1301 #define lrz_bonuslen lr_crtime[1]
1304 ztest_log_create(ztest_ds_t *zd, dmu_tx_t *tx, lr_create_t *lr)
1306 char *name = (void *)(lr + 1); /* name follows lr */
1307 size_t namesize = strlen(name) + 1;
1310 if (zil_replaying(zd->zd_zilog, tx))
1313 itx = zil_itx_create(TX_CREATE, sizeof (*lr) + namesize);
1314 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1315 sizeof (*lr) + namesize - sizeof (lr_t));
1317 zil_itx_assign(zd->zd_zilog, itx, tx);
1321 ztest_log_remove(ztest_ds_t *zd, dmu_tx_t *tx, lr_remove_t *lr, uint64_t object)
1323 char *name = (void *)(lr + 1); /* name follows lr */
1324 size_t namesize = strlen(name) + 1;
1327 if (zil_replaying(zd->zd_zilog, tx))
1330 itx = zil_itx_create(TX_REMOVE, sizeof (*lr) + namesize);
1331 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1332 sizeof (*lr) + namesize - sizeof (lr_t));
1334 itx->itx_oid = object;
1335 zil_itx_assign(zd->zd_zilog, itx, tx);
1339 ztest_log_write(ztest_ds_t *zd, dmu_tx_t *tx, lr_write_t *lr)
1342 itx_wr_state_t write_state = ztest_random(WR_NUM_STATES);
1344 if (zil_replaying(zd->zd_zilog, tx))
1347 if (lr->lr_length > ZIL_MAX_LOG_DATA)
1348 write_state = WR_INDIRECT;
1350 itx = zil_itx_create(TX_WRITE,
1351 sizeof (*lr) + (write_state == WR_COPIED ? lr->lr_length : 0));
1353 if (write_state == WR_COPIED &&
1354 dmu_read(zd->zd_os, lr->lr_foid, lr->lr_offset, lr->lr_length,
1355 ((lr_write_t *)&itx->itx_lr) + 1, DMU_READ_NO_PREFETCH) != 0) {
1356 zil_itx_destroy(itx);
1357 itx = zil_itx_create(TX_WRITE, sizeof (*lr));
1358 write_state = WR_NEED_COPY;
1360 itx->itx_private = zd;
1361 itx->itx_wr_state = write_state;
1362 itx->itx_sync = (ztest_random(8) == 0);
1363 itx->itx_sod += (write_state == WR_NEED_COPY ? lr->lr_length : 0);
1365 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1366 sizeof (*lr) - sizeof (lr_t));
1368 zil_itx_assign(zd->zd_zilog, itx, tx);
1372 ztest_log_truncate(ztest_ds_t *zd, dmu_tx_t *tx, lr_truncate_t *lr)
1376 if (zil_replaying(zd->zd_zilog, tx))
1379 itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr));
1380 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1381 sizeof (*lr) - sizeof (lr_t));
1383 itx->itx_sync = B_FALSE;
1384 zil_itx_assign(zd->zd_zilog, itx, tx);
1388 ztest_log_setattr(ztest_ds_t *zd, dmu_tx_t *tx, lr_setattr_t *lr)
1392 if (zil_replaying(zd->zd_zilog, tx))
1395 itx = zil_itx_create(TX_SETATTR, sizeof (*lr));
1396 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1397 sizeof (*lr) - sizeof (lr_t));
1399 itx->itx_sync = B_FALSE;
1400 zil_itx_assign(zd->zd_zilog, itx, tx);
1407 ztest_replay_create(ztest_ds_t *zd, lr_create_t *lr, boolean_t byteswap)
1409 char *name = (void *)(lr + 1); /* name follows lr */
1410 objset_t *os = zd->zd_os;
1411 ztest_block_tag_t *bbt;
1418 byteswap_uint64_array(lr, sizeof (*lr));
1420 ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1421 ASSERT(name[0] != '\0');
1423 tx = dmu_tx_create(os);
1425 dmu_tx_hold_zap(tx, lr->lr_doid, B_TRUE, name);
1427 if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1428 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, B_TRUE, NULL);
1430 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1433 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1437 ASSERT(dmu_objset_zil(os)->zl_replay == !!lr->lr_foid);
1439 if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1440 if (lr->lr_foid == 0) {
1441 lr->lr_foid = zap_create(os,
1442 lr->lrz_type, lr->lrz_bonustype,
1443 lr->lrz_bonuslen, tx);
1445 error = zap_create_claim(os, lr->lr_foid,
1446 lr->lrz_type, lr->lrz_bonustype,
1447 lr->lrz_bonuslen, tx);
1450 if (lr->lr_foid == 0) {
1451 lr->lr_foid = dmu_object_alloc(os,
1452 lr->lrz_type, 0, lr->lrz_bonustype,
1453 lr->lrz_bonuslen, tx);
1455 error = dmu_object_claim(os, lr->lr_foid,
1456 lr->lrz_type, 0, lr->lrz_bonustype,
1457 lr->lrz_bonuslen, tx);
1462 ASSERT3U(error, ==, EEXIST);
1463 ASSERT(zd->zd_zilog->zl_replay);
1468 ASSERT(lr->lr_foid != 0);
1470 if (lr->lrz_type != DMU_OT_ZAP_OTHER)
1471 VERIFY3U(0, ==, dmu_object_set_blocksize(os, lr->lr_foid,
1472 lr->lrz_blocksize, lr->lrz_ibshift, tx));
1474 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1475 bbt = ztest_bt_bonus(db);
1476 dmu_buf_will_dirty(db, tx);
1477 ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_gen, txg, txg);
1478 dmu_buf_rele(db, FTAG);
1480 VERIFY3U(0, ==, zap_add(os, lr->lr_doid, name, sizeof (uint64_t), 1,
1483 (void) ztest_log_create(zd, tx, lr);
1491 ztest_replay_remove(ztest_ds_t *zd, lr_remove_t *lr, boolean_t byteswap)
1493 char *name = (void *)(lr + 1); /* name follows lr */
1494 objset_t *os = zd->zd_os;
1495 dmu_object_info_t doi;
1497 uint64_t object, txg;
1500 byteswap_uint64_array(lr, sizeof (*lr));
1502 ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1503 ASSERT(name[0] != '\0');
1506 zap_lookup(os, lr->lr_doid, name, sizeof (object), 1, &object));
1507 ASSERT(object != 0);
1509 ztest_object_lock(zd, object, RL_WRITER);
1511 VERIFY3U(0, ==, dmu_object_info(os, object, &doi));
1513 tx = dmu_tx_create(os);
1515 dmu_tx_hold_zap(tx, lr->lr_doid, B_FALSE, name);
1516 dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END);
1518 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1520 ztest_object_unlock(zd, object);
1524 if (doi.doi_type == DMU_OT_ZAP_OTHER) {
1525 VERIFY3U(0, ==, zap_destroy(os, object, tx));
1527 VERIFY3U(0, ==, dmu_object_free(os, object, tx));
1530 VERIFY3U(0, ==, zap_remove(os, lr->lr_doid, name, tx));
1532 (void) ztest_log_remove(zd, tx, lr, object);
1536 ztest_object_unlock(zd, object);
1542 ztest_replay_write(ztest_ds_t *zd, lr_write_t *lr, boolean_t byteswap)
1544 objset_t *os = zd->zd_os;
1545 void *data = lr + 1; /* data follows lr */
1546 uint64_t offset, length;
1547 ztest_block_tag_t *bt = data;
1548 ztest_block_tag_t *bbt;
1549 uint64_t gen, txg, lrtxg, crtxg;
1550 dmu_object_info_t doi;
1553 arc_buf_t *abuf = NULL;
1557 byteswap_uint64_array(lr, sizeof (*lr));
1559 offset = lr->lr_offset;
1560 length = lr->lr_length;
1562 /* If it's a dmu_sync() block, write the whole block */
1563 if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
1564 uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
1565 if (length < blocksize) {
1566 offset -= offset % blocksize;
1571 if (bt->bt_magic == BSWAP_64(BT_MAGIC))
1572 byteswap_uint64_array(bt, sizeof (*bt));
1574 if (bt->bt_magic != BT_MAGIC)
1577 ztest_object_lock(zd, lr->lr_foid, RL_READER);
1578 rl = ztest_range_lock(zd, lr->lr_foid, offset, length, RL_WRITER);
1580 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1582 dmu_object_info_from_db(db, &doi);
1584 bbt = ztest_bt_bonus(db);
1585 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1587 crtxg = bbt->bt_crtxg;
1588 lrtxg = lr->lr_common.lrc_txg;
1590 tx = dmu_tx_create(os);
1592 dmu_tx_hold_write(tx, lr->lr_foid, offset, length);
1594 if (ztest_random(8) == 0 && length == doi.doi_data_block_size &&
1595 P2PHASE(offset, length) == 0)
1596 abuf = dmu_request_arcbuf(db, length);
1598 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1601 dmu_return_arcbuf(abuf);
1602 dmu_buf_rele(db, FTAG);
1603 ztest_range_unlock(rl);
1604 ztest_object_unlock(zd, lr->lr_foid);
1610 * Usually, verify the old data before writing new data --
1611 * but not always, because we also want to verify correct
1612 * behavior when the data was not recently read into cache.
1614 ASSERT(offset % doi.doi_data_block_size == 0);
1615 if (ztest_random(4) != 0) {
1616 int prefetch = ztest_random(2) ?
1617 DMU_READ_PREFETCH : DMU_READ_NO_PREFETCH;
1618 ztest_block_tag_t rbt;
1620 VERIFY(dmu_read(os, lr->lr_foid, offset,
1621 sizeof (rbt), &rbt, prefetch) == 0);
1622 if (rbt.bt_magic == BT_MAGIC) {
1623 ztest_bt_verify(&rbt, os, lr->lr_foid,
1624 offset, gen, txg, crtxg);
1629 * Writes can appear to be newer than the bonus buffer because
1630 * the ztest_get_data() callback does a dmu_read() of the
1631 * open-context data, which may be different than the data
1632 * as it was when the write was generated.
1634 if (zd->zd_zilog->zl_replay) {
1635 ztest_bt_verify(bt, os, lr->lr_foid, offset,
1636 MAX(gen, bt->bt_gen), MAX(txg, lrtxg),
1641 * Set the bt's gen/txg to the bonus buffer's gen/txg
1642 * so that all of the usual ASSERTs will work.
1644 ztest_bt_generate(bt, os, lr->lr_foid, offset, gen, txg, crtxg);
1648 dmu_write(os, lr->lr_foid, offset, length, data, tx);
1650 bcopy(data, abuf->b_data, length);
1651 dmu_assign_arcbuf(db, offset, abuf, tx);
1654 (void) ztest_log_write(zd, tx, lr);
1656 dmu_buf_rele(db, FTAG);
1660 ztest_range_unlock(rl);
1661 ztest_object_unlock(zd, lr->lr_foid);
1667 ztest_replay_truncate(ztest_ds_t *zd, lr_truncate_t *lr, boolean_t byteswap)
1669 objset_t *os = zd->zd_os;
1675 byteswap_uint64_array(lr, sizeof (*lr));
1677 ztest_object_lock(zd, lr->lr_foid, RL_READER);
1678 rl = ztest_range_lock(zd, lr->lr_foid, lr->lr_offset, lr->lr_length,
1681 tx = dmu_tx_create(os);
1683 dmu_tx_hold_free(tx, lr->lr_foid, lr->lr_offset, lr->lr_length);
1685 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1687 ztest_range_unlock(rl);
1688 ztest_object_unlock(zd, lr->lr_foid);
1692 VERIFY(dmu_free_range(os, lr->lr_foid, lr->lr_offset,
1693 lr->lr_length, tx) == 0);
1695 (void) ztest_log_truncate(zd, tx, lr);
1699 ztest_range_unlock(rl);
1700 ztest_object_unlock(zd, lr->lr_foid);
1706 ztest_replay_setattr(ztest_ds_t *zd, lr_setattr_t *lr, boolean_t byteswap)
1708 objset_t *os = zd->zd_os;
1711 ztest_block_tag_t *bbt;
1712 uint64_t txg, lrtxg, crtxg;
1715 byteswap_uint64_array(lr, sizeof (*lr));
1717 ztest_object_lock(zd, lr->lr_foid, RL_WRITER);
1719 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1721 tx = dmu_tx_create(os);
1722 dmu_tx_hold_bonus(tx, lr->lr_foid);
1724 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1726 dmu_buf_rele(db, FTAG);
1727 ztest_object_unlock(zd, lr->lr_foid);
1731 bbt = ztest_bt_bonus(db);
1732 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1733 crtxg = bbt->bt_crtxg;
1734 lrtxg = lr->lr_common.lrc_txg;
1736 if (zd->zd_zilog->zl_replay) {
1737 ASSERT(lr->lr_size != 0);
1738 ASSERT(lr->lr_mode != 0);
1742 * Randomly change the size and increment the generation.
1744 lr->lr_size = (ztest_random(db->db_size / sizeof (*bbt)) + 1) *
1746 lr->lr_mode = bbt->bt_gen + 1;
1751 * Verify that the current bonus buffer is not newer than our txg.
1753 ztest_bt_verify(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode,
1754 MAX(txg, lrtxg), crtxg);
1756 dmu_buf_will_dirty(db, tx);
1758 ASSERT3U(lr->lr_size, >=, sizeof (*bbt));
1759 ASSERT3U(lr->lr_size, <=, db->db_size);
1760 VERIFY0(dmu_set_bonus(db, lr->lr_size, tx));
1761 bbt = ztest_bt_bonus(db);
1763 ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode, txg, crtxg);
1765 dmu_buf_rele(db, FTAG);
1767 (void) ztest_log_setattr(zd, tx, lr);
1771 ztest_object_unlock(zd, lr->lr_foid);
1776 zil_replay_func_t *ztest_replay_vector[TX_MAX_TYPE] = {
1777 NULL, /* 0 no such transaction type */
1778 ztest_replay_create, /* TX_CREATE */
1779 NULL, /* TX_MKDIR */
1780 NULL, /* TX_MKXATTR */
1781 NULL, /* TX_SYMLINK */
1782 ztest_replay_remove, /* TX_REMOVE */
1783 NULL, /* TX_RMDIR */
1785 NULL, /* TX_RENAME */
1786 ztest_replay_write, /* TX_WRITE */
1787 ztest_replay_truncate, /* TX_TRUNCATE */
1788 ztest_replay_setattr, /* TX_SETATTR */
1790 NULL, /* TX_CREATE_ACL */
1791 NULL, /* TX_CREATE_ATTR */
1792 NULL, /* TX_CREATE_ACL_ATTR */
1793 NULL, /* TX_MKDIR_ACL */
1794 NULL, /* TX_MKDIR_ATTR */
1795 NULL, /* TX_MKDIR_ACL_ATTR */
1796 NULL, /* TX_WRITE2 */
1800 * ZIL get_data callbacks
1804 ztest_get_done(zgd_t *zgd, int error)
1806 ztest_ds_t *zd = zgd->zgd_private;
1807 uint64_t object = zgd->zgd_rl->rl_object;
1810 dmu_buf_rele(zgd->zgd_db, zgd);
1812 ztest_range_unlock(zgd->zgd_rl);
1813 ztest_object_unlock(zd, object);
1815 if (error == 0 && zgd->zgd_bp)
1816 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
1818 umem_free(zgd, sizeof (*zgd));
1822 ztest_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
1824 ztest_ds_t *zd = arg;
1825 objset_t *os = zd->zd_os;
1826 uint64_t object = lr->lr_foid;
1827 uint64_t offset = lr->lr_offset;
1828 uint64_t size = lr->lr_length;
1829 blkptr_t *bp = &lr->lr_blkptr;
1830 uint64_t txg = lr->lr_common.lrc_txg;
1832 dmu_object_info_t doi;
1837 ztest_object_lock(zd, object, RL_READER);
1838 error = dmu_bonus_hold(os, object, FTAG, &db);
1840 ztest_object_unlock(zd, object);
1844 crtxg = ztest_bt_bonus(db)->bt_crtxg;
1846 if (crtxg == 0 || crtxg > txg) {
1847 dmu_buf_rele(db, FTAG);
1848 ztest_object_unlock(zd, object);
1852 dmu_object_info_from_db(db, &doi);
1853 dmu_buf_rele(db, FTAG);
1856 zgd = umem_zalloc(sizeof (*zgd), UMEM_NOFAIL);
1857 zgd->zgd_zilog = zd->zd_zilog;
1858 zgd->zgd_private = zd;
1860 if (buf != NULL) { /* immediate write */
1861 zgd->zgd_rl = ztest_range_lock(zd, object, offset, size,
1864 error = dmu_read(os, object, offset, size, buf,
1865 DMU_READ_NO_PREFETCH);
1868 size = doi.doi_data_block_size;
1870 offset = P2ALIGN(offset, size);
1872 ASSERT(offset < size);
1876 zgd->zgd_rl = ztest_range_lock(zd, object, offset, size,
1879 error = dmu_buf_hold(os, object, offset, zgd, &db,
1880 DMU_READ_NO_PREFETCH);
1883 blkptr_t *obp = dmu_buf_get_blkptr(db);
1885 ASSERT(BP_IS_HOLE(bp));
1892 ASSERT(db->db_offset == offset);
1893 ASSERT(db->db_size == size);
1895 error = dmu_sync(zio, lr->lr_common.lrc_txg,
1896 ztest_get_done, zgd);
1903 ztest_get_done(zgd, error);
1909 ztest_lr_alloc(size_t lrsize, char *name)
1912 size_t namesize = name ? strlen(name) + 1 : 0;
1914 lr = umem_zalloc(lrsize + namesize, UMEM_NOFAIL);
1917 bcopy(name, lr + lrsize, namesize);
1923 ztest_lr_free(void *lr, size_t lrsize, char *name)
1925 size_t namesize = name ? strlen(name) + 1 : 0;
1927 umem_free(lr, lrsize + namesize);
1931 * Lookup a bunch of objects. Returns the number of objects not found.
1934 ztest_lookup(ztest_ds_t *zd, ztest_od_t *od, int count)
1939 ASSERT(_mutex_held(&zd->zd_dirobj_lock));
1941 for (int i = 0; i < count; i++, od++) {
1943 error = zap_lookup(zd->zd_os, od->od_dir, od->od_name,
1944 sizeof (uint64_t), 1, &od->od_object);
1946 ASSERT(error == ENOENT);
1947 ASSERT(od->od_object == 0);
1951 ztest_block_tag_t *bbt;
1952 dmu_object_info_t doi;
1954 ASSERT(od->od_object != 0);
1955 ASSERT(missing == 0); /* there should be no gaps */
1957 ztest_object_lock(zd, od->od_object, RL_READER);
1958 VERIFY3U(0, ==, dmu_bonus_hold(zd->zd_os,
1959 od->od_object, FTAG, &db));
1960 dmu_object_info_from_db(db, &doi);
1961 bbt = ztest_bt_bonus(db);
1962 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1963 od->od_type = doi.doi_type;
1964 od->od_blocksize = doi.doi_data_block_size;
1965 od->od_gen = bbt->bt_gen;
1966 dmu_buf_rele(db, FTAG);
1967 ztest_object_unlock(zd, od->od_object);
1975 ztest_create(ztest_ds_t *zd, ztest_od_t *od, int count)
1979 ASSERT(_mutex_held(&zd->zd_dirobj_lock));
1981 for (int i = 0; i < count; i++, od++) {
1988 lr_create_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
1990 lr->lr_doid = od->od_dir;
1991 lr->lr_foid = 0; /* 0 to allocate, > 0 to claim */
1992 lr->lrz_type = od->od_crtype;
1993 lr->lrz_blocksize = od->od_crblocksize;
1994 lr->lrz_ibshift = ztest_random_ibshift();
1995 lr->lrz_bonustype = DMU_OT_UINT64_OTHER;
1996 lr->lrz_bonuslen = dmu_bonus_max();
1997 lr->lr_gen = od->od_crgen;
1998 lr->lr_crtime[0] = time(NULL);
2000 if (ztest_replay_create(zd, lr, B_FALSE) != 0) {
2001 ASSERT(missing == 0);
2005 od->od_object = lr->lr_foid;
2006 od->od_type = od->od_crtype;
2007 od->od_blocksize = od->od_crblocksize;
2008 od->od_gen = od->od_crgen;
2009 ASSERT(od->od_object != 0);
2012 ztest_lr_free(lr, sizeof (*lr), od->od_name);
2019 ztest_remove(ztest_ds_t *zd, ztest_od_t *od, int count)
2024 ASSERT(_mutex_held(&zd->zd_dirobj_lock));
2028 for (int i = count - 1; i >= 0; i--, od--) {
2035 * No object was found.
2037 if (od->od_object == 0)
2040 lr_remove_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
2042 lr->lr_doid = od->od_dir;
2044 if ((error = ztest_replay_remove(zd, lr, B_FALSE)) != 0) {
2045 ASSERT3U(error, ==, ENOSPC);
2050 ztest_lr_free(lr, sizeof (*lr), od->od_name);
2057 ztest_write(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size,
2063 lr = ztest_lr_alloc(sizeof (*lr) + size, NULL);
2065 lr->lr_foid = object;
2066 lr->lr_offset = offset;
2067 lr->lr_length = size;
2069 BP_ZERO(&lr->lr_blkptr);
2071 bcopy(data, lr + 1, size);
2073 error = ztest_replay_write(zd, lr, B_FALSE);
2075 ztest_lr_free(lr, sizeof (*lr) + size, NULL);
2081 ztest_truncate(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2086 lr = ztest_lr_alloc(sizeof (*lr), NULL);
2088 lr->lr_foid = object;
2089 lr->lr_offset = offset;
2090 lr->lr_length = size;
2092 error = ztest_replay_truncate(zd, lr, B_FALSE);
2094 ztest_lr_free(lr, sizeof (*lr), NULL);
2100 ztest_setattr(ztest_ds_t *zd, uint64_t object)
2105 lr = ztest_lr_alloc(sizeof (*lr), NULL);
2107 lr->lr_foid = object;
2111 error = ztest_replay_setattr(zd, lr, B_FALSE);
2113 ztest_lr_free(lr, sizeof (*lr), NULL);
2119 ztest_prealloc(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2121 objset_t *os = zd->zd_os;
2126 txg_wait_synced(dmu_objset_pool(os), 0);
2128 ztest_object_lock(zd, object, RL_READER);
2129 rl = ztest_range_lock(zd, object, offset, size, RL_WRITER);
2131 tx = dmu_tx_create(os);
2133 dmu_tx_hold_write(tx, object, offset, size);
2135 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
2138 dmu_prealloc(os, object, offset, size, tx);
2140 txg_wait_synced(dmu_objset_pool(os), txg);
2142 (void) dmu_free_long_range(os, object, offset, size);
2145 ztest_range_unlock(rl);
2146 ztest_object_unlock(zd, object);
2150 ztest_io(ztest_ds_t *zd, uint64_t object, uint64_t offset)
2153 ztest_block_tag_t wbt;
2154 dmu_object_info_t doi;
2155 enum ztest_io_type io_type;
2159 VERIFY(dmu_object_info(zd->zd_os, object, &doi) == 0);
2160 blocksize = doi.doi_data_block_size;
2161 data = umem_alloc(blocksize, UMEM_NOFAIL);
2164 * Pick an i/o type at random, biased toward writing block tags.
2166 io_type = ztest_random(ZTEST_IO_TYPES);
2167 if (ztest_random(2) == 0)
2168 io_type = ZTEST_IO_WRITE_TAG;
2170 (void) rw_rdlock(&zd->zd_zilog_lock);
2174 case ZTEST_IO_WRITE_TAG:
2175 ztest_bt_generate(&wbt, zd->zd_os, object, offset, 0, 0, 0);
2176 (void) ztest_write(zd, object, offset, sizeof (wbt), &wbt);
2179 case ZTEST_IO_WRITE_PATTERN:
2180 (void) memset(data, 'a' + (object + offset) % 5, blocksize);
2181 if (ztest_random(2) == 0) {
2183 * Induce fletcher2 collisions to ensure that
2184 * zio_ddt_collision() detects and resolves them
2185 * when using fletcher2-verify for deduplication.
2187 ((uint64_t *)data)[0] ^= 1ULL << 63;
2188 ((uint64_t *)data)[4] ^= 1ULL << 63;
2190 (void) ztest_write(zd, object, offset, blocksize, data);
2193 case ZTEST_IO_WRITE_ZEROES:
2194 bzero(data, blocksize);
2195 (void) ztest_write(zd, object, offset, blocksize, data);
2198 case ZTEST_IO_TRUNCATE:
2199 (void) ztest_truncate(zd, object, offset, blocksize);
2202 case ZTEST_IO_SETATTR:
2203 (void) ztest_setattr(zd, object);
2206 case ZTEST_IO_REWRITE:
2207 (void) rw_rdlock(&ztest_name_lock);
2208 err = ztest_dsl_prop_set_uint64(zd->zd_name,
2209 ZFS_PROP_CHECKSUM, spa_dedup_checksum(ztest_spa),
2211 VERIFY(err == 0 || err == ENOSPC);
2212 err = ztest_dsl_prop_set_uint64(zd->zd_name,
2213 ZFS_PROP_COMPRESSION,
2214 ztest_random_dsl_prop(ZFS_PROP_COMPRESSION),
2216 VERIFY(err == 0 || err == ENOSPC);
2217 (void) rw_unlock(&ztest_name_lock);
2219 VERIFY0(dmu_read(zd->zd_os, object, offset, blocksize, data,
2220 DMU_READ_NO_PREFETCH));
2222 (void) ztest_write(zd, object, offset, blocksize, data);
2226 (void) rw_unlock(&zd->zd_zilog_lock);
2228 umem_free(data, blocksize);
2232 * Initialize an object description template.
2235 ztest_od_init(ztest_od_t *od, uint64_t id, char *tag, uint64_t index,
2236 dmu_object_type_t type, uint64_t blocksize, uint64_t gen)
2238 od->od_dir = ZTEST_DIROBJ;
2241 od->od_crtype = type;
2242 od->od_crblocksize = blocksize ? blocksize : ztest_random_blocksize();
2245 od->od_type = DMU_OT_NONE;
2246 od->od_blocksize = 0;
2249 (void) snprintf(od->od_name, sizeof (od->od_name), "%s(%lld)[%llu]",
2250 tag, (int64_t)id, index);
2254 * Lookup or create the objects for a test using the od template.
2255 * If the objects do not all exist, or if 'remove' is specified,
2256 * remove any existing objects and create new ones. Otherwise,
2257 * use the existing objects.
2260 ztest_object_init(ztest_ds_t *zd, ztest_od_t *od, size_t size, boolean_t remove)
2262 int count = size / sizeof (*od);
2265 VERIFY(mutex_lock(&zd->zd_dirobj_lock) == 0);
2266 if ((ztest_lookup(zd, od, count) != 0 || remove) &&
2267 (ztest_remove(zd, od, count) != 0 ||
2268 ztest_create(zd, od, count) != 0))
2271 VERIFY(mutex_unlock(&zd->zd_dirobj_lock) == 0);
2278 ztest_zil_commit(ztest_ds_t *zd, uint64_t id)
2280 zilog_t *zilog = zd->zd_zilog;
2282 (void) rw_rdlock(&zd->zd_zilog_lock);
2284 zil_commit(zilog, ztest_random(ZTEST_OBJECTS));
2287 * Remember the committed values in zd, which is in parent/child
2288 * shared memory. If we die, the next iteration of ztest_run()
2289 * will verify that the log really does contain this record.
2291 mutex_enter(&zilog->zl_lock);
2292 ASSERT(zd->zd_shared != NULL);
2293 ASSERT3U(zd->zd_shared->zd_seq, <=, zilog->zl_commit_lr_seq);
2294 zd->zd_shared->zd_seq = zilog->zl_commit_lr_seq;
2295 mutex_exit(&zilog->zl_lock);
2297 (void) rw_unlock(&zd->zd_zilog_lock);
2301 * This function is designed to simulate the operations that occur during a
2302 * mount/unmount operation. We hold the dataset across these operations in an
2303 * attempt to expose any implicit assumptions about ZIL management.
2307 ztest_zil_remount(ztest_ds_t *zd, uint64_t id)
2309 objset_t *os = zd->zd_os;
2312 * We grab the zd_dirobj_lock to ensure that no other thread is
2313 * updating the zil (i.e. adding in-memory log records) and the
2314 * zd_zilog_lock to block any I/O.
2316 VERIFY0(mutex_lock(&zd->zd_dirobj_lock));
2317 (void) rw_wrlock(&zd->zd_zilog_lock);
2319 /* zfsvfs_teardown() */
2320 zil_close(zd->zd_zilog);
2322 /* zfsvfs_setup() */
2323 VERIFY(zil_open(os, ztest_get_data) == zd->zd_zilog);
2324 zil_replay(os, zd, ztest_replay_vector);
2326 (void) rw_unlock(&zd->zd_zilog_lock);
2327 VERIFY(mutex_unlock(&zd->zd_dirobj_lock) == 0);
2331 * Verify that we can't destroy an active pool, create an existing pool,
2332 * or create a pool with a bad vdev spec.
2336 ztest_spa_create_destroy(ztest_ds_t *zd, uint64_t id)
2338 ztest_shared_opts_t *zo = &ztest_opts;
2343 * Attempt to create using a bad file.
2345 nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1);
2346 VERIFY3U(ENOENT, ==,
2347 spa_create("ztest_bad_file", nvroot, NULL, NULL));
2348 nvlist_free(nvroot);
2351 * Attempt to create using a bad mirror.
2353 nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 2, 1);
2354 VERIFY3U(ENOENT, ==,
2355 spa_create("ztest_bad_mirror", nvroot, NULL, NULL));
2356 nvlist_free(nvroot);
2359 * Attempt to create an existing pool. It shouldn't matter
2360 * what's in the nvroot; we should fail with EEXIST.
2362 (void) rw_rdlock(&ztest_name_lock);
2363 nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1);
2364 VERIFY3U(EEXIST, ==, spa_create(zo->zo_pool, nvroot, NULL, NULL));
2365 nvlist_free(nvroot);
2366 VERIFY3U(0, ==, spa_open(zo->zo_pool, &spa, FTAG));
2367 VERIFY3U(EBUSY, ==, spa_destroy(zo->zo_pool));
2368 spa_close(spa, FTAG);
2370 (void) rw_unlock(&ztest_name_lock);
2375 ztest_spa_upgrade(ztest_ds_t *zd, uint64_t id)
2378 uint64_t initial_version = SPA_VERSION_INITIAL;
2379 uint64_t version, newversion;
2380 nvlist_t *nvroot, *props;
2383 VERIFY0(mutex_lock(&ztest_vdev_lock));
2384 name = kmem_asprintf("%s_upgrade", ztest_opts.zo_pool);
2387 * Clean up from previous runs.
2389 (void) spa_destroy(name);
2391 nvroot = make_vdev_root(NULL, NULL, name, ztest_opts.zo_vdev_size, 0,
2392 0, ztest_opts.zo_raidz, ztest_opts.zo_mirrors, 1);
2395 * If we're configuring a RAIDZ device then make sure that the
2396 * the initial version is capable of supporting that feature.
2398 switch (ztest_opts.zo_raidz_parity) {
2401 initial_version = SPA_VERSION_INITIAL;
2404 initial_version = SPA_VERSION_RAIDZ2;
2407 initial_version = SPA_VERSION_RAIDZ3;
2412 * Create a pool with a spa version that can be upgraded. Pick
2413 * a value between initial_version and SPA_VERSION_BEFORE_FEATURES.
2416 version = ztest_random_spa_version(initial_version);
2417 } while (version > SPA_VERSION_BEFORE_FEATURES);
2419 props = fnvlist_alloc();
2420 fnvlist_add_uint64(props,
2421 zpool_prop_to_name(ZPOOL_PROP_VERSION), version);
2422 VERIFY0(spa_create(name, nvroot, props, NULL));
2423 fnvlist_free(nvroot);
2424 fnvlist_free(props);
2426 VERIFY0(spa_open(name, &spa, FTAG));
2427 VERIFY3U(spa_version(spa), ==, version);
2428 newversion = ztest_random_spa_version(version + 1);
2430 if (ztest_opts.zo_verbose >= 4) {
2431 (void) printf("upgrading spa version from %llu to %llu\n",
2432 (u_longlong_t)version, (u_longlong_t)newversion);
2435 spa_upgrade(spa, newversion);
2436 VERIFY3U(spa_version(spa), >, version);
2437 VERIFY3U(spa_version(spa), ==, fnvlist_lookup_uint64(spa->spa_config,
2438 zpool_prop_to_name(ZPOOL_PROP_VERSION)));
2439 spa_close(spa, FTAG);
2442 VERIFY0(mutex_unlock(&ztest_vdev_lock));
2446 vdev_lookup_by_path(vdev_t *vd, const char *path)
2450 if (vd->vdev_path != NULL && strcmp(path, vd->vdev_path) == 0)
2453 for (int c = 0; c < vd->vdev_children; c++)
2454 if ((mvd = vdev_lookup_by_path(vd->vdev_child[c], path)) !=
2462 * Find the first available hole which can be used as a top-level.
2465 find_vdev_hole(spa_t *spa)
2467 vdev_t *rvd = spa->spa_root_vdev;
2470 ASSERT(spa_config_held(spa, SCL_VDEV, RW_READER) == SCL_VDEV);
2472 for (c = 0; c < rvd->vdev_children; c++) {
2473 vdev_t *cvd = rvd->vdev_child[c];
2475 if (cvd->vdev_ishole)
2482 * Verify that vdev_add() works as expected.
2486 ztest_vdev_add_remove(ztest_ds_t *zd, uint64_t id)
2488 ztest_shared_t *zs = ztest_shared;
2489 spa_t *spa = ztest_spa;
2495 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2496 leaves = MAX(zs->zs_mirrors + zs->zs_splits, 1) * ztest_opts.zo_raidz;
2498 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2500 ztest_shared->zs_vdev_next_leaf = find_vdev_hole(spa) * leaves;
2503 * If we have slogs then remove them 1/4 of the time.
2505 if (spa_has_slogs(spa) && ztest_random(4) == 0) {
2507 * Grab the guid from the head of the log class rotor.
2509 guid = spa_log_class(spa)->mc_rotor->mg_vd->vdev_guid;
2511 spa_config_exit(spa, SCL_VDEV, FTAG);
2514 * We have to grab the zs_name_lock as writer to
2515 * prevent a race between removing a slog (dmu_objset_find)
2516 * and destroying a dataset. Removing the slog will
2517 * grab a reference on the dataset which may cause
2518 * dmu_objset_destroy() to fail with EBUSY thus
2519 * leaving the dataset in an inconsistent state.
2521 VERIFY(rw_wrlock(&ztest_name_lock) == 0);
2522 error = spa_vdev_remove(spa, guid, B_FALSE);
2523 VERIFY(rw_unlock(&ztest_name_lock) == 0);
2525 if (error && error != EEXIST)
2526 fatal(0, "spa_vdev_remove() = %d", error);
2528 spa_config_exit(spa, SCL_VDEV, FTAG);
2531 * Make 1/4 of the devices be log devices.
2533 nvroot = make_vdev_root(NULL, NULL, NULL,
2534 ztest_opts.zo_vdev_size, 0,
2535 ztest_random(4) == 0, ztest_opts.zo_raidz,
2538 error = spa_vdev_add(spa, nvroot);
2539 nvlist_free(nvroot);
2541 if (error == ENOSPC)
2542 ztest_record_enospc("spa_vdev_add");
2543 else if (error != 0)
2544 fatal(0, "spa_vdev_add() = %d", error);
2547 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2551 * Verify that adding/removing aux devices (l2arc, hot spare) works as expected.
2555 ztest_vdev_aux_add_remove(ztest_ds_t *zd, uint64_t id)
2557 ztest_shared_t *zs = ztest_shared;
2558 spa_t *spa = ztest_spa;
2559 vdev_t *rvd = spa->spa_root_vdev;
2560 spa_aux_vdev_t *sav;
2565 if (ztest_random(2) == 0) {
2566 sav = &spa->spa_spares;
2567 aux = ZPOOL_CONFIG_SPARES;
2569 sav = &spa->spa_l2cache;
2570 aux = ZPOOL_CONFIG_L2CACHE;
2573 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2575 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2577 if (sav->sav_count != 0 && ztest_random(4) == 0) {
2579 * Pick a random device to remove.
2581 guid = sav->sav_vdevs[ztest_random(sav->sav_count)]->vdev_guid;
2584 * Find an unused device we can add.
2586 zs->zs_vdev_aux = 0;
2588 char path[MAXPATHLEN];
2590 (void) snprintf(path, sizeof (path), ztest_aux_template,
2591 ztest_opts.zo_dir, ztest_opts.zo_pool, aux,
2593 for (c = 0; c < sav->sav_count; c++)
2594 if (strcmp(sav->sav_vdevs[c]->vdev_path,
2597 if (c == sav->sav_count &&
2598 vdev_lookup_by_path(rvd, path) == NULL)
2604 spa_config_exit(spa, SCL_VDEV, FTAG);
2610 nvlist_t *nvroot = make_vdev_root(NULL, aux, NULL,
2611 (ztest_opts.zo_vdev_size * 5) / 4, 0, 0, 0, 0, 1);
2612 error = spa_vdev_add(spa, nvroot);
2614 fatal(0, "spa_vdev_add(%p) = %d", nvroot, error);
2615 nvlist_free(nvroot);
2618 * Remove an existing device. Sometimes, dirty its
2619 * vdev state first to make sure we handle removal
2620 * of devices that have pending state changes.
2622 if (ztest_random(2) == 0)
2623 (void) vdev_online(spa, guid, 0, NULL);
2625 error = spa_vdev_remove(spa, guid, B_FALSE);
2626 if (error != 0 && error != EBUSY)
2627 fatal(0, "spa_vdev_remove(%llu) = %d", guid, error);
2630 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2634 * split a pool if it has mirror tlvdevs
2638 ztest_split_pool(ztest_ds_t *zd, uint64_t id)
2640 ztest_shared_t *zs = ztest_shared;
2641 spa_t *spa = ztest_spa;
2642 vdev_t *rvd = spa->spa_root_vdev;
2643 nvlist_t *tree, **child, *config, *split, **schild;
2644 uint_t c, children, schildren = 0, lastlogid = 0;
2647 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2649 /* ensure we have a useable config; mirrors of raidz aren't supported */
2650 if (zs->zs_mirrors < 3 || ztest_opts.zo_raidz > 1) {
2651 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2655 /* clean up the old pool, if any */
2656 (void) spa_destroy("splitp");
2658 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2660 /* generate a config from the existing config */
2661 mutex_enter(&spa->spa_props_lock);
2662 VERIFY(nvlist_lookup_nvlist(spa->spa_config, ZPOOL_CONFIG_VDEV_TREE,
2664 mutex_exit(&spa->spa_props_lock);
2666 VERIFY(nvlist_lookup_nvlist_array(tree, ZPOOL_CONFIG_CHILDREN, &child,
2669 schild = malloc(rvd->vdev_children * sizeof (nvlist_t *));
2670 for (c = 0; c < children; c++) {
2671 vdev_t *tvd = rvd->vdev_child[c];
2675 if (tvd->vdev_islog || tvd->vdev_ops == &vdev_hole_ops) {
2676 VERIFY(nvlist_alloc(&schild[schildren], NV_UNIQUE_NAME,
2678 VERIFY(nvlist_add_string(schild[schildren],
2679 ZPOOL_CONFIG_TYPE, VDEV_TYPE_HOLE) == 0);
2680 VERIFY(nvlist_add_uint64(schild[schildren],
2681 ZPOOL_CONFIG_IS_HOLE, 1) == 0);
2683 lastlogid = schildren;
2688 VERIFY(nvlist_lookup_nvlist_array(child[c],
2689 ZPOOL_CONFIG_CHILDREN, &mchild, &mchildren) == 0);
2690 VERIFY(nvlist_dup(mchild[0], &schild[schildren++], 0) == 0);
2693 /* OK, create a config that can be used to split */
2694 VERIFY(nvlist_alloc(&split, NV_UNIQUE_NAME, 0) == 0);
2695 VERIFY(nvlist_add_string(split, ZPOOL_CONFIG_TYPE,
2696 VDEV_TYPE_ROOT) == 0);
2697 VERIFY(nvlist_add_nvlist_array(split, ZPOOL_CONFIG_CHILDREN, schild,
2698 lastlogid != 0 ? lastlogid : schildren) == 0);
2700 VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, 0) == 0);
2701 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, split) == 0);
2703 for (c = 0; c < schildren; c++)
2704 nvlist_free(schild[c]);
2708 spa_config_exit(spa, SCL_VDEV, FTAG);
2710 (void) rw_wrlock(&ztest_name_lock);
2711 error = spa_vdev_split_mirror(spa, "splitp", config, NULL, B_FALSE);
2712 (void) rw_unlock(&ztest_name_lock);
2714 nvlist_free(config);
2717 (void) printf("successful split - results:\n");
2718 mutex_enter(&spa_namespace_lock);
2719 show_pool_stats(spa);
2720 show_pool_stats(spa_lookup("splitp"));
2721 mutex_exit(&spa_namespace_lock);
2725 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2730 * Verify that we can attach and detach devices.
2734 ztest_vdev_attach_detach(ztest_ds_t *zd, uint64_t id)
2736 ztest_shared_t *zs = ztest_shared;
2737 spa_t *spa = ztest_spa;
2738 spa_aux_vdev_t *sav = &spa->spa_spares;
2739 vdev_t *rvd = spa->spa_root_vdev;
2740 vdev_t *oldvd, *newvd, *pvd;
2744 uint64_t ashift = ztest_get_ashift();
2745 uint64_t oldguid, pguid;
2746 uint64_t oldsize, newsize;
2747 char oldpath[MAXPATHLEN], newpath[MAXPATHLEN];
2749 int oldvd_has_siblings = B_FALSE;
2750 int newvd_is_spare = B_FALSE;
2752 int error, expected_error;
2754 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2755 leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz;
2757 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2760 * Decide whether to do an attach or a replace.
2762 replacing = ztest_random(2);
2765 * Pick a random top-level vdev.
2767 top = ztest_random_vdev_top(spa, B_TRUE);
2770 * Pick a random leaf within it.
2772 leaf = ztest_random(leaves);
2777 oldvd = rvd->vdev_child[top];
2778 if (zs->zs_mirrors >= 1) {
2779 ASSERT(oldvd->vdev_ops == &vdev_mirror_ops);
2780 ASSERT(oldvd->vdev_children >= zs->zs_mirrors);
2781 oldvd = oldvd->vdev_child[leaf / ztest_opts.zo_raidz];
2783 if (ztest_opts.zo_raidz > 1) {
2784 ASSERT(oldvd->vdev_ops == &vdev_raidz_ops);
2785 ASSERT(oldvd->vdev_children == ztest_opts.zo_raidz);
2786 oldvd = oldvd->vdev_child[leaf % ztest_opts.zo_raidz];
2790 * If we're already doing an attach or replace, oldvd may be a
2791 * mirror vdev -- in which case, pick a random child.
2793 while (oldvd->vdev_children != 0) {
2794 oldvd_has_siblings = B_TRUE;
2795 ASSERT(oldvd->vdev_children >= 2);
2796 oldvd = oldvd->vdev_child[ztest_random(oldvd->vdev_children)];
2799 oldguid = oldvd->vdev_guid;
2800 oldsize = vdev_get_min_asize(oldvd);
2801 oldvd_is_log = oldvd->vdev_top->vdev_islog;
2802 (void) strcpy(oldpath, oldvd->vdev_path);
2803 pvd = oldvd->vdev_parent;
2804 pguid = pvd->vdev_guid;
2807 * If oldvd has siblings, then half of the time, detach it.
2809 if (oldvd_has_siblings && ztest_random(2) == 0) {
2810 spa_config_exit(spa, SCL_VDEV, FTAG);
2811 error = spa_vdev_detach(spa, oldguid, pguid, B_FALSE);
2812 if (error != 0 && error != ENODEV && error != EBUSY &&
2814 fatal(0, "detach (%s) returned %d", oldpath, error);
2815 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2820 * For the new vdev, choose with equal probability between the two
2821 * standard paths (ending in either 'a' or 'b') or a random hot spare.
2823 if (sav->sav_count != 0 && ztest_random(3) == 0) {
2824 newvd = sav->sav_vdevs[ztest_random(sav->sav_count)];
2825 newvd_is_spare = B_TRUE;
2826 (void) strcpy(newpath, newvd->vdev_path);
2828 (void) snprintf(newpath, sizeof (newpath), ztest_dev_template,
2829 ztest_opts.zo_dir, ztest_opts.zo_pool,
2830 top * leaves + leaf);
2831 if (ztest_random(2) == 0)
2832 newpath[strlen(newpath) - 1] = 'b';
2833 newvd = vdev_lookup_by_path(rvd, newpath);
2837 newsize = vdev_get_min_asize(newvd);
2840 * Make newsize a little bigger or smaller than oldsize.
2841 * If it's smaller, the attach should fail.
2842 * If it's larger, and we're doing a replace,
2843 * we should get dynamic LUN growth when we're done.
2845 newsize = 10 * oldsize / (9 + ztest_random(3));
2849 * If pvd is not a mirror or root, the attach should fail with ENOTSUP,
2850 * unless it's a replace; in that case any non-replacing parent is OK.
2852 * If newvd is already part of the pool, it should fail with EBUSY.
2854 * If newvd is too small, it should fail with EOVERFLOW.
2856 if (pvd->vdev_ops != &vdev_mirror_ops &&
2857 pvd->vdev_ops != &vdev_root_ops && (!replacing ||
2858 pvd->vdev_ops == &vdev_replacing_ops ||
2859 pvd->vdev_ops == &vdev_spare_ops))
2860 expected_error = ENOTSUP;
2861 else if (newvd_is_spare && (!replacing || oldvd_is_log))
2862 expected_error = ENOTSUP;
2863 else if (newvd == oldvd)
2864 expected_error = replacing ? 0 : EBUSY;
2865 else if (vdev_lookup_by_path(rvd, newpath) != NULL)
2866 expected_error = EBUSY;
2867 else if (newsize < oldsize)
2868 expected_error = EOVERFLOW;
2869 else if (ashift > oldvd->vdev_top->vdev_ashift)
2870 expected_error = EDOM;
2874 spa_config_exit(spa, SCL_VDEV, FTAG);
2877 * Build the nvlist describing newpath.
2879 root = make_vdev_root(newpath, NULL, NULL, newvd == NULL ? newsize : 0,
2880 ashift, 0, 0, 0, 1);
2882 error = spa_vdev_attach(spa, oldguid, root, replacing);
2887 * If our parent was the replacing vdev, but the replace completed,
2888 * then instead of failing with ENOTSUP we may either succeed,
2889 * fail with ENODEV, or fail with EOVERFLOW.
2891 if (expected_error == ENOTSUP &&
2892 (error == 0 || error == ENODEV || error == EOVERFLOW))
2893 expected_error = error;
2896 * If someone grew the LUN, the replacement may be too small.
2898 if (error == EOVERFLOW || error == EBUSY)
2899 expected_error = error;
2901 /* XXX workaround 6690467 */
2902 if (error != expected_error && expected_error != EBUSY) {
2903 fatal(0, "attach (%s %llu, %s %llu, %d) "
2904 "returned %d, expected %d",
2905 oldpath, oldsize, newpath,
2906 newsize, replacing, error, expected_error);
2909 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2913 * Callback function which expands the physical size of the vdev.
2916 grow_vdev(vdev_t *vd, void *arg)
2918 spa_t *spa = vd->vdev_spa;
2919 size_t *newsize = arg;
2923 ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
2924 ASSERT(vd->vdev_ops->vdev_op_leaf);
2926 if ((fd = open(vd->vdev_path, O_RDWR)) == -1)
2929 fsize = lseek(fd, 0, SEEK_END);
2930 (void) ftruncate(fd, *newsize);
2932 if (ztest_opts.zo_verbose >= 6) {
2933 (void) printf("%s grew from %lu to %lu bytes\n",
2934 vd->vdev_path, (ulong_t)fsize, (ulong_t)*newsize);
2941 * Callback function which expands a given vdev by calling vdev_online().
2945 online_vdev(vdev_t *vd, void *arg)
2947 spa_t *spa = vd->vdev_spa;
2948 vdev_t *tvd = vd->vdev_top;
2949 uint64_t guid = vd->vdev_guid;
2950 uint64_t generation = spa->spa_config_generation + 1;
2951 vdev_state_t newstate = VDEV_STATE_UNKNOWN;
2954 ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
2955 ASSERT(vd->vdev_ops->vdev_op_leaf);
2957 /* Calling vdev_online will initialize the new metaslabs */
2958 spa_config_exit(spa, SCL_STATE, spa);
2959 error = vdev_online(spa, guid, ZFS_ONLINE_EXPAND, &newstate);
2960 spa_config_enter(spa, SCL_STATE, spa, RW_READER);
2963 * If vdev_online returned an error or the underlying vdev_open
2964 * failed then we abort the expand. The only way to know that
2965 * vdev_open fails is by checking the returned newstate.
2967 if (error || newstate != VDEV_STATE_HEALTHY) {
2968 if (ztest_opts.zo_verbose >= 5) {
2969 (void) printf("Unable to expand vdev, state %llu, "
2970 "error %d\n", (u_longlong_t)newstate, error);
2974 ASSERT3U(newstate, ==, VDEV_STATE_HEALTHY);
2977 * Since we dropped the lock we need to ensure that we're
2978 * still talking to the original vdev. It's possible this
2979 * vdev may have been detached/replaced while we were
2980 * trying to online it.
2982 if (generation != spa->spa_config_generation) {
2983 if (ztest_opts.zo_verbose >= 5) {
2984 (void) printf("vdev configuration has changed, "
2985 "guid %llu, state %llu, expected gen %llu, "
2988 (u_longlong_t)tvd->vdev_state,
2989 (u_longlong_t)generation,
2990 (u_longlong_t)spa->spa_config_generation);
2998 * Traverse the vdev tree calling the supplied function.
2999 * We continue to walk the tree until we either have walked all
3000 * children or we receive a non-NULL return from the callback.
3001 * If a NULL callback is passed, then we just return back the first
3002 * leaf vdev we encounter.
3005 vdev_walk_tree(vdev_t *vd, vdev_t *(*func)(vdev_t *, void *), void *arg)
3007 if (vd->vdev_ops->vdev_op_leaf) {
3011 return (func(vd, arg));
3014 for (uint_t c = 0; c < vd->vdev_children; c++) {
3015 vdev_t *cvd = vd->vdev_child[c];
3016 if ((cvd = vdev_walk_tree(cvd, func, arg)) != NULL)
3023 * Verify that dynamic LUN growth works as expected.
3027 ztest_vdev_LUN_growth(ztest_ds_t *zd, uint64_t id)
3029 spa_t *spa = ztest_spa;
3031 metaslab_class_t *mc;
3032 metaslab_group_t *mg;
3033 size_t psize, newsize;
3035 uint64_t old_class_space, new_class_space, old_ms_count, new_ms_count;
3037 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
3038 spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3040 top = ztest_random_vdev_top(spa, B_TRUE);
3042 tvd = spa->spa_root_vdev->vdev_child[top];
3045 old_ms_count = tvd->vdev_ms_count;
3046 old_class_space = metaslab_class_get_space(mc);
3049 * Determine the size of the first leaf vdev associated with
3050 * our top-level device.
3052 vd = vdev_walk_tree(tvd, NULL, NULL);
3053 ASSERT3P(vd, !=, NULL);
3054 ASSERT(vd->vdev_ops->vdev_op_leaf);
3056 psize = vd->vdev_psize;
3059 * We only try to expand the vdev if it's healthy, less than 4x its
3060 * original size, and it has a valid psize.
3062 if (tvd->vdev_state != VDEV_STATE_HEALTHY ||
3063 psize == 0 || psize >= 4 * ztest_opts.zo_vdev_size) {
3064 spa_config_exit(spa, SCL_STATE, spa);
3065 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3069 newsize = psize + psize / 8;
3070 ASSERT3U(newsize, >, psize);
3072 if (ztest_opts.zo_verbose >= 6) {
3073 (void) printf("Expanding LUN %s from %lu to %lu\n",
3074 vd->vdev_path, (ulong_t)psize, (ulong_t)newsize);
3078 * Growing the vdev is a two step process:
3079 * 1). expand the physical size (i.e. relabel)
3080 * 2). online the vdev to create the new metaslabs
3082 if (vdev_walk_tree(tvd, grow_vdev, &newsize) != NULL ||
3083 vdev_walk_tree(tvd, online_vdev, NULL) != NULL ||
3084 tvd->vdev_state != VDEV_STATE_HEALTHY) {
3085 if (ztest_opts.zo_verbose >= 5) {
3086 (void) printf("Could not expand LUN because "
3087 "the vdev configuration changed.\n");
3089 spa_config_exit(spa, SCL_STATE, spa);
3090 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3094 spa_config_exit(spa, SCL_STATE, spa);
3097 * Expanding the LUN will update the config asynchronously,
3098 * thus we must wait for the async thread to complete any
3099 * pending tasks before proceeding.
3103 mutex_enter(&spa->spa_async_lock);
3104 done = (spa->spa_async_thread == NULL && !spa->spa_async_tasks);
3105 mutex_exit(&spa->spa_async_lock);
3108 txg_wait_synced(spa_get_dsl(spa), 0);
3109 (void) poll(NULL, 0, 100);
3112 spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3114 tvd = spa->spa_root_vdev->vdev_child[top];
3115 new_ms_count = tvd->vdev_ms_count;
3116 new_class_space = metaslab_class_get_space(mc);
3118 if (tvd->vdev_mg != mg || mg->mg_class != mc) {
3119 if (ztest_opts.zo_verbose >= 5) {
3120 (void) printf("Could not verify LUN expansion due to "
3121 "intervening vdev offline or remove.\n");
3123 spa_config_exit(spa, SCL_STATE, spa);
3124 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3129 * Make sure we were able to grow the vdev.
3131 if (new_ms_count <= old_ms_count)
3132 fatal(0, "LUN expansion failed: ms_count %llu <= %llu\n",
3133 old_ms_count, new_ms_count);
3136 * Make sure we were able to grow the pool.
3138 if (new_class_space <= old_class_space)
3139 fatal(0, "LUN expansion failed: class_space %llu <= %llu\n",
3140 old_class_space, new_class_space);
3142 if (ztest_opts.zo_verbose >= 5) {
3143 char oldnumbuf[6], newnumbuf[6];
3145 nicenum(old_class_space, oldnumbuf);
3146 nicenum(new_class_space, newnumbuf);
3147 (void) printf("%s grew from %s to %s\n",
3148 spa->spa_name, oldnumbuf, newnumbuf);
3151 spa_config_exit(spa, SCL_STATE, spa);
3152 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3156 * Verify that dmu_objset_{create,destroy,open,close} work as expected.
3160 ztest_objset_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
3163 * Create the objects common to all ztest datasets.
3165 VERIFY(zap_create_claim(os, ZTEST_DIROBJ,
3166 DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx) == 0);
3170 ztest_dataset_create(char *dsname)
3172 uint64_t zilset = ztest_random(100);
3173 int err = dmu_objset_create(dsname, DMU_OST_OTHER, 0,
3174 ztest_objset_create_cb, NULL);
3176 if (err || zilset < 80)
3179 if (ztest_opts.zo_verbose >= 6)
3180 (void) printf("Setting dataset %s to sync always\n", dsname);
3181 return (ztest_dsl_prop_set_uint64(dsname, ZFS_PROP_SYNC,
3182 ZFS_SYNC_ALWAYS, B_FALSE));
3187 ztest_objset_destroy_cb(const char *name, void *arg)
3190 dmu_object_info_t doi;
3194 * Verify that the dataset contains a directory object.
3196 VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_TRUE, FTAG, &os));
3197 error = dmu_object_info(os, ZTEST_DIROBJ, &doi);
3198 if (error != ENOENT) {
3199 /* We could have crashed in the middle of destroying it */
3201 ASSERT3U(doi.doi_type, ==, DMU_OT_ZAP_OTHER);
3202 ASSERT3S(doi.doi_physical_blocks_512, >=, 0);
3204 dmu_objset_disown(os, FTAG);
3207 * Destroy the dataset.
3209 if (strchr(name, '@') != NULL) {
3210 VERIFY0(dsl_destroy_snapshot(name, B_FALSE));
3212 VERIFY0(dsl_destroy_head(name));
3218 ztest_snapshot_create(char *osname, uint64_t id)
3220 char snapname[MAXNAMELEN];
3223 (void) snprintf(snapname, sizeof (snapname), "%llu", (u_longlong_t)id);
3225 error = dmu_objset_snapshot_one(osname, snapname);
3226 if (error == ENOSPC) {
3227 ztest_record_enospc(FTAG);
3230 if (error != 0 && error != EEXIST) {
3231 fatal(0, "ztest_snapshot_create(%s@%s) = %d", osname,
3238 ztest_snapshot_destroy(char *osname, uint64_t id)
3240 char snapname[MAXNAMELEN];
3243 (void) snprintf(snapname, MAXNAMELEN, "%s@%llu", osname,
3246 error = dsl_destroy_snapshot(snapname, B_FALSE);
3247 if (error != 0 && error != ENOENT)
3248 fatal(0, "ztest_snapshot_destroy(%s) = %d", snapname, error);
3254 ztest_dmu_objset_create_destroy(ztest_ds_t *zd, uint64_t id)
3260 char name[MAXNAMELEN];
3263 (void) rw_rdlock(&ztest_name_lock);
3265 (void) snprintf(name, MAXNAMELEN, "%s/temp_%llu",
3266 ztest_opts.zo_pool, (u_longlong_t)id);
3269 * If this dataset exists from a previous run, process its replay log
3270 * half of the time. If we don't replay it, then dmu_objset_destroy()
3271 * (invoked from ztest_objset_destroy_cb()) should just throw it away.
3273 if (ztest_random(2) == 0 &&
3274 dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os) == 0) {
3275 ztest_zd_init(&zdtmp, NULL, os);
3276 zil_replay(os, &zdtmp, ztest_replay_vector);
3277 ztest_zd_fini(&zdtmp);
3278 dmu_objset_disown(os, FTAG);
3282 * There may be an old instance of the dataset we're about to
3283 * create lying around from a previous run. If so, destroy it
3284 * and all of its snapshots.
3286 (void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
3287 DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
3290 * Verify that the destroyed dataset is no longer in the namespace.
3292 VERIFY3U(ENOENT, ==, dmu_objset_own(name, DMU_OST_OTHER, B_TRUE,
3296 * Verify that we can create a new dataset.
3298 error = ztest_dataset_create(name);
3300 if (error == ENOSPC) {
3301 ztest_record_enospc(FTAG);
3302 (void) rw_unlock(&ztest_name_lock);
3305 fatal(0, "dmu_objset_create(%s) = %d", name, error);
3308 VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os));
3310 ztest_zd_init(&zdtmp, NULL, os);
3313 * Open the intent log for it.
3315 zilog = zil_open(os, ztest_get_data);
3318 * Put some objects in there, do a little I/O to them,
3319 * and randomly take a couple of snapshots along the way.
3321 iters = ztest_random(5);
3322 for (int i = 0; i < iters; i++) {
3323 ztest_dmu_object_alloc_free(&zdtmp, id);
3324 if (ztest_random(iters) == 0)
3325 (void) ztest_snapshot_create(name, i);
3329 * Verify that we cannot create an existing dataset.
3331 VERIFY3U(EEXIST, ==,
3332 dmu_objset_create(name, DMU_OST_OTHER, 0, NULL, NULL));
3335 * Verify that we can hold an objset that is also owned.
3337 VERIFY3U(0, ==, dmu_objset_hold(name, FTAG, &os2));
3338 dmu_objset_rele(os2, FTAG);
3341 * Verify that we cannot own an objset that is already owned.
3344 dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os2));
3347 dmu_objset_disown(os, FTAG);
3348 ztest_zd_fini(&zdtmp);
3350 (void) rw_unlock(&ztest_name_lock);
3354 * Verify that dmu_snapshot_{create,destroy,open,close} work as expected.
3357 ztest_dmu_snapshot_create_destroy(ztest_ds_t *zd, uint64_t id)
3359 (void) rw_rdlock(&ztest_name_lock);
3360 (void) ztest_snapshot_destroy(zd->zd_name, id);
3361 (void) ztest_snapshot_create(zd->zd_name, id);
3362 (void) rw_unlock(&ztest_name_lock);
3366 * Cleanup non-standard snapshots and clones.
3369 ztest_dsl_dataset_cleanup(char *osname, uint64_t id)
3371 char snap1name[MAXNAMELEN];
3372 char clone1name[MAXNAMELEN];
3373 char snap2name[MAXNAMELEN];
3374 char clone2name[MAXNAMELEN];
3375 char snap3name[MAXNAMELEN];
3378 (void) snprintf(snap1name, MAXNAMELEN, "%s@s1_%llu", osname, id);
3379 (void) snprintf(clone1name, MAXNAMELEN, "%s/c1_%llu", osname, id);
3380 (void) snprintf(snap2name, MAXNAMELEN, "%s@s2_%llu", clone1name, id);
3381 (void) snprintf(clone2name, MAXNAMELEN, "%s/c2_%llu", osname, id);
3382 (void) snprintf(snap3name, MAXNAMELEN, "%s@s3_%llu", clone1name, id);
3384 error = dsl_destroy_head(clone2name);
3385 if (error && error != ENOENT)
3386 fatal(0, "dsl_destroy_head(%s) = %d", clone2name, error);
3387 error = dsl_destroy_snapshot(snap3name, B_FALSE);
3388 if (error && error != ENOENT)
3389 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap3name, error);
3390 error = dsl_destroy_snapshot(snap2name, B_FALSE);
3391 if (error && error != ENOENT)
3392 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap2name, error);
3393 error = dsl_destroy_head(clone1name);
3394 if (error && error != ENOENT)
3395 fatal(0, "dsl_destroy_head(%s) = %d", clone1name, error);
3396 error = dsl_destroy_snapshot(snap1name, B_FALSE);
3397 if (error && error != ENOENT)
3398 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap1name, error);
3402 * Verify dsl_dataset_promote handles EBUSY
3405 ztest_dsl_dataset_promote_busy(ztest_ds_t *zd, uint64_t id)
3408 char snap1name[MAXNAMELEN];
3409 char clone1name[MAXNAMELEN];
3410 char snap2name[MAXNAMELEN];
3411 char clone2name[MAXNAMELEN];
3412 char snap3name[MAXNAMELEN];
3413 char *osname = zd->zd_name;
3416 (void) rw_rdlock(&ztest_name_lock);
3418 ztest_dsl_dataset_cleanup(osname, id);
3420 (void) snprintf(snap1name, MAXNAMELEN, "%s@s1_%llu", osname, id);
3421 (void) snprintf(clone1name, MAXNAMELEN, "%s/c1_%llu", osname, id);
3422 (void) snprintf(snap2name, MAXNAMELEN, "%s@s2_%llu", clone1name, id);
3423 (void) snprintf(clone2name, MAXNAMELEN, "%s/c2_%llu", osname, id);
3424 (void) snprintf(snap3name, MAXNAMELEN, "%s@s3_%llu", clone1name, id);
3426 error = dmu_objset_snapshot_one(osname, strchr(snap1name, '@') + 1);
3427 if (error && error != EEXIST) {
3428 if (error == ENOSPC) {
3429 ztest_record_enospc(FTAG);
3432 fatal(0, "dmu_take_snapshot(%s) = %d", snap1name, error);
3435 error = dmu_objset_clone(clone1name, snap1name);
3437 if (error == ENOSPC) {
3438 ztest_record_enospc(FTAG);
3441 fatal(0, "dmu_objset_create(%s) = %d", clone1name, error);
3444 error = dmu_objset_snapshot_one(clone1name, strchr(snap2name, '@') + 1);
3445 if (error && error != EEXIST) {
3446 if (error == ENOSPC) {
3447 ztest_record_enospc(FTAG);
3450 fatal(0, "dmu_open_snapshot(%s) = %d", snap2name, error);
3453 error = dmu_objset_snapshot_one(clone1name, strchr(snap3name, '@') + 1);
3454 if (error && error != EEXIST) {
3455 if (error == ENOSPC) {
3456 ztest_record_enospc(FTAG);
3459 fatal(0, "dmu_open_snapshot(%s) = %d", snap3name, error);
3462 error = dmu_objset_clone(clone2name, snap3name);
3464 if (error == ENOSPC) {
3465 ztest_record_enospc(FTAG);
3468 fatal(0, "dmu_objset_create(%s) = %d", clone2name, error);
3471 error = dmu_objset_own(snap2name, DMU_OST_ANY, B_TRUE, FTAG, &os);
3473 fatal(0, "dmu_objset_own(%s) = %d", snap2name, error);
3474 error = dsl_dataset_promote(clone2name, NULL);
3476 fatal(0, "dsl_dataset_promote(%s), %d, not EBUSY", clone2name,
3478 dmu_objset_disown(os, FTAG);
3481 ztest_dsl_dataset_cleanup(osname, id);
3483 (void) rw_unlock(&ztest_name_lock);
3487 * Verify that dmu_object_{alloc,free} work as expected.
3490 ztest_dmu_object_alloc_free(ztest_ds_t *zd, uint64_t id)
3493 int batchsize = sizeof (od) / sizeof (od[0]);
3495 for (int b = 0; b < batchsize; b++)
3496 ztest_od_init(&od[b], id, FTAG, b, DMU_OT_UINT64_OTHER, 0, 0);
3499 * Destroy the previous batch of objects, create a new batch,
3500 * and do some I/O on the new objects.
3502 if (ztest_object_init(zd, od, sizeof (od), B_TRUE) != 0)
3505 while (ztest_random(4 * batchsize) != 0)
3506 ztest_io(zd, od[ztest_random(batchsize)].od_object,
3507 ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
3511 * Verify that dmu_{read,write} work as expected.
3514 ztest_dmu_read_write(ztest_ds_t *zd, uint64_t id)
3516 objset_t *os = zd->zd_os;
3519 int i, freeit, error;
3521 bufwad_t *packbuf, *bigbuf, *pack, *bigH, *bigT;
3522 uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
3523 uint64_t chunksize = (1000 + ztest_random(1000)) * sizeof (uint64_t);
3524 uint64_t regions = 997;
3525 uint64_t stride = 123456789ULL;
3526 uint64_t width = 40;
3527 int free_percent = 5;
3530 * This test uses two objects, packobj and bigobj, that are always
3531 * updated together (i.e. in the same tx) so that their contents are
3532 * in sync and can be compared. Their contents relate to each other
3533 * in a simple way: packobj is a dense array of 'bufwad' structures,
3534 * while bigobj is a sparse array of the same bufwads. Specifically,
3535 * for any index n, there are three bufwads that should be identical:
3537 * packobj, at offset n * sizeof (bufwad_t)
3538 * bigobj, at the head of the nth chunk
3539 * bigobj, at the tail of the nth chunk
3541 * The chunk size is arbitrary. It doesn't have to be a power of two,
3542 * and it doesn't have any relation to the object blocksize.
3543 * The only requirement is that it can hold at least two bufwads.
3545 * Normally, we write the bufwad to each of these locations.
3546 * However, free_percent of the time we instead write zeroes to
3547 * packobj and perform a dmu_free_range() on bigobj. By comparing
3548 * bigobj to packobj, we can verify that the DMU is correctly
3549 * tracking which parts of an object are allocated and free,
3550 * and that the contents of the allocated blocks are correct.
3554 * Read the directory info. If it's the first time, set things up.
3556 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, chunksize);
3557 ztest_od_init(&od[1], id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize);
3559 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
3562 bigobj = od[0].od_object;
3563 packobj = od[1].od_object;
3564 chunksize = od[0].od_gen;
3565 ASSERT(chunksize == od[1].od_gen);
3568 * Prefetch a random chunk of the big object.
3569 * Our aim here is to get some async reads in flight
3570 * for blocks that we may free below; the DMU should
3571 * handle this race correctly.
3573 n = ztest_random(regions) * stride + ztest_random(width);
3574 s = 1 + ztest_random(2 * width - 1);
3575 dmu_prefetch(os, bigobj, n * chunksize, s * chunksize);
3578 * Pick a random index and compute the offsets into packobj and bigobj.
3580 n = ztest_random(regions) * stride + ztest_random(width);
3581 s = 1 + ztest_random(width - 1);
3583 packoff = n * sizeof (bufwad_t);
3584 packsize = s * sizeof (bufwad_t);
3586 bigoff = n * chunksize;
3587 bigsize = s * chunksize;
3589 packbuf = umem_alloc(packsize, UMEM_NOFAIL);
3590 bigbuf = umem_alloc(bigsize, UMEM_NOFAIL);
3593 * free_percent of the time, free a range of bigobj rather than
3596 freeit = (ztest_random(100) < free_percent);
3599 * Read the current contents of our objects.
3601 error = dmu_read(os, packobj, packoff, packsize, packbuf,
3604 error = dmu_read(os, bigobj, bigoff, bigsize, bigbuf,
3609 * Get a tx for the mods to both packobj and bigobj.
3611 tx = dmu_tx_create(os);
3613 dmu_tx_hold_write(tx, packobj, packoff, packsize);
3616 dmu_tx_hold_free(tx, bigobj, bigoff, bigsize);
3618 dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
3620 /* This accounts for setting the checksum/compression. */
3621 dmu_tx_hold_bonus(tx, bigobj);
3623 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3625 umem_free(packbuf, packsize);
3626 umem_free(bigbuf, bigsize);
3630 dmu_object_set_checksum(os, bigobj,
3631 (enum zio_checksum)ztest_random_dsl_prop(ZFS_PROP_CHECKSUM), tx);
3633 dmu_object_set_compress(os, bigobj,
3634 (enum zio_compress)ztest_random_dsl_prop(ZFS_PROP_COMPRESSION), tx);
3637 * For each index from n to n + s, verify that the existing bufwad
3638 * in packobj matches the bufwads at the head and tail of the
3639 * corresponding chunk in bigobj. Then update all three bufwads
3640 * with the new values we want to write out.
3642 for (i = 0; i < s; i++) {
3644 pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
3646 bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
3648 bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
3650 ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
3651 ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
3653 if (pack->bw_txg > txg)
3654 fatal(0, "future leak: got %llx, open txg is %llx",
3657 if (pack->bw_data != 0 && pack->bw_index != n + i)
3658 fatal(0, "wrong index: got %llx, wanted %llx+%llx",
3659 pack->bw_index, n, i);
3661 if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
3662 fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
3664 if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
3665 fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
3668 bzero(pack, sizeof (bufwad_t));
3670 pack->bw_index = n + i;
3672 pack->bw_data = 1 + ztest_random(-2ULL);
3679 * We've verified all the old bufwads, and made new ones.
3680 * Now write them out.
3682 dmu_write(os, packobj, packoff, packsize, packbuf, tx);
3685 if (ztest_opts.zo_verbose >= 7) {
3686 (void) printf("freeing offset %llx size %llx"
3688 (u_longlong_t)bigoff,
3689 (u_longlong_t)bigsize,
3692 VERIFY(0 == dmu_free_range(os, bigobj, bigoff, bigsize, tx));
3694 if (ztest_opts.zo_verbose >= 7) {
3695 (void) printf("writing offset %llx size %llx"
3697 (u_longlong_t)bigoff,
3698 (u_longlong_t)bigsize,
3701 dmu_write(os, bigobj, bigoff, bigsize, bigbuf, tx);
3707 * Sanity check the stuff we just wrote.
3710 void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
3711 void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
3713 VERIFY(0 == dmu_read(os, packobj, packoff,
3714 packsize, packcheck, DMU_READ_PREFETCH));
3715 VERIFY(0 == dmu_read(os, bigobj, bigoff,
3716 bigsize, bigcheck, DMU_READ_PREFETCH));
3718 ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
3719 ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
3721 umem_free(packcheck, packsize);
3722 umem_free(bigcheck, bigsize);
3725 umem_free(packbuf, packsize);
3726 umem_free(bigbuf, bigsize);
3730 compare_and_update_pbbufs(uint64_t s, bufwad_t *packbuf, bufwad_t *bigbuf,
3731 uint64_t bigsize, uint64_t n, uint64_t chunksize, uint64_t txg)
3739 * For each index from n to n + s, verify that the existing bufwad
3740 * in packobj matches the bufwads at the head and tail of the
3741 * corresponding chunk in bigobj. Then update all three bufwads
3742 * with the new values we want to write out.
3744 for (i = 0; i < s; i++) {
3746 pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
3748 bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
3750 bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
3752 ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
3753 ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
3755 if (pack->bw_txg > txg)
3756 fatal(0, "future leak: got %llx, open txg is %llx",
3759 if (pack->bw_data != 0 && pack->bw_index != n + i)
3760 fatal(0, "wrong index: got %llx, wanted %llx+%llx",
3761 pack->bw_index, n, i);
3763 if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
3764 fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
3766 if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
3767 fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
3769 pack->bw_index = n + i;
3771 pack->bw_data = 1 + ztest_random(-2ULL);
3779 ztest_dmu_read_write_zcopy(ztest_ds_t *zd, uint64_t id)
3781 objset_t *os = zd->zd_os;
3787 bufwad_t *packbuf, *bigbuf;
3788 uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
3789 uint64_t blocksize = ztest_random_blocksize();
3790 uint64_t chunksize = blocksize;
3791 uint64_t regions = 997;
3792 uint64_t stride = 123456789ULL;
3794 dmu_buf_t *bonus_db;
3795 arc_buf_t **bigbuf_arcbufs;
3796 dmu_object_info_t doi;
3799 * This test uses two objects, packobj and bigobj, that are always
3800 * updated together (i.e. in the same tx) so that their contents are
3801 * in sync and can be compared. Their contents relate to each other
3802 * in a simple way: packobj is a dense array of 'bufwad' structures,
3803 * while bigobj is a sparse array of the same bufwads. Specifically,
3804 * for any index n, there are three bufwads that should be identical:
3806 * packobj, at offset n * sizeof (bufwad_t)
3807 * bigobj, at the head of the nth chunk
3808 * bigobj, at the tail of the nth chunk
3810 * The chunk size is set equal to bigobj block size so that
3811 * dmu_assign_arcbuf() can be tested for object updates.
3815 * Read the directory info. If it's the first time, set things up.
3817 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
3818 ztest_od_init(&od[1], id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize);
3820 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
3823 bigobj = od[0].od_object;
3824 packobj = od[1].od_object;
3825 blocksize = od[0].od_blocksize;
3826 chunksize = blocksize;
3827 ASSERT(chunksize == od[1].od_gen);
3829 VERIFY(dmu_object_info(os, bigobj, &doi) == 0);
3830 VERIFY(ISP2(doi.doi_data_block_size));
3831 VERIFY(chunksize == doi.doi_data_block_size);
3832 VERIFY(chunksize >= 2 * sizeof (bufwad_t));
3835 * Pick a random index and compute the offsets into packobj and bigobj.
3837 n = ztest_random(regions) * stride + ztest_random(width);
3838 s = 1 + ztest_random(width - 1);
3840 packoff = n * sizeof (bufwad_t);
3841 packsize = s * sizeof (bufwad_t);
3843 bigoff = n * chunksize;
3844 bigsize = s * chunksize;
3846 packbuf = umem_zalloc(packsize, UMEM_NOFAIL);
3847 bigbuf = umem_zalloc(bigsize, UMEM_NOFAIL);
3849 VERIFY3U(0, ==, dmu_bonus_hold(os, bigobj, FTAG, &bonus_db));
3851 bigbuf_arcbufs = umem_zalloc(2 * s * sizeof (arc_buf_t *), UMEM_NOFAIL);
3854 * Iteration 0 test zcopy for DB_UNCACHED dbufs.
3855 * Iteration 1 test zcopy to already referenced dbufs.
3856 * Iteration 2 test zcopy to dirty dbuf in the same txg.
3857 * Iteration 3 test zcopy to dbuf dirty in previous txg.
3858 * Iteration 4 test zcopy when dbuf is no longer dirty.
3859 * Iteration 5 test zcopy when it can't be done.
3860 * Iteration 6 one more zcopy write.
3862 for (i = 0; i < 7; i++) {
3867 * In iteration 5 (i == 5) use arcbufs
3868 * that don't match bigobj blksz to test
3869 * dmu_assign_arcbuf() when it can't directly
3870 * assign an arcbuf to a dbuf.
3872 for (j = 0; j < s; j++) {
3875 dmu_request_arcbuf(bonus_db, chunksize);
3877 bigbuf_arcbufs[2 * j] =
3878 dmu_request_arcbuf(bonus_db, chunksize / 2);
3879 bigbuf_arcbufs[2 * j + 1] =
3880 dmu_request_arcbuf(bonus_db, chunksize / 2);
3885 * Get a tx for the mods to both packobj and bigobj.
3887 tx = dmu_tx_create(os);
3889 dmu_tx_hold_write(tx, packobj, packoff, packsize);
3890 dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
3892 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3894 umem_free(packbuf, packsize);
3895 umem_free(bigbuf, bigsize);
3896 for (j = 0; j < s; j++) {
3898 dmu_return_arcbuf(bigbuf_arcbufs[j]);
3901 bigbuf_arcbufs[2 * j]);
3903 bigbuf_arcbufs[2 * j + 1]);
3906 umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
3907 dmu_buf_rele(bonus_db, FTAG);
3912 * 50% of the time don't read objects in the 1st iteration to
3913 * test dmu_assign_arcbuf() for the case when there're no
3914 * existing dbufs for the specified offsets.
3916 if (i != 0 || ztest_random(2) != 0) {
3917 error = dmu_read(os, packobj, packoff,
3918 packsize, packbuf, DMU_READ_PREFETCH);
3920 error = dmu_read(os, bigobj, bigoff, bigsize,
3921 bigbuf, DMU_READ_PREFETCH);
3924 compare_and_update_pbbufs(s, packbuf, bigbuf, bigsize,
3928 * We've verified all the old bufwads, and made new ones.
3929 * Now write them out.
3931 dmu_write(os, packobj, packoff, packsize, packbuf, tx);
3932 if (ztest_opts.zo_verbose >= 7) {
3933 (void) printf("writing offset %llx size %llx"
3935 (u_longlong_t)bigoff,
3936 (u_longlong_t)bigsize,
3939 for (off = bigoff, j = 0; j < s; j++, off += chunksize) {
3942 bcopy((caddr_t)bigbuf + (off - bigoff),
3943 bigbuf_arcbufs[j]->b_data, chunksize);
3945 bcopy((caddr_t)bigbuf + (off - bigoff),
3946 bigbuf_arcbufs[2 * j]->b_data,
3948 bcopy((caddr_t)bigbuf + (off - bigoff) +
3950 bigbuf_arcbufs[2 * j + 1]->b_data,
3955 VERIFY(dmu_buf_hold(os, bigobj, off,
3956 FTAG, &dbt, DMU_READ_NO_PREFETCH) == 0);
3959 dmu_assign_arcbuf(bonus_db, off,
3960 bigbuf_arcbufs[j], tx);
3962 dmu_assign_arcbuf(bonus_db, off,
3963 bigbuf_arcbufs[2 * j], tx);
3964 dmu_assign_arcbuf(bonus_db,
3965 off + chunksize / 2,
3966 bigbuf_arcbufs[2 * j + 1], tx);
3969 dmu_buf_rele(dbt, FTAG);
3975 * Sanity check the stuff we just wrote.
3978 void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
3979 void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
3981 VERIFY(0 == dmu_read(os, packobj, packoff,
3982 packsize, packcheck, DMU_READ_PREFETCH));
3983 VERIFY(0 == dmu_read(os, bigobj, bigoff,
3984 bigsize, bigcheck, DMU_READ_PREFETCH));
3986 ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
3987 ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
3989 umem_free(packcheck, packsize);
3990 umem_free(bigcheck, bigsize);
3993 txg_wait_open(dmu_objset_pool(os), 0);
3994 } else if (i == 3) {
3995 txg_wait_synced(dmu_objset_pool(os), 0);
3999 dmu_buf_rele(bonus_db, FTAG);
4000 umem_free(packbuf, packsize);
4001 umem_free(bigbuf, bigsize);
4002 umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
4007 ztest_dmu_write_parallel(ztest_ds_t *zd, uint64_t id)
4010 uint64_t offset = (1ULL << (ztest_random(20) + 43)) +
4011 (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
4014 * Have multiple threads write to large offsets in an object
4015 * to verify that parallel writes to an object -- even to the
4016 * same blocks within the object -- doesn't cause any trouble.
4018 ztest_od_init(&od[0], ID_PARALLEL, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
4020 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4023 while (ztest_random(10) != 0)
4024 ztest_io(zd, od[0].od_object, offset);
4028 ztest_dmu_prealloc(ztest_ds_t *zd, uint64_t id)
4031 uint64_t offset = (1ULL << (ztest_random(4) + SPA_MAXBLOCKSHIFT)) +
4032 (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
4033 uint64_t count = ztest_random(20) + 1;
4034 uint64_t blocksize = ztest_random_blocksize();
4037 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
4039 if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4042 if (ztest_truncate(zd, od[0].od_object, offset, count * blocksize) != 0)
4045 ztest_prealloc(zd, od[0].od_object, offset, count * blocksize);
4047 data = umem_zalloc(blocksize, UMEM_NOFAIL);
4049 while (ztest_random(count) != 0) {
4050 uint64_t randoff = offset + (ztest_random(count) * blocksize);
4051 if (ztest_write(zd, od[0].od_object, randoff, blocksize,
4054 while (ztest_random(4) != 0)
4055 ztest_io(zd, od[0].od_object, randoff);
4058 umem_free(data, blocksize);
4062 * Verify that zap_{create,destroy,add,remove,update} work as expected.
4064 #define ZTEST_ZAP_MIN_INTS 1
4065 #define ZTEST_ZAP_MAX_INTS 4
4066 #define ZTEST_ZAP_MAX_PROPS 1000
4069 ztest_zap(ztest_ds_t *zd, uint64_t id)
4071 objset_t *os = zd->zd_os;
4074 uint64_t txg, last_txg;
4075 uint64_t value[ZTEST_ZAP_MAX_INTS];
4076 uint64_t zl_ints, zl_intsize, prop;
4079 char propname[100], txgname[100];
4081 char *hc[2] = { "s.acl.h", ".s.open.h.hyLZlg" };
4083 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0);
4085 if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4088 object = od[0].od_object;
4091 * Generate a known hash collision, and verify that
4092 * we can lookup and remove both entries.
4094 tx = dmu_tx_create(os);
4095 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4096 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4099 for (i = 0; i < 2; i++) {
4101 VERIFY3U(0, ==, zap_add(os, object, hc[i], sizeof (uint64_t),
4104 for (i = 0; i < 2; i++) {
4105 VERIFY3U(EEXIST, ==, zap_add(os, object, hc[i],
4106 sizeof (uint64_t), 1, &value[i], tx));
4108 zap_length(os, object, hc[i], &zl_intsize, &zl_ints));
4109 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4110 ASSERT3U(zl_ints, ==, 1);
4112 for (i = 0; i < 2; i++) {
4113 VERIFY3U(0, ==, zap_remove(os, object, hc[i], tx));
4118 * Generate a buch of random entries.
4120 ints = MAX(ZTEST_ZAP_MIN_INTS, object % ZTEST_ZAP_MAX_INTS);
4122 prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
4123 (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
4124 (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
4125 bzero(value, sizeof (value));
4129 * If these zap entries already exist, validate their contents.
4131 error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
4133 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4134 ASSERT3U(zl_ints, ==, 1);
4136 VERIFY(zap_lookup(os, object, txgname, zl_intsize,
4137 zl_ints, &last_txg) == 0);
4139 VERIFY(zap_length(os, object, propname, &zl_intsize,
4142 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4143 ASSERT3U(zl_ints, ==, ints);
4145 VERIFY(zap_lookup(os, object, propname, zl_intsize,
4146 zl_ints, value) == 0);
4148 for (i = 0; i < ints; i++) {
4149 ASSERT3U(value[i], ==, last_txg + object + i);
4152 ASSERT3U(error, ==, ENOENT);
4156 * Atomically update two entries in our zap object.
4157 * The first is named txg_%llu, and contains the txg
4158 * in which the property was last updated. The second
4159 * is named prop_%llu, and the nth element of its value
4160 * should be txg + object + n.
4162 tx = dmu_tx_create(os);
4163 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4164 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4169 fatal(0, "zap future leak: old %llu new %llu", last_txg, txg);
4171 for (i = 0; i < ints; i++)
4172 value[i] = txg + object + i;
4174 VERIFY3U(0, ==, zap_update(os, object, txgname, sizeof (uint64_t),
4176 VERIFY3U(0, ==, zap_update(os, object, propname, sizeof (uint64_t),
4182 * Remove a random pair of entries.
4184 prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
4185 (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
4186 (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
4188 error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
4190 if (error == ENOENT)
4195 tx = dmu_tx_create(os);
4196 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4197 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4200 VERIFY3U(0, ==, zap_remove(os, object, txgname, tx));
4201 VERIFY3U(0, ==, zap_remove(os, object, propname, tx));
4206 * Testcase to test the upgrading of a microzap to fatzap.
4209 ztest_fzap(ztest_ds_t *zd, uint64_t id)
4211 objset_t *os = zd->zd_os;
4213 uint64_t object, txg;
4215 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0);
4217 if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4220 object = od[0].od_object;
4223 * Add entries to this ZAP and make sure it spills over
4224 * and gets upgraded to a fatzap. Also, since we are adding
4225 * 2050 entries we should see ptrtbl growth and leaf-block split.
4227 for (int i = 0; i < 2050; i++) {
4228 char name[MAXNAMELEN];
4233 (void) snprintf(name, sizeof (name), "fzap-%llu-%llu",
4236 tx = dmu_tx_create(os);
4237 dmu_tx_hold_zap(tx, object, B_TRUE, name);
4238 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4241 error = zap_add(os, object, name, sizeof (uint64_t), 1,
4243 ASSERT(error == 0 || error == EEXIST);
4250 ztest_zap_parallel(ztest_ds_t *zd, uint64_t id)
4252 objset_t *os = zd->zd_os;
4254 uint64_t txg, object, count, wsize, wc, zl_wsize, zl_wc;
4256 int i, namelen, error;
4257 int micro = ztest_random(2);
4258 char name[20], string_value[20];
4261 ztest_od_init(&od[0], ID_PARALLEL, FTAG, micro, DMU_OT_ZAP_OTHER, 0, 0);
4263 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4266 object = od[0].od_object;
4269 * Generate a random name of the form 'xxx.....' where each
4270 * x is a random printable character and the dots are dots.
4271 * There are 94 such characters, and the name length goes from
4272 * 6 to 20, so there are 94^3 * 15 = 12,458,760 possible names.
4274 namelen = ztest_random(sizeof (name) - 5) + 5 + 1;
4276 for (i = 0; i < 3; i++)
4277 name[i] = '!' + ztest_random('~' - '!' + 1);
4278 for (; i < namelen - 1; i++)
4282 if ((namelen & 1) || micro) {
4283 wsize = sizeof (txg);
4289 data = string_value;
4293 VERIFY0(zap_count(os, object, &count));
4294 ASSERT(count != -1ULL);
4297 * Select an operation: length, lookup, add, update, remove.
4299 i = ztest_random(5);
4302 tx = dmu_tx_create(os);
4303 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4304 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4307 bcopy(name, string_value, namelen);
4311 bzero(string_value, namelen);
4317 error = zap_length(os, object, name, &zl_wsize, &zl_wc);
4319 ASSERT3U(wsize, ==, zl_wsize);
4320 ASSERT3U(wc, ==, zl_wc);
4322 ASSERT3U(error, ==, ENOENT);
4327 error = zap_lookup(os, object, name, wsize, wc, data);
4329 if (data == string_value &&
4330 bcmp(name, data, namelen) != 0)
4331 fatal(0, "name '%s' != val '%s' len %d",
4332 name, data, namelen);
4334 ASSERT3U(error, ==, ENOENT);
4339 error = zap_add(os, object, name, wsize, wc, data, tx);
4340 ASSERT(error == 0 || error == EEXIST);
4344 VERIFY(zap_update(os, object, name, wsize, wc, data, tx) == 0);
4348 error = zap_remove(os, object, name, tx);
4349 ASSERT(error == 0 || error == ENOENT);
4358 * Commit callback data.
4360 typedef struct ztest_cb_data {
4361 list_node_t zcd_node;
4363 int zcd_expected_err;
4364 boolean_t zcd_added;
4365 boolean_t zcd_called;
4369 /* This is the actual commit callback function */
4371 ztest_commit_callback(void *arg, int error)
4373 ztest_cb_data_t *data = arg;
4374 uint64_t synced_txg;
4376 VERIFY(data != NULL);
4377 VERIFY3S(data->zcd_expected_err, ==, error);
4378 VERIFY(!data->zcd_called);
4380 synced_txg = spa_last_synced_txg(data->zcd_spa);
4381 if (data->zcd_txg > synced_txg)
4382 fatal(0, "commit callback of txg %" PRIu64 " called prematurely"
4383 ", last synced txg = %" PRIu64 "\n", data->zcd_txg,
4386 data->zcd_called = B_TRUE;
4388 if (error == ECANCELED) {
4389 ASSERT0(data->zcd_txg);
4390 ASSERT(!data->zcd_added);
4393 * The private callback data should be destroyed here, but
4394 * since we are going to check the zcd_called field after
4395 * dmu_tx_abort(), we will destroy it there.
4400 /* Was this callback added to the global callback list? */
4401 if (!data->zcd_added)
4404 ASSERT3U(data->zcd_txg, !=, 0);
4406 /* Remove our callback from the list */
4407 (void) mutex_lock(&zcl.zcl_callbacks_lock);
4408 list_remove(&zcl.zcl_callbacks, data);
4409 (void) mutex_unlock(&zcl.zcl_callbacks_lock);
4412 umem_free(data, sizeof (ztest_cb_data_t));
4415 /* Allocate and initialize callback data structure */
4416 static ztest_cb_data_t *
4417 ztest_create_cb_data(objset_t *os, uint64_t txg)
4419 ztest_cb_data_t *cb_data;
4421 cb_data = umem_zalloc(sizeof (ztest_cb_data_t), UMEM_NOFAIL);
4423 cb_data->zcd_txg = txg;
4424 cb_data->zcd_spa = dmu_objset_spa(os);
4430 * If a number of txgs equal to this threshold have been created after a commit
4431 * callback has been registered but not called, then we assume there is an
4432 * implementation bug.
4434 #define ZTEST_COMMIT_CALLBACK_THRESH (TXG_CONCURRENT_STATES + 2)
4437 * Commit callback test.
4440 ztest_dmu_commit_callbacks(ztest_ds_t *zd, uint64_t id)
4442 objset_t *os = zd->zd_os;
4445 ztest_cb_data_t *cb_data[3], *tmp_cb;
4446 uint64_t old_txg, txg;
4449 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
4451 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4454 tx = dmu_tx_create(os);
4456 cb_data[0] = ztest_create_cb_data(os, 0);
4457 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[0]);
4459 dmu_tx_hold_write(tx, od[0].od_object, 0, sizeof (uint64_t));
4461 /* Every once in a while, abort the transaction on purpose */
4462 if (ztest_random(100) == 0)
4466 error = dmu_tx_assign(tx, TXG_NOWAIT);
4468 txg = error ? 0 : dmu_tx_get_txg(tx);
4470 cb_data[0]->zcd_txg = txg;
4471 cb_data[1] = ztest_create_cb_data(os, txg);
4472 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[1]);
4476 * It's not a strict requirement to call the registered
4477 * callbacks from inside dmu_tx_abort(), but that's what
4478 * it's supposed to happen in the current implementation
4479 * so we will check for that.
4481 for (i = 0; i < 2; i++) {
4482 cb_data[i]->zcd_expected_err = ECANCELED;
4483 VERIFY(!cb_data[i]->zcd_called);
4488 for (i = 0; i < 2; i++) {
4489 VERIFY(cb_data[i]->zcd_called);
4490 umem_free(cb_data[i], sizeof (ztest_cb_data_t));
4496 cb_data[2] = ztest_create_cb_data(os, txg);
4497 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[2]);
4500 * Read existing data to make sure there isn't a future leak.
4502 VERIFY(0 == dmu_read(os, od[0].od_object, 0, sizeof (uint64_t),
4503 &old_txg, DMU_READ_PREFETCH));
4506 fatal(0, "future leak: got %" PRIu64 ", open txg is %" PRIu64,
4509 dmu_write(os, od[0].od_object, 0, sizeof (uint64_t), &txg, tx);
4511 (void) mutex_lock(&zcl.zcl_callbacks_lock);
4514 * Since commit callbacks don't have any ordering requirement and since
4515 * it is theoretically possible for a commit callback to be called
4516 * after an arbitrary amount of time has elapsed since its txg has been
4517 * synced, it is difficult to reliably determine whether a commit
4518 * callback hasn't been called due to high load or due to a flawed
4521 * In practice, we will assume that if after a certain number of txgs a
4522 * commit callback hasn't been called, then most likely there's an
4523 * implementation bug..
4525 tmp_cb = list_head(&zcl.zcl_callbacks);
4526 if (tmp_cb != NULL &&
4527 (txg - ZTEST_COMMIT_CALLBACK_THRESH) > tmp_cb->zcd_txg) {
4528 fatal(0, "Commit callback threshold exceeded, oldest txg: %"
4529 PRIu64 ", open txg: %" PRIu64 "\n", tmp_cb->zcd_txg, txg);
4533 * Let's find the place to insert our callbacks.
4535 * Even though the list is ordered by txg, it is possible for the
4536 * insertion point to not be the end because our txg may already be
4537 * quiescing at this point and other callbacks in the open txg
4538 * (from other objsets) may have sneaked in.
4540 tmp_cb = list_tail(&zcl.zcl_callbacks);
4541 while (tmp_cb != NULL && tmp_cb->zcd_txg > txg)
4542 tmp_cb = list_prev(&zcl.zcl_callbacks, tmp_cb);
4544 /* Add the 3 callbacks to the list */
4545 for (i = 0; i < 3; i++) {
4547 list_insert_head(&zcl.zcl_callbacks, cb_data[i]);
4549 list_insert_after(&zcl.zcl_callbacks, tmp_cb,
4552 cb_data[i]->zcd_added = B_TRUE;
4553 VERIFY(!cb_data[i]->zcd_called);
4555 tmp_cb = cb_data[i];
4558 (void) mutex_unlock(&zcl.zcl_callbacks_lock);
4565 ztest_dsl_prop_get_set(ztest_ds_t *zd, uint64_t id)
4567 zfs_prop_t proplist[] = {
4569 ZFS_PROP_COMPRESSION,
4574 (void) rw_rdlock(&ztest_name_lock);
4576 for (int p = 0; p < sizeof (proplist) / sizeof (proplist[0]); p++)
4577 (void) ztest_dsl_prop_set_uint64(zd->zd_name, proplist[p],
4578 ztest_random_dsl_prop(proplist[p]), (int)ztest_random(2));
4580 (void) rw_unlock(&ztest_name_lock);
4585 ztest_spa_prop_get_set(ztest_ds_t *zd, uint64_t id)
4587 nvlist_t *props = NULL;
4589 (void) rw_rdlock(&ztest_name_lock);
4591 (void) ztest_spa_prop_set_uint64(ZPOOL_PROP_DEDUPDITTO,
4592 ZIO_DEDUPDITTO_MIN + ztest_random(ZIO_DEDUPDITTO_MIN));
4594 VERIFY0(spa_prop_get(ztest_spa, &props));
4596 if (ztest_opts.zo_verbose >= 6)
4597 dump_nvlist(props, 4);
4601 (void) rw_unlock(&ztest_name_lock);
4605 user_release_one(const char *snapname, const char *holdname)
4607 nvlist_t *snaps, *holds;
4610 snaps = fnvlist_alloc();
4611 holds = fnvlist_alloc();
4612 fnvlist_add_boolean(holds, holdname);
4613 fnvlist_add_nvlist(snaps, snapname, holds);
4614 fnvlist_free(holds);
4615 error = dsl_dataset_user_release(snaps, NULL);
4616 fnvlist_free(snaps);
4621 * Test snapshot hold/release and deferred destroy.
4624 ztest_dmu_snapshot_hold(ztest_ds_t *zd, uint64_t id)
4627 objset_t *os = zd->zd_os;
4631 char clonename[100];
4633 char osname[MAXNAMELEN];
4636 (void) rw_rdlock(&ztest_name_lock);
4638 dmu_objset_name(os, osname);
4640 (void) snprintf(snapname, sizeof (snapname), "sh1_%llu", id);
4641 (void) snprintf(fullname, sizeof (fullname), "%s@%s", osname, snapname);
4642 (void) snprintf(clonename, sizeof (clonename),
4643 "%s/ch1_%llu", osname, id);
4644 (void) snprintf(tag, sizeof (tag), "tag_%llu", id);
4647 * Clean up from any previous run.
4649 error = dsl_destroy_head(clonename);
4650 if (error != ENOENT)
4652 error = user_release_one(fullname, tag);
4653 if (error != ESRCH && error != ENOENT)
4655 error = dsl_destroy_snapshot(fullname, B_FALSE);
4656 if (error != ENOENT)
4660 * Create snapshot, clone it, mark snap for deferred destroy,
4661 * destroy clone, verify snap was also destroyed.
4663 error = dmu_objset_snapshot_one(osname, snapname);
4665 if (error == ENOSPC) {
4666 ztest_record_enospc("dmu_objset_snapshot");
4669 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
4672 error = dmu_objset_clone(clonename, fullname);
4674 if (error == ENOSPC) {
4675 ztest_record_enospc("dmu_objset_clone");
4678 fatal(0, "dmu_objset_clone(%s) = %d", clonename, error);
4681 error = dsl_destroy_snapshot(fullname, B_TRUE);
4683 fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
4687 error = dsl_destroy_head(clonename);
4689 fatal(0, "dsl_destroy_head(%s) = %d", clonename, error);
4691 error = dmu_objset_hold(fullname, FTAG, &origin);
4692 if (error != ENOENT)
4693 fatal(0, "dmu_objset_hold(%s) = %d", fullname, error);
4696 * Create snapshot, add temporary hold, verify that we can't
4697 * destroy a held snapshot, mark for deferred destroy,
4698 * release hold, verify snapshot was destroyed.
4700 error = dmu_objset_snapshot_one(osname, snapname);
4702 if (error == ENOSPC) {
4703 ztest_record_enospc("dmu_objset_snapshot");
4706 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
4709 holds = fnvlist_alloc();
4710 fnvlist_add_string(holds, fullname, tag);
4711 error = dsl_dataset_user_hold(holds, 0, NULL);
4712 fnvlist_free(holds);
4715 fatal(0, "dsl_dataset_user_hold(%s)", fullname, tag);
4717 error = dsl_destroy_snapshot(fullname, B_FALSE);
4718 if (error != EBUSY) {
4719 fatal(0, "dsl_destroy_snapshot(%s, B_FALSE) = %d",
4723 error = dsl_destroy_snapshot(fullname, B_TRUE);
4725 fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
4729 error = user_release_one(fullname, tag);
4731 fatal(0, "user_release_one(%s, %s) = %d", fullname, tag, error);
4733 VERIFY3U(dmu_objset_hold(fullname, FTAG, &origin), ==, ENOENT);
4736 (void) rw_unlock(&ztest_name_lock);
4740 * Inject random faults into the on-disk data.
4744 ztest_fault_inject(ztest_ds_t *zd, uint64_t id)
4746 ztest_shared_t *zs = ztest_shared;
4747 spa_t *spa = ztest_spa;
4751 uint64_t bad = 0x1990c0ffeedecadeULL;
4753 char path0[MAXPATHLEN];
4754 char pathrand[MAXPATHLEN];
4756 int bshift = SPA_MAXBLOCKSHIFT + 2; /* don't scrog all labels */
4762 boolean_t islog = B_FALSE;
4764 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
4765 maxfaults = MAXFAULTS();
4766 leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz;
4767 mirror_save = zs->zs_mirrors;
4768 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4770 ASSERT(leaves >= 1);
4773 * Grab the name lock as reader. There are some operations
4774 * which don't like to have their vdevs changed while
4775 * they are in progress (i.e. spa_change_guid). Those
4776 * operations will have grabbed the name lock as writer.
4778 (void) rw_rdlock(&ztest_name_lock);
4781 * We need SCL_STATE here because we're going to look at vd0->vdev_tsd.
4783 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
4785 if (ztest_random(2) == 0) {
4787 * Inject errors on a normal data device or slog device.
4789 top = ztest_random_vdev_top(spa, B_TRUE);
4790 leaf = ztest_random(leaves) + zs->zs_splits;
4793 * Generate paths to the first leaf in this top-level vdev,
4794 * and to the random leaf we selected. We'll induce transient
4795 * write failures and random online/offline activity on leaf 0,
4796 * and we'll write random garbage to the randomly chosen leaf.
4798 (void) snprintf(path0, sizeof (path0), ztest_dev_template,
4799 ztest_opts.zo_dir, ztest_opts.zo_pool,
4800 top * leaves + zs->zs_splits);
4801 (void) snprintf(pathrand, sizeof (pathrand), ztest_dev_template,
4802 ztest_opts.zo_dir, ztest_opts.zo_pool,
4803 top * leaves + leaf);
4805 vd0 = vdev_lookup_by_path(spa->spa_root_vdev, path0);
4806 if (vd0 != NULL && vd0->vdev_top->vdev_islog)
4810 * If the top-level vdev needs to be resilvered
4811 * then we only allow faults on the device that is
4814 if (vd0 != NULL && maxfaults != 1 &&
4815 (!vdev_resilver_needed(vd0->vdev_top, NULL, NULL) ||
4816 vd0->vdev_resilver_txg != 0)) {
4818 * Make vd0 explicitly claim to be unreadable,
4819 * or unwriteable, or reach behind its back
4820 * and close the underlying fd. We can do this if
4821 * maxfaults == 0 because we'll fail and reexecute,
4822 * and we can do it if maxfaults >= 2 because we'll
4823 * have enough redundancy. If maxfaults == 1, the
4824 * combination of this with injection of random data
4825 * corruption below exceeds the pool's fault tolerance.
4827 vdev_file_t *vf = vd0->vdev_tsd;
4829 if (vf != NULL && ztest_random(3) == 0) {
4830 (void) close(vf->vf_vnode->v_fd);
4831 vf->vf_vnode->v_fd = -1;
4832 } else if (ztest_random(2) == 0) {
4833 vd0->vdev_cant_read = B_TRUE;
4835 vd0->vdev_cant_write = B_TRUE;
4837 guid0 = vd0->vdev_guid;
4841 * Inject errors on an l2cache device.
4843 spa_aux_vdev_t *sav = &spa->spa_l2cache;
4845 if (sav->sav_count == 0) {
4846 spa_config_exit(spa, SCL_STATE, FTAG);
4847 (void) rw_unlock(&ztest_name_lock);
4850 vd0 = sav->sav_vdevs[ztest_random(sav->sav_count)];
4851 guid0 = vd0->vdev_guid;
4852 (void) strcpy(path0, vd0->vdev_path);
4853 (void) strcpy(pathrand, vd0->vdev_path);
4857 maxfaults = INT_MAX; /* no limit on cache devices */
4860 spa_config_exit(spa, SCL_STATE, FTAG);
4861 (void) rw_unlock(&ztest_name_lock);
4864 * If we can tolerate two or more faults, or we're dealing
4865 * with a slog, randomly online/offline vd0.
4867 if ((maxfaults >= 2 || islog) && guid0 != 0) {
4868 if (ztest_random(10) < 6) {
4869 int flags = (ztest_random(2) == 0 ?
4870 ZFS_OFFLINE_TEMPORARY : 0);
4873 * We have to grab the zs_name_lock as writer to
4874 * prevent a race between offlining a slog and
4875 * destroying a dataset. Offlining the slog will
4876 * grab a reference on the dataset which may cause
4877 * dmu_objset_destroy() to fail with EBUSY thus
4878 * leaving the dataset in an inconsistent state.
4881 (void) rw_wrlock(&ztest_name_lock);
4883 VERIFY(vdev_offline(spa, guid0, flags) != EBUSY);
4886 (void) rw_unlock(&ztest_name_lock);
4889 * Ideally we would like to be able to randomly
4890 * call vdev_[on|off]line without holding locks
4891 * to force unpredictable failures but the side
4892 * effects of vdev_[on|off]line prevent us from
4893 * doing so. We grab the ztest_vdev_lock here to
4894 * prevent a race between injection testing and
4897 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
4898 (void) vdev_online(spa, guid0, 0, NULL);
4899 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4907 * We have at least single-fault tolerance, so inject data corruption.
4909 fd = open(pathrand, O_RDWR);
4911 if (fd == -1) /* we hit a gap in the device namespace */
4914 fsize = lseek(fd, 0, SEEK_END);
4916 while (--iters != 0) {
4917 offset = ztest_random(fsize / (leaves << bshift)) *
4918 (leaves << bshift) + (leaf << bshift) +
4919 (ztest_random(1ULL << (bshift - 1)) & -8ULL);
4921 if (offset >= fsize)
4924 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
4925 if (mirror_save != zs->zs_mirrors) {
4926 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4931 if (pwrite(fd, &bad, sizeof (bad), offset) != sizeof (bad))
4932 fatal(1, "can't inject bad word at 0x%llx in %s",
4935 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4937 if (ztest_opts.zo_verbose >= 7)
4938 (void) printf("injected bad word into %s,"
4939 " offset 0x%llx\n", pathrand, (u_longlong_t)offset);
4946 * Verify that DDT repair works as expected.
4949 ztest_ddt_repair(ztest_ds_t *zd, uint64_t id)
4951 ztest_shared_t *zs = ztest_shared;
4952 spa_t *spa = ztest_spa;
4953 objset_t *os = zd->zd_os;
4955 uint64_t object, blocksize, txg, pattern, psize;
4956 enum zio_checksum checksum = spa_dedup_checksum(spa);
4961 int copies = 2 * ZIO_DEDUPDITTO_MIN;
4963 blocksize = ztest_random_blocksize();
4964 blocksize = MIN(blocksize, 2048); /* because we write so many */
4966 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
4968 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4972 * Take the name lock as writer to prevent anyone else from changing
4973 * the pool and dataset properies we need to maintain during this test.
4975 (void) rw_wrlock(&ztest_name_lock);
4977 if (ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_DEDUP, checksum,
4979 ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_COPIES, 1,
4981 (void) rw_unlock(&ztest_name_lock);
4985 object = od[0].od_object;
4986 blocksize = od[0].od_blocksize;
4987 pattern = zs->zs_guid ^ dmu_objset_fsid_guid(os);
4989 ASSERT(object != 0);
4991 tx = dmu_tx_create(os);
4992 dmu_tx_hold_write(tx, object, 0, copies * blocksize);
4993 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
4995 (void) rw_unlock(&ztest_name_lock);
5000 * Write all the copies of our block.
5002 for (int i = 0; i < copies; i++) {
5003 uint64_t offset = i * blocksize;
5004 int error = dmu_buf_hold(os, object, offset, FTAG, &db,
5005 DMU_READ_NO_PREFETCH);
5007 fatal(B_FALSE, "dmu_buf_hold(%p, %llu, %llu) = %u",
5008 os, (long long)object, (long long) offset, error);
5010 ASSERT(db->db_offset == offset);
5011 ASSERT(db->db_size == blocksize);
5012 ASSERT(ztest_pattern_match(db->db_data, db->db_size, pattern) ||
5013 ztest_pattern_match(db->db_data, db->db_size, 0ULL));
5014 dmu_buf_will_fill(db, tx);
5015 ztest_pattern_set(db->db_data, db->db_size, pattern);
5016 dmu_buf_rele(db, FTAG);
5020 txg_wait_synced(spa_get_dsl(spa), txg);
5023 * Find out what block we got.
5025 VERIFY0(dmu_buf_hold(os, object, 0, FTAG, &db,
5026 DMU_READ_NO_PREFETCH));
5027 blk = *((dmu_buf_impl_t *)db)->db_blkptr;
5028 dmu_buf_rele(db, FTAG);
5031 * Damage the block. Dedup-ditto will save us when we read it later.
5033 psize = BP_GET_PSIZE(&blk);
5034 buf = zio_buf_alloc(psize);
5035 ztest_pattern_set(buf, psize, ~pattern);
5037 (void) zio_wait(zio_rewrite(NULL, spa, 0, &blk,
5038 buf, psize, NULL, NULL, ZIO_PRIORITY_SYNC_WRITE,
5039 ZIO_FLAG_CANFAIL | ZIO_FLAG_INDUCE_DAMAGE, NULL));
5041 zio_buf_free(buf, psize);
5043 (void) rw_unlock(&ztest_name_lock);
5051 ztest_scrub(ztest_ds_t *zd, uint64_t id)
5053 spa_t *spa = ztest_spa;
5055 (void) spa_scan(spa, POOL_SCAN_SCRUB);
5056 (void) poll(NULL, 0, 100); /* wait a moment, then force a restart */
5057 (void) spa_scan(spa, POOL_SCAN_SCRUB);
5061 * Change the guid for the pool.
5065 ztest_reguid(ztest_ds_t *zd, uint64_t id)
5067 spa_t *spa = ztest_spa;
5068 uint64_t orig, load;
5071 orig = spa_guid(spa);
5072 load = spa_load_guid(spa);
5074 (void) rw_wrlock(&ztest_name_lock);
5075 error = spa_change_guid(spa);
5076 (void) rw_unlock(&ztest_name_lock);
5081 if (ztest_opts.zo_verbose >= 4) {
5082 (void) printf("Changed guid old %llu -> %llu\n",
5083 (u_longlong_t)orig, (u_longlong_t)spa_guid(spa));
5086 VERIFY3U(orig, !=, spa_guid(spa));
5087 VERIFY3U(load, ==, spa_load_guid(spa));
5091 * Rename the pool to a different name and then rename it back.
5095 ztest_spa_rename(ztest_ds_t *zd, uint64_t id)
5097 char *oldname, *newname;
5100 (void) rw_wrlock(&ztest_name_lock);
5102 oldname = ztest_opts.zo_pool;
5103 newname = umem_alloc(strlen(oldname) + 5, UMEM_NOFAIL);
5104 (void) strcpy(newname, oldname);
5105 (void) strcat(newname, "_tmp");
5110 VERIFY3U(0, ==, spa_rename(oldname, newname));
5113 * Try to open it under the old name, which shouldn't exist
5115 VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
5118 * Open it under the new name and make sure it's still the same spa_t.
5120 VERIFY3U(0, ==, spa_open(newname, &spa, FTAG));
5122 ASSERT(spa == ztest_spa);
5123 spa_close(spa, FTAG);
5126 * Rename it back to the original
5128 VERIFY3U(0, ==, spa_rename(newname, oldname));
5131 * Make sure it can still be opened
5133 VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG));
5135 ASSERT(spa == ztest_spa);
5136 spa_close(spa, FTAG);
5138 umem_free(newname, strlen(newname) + 1);
5140 (void) rw_unlock(&ztest_name_lock);
5144 * Verify pool integrity by running zdb.
5147 ztest_run_zdb(char *pool)
5150 char zdb[MAXPATHLEN + MAXNAMELEN + 20];
5158 strlcpy(zdb, "/usr/bin/ztest", sizeof(zdb));
5160 /* zdb lives in /usr/sbin, while ztest lives in /usr/bin */
5161 bin = strstr(zdb, "/usr/bin/");
5162 ztest = strstr(bin, "/ztest");
5164 isalen = ztest - isa;
5168 "/usr/sbin%.*s/zdb -bcc%s%s -U %s %s",
5171 ztest_opts.zo_verbose >= 3 ? "s" : "",
5172 ztest_opts.zo_verbose >= 4 ? "v" : "",
5177 if (ztest_opts.zo_verbose >= 5)
5178 (void) printf("Executing %s\n", strstr(zdb, "zdb "));
5180 fp = popen(zdb, "r");
5183 while (fgets(zbuf, sizeof (zbuf), fp) != NULL)
5184 if (ztest_opts.zo_verbose >= 3)
5185 (void) printf("%s", zbuf);
5187 status = pclose(fp);
5192 ztest_dump_core = 0;
5193 if (WIFEXITED(status))
5194 fatal(0, "'%s' exit code %d", zdb, WEXITSTATUS(status));
5196 fatal(0, "'%s' died with signal %d", zdb, WTERMSIG(status));
5200 ztest_walk_pool_directory(char *header)
5204 if (ztest_opts.zo_verbose >= 6)
5205 (void) printf("%s\n", header);
5207 mutex_enter(&spa_namespace_lock);
5208 while ((spa = spa_next(spa)) != NULL)
5209 if (ztest_opts.zo_verbose >= 6)
5210 (void) printf("\t%s\n", spa_name(spa));
5211 mutex_exit(&spa_namespace_lock);
5215 ztest_spa_import_export(char *oldname, char *newname)
5217 nvlist_t *config, *newconfig;
5222 if (ztest_opts.zo_verbose >= 4) {
5223 (void) printf("import/export: old = %s, new = %s\n",
5228 * Clean up from previous runs.
5230 (void) spa_destroy(newname);
5233 * Get the pool's configuration and guid.
5235 VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG));
5238 * Kick off a scrub to tickle scrub/export races.
5240 if (ztest_random(2) == 0)
5241 (void) spa_scan(spa, POOL_SCAN_SCRUB);
5243 pool_guid = spa_guid(spa);
5244 spa_close(spa, FTAG);
5246 ztest_walk_pool_directory("pools before export");
5251 VERIFY3U(0, ==, spa_export(oldname, &config, B_FALSE, B_FALSE));
5253 ztest_walk_pool_directory("pools after export");
5258 newconfig = spa_tryimport(config);
5259 ASSERT(newconfig != NULL);
5260 nvlist_free(newconfig);
5263 * Import it under the new name.
5265 error = spa_import(newname, config, NULL, 0);
5267 dump_nvlist(config, 0);
5268 fatal(B_FALSE, "couldn't import pool %s as %s: error %u",
5269 oldname, newname, error);
5272 ztest_walk_pool_directory("pools after import");
5275 * Try to import it again -- should fail with EEXIST.
5277 VERIFY3U(EEXIST, ==, spa_import(newname, config, NULL, 0));
5280 * Try to import it under a different name -- should fail with EEXIST.
5282 VERIFY3U(EEXIST, ==, spa_import(oldname, config, NULL, 0));
5285 * Verify that the pool is no longer visible under the old name.
5287 VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
5290 * Verify that we can open and close the pool using the new name.
5292 VERIFY3U(0, ==, spa_open(newname, &spa, FTAG));
5293 ASSERT(pool_guid == spa_guid(spa));
5294 spa_close(spa, FTAG);
5296 nvlist_free(config);
5300 ztest_resume(spa_t *spa)
5302 if (spa_suspended(spa) && ztest_opts.zo_verbose >= 6)
5303 (void) printf("resuming from suspended state\n");
5304 spa_vdev_state_enter(spa, SCL_NONE);
5305 vdev_clear(spa, NULL);
5306 (void) spa_vdev_state_exit(spa, NULL, 0);
5307 (void) zio_resume(spa);
5311 ztest_resume_thread(void *arg)
5315 while (!ztest_exiting) {
5316 if (spa_suspended(spa))
5318 (void) poll(NULL, 0, 100);
5324 ztest_deadman_thread(void *arg)
5326 ztest_shared_t *zs = arg;
5327 spa_t *spa = ztest_spa;
5328 hrtime_t delta, total = 0;
5331 delta = zs->zs_thread_stop - zs->zs_thread_start +
5332 MSEC2NSEC(zfs_deadman_synctime_ms);
5334 (void) poll(NULL, 0, (int)NSEC2MSEC(delta));
5337 * If the pool is suspended then fail immediately. Otherwise,
5338 * check to see if the pool is making any progress. If
5339 * vdev_deadman() discovers that there hasn't been any recent
5340 * I/Os then it will end up aborting the tests.
5342 if (spa_suspended(spa)) {
5343 fatal(0, "aborting test after %llu seconds because "
5344 "pool has transitioned to a suspended state.",
5345 zfs_deadman_synctime_ms / 1000);
5348 vdev_deadman(spa->spa_root_vdev);
5350 total += zfs_deadman_synctime_ms/1000;
5351 (void) printf("ztest has been running for %lld seconds\n",
5357 ztest_execute(int test, ztest_info_t *zi, uint64_t id)
5359 ztest_ds_t *zd = &ztest_ds[id % ztest_opts.zo_datasets];
5360 ztest_shared_callstate_t *zc = ZTEST_GET_SHARED_CALLSTATE(test);
5361 hrtime_t functime = gethrtime();
5363 for (int i = 0; i < zi->zi_iters; i++)
5364 zi->zi_func(zd, id);
5366 functime = gethrtime() - functime;
5368 atomic_add_64(&zc->zc_count, 1);
5369 atomic_add_64(&zc->zc_time, functime);
5371 if (ztest_opts.zo_verbose >= 4) {
5373 (void) dladdr((void *)zi->zi_func, &dli);
5374 (void) printf("%6.2f sec in %s\n",
5375 (double)functime / NANOSEC, dli.dli_sname);
5380 ztest_thread(void *arg)
5383 uint64_t id = (uintptr_t)arg;
5384 ztest_shared_t *zs = ztest_shared;
5388 ztest_shared_callstate_t *zc;
5390 while ((now = gethrtime()) < zs->zs_thread_stop) {
5392 * See if it's time to force a crash.
5394 if (now > zs->zs_thread_kill)
5398 * If we're getting ENOSPC with some regularity, stop.
5400 if (zs->zs_enospc_count > 10)
5404 * Pick a random function to execute.
5406 rand = ztest_random(ZTEST_FUNCS);
5407 zi = &ztest_info[rand];
5408 zc = ZTEST_GET_SHARED_CALLSTATE(rand);
5409 call_next = zc->zc_next;
5411 if (now >= call_next &&
5412 atomic_cas_64(&zc->zc_next, call_next, call_next +
5413 ztest_random(2 * zi->zi_interval[0] + 1)) == call_next) {
5414 ztest_execute(rand, zi, id);
5422 ztest_dataset_name(char *dsname, char *pool, int d)
5424 (void) snprintf(dsname, MAXNAMELEN, "%s/ds_%d", pool, d);
5428 ztest_dataset_destroy(int d)
5430 char name[MAXNAMELEN];
5432 ztest_dataset_name(name, ztest_opts.zo_pool, d);
5434 if (ztest_opts.zo_verbose >= 3)
5435 (void) printf("Destroying %s to free up space\n", name);
5438 * Cleanup any non-standard clones and snapshots. In general,
5439 * ztest thread t operates on dataset (t % zopt_datasets),
5440 * so there may be more than one thing to clean up.
5442 for (int t = d; t < ztest_opts.zo_threads;
5443 t += ztest_opts.zo_datasets) {
5444 ztest_dsl_dataset_cleanup(name, t);
5447 (void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
5448 DS_FIND_SNAPSHOTS | DS_FIND_CHILDREN);
5452 ztest_dataset_dirobj_verify(ztest_ds_t *zd)
5454 uint64_t usedobjs, dirobjs, scratch;
5457 * ZTEST_DIROBJ is the object directory for the entire dataset.
5458 * Therefore, the number of objects in use should equal the
5459 * number of ZTEST_DIROBJ entries, +1 for ZTEST_DIROBJ itself.
5460 * If not, we have an object leak.
5462 * Note that we can only check this in ztest_dataset_open(),
5463 * when the open-context and syncing-context values agree.
5464 * That's because zap_count() returns the open-context value,
5465 * while dmu_objset_space() returns the rootbp fill count.
5467 VERIFY3U(0, ==, zap_count(zd->zd_os, ZTEST_DIROBJ, &dirobjs));
5468 dmu_objset_space(zd->zd_os, &scratch, &scratch, &usedobjs, &scratch);
5469 ASSERT3U(dirobjs + 1, ==, usedobjs);
5473 ztest_dataset_open(int d)
5475 ztest_ds_t *zd = &ztest_ds[d];
5476 uint64_t committed_seq = ZTEST_GET_SHARED_DS(d)->zd_seq;
5479 char name[MAXNAMELEN];
5482 ztest_dataset_name(name, ztest_opts.zo_pool, d);
5484 (void) rw_rdlock(&ztest_name_lock);
5486 error = ztest_dataset_create(name);
5487 if (error == ENOSPC) {
5488 (void) rw_unlock(&ztest_name_lock);
5489 ztest_record_enospc(FTAG);
5492 ASSERT(error == 0 || error == EEXIST);
5494 VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, zd, &os));
5495 (void) rw_unlock(&ztest_name_lock);
5497 ztest_zd_init(zd, ZTEST_GET_SHARED_DS(d), os);
5499 zilog = zd->zd_zilog;
5501 if (zilog->zl_header->zh_claim_lr_seq != 0 &&
5502 zilog->zl_header->zh_claim_lr_seq < committed_seq)
5503 fatal(0, "missing log records: claimed %llu < committed %llu",
5504 zilog->zl_header->zh_claim_lr_seq, committed_seq);
5506 ztest_dataset_dirobj_verify(zd);
5508 zil_replay(os, zd, ztest_replay_vector);
5510 ztest_dataset_dirobj_verify(zd);
5512 if (ztest_opts.zo_verbose >= 6)
5513 (void) printf("%s replay %llu blocks, %llu records, seq %llu\n",
5515 (u_longlong_t)zilog->zl_parse_blk_count,
5516 (u_longlong_t)zilog->zl_parse_lr_count,
5517 (u_longlong_t)zilog->zl_replaying_seq);
5519 zilog = zil_open(os, ztest_get_data);
5521 if (zilog->zl_replaying_seq != 0 &&
5522 zilog->zl_replaying_seq < committed_seq)
5523 fatal(0, "missing log records: replayed %llu < committed %llu",
5524 zilog->zl_replaying_seq, committed_seq);
5530 ztest_dataset_close(int d)
5532 ztest_ds_t *zd = &ztest_ds[d];
5534 zil_close(zd->zd_zilog);
5535 dmu_objset_disown(zd->zd_os, zd);
5541 * Kick off threads to run tests on all datasets in parallel.
5544 ztest_run(ztest_shared_t *zs)
5549 thread_t resume_tid;
5552 ztest_exiting = B_FALSE;
5555 * Initialize parent/child shared state.
5557 VERIFY(_mutex_init(&ztest_vdev_lock, USYNC_THREAD, NULL) == 0);
5558 VERIFY(rwlock_init(&ztest_name_lock, USYNC_THREAD, NULL) == 0);
5560 zs->zs_thread_start = gethrtime();
5561 zs->zs_thread_stop =
5562 zs->zs_thread_start + ztest_opts.zo_passtime * NANOSEC;
5563 zs->zs_thread_stop = MIN(zs->zs_thread_stop, zs->zs_proc_stop);
5564 zs->zs_thread_kill = zs->zs_thread_stop;
5565 if (ztest_random(100) < ztest_opts.zo_killrate) {
5566 zs->zs_thread_kill -=
5567 ztest_random(ztest_opts.zo_passtime * NANOSEC);
5570 (void) _mutex_init(&zcl.zcl_callbacks_lock, USYNC_THREAD, NULL);
5572 list_create(&zcl.zcl_callbacks, sizeof (ztest_cb_data_t),
5573 offsetof(ztest_cb_data_t, zcd_node));
5578 kernel_init(FREAD | FWRITE);
5579 VERIFY0(spa_open(ztest_opts.zo_pool, &spa, FTAG));
5580 spa->spa_debug = B_TRUE;
5583 VERIFY0(dmu_objset_own(ztest_opts.zo_pool,
5584 DMU_OST_ANY, B_TRUE, FTAG, &os));
5585 zs->zs_guid = dmu_objset_fsid_guid(os);
5586 dmu_objset_disown(os, FTAG);
5588 spa->spa_dedup_ditto = 2 * ZIO_DEDUPDITTO_MIN;
5591 * We don't expect the pool to suspend unless maxfaults == 0,
5592 * in which case ztest_fault_inject() temporarily takes away
5593 * the only valid replica.
5595 if (MAXFAULTS() == 0)
5596 spa->spa_failmode = ZIO_FAILURE_MODE_WAIT;
5598 spa->spa_failmode = ZIO_FAILURE_MODE_PANIC;
5601 * Create a thread to periodically resume suspended I/O.
5603 VERIFY(thr_create(0, 0, ztest_resume_thread, spa, THR_BOUND,
5607 * Create a deadman thread to abort() if we hang.
5609 VERIFY(thr_create(0, 0, ztest_deadman_thread, zs, THR_BOUND,
5613 * Verify that we can safely inquire about about any object,
5614 * whether it's allocated or not. To make it interesting,
5615 * we probe a 5-wide window around each power of two.
5616 * This hits all edge cases, including zero and the max.
5618 for (int t = 0; t < 64; t++) {
5619 for (int d = -5; d <= 5; d++) {
5620 error = dmu_object_info(spa->spa_meta_objset,
5621 (1ULL << t) + d, NULL);
5622 ASSERT(error == 0 || error == ENOENT ||
5628 * If we got any ENOSPC errors on the previous run, destroy something.
5630 if (zs->zs_enospc_count != 0) {
5631 int d = ztest_random(ztest_opts.zo_datasets);
5632 ztest_dataset_destroy(d);
5634 zs->zs_enospc_count = 0;
5636 tid = umem_zalloc(ztest_opts.zo_threads * sizeof (thread_t),
5639 if (ztest_opts.zo_verbose >= 4)
5640 (void) printf("starting main threads...\n");
5643 * Kick off all the tests that run in parallel.
5645 for (int t = 0; t < ztest_opts.zo_threads; t++) {
5646 if (t < ztest_opts.zo_datasets &&
5647 ztest_dataset_open(t) != 0)
5649 VERIFY(thr_create(0, 0, ztest_thread, (void *)(uintptr_t)t,
5650 THR_BOUND, &tid[t]) == 0);
5654 * Wait for all of the tests to complete. We go in reverse order
5655 * so we don't close datasets while threads are still using them.
5657 for (int t = ztest_opts.zo_threads - 1; t >= 0; t--) {
5658 VERIFY(thr_join(tid[t], NULL, NULL) == 0);
5659 if (t < ztest_opts.zo_datasets)
5660 ztest_dataset_close(t);
5663 txg_wait_synced(spa_get_dsl(spa), 0);
5665 zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(spa));
5666 zs->zs_space = metaslab_class_get_space(spa_normal_class(spa));
5667 zfs_dbgmsg_print(FTAG);
5669 umem_free(tid, ztest_opts.zo_threads * sizeof (thread_t));
5671 /* Kill the resume thread */
5672 ztest_exiting = B_TRUE;
5673 VERIFY(thr_join(resume_tid, NULL, NULL) == 0);
5677 * Right before closing the pool, kick off a bunch of async I/O;
5678 * spa_close() should wait for it to complete.
5680 for (uint64_t object = 1; object < 50; object++)
5681 dmu_prefetch(spa->spa_meta_objset, object, 0, 1ULL << 20);
5683 spa_close(spa, FTAG);
5686 * Verify that we can loop over all pools.
5688 mutex_enter(&spa_namespace_lock);
5689 for (spa = spa_next(NULL); spa != NULL; spa = spa_next(spa))
5690 if (ztest_opts.zo_verbose > 3)
5691 (void) printf("spa_next: found %s\n", spa_name(spa));
5692 mutex_exit(&spa_namespace_lock);
5695 * Verify that we can export the pool and reimport it under a
5698 if (ztest_random(2) == 0) {
5699 char name[MAXNAMELEN];
5700 (void) snprintf(name, MAXNAMELEN, "%s_import",
5701 ztest_opts.zo_pool);
5702 ztest_spa_import_export(ztest_opts.zo_pool, name);
5703 ztest_spa_import_export(name, ztest_opts.zo_pool);
5708 list_destroy(&zcl.zcl_callbacks);
5710 (void) _mutex_destroy(&zcl.zcl_callbacks_lock);
5712 (void) rwlock_destroy(&ztest_name_lock);
5713 (void) _mutex_destroy(&ztest_vdev_lock);
5719 ztest_ds_t *zd = &ztest_ds[0];
5723 if (ztest_opts.zo_verbose >= 3)
5724 (void) printf("testing spa_freeze()...\n");
5726 kernel_init(FREAD | FWRITE);
5727 VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
5728 VERIFY3U(0, ==, ztest_dataset_open(0));
5729 spa->spa_debug = B_TRUE;
5733 * Force the first log block to be transactionally allocated.
5734 * We have to do this before we freeze the pool -- otherwise
5735 * the log chain won't be anchored.
5737 while (BP_IS_HOLE(&zd->zd_zilog->zl_header->zh_log)) {
5738 ztest_dmu_object_alloc_free(zd, 0);
5739 zil_commit(zd->zd_zilog, 0);
5742 txg_wait_synced(spa_get_dsl(spa), 0);
5745 * Freeze the pool. This stops spa_sync() from doing anything,
5746 * so that the only way to record changes from now on is the ZIL.
5751 * Run tests that generate log records but don't alter the pool config
5752 * or depend on DSL sync tasks (snapshots, objset create/destroy, etc).
5753 * We do a txg_wait_synced() after each iteration to force the txg
5754 * to increase well beyond the last synced value in the uberblock.
5755 * The ZIL should be OK with that.
5757 while (ztest_random(10) != 0 &&
5758 numloops++ < ztest_opts.zo_maxloops) {
5759 ztest_dmu_write_parallel(zd, 0);
5760 ztest_dmu_object_alloc_free(zd, 0);
5761 txg_wait_synced(spa_get_dsl(spa), 0);
5765 * Commit all of the changes we just generated.
5767 zil_commit(zd->zd_zilog, 0);
5768 txg_wait_synced(spa_get_dsl(spa), 0);
5771 * Close our dataset and close the pool.
5773 ztest_dataset_close(0);
5774 spa_close(spa, FTAG);
5778 * Open and close the pool and dataset to induce log replay.
5780 kernel_init(FREAD | FWRITE);
5781 VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
5782 ASSERT(spa_freeze_txg(spa) == UINT64_MAX);
5783 VERIFY3U(0, ==, ztest_dataset_open(0));
5784 ztest_dataset_close(0);
5786 spa->spa_debug = B_TRUE;
5788 txg_wait_synced(spa_get_dsl(spa), 0);
5789 ztest_reguid(NULL, 0);
5791 spa_close(spa, FTAG);
5796 print_time(hrtime_t t, char *timebuf)
5798 hrtime_t s = t / NANOSEC;
5799 hrtime_t m = s / 60;
5800 hrtime_t h = m / 60;
5801 hrtime_t d = h / 24;
5810 (void) sprintf(timebuf,
5811 "%llud%02lluh%02llum%02llus", d, h, m, s);
5813 (void) sprintf(timebuf, "%lluh%02llum%02llus", h, m, s);
5815 (void) sprintf(timebuf, "%llum%02llus", m, s);
5817 (void) sprintf(timebuf, "%llus", s);
5825 VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
5826 if (ztest_random(2) == 0)
5828 VERIFY(nvlist_add_uint64(props, "autoreplace", 1) == 0);
5834 * Create a storage pool with the given name and initial vdev size.
5835 * Then test spa_freeze() functionality.
5838 ztest_init(ztest_shared_t *zs)
5841 nvlist_t *nvroot, *props;
5843 VERIFY(_mutex_init(&ztest_vdev_lock, USYNC_THREAD, NULL) == 0);
5844 VERIFY(rwlock_init(&ztest_name_lock, USYNC_THREAD, NULL) == 0);
5846 kernel_init(FREAD | FWRITE);
5849 * Create the storage pool.
5851 (void) spa_destroy(ztest_opts.zo_pool);
5852 ztest_shared->zs_vdev_next_leaf = 0;
5854 zs->zs_mirrors = ztest_opts.zo_mirrors;
5855 nvroot = make_vdev_root(NULL, NULL, NULL, ztest_opts.zo_vdev_size, 0,
5856 0, ztest_opts.zo_raidz, zs->zs_mirrors, 1);
5857 props = make_random_props();
5858 for (int i = 0; i < SPA_FEATURES; i++) {
5860 (void) snprintf(buf, sizeof (buf), "feature@%s",
5861 spa_feature_table[i].fi_uname);
5862 VERIFY3U(0, ==, nvlist_add_uint64(props, buf, 0));
5864 VERIFY3U(0, ==, spa_create(ztest_opts.zo_pool, nvroot, props, NULL));
5865 nvlist_free(nvroot);
5867 VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
5868 zs->zs_metaslab_sz =
5869 1ULL << spa->spa_root_vdev->vdev_child[0]->vdev_ms_shift;
5871 spa_close(spa, FTAG);
5875 ztest_run_zdb(ztest_opts.zo_pool);
5879 ztest_run_zdb(ztest_opts.zo_pool);
5881 (void) rwlock_destroy(&ztest_name_lock);
5882 (void) _mutex_destroy(&ztest_vdev_lock);
5888 static char ztest_name_data[] = "/tmp/ztest.data.XXXXXX";
5890 ztest_fd_data = mkstemp(ztest_name_data);
5891 ASSERT3S(ztest_fd_data, >=, 0);
5892 (void) unlink(ztest_name_data);
5897 shared_data_size(ztest_shared_hdr_t *hdr)
5901 size = hdr->zh_hdr_size;
5902 size += hdr->zh_opts_size;
5903 size += hdr->zh_size;
5904 size += hdr->zh_stats_size * hdr->zh_stats_count;
5905 size += hdr->zh_ds_size * hdr->zh_ds_count;
5914 ztest_shared_hdr_t *hdr;
5916 hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
5917 PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
5918 ASSERT(hdr != MAP_FAILED);
5920 VERIFY3U(0, ==, ftruncate(ztest_fd_data, sizeof (ztest_shared_hdr_t)));
5922 hdr->zh_hdr_size = sizeof (ztest_shared_hdr_t);
5923 hdr->zh_opts_size = sizeof (ztest_shared_opts_t);
5924 hdr->zh_size = sizeof (ztest_shared_t);
5925 hdr->zh_stats_size = sizeof (ztest_shared_callstate_t);
5926 hdr->zh_stats_count = ZTEST_FUNCS;
5927 hdr->zh_ds_size = sizeof (ztest_shared_ds_t);
5928 hdr->zh_ds_count = ztest_opts.zo_datasets;
5930 size = shared_data_size(hdr);
5931 VERIFY3U(0, ==, ftruncate(ztest_fd_data, size));
5933 (void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
5940 ztest_shared_hdr_t *hdr;
5943 hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
5944 PROT_READ, MAP_SHARED, ztest_fd_data, 0);
5945 ASSERT(hdr != MAP_FAILED);
5947 size = shared_data_size(hdr);
5949 (void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
5950 hdr = ztest_shared_hdr = (void *)mmap(0, P2ROUNDUP(size, getpagesize()),
5951 PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
5952 ASSERT(hdr != MAP_FAILED);
5953 buf = (uint8_t *)hdr;
5955 offset = hdr->zh_hdr_size;
5956 ztest_shared_opts = (void *)&buf[offset];
5957 offset += hdr->zh_opts_size;
5958 ztest_shared = (void *)&buf[offset];
5959 offset += hdr->zh_size;
5960 ztest_shared_callstate = (void *)&buf[offset];
5961 offset += hdr->zh_stats_size * hdr->zh_stats_count;
5962 ztest_shared_ds = (void *)&buf[offset];
5966 exec_child(char *cmd, char *libpath, boolean_t ignorekill, int *statusp)
5970 char *cmdbuf = NULL;
5975 cmdbuf = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
5976 (void) strlcpy(cmdbuf, getexecname(), MAXPATHLEN);
5981 fatal(1, "fork failed");
5983 if (pid == 0) { /* child */
5984 char *emptyargv[2] = { cmd, NULL };
5985 char fd_data_str[12];
5987 struct rlimit rl = { 1024, 1024 };
5988 (void) setrlimit(RLIMIT_NOFILE, &rl);
5990 (void) close(ztest_fd_rand);
5992 snprintf(fd_data_str, 12, "%d", ztest_fd_data));
5993 VERIFY0(setenv("ZTEST_FD_DATA", fd_data_str, 1));
5995 (void) enable_extended_FILE_stdio(-1, -1);
5996 if (libpath != NULL)
5997 VERIFY(0 == setenv("LD_LIBRARY_PATH", libpath, 1));
5999 (void) execv(cmd, emptyargv);
6001 (void) execvp(cmd, emptyargv);
6003 ztest_dump_core = B_FALSE;
6004 fatal(B_TRUE, "exec failed: %s", cmd);
6007 if (cmdbuf != NULL) {
6008 umem_free(cmdbuf, MAXPATHLEN);
6012 while (waitpid(pid, &status, 0) != pid)
6014 if (statusp != NULL)
6017 if (WIFEXITED(status)) {
6018 if (WEXITSTATUS(status) != 0) {
6019 (void) fprintf(stderr, "child exited with code %d\n",
6020 WEXITSTATUS(status));
6024 } else if (WIFSIGNALED(status)) {
6025 if (!ignorekill || WTERMSIG(status) != SIGKILL) {
6026 (void) fprintf(stderr, "child died with signal %d\n",
6032 (void) fprintf(stderr, "something strange happened to child\n");
6039 ztest_run_init(void)
6041 ztest_shared_t *zs = ztest_shared;
6043 ASSERT(ztest_opts.zo_init != 0);
6046 * Blow away any existing copy of zpool.cache
6048 (void) remove(spa_config_path);
6051 * Create and initialize our storage pool.
6053 for (int i = 1; i <= ztest_opts.zo_init; i++) {
6054 bzero(zs, sizeof (ztest_shared_t));
6055 if (ztest_opts.zo_verbose >= 3 &&
6056 ztest_opts.zo_init != 1) {
6057 (void) printf("ztest_init(), pass %d\n", i);
6064 main(int argc, char **argv)
6072 ztest_shared_callstate_t *zc;
6078 char *fd_data_str = getenv("ZTEST_FD_DATA");
6080 (void) setvbuf(stdout, NULL, _IOLBF, 0);
6082 dprintf_setup(&argc, argv);
6083 zfs_deadman_synctime_ms = 300000;
6085 ztest_fd_rand = open("/dev/urandom", O_RDONLY);
6086 ASSERT3S(ztest_fd_rand, >=, 0);
6089 process_options(argc, argv);
6094 bcopy(&ztest_opts, ztest_shared_opts,
6095 sizeof (*ztest_shared_opts));
6097 ztest_fd_data = atoi(fd_data_str);
6099 bcopy(ztest_shared_opts, &ztest_opts, sizeof (ztest_opts));
6101 ASSERT3U(ztest_opts.zo_datasets, ==, ztest_shared_hdr->zh_ds_count);
6103 /* Override location of zpool.cache */
6104 VERIFY3U(asprintf((char **)&spa_config_path, "%s/zpool.cache",
6105 ztest_opts.zo_dir), !=, -1);
6107 ztest_ds = umem_alloc(ztest_opts.zo_datasets * sizeof (ztest_ds_t),
6112 metaslab_gang_bang = ztest_opts.zo_metaslab_gang_bang;
6113 metaslab_df_alloc_threshold =
6114 zs->zs_metaslab_df_alloc_threshold;
6123 hasalt = (strlen(ztest_opts.zo_alt_ztest) != 0);
6125 if (ztest_opts.zo_verbose >= 1) {
6126 (void) printf("%llu vdevs, %d datasets, %d threads,"
6127 " %llu seconds...\n",
6128 (u_longlong_t)ztest_opts.zo_vdevs,
6129 ztest_opts.zo_datasets,
6130 ztest_opts.zo_threads,
6131 (u_longlong_t)ztest_opts.zo_time);
6134 cmd = umem_alloc(MAXNAMELEN, UMEM_NOFAIL);
6135 (void) strlcpy(cmd, getexecname(), MAXNAMELEN);
6137 zs->zs_do_init = B_TRUE;
6138 if (strlen(ztest_opts.zo_alt_ztest) != 0) {
6139 if (ztest_opts.zo_verbose >= 1) {
6140 (void) printf("Executing older ztest for "
6141 "initialization: %s\n", ztest_opts.zo_alt_ztest);
6143 VERIFY(!exec_child(ztest_opts.zo_alt_ztest,
6144 ztest_opts.zo_alt_libpath, B_FALSE, NULL));
6146 VERIFY(!exec_child(NULL, NULL, B_FALSE, NULL));
6148 zs->zs_do_init = B_FALSE;
6150 zs->zs_proc_start = gethrtime();
6151 zs->zs_proc_stop = zs->zs_proc_start + ztest_opts.zo_time * NANOSEC;
6153 for (int f = 0; f < ZTEST_FUNCS; f++) {
6154 zi = &ztest_info[f];
6155 zc = ZTEST_GET_SHARED_CALLSTATE(f);
6156 if (zs->zs_proc_start + zi->zi_interval[0] > zs->zs_proc_stop)
6157 zc->zc_next = UINT64_MAX;
6159 zc->zc_next = zs->zs_proc_start +
6160 ztest_random(2 * zi->zi_interval[0] + 1);
6164 * Run the tests in a loop. These tests include fault injection
6165 * to verify that self-healing data works, and forced crashes
6166 * to verify that we never lose on-disk consistency.
6168 while (gethrtime() < zs->zs_proc_stop) {
6173 * Initialize the workload counters for each function.
6175 for (int f = 0; f < ZTEST_FUNCS; f++) {
6176 zc = ZTEST_GET_SHARED_CALLSTATE(f);
6181 /* Set the allocation switch size */
6182 zs->zs_metaslab_df_alloc_threshold =
6183 ztest_random(zs->zs_metaslab_sz / 4) + 1;
6185 if (!hasalt || ztest_random(2) == 0) {
6186 if (hasalt && ztest_opts.zo_verbose >= 1) {
6187 (void) printf("Executing newer ztest: %s\n",
6191 killed = exec_child(cmd, NULL, B_TRUE, &status);
6193 if (hasalt && ztest_opts.zo_verbose >= 1) {
6194 (void) printf("Executing older ztest: %s\n",
6195 ztest_opts.zo_alt_ztest);
6198 killed = exec_child(ztest_opts.zo_alt_ztest,
6199 ztest_opts.zo_alt_libpath, B_TRUE, &status);
6206 if (ztest_opts.zo_verbose >= 1) {
6207 hrtime_t now = gethrtime();
6209 now = MIN(now, zs->zs_proc_stop);
6210 print_time(zs->zs_proc_stop - now, timebuf);
6211 nicenum(zs->zs_space, numbuf);
6213 (void) printf("Pass %3d, %8s, %3llu ENOSPC, "
6214 "%4.1f%% of %5s used, %3.0f%% done, %8s to go\n",
6216 WIFEXITED(status) ? "Complete" : "SIGKILL",
6217 (u_longlong_t)zs->zs_enospc_count,
6218 100.0 * zs->zs_alloc / zs->zs_space,
6220 100.0 * (now - zs->zs_proc_start) /
6221 (ztest_opts.zo_time * NANOSEC), timebuf);
6224 if (ztest_opts.zo_verbose >= 2) {
6225 (void) printf("\nWorkload summary:\n\n");
6226 (void) printf("%7s %9s %s\n",
6227 "Calls", "Time", "Function");
6228 (void) printf("%7s %9s %s\n",
6229 "-----", "----", "--------");
6230 for (int f = 0; f < ZTEST_FUNCS; f++) {
6233 zi = &ztest_info[f];
6234 zc = ZTEST_GET_SHARED_CALLSTATE(f);
6235 print_time(zc->zc_time, timebuf);
6236 (void) dladdr((void *)zi->zi_func, &dli);
6237 (void) printf("%7llu %9s %s\n",
6238 (u_longlong_t)zc->zc_count, timebuf,
6241 (void) printf("\n");
6245 * It's possible that we killed a child during a rename test,
6246 * in which case we'll have a 'ztest_tmp' pool lying around
6247 * instead of 'ztest'. Do a blind rename in case this happened.
6250 if (spa_open(ztest_opts.zo_pool, &spa, FTAG) == 0) {
6251 spa_close(spa, FTAG);
6253 char tmpname[MAXNAMELEN];
6255 kernel_init(FREAD | FWRITE);
6256 (void) snprintf(tmpname, sizeof (tmpname), "%s_tmp",
6257 ztest_opts.zo_pool);
6258 (void) spa_rename(tmpname, ztest_opts.zo_pool);
6262 ztest_run_zdb(ztest_opts.zo_pool);
6265 if (ztest_opts.zo_verbose >= 1) {
6267 (void) printf("%d runs of older ztest: %s\n", older,
6268 ztest_opts.zo_alt_ztest);
6269 (void) printf("%d runs of newer ztest: %s\n", newer,
6272 (void) printf("%d killed, %d completed, %.0f%% kill rate\n",
6273 kills, iters - kills, (100.0 * kills) / MAX(1, iters));
6276 umem_free(cmd, MAXNAMELEN);