4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2011, 2016 by Delphix. All rights reserved.
24 * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
25 * Copyright (c) 2013 Steven Hartland. All rights reserved.
26 * Copyright (c) 2014 Integros [integros.com]
27 * Copyright 2017 Joyent, Inc.
28 * Copyright 2017 RackTop Systems.
32 * The objective of this program is to provide a DMU/ZAP/SPA stress test
33 * that runs entirely in userland, is easy to use, and easy to extend.
35 * The overall design of the ztest program is as follows:
37 * (1) For each major functional area (e.g. adding vdevs to a pool,
38 * creating and destroying datasets, reading and writing objects, etc)
39 * we have a simple routine to test that functionality. These
40 * individual routines do not have to do anything "stressful".
42 * (2) We turn these simple functionality tests into a stress test by
43 * running them all in parallel, with as many threads as desired,
44 * and spread across as many datasets, objects, and vdevs as desired.
46 * (3) While all this is happening, we inject faults into the pool to
47 * verify that self-healing data really works.
49 * (4) Every time we open a dataset, we change its checksum and compression
50 * functions. Thus even individual objects vary from block to block
51 * in which checksum they use and whether they're compressed.
53 * (5) To verify that we never lose on-disk consistency after a crash,
54 * we run the entire test in a child of the main process.
55 * At random times, the child self-immolates with a SIGKILL.
56 * This is the software equivalent of pulling the power cord.
57 * The parent then runs the test again, using the existing
58 * storage pool, as many times as desired. If backwards compatibility
59 * testing is enabled ztest will sometimes run the "older" version
60 * of ztest after a SIGKILL.
62 * (6) To verify that we don't have future leaks or temporal incursions,
63 * many of the functional tests record the transaction group number
64 * as part of their data. When reading old data, they verify that
65 * the transaction group number is less than the current, open txg.
66 * If you add a new test, please do this if applicable.
68 * When run with no arguments, ztest runs for about five minutes and
69 * produces no output if successful. To get a little bit of information,
70 * specify -V. To get more information, specify -VV, and so on.
72 * To turn this into an overnight stress test, use -T to specify run time.
74 * You can ask more more vdevs [-v], datasets [-d], or threads [-t]
75 * to increase the pool capacity, fanout, and overall stress level.
77 * Use the -k option to set the desired frequency of kills.
79 * When ztest invokes itself it passes all relevant information through a
80 * temporary file which is mmap-ed in the child process. This allows shared
81 * memory to survive the exec syscall. The ztest_shared_hdr_t struct is always
82 * stored at offset 0 of this file and contains information on the size and
83 * number of shared structures in the file. The information stored in this file
84 * must remain backwards compatible with older versions of ztest so that
85 * ztest can invoke them during backwards compatibility testing (-B).
88 #include <sys/zfs_context.h>
94 #include <sys/dmu_objset.h>
100 #include <sys/resource.h>
103 #include <sys/zil_impl.h>
104 #include <sys/vdev_impl.h>
105 #include <sys/vdev_file.h>
106 #include <sys/spa_impl.h>
107 #include <sys/metaslab_impl.h>
108 #include <sys/dsl_prop.h>
109 #include <sys/dsl_dataset.h>
110 #include <sys/dsl_destroy.h>
111 #include <sys/dsl_scan.h>
112 #include <sys/zio_checksum.h>
113 #include <sys/refcount.h>
114 #include <sys/zfeature.h>
115 #include <sys/dsl_userhold.h>
118 #include <stdio_ext.h>
126 #include <sys/fs/zfs.h>
127 #include <libnvpair.h>
128 #include <libcmdutils.h>
130 static int ztest_fd_data
= -1;
131 static int ztest_fd_rand
= -1;
133 typedef struct ztest_shared_hdr
{
134 uint64_t zh_hdr_size
;
135 uint64_t zh_opts_size
;
137 uint64_t zh_stats_size
;
138 uint64_t zh_stats_count
;
140 uint64_t zh_ds_count
;
141 } ztest_shared_hdr_t
;
143 static ztest_shared_hdr_t
*ztest_shared_hdr
;
145 typedef struct ztest_shared_opts
{
146 char zo_pool
[ZFS_MAX_DATASET_NAME_LEN
];
147 char zo_dir
[ZFS_MAX_DATASET_NAME_LEN
];
148 char zo_alt_ztest
[MAXNAMELEN
];
149 char zo_alt_libpath
[MAXNAMELEN
];
151 uint64_t zo_vdevtime
;
159 uint64_t zo_passtime
;
160 uint64_t zo_killrate
;
164 uint64_t zo_maxloops
;
165 uint64_t zo_metaslab_gang_bang
;
166 } ztest_shared_opts_t
;
168 static const ztest_shared_opts_t ztest_opts_defaults
= {
169 .zo_pool
= { 'z', 't', 'e', 's', 't', '\0' },
170 .zo_dir
= { '/', 't', 'm', 'p', '\0' },
171 .zo_alt_ztest
= { '\0' },
172 .zo_alt_libpath
= { '\0' },
174 .zo_ashift
= SPA_MINBLOCKSHIFT
,
177 .zo_raidz_parity
= 1,
178 .zo_vdev_size
= SPA_MINDEVSIZE
* 4, /* 256m default size */
181 .zo_passtime
= 60, /* 60 seconds */
182 .zo_killrate
= 70, /* 70% kill rate */
185 .zo_time
= 300, /* 5 minutes */
186 .zo_maxloops
= 50, /* max loops during spa_freeze() */
187 .zo_metaslab_gang_bang
= 32 << 10
190 extern uint64_t metaslab_gang_bang
;
191 extern uint64_t metaslab_df_alloc_threshold
;
192 extern uint64_t zfs_deadman_synctime_ms
;
193 extern int metaslab_preload_limit
;
194 extern boolean_t zfs_compressed_arc_enabled
;
195 extern boolean_t zfs_abd_scatter_enabled
;
197 static ztest_shared_opts_t
*ztest_shared_opts
;
198 static ztest_shared_opts_t ztest_opts
;
200 typedef struct ztest_shared_ds
{
204 static ztest_shared_ds_t
*ztest_shared_ds
;
205 #define ZTEST_GET_SHARED_DS(d) (&ztest_shared_ds[d])
207 #define BT_MAGIC 0x123456789abcdefULL
208 #define MAXFAULTS() \
209 (MAX(zs->zs_mirrors, 1) * (ztest_opts.zo_raidz_parity + 1) - 1)
213 ZTEST_IO_WRITE_PATTERN
,
214 ZTEST_IO_WRITE_ZEROES
,
221 typedef struct ztest_block_tag
{
231 typedef struct bufwad
{
238 * XXX -- fix zfs range locks to be generic so we can use them here.
260 #define ZTEST_RANGE_LOCKS 64
261 #define ZTEST_OBJECT_LOCKS 64
264 * Object descriptor. Used as a template for object lookup/create/remove.
266 typedef struct ztest_od
{
269 dmu_object_type_t od_type
;
270 dmu_object_type_t od_crtype
;
271 uint64_t od_blocksize
;
272 uint64_t od_crblocksize
;
275 char od_name
[ZFS_MAX_DATASET_NAME_LEN
];
281 typedef struct ztest_ds
{
282 ztest_shared_ds_t
*zd_shared
;
284 krwlock_t zd_zilog_lock
;
286 ztest_od_t
*zd_od
; /* debugging aid */
287 char zd_name
[ZFS_MAX_DATASET_NAME_LEN
];
288 kmutex_t zd_dirobj_lock
;
289 rll_t zd_object_lock
[ZTEST_OBJECT_LOCKS
];
290 rll_t zd_range_lock
[ZTEST_RANGE_LOCKS
];
294 * Per-iteration state.
296 typedef void ztest_func_t(ztest_ds_t
*zd
, uint64_t id
);
298 typedef struct ztest_info
{
299 ztest_func_t
*zi_func
; /* test function */
300 uint64_t zi_iters
; /* iterations per execution */
301 uint64_t *zi_interval
; /* execute every <interval> seconds */
304 typedef struct ztest_shared_callstate
{
305 uint64_t zc_count
; /* per-pass count */
306 uint64_t zc_time
; /* per-pass time */
307 uint64_t zc_next
; /* next time to call this function */
308 } ztest_shared_callstate_t
;
310 static ztest_shared_callstate_t
*ztest_shared_callstate
;
311 #define ZTEST_GET_SHARED_CALLSTATE(c) (&ztest_shared_callstate[c])
314 * Note: these aren't static because we want dladdr() to work.
316 ztest_func_t ztest_dmu_read_write
;
317 ztest_func_t ztest_dmu_write_parallel
;
318 ztest_func_t ztest_dmu_object_alloc_free
;
319 ztest_func_t ztest_dmu_commit_callbacks
;
320 ztest_func_t ztest_zap
;
321 ztest_func_t ztest_zap_parallel
;
322 ztest_func_t ztest_zil_commit
;
323 ztest_func_t ztest_zil_remount
;
324 ztest_func_t ztest_dmu_read_write_zcopy
;
325 ztest_func_t ztest_dmu_objset_create_destroy
;
326 ztest_func_t ztest_dmu_prealloc
;
327 ztest_func_t ztest_fzap
;
328 ztest_func_t ztest_dmu_snapshot_create_destroy
;
329 ztest_func_t ztest_dsl_prop_get_set
;
330 ztest_func_t ztest_spa_prop_get_set
;
331 ztest_func_t ztest_spa_create_destroy
;
332 ztest_func_t ztest_fault_inject
;
333 ztest_func_t ztest_ddt_repair
;
334 ztest_func_t ztest_dmu_snapshot_hold
;
335 ztest_func_t ztest_spa_rename
;
336 ztest_func_t ztest_scrub
;
337 ztest_func_t ztest_dsl_dataset_promote_busy
;
338 ztest_func_t ztest_vdev_attach_detach
;
339 ztest_func_t ztest_vdev_LUN_growth
;
340 ztest_func_t ztest_vdev_add_remove
;
341 ztest_func_t ztest_vdev_aux_add_remove
;
342 ztest_func_t ztest_split_pool
;
343 ztest_func_t ztest_reguid
;
344 ztest_func_t ztest_spa_upgrade
;
345 ztest_func_t ztest_device_removal
;
346 ztest_func_t ztest_remap_blocks
;
348 uint64_t zopt_always
= 0ULL * NANOSEC
; /* all the time */
349 uint64_t zopt_incessant
= 1ULL * NANOSEC
/ 10; /* every 1/10 second */
350 uint64_t zopt_often
= 1ULL * NANOSEC
; /* every second */
351 uint64_t zopt_sometimes
= 10ULL * NANOSEC
; /* every 10 seconds */
352 uint64_t zopt_rarely
= 60ULL * NANOSEC
; /* every 60 seconds */
354 ztest_info_t ztest_info
[] = {
355 { ztest_dmu_read_write
, 1, &zopt_always
},
356 { ztest_dmu_write_parallel
, 10, &zopt_always
},
357 { ztest_dmu_object_alloc_free
, 1, &zopt_always
},
358 { ztest_dmu_commit_callbacks
, 1, &zopt_always
},
359 { ztest_zap
, 30, &zopt_always
},
360 { ztest_zap_parallel
, 100, &zopt_always
},
361 { ztest_split_pool
, 1, &zopt_always
},
362 { ztest_zil_commit
, 1, &zopt_incessant
},
363 { ztest_zil_remount
, 1, &zopt_sometimes
},
364 { ztest_dmu_read_write_zcopy
, 1, &zopt_often
},
365 { ztest_dmu_objset_create_destroy
, 1, &zopt_often
},
366 { ztest_dsl_prop_get_set
, 1, &zopt_often
},
367 { ztest_spa_prop_get_set
, 1, &zopt_sometimes
},
369 { ztest_dmu_prealloc
, 1, &zopt_sometimes
},
371 { ztest_fzap
, 1, &zopt_sometimes
},
372 { ztest_dmu_snapshot_create_destroy
, 1, &zopt_sometimes
},
373 { ztest_spa_create_destroy
, 1, &zopt_sometimes
},
374 { ztest_fault_inject
, 1, &zopt_sometimes
},
375 { ztest_ddt_repair
, 1, &zopt_sometimes
},
376 { ztest_dmu_snapshot_hold
, 1, &zopt_sometimes
},
377 { ztest_reguid
, 1, &zopt_rarely
},
378 { ztest_spa_rename
, 1, &zopt_rarely
},
379 { ztest_scrub
, 1, &zopt_rarely
},
380 { ztest_spa_upgrade
, 1, &zopt_rarely
},
381 { ztest_dsl_dataset_promote_busy
, 1, &zopt_rarely
},
382 { ztest_vdev_attach_detach
, 1, &zopt_sometimes
},
383 { ztest_vdev_LUN_growth
, 1, &zopt_rarely
},
384 { ztest_vdev_add_remove
, 1,
385 &ztest_opts
.zo_vdevtime
},
386 { ztest_vdev_aux_add_remove
, 1,
387 &ztest_opts
.zo_vdevtime
},
388 { ztest_device_removal
, 1, &zopt_sometimes
},
389 { ztest_remap_blocks
, 1, &zopt_sometimes
}
392 #define ZTEST_FUNCS (sizeof (ztest_info) / sizeof (ztest_info_t))
395 * The following struct is used to hold a list of uncalled commit callbacks.
396 * The callbacks are ordered by txg number.
398 typedef struct ztest_cb_list
{
399 kmutex_t zcl_callbacks_lock
;
400 list_t zcl_callbacks
;
404 * Stuff we need to share writably between parent and child.
406 typedef struct ztest_shared
{
407 boolean_t zs_do_init
;
408 hrtime_t zs_proc_start
;
409 hrtime_t zs_proc_stop
;
410 hrtime_t zs_thread_start
;
411 hrtime_t zs_thread_stop
;
412 hrtime_t zs_thread_kill
;
413 uint64_t zs_enospc_count
;
414 uint64_t zs_vdev_next_leaf
;
415 uint64_t zs_vdev_aux
;
420 uint64_t zs_metaslab_sz
;
421 uint64_t zs_metaslab_df_alloc_threshold
;
425 #define ID_PARALLEL -1ULL
427 static char ztest_dev_template
[] = "%s/%s.%llua";
428 static char ztest_aux_template
[] = "%s/%s.%s.%llu";
429 ztest_shared_t
*ztest_shared
;
431 static spa_t
*ztest_spa
= NULL
;
432 static ztest_ds_t
*ztest_ds
;
434 static kmutex_t ztest_vdev_lock
;
437 * The ztest_name_lock protects the pool and dataset namespace used by
438 * the individual tests. To modify the namespace, consumers must grab
439 * this lock as writer. Grabbing the lock as reader will ensure that the
440 * namespace does not change while the lock is held.
442 static krwlock_t ztest_name_lock
;
444 static boolean_t ztest_dump_core
= B_TRUE
;
445 static boolean_t ztest_exiting
;
447 /* Global commit callback list */
448 static ztest_cb_list_t zcl
;
451 ZTEST_META_DNODE
= 0,
456 static void usage(boolean_t
) __NORETURN
;
459 * These libumem hooks provide a reasonable set of defaults for the allocator's
460 * debugging facilities.
465 return ("default,verbose"); /* $UMEM_DEBUG setting */
469 _umem_logging_init(void)
471 return ("fail,contents"); /* $UMEM_LOGGING setting */
474 #define FATAL_MSG_SZ 1024
479 fatal(int do_perror
, char *message
, ...)
482 int save_errno
= errno
;
483 char buf
[FATAL_MSG_SZ
];
485 (void) fflush(stdout
);
487 va_start(args
, message
);
488 (void) sprintf(buf
, "ztest: ");
490 (void) vsprintf(buf
+ strlen(buf
), message
, args
);
493 (void) snprintf(buf
+ strlen(buf
), FATAL_MSG_SZ
- strlen(buf
),
494 ": %s", strerror(save_errno
));
496 (void) fprintf(stderr
, "%s\n", buf
);
497 fatal_msg
= buf
; /* to ease debugging */
504 str2shift(const char *buf
)
506 const char *ends
= "BKMGTPEZ";
511 for (i
= 0; i
< strlen(ends
); i
++) {
512 if (toupper(buf
[0]) == ends
[i
])
515 if (i
== strlen(ends
)) {
516 (void) fprintf(stderr
, "ztest: invalid bytes suffix: %s\n",
520 if (buf
[1] == '\0' || (toupper(buf
[1]) == 'B' && buf
[2] == '\0')) {
523 (void) fprintf(stderr
, "ztest: invalid bytes suffix: %s\n", buf
);
529 nicenumtoull(const char *buf
)
534 val
= strtoull(buf
, &end
, 0);
536 (void) fprintf(stderr
, "ztest: bad numeric value: %s\n", buf
);
538 } else if (end
[0] == '.') {
539 double fval
= strtod(buf
, &end
);
540 fval
*= pow(2, str2shift(end
));
541 if (fval
> UINT64_MAX
) {
542 (void) fprintf(stderr
, "ztest: value too large: %s\n",
546 val
= (uint64_t)fval
;
548 int shift
= str2shift(end
);
549 if (shift
>= 64 || (val
<< shift
) >> shift
!= val
) {
550 (void) fprintf(stderr
, "ztest: value too large: %s\n",
560 usage(boolean_t requested
)
562 const ztest_shared_opts_t
*zo
= &ztest_opts_defaults
;
564 char nice_vdev_size
[NN_NUMBUF_SZ
];
565 char nice_gang_bang
[NN_NUMBUF_SZ
];
566 FILE *fp
= requested
? stdout
: stderr
;
568 nicenum(zo
->zo_vdev_size
, nice_vdev_size
, sizeof (nice_vdev_size
));
569 nicenum(zo
->zo_metaslab_gang_bang
, nice_gang_bang
,
570 sizeof (nice_gang_bang
));
572 (void) fprintf(fp
, "Usage: %s\n"
573 "\t[-v vdevs (default: %llu)]\n"
574 "\t[-s size_of_each_vdev (default: %s)]\n"
575 "\t[-a alignment_shift (default: %d)] use 0 for random\n"
576 "\t[-m mirror_copies (default: %d)]\n"
577 "\t[-r raidz_disks (default: %d)]\n"
578 "\t[-R raidz_parity (default: %d)]\n"
579 "\t[-d datasets (default: %d)]\n"
580 "\t[-t threads (default: %d)]\n"
581 "\t[-g gang_block_threshold (default: %s)]\n"
582 "\t[-i init_count (default: %d)] initialize pool i times\n"
583 "\t[-k kill_percentage (default: %llu%%)]\n"
584 "\t[-p pool_name (default: %s)]\n"
585 "\t[-f dir (default: %s)] file directory for vdev files\n"
586 "\t[-V] verbose (use multiple times for ever more blather)\n"
587 "\t[-E] use existing pool instead of creating new one\n"
588 "\t[-T time (default: %llu sec)] total run time\n"
589 "\t[-F freezeloops (default: %llu)] max loops in spa_freeze()\n"
590 "\t[-P passtime (default: %llu sec)] time per pass\n"
591 "\t[-B alt_ztest (default: <none>)] alternate ztest path\n"
592 "\t[-o variable=value] ... set global variable to an unsigned\n"
593 "\t 32-bit integer value\n"
594 "\t[-h] (print help)\n"
597 (u_longlong_t
)zo
->zo_vdevs
, /* -v */
598 nice_vdev_size
, /* -s */
599 zo
->zo_ashift
, /* -a */
600 zo
->zo_mirrors
, /* -m */
601 zo
->zo_raidz
, /* -r */
602 zo
->zo_raidz_parity
, /* -R */
603 zo
->zo_datasets
, /* -d */
604 zo
->zo_threads
, /* -t */
605 nice_gang_bang
, /* -g */
606 zo
->zo_init
, /* -i */
607 (u_longlong_t
)zo
->zo_killrate
, /* -k */
608 zo
->zo_pool
, /* -p */
610 (u_longlong_t
)zo
->zo_time
, /* -T */
611 (u_longlong_t
)zo
->zo_maxloops
, /* -F */
612 (u_longlong_t
)zo
->zo_passtime
);
613 exit(requested
? 0 : 1);
617 process_options(int argc
, char **argv
)
620 ztest_shared_opts_t
*zo
= &ztest_opts
;
624 char altdir
[MAXNAMELEN
] = { 0 };
626 bcopy(&ztest_opts_defaults
, zo
, sizeof (*zo
));
628 while ((opt
= getopt(argc
, argv
,
629 "v:s:a:m:r:R:d:t:g:i:k:p:f:VET:P:hF:B:o:")) != EOF
) {
646 value
= nicenumtoull(optarg
);
650 zo
->zo_vdevs
= value
;
653 zo
->zo_vdev_size
= MAX(SPA_MINDEVSIZE
, value
);
656 zo
->zo_ashift
= value
;
659 zo
->zo_mirrors
= value
;
662 zo
->zo_raidz
= MAX(1, value
);
665 zo
->zo_raidz_parity
= MIN(MAX(value
, 1), 3);
668 zo
->zo_datasets
= MAX(1, value
);
671 zo
->zo_threads
= MAX(1, value
);
674 zo
->zo_metaslab_gang_bang
= MAX(SPA_MINBLOCKSIZE
<< 1,
681 zo
->zo_killrate
= value
;
684 (void) strlcpy(zo
->zo_pool
, optarg
,
685 sizeof (zo
->zo_pool
));
688 path
= realpath(optarg
, NULL
);
690 (void) fprintf(stderr
, "error: %s: %s\n",
691 optarg
, strerror(errno
));
694 (void) strlcpy(zo
->zo_dir
, path
,
695 sizeof (zo
->zo_dir
));
708 zo
->zo_passtime
= MAX(1, value
);
711 zo
->zo_maxloops
= MAX(1, value
);
714 (void) strlcpy(altdir
, optarg
, sizeof (altdir
));
717 if (set_global_var(optarg
) != 0)
730 zo
->zo_raidz_parity
= MIN(zo
->zo_raidz_parity
, zo
->zo_raidz
- 1);
733 (zo
->zo_vdevs
> 0 ? zo
->zo_time
* NANOSEC
/ zo
->zo_vdevs
:
736 if (strlen(altdir
) > 0) {
744 cmd
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
745 realaltdir
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
747 VERIFY(NULL
!= realpath(getexecname(), cmd
));
748 if (0 != access(altdir
, F_OK
)) {
749 ztest_dump_core
= B_FALSE
;
750 fatal(B_TRUE
, "invalid alternate ztest path: %s",
753 VERIFY(NULL
!= realpath(altdir
, realaltdir
));
756 * 'cmd' should be of the form "<anything>/usr/bin/<isa>/ztest".
757 * We want to extract <isa> to determine if we should use
758 * 32 or 64 bit binaries.
760 bin
= strstr(cmd
, "/usr/bin/");
761 ztest
= strstr(bin
, "/ztest");
763 isalen
= ztest
- isa
;
764 (void) snprintf(zo
->zo_alt_ztest
, sizeof (zo
->zo_alt_ztest
),
765 "%s/usr/bin/%.*s/ztest", realaltdir
, isalen
, isa
);
766 (void) snprintf(zo
->zo_alt_libpath
, sizeof (zo
->zo_alt_libpath
),
767 "%s/usr/lib/%.*s", realaltdir
, isalen
, isa
);
769 if (0 != access(zo
->zo_alt_ztest
, X_OK
)) {
770 ztest_dump_core
= B_FALSE
;
771 fatal(B_TRUE
, "invalid alternate ztest: %s",
773 } else if (0 != access(zo
->zo_alt_libpath
, X_OK
)) {
774 ztest_dump_core
= B_FALSE
;
775 fatal(B_TRUE
, "invalid alternate lib directory %s",
779 umem_free(cmd
, MAXPATHLEN
);
780 umem_free(realaltdir
, MAXPATHLEN
);
785 ztest_kill(ztest_shared_t
*zs
)
787 zs
->zs_alloc
= metaslab_class_get_alloc(spa_normal_class(ztest_spa
));
788 zs
->zs_space
= metaslab_class_get_space(spa_normal_class(ztest_spa
));
791 * Before we kill off ztest, make sure that the config is updated.
792 * See comment above spa_write_cachefile().
794 mutex_enter(&spa_namespace_lock
);
795 spa_write_cachefile(ztest_spa
, B_FALSE
, B_FALSE
);
796 mutex_exit(&spa_namespace_lock
);
798 zfs_dbgmsg_print(FTAG
);
799 (void) kill(getpid(), SIGKILL
);
803 ztest_random(uint64_t range
)
807 ASSERT3S(ztest_fd_rand
, >=, 0);
812 if (read(ztest_fd_rand
, &r
, sizeof (r
)) != sizeof (r
))
813 fatal(1, "short read from /dev/urandom");
820 ztest_record_enospc(const char *s
)
822 ztest_shared
->zs_enospc_count
++;
826 ztest_get_ashift(void)
828 if (ztest_opts
.zo_ashift
== 0)
829 return (SPA_MINBLOCKSHIFT
+ ztest_random(5));
830 return (ztest_opts
.zo_ashift
);
834 make_vdev_file(char *path
, char *aux
, char *pool
, size_t size
, uint64_t ashift
)
836 char pathbuf
[MAXPATHLEN
];
841 ashift
= ztest_get_ashift();
847 vdev
= ztest_shared
->zs_vdev_aux
;
848 (void) snprintf(path
, sizeof (pathbuf
),
849 ztest_aux_template
, ztest_opts
.zo_dir
,
850 pool
== NULL
? ztest_opts
.zo_pool
: pool
,
853 vdev
= ztest_shared
->zs_vdev_next_leaf
++;
854 (void) snprintf(path
, sizeof (pathbuf
),
855 ztest_dev_template
, ztest_opts
.zo_dir
,
856 pool
== NULL
? ztest_opts
.zo_pool
: pool
, vdev
);
861 int fd
= open(path
, O_RDWR
| O_CREAT
| O_TRUNC
, 0666);
863 fatal(1, "can't open %s", path
);
864 if (ftruncate(fd
, size
) != 0)
865 fatal(1, "can't ftruncate %s", path
);
869 VERIFY(nvlist_alloc(&file
, NV_UNIQUE_NAME
, 0) == 0);
870 VERIFY(nvlist_add_string(file
, ZPOOL_CONFIG_TYPE
, VDEV_TYPE_FILE
) == 0);
871 VERIFY(nvlist_add_string(file
, ZPOOL_CONFIG_PATH
, path
) == 0);
872 VERIFY(nvlist_add_uint64(file
, ZPOOL_CONFIG_ASHIFT
, ashift
) == 0);
878 make_vdev_raidz(char *path
, char *aux
, char *pool
, size_t size
,
879 uint64_t ashift
, int r
)
881 nvlist_t
*raidz
, **child
;
885 return (make_vdev_file(path
, aux
, pool
, size
, ashift
));
886 child
= umem_alloc(r
* sizeof (nvlist_t
*), UMEM_NOFAIL
);
888 for (c
= 0; c
< r
; c
++)
889 child
[c
] = make_vdev_file(path
, aux
, pool
, size
, ashift
);
891 VERIFY(nvlist_alloc(&raidz
, NV_UNIQUE_NAME
, 0) == 0);
892 VERIFY(nvlist_add_string(raidz
, ZPOOL_CONFIG_TYPE
,
893 VDEV_TYPE_RAIDZ
) == 0);
894 VERIFY(nvlist_add_uint64(raidz
, ZPOOL_CONFIG_NPARITY
,
895 ztest_opts
.zo_raidz_parity
) == 0);
896 VERIFY(nvlist_add_nvlist_array(raidz
, ZPOOL_CONFIG_CHILDREN
,
899 for (c
= 0; c
< r
; c
++)
900 nvlist_free(child
[c
]);
902 umem_free(child
, r
* sizeof (nvlist_t
*));
908 make_vdev_mirror(char *path
, char *aux
, char *pool
, size_t size
,
909 uint64_t ashift
, int r
, int m
)
911 nvlist_t
*mirror
, **child
;
915 return (make_vdev_raidz(path
, aux
, pool
, size
, ashift
, r
));
917 child
= umem_alloc(m
* sizeof (nvlist_t
*), UMEM_NOFAIL
);
919 for (c
= 0; c
< m
; c
++)
920 child
[c
] = make_vdev_raidz(path
, aux
, pool
, size
, ashift
, r
);
922 VERIFY(nvlist_alloc(&mirror
, NV_UNIQUE_NAME
, 0) == 0);
923 VERIFY(nvlist_add_string(mirror
, ZPOOL_CONFIG_TYPE
,
924 VDEV_TYPE_MIRROR
) == 0);
925 VERIFY(nvlist_add_nvlist_array(mirror
, ZPOOL_CONFIG_CHILDREN
,
928 for (c
= 0; c
< m
; c
++)
929 nvlist_free(child
[c
]);
931 umem_free(child
, m
* sizeof (nvlist_t
*));
937 make_vdev_root(char *path
, char *aux
, char *pool
, size_t size
, uint64_t ashift
,
938 int log
, int r
, int m
, int t
)
940 nvlist_t
*root
, **child
;
945 child
= umem_alloc(t
* sizeof (nvlist_t
*), UMEM_NOFAIL
);
947 for (c
= 0; c
< t
; c
++) {
948 child
[c
] = make_vdev_mirror(path
, aux
, pool
, size
, ashift
,
950 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_IS_LOG
,
954 VERIFY(nvlist_alloc(&root
, NV_UNIQUE_NAME
, 0) == 0);
955 VERIFY(nvlist_add_string(root
, ZPOOL_CONFIG_TYPE
, VDEV_TYPE_ROOT
) == 0);
956 VERIFY(nvlist_add_nvlist_array(root
, aux
? aux
: ZPOOL_CONFIG_CHILDREN
,
959 for (c
= 0; c
< t
; c
++)
960 nvlist_free(child
[c
]);
962 umem_free(child
, t
* sizeof (nvlist_t
*));
968 * Find a random spa version. Returns back a random spa version in the
969 * range [initial_version, SPA_VERSION_FEATURES].
972 ztest_random_spa_version(uint64_t initial_version
)
974 uint64_t version
= initial_version
;
976 if (version
<= SPA_VERSION_BEFORE_FEATURES
) {
978 ztest_random(SPA_VERSION_BEFORE_FEATURES
- version
+ 1);
981 if (version
> SPA_VERSION_BEFORE_FEATURES
)
982 version
= SPA_VERSION_FEATURES
;
984 ASSERT(SPA_VERSION_IS_SUPPORTED(version
));
989 ztest_random_blocksize(void)
991 uint64_t block_shift
;
993 * Choose a block size >= the ashift.
994 * If the SPA supports new MAXBLOCKSIZE, test up to 1MB blocks.
996 int maxbs
= SPA_OLD_MAXBLOCKSHIFT
;
997 if (spa_maxblocksize(ztest_spa
) == SPA_MAXBLOCKSIZE
)
999 block_shift
= ztest_random(maxbs
- ztest_spa
->spa_max_ashift
+ 1);
1000 return (1 << (SPA_MINBLOCKSHIFT
+ block_shift
));
1004 ztest_random_ibshift(void)
1006 return (DN_MIN_INDBLKSHIFT
+
1007 ztest_random(DN_MAX_INDBLKSHIFT
- DN_MIN_INDBLKSHIFT
+ 1));
1011 ztest_random_vdev_top(spa_t
*spa
, boolean_t log_ok
)
1014 vdev_t
*rvd
= spa
->spa_root_vdev
;
1017 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_READER
) != 0);
1020 top
= ztest_random(rvd
->vdev_children
);
1021 tvd
= rvd
->vdev_child
[top
];
1022 } while (!vdev_is_concrete(tvd
) || (tvd
->vdev_islog
&& !log_ok
) ||
1023 tvd
->vdev_mg
== NULL
|| tvd
->vdev_mg
->mg_class
== NULL
);
1029 ztest_random_dsl_prop(zfs_prop_t prop
)
1034 value
= zfs_prop_random_value(prop
, ztest_random(-1ULL));
1035 } while (prop
== ZFS_PROP_CHECKSUM
&& value
== ZIO_CHECKSUM_OFF
);
1041 ztest_dsl_prop_set_uint64(char *osname
, zfs_prop_t prop
, uint64_t value
,
1044 const char *propname
= zfs_prop_to_name(prop
);
1045 const char *valname
;
1046 char setpoint
[MAXPATHLEN
];
1050 error
= dsl_prop_set_int(osname
, propname
,
1051 (inherit
? ZPROP_SRC_NONE
: ZPROP_SRC_LOCAL
), value
);
1053 if (error
== ENOSPC
) {
1054 ztest_record_enospc(FTAG
);
1059 VERIFY0(dsl_prop_get_integer(osname
, propname
, &curval
, setpoint
));
1061 if (ztest_opts
.zo_verbose
>= 6) {
1062 VERIFY(zfs_prop_index_to_string(prop
, curval
, &valname
) == 0);
1063 (void) printf("%s %s = %s at '%s'\n",
1064 osname
, propname
, valname
, setpoint
);
1071 ztest_spa_prop_set_uint64(zpool_prop_t prop
, uint64_t value
)
1073 spa_t
*spa
= ztest_spa
;
1074 nvlist_t
*props
= NULL
;
1077 VERIFY(nvlist_alloc(&props
, NV_UNIQUE_NAME
, 0) == 0);
1078 VERIFY(nvlist_add_uint64(props
, zpool_prop_to_name(prop
), value
) == 0);
1080 error
= spa_prop_set(spa
, props
);
1084 if (error
== ENOSPC
) {
1085 ztest_record_enospc(FTAG
);
1094 ztest_rll_init(rll_t
*rll
)
1096 rll
->rll_writer
= NULL
;
1097 rll
->rll_readers
= 0;
1098 mutex_init(&rll
->rll_lock
, NULL
, USYNC_THREAD
, NULL
);
1099 cv_init(&rll
->rll_cv
, NULL
, USYNC_THREAD
, NULL
);
1103 ztest_rll_destroy(rll_t
*rll
)
1105 ASSERT(rll
->rll_writer
== NULL
);
1106 ASSERT(rll
->rll_readers
== 0);
1107 mutex_destroy(&rll
->rll_lock
);
1108 cv_destroy(&rll
->rll_cv
);
1112 ztest_rll_lock(rll_t
*rll
, rl_type_t type
)
1114 mutex_enter(&rll
->rll_lock
);
1116 if (type
== RL_READER
) {
1117 while (rll
->rll_writer
!= NULL
)
1118 cv_wait(&rll
->rll_cv
, &rll
->rll_lock
);
1121 while (rll
->rll_writer
!= NULL
|| rll
->rll_readers
)
1122 cv_wait(&rll
->rll_cv
, &rll
->rll_lock
);
1123 rll
->rll_writer
= curthread
;
1126 mutex_exit(&rll
->rll_lock
);
1130 ztest_rll_unlock(rll_t
*rll
)
1132 mutex_enter(&rll
->rll_lock
);
1134 if (rll
->rll_writer
) {
1135 ASSERT(rll
->rll_readers
== 0);
1136 rll
->rll_writer
= NULL
;
1138 ASSERT(rll
->rll_readers
!= 0);
1139 ASSERT(rll
->rll_writer
== NULL
);
1143 if (rll
->rll_writer
== NULL
&& rll
->rll_readers
== 0)
1144 cv_broadcast(&rll
->rll_cv
);
1146 mutex_exit(&rll
->rll_lock
);
1150 ztest_object_lock(ztest_ds_t
*zd
, uint64_t object
, rl_type_t type
)
1152 rll_t
*rll
= &zd
->zd_object_lock
[object
& (ZTEST_OBJECT_LOCKS
- 1)];
1154 ztest_rll_lock(rll
, type
);
1158 ztest_object_unlock(ztest_ds_t
*zd
, uint64_t object
)
1160 rll_t
*rll
= &zd
->zd_object_lock
[object
& (ZTEST_OBJECT_LOCKS
- 1)];
1162 ztest_rll_unlock(rll
);
1166 ztest_range_lock(ztest_ds_t
*zd
, uint64_t object
, uint64_t offset
,
1167 uint64_t size
, rl_type_t type
)
1169 uint64_t hash
= object
^ (offset
% (ZTEST_RANGE_LOCKS
+ 1));
1170 rll_t
*rll
= &zd
->zd_range_lock
[hash
& (ZTEST_RANGE_LOCKS
- 1)];
1173 rl
= umem_alloc(sizeof (*rl
), UMEM_NOFAIL
);
1174 rl
->rl_object
= object
;
1175 rl
->rl_offset
= offset
;
1179 ztest_rll_lock(rll
, type
);
1185 ztest_range_unlock(rl_t
*rl
)
1187 rll_t
*rll
= rl
->rl_lock
;
1189 ztest_rll_unlock(rll
);
1191 umem_free(rl
, sizeof (*rl
));
1195 ztest_zd_init(ztest_ds_t
*zd
, ztest_shared_ds_t
*szd
, objset_t
*os
)
1198 zd
->zd_zilog
= dmu_objset_zil(os
);
1199 zd
->zd_shared
= szd
;
1200 dmu_objset_name(os
, zd
->zd_name
);
1202 if (zd
->zd_shared
!= NULL
)
1203 zd
->zd_shared
->zd_seq
= 0;
1205 rw_init(&zd
->zd_zilog_lock
, NULL
, USYNC_THREAD
, NULL
);
1206 mutex_init(&zd
->zd_dirobj_lock
, NULL
, USYNC_THREAD
, NULL
);
1208 for (int l
= 0; l
< ZTEST_OBJECT_LOCKS
; l
++)
1209 ztest_rll_init(&zd
->zd_object_lock
[l
]);
1211 for (int l
= 0; l
< ZTEST_RANGE_LOCKS
; l
++)
1212 ztest_rll_init(&zd
->zd_range_lock
[l
]);
1216 ztest_zd_fini(ztest_ds_t
*zd
)
1218 mutex_destroy(&zd
->zd_dirobj_lock
);
1220 for (int l
= 0; l
< ZTEST_OBJECT_LOCKS
; l
++)
1221 ztest_rll_destroy(&zd
->zd_object_lock
[l
]);
1223 for (int l
= 0; l
< ZTEST_RANGE_LOCKS
; l
++)
1224 ztest_rll_destroy(&zd
->zd_range_lock
[l
]);
1227 #define TXG_MIGHTWAIT (ztest_random(10) == 0 ? TXG_NOWAIT : TXG_WAIT)
1230 ztest_tx_assign(dmu_tx_t
*tx
, uint64_t txg_how
, const char *tag
)
1236 * Attempt to assign tx to some transaction group.
1238 error
= dmu_tx_assign(tx
, txg_how
);
1240 if (error
== ERESTART
) {
1241 ASSERT(txg_how
== TXG_NOWAIT
);
1244 ASSERT3U(error
, ==, ENOSPC
);
1245 ztest_record_enospc(tag
);
1250 txg
= dmu_tx_get_txg(tx
);
1256 ztest_pattern_set(void *buf
, uint64_t size
, uint64_t value
)
1259 uint64_t *ip_end
= (uint64_t *)((uintptr_t)buf
+ (uintptr_t)size
);
1266 ztest_pattern_match(void *buf
, uint64_t size
, uint64_t value
)
1269 uint64_t *ip_end
= (uint64_t *)((uintptr_t)buf
+ (uintptr_t)size
);
1273 diff
|= (value
- *ip
++);
1279 ztest_bt_generate(ztest_block_tag_t
*bt
, objset_t
*os
, uint64_t object
,
1280 uint64_t offset
, uint64_t gen
, uint64_t txg
, uint64_t crtxg
)
1282 bt
->bt_magic
= BT_MAGIC
;
1283 bt
->bt_objset
= dmu_objset_id(os
);
1284 bt
->bt_object
= object
;
1285 bt
->bt_offset
= offset
;
1288 bt
->bt_crtxg
= crtxg
;
1292 ztest_bt_verify(ztest_block_tag_t
*bt
, objset_t
*os
, uint64_t object
,
1293 uint64_t offset
, uint64_t gen
, uint64_t txg
, uint64_t crtxg
)
1295 ASSERT3U(bt
->bt_magic
, ==, BT_MAGIC
);
1296 ASSERT3U(bt
->bt_objset
, ==, dmu_objset_id(os
));
1297 ASSERT3U(bt
->bt_object
, ==, object
);
1298 ASSERT3U(bt
->bt_offset
, ==, offset
);
1299 ASSERT3U(bt
->bt_gen
, <=, gen
);
1300 ASSERT3U(bt
->bt_txg
, <=, txg
);
1301 ASSERT3U(bt
->bt_crtxg
, ==, crtxg
);
1304 static ztest_block_tag_t
*
1305 ztest_bt_bonus(dmu_buf_t
*db
)
1307 dmu_object_info_t doi
;
1308 ztest_block_tag_t
*bt
;
1310 dmu_object_info_from_db(db
, &doi
);
1311 ASSERT3U(doi
.doi_bonus_size
, <=, db
->db_size
);
1312 ASSERT3U(doi
.doi_bonus_size
, >=, sizeof (*bt
));
1313 bt
= (void *)((char *)db
->db_data
+ doi
.doi_bonus_size
- sizeof (*bt
));
1322 #define lrz_type lr_mode
1323 #define lrz_blocksize lr_uid
1324 #define lrz_ibshift lr_gid
1325 #define lrz_bonustype lr_rdev
1326 #define lrz_bonuslen lr_crtime[1]
1329 ztest_log_create(ztest_ds_t
*zd
, dmu_tx_t
*tx
, lr_create_t
*lr
)
1331 char *name
= (void *)(lr
+ 1); /* name follows lr */
1332 size_t namesize
= strlen(name
) + 1;
1335 if (zil_replaying(zd
->zd_zilog
, tx
))
1338 itx
= zil_itx_create(TX_CREATE
, sizeof (*lr
) + namesize
);
1339 bcopy(&lr
->lr_common
+ 1, &itx
->itx_lr
+ 1,
1340 sizeof (*lr
) + namesize
- sizeof (lr_t
));
1342 zil_itx_assign(zd
->zd_zilog
, itx
, tx
);
1346 ztest_log_remove(ztest_ds_t
*zd
, dmu_tx_t
*tx
, lr_remove_t
*lr
, uint64_t object
)
1348 char *name
= (void *)(lr
+ 1); /* name follows lr */
1349 size_t namesize
= strlen(name
) + 1;
1352 if (zil_replaying(zd
->zd_zilog
, tx
))
1355 itx
= zil_itx_create(TX_REMOVE
, sizeof (*lr
) + namesize
);
1356 bcopy(&lr
->lr_common
+ 1, &itx
->itx_lr
+ 1,
1357 sizeof (*lr
) + namesize
- sizeof (lr_t
));
1359 itx
->itx_oid
= object
;
1360 zil_itx_assign(zd
->zd_zilog
, itx
, tx
);
1364 ztest_log_write(ztest_ds_t
*zd
, dmu_tx_t
*tx
, lr_write_t
*lr
)
1367 itx_wr_state_t write_state
= ztest_random(WR_NUM_STATES
);
1369 if (zil_replaying(zd
->zd_zilog
, tx
))
1372 if (lr
->lr_length
> ZIL_MAX_LOG_DATA
)
1373 write_state
= WR_INDIRECT
;
1375 itx
= zil_itx_create(TX_WRITE
,
1376 sizeof (*lr
) + (write_state
== WR_COPIED
? lr
->lr_length
: 0));
1378 if (write_state
== WR_COPIED
&&
1379 dmu_read(zd
->zd_os
, lr
->lr_foid
, lr
->lr_offset
, lr
->lr_length
,
1380 ((lr_write_t
*)&itx
->itx_lr
) + 1, DMU_READ_NO_PREFETCH
) != 0) {
1381 zil_itx_destroy(itx
);
1382 itx
= zil_itx_create(TX_WRITE
, sizeof (*lr
));
1383 write_state
= WR_NEED_COPY
;
1385 itx
->itx_private
= zd
;
1386 itx
->itx_wr_state
= write_state
;
1387 itx
->itx_sync
= (ztest_random(8) == 0);
1389 bcopy(&lr
->lr_common
+ 1, &itx
->itx_lr
+ 1,
1390 sizeof (*lr
) - sizeof (lr_t
));
1392 zil_itx_assign(zd
->zd_zilog
, itx
, tx
);
1396 ztest_log_truncate(ztest_ds_t
*zd
, dmu_tx_t
*tx
, lr_truncate_t
*lr
)
1400 if (zil_replaying(zd
->zd_zilog
, tx
))
1403 itx
= zil_itx_create(TX_TRUNCATE
, sizeof (*lr
));
1404 bcopy(&lr
->lr_common
+ 1, &itx
->itx_lr
+ 1,
1405 sizeof (*lr
) - sizeof (lr_t
));
1407 itx
->itx_sync
= B_FALSE
;
1408 zil_itx_assign(zd
->zd_zilog
, itx
, tx
);
1412 ztest_log_setattr(ztest_ds_t
*zd
, dmu_tx_t
*tx
, lr_setattr_t
*lr
)
1416 if (zil_replaying(zd
->zd_zilog
, tx
))
1419 itx
= zil_itx_create(TX_SETATTR
, sizeof (*lr
));
1420 bcopy(&lr
->lr_common
+ 1, &itx
->itx_lr
+ 1,
1421 sizeof (*lr
) - sizeof (lr_t
));
1423 itx
->itx_sync
= B_FALSE
;
1424 zil_itx_assign(zd
->zd_zilog
, itx
, tx
);
1431 ztest_replay_create(void *arg1
, void *arg2
, boolean_t byteswap
)
1433 ztest_ds_t
*zd
= arg1
;
1434 lr_create_t
*lr
= arg2
;
1435 char *name
= (void *)(lr
+ 1); /* name follows lr */
1436 objset_t
*os
= zd
->zd_os
;
1437 ztest_block_tag_t
*bbt
;
1444 byteswap_uint64_array(lr
, sizeof (*lr
));
1446 ASSERT(lr
->lr_doid
== ZTEST_DIROBJ
);
1447 ASSERT(name
[0] != '\0');
1449 tx
= dmu_tx_create(os
);
1451 dmu_tx_hold_zap(tx
, lr
->lr_doid
, B_TRUE
, name
);
1453 if (lr
->lrz_type
== DMU_OT_ZAP_OTHER
) {
1454 dmu_tx_hold_zap(tx
, DMU_NEW_OBJECT
, B_TRUE
, NULL
);
1456 dmu_tx_hold_bonus(tx
, DMU_NEW_OBJECT
);
1459 txg
= ztest_tx_assign(tx
, TXG_WAIT
, FTAG
);
1463 ASSERT(dmu_objset_zil(os
)->zl_replay
== !!lr
->lr_foid
);
1465 if (lr
->lrz_type
== DMU_OT_ZAP_OTHER
) {
1466 if (lr
->lr_foid
== 0) {
1467 lr
->lr_foid
= zap_create(os
,
1468 lr
->lrz_type
, lr
->lrz_bonustype
,
1469 lr
->lrz_bonuslen
, tx
);
1471 error
= zap_create_claim(os
, lr
->lr_foid
,
1472 lr
->lrz_type
, lr
->lrz_bonustype
,
1473 lr
->lrz_bonuslen
, tx
);
1476 if (lr
->lr_foid
== 0) {
1477 lr
->lr_foid
= dmu_object_alloc(os
,
1478 lr
->lrz_type
, 0, lr
->lrz_bonustype
,
1479 lr
->lrz_bonuslen
, tx
);
1481 error
= dmu_object_claim(os
, lr
->lr_foid
,
1482 lr
->lrz_type
, 0, lr
->lrz_bonustype
,
1483 lr
->lrz_bonuslen
, tx
);
1488 ASSERT3U(error
, ==, EEXIST
);
1489 ASSERT(zd
->zd_zilog
->zl_replay
);
1494 ASSERT(lr
->lr_foid
!= 0);
1496 if (lr
->lrz_type
!= DMU_OT_ZAP_OTHER
)
1497 VERIFY3U(0, ==, dmu_object_set_blocksize(os
, lr
->lr_foid
,
1498 lr
->lrz_blocksize
, lr
->lrz_ibshift
, tx
));
1500 VERIFY3U(0, ==, dmu_bonus_hold(os
, lr
->lr_foid
, FTAG
, &db
));
1501 bbt
= ztest_bt_bonus(db
);
1502 dmu_buf_will_dirty(db
, tx
);
1503 ztest_bt_generate(bbt
, os
, lr
->lr_foid
, -1ULL, lr
->lr_gen
, txg
, txg
);
1504 dmu_buf_rele(db
, FTAG
);
1506 VERIFY3U(0, ==, zap_add(os
, lr
->lr_doid
, name
, sizeof (uint64_t), 1,
1509 (void) ztest_log_create(zd
, tx
, lr
);
1517 ztest_replay_remove(void *arg1
, void *arg2
, boolean_t byteswap
)
1519 ztest_ds_t
*zd
= arg1
;
1520 lr_remove_t
*lr
= arg2
;
1521 char *name
= (void *)(lr
+ 1); /* name follows lr */
1522 objset_t
*os
= zd
->zd_os
;
1523 dmu_object_info_t doi
;
1525 uint64_t object
, txg
;
1528 byteswap_uint64_array(lr
, sizeof (*lr
));
1530 ASSERT(lr
->lr_doid
== ZTEST_DIROBJ
);
1531 ASSERT(name
[0] != '\0');
1534 zap_lookup(os
, lr
->lr_doid
, name
, sizeof (object
), 1, &object
));
1535 ASSERT(object
!= 0);
1537 ztest_object_lock(zd
, object
, RL_WRITER
);
1539 VERIFY3U(0, ==, dmu_object_info(os
, object
, &doi
));
1541 tx
= dmu_tx_create(os
);
1543 dmu_tx_hold_zap(tx
, lr
->lr_doid
, B_FALSE
, name
);
1544 dmu_tx_hold_free(tx
, object
, 0, DMU_OBJECT_END
);
1546 txg
= ztest_tx_assign(tx
, TXG_WAIT
, FTAG
);
1548 ztest_object_unlock(zd
, object
);
1552 if (doi
.doi_type
== DMU_OT_ZAP_OTHER
) {
1553 VERIFY3U(0, ==, zap_destroy(os
, object
, tx
));
1555 VERIFY3U(0, ==, dmu_object_free(os
, object
, tx
));
1558 VERIFY3U(0, ==, zap_remove(os
, lr
->lr_doid
, name
, tx
));
1560 (void) ztest_log_remove(zd
, tx
, lr
, object
);
1564 ztest_object_unlock(zd
, object
);
1570 ztest_replay_write(void *arg1
, void *arg2
, boolean_t byteswap
)
1572 ztest_ds_t
*zd
= arg1
;
1573 lr_write_t
*lr
= arg2
;
1574 objset_t
*os
= zd
->zd_os
;
1575 void *data
= lr
+ 1; /* data follows lr */
1576 uint64_t offset
, length
;
1577 ztest_block_tag_t
*bt
= data
;
1578 ztest_block_tag_t
*bbt
;
1579 uint64_t gen
, txg
, lrtxg
, crtxg
;
1580 dmu_object_info_t doi
;
1583 arc_buf_t
*abuf
= NULL
;
1587 byteswap_uint64_array(lr
, sizeof (*lr
));
1589 offset
= lr
->lr_offset
;
1590 length
= lr
->lr_length
;
1592 /* If it's a dmu_sync() block, write the whole block */
1593 if (lr
->lr_common
.lrc_reclen
== sizeof (lr_write_t
)) {
1594 uint64_t blocksize
= BP_GET_LSIZE(&lr
->lr_blkptr
);
1595 if (length
< blocksize
) {
1596 offset
-= offset
% blocksize
;
1601 if (bt
->bt_magic
== BSWAP_64(BT_MAGIC
))
1602 byteswap_uint64_array(bt
, sizeof (*bt
));
1604 if (bt
->bt_magic
!= BT_MAGIC
)
1607 ztest_object_lock(zd
, lr
->lr_foid
, RL_READER
);
1608 rl
= ztest_range_lock(zd
, lr
->lr_foid
, offset
, length
, RL_WRITER
);
1610 VERIFY3U(0, ==, dmu_bonus_hold(os
, lr
->lr_foid
, FTAG
, &db
));
1612 dmu_object_info_from_db(db
, &doi
);
1614 bbt
= ztest_bt_bonus(db
);
1615 ASSERT3U(bbt
->bt_magic
, ==, BT_MAGIC
);
1617 crtxg
= bbt
->bt_crtxg
;
1618 lrtxg
= lr
->lr_common
.lrc_txg
;
1620 tx
= dmu_tx_create(os
);
1622 dmu_tx_hold_write(tx
, lr
->lr_foid
, offset
, length
);
1624 if (ztest_random(8) == 0 && length
== doi
.doi_data_block_size
&&
1625 P2PHASE(offset
, length
) == 0)
1626 abuf
= dmu_request_arcbuf(db
, length
);
1628 txg
= ztest_tx_assign(tx
, TXG_WAIT
, FTAG
);
1631 dmu_return_arcbuf(abuf
);
1632 dmu_buf_rele(db
, FTAG
);
1633 ztest_range_unlock(rl
);
1634 ztest_object_unlock(zd
, lr
->lr_foid
);
1640 * Usually, verify the old data before writing new data --
1641 * but not always, because we also want to verify correct
1642 * behavior when the data was not recently read into cache.
1644 ASSERT(offset
% doi
.doi_data_block_size
== 0);
1645 if (ztest_random(4) != 0) {
1646 int prefetch
= ztest_random(2) ?
1647 DMU_READ_PREFETCH
: DMU_READ_NO_PREFETCH
;
1648 ztest_block_tag_t rbt
;
1650 VERIFY(dmu_read(os
, lr
->lr_foid
, offset
,
1651 sizeof (rbt
), &rbt
, prefetch
) == 0);
1652 if (rbt
.bt_magic
== BT_MAGIC
) {
1653 ztest_bt_verify(&rbt
, os
, lr
->lr_foid
,
1654 offset
, gen
, txg
, crtxg
);
1659 * Writes can appear to be newer than the bonus buffer because
1660 * the ztest_get_data() callback does a dmu_read() of the
1661 * open-context data, which may be different than the data
1662 * as it was when the write was generated.
1664 if (zd
->zd_zilog
->zl_replay
) {
1665 ztest_bt_verify(bt
, os
, lr
->lr_foid
, offset
,
1666 MAX(gen
, bt
->bt_gen
), MAX(txg
, lrtxg
),
1671 * Set the bt's gen/txg to the bonus buffer's gen/txg
1672 * so that all of the usual ASSERTs will work.
1674 ztest_bt_generate(bt
, os
, lr
->lr_foid
, offset
, gen
, txg
, crtxg
);
1678 dmu_write(os
, lr
->lr_foid
, offset
, length
, data
, tx
);
1680 bcopy(data
, abuf
->b_data
, length
);
1681 dmu_assign_arcbuf(db
, offset
, abuf
, tx
);
1684 (void) ztest_log_write(zd
, tx
, lr
);
1686 dmu_buf_rele(db
, FTAG
);
1690 ztest_range_unlock(rl
);
1691 ztest_object_unlock(zd
, lr
->lr_foid
);
1697 ztest_replay_truncate(void *arg1
, void *arg2
, boolean_t byteswap
)
1699 ztest_ds_t
*zd
= arg1
;
1700 lr_truncate_t
*lr
= arg2
;
1701 objset_t
*os
= zd
->zd_os
;
1707 byteswap_uint64_array(lr
, sizeof (*lr
));
1709 ztest_object_lock(zd
, lr
->lr_foid
, RL_READER
);
1710 rl
= ztest_range_lock(zd
, lr
->lr_foid
, lr
->lr_offset
, lr
->lr_length
,
1713 tx
= dmu_tx_create(os
);
1715 dmu_tx_hold_free(tx
, lr
->lr_foid
, lr
->lr_offset
, lr
->lr_length
);
1717 txg
= ztest_tx_assign(tx
, TXG_WAIT
, FTAG
);
1719 ztest_range_unlock(rl
);
1720 ztest_object_unlock(zd
, lr
->lr_foid
);
1724 VERIFY(dmu_free_range(os
, lr
->lr_foid
, lr
->lr_offset
,
1725 lr
->lr_length
, tx
) == 0);
1727 (void) ztest_log_truncate(zd
, tx
, lr
);
1731 ztest_range_unlock(rl
);
1732 ztest_object_unlock(zd
, lr
->lr_foid
);
1738 ztest_replay_setattr(void *arg1
, void *arg2
, boolean_t byteswap
)
1740 ztest_ds_t
*zd
= arg1
;
1741 lr_setattr_t
*lr
= arg2
;
1742 objset_t
*os
= zd
->zd_os
;
1745 ztest_block_tag_t
*bbt
;
1746 uint64_t txg
, lrtxg
, crtxg
;
1749 byteswap_uint64_array(lr
, sizeof (*lr
));
1751 ztest_object_lock(zd
, lr
->lr_foid
, RL_WRITER
);
1753 VERIFY3U(0, ==, dmu_bonus_hold(os
, lr
->lr_foid
, FTAG
, &db
));
1755 tx
= dmu_tx_create(os
);
1756 dmu_tx_hold_bonus(tx
, lr
->lr_foid
);
1758 txg
= ztest_tx_assign(tx
, TXG_WAIT
, FTAG
);
1760 dmu_buf_rele(db
, FTAG
);
1761 ztest_object_unlock(zd
, lr
->lr_foid
);
1765 bbt
= ztest_bt_bonus(db
);
1766 ASSERT3U(bbt
->bt_magic
, ==, BT_MAGIC
);
1767 crtxg
= bbt
->bt_crtxg
;
1768 lrtxg
= lr
->lr_common
.lrc_txg
;
1770 if (zd
->zd_zilog
->zl_replay
) {
1771 ASSERT(lr
->lr_size
!= 0);
1772 ASSERT(lr
->lr_mode
!= 0);
1776 * Randomly change the size and increment the generation.
1778 lr
->lr_size
= (ztest_random(db
->db_size
/ sizeof (*bbt
)) + 1) *
1780 lr
->lr_mode
= bbt
->bt_gen
+ 1;
1785 * Verify that the current bonus buffer is not newer than our txg.
1787 ztest_bt_verify(bbt
, os
, lr
->lr_foid
, -1ULL, lr
->lr_mode
,
1788 MAX(txg
, lrtxg
), crtxg
);
1790 dmu_buf_will_dirty(db
, tx
);
1792 ASSERT3U(lr
->lr_size
, >=, sizeof (*bbt
));
1793 ASSERT3U(lr
->lr_size
, <=, db
->db_size
);
1794 VERIFY0(dmu_set_bonus(db
, lr
->lr_size
, tx
));
1795 bbt
= ztest_bt_bonus(db
);
1797 ztest_bt_generate(bbt
, os
, lr
->lr_foid
, -1ULL, lr
->lr_mode
, txg
, crtxg
);
1799 dmu_buf_rele(db
, FTAG
);
1801 (void) ztest_log_setattr(zd
, tx
, lr
);
1805 ztest_object_unlock(zd
, lr
->lr_foid
);
1810 zil_replay_func_t
*ztest_replay_vector
[TX_MAX_TYPE
] = {
1811 NULL
, /* 0 no such transaction type */
1812 ztest_replay_create
, /* TX_CREATE */
1813 NULL
, /* TX_MKDIR */
1814 NULL
, /* TX_MKXATTR */
1815 NULL
, /* TX_SYMLINK */
1816 ztest_replay_remove
, /* TX_REMOVE */
1817 NULL
, /* TX_RMDIR */
1819 NULL
, /* TX_RENAME */
1820 ztest_replay_write
, /* TX_WRITE */
1821 ztest_replay_truncate
, /* TX_TRUNCATE */
1822 ztest_replay_setattr
, /* TX_SETATTR */
1824 NULL
, /* TX_CREATE_ACL */
1825 NULL
, /* TX_CREATE_ATTR */
1826 NULL
, /* TX_CREATE_ACL_ATTR */
1827 NULL
, /* TX_MKDIR_ACL */
1828 NULL
, /* TX_MKDIR_ATTR */
1829 NULL
, /* TX_MKDIR_ACL_ATTR */
1830 NULL
, /* TX_WRITE2 */
1834 * ZIL get_data callbacks
1838 ztest_get_done(zgd_t
*zgd
, int error
)
1840 ztest_ds_t
*zd
= zgd
->zgd_private
;
1841 uint64_t object
= zgd
->zgd_rl
->rl_object
;
1844 dmu_buf_rele(zgd
->zgd_db
, zgd
);
1846 ztest_range_unlock(zgd
->zgd_rl
);
1847 ztest_object_unlock(zd
, object
);
1849 if (error
== 0 && zgd
->zgd_bp
)
1850 zil_lwb_add_block(zgd
->zgd_lwb
, zgd
->zgd_bp
);
1852 umem_free(zgd
, sizeof (*zgd
));
1856 ztest_get_data(void *arg
, lr_write_t
*lr
, char *buf
, struct lwb
*lwb
,
1859 ztest_ds_t
*zd
= arg
;
1860 objset_t
*os
= zd
->zd_os
;
1861 uint64_t object
= lr
->lr_foid
;
1862 uint64_t offset
= lr
->lr_offset
;
1863 uint64_t size
= lr
->lr_length
;
1864 uint64_t txg
= lr
->lr_common
.lrc_txg
;
1866 dmu_object_info_t doi
;
1871 ASSERT3P(lwb
, !=, NULL
);
1872 ASSERT3P(zio
, !=, NULL
);
1873 ASSERT3U(size
, !=, 0);
1875 ztest_object_lock(zd
, object
, RL_READER
);
1876 error
= dmu_bonus_hold(os
, object
, FTAG
, &db
);
1878 ztest_object_unlock(zd
, object
);
1882 crtxg
= ztest_bt_bonus(db
)->bt_crtxg
;
1884 if (crtxg
== 0 || crtxg
> txg
) {
1885 dmu_buf_rele(db
, FTAG
);
1886 ztest_object_unlock(zd
, object
);
1890 dmu_object_info_from_db(db
, &doi
);
1891 dmu_buf_rele(db
, FTAG
);
1894 zgd
= umem_zalloc(sizeof (*zgd
), UMEM_NOFAIL
);
1896 zgd
->zgd_private
= zd
;
1898 if (buf
!= NULL
) { /* immediate write */
1899 zgd
->zgd_rl
= ztest_range_lock(zd
, object
, offset
, size
,
1902 error
= dmu_read(os
, object
, offset
, size
, buf
,
1903 DMU_READ_NO_PREFETCH
);
1906 size
= doi
.doi_data_block_size
;
1908 offset
= P2ALIGN(offset
, size
);
1910 ASSERT(offset
< size
);
1914 zgd
->zgd_rl
= ztest_range_lock(zd
, object
, offset
, size
,
1917 error
= dmu_buf_hold(os
, object
, offset
, zgd
, &db
,
1918 DMU_READ_NO_PREFETCH
);
1921 blkptr_t
*bp
= &lr
->lr_blkptr
;
1926 ASSERT(db
->db_offset
== offset
);
1927 ASSERT(db
->db_size
== size
);
1929 error
= dmu_sync(zio
, lr
->lr_common
.lrc_txg
,
1930 ztest_get_done
, zgd
);
1937 ztest_get_done(zgd
, error
);
1943 ztest_lr_alloc(size_t lrsize
, char *name
)
1946 size_t namesize
= name
? strlen(name
) + 1 : 0;
1948 lr
= umem_zalloc(lrsize
+ namesize
, UMEM_NOFAIL
);
1951 bcopy(name
, lr
+ lrsize
, namesize
);
1957 ztest_lr_free(void *lr
, size_t lrsize
, char *name
)
1959 size_t namesize
= name
? strlen(name
) + 1 : 0;
1961 umem_free(lr
, lrsize
+ namesize
);
1965 * Lookup a bunch of objects. Returns the number of objects not found.
1968 ztest_lookup(ztest_ds_t
*zd
, ztest_od_t
*od
, int count
)
1973 ASSERT(MUTEX_HELD(&zd
->zd_dirobj_lock
));
1975 for (int i
= 0; i
< count
; i
++, od
++) {
1977 error
= zap_lookup(zd
->zd_os
, od
->od_dir
, od
->od_name
,
1978 sizeof (uint64_t), 1, &od
->od_object
);
1980 ASSERT(error
== ENOENT
);
1981 ASSERT(od
->od_object
== 0);
1985 ztest_block_tag_t
*bbt
;
1986 dmu_object_info_t doi
;
1988 ASSERT(od
->od_object
!= 0);
1989 ASSERT(missing
== 0); /* there should be no gaps */
1991 ztest_object_lock(zd
, od
->od_object
, RL_READER
);
1992 VERIFY3U(0, ==, dmu_bonus_hold(zd
->zd_os
,
1993 od
->od_object
, FTAG
, &db
));
1994 dmu_object_info_from_db(db
, &doi
);
1995 bbt
= ztest_bt_bonus(db
);
1996 ASSERT3U(bbt
->bt_magic
, ==, BT_MAGIC
);
1997 od
->od_type
= doi
.doi_type
;
1998 od
->od_blocksize
= doi
.doi_data_block_size
;
1999 od
->od_gen
= bbt
->bt_gen
;
2000 dmu_buf_rele(db
, FTAG
);
2001 ztest_object_unlock(zd
, od
->od_object
);
2009 ztest_create(ztest_ds_t
*zd
, ztest_od_t
*od
, int count
)
2013 ASSERT(MUTEX_HELD(&zd
->zd_dirobj_lock
));
2015 for (int i
= 0; i
< count
; i
++, od
++) {
2022 lr_create_t
*lr
= ztest_lr_alloc(sizeof (*lr
), od
->od_name
);
2024 lr
->lr_doid
= od
->od_dir
;
2025 lr
->lr_foid
= 0; /* 0 to allocate, > 0 to claim */
2026 lr
->lrz_type
= od
->od_crtype
;
2027 lr
->lrz_blocksize
= od
->od_crblocksize
;
2028 lr
->lrz_ibshift
= ztest_random_ibshift();
2029 lr
->lrz_bonustype
= DMU_OT_UINT64_OTHER
;
2030 lr
->lrz_bonuslen
= dmu_bonus_max();
2031 lr
->lr_gen
= od
->od_crgen
;
2032 lr
->lr_crtime
[0] = time(NULL
);
2034 if (ztest_replay_create(zd
, lr
, B_FALSE
) != 0) {
2035 ASSERT(missing
== 0);
2039 od
->od_object
= lr
->lr_foid
;
2040 od
->od_type
= od
->od_crtype
;
2041 od
->od_blocksize
= od
->od_crblocksize
;
2042 od
->od_gen
= od
->od_crgen
;
2043 ASSERT(od
->od_object
!= 0);
2046 ztest_lr_free(lr
, sizeof (*lr
), od
->od_name
);
2053 ztest_remove(ztest_ds_t
*zd
, ztest_od_t
*od
, int count
)
2058 ASSERT(MUTEX_HELD(&zd
->zd_dirobj_lock
));
2062 for (int i
= count
- 1; i
>= 0; i
--, od
--) {
2069 * No object was found.
2071 if (od
->od_object
== 0)
2074 lr_remove_t
*lr
= ztest_lr_alloc(sizeof (*lr
), od
->od_name
);
2076 lr
->lr_doid
= od
->od_dir
;
2078 if ((error
= ztest_replay_remove(zd
, lr
, B_FALSE
)) != 0) {
2079 ASSERT3U(error
, ==, ENOSPC
);
2084 ztest_lr_free(lr
, sizeof (*lr
), od
->od_name
);
2091 ztest_write(ztest_ds_t
*zd
, uint64_t object
, uint64_t offset
, uint64_t size
,
2097 lr
= ztest_lr_alloc(sizeof (*lr
) + size
, NULL
);
2099 lr
->lr_foid
= object
;
2100 lr
->lr_offset
= offset
;
2101 lr
->lr_length
= size
;
2103 BP_ZERO(&lr
->lr_blkptr
);
2105 bcopy(data
, lr
+ 1, size
);
2107 error
= ztest_replay_write(zd
, lr
, B_FALSE
);
2109 ztest_lr_free(lr
, sizeof (*lr
) + size
, NULL
);
2115 ztest_truncate(ztest_ds_t
*zd
, uint64_t object
, uint64_t offset
, uint64_t size
)
2120 lr
= ztest_lr_alloc(sizeof (*lr
), NULL
);
2122 lr
->lr_foid
= object
;
2123 lr
->lr_offset
= offset
;
2124 lr
->lr_length
= size
;
2126 error
= ztest_replay_truncate(zd
, lr
, B_FALSE
);
2128 ztest_lr_free(lr
, sizeof (*lr
), NULL
);
2134 ztest_setattr(ztest_ds_t
*zd
, uint64_t object
)
2139 lr
= ztest_lr_alloc(sizeof (*lr
), NULL
);
2141 lr
->lr_foid
= object
;
2145 error
= ztest_replay_setattr(zd
, lr
, B_FALSE
);
2147 ztest_lr_free(lr
, sizeof (*lr
), NULL
);
2153 ztest_prealloc(ztest_ds_t
*zd
, uint64_t object
, uint64_t offset
, uint64_t size
)
2155 objset_t
*os
= zd
->zd_os
;
2160 txg_wait_synced(dmu_objset_pool(os
), 0);
2162 ztest_object_lock(zd
, object
, RL_READER
);
2163 rl
= ztest_range_lock(zd
, object
, offset
, size
, RL_WRITER
);
2165 tx
= dmu_tx_create(os
);
2167 dmu_tx_hold_write(tx
, object
, offset
, size
);
2169 txg
= ztest_tx_assign(tx
, TXG_WAIT
, FTAG
);
2172 dmu_prealloc(os
, object
, offset
, size
, tx
);
2174 txg_wait_synced(dmu_objset_pool(os
), txg
);
2176 (void) dmu_free_long_range(os
, object
, offset
, size
);
2179 ztest_range_unlock(rl
);
2180 ztest_object_unlock(zd
, object
);
2184 ztest_io(ztest_ds_t
*zd
, uint64_t object
, uint64_t offset
)
2187 ztest_block_tag_t wbt
;
2188 dmu_object_info_t doi
;
2189 enum ztest_io_type io_type
;
2193 VERIFY(dmu_object_info(zd
->zd_os
, object
, &doi
) == 0);
2194 blocksize
= doi
.doi_data_block_size
;
2195 data
= umem_alloc(blocksize
, UMEM_NOFAIL
);
2198 * Pick an i/o type at random, biased toward writing block tags.
2200 io_type
= ztest_random(ZTEST_IO_TYPES
);
2201 if (ztest_random(2) == 0)
2202 io_type
= ZTEST_IO_WRITE_TAG
;
2204 rw_enter(&zd
->zd_zilog_lock
, RW_READER
);
2208 case ZTEST_IO_WRITE_TAG
:
2209 ztest_bt_generate(&wbt
, zd
->zd_os
, object
, offset
, 0, 0, 0);
2210 (void) ztest_write(zd
, object
, offset
, sizeof (wbt
), &wbt
);
2213 case ZTEST_IO_WRITE_PATTERN
:
2214 (void) memset(data
, 'a' + (object
+ offset
) % 5, blocksize
);
2215 if (ztest_random(2) == 0) {
2217 * Induce fletcher2 collisions to ensure that
2218 * zio_ddt_collision() detects and resolves them
2219 * when using fletcher2-verify for deduplication.
2221 ((uint64_t *)data
)[0] ^= 1ULL << 63;
2222 ((uint64_t *)data
)[4] ^= 1ULL << 63;
2224 (void) ztest_write(zd
, object
, offset
, blocksize
, data
);
2227 case ZTEST_IO_WRITE_ZEROES
:
2228 bzero(data
, blocksize
);
2229 (void) ztest_write(zd
, object
, offset
, blocksize
, data
);
2232 case ZTEST_IO_TRUNCATE
:
2233 (void) ztest_truncate(zd
, object
, offset
, blocksize
);
2236 case ZTEST_IO_SETATTR
:
2237 (void) ztest_setattr(zd
, object
);
2240 case ZTEST_IO_REWRITE
:
2241 rw_enter(&ztest_name_lock
, RW_READER
);
2242 err
= ztest_dsl_prop_set_uint64(zd
->zd_name
,
2243 ZFS_PROP_CHECKSUM
, spa_dedup_checksum(ztest_spa
),
2245 VERIFY(err
== 0 || err
== ENOSPC
);
2246 err
= ztest_dsl_prop_set_uint64(zd
->zd_name
,
2247 ZFS_PROP_COMPRESSION
,
2248 ztest_random_dsl_prop(ZFS_PROP_COMPRESSION
),
2250 VERIFY(err
== 0 || err
== ENOSPC
);
2251 rw_exit(&ztest_name_lock
);
2253 VERIFY0(dmu_read(zd
->zd_os
, object
, offset
, blocksize
, data
,
2254 DMU_READ_NO_PREFETCH
));
2256 (void) ztest_write(zd
, object
, offset
, blocksize
, data
);
2260 rw_exit(&zd
->zd_zilog_lock
);
2262 umem_free(data
, blocksize
);
2266 * Initialize an object description template.
2269 ztest_od_init(ztest_od_t
*od
, uint64_t id
, char *tag
, uint64_t index
,
2270 dmu_object_type_t type
, uint64_t blocksize
, uint64_t gen
)
2272 od
->od_dir
= ZTEST_DIROBJ
;
2275 od
->od_crtype
= type
;
2276 od
->od_crblocksize
= blocksize
? blocksize
: ztest_random_blocksize();
2279 od
->od_type
= DMU_OT_NONE
;
2280 od
->od_blocksize
= 0;
2283 (void) snprintf(od
->od_name
, sizeof (od
->od_name
), "%s(%lld)[%llu]",
2284 tag
, (int64_t)id
, index
);
2288 * Lookup or create the objects for a test using the od template.
2289 * If the objects do not all exist, or if 'remove' is specified,
2290 * remove any existing objects and create new ones. Otherwise,
2291 * use the existing objects.
2294 ztest_object_init(ztest_ds_t
*zd
, ztest_od_t
*od
, size_t size
, boolean_t remove
)
2296 int count
= size
/ sizeof (*od
);
2299 mutex_enter(&zd
->zd_dirobj_lock
);
2300 if ((ztest_lookup(zd
, od
, count
) != 0 || remove
) &&
2301 (ztest_remove(zd
, od
, count
) != 0 ||
2302 ztest_create(zd
, od
, count
) != 0))
2305 mutex_exit(&zd
->zd_dirobj_lock
);
2312 ztest_zil_commit(ztest_ds_t
*zd
, uint64_t id
)
2314 zilog_t
*zilog
= zd
->zd_zilog
;
2316 rw_enter(&zd
->zd_zilog_lock
, RW_READER
);
2318 zil_commit(zilog
, ztest_random(ZTEST_OBJECTS
));
2321 * Remember the committed values in zd, which is in parent/child
2322 * shared memory. If we die, the next iteration of ztest_run()
2323 * will verify that the log really does contain this record.
2325 mutex_enter(&zilog
->zl_lock
);
2326 ASSERT(zd
->zd_shared
!= NULL
);
2327 ASSERT3U(zd
->zd_shared
->zd_seq
, <=, zilog
->zl_commit_lr_seq
);
2328 zd
->zd_shared
->zd_seq
= zilog
->zl_commit_lr_seq
;
2329 mutex_exit(&zilog
->zl_lock
);
2331 rw_exit(&zd
->zd_zilog_lock
);
2335 * This function is designed to simulate the operations that occur during a
2336 * mount/unmount operation. We hold the dataset across these operations in an
2337 * attempt to expose any implicit assumptions about ZIL management.
2341 ztest_zil_remount(ztest_ds_t
*zd
, uint64_t id
)
2343 objset_t
*os
= zd
->zd_os
;
2346 * We grab the zd_dirobj_lock to ensure that no other thread is
2347 * updating the zil (i.e. adding in-memory log records) and the
2348 * zd_zilog_lock to block any I/O.
2350 mutex_enter(&zd
->zd_dirobj_lock
);
2351 rw_enter(&zd
->zd_zilog_lock
, RW_WRITER
);
2353 /* zfsvfs_teardown() */
2354 zil_close(zd
->zd_zilog
);
2356 /* zfsvfs_setup() */
2357 VERIFY(zil_open(os
, ztest_get_data
) == zd
->zd_zilog
);
2358 zil_replay(os
, zd
, ztest_replay_vector
);
2360 rw_exit(&zd
->zd_zilog_lock
);
2361 mutex_exit(&zd
->zd_dirobj_lock
);
2365 * Verify that we can't destroy an active pool, create an existing pool,
2366 * or create a pool with a bad vdev spec.
2370 ztest_spa_create_destroy(ztest_ds_t
*zd
, uint64_t id
)
2372 ztest_shared_opts_t
*zo
= &ztest_opts
;
2377 * Attempt to create using a bad file.
2379 nvroot
= make_vdev_root("/dev/bogus", NULL
, NULL
, 0, 0, 0, 0, 0, 1);
2380 VERIFY3U(ENOENT
, ==,
2381 spa_create("ztest_bad_file", nvroot
, NULL
, NULL
));
2382 nvlist_free(nvroot
);
2385 * Attempt to create using a bad mirror.
2387 nvroot
= make_vdev_root("/dev/bogus", NULL
, NULL
, 0, 0, 0, 0, 2, 1);
2388 VERIFY3U(ENOENT
, ==,
2389 spa_create("ztest_bad_mirror", nvroot
, NULL
, NULL
));
2390 nvlist_free(nvroot
);
2393 * Attempt to create an existing pool. It shouldn't matter
2394 * what's in the nvroot; we should fail with EEXIST.
2396 rw_enter(&ztest_name_lock
, RW_READER
);
2397 nvroot
= make_vdev_root("/dev/bogus", NULL
, NULL
, 0, 0, 0, 0, 0, 1);
2398 VERIFY3U(EEXIST
, ==, spa_create(zo
->zo_pool
, nvroot
, NULL
, NULL
));
2399 nvlist_free(nvroot
);
2400 VERIFY3U(0, ==, spa_open(zo
->zo_pool
, &spa
, FTAG
));
2401 VERIFY3U(EBUSY
, ==, spa_destroy(zo
->zo_pool
));
2402 spa_close(spa
, FTAG
);
2404 rw_exit(&ztest_name_lock
);
2409 ztest_spa_upgrade(ztest_ds_t
*zd
, uint64_t id
)
2412 uint64_t initial_version
= SPA_VERSION_INITIAL
;
2413 uint64_t version
, newversion
;
2414 nvlist_t
*nvroot
, *props
;
2417 mutex_enter(&ztest_vdev_lock
);
2418 name
= kmem_asprintf("%s_upgrade", ztest_opts
.zo_pool
);
2421 * Clean up from previous runs.
2423 (void) spa_destroy(name
);
2425 nvroot
= make_vdev_root(NULL
, NULL
, name
, ztest_opts
.zo_vdev_size
, 0,
2426 0, ztest_opts
.zo_raidz
, ztest_opts
.zo_mirrors
, 1);
2429 * If we're configuring a RAIDZ device then make sure that the
2430 * the initial version is capable of supporting that feature.
2432 switch (ztest_opts
.zo_raidz_parity
) {
2435 initial_version
= SPA_VERSION_INITIAL
;
2438 initial_version
= SPA_VERSION_RAIDZ2
;
2441 initial_version
= SPA_VERSION_RAIDZ3
;
2446 * Create a pool with a spa version that can be upgraded. Pick
2447 * a value between initial_version and SPA_VERSION_BEFORE_FEATURES.
2450 version
= ztest_random_spa_version(initial_version
);
2451 } while (version
> SPA_VERSION_BEFORE_FEATURES
);
2453 props
= fnvlist_alloc();
2454 fnvlist_add_uint64(props
,
2455 zpool_prop_to_name(ZPOOL_PROP_VERSION
), version
);
2456 VERIFY0(spa_create(name
, nvroot
, props
, NULL
));
2457 fnvlist_free(nvroot
);
2458 fnvlist_free(props
);
2460 VERIFY0(spa_open(name
, &spa
, FTAG
));
2461 VERIFY3U(spa_version(spa
), ==, version
);
2462 newversion
= ztest_random_spa_version(version
+ 1);
2464 if (ztest_opts
.zo_verbose
>= 4) {
2465 (void) printf("upgrading spa version from %llu to %llu\n",
2466 (u_longlong_t
)version
, (u_longlong_t
)newversion
);
2469 spa_upgrade(spa
, newversion
);
2470 VERIFY3U(spa_version(spa
), >, version
);
2471 VERIFY3U(spa_version(spa
), ==, fnvlist_lookup_uint64(spa
->spa_config
,
2472 zpool_prop_to_name(ZPOOL_PROP_VERSION
)));
2473 spa_close(spa
, FTAG
);
2476 mutex_exit(&ztest_vdev_lock
);
2480 vdev_lookup_by_path(vdev_t
*vd
, const char *path
)
2484 if (vd
->vdev_path
!= NULL
&& strcmp(path
, vd
->vdev_path
) == 0)
2487 for (int c
= 0; c
< vd
->vdev_children
; c
++)
2488 if ((mvd
= vdev_lookup_by_path(vd
->vdev_child
[c
], path
)) !=
2496 * Find the first available hole which can be used as a top-level.
2499 find_vdev_hole(spa_t
*spa
)
2501 vdev_t
*rvd
= spa
->spa_root_vdev
;
2504 ASSERT(spa_config_held(spa
, SCL_VDEV
, RW_READER
) == SCL_VDEV
);
2506 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
2507 vdev_t
*cvd
= rvd
->vdev_child
[c
];
2509 if (cvd
->vdev_ishole
)
2516 * Verify that vdev_add() works as expected.
2520 ztest_vdev_add_remove(ztest_ds_t
*zd
, uint64_t id
)
2522 ztest_shared_t
*zs
= ztest_shared
;
2523 spa_t
*spa
= ztest_spa
;
2529 mutex_enter(&ztest_vdev_lock
);
2530 leaves
= MAX(zs
->zs_mirrors
+ zs
->zs_splits
, 1) * ztest_opts
.zo_raidz
;
2532 spa_config_enter(spa
, SCL_VDEV
, FTAG
, RW_READER
);
2534 ztest_shared
->zs_vdev_next_leaf
= find_vdev_hole(spa
) * leaves
;
2537 * If we have slogs then remove them 1/4 of the time.
2539 if (spa_has_slogs(spa
) && ztest_random(4) == 0) {
2541 * Grab the guid from the head of the log class rotor.
2543 guid
= spa_log_class(spa
)->mc_rotor
->mg_vd
->vdev_guid
;
2545 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
2548 * We have to grab the zs_name_lock as writer to
2549 * prevent a race between removing a slog (dmu_objset_find)
2550 * and destroying a dataset. Removing the slog will
2551 * grab a reference on the dataset which may cause
2552 * dmu_objset_destroy() to fail with EBUSY thus
2553 * leaving the dataset in an inconsistent state.
2555 rw_enter(&ztest_name_lock
, RW_WRITER
);
2556 error
= spa_vdev_remove(spa
, guid
, B_FALSE
);
2557 rw_exit(&ztest_name_lock
);
2559 if (error
&& error
!= EEXIST
)
2560 fatal(0, "spa_vdev_remove() = %d", error
);
2562 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
2565 * Make 1/4 of the devices be log devices.
2567 nvroot
= make_vdev_root(NULL
, NULL
, NULL
,
2568 ztest_opts
.zo_vdev_size
, 0,
2569 ztest_random(4) == 0, ztest_opts
.zo_raidz
,
2572 error
= spa_vdev_add(spa
, nvroot
);
2573 nvlist_free(nvroot
);
2575 if (error
== ENOSPC
)
2576 ztest_record_enospc("spa_vdev_add");
2577 else if (error
!= 0)
2578 fatal(0, "spa_vdev_add() = %d", error
);
2581 mutex_exit(&ztest_vdev_lock
);
2585 * Verify that adding/removing aux devices (l2arc, hot spare) works as expected.
2589 ztest_vdev_aux_add_remove(ztest_ds_t
*zd
, uint64_t id
)
2591 ztest_shared_t
*zs
= ztest_shared
;
2592 spa_t
*spa
= ztest_spa
;
2593 vdev_t
*rvd
= spa
->spa_root_vdev
;
2594 spa_aux_vdev_t
*sav
;
2599 if (ztest_random(2) == 0) {
2600 sav
= &spa
->spa_spares
;
2601 aux
= ZPOOL_CONFIG_SPARES
;
2603 sav
= &spa
->spa_l2cache
;
2604 aux
= ZPOOL_CONFIG_L2CACHE
;
2607 mutex_enter(&ztest_vdev_lock
);
2609 spa_config_enter(spa
, SCL_VDEV
, FTAG
, RW_READER
);
2611 if (sav
->sav_count
!= 0 && ztest_random(4) == 0) {
2613 * Pick a random device to remove.
2615 guid
= sav
->sav_vdevs
[ztest_random(sav
->sav_count
)]->vdev_guid
;
2618 * Find an unused device we can add.
2620 zs
->zs_vdev_aux
= 0;
2622 char path
[MAXPATHLEN
];
2624 (void) snprintf(path
, sizeof (path
), ztest_aux_template
,
2625 ztest_opts
.zo_dir
, ztest_opts
.zo_pool
, aux
,
2627 for (c
= 0; c
< sav
->sav_count
; c
++)
2628 if (strcmp(sav
->sav_vdevs
[c
]->vdev_path
,
2631 if (c
== sav
->sav_count
&&
2632 vdev_lookup_by_path(rvd
, path
) == NULL
)
2638 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
2644 nvlist_t
*nvroot
= make_vdev_root(NULL
, aux
, NULL
,
2645 (ztest_opts
.zo_vdev_size
* 5) / 4, 0, 0, 0, 0, 1);
2646 error
= spa_vdev_add(spa
, nvroot
);
2648 fatal(0, "spa_vdev_add(%p) = %d", nvroot
, error
);
2649 nvlist_free(nvroot
);
2652 * Remove an existing device. Sometimes, dirty its
2653 * vdev state first to make sure we handle removal
2654 * of devices that have pending state changes.
2656 if (ztest_random(2) == 0)
2657 (void) vdev_online(spa
, guid
, 0, NULL
);
2659 error
= spa_vdev_remove(spa
, guid
, B_FALSE
);
2660 if (error
!= 0 && error
!= EBUSY
)
2661 fatal(0, "spa_vdev_remove(%llu) = %d", guid
, error
);
2664 mutex_exit(&ztest_vdev_lock
);
2668 * split a pool if it has mirror tlvdevs
2672 ztest_split_pool(ztest_ds_t
*zd
, uint64_t id
)
2674 ztest_shared_t
*zs
= ztest_shared
;
2675 spa_t
*spa
= ztest_spa
;
2676 vdev_t
*rvd
= spa
->spa_root_vdev
;
2677 nvlist_t
*tree
, **child
, *config
, *split
, **schild
;
2678 uint_t c
, children
, schildren
= 0, lastlogid
= 0;
2681 mutex_enter(&ztest_vdev_lock
);
2683 /* ensure we have a useable config; mirrors of raidz aren't supported */
2684 if (zs
->zs_mirrors
< 3 || ztest_opts
.zo_raidz
> 1) {
2685 mutex_exit(&ztest_vdev_lock
);
2689 /* clean up the old pool, if any */
2690 (void) spa_destroy("splitp");
2692 spa_config_enter(spa
, SCL_VDEV
, FTAG
, RW_READER
);
2694 /* generate a config from the existing config */
2695 mutex_enter(&spa
->spa_props_lock
);
2696 VERIFY(nvlist_lookup_nvlist(spa
->spa_config
, ZPOOL_CONFIG_VDEV_TREE
,
2698 mutex_exit(&spa
->spa_props_lock
);
2700 VERIFY(nvlist_lookup_nvlist_array(tree
, ZPOOL_CONFIG_CHILDREN
, &child
,
2703 schild
= malloc(rvd
->vdev_children
* sizeof (nvlist_t
*));
2704 for (c
= 0; c
< children
; c
++) {
2705 vdev_t
*tvd
= rvd
->vdev_child
[c
];
2709 if (tvd
->vdev_islog
|| tvd
->vdev_ops
== &vdev_hole_ops
) {
2710 VERIFY(nvlist_alloc(&schild
[schildren
], NV_UNIQUE_NAME
,
2712 VERIFY(nvlist_add_string(schild
[schildren
],
2713 ZPOOL_CONFIG_TYPE
, VDEV_TYPE_HOLE
) == 0);
2714 VERIFY(nvlist_add_uint64(schild
[schildren
],
2715 ZPOOL_CONFIG_IS_HOLE
, 1) == 0);
2717 lastlogid
= schildren
;
2722 VERIFY(nvlist_lookup_nvlist_array(child
[c
],
2723 ZPOOL_CONFIG_CHILDREN
, &mchild
, &mchildren
) == 0);
2724 VERIFY(nvlist_dup(mchild
[0], &schild
[schildren
++], 0) == 0);
2727 /* OK, create a config that can be used to split */
2728 VERIFY(nvlist_alloc(&split
, NV_UNIQUE_NAME
, 0) == 0);
2729 VERIFY(nvlist_add_string(split
, ZPOOL_CONFIG_TYPE
,
2730 VDEV_TYPE_ROOT
) == 0);
2731 VERIFY(nvlist_add_nvlist_array(split
, ZPOOL_CONFIG_CHILDREN
, schild
,
2732 lastlogid
!= 0 ? lastlogid
: schildren
) == 0);
2734 VERIFY(nvlist_alloc(&config
, NV_UNIQUE_NAME
, 0) == 0);
2735 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, split
) == 0);
2737 for (c
= 0; c
< schildren
; c
++)
2738 nvlist_free(schild
[c
]);
2742 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
2744 rw_enter(&ztest_name_lock
, RW_WRITER
);
2745 error
= spa_vdev_split_mirror(spa
, "splitp", config
, NULL
, B_FALSE
);
2746 rw_exit(&ztest_name_lock
);
2748 nvlist_free(config
);
2751 (void) printf("successful split - results:\n");
2752 mutex_enter(&spa_namespace_lock
);
2753 show_pool_stats(spa
);
2754 show_pool_stats(spa_lookup("splitp"));
2755 mutex_exit(&spa_namespace_lock
);
2759 mutex_exit(&ztest_vdev_lock
);
2764 * Verify that we can attach and detach devices.
2768 ztest_vdev_attach_detach(ztest_ds_t
*zd
, uint64_t id
)
2770 ztest_shared_t
*zs
= ztest_shared
;
2771 spa_t
*spa
= ztest_spa
;
2772 spa_aux_vdev_t
*sav
= &spa
->spa_spares
;
2773 vdev_t
*rvd
= spa
->spa_root_vdev
;
2774 vdev_t
*oldvd
, *newvd
, *pvd
;
2778 uint64_t ashift
= ztest_get_ashift();
2779 uint64_t oldguid
, pguid
;
2780 uint64_t oldsize
, newsize
;
2781 char oldpath
[MAXPATHLEN
], newpath
[MAXPATHLEN
];
2783 int oldvd_has_siblings
= B_FALSE
;
2784 int newvd_is_spare
= B_FALSE
;
2786 int error
, expected_error
;
2788 mutex_enter(&ztest_vdev_lock
);
2789 leaves
= MAX(zs
->zs_mirrors
, 1) * ztest_opts
.zo_raidz
;
2791 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
2794 * If a vdev is in the process of being removed, its removal may
2795 * finish while we are in progress, leading to an unexpected error
2796 * value. Don't bother trying to attach while we are in the middle
2799 if (spa
->spa_vdev_removal
!= NULL
) {
2800 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2801 mutex_exit(&ztest_vdev_lock
);
2806 * Decide whether to do an attach or a replace.
2808 replacing
= ztest_random(2);
2811 * Pick a random top-level vdev.
2813 top
= ztest_random_vdev_top(spa
, B_TRUE
);
2816 * Pick a random leaf within it.
2818 leaf
= ztest_random(leaves
);
2823 oldvd
= rvd
->vdev_child
[top
];
2824 if (zs
->zs_mirrors
>= 1) {
2825 ASSERT(oldvd
->vdev_ops
== &vdev_mirror_ops
);
2826 ASSERT(oldvd
->vdev_children
>= zs
->zs_mirrors
);
2827 oldvd
= oldvd
->vdev_child
[leaf
/ ztest_opts
.zo_raidz
];
2829 if (ztest_opts
.zo_raidz
> 1) {
2830 ASSERT(oldvd
->vdev_ops
== &vdev_raidz_ops
);
2831 ASSERT(oldvd
->vdev_children
== ztest_opts
.zo_raidz
);
2832 oldvd
= oldvd
->vdev_child
[leaf
% ztest_opts
.zo_raidz
];
2836 * If we're already doing an attach or replace, oldvd may be a
2837 * mirror vdev -- in which case, pick a random child.
2839 while (oldvd
->vdev_children
!= 0) {
2840 oldvd_has_siblings
= B_TRUE
;
2841 ASSERT(oldvd
->vdev_children
>= 2);
2842 oldvd
= oldvd
->vdev_child
[ztest_random(oldvd
->vdev_children
)];
2845 oldguid
= oldvd
->vdev_guid
;
2846 oldsize
= vdev_get_min_asize(oldvd
);
2847 oldvd_is_log
= oldvd
->vdev_top
->vdev_islog
;
2848 (void) strcpy(oldpath
, oldvd
->vdev_path
);
2849 pvd
= oldvd
->vdev_parent
;
2850 pguid
= pvd
->vdev_guid
;
2853 * If oldvd has siblings, then half of the time, detach it.
2855 if (oldvd_has_siblings
&& ztest_random(2) == 0) {
2856 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2857 error
= spa_vdev_detach(spa
, oldguid
, pguid
, B_FALSE
);
2858 if (error
!= 0 && error
!= ENODEV
&& error
!= EBUSY
&&
2860 fatal(0, "detach (%s) returned %d", oldpath
, error
);
2861 mutex_exit(&ztest_vdev_lock
);
2866 * For the new vdev, choose with equal probability between the two
2867 * standard paths (ending in either 'a' or 'b') or a random hot spare.
2869 if (sav
->sav_count
!= 0 && ztest_random(3) == 0) {
2870 newvd
= sav
->sav_vdevs
[ztest_random(sav
->sav_count
)];
2871 newvd_is_spare
= B_TRUE
;
2872 (void) strcpy(newpath
, newvd
->vdev_path
);
2874 (void) snprintf(newpath
, sizeof (newpath
), ztest_dev_template
,
2875 ztest_opts
.zo_dir
, ztest_opts
.zo_pool
,
2876 top
* leaves
+ leaf
);
2877 if (ztest_random(2) == 0)
2878 newpath
[strlen(newpath
) - 1] = 'b';
2879 newvd
= vdev_lookup_by_path(rvd
, newpath
);
2884 * Reopen to ensure the vdev's asize field isn't stale.
2887 newsize
= vdev_get_min_asize(newvd
);
2890 * Make newsize a little bigger or smaller than oldsize.
2891 * If it's smaller, the attach should fail.
2892 * If it's larger, and we're doing a replace,
2893 * we should get dynamic LUN growth when we're done.
2895 newsize
= 10 * oldsize
/ (9 + ztest_random(3));
2899 * If pvd is not a mirror or root, the attach should fail with ENOTSUP,
2900 * unless it's a replace; in that case any non-replacing parent is OK.
2902 * If newvd is already part of the pool, it should fail with EBUSY.
2904 * If newvd is too small, it should fail with EOVERFLOW.
2906 if (pvd
->vdev_ops
!= &vdev_mirror_ops
&&
2907 pvd
->vdev_ops
!= &vdev_root_ops
&& (!replacing
||
2908 pvd
->vdev_ops
== &vdev_replacing_ops
||
2909 pvd
->vdev_ops
== &vdev_spare_ops
))
2910 expected_error
= ENOTSUP
;
2911 else if (newvd_is_spare
&& (!replacing
|| oldvd_is_log
))
2912 expected_error
= ENOTSUP
;
2913 else if (newvd
== oldvd
)
2914 expected_error
= replacing
? 0 : EBUSY
;
2915 else if (vdev_lookup_by_path(rvd
, newpath
) != NULL
)
2916 expected_error
= EBUSY
;
2917 else if (newsize
< oldsize
)
2918 expected_error
= EOVERFLOW
;
2919 else if (ashift
> oldvd
->vdev_top
->vdev_ashift
)
2920 expected_error
= EDOM
;
2924 spa_config_exit(spa
, SCL_ALL
, FTAG
);
2927 * Build the nvlist describing newpath.
2929 root
= make_vdev_root(newpath
, NULL
, NULL
, newvd
== NULL
? newsize
: 0,
2930 ashift
, 0, 0, 0, 1);
2932 error
= spa_vdev_attach(spa
, oldguid
, root
, replacing
);
2937 * If our parent was the replacing vdev, but the replace completed,
2938 * then instead of failing with ENOTSUP we may either succeed,
2939 * fail with ENODEV, or fail with EOVERFLOW.
2941 if (expected_error
== ENOTSUP
&&
2942 (error
== 0 || error
== ENODEV
|| error
== EOVERFLOW
))
2943 expected_error
= error
;
2946 * If someone grew the LUN, the replacement may be too small.
2948 if (error
== EOVERFLOW
|| error
== EBUSY
)
2949 expected_error
= error
;
2951 /* XXX workaround 6690467 */
2952 if (error
!= expected_error
&& expected_error
!= EBUSY
) {
2953 fatal(0, "attach (%s %llu, %s %llu, %d) "
2954 "returned %d, expected %d",
2955 oldpath
, oldsize
, newpath
,
2956 newsize
, replacing
, error
, expected_error
);
2959 mutex_exit(&ztest_vdev_lock
);
2964 ztest_device_removal(ztest_ds_t
*zd
, uint64_t id
)
2966 spa_t
*spa
= ztest_spa
;
2970 mutex_enter(&ztest_vdev_lock
);
2972 spa_config_enter(spa
, SCL_VDEV
, FTAG
, RW_READER
);
2973 vd
= vdev_lookup_top(spa
, ztest_random_vdev_top(spa
, B_FALSE
));
2974 guid
= vd
->vdev_guid
;
2975 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
2977 (void) spa_vdev_remove(spa
, guid
, B_FALSE
);
2979 mutex_exit(&ztest_vdev_lock
);
2983 * Callback function which expands the physical size of the vdev.
2986 grow_vdev(vdev_t
*vd
, void *arg
)
2988 spa_t
*spa
= vd
->vdev_spa
;
2989 size_t *newsize
= arg
;
2993 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_READER
) == SCL_STATE
);
2994 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
2996 if ((fd
= open(vd
->vdev_path
, O_RDWR
)) == -1)
2999 fsize
= lseek(fd
, 0, SEEK_END
);
3000 (void) ftruncate(fd
, *newsize
);
3002 if (ztest_opts
.zo_verbose
>= 6) {
3003 (void) printf("%s grew from %lu to %lu bytes\n",
3004 vd
->vdev_path
, (ulong_t
)fsize
, (ulong_t
)*newsize
);
3011 * Callback function which expands a given vdev by calling vdev_online().
3015 online_vdev(vdev_t
*vd
, void *arg
)
3017 spa_t
*spa
= vd
->vdev_spa
;
3018 vdev_t
*tvd
= vd
->vdev_top
;
3019 uint64_t guid
= vd
->vdev_guid
;
3020 uint64_t generation
= spa
->spa_config_generation
+ 1;
3021 vdev_state_t newstate
= VDEV_STATE_UNKNOWN
;
3024 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_READER
) == SCL_STATE
);
3025 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
3027 /* Calling vdev_online will initialize the new metaslabs */
3028 spa_config_exit(spa
, SCL_STATE
, spa
);
3029 error
= vdev_online(spa
, guid
, ZFS_ONLINE_EXPAND
, &newstate
);
3030 spa_config_enter(spa
, SCL_STATE
, spa
, RW_READER
);
3033 * If vdev_online returned an error or the underlying vdev_open
3034 * failed then we abort the expand. The only way to know that
3035 * vdev_open fails is by checking the returned newstate.
3037 if (error
|| newstate
!= VDEV_STATE_HEALTHY
) {
3038 if (ztest_opts
.zo_verbose
>= 5) {
3039 (void) printf("Unable to expand vdev, state %llu, "
3040 "error %d\n", (u_longlong_t
)newstate
, error
);
3044 ASSERT3U(newstate
, ==, VDEV_STATE_HEALTHY
);
3047 * Since we dropped the lock we need to ensure that we're
3048 * still talking to the original vdev. It's possible this
3049 * vdev may have been detached/replaced while we were
3050 * trying to online it.
3052 if (generation
!= spa
->spa_config_generation
) {
3053 if (ztest_opts
.zo_verbose
>= 5) {
3054 (void) printf("vdev configuration has changed, "
3055 "guid %llu, state %llu, expected gen %llu, "
3058 (u_longlong_t
)tvd
->vdev_state
,
3059 (u_longlong_t
)generation
,
3060 (u_longlong_t
)spa
->spa_config_generation
);
3068 * Traverse the vdev tree calling the supplied function.
3069 * We continue to walk the tree until we either have walked all
3070 * children or we receive a non-NULL return from the callback.
3071 * If a NULL callback is passed, then we just return back the first
3072 * leaf vdev we encounter.
3075 vdev_walk_tree(vdev_t
*vd
, vdev_t
*(*func
)(vdev_t
*, void *), void *arg
)
3077 if (vd
->vdev_ops
->vdev_op_leaf
) {
3081 return (func(vd
, arg
));
3084 for (uint_t c
= 0; c
< vd
->vdev_children
; c
++) {
3085 vdev_t
*cvd
= vd
->vdev_child
[c
];
3086 if ((cvd
= vdev_walk_tree(cvd
, func
, arg
)) != NULL
)
3093 * Verify that dynamic LUN growth works as expected.
3097 ztest_vdev_LUN_growth(ztest_ds_t
*zd
, uint64_t id
)
3099 spa_t
*spa
= ztest_spa
;
3101 metaslab_class_t
*mc
;
3102 metaslab_group_t
*mg
;
3103 size_t psize
, newsize
;
3105 uint64_t old_class_space
, new_class_space
, old_ms_count
, new_ms_count
;
3107 mutex_enter(&ztest_vdev_lock
);
3108 spa_config_enter(spa
, SCL_STATE
, spa
, RW_READER
);
3111 * If there is a vdev removal in progress, it could complete while
3112 * we are running, in which case we would not be able to verify
3113 * that the metaslab_class space increased (because it decreases
3114 * when the device removal completes).
3116 if (spa
->spa_vdev_removal
!= NULL
) {
3117 spa_config_exit(spa
, SCL_STATE
, FTAG
);
3118 mutex_exit(&ztest_vdev_lock
);
3122 top
= ztest_random_vdev_top(spa
, B_TRUE
);
3124 tvd
= spa
->spa_root_vdev
->vdev_child
[top
];
3127 old_ms_count
= tvd
->vdev_ms_count
;
3128 old_class_space
= metaslab_class_get_space(mc
);
3131 * Determine the size of the first leaf vdev associated with
3132 * our top-level device.
3134 vd
= vdev_walk_tree(tvd
, NULL
, NULL
);
3135 ASSERT3P(vd
, !=, NULL
);
3136 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
3138 psize
= vd
->vdev_psize
;
3141 * We only try to expand the vdev if it's healthy, less than 4x its
3142 * original size, and it has a valid psize.
3144 if (tvd
->vdev_state
!= VDEV_STATE_HEALTHY
||
3145 psize
== 0 || psize
>= 4 * ztest_opts
.zo_vdev_size
) {
3146 spa_config_exit(spa
, SCL_STATE
, spa
);
3147 mutex_exit(&ztest_vdev_lock
);
3151 newsize
= psize
+ psize
/ 8;
3152 ASSERT3U(newsize
, >, psize
);
3154 if (ztest_opts
.zo_verbose
>= 6) {
3155 (void) printf("Expanding LUN %s from %lu to %lu\n",
3156 vd
->vdev_path
, (ulong_t
)psize
, (ulong_t
)newsize
);
3160 * Growing the vdev is a two step process:
3161 * 1). expand the physical size (i.e. relabel)
3162 * 2). online the vdev to create the new metaslabs
3164 if (vdev_walk_tree(tvd
, grow_vdev
, &newsize
) != NULL
||
3165 vdev_walk_tree(tvd
, online_vdev
, NULL
) != NULL
||
3166 tvd
->vdev_state
!= VDEV_STATE_HEALTHY
) {
3167 if (ztest_opts
.zo_verbose
>= 5) {
3168 (void) printf("Could not expand LUN because "
3169 "the vdev configuration changed.\n");
3171 spa_config_exit(spa
, SCL_STATE
, spa
);
3172 mutex_exit(&ztest_vdev_lock
);
3176 spa_config_exit(spa
, SCL_STATE
, spa
);
3179 * Expanding the LUN will update the config asynchronously,
3180 * thus we must wait for the async thread to complete any
3181 * pending tasks before proceeding.
3185 mutex_enter(&spa
->spa_async_lock
);
3186 done
= (spa
->spa_async_thread
== NULL
&& !spa
->spa_async_tasks
);
3187 mutex_exit(&spa
->spa_async_lock
);
3190 txg_wait_synced(spa_get_dsl(spa
), 0);
3191 (void) poll(NULL
, 0, 100);
3194 spa_config_enter(spa
, SCL_STATE
, spa
, RW_READER
);
3196 tvd
= spa
->spa_root_vdev
->vdev_child
[top
];
3197 new_ms_count
= tvd
->vdev_ms_count
;
3198 new_class_space
= metaslab_class_get_space(mc
);
3200 if (tvd
->vdev_mg
!= mg
|| mg
->mg_class
!= mc
) {
3201 if (ztest_opts
.zo_verbose
>= 5) {
3202 (void) printf("Could not verify LUN expansion due to "
3203 "intervening vdev offline or remove.\n");
3205 spa_config_exit(spa
, SCL_STATE
, spa
);
3206 mutex_exit(&ztest_vdev_lock
);
3211 * Make sure we were able to grow the vdev.
3213 if (new_ms_count
<= old_ms_count
) {
3214 fatal(0, "LUN expansion failed: ms_count %llu < %llu\n",
3215 old_ms_count
, new_ms_count
);
3219 * Make sure we were able to grow the pool.
3221 if (new_class_space
<= old_class_space
) {
3222 fatal(0, "LUN expansion failed: class_space %llu < %llu\n",
3223 old_class_space
, new_class_space
);
3226 if (ztest_opts
.zo_verbose
>= 5) {
3227 char oldnumbuf
[NN_NUMBUF_SZ
], newnumbuf
[NN_NUMBUF_SZ
];
3229 nicenum(old_class_space
, oldnumbuf
, sizeof (oldnumbuf
));
3230 nicenum(new_class_space
, newnumbuf
, sizeof (newnumbuf
));
3231 (void) printf("%s grew from %s to %s\n",
3232 spa
->spa_name
, oldnumbuf
, newnumbuf
);
3235 spa_config_exit(spa
, SCL_STATE
, spa
);
3236 mutex_exit(&ztest_vdev_lock
);
3240 * Verify that dmu_objset_{create,destroy,open,close} work as expected.
3244 ztest_objset_create_cb(objset_t
*os
, void *arg
, cred_t
*cr
, dmu_tx_t
*tx
)
3247 * Create the objects common to all ztest datasets.
3249 VERIFY(zap_create_claim(os
, ZTEST_DIROBJ
,
3250 DMU_OT_ZAP_OTHER
, DMU_OT_NONE
, 0, tx
) == 0);
3254 ztest_dataset_create(char *dsname
)
3256 uint64_t zilset
= ztest_random(100);
3257 int err
= dmu_objset_create(dsname
, DMU_OST_OTHER
, 0,
3258 ztest_objset_create_cb
, NULL
);
3260 if (err
|| zilset
< 80)
3263 if (ztest_opts
.zo_verbose
>= 6)
3264 (void) printf("Setting dataset %s to sync always\n", dsname
);
3265 return (ztest_dsl_prop_set_uint64(dsname
, ZFS_PROP_SYNC
,
3266 ZFS_SYNC_ALWAYS
, B_FALSE
));
3271 ztest_objset_destroy_cb(const char *name
, void *arg
)
3274 dmu_object_info_t doi
;
3278 * Verify that the dataset contains a directory object.
3280 VERIFY0(dmu_objset_own(name
, DMU_OST_OTHER
, B_TRUE
, FTAG
, &os
));
3281 error
= dmu_object_info(os
, ZTEST_DIROBJ
, &doi
);
3282 if (error
!= ENOENT
) {
3283 /* We could have crashed in the middle of destroying it */
3285 ASSERT3U(doi
.doi_type
, ==, DMU_OT_ZAP_OTHER
);
3286 ASSERT3S(doi
.doi_physical_blocks_512
, >=, 0);
3288 dmu_objset_disown(os
, FTAG
);
3291 * Destroy the dataset.
3293 if (strchr(name
, '@') != NULL
) {
3294 VERIFY0(dsl_destroy_snapshot(name
, B_TRUE
));
3296 error
= dsl_destroy_head(name
);
3297 /* There could be a hold on this dataset */
3305 ztest_snapshot_create(char *osname
, uint64_t id
)
3307 char snapname
[ZFS_MAX_DATASET_NAME_LEN
];
3310 (void) snprintf(snapname
, sizeof (snapname
), "%llu", (u_longlong_t
)id
);
3312 error
= dmu_objset_snapshot_one(osname
, snapname
);
3313 if (error
== ENOSPC
) {
3314 ztest_record_enospc(FTAG
);
3317 if (error
!= 0 && error
!= EEXIST
) {
3318 fatal(0, "ztest_snapshot_create(%s@%s) = %d", osname
,
3325 ztest_snapshot_destroy(char *osname
, uint64_t id
)
3327 char snapname
[ZFS_MAX_DATASET_NAME_LEN
];
3330 (void) snprintf(snapname
, sizeof (snapname
), "%s@%llu", osname
,
3333 error
= dsl_destroy_snapshot(snapname
, B_FALSE
);
3334 if (error
!= 0 && error
!= ENOENT
)
3335 fatal(0, "ztest_snapshot_destroy(%s) = %d", snapname
, error
);
3341 ztest_dmu_objset_create_destroy(ztest_ds_t
*zd
, uint64_t id
)
3347 char name
[ZFS_MAX_DATASET_NAME_LEN
];
3350 rw_enter(&ztest_name_lock
, RW_READER
);
3352 (void) snprintf(name
, sizeof (name
), "%s/temp_%llu",
3353 ztest_opts
.zo_pool
, (u_longlong_t
)id
);
3356 * If this dataset exists from a previous run, process its replay log
3357 * half of the time. If we don't replay it, then dmu_objset_destroy()
3358 * (invoked from ztest_objset_destroy_cb()) should just throw it away.
3360 if (ztest_random(2) == 0 &&
3361 dmu_objset_own(name
, DMU_OST_OTHER
, B_FALSE
, FTAG
, &os
) == 0) {
3362 ztest_zd_init(&zdtmp
, NULL
, os
);
3363 zil_replay(os
, &zdtmp
, ztest_replay_vector
);
3364 ztest_zd_fini(&zdtmp
);
3365 dmu_objset_disown(os
, FTAG
);
3369 * There may be an old instance of the dataset we're about to
3370 * create lying around from a previous run. If so, destroy it
3371 * and all of its snapshots.
3373 (void) dmu_objset_find(name
, ztest_objset_destroy_cb
, NULL
,
3374 DS_FIND_CHILDREN
| DS_FIND_SNAPSHOTS
);
3377 * Verify that the destroyed dataset is no longer in the namespace.
3379 VERIFY3U(ENOENT
, ==, dmu_objset_own(name
, DMU_OST_OTHER
, B_TRUE
,
3383 * Verify that we can create a new dataset.
3385 error
= ztest_dataset_create(name
);
3387 if (error
== ENOSPC
) {
3388 ztest_record_enospc(FTAG
);
3389 rw_exit(&ztest_name_lock
);
3392 fatal(0, "dmu_objset_create(%s) = %d", name
, error
);
3395 VERIFY0(dmu_objset_own(name
, DMU_OST_OTHER
, B_FALSE
, FTAG
, &os
));
3397 ztest_zd_init(&zdtmp
, NULL
, os
);
3400 * Open the intent log for it.
3402 zilog
= zil_open(os
, ztest_get_data
);
3405 * Put some objects in there, do a little I/O to them,
3406 * and randomly take a couple of snapshots along the way.
3408 iters
= ztest_random(5);
3409 for (int i
= 0; i
< iters
; i
++) {
3410 ztest_dmu_object_alloc_free(&zdtmp
, id
);
3411 if (ztest_random(iters
) == 0)
3412 (void) ztest_snapshot_create(name
, i
);
3416 * Verify that we cannot create an existing dataset.
3418 VERIFY3U(EEXIST
, ==,
3419 dmu_objset_create(name
, DMU_OST_OTHER
, 0, NULL
, NULL
));
3422 * Verify that we can hold an objset that is also owned.
3424 VERIFY3U(0, ==, dmu_objset_hold(name
, FTAG
, &os2
));
3425 dmu_objset_rele(os2
, FTAG
);
3428 * Verify that we cannot own an objset that is already owned.
3431 dmu_objset_own(name
, DMU_OST_OTHER
, B_FALSE
, FTAG
, &os2
));
3434 dmu_objset_disown(os
, FTAG
);
3435 ztest_zd_fini(&zdtmp
);
3437 rw_exit(&ztest_name_lock
);
3441 * Verify that dmu_snapshot_{create,destroy,open,close} work as expected.
3444 ztest_dmu_snapshot_create_destroy(ztest_ds_t
*zd
, uint64_t id
)
3446 rw_enter(&ztest_name_lock
, RW_READER
);
3447 (void) ztest_snapshot_destroy(zd
->zd_name
, id
);
3448 (void) ztest_snapshot_create(zd
->zd_name
, id
);
3449 rw_exit(&ztest_name_lock
);
3453 * Cleanup non-standard snapshots and clones.
3456 ztest_dsl_dataset_cleanup(char *osname
, uint64_t id
)
3458 char snap1name
[ZFS_MAX_DATASET_NAME_LEN
];
3459 char clone1name
[ZFS_MAX_DATASET_NAME_LEN
];
3460 char snap2name
[ZFS_MAX_DATASET_NAME_LEN
];
3461 char clone2name
[ZFS_MAX_DATASET_NAME_LEN
];
3462 char snap3name
[ZFS_MAX_DATASET_NAME_LEN
];
3465 (void) snprintf(snap1name
, sizeof (snap1name
),
3466 "%s@s1_%llu", osname
, id
);
3467 (void) snprintf(clone1name
, sizeof (clone1name
),
3468 "%s/c1_%llu", osname
, id
);
3469 (void) snprintf(snap2name
, sizeof (snap2name
),
3470 "%s@s2_%llu", clone1name
, id
);
3471 (void) snprintf(clone2name
, sizeof (clone2name
),
3472 "%s/c2_%llu", osname
, id
);
3473 (void) snprintf(snap3name
, sizeof (snap3name
),
3474 "%s@s3_%llu", clone1name
, id
);
3476 error
= dsl_destroy_head(clone2name
);
3477 if (error
&& error
!= ENOENT
)
3478 fatal(0, "dsl_destroy_head(%s) = %d", clone2name
, error
);
3479 error
= dsl_destroy_snapshot(snap3name
, B_FALSE
);
3480 if (error
&& error
!= ENOENT
)
3481 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap3name
, error
);
3482 error
= dsl_destroy_snapshot(snap2name
, B_FALSE
);
3483 if (error
&& error
!= ENOENT
)
3484 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap2name
, error
);
3485 error
= dsl_destroy_head(clone1name
);
3486 if (error
&& error
!= ENOENT
)
3487 fatal(0, "dsl_destroy_head(%s) = %d", clone1name
, error
);
3488 error
= dsl_destroy_snapshot(snap1name
, B_FALSE
);
3489 if (error
&& error
!= ENOENT
)
3490 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap1name
, error
);
3494 * Verify dsl_dataset_promote handles EBUSY
3497 ztest_dsl_dataset_promote_busy(ztest_ds_t
*zd
, uint64_t id
)
3500 char snap1name
[ZFS_MAX_DATASET_NAME_LEN
];
3501 char clone1name
[ZFS_MAX_DATASET_NAME_LEN
];
3502 char snap2name
[ZFS_MAX_DATASET_NAME_LEN
];
3503 char clone2name
[ZFS_MAX_DATASET_NAME_LEN
];
3504 char snap3name
[ZFS_MAX_DATASET_NAME_LEN
];
3505 char *osname
= zd
->zd_name
;
3508 rw_enter(&ztest_name_lock
, RW_READER
);
3510 ztest_dsl_dataset_cleanup(osname
, id
);
3512 (void) snprintf(snap1name
, sizeof (snap1name
),
3513 "%s@s1_%llu", osname
, id
);
3514 (void) snprintf(clone1name
, sizeof (clone1name
),
3515 "%s/c1_%llu", osname
, id
);
3516 (void) snprintf(snap2name
, sizeof (snap2name
),
3517 "%s@s2_%llu", clone1name
, id
);
3518 (void) snprintf(clone2name
, sizeof (clone2name
),
3519 "%s/c2_%llu", osname
, id
);
3520 (void) snprintf(snap3name
, sizeof (snap3name
),
3521 "%s@s3_%llu", clone1name
, id
);
3523 error
= dmu_objset_snapshot_one(osname
, strchr(snap1name
, '@') + 1);
3524 if (error
&& error
!= EEXIST
) {
3525 if (error
== ENOSPC
) {
3526 ztest_record_enospc(FTAG
);
3529 fatal(0, "dmu_take_snapshot(%s) = %d", snap1name
, error
);
3532 error
= dmu_objset_clone(clone1name
, snap1name
);
3534 if (error
== ENOSPC
) {
3535 ztest_record_enospc(FTAG
);
3538 fatal(0, "dmu_objset_create(%s) = %d", clone1name
, error
);
3541 error
= dmu_objset_snapshot_one(clone1name
, strchr(snap2name
, '@') + 1);
3542 if (error
&& error
!= EEXIST
) {
3543 if (error
== ENOSPC
) {
3544 ztest_record_enospc(FTAG
);
3547 fatal(0, "dmu_open_snapshot(%s) = %d", snap2name
, error
);
3550 error
= dmu_objset_snapshot_one(clone1name
, strchr(snap3name
, '@') + 1);
3551 if (error
&& error
!= EEXIST
) {
3552 if (error
== ENOSPC
) {
3553 ztest_record_enospc(FTAG
);
3556 fatal(0, "dmu_open_snapshot(%s) = %d", snap3name
, error
);
3559 error
= dmu_objset_clone(clone2name
, snap3name
);
3561 if (error
== ENOSPC
) {
3562 ztest_record_enospc(FTAG
);
3565 fatal(0, "dmu_objset_create(%s) = %d", clone2name
, error
);
3568 error
= dmu_objset_own(snap2name
, DMU_OST_ANY
, B_TRUE
, FTAG
, &os
);
3570 fatal(0, "dmu_objset_own(%s) = %d", snap2name
, error
);
3571 error
= dsl_dataset_promote(clone2name
, NULL
);
3572 if (error
== ENOSPC
) {
3573 dmu_objset_disown(os
, FTAG
);
3574 ztest_record_enospc(FTAG
);
3578 fatal(0, "dsl_dataset_promote(%s), %d, not EBUSY", clone2name
,
3580 dmu_objset_disown(os
, FTAG
);
3583 ztest_dsl_dataset_cleanup(osname
, id
);
3585 rw_exit(&ztest_name_lock
);
3589 * Verify that dmu_object_{alloc,free} work as expected.
3592 ztest_dmu_object_alloc_free(ztest_ds_t
*zd
, uint64_t id
)
3595 int batchsize
= sizeof (od
) / sizeof (od
[0]);
3597 for (int b
= 0; b
< batchsize
; b
++)
3598 ztest_od_init(&od
[b
], id
, FTAG
, b
, DMU_OT_UINT64_OTHER
, 0, 0);
3601 * Destroy the previous batch of objects, create a new batch,
3602 * and do some I/O on the new objects.
3604 if (ztest_object_init(zd
, od
, sizeof (od
), B_TRUE
) != 0)
3607 while (ztest_random(4 * batchsize
) != 0)
3608 ztest_io(zd
, od
[ztest_random(batchsize
)].od_object
,
3609 ztest_random(ZTEST_RANGE_LOCKS
) << SPA_MAXBLOCKSHIFT
);
3613 * Verify that dmu_{read,write} work as expected.
3616 ztest_dmu_read_write(ztest_ds_t
*zd
, uint64_t id
)
3618 objset_t
*os
= zd
->zd_os
;
3621 int i
, freeit
, error
;
3623 bufwad_t
*packbuf
, *bigbuf
, *pack
, *bigH
, *bigT
;
3624 uint64_t packobj
, packoff
, packsize
, bigobj
, bigoff
, bigsize
;
3625 uint64_t chunksize
= (1000 + ztest_random(1000)) * sizeof (uint64_t);
3626 uint64_t regions
= 997;
3627 uint64_t stride
= 123456789ULL;
3628 uint64_t width
= 40;
3629 int free_percent
= 5;
3632 * This test uses two objects, packobj and bigobj, that are always
3633 * updated together (i.e. in the same tx) so that their contents are
3634 * in sync and can be compared. Their contents relate to each other
3635 * in a simple way: packobj is a dense array of 'bufwad' structures,
3636 * while bigobj is a sparse array of the same bufwads. Specifically,
3637 * for any index n, there are three bufwads that should be identical:
3639 * packobj, at offset n * sizeof (bufwad_t)
3640 * bigobj, at the head of the nth chunk
3641 * bigobj, at the tail of the nth chunk
3643 * The chunk size is arbitrary. It doesn't have to be a power of two,
3644 * and it doesn't have any relation to the object blocksize.
3645 * The only requirement is that it can hold at least two bufwads.
3647 * Normally, we write the bufwad to each of these locations.
3648 * However, free_percent of the time we instead write zeroes to
3649 * packobj and perform a dmu_free_range() on bigobj. By comparing
3650 * bigobj to packobj, we can verify that the DMU is correctly
3651 * tracking which parts of an object are allocated and free,
3652 * and that the contents of the allocated blocks are correct.
3656 * Read the directory info. If it's the first time, set things up.
3658 ztest_od_init(&od
[0], id
, FTAG
, 0, DMU_OT_UINT64_OTHER
, 0, chunksize
);
3659 ztest_od_init(&od
[1], id
, FTAG
, 1, DMU_OT_UINT64_OTHER
, 0, chunksize
);
3661 if (ztest_object_init(zd
, od
, sizeof (od
), B_FALSE
) != 0)
3664 bigobj
= od
[0].od_object
;
3665 packobj
= od
[1].od_object
;
3666 chunksize
= od
[0].od_gen
;
3667 ASSERT(chunksize
== od
[1].od_gen
);
3670 * Prefetch a random chunk of the big object.
3671 * Our aim here is to get some async reads in flight
3672 * for blocks that we may free below; the DMU should
3673 * handle this race correctly.
3675 n
= ztest_random(regions
) * stride
+ ztest_random(width
);
3676 s
= 1 + ztest_random(2 * width
- 1);
3677 dmu_prefetch(os
, bigobj
, 0, n
* chunksize
, s
* chunksize
,
3678 ZIO_PRIORITY_SYNC_READ
);
3681 * Pick a random index and compute the offsets into packobj and bigobj.
3683 n
= ztest_random(regions
) * stride
+ ztest_random(width
);
3684 s
= 1 + ztest_random(width
- 1);
3686 packoff
= n
* sizeof (bufwad_t
);
3687 packsize
= s
* sizeof (bufwad_t
);
3689 bigoff
= n
* chunksize
;
3690 bigsize
= s
* chunksize
;
3692 packbuf
= umem_alloc(packsize
, UMEM_NOFAIL
);
3693 bigbuf
= umem_alloc(bigsize
, UMEM_NOFAIL
);
3696 * free_percent of the time, free a range of bigobj rather than
3699 freeit
= (ztest_random(100) < free_percent
);
3702 * Read the current contents of our objects.
3704 error
= dmu_read(os
, packobj
, packoff
, packsize
, packbuf
,
3707 error
= dmu_read(os
, bigobj
, bigoff
, bigsize
, bigbuf
,
3712 * Get a tx for the mods to both packobj and bigobj.
3714 tx
= dmu_tx_create(os
);
3716 dmu_tx_hold_write(tx
, packobj
, packoff
, packsize
);
3719 dmu_tx_hold_free(tx
, bigobj
, bigoff
, bigsize
);
3721 dmu_tx_hold_write(tx
, bigobj
, bigoff
, bigsize
);
3723 /* This accounts for setting the checksum/compression. */
3724 dmu_tx_hold_bonus(tx
, bigobj
);
3726 txg
= ztest_tx_assign(tx
, TXG_MIGHTWAIT
, FTAG
);
3728 umem_free(packbuf
, packsize
);
3729 umem_free(bigbuf
, bigsize
);
3733 enum zio_checksum cksum
;
3735 cksum
= (enum zio_checksum
)
3736 ztest_random_dsl_prop(ZFS_PROP_CHECKSUM
);
3737 } while (cksum
>= ZIO_CHECKSUM_LEGACY_FUNCTIONS
);
3738 dmu_object_set_checksum(os
, bigobj
, cksum
, tx
);
3740 enum zio_compress comp
;
3742 comp
= (enum zio_compress
)
3743 ztest_random_dsl_prop(ZFS_PROP_COMPRESSION
);
3744 } while (comp
>= ZIO_COMPRESS_LEGACY_FUNCTIONS
);
3745 dmu_object_set_compress(os
, bigobj
, comp
, tx
);
3748 * For each index from n to n + s, verify that the existing bufwad
3749 * in packobj matches the bufwads at the head and tail of the
3750 * corresponding chunk in bigobj. Then update all three bufwads
3751 * with the new values we want to write out.
3753 for (i
= 0; i
< s
; i
++) {
3755 pack
= (bufwad_t
*)((char *)packbuf
+ i
* sizeof (bufwad_t
));
3757 bigH
= (bufwad_t
*)((char *)bigbuf
+ i
* chunksize
);
3759 bigT
= (bufwad_t
*)((char *)bigH
+ chunksize
) - 1;
3761 ASSERT((uintptr_t)bigH
- (uintptr_t)bigbuf
< bigsize
);
3762 ASSERT((uintptr_t)bigT
- (uintptr_t)bigbuf
< bigsize
);
3764 if (pack
->bw_txg
> txg
)
3765 fatal(0, "future leak: got %llx, open txg is %llx",
3768 if (pack
->bw_data
!= 0 && pack
->bw_index
!= n
+ i
)
3769 fatal(0, "wrong index: got %llx, wanted %llx+%llx",
3770 pack
->bw_index
, n
, i
);
3772 if (bcmp(pack
, bigH
, sizeof (bufwad_t
)) != 0)
3773 fatal(0, "pack/bigH mismatch in %p/%p", pack
, bigH
);
3775 if (bcmp(pack
, bigT
, sizeof (bufwad_t
)) != 0)
3776 fatal(0, "pack/bigT mismatch in %p/%p", pack
, bigT
);
3779 bzero(pack
, sizeof (bufwad_t
));
3781 pack
->bw_index
= n
+ i
;
3783 pack
->bw_data
= 1 + ztest_random(-2ULL);
3790 * We've verified all the old bufwads, and made new ones.
3791 * Now write them out.
3793 dmu_write(os
, packobj
, packoff
, packsize
, packbuf
, tx
);
3796 if (ztest_opts
.zo_verbose
>= 7) {
3797 (void) printf("freeing offset %llx size %llx"
3799 (u_longlong_t
)bigoff
,
3800 (u_longlong_t
)bigsize
,
3803 VERIFY(0 == dmu_free_range(os
, bigobj
, bigoff
, bigsize
, tx
));
3805 if (ztest_opts
.zo_verbose
>= 7) {
3806 (void) printf("writing offset %llx size %llx"
3808 (u_longlong_t
)bigoff
,
3809 (u_longlong_t
)bigsize
,
3812 dmu_write(os
, bigobj
, bigoff
, bigsize
, bigbuf
, tx
);
3818 * Sanity check the stuff we just wrote.
3821 void *packcheck
= umem_alloc(packsize
, UMEM_NOFAIL
);
3822 void *bigcheck
= umem_alloc(bigsize
, UMEM_NOFAIL
);
3824 VERIFY(0 == dmu_read(os
, packobj
, packoff
,
3825 packsize
, packcheck
, DMU_READ_PREFETCH
));
3826 VERIFY(0 == dmu_read(os
, bigobj
, bigoff
,
3827 bigsize
, bigcheck
, DMU_READ_PREFETCH
));
3829 ASSERT(bcmp(packbuf
, packcheck
, packsize
) == 0);
3830 ASSERT(bcmp(bigbuf
, bigcheck
, bigsize
) == 0);
3832 umem_free(packcheck
, packsize
);
3833 umem_free(bigcheck
, bigsize
);
3836 umem_free(packbuf
, packsize
);
3837 umem_free(bigbuf
, bigsize
);
3841 compare_and_update_pbbufs(uint64_t s
, bufwad_t
*packbuf
, bufwad_t
*bigbuf
,
3842 uint64_t bigsize
, uint64_t n
, uint64_t chunksize
, uint64_t txg
)
3850 * For each index from n to n + s, verify that the existing bufwad
3851 * in packobj matches the bufwads at the head and tail of the
3852 * corresponding chunk in bigobj. Then update all three bufwads
3853 * with the new values we want to write out.
3855 for (i
= 0; i
< s
; i
++) {
3857 pack
= (bufwad_t
*)((char *)packbuf
+ i
* sizeof (bufwad_t
));
3859 bigH
= (bufwad_t
*)((char *)bigbuf
+ i
* chunksize
);
3861 bigT
= (bufwad_t
*)((char *)bigH
+ chunksize
) - 1;
3863 ASSERT((uintptr_t)bigH
- (uintptr_t)bigbuf
< bigsize
);
3864 ASSERT((uintptr_t)bigT
- (uintptr_t)bigbuf
< bigsize
);
3866 if (pack
->bw_txg
> txg
)
3867 fatal(0, "future leak: got %llx, open txg is %llx",
3870 if (pack
->bw_data
!= 0 && pack
->bw_index
!= n
+ i
)
3871 fatal(0, "wrong index: got %llx, wanted %llx+%llx",
3872 pack
->bw_index
, n
, i
);
3874 if (bcmp(pack
, bigH
, sizeof (bufwad_t
)) != 0)
3875 fatal(0, "pack/bigH mismatch in %p/%p", pack
, bigH
);
3877 if (bcmp(pack
, bigT
, sizeof (bufwad_t
)) != 0)
3878 fatal(0, "pack/bigT mismatch in %p/%p", pack
, bigT
);
3880 pack
->bw_index
= n
+ i
;
3882 pack
->bw_data
= 1 + ztest_random(-2ULL);
3890 ztest_dmu_read_write_zcopy(ztest_ds_t
*zd
, uint64_t id
)
3892 objset_t
*os
= zd
->zd_os
;
3898 bufwad_t
*packbuf
, *bigbuf
;
3899 uint64_t packobj
, packoff
, packsize
, bigobj
, bigoff
, bigsize
;
3900 uint64_t blocksize
= ztest_random_blocksize();
3901 uint64_t chunksize
= blocksize
;
3902 uint64_t regions
= 997;
3903 uint64_t stride
= 123456789ULL;
3905 dmu_buf_t
*bonus_db
;
3906 arc_buf_t
**bigbuf_arcbufs
;
3907 dmu_object_info_t doi
;
3910 * This test uses two objects, packobj and bigobj, that are always
3911 * updated together (i.e. in the same tx) so that their contents are
3912 * in sync and can be compared. Their contents relate to each other
3913 * in a simple way: packobj is a dense array of 'bufwad' structures,
3914 * while bigobj is a sparse array of the same bufwads. Specifically,
3915 * for any index n, there are three bufwads that should be identical:
3917 * packobj, at offset n * sizeof (bufwad_t)
3918 * bigobj, at the head of the nth chunk
3919 * bigobj, at the tail of the nth chunk
3921 * The chunk size is set equal to bigobj block size so that
3922 * dmu_assign_arcbuf() can be tested for object updates.
3926 * Read the directory info. If it's the first time, set things up.
3928 ztest_od_init(&od
[0], id
, FTAG
, 0, DMU_OT_UINT64_OTHER
, blocksize
, 0);
3929 ztest_od_init(&od
[1], id
, FTAG
, 1, DMU_OT_UINT64_OTHER
, 0, chunksize
);
3931 if (ztest_object_init(zd
, od
, sizeof (od
), B_FALSE
) != 0)
3934 bigobj
= od
[0].od_object
;
3935 packobj
= od
[1].od_object
;
3936 blocksize
= od
[0].od_blocksize
;
3937 chunksize
= blocksize
;
3938 ASSERT(chunksize
== od
[1].od_gen
);
3940 VERIFY(dmu_object_info(os
, bigobj
, &doi
) == 0);
3941 VERIFY(ISP2(doi
.doi_data_block_size
));
3942 VERIFY(chunksize
== doi
.doi_data_block_size
);
3943 VERIFY(chunksize
>= 2 * sizeof (bufwad_t
));
3946 * Pick a random index and compute the offsets into packobj and bigobj.
3948 n
= ztest_random(regions
) * stride
+ ztest_random(width
);
3949 s
= 1 + ztest_random(width
- 1);
3951 packoff
= n
* sizeof (bufwad_t
);
3952 packsize
= s
* sizeof (bufwad_t
);
3954 bigoff
= n
* chunksize
;
3955 bigsize
= s
* chunksize
;
3957 packbuf
= umem_zalloc(packsize
, UMEM_NOFAIL
);
3958 bigbuf
= umem_zalloc(bigsize
, UMEM_NOFAIL
);
3960 VERIFY3U(0, ==, dmu_bonus_hold(os
, bigobj
, FTAG
, &bonus_db
));
3962 bigbuf_arcbufs
= umem_zalloc(2 * s
* sizeof (arc_buf_t
*), UMEM_NOFAIL
);
3965 * Iteration 0 test zcopy for DB_UNCACHED dbufs.
3966 * Iteration 1 test zcopy to already referenced dbufs.
3967 * Iteration 2 test zcopy to dirty dbuf in the same txg.
3968 * Iteration 3 test zcopy to dbuf dirty in previous txg.
3969 * Iteration 4 test zcopy when dbuf is no longer dirty.
3970 * Iteration 5 test zcopy when it can't be done.
3971 * Iteration 6 one more zcopy write.
3973 for (i
= 0; i
< 7; i
++) {
3978 * In iteration 5 (i == 5) use arcbufs
3979 * that don't match bigobj blksz to test
3980 * dmu_assign_arcbuf() when it can't directly
3981 * assign an arcbuf to a dbuf.
3983 for (j
= 0; j
< s
; j
++) {
3984 if (i
!= 5 || chunksize
< (SPA_MINBLOCKSIZE
* 2)) {
3986 dmu_request_arcbuf(bonus_db
, chunksize
);
3988 bigbuf_arcbufs
[2 * j
] =
3989 dmu_request_arcbuf(bonus_db
, chunksize
/ 2);
3990 bigbuf_arcbufs
[2 * j
+ 1] =
3991 dmu_request_arcbuf(bonus_db
, chunksize
/ 2);
3996 * Get a tx for the mods to both packobj and bigobj.
3998 tx
= dmu_tx_create(os
);
4000 dmu_tx_hold_write(tx
, packobj
, packoff
, packsize
);
4001 dmu_tx_hold_write(tx
, bigobj
, bigoff
, bigsize
);
4003 txg
= ztest_tx_assign(tx
, TXG_MIGHTWAIT
, FTAG
);
4005 umem_free(packbuf
, packsize
);
4006 umem_free(bigbuf
, bigsize
);
4007 for (j
= 0; j
< s
; j
++) {
4009 chunksize
< (SPA_MINBLOCKSIZE
* 2)) {
4010 dmu_return_arcbuf(bigbuf_arcbufs
[j
]);
4013 bigbuf_arcbufs
[2 * j
]);
4015 bigbuf_arcbufs
[2 * j
+ 1]);
4018 umem_free(bigbuf_arcbufs
, 2 * s
* sizeof (arc_buf_t
*));
4019 dmu_buf_rele(bonus_db
, FTAG
);
4024 * 50% of the time don't read objects in the 1st iteration to
4025 * test dmu_assign_arcbuf() for the case when there're no
4026 * existing dbufs for the specified offsets.
4028 if (i
!= 0 || ztest_random(2) != 0) {
4029 error
= dmu_read(os
, packobj
, packoff
,
4030 packsize
, packbuf
, DMU_READ_PREFETCH
);
4032 error
= dmu_read(os
, bigobj
, bigoff
, bigsize
,
4033 bigbuf
, DMU_READ_PREFETCH
);
4036 compare_and_update_pbbufs(s
, packbuf
, bigbuf
, bigsize
,
4040 * We've verified all the old bufwads, and made new ones.
4041 * Now write them out.
4043 dmu_write(os
, packobj
, packoff
, packsize
, packbuf
, tx
);
4044 if (ztest_opts
.zo_verbose
>= 7) {
4045 (void) printf("writing offset %llx size %llx"
4047 (u_longlong_t
)bigoff
,
4048 (u_longlong_t
)bigsize
,
4051 for (off
= bigoff
, j
= 0; j
< s
; j
++, off
+= chunksize
) {
4053 if (i
!= 5 || chunksize
< (SPA_MINBLOCKSIZE
* 2)) {
4054 bcopy((caddr_t
)bigbuf
+ (off
- bigoff
),
4055 bigbuf_arcbufs
[j
]->b_data
, chunksize
);
4057 bcopy((caddr_t
)bigbuf
+ (off
- bigoff
),
4058 bigbuf_arcbufs
[2 * j
]->b_data
,
4060 bcopy((caddr_t
)bigbuf
+ (off
- bigoff
) +
4062 bigbuf_arcbufs
[2 * j
+ 1]->b_data
,
4067 VERIFY(dmu_buf_hold(os
, bigobj
, off
,
4068 FTAG
, &dbt
, DMU_READ_NO_PREFETCH
) == 0);
4070 if (i
!= 5 || chunksize
< (SPA_MINBLOCKSIZE
* 2)) {
4071 dmu_assign_arcbuf(bonus_db
, off
,
4072 bigbuf_arcbufs
[j
], tx
);
4074 dmu_assign_arcbuf(bonus_db
, off
,
4075 bigbuf_arcbufs
[2 * j
], tx
);
4076 dmu_assign_arcbuf(bonus_db
,
4077 off
+ chunksize
/ 2,
4078 bigbuf_arcbufs
[2 * j
+ 1], tx
);
4081 dmu_buf_rele(dbt
, FTAG
);
4087 * Sanity check the stuff we just wrote.
4090 void *packcheck
= umem_alloc(packsize
, UMEM_NOFAIL
);
4091 void *bigcheck
= umem_alloc(bigsize
, UMEM_NOFAIL
);
4093 VERIFY(0 == dmu_read(os
, packobj
, packoff
,
4094 packsize
, packcheck
, DMU_READ_PREFETCH
));
4095 VERIFY(0 == dmu_read(os
, bigobj
, bigoff
,
4096 bigsize
, bigcheck
, DMU_READ_PREFETCH
));
4098 ASSERT(bcmp(packbuf
, packcheck
, packsize
) == 0);
4099 ASSERT(bcmp(bigbuf
, bigcheck
, bigsize
) == 0);
4101 umem_free(packcheck
, packsize
);
4102 umem_free(bigcheck
, bigsize
);
4105 txg_wait_open(dmu_objset_pool(os
), 0);
4106 } else if (i
== 3) {
4107 txg_wait_synced(dmu_objset_pool(os
), 0);
4111 dmu_buf_rele(bonus_db
, FTAG
);
4112 umem_free(packbuf
, packsize
);
4113 umem_free(bigbuf
, bigsize
);
4114 umem_free(bigbuf_arcbufs
, 2 * s
* sizeof (arc_buf_t
*));
4119 ztest_dmu_write_parallel(ztest_ds_t
*zd
, uint64_t id
)
4122 uint64_t offset
= (1ULL << (ztest_random(20) + 43)) +
4123 (ztest_random(ZTEST_RANGE_LOCKS
) << SPA_MAXBLOCKSHIFT
);
4126 * Have multiple threads write to large offsets in an object
4127 * to verify that parallel writes to an object -- even to the
4128 * same blocks within the object -- doesn't cause any trouble.
4130 ztest_od_init(&od
[0], ID_PARALLEL
, FTAG
, 0, DMU_OT_UINT64_OTHER
, 0, 0);
4132 if (ztest_object_init(zd
, od
, sizeof (od
), B_FALSE
) != 0)
4135 while (ztest_random(10) != 0)
4136 ztest_io(zd
, od
[0].od_object
, offset
);
4140 ztest_dmu_prealloc(ztest_ds_t
*zd
, uint64_t id
)
4143 uint64_t offset
= (1ULL << (ztest_random(4) + SPA_MAXBLOCKSHIFT
)) +
4144 (ztest_random(ZTEST_RANGE_LOCKS
) << SPA_MAXBLOCKSHIFT
);
4145 uint64_t count
= ztest_random(20) + 1;
4146 uint64_t blocksize
= ztest_random_blocksize();
4149 ztest_od_init(&od
[0], id
, FTAG
, 0, DMU_OT_UINT64_OTHER
, blocksize
, 0);
4151 if (ztest_object_init(zd
, od
, sizeof (od
), !ztest_random(2)) != 0)
4154 if (ztest_truncate(zd
, od
[0].od_object
, offset
, count
* blocksize
) != 0)
4157 ztest_prealloc(zd
, od
[0].od_object
, offset
, count
* blocksize
);
4159 data
= umem_zalloc(blocksize
, UMEM_NOFAIL
);
4161 while (ztest_random(count
) != 0) {
4162 uint64_t randoff
= offset
+ (ztest_random(count
) * blocksize
);
4163 if (ztest_write(zd
, od
[0].od_object
, randoff
, blocksize
,
4166 while (ztest_random(4) != 0)
4167 ztest_io(zd
, od
[0].od_object
, randoff
);
4170 umem_free(data
, blocksize
);
4174 * Verify that zap_{create,destroy,add,remove,update} work as expected.
4176 #define ZTEST_ZAP_MIN_INTS 1
4177 #define ZTEST_ZAP_MAX_INTS 4
4178 #define ZTEST_ZAP_MAX_PROPS 1000
4181 ztest_zap(ztest_ds_t
*zd
, uint64_t id
)
4183 objset_t
*os
= zd
->zd_os
;
4186 uint64_t txg
, last_txg
;
4187 uint64_t value
[ZTEST_ZAP_MAX_INTS
];
4188 uint64_t zl_ints
, zl_intsize
, prop
;
4191 char propname
[100], txgname
[100];
4193 char *hc
[2] = { "s.acl.h", ".s.open.h.hyLZlg" };
4195 ztest_od_init(&od
[0], id
, FTAG
, 0, DMU_OT_ZAP_OTHER
, 0, 0);
4197 if (ztest_object_init(zd
, od
, sizeof (od
), !ztest_random(2)) != 0)
4200 object
= od
[0].od_object
;
4203 * Generate a known hash collision, and verify that
4204 * we can lookup and remove both entries.
4206 tx
= dmu_tx_create(os
);
4207 dmu_tx_hold_zap(tx
, object
, B_TRUE
, NULL
);
4208 txg
= ztest_tx_assign(tx
, TXG_MIGHTWAIT
, FTAG
);
4211 for (i
= 0; i
< 2; i
++) {
4213 VERIFY3U(0, ==, zap_add(os
, object
, hc
[i
], sizeof (uint64_t),
4216 for (i
= 0; i
< 2; i
++) {
4217 VERIFY3U(EEXIST
, ==, zap_add(os
, object
, hc
[i
],
4218 sizeof (uint64_t), 1, &value
[i
], tx
));
4220 zap_length(os
, object
, hc
[i
], &zl_intsize
, &zl_ints
));
4221 ASSERT3U(zl_intsize
, ==, sizeof (uint64_t));
4222 ASSERT3U(zl_ints
, ==, 1);
4224 for (i
= 0; i
< 2; i
++) {
4225 VERIFY3U(0, ==, zap_remove(os
, object
, hc
[i
], tx
));
4230 * Generate a buch of random entries.
4232 ints
= MAX(ZTEST_ZAP_MIN_INTS
, object
% ZTEST_ZAP_MAX_INTS
);
4234 prop
= ztest_random(ZTEST_ZAP_MAX_PROPS
);
4235 (void) sprintf(propname
, "prop_%llu", (u_longlong_t
)prop
);
4236 (void) sprintf(txgname
, "txg_%llu", (u_longlong_t
)prop
);
4237 bzero(value
, sizeof (value
));
4241 * If these zap entries already exist, validate their contents.
4243 error
= zap_length(os
, object
, txgname
, &zl_intsize
, &zl_ints
);
4245 ASSERT3U(zl_intsize
, ==, sizeof (uint64_t));
4246 ASSERT3U(zl_ints
, ==, 1);
4248 VERIFY(zap_lookup(os
, object
, txgname
, zl_intsize
,
4249 zl_ints
, &last_txg
) == 0);
4251 VERIFY(zap_length(os
, object
, propname
, &zl_intsize
,
4254 ASSERT3U(zl_intsize
, ==, sizeof (uint64_t));
4255 ASSERT3U(zl_ints
, ==, ints
);
4257 VERIFY(zap_lookup(os
, object
, propname
, zl_intsize
,
4258 zl_ints
, value
) == 0);
4260 for (i
= 0; i
< ints
; i
++) {
4261 ASSERT3U(value
[i
], ==, last_txg
+ object
+ i
);
4264 ASSERT3U(error
, ==, ENOENT
);
4268 * Atomically update two entries in our zap object.
4269 * The first is named txg_%llu, and contains the txg
4270 * in which the property was last updated. The second
4271 * is named prop_%llu, and the nth element of its value
4272 * should be txg + object + n.
4274 tx
= dmu_tx_create(os
);
4275 dmu_tx_hold_zap(tx
, object
, B_TRUE
, NULL
);
4276 txg
= ztest_tx_assign(tx
, TXG_MIGHTWAIT
, FTAG
);
4281 fatal(0, "zap future leak: old %llu new %llu", last_txg
, txg
);
4283 for (i
= 0; i
< ints
; i
++)
4284 value
[i
] = txg
+ object
+ i
;
4286 VERIFY3U(0, ==, zap_update(os
, object
, txgname
, sizeof (uint64_t),
4288 VERIFY3U(0, ==, zap_update(os
, object
, propname
, sizeof (uint64_t),
4294 * Remove a random pair of entries.
4296 prop
= ztest_random(ZTEST_ZAP_MAX_PROPS
);
4297 (void) sprintf(propname
, "prop_%llu", (u_longlong_t
)prop
);
4298 (void) sprintf(txgname
, "txg_%llu", (u_longlong_t
)prop
);
4300 error
= zap_length(os
, object
, txgname
, &zl_intsize
, &zl_ints
);
4302 if (error
== ENOENT
)
4307 tx
= dmu_tx_create(os
);
4308 dmu_tx_hold_zap(tx
, object
, B_TRUE
, NULL
);
4309 txg
= ztest_tx_assign(tx
, TXG_MIGHTWAIT
, FTAG
);
4312 VERIFY3U(0, ==, zap_remove(os
, object
, txgname
, tx
));
4313 VERIFY3U(0, ==, zap_remove(os
, object
, propname
, tx
));
4318 * Testcase to test the upgrading of a microzap to fatzap.
4321 ztest_fzap(ztest_ds_t
*zd
, uint64_t id
)
4323 objset_t
*os
= zd
->zd_os
;
4325 uint64_t object
, txg
;
4327 ztest_od_init(&od
[0], id
, FTAG
, 0, DMU_OT_ZAP_OTHER
, 0, 0);
4329 if (ztest_object_init(zd
, od
, sizeof (od
), !ztest_random(2)) != 0)
4332 object
= od
[0].od_object
;
4335 * Add entries to this ZAP and make sure it spills over
4336 * and gets upgraded to a fatzap. Also, since we are adding
4337 * 2050 entries we should see ptrtbl growth and leaf-block split.
4339 for (int i
= 0; i
< 2050; i
++) {
4340 char name
[ZFS_MAX_DATASET_NAME_LEN
];
4345 (void) snprintf(name
, sizeof (name
), "fzap-%llu-%llu",
4348 tx
= dmu_tx_create(os
);
4349 dmu_tx_hold_zap(tx
, object
, B_TRUE
, name
);
4350 txg
= ztest_tx_assign(tx
, TXG_MIGHTWAIT
, FTAG
);
4353 error
= zap_add(os
, object
, name
, sizeof (uint64_t), 1,
4355 ASSERT(error
== 0 || error
== EEXIST
);
4362 ztest_zap_parallel(ztest_ds_t
*zd
, uint64_t id
)
4364 objset_t
*os
= zd
->zd_os
;
4366 uint64_t txg
, object
, count
, wsize
, wc
, zl_wsize
, zl_wc
;
4368 int i
, namelen
, error
;
4369 int micro
= ztest_random(2);
4370 char name
[20], string_value
[20];
4373 ztest_od_init(&od
[0], ID_PARALLEL
, FTAG
, micro
, DMU_OT_ZAP_OTHER
, 0, 0);
4375 if (ztest_object_init(zd
, od
, sizeof (od
), B_FALSE
) != 0)
4378 object
= od
[0].od_object
;
4381 * Generate a random name of the form 'xxx.....' where each
4382 * x is a random printable character and the dots are dots.
4383 * There are 94 such characters, and the name length goes from
4384 * 6 to 20, so there are 94^3 * 15 = 12,458,760 possible names.
4386 namelen
= ztest_random(sizeof (name
) - 5) + 5 + 1;
4388 for (i
= 0; i
< 3; i
++)
4389 name
[i
] = '!' + ztest_random('~' - '!' + 1);
4390 for (; i
< namelen
- 1; i
++)
4394 if ((namelen
& 1) || micro
) {
4395 wsize
= sizeof (txg
);
4401 data
= string_value
;
4405 VERIFY0(zap_count(os
, object
, &count
));
4406 ASSERT(count
!= -1ULL);
4409 * Select an operation: length, lookup, add, update, remove.
4411 i
= ztest_random(5);
4414 tx
= dmu_tx_create(os
);
4415 dmu_tx_hold_zap(tx
, object
, B_TRUE
, NULL
);
4416 txg
= ztest_tx_assign(tx
, TXG_MIGHTWAIT
, FTAG
);
4419 bcopy(name
, string_value
, namelen
);
4423 bzero(string_value
, namelen
);
4429 error
= zap_length(os
, object
, name
, &zl_wsize
, &zl_wc
);
4431 ASSERT3U(wsize
, ==, zl_wsize
);
4432 ASSERT3U(wc
, ==, zl_wc
);
4434 ASSERT3U(error
, ==, ENOENT
);
4439 error
= zap_lookup(os
, object
, name
, wsize
, wc
, data
);
4441 if (data
== string_value
&&
4442 bcmp(name
, data
, namelen
) != 0)
4443 fatal(0, "name '%s' != val '%s' len %d",
4444 name
, data
, namelen
);
4446 ASSERT3U(error
, ==, ENOENT
);
4451 error
= zap_add(os
, object
, name
, wsize
, wc
, data
, tx
);
4452 ASSERT(error
== 0 || error
== EEXIST
);
4456 VERIFY(zap_update(os
, object
, name
, wsize
, wc
, data
, tx
) == 0);
4460 error
= zap_remove(os
, object
, name
, tx
);
4461 ASSERT(error
== 0 || error
== ENOENT
);
4470 * Commit callback data.
4472 typedef struct ztest_cb_data
{
4473 list_node_t zcd_node
;
4475 int zcd_expected_err
;
4476 boolean_t zcd_added
;
4477 boolean_t zcd_called
;
4481 /* This is the actual commit callback function */
4483 ztest_commit_callback(void *arg
, int error
)
4485 ztest_cb_data_t
*data
= arg
;
4486 uint64_t synced_txg
;
4488 VERIFY(data
!= NULL
);
4489 VERIFY3S(data
->zcd_expected_err
, ==, error
);
4490 VERIFY(!data
->zcd_called
);
4492 synced_txg
= spa_last_synced_txg(data
->zcd_spa
);
4493 if (data
->zcd_txg
> synced_txg
)
4494 fatal(0, "commit callback of txg %" PRIu64
" called prematurely"
4495 ", last synced txg = %" PRIu64
"\n", data
->zcd_txg
,
4498 data
->zcd_called
= B_TRUE
;
4500 if (error
== ECANCELED
) {
4501 ASSERT0(data
->zcd_txg
);
4502 ASSERT(!data
->zcd_added
);
4505 * The private callback data should be destroyed here, but
4506 * since we are going to check the zcd_called field after
4507 * dmu_tx_abort(), we will destroy it there.
4512 /* Was this callback added to the global callback list? */
4513 if (!data
->zcd_added
)
4516 ASSERT3U(data
->zcd_txg
, !=, 0);
4518 /* Remove our callback from the list */
4519 mutex_enter(&zcl
.zcl_callbacks_lock
);
4520 list_remove(&zcl
.zcl_callbacks
, data
);
4521 mutex_exit(&zcl
.zcl_callbacks_lock
);
4524 umem_free(data
, sizeof (ztest_cb_data_t
));
4527 /* Allocate and initialize callback data structure */
4528 static ztest_cb_data_t
*
4529 ztest_create_cb_data(objset_t
*os
, uint64_t txg
)
4531 ztest_cb_data_t
*cb_data
;
4533 cb_data
= umem_zalloc(sizeof (ztest_cb_data_t
), UMEM_NOFAIL
);
4535 cb_data
->zcd_txg
= txg
;
4536 cb_data
->zcd_spa
= dmu_objset_spa(os
);
4542 * If a number of txgs equal to this threshold have been created after a commit
4543 * callback has been registered but not called, then we assume there is an
4544 * implementation bug.
4546 #define ZTEST_COMMIT_CALLBACK_THRESH (TXG_CONCURRENT_STATES + 2)
4549 * Commit callback test.
4552 ztest_dmu_commit_callbacks(ztest_ds_t
*zd
, uint64_t id
)
4554 objset_t
*os
= zd
->zd_os
;
4557 ztest_cb_data_t
*cb_data
[3], *tmp_cb
;
4558 uint64_t old_txg
, txg
;
4561 ztest_od_init(&od
[0], id
, FTAG
, 0, DMU_OT_UINT64_OTHER
, 0, 0);
4563 if (ztest_object_init(zd
, od
, sizeof (od
), B_FALSE
) != 0)
4566 tx
= dmu_tx_create(os
);
4568 cb_data
[0] = ztest_create_cb_data(os
, 0);
4569 dmu_tx_callback_register(tx
, ztest_commit_callback
, cb_data
[0]);
4571 dmu_tx_hold_write(tx
, od
[0].od_object
, 0, sizeof (uint64_t));
4573 /* Every once in a while, abort the transaction on purpose */
4574 if (ztest_random(100) == 0)
4578 error
= dmu_tx_assign(tx
, TXG_NOWAIT
);
4580 txg
= error
? 0 : dmu_tx_get_txg(tx
);
4582 cb_data
[0]->zcd_txg
= txg
;
4583 cb_data
[1] = ztest_create_cb_data(os
, txg
);
4584 dmu_tx_callback_register(tx
, ztest_commit_callback
, cb_data
[1]);
4588 * It's not a strict requirement to call the registered
4589 * callbacks from inside dmu_tx_abort(), but that's what
4590 * it's supposed to happen in the current implementation
4591 * so we will check for that.
4593 for (i
= 0; i
< 2; i
++) {
4594 cb_data
[i
]->zcd_expected_err
= ECANCELED
;
4595 VERIFY(!cb_data
[i
]->zcd_called
);
4600 for (i
= 0; i
< 2; i
++) {
4601 VERIFY(cb_data
[i
]->zcd_called
);
4602 umem_free(cb_data
[i
], sizeof (ztest_cb_data_t
));
4608 cb_data
[2] = ztest_create_cb_data(os
, txg
);
4609 dmu_tx_callback_register(tx
, ztest_commit_callback
, cb_data
[2]);
4612 * Read existing data to make sure there isn't a future leak.
4614 VERIFY(0 == dmu_read(os
, od
[0].od_object
, 0, sizeof (uint64_t),
4615 &old_txg
, DMU_READ_PREFETCH
));
4618 fatal(0, "future leak: got %" PRIu64
", open txg is %" PRIu64
,
4621 dmu_write(os
, od
[0].od_object
, 0, sizeof (uint64_t), &txg
, tx
);
4623 mutex_enter(&zcl
.zcl_callbacks_lock
);
4626 * Since commit callbacks don't have any ordering requirement and since
4627 * it is theoretically possible for a commit callback to be called
4628 * after an arbitrary amount of time has elapsed since its txg has been
4629 * synced, it is difficult to reliably determine whether a commit
4630 * callback hasn't been called due to high load or due to a flawed
4633 * In practice, we will assume that if after a certain number of txgs a
4634 * commit callback hasn't been called, then most likely there's an
4635 * implementation bug..
4637 tmp_cb
= list_head(&zcl
.zcl_callbacks
);
4638 if (tmp_cb
!= NULL
&&
4639 (txg
- ZTEST_COMMIT_CALLBACK_THRESH
) > tmp_cb
->zcd_txg
) {
4640 fatal(0, "Commit callback threshold exceeded, oldest txg: %"
4641 PRIu64
", open txg: %" PRIu64
"\n", tmp_cb
->zcd_txg
, txg
);
4645 * Let's find the place to insert our callbacks.
4647 * Even though the list is ordered by txg, it is possible for the
4648 * insertion point to not be the end because our txg may already be
4649 * quiescing at this point and other callbacks in the open txg
4650 * (from other objsets) may have sneaked in.
4652 tmp_cb
= list_tail(&zcl
.zcl_callbacks
);
4653 while (tmp_cb
!= NULL
&& tmp_cb
->zcd_txg
> txg
)
4654 tmp_cb
= list_prev(&zcl
.zcl_callbacks
, tmp_cb
);
4656 /* Add the 3 callbacks to the list */
4657 for (i
= 0; i
< 3; i
++) {
4659 list_insert_head(&zcl
.zcl_callbacks
, cb_data
[i
]);
4661 list_insert_after(&zcl
.zcl_callbacks
, tmp_cb
,
4664 cb_data
[i
]->zcd_added
= B_TRUE
;
4665 VERIFY(!cb_data
[i
]->zcd_called
);
4667 tmp_cb
= cb_data
[i
];
4670 mutex_exit(&zcl
.zcl_callbacks_lock
);
4677 ztest_dsl_prop_get_set(ztest_ds_t
*zd
, uint64_t id
)
4679 zfs_prop_t proplist
[] = {
4681 ZFS_PROP_COMPRESSION
,
4686 rw_enter(&ztest_name_lock
, RW_READER
);
4688 for (int p
= 0; p
< sizeof (proplist
) / sizeof (proplist
[0]); p
++)
4689 (void) ztest_dsl_prop_set_uint64(zd
->zd_name
, proplist
[p
],
4690 ztest_random_dsl_prop(proplist
[p
]), (int)ztest_random(2));
4692 rw_exit(&ztest_name_lock
);
4697 ztest_remap_blocks(ztest_ds_t
*zd
, uint64_t id
)
4699 rw_enter(&ztest_name_lock
, RW_READER
);
4701 int error
= dmu_objset_remap_indirects(zd
->zd_name
);
4702 if (error
== ENOSPC
)
4706 rw_exit(&ztest_name_lock
);
4711 ztest_spa_prop_get_set(ztest_ds_t
*zd
, uint64_t id
)
4713 nvlist_t
*props
= NULL
;
4715 rw_enter(&ztest_name_lock
, RW_READER
);
4717 (void) ztest_spa_prop_set_uint64(ZPOOL_PROP_DEDUPDITTO
,
4718 ZIO_DEDUPDITTO_MIN
+ ztest_random(ZIO_DEDUPDITTO_MIN
));
4720 VERIFY0(spa_prop_get(ztest_spa
, &props
));
4722 if (ztest_opts
.zo_verbose
>= 6)
4723 dump_nvlist(props
, 4);
4727 rw_exit(&ztest_name_lock
);
4731 user_release_one(const char *snapname
, const char *holdname
)
4733 nvlist_t
*snaps
, *holds
;
4736 snaps
= fnvlist_alloc();
4737 holds
= fnvlist_alloc();
4738 fnvlist_add_boolean(holds
, holdname
);
4739 fnvlist_add_nvlist(snaps
, snapname
, holds
);
4740 fnvlist_free(holds
);
4741 error
= dsl_dataset_user_release(snaps
, NULL
);
4742 fnvlist_free(snaps
);
4747 * Test snapshot hold/release and deferred destroy.
4750 ztest_dmu_snapshot_hold(ztest_ds_t
*zd
, uint64_t id
)
4753 objset_t
*os
= zd
->zd_os
;
4757 char clonename
[100];
4759 char osname
[ZFS_MAX_DATASET_NAME_LEN
];
4762 rw_enter(&ztest_name_lock
, RW_READER
);
4764 dmu_objset_name(os
, osname
);
4766 (void) snprintf(snapname
, sizeof (snapname
), "sh1_%llu", id
);
4767 (void) snprintf(fullname
, sizeof (fullname
), "%s@%s", osname
, snapname
);
4768 (void) snprintf(clonename
, sizeof (clonename
),
4769 "%s/ch1_%llu", osname
, id
);
4770 (void) snprintf(tag
, sizeof (tag
), "tag_%llu", id
);
4773 * Clean up from any previous run.
4775 error
= dsl_destroy_head(clonename
);
4776 if (error
!= ENOENT
)
4778 error
= user_release_one(fullname
, tag
);
4779 if (error
!= ESRCH
&& error
!= ENOENT
)
4781 error
= dsl_destroy_snapshot(fullname
, B_FALSE
);
4782 if (error
!= ENOENT
)
4786 * Create snapshot, clone it, mark snap for deferred destroy,
4787 * destroy clone, verify snap was also destroyed.
4789 error
= dmu_objset_snapshot_one(osname
, snapname
);
4791 if (error
== ENOSPC
) {
4792 ztest_record_enospc("dmu_objset_snapshot");
4795 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname
, error
);
4798 error
= dmu_objset_clone(clonename
, fullname
);
4800 if (error
== ENOSPC
) {
4801 ztest_record_enospc("dmu_objset_clone");
4804 fatal(0, "dmu_objset_clone(%s) = %d", clonename
, error
);
4807 error
= dsl_destroy_snapshot(fullname
, B_TRUE
);
4809 fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
4813 error
= dsl_destroy_head(clonename
);
4815 fatal(0, "dsl_destroy_head(%s) = %d", clonename
, error
);
4817 error
= dmu_objset_hold(fullname
, FTAG
, &origin
);
4818 if (error
!= ENOENT
)
4819 fatal(0, "dmu_objset_hold(%s) = %d", fullname
, error
);
4822 * Create snapshot, add temporary hold, verify that we can't
4823 * destroy a held snapshot, mark for deferred destroy,
4824 * release hold, verify snapshot was destroyed.
4826 error
= dmu_objset_snapshot_one(osname
, snapname
);
4828 if (error
== ENOSPC
) {
4829 ztest_record_enospc("dmu_objset_snapshot");
4832 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname
, error
);
4835 holds
= fnvlist_alloc();
4836 fnvlist_add_string(holds
, fullname
, tag
);
4837 error
= dsl_dataset_user_hold(holds
, 0, NULL
);
4838 fnvlist_free(holds
);
4840 if (error
== ENOSPC
) {
4841 ztest_record_enospc("dsl_dataset_user_hold");
4844 fatal(0, "dsl_dataset_user_hold(%s, %s) = %u",
4845 fullname
, tag
, error
);
4848 error
= dsl_destroy_snapshot(fullname
, B_FALSE
);
4849 if (error
!= EBUSY
) {
4850 fatal(0, "dsl_destroy_snapshot(%s, B_FALSE) = %d",
4854 error
= dsl_destroy_snapshot(fullname
, B_TRUE
);
4856 fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
4860 error
= user_release_one(fullname
, tag
);
4862 fatal(0, "user_release_one(%s, %s) = %d", fullname
, tag
, error
);
4864 VERIFY3U(dmu_objset_hold(fullname
, FTAG
, &origin
), ==, ENOENT
);
4867 rw_exit(&ztest_name_lock
);
4871 * Inject random faults into the on-disk data.
4875 ztest_fault_inject(ztest_ds_t
*zd
, uint64_t id
)
4877 ztest_shared_t
*zs
= ztest_shared
;
4878 spa_t
*spa
= ztest_spa
;
4882 uint64_t bad
= 0x1990c0ffeedecade;
4884 char path0
[MAXPATHLEN
];
4885 char pathrand
[MAXPATHLEN
];
4887 int bshift
= SPA_MAXBLOCKSHIFT
+ 2;
4893 boolean_t islog
= B_FALSE
;
4895 mutex_enter(&ztest_vdev_lock
);
4896 maxfaults
= MAXFAULTS();
4897 leaves
= MAX(zs
->zs_mirrors
, 1) * ztest_opts
.zo_raidz
;
4898 mirror_save
= zs
->zs_mirrors
;
4899 mutex_exit(&ztest_vdev_lock
);
4901 ASSERT(leaves
>= 1);
4904 * Grab the name lock as reader. There are some operations
4905 * which don't like to have their vdevs changed while
4906 * they are in progress (i.e. spa_change_guid). Those
4907 * operations will have grabbed the name lock as writer.
4909 rw_enter(&ztest_name_lock
, RW_READER
);
4912 * We need SCL_STATE here because we're going to look at vd0->vdev_tsd.
4914 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
4916 if (ztest_random(2) == 0) {
4918 * Inject errors on a normal data device or slog device.
4920 top
= ztest_random_vdev_top(spa
, B_TRUE
);
4921 leaf
= ztest_random(leaves
) + zs
->zs_splits
;
4924 * Generate paths to the first leaf in this top-level vdev,
4925 * and to the random leaf we selected. We'll induce transient
4926 * write failures and random online/offline activity on leaf 0,
4927 * and we'll write random garbage to the randomly chosen leaf.
4929 (void) snprintf(path0
, sizeof (path0
), ztest_dev_template
,
4930 ztest_opts
.zo_dir
, ztest_opts
.zo_pool
,
4931 top
* leaves
+ zs
->zs_splits
);
4932 (void) snprintf(pathrand
, sizeof (pathrand
), ztest_dev_template
,
4933 ztest_opts
.zo_dir
, ztest_opts
.zo_pool
,
4934 top
* leaves
+ leaf
);
4936 vd0
= vdev_lookup_by_path(spa
->spa_root_vdev
, path0
);
4937 if (vd0
!= NULL
&& vd0
->vdev_top
->vdev_islog
)
4941 * If the top-level vdev needs to be resilvered
4942 * then we only allow faults on the device that is
4945 if (vd0
!= NULL
&& maxfaults
!= 1 &&
4946 (!vdev_resilver_needed(vd0
->vdev_top
, NULL
, NULL
) ||
4947 vd0
->vdev_resilver_txg
!= 0)) {
4949 * Make vd0 explicitly claim to be unreadable,
4950 * or unwriteable, or reach behind its back
4951 * and close the underlying fd. We can do this if
4952 * maxfaults == 0 because we'll fail and reexecute,
4953 * and we can do it if maxfaults >= 2 because we'll
4954 * have enough redundancy. If maxfaults == 1, the
4955 * combination of this with injection of random data
4956 * corruption below exceeds the pool's fault tolerance.
4958 vdev_file_t
*vf
= vd0
->vdev_tsd
;
4960 zfs_dbgmsg("injecting fault to vdev %llu; maxfaults=%d",
4961 (long long)vd0
->vdev_id
, (int)maxfaults
);
4963 if (vf
!= NULL
&& ztest_random(3) == 0) {
4964 (void) close(vf
->vf_vnode
->v_fd
);
4965 vf
->vf_vnode
->v_fd
= -1;
4966 } else if (ztest_random(2) == 0) {
4967 vd0
->vdev_cant_read
= B_TRUE
;
4969 vd0
->vdev_cant_write
= B_TRUE
;
4971 guid0
= vd0
->vdev_guid
;
4975 * Inject errors on an l2cache device.
4977 spa_aux_vdev_t
*sav
= &spa
->spa_l2cache
;
4979 if (sav
->sav_count
== 0) {
4980 spa_config_exit(spa
, SCL_STATE
, FTAG
);
4981 rw_exit(&ztest_name_lock
);
4984 vd0
= sav
->sav_vdevs
[ztest_random(sav
->sav_count
)];
4985 guid0
= vd0
->vdev_guid
;
4986 (void) strcpy(path0
, vd0
->vdev_path
);
4987 (void) strcpy(pathrand
, vd0
->vdev_path
);
4991 maxfaults
= INT_MAX
; /* no limit on cache devices */
4994 spa_config_exit(spa
, SCL_STATE
, FTAG
);
4995 rw_exit(&ztest_name_lock
);
4998 * If we can tolerate two or more faults, or we're dealing
4999 * with a slog, randomly online/offline vd0.
5001 if ((maxfaults
>= 2 || islog
) && guid0
!= 0) {
5002 if (ztest_random(10) < 6) {
5003 int flags
= (ztest_random(2) == 0 ?
5004 ZFS_OFFLINE_TEMPORARY
: 0);
5007 * We have to grab the zs_name_lock as writer to
5008 * prevent a race between offlining a slog and
5009 * destroying a dataset. Offlining the slog will
5010 * grab a reference on the dataset which may cause
5011 * dmu_objset_destroy() to fail with EBUSY thus
5012 * leaving the dataset in an inconsistent state.
5015 rw_enter(&ztest_name_lock
, RW_WRITER
);
5017 VERIFY(vdev_offline(spa
, guid0
, flags
) != EBUSY
);
5020 rw_exit(&ztest_name_lock
);
5023 * Ideally we would like to be able to randomly
5024 * call vdev_[on|off]line without holding locks
5025 * to force unpredictable failures but the side
5026 * effects of vdev_[on|off]line prevent us from
5027 * doing so. We grab the ztest_vdev_lock here to
5028 * prevent a race between injection testing and
5031 mutex_enter(&ztest_vdev_lock
);
5032 (void) vdev_online(spa
, guid0
, 0, NULL
);
5033 mutex_exit(&ztest_vdev_lock
);
5041 * We have at least single-fault tolerance, so inject data corruption.
5043 fd
= open(pathrand
, O_RDWR
);
5045 if (fd
== -1) /* we hit a gap in the device namespace */
5048 fsize
= lseek(fd
, 0, SEEK_END
);
5050 while (--iters
!= 0) {
5052 * The offset must be chosen carefully to ensure that
5053 * we do not inject a given logical block with errors
5054 * on two different leaf devices, because ZFS can not
5055 * tolerate that (if maxfaults==1).
5057 * We divide each leaf into chunks of size
5058 * (# leaves * SPA_MAXBLOCKSIZE * 4). Within each chunk
5059 * there is a series of ranges to which we can inject errors.
5060 * Each range can accept errors on only a single leaf vdev.
5061 * The error injection ranges are separated by ranges
5062 * which we will not inject errors on any device (DMZs).
5063 * Each DMZ must be large enough such that a single block
5064 * can not straddle it, so that a single block can not be
5065 * a target in two different injection ranges (on different
5068 * For example, with 3 leaves, each chunk looks like:
5069 * 0 to 32M: injection range for leaf 0
5070 * 32M to 64M: DMZ - no injection allowed
5071 * 64M to 96M: injection range for leaf 1
5072 * 96M to 128M: DMZ - no injection allowed
5073 * 128M to 160M: injection range for leaf 2
5074 * 160M to 192M: DMZ - no injection allowed
5076 offset
= ztest_random(fsize
/ (leaves
<< bshift
)) *
5077 (leaves
<< bshift
) + (leaf
<< bshift
) +
5078 (ztest_random(1ULL << (bshift
- 1)) & -8ULL);
5081 * Only allow damage to the labels at one end of the vdev.
5083 * If all labels are damaged, the device will be totally
5084 * inaccessible, which will result in loss of data,
5085 * because we also damage (parts of) the other side of
5088 * Additionally, we will always have both an even and an
5089 * odd label, so that we can handle crashes in the
5090 * middle of vdev_config_sync().
5092 if ((leaf
& 1) == 0 && offset
< VDEV_LABEL_START_SIZE
)
5096 * The two end labels are stored at the "end" of the disk, but
5097 * the end of the disk (vdev_psize) is aligned to
5098 * sizeof (vdev_label_t).
5100 uint64_t psize
= P2ALIGN(fsize
, sizeof (vdev_label_t
));
5101 if ((leaf
& 1) == 1 &&
5102 offset
+ sizeof (bad
) > psize
- VDEV_LABEL_END_SIZE
)
5105 mutex_enter(&ztest_vdev_lock
);
5106 if (mirror_save
!= zs
->zs_mirrors
) {
5107 mutex_exit(&ztest_vdev_lock
);
5112 if (pwrite(fd
, &bad
, sizeof (bad
), offset
) != sizeof (bad
))
5113 fatal(1, "can't inject bad word at 0x%llx in %s",
5116 mutex_exit(&ztest_vdev_lock
);
5118 if (ztest_opts
.zo_verbose
>= 7)
5119 (void) printf("injected bad word into %s,"
5120 " offset 0x%llx\n", pathrand
, (u_longlong_t
)offset
);
5127 * Verify that DDT repair works as expected.
5130 ztest_ddt_repair(ztest_ds_t
*zd
, uint64_t id
)
5132 ztest_shared_t
*zs
= ztest_shared
;
5133 spa_t
*spa
= ztest_spa
;
5134 objset_t
*os
= zd
->zd_os
;
5136 uint64_t object
, blocksize
, txg
, pattern
, psize
;
5137 enum zio_checksum checksum
= spa_dedup_checksum(spa
);
5142 int copies
= 2 * ZIO_DEDUPDITTO_MIN
;
5144 blocksize
= ztest_random_blocksize();
5145 blocksize
= MIN(blocksize
, 2048); /* because we write so many */
5147 ztest_od_init(&od
[0], id
, FTAG
, 0, DMU_OT_UINT64_OTHER
, blocksize
, 0);
5149 if (ztest_object_init(zd
, od
, sizeof (od
), B_FALSE
) != 0)
5153 * Take the name lock as writer to prevent anyone else from changing
5154 * the pool and dataset properies we need to maintain during this test.
5156 rw_enter(&ztest_name_lock
, RW_WRITER
);
5158 if (ztest_dsl_prop_set_uint64(zd
->zd_name
, ZFS_PROP_DEDUP
, checksum
,
5160 ztest_dsl_prop_set_uint64(zd
->zd_name
, ZFS_PROP_COPIES
, 1,
5162 rw_exit(&ztest_name_lock
);
5166 dmu_objset_stats_t dds
;
5167 dsl_pool_config_enter(dmu_objset_pool(os
), FTAG
);
5168 dmu_objset_fast_stat(os
, &dds
);
5169 dsl_pool_config_exit(dmu_objset_pool(os
), FTAG
);
5171 object
= od
[0].od_object
;
5172 blocksize
= od
[0].od_blocksize
;
5173 pattern
= zs
->zs_guid
^ dds
.dds_guid
;
5175 ASSERT(object
!= 0);
5177 tx
= dmu_tx_create(os
);
5178 dmu_tx_hold_write(tx
, object
, 0, copies
* blocksize
);
5179 txg
= ztest_tx_assign(tx
, TXG_WAIT
, FTAG
);
5181 rw_exit(&ztest_name_lock
);
5186 * Write all the copies of our block.
5188 for (int i
= 0; i
< copies
; i
++) {
5189 uint64_t offset
= i
* blocksize
;
5190 int error
= dmu_buf_hold(os
, object
, offset
, FTAG
, &db
,
5191 DMU_READ_NO_PREFETCH
);
5193 fatal(B_FALSE
, "dmu_buf_hold(%p, %llu, %llu) = %u",
5194 os
, (long long)object
, (long long) offset
, error
);
5196 ASSERT(db
->db_offset
== offset
);
5197 ASSERT(db
->db_size
== blocksize
);
5198 ASSERT(ztest_pattern_match(db
->db_data
, db
->db_size
, pattern
) ||
5199 ztest_pattern_match(db
->db_data
, db
->db_size
, 0ULL));
5200 dmu_buf_will_fill(db
, tx
);
5201 ztest_pattern_set(db
->db_data
, db
->db_size
, pattern
);
5202 dmu_buf_rele(db
, FTAG
);
5206 txg_wait_synced(spa_get_dsl(spa
), txg
);
5209 * Find out what block we got.
5211 VERIFY0(dmu_buf_hold(os
, object
, 0, FTAG
, &db
,
5212 DMU_READ_NO_PREFETCH
));
5213 blk
= *((dmu_buf_impl_t
*)db
)->db_blkptr
;
5214 dmu_buf_rele(db
, FTAG
);
5217 * Damage the block. Dedup-ditto will save us when we read it later.
5219 psize
= BP_GET_PSIZE(&blk
);
5220 abd
= abd_alloc_linear(psize
, B_TRUE
);
5221 ztest_pattern_set(abd_to_buf(abd
), psize
, ~pattern
);
5223 (void) zio_wait(zio_rewrite(NULL
, spa
, 0, &blk
,
5224 abd
, psize
, NULL
, NULL
, ZIO_PRIORITY_SYNC_WRITE
,
5225 ZIO_FLAG_CANFAIL
| ZIO_FLAG_INDUCE_DAMAGE
, NULL
));
5229 rw_exit(&ztest_name_lock
);
5237 ztest_scrub(ztest_ds_t
*zd
, uint64_t id
)
5239 spa_t
*spa
= ztest_spa
;
5241 (void) spa_scan(spa
, POOL_SCAN_SCRUB
);
5242 (void) poll(NULL
, 0, 100); /* wait a moment, then force a restart */
5243 (void) spa_scan(spa
, POOL_SCAN_SCRUB
);
5247 * Change the guid for the pool.
5251 ztest_reguid(ztest_ds_t
*zd
, uint64_t id
)
5253 spa_t
*spa
= ztest_spa
;
5254 uint64_t orig
, load
;
5257 orig
= spa_guid(spa
);
5258 load
= spa_load_guid(spa
);
5260 rw_enter(&ztest_name_lock
, RW_WRITER
);
5261 error
= spa_change_guid(spa
);
5262 rw_exit(&ztest_name_lock
);
5267 if (ztest_opts
.zo_verbose
>= 4) {
5268 (void) printf("Changed guid old %llu -> %llu\n",
5269 (u_longlong_t
)orig
, (u_longlong_t
)spa_guid(spa
));
5272 VERIFY3U(orig
, !=, spa_guid(spa
));
5273 VERIFY3U(load
, ==, spa_load_guid(spa
));
5277 * Rename the pool to a different name and then rename it back.
5281 ztest_spa_rename(ztest_ds_t
*zd
, uint64_t id
)
5283 char *oldname
, *newname
;
5286 rw_enter(&ztest_name_lock
, RW_WRITER
);
5288 oldname
= ztest_opts
.zo_pool
;
5289 newname
= umem_alloc(strlen(oldname
) + 5, UMEM_NOFAIL
);
5290 (void) strcpy(newname
, oldname
);
5291 (void) strcat(newname
, "_tmp");
5296 VERIFY3U(0, ==, spa_rename(oldname
, newname
));
5299 * Try to open it under the old name, which shouldn't exist
5301 VERIFY3U(ENOENT
, ==, spa_open(oldname
, &spa
, FTAG
));
5304 * Open it under the new name and make sure it's still the same spa_t.
5306 VERIFY3U(0, ==, spa_open(newname
, &spa
, FTAG
));
5308 ASSERT(spa
== ztest_spa
);
5309 spa_close(spa
, FTAG
);
5312 * Rename it back to the original
5314 VERIFY3U(0, ==, spa_rename(newname
, oldname
));
5317 * Make sure it can still be opened
5319 VERIFY3U(0, ==, spa_open(oldname
, &spa
, FTAG
));
5321 ASSERT(spa
== ztest_spa
);
5322 spa_close(spa
, FTAG
);
5324 umem_free(newname
, strlen(newname
) + 1);
5326 rw_exit(&ztest_name_lock
);
5330 * Verify pool integrity by running zdb.
5333 ztest_run_zdb(char *pool
)
5336 char zdb
[MAXPATHLEN
+ MAXNAMELEN
+ 20];
5344 (void) realpath(getexecname(), zdb
);
5346 /* zdb lives in /usr/sbin, while ztest lives in /usr/bin */
5347 bin
= strstr(zdb
, "/usr/bin/");
5348 ztest
= strstr(bin
, "/ztest");
5350 isalen
= ztest
- isa
;
5354 "/usr/sbin%.*s/zdb -bcc%s%s -G -d -U %s %s",
5357 ztest_opts
.zo_verbose
>= 3 ? "s" : "",
5358 ztest_opts
.zo_verbose
>= 4 ? "v" : "",
5363 if (ztest_opts
.zo_verbose
>= 5)
5364 (void) printf("Executing %s\n", strstr(zdb
, "zdb "));
5366 fp
= popen(zdb
, "r");
5368 while (fgets(zbuf
, sizeof (zbuf
), fp
) != NULL
)
5369 if (ztest_opts
.zo_verbose
>= 3)
5370 (void) printf("%s", zbuf
);
5372 status
= pclose(fp
);
5377 ztest_dump_core
= 0;
5378 if (WIFEXITED(status
))
5379 fatal(0, "'%s' exit code %d", zdb
, WEXITSTATUS(status
));
5381 fatal(0, "'%s' died with signal %d", zdb
, WTERMSIG(status
));
5385 ztest_walk_pool_directory(char *header
)
5389 if (ztest_opts
.zo_verbose
>= 6)
5390 (void) printf("%s\n", header
);
5392 mutex_enter(&spa_namespace_lock
);
5393 while ((spa
= spa_next(spa
)) != NULL
)
5394 if (ztest_opts
.zo_verbose
>= 6)
5395 (void) printf("\t%s\n", spa_name(spa
));
5396 mutex_exit(&spa_namespace_lock
);
5400 ztest_spa_import_export(char *oldname
, char *newname
)
5402 nvlist_t
*config
, *newconfig
;
5407 if (ztest_opts
.zo_verbose
>= 4) {
5408 (void) printf("import/export: old = %s, new = %s\n",
5413 * Clean up from previous runs.
5415 (void) spa_destroy(newname
);
5418 * Get the pool's configuration and guid.
5420 VERIFY3U(0, ==, spa_open(oldname
, &spa
, FTAG
));
5423 * Kick off a scrub to tickle scrub/export races.
5425 if (ztest_random(2) == 0)
5426 (void) spa_scan(spa
, POOL_SCAN_SCRUB
);
5428 pool_guid
= spa_guid(spa
);
5429 spa_close(spa
, FTAG
);
5431 ztest_walk_pool_directory("pools before export");
5436 VERIFY3U(0, ==, spa_export(oldname
, &config
, B_FALSE
, B_FALSE
));
5438 ztest_walk_pool_directory("pools after export");
5443 newconfig
= spa_tryimport(config
);
5444 ASSERT(newconfig
!= NULL
);
5445 nvlist_free(newconfig
);
5448 * Import it under the new name.
5450 error
= spa_import(newname
, config
, NULL
, 0);
5452 dump_nvlist(config
, 0);
5453 fatal(B_FALSE
, "couldn't import pool %s as %s: error %u",
5454 oldname
, newname
, error
);
5457 ztest_walk_pool_directory("pools after import");
5460 * Try to import it again -- should fail with EEXIST.
5462 VERIFY3U(EEXIST
, ==, spa_import(newname
, config
, NULL
, 0));
5465 * Try to import it under a different name -- should fail with EEXIST.
5467 VERIFY3U(EEXIST
, ==, spa_import(oldname
, config
, NULL
, 0));
5470 * Verify that the pool is no longer visible under the old name.
5472 VERIFY3U(ENOENT
, ==, spa_open(oldname
, &spa
, FTAG
));
5475 * Verify that we can open and close the pool using the new name.
5477 VERIFY3U(0, ==, spa_open(newname
, &spa
, FTAG
));
5478 ASSERT(pool_guid
== spa_guid(spa
));
5479 spa_close(spa
, FTAG
);
5481 nvlist_free(config
);
5485 ztest_resume(spa_t
*spa
)
5487 if (spa_suspended(spa
) && ztest_opts
.zo_verbose
>= 6)
5488 (void) printf("resuming from suspended state\n");
5489 spa_vdev_state_enter(spa
, SCL_NONE
);
5490 vdev_clear(spa
, NULL
);
5491 (void) spa_vdev_state_exit(spa
, NULL
, 0);
5492 (void) zio_resume(spa
);
5496 ztest_resume_thread(void *arg
)
5500 while (!ztest_exiting
) {
5501 if (spa_suspended(spa
))
5503 (void) poll(NULL
, 0, 100);
5506 * Periodically change the zfs_compressed_arc_enabled setting.
5508 if (ztest_random(10) == 0)
5509 zfs_compressed_arc_enabled
= ztest_random(2);
5512 * Periodically change the zfs_abd_scatter_enabled setting.
5514 if (ztest_random(10) == 0)
5515 zfs_abd_scatter_enabled
= ztest_random(2);
5521 ztest_deadman_thread(void *arg
)
5523 ztest_shared_t
*zs
= arg
;
5524 spa_t
*spa
= ztest_spa
;
5525 hrtime_t delta
, total
= 0;
5528 delta
= zs
->zs_thread_stop
- zs
->zs_thread_start
+
5529 MSEC2NSEC(zfs_deadman_synctime_ms
);
5531 (void) poll(NULL
, 0, (int)NSEC2MSEC(delta
));
5534 * If the pool is suspended then fail immediately. Otherwise,
5535 * check to see if the pool is making any progress. If
5536 * vdev_deadman() discovers that there hasn't been any recent
5537 * I/Os then it will end up aborting the tests.
5539 if (spa_suspended(spa
) || spa
->spa_root_vdev
== NULL
) {
5540 fatal(0, "aborting test after %llu seconds because "
5541 "pool has transitioned to a suspended state.",
5542 zfs_deadman_synctime_ms
/ 1000);
5545 vdev_deadman(spa
->spa_root_vdev
);
5547 total
+= zfs_deadman_synctime_ms
/1000;
5548 (void) printf("ztest has been running for %lld seconds\n",
5554 ztest_execute(int test
, ztest_info_t
*zi
, uint64_t id
)
5556 ztest_ds_t
*zd
= &ztest_ds
[id
% ztest_opts
.zo_datasets
];
5557 ztest_shared_callstate_t
*zc
= ZTEST_GET_SHARED_CALLSTATE(test
);
5558 hrtime_t functime
= gethrtime();
5560 for (int i
= 0; i
< zi
->zi_iters
; i
++)
5561 zi
->zi_func(zd
, id
);
5563 functime
= gethrtime() - functime
;
5565 atomic_add_64(&zc
->zc_count
, 1);
5566 atomic_add_64(&zc
->zc_time
, functime
);
5568 if (ztest_opts
.zo_verbose
>= 4) {
5570 (void) dladdr((void *)zi
->zi_func
, &dli
);
5571 (void) printf("%6.2f sec in %s\n",
5572 (double)functime
/ NANOSEC
, dli
.dli_sname
);
5577 ztest_thread(void *arg
)
5580 uint64_t id
= (uintptr_t)arg
;
5581 ztest_shared_t
*zs
= ztest_shared
;
5585 ztest_shared_callstate_t
*zc
;
5587 while ((now
= gethrtime()) < zs
->zs_thread_stop
) {
5589 * See if it's time to force a crash.
5591 if (now
> zs
->zs_thread_kill
)
5595 * If we're getting ENOSPC with some regularity, stop.
5597 if (zs
->zs_enospc_count
> 10)
5601 * Pick a random function to execute.
5603 rand
= ztest_random(ZTEST_FUNCS
);
5604 zi
= &ztest_info
[rand
];
5605 zc
= ZTEST_GET_SHARED_CALLSTATE(rand
);
5606 call_next
= zc
->zc_next
;
5608 if (now
>= call_next
&&
5609 atomic_cas_64(&zc
->zc_next
, call_next
, call_next
+
5610 ztest_random(2 * zi
->zi_interval
[0] + 1)) == call_next
) {
5611 ztest_execute(rand
, zi
, id
);
5619 ztest_dataset_name(char *dsname
, char *pool
, int d
)
5621 (void) snprintf(dsname
, ZFS_MAX_DATASET_NAME_LEN
, "%s/ds_%d", pool
, d
);
5625 ztest_dataset_destroy(int d
)
5627 char name
[ZFS_MAX_DATASET_NAME_LEN
];
5629 ztest_dataset_name(name
, ztest_opts
.zo_pool
, d
);
5631 if (ztest_opts
.zo_verbose
>= 3)
5632 (void) printf("Destroying %s to free up space\n", name
);
5635 * Cleanup any non-standard clones and snapshots. In general,
5636 * ztest thread t operates on dataset (t % zopt_datasets),
5637 * so there may be more than one thing to clean up.
5639 for (int t
= d
; t
< ztest_opts
.zo_threads
;
5640 t
+= ztest_opts
.zo_datasets
) {
5641 ztest_dsl_dataset_cleanup(name
, t
);
5644 (void) dmu_objset_find(name
, ztest_objset_destroy_cb
, NULL
,
5645 DS_FIND_SNAPSHOTS
| DS_FIND_CHILDREN
);
5649 ztest_dataset_dirobj_verify(ztest_ds_t
*zd
)
5651 uint64_t usedobjs
, dirobjs
, scratch
;
5654 * ZTEST_DIROBJ is the object directory for the entire dataset.
5655 * Therefore, the number of objects in use should equal the
5656 * number of ZTEST_DIROBJ entries, +1 for ZTEST_DIROBJ itself.
5657 * If not, we have an object leak.
5659 * Note that we can only check this in ztest_dataset_open(),
5660 * when the open-context and syncing-context values agree.
5661 * That's because zap_count() returns the open-context value,
5662 * while dmu_objset_space() returns the rootbp fill count.
5664 VERIFY3U(0, ==, zap_count(zd
->zd_os
, ZTEST_DIROBJ
, &dirobjs
));
5665 dmu_objset_space(zd
->zd_os
, &scratch
, &scratch
, &usedobjs
, &scratch
);
5666 ASSERT3U(dirobjs
+ 1, ==, usedobjs
);
5670 ztest_dataset_open(int d
)
5672 ztest_ds_t
*zd
= &ztest_ds
[d
];
5673 uint64_t committed_seq
= ZTEST_GET_SHARED_DS(d
)->zd_seq
;
5676 char name
[ZFS_MAX_DATASET_NAME_LEN
];
5679 ztest_dataset_name(name
, ztest_opts
.zo_pool
, d
);
5681 rw_enter(&ztest_name_lock
, RW_READER
);
5683 error
= ztest_dataset_create(name
);
5684 if (error
== ENOSPC
) {
5685 rw_exit(&ztest_name_lock
);
5686 ztest_record_enospc(FTAG
);
5689 ASSERT(error
== 0 || error
== EEXIST
);
5691 VERIFY0(dmu_objset_own(name
, DMU_OST_OTHER
, B_FALSE
, zd
, &os
));
5692 rw_exit(&ztest_name_lock
);
5694 ztest_zd_init(zd
, ZTEST_GET_SHARED_DS(d
), os
);
5696 zilog
= zd
->zd_zilog
;
5698 if (zilog
->zl_header
->zh_claim_lr_seq
!= 0 &&
5699 zilog
->zl_header
->zh_claim_lr_seq
< committed_seq
)
5700 fatal(0, "missing log records: claimed %llu < committed %llu",
5701 zilog
->zl_header
->zh_claim_lr_seq
, committed_seq
);
5703 ztest_dataset_dirobj_verify(zd
);
5705 zil_replay(os
, zd
, ztest_replay_vector
);
5707 ztest_dataset_dirobj_verify(zd
);
5709 if (ztest_opts
.zo_verbose
>= 6)
5710 (void) printf("%s replay %llu blocks, %llu records, seq %llu\n",
5712 (u_longlong_t
)zilog
->zl_parse_blk_count
,
5713 (u_longlong_t
)zilog
->zl_parse_lr_count
,
5714 (u_longlong_t
)zilog
->zl_replaying_seq
);
5716 zilog
= zil_open(os
, ztest_get_data
);
5718 if (zilog
->zl_replaying_seq
!= 0 &&
5719 zilog
->zl_replaying_seq
< committed_seq
)
5720 fatal(0, "missing log records: replayed %llu < committed %llu",
5721 zilog
->zl_replaying_seq
, committed_seq
);
5727 ztest_dataset_close(int d
)
5729 ztest_ds_t
*zd
= &ztest_ds
[d
];
5731 zil_close(zd
->zd_zilog
);
5732 dmu_objset_disown(zd
->zd_os
, zd
);
5738 * Kick off threads to run tests on all datasets in parallel.
5741 ztest_run(ztest_shared_t
*zs
)
5746 thread_t resume_tid
;
5749 ztest_exiting
= B_FALSE
;
5752 * Initialize parent/child shared state.
5754 mutex_init(&ztest_vdev_lock
, NULL
, USYNC_THREAD
, NULL
);
5755 rw_init(&ztest_name_lock
, NULL
, USYNC_THREAD
, NULL
);
5757 zs
->zs_thread_start
= gethrtime();
5758 zs
->zs_thread_stop
=
5759 zs
->zs_thread_start
+ ztest_opts
.zo_passtime
* NANOSEC
;
5760 zs
->zs_thread_stop
= MIN(zs
->zs_thread_stop
, zs
->zs_proc_stop
);
5761 zs
->zs_thread_kill
= zs
->zs_thread_stop
;
5762 if (ztest_random(100) < ztest_opts
.zo_killrate
) {
5763 zs
->zs_thread_kill
-=
5764 ztest_random(ztest_opts
.zo_passtime
* NANOSEC
);
5767 mutex_init(&zcl
.zcl_callbacks_lock
, NULL
, USYNC_THREAD
, NULL
);
5769 list_create(&zcl
.zcl_callbacks
, sizeof (ztest_cb_data_t
),
5770 offsetof(ztest_cb_data_t
, zcd_node
));
5775 kernel_init(FREAD
| FWRITE
);
5776 VERIFY0(spa_open(ztest_opts
.zo_pool
, &spa
, FTAG
));
5777 spa
->spa_debug
= B_TRUE
;
5778 metaslab_preload_limit
= ztest_random(20) + 1;
5781 dmu_objset_stats_t dds
;
5782 VERIFY0(dmu_objset_own(ztest_opts
.zo_pool
,
5783 DMU_OST_ANY
, B_TRUE
, FTAG
, &os
));
5784 dsl_pool_config_enter(dmu_objset_pool(os
), FTAG
);
5785 dmu_objset_fast_stat(os
, &dds
);
5786 dsl_pool_config_exit(dmu_objset_pool(os
), FTAG
);
5787 zs
->zs_guid
= dds
.dds_guid
;
5788 dmu_objset_disown(os
, FTAG
);
5790 spa
->spa_dedup_ditto
= 2 * ZIO_DEDUPDITTO_MIN
;
5793 * We don't expect the pool to suspend unless maxfaults == 0,
5794 * in which case ztest_fault_inject() temporarily takes away
5795 * the only valid replica.
5797 if (MAXFAULTS() == 0)
5798 spa
->spa_failmode
= ZIO_FAILURE_MODE_WAIT
;
5800 spa
->spa_failmode
= ZIO_FAILURE_MODE_PANIC
;
5803 * Create a thread to periodically resume suspended I/O.
5805 VERIFY(thr_create(0, 0, ztest_resume_thread
, spa
, THR_BOUND
,
5809 * Create a deadman thread to abort() if we hang.
5811 VERIFY(thr_create(0, 0, ztest_deadman_thread
, zs
, THR_BOUND
,
5815 * Verify that we can safely inquire about about any object,
5816 * whether it's allocated or not. To make it interesting,
5817 * we probe a 5-wide window around each power of two.
5818 * This hits all edge cases, including zero and the max.
5820 for (int t
= 0; t
< 64; t
++) {
5821 for (int d
= -5; d
<= 5; d
++) {
5822 error
= dmu_object_info(spa
->spa_meta_objset
,
5823 (1ULL << t
) + d
, NULL
);
5824 ASSERT(error
== 0 || error
== ENOENT
||
5830 * If we got any ENOSPC errors on the previous run, destroy something.
5832 if (zs
->zs_enospc_count
!= 0) {
5833 int d
= ztest_random(ztest_opts
.zo_datasets
);
5834 ztest_dataset_destroy(d
);
5836 zs
->zs_enospc_count
= 0;
5838 tid
= umem_zalloc(ztest_opts
.zo_threads
* sizeof (thread_t
),
5841 if (ztest_opts
.zo_verbose
>= 4)
5842 (void) printf("starting main threads...\n");
5845 * Kick off all the tests that run in parallel.
5847 for (int t
= 0; t
< ztest_opts
.zo_threads
; t
++) {
5848 if (t
< ztest_opts
.zo_datasets
&&
5849 ztest_dataset_open(t
) != 0)
5851 VERIFY(thr_create(0, 0, ztest_thread
, (void *)(uintptr_t)t
,
5852 THR_BOUND
, &tid
[t
]) == 0);
5856 * Wait for all of the tests to complete. We go in reverse order
5857 * so we don't close datasets while threads are still using them.
5859 for (int t
= ztest_opts
.zo_threads
- 1; t
>= 0; t
--) {
5860 VERIFY(thr_join(tid
[t
], NULL
, NULL
) == 0);
5861 if (t
< ztest_opts
.zo_datasets
)
5862 ztest_dataset_close(t
);
5865 txg_wait_synced(spa_get_dsl(spa
), 0);
5867 zs
->zs_alloc
= metaslab_class_get_alloc(spa_normal_class(spa
));
5868 zs
->zs_space
= metaslab_class_get_space(spa_normal_class(spa
));
5869 zfs_dbgmsg_print(FTAG
);
5871 umem_free(tid
, ztest_opts
.zo_threads
* sizeof (thread_t
));
5873 /* Kill the resume thread */
5874 ztest_exiting
= B_TRUE
;
5875 VERIFY(thr_join(resume_tid
, NULL
, NULL
) == 0);
5879 * Right before closing the pool, kick off a bunch of async I/O;
5880 * spa_close() should wait for it to complete.
5882 for (uint64_t object
= 1; object
< 50; object
++) {
5883 dmu_prefetch(spa
->spa_meta_objset
, object
, 0, 0, 1ULL << 20,
5884 ZIO_PRIORITY_SYNC_READ
);
5887 spa_close(spa
, FTAG
);
5890 * Verify that we can loop over all pools.
5892 mutex_enter(&spa_namespace_lock
);
5893 for (spa
= spa_next(NULL
); spa
!= NULL
; spa
= spa_next(spa
))
5894 if (ztest_opts
.zo_verbose
> 3)
5895 (void) printf("spa_next: found %s\n", spa_name(spa
));
5896 mutex_exit(&spa_namespace_lock
);
5899 * Verify that we can export the pool and reimport it under a
5902 if (ztest_random(2) == 0) {
5903 char name
[ZFS_MAX_DATASET_NAME_LEN
];
5904 (void) snprintf(name
, sizeof (name
), "%s_import",
5905 ztest_opts
.zo_pool
);
5906 ztest_spa_import_export(ztest_opts
.zo_pool
, name
);
5907 ztest_spa_import_export(name
, ztest_opts
.zo_pool
);
5912 list_destroy(&zcl
.zcl_callbacks
);
5914 mutex_destroy(&zcl
.zcl_callbacks_lock
);
5916 rw_destroy(&ztest_name_lock
);
5917 mutex_destroy(&ztest_vdev_lock
);
5923 ztest_ds_t
*zd
= &ztest_ds
[0];
5927 if (ztest_opts
.zo_verbose
>= 3)
5928 (void) printf("testing spa_freeze()...\n");
5930 kernel_init(FREAD
| FWRITE
);
5931 VERIFY3U(0, ==, spa_open(ztest_opts
.zo_pool
, &spa
, FTAG
));
5932 VERIFY3U(0, ==, ztest_dataset_open(0));
5933 spa
->spa_debug
= B_TRUE
;
5937 * Force the first log block to be transactionally allocated.
5938 * We have to do this before we freeze the pool -- otherwise
5939 * the log chain won't be anchored.
5941 while (BP_IS_HOLE(&zd
->zd_zilog
->zl_header
->zh_log
)) {
5942 ztest_dmu_object_alloc_free(zd
, 0);
5943 zil_commit(zd
->zd_zilog
, 0);
5946 txg_wait_synced(spa_get_dsl(spa
), 0);
5949 * Freeze the pool. This stops spa_sync() from doing anything,
5950 * so that the only way to record changes from now on is the ZIL.
5955 * Because it is hard to predict how much space a write will actually
5956 * require beforehand, we leave ourselves some fudge space to write over
5959 uint64_t capacity
= metaslab_class_get_space(spa_normal_class(spa
)) / 2;
5962 * Run tests that generate log records but don't alter the pool config
5963 * or depend on DSL sync tasks (snapshots, objset create/destroy, etc).
5964 * We do a txg_wait_synced() after each iteration to force the txg
5965 * to increase well beyond the last synced value in the uberblock.
5966 * The ZIL should be OK with that.
5968 * Run a random number of times less than zo_maxloops and ensure we do
5969 * not run out of space on the pool.
5971 while (ztest_random(10) != 0 &&
5972 numloops
++ < ztest_opts
.zo_maxloops
&&
5973 metaslab_class_get_alloc(spa_normal_class(spa
)) < capacity
) {
5975 ztest_od_init(&od
, 0, FTAG
, 0, DMU_OT_UINT64_OTHER
, 0, 0);
5976 VERIFY0(ztest_object_init(zd
, &od
, sizeof (od
), B_FALSE
));
5977 ztest_io(zd
, od
.od_object
,
5978 ztest_random(ZTEST_RANGE_LOCKS
) << SPA_MAXBLOCKSHIFT
);
5979 txg_wait_synced(spa_get_dsl(spa
), 0);
5983 * Commit all of the changes we just generated.
5985 zil_commit(zd
->zd_zilog
, 0);
5986 txg_wait_synced(spa_get_dsl(spa
), 0);
5989 * Close our dataset and close the pool.
5991 ztest_dataset_close(0);
5992 spa_close(spa
, FTAG
);
5996 * Open and close the pool and dataset to induce log replay.
5998 kernel_init(FREAD
| FWRITE
);
5999 VERIFY3U(0, ==, spa_open(ztest_opts
.zo_pool
, &spa
, FTAG
));
6000 ASSERT(spa_freeze_txg(spa
) == UINT64_MAX
);
6001 VERIFY3U(0, ==, ztest_dataset_open(0));
6002 ztest_dataset_close(0);
6004 spa
->spa_debug
= B_TRUE
;
6006 txg_wait_synced(spa_get_dsl(spa
), 0);
6007 ztest_reguid(NULL
, 0);
6009 spa_close(spa
, FTAG
);
6014 print_time(hrtime_t t
, char *timebuf
)
6016 hrtime_t s
= t
/ NANOSEC
;
6017 hrtime_t m
= s
/ 60;
6018 hrtime_t h
= m
/ 60;
6019 hrtime_t d
= h
/ 24;
6028 (void) sprintf(timebuf
,
6029 "%llud%02lluh%02llum%02llus", d
, h
, m
, s
);
6031 (void) sprintf(timebuf
, "%lluh%02llum%02llus", h
, m
, s
);
6033 (void) sprintf(timebuf
, "%llum%02llus", m
, s
);
6035 (void) sprintf(timebuf
, "%llus", s
);
6043 VERIFY(nvlist_alloc(&props
, NV_UNIQUE_NAME
, 0) == 0);
6044 if (ztest_random(2) == 0)
6046 VERIFY(nvlist_add_uint64(props
, "autoreplace", 1) == 0);
6052 * Create a storage pool with the given name and initial vdev size.
6053 * Then test spa_freeze() functionality.
6056 ztest_init(ztest_shared_t
*zs
)
6059 nvlist_t
*nvroot
, *props
;
6061 mutex_init(&ztest_vdev_lock
, NULL
, USYNC_THREAD
, NULL
);
6062 rw_init(&ztest_name_lock
, NULL
, USYNC_THREAD
, NULL
);
6064 kernel_init(FREAD
| FWRITE
);
6067 * Create the storage pool.
6069 (void) spa_destroy(ztest_opts
.zo_pool
);
6070 ztest_shared
->zs_vdev_next_leaf
= 0;
6072 zs
->zs_mirrors
= ztest_opts
.zo_mirrors
;
6073 nvroot
= make_vdev_root(NULL
, NULL
, NULL
, ztest_opts
.zo_vdev_size
, 0,
6074 0, ztest_opts
.zo_raidz
, zs
->zs_mirrors
, 1);
6075 props
= make_random_props();
6076 for (int i
= 0; i
< SPA_FEATURES
; i
++) {
6078 (void) snprintf(buf
, sizeof (buf
), "feature@%s",
6079 spa_feature_table
[i
].fi_uname
);
6080 VERIFY3U(0, ==, nvlist_add_uint64(props
, buf
, 0));
6082 VERIFY3U(0, ==, spa_create(ztest_opts
.zo_pool
, nvroot
, props
, NULL
));
6083 nvlist_free(nvroot
);
6086 VERIFY3U(0, ==, spa_open(ztest_opts
.zo_pool
, &spa
, FTAG
));
6087 zs
->zs_metaslab_sz
=
6088 1ULL << spa
->spa_root_vdev
->vdev_child
[0]->vdev_ms_shift
;
6090 spa_close(spa
, FTAG
);
6094 ztest_run_zdb(ztest_opts
.zo_pool
);
6098 ztest_run_zdb(ztest_opts
.zo_pool
);
6100 rw_destroy(&ztest_name_lock
);
6101 mutex_destroy(&ztest_vdev_lock
);
6107 static char ztest_name_data
[] = "/tmp/ztest.data.XXXXXX";
6109 ztest_fd_data
= mkstemp(ztest_name_data
);
6110 ASSERT3S(ztest_fd_data
, >=, 0);
6111 (void) unlink(ztest_name_data
);
6116 shared_data_size(ztest_shared_hdr_t
*hdr
)
6120 size
= hdr
->zh_hdr_size
;
6121 size
+= hdr
->zh_opts_size
;
6122 size
+= hdr
->zh_size
;
6123 size
+= hdr
->zh_stats_size
* hdr
->zh_stats_count
;
6124 size
+= hdr
->zh_ds_size
* hdr
->zh_ds_count
;
6133 ztest_shared_hdr_t
*hdr
;
6135 hdr
= (void *)mmap(0, P2ROUNDUP(sizeof (*hdr
), getpagesize()),
6136 PROT_READ
| PROT_WRITE
, MAP_SHARED
, ztest_fd_data
, 0);
6137 ASSERT(hdr
!= MAP_FAILED
);
6139 VERIFY3U(0, ==, ftruncate(ztest_fd_data
, sizeof (ztest_shared_hdr_t
)));
6141 hdr
->zh_hdr_size
= sizeof (ztest_shared_hdr_t
);
6142 hdr
->zh_opts_size
= sizeof (ztest_shared_opts_t
);
6143 hdr
->zh_size
= sizeof (ztest_shared_t
);
6144 hdr
->zh_stats_size
= sizeof (ztest_shared_callstate_t
);
6145 hdr
->zh_stats_count
= ZTEST_FUNCS
;
6146 hdr
->zh_ds_size
= sizeof (ztest_shared_ds_t
);
6147 hdr
->zh_ds_count
= ztest_opts
.zo_datasets
;
6149 size
= shared_data_size(hdr
);
6150 VERIFY3U(0, ==, ftruncate(ztest_fd_data
, size
));
6152 (void) munmap((caddr_t
)hdr
, P2ROUNDUP(sizeof (*hdr
), getpagesize()));
6159 ztest_shared_hdr_t
*hdr
;
6162 hdr
= (void *)mmap(0, P2ROUNDUP(sizeof (*hdr
), getpagesize()),
6163 PROT_READ
, MAP_SHARED
, ztest_fd_data
, 0);
6164 ASSERT(hdr
!= MAP_FAILED
);
6166 size
= shared_data_size(hdr
);
6168 (void) munmap((caddr_t
)hdr
, P2ROUNDUP(sizeof (*hdr
), getpagesize()));
6169 hdr
= ztest_shared_hdr
= (void *)mmap(0, P2ROUNDUP(size
, getpagesize()),
6170 PROT_READ
| PROT_WRITE
, MAP_SHARED
, ztest_fd_data
, 0);
6171 ASSERT(hdr
!= MAP_FAILED
);
6172 buf
= (uint8_t *)hdr
;
6174 offset
= hdr
->zh_hdr_size
;
6175 ztest_shared_opts
= (void *)&buf
[offset
];
6176 offset
+= hdr
->zh_opts_size
;
6177 ztest_shared
= (void *)&buf
[offset
];
6178 offset
+= hdr
->zh_size
;
6179 ztest_shared_callstate
= (void *)&buf
[offset
];
6180 offset
+= hdr
->zh_stats_size
* hdr
->zh_stats_count
;
6181 ztest_shared_ds
= (void *)&buf
[offset
];
6185 exec_child(char *cmd
, char *libpath
, boolean_t ignorekill
, int *statusp
)
6189 char *cmdbuf
= NULL
;
6194 cmdbuf
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
6195 (void) strlcpy(cmdbuf
, getexecname(), MAXPATHLEN
);
6200 fatal(1, "fork failed");
6202 if (pid
== 0) { /* child */
6203 char *emptyargv
[2] = { cmd
, NULL
};
6204 char fd_data_str
[12];
6206 struct rlimit rl
= { 1024, 1024 };
6207 (void) setrlimit(RLIMIT_NOFILE
, &rl
);
6209 (void) close(ztest_fd_rand
);
6211 snprintf(fd_data_str
, 12, "%d", ztest_fd_data
));
6212 VERIFY0(setenv("ZTEST_FD_DATA", fd_data_str
, 1));
6214 (void) enable_extended_FILE_stdio(-1, -1);
6215 if (libpath
!= NULL
)
6216 VERIFY(0 == setenv("LD_LIBRARY_PATH", libpath
, 1));
6217 (void) execv(cmd
, emptyargv
);
6218 ztest_dump_core
= B_FALSE
;
6219 fatal(B_TRUE
, "exec failed: %s", cmd
);
6222 if (cmdbuf
!= NULL
) {
6223 umem_free(cmdbuf
, MAXPATHLEN
);
6227 while (waitpid(pid
, &status
, 0) != pid
)
6229 if (statusp
!= NULL
)
6232 if (WIFEXITED(status
)) {
6233 if (WEXITSTATUS(status
) != 0) {
6234 (void) fprintf(stderr
, "child exited with code %d\n",
6235 WEXITSTATUS(status
));
6239 } else if (WIFSIGNALED(status
)) {
6240 if (!ignorekill
|| WTERMSIG(status
) != SIGKILL
) {
6241 (void) fprintf(stderr
, "child died with signal %d\n",
6247 (void) fprintf(stderr
, "something strange happened to child\n");
6254 ztest_run_init(void)
6256 ztest_shared_t
*zs
= ztest_shared
;
6258 ASSERT(ztest_opts
.zo_init
!= 0);
6261 * Blow away any existing copy of zpool.cache
6263 (void) remove(spa_config_path
);
6266 * Create and initialize our storage pool.
6268 for (int i
= 1; i
<= ztest_opts
.zo_init
; i
++) {
6269 bzero(zs
, sizeof (ztest_shared_t
));
6270 if (ztest_opts
.zo_verbose
>= 3 &&
6271 ztest_opts
.zo_init
!= 1) {
6272 (void) printf("ztest_init(), pass %d\n", i
);
6279 main(int argc
, char **argv
)
6287 ztest_shared_callstate_t
*zc
;
6289 char numbuf
[NN_NUMBUF_SZ
];
6293 char *fd_data_str
= getenv("ZTEST_FD_DATA");
6295 (void) setvbuf(stdout
, NULL
, _IOLBF
, 0);
6297 dprintf_setup(&argc
, argv
);
6298 zfs_deadman_synctime_ms
= 300000;
6300 ztest_fd_rand
= open("/dev/urandom", O_RDONLY
);
6301 ASSERT3S(ztest_fd_rand
, >=, 0);
6304 process_options(argc
, argv
);
6309 bcopy(&ztest_opts
, ztest_shared_opts
,
6310 sizeof (*ztest_shared_opts
));
6312 ztest_fd_data
= atoi(fd_data_str
);
6314 bcopy(ztest_shared_opts
, &ztest_opts
, sizeof (ztest_opts
));
6316 ASSERT3U(ztest_opts
.zo_datasets
, ==, ztest_shared_hdr
->zh_ds_count
);
6318 /* Override location of zpool.cache */
6319 VERIFY3U(asprintf((char **)&spa_config_path
, "%s/zpool.cache",
6320 ztest_opts
.zo_dir
), !=, -1);
6322 ztest_ds
= umem_alloc(ztest_opts
.zo_datasets
* sizeof (ztest_ds_t
),
6327 metaslab_gang_bang
= ztest_opts
.zo_metaslab_gang_bang
;
6328 metaslab_df_alloc_threshold
=
6329 zs
->zs_metaslab_df_alloc_threshold
;
6338 hasalt
= (strlen(ztest_opts
.zo_alt_ztest
) != 0);
6340 if (ztest_opts
.zo_verbose
>= 1) {
6341 (void) printf("%llu vdevs, %d datasets, %d threads,"
6342 " %llu seconds...\n",
6343 (u_longlong_t
)ztest_opts
.zo_vdevs
,
6344 ztest_opts
.zo_datasets
,
6345 ztest_opts
.zo_threads
,
6346 (u_longlong_t
)ztest_opts
.zo_time
);
6349 cmd
= umem_alloc(MAXNAMELEN
, UMEM_NOFAIL
);
6350 (void) strlcpy(cmd
, getexecname(), MAXNAMELEN
);
6352 zs
->zs_do_init
= B_TRUE
;
6353 if (strlen(ztest_opts
.zo_alt_ztest
) != 0) {
6354 if (ztest_opts
.zo_verbose
>= 1) {
6355 (void) printf("Executing older ztest for "
6356 "initialization: %s\n", ztest_opts
.zo_alt_ztest
);
6358 VERIFY(!exec_child(ztest_opts
.zo_alt_ztest
,
6359 ztest_opts
.zo_alt_libpath
, B_FALSE
, NULL
));
6361 VERIFY(!exec_child(NULL
, NULL
, B_FALSE
, NULL
));
6363 zs
->zs_do_init
= B_FALSE
;
6365 zs
->zs_proc_start
= gethrtime();
6366 zs
->zs_proc_stop
= zs
->zs_proc_start
+ ztest_opts
.zo_time
* NANOSEC
;
6368 for (int f
= 0; f
< ZTEST_FUNCS
; f
++) {
6369 zi
= &ztest_info
[f
];
6370 zc
= ZTEST_GET_SHARED_CALLSTATE(f
);
6371 if (zs
->zs_proc_start
+ zi
->zi_interval
[0] > zs
->zs_proc_stop
)
6372 zc
->zc_next
= UINT64_MAX
;
6374 zc
->zc_next
= zs
->zs_proc_start
+
6375 ztest_random(2 * zi
->zi_interval
[0] + 1);
6379 * Run the tests in a loop. These tests include fault injection
6380 * to verify that self-healing data works, and forced crashes
6381 * to verify that we never lose on-disk consistency.
6383 while (gethrtime() < zs
->zs_proc_stop
) {
6388 * Initialize the workload counters for each function.
6390 for (int f
= 0; f
< ZTEST_FUNCS
; f
++) {
6391 zc
= ZTEST_GET_SHARED_CALLSTATE(f
);
6396 /* Set the allocation switch size */
6397 zs
->zs_metaslab_df_alloc_threshold
=
6398 ztest_random(zs
->zs_metaslab_sz
/ 4) + 1;
6400 if (!hasalt
|| ztest_random(2) == 0) {
6401 if (hasalt
&& ztest_opts
.zo_verbose
>= 1) {
6402 (void) printf("Executing newer ztest: %s\n",
6406 killed
= exec_child(cmd
, NULL
, B_TRUE
, &status
);
6408 if (hasalt
&& ztest_opts
.zo_verbose
>= 1) {
6409 (void) printf("Executing older ztest: %s\n",
6410 ztest_opts
.zo_alt_ztest
);
6413 killed
= exec_child(ztest_opts
.zo_alt_ztest
,
6414 ztest_opts
.zo_alt_libpath
, B_TRUE
, &status
);
6421 if (ztest_opts
.zo_verbose
>= 1) {
6422 hrtime_t now
= gethrtime();
6424 now
= MIN(now
, zs
->zs_proc_stop
);
6425 print_time(zs
->zs_proc_stop
- now
, timebuf
);
6426 nicenum(zs
->zs_space
, numbuf
, sizeof (numbuf
));
6428 (void) printf("Pass %3d, %8s, %3llu ENOSPC, "
6429 "%4.1f%% of %5s used, %3.0f%% done, %8s to go\n",
6431 WIFEXITED(status
) ? "Complete" : "SIGKILL",
6432 (u_longlong_t
)zs
->zs_enospc_count
,
6433 100.0 * zs
->zs_alloc
/ zs
->zs_space
,
6435 100.0 * (now
- zs
->zs_proc_start
) /
6436 (ztest_opts
.zo_time
* NANOSEC
), timebuf
);
6439 if (ztest_opts
.zo_verbose
>= 2) {
6440 (void) printf("\nWorkload summary:\n\n");
6441 (void) printf("%7s %9s %s\n",
6442 "Calls", "Time", "Function");
6443 (void) printf("%7s %9s %s\n",
6444 "-----", "----", "--------");
6445 for (int f
= 0; f
< ZTEST_FUNCS
; f
++) {
6448 zi
= &ztest_info
[f
];
6449 zc
= ZTEST_GET_SHARED_CALLSTATE(f
);
6450 print_time(zc
->zc_time
, timebuf
);
6451 (void) dladdr((void *)zi
->zi_func
, &dli
);
6452 (void) printf("%7llu %9s %s\n",
6453 (u_longlong_t
)zc
->zc_count
, timebuf
,
6456 (void) printf("\n");
6460 * It's possible that we killed a child during a rename test,
6461 * in which case we'll have a 'ztest_tmp' pool lying around
6462 * instead of 'ztest'. Do a blind rename in case this happened.
6465 if (spa_open(ztest_opts
.zo_pool
, &spa
, FTAG
) == 0) {
6466 spa_close(spa
, FTAG
);
6468 char tmpname
[ZFS_MAX_DATASET_NAME_LEN
];
6470 kernel_init(FREAD
| FWRITE
);
6471 (void) snprintf(tmpname
, sizeof (tmpname
), "%s_tmp",
6472 ztest_opts
.zo_pool
);
6473 (void) spa_rename(tmpname
, ztest_opts
.zo_pool
);
6477 ztest_run_zdb(ztest_opts
.zo_pool
);
6480 if (ztest_opts
.zo_verbose
>= 1) {
6482 (void) printf("%d runs of older ztest: %s\n", older
,
6483 ztest_opts
.zo_alt_ztest
);
6484 (void) printf("%d runs of newer ztest: %s\n", newer
,
6487 (void) printf("%d killed, %d completed, %.0f%% kill rate\n",
6488 kills
, iters
- kills
, (100.0 * kills
) / MAX(1, iters
));
6491 umem_free(cmd
, MAXNAMELEN
);