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.
26 * The objective of this program is to provide a DMU/ZAP/SPA stress test
27 * that runs entirely in userland, is easy to use, and easy to extend.
29 * The overall design of the ztest program is as follows:
31 * (1) For each major functional area (e.g. adding vdevs to a pool,
32 * creating and destroying datasets, reading and writing objects, etc)
33 * we have a simple routine to test that functionality. These
34 * individual routines do not have to do anything "stressful".
36 * (2) We turn these simple functionality tests into a stress test by
37 * running them all in parallel, with as many threads as desired,
38 * and spread across as many datasets, objects, and vdevs as desired.
40 * (3) While all this is happening, we inject faults into the pool to
41 * verify that self-healing data really works.
43 * (4) Every time we open a dataset, we change its checksum and compression
44 * functions. Thus even individual objects vary from block to block
45 * in which checksum they use and whether they're compressed.
47 * (5) To verify that we never lose on-disk consistency after a crash,
48 * we run the entire test in a child of the main process.
49 * At random times, the child self-immolates with a SIGKILL.
50 * This is the software equivalent of pulling the power cord.
51 * The parent then runs the test again, using the existing
52 * storage pool, as many times as desired.
54 * (6) To verify that we don't have future leaks or temporal incursions,
55 * many of the functional tests record the transaction group number
56 * as part of their data. When reading old data, they verify that
57 * the transaction group number is less than the current, open txg.
58 * If you add a new test, please do this if applicable.
60 * When run with no arguments, ztest runs for about five minutes and
61 * produces no output if successful. To get a little bit of information,
62 * specify -V. To get more information, specify -VV, and so on.
64 * To turn this into an overnight stress test, use -T to specify run time.
66 * You can ask more more vdevs [-v], datasets [-d], or threads [-t]
67 * to increase the pool capacity, fanout, and overall stress level.
69 * The -N(okill) option will suppress kills, so each child runs to completion.
70 * This can be useful when you're trying to distinguish temporal incursions
71 * from plain old race conditions.
74 #include <sys/zfs_context.h>
80 #include <sys/dmu_objset.h>
86 #include <sys/resource.h>
89 #include <sys/zil_impl.h>
90 #include <sys/vdev_impl.h>
91 #include <sys/vdev_file.h>
92 #include <sys/spa_impl.h>
93 #include <sys/metaslab_impl.h>
94 #include <sys/dsl_prop.h>
95 #include <sys/dsl_dataset.h>
96 #include <sys/dsl_scan.h>
97 #include <sys/zio_checksum.h>
98 #include <sys/refcount.h>
100 #include <stdio_ext.h>
108 #include <sys/fs/zfs.h>
109 #include <libnvpair.h>
111 static char cmdname
[] = "ztest";
112 static char *zopt_pool
= cmdname
;
114 static uint64_t zopt_vdevs
= 5;
115 static uint64_t zopt_vdevtime
;
116 static int zopt_ashift
= SPA_MINBLOCKSHIFT
;
117 static int zopt_mirrors
= 2;
118 static int zopt_raidz
= 4;
119 static int zopt_raidz_parity
= 1;
120 static size_t zopt_vdev_size
= SPA_MINDEVSIZE
;
121 static int zopt_datasets
= 7;
122 static int zopt_threads
= 23;
123 static uint64_t zopt_passtime
= 60; /* 60 seconds */
124 static uint64_t zopt_killrate
= 70; /* 70% kill rate */
125 static int zopt_verbose
= 0;
126 static int zopt_init
= 1;
127 static char *zopt_dir
= "/tmp";
128 static uint64_t zopt_time
= 300; /* 5 minutes */
129 static uint64_t zopt_maxloops
= 50; /* max loops during spa_freeze() */
131 #define BT_MAGIC 0x123456789abcdefULL
132 #define MAXFAULTS() (MAX(zs->zs_mirrors, 1) * (zopt_raidz_parity + 1) - 1)
136 ZTEST_IO_WRITE_PATTERN
,
137 ZTEST_IO_WRITE_ZEROES
,
143 typedef struct ztest_block_tag
{
153 typedef struct bufwad
{
160 * XXX -- fix zfs range locks to be generic so we can use them here.
182 #define ZTEST_RANGE_LOCKS 64
183 #define ZTEST_OBJECT_LOCKS 64
186 * Object descriptor. Used as a template for object lookup/create/remove.
188 typedef struct ztest_od
{
191 dmu_object_type_t od_type
;
192 dmu_object_type_t od_crtype
;
193 uint64_t od_blocksize
;
194 uint64_t od_crblocksize
;
197 char od_name
[MAXNAMELEN
];
203 typedef struct ztest_ds
{
207 ztest_od_t
*zd_od
; /* debugging aid */
208 char zd_name
[MAXNAMELEN
];
209 mutex_t zd_dirobj_lock
;
210 rll_t zd_object_lock
[ZTEST_OBJECT_LOCKS
];
211 rll_t zd_range_lock
[ZTEST_RANGE_LOCKS
];
215 * Per-iteration state.
217 typedef void ztest_func_t(ztest_ds_t
*zd
, uint64_t id
);
219 typedef struct ztest_info
{
220 ztest_func_t
*zi_func
; /* test function */
221 uint64_t zi_iters
; /* iterations per execution */
222 uint64_t *zi_interval
; /* execute every <interval> seconds */
223 uint64_t zi_call_count
; /* per-pass count */
224 uint64_t zi_call_time
; /* per-pass time */
225 uint64_t zi_call_next
; /* next time to call this function */
229 * Note: these aren't static because we want dladdr() to work.
231 ztest_func_t ztest_dmu_read_write
;
232 ztest_func_t ztest_dmu_write_parallel
;
233 ztest_func_t ztest_dmu_object_alloc_free
;
234 ztest_func_t ztest_dmu_commit_callbacks
;
235 ztest_func_t ztest_zap
;
236 ztest_func_t ztest_zap_parallel
;
237 ztest_func_t ztest_zil_commit
;
238 ztest_func_t ztest_dmu_read_write_zcopy
;
239 ztest_func_t ztest_dmu_objset_create_destroy
;
240 ztest_func_t ztest_dmu_prealloc
;
241 ztest_func_t ztest_fzap
;
242 ztest_func_t ztest_dmu_snapshot_create_destroy
;
243 ztest_func_t ztest_dsl_prop_get_set
;
244 ztest_func_t ztest_spa_prop_get_set
;
245 ztest_func_t ztest_spa_create_destroy
;
246 ztest_func_t ztest_fault_inject
;
247 ztest_func_t ztest_ddt_repair
;
248 ztest_func_t ztest_dmu_snapshot_hold
;
249 ztest_func_t ztest_spa_rename
;
250 ztest_func_t ztest_scrub
;
251 ztest_func_t ztest_dsl_dataset_promote_busy
;
252 ztest_func_t ztest_vdev_attach_detach
;
253 ztest_func_t ztest_vdev_LUN_growth
;
254 ztest_func_t ztest_vdev_add_remove
;
255 ztest_func_t ztest_vdev_aux_add_remove
;
256 ztest_func_t ztest_split_pool
;
258 uint64_t zopt_always
= 0ULL * NANOSEC
; /* all the time */
259 uint64_t zopt_incessant
= 1ULL * NANOSEC
/ 10; /* every 1/10 second */
260 uint64_t zopt_often
= 1ULL * NANOSEC
; /* every second */
261 uint64_t zopt_sometimes
= 10ULL * NANOSEC
; /* every 10 seconds */
262 uint64_t zopt_rarely
= 60ULL * NANOSEC
; /* every 60 seconds */
264 ztest_info_t ztest_info
[] = {
265 { ztest_dmu_read_write
, 1, &zopt_always
},
266 { ztest_dmu_write_parallel
, 10, &zopt_always
},
267 { ztest_dmu_object_alloc_free
, 1, &zopt_always
},
268 { ztest_dmu_commit_callbacks
, 1, &zopt_always
},
269 { ztest_zap
, 30, &zopt_always
},
270 { ztest_zap_parallel
, 100, &zopt_always
},
271 { ztest_split_pool
, 1, &zopt_always
},
272 { ztest_zil_commit
, 1, &zopt_incessant
},
273 { ztest_dmu_read_write_zcopy
, 1, &zopt_often
},
274 { ztest_dmu_objset_create_destroy
, 1, &zopt_often
},
275 { ztest_dsl_prop_get_set
, 1, &zopt_often
},
276 { ztest_spa_prop_get_set
, 1, &zopt_sometimes
},
278 { ztest_dmu_prealloc
, 1, &zopt_sometimes
},
280 { ztest_fzap
, 1, &zopt_sometimes
},
281 { ztest_dmu_snapshot_create_destroy
, 1, &zopt_sometimes
},
282 { ztest_spa_create_destroy
, 1, &zopt_sometimes
},
283 { ztest_fault_inject
, 1, &zopt_sometimes
},
284 { ztest_ddt_repair
, 1, &zopt_sometimes
},
285 { ztest_dmu_snapshot_hold
, 1, &zopt_sometimes
},
286 { ztest_spa_rename
, 1, &zopt_rarely
},
287 { ztest_scrub
, 1, &zopt_rarely
},
288 { ztest_dsl_dataset_promote_busy
, 1, &zopt_rarely
},
289 { ztest_vdev_attach_detach
, 1, &zopt_rarely
},
290 { ztest_vdev_LUN_growth
, 1, &zopt_rarely
},
291 { ztest_vdev_add_remove
, 1, &zopt_vdevtime
},
292 { ztest_vdev_aux_add_remove
, 1, &zopt_vdevtime
},
295 #define ZTEST_FUNCS (sizeof (ztest_info) / sizeof (ztest_info_t))
298 * The following struct is used to hold a list of uncalled commit callbacks.
299 * The callbacks are ordered by txg number.
301 typedef struct ztest_cb_list
{
302 mutex_t zcl_callbacks_lock
;
303 list_t zcl_callbacks
;
307 * Stuff we need to share writably between parent and child.
309 typedef struct ztest_shared
{
312 hrtime_t zs_proc_start
;
313 hrtime_t zs_proc_stop
;
314 hrtime_t zs_thread_start
;
315 hrtime_t zs_thread_stop
;
316 hrtime_t zs_thread_kill
;
317 uint64_t zs_enospc_count
;
318 uint64_t zs_vdev_next_leaf
;
319 uint64_t zs_vdev_aux
;
322 mutex_t zs_vdev_lock
;
323 rwlock_t zs_name_lock
;
324 ztest_info_t zs_info
[ZTEST_FUNCS
];
330 #define ID_PARALLEL -1ULL
332 static char ztest_dev_template
[] = "%s/%s.%llua";
333 static char ztest_aux_template
[] = "%s/%s.%s.%llu";
334 ztest_shared_t
*ztest_shared
;
337 static int ztest_random_fd
;
338 static int ztest_dump_core
= 1;
340 static boolean_t ztest_exiting
;
342 /* Global commit callback list */
343 static ztest_cb_list_t zcl
;
345 extern uint64_t metaslab_gang_bang
;
346 extern uint64_t metaslab_df_alloc_threshold
;
347 static uint64_t metaslab_sz
;
350 ZTEST_META_DNODE
= 0,
355 static void usage(boolean_t
) __NORETURN
;
358 * These libumem hooks provide a reasonable set of defaults for the allocator's
359 * debugging facilities.
362 _umem_debug_init(void)
364 return ("default,verbose"); /* $UMEM_DEBUG setting */
368 _umem_logging_init(void)
370 return ("fail,contents"); /* $UMEM_LOGGING setting */
373 #define FATAL_MSG_SZ 1024
378 fatal(int do_perror
, char *message
, ...)
381 int save_errno
= errno
;
382 char buf
[FATAL_MSG_SZ
];
384 (void) fflush(stdout
);
386 va_start(args
, message
);
387 (void) sprintf(buf
, "ztest: ");
389 (void) vsprintf(buf
+ strlen(buf
), message
, args
);
392 (void) snprintf(buf
+ strlen(buf
), FATAL_MSG_SZ
- strlen(buf
),
393 ": %s", strerror(save_errno
));
395 (void) fprintf(stderr
, "%s\n", buf
);
396 fatal_msg
= buf
; /* to ease debugging */
403 str2shift(const char *buf
)
405 const char *ends
= "BKMGTPEZ";
410 for (i
= 0; i
< strlen(ends
); i
++) {
411 if (toupper(buf
[0]) == ends
[i
])
414 if (i
== strlen(ends
)) {
415 (void) fprintf(stderr
, "ztest: invalid bytes suffix: %s\n",
419 if (buf
[1] == '\0' || (toupper(buf
[1]) == 'B' && buf
[2] == '\0')) {
422 (void) fprintf(stderr
, "ztest: invalid bytes suffix: %s\n", buf
);
428 nicenumtoull(const char *buf
)
433 val
= strtoull(buf
, &end
, 0);
435 (void) fprintf(stderr
, "ztest: bad numeric value: %s\n", buf
);
437 } else if (end
[0] == '.') {
438 double fval
= strtod(buf
, &end
);
439 fval
*= pow(2, str2shift(end
));
440 if (fval
> UINT64_MAX
) {
441 (void) fprintf(stderr
, "ztest: value too large: %s\n",
445 val
= (uint64_t)fval
;
447 int shift
= str2shift(end
);
448 if (shift
>= 64 || (val
<< shift
) >> shift
!= val
) {
449 (void) fprintf(stderr
, "ztest: value too large: %s\n",
459 usage(boolean_t requested
)
461 char nice_vdev_size
[10];
462 char nice_gang_bang
[10];
463 FILE *fp
= requested
? stdout
: stderr
;
465 nicenum(zopt_vdev_size
, nice_vdev_size
);
466 nicenum(metaslab_gang_bang
, nice_gang_bang
);
468 (void) fprintf(fp
, "Usage: %s\n"
469 "\t[-v vdevs (default: %llu)]\n"
470 "\t[-s size_of_each_vdev (default: %s)]\n"
471 "\t[-a alignment_shift (default: %d)] use 0 for random\n"
472 "\t[-m mirror_copies (default: %d)]\n"
473 "\t[-r raidz_disks (default: %d)]\n"
474 "\t[-R raidz_parity (default: %d)]\n"
475 "\t[-d datasets (default: %d)]\n"
476 "\t[-t threads (default: %d)]\n"
477 "\t[-g gang_block_threshold (default: %s)]\n"
478 "\t[-i init_count (default: %d)] initialize pool i times\n"
479 "\t[-k kill_percentage (default: %llu%%)]\n"
480 "\t[-p pool_name (default: %s)]\n"
481 "\t[-f dir (default: %s)] file directory for vdev files\n"
482 "\t[-V] verbose (use multiple times for ever more blather)\n"
483 "\t[-E] use existing pool instead of creating new one\n"
484 "\t[-T time (default: %llu sec)] total run time\n"
485 "\t[-F freezeloops (default: %llu)] max loops in spa_freeze()\n"
486 "\t[-P passtime (default: %llu sec)] time per pass\n"
487 "\t[-h] (print help)\n"
490 (u_longlong_t
)zopt_vdevs
, /* -v */
491 nice_vdev_size
, /* -s */
492 zopt_ashift
, /* -a */
493 zopt_mirrors
, /* -m */
495 zopt_raidz_parity
, /* -R */
496 zopt_datasets
, /* -d */
497 zopt_threads
, /* -t */
498 nice_gang_bang
, /* -g */
500 (u_longlong_t
)zopt_killrate
, /* -k */
503 (u_longlong_t
)zopt_time
, /* -T */
504 (u_longlong_t
)zopt_maxloops
, /* -F */
505 (u_longlong_t
)zopt_passtime
); /* -P */
506 exit(requested
? 0 : 1);
510 process_options(int argc
, char **argv
)
515 /* By default, test gang blocks for blocks 32K and greater */
516 metaslab_gang_bang
= 32 << 10;
518 while ((opt
= getopt(argc
, argv
,
519 "v:s:a:m:r:R:d:t:g:i:k:p:f:VET:P:hF:")) != EOF
) {
536 value
= nicenumtoull(optarg
);
543 zopt_vdev_size
= MAX(SPA_MINDEVSIZE
, value
);
549 zopt_mirrors
= value
;
552 zopt_raidz
= MAX(1, value
);
555 zopt_raidz_parity
= MIN(MAX(value
, 1), 3);
558 zopt_datasets
= MAX(1, value
);
561 zopt_threads
= MAX(1, value
);
564 metaslab_gang_bang
= MAX(SPA_MINBLOCKSIZE
<< 1, value
);
570 zopt_killrate
= value
;
573 zopt_pool
= strdup(optarg
);
576 zopt_dir
= strdup(optarg
);
588 zopt_passtime
= MAX(1, value
);
591 zopt_maxloops
= MAX(1, value
);
603 zopt_raidz_parity
= MIN(zopt_raidz_parity
, zopt_raidz
- 1);
605 zopt_vdevtime
= (zopt_vdevs
> 0 ? zopt_time
* NANOSEC
/ zopt_vdevs
:
610 ztest_kill(ztest_shared_t
*zs
)
612 zs
->zs_alloc
= metaslab_class_get_alloc(spa_normal_class(zs
->zs_spa
));
613 zs
->zs_space
= metaslab_class_get_space(spa_normal_class(zs
->zs_spa
));
614 (void) kill(getpid(), SIGKILL
);
618 ztest_random(uint64_t range
)
625 if (read(ztest_random_fd
, &r
, sizeof (r
)) != sizeof (r
))
626 fatal(1, "short read from /dev/urandom");
633 ztest_record_enospc(const char *s
)
635 ztest_shared
->zs_enospc_count
++;
639 ztest_get_ashift(void)
641 if (zopt_ashift
== 0)
642 return (SPA_MINBLOCKSHIFT
+ ztest_random(3));
643 return (zopt_ashift
);
647 make_vdev_file(char *path
, char *aux
, size_t size
, uint64_t ashift
)
649 char pathbuf
[MAXPATHLEN
];
654 ashift
= ztest_get_ashift();
660 vdev
= ztest_shared
->zs_vdev_aux
;
661 (void) sprintf(path
, ztest_aux_template
,
662 zopt_dir
, zopt_pool
, aux
, vdev
);
664 vdev
= ztest_shared
->zs_vdev_next_leaf
++;
665 (void) sprintf(path
, ztest_dev_template
,
666 zopt_dir
, zopt_pool
, vdev
);
671 int fd
= open(path
, O_RDWR
| O_CREAT
| O_TRUNC
, 0666);
673 fatal(1, "can't open %s", path
);
674 if (ftruncate(fd
, size
) != 0)
675 fatal(1, "can't ftruncate %s", path
);
679 VERIFY(nvlist_alloc(&file
, NV_UNIQUE_NAME
, 0) == 0);
680 VERIFY(nvlist_add_string(file
, ZPOOL_CONFIG_TYPE
, VDEV_TYPE_FILE
) == 0);
681 VERIFY(nvlist_add_string(file
, ZPOOL_CONFIG_PATH
, path
) == 0);
682 VERIFY(nvlist_add_uint64(file
, ZPOOL_CONFIG_ASHIFT
, ashift
) == 0);
688 make_vdev_raidz(char *path
, char *aux
, size_t size
, uint64_t ashift
, int r
)
690 nvlist_t
*raidz
, **child
;
694 return (make_vdev_file(path
, aux
, size
, ashift
));
695 child
= umem_alloc(r
* sizeof (nvlist_t
*), UMEM_NOFAIL
);
697 for (c
= 0; c
< r
; c
++)
698 child
[c
] = make_vdev_file(path
, aux
, size
, ashift
);
700 VERIFY(nvlist_alloc(&raidz
, NV_UNIQUE_NAME
, 0) == 0);
701 VERIFY(nvlist_add_string(raidz
, ZPOOL_CONFIG_TYPE
,
702 VDEV_TYPE_RAIDZ
) == 0);
703 VERIFY(nvlist_add_uint64(raidz
, ZPOOL_CONFIG_NPARITY
,
704 zopt_raidz_parity
) == 0);
705 VERIFY(nvlist_add_nvlist_array(raidz
, ZPOOL_CONFIG_CHILDREN
,
708 for (c
= 0; c
< r
; c
++)
709 nvlist_free(child
[c
]);
711 umem_free(child
, r
* sizeof (nvlist_t
*));
717 make_vdev_mirror(char *path
, char *aux
, size_t size
, uint64_t ashift
,
720 nvlist_t
*mirror
, **child
;
724 return (make_vdev_raidz(path
, aux
, size
, ashift
, r
));
726 child
= umem_alloc(m
* sizeof (nvlist_t
*), UMEM_NOFAIL
);
728 for (c
= 0; c
< m
; c
++)
729 child
[c
] = make_vdev_raidz(path
, aux
, size
, ashift
, r
);
731 VERIFY(nvlist_alloc(&mirror
, NV_UNIQUE_NAME
, 0) == 0);
732 VERIFY(nvlist_add_string(mirror
, ZPOOL_CONFIG_TYPE
,
733 VDEV_TYPE_MIRROR
) == 0);
734 VERIFY(nvlist_add_nvlist_array(mirror
, ZPOOL_CONFIG_CHILDREN
,
737 for (c
= 0; c
< m
; c
++)
738 nvlist_free(child
[c
]);
740 umem_free(child
, m
* sizeof (nvlist_t
*));
746 make_vdev_root(char *path
, char *aux
, size_t size
, uint64_t ashift
,
747 int log
, int r
, int m
, int t
)
749 nvlist_t
*root
, **child
;
754 child
= umem_alloc(t
* sizeof (nvlist_t
*), UMEM_NOFAIL
);
756 for (c
= 0; c
< t
; c
++) {
757 child
[c
] = make_vdev_mirror(path
, aux
, size
, ashift
, r
, m
);
758 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_IS_LOG
,
762 VERIFY(nvlist_alloc(&root
, NV_UNIQUE_NAME
, 0) == 0);
763 VERIFY(nvlist_add_string(root
, ZPOOL_CONFIG_TYPE
, VDEV_TYPE_ROOT
) == 0);
764 VERIFY(nvlist_add_nvlist_array(root
, aux
? aux
: ZPOOL_CONFIG_CHILDREN
,
767 for (c
= 0; c
< t
; c
++)
768 nvlist_free(child
[c
]);
770 umem_free(child
, t
* sizeof (nvlist_t
*));
776 ztest_random_blocksize(void)
778 return (1 << (SPA_MINBLOCKSHIFT
+
779 ztest_random(SPA_MAXBLOCKSHIFT
- SPA_MINBLOCKSHIFT
+ 1)));
783 ztest_random_ibshift(void)
785 return (DN_MIN_INDBLKSHIFT
+
786 ztest_random(DN_MAX_INDBLKSHIFT
- DN_MIN_INDBLKSHIFT
+ 1));
790 ztest_random_vdev_top(spa_t
*spa
, boolean_t log_ok
)
793 vdev_t
*rvd
= spa
->spa_root_vdev
;
796 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_READER
) != 0);
799 top
= ztest_random(rvd
->vdev_children
);
800 tvd
= rvd
->vdev_child
[top
];
801 } while (tvd
->vdev_ishole
|| (tvd
->vdev_islog
&& !log_ok
) ||
802 tvd
->vdev_mg
== NULL
|| tvd
->vdev_mg
->mg_class
== NULL
);
808 ztest_random_dsl_prop(zfs_prop_t prop
)
813 value
= zfs_prop_random_value(prop
, ztest_random(-1ULL));
814 } while (prop
== ZFS_PROP_CHECKSUM
&& value
== ZIO_CHECKSUM_OFF
);
820 ztest_dsl_prop_set_uint64(char *osname
, zfs_prop_t prop
, uint64_t value
,
823 const char *propname
= zfs_prop_to_name(prop
);
825 char setpoint
[MAXPATHLEN
];
829 error
= dsl_prop_set(osname
, propname
,
830 (inherit
? ZPROP_SRC_NONE
: ZPROP_SRC_LOCAL
),
831 sizeof (value
), 1, &value
);
833 if (error
== ENOSPC
) {
834 ztest_record_enospc(FTAG
);
837 ASSERT3U(error
, ==, 0);
839 VERIFY3U(dsl_prop_get(osname
, propname
, sizeof (curval
),
840 1, &curval
, setpoint
), ==, 0);
842 if (zopt_verbose
>= 6) {
843 VERIFY(zfs_prop_index_to_string(prop
, curval
, &valname
) == 0);
844 (void) printf("%s %s = %s at '%s'\n",
845 osname
, propname
, valname
, setpoint
);
852 ztest_spa_prop_set_uint64(ztest_shared_t
*zs
, zpool_prop_t prop
, uint64_t value
)
854 spa_t
*spa
= zs
->zs_spa
;
855 nvlist_t
*props
= NULL
;
858 VERIFY(nvlist_alloc(&props
, NV_UNIQUE_NAME
, 0) == 0);
859 VERIFY(nvlist_add_uint64(props
, zpool_prop_to_name(prop
), value
) == 0);
861 error
= spa_prop_set(spa
, props
);
865 if (error
== ENOSPC
) {
866 ztest_record_enospc(FTAG
);
869 ASSERT3U(error
, ==, 0);
875 ztest_rll_init(rll_t
*rll
)
877 rll
->rll_writer
= NULL
;
878 rll
->rll_readers
= 0;
879 VERIFY(_mutex_init(&rll
->rll_lock
, USYNC_THREAD
, NULL
) == 0);
880 VERIFY(cond_init(&rll
->rll_cv
, USYNC_THREAD
, NULL
) == 0);
884 ztest_rll_destroy(rll_t
*rll
)
886 ASSERT(rll
->rll_writer
== NULL
);
887 ASSERT(rll
->rll_readers
== 0);
888 VERIFY(_mutex_destroy(&rll
->rll_lock
) == 0);
889 VERIFY(cond_destroy(&rll
->rll_cv
) == 0);
893 ztest_rll_lock(rll_t
*rll
, rl_type_t type
)
895 VERIFY(mutex_lock(&rll
->rll_lock
) == 0);
897 if (type
== RL_READER
) {
898 while (rll
->rll_writer
!= NULL
)
899 (void) cond_wait(&rll
->rll_cv
, &rll
->rll_lock
);
902 while (rll
->rll_writer
!= NULL
|| rll
->rll_readers
)
903 (void) cond_wait(&rll
->rll_cv
, &rll
->rll_lock
);
904 rll
->rll_writer
= curthread
;
907 VERIFY(mutex_unlock(&rll
->rll_lock
) == 0);
911 ztest_rll_unlock(rll_t
*rll
)
913 VERIFY(mutex_lock(&rll
->rll_lock
) == 0);
915 if (rll
->rll_writer
) {
916 ASSERT(rll
->rll_readers
== 0);
917 rll
->rll_writer
= NULL
;
919 ASSERT(rll
->rll_readers
!= 0);
920 ASSERT(rll
->rll_writer
== NULL
);
924 if (rll
->rll_writer
== NULL
&& rll
->rll_readers
== 0)
925 VERIFY(cond_broadcast(&rll
->rll_cv
) == 0);
927 VERIFY(mutex_unlock(&rll
->rll_lock
) == 0);
931 ztest_object_lock(ztest_ds_t
*zd
, uint64_t object
, rl_type_t type
)
933 rll_t
*rll
= &zd
->zd_object_lock
[object
& (ZTEST_OBJECT_LOCKS
- 1)];
935 ztest_rll_lock(rll
, type
);
939 ztest_object_unlock(ztest_ds_t
*zd
, uint64_t object
)
941 rll_t
*rll
= &zd
->zd_object_lock
[object
& (ZTEST_OBJECT_LOCKS
- 1)];
943 ztest_rll_unlock(rll
);
947 ztest_range_lock(ztest_ds_t
*zd
, uint64_t object
, uint64_t offset
,
948 uint64_t size
, rl_type_t type
)
950 uint64_t hash
= object
^ (offset
% (ZTEST_RANGE_LOCKS
+ 1));
951 rll_t
*rll
= &zd
->zd_range_lock
[hash
& (ZTEST_RANGE_LOCKS
- 1)];
954 rl
= umem_alloc(sizeof (*rl
), UMEM_NOFAIL
);
955 rl
->rl_object
= object
;
956 rl
->rl_offset
= offset
;
960 ztest_rll_lock(rll
, type
);
966 ztest_range_unlock(rl_t
*rl
)
968 rll_t
*rll
= rl
->rl_lock
;
970 ztest_rll_unlock(rll
);
972 umem_free(rl
, sizeof (*rl
));
976 ztest_zd_init(ztest_ds_t
*zd
, objset_t
*os
)
979 zd
->zd_zilog
= dmu_objset_zil(os
);
981 dmu_objset_name(os
, zd
->zd_name
);
984 VERIFY(_mutex_init(&zd
->zd_dirobj_lock
, USYNC_THREAD
, NULL
) == 0);
986 for (l
= 0; l
< ZTEST_OBJECT_LOCKS
; l
++)
987 ztest_rll_init(&zd
->zd_object_lock
[l
]);
989 for (l
= 0; l
< ZTEST_RANGE_LOCKS
; l
++)
990 ztest_rll_init(&zd
->zd_range_lock
[l
]);
994 ztest_zd_fini(ztest_ds_t
*zd
)
998 VERIFY(_mutex_destroy(&zd
->zd_dirobj_lock
) == 0);
1000 for (l
= 0; l
< ZTEST_OBJECT_LOCKS
; l
++)
1001 ztest_rll_destroy(&zd
->zd_object_lock
[l
]);
1003 for (l
= 0; l
< ZTEST_RANGE_LOCKS
; l
++)
1004 ztest_rll_destroy(&zd
->zd_range_lock
[l
]);
1007 #define TXG_MIGHTWAIT (ztest_random(10) == 0 ? TXG_NOWAIT : TXG_WAIT)
1010 ztest_tx_assign(dmu_tx_t
*tx
, uint64_t txg_how
, const char *tag
)
1016 * Attempt to assign tx to some transaction group.
1018 error
= dmu_tx_assign(tx
, txg_how
);
1020 if (error
== ERESTART
) {
1021 ASSERT(txg_how
== TXG_NOWAIT
);
1024 ASSERT3U(error
, ==, ENOSPC
);
1025 ztest_record_enospc(tag
);
1030 txg
= dmu_tx_get_txg(tx
);
1036 ztest_pattern_set(void *buf
, uint64_t size
, uint64_t value
)
1039 uint64_t *ip_end
= (uint64_t *)((uintptr_t)buf
+ (uintptr_t)size
);
1047 ztest_pattern_match(void *buf
, uint64_t size
, uint64_t value
)
1050 uint64_t *ip_end
= (uint64_t *)((uintptr_t)buf
+ (uintptr_t)size
);
1054 diff
|= (value
- *ip
++);
1061 ztest_bt_generate(ztest_block_tag_t
*bt
, objset_t
*os
, uint64_t object
,
1062 uint64_t offset
, uint64_t gen
, uint64_t txg
, uint64_t crtxg
)
1064 bt
->bt_magic
= BT_MAGIC
;
1065 bt
->bt_objset
= dmu_objset_id(os
);
1066 bt
->bt_object
= object
;
1067 bt
->bt_offset
= offset
;
1070 bt
->bt_crtxg
= crtxg
;
1074 ztest_bt_verify(ztest_block_tag_t
*bt
, objset_t
*os
, uint64_t object
,
1075 uint64_t offset
, uint64_t gen
, uint64_t txg
, uint64_t crtxg
)
1077 ASSERT(bt
->bt_magic
== BT_MAGIC
);
1078 ASSERT(bt
->bt_objset
== dmu_objset_id(os
));
1079 ASSERT(bt
->bt_object
== object
);
1080 ASSERT(bt
->bt_offset
== offset
);
1081 ASSERT(bt
->bt_gen
<= gen
);
1082 ASSERT(bt
->bt_txg
<= txg
);
1083 ASSERT(bt
->bt_crtxg
== crtxg
);
1086 static ztest_block_tag_t
*
1087 ztest_bt_bonus(dmu_buf_t
*db
)
1089 dmu_object_info_t doi
;
1090 ztest_block_tag_t
*bt
;
1092 dmu_object_info_from_db(db
, &doi
);
1093 ASSERT3U(doi
.doi_bonus_size
, <=, db
->db_size
);
1094 ASSERT3U(doi
.doi_bonus_size
, >=, sizeof (*bt
));
1095 bt
= (void *)((char *)db
->db_data
+ doi
.doi_bonus_size
- sizeof (*bt
));
1104 #define lrz_type lr_mode
1105 #define lrz_blocksize lr_uid
1106 #define lrz_ibshift lr_gid
1107 #define lrz_bonustype lr_rdev
1108 #define lrz_bonuslen lr_crtime[1]
1111 ztest_log_create(ztest_ds_t
*zd
, dmu_tx_t
*tx
, lr_create_t
*lr
)
1113 char *name
= (void *)(lr
+ 1); /* name follows lr */
1114 size_t namesize
= strlen(name
) + 1;
1117 if (zil_replaying(zd
->zd_zilog
, tx
))
1120 itx
= zil_itx_create(TX_CREATE
, sizeof (*lr
) + namesize
);
1121 bcopy(&lr
->lr_common
+ 1, &itx
->itx_lr
+ 1,
1122 sizeof (*lr
) + namesize
- sizeof (lr_t
));
1124 zil_itx_assign(zd
->zd_zilog
, itx
, tx
);
1128 ztest_log_remove(ztest_ds_t
*zd
, dmu_tx_t
*tx
, lr_remove_t
*lr
, uint64_t object
)
1130 char *name
= (void *)(lr
+ 1); /* name follows lr */
1131 size_t namesize
= strlen(name
) + 1;
1134 if (zil_replaying(zd
->zd_zilog
, tx
))
1137 itx
= zil_itx_create(TX_REMOVE
, sizeof (*lr
) + namesize
);
1138 bcopy(&lr
->lr_common
+ 1, &itx
->itx_lr
+ 1,
1139 sizeof (*lr
) + namesize
- sizeof (lr_t
));
1141 itx
->itx_oid
= object
;
1142 zil_itx_assign(zd
->zd_zilog
, itx
, tx
);
1146 ztest_log_write(ztest_ds_t
*zd
, dmu_tx_t
*tx
, lr_write_t
*lr
)
1149 itx_wr_state_t write_state
= ztest_random(WR_NUM_STATES
);
1151 if (zil_replaying(zd
->zd_zilog
, tx
))
1154 if (lr
->lr_length
> ZIL_MAX_LOG_DATA
)
1155 write_state
= WR_INDIRECT
;
1157 itx
= zil_itx_create(TX_WRITE
,
1158 sizeof (*lr
) + (write_state
== WR_COPIED
? lr
->lr_length
: 0));
1160 if (write_state
== WR_COPIED
&&
1161 dmu_read(zd
->zd_os
, lr
->lr_foid
, lr
->lr_offset
, lr
->lr_length
,
1162 ((lr_write_t
*)&itx
->itx_lr
) + 1, DMU_READ_NO_PREFETCH
) != 0) {
1163 zil_itx_destroy(itx
);
1164 itx
= zil_itx_create(TX_WRITE
, sizeof (*lr
));
1165 write_state
= WR_NEED_COPY
;
1167 itx
->itx_private
= zd
;
1168 itx
->itx_wr_state
= write_state
;
1169 itx
->itx_sync
= (ztest_random(8) == 0);
1170 itx
->itx_sod
+= (write_state
== WR_NEED_COPY
? lr
->lr_length
: 0);
1172 bcopy(&lr
->lr_common
+ 1, &itx
->itx_lr
+ 1,
1173 sizeof (*lr
) - sizeof (lr_t
));
1175 zil_itx_assign(zd
->zd_zilog
, itx
, tx
);
1179 ztest_log_truncate(ztest_ds_t
*zd
, dmu_tx_t
*tx
, lr_truncate_t
*lr
)
1183 if (zil_replaying(zd
->zd_zilog
, tx
))
1186 itx
= zil_itx_create(TX_TRUNCATE
, sizeof (*lr
));
1187 bcopy(&lr
->lr_common
+ 1, &itx
->itx_lr
+ 1,
1188 sizeof (*lr
) - sizeof (lr_t
));
1190 itx
->itx_sync
= B_FALSE
;
1191 zil_itx_assign(zd
->zd_zilog
, itx
, tx
);
1195 ztest_log_setattr(ztest_ds_t
*zd
, dmu_tx_t
*tx
, lr_setattr_t
*lr
)
1199 if (zil_replaying(zd
->zd_zilog
, tx
))
1202 itx
= zil_itx_create(TX_SETATTR
, sizeof (*lr
));
1203 bcopy(&lr
->lr_common
+ 1, &itx
->itx_lr
+ 1,
1204 sizeof (*lr
) - sizeof (lr_t
));
1206 itx
->itx_sync
= B_FALSE
;
1207 zil_itx_assign(zd
->zd_zilog
, itx
, tx
);
1214 ztest_replay_create(ztest_ds_t
*zd
, lr_create_t
*lr
, boolean_t byteswap
)
1216 char *name
= (void *)(lr
+ 1); /* name follows lr */
1217 objset_t
*os
= zd
->zd_os
;
1218 ztest_block_tag_t
*bbt
;
1225 byteswap_uint64_array(lr
, sizeof (*lr
));
1227 ASSERT(lr
->lr_doid
== ZTEST_DIROBJ
);
1228 ASSERT(name
[0] != '\0');
1230 tx
= dmu_tx_create(os
);
1232 dmu_tx_hold_zap(tx
, lr
->lr_doid
, B_TRUE
, name
);
1234 if (lr
->lrz_type
== DMU_OT_ZAP_OTHER
) {
1235 dmu_tx_hold_zap(tx
, DMU_NEW_OBJECT
, B_TRUE
, NULL
);
1237 dmu_tx_hold_bonus(tx
, DMU_NEW_OBJECT
);
1240 txg
= ztest_tx_assign(tx
, TXG_WAIT
, FTAG
);
1244 ASSERT(dmu_objset_zil(os
)->zl_replay
== !!lr
->lr_foid
);
1246 if (lr
->lrz_type
== DMU_OT_ZAP_OTHER
) {
1247 if (lr
->lr_foid
== 0) {
1248 lr
->lr_foid
= zap_create(os
,
1249 lr
->lrz_type
, lr
->lrz_bonustype
,
1250 lr
->lrz_bonuslen
, tx
);
1252 error
= zap_create_claim(os
, lr
->lr_foid
,
1253 lr
->lrz_type
, lr
->lrz_bonustype
,
1254 lr
->lrz_bonuslen
, tx
);
1257 if (lr
->lr_foid
== 0) {
1258 lr
->lr_foid
= dmu_object_alloc(os
,
1259 lr
->lrz_type
, 0, lr
->lrz_bonustype
,
1260 lr
->lrz_bonuslen
, tx
);
1262 error
= dmu_object_claim(os
, lr
->lr_foid
,
1263 lr
->lrz_type
, 0, lr
->lrz_bonustype
,
1264 lr
->lrz_bonuslen
, tx
);
1269 ASSERT3U(error
, ==, EEXIST
);
1270 ASSERT(zd
->zd_zilog
->zl_replay
);
1275 ASSERT(lr
->lr_foid
!= 0);
1277 if (lr
->lrz_type
!= DMU_OT_ZAP_OTHER
)
1278 VERIFY3U(0, ==, dmu_object_set_blocksize(os
, lr
->lr_foid
,
1279 lr
->lrz_blocksize
, lr
->lrz_ibshift
, tx
));
1281 VERIFY3U(0, ==, dmu_bonus_hold(os
, lr
->lr_foid
, FTAG
, &db
));
1282 bbt
= ztest_bt_bonus(db
);
1283 dmu_buf_will_dirty(db
, tx
);
1284 ztest_bt_generate(bbt
, os
, lr
->lr_foid
, -1ULL, lr
->lr_gen
, txg
, txg
);
1285 dmu_buf_rele(db
, FTAG
);
1287 VERIFY3U(0, ==, zap_add(os
, lr
->lr_doid
, name
, sizeof (uint64_t), 1,
1290 (void) ztest_log_create(zd
, tx
, lr
);
1298 ztest_replay_remove(ztest_ds_t
*zd
, lr_remove_t
*lr
, boolean_t byteswap
)
1300 char *name
= (void *)(lr
+ 1); /* name follows lr */
1301 objset_t
*os
= zd
->zd_os
;
1302 dmu_object_info_t doi
;
1304 uint64_t object
, txg
;
1307 byteswap_uint64_array(lr
, sizeof (*lr
));
1309 ASSERT(lr
->lr_doid
== ZTEST_DIROBJ
);
1310 ASSERT(name
[0] != '\0');
1313 zap_lookup(os
, lr
->lr_doid
, name
, sizeof (object
), 1, &object
));
1314 ASSERT(object
!= 0);
1316 ztest_object_lock(zd
, object
, RL_WRITER
);
1318 VERIFY3U(0, ==, dmu_object_info(os
, object
, &doi
));
1320 tx
= dmu_tx_create(os
);
1322 dmu_tx_hold_zap(tx
, lr
->lr_doid
, B_FALSE
, name
);
1323 dmu_tx_hold_free(tx
, object
, 0, DMU_OBJECT_END
);
1325 txg
= ztest_tx_assign(tx
, TXG_WAIT
, FTAG
);
1327 ztest_object_unlock(zd
, object
);
1331 if (doi
.doi_type
== DMU_OT_ZAP_OTHER
) {
1332 VERIFY3U(0, ==, zap_destroy(os
, object
, tx
));
1334 VERIFY3U(0, ==, dmu_object_free(os
, object
, tx
));
1337 VERIFY3U(0, ==, zap_remove(os
, lr
->lr_doid
, name
, tx
));
1339 (void) ztest_log_remove(zd
, tx
, lr
, object
);
1343 ztest_object_unlock(zd
, object
);
1349 ztest_replay_write(ztest_ds_t
*zd
, lr_write_t
*lr
, boolean_t byteswap
)
1351 objset_t
*os
= zd
->zd_os
;
1352 void *data
= lr
+ 1; /* data follows lr */
1353 uint64_t offset
, length
;
1354 ztest_block_tag_t
*bt
= data
;
1355 ztest_block_tag_t
*bbt
;
1356 uint64_t gen
, txg
, lrtxg
, crtxg
;
1357 dmu_object_info_t doi
;
1360 arc_buf_t
*abuf
= NULL
;
1364 byteswap_uint64_array(lr
, sizeof (*lr
));
1366 offset
= lr
->lr_offset
;
1367 length
= lr
->lr_length
;
1369 /* If it's a dmu_sync() block, write the whole block */
1370 if (lr
->lr_common
.lrc_reclen
== sizeof (lr_write_t
)) {
1371 uint64_t blocksize
= BP_GET_LSIZE(&lr
->lr_blkptr
);
1372 if (length
< blocksize
) {
1373 offset
-= offset
% blocksize
;
1378 if (bt
->bt_magic
== BSWAP_64(BT_MAGIC
))
1379 byteswap_uint64_array(bt
, sizeof (*bt
));
1381 if (bt
->bt_magic
!= BT_MAGIC
)
1384 ztest_object_lock(zd
, lr
->lr_foid
, RL_READER
);
1385 rl
= ztest_range_lock(zd
, lr
->lr_foid
, offset
, length
, RL_WRITER
);
1387 VERIFY3U(0, ==, dmu_bonus_hold(os
, lr
->lr_foid
, FTAG
, &db
));
1389 dmu_object_info_from_db(db
, &doi
);
1391 bbt
= ztest_bt_bonus(db
);
1392 ASSERT3U(bbt
->bt_magic
, ==, BT_MAGIC
);
1394 crtxg
= bbt
->bt_crtxg
;
1395 lrtxg
= lr
->lr_common
.lrc_txg
;
1397 tx
= dmu_tx_create(os
);
1399 dmu_tx_hold_write(tx
, lr
->lr_foid
, offset
, length
);
1401 if (ztest_random(8) == 0 && length
== doi
.doi_data_block_size
&&
1402 P2PHASE(offset
, length
) == 0)
1403 abuf
= dmu_request_arcbuf(db
, length
);
1405 txg
= ztest_tx_assign(tx
, TXG_WAIT
, FTAG
);
1408 dmu_return_arcbuf(abuf
);
1409 dmu_buf_rele(db
, FTAG
);
1410 ztest_range_unlock(rl
);
1411 ztest_object_unlock(zd
, lr
->lr_foid
);
1417 * Usually, verify the old data before writing new data --
1418 * but not always, because we also want to verify correct
1419 * behavior when the data was not recently read into cache.
1421 ASSERT(offset
% doi
.doi_data_block_size
== 0);
1422 if (ztest_random(4) != 0) {
1423 int prefetch
= ztest_random(2) ?
1424 DMU_READ_PREFETCH
: DMU_READ_NO_PREFETCH
;
1425 ztest_block_tag_t rbt
;
1427 VERIFY(dmu_read(os
, lr
->lr_foid
, offset
,
1428 sizeof (rbt
), &rbt
, prefetch
) == 0);
1429 if (rbt
.bt_magic
== BT_MAGIC
) {
1430 ztest_bt_verify(&rbt
, os
, lr
->lr_foid
,
1431 offset
, gen
, txg
, crtxg
);
1436 * Writes can appear to be newer than the bonus buffer because
1437 * the ztest_get_data() callback does a dmu_read() of the
1438 * open-context data, which may be different than the data
1439 * as it was when the write was generated.
1441 if (zd
->zd_zilog
->zl_replay
) {
1442 ztest_bt_verify(bt
, os
, lr
->lr_foid
, offset
,
1443 MAX(gen
, bt
->bt_gen
), MAX(txg
, lrtxg
),
1448 * Set the bt's gen/txg to the bonus buffer's gen/txg
1449 * so that all of the usual ASSERTs will work.
1451 ztest_bt_generate(bt
, os
, lr
->lr_foid
, offset
, gen
, txg
, crtxg
);
1455 dmu_write(os
, lr
->lr_foid
, offset
, length
, data
, tx
);
1457 bcopy(data
, abuf
->b_data
, length
);
1458 dmu_assign_arcbuf(db
, offset
, abuf
, tx
);
1461 (void) ztest_log_write(zd
, tx
, lr
);
1463 dmu_buf_rele(db
, FTAG
);
1467 ztest_range_unlock(rl
);
1468 ztest_object_unlock(zd
, lr
->lr_foid
);
1474 ztest_replay_truncate(ztest_ds_t
*zd
, lr_truncate_t
*lr
, boolean_t byteswap
)
1476 objset_t
*os
= zd
->zd_os
;
1482 byteswap_uint64_array(lr
, sizeof (*lr
));
1484 ztest_object_lock(zd
, lr
->lr_foid
, RL_READER
);
1485 rl
= ztest_range_lock(zd
, lr
->lr_foid
, lr
->lr_offset
, lr
->lr_length
,
1488 tx
= dmu_tx_create(os
);
1490 dmu_tx_hold_free(tx
, lr
->lr_foid
, lr
->lr_offset
, lr
->lr_length
);
1492 txg
= ztest_tx_assign(tx
, TXG_WAIT
, FTAG
);
1494 ztest_range_unlock(rl
);
1495 ztest_object_unlock(zd
, lr
->lr_foid
);
1499 VERIFY(dmu_free_range(os
, lr
->lr_foid
, lr
->lr_offset
,
1500 lr
->lr_length
, tx
) == 0);
1502 (void) ztest_log_truncate(zd
, tx
, lr
);
1506 ztest_range_unlock(rl
);
1507 ztest_object_unlock(zd
, lr
->lr_foid
);
1513 ztest_replay_setattr(ztest_ds_t
*zd
, lr_setattr_t
*lr
, boolean_t byteswap
)
1515 objset_t
*os
= zd
->zd_os
;
1518 ztest_block_tag_t
*bbt
;
1519 uint64_t txg
, lrtxg
, crtxg
;
1522 byteswap_uint64_array(lr
, sizeof (*lr
));
1524 ztest_object_lock(zd
, lr
->lr_foid
, RL_WRITER
);
1526 VERIFY3U(0, ==, dmu_bonus_hold(os
, lr
->lr_foid
, FTAG
, &db
));
1528 tx
= dmu_tx_create(os
);
1529 dmu_tx_hold_bonus(tx
, lr
->lr_foid
);
1531 txg
= ztest_tx_assign(tx
, TXG_WAIT
, FTAG
);
1533 dmu_buf_rele(db
, FTAG
);
1534 ztest_object_unlock(zd
, lr
->lr_foid
);
1538 bbt
= ztest_bt_bonus(db
);
1539 ASSERT3U(bbt
->bt_magic
, ==, BT_MAGIC
);
1540 crtxg
= bbt
->bt_crtxg
;
1541 lrtxg
= lr
->lr_common
.lrc_txg
;
1543 if (zd
->zd_zilog
->zl_replay
) {
1544 ASSERT(lr
->lr_size
!= 0);
1545 ASSERT(lr
->lr_mode
!= 0);
1549 * Randomly change the size and increment the generation.
1551 lr
->lr_size
= (ztest_random(db
->db_size
/ sizeof (*bbt
)) + 1) *
1553 lr
->lr_mode
= bbt
->bt_gen
+ 1;
1558 * Verify that the current bonus buffer is not newer than our txg.
1560 ztest_bt_verify(bbt
, os
, lr
->lr_foid
, -1ULL, lr
->lr_mode
,
1561 MAX(txg
, lrtxg
), crtxg
);
1563 dmu_buf_will_dirty(db
, tx
);
1565 ASSERT3U(lr
->lr_size
, >=, sizeof (*bbt
));
1566 ASSERT3U(lr
->lr_size
, <=, db
->db_size
);
1567 VERIFY3U(dmu_set_bonus(db
, lr
->lr_size
, tx
), ==, 0);
1568 bbt
= ztest_bt_bonus(db
);
1570 ztest_bt_generate(bbt
, os
, lr
->lr_foid
, -1ULL, lr
->lr_mode
, txg
, crtxg
);
1572 dmu_buf_rele(db
, FTAG
);
1574 (void) ztest_log_setattr(zd
, tx
, lr
);
1578 ztest_object_unlock(zd
, lr
->lr_foid
);
1583 zil_replay_func_t
*ztest_replay_vector
[TX_MAX_TYPE
] = {
1584 NULL
, /* 0 no such transaction type */
1585 (zil_replay_func_t
*)ztest_replay_create
, /* TX_CREATE */
1586 NULL
, /* TX_MKDIR */
1587 NULL
, /* TX_MKXATTR */
1588 NULL
, /* TX_SYMLINK */
1589 (zil_replay_func_t
*)ztest_replay_remove
, /* TX_REMOVE */
1590 NULL
, /* TX_RMDIR */
1592 NULL
, /* TX_RENAME */
1593 (zil_replay_func_t
*)ztest_replay_write
, /* TX_WRITE */
1594 (zil_replay_func_t
*)ztest_replay_truncate
, /* TX_TRUNCATE */
1595 (zil_replay_func_t
*)ztest_replay_setattr
, /* TX_SETATTR */
1597 NULL
, /* TX_CREATE_ACL */
1598 NULL
, /* TX_CREATE_ATTR */
1599 NULL
, /* TX_CREATE_ACL_ATTR */
1600 NULL
, /* TX_MKDIR_ACL */
1601 NULL
, /* TX_MKDIR_ATTR */
1602 NULL
, /* TX_MKDIR_ACL_ATTR */
1603 NULL
, /* TX_WRITE2 */
1607 * ZIL get_data callbacks
1611 ztest_get_done(zgd_t
*zgd
, int error
)
1613 ztest_ds_t
*zd
= zgd
->zgd_private
;
1614 uint64_t object
= zgd
->zgd_rl
->rl_object
;
1617 dmu_buf_rele(zgd
->zgd_db
, zgd
);
1619 ztest_range_unlock(zgd
->zgd_rl
);
1620 ztest_object_unlock(zd
, object
);
1622 if (error
== 0 && zgd
->zgd_bp
)
1623 zil_add_block(zgd
->zgd_zilog
, zgd
->zgd_bp
);
1625 umem_free(zgd
, sizeof (*zgd
));
1629 ztest_get_data(void *arg
, lr_write_t
*lr
, char *buf
, zio_t
*zio
)
1631 ztest_ds_t
*zd
= arg
;
1632 objset_t
*os
= zd
->zd_os
;
1633 uint64_t object
= lr
->lr_foid
;
1634 uint64_t offset
= lr
->lr_offset
;
1635 uint64_t size
= lr
->lr_length
;
1636 blkptr_t
*bp
= &lr
->lr_blkptr
;
1637 uint64_t txg
= lr
->lr_common
.lrc_txg
;
1639 dmu_object_info_t doi
;
1644 ztest_object_lock(zd
, object
, RL_READER
);
1645 error
= dmu_bonus_hold(os
, object
, FTAG
, &db
);
1647 ztest_object_unlock(zd
, object
);
1651 crtxg
= ztest_bt_bonus(db
)->bt_crtxg
;
1653 if (crtxg
== 0 || crtxg
> txg
) {
1654 dmu_buf_rele(db
, FTAG
);
1655 ztest_object_unlock(zd
, object
);
1659 dmu_object_info_from_db(db
, &doi
);
1660 dmu_buf_rele(db
, FTAG
);
1663 zgd
= umem_zalloc(sizeof (*zgd
), UMEM_NOFAIL
);
1664 zgd
->zgd_zilog
= zd
->zd_zilog
;
1665 zgd
->zgd_private
= zd
;
1667 if (buf
!= NULL
) { /* immediate write */
1668 zgd
->zgd_rl
= ztest_range_lock(zd
, object
, offset
, size
,
1671 error
= dmu_read(os
, object
, offset
, size
, buf
,
1672 DMU_READ_NO_PREFETCH
);
1675 size
= doi
.doi_data_block_size
;
1677 offset
= P2ALIGN(offset
, size
);
1679 ASSERT(offset
< size
);
1683 zgd
->zgd_rl
= ztest_range_lock(zd
, object
, offset
, size
,
1686 error
= dmu_buf_hold(os
, object
, offset
, zgd
, &db
,
1687 DMU_READ_NO_PREFETCH
);
1693 ASSERT(db
->db_offset
== offset
);
1694 ASSERT(db
->db_size
== size
);
1696 error
= dmu_sync(zio
, lr
->lr_common
.lrc_txg
,
1697 ztest_get_done
, zgd
);
1704 ztest_get_done(zgd
, error
);
1710 ztest_lr_alloc(size_t lrsize
, char *name
)
1713 size_t namesize
= name
? strlen(name
) + 1 : 0;
1715 lr
= umem_zalloc(lrsize
+ namesize
, UMEM_NOFAIL
);
1718 bcopy(name
, lr
+ lrsize
, namesize
);
1724 ztest_lr_free(void *lr
, size_t lrsize
, char *name
)
1726 size_t namesize
= name
? strlen(name
) + 1 : 0;
1728 umem_free(lr
, lrsize
+ namesize
);
1732 * Lookup a bunch of objects. Returns the number of objects not found.
1735 ztest_lookup(ztest_ds_t
*zd
, ztest_od_t
*od
, int count
)
1741 ASSERT(_mutex_held(&zd
->zd_dirobj_lock
));
1743 for (i
= 0; i
< count
; i
++, od
++) {
1745 error
= zap_lookup(zd
->zd_os
, od
->od_dir
, od
->od_name
,
1746 sizeof (uint64_t), 1, &od
->od_object
);
1748 ASSERT(error
== ENOENT
);
1749 ASSERT(od
->od_object
== 0);
1753 ztest_block_tag_t
*bbt
;
1754 dmu_object_info_t doi
;
1756 ASSERT(od
->od_object
!= 0);
1757 ASSERT(missing
== 0); /* there should be no gaps */
1759 ztest_object_lock(zd
, od
->od_object
, RL_READER
);
1760 VERIFY3U(0, ==, dmu_bonus_hold(zd
->zd_os
,
1761 od
->od_object
, FTAG
, &db
));
1762 dmu_object_info_from_db(db
, &doi
);
1763 bbt
= ztest_bt_bonus(db
);
1764 ASSERT3U(bbt
->bt_magic
, ==, BT_MAGIC
);
1765 od
->od_type
= doi
.doi_type
;
1766 od
->od_blocksize
= doi
.doi_data_block_size
;
1767 od
->od_gen
= bbt
->bt_gen
;
1768 dmu_buf_rele(db
, FTAG
);
1769 ztest_object_unlock(zd
, od
->od_object
);
1777 ztest_create(ztest_ds_t
*zd
, ztest_od_t
*od
, int count
)
1782 ASSERT(_mutex_held(&zd
->zd_dirobj_lock
));
1784 for (i
= 0; i
< count
; i
++, od
++) {
1791 lr_create_t
*lr
= ztest_lr_alloc(sizeof (*lr
), od
->od_name
);
1793 lr
->lr_doid
= od
->od_dir
;
1794 lr
->lr_foid
= 0; /* 0 to allocate, > 0 to claim */
1795 lr
->lrz_type
= od
->od_crtype
;
1796 lr
->lrz_blocksize
= od
->od_crblocksize
;
1797 lr
->lrz_ibshift
= ztest_random_ibshift();
1798 lr
->lrz_bonustype
= DMU_OT_UINT64_OTHER
;
1799 lr
->lrz_bonuslen
= dmu_bonus_max();
1800 lr
->lr_gen
= od
->od_crgen
;
1801 lr
->lr_crtime
[0] = time(NULL
);
1803 if (ztest_replay_create(zd
, lr
, B_FALSE
) != 0) {
1804 ASSERT(missing
== 0);
1808 od
->od_object
= lr
->lr_foid
;
1809 od
->od_type
= od
->od_crtype
;
1810 od
->od_blocksize
= od
->od_crblocksize
;
1811 od
->od_gen
= od
->od_crgen
;
1812 ASSERT(od
->od_object
!= 0);
1815 ztest_lr_free(lr
, sizeof (*lr
), od
->od_name
);
1822 ztest_remove(ztest_ds_t
*zd
, ztest_od_t
*od
, int count
)
1828 ASSERT(_mutex_held(&zd
->zd_dirobj_lock
));
1832 for (i
= count
- 1; i
>= 0; i
--, od
--) {
1838 if (od
->od_object
== 0)
1841 lr_remove_t
*lr
= ztest_lr_alloc(sizeof (*lr
), od
->od_name
);
1843 lr
->lr_doid
= od
->od_dir
;
1845 if ((error
= ztest_replay_remove(zd
, lr
, B_FALSE
)) != 0) {
1846 ASSERT3U(error
, ==, ENOSPC
);
1851 ztest_lr_free(lr
, sizeof (*lr
), od
->od_name
);
1858 ztest_write(ztest_ds_t
*zd
, uint64_t object
, uint64_t offset
, uint64_t size
,
1864 lr
= ztest_lr_alloc(sizeof (*lr
) + size
, NULL
);
1866 lr
->lr_foid
= object
;
1867 lr
->lr_offset
= offset
;
1868 lr
->lr_length
= size
;
1870 BP_ZERO(&lr
->lr_blkptr
);
1872 bcopy(data
, lr
+ 1, size
);
1874 error
= ztest_replay_write(zd
, lr
, B_FALSE
);
1876 ztest_lr_free(lr
, sizeof (*lr
) + size
, NULL
);
1882 ztest_truncate(ztest_ds_t
*zd
, uint64_t object
, uint64_t offset
, uint64_t size
)
1887 lr
= ztest_lr_alloc(sizeof (*lr
), NULL
);
1889 lr
->lr_foid
= object
;
1890 lr
->lr_offset
= offset
;
1891 lr
->lr_length
= size
;
1893 error
= ztest_replay_truncate(zd
, lr
, B_FALSE
);
1895 ztest_lr_free(lr
, sizeof (*lr
), NULL
);
1901 ztest_setattr(ztest_ds_t
*zd
, uint64_t object
)
1906 lr
= ztest_lr_alloc(sizeof (*lr
), NULL
);
1908 lr
->lr_foid
= object
;
1912 error
= ztest_replay_setattr(zd
, lr
, B_FALSE
);
1914 ztest_lr_free(lr
, sizeof (*lr
), NULL
);
1920 ztest_prealloc(ztest_ds_t
*zd
, uint64_t object
, uint64_t offset
, uint64_t size
)
1922 objset_t
*os
= zd
->zd_os
;
1927 txg_wait_synced(dmu_objset_pool(os
), 0);
1929 ztest_object_lock(zd
, object
, RL_READER
);
1930 rl
= ztest_range_lock(zd
, object
, offset
, size
, RL_WRITER
);
1932 tx
= dmu_tx_create(os
);
1934 dmu_tx_hold_write(tx
, object
, offset
, size
);
1936 txg
= ztest_tx_assign(tx
, TXG_WAIT
, FTAG
);
1939 dmu_prealloc(os
, object
, offset
, size
, tx
);
1941 txg_wait_synced(dmu_objset_pool(os
), txg
);
1943 (void) dmu_free_long_range(os
, object
, offset
, size
);
1946 ztest_range_unlock(rl
);
1947 ztest_object_unlock(zd
, object
);
1951 ztest_io(ztest_ds_t
*zd
, uint64_t object
, uint64_t offset
)
1953 ztest_block_tag_t wbt
;
1954 dmu_object_info_t doi
;
1955 enum ztest_io_type io_type
;
1959 VERIFY(dmu_object_info(zd
->zd_os
, object
, &doi
) == 0);
1960 blocksize
= doi
.doi_data_block_size
;
1961 data
= umem_alloc(blocksize
, UMEM_NOFAIL
);
1964 * Pick an i/o type at random, biased toward writing block tags.
1966 io_type
= ztest_random(ZTEST_IO_TYPES
);
1967 if (ztest_random(2) == 0)
1968 io_type
= ZTEST_IO_WRITE_TAG
;
1972 case ZTEST_IO_WRITE_TAG
:
1973 ztest_bt_generate(&wbt
, zd
->zd_os
, object
, offset
, 0, 0, 0);
1974 (void) ztest_write(zd
, object
, offset
, sizeof (wbt
), &wbt
);
1977 case ZTEST_IO_WRITE_PATTERN
:
1978 (void) memset(data
, 'a' + (object
+ offset
) % 5, blocksize
);
1979 if (ztest_random(2) == 0) {
1981 * Induce fletcher2 collisions to ensure that
1982 * zio_ddt_collision() detects and resolves them
1983 * when using fletcher2-verify for deduplication.
1985 ((uint64_t *)data
)[0] ^= 1ULL << 63;
1986 ((uint64_t *)data
)[4] ^= 1ULL << 63;
1988 (void) ztest_write(zd
, object
, offset
, blocksize
, data
);
1991 case ZTEST_IO_WRITE_ZEROES
:
1992 bzero(data
, blocksize
);
1993 (void) ztest_write(zd
, object
, offset
, blocksize
, data
);
1996 case ZTEST_IO_TRUNCATE
:
1997 (void) ztest_truncate(zd
, object
, offset
, blocksize
);
2000 case ZTEST_IO_SETATTR
:
2001 (void) ztest_setattr(zd
, object
);
2007 umem_free(data
, blocksize
);
2011 * Initialize an object description template.
2014 ztest_od_init(ztest_od_t
*od
, uint64_t id
, char *tag
, uint64_t index
,
2015 dmu_object_type_t type
, uint64_t blocksize
, uint64_t gen
)
2017 od
->od_dir
= ZTEST_DIROBJ
;
2020 od
->od_crtype
= type
;
2021 od
->od_crblocksize
= blocksize
? blocksize
: ztest_random_blocksize();
2024 od
->od_type
= DMU_OT_NONE
;
2025 od
->od_blocksize
= 0;
2028 (void) snprintf(od
->od_name
, sizeof (od
->od_name
), "%s(%lld)[%llu]",
2029 tag
, (longlong_t
)id
, (u_longlong_t
)index
);
2033 * Lookup or create the objects for a test using the od template.
2034 * If the objects do not all exist, or if 'remove' is specified,
2035 * remove any existing objects and create new ones. Otherwise,
2036 * use the existing objects.
2039 ztest_object_init(ztest_ds_t
*zd
, ztest_od_t
*od
, size_t size
, boolean_t remove
)
2041 int count
= size
/ sizeof (*od
);
2044 VERIFY(mutex_lock(&zd
->zd_dirobj_lock
) == 0);
2045 if ((ztest_lookup(zd
, od
, count
) != 0 || remove
) &&
2046 (ztest_remove(zd
, od
, count
) != 0 ||
2047 ztest_create(zd
, od
, count
) != 0))
2050 VERIFY(mutex_unlock(&zd
->zd_dirobj_lock
) == 0);
2057 ztest_zil_commit(ztest_ds_t
*zd
, uint64_t id
)
2059 zilog_t
*zilog
= zd
->zd_zilog
;
2061 zil_commit(zilog
, ztest_random(ZTEST_OBJECTS
));
2064 * Remember the committed values in zd, which is in parent/child
2065 * shared memory. If we die, the next iteration of ztest_run()
2066 * will verify that the log really does contain this record.
2068 mutex_enter(&zilog
->zl_lock
);
2069 ASSERT(zd
->zd_seq
<= zilog
->zl_commit_lr_seq
);
2070 zd
->zd_seq
= zilog
->zl_commit_lr_seq
;
2071 mutex_exit(&zilog
->zl_lock
);
2075 * Verify that we can't destroy an active pool, create an existing pool,
2076 * or create a pool with a bad vdev spec.
2080 ztest_spa_create_destroy(ztest_ds_t
*zd
, uint64_t id
)
2082 ztest_shared_t
*zs
= ztest_shared
;
2087 * Attempt to create using a bad file.
2089 nvroot
= make_vdev_root("/dev/bogus", NULL
, 0, 0, 0, 0, 0, 1);
2090 VERIFY3U(ENOENT
, ==,
2091 spa_create("ztest_bad_file", nvroot
, NULL
, NULL
, NULL
));
2092 nvlist_free(nvroot
);
2095 * Attempt to create using a bad mirror.
2097 nvroot
= make_vdev_root("/dev/bogus", NULL
, 0, 0, 0, 0, 2, 1);
2098 VERIFY3U(ENOENT
, ==,
2099 spa_create("ztest_bad_mirror", nvroot
, NULL
, NULL
, NULL
));
2100 nvlist_free(nvroot
);
2103 * Attempt to create an existing pool. It shouldn't matter
2104 * what's in the nvroot; we should fail with EEXIST.
2106 (void) rw_rdlock(&zs
->zs_name_lock
);
2107 nvroot
= make_vdev_root("/dev/bogus", NULL
, 0, 0, 0, 0, 0, 1);
2108 VERIFY3U(EEXIST
, ==, spa_create(zs
->zs_pool
, nvroot
, NULL
, NULL
, NULL
));
2109 nvlist_free(nvroot
);
2110 VERIFY3U(0, ==, spa_open(zs
->zs_pool
, &spa
, FTAG
));
2111 VERIFY3U(EBUSY
, ==, spa_destroy(zs
->zs_pool
));
2112 spa_close(spa
, FTAG
);
2114 (void) rw_unlock(&zs
->zs_name_lock
);
2118 vdev_lookup_by_path(vdev_t
*vd
, const char *path
)
2123 if (vd
->vdev_path
!= NULL
&& strcmp(path
, vd
->vdev_path
) == 0)
2126 for (c
= 0; c
< vd
->vdev_children
; c
++)
2127 if ((mvd
= vdev_lookup_by_path(vd
->vdev_child
[c
], path
)) !=
2135 * Find the first available hole which can be used as a top-level.
2138 find_vdev_hole(spa_t
*spa
)
2140 vdev_t
*rvd
= spa
->spa_root_vdev
;
2143 ASSERT(spa_config_held(spa
, SCL_VDEV
, RW_READER
) == SCL_VDEV
);
2145 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
2146 vdev_t
*cvd
= rvd
->vdev_child
[c
];
2148 if (cvd
->vdev_ishole
)
2155 * Verify that vdev_add() works as expected.
2159 ztest_vdev_add_remove(ztest_ds_t
*zd
, uint64_t id
)
2161 ztest_shared_t
*zs
= ztest_shared
;
2162 spa_t
*spa
= zs
->zs_spa
;
2168 VERIFY(mutex_lock(&zs
->zs_vdev_lock
) == 0);
2169 leaves
= MAX(zs
->zs_mirrors
+ zs
->zs_splits
, 1) * zopt_raidz
;
2171 spa_config_enter(spa
, SCL_VDEV
, FTAG
, RW_READER
);
2173 ztest_shared
->zs_vdev_next_leaf
= find_vdev_hole(spa
) * leaves
;
2176 * If we have slogs then remove them 1/4 of the time.
2178 if (spa_has_slogs(spa
) && ztest_random(4) == 0) {
2180 * Grab the guid from the head of the log class rotor.
2182 guid
= spa_log_class(spa
)->mc_rotor
->mg_vd
->vdev_guid
;
2184 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
2187 * We have to grab the zs_name_lock as writer to
2188 * prevent a race between removing a slog (dmu_objset_find)
2189 * and destroying a dataset. Removing the slog will
2190 * grab a reference on the dataset which may cause
2191 * dmu_objset_destroy() to fail with EBUSY thus
2192 * leaving the dataset in an inconsistent state.
2194 VERIFY(rw_wrlock(&ztest_shared
->zs_name_lock
) == 0);
2195 error
= spa_vdev_remove(spa
, guid
, B_FALSE
);
2196 VERIFY(rw_unlock(&ztest_shared
->zs_name_lock
) == 0);
2198 if (error
&& error
!= EEXIST
)
2199 fatal(0, "spa_vdev_remove() = %d", error
);
2201 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
2204 * Make 1/4 of the devices be log devices.
2206 nvroot
= make_vdev_root(NULL
, NULL
, zopt_vdev_size
, 0,
2207 ztest_random(4) == 0, zopt_raidz
, zs
->zs_mirrors
, 1);
2209 error
= spa_vdev_add(spa
, nvroot
);
2210 nvlist_free(nvroot
);
2212 if (error
== ENOSPC
)
2213 ztest_record_enospc("spa_vdev_add");
2214 else if (error
!= 0)
2215 fatal(0, "spa_vdev_add() = %d", error
);
2218 VERIFY(mutex_unlock(&ztest_shared
->zs_vdev_lock
) == 0);
2222 * Verify that adding/removing aux devices (l2arc, hot spare) works as expected.
2226 ztest_vdev_aux_add_remove(ztest_ds_t
*zd
, uint64_t id
)
2228 ztest_shared_t
*zs
= ztest_shared
;
2229 spa_t
*spa
= zs
->zs_spa
;
2230 vdev_t
*rvd
= spa
->spa_root_vdev
;
2231 spa_aux_vdev_t
*sav
;
2236 if (ztest_random(2) == 0) {
2237 sav
= &spa
->spa_spares
;
2238 aux
= ZPOOL_CONFIG_SPARES
;
2240 sav
= &spa
->spa_l2cache
;
2241 aux
= ZPOOL_CONFIG_L2CACHE
;
2244 VERIFY(mutex_lock(&zs
->zs_vdev_lock
) == 0);
2246 spa_config_enter(spa
, SCL_VDEV
, FTAG
, RW_READER
);
2248 if (sav
->sav_count
!= 0 && ztest_random(4) == 0) {
2250 * Pick a random device to remove.
2252 guid
= sav
->sav_vdevs
[ztest_random(sav
->sav_count
)]->vdev_guid
;
2255 * Find an unused device we can add.
2257 zs
->zs_vdev_aux
= 0;
2259 char path
[MAXPATHLEN
];
2261 (void) sprintf(path
, ztest_aux_template
, zopt_dir
,
2262 zopt_pool
, aux
, zs
->zs_vdev_aux
);
2263 for (c
= 0; c
< sav
->sav_count
; c
++)
2264 if (strcmp(sav
->sav_vdevs
[c
]->vdev_path
,
2267 if (c
== sav
->sav_count
&&
2268 vdev_lookup_by_path(rvd
, path
) == NULL
)
2274 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
2280 nvlist_t
*nvroot
= make_vdev_root(NULL
, aux
,
2281 (zopt_vdev_size
* 5) / 4, 0, 0, 0, 0, 1);
2282 error
= spa_vdev_add(spa
, nvroot
);
2284 fatal(0, "spa_vdev_add(%p) = %d", nvroot
, error
);
2285 nvlist_free(nvroot
);
2288 * Remove an existing device. Sometimes, dirty its
2289 * vdev state first to make sure we handle removal
2290 * of devices that have pending state changes.
2292 if (ztest_random(2) == 0)
2293 (void) vdev_online(spa
, guid
, 0, NULL
);
2295 error
= spa_vdev_remove(spa
, guid
, B_FALSE
);
2296 if (error
!= 0 && error
!= EBUSY
)
2297 fatal(0, "spa_vdev_remove(%llu) = %d", guid
, error
);
2300 VERIFY(mutex_unlock(&zs
->zs_vdev_lock
) == 0);
2304 * split a pool if it has mirror tlvdevs
2308 ztest_split_pool(ztest_ds_t
*zd
, uint64_t id
)
2310 ztest_shared_t
*zs
= ztest_shared
;
2311 spa_t
*spa
= zs
->zs_spa
;
2312 vdev_t
*rvd
= spa
->spa_root_vdev
;
2313 nvlist_t
*tree
, **child
, *config
, *split
, **schild
;
2314 uint_t c
, children
, schildren
= 0, lastlogid
= 0;
2317 VERIFY(mutex_lock(&zs
->zs_vdev_lock
) == 0);
2319 /* ensure we have a useable config; mirrors of raidz aren't supported */
2320 if (zs
->zs_mirrors
< 3 || zopt_raidz
> 1) {
2321 VERIFY(mutex_unlock(&zs
->zs_vdev_lock
) == 0);
2325 /* clean up the old pool, if any */
2326 (void) spa_destroy("splitp");
2328 spa_config_enter(spa
, SCL_VDEV
, FTAG
, RW_READER
);
2330 /* generate a config from the existing config */
2331 mutex_enter(&spa
->spa_props_lock
);
2332 VERIFY(nvlist_lookup_nvlist(spa
->spa_config
, ZPOOL_CONFIG_VDEV_TREE
,
2334 mutex_exit(&spa
->spa_props_lock
);
2336 VERIFY(nvlist_lookup_nvlist_array(tree
, ZPOOL_CONFIG_CHILDREN
, &child
,
2339 schild
= malloc(rvd
->vdev_children
* sizeof (nvlist_t
*));
2340 for (c
= 0; c
< children
; c
++) {
2341 vdev_t
*tvd
= rvd
->vdev_child
[c
];
2345 if (tvd
->vdev_islog
|| tvd
->vdev_ops
== &vdev_hole_ops
) {
2346 VERIFY(nvlist_alloc(&schild
[schildren
], NV_UNIQUE_NAME
,
2348 VERIFY(nvlist_add_string(schild
[schildren
],
2349 ZPOOL_CONFIG_TYPE
, VDEV_TYPE_HOLE
) == 0);
2350 VERIFY(nvlist_add_uint64(schild
[schildren
],
2351 ZPOOL_CONFIG_IS_HOLE
, 1) == 0);
2353 lastlogid
= schildren
;
2358 VERIFY(nvlist_lookup_nvlist_array(child
[c
],
2359 ZPOOL_CONFIG_CHILDREN
, &mchild
, &mchildren
) == 0);
2360 VERIFY(nvlist_dup(mchild
[0], &schild
[schildren
++], 0) == 0);
2363 /* OK, create a config that can be used to split */
2364 VERIFY(nvlist_alloc(&split
, NV_UNIQUE_NAME
, 0) == 0);
2365 VERIFY(nvlist_add_string(split
, ZPOOL_CONFIG_TYPE
,
2366 VDEV_TYPE_ROOT
) == 0);
2367 VERIFY(nvlist_add_nvlist_array(split
, ZPOOL_CONFIG_CHILDREN
, schild
,
2368 lastlogid
!= 0 ? lastlogid
: schildren
) == 0);
2370 VERIFY(nvlist_alloc(&config
, NV_UNIQUE_NAME
, 0) == 0);
2371 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, split
) == 0);
2373 for (c
= 0; c
< schildren
; c
++)
2374 nvlist_free(schild
[c
]);
2378 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
2380 (void) rw_wrlock(&zs
->zs_name_lock
);
2381 error
= spa_vdev_split_mirror(spa
, "splitp", config
, NULL
, B_FALSE
);
2382 (void) rw_unlock(&zs
->zs_name_lock
);
2384 nvlist_free(config
);
2387 (void) printf("successful split - results:\n");
2388 mutex_enter(&spa_namespace_lock
);
2389 show_pool_stats(spa
);
2390 show_pool_stats(spa_lookup("splitp"));
2391 mutex_exit(&spa_namespace_lock
);
2395 VERIFY(mutex_unlock(&zs
->zs_vdev_lock
) == 0);
2400 * Verify that we can attach and detach devices.
2404 ztest_vdev_attach_detach(ztest_ds_t
*zd
, uint64_t id
)
2406 ztest_shared_t
*zs
= ztest_shared
;
2407 spa_t
*spa
= zs
->zs_spa
;
2408 spa_aux_vdev_t
*sav
= &spa
->spa_spares
;
2409 vdev_t
*rvd
= spa
->spa_root_vdev
;
2410 vdev_t
*oldvd
, *newvd
, *pvd
;
2414 uint64_t ashift
= ztest_get_ashift();
2415 uint64_t oldguid
, pguid
;
2416 size_t oldsize
, newsize
;
2417 char oldpath
[MAXPATHLEN
], newpath
[MAXPATHLEN
];
2419 int oldvd_has_siblings
= B_FALSE
;
2420 int newvd_is_spare
= B_FALSE
;
2422 int error
, expected_error
;
2424 VERIFY(mutex_lock(&zs
->zs_vdev_lock
) == 0);
2425 leaves
= MAX(zs
->zs_mirrors
, 1) * zopt_raidz
;
2427 spa_config_enter(spa
, SCL_VDEV
, FTAG
, RW_READER
);
2430 * Decide whether to do an attach or a replace.
2432 replacing
= ztest_random(2);
2435 * Pick a random top-level vdev.
2437 top
= ztest_random_vdev_top(spa
, B_TRUE
);
2440 * Pick a random leaf within it.
2442 leaf
= ztest_random(leaves
);
2447 oldvd
= rvd
->vdev_child
[top
];
2448 if (zs
->zs_mirrors
>= 1) {
2449 ASSERT(oldvd
->vdev_ops
== &vdev_mirror_ops
);
2450 ASSERT(oldvd
->vdev_children
>= zs
->zs_mirrors
);
2451 oldvd
= oldvd
->vdev_child
[leaf
/ zopt_raidz
];
2453 if (zopt_raidz
> 1) {
2454 ASSERT(oldvd
->vdev_ops
== &vdev_raidz_ops
);
2455 ASSERT(oldvd
->vdev_children
== zopt_raidz
);
2456 oldvd
= oldvd
->vdev_child
[leaf
% zopt_raidz
];
2460 * If we're already doing an attach or replace, oldvd may be a
2461 * mirror vdev -- in which case, pick a random child.
2463 while (oldvd
->vdev_children
!= 0) {
2464 oldvd_has_siblings
= B_TRUE
;
2465 ASSERT(oldvd
->vdev_children
>= 2);
2466 oldvd
= oldvd
->vdev_child
[ztest_random(oldvd
->vdev_children
)];
2469 oldguid
= oldvd
->vdev_guid
;
2470 oldsize
= vdev_get_min_asize(oldvd
);
2471 oldvd_is_log
= oldvd
->vdev_top
->vdev_islog
;
2472 (void) strcpy(oldpath
, oldvd
->vdev_path
);
2473 pvd
= oldvd
->vdev_parent
;
2474 pguid
= pvd
->vdev_guid
;
2477 * If oldvd has siblings, then half of the time, detach it.
2479 if (oldvd_has_siblings
&& ztest_random(2) == 0) {
2480 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
2481 error
= spa_vdev_detach(spa
, oldguid
, pguid
, B_FALSE
);
2482 if (error
!= 0 && error
!= ENODEV
&& error
!= EBUSY
&&
2484 fatal(0, "detach (%s) returned %d", oldpath
, error
);
2485 VERIFY(mutex_unlock(&zs
->zs_vdev_lock
) == 0);
2490 * For the new vdev, choose with equal probability between the two
2491 * standard paths (ending in either 'a' or 'b') or a random hot spare.
2493 if (sav
->sav_count
!= 0 && ztest_random(3) == 0) {
2494 newvd
= sav
->sav_vdevs
[ztest_random(sav
->sav_count
)];
2495 newvd_is_spare
= B_TRUE
;
2496 (void) strcpy(newpath
, newvd
->vdev_path
);
2498 (void) snprintf(newpath
, sizeof (newpath
), ztest_dev_template
,
2499 zopt_dir
, zopt_pool
, top
* leaves
+ leaf
);
2500 if (ztest_random(2) == 0)
2501 newpath
[strlen(newpath
) - 1] = 'b';
2502 newvd
= vdev_lookup_by_path(rvd
, newpath
);
2506 newsize
= vdev_get_min_asize(newvd
);
2509 * Make newsize a little bigger or smaller than oldsize.
2510 * If it's smaller, the attach should fail.
2511 * If it's larger, and we're doing a replace,
2512 * we should get dynamic LUN growth when we're done.
2514 newsize
= 10 * oldsize
/ (9 + ztest_random(3));
2518 * If pvd is not a mirror or root, the attach should fail with ENOTSUP,
2519 * unless it's a replace; in that case any non-replacing parent is OK.
2521 * If newvd is already part of the pool, it should fail with EBUSY.
2523 * If newvd is too small, it should fail with EOVERFLOW.
2525 if (pvd
->vdev_ops
!= &vdev_mirror_ops
&&
2526 pvd
->vdev_ops
!= &vdev_root_ops
&& (!replacing
||
2527 pvd
->vdev_ops
== &vdev_replacing_ops
||
2528 pvd
->vdev_ops
== &vdev_spare_ops
))
2529 expected_error
= ENOTSUP
;
2530 else if (newvd_is_spare
&& (!replacing
|| oldvd_is_log
))
2531 expected_error
= ENOTSUP
;
2532 else if (newvd
== oldvd
)
2533 expected_error
= replacing
? 0 : EBUSY
;
2534 else if (vdev_lookup_by_path(rvd
, newpath
) != NULL
)
2535 expected_error
= EBUSY
;
2536 else if (newsize
< oldsize
)
2537 expected_error
= EOVERFLOW
;
2538 else if (ashift
> oldvd
->vdev_top
->vdev_ashift
)
2539 expected_error
= EDOM
;
2543 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
2546 * Build the nvlist describing newpath.
2548 root
= make_vdev_root(newpath
, NULL
, newvd
== NULL
? newsize
: 0,
2549 ashift
, 0, 0, 0, 1);
2551 error
= spa_vdev_attach(spa
, oldguid
, root
, replacing
);
2556 * If our parent was the replacing vdev, but the replace completed,
2557 * then instead of failing with ENOTSUP we may either succeed,
2558 * fail with ENODEV, or fail with EOVERFLOW.
2560 if (expected_error
== ENOTSUP
&&
2561 (error
== 0 || error
== ENODEV
|| error
== EOVERFLOW
))
2562 expected_error
= error
;
2565 * If someone grew the LUN, the replacement may be too small.
2567 if (error
== EOVERFLOW
|| error
== EBUSY
)
2568 expected_error
= error
;
2570 /* XXX workaround 6690467 */
2571 if (error
!= expected_error
&& expected_error
!= EBUSY
) {
2572 fatal(0, "attach (%s %llu, %s %llu, %d) "
2573 "returned %d, expected %d",
2574 oldpath
, (longlong_t
)oldsize
, newpath
,
2575 (longlong_t
)newsize
, replacing
, error
, expected_error
);
2578 VERIFY(mutex_unlock(&zs
->zs_vdev_lock
) == 0);
2582 * Callback function which expands the physical size of the vdev.
2585 grow_vdev(vdev_t
*vd
, void *arg
)
2587 ASSERTV(spa_t
*spa
= vd
->vdev_spa
);
2588 size_t *newsize
= arg
;
2592 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_READER
) == SCL_STATE
);
2593 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
2595 if ((fd
= open(vd
->vdev_path
, O_RDWR
)) == -1)
2598 fsize
= lseek(fd
, 0, SEEK_END
);
2599 VERIFY(ftruncate(fd
, *newsize
) == 0);
2601 if (zopt_verbose
>= 6) {
2602 (void) printf("%s grew from %lu to %lu bytes\n",
2603 vd
->vdev_path
, (ulong_t
)fsize
, (ulong_t
)*newsize
);
2610 * Callback function which expands a given vdev by calling vdev_online().
2614 online_vdev(vdev_t
*vd
, void *arg
)
2616 spa_t
*spa
= vd
->vdev_spa
;
2617 vdev_t
*tvd
= vd
->vdev_top
;
2618 uint64_t guid
= vd
->vdev_guid
;
2619 uint64_t generation
= spa
->spa_config_generation
+ 1;
2620 vdev_state_t newstate
= VDEV_STATE_UNKNOWN
;
2623 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_READER
) == SCL_STATE
);
2624 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
2626 /* Calling vdev_online will initialize the new metaslabs */
2627 spa_config_exit(spa
, SCL_STATE
, spa
);
2628 error
= vdev_online(spa
, guid
, ZFS_ONLINE_EXPAND
, &newstate
);
2629 spa_config_enter(spa
, SCL_STATE
, spa
, RW_READER
);
2632 * If vdev_online returned an error or the underlying vdev_open
2633 * failed then we abort the expand. The only way to know that
2634 * vdev_open fails is by checking the returned newstate.
2636 if (error
|| newstate
!= VDEV_STATE_HEALTHY
) {
2637 if (zopt_verbose
>= 5) {
2638 (void) printf("Unable to expand vdev, state %llu, "
2639 "error %d\n", (u_longlong_t
)newstate
, error
);
2643 ASSERT3U(newstate
, ==, VDEV_STATE_HEALTHY
);
2646 * Since we dropped the lock we need to ensure that we're
2647 * still talking to the original vdev. It's possible this
2648 * vdev may have been detached/replaced while we were
2649 * trying to online it.
2651 if (generation
!= spa
->spa_config_generation
) {
2652 if (zopt_verbose
>= 5) {
2653 (void) printf("vdev configuration has changed, "
2654 "guid %llu, state %llu, expected gen %llu, "
2657 (u_longlong_t
)tvd
->vdev_state
,
2658 (u_longlong_t
)generation
,
2659 (u_longlong_t
)spa
->spa_config_generation
);
2667 * Traverse the vdev tree calling the supplied function.
2668 * We continue to walk the tree until we either have walked all
2669 * children or we receive a non-NULL return from the callback.
2670 * If a NULL callback is passed, then we just return back the first
2671 * leaf vdev we encounter.
2674 vdev_walk_tree(vdev_t
*vd
, vdev_t
*(*func
)(vdev_t
*, void *), void *arg
)
2678 if (vd
->vdev_ops
->vdev_op_leaf
) {
2682 return (func(vd
, arg
));
2685 for (c
= 0; c
< vd
->vdev_children
; c
++) {
2686 vdev_t
*cvd
= vd
->vdev_child
[c
];
2687 if ((cvd
= vdev_walk_tree(cvd
, func
, arg
)) != NULL
)
2694 * Verify that dynamic LUN growth works as expected.
2698 ztest_vdev_LUN_growth(ztest_ds_t
*zd
, uint64_t id
)
2700 ztest_shared_t
*zs
= ztest_shared
;
2701 spa_t
*spa
= zs
->zs_spa
;
2703 metaslab_class_t
*mc
;
2704 metaslab_group_t
*mg
;
2705 size_t psize
, newsize
;
2707 uint64_t old_class_space
, new_class_space
, old_ms_count
, new_ms_count
;
2709 VERIFY(mutex_lock(&zs
->zs_vdev_lock
) == 0);
2710 spa_config_enter(spa
, SCL_STATE
, spa
, RW_READER
);
2712 top
= ztest_random_vdev_top(spa
, B_TRUE
);
2714 tvd
= spa
->spa_root_vdev
->vdev_child
[top
];
2717 old_ms_count
= tvd
->vdev_ms_count
;
2718 old_class_space
= metaslab_class_get_space(mc
);
2721 * Determine the size of the first leaf vdev associated with
2722 * our top-level device.
2724 vd
= vdev_walk_tree(tvd
, NULL
, NULL
);
2725 ASSERT3P(vd
, !=, NULL
);
2726 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
2728 psize
= vd
->vdev_psize
;
2731 * We only try to expand the vdev if it's healthy, less than 4x its
2732 * original size, and it has a valid psize.
2734 if (tvd
->vdev_state
!= VDEV_STATE_HEALTHY
||
2735 psize
== 0 || psize
>= 4 * zopt_vdev_size
) {
2736 spa_config_exit(spa
, SCL_STATE
, spa
);
2737 VERIFY(mutex_unlock(&zs
->zs_vdev_lock
) == 0);
2741 newsize
= psize
+ psize
/ 8;
2742 ASSERT3U(newsize
, >, psize
);
2744 if (zopt_verbose
>= 6) {
2745 (void) printf("Expanding LUN %s from %lu to %lu\n",
2746 vd
->vdev_path
, (ulong_t
)psize
, (ulong_t
)newsize
);
2750 * Growing the vdev is a two step process:
2751 * 1). expand the physical size (i.e. relabel)
2752 * 2). online the vdev to create the new metaslabs
2754 if (vdev_walk_tree(tvd
, grow_vdev
, &newsize
) != NULL
||
2755 vdev_walk_tree(tvd
, online_vdev
, NULL
) != NULL
||
2756 tvd
->vdev_state
!= VDEV_STATE_HEALTHY
) {
2757 if (zopt_verbose
>= 5) {
2758 (void) printf("Could not expand LUN because "
2759 "the vdev configuration changed.\n");
2761 spa_config_exit(spa
, SCL_STATE
, spa
);
2762 VERIFY(mutex_unlock(&zs
->zs_vdev_lock
) == 0);
2766 spa_config_exit(spa
, SCL_STATE
, spa
);
2769 * Expanding the LUN will update the config asynchronously,
2770 * thus we must wait for the async thread to complete any
2771 * pending tasks before proceeding.
2775 mutex_enter(&spa
->spa_async_lock
);
2776 done
= (spa
->spa_async_thread
== NULL
&& !spa
->spa_async_tasks
);
2777 mutex_exit(&spa
->spa_async_lock
);
2780 txg_wait_synced(spa_get_dsl(spa
), 0);
2781 (void) poll(NULL
, 0, 100);
2784 spa_config_enter(spa
, SCL_STATE
, spa
, RW_READER
);
2786 tvd
= spa
->spa_root_vdev
->vdev_child
[top
];
2787 new_ms_count
= tvd
->vdev_ms_count
;
2788 new_class_space
= metaslab_class_get_space(mc
);
2790 if (tvd
->vdev_mg
!= mg
|| mg
->mg_class
!= mc
) {
2791 if (zopt_verbose
>= 5) {
2792 (void) printf("Could not verify LUN expansion due to "
2793 "intervening vdev offline or remove.\n");
2795 spa_config_exit(spa
, SCL_STATE
, spa
);
2796 VERIFY(mutex_unlock(&zs
->zs_vdev_lock
) == 0);
2801 * Make sure we were able to grow the vdev.
2803 if (new_ms_count
<= old_ms_count
)
2804 fatal(0, "LUN expansion failed: ms_count %llu <= %llu\n",
2805 old_ms_count
, new_ms_count
);
2808 * Make sure we were able to grow the pool.
2810 if (new_class_space
<= old_class_space
)
2811 fatal(0, "LUN expansion failed: class_space %llu <= %llu\n",
2812 old_class_space
, new_class_space
);
2814 if (zopt_verbose
>= 5) {
2815 char oldnumbuf
[6], newnumbuf
[6];
2817 nicenum(old_class_space
, oldnumbuf
);
2818 nicenum(new_class_space
, newnumbuf
);
2819 (void) printf("%s grew from %s to %s\n",
2820 spa
->spa_name
, oldnumbuf
, newnumbuf
);
2823 spa_config_exit(spa
, SCL_STATE
, spa
);
2824 VERIFY(mutex_unlock(&zs
->zs_vdev_lock
) == 0);
2828 * Verify that dmu_objset_{create,destroy,open,close} work as expected.
2832 ztest_objset_create_cb(objset_t
*os
, void *arg
, cred_t
*cr
, dmu_tx_t
*tx
)
2835 * Create the objects common to all ztest datasets.
2837 VERIFY(zap_create_claim(os
, ZTEST_DIROBJ
,
2838 DMU_OT_ZAP_OTHER
, DMU_OT_NONE
, 0, tx
) == 0);
2842 ztest_dataset_create(char *dsname
)
2844 uint64_t zilset
= ztest_random(100);
2845 int err
= dmu_objset_create(dsname
, DMU_OST_OTHER
, 0,
2846 ztest_objset_create_cb
, NULL
);
2848 if (err
|| zilset
< 80)
2851 (void) printf("Setting dataset %s to sync always\n", dsname
);
2852 return (ztest_dsl_prop_set_uint64(dsname
, ZFS_PROP_SYNC
,
2853 ZFS_SYNC_ALWAYS
, B_FALSE
));
2858 ztest_objset_destroy_cb(const char *name
, void *arg
)
2861 dmu_object_info_t doi
;
2865 * Verify that the dataset contains a directory object.
2867 VERIFY3U(0, ==, dmu_objset_hold(name
, FTAG
, &os
));
2868 error
= dmu_object_info(os
, ZTEST_DIROBJ
, &doi
);
2869 if (error
!= ENOENT
) {
2870 /* We could have crashed in the middle of destroying it */
2871 ASSERT3U(error
, ==, 0);
2872 ASSERT3U(doi
.doi_type
, ==, DMU_OT_ZAP_OTHER
);
2873 ASSERT3S(doi
.doi_physical_blocks_512
, >=, 0);
2875 dmu_objset_rele(os
, FTAG
);
2878 * Destroy the dataset.
2880 VERIFY3U(0, ==, dmu_objset_destroy(name
, B_FALSE
));
2885 ztest_snapshot_create(char *osname
, uint64_t id
)
2887 char snapname
[MAXNAMELEN
];
2890 (void) snprintf(snapname
, MAXNAMELEN
, "%s@%llu", osname
,
2893 error
= dmu_objset_snapshot(osname
, strchr(snapname
, '@') + 1,
2894 NULL
, NULL
, B_FALSE
, B_FALSE
, -1);
2895 if (error
== ENOSPC
) {
2896 ztest_record_enospc(FTAG
);
2899 if (error
!= 0 && error
!= EEXIST
)
2900 fatal(0, "ztest_snapshot_create(%s) = %d", snapname
, error
);
2905 ztest_snapshot_destroy(char *osname
, uint64_t id
)
2907 char snapname
[MAXNAMELEN
];
2910 (void) snprintf(snapname
, MAXNAMELEN
, "%s@%llu", osname
,
2913 error
= dmu_objset_destroy(snapname
, B_FALSE
);
2914 if (error
!= 0 && error
!= ENOENT
)
2915 fatal(0, "ztest_snapshot_destroy(%s) = %d", snapname
, error
);
2921 ztest_dmu_objset_create_destroy(ztest_ds_t
*zd
, uint64_t id
)
2923 ztest_shared_t
*zs
= ztest_shared
;
2928 char name
[MAXNAMELEN
];
2932 (void) rw_rdlock(&zs
->zs_name_lock
);
2934 (void) snprintf(name
, MAXNAMELEN
, "%s/temp_%llu",
2935 zs
->zs_pool
, (u_longlong_t
)id
);
2938 * If this dataset exists from a previous run, process its replay log
2939 * half of the time. If we don't replay it, then dmu_objset_destroy()
2940 * (invoked from ztest_objset_destroy_cb()) should just throw it away.
2942 if (ztest_random(2) == 0 &&
2943 dmu_objset_own(name
, DMU_OST_OTHER
, B_FALSE
, FTAG
, &os
) == 0) {
2944 ztest_zd_init(&zdtmp
, os
);
2945 zil_replay(os
, &zdtmp
, ztest_replay_vector
);
2946 ztest_zd_fini(&zdtmp
);
2947 dmu_objset_disown(os
, FTAG
);
2951 * There may be an old instance of the dataset we're about to
2952 * create lying around from a previous run. If so, destroy it
2953 * and all of its snapshots.
2955 (void) dmu_objset_find(name
, ztest_objset_destroy_cb
, NULL
,
2956 DS_FIND_CHILDREN
| DS_FIND_SNAPSHOTS
);
2959 * Verify that the destroyed dataset is no longer in the namespace.
2961 VERIFY3U(ENOENT
, ==, dmu_objset_hold(name
, FTAG
, &os
));
2964 * Verify that we can create a new dataset.
2966 error
= ztest_dataset_create(name
);
2968 if (error
== ENOSPC
) {
2969 ztest_record_enospc(FTAG
);
2970 (void) rw_unlock(&zs
->zs_name_lock
);
2973 fatal(0, "dmu_objset_create(%s) = %d", name
, error
);
2977 dmu_objset_own(name
, DMU_OST_OTHER
, B_FALSE
, FTAG
, &os
));
2979 ztest_zd_init(&zdtmp
, os
);
2982 * Open the intent log for it.
2984 zilog
= zil_open(os
, ztest_get_data
);
2987 * Put some objects in there, do a little I/O to them,
2988 * and randomly take a couple of snapshots along the way.
2990 iters
= ztest_random(5);
2991 for (i
= 0; i
< iters
; i
++) {
2992 ztest_dmu_object_alloc_free(&zdtmp
, id
);
2993 if (ztest_random(iters
) == 0)
2994 (void) ztest_snapshot_create(name
, i
);
2998 * Verify that we cannot create an existing dataset.
3000 VERIFY3U(EEXIST
, ==,
3001 dmu_objset_create(name
, DMU_OST_OTHER
, 0, NULL
, NULL
));
3004 * Verify that we can hold an objset that is also owned.
3006 VERIFY3U(0, ==, dmu_objset_hold(name
, FTAG
, &os2
));
3007 dmu_objset_rele(os2
, FTAG
);
3010 * Verify that we cannot own an objset that is already owned.
3013 dmu_objset_own(name
, DMU_OST_OTHER
, B_FALSE
, FTAG
, &os2
));
3016 dmu_objset_disown(os
, FTAG
);
3017 ztest_zd_fini(&zdtmp
);
3019 (void) rw_unlock(&zs
->zs_name_lock
);
3023 * Verify that dmu_snapshot_{create,destroy,open,close} work as expected.
3026 ztest_dmu_snapshot_create_destroy(ztest_ds_t
*zd
, uint64_t id
)
3028 ztest_shared_t
*zs
= ztest_shared
;
3030 (void) rw_rdlock(&zs
->zs_name_lock
);
3031 (void) ztest_snapshot_destroy(zd
->zd_name
, id
);
3032 (void) ztest_snapshot_create(zd
->zd_name
, id
);
3033 (void) rw_unlock(&zs
->zs_name_lock
);
3037 * Cleanup non-standard snapshots and clones.
3040 ztest_dsl_dataset_cleanup(char *osname
, uint64_t id
)
3042 char snap1name
[MAXNAMELEN
];
3043 char clone1name
[MAXNAMELEN
];
3044 char snap2name
[MAXNAMELEN
];
3045 char clone2name
[MAXNAMELEN
];
3046 char snap3name
[MAXNAMELEN
];
3049 (void) snprintf(snap1name
, MAXNAMELEN
, "%s@s1_%llu",
3050 osname
, (u_longlong_t
)id
);
3051 (void) snprintf(clone1name
, MAXNAMELEN
, "%s/c1_%llu",
3052 osname
, (u_longlong_t
)id
);
3053 (void) snprintf(snap2name
, MAXNAMELEN
, "%s@s2_%llu",
3054 clone1name
, (u_longlong_t
)id
);
3055 (void) snprintf(clone2name
, MAXNAMELEN
, "%s/c2_%llu",
3056 osname
, (u_longlong_t
)id
);
3057 (void) snprintf(snap3name
, MAXNAMELEN
, "%s@s3_%llu",
3058 clone1name
, (u_longlong_t
)id
);
3060 error
= dmu_objset_destroy(clone2name
, B_FALSE
);
3061 if (error
&& error
!= ENOENT
)
3062 fatal(0, "dmu_objset_destroy(%s) = %d", clone2name
, error
);
3063 error
= dmu_objset_destroy(snap3name
, B_FALSE
);
3064 if (error
&& error
!= ENOENT
)
3065 fatal(0, "dmu_objset_destroy(%s) = %d", snap3name
, error
);
3066 error
= dmu_objset_destroy(snap2name
, B_FALSE
);
3067 if (error
&& error
!= ENOENT
)
3068 fatal(0, "dmu_objset_destroy(%s) = %d", snap2name
, error
);
3069 error
= dmu_objset_destroy(clone1name
, B_FALSE
);
3070 if (error
&& error
!= ENOENT
)
3071 fatal(0, "dmu_objset_destroy(%s) = %d", clone1name
, error
);
3072 error
= dmu_objset_destroy(snap1name
, B_FALSE
);
3073 if (error
&& error
!= ENOENT
)
3074 fatal(0, "dmu_objset_destroy(%s) = %d", snap1name
, error
);
3078 * Verify dsl_dataset_promote handles EBUSY
3081 ztest_dsl_dataset_promote_busy(ztest_ds_t
*zd
, uint64_t id
)
3083 ztest_shared_t
*zs
= ztest_shared
;
3086 char snap1name
[MAXNAMELEN
];
3087 char clone1name
[MAXNAMELEN
];
3088 char snap2name
[MAXNAMELEN
];
3089 char clone2name
[MAXNAMELEN
];
3090 char snap3name
[MAXNAMELEN
];
3091 char *osname
= zd
->zd_name
;
3094 (void) rw_rdlock(&zs
->zs_name_lock
);
3096 ztest_dsl_dataset_cleanup(osname
, id
);
3098 (void) snprintf(snap1name
, MAXNAMELEN
, "%s@s1_%llu",
3099 osname
, (u_longlong_t
)id
);
3100 (void) snprintf(clone1name
, MAXNAMELEN
, "%s/c1_%llu",
3101 osname
, (u_longlong_t
)id
);
3102 (void) snprintf(snap2name
, MAXNAMELEN
, "%s@s2_%llu",
3103 clone1name
, (u_longlong_t
)id
);
3104 (void) snprintf(clone2name
, MAXNAMELEN
, "%s/c2_%llu",
3105 osname
, (u_longlong_t
)id
);
3106 (void) snprintf(snap3name
, MAXNAMELEN
, "%s@s3_%llu",
3107 clone1name
, (u_longlong_t
)id
);
3109 error
= dmu_objset_snapshot(osname
, strchr(snap1name
, '@')+1,
3110 NULL
, NULL
, B_FALSE
, B_FALSE
, -1);
3111 if (error
&& error
!= EEXIST
) {
3112 if (error
== ENOSPC
) {
3113 ztest_record_enospc(FTAG
);
3116 fatal(0, "dmu_take_snapshot(%s) = %d", snap1name
, error
);
3119 error
= dmu_objset_hold(snap1name
, FTAG
, &clone
);
3121 fatal(0, "dmu_open_snapshot(%s) = %d", snap1name
, error
);
3123 error
= dmu_objset_clone(clone1name
, dmu_objset_ds(clone
), 0);
3124 dmu_objset_rele(clone
, FTAG
);
3126 if (error
== ENOSPC
) {
3127 ztest_record_enospc(FTAG
);
3130 fatal(0, "dmu_objset_create(%s) = %d", clone1name
, error
);
3133 error
= dmu_objset_snapshot(clone1name
, strchr(snap2name
, '@')+1,
3134 NULL
, NULL
, B_FALSE
, B_FALSE
, -1);
3135 if (error
&& error
!= EEXIST
) {
3136 if (error
== ENOSPC
) {
3137 ztest_record_enospc(FTAG
);
3140 fatal(0, "dmu_open_snapshot(%s) = %d", snap2name
, error
);
3143 error
= dmu_objset_snapshot(clone1name
, strchr(snap3name
, '@')+1,
3144 NULL
, NULL
, B_FALSE
, B_FALSE
, -1);
3145 if (error
&& error
!= EEXIST
) {
3146 if (error
== ENOSPC
) {
3147 ztest_record_enospc(FTAG
);
3150 fatal(0, "dmu_open_snapshot(%s) = %d", snap3name
, error
);
3153 error
= dmu_objset_hold(snap3name
, FTAG
, &clone
);
3155 fatal(0, "dmu_open_snapshot(%s) = %d", snap3name
, error
);
3157 error
= dmu_objset_clone(clone2name
, dmu_objset_ds(clone
), 0);
3158 dmu_objset_rele(clone
, FTAG
);
3160 if (error
== ENOSPC
) {
3161 ztest_record_enospc(FTAG
);
3164 fatal(0, "dmu_objset_create(%s) = %d", clone2name
, error
);
3167 error
= dsl_dataset_own(snap2name
, B_FALSE
, FTAG
, &ds
);
3169 fatal(0, "dsl_dataset_own(%s) = %d", snap2name
, error
);
3170 error
= dsl_dataset_promote(clone2name
, NULL
);
3172 fatal(0, "dsl_dataset_promote(%s), %d, not EBUSY", clone2name
,
3174 dsl_dataset_disown(ds
, FTAG
);
3177 ztest_dsl_dataset_cleanup(osname
, id
);
3179 (void) rw_unlock(&zs
->zs_name_lock
);
3183 * Verify that dmu_object_{alloc,free} work as expected.
3186 ztest_dmu_object_alloc_free(ztest_ds_t
*zd
, uint64_t id
)
3189 int batchsize
= sizeof (od
) / sizeof (od
[0]);
3192 for (b
= 0; b
< batchsize
; b
++)
3193 ztest_od_init(&od
[b
], id
, FTAG
, b
, DMU_OT_UINT64_OTHER
, 0, 0);
3196 * Destroy the previous batch of objects, create a new batch,
3197 * and do some I/O on the new objects.
3199 if (ztest_object_init(zd
, od
, sizeof (od
), B_TRUE
) != 0)
3202 while (ztest_random(4 * batchsize
) != 0)
3203 ztest_io(zd
, od
[ztest_random(batchsize
)].od_object
,
3204 ztest_random(ZTEST_RANGE_LOCKS
) << SPA_MAXBLOCKSHIFT
);
3208 * Verify that dmu_{read,write} work as expected.
3211 ztest_dmu_read_write(ztest_ds_t
*zd
, uint64_t id
)
3213 objset_t
*os
= zd
->zd_os
;
3216 int i
, freeit
, error
;
3218 bufwad_t
*packbuf
, *bigbuf
, *pack
, *bigH
, *bigT
;
3219 uint64_t packobj
, packoff
, packsize
, bigobj
, bigoff
, bigsize
;
3220 uint64_t chunksize
= (1000 + ztest_random(1000)) * sizeof (uint64_t);
3221 uint64_t regions
= 997;
3222 uint64_t stride
= 123456789ULL;
3223 uint64_t width
= 40;
3224 int free_percent
= 5;
3227 * This test uses two objects, packobj and bigobj, that are always
3228 * updated together (i.e. in the same tx) so that their contents are
3229 * in sync and can be compared. Their contents relate to each other
3230 * in a simple way: packobj is a dense array of 'bufwad' structures,
3231 * while bigobj is a sparse array of the same bufwads. Specifically,
3232 * for any index n, there are three bufwads that should be identical:
3234 * packobj, at offset n * sizeof (bufwad_t)
3235 * bigobj, at the head of the nth chunk
3236 * bigobj, at the tail of the nth chunk
3238 * The chunk size is arbitrary. It doesn't have to be a power of two,
3239 * and it doesn't have any relation to the object blocksize.
3240 * The only requirement is that it can hold at least two bufwads.
3242 * Normally, we write the bufwad to each of these locations.
3243 * However, free_percent of the time we instead write zeroes to
3244 * packobj and perform a dmu_free_range() on bigobj. By comparing
3245 * bigobj to packobj, we can verify that the DMU is correctly
3246 * tracking which parts of an object are allocated and free,
3247 * and that the contents of the allocated blocks are correct.
3251 * Read the directory info. If it's the first time, set things up.
3253 ztest_od_init(&od
[0], id
, FTAG
, 0, DMU_OT_UINT64_OTHER
, 0, chunksize
);
3254 ztest_od_init(&od
[1], id
, FTAG
, 1, DMU_OT_UINT64_OTHER
, 0, chunksize
);
3256 if (ztest_object_init(zd
, od
, sizeof (od
), B_FALSE
) != 0)
3259 bigobj
= od
[0].od_object
;
3260 packobj
= od
[1].od_object
;
3261 chunksize
= od
[0].od_gen
;
3262 ASSERT(chunksize
== od
[1].od_gen
);
3265 * Prefetch a random chunk of the big object.
3266 * Our aim here is to get some async reads in flight
3267 * for blocks that we may free below; the DMU should
3268 * handle this race correctly.
3270 n
= ztest_random(regions
) * stride
+ ztest_random(width
);
3271 s
= 1 + ztest_random(2 * width
- 1);
3272 dmu_prefetch(os
, bigobj
, n
* chunksize
, s
* chunksize
);
3275 * Pick a random index and compute the offsets into packobj and bigobj.
3277 n
= ztest_random(regions
) * stride
+ ztest_random(width
);
3278 s
= 1 + ztest_random(width
- 1);
3280 packoff
= n
* sizeof (bufwad_t
);
3281 packsize
= s
* sizeof (bufwad_t
);
3283 bigoff
= n
* chunksize
;
3284 bigsize
= s
* chunksize
;
3286 packbuf
= umem_alloc(packsize
, UMEM_NOFAIL
);
3287 bigbuf
= umem_alloc(bigsize
, UMEM_NOFAIL
);
3290 * free_percent of the time, free a range of bigobj rather than
3293 freeit
= (ztest_random(100) < free_percent
);
3296 * Read the current contents of our objects.
3298 error
= dmu_read(os
, packobj
, packoff
, packsize
, packbuf
,
3300 ASSERT3U(error
, ==, 0);
3301 error
= dmu_read(os
, bigobj
, bigoff
, bigsize
, bigbuf
,
3303 ASSERT3U(error
, ==, 0);
3306 * Get a tx for the mods to both packobj and bigobj.
3308 tx
= dmu_tx_create(os
);
3310 dmu_tx_hold_write(tx
, packobj
, packoff
, packsize
);
3313 dmu_tx_hold_free(tx
, bigobj
, bigoff
, bigsize
);
3315 dmu_tx_hold_write(tx
, bigobj
, bigoff
, bigsize
);
3317 txg
= ztest_tx_assign(tx
, TXG_MIGHTWAIT
, FTAG
);
3319 umem_free(packbuf
, packsize
);
3320 umem_free(bigbuf
, bigsize
);
3324 dmu_object_set_checksum(os
, bigobj
,
3325 (enum zio_checksum
)ztest_random_dsl_prop(ZFS_PROP_CHECKSUM
), tx
);
3327 dmu_object_set_compress(os
, bigobj
,
3328 (enum zio_compress
)ztest_random_dsl_prop(ZFS_PROP_COMPRESSION
), tx
);
3331 * For each index from n to n + s, verify that the existing bufwad
3332 * in packobj matches the bufwads at the head and tail of the
3333 * corresponding chunk in bigobj. Then update all three bufwads
3334 * with the new values we want to write out.
3336 for (i
= 0; i
< s
; i
++) {
3338 pack
= (bufwad_t
*)((char *)packbuf
+ i
* sizeof (bufwad_t
));
3340 bigH
= (bufwad_t
*)((char *)bigbuf
+ i
* chunksize
);
3342 bigT
= (bufwad_t
*)((char *)bigH
+ chunksize
) - 1;
3344 ASSERT((uintptr_t)bigH
- (uintptr_t)bigbuf
< bigsize
);
3345 ASSERT((uintptr_t)bigT
- (uintptr_t)bigbuf
< bigsize
);
3347 if (pack
->bw_txg
> txg
)
3348 fatal(0, "future leak: got %llx, open txg is %llx",
3351 if (pack
->bw_data
!= 0 && pack
->bw_index
!= n
+ i
)
3352 fatal(0, "wrong index: got %llx, wanted %llx+%llx",
3353 pack
->bw_index
, n
, i
);
3355 if (bcmp(pack
, bigH
, sizeof (bufwad_t
)) != 0)
3356 fatal(0, "pack/bigH mismatch in %p/%p", pack
, bigH
);
3358 if (bcmp(pack
, bigT
, sizeof (bufwad_t
)) != 0)
3359 fatal(0, "pack/bigT mismatch in %p/%p", pack
, bigT
);
3362 bzero(pack
, sizeof (bufwad_t
));
3364 pack
->bw_index
= n
+ i
;
3366 pack
->bw_data
= 1 + ztest_random(-2ULL);
3373 * We've verified all the old bufwads, and made new ones.
3374 * Now write them out.
3376 dmu_write(os
, packobj
, packoff
, packsize
, packbuf
, tx
);
3379 if (zopt_verbose
>= 7) {
3380 (void) printf("freeing offset %llx size %llx"
3382 (u_longlong_t
)bigoff
,
3383 (u_longlong_t
)bigsize
,
3386 VERIFY(0 == dmu_free_range(os
, bigobj
, bigoff
, bigsize
, tx
));
3388 if (zopt_verbose
>= 7) {
3389 (void) printf("writing offset %llx size %llx"
3391 (u_longlong_t
)bigoff
,
3392 (u_longlong_t
)bigsize
,
3395 dmu_write(os
, bigobj
, bigoff
, bigsize
, bigbuf
, tx
);
3401 * Sanity check the stuff we just wrote.
3404 void *packcheck
= umem_alloc(packsize
, UMEM_NOFAIL
);
3405 void *bigcheck
= umem_alloc(bigsize
, UMEM_NOFAIL
);
3407 VERIFY(0 == dmu_read(os
, packobj
, packoff
,
3408 packsize
, packcheck
, DMU_READ_PREFETCH
));
3409 VERIFY(0 == dmu_read(os
, bigobj
, bigoff
,
3410 bigsize
, bigcheck
, DMU_READ_PREFETCH
));
3412 ASSERT(bcmp(packbuf
, packcheck
, packsize
) == 0);
3413 ASSERT(bcmp(bigbuf
, bigcheck
, bigsize
) == 0);
3415 umem_free(packcheck
, packsize
);
3416 umem_free(bigcheck
, bigsize
);
3419 umem_free(packbuf
, packsize
);
3420 umem_free(bigbuf
, bigsize
);
3424 compare_and_update_pbbufs(uint64_t s
, bufwad_t
*packbuf
, bufwad_t
*bigbuf
,
3425 uint64_t bigsize
, uint64_t n
, uint64_t chunksize
, uint64_t txg
)
3433 * For each index from n to n + s, verify that the existing bufwad
3434 * in packobj matches the bufwads at the head and tail of the
3435 * corresponding chunk in bigobj. Then update all three bufwads
3436 * with the new values we want to write out.
3438 for (i
= 0; i
< s
; i
++) {
3440 pack
= (bufwad_t
*)((char *)packbuf
+ i
* sizeof (bufwad_t
));
3442 bigH
= (bufwad_t
*)((char *)bigbuf
+ i
* chunksize
);
3444 bigT
= (bufwad_t
*)((char *)bigH
+ chunksize
) - 1;
3446 ASSERT((uintptr_t)bigH
- (uintptr_t)bigbuf
< bigsize
);
3447 ASSERT((uintptr_t)bigT
- (uintptr_t)bigbuf
< bigsize
);
3449 if (pack
->bw_txg
> txg
)
3450 fatal(0, "future leak: got %llx, open txg is %llx",
3453 if (pack
->bw_data
!= 0 && pack
->bw_index
!= n
+ i
)
3454 fatal(0, "wrong index: got %llx, wanted %llx+%llx",
3455 pack
->bw_index
, n
, i
);
3457 if (bcmp(pack
, bigH
, sizeof (bufwad_t
)) != 0)
3458 fatal(0, "pack/bigH mismatch in %p/%p", pack
, bigH
);
3460 if (bcmp(pack
, bigT
, sizeof (bufwad_t
)) != 0)
3461 fatal(0, "pack/bigT mismatch in %p/%p", pack
, bigT
);
3463 pack
->bw_index
= n
+ i
;
3465 pack
->bw_data
= 1 + ztest_random(-2ULL);
3473 ztest_dmu_read_write_zcopy(ztest_ds_t
*zd
, uint64_t id
)
3475 objset_t
*os
= zd
->zd_os
;
3481 bufwad_t
*packbuf
, *bigbuf
;
3482 uint64_t packobj
, packoff
, packsize
, bigobj
, bigoff
, bigsize
;
3483 uint64_t blocksize
= ztest_random_blocksize();
3484 uint64_t chunksize
= blocksize
;
3485 uint64_t regions
= 997;
3486 uint64_t stride
= 123456789ULL;
3488 dmu_buf_t
*bonus_db
;
3489 arc_buf_t
**bigbuf_arcbufs
;
3490 dmu_object_info_t doi
;
3493 * This test uses two objects, packobj and bigobj, that are always
3494 * updated together (i.e. in the same tx) so that their contents are
3495 * in sync and can be compared. Their contents relate to each other
3496 * in a simple way: packobj is a dense array of 'bufwad' structures,
3497 * while bigobj is a sparse array of the same bufwads. Specifically,
3498 * for any index n, there are three bufwads that should be identical:
3500 * packobj, at offset n * sizeof (bufwad_t)
3501 * bigobj, at the head of the nth chunk
3502 * bigobj, at the tail of the nth chunk
3504 * The chunk size is set equal to bigobj block size so that
3505 * dmu_assign_arcbuf() can be tested for object updates.
3509 * Read the directory info. If it's the first time, set things up.
3511 ztest_od_init(&od
[0], id
, FTAG
, 0, DMU_OT_UINT64_OTHER
, blocksize
, 0);
3512 ztest_od_init(&od
[1], id
, FTAG
, 1, DMU_OT_UINT64_OTHER
, 0, chunksize
);
3514 if (ztest_object_init(zd
, od
, sizeof (od
), B_FALSE
) != 0)
3517 bigobj
= od
[0].od_object
;
3518 packobj
= od
[1].od_object
;
3519 blocksize
= od
[0].od_blocksize
;
3520 chunksize
= blocksize
;
3521 ASSERT(chunksize
== od
[1].od_gen
);
3523 VERIFY(dmu_object_info(os
, bigobj
, &doi
) == 0);
3524 VERIFY(ISP2(doi
.doi_data_block_size
));
3525 VERIFY(chunksize
== doi
.doi_data_block_size
);
3526 VERIFY(chunksize
>= 2 * sizeof (bufwad_t
));
3529 * Pick a random index and compute the offsets into packobj and bigobj.
3531 n
= ztest_random(regions
) * stride
+ ztest_random(width
);
3532 s
= 1 + ztest_random(width
- 1);
3534 packoff
= n
* sizeof (bufwad_t
);
3535 packsize
= s
* sizeof (bufwad_t
);
3537 bigoff
= n
* chunksize
;
3538 bigsize
= s
* chunksize
;
3540 packbuf
= umem_zalloc(packsize
, UMEM_NOFAIL
);
3541 bigbuf
= umem_zalloc(bigsize
, UMEM_NOFAIL
);
3543 VERIFY3U(0, ==, dmu_bonus_hold(os
, bigobj
, FTAG
, &bonus_db
));
3545 bigbuf_arcbufs
= umem_zalloc(2 * s
* sizeof (arc_buf_t
*), UMEM_NOFAIL
);
3548 * Iteration 0 test zcopy for DB_UNCACHED dbufs.
3549 * Iteration 1 test zcopy to already referenced dbufs.
3550 * Iteration 2 test zcopy to dirty dbuf in the same txg.
3551 * Iteration 3 test zcopy to dbuf dirty in previous txg.
3552 * Iteration 4 test zcopy when dbuf is no longer dirty.
3553 * Iteration 5 test zcopy when it can't be done.
3554 * Iteration 6 one more zcopy write.
3556 for (i
= 0; i
< 7; i
++) {
3561 * In iteration 5 (i == 5) use arcbufs
3562 * that don't match bigobj blksz to test
3563 * dmu_assign_arcbuf() when it can't directly
3564 * assign an arcbuf to a dbuf.
3566 for (j
= 0; j
< s
; j
++) {
3569 dmu_request_arcbuf(bonus_db
, chunksize
);
3571 bigbuf_arcbufs
[2 * j
] =
3572 dmu_request_arcbuf(bonus_db
, chunksize
/ 2);
3573 bigbuf_arcbufs
[2 * j
+ 1] =
3574 dmu_request_arcbuf(bonus_db
, chunksize
/ 2);
3579 * Get a tx for the mods to both packobj and bigobj.
3581 tx
= dmu_tx_create(os
);
3583 dmu_tx_hold_write(tx
, packobj
, packoff
, packsize
);
3584 dmu_tx_hold_write(tx
, bigobj
, bigoff
, bigsize
);
3586 txg
= ztest_tx_assign(tx
, TXG_MIGHTWAIT
, FTAG
);
3588 umem_free(packbuf
, packsize
);
3589 umem_free(bigbuf
, bigsize
);
3590 for (j
= 0; j
< s
; j
++) {
3592 dmu_return_arcbuf(bigbuf_arcbufs
[j
]);
3595 bigbuf_arcbufs
[2 * j
]);
3597 bigbuf_arcbufs
[2 * j
+ 1]);
3600 umem_free(bigbuf_arcbufs
, 2 * s
* sizeof (arc_buf_t
*));
3601 dmu_buf_rele(bonus_db
, FTAG
);
3606 * 50% of the time don't read objects in the 1st iteration to
3607 * test dmu_assign_arcbuf() for the case when there're no
3608 * existing dbufs for the specified offsets.
3610 if (i
!= 0 || ztest_random(2) != 0) {
3611 error
= dmu_read(os
, packobj
, packoff
,
3612 packsize
, packbuf
, DMU_READ_PREFETCH
);
3613 ASSERT3U(error
, ==, 0);
3614 error
= dmu_read(os
, bigobj
, bigoff
, bigsize
,
3615 bigbuf
, DMU_READ_PREFETCH
);
3616 ASSERT3U(error
, ==, 0);
3618 compare_and_update_pbbufs(s
, packbuf
, bigbuf
, bigsize
,
3622 * We've verified all the old bufwads, and made new ones.
3623 * Now write them out.
3625 dmu_write(os
, packobj
, packoff
, packsize
, packbuf
, tx
);
3626 if (zopt_verbose
>= 7) {
3627 (void) printf("writing offset %llx size %llx"
3629 (u_longlong_t
)bigoff
,
3630 (u_longlong_t
)bigsize
,
3633 for (off
= bigoff
, j
= 0; j
< s
; j
++, off
+= chunksize
) {
3636 bcopy((caddr_t
)bigbuf
+ (off
- bigoff
),
3637 bigbuf_arcbufs
[j
]->b_data
, chunksize
);
3639 bcopy((caddr_t
)bigbuf
+ (off
- bigoff
),
3640 bigbuf_arcbufs
[2 * j
]->b_data
,
3642 bcopy((caddr_t
)bigbuf
+ (off
- bigoff
) +
3644 bigbuf_arcbufs
[2 * j
+ 1]->b_data
,
3649 VERIFY(dmu_buf_hold(os
, bigobj
, off
,
3650 FTAG
, &dbt
, DMU_READ_NO_PREFETCH
) == 0);
3653 dmu_assign_arcbuf(bonus_db
, off
,
3654 bigbuf_arcbufs
[j
], tx
);
3656 dmu_assign_arcbuf(bonus_db
, off
,
3657 bigbuf_arcbufs
[2 * j
], tx
);
3658 dmu_assign_arcbuf(bonus_db
,
3659 off
+ chunksize
/ 2,
3660 bigbuf_arcbufs
[2 * j
+ 1], tx
);
3663 dmu_buf_rele(dbt
, FTAG
);
3669 * Sanity check the stuff we just wrote.
3672 void *packcheck
= umem_alloc(packsize
, UMEM_NOFAIL
);
3673 void *bigcheck
= umem_alloc(bigsize
, UMEM_NOFAIL
);
3675 VERIFY(0 == dmu_read(os
, packobj
, packoff
,
3676 packsize
, packcheck
, DMU_READ_PREFETCH
));
3677 VERIFY(0 == dmu_read(os
, bigobj
, bigoff
,
3678 bigsize
, bigcheck
, DMU_READ_PREFETCH
));
3680 ASSERT(bcmp(packbuf
, packcheck
, packsize
) == 0);
3681 ASSERT(bcmp(bigbuf
, bigcheck
, bigsize
) == 0);
3683 umem_free(packcheck
, packsize
);
3684 umem_free(bigcheck
, bigsize
);
3687 txg_wait_open(dmu_objset_pool(os
), 0);
3688 } else if (i
== 3) {
3689 txg_wait_synced(dmu_objset_pool(os
), 0);
3693 dmu_buf_rele(bonus_db
, FTAG
);
3694 umem_free(packbuf
, packsize
);
3695 umem_free(bigbuf
, bigsize
);
3696 umem_free(bigbuf_arcbufs
, 2 * s
* sizeof (arc_buf_t
*));
3701 ztest_dmu_write_parallel(ztest_ds_t
*zd
, uint64_t id
)
3704 uint64_t offset
= (1ULL << (ztest_random(20) + 43)) +
3705 (ztest_random(ZTEST_RANGE_LOCKS
) << SPA_MAXBLOCKSHIFT
);
3708 * Have multiple threads write to large offsets in an object
3709 * to verify that parallel writes to an object -- even to the
3710 * same blocks within the object -- doesn't cause any trouble.
3712 ztest_od_init(&od
[0], ID_PARALLEL
, FTAG
, 0, DMU_OT_UINT64_OTHER
, 0, 0);
3714 if (ztest_object_init(zd
, od
, sizeof (od
), B_FALSE
) != 0)
3717 while (ztest_random(10) != 0)
3718 ztest_io(zd
, od
[0].od_object
, offset
);
3722 ztest_dmu_prealloc(ztest_ds_t
*zd
, uint64_t id
)
3725 uint64_t offset
= (1ULL << (ztest_random(4) + SPA_MAXBLOCKSHIFT
)) +
3726 (ztest_random(ZTEST_RANGE_LOCKS
) << SPA_MAXBLOCKSHIFT
);
3727 uint64_t count
= ztest_random(20) + 1;
3728 uint64_t blocksize
= ztest_random_blocksize();
3731 ztest_od_init(&od
[0], id
, FTAG
, 0, DMU_OT_UINT64_OTHER
, blocksize
, 0);
3733 if (ztest_object_init(zd
, od
, sizeof (od
), !ztest_random(2)) != 0)
3736 if (ztest_truncate(zd
, od
[0].od_object
, offset
, count
* blocksize
) != 0)
3739 ztest_prealloc(zd
, od
[0].od_object
, offset
, count
* blocksize
);
3741 data
= umem_zalloc(blocksize
, UMEM_NOFAIL
);
3743 while (ztest_random(count
) != 0) {
3744 uint64_t randoff
= offset
+ (ztest_random(count
) * blocksize
);
3745 if (ztest_write(zd
, od
[0].od_object
, randoff
, blocksize
,
3748 while (ztest_random(4) != 0)
3749 ztest_io(zd
, od
[0].od_object
, randoff
);
3752 umem_free(data
, blocksize
);
3756 * Verify that zap_{create,destroy,add,remove,update} work as expected.
3758 #define ZTEST_ZAP_MIN_INTS 1
3759 #define ZTEST_ZAP_MAX_INTS 4
3760 #define ZTEST_ZAP_MAX_PROPS 1000
3763 ztest_zap(ztest_ds_t
*zd
, uint64_t id
)
3765 objset_t
*os
= zd
->zd_os
;
3768 uint64_t txg
, last_txg
;
3769 uint64_t value
[ZTEST_ZAP_MAX_INTS
];
3770 uint64_t zl_ints
, zl_intsize
, prop
;
3773 char propname
[100], txgname
[100];
3775 char *hc
[2] = { "s.acl.h", ".s.open.h.hyLZlg" };
3777 ztest_od_init(&od
[0], id
, FTAG
, 0, DMU_OT_ZAP_OTHER
, 0, 0);
3779 if (ztest_object_init(zd
, od
, sizeof (od
), !ztest_random(2)) != 0)
3782 object
= od
[0].od_object
;
3785 * Generate a known hash collision, and verify that
3786 * we can lookup and remove both entries.
3788 tx
= dmu_tx_create(os
);
3789 dmu_tx_hold_zap(tx
, object
, B_TRUE
, NULL
);
3790 txg
= ztest_tx_assign(tx
, TXG_MIGHTWAIT
, FTAG
);
3793 for (i
= 0; i
< 2; i
++) {
3795 VERIFY3U(0, ==, zap_add(os
, object
, hc
[i
], sizeof (uint64_t),
3798 for (i
= 0; i
< 2; i
++) {
3799 VERIFY3U(EEXIST
, ==, zap_add(os
, object
, hc
[i
],
3800 sizeof (uint64_t), 1, &value
[i
], tx
));
3802 zap_length(os
, object
, hc
[i
], &zl_intsize
, &zl_ints
));
3803 ASSERT3U(zl_intsize
, ==, sizeof (uint64_t));
3804 ASSERT3U(zl_ints
, ==, 1);
3806 for (i
= 0; i
< 2; i
++) {
3807 VERIFY3U(0, ==, zap_remove(os
, object
, hc
[i
], tx
));
3812 * Generate a buch of random entries.
3814 ints
= MAX(ZTEST_ZAP_MIN_INTS
, object
% ZTEST_ZAP_MAX_INTS
);
3816 prop
= ztest_random(ZTEST_ZAP_MAX_PROPS
);
3817 (void) sprintf(propname
, "prop_%llu", (u_longlong_t
)prop
);
3818 (void) sprintf(txgname
, "txg_%llu", (u_longlong_t
)prop
);
3819 bzero(value
, sizeof (value
));
3823 * If these zap entries already exist, validate their contents.
3825 error
= zap_length(os
, object
, txgname
, &zl_intsize
, &zl_ints
);
3827 ASSERT3U(zl_intsize
, ==, sizeof (uint64_t));
3828 ASSERT3U(zl_ints
, ==, 1);
3830 VERIFY(zap_lookup(os
, object
, txgname
, zl_intsize
,
3831 zl_ints
, &last_txg
) == 0);
3833 VERIFY(zap_length(os
, object
, propname
, &zl_intsize
,
3836 ASSERT3U(zl_intsize
, ==, sizeof (uint64_t));
3837 ASSERT3U(zl_ints
, ==, ints
);
3839 VERIFY(zap_lookup(os
, object
, propname
, zl_intsize
,
3840 zl_ints
, value
) == 0);
3842 for (i
= 0; i
< ints
; i
++) {
3843 ASSERT3U(value
[i
], ==, last_txg
+ object
+ i
);
3846 ASSERT3U(error
, ==, ENOENT
);
3850 * Atomically update two entries in our zap object.
3851 * The first is named txg_%llu, and contains the txg
3852 * in which the property was last updated. The second
3853 * is named prop_%llu, and the nth element of its value
3854 * should be txg + object + n.
3856 tx
= dmu_tx_create(os
);
3857 dmu_tx_hold_zap(tx
, object
, B_TRUE
, NULL
);
3858 txg
= ztest_tx_assign(tx
, TXG_MIGHTWAIT
, FTAG
);
3863 fatal(0, "zap future leak: old %llu new %llu", last_txg
, txg
);
3865 for (i
= 0; i
< ints
; i
++)
3866 value
[i
] = txg
+ object
+ i
;
3868 VERIFY3U(0, ==, zap_update(os
, object
, txgname
, sizeof (uint64_t),
3870 VERIFY3U(0, ==, zap_update(os
, object
, propname
, sizeof (uint64_t),
3876 * Remove a random pair of entries.
3878 prop
= ztest_random(ZTEST_ZAP_MAX_PROPS
);
3879 (void) sprintf(propname
, "prop_%llu", (u_longlong_t
)prop
);
3880 (void) sprintf(txgname
, "txg_%llu", (u_longlong_t
)prop
);
3882 error
= zap_length(os
, object
, txgname
, &zl_intsize
, &zl_ints
);
3884 if (error
== ENOENT
)
3887 ASSERT3U(error
, ==, 0);
3889 tx
= dmu_tx_create(os
);
3890 dmu_tx_hold_zap(tx
, object
, B_TRUE
, NULL
);
3891 txg
= ztest_tx_assign(tx
, TXG_MIGHTWAIT
, FTAG
);
3894 VERIFY3U(0, ==, zap_remove(os
, object
, txgname
, tx
));
3895 VERIFY3U(0, ==, zap_remove(os
, object
, propname
, tx
));
3900 * Testcase to test the upgrading of a microzap to fatzap.
3903 ztest_fzap(ztest_ds_t
*zd
, uint64_t id
)
3905 objset_t
*os
= zd
->zd_os
;
3907 uint64_t object
, txg
;
3910 ztest_od_init(&od
[0], id
, FTAG
, 0, DMU_OT_ZAP_OTHER
, 0, 0);
3912 if (ztest_object_init(zd
, od
, sizeof (od
), !ztest_random(2)) != 0)
3915 object
= od
[0].od_object
;
3918 * Add entries to this ZAP and make sure it spills over
3919 * and gets upgraded to a fatzap. Also, since we are adding
3920 * 2050 entries we should see ptrtbl growth and leaf-block split.
3922 for (i
= 0; i
< 2050; i
++) {
3923 char name
[MAXNAMELEN
];
3928 (void) snprintf(name
, sizeof (name
), "fzap-%llu-%llu",
3929 (u_longlong_t
)id
, (u_longlong_t
)value
);
3931 tx
= dmu_tx_create(os
);
3932 dmu_tx_hold_zap(tx
, object
, B_TRUE
, name
);
3933 txg
= ztest_tx_assign(tx
, TXG_MIGHTWAIT
, FTAG
);
3936 error
= zap_add(os
, object
, name
, sizeof (uint64_t), 1,
3938 ASSERT(error
== 0 || error
== EEXIST
);
3945 ztest_zap_parallel(ztest_ds_t
*zd
, uint64_t id
)
3947 objset_t
*os
= zd
->zd_os
;
3949 uint64_t txg
, object
, count
, wsize
, wc
, zl_wsize
, zl_wc
;
3951 int i
, namelen
, error
;
3952 int micro
= ztest_random(2);
3953 char name
[20], string_value
[20];
3956 ztest_od_init(&od
[0], ID_PARALLEL
, FTAG
, micro
, DMU_OT_ZAP_OTHER
, 0, 0);
3958 if (ztest_object_init(zd
, od
, sizeof (od
), B_FALSE
) != 0)
3961 object
= od
[0].od_object
;
3964 * Generate a random name of the form 'xxx.....' where each
3965 * x is a random printable character and the dots are dots.
3966 * There are 94 such characters, and the name length goes from
3967 * 6 to 20, so there are 94^3 * 15 = 12,458,760 possible names.
3969 namelen
= ztest_random(sizeof (name
) - 5) + 5 + 1;
3971 for (i
= 0; i
< 3; i
++)
3972 name
[i
] = '!' + ztest_random('~' - '!' + 1);
3973 for (; i
< namelen
- 1; i
++)
3977 if ((namelen
& 1) || micro
) {
3978 wsize
= sizeof (txg
);
3984 data
= string_value
;
3988 VERIFY(zap_count(os
, object
, &count
) == 0);
3989 ASSERT(count
!= -1ULL);
3992 * Select an operation: length, lookup, add, update, remove.
3994 i
= ztest_random(5);
3997 tx
= dmu_tx_create(os
);
3998 dmu_tx_hold_zap(tx
, object
, B_TRUE
, NULL
);
3999 txg
= ztest_tx_assign(tx
, TXG_MIGHTWAIT
, FTAG
);
4002 bcopy(name
, string_value
, namelen
);
4006 bzero(string_value
, namelen
);
4012 error
= zap_length(os
, object
, name
, &zl_wsize
, &zl_wc
);
4014 ASSERT3U(wsize
, ==, zl_wsize
);
4015 ASSERT3U(wc
, ==, zl_wc
);
4017 ASSERT3U(error
, ==, ENOENT
);
4022 error
= zap_lookup(os
, object
, name
, wsize
, wc
, data
);
4024 if (data
== string_value
&&
4025 bcmp(name
, data
, namelen
) != 0)
4026 fatal(0, "name '%s' != val '%s' len %d",
4027 name
, data
, namelen
);
4029 ASSERT3U(error
, ==, ENOENT
);
4034 error
= zap_add(os
, object
, name
, wsize
, wc
, data
, tx
);
4035 ASSERT(error
== 0 || error
== EEXIST
);
4039 VERIFY(zap_update(os
, object
, name
, wsize
, wc
, data
, tx
) == 0);
4043 error
= zap_remove(os
, object
, name
, tx
);
4044 ASSERT(error
== 0 || error
== ENOENT
);
4053 * Commit callback data.
4055 typedef struct ztest_cb_data
{
4056 list_node_t zcd_node
;
4058 int zcd_expected_err
;
4059 boolean_t zcd_added
;
4060 boolean_t zcd_called
;
4064 /* This is the actual commit callback function */
4066 ztest_commit_callback(void *arg
, int error
)
4068 ztest_cb_data_t
*data
= arg
;
4069 uint64_t synced_txg
;
4071 VERIFY(data
!= NULL
);
4072 VERIFY3S(data
->zcd_expected_err
, ==, error
);
4073 VERIFY(!data
->zcd_called
);
4075 synced_txg
= spa_last_synced_txg(data
->zcd_spa
);
4076 if (data
->zcd_txg
> synced_txg
)
4077 fatal(0, "commit callback of txg %" PRIu64
" called prematurely"
4078 ", last synced txg = %" PRIu64
"\n", data
->zcd_txg
,
4081 data
->zcd_called
= B_TRUE
;
4083 if (error
== ECANCELED
) {
4084 ASSERT3U(data
->zcd_txg
, ==, 0);
4085 ASSERT(!data
->zcd_added
);
4088 * The private callback data should be destroyed here, but
4089 * since we are going to check the zcd_called field after
4090 * dmu_tx_abort(), we will destroy it there.
4095 /* Was this callback added to the global callback list? */
4096 if (!data
->zcd_added
)
4099 ASSERT3U(data
->zcd_txg
, !=, 0);
4101 /* Remove our callback from the list */
4102 (void) mutex_lock(&zcl
.zcl_callbacks_lock
);
4103 list_remove(&zcl
.zcl_callbacks
, data
);
4104 (void) mutex_unlock(&zcl
.zcl_callbacks_lock
);
4107 umem_free(data
, sizeof (ztest_cb_data_t
));
4110 /* Allocate and initialize callback data structure */
4111 static ztest_cb_data_t
*
4112 ztest_create_cb_data(objset_t
*os
, uint64_t txg
)
4114 ztest_cb_data_t
*cb_data
;
4116 cb_data
= umem_zalloc(sizeof (ztest_cb_data_t
), UMEM_NOFAIL
);
4118 cb_data
->zcd_txg
= txg
;
4119 cb_data
->zcd_spa
= dmu_objset_spa(os
);
4125 * If a number of txgs equal to this threshold have been created after a commit
4126 * callback has been registered but not called, then we assume there is an
4127 * implementation bug.
4129 #define ZTEST_COMMIT_CALLBACK_THRESH (TXG_CONCURRENT_STATES + 2)
4132 * Commit callback test.
4135 ztest_dmu_commit_callbacks(ztest_ds_t
*zd
, uint64_t id
)
4137 objset_t
*os
= zd
->zd_os
;
4140 ztest_cb_data_t
*cb_data
[3], *tmp_cb
;
4141 uint64_t old_txg
, txg
;
4144 ztest_od_init(&od
[0], id
, FTAG
, 0, DMU_OT_UINT64_OTHER
, 0, 0);
4146 if (ztest_object_init(zd
, od
, sizeof (od
), B_FALSE
) != 0)
4149 tx
= dmu_tx_create(os
);
4151 cb_data
[0] = ztest_create_cb_data(os
, 0);
4152 dmu_tx_callback_register(tx
, ztest_commit_callback
, cb_data
[0]);
4154 dmu_tx_hold_write(tx
, od
[0].od_object
, 0, sizeof (uint64_t));
4156 /* Every once in a while, abort the transaction on purpose */
4157 if (ztest_random(100) == 0)
4161 error
= dmu_tx_assign(tx
, TXG_NOWAIT
);
4163 txg
= error
? 0 : dmu_tx_get_txg(tx
);
4165 cb_data
[0]->zcd_txg
= txg
;
4166 cb_data
[1] = ztest_create_cb_data(os
, txg
);
4167 dmu_tx_callback_register(tx
, ztest_commit_callback
, cb_data
[1]);
4171 * It's not a strict requirement to call the registered
4172 * callbacks from inside dmu_tx_abort(), but that's what
4173 * it's supposed to happen in the current implementation
4174 * so we will check for that.
4176 for (i
= 0; i
< 2; i
++) {
4177 cb_data
[i
]->zcd_expected_err
= ECANCELED
;
4178 VERIFY(!cb_data
[i
]->zcd_called
);
4183 for (i
= 0; i
< 2; i
++) {
4184 VERIFY(cb_data
[i
]->zcd_called
);
4185 umem_free(cb_data
[i
], sizeof (ztest_cb_data_t
));
4191 cb_data
[2] = ztest_create_cb_data(os
, txg
);
4192 dmu_tx_callback_register(tx
, ztest_commit_callback
, cb_data
[2]);
4195 * Read existing data to make sure there isn't a future leak.
4197 VERIFY(0 == dmu_read(os
, od
[0].od_object
, 0, sizeof (uint64_t),
4198 &old_txg
, DMU_READ_PREFETCH
));
4201 fatal(0, "future leak: got %" PRIu64
", open txg is %" PRIu64
,
4204 dmu_write(os
, od
[0].od_object
, 0, sizeof (uint64_t), &txg
, tx
);
4206 (void) mutex_lock(&zcl
.zcl_callbacks_lock
);
4209 * Since commit callbacks don't have any ordering requirement and since
4210 * it is theoretically possible for a commit callback to be called
4211 * after an arbitrary amount of time has elapsed since its txg has been
4212 * synced, it is difficult to reliably determine whether a commit
4213 * callback hasn't been called due to high load or due to a flawed
4216 * In practice, we will assume that if after a certain number of txgs a
4217 * commit callback hasn't been called, then most likely there's an
4218 * implementation bug..
4220 tmp_cb
= list_head(&zcl
.zcl_callbacks
);
4221 if (tmp_cb
!= NULL
&&
4222 tmp_cb
->zcd_txg
> txg
- ZTEST_COMMIT_CALLBACK_THRESH
) {
4223 fatal(0, "Commit callback threshold exceeded, oldest txg: %"
4224 PRIu64
", open txg: %" PRIu64
"\n", tmp_cb
->zcd_txg
, txg
);
4228 * Let's find the place to insert our callbacks.
4230 * Even though the list is ordered by txg, it is possible for the
4231 * insertion point to not be the end because our txg may already be
4232 * quiescing at this point and other callbacks in the open txg
4233 * (from other objsets) may have sneaked in.
4235 tmp_cb
= list_tail(&zcl
.zcl_callbacks
);
4236 while (tmp_cb
!= NULL
&& tmp_cb
->zcd_txg
> txg
)
4237 tmp_cb
= list_prev(&zcl
.zcl_callbacks
, tmp_cb
);
4239 /* Add the 3 callbacks to the list */
4240 for (i
= 0; i
< 3; i
++) {
4242 list_insert_head(&zcl
.zcl_callbacks
, cb_data
[i
]);
4244 list_insert_after(&zcl
.zcl_callbacks
, tmp_cb
,
4247 cb_data
[i
]->zcd_added
= B_TRUE
;
4248 VERIFY(!cb_data
[i
]->zcd_called
);
4250 tmp_cb
= cb_data
[i
];
4253 (void) mutex_unlock(&zcl
.zcl_callbacks_lock
);
4260 ztest_dsl_prop_get_set(ztest_ds_t
*zd
, uint64_t id
)
4262 zfs_prop_t proplist
[] = {
4264 ZFS_PROP_COMPRESSION
,
4268 ztest_shared_t
*zs
= ztest_shared
;
4271 (void) rw_rdlock(&zs
->zs_name_lock
);
4273 for (p
= 0; p
< sizeof (proplist
) / sizeof (proplist
[0]); p
++)
4274 (void) ztest_dsl_prop_set_uint64(zd
->zd_name
, proplist
[p
],
4275 ztest_random_dsl_prop(proplist
[p
]), (int)ztest_random(2));
4277 (void) rw_unlock(&zs
->zs_name_lock
);
4282 ztest_spa_prop_get_set(ztest_ds_t
*zd
, uint64_t id
)
4284 ztest_shared_t
*zs
= ztest_shared
;
4285 nvlist_t
*props
= NULL
;
4287 (void) rw_rdlock(&zs
->zs_name_lock
);
4289 (void) ztest_spa_prop_set_uint64(zs
, ZPOOL_PROP_DEDUPDITTO
,
4290 ZIO_DEDUPDITTO_MIN
+ ztest_random(ZIO_DEDUPDITTO_MIN
));
4292 VERIFY3U(spa_prop_get(zs
->zs_spa
, &props
), ==, 0);
4294 if (zopt_verbose
>= 6)
4295 dump_nvlist(props
, 4);
4299 (void) rw_unlock(&zs
->zs_name_lock
);
4303 * Test snapshot hold/release and deferred destroy.
4306 ztest_dmu_snapshot_hold(ztest_ds_t
*zd
, uint64_t id
)
4309 objset_t
*os
= zd
->zd_os
;
4313 char clonename
[100];
4315 char osname
[MAXNAMELEN
];
4317 (void) rw_rdlock(&ztest_shared
->zs_name_lock
);
4319 dmu_objset_name(os
, osname
);
4321 (void) snprintf(snapname
, 100, "sh1_%llu", id
);
4322 (void) snprintf(fullname
, 100, "%s@%s", osname
, snapname
);
4323 (void) snprintf(clonename
, 100, "%s/ch1_%llu", osname
, id
);
4324 (void) snprintf(tag
, 100, "%tag_%llu", id
);
4327 * Clean up from any previous run.
4329 (void) dmu_objset_destroy(clonename
, B_FALSE
);
4330 (void) dsl_dataset_user_release(osname
, snapname
, tag
, B_FALSE
);
4331 (void) dmu_objset_destroy(fullname
, B_FALSE
);
4334 * Create snapshot, clone it, mark snap for deferred destroy,
4335 * destroy clone, verify snap was also destroyed.
4337 error
= dmu_objset_snapshot(osname
, snapname
, NULL
, NULL
, FALSE
,
4340 if (error
== ENOSPC
) {
4341 ztest_record_enospc("dmu_objset_snapshot");
4344 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname
, error
);
4347 error
= dmu_objset_hold(fullname
, FTAG
, &origin
);
4349 fatal(0, "dmu_objset_hold(%s) = %d", fullname
, error
);
4351 error
= dmu_objset_clone(clonename
, dmu_objset_ds(origin
), 0);
4352 dmu_objset_rele(origin
, FTAG
);
4354 if (error
== ENOSPC
) {
4355 ztest_record_enospc("dmu_objset_clone");
4358 fatal(0, "dmu_objset_clone(%s) = %d", clonename
, error
);
4361 error
= dmu_objset_destroy(fullname
, B_TRUE
);
4363 fatal(0, "dmu_objset_destroy(%s, B_TRUE) = %d",
4367 error
= dmu_objset_destroy(clonename
, B_FALSE
);
4369 fatal(0, "dmu_objset_destroy(%s) = %d", clonename
, error
);
4371 error
= dmu_objset_hold(fullname
, FTAG
, &origin
);
4372 if (error
!= ENOENT
)
4373 fatal(0, "dmu_objset_hold(%s) = %d", fullname
, error
);
4376 * Create snapshot, add temporary hold, verify that we can't
4377 * destroy a held snapshot, mark for deferred destroy,
4378 * release hold, verify snapshot was destroyed.
4380 error
= dmu_objset_snapshot(osname
, snapname
, NULL
, NULL
, FALSE
,
4383 if (error
== ENOSPC
) {
4384 ztest_record_enospc("dmu_objset_snapshot");
4387 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname
, error
);
4390 error
= dsl_dataset_user_hold(osname
, snapname
, tag
, B_FALSE
,
4393 fatal(0, "dsl_dataset_user_hold(%s)", fullname
, tag
);
4395 error
= dmu_objset_destroy(fullname
, B_FALSE
);
4396 if (error
!= EBUSY
) {
4397 fatal(0, "dmu_objset_destroy(%s, B_FALSE) = %d",
4401 error
= dmu_objset_destroy(fullname
, B_TRUE
);
4403 fatal(0, "dmu_objset_destroy(%s, B_TRUE) = %d",
4407 error
= dsl_dataset_user_release(osname
, snapname
, tag
, B_FALSE
);
4409 fatal(0, "dsl_dataset_user_release(%s)", fullname
, tag
);
4411 VERIFY(dmu_objset_hold(fullname
, FTAG
, &origin
) == ENOENT
);
4414 (void) rw_unlock(&ztest_shared
->zs_name_lock
);
4418 * Inject random faults into the on-disk data.
4422 ztest_fault_inject(ztest_ds_t
*zd
, uint64_t id
)
4424 ztest_shared_t
*zs
= ztest_shared
;
4425 spa_t
*spa
= zs
->zs_spa
;
4429 uint64_t bad
= 0x1990c0ffeedecadeull
;
4431 char path0
[MAXPATHLEN
];
4432 char pathrand
[MAXPATHLEN
];
4434 int bshift
= SPA_MAXBLOCKSHIFT
+ 2; /* don't scrog all labels */
4440 boolean_t islog
= B_FALSE
;
4442 VERIFY(mutex_lock(&zs
->zs_vdev_lock
) == 0);
4443 maxfaults
= MAXFAULTS();
4444 leaves
= MAX(zs
->zs_mirrors
, 1) * zopt_raidz
;
4445 mirror_save
= zs
->zs_mirrors
;
4446 VERIFY(mutex_unlock(&zs
->zs_vdev_lock
) == 0);
4448 ASSERT(leaves
>= 1);
4451 * We need SCL_STATE here because we're going to look at vd0->vdev_tsd.
4453 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
4455 if (ztest_random(2) == 0) {
4457 * Inject errors on a normal data device or slog device.
4459 top
= ztest_random_vdev_top(spa
, B_TRUE
);
4460 leaf
= ztest_random(leaves
) + zs
->zs_splits
;
4463 * Generate paths to the first leaf in this top-level vdev,
4464 * and to the random leaf we selected. We'll induce transient
4465 * write failures and random online/offline activity on leaf 0,
4466 * and we'll write random garbage to the randomly chosen leaf.
4468 (void) snprintf(path0
, sizeof (path0
), ztest_dev_template
,
4469 zopt_dir
, zopt_pool
, top
* leaves
+ zs
->zs_splits
);
4470 (void) snprintf(pathrand
, sizeof (pathrand
), ztest_dev_template
,
4471 zopt_dir
, zopt_pool
, top
* leaves
+ leaf
);
4473 vd0
= vdev_lookup_by_path(spa
->spa_root_vdev
, path0
);
4474 if (vd0
!= NULL
&& vd0
->vdev_top
->vdev_islog
)
4477 if (vd0
!= NULL
&& maxfaults
!= 1) {
4479 * Make vd0 explicitly claim to be unreadable,
4480 * or unwriteable, or reach behind its back
4481 * and close the underlying fd. We can do this if
4482 * maxfaults == 0 because we'll fail and reexecute,
4483 * and we can do it if maxfaults >= 2 because we'll
4484 * have enough redundancy. If maxfaults == 1, the
4485 * combination of this with injection of random data
4486 * corruption below exceeds the pool's fault tolerance.
4488 vdev_file_t
*vf
= vd0
->vdev_tsd
;
4490 if (vf
!= NULL
&& ztest_random(3) == 0) {
4491 (void) close(vf
->vf_vnode
->v_fd
);
4492 vf
->vf_vnode
->v_fd
= -1;
4493 } else if (ztest_random(2) == 0) {
4494 vd0
->vdev_cant_read
= B_TRUE
;
4496 vd0
->vdev_cant_write
= B_TRUE
;
4498 guid0
= vd0
->vdev_guid
;
4502 * Inject errors on an l2cache device.
4504 spa_aux_vdev_t
*sav
= &spa
->spa_l2cache
;
4506 if (sav
->sav_count
== 0) {
4507 spa_config_exit(spa
, SCL_STATE
, FTAG
);
4510 vd0
= sav
->sav_vdevs
[ztest_random(sav
->sav_count
)];
4511 guid0
= vd0
->vdev_guid
;
4512 (void) strcpy(path0
, vd0
->vdev_path
);
4513 (void) strcpy(pathrand
, vd0
->vdev_path
);
4517 maxfaults
= INT_MAX
; /* no limit on cache devices */
4520 spa_config_exit(spa
, SCL_STATE
, FTAG
);
4523 * If we can tolerate two or more faults, or we're dealing
4524 * with a slog, randomly online/offline vd0.
4526 if ((maxfaults
>= 2 || islog
) && guid0
!= 0) {
4527 if (ztest_random(10) < 6) {
4528 int flags
= (ztest_random(2) == 0 ?
4529 ZFS_OFFLINE_TEMPORARY
: 0);
4532 * We have to grab the zs_name_lock as writer to
4533 * prevent a race between offlining a slog and
4534 * destroying a dataset. Offlining the slog will
4535 * grab a reference on the dataset which may cause
4536 * dmu_objset_destroy() to fail with EBUSY thus
4537 * leaving the dataset in an inconsistent state.
4540 (void) rw_wrlock(&ztest_shared
->zs_name_lock
);
4542 VERIFY(vdev_offline(spa
, guid0
, flags
) != EBUSY
);
4545 (void) rw_unlock(&ztest_shared
->zs_name_lock
);
4547 (void) vdev_online(spa
, guid0
, 0, NULL
);
4555 * We have at least single-fault tolerance, so inject data corruption.
4557 fd
= open(pathrand
, O_RDWR
);
4559 if (fd
== -1) /* we hit a gap in the device namespace */
4562 fsize
= lseek(fd
, 0, SEEK_END
);
4564 while (--iters
!= 0) {
4565 offset
= ztest_random(fsize
/ (leaves
<< bshift
)) *
4566 (leaves
<< bshift
) + (leaf
<< bshift
) +
4567 (ztest_random(1ULL << (bshift
- 1)) & -8ULL);
4569 if (offset
>= fsize
)
4572 VERIFY(mutex_lock(&zs
->zs_vdev_lock
) == 0);
4573 if (mirror_save
!= zs
->zs_mirrors
) {
4574 VERIFY(mutex_unlock(&zs
->zs_vdev_lock
) == 0);
4579 if (pwrite(fd
, &bad
, sizeof (bad
), offset
) != sizeof (bad
))
4580 fatal(1, "can't inject bad word at 0x%llx in %s",
4583 VERIFY(mutex_unlock(&zs
->zs_vdev_lock
) == 0);
4585 if (zopt_verbose
>= 7)
4586 (void) printf("injected bad word into %s,"
4587 " offset 0x%llx\n", pathrand
, (u_longlong_t
)offset
);
4594 * Verify that DDT repair works as expected.
4597 ztest_ddt_repair(ztest_ds_t
*zd
, uint64_t id
)
4599 ztest_shared_t
*zs
= ztest_shared
;
4600 spa_t
*spa
= zs
->zs_spa
;
4601 objset_t
*os
= zd
->zd_os
;
4603 uint64_t object
, blocksize
, txg
, pattern
, psize
;
4604 enum zio_checksum checksum
= spa_dedup_checksum(spa
);
4609 int copies
= 2 * ZIO_DEDUPDITTO_MIN
;
4612 blocksize
= ztest_random_blocksize();
4613 blocksize
= MIN(blocksize
, 2048); /* because we write so many */
4615 ztest_od_init(&od
[0], id
, FTAG
, 0, DMU_OT_UINT64_OTHER
, blocksize
, 0);
4617 if (ztest_object_init(zd
, od
, sizeof (od
), B_FALSE
) != 0)
4621 * Take the name lock as writer to prevent anyone else from changing
4622 * the pool and dataset properies we need to maintain during this test.
4624 (void) rw_wrlock(&zs
->zs_name_lock
);
4626 if (ztest_dsl_prop_set_uint64(zd
->zd_name
, ZFS_PROP_DEDUP
, checksum
,
4628 ztest_dsl_prop_set_uint64(zd
->zd_name
, ZFS_PROP_COPIES
, 1,
4630 (void) rw_unlock(&zs
->zs_name_lock
);
4634 object
= od
[0].od_object
;
4635 blocksize
= od
[0].od_blocksize
;
4636 pattern
= spa_guid(spa
) ^ dmu_objset_fsid_guid(os
);
4638 ASSERT(object
!= 0);
4640 tx
= dmu_tx_create(os
);
4641 dmu_tx_hold_write(tx
, object
, 0, copies
* blocksize
);
4642 txg
= ztest_tx_assign(tx
, TXG_WAIT
, FTAG
);
4644 (void) rw_unlock(&zs
->zs_name_lock
);
4649 * Write all the copies of our block.
4651 for (i
= 0; i
< copies
; i
++) {
4652 uint64_t offset
= i
* blocksize
;
4653 VERIFY(dmu_buf_hold(os
, object
, offset
, FTAG
, &db
,
4654 DMU_READ_NO_PREFETCH
) == 0);
4655 ASSERT(db
->db_offset
== offset
);
4656 ASSERT(db
->db_size
== blocksize
);
4657 ASSERT(ztest_pattern_match(db
->db_data
, db
->db_size
, pattern
) ||
4658 ztest_pattern_match(db
->db_data
, db
->db_size
, 0ULL));
4659 dmu_buf_will_fill(db
, tx
);
4660 ztest_pattern_set(db
->db_data
, db
->db_size
, pattern
);
4661 dmu_buf_rele(db
, FTAG
);
4665 txg_wait_synced(spa_get_dsl(spa
), txg
);
4668 * Find out what block we got.
4670 VERIFY(dmu_buf_hold(os
, object
, 0, FTAG
, &db
,
4671 DMU_READ_NO_PREFETCH
) == 0);
4672 blk
= *((dmu_buf_impl_t
*)db
)->db_blkptr
;
4673 dmu_buf_rele(db
, FTAG
);
4676 * Damage the block. Dedup-ditto will save us when we read it later.
4678 psize
= BP_GET_PSIZE(&blk
);
4679 buf
= zio_buf_alloc(psize
);
4680 ztest_pattern_set(buf
, psize
, ~pattern
);
4682 (void) zio_wait(zio_rewrite(NULL
, spa
, 0, &blk
,
4683 buf
, psize
, NULL
, NULL
, ZIO_PRIORITY_SYNC_WRITE
,
4684 ZIO_FLAG_CANFAIL
| ZIO_FLAG_INDUCE_DAMAGE
, NULL
));
4686 zio_buf_free(buf
, psize
);
4688 (void) rw_unlock(&zs
->zs_name_lock
);
4696 ztest_scrub(ztest_ds_t
*zd
, uint64_t id
)
4698 ztest_shared_t
*zs
= ztest_shared
;
4699 spa_t
*spa
= zs
->zs_spa
;
4701 (void) spa_scan(spa
, POOL_SCAN_SCRUB
);
4702 (void) poll(NULL
, 0, 100); /* wait a moment, then force a restart */
4703 (void) spa_scan(spa
, POOL_SCAN_SCRUB
);
4707 * Rename the pool to a different name and then rename it back.
4711 ztest_spa_rename(ztest_ds_t
*zd
, uint64_t id
)
4713 ztest_shared_t
*zs
= ztest_shared
;
4714 char *oldname
, *newname
;
4717 (void) rw_wrlock(&zs
->zs_name_lock
);
4719 oldname
= zs
->zs_pool
;
4720 newname
= umem_alloc(strlen(oldname
) + 5, UMEM_NOFAIL
);
4721 (void) strcpy(newname
, oldname
);
4722 (void) strcat(newname
, "_tmp");
4727 VERIFY3U(0, ==, spa_rename(oldname
, newname
));
4730 * Try to open it under the old name, which shouldn't exist
4732 VERIFY3U(ENOENT
, ==, spa_open(oldname
, &spa
, FTAG
));
4735 * Open it under the new name and make sure it's still the same spa_t.
4737 VERIFY3U(0, ==, spa_open(newname
, &spa
, FTAG
));
4739 ASSERT(spa
== zs
->zs_spa
);
4740 spa_close(spa
, FTAG
);
4743 * Rename it back to the original
4745 VERIFY3U(0, ==, spa_rename(newname
, oldname
));
4748 * Make sure it can still be opened
4750 VERIFY3U(0, ==, spa_open(oldname
, &spa
, FTAG
));
4752 ASSERT(spa
== zs
->zs_spa
);
4753 spa_close(spa
, FTAG
);
4755 umem_free(newname
, strlen(newname
) + 1);
4757 (void) rw_unlock(&zs
->zs_name_lock
);
4761 * Verify pool integrity by running zdb.
4764 ztest_run_zdb(char *pool
)
4767 char zdb
[MAXPATHLEN
+ MAXNAMELEN
+ 20];
4775 (void) realpath(getexecname(), zdb
);
4777 /* zdb lives in /usr/sbin, while ztest lives in /usr/bin */
4778 bin
= strstr(zdb
, "/usr/bin/");
4779 ztest
= strstr(bin
, "/ztest");
4781 isalen
= ztest
- isa
;
4785 "/usr/sbin%.*s/zdb -bcc%s%s -U %s %s",
4788 zopt_verbose
>= 3 ? "s" : "",
4789 zopt_verbose
>= 4 ? "v" : "",
4794 if (zopt_verbose
>= 5)
4795 (void) printf("Executing %s\n", strstr(zdb
, "zdb "));
4797 fp
= popen(zdb
, "r");
4799 while (fgets(zbuf
, sizeof (zbuf
), fp
) != NULL
)
4800 if (zopt_verbose
>= 3)
4801 (void) printf("%s", zbuf
);
4803 status
= pclose(fp
);
4808 ztest_dump_core
= 0;
4809 if (WIFEXITED(status
))
4810 fatal(0, "'%s' exit code %d", zdb
, WEXITSTATUS(status
));
4812 fatal(0, "'%s' died with signal %d", zdb
, WTERMSIG(status
));
4816 ztest_walk_pool_directory(char *header
)
4820 if (zopt_verbose
>= 6)
4821 (void) printf("%s\n", header
);
4823 mutex_enter(&spa_namespace_lock
);
4824 while ((spa
= spa_next(spa
)) != NULL
)
4825 if (zopt_verbose
>= 6)
4826 (void) printf("\t%s\n", spa_name(spa
));
4827 mutex_exit(&spa_namespace_lock
);
4831 ztest_spa_import_export(char *oldname
, char *newname
)
4833 nvlist_t
*config
, *newconfig
;
4837 if (zopt_verbose
>= 4) {
4838 (void) printf("import/export: old = %s, new = %s\n",
4843 * Clean up from previous runs.
4845 (void) spa_destroy(newname
);
4848 * Get the pool's configuration and guid.
4850 VERIFY3U(0, ==, spa_open(oldname
, &spa
, FTAG
));
4853 * Kick off a scrub to tickle scrub/export races.
4855 if (ztest_random(2) == 0)
4856 (void) spa_scan(spa
, POOL_SCAN_SCRUB
);
4858 pool_guid
= spa_guid(spa
);
4859 spa_close(spa
, FTAG
);
4861 ztest_walk_pool_directory("pools before export");
4866 VERIFY3U(0, ==, spa_export(oldname
, &config
, B_FALSE
, B_FALSE
));
4868 ztest_walk_pool_directory("pools after export");
4873 newconfig
= spa_tryimport(config
);
4874 ASSERT(newconfig
!= NULL
);
4875 nvlist_free(newconfig
);
4878 * Import it under the new name.
4880 VERIFY3U(0, ==, spa_import(newname
, config
, NULL
, 0));
4882 ztest_walk_pool_directory("pools after import");
4885 * Try to import it again -- should fail with EEXIST.
4887 VERIFY3U(EEXIST
, ==, spa_import(newname
, config
, NULL
, 0));
4890 * Try to import it under a different name -- should fail with EEXIST.
4892 VERIFY3U(EEXIST
, ==, spa_import(oldname
, config
, NULL
, 0));
4895 * Verify that the pool is no longer visible under the old name.
4897 VERIFY3U(ENOENT
, ==, spa_open(oldname
, &spa
, FTAG
));
4900 * Verify that we can open and close the pool using the new name.
4902 VERIFY3U(0, ==, spa_open(newname
, &spa
, FTAG
));
4903 ASSERT(pool_guid
== spa_guid(spa
));
4904 spa_close(spa
, FTAG
);
4906 nvlist_free(config
);
4910 ztest_resume(spa_t
*spa
)
4912 if (spa_suspended(spa
) && zopt_verbose
>= 6)
4913 (void) printf("resuming from suspended state\n");
4914 spa_vdev_state_enter(spa
, SCL_NONE
);
4915 vdev_clear(spa
, NULL
);
4916 (void) spa_vdev_state_exit(spa
, NULL
, 0);
4917 (void) zio_resume(spa
);
4921 ztest_resume_thread(void *arg
)
4925 while (!ztest_exiting
) {
4926 if (spa_suspended(spa
))
4928 (void) poll(NULL
, 0, 100);
4934 ztest_deadman_thread(void *arg
)
4936 ztest_shared_t
*zs
= arg
;
4940 delta
= (zs
->zs_thread_stop
- zs
->zs_thread_start
) / NANOSEC
+ grace
;
4942 (void) poll(NULL
, 0, (int)(1000 * delta
));
4944 fatal(0, "failed to complete within %d seconds of deadline", grace
);
4950 ztest_execute(ztest_info_t
*zi
, uint64_t id
)
4952 ztest_shared_t
*zs
= ztest_shared
;
4953 ztest_ds_t
*zd
= &zs
->zs_zd
[id
% zopt_datasets
];
4954 hrtime_t functime
= gethrtime();
4957 for (i
= 0; i
< zi
->zi_iters
; i
++)
4958 zi
->zi_func(zd
, id
);
4960 functime
= gethrtime() - functime
;
4962 atomic_add_64(&zi
->zi_call_count
, 1);
4963 atomic_add_64(&zi
->zi_call_time
, functime
);
4965 if (zopt_verbose
>= 4) {
4967 (void) dladdr((void *)zi
->zi_func
, &dli
);
4968 (void) printf("%6.2f sec in %s\n",
4969 (double)functime
/ NANOSEC
, dli
.dli_sname
);
4974 ztest_thread(void *arg
)
4976 uint64_t id
= (uintptr_t)arg
;
4977 ztest_shared_t
*zs
= ztest_shared
;
4982 while ((now
= gethrtime()) < zs
->zs_thread_stop
) {
4984 * See if it's time to force a crash.
4986 if (now
> zs
->zs_thread_kill
)
4990 * If we're getting ENOSPC with some regularity, stop.
4992 if (zs
->zs_enospc_count
> 10)
4996 * Pick a random function to execute.
4998 zi
= &zs
->zs_info
[ztest_random(ZTEST_FUNCS
)];
4999 call_next
= zi
->zi_call_next
;
5001 if (now
>= call_next
&&
5002 atomic_cas_64(&zi
->zi_call_next
, call_next
, call_next
+
5003 ztest_random(2 * zi
->zi_interval
[0] + 1)) == call_next
)
5004 ztest_execute(zi
, id
);
5011 ztest_dataset_name(char *dsname
, char *pool
, int d
)
5013 (void) snprintf(dsname
, MAXNAMELEN
, "%s/ds_%d", pool
, d
);
5017 ztest_dataset_destroy(ztest_shared_t
*zs
, int d
)
5019 char name
[MAXNAMELEN
];
5022 ztest_dataset_name(name
, zs
->zs_pool
, d
);
5024 if (zopt_verbose
>= 3)
5025 (void) printf("Destroying %s to free up space\n", name
);
5028 * Cleanup any non-standard clones and snapshots. In general,
5029 * ztest thread t operates on dataset (t % zopt_datasets),
5030 * so there may be more than one thing to clean up.
5032 for (t
= d
; t
< zopt_threads
; t
+= zopt_datasets
)
5033 ztest_dsl_dataset_cleanup(name
, t
);
5035 (void) dmu_objset_find(name
, ztest_objset_destroy_cb
, NULL
,
5036 DS_FIND_SNAPSHOTS
| DS_FIND_CHILDREN
);
5040 ztest_dataset_dirobj_verify(ztest_ds_t
*zd
)
5042 uint64_t usedobjs
, dirobjs
, scratch
;
5045 * ZTEST_DIROBJ is the object directory for the entire dataset.
5046 * Therefore, the number of objects in use should equal the
5047 * number of ZTEST_DIROBJ entries, +1 for ZTEST_DIROBJ itself.
5048 * If not, we have an object leak.
5050 * Note that we can only check this in ztest_dataset_open(),
5051 * when the open-context and syncing-context values agree.
5052 * That's because zap_count() returns the open-context value,
5053 * while dmu_objset_space() returns the rootbp fill count.
5055 VERIFY3U(0, ==, zap_count(zd
->zd_os
, ZTEST_DIROBJ
, &dirobjs
));
5056 dmu_objset_space(zd
->zd_os
, &scratch
, &scratch
, &usedobjs
, &scratch
);
5057 ASSERT3U(dirobjs
+ 1, ==, usedobjs
);
5061 ztest_dataset_open(ztest_shared_t
*zs
, int d
)
5063 ztest_ds_t
*zd
= &zs
->zs_zd
[d
];
5064 uint64_t committed_seq
= zd
->zd_seq
;
5067 char name
[MAXNAMELEN
];
5070 ztest_dataset_name(name
, zs
->zs_pool
, d
);
5072 (void) rw_rdlock(&zs
->zs_name_lock
);
5074 error
= ztest_dataset_create(name
);
5075 if (error
== ENOSPC
) {
5076 (void) rw_unlock(&zs
->zs_name_lock
);
5077 ztest_record_enospc(FTAG
);
5080 ASSERT(error
== 0 || error
== EEXIST
);
5082 VERIFY3U(dmu_objset_hold(name
, zd
, &os
), ==, 0);
5083 (void) rw_unlock(&zs
->zs_name_lock
);
5085 ztest_zd_init(zd
, os
);
5087 zilog
= zd
->zd_zilog
;
5089 if (zilog
->zl_header
->zh_claim_lr_seq
!= 0 &&
5090 zilog
->zl_header
->zh_claim_lr_seq
< committed_seq
)
5091 fatal(0, "missing log records: claimed %llu < committed %llu",
5092 zilog
->zl_header
->zh_claim_lr_seq
, committed_seq
);
5094 ztest_dataset_dirobj_verify(zd
);
5096 zil_replay(os
, zd
, ztest_replay_vector
);
5098 ztest_dataset_dirobj_verify(zd
);
5100 if (zopt_verbose
>= 6)
5101 (void) printf("%s replay %llu blocks, %llu records, seq %llu\n",
5103 (u_longlong_t
)zilog
->zl_parse_blk_count
,
5104 (u_longlong_t
)zilog
->zl_parse_lr_count
,
5105 (u_longlong_t
)zilog
->zl_replaying_seq
);
5107 zilog
= zil_open(os
, ztest_get_data
);
5109 if (zilog
->zl_replaying_seq
!= 0 &&
5110 zilog
->zl_replaying_seq
< committed_seq
)
5111 fatal(0, "missing log records: replayed %llu < committed %llu",
5112 zilog
->zl_replaying_seq
, committed_seq
);
5118 ztest_dataset_close(ztest_shared_t
*zs
, int d
)
5120 ztest_ds_t
*zd
= &zs
->zs_zd
[d
];
5122 zil_close(zd
->zd_zilog
);
5123 dmu_objset_rele(zd
->zd_os
, zd
);
5129 * Kick off threads to run tests on all datasets in parallel.
5132 ztest_run(ztest_shared_t
*zs
)
5136 thread_t resume_tid
;
5140 ztest_exiting
= B_FALSE
;
5143 * Initialize parent/child shared state.
5145 VERIFY(_mutex_init(&zs
->zs_vdev_lock
, USYNC_THREAD
, NULL
) == 0);
5146 VERIFY(rwlock_init(&zs
->zs_name_lock
, USYNC_THREAD
, NULL
) == 0);
5148 zs
->zs_thread_start
= gethrtime();
5149 zs
->zs_thread_stop
= zs
->zs_thread_start
+ zopt_passtime
* NANOSEC
;
5150 zs
->zs_thread_stop
= MIN(zs
->zs_thread_stop
, zs
->zs_proc_stop
);
5151 zs
->zs_thread_kill
= zs
->zs_thread_stop
;
5152 if (ztest_random(100) < zopt_killrate
)
5153 zs
->zs_thread_kill
-= ztest_random(zopt_passtime
* NANOSEC
);
5155 (void) _mutex_init(&zcl
.zcl_callbacks_lock
, USYNC_THREAD
, NULL
);
5157 list_create(&zcl
.zcl_callbacks
, sizeof (ztest_cb_data_t
),
5158 offsetof(ztest_cb_data_t
, zcd_node
));
5163 kernel_init(FREAD
| FWRITE
);
5164 VERIFY(spa_open(zs
->zs_pool
, &spa
, FTAG
) == 0);
5167 spa
->spa_dedup_ditto
= 2 * ZIO_DEDUPDITTO_MIN
;
5170 * We don't expect the pool to suspend unless maxfaults == 0,
5171 * in which case ztest_fault_inject() temporarily takes away
5172 * the only valid replica.
5174 if (MAXFAULTS() == 0)
5175 spa
->spa_failmode
= ZIO_FAILURE_MODE_WAIT
;
5177 spa
->spa_failmode
= ZIO_FAILURE_MODE_PANIC
;
5180 * Create a thread to periodically resume suspended I/O.
5182 VERIFY(thr_create(0, 0, ztest_resume_thread
, spa
, THR_BOUND
,
5186 * Create a deadman thread to abort() if we hang.
5188 VERIFY(thr_create(0, 0, ztest_deadman_thread
, zs
, THR_BOUND
,
5192 * Verify that we can safely inquire about about any object,
5193 * whether it's allocated or not. To make it interesting,
5194 * we probe a 5-wide window around each power of two.
5195 * This hits all edge cases, including zero and the max.
5197 for (t
= 0; t
< 64; t
++) {
5198 for (d
= -5; d
<= 5; d
++) {
5199 error
= dmu_object_info(spa
->spa_meta_objset
,
5200 (1ULL << t
) + d
, NULL
);
5201 ASSERT(error
== 0 || error
== ENOENT
||
5207 * If we got any ENOSPC errors on the previous run, destroy something.
5209 if (zs
->zs_enospc_count
!= 0) {
5210 int d
= ztest_random(zopt_datasets
);
5211 ztest_dataset_destroy(zs
, d
);
5213 zs
->zs_enospc_count
= 0;
5215 tid
= umem_zalloc(zopt_threads
* sizeof (thread_t
), UMEM_NOFAIL
);
5217 if (zopt_verbose
>= 4)
5218 (void) printf("starting main threads...\n");
5221 * Kick off all the tests that run in parallel.
5223 for (t
= 0; t
< zopt_threads
; t
++) {
5224 if (t
< zopt_datasets
&& ztest_dataset_open(zs
, t
) != 0)
5226 VERIFY(thr_create(0, 0, ztest_thread
, (void *)(uintptr_t)t
,
5227 THR_BOUND
, &tid
[t
]) == 0);
5231 * Wait for all of the tests to complete. We go in reverse order
5232 * so we don't close datasets while threads are still using them.
5234 for (t
= zopt_threads
- 1; t
>= 0; t
--) {
5235 VERIFY(thr_join(tid
[t
], NULL
, NULL
) == 0);
5236 if (t
< zopt_datasets
)
5237 ztest_dataset_close(zs
, t
);
5240 txg_wait_synced(spa_get_dsl(spa
), 0);
5242 zs
->zs_alloc
= metaslab_class_get_alloc(spa_normal_class(spa
));
5243 zs
->zs_space
= metaslab_class_get_space(spa_normal_class(spa
));
5245 umem_free(tid
, zopt_threads
* sizeof (thread_t
));
5247 /* Kill the resume thread */
5248 ztest_exiting
= B_TRUE
;
5249 VERIFY(thr_join(resume_tid
, NULL
, NULL
) == 0);
5253 * Right before closing the pool, kick off a bunch of async I/O;
5254 * spa_close() should wait for it to complete.
5256 for (uint64_t object
= 1; object
< 50; object
++)
5257 dmu_prefetch(spa
->spa_meta_objset
, object
, 0, 1ULL << 20);
5259 spa_close(spa
, FTAG
);
5262 * Verify that we can loop over all pools.
5264 mutex_enter(&spa_namespace_lock
);
5265 for (spa
= spa_next(NULL
); spa
!= NULL
; spa
= spa_next(spa
))
5266 if (zopt_verbose
> 3)
5267 (void) printf("spa_next: found %s\n", spa_name(spa
));
5268 mutex_exit(&spa_namespace_lock
);
5271 * Verify that we can export the pool and reimport it under a
5274 if (ztest_random(2) == 0) {
5275 char name
[MAXNAMELEN
];
5276 (void) snprintf(name
, MAXNAMELEN
, "%s_import", zs
->zs_pool
);
5277 ztest_spa_import_export(zs
->zs_pool
, name
);
5278 ztest_spa_import_export(name
, zs
->zs_pool
);
5283 list_destroy(&zcl
.zcl_callbacks
);
5285 (void) _mutex_destroy(&zcl
.zcl_callbacks_lock
);
5287 (void) rwlock_destroy(&zs
->zs_name_lock
);
5288 (void) _mutex_destroy(&zs
->zs_vdev_lock
);
5292 ztest_freeze(ztest_shared_t
*zs
)
5294 ztest_ds_t
*zd
= &zs
->zs_zd
[0];
5298 if (zopt_verbose
>= 3)
5299 (void) printf("testing spa_freeze()...\n");
5301 kernel_init(FREAD
| FWRITE
);
5302 VERIFY3U(0, ==, spa_open(zs
->zs_pool
, &spa
, FTAG
));
5303 VERIFY3U(0, ==, ztest_dataset_open(zs
, 0));
5306 * Force the first log block to be transactionally allocated.
5307 * We have to do this before we freeze the pool -- otherwise
5308 * the log chain won't be anchored.
5310 while (BP_IS_HOLE(&zd
->zd_zilog
->zl_header
->zh_log
)) {
5311 ztest_dmu_object_alloc_free(zd
, 0);
5312 zil_commit(zd
->zd_zilog
, 0);
5315 txg_wait_synced(spa_get_dsl(spa
), 0);
5318 * Freeze the pool. This stops spa_sync() from doing anything,
5319 * so that the only way to record changes from now on is the ZIL.
5324 * Run tests that generate log records but don't alter the pool config
5325 * or depend on DSL sync tasks (snapshots, objset create/destroy, etc).
5326 * We do a txg_wait_synced() after each iteration to force the txg
5327 * to increase well beyond the last synced value in the uberblock.
5328 * The ZIL should be OK with that.
5330 while (ztest_random(10) != 0 && numloops
++ < zopt_maxloops
) {
5331 ztest_dmu_write_parallel(zd
, 0);
5332 ztest_dmu_object_alloc_free(zd
, 0);
5333 txg_wait_synced(spa_get_dsl(spa
), 0);
5337 * Commit all of the changes we just generated.
5339 zil_commit(zd
->zd_zilog
, 0);
5340 txg_wait_synced(spa_get_dsl(spa
), 0);
5343 * Close our dataset and close the pool.
5345 ztest_dataset_close(zs
, 0);
5346 spa_close(spa
, FTAG
);
5350 * Open and close the pool and dataset to induce log replay.
5352 kernel_init(FREAD
| FWRITE
);
5353 VERIFY3U(0, ==, spa_open(zs
->zs_pool
, &spa
, FTAG
));
5354 VERIFY3U(0, ==, ztest_dataset_open(zs
, 0));
5355 ztest_dataset_close(zs
, 0);
5356 spa_close(spa
, FTAG
);
5361 print_time(hrtime_t t
, char *timebuf
)
5363 hrtime_t s
= t
/ NANOSEC
;
5364 hrtime_t m
= s
/ 60;
5365 hrtime_t h
= m
/ 60;
5366 hrtime_t d
= h
/ 24;
5375 (void) sprintf(timebuf
,
5376 "%llud%02lluh%02llum%02llus", d
, h
, m
, s
);
5378 (void) sprintf(timebuf
, "%lluh%02llum%02llus", h
, m
, s
);
5380 (void) sprintf(timebuf
, "%llum%02llus", m
, s
);
5382 (void) sprintf(timebuf
, "%llus", s
);
5386 make_random_props(void)
5390 if (ztest_random(2) == 0)
5393 VERIFY(nvlist_alloc(&props
, NV_UNIQUE_NAME
, 0) == 0);
5394 VERIFY(nvlist_add_uint64(props
, "autoreplace", 1) == 0);
5396 (void) printf("props:\n");
5397 dump_nvlist(props
, 4);
5403 * Create a storage pool with the given name and initial vdev size.
5404 * Then test spa_freeze() functionality.
5407 ztest_init(ztest_shared_t
*zs
)
5410 nvlist_t
*nvroot
, *props
;
5412 VERIFY(_mutex_init(&zs
->zs_vdev_lock
, USYNC_THREAD
, NULL
) == 0);
5413 VERIFY(rwlock_init(&zs
->zs_name_lock
, USYNC_THREAD
, NULL
) == 0);
5415 kernel_init(FREAD
| FWRITE
);
5418 * Create the storage pool.
5420 (void) spa_destroy(zs
->zs_pool
);
5421 ztest_shared
->zs_vdev_next_leaf
= 0;
5423 zs
->zs_mirrors
= zopt_mirrors
;
5424 nvroot
= make_vdev_root(NULL
, NULL
, zopt_vdev_size
, 0,
5425 0, zopt_raidz
, zs
->zs_mirrors
, 1);
5426 props
= make_random_props();
5427 VERIFY3U(0, ==, spa_create(zs
->zs_pool
, nvroot
, props
, NULL
, NULL
));
5428 nvlist_free(nvroot
);
5430 VERIFY3U(0, ==, spa_open(zs
->zs_pool
, &spa
, FTAG
));
5431 metaslab_sz
= 1ULL << spa
->spa_root_vdev
->vdev_child
[0]->vdev_ms_shift
;
5432 spa_close(spa
, FTAG
);
5436 ztest_run_zdb(zs
->zs_pool
);
5440 ztest_run_zdb(zs
->zs_pool
);
5442 (void) rwlock_destroy(&zs
->zs_name_lock
);
5443 (void) _mutex_destroy(&zs
->zs_vdev_lock
);
5447 main(int argc
, char **argv
)
5459 (void) setvbuf(stdout
, NULL
, _IOLBF
, 0);
5461 ztest_random_fd
= open("/dev/urandom", O_RDONLY
);
5463 process_options(argc
, argv
);
5465 /* Override location of zpool.cache */
5466 VERIFY(asprintf((char **)&spa_config_path
, "%s/zpool.cache",
5470 * Blow away any existing copy of zpool.cache
5473 (void) remove(spa_config_path
);
5475 shared_size
= sizeof (*zs
) + zopt_datasets
* sizeof (ztest_ds_t
);
5477 zs
= ztest_shared
= (void *)mmap(0,
5478 P2ROUNDUP(shared_size
, getpagesize()),
5479 PROT_READ
| PROT_WRITE
, MAP_SHARED
| MAP_ANON
, -1, 0);
5481 if (zopt_verbose
>= 1) {
5482 (void) printf("%llu vdevs, %d datasets, %d threads,"
5483 " %llu seconds...\n",
5484 (u_longlong_t
)zopt_vdevs
, zopt_datasets
, zopt_threads
,
5485 (u_longlong_t
)zopt_time
);
5489 * Create and initialize our storage pool.
5491 for (i
= 1; i
<= zopt_init
; i
++) {
5492 bzero(zs
, sizeof (ztest_shared_t
));
5493 if (zopt_verbose
>= 3 && zopt_init
!= 1)
5494 (void) printf("ztest_init(), pass %d\n", i
);
5495 zs
->zs_pool
= zopt_pool
;
5499 zs
->zs_pool
= zopt_pool
;
5500 zs
->zs_proc_start
= gethrtime();
5501 zs
->zs_proc_stop
= zs
->zs_proc_start
+ zopt_time
* NANOSEC
;
5503 for (f
= 0; f
< ZTEST_FUNCS
; f
++) {
5504 zi
= &zs
->zs_info
[f
];
5505 *zi
= ztest_info
[f
];
5506 if (zs
->zs_proc_start
+ zi
->zi_interval
[0] > zs
->zs_proc_stop
)
5507 zi
->zi_call_next
= UINT64_MAX
;
5509 zi
->zi_call_next
= zs
->zs_proc_start
+
5510 ztest_random(2 * zi
->zi_interval
[0] + 1);
5514 * Run the tests in a loop. These tests include fault injection
5515 * to verify that self-healing data works, and forced crashes
5516 * to verify that we never lose on-disk consistency.
5518 while (gethrtime() < zs
->zs_proc_stop
) {
5523 * Initialize the workload counters for each function.
5525 for (f
= 0; f
< ZTEST_FUNCS
; f
++) {
5526 zi
= &zs
->zs_info
[f
];
5527 zi
->zi_call_count
= 0;
5528 zi
->zi_call_time
= 0;
5531 /* Set the allocation switch size */
5532 metaslab_df_alloc_threshold
= ztest_random(metaslab_sz
/ 4) + 1;
5537 fatal(1, "fork failed");
5539 if (pid
== 0) { /* child */
5540 struct rlimit rl
= { 1024, 1024 };
5541 (void) setrlimit(RLIMIT_NOFILE
, &rl
);
5542 (void) enable_extended_FILE_stdio(-1, -1);
5547 while (waitpid(pid
, &status
, 0) != pid
)
5550 if (WIFEXITED(status
)) {
5551 if (WEXITSTATUS(status
) != 0) {
5552 (void) fprintf(stderr
,
5553 "child exited with code %d\n",
5554 WEXITSTATUS(status
));
5557 } else if (WIFSIGNALED(status
)) {
5558 if (WTERMSIG(status
) != SIGKILL
) {
5559 (void) fprintf(stderr
,
5560 "child died with signal %d\n",
5566 (void) fprintf(stderr
, "something strange happened "
5573 if (zopt_verbose
>= 1) {
5574 hrtime_t now
= gethrtime();
5576 now
= MIN(now
, zs
->zs_proc_stop
);
5577 print_time(zs
->zs_proc_stop
- now
, timebuf
);
5578 nicenum(zs
->zs_space
, numbuf
);
5580 (void) printf("Pass %3d, %8s, %3llu ENOSPC, "
5581 "%4.1f%% of %5s used, %3.0f%% done, %8s to go\n",
5583 WIFEXITED(status
) ? "Complete" : "SIGKILL",
5584 (u_longlong_t
)zs
->zs_enospc_count
,
5585 100.0 * zs
->zs_alloc
/ zs
->zs_space
,
5587 100.0 * (now
- zs
->zs_proc_start
) /
5588 (zopt_time
* NANOSEC
), timebuf
);
5591 if (zopt_verbose
>= 2) {
5592 (void) printf("\nWorkload summary:\n\n");
5593 (void) printf("%7s %9s %s\n",
5594 "Calls", "Time", "Function");
5595 (void) printf("%7s %9s %s\n",
5596 "-----", "----", "--------");
5597 for (f
= 0; f
< ZTEST_FUNCS
; f
++) {
5600 zi
= &zs
->zs_info
[f
];
5601 print_time(zi
->zi_call_time
, timebuf
);
5602 (void) dladdr((void *)zi
->zi_func
, &dli
);
5603 (void) printf("%7llu %9s %s\n",
5604 (u_longlong_t
)zi
->zi_call_count
, timebuf
,
5607 (void) printf("\n");
5611 * It's possible that we killed a child during a rename test,
5612 * in which case we'll have a 'ztest_tmp' pool lying around
5613 * instead of 'ztest'. Do a blind rename in case this happened.
5616 if (spa_open(zopt_pool
, &spa
, FTAG
) == 0) {
5617 spa_close(spa
, FTAG
);
5619 char tmpname
[MAXNAMELEN
];
5621 kernel_init(FREAD
| FWRITE
);
5622 (void) snprintf(tmpname
, sizeof (tmpname
), "%s_tmp",
5624 (void) spa_rename(tmpname
, zopt_pool
);
5628 ztest_run_zdb(zopt_pool
);
5631 if (zopt_verbose
>= 1) {
5632 (void) printf("%d killed, %d completed, %.0f%% kill rate\n",
5633 kills
, iters
- kills
, (100.0 * kills
) / MAX(1, iters
));