4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2011, 2018 by Delphix. All rights reserved.
24 * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
25 * Copyright (c) 2013 Steven Hartland. All rights reserved.
26 * Copyright (c) 2014 Integros [integros.com]
27 * Copyright 2017 Joyent, Inc.
28 * Copyright (c) 2017, Intel Corporation.
32 * The objective of this program is to provide a DMU/ZAP/SPA stress test
33 * that runs entirely in userland, is easy to use, and easy to extend.
35 * The overall design of the ztest program is as follows:
37 * (1) For each major functional area (e.g. adding vdevs to a pool,
38 * creating and destroying datasets, reading and writing objects, etc)
39 * we have a simple routine to test that functionality. These
40 * individual routines do not have to do anything "stressful".
42 * (2) We turn these simple functionality tests into a stress test by
43 * running them all in parallel, with as many threads as desired,
44 * and spread across as many datasets, objects, and vdevs as desired.
46 * (3) While all this is happening, we inject faults into the pool to
47 * verify that self-healing data really works.
49 * (4) Every time we open a dataset, we change its checksum and compression
50 * functions. Thus even individual objects vary from block to block
51 * in which checksum they use and whether they're compressed.
53 * (5) To verify that we never lose on-disk consistency after a crash,
54 * we run the entire test in a child of the main process.
55 * At random times, the child self-immolates with a SIGKILL.
56 * This is the software equivalent of pulling the power cord.
57 * The parent then runs the test again, using the existing
58 * storage pool, as many times as desired. If backwards compatibility
59 * testing is enabled ztest will sometimes run the "older" version
60 * of ztest after a SIGKILL.
62 * (6) To verify that we don't have future leaks or temporal incursions,
63 * many of the functional tests record the transaction group number
64 * as part of their data. When reading old data, they verify that
65 * the transaction group number is less than the current, open txg.
66 * If you add a new test, please do this if applicable.
68 * (7) Threads are created with a reduced stack size, for sanity checking.
69 * Therefore, it's important not to allocate huge buffers on the stack.
71 * When run with no arguments, ztest runs for about five minutes and
72 * produces no output if successful. To get a little bit of information,
73 * specify -V. To get more information, specify -VV, and so on.
75 * To turn this into an overnight stress test, use -T to specify run time.
77 * You can ask more vdevs [-v], datasets [-d], or threads [-t]
78 * to increase the pool capacity, fanout, and overall stress level.
80 * Use the -k option to set the desired frequency of kills.
82 * When ztest invokes itself it passes all relevant information through a
83 * temporary file which is mmap-ed in the child process. This allows shared
84 * memory to survive the exec syscall. The ztest_shared_hdr_t struct is always
85 * stored at offset 0 of this file and contains information on the size and
86 * number of shared structures in the file. The information stored in this file
87 * must remain backwards compatible with older versions of ztest so that
88 * ztest can invoke them during backwards compatibility testing (-B).
91 #include <sys/zfs_context.h>
97 #include <sys/dmu_objset.h>
100 #include <sys/time.h>
101 #include <sys/wait.h>
102 #include <sys/mman.h>
103 #include <sys/resource.h>
106 #include <sys/zil_impl.h>
107 #include <sys/vdev_draid.h>
108 #include <sys/vdev_impl.h>
109 #include <sys/vdev_file.h>
110 #include <sys/vdev_initialize.h>
111 #include <sys/vdev_raidz.h>
112 #include <sys/vdev_trim.h>
113 #include <sys/spa_impl.h>
114 #include <sys/metaslab_impl.h>
115 #include <sys/dsl_prop.h>
116 #include <sys/dsl_dataset.h>
117 #include <sys/dsl_destroy.h>
118 #include <sys/dsl_scan.h>
119 #include <sys/zio_checksum.h>
120 #include <sys/zfs_refcount.h>
121 #include <sys/zfeature.h>
122 #include <sys/dsl_userhold.h>
132 #include <sys/fs/zfs.h>
133 #include <zfs_fletcher.h>
134 #include <libnvpair.h>
135 #include <libzutil.h>
136 #include <sys/crypto/icp.h>
137 #if (__GLIBC__ && !__UCLIBC__)
138 #include <execinfo.h> /* for backtrace() */
141 static int ztest_fd_data
= -1;
142 static int ztest_fd_rand
= -1;
144 typedef struct ztest_shared_hdr
{
145 uint64_t zh_hdr_size
;
146 uint64_t zh_opts_size
;
148 uint64_t zh_stats_size
;
149 uint64_t zh_stats_count
;
151 uint64_t zh_ds_count
;
152 } ztest_shared_hdr_t
;
154 static ztest_shared_hdr_t
*ztest_shared_hdr
;
156 enum ztest_class_state
{
157 ZTEST_VDEV_CLASS_OFF
,
162 #define ZO_GVARS_MAX_ARGLEN ((size_t)64)
163 #define ZO_GVARS_MAX_COUNT ((size_t)10)
165 typedef struct ztest_shared_opts
{
166 char zo_pool
[ZFS_MAX_DATASET_NAME_LEN
];
167 char zo_dir
[ZFS_MAX_DATASET_NAME_LEN
];
168 char zo_alt_ztest
[MAXNAMELEN
];
169 char zo_alt_libpath
[MAXNAMELEN
];
171 uint64_t zo_vdevtime
;
175 int zo_raid_children
;
177 char zo_raid_type
[8];
182 uint64_t zo_passtime
;
183 uint64_t zo_killrate
;
187 uint64_t zo_maxloops
;
188 uint64_t zo_metaslab_force_ganging
;
190 int zo_special_vdevs
;
193 char zo_gvars
[ZO_GVARS_MAX_COUNT
][ZO_GVARS_MAX_ARGLEN
];
194 } ztest_shared_opts_t
;
196 /* Default values for command line options. */
197 #define DEFAULT_POOL "ztest"
198 #define DEFAULT_VDEV_DIR "/tmp"
199 #define DEFAULT_VDEV_COUNT 5
200 #define DEFAULT_VDEV_SIZE (SPA_MINDEVSIZE * 4) /* 256m default size */
201 #define DEFAULT_VDEV_SIZE_STR "256M"
202 #define DEFAULT_ASHIFT SPA_MINBLOCKSHIFT
203 #define DEFAULT_MIRRORS 2
204 #define DEFAULT_RAID_CHILDREN 4
205 #define DEFAULT_RAID_PARITY 1
206 #define DEFAULT_DRAID_DATA 4
207 #define DEFAULT_DRAID_SPARES 1
208 #define DEFAULT_DATASETS_COUNT 7
209 #define DEFAULT_THREADS 23
210 #define DEFAULT_RUN_TIME 300 /* 300 seconds */
211 #define DEFAULT_RUN_TIME_STR "300 sec"
212 #define DEFAULT_PASS_TIME 60 /* 60 seconds */
213 #define DEFAULT_PASS_TIME_STR "60 sec"
214 #define DEFAULT_KILL_RATE 70 /* 70% kill rate */
215 #define DEFAULT_KILLRATE_STR "70%"
216 #define DEFAULT_INITS 1
217 #define DEFAULT_MAX_LOOPS 50 /* 5 minutes */
218 #define DEFAULT_FORCE_GANGING (64 << 10)
219 #define DEFAULT_FORCE_GANGING_STR "64K"
221 /* Simplifying assumption: -1 is not a valid default. */
222 #define NO_DEFAULT -1
224 static const ztest_shared_opts_t ztest_opts_defaults
= {
225 .zo_pool
= DEFAULT_POOL
,
226 .zo_dir
= DEFAULT_VDEV_DIR
,
227 .zo_alt_ztest
= { '\0' },
228 .zo_alt_libpath
= { '\0' },
229 .zo_vdevs
= DEFAULT_VDEV_COUNT
,
230 .zo_ashift
= DEFAULT_ASHIFT
,
231 .zo_mirrors
= DEFAULT_MIRRORS
,
232 .zo_raid_children
= DEFAULT_RAID_CHILDREN
,
233 .zo_raid_parity
= DEFAULT_RAID_PARITY
,
234 .zo_raid_type
= VDEV_TYPE_RAIDZ
,
235 .zo_vdev_size
= DEFAULT_VDEV_SIZE
,
236 .zo_draid_data
= DEFAULT_DRAID_DATA
, /* data drives */
237 .zo_draid_spares
= DEFAULT_DRAID_SPARES
, /* distributed spares */
238 .zo_datasets
= DEFAULT_DATASETS_COUNT
,
239 .zo_threads
= DEFAULT_THREADS
,
240 .zo_passtime
= DEFAULT_PASS_TIME
,
241 .zo_killrate
= DEFAULT_KILL_RATE
,
244 .zo_init
= DEFAULT_INITS
,
245 .zo_time
= DEFAULT_RUN_TIME
,
246 .zo_maxloops
= DEFAULT_MAX_LOOPS
, /* max loops during spa_freeze() */
247 .zo_metaslab_force_ganging
= DEFAULT_FORCE_GANGING
,
248 .zo_special_vdevs
= ZTEST_VDEV_CLASS_RND
,
252 extern uint64_t metaslab_force_ganging
;
253 extern uint64_t metaslab_df_alloc_threshold
;
254 extern unsigned long zfs_deadman_synctime_ms
;
255 extern int metaslab_preload_limit
;
256 extern int zfs_compressed_arc_enabled
;
257 extern int zfs_abd_scatter_enabled
;
258 extern int dmu_object_alloc_chunk_shift
;
259 extern boolean_t zfs_force_some_double_word_sm_entries
;
260 extern unsigned long zio_decompress_fail_fraction
;
261 extern unsigned long zfs_reconstruct_indirect_damage_fraction
;
264 static ztest_shared_opts_t
*ztest_shared_opts
;
265 static ztest_shared_opts_t ztest_opts
;
266 static char *ztest_wkeydata
= "abcdefghijklmnopqrstuvwxyz012345";
268 typedef struct ztest_shared_ds
{
272 static ztest_shared_ds_t
*ztest_shared_ds
;
273 #define ZTEST_GET_SHARED_DS(d) (&ztest_shared_ds[d])
275 #define BT_MAGIC 0x123456789abcdefULL
276 #define MAXFAULTS(zs) \
277 (MAX((zs)->zs_mirrors, 1) * (ztest_opts.zo_raid_parity + 1) - 1)
281 ZTEST_IO_WRITE_PATTERN
,
282 ZTEST_IO_WRITE_ZEROES
,
289 typedef struct ztest_block_tag
{
293 uint64_t bt_dnodesize
;
300 typedef struct bufwad
{
307 * It would be better to use a rangelock_t per object. Unfortunately
308 * the rangelock_t is not a drop-in replacement for rl_t, because we
309 * still need to map from object ID to rangelock_t.
331 #define ZTEST_RANGE_LOCKS 64
332 #define ZTEST_OBJECT_LOCKS 64
335 * Object descriptor. Used as a template for object lookup/create/remove.
337 typedef struct ztest_od
{
340 dmu_object_type_t od_type
;
341 dmu_object_type_t od_crtype
;
342 uint64_t od_blocksize
;
343 uint64_t od_crblocksize
;
344 uint64_t od_crdnodesize
;
347 char od_name
[ZFS_MAX_DATASET_NAME_LEN
];
353 typedef struct ztest_ds
{
354 ztest_shared_ds_t
*zd_shared
;
356 pthread_rwlock_t zd_zilog_lock
;
358 ztest_od_t
*zd_od
; /* debugging aid */
359 char zd_name
[ZFS_MAX_DATASET_NAME_LEN
];
360 kmutex_t zd_dirobj_lock
;
361 rll_t zd_object_lock
[ZTEST_OBJECT_LOCKS
];
362 rll_t zd_range_lock
[ZTEST_RANGE_LOCKS
];
366 * Per-iteration state.
368 typedef void ztest_func_t(ztest_ds_t
*zd
, uint64_t id
);
370 typedef struct ztest_info
{
371 ztest_func_t
*zi_func
; /* test function */
372 uint64_t zi_iters
; /* iterations per execution */
373 uint64_t *zi_interval
; /* execute every <interval> seconds */
374 const char *zi_funcname
; /* name of test function */
377 typedef struct ztest_shared_callstate
{
378 uint64_t zc_count
; /* per-pass count */
379 uint64_t zc_time
; /* per-pass time */
380 uint64_t zc_next
; /* next time to call this function */
381 } ztest_shared_callstate_t
;
383 static ztest_shared_callstate_t
*ztest_shared_callstate
;
384 #define ZTEST_GET_SHARED_CALLSTATE(c) (&ztest_shared_callstate[c])
386 ztest_func_t ztest_dmu_read_write
;
387 ztest_func_t ztest_dmu_write_parallel
;
388 ztest_func_t ztest_dmu_object_alloc_free
;
389 ztest_func_t ztest_dmu_object_next_chunk
;
390 ztest_func_t ztest_dmu_commit_callbacks
;
391 ztest_func_t ztest_zap
;
392 ztest_func_t ztest_zap_parallel
;
393 ztest_func_t ztest_zil_commit
;
394 ztest_func_t ztest_zil_remount
;
395 ztest_func_t ztest_dmu_read_write_zcopy
;
396 ztest_func_t ztest_dmu_objset_create_destroy
;
397 ztest_func_t ztest_dmu_prealloc
;
398 ztest_func_t ztest_fzap
;
399 ztest_func_t ztest_dmu_snapshot_create_destroy
;
400 ztest_func_t ztest_dsl_prop_get_set
;
401 ztest_func_t ztest_spa_prop_get_set
;
402 ztest_func_t ztest_spa_create_destroy
;
403 ztest_func_t ztest_fault_inject
;
404 ztest_func_t ztest_dmu_snapshot_hold
;
405 ztest_func_t ztest_mmp_enable_disable
;
406 ztest_func_t ztest_scrub
;
407 ztest_func_t ztest_dsl_dataset_promote_busy
;
408 ztest_func_t ztest_vdev_attach_detach
;
409 ztest_func_t ztest_vdev_LUN_growth
;
410 ztest_func_t ztest_vdev_add_remove
;
411 ztest_func_t ztest_vdev_class_add
;
412 ztest_func_t ztest_vdev_aux_add_remove
;
413 ztest_func_t ztest_split_pool
;
414 ztest_func_t ztest_reguid
;
415 ztest_func_t ztest_spa_upgrade
;
416 ztest_func_t ztest_device_removal
;
417 ztest_func_t ztest_spa_checkpoint_create_discard
;
418 ztest_func_t ztest_initialize
;
419 ztest_func_t ztest_trim
;
420 ztest_func_t ztest_fletcher
;
421 ztest_func_t ztest_fletcher_incr
;
422 ztest_func_t ztest_verify_dnode_bt
;
424 uint64_t zopt_always
= 0ULL * NANOSEC
; /* all the time */
425 uint64_t zopt_incessant
= 1ULL * NANOSEC
/ 10; /* every 1/10 second */
426 uint64_t zopt_often
= 1ULL * NANOSEC
; /* every second */
427 uint64_t zopt_sometimes
= 10ULL * NANOSEC
; /* every 10 seconds */
428 uint64_t zopt_rarely
= 60ULL * NANOSEC
; /* every 60 seconds */
430 #define ZTI_INIT(func, iters, interval) \
431 { .zi_func = (func), \
432 .zi_iters = (iters), \
433 .zi_interval = (interval), \
434 .zi_funcname = # func }
436 ztest_info_t ztest_info
[] = {
437 ZTI_INIT(ztest_dmu_read_write
, 1, &zopt_always
),
438 ZTI_INIT(ztest_dmu_write_parallel
, 10, &zopt_always
),
439 ZTI_INIT(ztest_dmu_object_alloc_free
, 1, &zopt_always
),
440 ZTI_INIT(ztest_dmu_object_next_chunk
, 1, &zopt_sometimes
),
441 ZTI_INIT(ztest_dmu_commit_callbacks
, 1, &zopt_always
),
442 ZTI_INIT(ztest_zap
, 30, &zopt_always
),
443 ZTI_INIT(ztest_zap_parallel
, 100, &zopt_always
),
444 ZTI_INIT(ztest_split_pool
, 1, &zopt_always
),
445 ZTI_INIT(ztest_zil_commit
, 1, &zopt_incessant
),
446 ZTI_INIT(ztest_zil_remount
, 1, &zopt_sometimes
),
447 ZTI_INIT(ztest_dmu_read_write_zcopy
, 1, &zopt_often
),
448 ZTI_INIT(ztest_dmu_objset_create_destroy
, 1, &zopt_often
),
449 ZTI_INIT(ztest_dsl_prop_get_set
, 1, &zopt_often
),
450 ZTI_INIT(ztest_spa_prop_get_set
, 1, &zopt_sometimes
),
452 ZTI_INIT(ztest_dmu_prealloc
, 1, &zopt_sometimes
),
454 ZTI_INIT(ztest_fzap
, 1, &zopt_sometimes
),
455 ZTI_INIT(ztest_dmu_snapshot_create_destroy
, 1, &zopt_sometimes
),
456 ZTI_INIT(ztest_spa_create_destroy
, 1, &zopt_sometimes
),
457 ZTI_INIT(ztest_fault_inject
, 1, &zopt_sometimes
),
458 ZTI_INIT(ztest_dmu_snapshot_hold
, 1, &zopt_sometimes
),
459 ZTI_INIT(ztest_mmp_enable_disable
, 1, &zopt_sometimes
),
460 ZTI_INIT(ztest_reguid
, 1, &zopt_rarely
),
461 ZTI_INIT(ztest_scrub
, 1, &zopt_rarely
),
462 ZTI_INIT(ztest_spa_upgrade
, 1, &zopt_rarely
),
463 ZTI_INIT(ztest_dsl_dataset_promote_busy
, 1, &zopt_rarely
),
464 ZTI_INIT(ztest_vdev_attach_detach
, 1, &zopt_sometimes
),
465 ZTI_INIT(ztest_vdev_LUN_growth
, 1, &zopt_rarely
),
466 ZTI_INIT(ztest_vdev_add_remove
, 1, &ztest_opts
.zo_vdevtime
),
467 ZTI_INIT(ztest_vdev_class_add
, 1, &ztest_opts
.zo_vdevtime
),
468 ZTI_INIT(ztest_vdev_aux_add_remove
, 1, &ztest_opts
.zo_vdevtime
),
469 ZTI_INIT(ztest_device_removal
, 1, &zopt_sometimes
),
470 ZTI_INIT(ztest_spa_checkpoint_create_discard
, 1, &zopt_rarely
),
471 ZTI_INIT(ztest_initialize
, 1, &zopt_sometimes
),
472 ZTI_INIT(ztest_trim
, 1, &zopt_sometimes
),
473 ZTI_INIT(ztest_fletcher
, 1, &zopt_rarely
),
474 ZTI_INIT(ztest_fletcher_incr
, 1, &zopt_rarely
),
475 ZTI_INIT(ztest_verify_dnode_bt
, 1, &zopt_sometimes
),
478 #define ZTEST_FUNCS (sizeof (ztest_info) / sizeof (ztest_info_t))
481 * The following struct is used to hold a list of uncalled commit callbacks.
482 * The callbacks are ordered by txg number.
484 typedef struct ztest_cb_list
{
485 kmutex_t zcl_callbacks_lock
;
486 list_t zcl_callbacks
;
490 * Stuff we need to share writably between parent and child.
492 typedef struct ztest_shared
{
493 boolean_t zs_do_init
;
494 hrtime_t zs_proc_start
;
495 hrtime_t zs_proc_stop
;
496 hrtime_t zs_thread_start
;
497 hrtime_t zs_thread_stop
;
498 hrtime_t zs_thread_kill
;
499 uint64_t zs_enospc_count
;
500 uint64_t zs_vdev_next_leaf
;
501 uint64_t zs_vdev_aux
;
506 uint64_t zs_metaslab_sz
;
507 uint64_t zs_metaslab_df_alloc_threshold
;
511 #define ID_PARALLEL -1ULL
513 static char ztest_dev_template
[] = "%s/%s.%llua";
514 static char ztest_aux_template
[] = "%s/%s.%s.%llu";
515 ztest_shared_t
*ztest_shared
;
517 static spa_t
*ztest_spa
= NULL
;
518 static ztest_ds_t
*ztest_ds
;
520 static kmutex_t ztest_vdev_lock
;
521 static boolean_t ztest_device_removal_active
= B_FALSE
;
522 static boolean_t ztest_pool_scrubbed
= B_FALSE
;
523 static kmutex_t ztest_checkpoint_lock
;
526 * The ztest_name_lock protects the pool and dataset namespace used by
527 * the individual tests. To modify the namespace, consumers must grab
528 * this lock as writer. Grabbing the lock as reader will ensure that the
529 * namespace does not change while the lock is held.
531 static pthread_rwlock_t ztest_name_lock
;
533 static boolean_t ztest_dump_core
= B_TRUE
;
534 static boolean_t ztest_exiting
;
536 /* Global commit callback list */
537 static ztest_cb_list_t zcl
;
538 /* Commit cb delay */
539 static uint64_t zc_min_txg_delay
= UINT64_MAX
;
540 static int zc_cb_counter
= 0;
543 * Minimum number of commit callbacks that need to be registered for us to check
544 * whether the minimum txg delay is acceptable.
546 #define ZTEST_COMMIT_CB_MIN_REG 100
549 * If a number of txgs equal to this threshold have been created after a commit
550 * callback has been registered but not called, then we assume there is an
551 * implementation bug.
553 #define ZTEST_COMMIT_CB_THRESH (TXG_CONCURRENT_STATES + 1000)
556 ZTEST_META_DNODE
= 0,
561 static _Noreturn
void usage(boolean_t
);
562 static int ztest_scrub_impl(spa_t
*spa
);
565 * These libumem hooks provide a reasonable set of defaults for the allocator's
566 * debugging facilities.
569 _umem_debug_init(void)
571 return ("default,verbose"); /* $UMEM_DEBUG setting */
575 _umem_logging_init(void)
577 return ("fail,contents"); /* $UMEM_LOGGING setting */
581 dump_debug_buffer(void)
583 ssize_t ret
__attribute__((unused
));
585 if (!ztest_opts
.zo_dump_dbgmsg
)
589 * We use write() instead of printf() so that this function
590 * is safe to call from a signal handler.
592 ret
= write(STDOUT_FILENO
, "\n", 1);
593 zfs_dbgmsg_print("ztest");
596 #define BACKTRACE_SZ 100
598 static void sig_handler(int signo
)
600 struct sigaction action
;
601 #if (__GLIBC__ && !__UCLIBC__) /* backtrace() is a GNU extension */
603 void *buffer
[BACKTRACE_SZ
];
605 nptrs
= backtrace(buffer
, BACKTRACE_SZ
);
606 backtrace_symbols_fd(buffer
, nptrs
, STDERR_FILENO
);
611 * Restore default action and re-raise signal so SIGSEGV and
612 * SIGABRT can trigger a core dump.
614 action
.sa_handler
= SIG_DFL
;
615 sigemptyset(&action
.sa_mask
);
617 (void) sigaction(signo
, &action
, NULL
);
621 #define FATAL_MSG_SZ 1024
625 static __attribute__((format(printf
, 2, 3))) _Noreturn
void
626 fatal(int do_perror
, char *message
, ...)
629 int save_errno
= errno
;
632 (void) fflush(stdout
);
633 buf
= umem_alloc(FATAL_MSG_SZ
, UMEM_NOFAIL
);
637 va_start(args
, message
);
638 (void) sprintf(buf
, "ztest: ");
640 (void) vsprintf(buf
+ strlen(buf
), message
, args
);
643 (void) snprintf(buf
+ strlen(buf
), FATAL_MSG_SZ
- strlen(buf
),
644 ": %s", strerror(save_errno
));
646 (void) fprintf(stderr
, "%s\n", buf
);
647 fatal_msg
= buf
; /* to ease debugging */
659 str2shift(const char *buf
)
661 const char *ends
= "BKMGTPEZ";
666 for (i
= 0; i
< strlen(ends
); i
++) {
667 if (toupper(buf
[0]) == ends
[i
])
670 if (i
== strlen(ends
)) {
671 (void) fprintf(stderr
, "ztest: invalid bytes suffix: %s\n",
675 if (buf
[1] == '\0' || (toupper(buf
[1]) == 'B' && buf
[2] == '\0')) {
678 (void) fprintf(stderr
, "ztest: invalid bytes suffix: %s\n", buf
);
683 nicenumtoull(const char *buf
)
688 val
= strtoull(buf
, &end
, 0);
690 (void) fprintf(stderr
, "ztest: bad numeric value: %s\n", buf
);
692 } else if (end
[0] == '.') {
693 double fval
= strtod(buf
, &end
);
694 fval
*= pow(2, str2shift(end
));
696 * UINT64_MAX is not exactly representable as a double.
697 * The closest representation is UINT64_MAX + 1, so we
698 * use a >= comparison instead of > for the bounds check.
700 if (fval
>= (double)UINT64_MAX
) {
701 (void) fprintf(stderr
, "ztest: value too large: %s\n",
705 val
= (uint64_t)fval
;
707 int shift
= str2shift(end
);
708 if (shift
>= 64 || (val
<< shift
) >> shift
!= val
) {
709 (void) fprintf(stderr
, "ztest: value too large: %s\n",
718 typedef struct ztest_option
{
719 const char short_opt
;
720 const char *long_opt
;
721 const char *long_opt_param
;
723 unsigned int default_int
;
728 * The following option_table is used for generating the usage info as well as
729 * the long and short option information for calling getopt_long().
731 static ztest_option_t option_table
[] = {
732 { 'v', "vdevs", "INTEGER", "Number of vdevs", DEFAULT_VDEV_COUNT
,
734 { 's', "vdev-size", "INTEGER", "Size of each vdev",
735 NO_DEFAULT
, DEFAULT_VDEV_SIZE_STR
},
736 { 'a', "alignment-shift", "INTEGER",
737 "Alignment shift; use 0 for random", DEFAULT_ASHIFT
, NULL
},
738 { 'm', "mirror-copies", "INTEGER", "Number of mirror copies",
739 DEFAULT_MIRRORS
, NULL
},
740 { 'r', "raid-disks", "INTEGER", "Number of raidz/draid disks",
741 DEFAULT_RAID_CHILDREN
, NULL
},
742 { 'R', "raid-parity", "INTEGER", "Raid parity",
743 DEFAULT_RAID_PARITY
, NULL
},
744 { 'K', "raid-kind", "raidz|draid|random", "Raid kind",
745 NO_DEFAULT
, "random"},
746 { 'D', "draid-data", "INTEGER", "Number of draid data drives",
747 DEFAULT_DRAID_DATA
, NULL
},
748 { 'S', "draid-spares", "INTEGER", "Number of draid spares",
749 DEFAULT_DRAID_SPARES
, NULL
},
750 { 'd', "datasets", "INTEGER", "Number of datasets",
751 DEFAULT_DATASETS_COUNT
, NULL
},
752 { 't', "threads", "INTEGER", "Number of ztest threads",
753 DEFAULT_THREADS
, NULL
},
754 { 'g', "gang-block-threshold", "INTEGER",
755 "Metaslab gang block threshold",
756 NO_DEFAULT
, DEFAULT_FORCE_GANGING_STR
},
757 { 'i', "init-count", "INTEGER", "Number of times to initialize pool",
758 DEFAULT_INITS
, NULL
},
759 { 'k', "kill-percentage", "INTEGER", "Kill percentage",
760 NO_DEFAULT
, DEFAULT_KILLRATE_STR
},
761 { 'p', "pool-name", "STRING", "Pool name",
762 NO_DEFAULT
, DEFAULT_POOL
},
763 { 'f', "vdev-file-directory", "PATH", "File directory for vdev files",
764 NO_DEFAULT
, DEFAULT_VDEV_DIR
},
765 { 'M', "multi-host", NULL
,
766 "Multi-host; simulate pool imported on remote host",
768 { 'E', "use-existing-pool", NULL
,
769 "Use existing pool instead of creating new one", NO_DEFAULT
, NULL
},
770 { 'T', "run-time", "INTEGER", "Total run time",
771 NO_DEFAULT
, DEFAULT_RUN_TIME_STR
},
772 { 'P', "pass-time", "INTEGER", "Time per pass",
773 NO_DEFAULT
, DEFAULT_PASS_TIME_STR
},
774 { 'F', "freeze-loops", "INTEGER", "Max loops in spa_freeze()",
775 DEFAULT_MAX_LOOPS
, NULL
},
776 { 'B', "alt-ztest", "PATH", "Alternate ztest path",
778 { 'C', "vdev-class-state", "on|off|random", "vdev class state",
779 NO_DEFAULT
, "random"},
780 { 'o', "option", "\"OPTION=INTEGER\"",
781 "Set global variable to an unsigned 32-bit integer value",
783 { 'G', "dump-debug-msg", NULL
,
784 "Dump zfs_dbgmsg buffer before exiting due to an error",
786 { 'V', "verbose", NULL
,
787 "Verbose (use multiple times for ever more verbosity)",
789 { 'h', "help", NULL
, "Show this help",
794 static struct option
*long_opts
= NULL
;
795 static char *short_opts
= NULL
;
800 ASSERT3P(long_opts
, ==, NULL
);
801 ASSERT3P(short_opts
, ==, NULL
);
803 int count
= sizeof (option_table
) / sizeof (option_table
[0]);
804 long_opts
= umem_alloc(sizeof (struct option
) * count
, UMEM_NOFAIL
);
806 short_opts
= umem_alloc(sizeof (char) * 2 * count
, UMEM_NOFAIL
);
807 int short_opt_index
= 0;
809 for (int i
= 0; i
< count
; i
++) {
810 long_opts
[i
].val
= option_table
[i
].short_opt
;
811 long_opts
[i
].name
= option_table
[i
].long_opt
;
812 long_opts
[i
].has_arg
= option_table
[i
].long_opt_param
!= NULL
813 ? required_argument
: no_argument
;
814 long_opts
[i
].flag
= NULL
;
815 short_opts
[short_opt_index
++] = option_table
[i
].short_opt
;
816 if (option_table
[i
].long_opt_param
!= NULL
) {
817 short_opts
[short_opt_index
++] = ':';
825 int count
= sizeof (option_table
) / sizeof (option_table
[0]);
827 umem_free(long_opts
, sizeof (struct option
) * count
);
828 umem_free(short_opts
, sizeof (char) * 2 * count
);
835 usage(boolean_t requested
)
838 FILE *fp
= requested
? stdout
: stderr
;
840 (void) fprintf(fp
, "Usage: %s [OPTIONS...]\n", DEFAULT_POOL
);
841 for (int i
= 0; option_table
[i
].short_opt
!= 0; i
++) {
842 if (option_table
[i
].long_opt_param
!= NULL
) {
843 (void) sprintf(option
, " -%c --%s=%s",
844 option_table
[i
].short_opt
,
845 option_table
[i
].long_opt
,
846 option_table
[i
].long_opt_param
);
848 (void) sprintf(option
, " -%c --%s",
849 option_table
[i
].short_opt
,
850 option_table
[i
].long_opt
);
852 (void) fprintf(fp
, " %-40s%s", option
,
853 option_table
[i
].comment
);
855 if (option_table
[i
].long_opt_param
!= NULL
) {
856 if (option_table
[i
].default_str
!= NULL
) {
857 (void) fprintf(fp
, " (default: %s)",
858 option_table
[i
].default_str
);
859 } else if (option_table
[i
].default_int
!= NO_DEFAULT
) {
860 (void) fprintf(fp
, " (default: %u)",
861 option_table
[i
].default_int
);
864 (void) fprintf(fp
, "\n");
866 exit(requested
? 0 : 1);
870 ztest_random(uint64_t range
)
874 ASSERT3S(ztest_fd_rand
, >=, 0);
879 if (read(ztest_fd_rand
, &r
, sizeof (r
)) != sizeof (r
))
880 fatal(B_TRUE
, "short read from /dev/urandom");
886 ztest_parse_name_value(const char *input
, ztest_shared_opts_t
*zo
)
890 int state
= ZTEST_VDEV_CLASS_RND
;
892 (void) strlcpy(name
, input
, sizeof (name
));
894 value
= strchr(name
, '=');
896 (void) fprintf(stderr
, "missing value in property=value "
897 "'-C' argument (%s)\n", input
);
903 if (strcmp(value
, "on") == 0) {
904 state
= ZTEST_VDEV_CLASS_ON
;
905 } else if (strcmp(value
, "off") == 0) {
906 state
= ZTEST_VDEV_CLASS_OFF
;
907 } else if (strcmp(value
, "random") == 0) {
908 state
= ZTEST_VDEV_CLASS_RND
;
910 (void) fprintf(stderr
, "invalid property value '%s'\n", value
);
914 if (strcmp(name
, "special") == 0) {
915 zo
->zo_special_vdevs
= state
;
917 (void) fprintf(stderr
, "invalid property name '%s'\n", name
);
920 if (zo
->zo_verbose
>= 3)
921 (void) printf("%s vdev state is '%s'\n", name
, value
);
925 process_options(int argc
, char **argv
)
928 ztest_shared_opts_t
*zo
= &ztest_opts
;
932 char altdir
[MAXNAMELEN
] = { 0 };
933 char raid_kind
[8] = "random";
935 memcpy(zo
, &ztest_opts_defaults
, sizeof (*zo
));
939 while ((opt
= getopt_long(argc
, argv
, short_opts
, long_opts
,
959 value
= nicenumtoull(optarg
);
963 zo
->zo_vdevs
= value
;
966 zo
->zo_vdev_size
= MAX(SPA_MINDEVSIZE
, value
);
969 zo
->zo_ashift
= value
;
972 zo
->zo_mirrors
= value
;
975 zo
->zo_raid_children
= MAX(1, value
);
978 zo
->zo_raid_parity
= MIN(MAX(value
, 1), 3);
981 (void) strlcpy(raid_kind
, optarg
, sizeof (raid_kind
));
984 zo
->zo_draid_data
= MAX(1, value
);
987 zo
->zo_draid_spares
= MAX(1, value
);
990 zo
->zo_datasets
= MAX(1, value
);
993 zo
->zo_threads
= MAX(1, value
);
996 zo
->zo_metaslab_force_ganging
=
997 MAX(SPA_MINBLOCKSIZE
<< 1, value
);
1000 zo
->zo_init
= value
;
1003 zo
->zo_killrate
= value
;
1006 (void) strlcpy(zo
->zo_pool
, optarg
,
1007 sizeof (zo
->zo_pool
));
1010 path
= realpath(optarg
, NULL
);
1012 (void) fprintf(stderr
, "error: %s: %s\n",
1013 optarg
, strerror(errno
));
1016 (void) strlcpy(zo
->zo_dir
, path
,
1017 sizeof (zo
->zo_dir
));
1022 zo
->zo_mmp_test
= 1;
1031 zo
->zo_time
= value
;
1034 zo
->zo_passtime
= MAX(1, value
);
1037 zo
->zo_maxloops
= MAX(1, value
);
1040 (void) strlcpy(altdir
, optarg
, sizeof (altdir
));
1043 ztest_parse_name_value(optarg
, zo
);
1046 if (zo
->zo_gvars_count
>= ZO_GVARS_MAX_COUNT
) {
1047 (void) fprintf(stderr
,
1048 "max global var count (%zu) exceeded\n",
1049 ZO_GVARS_MAX_COUNT
);
1052 char *v
= zo
->zo_gvars
[zo
->zo_gvars_count
];
1053 if (strlcpy(v
, optarg
, ZO_GVARS_MAX_ARGLEN
) >=
1054 ZO_GVARS_MAX_ARGLEN
) {
1055 (void) fprintf(stderr
,
1056 "global var option '%s' is too long\n",
1060 zo
->zo_gvars_count
++;
1063 zo
->zo_dump_dbgmsg
= 1;
1077 /* When raid choice is 'random' add a draid pool 50% of the time */
1078 if (strcmp(raid_kind
, "random") == 0) {
1079 (void) strlcpy(raid_kind
, (ztest_random(2) == 0) ?
1080 "draid" : "raidz", sizeof (raid_kind
));
1082 if (ztest_opts
.zo_verbose
>= 3)
1083 (void) printf("choosing RAID type '%s'\n", raid_kind
);
1086 if (strcmp(raid_kind
, "draid") == 0) {
1087 uint64_t min_devsize
;
1089 /* With fewer disk use 256M, otherwise 128M is OK */
1090 min_devsize
= (ztest_opts
.zo_raid_children
< 16) ?
1091 (256ULL << 20) : (128ULL << 20);
1093 /* No top-level mirrors with dRAID for now */
1096 /* Use more appropriate defaults for dRAID */
1097 if (zo
->zo_vdevs
== ztest_opts_defaults
.zo_vdevs
)
1099 if (zo
->zo_raid_children
==
1100 ztest_opts_defaults
.zo_raid_children
)
1101 zo
->zo_raid_children
= 16;
1102 if (zo
->zo_ashift
< 12)
1104 if (zo
->zo_vdev_size
< min_devsize
)
1105 zo
->zo_vdev_size
= min_devsize
;
1107 if (zo
->zo_draid_data
+ zo
->zo_raid_parity
>
1108 zo
->zo_raid_children
- zo
->zo_draid_spares
) {
1109 (void) fprintf(stderr
, "error: too few draid "
1110 "children (%d) for stripe width (%d)\n",
1111 zo
->zo_raid_children
,
1112 zo
->zo_draid_data
+ zo
->zo_raid_parity
);
1116 (void) strlcpy(zo
->zo_raid_type
, VDEV_TYPE_DRAID
,
1117 sizeof (zo
->zo_raid_type
));
1119 } else /* using raidz */ {
1120 ASSERT0(strcmp(raid_kind
, "raidz"));
1122 zo
->zo_raid_parity
= MIN(zo
->zo_raid_parity
,
1123 zo
->zo_raid_children
- 1);
1127 (zo
->zo_vdevs
> 0 ? zo
->zo_time
* NANOSEC
/ zo
->zo_vdevs
:
1130 if (strlen(altdir
) > 0) {
1138 cmd
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
1139 realaltdir
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
1141 VERIFY3P(NULL
, !=, realpath(getexecname(), cmd
));
1142 if (0 != access(altdir
, F_OK
)) {
1143 ztest_dump_core
= B_FALSE
;
1144 fatal(B_TRUE
, "invalid alternate ztest path: %s",
1147 VERIFY3P(NULL
, !=, realpath(altdir
, realaltdir
));
1150 * 'cmd' should be of the form "<anything>/usr/bin/<isa>/ztest".
1151 * We want to extract <isa> to determine if we should use
1152 * 32 or 64 bit binaries.
1154 bin
= strstr(cmd
, "/usr/bin/");
1155 ztest
= strstr(bin
, "/ztest");
1157 isalen
= ztest
- isa
;
1158 (void) snprintf(zo
->zo_alt_ztest
, sizeof (zo
->zo_alt_ztest
),
1159 "%s/usr/bin/%.*s/ztest", realaltdir
, isalen
, isa
);
1160 (void) snprintf(zo
->zo_alt_libpath
, sizeof (zo
->zo_alt_libpath
),
1161 "%s/usr/lib/%.*s", realaltdir
, isalen
, isa
);
1163 if (0 != access(zo
->zo_alt_ztest
, X_OK
)) {
1164 ztest_dump_core
= B_FALSE
;
1165 fatal(B_TRUE
, "invalid alternate ztest: %s",
1167 } else if (0 != access(zo
->zo_alt_libpath
, X_OK
)) {
1168 ztest_dump_core
= B_FALSE
;
1169 fatal(B_TRUE
, "invalid alternate lib directory %s",
1170 zo
->zo_alt_libpath
);
1173 umem_free(cmd
, MAXPATHLEN
);
1174 umem_free(realaltdir
, MAXPATHLEN
);
1179 ztest_kill(ztest_shared_t
*zs
)
1181 zs
->zs_alloc
= metaslab_class_get_alloc(spa_normal_class(ztest_spa
));
1182 zs
->zs_space
= metaslab_class_get_space(spa_normal_class(ztest_spa
));
1185 * Before we kill off ztest, make sure that the config is updated.
1186 * See comment above spa_write_cachefile().
1188 mutex_enter(&spa_namespace_lock
);
1189 spa_write_cachefile(ztest_spa
, B_FALSE
, B_FALSE
);
1190 mutex_exit(&spa_namespace_lock
);
1192 (void) kill(getpid(), SIGKILL
);
1196 ztest_record_enospc(const char *s
)
1199 ztest_shared
->zs_enospc_count
++;
1203 ztest_get_ashift(void)
1205 if (ztest_opts
.zo_ashift
== 0)
1206 return (SPA_MINBLOCKSHIFT
+ ztest_random(5));
1207 return (ztest_opts
.zo_ashift
);
1211 ztest_is_draid_spare(const char *name
)
1213 uint64_t spare_id
= 0, parity
= 0, vdev_id
= 0;
1215 if (sscanf(name
, VDEV_TYPE_DRAID
"%"PRIu64
"-%"PRIu64
"-%"PRIu64
"",
1216 &parity
, &vdev_id
, &spare_id
) == 3) {
1224 make_vdev_file(char *path
, char *aux
, char *pool
, size_t size
, uint64_t ashift
)
1229 boolean_t draid_spare
= B_FALSE
;
1231 pathbuf
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
1234 ashift
= ztest_get_ashift();
1240 vdev
= ztest_shared
->zs_vdev_aux
;
1241 (void) snprintf(path
, MAXPATHLEN
,
1242 ztest_aux_template
, ztest_opts
.zo_dir
,
1243 pool
== NULL
? ztest_opts
.zo_pool
: pool
,
1246 vdev
= ztest_shared
->zs_vdev_next_leaf
++;
1247 (void) snprintf(path
, MAXPATHLEN
,
1248 ztest_dev_template
, ztest_opts
.zo_dir
,
1249 pool
== NULL
? ztest_opts
.zo_pool
: pool
, vdev
);
1252 draid_spare
= ztest_is_draid_spare(path
);
1255 if (size
!= 0 && !draid_spare
) {
1256 int fd
= open(path
, O_RDWR
| O_CREAT
| O_TRUNC
, 0666);
1258 fatal(B_TRUE
, "can't open %s", path
);
1259 if (ftruncate(fd
, size
) != 0)
1260 fatal(B_TRUE
, "can't ftruncate %s", path
);
1264 file
= fnvlist_alloc();
1265 fnvlist_add_string(file
, ZPOOL_CONFIG_TYPE
,
1266 draid_spare
? VDEV_TYPE_DRAID_SPARE
: VDEV_TYPE_FILE
);
1267 fnvlist_add_string(file
, ZPOOL_CONFIG_PATH
, path
);
1268 fnvlist_add_uint64(file
, ZPOOL_CONFIG_ASHIFT
, ashift
);
1269 umem_free(pathbuf
, MAXPATHLEN
);
1275 make_vdev_raid(char *path
, char *aux
, char *pool
, size_t size
,
1276 uint64_t ashift
, int r
)
1278 nvlist_t
*raid
, **child
;
1282 return (make_vdev_file(path
, aux
, pool
, size
, ashift
));
1283 child
= umem_alloc(r
* sizeof (nvlist_t
*), UMEM_NOFAIL
);
1285 for (c
= 0; c
< r
; c
++)
1286 child
[c
] = make_vdev_file(path
, aux
, pool
, size
, ashift
);
1288 raid
= fnvlist_alloc();
1289 fnvlist_add_string(raid
, ZPOOL_CONFIG_TYPE
,
1290 ztest_opts
.zo_raid_type
);
1291 fnvlist_add_uint64(raid
, ZPOOL_CONFIG_NPARITY
,
1292 ztest_opts
.zo_raid_parity
);
1293 fnvlist_add_nvlist_array(raid
, ZPOOL_CONFIG_CHILDREN
,
1294 (const nvlist_t
**)child
, r
);
1296 if (strcmp(ztest_opts
.zo_raid_type
, VDEV_TYPE_DRAID
) == 0) {
1297 uint64_t ndata
= ztest_opts
.zo_draid_data
;
1298 uint64_t nparity
= ztest_opts
.zo_raid_parity
;
1299 uint64_t nspares
= ztest_opts
.zo_draid_spares
;
1300 uint64_t children
= ztest_opts
.zo_raid_children
;
1301 uint64_t ngroups
= 1;
1304 * Calculate the minimum number of groups required to fill a
1305 * slice. This is the LCM of the stripe width (data + parity)
1306 * and the number of data drives (children - spares).
1308 while (ngroups
* (ndata
+ nparity
) % (children
- nspares
) != 0)
1311 /* Store the basic dRAID configuration. */
1312 fnvlist_add_uint64(raid
, ZPOOL_CONFIG_DRAID_NDATA
, ndata
);
1313 fnvlist_add_uint64(raid
, ZPOOL_CONFIG_DRAID_NSPARES
, nspares
);
1314 fnvlist_add_uint64(raid
, ZPOOL_CONFIG_DRAID_NGROUPS
, ngroups
);
1317 for (c
= 0; c
< r
; c
++)
1318 fnvlist_free(child
[c
]);
1320 umem_free(child
, r
* sizeof (nvlist_t
*));
1326 make_vdev_mirror(char *path
, char *aux
, char *pool
, size_t size
,
1327 uint64_t ashift
, int r
, int m
)
1329 nvlist_t
*mirror
, **child
;
1333 return (make_vdev_raid(path
, aux
, pool
, size
, ashift
, r
));
1335 child
= umem_alloc(m
* sizeof (nvlist_t
*), UMEM_NOFAIL
);
1337 for (c
= 0; c
< m
; c
++)
1338 child
[c
] = make_vdev_raid(path
, aux
, pool
, size
, ashift
, r
);
1340 mirror
= fnvlist_alloc();
1341 fnvlist_add_string(mirror
, ZPOOL_CONFIG_TYPE
, VDEV_TYPE_MIRROR
);
1342 fnvlist_add_nvlist_array(mirror
, ZPOOL_CONFIG_CHILDREN
,
1343 (const nvlist_t
**)child
, m
);
1345 for (c
= 0; c
< m
; c
++)
1346 fnvlist_free(child
[c
]);
1348 umem_free(child
, m
* sizeof (nvlist_t
*));
1354 make_vdev_root(char *path
, char *aux
, char *pool
, size_t size
, uint64_t ashift
,
1355 const char *class, int r
, int m
, int t
)
1357 nvlist_t
*root
, **child
;
1363 log
= (class != NULL
&& strcmp(class, "log") == 0);
1365 child
= umem_alloc(t
* sizeof (nvlist_t
*), UMEM_NOFAIL
);
1367 for (c
= 0; c
< t
; c
++) {
1368 child
[c
] = make_vdev_mirror(path
, aux
, pool
, size
, ashift
,
1370 fnvlist_add_uint64(child
[c
], ZPOOL_CONFIG_IS_LOG
, log
);
1372 if (class != NULL
&& class[0] != '\0') {
1373 ASSERT(m
> 1 || log
); /* expecting a mirror */
1374 fnvlist_add_string(child
[c
],
1375 ZPOOL_CONFIG_ALLOCATION_BIAS
, class);
1379 root
= fnvlist_alloc();
1380 fnvlist_add_string(root
, ZPOOL_CONFIG_TYPE
, VDEV_TYPE_ROOT
);
1381 fnvlist_add_nvlist_array(root
, aux
? aux
: ZPOOL_CONFIG_CHILDREN
,
1382 (const nvlist_t
**)child
, t
);
1384 for (c
= 0; c
< t
; c
++)
1385 fnvlist_free(child
[c
]);
1387 umem_free(child
, t
* sizeof (nvlist_t
*));
1393 * Find a random spa version. Returns back a random spa version in the
1394 * range [initial_version, SPA_VERSION_FEATURES].
1397 ztest_random_spa_version(uint64_t initial_version
)
1399 uint64_t version
= initial_version
;
1401 if (version
<= SPA_VERSION_BEFORE_FEATURES
) {
1403 ztest_random(SPA_VERSION_BEFORE_FEATURES
- version
+ 1);
1406 if (version
> SPA_VERSION_BEFORE_FEATURES
)
1407 version
= SPA_VERSION_FEATURES
;
1409 ASSERT(SPA_VERSION_IS_SUPPORTED(version
));
1414 ztest_random_blocksize(void)
1416 ASSERT3U(ztest_spa
->spa_max_ashift
, !=, 0);
1419 * Choose a block size >= the ashift.
1420 * If the SPA supports new MAXBLOCKSIZE, test up to 1MB blocks.
1422 int maxbs
= SPA_OLD_MAXBLOCKSHIFT
;
1423 if (spa_maxblocksize(ztest_spa
) == SPA_MAXBLOCKSIZE
)
1425 uint64_t block_shift
=
1426 ztest_random(maxbs
- ztest_spa
->spa_max_ashift
+ 1);
1427 return (1 << (SPA_MINBLOCKSHIFT
+ block_shift
));
1431 ztest_random_dnodesize(void)
1434 int max_slots
= spa_maxdnodesize(ztest_spa
) >> DNODE_SHIFT
;
1436 if (max_slots
== DNODE_MIN_SLOTS
)
1437 return (DNODE_MIN_SIZE
);
1440 * Weight the random distribution more heavily toward smaller
1441 * dnode sizes since that is more likely to reflect real-world
1444 ASSERT3U(max_slots
, >, 4);
1445 switch (ztest_random(10)) {
1447 slots
= 5 + ztest_random(max_slots
- 4);
1450 slots
= 2 + ztest_random(3);
1457 return (slots
<< DNODE_SHIFT
);
1461 ztest_random_ibshift(void)
1463 return (DN_MIN_INDBLKSHIFT
+
1464 ztest_random(DN_MAX_INDBLKSHIFT
- DN_MIN_INDBLKSHIFT
+ 1));
1468 ztest_random_vdev_top(spa_t
*spa
, boolean_t log_ok
)
1471 vdev_t
*rvd
= spa
->spa_root_vdev
;
1474 ASSERT3U(spa_config_held(spa
, SCL_ALL
, RW_READER
), !=, 0);
1477 top
= ztest_random(rvd
->vdev_children
);
1478 tvd
= rvd
->vdev_child
[top
];
1479 } while (!vdev_is_concrete(tvd
) || (tvd
->vdev_islog
&& !log_ok
) ||
1480 tvd
->vdev_mg
== NULL
|| tvd
->vdev_mg
->mg_class
== NULL
);
1486 ztest_random_dsl_prop(zfs_prop_t prop
)
1491 value
= zfs_prop_random_value(prop
, ztest_random(-1ULL));
1492 } while (prop
== ZFS_PROP_CHECKSUM
&& value
== ZIO_CHECKSUM_OFF
);
1498 ztest_dsl_prop_set_uint64(char *osname
, zfs_prop_t prop
, uint64_t value
,
1501 const char *propname
= zfs_prop_to_name(prop
);
1502 const char *valname
;
1507 error
= dsl_prop_set_int(osname
, propname
,
1508 (inherit
? ZPROP_SRC_NONE
: ZPROP_SRC_LOCAL
), value
);
1510 if (error
== ENOSPC
) {
1511 ztest_record_enospc(FTAG
);
1516 setpoint
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
1517 VERIFY0(dsl_prop_get_integer(osname
, propname
, &curval
, setpoint
));
1519 if (ztest_opts
.zo_verbose
>= 6) {
1522 err
= zfs_prop_index_to_string(prop
, curval
, &valname
);
1524 (void) printf("%s %s = %llu at '%s'\n", osname
,
1525 propname
, (unsigned long long)curval
, setpoint
);
1527 (void) printf("%s %s = %s at '%s'\n",
1528 osname
, propname
, valname
, setpoint
);
1530 umem_free(setpoint
, MAXPATHLEN
);
1536 ztest_spa_prop_set_uint64(zpool_prop_t prop
, uint64_t value
)
1538 spa_t
*spa
= ztest_spa
;
1539 nvlist_t
*props
= NULL
;
1542 props
= fnvlist_alloc();
1543 fnvlist_add_uint64(props
, zpool_prop_to_name(prop
), value
);
1545 error
= spa_prop_set(spa
, props
);
1547 fnvlist_free(props
);
1549 if (error
== ENOSPC
) {
1550 ztest_record_enospc(FTAG
);
1559 ztest_dmu_objset_own(const char *name
, dmu_objset_type_t type
,
1560 boolean_t readonly
, boolean_t decrypt
, void *tag
, objset_t
**osp
)
1564 char ddname
[ZFS_MAX_DATASET_NAME_LEN
];
1566 strcpy(ddname
, name
);
1567 cp
= strchr(ddname
, '@');
1571 err
= dmu_objset_own(name
, type
, readonly
, decrypt
, tag
, osp
);
1572 while (decrypt
&& err
== EACCES
) {
1573 dsl_crypto_params_t
*dcp
;
1574 nvlist_t
*crypto_args
= fnvlist_alloc();
1576 fnvlist_add_uint8_array(crypto_args
, "wkeydata",
1577 (uint8_t *)ztest_wkeydata
, WRAPPING_KEY_LEN
);
1578 VERIFY0(dsl_crypto_params_create_nvlist(DCP_CMD_NONE
, NULL
,
1579 crypto_args
, &dcp
));
1580 err
= spa_keystore_load_wkey(ddname
, dcp
, B_FALSE
);
1582 * Note: if there was an error loading, the wkey was not
1583 * consumed, and needs to be freed.
1585 dsl_crypto_params_free(dcp
, (err
!= 0));
1586 fnvlist_free(crypto_args
);
1588 if (err
== EINVAL
) {
1590 * We couldn't load a key for this dataset so try
1591 * the parent. This loop will eventually hit the
1592 * encryption root since ztest only makes clones
1593 * as children of their origin datasets.
1595 cp
= strrchr(ddname
, '/');
1602 } else if (err
!= 0) {
1606 err
= dmu_objset_own(name
, type
, readonly
, decrypt
, tag
, osp
);
1614 ztest_rll_init(rll_t
*rll
)
1616 rll
->rll_writer
= NULL
;
1617 rll
->rll_readers
= 0;
1618 mutex_init(&rll
->rll_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
1619 cv_init(&rll
->rll_cv
, NULL
, CV_DEFAULT
, NULL
);
1623 ztest_rll_destroy(rll_t
*rll
)
1625 ASSERT3P(rll
->rll_writer
, ==, NULL
);
1626 ASSERT0(rll
->rll_readers
);
1627 mutex_destroy(&rll
->rll_lock
);
1628 cv_destroy(&rll
->rll_cv
);
1632 ztest_rll_lock(rll_t
*rll
, rl_type_t type
)
1634 mutex_enter(&rll
->rll_lock
);
1636 if (type
== RL_READER
) {
1637 while (rll
->rll_writer
!= NULL
)
1638 (void) cv_wait(&rll
->rll_cv
, &rll
->rll_lock
);
1641 while (rll
->rll_writer
!= NULL
|| rll
->rll_readers
)
1642 (void) cv_wait(&rll
->rll_cv
, &rll
->rll_lock
);
1643 rll
->rll_writer
= curthread
;
1646 mutex_exit(&rll
->rll_lock
);
1650 ztest_rll_unlock(rll_t
*rll
)
1652 mutex_enter(&rll
->rll_lock
);
1654 if (rll
->rll_writer
) {
1655 ASSERT0(rll
->rll_readers
);
1656 rll
->rll_writer
= NULL
;
1658 ASSERT3S(rll
->rll_readers
, >, 0);
1659 ASSERT3P(rll
->rll_writer
, ==, NULL
);
1663 if (rll
->rll_writer
== NULL
&& rll
->rll_readers
== 0)
1664 cv_broadcast(&rll
->rll_cv
);
1666 mutex_exit(&rll
->rll_lock
);
1670 ztest_object_lock(ztest_ds_t
*zd
, uint64_t object
, rl_type_t type
)
1672 rll_t
*rll
= &zd
->zd_object_lock
[object
& (ZTEST_OBJECT_LOCKS
- 1)];
1674 ztest_rll_lock(rll
, type
);
1678 ztest_object_unlock(ztest_ds_t
*zd
, uint64_t object
)
1680 rll_t
*rll
= &zd
->zd_object_lock
[object
& (ZTEST_OBJECT_LOCKS
- 1)];
1682 ztest_rll_unlock(rll
);
1686 ztest_range_lock(ztest_ds_t
*zd
, uint64_t object
, uint64_t offset
,
1687 uint64_t size
, rl_type_t type
)
1689 uint64_t hash
= object
^ (offset
% (ZTEST_RANGE_LOCKS
+ 1));
1690 rll_t
*rll
= &zd
->zd_range_lock
[hash
& (ZTEST_RANGE_LOCKS
- 1)];
1693 rl
= umem_alloc(sizeof (*rl
), UMEM_NOFAIL
);
1694 rl
->rl_object
= object
;
1695 rl
->rl_offset
= offset
;
1699 ztest_rll_lock(rll
, type
);
1705 ztest_range_unlock(rl_t
*rl
)
1707 rll_t
*rll
= rl
->rl_lock
;
1709 ztest_rll_unlock(rll
);
1711 umem_free(rl
, sizeof (*rl
));
1715 ztest_zd_init(ztest_ds_t
*zd
, ztest_shared_ds_t
*szd
, objset_t
*os
)
1718 zd
->zd_zilog
= dmu_objset_zil(os
);
1719 zd
->zd_shared
= szd
;
1720 dmu_objset_name(os
, zd
->zd_name
);
1723 if (zd
->zd_shared
!= NULL
)
1724 zd
->zd_shared
->zd_seq
= 0;
1726 VERIFY0(pthread_rwlock_init(&zd
->zd_zilog_lock
, NULL
));
1727 mutex_init(&zd
->zd_dirobj_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
1729 for (l
= 0; l
< ZTEST_OBJECT_LOCKS
; l
++)
1730 ztest_rll_init(&zd
->zd_object_lock
[l
]);
1732 for (l
= 0; l
< ZTEST_RANGE_LOCKS
; l
++)
1733 ztest_rll_init(&zd
->zd_range_lock
[l
]);
1737 ztest_zd_fini(ztest_ds_t
*zd
)
1741 mutex_destroy(&zd
->zd_dirobj_lock
);
1742 (void) pthread_rwlock_destroy(&zd
->zd_zilog_lock
);
1744 for (l
= 0; l
< ZTEST_OBJECT_LOCKS
; l
++)
1745 ztest_rll_destroy(&zd
->zd_object_lock
[l
]);
1747 for (l
= 0; l
< ZTEST_RANGE_LOCKS
; l
++)
1748 ztest_rll_destroy(&zd
->zd_range_lock
[l
]);
1751 #define TXG_MIGHTWAIT (ztest_random(10) == 0 ? TXG_NOWAIT : TXG_WAIT)
1754 ztest_tx_assign(dmu_tx_t
*tx
, uint64_t txg_how
, const char *tag
)
1760 * Attempt to assign tx to some transaction group.
1762 error
= dmu_tx_assign(tx
, txg_how
);
1764 if (error
== ERESTART
) {
1765 ASSERT3U(txg_how
, ==, TXG_NOWAIT
);
1768 ASSERT3U(error
, ==, ENOSPC
);
1769 ztest_record_enospc(tag
);
1774 txg
= dmu_tx_get_txg(tx
);
1775 ASSERT3U(txg
, !=, 0);
1780 ztest_bt_generate(ztest_block_tag_t
*bt
, objset_t
*os
, uint64_t object
,
1781 uint64_t dnodesize
, uint64_t offset
, uint64_t gen
, uint64_t txg
,
1784 bt
->bt_magic
= BT_MAGIC
;
1785 bt
->bt_objset
= dmu_objset_id(os
);
1786 bt
->bt_object
= object
;
1787 bt
->bt_dnodesize
= dnodesize
;
1788 bt
->bt_offset
= offset
;
1791 bt
->bt_crtxg
= crtxg
;
1795 ztest_bt_verify(ztest_block_tag_t
*bt
, objset_t
*os
, uint64_t object
,
1796 uint64_t dnodesize
, uint64_t offset
, uint64_t gen
, uint64_t txg
,
1799 ASSERT3U(bt
->bt_magic
, ==, BT_MAGIC
);
1800 ASSERT3U(bt
->bt_objset
, ==, dmu_objset_id(os
));
1801 ASSERT3U(bt
->bt_object
, ==, object
);
1802 ASSERT3U(bt
->bt_dnodesize
, ==, dnodesize
);
1803 ASSERT3U(bt
->bt_offset
, ==, offset
);
1804 ASSERT3U(bt
->bt_gen
, <=, gen
);
1805 ASSERT3U(bt
->bt_txg
, <=, txg
);
1806 ASSERT3U(bt
->bt_crtxg
, ==, crtxg
);
1809 static ztest_block_tag_t
*
1810 ztest_bt_bonus(dmu_buf_t
*db
)
1812 dmu_object_info_t doi
;
1813 ztest_block_tag_t
*bt
;
1815 dmu_object_info_from_db(db
, &doi
);
1816 ASSERT3U(doi
.doi_bonus_size
, <=, db
->db_size
);
1817 ASSERT3U(doi
.doi_bonus_size
, >=, sizeof (*bt
));
1818 bt
= (void *)((char *)db
->db_data
+ doi
.doi_bonus_size
- sizeof (*bt
));
1824 * Generate a token to fill up unused bonus buffer space. Try to make
1825 * it unique to the object, generation, and offset to verify that data
1826 * is not getting overwritten by data from other dnodes.
1828 #define ZTEST_BONUS_FILL_TOKEN(obj, ds, gen, offset) \
1829 (((ds) << 48) | ((gen) << 32) | ((obj) << 8) | (offset))
1832 * Fill up the unused bonus buffer region before the block tag with a
1833 * verifiable pattern. Filling the whole bonus area with non-zero data
1834 * helps ensure that all dnode traversal code properly skips the
1835 * interior regions of large dnodes.
1838 ztest_fill_unused_bonus(dmu_buf_t
*db
, void *end
, uint64_t obj
,
1839 objset_t
*os
, uint64_t gen
)
1843 ASSERT(IS_P2ALIGNED((char *)end
- (char *)db
->db_data
, 8));
1845 for (bonusp
= db
->db_data
; bonusp
< (uint64_t *)end
; bonusp
++) {
1846 uint64_t token
= ZTEST_BONUS_FILL_TOKEN(obj
, dmu_objset_id(os
),
1847 gen
, bonusp
- (uint64_t *)db
->db_data
);
1853 * Verify that the unused area of a bonus buffer is filled with the
1857 ztest_verify_unused_bonus(dmu_buf_t
*db
, void *end
, uint64_t obj
,
1858 objset_t
*os
, uint64_t gen
)
1862 for (bonusp
= db
->db_data
; bonusp
< (uint64_t *)end
; bonusp
++) {
1863 uint64_t token
= ZTEST_BONUS_FILL_TOKEN(obj
, dmu_objset_id(os
),
1864 gen
, bonusp
- (uint64_t *)db
->db_data
);
1865 VERIFY3U(*bonusp
, ==, token
);
1873 #define lrz_type lr_mode
1874 #define lrz_blocksize lr_uid
1875 #define lrz_ibshift lr_gid
1876 #define lrz_bonustype lr_rdev
1877 #define lrz_dnodesize lr_crtime[1]
1880 ztest_log_create(ztest_ds_t
*zd
, dmu_tx_t
*tx
, lr_create_t
*lr
)
1882 char *name
= (void *)(lr
+ 1); /* name follows lr */
1883 size_t namesize
= strlen(name
) + 1;
1886 if (zil_replaying(zd
->zd_zilog
, tx
))
1889 itx
= zil_itx_create(TX_CREATE
, sizeof (*lr
) + namesize
);
1890 memcpy(&itx
->itx_lr
+ 1, &lr
->lr_common
+ 1,
1891 sizeof (*lr
) + namesize
- sizeof (lr_t
));
1893 zil_itx_assign(zd
->zd_zilog
, itx
, tx
);
1897 ztest_log_remove(ztest_ds_t
*zd
, dmu_tx_t
*tx
, lr_remove_t
*lr
, uint64_t object
)
1899 char *name
= (void *)(lr
+ 1); /* name follows lr */
1900 size_t namesize
= strlen(name
) + 1;
1903 if (zil_replaying(zd
->zd_zilog
, tx
))
1906 itx
= zil_itx_create(TX_REMOVE
, sizeof (*lr
) + namesize
);
1907 memcpy(&itx
->itx_lr
+ 1, &lr
->lr_common
+ 1,
1908 sizeof (*lr
) + namesize
- sizeof (lr_t
));
1910 itx
->itx_oid
= object
;
1911 zil_itx_assign(zd
->zd_zilog
, itx
, tx
);
1915 ztest_log_write(ztest_ds_t
*zd
, dmu_tx_t
*tx
, lr_write_t
*lr
)
1918 itx_wr_state_t write_state
= ztest_random(WR_NUM_STATES
);
1920 if (zil_replaying(zd
->zd_zilog
, tx
))
1923 if (lr
->lr_length
> zil_max_log_data(zd
->zd_zilog
))
1924 write_state
= WR_INDIRECT
;
1926 itx
= zil_itx_create(TX_WRITE
,
1927 sizeof (*lr
) + (write_state
== WR_COPIED
? lr
->lr_length
: 0));
1929 if (write_state
== WR_COPIED
&&
1930 dmu_read(zd
->zd_os
, lr
->lr_foid
, lr
->lr_offset
, lr
->lr_length
,
1931 ((lr_write_t
*)&itx
->itx_lr
) + 1, DMU_READ_NO_PREFETCH
) != 0) {
1932 zil_itx_destroy(itx
);
1933 itx
= zil_itx_create(TX_WRITE
, sizeof (*lr
));
1934 write_state
= WR_NEED_COPY
;
1936 itx
->itx_private
= zd
;
1937 itx
->itx_wr_state
= write_state
;
1938 itx
->itx_sync
= (ztest_random(8) == 0);
1940 memcpy(&itx
->itx_lr
+ 1, &lr
->lr_common
+ 1,
1941 sizeof (*lr
) - sizeof (lr_t
));
1943 zil_itx_assign(zd
->zd_zilog
, itx
, tx
);
1947 ztest_log_truncate(ztest_ds_t
*zd
, dmu_tx_t
*tx
, lr_truncate_t
*lr
)
1951 if (zil_replaying(zd
->zd_zilog
, tx
))
1954 itx
= zil_itx_create(TX_TRUNCATE
, sizeof (*lr
));
1955 memcpy(&itx
->itx_lr
+ 1, &lr
->lr_common
+ 1,
1956 sizeof (*lr
) - sizeof (lr_t
));
1958 itx
->itx_sync
= B_FALSE
;
1959 zil_itx_assign(zd
->zd_zilog
, itx
, tx
);
1963 ztest_log_setattr(ztest_ds_t
*zd
, dmu_tx_t
*tx
, lr_setattr_t
*lr
)
1967 if (zil_replaying(zd
->zd_zilog
, tx
))
1970 itx
= zil_itx_create(TX_SETATTR
, sizeof (*lr
));
1971 memcpy(&itx
->itx_lr
+ 1, &lr
->lr_common
+ 1,
1972 sizeof (*lr
) - sizeof (lr_t
));
1974 itx
->itx_sync
= B_FALSE
;
1975 zil_itx_assign(zd
->zd_zilog
, itx
, tx
);
1982 ztest_replay_create(void *arg1
, void *arg2
, boolean_t byteswap
)
1984 ztest_ds_t
*zd
= arg1
;
1985 lr_create_t
*lr
= arg2
;
1986 char *name
= (void *)(lr
+ 1); /* name follows lr */
1987 objset_t
*os
= zd
->zd_os
;
1988 ztest_block_tag_t
*bbt
;
1996 byteswap_uint64_array(lr
, sizeof (*lr
));
1998 ASSERT3U(lr
->lr_doid
, ==, ZTEST_DIROBJ
);
1999 ASSERT3S(name
[0], !=, '\0');
2001 tx
= dmu_tx_create(os
);
2003 dmu_tx_hold_zap(tx
, lr
->lr_doid
, B_TRUE
, name
);
2005 if (lr
->lrz_type
== DMU_OT_ZAP_OTHER
) {
2006 dmu_tx_hold_zap(tx
, DMU_NEW_OBJECT
, B_TRUE
, NULL
);
2008 dmu_tx_hold_bonus(tx
, DMU_NEW_OBJECT
);
2011 txg
= ztest_tx_assign(tx
, TXG_WAIT
, FTAG
);
2015 ASSERT3U(dmu_objset_zil(os
)->zl_replay
, ==, !!lr
->lr_foid
);
2016 bonuslen
= DN_BONUS_SIZE(lr
->lrz_dnodesize
);
2018 if (lr
->lrz_type
== DMU_OT_ZAP_OTHER
) {
2019 if (lr
->lr_foid
== 0) {
2020 lr
->lr_foid
= zap_create_dnsize(os
,
2021 lr
->lrz_type
, lr
->lrz_bonustype
,
2022 bonuslen
, lr
->lrz_dnodesize
, tx
);
2024 error
= zap_create_claim_dnsize(os
, lr
->lr_foid
,
2025 lr
->lrz_type
, lr
->lrz_bonustype
,
2026 bonuslen
, lr
->lrz_dnodesize
, tx
);
2029 if (lr
->lr_foid
== 0) {
2030 lr
->lr_foid
= dmu_object_alloc_dnsize(os
,
2031 lr
->lrz_type
, 0, lr
->lrz_bonustype
,
2032 bonuslen
, lr
->lrz_dnodesize
, tx
);
2034 error
= dmu_object_claim_dnsize(os
, lr
->lr_foid
,
2035 lr
->lrz_type
, 0, lr
->lrz_bonustype
,
2036 bonuslen
, lr
->lrz_dnodesize
, tx
);
2041 ASSERT3U(error
, ==, EEXIST
);
2042 ASSERT(zd
->zd_zilog
->zl_replay
);
2047 ASSERT3U(lr
->lr_foid
, !=, 0);
2049 if (lr
->lrz_type
!= DMU_OT_ZAP_OTHER
)
2050 VERIFY0(dmu_object_set_blocksize(os
, lr
->lr_foid
,
2051 lr
->lrz_blocksize
, lr
->lrz_ibshift
, tx
));
2053 VERIFY0(dmu_bonus_hold(os
, lr
->lr_foid
, FTAG
, &db
));
2054 bbt
= ztest_bt_bonus(db
);
2055 dmu_buf_will_dirty(db
, tx
);
2056 ztest_bt_generate(bbt
, os
, lr
->lr_foid
, lr
->lrz_dnodesize
, -1ULL,
2057 lr
->lr_gen
, txg
, txg
);
2058 ztest_fill_unused_bonus(db
, bbt
, lr
->lr_foid
, os
, lr
->lr_gen
);
2059 dmu_buf_rele(db
, FTAG
);
2061 VERIFY0(zap_add(os
, lr
->lr_doid
, name
, sizeof (uint64_t), 1,
2064 (void) ztest_log_create(zd
, tx
, lr
);
2072 ztest_replay_remove(void *arg1
, void *arg2
, boolean_t byteswap
)
2074 ztest_ds_t
*zd
= arg1
;
2075 lr_remove_t
*lr
= arg2
;
2076 char *name
= (void *)(lr
+ 1); /* name follows lr */
2077 objset_t
*os
= zd
->zd_os
;
2078 dmu_object_info_t doi
;
2080 uint64_t object
, txg
;
2083 byteswap_uint64_array(lr
, sizeof (*lr
));
2085 ASSERT3U(lr
->lr_doid
, ==, ZTEST_DIROBJ
);
2086 ASSERT3S(name
[0], !=, '\0');
2089 zap_lookup(os
, lr
->lr_doid
, name
, sizeof (object
), 1, &object
));
2090 ASSERT3U(object
, !=, 0);
2092 ztest_object_lock(zd
, object
, RL_WRITER
);
2094 VERIFY0(dmu_object_info(os
, object
, &doi
));
2096 tx
= dmu_tx_create(os
);
2098 dmu_tx_hold_zap(tx
, lr
->lr_doid
, B_FALSE
, name
);
2099 dmu_tx_hold_free(tx
, object
, 0, DMU_OBJECT_END
);
2101 txg
= ztest_tx_assign(tx
, TXG_WAIT
, FTAG
);
2103 ztest_object_unlock(zd
, object
);
2107 if (doi
.doi_type
== DMU_OT_ZAP_OTHER
) {
2108 VERIFY0(zap_destroy(os
, object
, tx
));
2110 VERIFY0(dmu_object_free(os
, object
, tx
));
2113 VERIFY0(zap_remove(os
, lr
->lr_doid
, name
, tx
));
2115 (void) ztest_log_remove(zd
, tx
, lr
, object
);
2119 ztest_object_unlock(zd
, object
);
2125 ztest_replay_write(void *arg1
, void *arg2
, boolean_t byteswap
)
2127 ztest_ds_t
*zd
= arg1
;
2128 lr_write_t
*lr
= arg2
;
2129 objset_t
*os
= zd
->zd_os
;
2130 void *data
= lr
+ 1; /* data follows lr */
2131 uint64_t offset
, length
;
2132 ztest_block_tag_t
*bt
= data
;
2133 ztest_block_tag_t
*bbt
;
2134 uint64_t gen
, txg
, lrtxg
, crtxg
;
2135 dmu_object_info_t doi
;
2138 arc_buf_t
*abuf
= NULL
;
2142 byteswap_uint64_array(lr
, sizeof (*lr
));
2144 offset
= lr
->lr_offset
;
2145 length
= lr
->lr_length
;
2147 /* If it's a dmu_sync() block, write the whole block */
2148 if (lr
->lr_common
.lrc_reclen
== sizeof (lr_write_t
)) {
2149 uint64_t blocksize
= BP_GET_LSIZE(&lr
->lr_blkptr
);
2150 if (length
< blocksize
) {
2151 offset
-= offset
% blocksize
;
2156 if (bt
->bt_magic
== BSWAP_64(BT_MAGIC
))
2157 byteswap_uint64_array(bt
, sizeof (*bt
));
2159 if (bt
->bt_magic
!= BT_MAGIC
)
2162 ztest_object_lock(zd
, lr
->lr_foid
, RL_READER
);
2163 rl
= ztest_range_lock(zd
, lr
->lr_foid
, offset
, length
, RL_WRITER
);
2165 VERIFY0(dmu_bonus_hold(os
, lr
->lr_foid
, FTAG
, &db
));
2167 dmu_object_info_from_db(db
, &doi
);
2169 bbt
= ztest_bt_bonus(db
);
2170 ASSERT3U(bbt
->bt_magic
, ==, BT_MAGIC
);
2172 crtxg
= bbt
->bt_crtxg
;
2173 lrtxg
= lr
->lr_common
.lrc_txg
;
2175 tx
= dmu_tx_create(os
);
2177 dmu_tx_hold_write(tx
, lr
->lr_foid
, offset
, length
);
2179 if (ztest_random(8) == 0 && length
== doi
.doi_data_block_size
&&
2180 P2PHASE(offset
, length
) == 0)
2181 abuf
= dmu_request_arcbuf(db
, length
);
2183 txg
= ztest_tx_assign(tx
, TXG_WAIT
, FTAG
);
2186 dmu_return_arcbuf(abuf
);
2187 dmu_buf_rele(db
, FTAG
);
2188 ztest_range_unlock(rl
);
2189 ztest_object_unlock(zd
, lr
->lr_foid
);
2195 * Usually, verify the old data before writing new data --
2196 * but not always, because we also want to verify correct
2197 * behavior when the data was not recently read into cache.
2199 ASSERT0(offset
% doi
.doi_data_block_size
);
2200 if (ztest_random(4) != 0) {
2201 int prefetch
= ztest_random(2) ?
2202 DMU_READ_PREFETCH
: DMU_READ_NO_PREFETCH
;
2203 ztest_block_tag_t rbt
;
2205 VERIFY(dmu_read(os
, lr
->lr_foid
, offset
,
2206 sizeof (rbt
), &rbt
, prefetch
) == 0);
2207 if (rbt
.bt_magic
== BT_MAGIC
) {
2208 ztest_bt_verify(&rbt
, os
, lr
->lr_foid
, 0,
2209 offset
, gen
, txg
, crtxg
);
2214 * Writes can appear to be newer than the bonus buffer because
2215 * the ztest_get_data() callback does a dmu_read() of the
2216 * open-context data, which may be different than the data
2217 * as it was when the write was generated.
2219 if (zd
->zd_zilog
->zl_replay
) {
2220 ztest_bt_verify(bt
, os
, lr
->lr_foid
, 0, offset
,
2221 MAX(gen
, bt
->bt_gen
), MAX(txg
, lrtxg
),
2226 * Set the bt's gen/txg to the bonus buffer's gen/txg
2227 * so that all of the usual ASSERTs will work.
2229 ztest_bt_generate(bt
, os
, lr
->lr_foid
, 0, offset
, gen
, txg
,
2234 dmu_write(os
, lr
->lr_foid
, offset
, length
, data
, tx
);
2236 memcpy(abuf
->b_data
, data
, length
);
2237 dmu_assign_arcbuf_by_dbuf(db
, offset
, abuf
, tx
);
2240 (void) ztest_log_write(zd
, tx
, lr
);
2242 dmu_buf_rele(db
, FTAG
);
2246 ztest_range_unlock(rl
);
2247 ztest_object_unlock(zd
, lr
->lr_foid
);
2253 ztest_replay_truncate(void *arg1
, void *arg2
, boolean_t byteswap
)
2255 ztest_ds_t
*zd
= arg1
;
2256 lr_truncate_t
*lr
= arg2
;
2257 objset_t
*os
= zd
->zd_os
;
2263 byteswap_uint64_array(lr
, sizeof (*lr
));
2265 ztest_object_lock(zd
, lr
->lr_foid
, RL_READER
);
2266 rl
= ztest_range_lock(zd
, lr
->lr_foid
, lr
->lr_offset
, lr
->lr_length
,
2269 tx
= dmu_tx_create(os
);
2271 dmu_tx_hold_free(tx
, lr
->lr_foid
, lr
->lr_offset
, lr
->lr_length
);
2273 txg
= ztest_tx_assign(tx
, TXG_WAIT
, FTAG
);
2275 ztest_range_unlock(rl
);
2276 ztest_object_unlock(zd
, lr
->lr_foid
);
2280 VERIFY0(dmu_free_range(os
, lr
->lr_foid
, lr
->lr_offset
,
2281 lr
->lr_length
, tx
));
2283 (void) ztest_log_truncate(zd
, tx
, lr
);
2287 ztest_range_unlock(rl
);
2288 ztest_object_unlock(zd
, lr
->lr_foid
);
2294 ztest_replay_setattr(void *arg1
, void *arg2
, boolean_t byteswap
)
2296 ztest_ds_t
*zd
= arg1
;
2297 lr_setattr_t
*lr
= arg2
;
2298 objset_t
*os
= zd
->zd_os
;
2301 ztest_block_tag_t
*bbt
;
2302 uint64_t txg
, lrtxg
, crtxg
, dnodesize
;
2305 byteswap_uint64_array(lr
, sizeof (*lr
));
2307 ztest_object_lock(zd
, lr
->lr_foid
, RL_WRITER
);
2309 VERIFY0(dmu_bonus_hold(os
, lr
->lr_foid
, FTAG
, &db
));
2311 tx
= dmu_tx_create(os
);
2312 dmu_tx_hold_bonus(tx
, lr
->lr_foid
);
2314 txg
= ztest_tx_assign(tx
, TXG_WAIT
, FTAG
);
2316 dmu_buf_rele(db
, FTAG
);
2317 ztest_object_unlock(zd
, lr
->lr_foid
);
2321 bbt
= ztest_bt_bonus(db
);
2322 ASSERT3U(bbt
->bt_magic
, ==, BT_MAGIC
);
2323 crtxg
= bbt
->bt_crtxg
;
2324 lrtxg
= lr
->lr_common
.lrc_txg
;
2325 dnodesize
= bbt
->bt_dnodesize
;
2327 if (zd
->zd_zilog
->zl_replay
) {
2328 ASSERT3U(lr
->lr_size
, !=, 0);
2329 ASSERT3U(lr
->lr_mode
, !=, 0);
2330 ASSERT3U(lrtxg
, !=, 0);
2333 * Randomly change the size and increment the generation.
2335 lr
->lr_size
= (ztest_random(db
->db_size
/ sizeof (*bbt
)) + 1) *
2337 lr
->lr_mode
= bbt
->bt_gen
+ 1;
2342 * Verify that the current bonus buffer is not newer than our txg.
2344 ztest_bt_verify(bbt
, os
, lr
->lr_foid
, dnodesize
, -1ULL, lr
->lr_mode
,
2345 MAX(txg
, lrtxg
), crtxg
);
2347 dmu_buf_will_dirty(db
, tx
);
2349 ASSERT3U(lr
->lr_size
, >=, sizeof (*bbt
));
2350 ASSERT3U(lr
->lr_size
, <=, db
->db_size
);
2351 VERIFY0(dmu_set_bonus(db
, lr
->lr_size
, tx
));
2352 bbt
= ztest_bt_bonus(db
);
2354 ztest_bt_generate(bbt
, os
, lr
->lr_foid
, dnodesize
, -1ULL, lr
->lr_mode
,
2356 ztest_fill_unused_bonus(db
, bbt
, lr
->lr_foid
, os
, bbt
->bt_gen
);
2357 dmu_buf_rele(db
, FTAG
);
2359 (void) ztest_log_setattr(zd
, tx
, lr
);
2363 ztest_object_unlock(zd
, lr
->lr_foid
);
2368 zil_replay_func_t
*ztest_replay_vector
[TX_MAX_TYPE
] = {
2369 NULL
, /* 0 no such transaction type */
2370 ztest_replay_create
, /* TX_CREATE */
2371 NULL
, /* TX_MKDIR */
2372 NULL
, /* TX_MKXATTR */
2373 NULL
, /* TX_SYMLINK */
2374 ztest_replay_remove
, /* TX_REMOVE */
2375 NULL
, /* TX_RMDIR */
2377 NULL
, /* TX_RENAME */
2378 ztest_replay_write
, /* TX_WRITE */
2379 ztest_replay_truncate
, /* TX_TRUNCATE */
2380 ztest_replay_setattr
, /* TX_SETATTR */
2382 NULL
, /* TX_CREATE_ACL */
2383 NULL
, /* TX_CREATE_ATTR */
2384 NULL
, /* TX_CREATE_ACL_ATTR */
2385 NULL
, /* TX_MKDIR_ACL */
2386 NULL
, /* TX_MKDIR_ATTR */
2387 NULL
, /* TX_MKDIR_ACL_ATTR */
2388 NULL
, /* TX_WRITE2 */
2389 NULL
, /* TX_SETSAXATTR */
2393 * ZIL get_data callbacks
2397 ztest_get_done(zgd_t
*zgd
, int error
)
2400 ztest_ds_t
*zd
= zgd
->zgd_private
;
2401 uint64_t object
= ((rl_t
*)zgd
->zgd_lr
)->rl_object
;
2404 dmu_buf_rele(zgd
->zgd_db
, zgd
);
2406 ztest_range_unlock((rl_t
*)zgd
->zgd_lr
);
2407 ztest_object_unlock(zd
, object
);
2409 umem_free(zgd
, sizeof (*zgd
));
2413 ztest_get_data(void *arg
, uint64_t arg2
, lr_write_t
*lr
, char *buf
,
2414 struct lwb
*lwb
, zio_t
*zio
)
2417 ztest_ds_t
*zd
= arg
;
2418 objset_t
*os
= zd
->zd_os
;
2419 uint64_t object
= lr
->lr_foid
;
2420 uint64_t offset
= lr
->lr_offset
;
2421 uint64_t size
= lr
->lr_length
;
2422 uint64_t txg
= lr
->lr_common
.lrc_txg
;
2424 dmu_object_info_t doi
;
2429 ASSERT3P(lwb
, !=, NULL
);
2430 ASSERT3P(zio
, !=, NULL
);
2431 ASSERT3U(size
, !=, 0);
2433 ztest_object_lock(zd
, object
, RL_READER
);
2434 error
= dmu_bonus_hold(os
, object
, FTAG
, &db
);
2436 ztest_object_unlock(zd
, object
);
2440 crtxg
= ztest_bt_bonus(db
)->bt_crtxg
;
2442 if (crtxg
== 0 || crtxg
> txg
) {
2443 dmu_buf_rele(db
, FTAG
);
2444 ztest_object_unlock(zd
, object
);
2448 dmu_object_info_from_db(db
, &doi
);
2449 dmu_buf_rele(db
, FTAG
);
2452 zgd
= umem_zalloc(sizeof (*zgd
), UMEM_NOFAIL
);
2454 zgd
->zgd_private
= zd
;
2456 if (buf
!= NULL
) { /* immediate write */
2457 zgd
->zgd_lr
= (struct zfs_locked_range
*)ztest_range_lock(zd
,
2458 object
, offset
, size
, RL_READER
);
2460 error
= dmu_read(os
, object
, offset
, size
, buf
,
2461 DMU_READ_NO_PREFETCH
);
2464 size
= doi
.doi_data_block_size
;
2466 offset
= P2ALIGN(offset
, size
);
2468 ASSERT3U(offset
, <, size
);
2472 zgd
->zgd_lr
= (struct zfs_locked_range
*)ztest_range_lock(zd
,
2473 object
, offset
, size
, RL_READER
);
2475 error
= dmu_buf_hold(os
, object
, offset
, zgd
, &db
,
2476 DMU_READ_NO_PREFETCH
);
2479 blkptr_t
*bp
= &lr
->lr_blkptr
;
2484 ASSERT3U(db
->db_offset
, ==, offset
);
2485 ASSERT3U(db
->db_size
, ==, size
);
2487 error
= dmu_sync(zio
, lr
->lr_common
.lrc_txg
,
2488 ztest_get_done
, zgd
);
2495 ztest_get_done(zgd
, error
);
2501 ztest_lr_alloc(size_t lrsize
, char *name
)
2504 size_t namesize
= name
? strlen(name
) + 1 : 0;
2506 lr
= umem_zalloc(lrsize
+ namesize
, UMEM_NOFAIL
);
2509 memcpy(lr
+ lrsize
, name
, namesize
);
2515 ztest_lr_free(void *lr
, size_t lrsize
, char *name
)
2517 size_t namesize
= name
? strlen(name
) + 1 : 0;
2519 umem_free(lr
, lrsize
+ namesize
);
2523 * Lookup a bunch of objects. Returns the number of objects not found.
2526 ztest_lookup(ztest_ds_t
*zd
, ztest_od_t
*od
, int count
)
2532 ASSERT(MUTEX_HELD(&zd
->zd_dirobj_lock
));
2534 for (i
= 0; i
< count
; i
++, od
++) {
2536 error
= zap_lookup(zd
->zd_os
, od
->od_dir
, od
->od_name
,
2537 sizeof (uint64_t), 1, &od
->od_object
);
2539 ASSERT3S(error
, ==, ENOENT
);
2540 ASSERT0(od
->od_object
);
2544 ztest_block_tag_t
*bbt
;
2545 dmu_object_info_t doi
;
2547 ASSERT3U(od
->od_object
, !=, 0);
2548 ASSERT0(missing
); /* there should be no gaps */
2550 ztest_object_lock(zd
, od
->od_object
, RL_READER
);
2551 VERIFY0(dmu_bonus_hold(zd
->zd_os
, od
->od_object
,
2553 dmu_object_info_from_db(db
, &doi
);
2554 bbt
= ztest_bt_bonus(db
);
2555 ASSERT3U(bbt
->bt_magic
, ==, BT_MAGIC
);
2556 od
->od_type
= doi
.doi_type
;
2557 od
->od_blocksize
= doi
.doi_data_block_size
;
2558 od
->od_gen
= bbt
->bt_gen
;
2559 dmu_buf_rele(db
, FTAG
);
2560 ztest_object_unlock(zd
, od
->od_object
);
2568 ztest_create(ztest_ds_t
*zd
, ztest_od_t
*od
, int count
)
2573 ASSERT(MUTEX_HELD(&zd
->zd_dirobj_lock
));
2575 for (i
= 0; i
< count
; i
++, od
++) {
2582 lr_create_t
*lr
= ztest_lr_alloc(sizeof (*lr
), od
->od_name
);
2584 lr
->lr_doid
= od
->od_dir
;
2585 lr
->lr_foid
= 0; /* 0 to allocate, > 0 to claim */
2586 lr
->lrz_type
= od
->od_crtype
;
2587 lr
->lrz_blocksize
= od
->od_crblocksize
;
2588 lr
->lrz_ibshift
= ztest_random_ibshift();
2589 lr
->lrz_bonustype
= DMU_OT_UINT64_OTHER
;
2590 lr
->lrz_dnodesize
= od
->od_crdnodesize
;
2591 lr
->lr_gen
= od
->od_crgen
;
2592 lr
->lr_crtime
[0] = time(NULL
);
2594 if (ztest_replay_create(zd
, lr
, B_FALSE
) != 0) {
2599 od
->od_object
= lr
->lr_foid
;
2600 od
->od_type
= od
->od_crtype
;
2601 od
->od_blocksize
= od
->od_crblocksize
;
2602 od
->od_gen
= od
->od_crgen
;
2603 ASSERT3U(od
->od_object
, !=, 0);
2606 ztest_lr_free(lr
, sizeof (*lr
), od
->od_name
);
2613 ztest_remove(ztest_ds_t
*zd
, ztest_od_t
*od
, int count
)
2619 ASSERT(MUTEX_HELD(&zd
->zd_dirobj_lock
));
2623 for (i
= count
- 1; i
>= 0; i
--, od
--) {
2630 * No object was found.
2632 if (od
->od_object
== 0)
2635 lr_remove_t
*lr
= ztest_lr_alloc(sizeof (*lr
), od
->od_name
);
2637 lr
->lr_doid
= od
->od_dir
;
2639 if ((error
= ztest_replay_remove(zd
, lr
, B_FALSE
)) != 0) {
2640 ASSERT3U(error
, ==, ENOSPC
);
2645 ztest_lr_free(lr
, sizeof (*lr
), od
->od_name
);
2652 ztest_write(ztest_ds_t
*zd
, uint64_t object
, uint64_t offset
, uint64_t size
,
2658 lr
= ztest_lr_alloc(sizeof (*lr
) + size
, NULL
);
2660 lr
->lr_foid
= object
;
2661 lr
->lr_offset
= offset
;
2662 lr
->lr_length
= size
;
2664 BP_ZERO(&lr
->lr_blkptr
);
2666 memcpy(lr
+ 1, data
, size
);
2668 error
= ztest_replay_write(zd
, lr
, B_FALSE
);
2670 ztest_lr_free(lr
, sizeof (*lr
) + size
, NULL
);
2676 ztest_truncate(ztest_ds_t
*zd
, uint64_t object
, uint64_t offset
, uint64_t size
)
2681 lr
= ztest_lr_alloc(sizeof (*lr
), NULL
);
2683 lr
->lr_foid
= object
;
2684 lr
->lr_offset
= offset
;
2685 lr
->lr_length
= size
;
2687 error
= ztest_replay_truncate(zd
, lr
, B_FALSE
);
2689 ztest_lr_free(lr
, sizeof (*lr
), NULL
);
2695 ztest_setattr(ztest_ds_t
*zd
, uint64_t object
)
2700 lr
= ztest_lr_alloc(sizeof (*lr
), NULL
);
2702 lr
->lr_foid
= object
;
2706 error
= ztest_replay_setattr(zd
, lr
, B_FALSE
);
2708 ztest_lr_free(lr
, sizeof (*lr
), NULL
);
2714 ztest_prealloc(ztest_ds_t
*zd
, uint64_t object
, uint64_t offset
, uint64_t size
)
2716 objset_t
*os
= zd
->zd_os
;
2721 txg_wait_synced(dmu_objset_pool(os
), 0);
2723 ztest_object_lock(zd
, object
, RL_READER
);
2724 rl
= ztest_range_lock(zd
, object
, offset
, size
, RL_WRITER
);
2726 tx
= dmu_tx_create(os
);
2728 dmu_tx_hold_write(tx
, object
, offset
, size
);
2730 txg
= ztest_tx_assign(tx
, TXG_WAIT
, FTAG
);
2733 dmu_prealloc(os
, object
, offset
, size
, tx
);
2735 txg_wait_synced(dmu_objset_pool(os
), txg
);
2737 (void) dmu_free_long_range(os
, object
, offset
, size
);
2740 ztest_range_unlock(rl
);
2741 ztest_object_unlock(zd
, object
);
2745 ztest_io(ztest_ds_t
*zd
, uint64_t object
, uint64_t offset
)
2748 ztest_block_tag_t wbt
;
2749 dmu_object_info_t doi
;
2750 enum ztest_io_type io_type
;
2754 VERIFY0(dmu_object_info(zd
->zd_os
, object
, &doi
));
2755 blocksize
= doi
.doi_data_block_size
;
2756 data
= umem_alloc(blocksize
, UMEM_NOFAIL
);
2759 * Pick an i/o type at random, biased toward writing block tags.
2761 io_type
= ztest_random(ZTEST_IO_TYPES
);
2762 if (ztest_random(2) == 0)
2763 io_type
= ZTEST_IO_WRITE_TAG
;
2765 (void) pthread_rwlock_rdlock(&zd
->zd_zilog_lock
);
2769 case ZTEST_IO_WRITE_TAG
:
2770 ztest_bt_generate(&wbt
, zd
->zd_os
, object
, doi
.doi_dnodesize
,
2772 (void) ztest_write(zd
, object
, offset
, sizeof (wbt
), &wbt
);
2775 case ZTEST_IO_WRITE_PATTERN
:
2776 (void) memset(data
, 'a' + (object
+ offset
) % 5, blocksize
);
2777 if (ztest_random(2) == 0) {
2779 * Induce fletcher2 collisions to ensure that
2780 * zio_ddt_collision() detects and resolves them
2781 * when using fletcher2-verify for deduplication.
2783 ((uint64_t *)data
)[0] ^= 1ULL << 63;
2784 ((uint64_t *)data
)[4] ^= 1ULL << 63;
2786 (void) ztest_write(zd
, object
, offset
, blocksize
, data
);
2789 case ZTEST_IO_WRITE_ZEROES
:
2790 memset(data
, 0, blocksize
);
2791 (void) ztest_write(zd
, object
, offset
, blocksize
, data
);
2794 case ZTEST_IO_TRUNCATE
:
2795 (void) ztest_truncate(zd
, object
, offset
, blocksize
);
2798 case ZTEST_IO_SETATTR
:
2799 (void) ztest_setattr(zd
, object
);
2804 case ZTEST_IO_REWRITE
:
2805 (void) pthread_rwlock_rdlock(&ztest_name_lock
);
2806 err
= ztest_dsl_prop_set_uint64(zd
->zd_name
,
2807 ZFS_PROP_CHECKSUM
, spa_dedup_checksum(ztest_spa
),
2809 VERIFY(err
== 0 || err
== ENOSPC
);
2810 err
= ztest_dsl_prop_set_uint64(zd
->zd_name
,
2811 ZFS_PROP_COMPRESSION
,
2812 ztest_random_dsl_prop(ZFS_PROP_COMPRESSION
),
2814 VERIFY(err
== 0 || err
== ENOSPC
);
2815 (void) pthread_rwlock_unlock(&ztest_name_lock
);
2817 VERIFY0(dmu_read(zd
->zd_os
, object
, offset
, blocksize
, data
,
2818 DMU_READ_NO_PREFETCH
));
2820 (void) ztest_write(zd
, object
, offset
, blocksize
, data
);
2824 (void) pthread_rwlock_unlock(&zd
->zd_zilog_lock
);
2826 umem_free(data
, blocksize
);
2830 * Initialize an object description template.
2833 ztest_od_init(ztest_od_t
*od
, uint64_t id
, char *tag
, uint64_t index
,
2834 dmu_object_type_t type
, uint64_t blocksize
, uint64_t dnodesize
,
2837 od
->od_dir
= ZTEST_DIROBJ
;
2840 od
->od_crtype
= type
;
2841 od
->od_crblocksize
= blocksize
? blocksize
: ztest_random_blocksize();
2842 od
->od_crdnodesize
= dnodesize
? dnodesize
: ztest_random_dnodesize();
2845 od
->od_type
= DMU_OT_NONE
;
2846 od
->od_blocksize
= 0;
2849 (void) snprintf(od
->od_name
, sizeof (od
->od_name
),
2850 "%s(%"PRId64
")[%"PRIu64
"]",
2855 * Lookup or create the objects for a test using the od template.
2856 * If the objects do not all exist, or if 'remove' is specified,
2857 * remove any existing objects and create new ones. Otherwise,
2858 * use the existing objects.
2861 ztest_object_init(ztest_ds_t
*zd
, ztest_od_t
*od
, size_t size
, boolean_t remove
)
2863 int count
= size
/ sizeof (*od
);
2866 mutex_enter(&zd
->zd_dirobj_lock
);
2867 if ((ztest_lookup(zd
, od
, count
) != 0 || remove
) &&
2868 (ztest_remove(zd
, od
, count
) != 0 ||
2869 ztest_create(zd
, od
, count
) != 0))
2872 mutex_exit(&zd
->zd_dirobj_lock
);
2878 ztest_zil_commit(ztest_ds_t
*zd
, uint64_t id
)
2881 zilog_t
*zilog
= zd
->zd_zilog
;
2883 (void) pthread_rwlock_rdlock(&zd
->zd_zilog_lock
);
2885 zil_commit(zilog
, ztest_random(ZTEST_OBJECTS
));
2888 * Remember the committed values in zd, which is in parent/child
2889 * shared memory. If we die, the next iteration of ztest_run()
2890 * will verify that the log really does contain this record.
2892 mutex_enter(&zilog
->zl_lock
);
2893 ASSERT3P(zd
->zd_shared
, !=, NULL
);
2894 ASSERT3U(zd
->zd_shared
->zd_seq
, <=, zilog
->zl_commit_lr_seq
);
2895 zd
->zd_shared
->zd_seq
= zilog
->zl_commit_lr_seq
;
2896 mutex_exit(&zilog
->zl_lock
);
2898 (void) pthread_rwlock_unlock(&zd
->zd_zilog_lock
);
2902 * This function is designed to simulate the operations that occur during a
2903 * mount/unmount operation. We hold the dataset across these operations in an
2904 * attempt to expose any implicit assumptions about ZIL management.
2907 ztest_zil_remount(ztest_ds_t
*zd
, uint64_t id
)
2910 objset_t
*os
= zd
->zd_os
;
2913 * We hold the ztest_vdev_lock so we don't cause problems with
2914 * other threads that wish to remove a log device, such as
2915 * ztest_device_removal().
2917 mutex_enter(&ztest_vdev_lock
);
2920 * We grab the zd_dirobj_lock to ensure that no other thread is
2921 * updating the zil (i.e. adding in-memory log records) and the
2922 * zd_zilog_lock to block any I/O.
2924 mutex_enter(&zd
->zd_dirobj_lock
);
2925 (void) pthread_rwlock_wrlock(&zd
->zd_zilog_lock
);
2927 /* zfsvfs_teardown() */
2928 zil_close(zd
->zd_zilog
);
2930 /* zfsvfs_setup() */
2931 VERIFY3P(zil_open(os
, ztest_get_data
), ==, zd
->zd_zilog
);
2932 zil_replay(os
, zd
, ztest_replay_vector
);
2934 (void) pthread_rwlock_unlock(&zd
->zd_zilog_lock
);
2935 mutex_exit(&zd
->zd_dirobj_lock
);
2936 mutex_exit(&ztest_vdev_lock
);
2940 * Verify that we can't destroy an active pool, create an existing pool,
2941 * or create a pool with a bad vdev spec.
2944 ztest_spa_create_destroy(ztest_ds_t
*zd
, uint64_t id
)
2946 (void) zd
, (void) id
;
2947 ztest_shared_opts_t
*zo
= &ztest_opts
;
2951 if (zo
->zo_mmp_test
)
2955 * Attempt to create using a bad file.
2957 nvroot
= make_vdev_root("/dev/bogus", NULL
, NULL
, 0, 0, NULL
, 0, 0, 1);
2958 VERIFY3U(ENOENT
, ==,
2959 spa_create("ztest_bad_file", nvroot
, NULL
, NULL
, NULL
));
2960 fnvlist_free(nvroot
);
2963 * Attempt to create using a bad mirror.
2965 nvroot
= make_vdev_root("/dev/bogus", NULL
, NULL
, 0, 0, NULL
, 0, 2, 1);
2966 VERIFY3U(ENOENT
, ==,
2967 spa_create("ztest_bad_mirror", nvroot
, NULL
, NULL
, NULL
));
2968 fnvlist_free(nvroot
);
2971 * Attempt to create an existing pool. It shouldn't matter
2972 * what's in the nvroot; we should fail with EEXIST.
2974 (void) pthread_rwlock_rdlock(&ztest_name_lock
);
2975 nvroot
= make_vdev_root("/dev/bogus", NULL
, NULL
, 0, 0, NULL
, 0, 0, 1);
2976 VERIFY3U(EEXIST
, ==,
2977 spa_create(zo
->zo_pool
, nvroot
, NULL
, NULL
, NULL
));
2978 fnvlist_free(nvroot
);
2981 * We open a reference to the spa and then we try to export it
2982 * expecting one of the following errors:
2985 * Because of the reference we just opened.
2987 * ZFS_ERR_EXPORT_IN_PROGRESS
2988 * For the case that there is another ztest thread doing
2989 * an export concurrently.
2991 VERIFY0(spa_open(zo
->zo_pool
, &spa
, FTAG
));
2992 int error
= spa_destroy(zo
->zo_pool
);
2993 if (error
!= EBUSY
&& error
!= ZFS_ERR_EXPORT_IN_PROGRESS
) {
2994 fatal(B_FALSE
, "spa_destroy(%s) returned unexpected value %d",
2995 spa
->spa_name
, error
);
2997 spa_close(spa
, FTAG
);
2999 (void) pthread_rwlock_unlock(&ztest_name_lock
);
3003 * Start and then stop the MMP threads to ensure the startup and shutdown code
3004 * works properly. Actual protection and property-related code tested via ZTS.
3007 ztest_mmp_enable_disable(ztest_ds_t
*zd
, uint64_t id
)
3009 (void) zd
, (void) id
;
3010 ztest_shared_opts_t
*zo
= &ztest_opts
;
3011 spa_t
*spa
= ztest_spa
;
3013 if (zo
->zo_mmp_test
)
3017 * Since enabling MMP involves setting a property, it could not be done
3018 * while the pool is suspended.
3020 if (spa_suspended(spa
))
3023 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
3024 mutex_enter(&spa
->spa_props_lock
);
3026 zfs_multihost_fail_intervals
= 0;
3028 if (!spa_multihost(spa
)) {
3029 spa
->spa_multihost
= B_TRUE
;
3030 mmp_thread_start(spa
);
3033 mutex_exit(&spa
->spa_props_lock
);
3034 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
3036 txg_wait_synced(spa_get_dsl(spa
), 0);
3037 mmp_signal_all_threads();
3038 txg_wait_synced(spa_get_dsl(spa
), 0);
3040 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
3041 mutex_enter(&spa
->spa_props_lock
);
3043 if (spa_multihost(spa
)) {
3044 mmp_thread_stop(spa
);
3045 spa
->spa_multihost
= B_FALSE
;
3048 mutex_exit(&spa
->spa_props_lock
);
3049 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
3053 ztest_spa_upgrade(ztest_ds_t
*zd
, uint64_t id
)
3055 (void) zd
, (void) id
;
3057 uint64_t initial_version
= SPA_VERSION_INITIAL
;
3058 uint64_t version
, newversion
;
3059 nvlist_t
*nvroot
, *props
;
3062 if (ztest_opts
.zo_mmp_test
)
3065 /* dRAID added after feature flags, skip upgrade test. */
3066 if (strcmp(ztest_opts
.zo_raid_type
, VDEV_TYPE_DRAID
) == 0)
3069 mutex_enter(&ztest_vdev_lock
);
3070 name
= kmem_asprintf("%s_upgrade", ztest_opts
.zo_pool
);
3073 * Clean up from previous runs.
3075 (void) spa_destroy(name
);
3077 nvroot
= make_vdev_root(NULL
, NULL
, name
, ztest_opts
.zo_vdev_size
, 0,
3078 NULL
, ztest_opts
.zo_raid_children
, ztest_opts
.zo_mirrors
, 1);
3081 * If we're configuring a RAIDZ device then make sure that the
3082 * initial version is capable of supporting that feature.
3084 switch (ztest_opts
.zo_raid_parity
) {
3087 initial_version
= SPA_VERSION_INITIAL
;
3090 initial_version
= SPA_VERSION_RAIDZ2
;
3093 initial_version
= SPA_VERSION_RAIDZ3
;
3098 * Create a pool with a spa version that can be upgraded. Pick
3099 * a value between initial_version and SPA_VERSION_BEFORE_FEATURES.
3102 version
= ztest_random_spa_version(initial_version
);
3103 } while (version
> SPA_VERSION_BEFORE_FEATURES
);
3105 props
= fnvlist_alloc();
3106 fnvlist_add_uint64(props
,
3107 zpool_prop_to_name(ZPOOL_PROP_VERSION
), version
);
3108 VERIFY0(spa_create(name
, nvroot
, props
, NULL
, NULL
));
3109 fnvlist_free(nvroot
);
3110 fnvlist_free(props
);
3112 VERIFY0(spa_open(name
, &spa
, FTAG
));
3113 VERIFY3U(spa_version(spa
), ==, version
);
3114 newversion
= ztest_random_spa_version(version
+ 1);
3116 if (ztest_opts
.zo_verbose
>= 4) {
3117 (void) printf("upgrading spa version from "
3118 "%"PRIu64
" to %"PRIu64
"\n",
3119 version
, newversion
);
3122 spa_upgrade(spa
, newversion
);
3123 VERIFY3U(spa_version(spa
), >, version
);
3124 VERIFY3U(spa_version(spa
), ==, fnvlist_lookup_uint64(spa
->spa_config
,
3125 zpool_prop_to_name(ZPOOL_PROP_VERSION
)));
3126 spa_close(spa
, FTAG
);
3129 mutex_exit(&ztest_vdev_lock
);
3133 ztest_spa_checkpoint(spa_t
*spa
)
3135 ASSERT(MUTEX_HELD(&ztest_checkpoint_lock
));
3137 int error
= spa_checkpoint(spa
->spa_name
);
3141 case ZFS_ERR_DEVRM_IN_PROGRESS
:
3142 case ZFS_ERR_DISCARDING_CHECKPOINT
:
3143 case ZFS_ERR_CHECKPOINT_EXISTS
:
3146 ztest_record_enospc(FTAG
);
3149 fatal(B_FALSE
, "spa_checkpoint(%s) = %d", spa
->spa_name
, error
);
3154 ztest_spa_discard_checkpoint(spa_t
*spa
)
3156 ASSERT(MUTEX_HELD(&ztest_checkpoint_lock
));
3158 int error
= spa_checkpoint_discard(spa
->spa_name
);
3162 case ZFS_ERR_DISCARDING_CHECKPOINT
:
3163 case ZFS_ERR_NO_CHECKPOINT
:
3166 fatal(B_FALSE
, "spa_discard_checkpoint(%s) = %d",
3167 spa
->spa_name
, error
);
3173 ztest_spa_checkpoint_create_discard(ztest_ds_t
*zd
, uint64_t id
)
3175 (void) zd
, (void) id
;
3176 spa_t
*spa
= ztest_spa
;
3178 mutex_enter(&ztest_checkpoint_lock
);
3179 if (ztest_random(2) == 0) {
3180 ztest_spa_checkpoint(spa
);
3182 ztest_spa_discard_checkpoint(spa
);
3184 mutex_exit(&ztest_checkpoint_lock
);
3189 vdev_lookup_by_path(vdev_t
*vd
, const char *path
)
3194 if (vd
->vdev_path
!= NULL
&& strcmp(path
, vd
->vdev_path
) == 0)
3197 for (c
= 0; c
< vd
->vdev_children
; c
++)
3198 if ((mvd
= vdev_lookup_by_path(vd
->vdev_child
[c
], path
)) !=
3206 spa_num_top_vdevs(spa_t
*spa
)
3208 vdev_t
*rvd
= spa
->spa_root_vdev
;
3209 ASSERT3U(spa_config_held(spa
, SCL_VDEV
, RW_READER
), ==, SCL_VDEV
);
3210 return (rvd
->vdev_children
);
3214 * Verify that vdev_add() works as expected.
3217 ztest_vdev_add_remove(ztest_ds_t
*zd
, uint64_t id
)
3219 (void) zd
, (void) id
;
3220 ztest_shared_t
*zs
= ztest_shared
;
3221 spa_t
*spa
= ztest_spa
;
3227 if (ztest_opts
.zo_mmp_test
)
3230 mutex_enter(&ztest_vdev_lock
);
3231 leaves
= MAX(zs
->zs_mirrors
+ zs
->zs_splits
, 1) *
3232 ztest_opts
.zo_raid_children
;
3234 spa_config_enter(spa
, SCL_VDEV
, FTAG
, RW_READER
);
3236 ztest_shared
->zs_vdev_next_leaf
= spa_num_top_vdevs(spa
) * leaves
;
3239 * If we have slogs then remove them 1/4 of the time.
3241 if (spa_has_slogs(spa
) && ztest_random(4) == 0) {
3242 metaslab_group_t
*mg
;
3245 * find the first real slog in log allocation class
3247 mg
= spa_log_class(spa
)->mc_allocator
[0].mca_rotor
;
3248 while (!mg
->mg_vd
->vdev_islog
)
3251 guid
= mg
->mg_vd
->vdev_guid
;
3253 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
3256 * We have to grab the zs_name_lock as writer to
3257 * prevent a race between removing a slog (dmu_objset_find)
3258 * and destroying a dataset. Removing the slog will
3259 * grab a reference on the dataset which may cause
3260 * dsl_destroy_head() to fail with EBUSY thus
3261 * leaving the dataset in an inconsistent state.
3263 pthread_rwlock_wrlock(&ztest_name_lock
);
3264 error
= spa_vdev_remove(spa
, guid
, B_FALSE
);
3265 pthread_rwlock_unlock(&ztest_name_lock
);
3269 case EEXIST
: /* Generic zil_reset() error */
3270 case EBUSY
: /* Replay required */
3271 case EACCES
: /* Crypto key not loaded */
3272 case ZFS_ERR_CHECKPOINT_EXISTS
:
3273 case ZFS_ERR_DISCARDING_CHECKPOINT
:
3276 fatal(B_FALSE
, "spa_vdev_remove() = %d", error
);
3279 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
3282 * Make 1/4 of the devices be log devices
3284 nvroot
= make_vdev_root(NULL
, NULL
, NULL
,
3285 ztest_opts
.zo_vdev_size
, 0, (ztest_random(4) == 0) ?
3286 "log" : NULL
, ztest_opts
.zo_raid_children
, zs
->zs_mirrors
,
3289 error
= spa_vdev_add(spa
, nvroot
);
3290 fnvlist_free(nvroot
);
3296 ztest_record_enospc("spa_vdev_add");
3299 fatal(B_FALSE
, "spa_vdev_add() = %d", error
);
3303 mutex_exit(&ztest_vdev_lock
);
3307 ztest_vdev_class_add(ztest_ds_t
*zd
, uint64_t id
)
3309 (void) zd
, (void) id
;
3310 ztest_shared_t
*zs
= ztest_shared
;
3311 spa_t
*spa
= ztest_spa
;
3314 const char *class = (ztest_random(2) == 0) ?
3315 VDEV_ALLOC_BIAS_SPECIAL
: VDEV_ALLOC_BIAS_DEDUP
;
3319 * By default add a special vdev 50% of the time
3321 if ((ztest_opts
.zo_special_vdevs
== ZTEST_VDEV_CLASS_OFF
) ||
3322 (ztest_opts
.zo_special_vdevs
== ZTEST_VDEV_CLASS_RND
&&
3323 ztest_random(2) == 0)) {
3327 mutex_enter(&ztest_vdev_lock
);
3329 /* Only test with mirrors */
3330 if (zs
->zs_mirrors
< 2) {
3331 mutex_exit(&ztest_vdev_lock
);
3335 /* requires feature@allocation_classes */
3336 if (!spa_feature_is_enabled(spa
, SPA_FEATURE_ALLOCATION_CLASSES
)) {
3337 mutex_exit(&ztest_vdev_lock
);
3341 leaves
= MAX(zs
->zs_mirrors
+ zs
->zs_splits
, 1) *
3342 ztest_opts
.zo_raid_children
;
3344 spa_config_enter(spa
, SCL_VDEV
, FTAG
, RW_READER
);
3345 ztest_shared
->zs_vdev_next_leaf
= spa_num_top_vdevs(spa
) * leaves
;
3346 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
3348 nvroot
= make_vdev_root(NULL
, NULL
, NULL
, ztest_opts
.zo_vdev_size
, 0,
3349 class, ztest_opts
.zo_raid_children
, zs
->zs_mirrors
, 1);
3351 error
= spa_vdev_add(spa
, nvroot
);
3352 fnvlist_free(nvroot
);
3354 if (error
== ENOSPC
)
3355 ztest_record_enospc("spa_vdev_add");
3356 else if (error
!= 0)
3357 fatal(B_FALSE
, "spa_vdev_add() = %d", error
);
3360 * 50% of the time allow small blocks in the special class
3363 spa_special_class(spa
)->mc_groups
== 1 && ztest_random(2) == 0) {
3364 if (ztest_opts
.zo_verbose
>= 3)
3365 (void) printf("Enabling special VDEV small blocks\n");
3366 (void) ztest_dsl_prop_set_uint64(zd
->zd_name
,
3367 ZFS_PROP_SPECIAL_SMALL_BLOCKS
, 32768, B_FALSE
);
3370 mutex_exit(&ztest_vdev_lock
);
3372 if (ztest_opts
.zo_verbose
>= 3) {
3373 metaslab_class_t
*mc
;
3375 if (strcmp(class, VDEV_ALLOC_BIAS_SPECIAL
) == 0)
3376 mc
= spa_special_class(spa
);
3378 mc
= spa_dedup_class(spa
);
3379 (void) printf("Added a %s mirrored vdev (of %d)\n",
3380 class, (int)mc
->mc_groups
);
3385 * Verify that adding/removing aux devices (l2arc, hot spare) works as expected.
3388 ztest_vdev_aux_add_remove(ztest_ds_t
*zd
, uint64_t id
)
3390 (void) zd
, (void) id
;
3391 ztest_shared_t
*zs
= ztest_shared
;
3392 spa_t
*spa
= ztest_spa
;
3393 vdev_t
*rvd
= spa
->spa_root_vdev
;
3394 spa_aux_vdev_t
*sav
;
3398 int error
, ignore_err
= 0;
3400 if (ztest_opts
.zo_mmp_test
)
3403 path
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
3405 if (ztest_random(2) == 0) {
3406 sav
= &spa
->spa_spares
;
3407 aux
= ZPOOL_CONFIG_SPARES
;
3409 sav
= &spa
->spa_l2cache
;
3410 aux
= ZPOOL_CONFIG_L2CACHE
;
3413 mutex_enter(&ztest_vdev_lock
);
3415 spa_config_enter(spa
, SCL_VDEV
, FTAG
, RW_READER
);
3417 if (sav
->sav_count
!= 0 && ztest_random(4) == 0) {
3419 * Pick a random device to remove.
3421 vdev_t
*svd
= sav
->sav_vdevs
[ztest_random(sav
->sav_count
)];
3423 /* dRAID spares cannot be removed; try anyways to see ENOTSUP */
3424 if (strstr(svd
->vdev_path
, VDEV_TYPE_DRAID
) != NULL
)
3425 ignore_err
= ENOTSUP
;
3427 guid
= svd
->vdev_guid
;
3430 * Find an unused device we can add.
3432 zs
->zs_vdev_aux
= 0;
3435 (void) snprintf(path
, MAXPATHLEN
, ztest_aux_template
,
3436 ztest_opts
.zo_dir
, ztest_opts
.zo_pool
, aux
,
3438 for (c
= 0; c
< sav
->sav_count
; c
++)
3439 if (strcmp(sav
->sav_vdevs
[c
]->vdev_path
,
3442 if (c
== sav
->sav_count
&&
3443 vdev_lookup_by_path(rvd
, path
) == NULL
)
3449 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
3455 nvlist_t
*nvroot
= make_vdev_root(NULL
, aux
, NULL
,
3456 (ztest_opts
.zo_vdev_size
* 5) / 4, 0, NULL
, 0, 0, 1);
3457 error
= spa_vdev_add(spa
, nvroot
);
3463 fatal(B_FALSE
, "spa_vdev_add(%p) = %d", nvroot
, error
);
3465 fnvlist_free(nvroot
);
3468 * Remove an existing device. Sometimes, dirty its
3469 * vdev state first to make sure we handle removal
3470 * of devices that have pending state changes.
3472 if (ztest_random(2) == 0)
3473 (void) vdev_online(spa
, guid
, 0, NULL
);
3475 error
= spa_vdev_remove(spa
, guid
, B_FALSE
);
3480 case ZFS_ERR_CHECKPOINT_EXISTS
:
3481 case ZFS_ERR_DISCARDING_CHECKPOINT
:
3484 if (error
!= ignore_err
)
3486 "spa_vdev_remove(%"PRIu64
") = %d",
3491 mutex_exit(&ztest_vdev_lock
);
3493 umem_free(path
, MAXPATHLEN
);
3497 * split a pool if it has mirror tlvdevs
3500 ztest_split_pool(ztest_ds_t
*zd
, uint64_t id
)
3502 (void) zd
, (void) id
;
3503 ztest_shared_t
*zs
= ztest_shared
;
3504 spa_t
*spa
= ztest_spa
;
3505 vdev_t
*rvd
= spa
->spa_root_vdev
;
3506 nvlist_t
*tree
, **child
, *config
, *split
, **schild
;
3507 uint_t c
, children
, schildren
= 0, lastlogid
= 0;
3510 if (ztest_opts
.zo_mmp_test
)
3513 mutex_enter(&ztest_vdev_lock
);
3515 /* ensure we have a usable config; mirrors of raidz aren't supported */
3516 if (zs
->zs_mirrors
< 3 || ztest_opts
.zo_raid_children
> 1) {
3517 mutex_exit(&ztest_vdev_lock
);
3521 /* clean up the old pool, if any */
3522 (void) spa_destroy("splitp");
3524 spa_config_enter(spa
, SCL_VDEV
, FTAG
, RW_READER
);
3526 /* generate a config from the existing config */
3527 mutex_enter(&spa
->spa_props_lock
);
3528 tree
= fnvlist_lookup_nvlist(spa
->spa_config
, ZPOOL_CONFIG_VDEV_TREE
);
3529 mutex_exit(&spa
->spa_props_lock
);
3531 VERIFY0(nvlist_lookup_nvlist_array(tree
, ZPOOL_CONFIG_CHILDREN
,
3532 &child
, &children
));
3534 schild
= malloc(rvd
->vdev_children
* sizeof (nvlist_t
*));
3535 for (c
= 0; c
< children
; c
++) {
3536 vdev_t
*tvd
= rvd
->vdev_child
[c
];
3540 if (tvd
->vdev_islog
|| tvd
->vdev_ops
== &vdev_hole_ops
) {
3541 schild
[schildren
] = fnvlist_alloc();
3542 fnvlist_add_string(schild
[schildren
],
3543 ZPOOL_CONFIG_TYPE
, VDEV_TYPE_HOLE
);
3544 fnvlist_add_uint64(schild
[schildren
],
3545 ZPOOL_CONFIG_IS_HOLE
, 1);
3547 lastlogid
= schildren
;
3552 VERIFY0(nvlist_lookup_nvlist_array(child
[c
],
3553 ZPOOL_CONFIG_CHILDREN
, &mchild
, &mchildren
));
3554 schild
[schildren
++] = fnvlist_dup(mchild
[0]);
3557 /* OK, create a config that can be used to split */
3558 split
= fnvlist_alloc();
3559 fnvlist_add_string(split
, ZPOOL_CONFIG_TYPE
, VDEV_TYPE_ROOT
);
3560 fnvlist_add_nvlist_array(split
, ZPOOL_CONFIG_CHILDREN
,
3561 (const nvlist_t
**)schild
, lastlogid
!= 0 ? lastlogid
: schildren
);
3563 config
= fnvlist_alloc();
3564 fnvlist_add_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, split
);
3566 for (c
= 0; c
< schildren
; c
++)
3567 fnvlist_free(schild
[c
]);
3569 fnvlist_free(split
);
3571 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
3573 (void) pthread_rwlock_wrlock(&ztest_name_lock
);
3574 error
= spa_vdev_split_mirror(spa
, "splitp", config
, NULL
, B_FALSE
);
3575 (void) pthread_rwlock_unlock(&ztest_name_lock
);
3577 fnvlist_free(config
);
3580 (void) printf("successful split - results:\n");
3581 mutex_enter(&spa_namespace_lock
);
3582 show_pool_stats(spa
);
3583 show_pool_stats(spa_lookup("splitp"));
3584 mutex_exit(&spa_namespace_lock
);
3588 mutex_exit(&ztest_vdev_lock
);
3592 * Verify that we can attach and detach devices.
3595 ztest_vdev_attach_detach(ztest_ds_t
*zd
, uint64_t id
)
3597 (void) zd
, (void) id
;
3598 ztest_shared_t
*zs
= ztest_shared
;
3599 spa_t
*spa
= ztest_spa
;
3600 spa_aux_vdev_t
*sav
= &spa
->spa_spares
;
3601 vdev_t
*rvd
= spa
->spa_root_vdev
;
3602 vdev_t
*oldvd
, *newvd
, *pvd
;
3606 uint64_t ashift
= ztest_get_ashift();
3607 uint64_t oldguid
, pguid
;
3608 uint64_t oldsize
, newsize
;
3609 char *oldpath
, *newpath
;
3611 int oldvd_has_siblings
= B_FALSE
;
3612 int newvd_is_spare
= B_FALSE
;
3613 int newvd_is_dspare
= B_FALSE
;
3615 int error
, expected_error
;
3617 if (ztest_opts
.zo_mmp_test
)
3620 oldpath
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
3621 newpath
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
3623 mutex_enter(&ztest_vdev_lock
);
3624 leaves
= MAX(zs
->zs_mirrors
, 1) * ztest_opts
.zo_raid_children
;
3626 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
3629 * If a vdev is in the process of being removed, its removal may
3630 * finish while we are in progress, leading to an unexpected error
3631 * value. Don't bother trying to attach while we are in the middle
3634 if (ztest_device_removal_active
) {
3635 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3640 * Decide whether to do an attach or a replace.
3642 replacing
= ztest_random(2);
3645 * Pick a random top-level vdev.
3647 top
= ztest_random_vdev_top(spa
, B_TRUE
);
3650 * Pick a random leaf within it.
3652 leaf
= ztest_random(leaves
);
3657 oldvd
= rvd
->vdev_child
[top
];
3659 /* pick a child from the mirror */
3660 if (zs
->zs_mirrors
>= 1) {
3661 ASSERT3P(oldvd
->vdev_ops
, ==, &vdev_mirror_ops
);
3662 ASSERT3U(oldvd
->vdev_children
, >=, zs
->zs_mirrors
);
3663 oldvd
= oldvd
->vdev_child
[leaf
/ ztest_opts
.zo_raid_children
];
3666 /* pick a child out of the raidz group */
3667 if (ztest_opts
.zo_raid_children
> 1) {
3668 if (strcmp(oldvd
->vdev_ops
->vdev_op_type
, "raidz") == 0)
3669 ASSERT3P(oldvd
->vdev_ops
, ==, &vdev_raidz_ops
);
3671 ASSERT3P(oldvd
->vdev_ops
, ==, &vdev_draid_ops
);
3672 ASSERT3U(oldvd
->vdev_children
, ==, ztest_opts
.zo_raid_children
);
3673 oldvd
= oldvd
->vdev_child
[leaf
% ztest_opts
.zo_raid_children
];
3677 * If we're already doing an attach or replace, oldvd may be a
3678 * mirror vdev -- in which case, pick a random child.
3680 while (oldvd
->vdev_children
!= 0) {
3681 oldvd_has_siblings
= B_TRUE
;
3682 ASSERT3U(oldvd
->vdev_children
, >=, 2);
3683 oldvd
= oldvd
->vdev_child
[ztest_random(oldvd
->vdev_children
)];
3686 oldguid
= oldvd
->vdev_guid
;
3687 oldsize
= vdev_get_min_asize(oldvd
);
3688 oldvd_is_log
= oldvd
->vdev_top
->vdev_islog
;
3689 (void) strcpy(oldpath
, oldvd
->vdev_path
);
3690 pvd
= oldvd
->vdev_parent
;
3691 pguid
= pvd
->vdev_guid
;
3694 * If oldvd has siblings, then half of the time, detach it. Prior
3695 * to the detach the pool is scrubbed in order to prevent creating
3696 * unrepairable blocks as a result of the data corruption injection.
3698 if (oldvd_has_siblings
&& ztest_random(2) == 0) {
3699 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3701 error
= ztest_scrub_impl(spa
);
3705 error
= spa_vdev_detach(spa
, oldguid
, pguid
, B_FALSE
);
3706 if (error
!= 0 && error
!= ENODEV
&& error
!= EBUSY
&&
3707 error
!= ENOTSUP
&& error
!= ZFS_ERR_CHECKPOINT_EXISTS
&&
3708 error
!= ZFS_ERR_DISCARDING_CHECKPOINT
)
3709 fatal(B_FALSE
, "detach (%s) returned %d",
3715 * For the new vdev, choose with equal probability between the two
3716 * standard paths (ending in either 'a' or 'b') or a random hot spare.
3718 if (sav
->sav_count
!= 0 && ztest_random(3) == 0) {
3719 newvd
= sav
->sav_vdevs
[ztest_random(sav
->sav_count
)];
3720 newvd_is_spare
= B_TRUE
;
3722 if (newvd
->vdev_ops
== &vdev_draid_spare_ops
)
3723 newvd_is_dspare
= B_TRUE
;
3725 (void) strcpy(newpath
, newvd
->vdev_path
);
3727 (void) snprintf(newpath
, MAXPATHLEN
, ztest_dev_template
,
3728 ztest_opts
.zo_dir
, ztest_opts
.zo_pool
,
3729 top
* leaves
+ leaf
);
3730 if (ztest_random(2) == 0)
3731 newpath
[strlen(newpath
) - 1] = 'b';
3732 newvd
= vdev_lookup_by_path(rvd
, newpath
);
3737 * Reopen to ensure the vdev's asize field isn't stale.
3740 newsize
= vdev_get_min_asize(newvd
);
3743 * Make newsize a little bigger or smaller than oldsize.
3744 * If it's smaller, the attach should fail.
3745 * If it's larger, and we're doing a replace,
3746 * we should get dynamic LUN growth when we're done.
3748 newsize
= 10 * oldsize
/ (9 + ztest_random(3));
3752 * If pvd is not a mirror or root, the attach should fail with ENOTSUP,
3753 * unless it's a replace; in that case any non-replacing parent is OK.
3755 * If newvd is already part of the pool, it should fail with EBUSY.
3757 * If newvd is too small, it should fail with EOVERFLOW.
3759 * If newvd is a distributed spare and it's being attached to a
3760 * dRAID which is not its parent it should fail with EINVAL.
3762 if (pvd
->vdev_ops
!= &vdev_mirror_ops
&&
3763 pvd
->vdev_ops
!= &vdev_root_ops
&& (!replacing
||
3764 pvd
->vdev_ops
== &vdev_replacing_ops
||
3765 pvd
->vdev_ops
== &vdev_spare_ops
))
3766 expected_error
= ENOTSUP
;
3767 else if (newvd_is_spare
&& (!replacing
|| oldvd_is_log
))
3768 expected_error
= ENOTSUP
;
3769 else if (newvd
== oldvd
)
3770 expected_error
= replacing
? 0 : EBUSY
;
3771 else if (vdev_lookup_by_path(rvd
, newpath
) != NULL
)
3772 expected_error
= EBUSY
;
3773 else if (!newvd_is_dspare
&& newsize
< oldsize
)
3774 expected_error
= EOVERFLOW
;
3775 else if (ashift
> oldvd
->vdev_top
->vdev_ashift
)
3776 expected_error
= EDOM
;
3777 else if (newvd_is_dspare
&& pvd
!= vdev_draid_spare_get_parent(newvd
))
3778 expected_error
= ENOTSUP
;
3782 spa_config_exit(spa
, SCL_ALL
, FTAG
);
3785 * Build the nvlist describing newpath.
3787 root
= make_vdev_root(newpath
, NULL
, NULL
, newvd
== NULL
? newsize
: 0,
3788 ashift
, NULL
, 0, 0, 1);
3791 * When supported select either a healing or sequential resilver.
3793 boolean_t rebuilding
= B_FALSE
;
3794 if (pvd
->vdev_ops
== &vdev_mirror_ops
||
3795 pvd
->vdev_ops
== &vdev_root_ops
) {
3796 rebuilding
= !!ztest_random(2);
3799 error
= spa_vdev_attach(spa
, oldguid
, root
, replacing
, rebuilding
);
3804 * If our parent was the replacing vdev, but the replace completed,
3805 * then instead of failing with ENOTSUP we may either succeed,
3806 * fail with ENODEV, or fail with EOVERFLOW.
3808 if (expected_error
== ENOTSUP
&&
3809 (error
== 0 || error
== ENODEV
|| error
== EOVERFLOW
))
3810 expected_error
= error
;
3813 * If someone grew the LUN, the replacement may be too small.
3815 if (error
== EOVERFLOW
|| error
== EBUSY
)
3816 expected_error
= error
;
3818 if (error
== ZFS_ERR_CHECKPOINT_EXISTS
||
3819 error
== ZFS_ERR_DISCARDING_CHECKPOINT
||
3820 error
== ZFS_ERR_RESILVER_IN_PROGRESS
||
3821 error
== ZFS_ERR_REBUILD_IN_PROGRESS
)
3822 expected_error
= error
;
3824 if (error
!= expected_error
&& expected_error
!= EBUSY
) {
3825 fatal(B_FALSE
, "attach (%s %"PRIu64
", %s %"PRIu64
", %d) "
3826 "returned %d, expected %d",
3827 oldpath
, oldsize
, newpath
,
3828 newsize
, replacing
, error
, expected_error
);
3831 mutex_exit(&ztest_vdev_lock
);
3833 umem_free(oldpath
, MAXPATHLEN
);
3834 umem_free(newpath
, MAXPATHLEN
);
3838 ztest_device_removal(ztest_ds_t
*zd
, uint64_t id
)
3840 (void) zd
, (void) id
;
3841 spa_t
*spa
= ztest_spa
;
3846 mutex_enter(&ztest_vdev_lock
);
3848 if (ztest_device_removal_active
) {
3849 mutex_exit(&ztest_vdev_lock
);
3854 * Remove a random top-level vdev and wait for removal to finish.
3856 spa_config_enter(spa
, SCL_VDEV
, FTAG
, RW_READER
);
3857 vd
= vdev_lookup_top(spa
, ztest_random_vdev_top(spa
, B_FALSE
));
3858 guid
= vd
->vdev_guid
;
3859 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
3861 error
= spa_vdev_remove(spa
, guid
, B_FALSE
);
3863 ztest_device_removal_active
= B_TRUE
;
3864 mutex_exit(&ztest_vdev_lock
);
3867 * spa->spa_vdev_removal is created in a sync task that
3868 * is initiated via dsl_sync_task_nowait(). Since the
3869 * task may not run before spa_vdev_remove() returns, we
3870 * must wait at least 1 txg to ensure that the removal
3871 * struct has been created.
3873 txg_wait_synced(spa_get_dsl(spa
), 0);
3875 while (spa
->spa_removing_phys
.sr_state
== DSS_SCANNING
)
3876 txg_wait_synced(spa_get_dsl(spa
), 0);
3878 mutex_exit(&ztest_vdev_lock
);
3883 * The pool needs to be scrubbed after completing device removal.
3884 * Failure to do so may result in checksum errors due to the
3885 * strategy employed by ztest_fault_inject() when selecting which
3886 * offset are redundant and can be damaged.
3888 error
= spa_scan(spa
, POOL_SCAN_SCRUB
);
3890 while (dsl_scan_scrubbing(spa_get_dsl(spa
)))
3891 txg_wait_synced(spa_get_dsl(spa
), 0);
3894 mutex_enter(&ztest_vdev_lock
);
3895 ztest_device_removal_active
= B_FALSE
;
3896 mutex_exit(&ztest_vdev_lock
);
3900 * Callback function which expands the physical size of the vdev.
3903 grow_vdev(vdev_t
*vd
, void *arg
)
3905 spa_t
*spa __maybe_unused
= vd
->vdev_spa
;
3906 size_t *newsize
= arg
;
3910 ASSERT3S(spa_config_held(spa
, SCL_STATE
, RW_READER
), ==, SCL_STATE
);
3911 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
3913 if ((fd
= open(vd
->vdev_path
, O_RDWR
)) == -1)
3916 fsize
= lseek(fd
, 0, SEEK_END
);
3917 VERIFY0(ftruncate(fd
, *newsize
));
3919 if (ztest_opts
.zo_verbose
>= 6) {
3920 (void) printf("%s grew from %lu to %lu bytes\n",
3921 vd
->vdev_path
, (ulong_t
)fsize
, (ulong_t
)*newsize
);
3928 * Callback function which expands a given vdev by calling vdev_online().
3931 online_vdev(vdev_t
*vd
, void *arg
)
3934 spa_t
*spa
= vd
->vdev_spa
;
3935 vdev_t
*tvd
= vd
->vdev_top
;
3936 uint64_t guid
= vd
->vdev_guid
;
3937 uint64_t generation
= spa
->spa_config_generation
+ 1;
3938 vdev_state_t newstate
= VDEV_STATE_UNKNOWN
;
3941 ASSERT3S(spa_config_held(spa
, SCL_STATE
, RW_READER
), ==, SCL_STATE
);
3942 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
3944 /* Calling vdev_online will initialize the new metaslabs */
3945 spa_config_exit(spa
, SCL_STATE
, spa
);
3946 error
= vdev_online(spa
, guid
, ZFS_ONLINE_EXPAND
, &newstate
);
3947 spa_config_enter(spa
, SCL_STATE
, spa
, RW_READER
);
3950 * If vdev_online returned an error or the underlying vdev_open
3951 * failed then we abort the expand. The only way to know that
3952 * vdev_open fails is by checking the returned newstate.
3954 if (error
|| newstate
!= VDEV_STATE_HEALTHY
) {
3955 if (ztest_opts
.zo_verbose
>= 5) {
3956 (void) printf("Unable to expand vdev, state %u, "
3957 "error %d\n", newstate
, error
);
3961 ASSERT3U(newstate
, ==, VDEV_STATE_HEALTHY
);
3964 * Since we dropped the lock we need to ensure that we're
3965 * still talking to the original vdev. It's possible this
3966 * vdev may have been detached/replaced while we were
3967 * trying to online it.
3969 if (generation
!= spa
->spa_config_generation
) {
3970 if (ztest_opts
.zo_verbose
>= 5) {
3971 (void) printf("vdev configuration has changed, "
3972 "guid %"PRIu64
", state %"PRIu64
", "
3973 "expected gen %"PRIu64
", got gen %"PRIu64
"\n",
3977 spa
->spa_config_generation
);
3985 * Traverse the vdev tree calling the supplied function.
3986 * We continue to walk the tree until we either have walked all
3987 * children or we receive a non-NULL return from the callback.
3988 * If a NULL callback is passed, then we just return back the first
3989 * leaf vdev we encounter.
3992 vdev_walk_tree(vdev_t
*vd
, vdev_t
*(*func
)(vdev_t
*, void *), void *arg
)
3996 if (vd
->vdev_ops
->vdev_op_leaf
) {
4000 return (func(vd
, arg
));
4003 for (c
= 0; c
< vd
->vdev_children
; c
++) {
4004 vdev_t
*cvd
= vd
->vdev_child
[c
];
4005 if ((cvd
= vdev_walk_tree(cvd
, func
, arg
)) != NULL
)
4012 * Verify that dynamic LUN growth works as expected.
4015 ztest_vdev_LUN_growth(ztest_ds_t
*zd
, uint64_t id
)
4017 (void) zd
, (void) id
;
4018 spa_t
*spa
= ztest_spa
;
4020 metaslab_class_t
*mc
;
4021 metaslab_group_t
*mg
;
4022 size_t psize
, newsize
;
4024 uint64_t old_class_space
, new_class_space
, old_ms_count
, new_ms_count
;
4026 mutex_enter(&ztest_checkpoint_lock
);
4027 mutex_enter(&ztest_vdev_lock
);
4028 spa_config_enter(spa
, SCL_STATE
, spa
, RW_READER
);
4031 * If there is a vdev removal in progress, it could complete while
4032 * we are running, in which case we would not be able to verify
4033 * that the metaslab_class space increased (because it decreases
4034 * when the device removal completes).
4036 if (ztest_device_removal_active
) {
4037 spa_config_exit(spa
, SCL_STATE
, spa
);
4038 mutex_exit(&ztest_vdev_lock
);
4039 mutex_exit(&ztest_checkpoint_lock
);
4043 top
= ztest_random_vdev_top(spa
, B_TRUE
);
4045 tvd
= spa
->spa_root_vdev
->vdev_child
[top
];
4048 old_ms_count
= tvd
->vdev_ms_count
;
4049 old_class_space
= metaslab_class_get_space(mc
);
4052 * Determine the size of the first leaf vdev associated with
4053 * our top-level device.
4055 vd
= vdev_walk_tree(tvd
, NULL
, NULL
);
4056 ASSERT3P(vd
, !=, NULL
);
4057 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
4059 psize
= vd
->vdev_psize
;
4062 * We only try to expand the vdev if it's healthy, less than 4x its
4063 * original size, and it has a valid psize.
4065 if (tvd
->vdev_state
!= VDEV_STATE_HEALTHY
||
4066 psize
== 0 || psize
>= 4 * ztest_opts
.zo_vdev_size
) {
4067 spa_config_exit(spa
, SCL_STATE
, spa
);
4068 mutex_exit(&ztest_vdev_lock
);
4069 mutex_exit(&ztest_checkpoint_lock
);
4072 ASSERT3U(psize
, >, 0);
4073 newsize
= psize
+ MAX(psize
/ 8, SPA_MAXBLOCKSIZE
);
4074 ASSERT3U(newsize
, >, psize
);
4076 if (ztest_opts
.zo_verbose
>= 6) {
4077 (void) printf("Expanding LUN %s from %lu to %lu\n",
4078 vd
->vdev_path
, (ulong_t
)psize
, (ulong_t
)newsize
);
4082 * Growing the vdev is a two step process:
4083 * 1). expand the physical size (i.e. relabel)
4084 * 2). online the vdev to create the new metaslabs
4086 if (vdev_walk_tree(tvd
, grow_vdev
, &newsize
) != NULL
||
4087 vdev_walk_tree(tvd
, online_vdev
, NULL
) != NULL
||
4088 tvd
->vdev_state
!= VDEV_STATE_HEALTHY
) {
4089 if (ztest_opts
.zo_verbose
>= 5) {
4090 (void) printf("Could not expand LUN because "
4091 "the vdev configuration changed.\n");
4093 spa_config_exit(spa
, SCL_STATE
, spa
);
4094 mutex_exit(&ztest_vdev_lock
);
4095 mutex_exit(&ztest_checkpoint_lock
);
4099 spa_config_exit(spa
, SCL_STATE
, spa
);
4102 * Expanding the LUN will update the config asynchronously,
4103 * thus we must wait for the async thread to complete any
4104 * pending tasks before proceeding.
4108 mutex_enter(&spa
->spa_async_lock
);
4109 done
= (spa
->spa_async_thread
== NULL
&& !spa
->spa_async_tasks
);
4110 mutex_exit(&spa
->spa_async_lock
);
4113 txg_wait_synced(spa_get_dsl(spa
), 0);
4114 (void) poll(NULL
, 0, 100);
4117 spa_config_enter(spa
, SCL_STATE
, spa
, RW_READER
);
4119 tvd
= spa
->spa_root_vdev
->vdev_child
[top
];
4120 new_ms_count
= tvd
->vdev_ms_count
;
4121 new_class_space
= metaslab_class_get_space(mc
);
4123 if (tvd
->vdev_mg
!= mg
|| mg
->mg_class
!= mc
) {
4124 if (ztest_opts
.zo_verbose
>= 5) {
4125 (void) printf("Could not verify LUN expansion due to "
4126 "intervening vdev offline or remove.\n");
4128 spa_config_exit(spa
, SCL_STATE
, spa
);
4129 mutex_exit(&ztest_vdev_lock
);
4130 mutex_exit(&ztest_checkpoint_lock
);
4135 * Make sure we were able to grow the vdev.
4137 if (new_ms_count
<= old_ms_count
) {
4139 "LUN expansion failed: ms_count %"PRIu64
" < %"PRIu64
"\n",
4140 old_ms_count
, new_ms_count
);
4144 * Make sure we were able to grow the pool.
4146 if (new_class_space
<= old_class_space
) {
4148 "LUN expansion failed: class_space %"PRIu64
" < %"PRIu64
"\n",
4149 old_class_space
, new_class_space
);
4152 if (ztest_opts
.zo_verbose
>= 5) {
4153 char oldnumbuf
[NN_NUMBUF_SZ
], newnumbuf
[NN_NUMBUF_SZ
];
4155 nicenum(old_class_space
, oldnumbuf
, sizeof (oldnumbuf
));
4156 nicenum(new_class_space
, newnumbuf
, sizeof (newnumbuf
));
4157 (void) printf("%s grew from %s to %s\n",
4158 spa
->spa_name
, oldnumbuf
, newnumbuf
);
4161 spa_config_exit(spa
, SCL_STATE
, spa
);
4162 mutex_exit(&ztest_vdev_lock
);
4163 mutex_exit(&ztest_checkpoint_lock
);
4167 * Verify that dmu_objset_{create,destroy,open,close} work as expected.
4170 ztest_objset_create_cb(objset_t
*os
, void *arg
, cred_t
*cr
, dmu_tx_t
*tx
)
4172 (void) arg
, (void) cr
;
4175 * Create the objects common to all ztest datasets.
4177 VERIFY0(zap_create_claim(os
, ZTEST_DIROBJ
,
4178 DMU_OT_ZAP_OTHER
, DMU_OT_NONE
, 0, tx
));
4182 ztest_dataset_create(char *dsname
)
4186 dsl_crypto_params_t
*dcp
= NULL
;
4189 * 50% of the time, we create encrypted datasets
4190 * using a random cipher suite and a hard-coded
4193 rand
= ztest_random(2);
4195 nvlist_t
*crypto_args
= fnvlist_alloc();
4196 nvlist_t
*props
= fnvlist_alloc();
4198 /* slight bias towards the default cipher suite */
4199 rand
= ztest_random(ZIO_CRYPT_FUNCTIONS
);
4200 if (rand
< ZIO_CRYPT_AES_128_CCM
)
4201 rand
= ZIO_CRYPT_ON
;
4203 fnvlist_add_uint64(props
,
4204 zfs_prop_to_name(ZFS_PROP_ENCRYPTION
), rand
);
4205 fnvlist_add_uint8_array(crypto_args
, "wkeydata",
4206 (uint8_t *)ztest_wkeydata
, WRAPPING_KEY_LEN
);
4209 * These parameters aren't really used by the kernel. They
4210 * are simply stored so that userspace knows how to load
4213 fnvlist_add_uint64(props
,
4214 zfs_prop_to_name(ZFS_PROP_KEYFORMAT
), ZFS_KEYFORMAT_RAW
);
4215 fnvlist_add_string(props
,
4216 zfs_prop_to_name(ZFS_PROP_KEYLOCATION
), "prompt");
4217 fnvlist_add_uint64(props
,
4218 zfs_prop_to_name(ZFS_PROP_PBKDF2_SALT
), 0ULL);
4219 fnvlist_add_uint64(props
,
4220 zfs_prop_to_name(ZFS_PROP_PBKDF2_ITERS
), 0ULL);
4222 VERIFY0(dsl_crypto_params_create_nvlist(DCP_CMD_NONE
, props
,
4223 crypto_args
, &dcp
));
4226 * Cycle through all available encryption implementations
4227 * to verify interoperability.
4229 VERIFY0(gcm_impl_set("cycle"));
4230 VERIFY0(aes_impl_set("cycle"));
4232 fnvlist_free(crypto_args
);
4233 fnvlist_free(props
);
4236 err
= dmu_objset_create(dsname
, DMU_OST_OTHER
, 0, dcp
,
4237 ztest_objset_create_cb
, NULL
);
4238 dsl_crypto_params_free(dcp
, !!err
);
4240 rand
= ztest_random(100);
4241 if (err
|| rand
< 80)
4244 if (ztest_opts
.zo_verbose
>= 5)
4245 (void) printf("Setting dataset %s to sync always\n", dsname
);
4246 return (ztest_dsl_prop_set_uint64(dsname
, ZFS_PROP_SYNC
,
4247 ZFS_SYNC_ALWAYS
, B_FALSE
));
4251 ztest_objset_destroy_cb(const char *name
, void *arg
)
4255 dmu_object_info_t doi
;
4259 * Verify that the dataset contains a directory object.
4261 VERIFY0(ztest_dmu_objset_own(name
, DMU_OST_OTHER
, B_TRUE
,
4262 B_TRUE
, FTAG
, &os
));
4263 error
= dmu_object_info(os
, ZTEST_DIROBJ
, &doi
);
4264 if (error
!= ENOENT
) {
4265 /* We could have crashed in the middle of destroying it */
4267 ASSERT3U(doi
.doi_type
, ==, DMU_OT_ZAP_OTHER
);
4268 ASSERT3S(doi
.doi_physical_blocks_512
, >=, 0);
4270 dmu_objset_disown(os
, B_TRUE
, FTAG
);
4273 * Destroy the dataset.
4275 if (strchr(name
, '@') != NULL
) {
4276 error
= dsl_destroy_snapshot(name
, B_TRUE
);
4277 if (error
!= ECHRNG
) {
4279 * The program was executed, but encountered a runtime
4280 * error, such as insufficient slop, or a hold on the
4286 error
= dsl_destroy_head(name
);
4287 if (error
== ENOSPC
) {
4288 /* There could be checkpoint or insufficient slop */
4289 ztest_record_enospc(FTAG
);
4290 } else if (error
!= EBUSY
) {
4291 /* There could be a hold on this dataset */
4299 ztest_snapshot_create(char *osname
, uint64_t id
)
4301 char snapname
[ZFS_MAX_DATASET_NAME_LEN
];
4304 (void) snprintf(snapname
, sizeof (snapname
), "%"PRIu64
"", id
);
4306 error
= dmu_objset_snapshot_one(osname
, snapname
);
4307 if (error
== ENOSPC
) {
4308 ztest_record_enospc(FTAG
);
4311 if (error
!= 0 && error
!= EEXIST
) {
4312 fatal(B_FALSE
, "ztest_snapshot_create(%s@%s) = %d", osname
,
4319 ztest_snapshot_destroy(char *osname
, uint64_t id
)
4321 char snapname
[ZFS_MAX_DATASET_NAME_LEN
];
4324 (void) snprintf(snapname
, sizeof (snapname
), "%s@%"PRIu64
"",
4327 error
= dsl_destroy_snapshot(snapname
, B_FALSE
);
4328 if (error
!= 0 && error
!= ENOENT
)
4329 fatal(B_FALSE
, "ztest_snapshot_destroy(%s) = %d",
4335 ztest_dmu_objset_create_destroy(ztest_ds_t
*zd
, uint64_t id
)
4342 char name
[ZFS_MAX_DATASET_NAME_LEN
];
4346 zdtmp
= umem_alloc(sizeof (ztest_ds_t
), UMEM_NOFAIL
);
4348 (void) pthread_rwlock_rdlock(&ztest_name_lock
);
4350 (void) snprintf(name
, sizeof (name
), "%s/temp_%"PRIu64
"",
4351 ztest_opts
.zo_pool
, id
);
4354 * If this dataset exists from a previous run, process its replay log
4355 * half of the time. If we don't replay it, then dsl_destroy_head()
4356 * (invoked from ztest_objset_destroy_cb()) should just throw it away.
4358 if (ztest_random(2) == 0 &&
4359 ztest_dmu_objset_own(name
, DMU_OST_OTHER
, B_FALSE
,
4360 B_TRUE
, FTAG
, &os
) == 0) {
4361 ztest_zd_init(zdtmp
, NULL
, os
);
4362 zil_replay(os
, zdtmp
, ztest_replay_vector
);
4363 ztest_zd_fini(zdtmp
);
4364 dmu_objset_disown(os
, B_TRUE
, FTAG
);
4368 * There may be an old instance of the dataset we're about to
4369 * create lying around from a previous run. If so, destroy it
4370 * and all of its snapshots.
4372 (void) dmu_objset_find(name
, ztest_objset_destroy_cb
, NULL
,
4373 DS_FIND_CHILDREN
| DS_FIND_SNAPSHOTS
);
4376 * Verify that the destroyed dataset is no longer in the namespace.
4378 VERIFY3U(ENOENT
, ==, ztest_dmu_objset_own(name
, DMU_OST_OTHER
, B_TRUE
,
4379 B_TRUE
, FTAG
, &os
));
4382 * Verify that we can create a new dataset.
4384 error
= ztest_dataset_create(name
);
4386 if (error
== ENOSPC
) {
4387 ztest_record_enospc(FTAG
);
4390 fatal(B_FALSE
, "dmu_objset_create(%s) = %d", name
, error
);
4393 VERIFY0(ztest_dmu_objset_own(name
, DMU_OST_OTHER
, B_FALSE
, B_TRUE
,
4396 ztest_zd_init(zdtmp
, NULL
, os
);
4399 * Open the intent log for it.
4401 zilog
= zil_open(os
, ztest_get_data
);
4404 * Put some objects in there, do a little I/O to them,
4405 * and randomly take a couple of snapshots along the way.
4407 iters
= ztest_random(5);
4408 for (i
= 0; i
< iters
; i
++) {
4409 ztest_dmu_object_alloc_free(zdtmp
, id
);
4410 if (ztest_random(iters
) == 0)
4411 (void) ztest_snapshot_create(name
, i
);
4415 * Verify that we cannot create an existing dataset.
4417 VERIFY3U(EEXIST
, ==,
4418 dmu_objset_create(name
, DMU_OST_OTHER
, 0, NULL
, NULL
, NULL
));
4421 * Verify that we can hold an objset that is also owned.
4423 VERIFY0(dmu_objset_hold(name
, FTAG
, &os2
));
4424 dmu_objset_rele(os2
, FTAG
);
4427 * Verify that we cannot own an objset that is already owned.
4429 VERIFY3U(EBUSY
, ==, ztest_dmu_objset_own(name
, DMU_OST_OTHER
,
4430 B_FALSE
, B_TRUE
, FTAG
, &os2
));
4433 dmu_objset_disown(os
, B_TRUE
, FTAG
);
4434 ztest_zd_fini(zdtmp
);
4436 (void) pthread_rwlock_unlock(&ztest_name_lock
);
4438 umem_free(zdtmp
, sizeof (ztest_ds_t
));
4442 * Verify that dmu_snapshot_{create,destroy,open,close} work as expected.
4445 ztest_dmu_snapshot_create_destroy(ztest_ds_t
*zd
, uint64_t id
)
4447 (void) pthread_rwlock_rdlock(&ztest_name_lock
);
4448 (void) ztest_snapshot_destroy(zd
->zd_name
, id
);
4449 (void) ztest_snapshot_create(zd
->zd_name
, id
);
4450 (void) pthread_rwlock_unlock(&ztest_name_lock
);
4454 * Cleanup non-standard snapshots and clones.
4457 ztest_dsl_dataset_cleanup(char *osname
, uint64_t id
)
4466 snap1name
= umem_alloc(ZFS_MAX_DATASET_NAME_LEN
, UMEM_NOFAIL
);
4467 clone1name
= umem_alloc(ZFS_MAX_DATASET_NAME_LEN
, UMEM_NOFAIL
);
4468 snap2name
= umem_alloc(ZFS_MAX_DATASET_NAME_LEN
, UMEM_NOFAIL
);
4469 clone2name
= umem_alloc(ZFS_MAX_DATASET_NAME_LEN
, UMEM_NOFAIL
);
4470 snap3name
= umem_alloc(ZFS_MAX_DATASET_NAME_LEN
, UMEM_NOFAIL
);
4472 (void) snprintf(snap1name
, ZFS_MAX_DATASET_NAME_LEN
, "%s@s1_%"PRIu64
"",
4474 (void) snprintf(clone1name
, ZFS_MAX_DATASET_NAME_LEN
, "%s/c1_%"PRIu64
"",
4476 (void) snprintf(snap2name
, ZFS_MAX_DATASET_NAME_LEN
, "%s@s2_%"PRIu64
"",
4478 (void) snprintf(clone2name
, ZFS_MAX_DATASET_NAME_LEN
, "%s/c2_%"PRIu64
"",
4480 (void) snprintf(snap3name
, ZFS_MAX_DATASET_NAME_LEN
, "%s@s3_%"PRIu64
"",
4483 error
= dsl_destroy_head(clone2name
);
4484 if (error
&& error
!= ENOENT
)
4485 fatal(B_FALSE
, "dsl_destroy_head(%s) = %d", clone2name
, error
);
4486 error
= dsl_destroy_snapshot(snap3name
, B_FALSE
);
4487 if (error
&& error
!= ENOENT
)
4488 fatal(B_FALSE
, "dsl_destroy_snapshot(%s) = %d",
4490 error
= dsl_destroy_snapshot(snap2name
, B_FALSE
);
4491 if (error
&& error
!= ENOENT
)
4492 fatal(B_FALSE
, "dsl_destroy_snapshot(%s) = %d",
4494 error
= dsl_destroy_head(clone1name
);
4495 if (error
&& error
!= ENOENT
)
4496 fatal(B_FALSE
, "dsl_destroy_head(%s) = %d", clone1name
, error
);
4497 error
= dsl_destroy_snapshot(snap1name
, B_FALSE
);
4498 if (error
&& error
!= ENOENT
)
4499 fatal(B_FALSE
, "dsl_destroy_snapshot(%s) = %d",
4502 umem_free(snap1name
, ZFS_MAX_DATASET_NAME_LEN
);
4503 umem_free(clone1name
, ZFS_MAX_DATASET_NAME_LEN
);
4504 umem_free(snap2name
, ZFS_MAX_DATASET_NAME_LEN
);
4505 umem_free(clone2name
, ZFS_MAX_DATASET_NAME_LEN
);
4506 umem_free(snap3name
, ZFS_MAX_DATASET_NAME_LEN
);
4510 * Verify dsl_dataset_promote handles EBUSY
4513 ztest_dsl_dataset_promote_busy(ztest_ds_t
*zd
, uint64_t id
)
4521 char *osname
= zd
->zd_name
;
4524 snap1name
= umem_alloc(ZFS_MAX_DATASET_NAME_LEN
, UMEM_NOFAIL
);
4525 clone1name
= umem_alloc(ZFS_MAX_DATASET_NAME_LEN
, UMEM_NOFAIL
);
4526 snap2name
= umem_alloc(ZFS_MAX_DATASET_NAME_LEN
, UMEM_NOFAIL
);
4527 clone2name
= umem_alloc(ZFS_MAX_DATASET_NAME_LEN
, UMEM_NOFAIL
);
4528 snap3name
= umem_alloc(ZFS_MAX_DATASET_NAME_LEN
, UMEM_NOFAIL
);
4530 (void) pthread_rwlock_rdlock(&ztest_name_lock
);
4532 ztest_dsl_dataset_cleanup(osname
, id
);
4534 (void) snprintf(snap1name
, ZFS_MAX_DATASET_NAME_LEN
, "%s@s1_%"PRIu64
"",
4536 (void) snprintf(clone1name
, ZFS_MAX_DATASET_NAME_LEN
, "%s/c1_%"PRIu64
"",
4538 (void) snprintf(snap2name
, ZFS_MAX_DATASET_NAME_LEN
, "%s@s2_%"PRIu64
"",
4540 (void) snprintf(clone2name
, ZFS_MAX_DATASET_NAME_LEN
, "%s/c2_%"PRIu64
"",
4542 (void) snprintf(snap3name
, ZFS_MAX_DATASET_NAME_LEN
, "%s@s3_%"PRIu64
"",
4545 error
= dmu_objset_snapshot_one(osname
, strchr(snap1name
, '@') + 1);
4546 if (error
&& error
!= EEXIST
) {
4547 if (error
== ENOSPC
) {
4548 ztest_record_enospc(FTAG
);
4551 fatal(B_FALSE
, "dmu_take_snapshot(%s) = %d", snap1name
, error
);
4554 error
= dmu_objset_clone(clone1name
, snap1name
);
4556 if (error
== ENOSPC
) {
4557 ztest_record_enospc(FTAG
);
4560 fatal(B_FALSE
, "dmu_objset_create(%s) = %d", clone1name
, error
);
4563 error
= dmu_objset_snapshot_one(clone1name
, strchr(snap2name
, '@') + 1);
4564 if (error
&& error
!= EEXIST
) {
4565 if (error
== ENOSPC
) {
4566 ztest_record_enospc(FTAG
);
4569 fatal(B_FALSE
, "dmu_open_snapshot(%s) = %d", snap2name
, error
);
4572 error
= dmu_objset_snapshot_one(clone1name
, strchr(snap3name
, '@') + 1);
4573 if (error
&& error
!= EEXIST
) {
4574 if (error
== ENOSPC
) {
4575 ztest_record_enospc(FTAG
);
4578 fatal(B_FALSE
, "dmu_open_snapshot(%s) = %d", snap3name
, error
);
4581 error
= dmu_objset_clone(clone2name
, snap3name
);
4583 if (error
== ENOSPC
) {
4584 ztest_record_enospc(FTAG
);
4587 fatal(B_FALSE
, "dmu_objset_create(%s) = %d", clone2name
, error
);
4590 error
= ztest_dmu_objset_own(snap2name
, DMU_OST_ANY
, B_TRUE
, B_TRUE
,
4593 fatal(B_FALSE
, "dmu_objset_own(%s) = %d", snap2name
, error
);
4594 error
= dsl_dataset_promote(clone2name
, NULL
);
4595 if (error
== ENOSPC
) {
4596 dmu_objset_disown(os
, B_TRUE
, FTAG
);
4597 ztest_record_enospc(FTAG
);
4601 fatal(B_FALSE
, "dsl_dataset_promote(%s), %d, not EBUSY",
4603 dmu_objset_disown(os
, B_TRUE
, FTAG
);
4606 ztest_dsl_dataset_cleanup(osname
, id
);
4608 (void) pthread_rwlock_unlock(&ztest_name_lock
);
4610 umem_free(snap1name
, ZFS_MAX_DATASET_NAME_LEN
);
4611 umem_free(clone1name
, ZFS_MAX_DATASET_NAME_LEN
);
4612 umem_free(snap2name
, ZFS_MAX_DATASET_NAME_LEN
);
4613 umem_free(clone2name
, ZFS_MAX_DATASET_NAME_LEN
);
4614 umem_free(snap3name
, ZFS_MAX_DATASET_NAME_LEN
);
4617 #undef OD_ARRAY_SIZE
4618 #define OD_ARRAY_SIZE 4
4621 * Verify that dmu_object_{alloc,free} work as expected.
4624 ztest_dmu_object_alloc_free(ztest_ds_t
*zd
, uint64_t id
)
4631 size
= sizeof (ztest_od_t
) * OD_ARRAY_SIZE
;
4632 od
= umem_alloc(size
, UMEM_NOFAIL
);
4633 batchsize
= OD_ARRAY_SIZE
;
4635 for (b
= 0; b
< batchsize
; b
++)
4636 ztest_od_init(od
+ b
, id
, FTAG
, b
, DMU_OT_UINT64_OTHER
,
4640 * Destroy the previous batch of objects, create a new batch,
4641 * and do some I/O on the new objects.
4643 if (ztest_object_init(zd
, od
, size
, B_TRUE
) != 0)
4646 while (ztest_random(4 * batchsize
) != 0)
4647 ztest_io(zd
, od
[ztest_random(batchsize
)].od_object
,
4648 ztest_random(ZTEST_RANGE_LOCKS
) << SPA_MAXBLOCKSHIFT
);
4650 umem_free(od
, size
);
4654 * Rewind the global allocator to verify object allocation backfilling.
4657 ztest_dmu_object_next_chunk(ztest_ds_t
*zd
, uint64_t id
)
4660 objset_t
*os
= zd
->zd_os
;
4661 int dnodes_per_chunk
= 1 << dmu_object_alloc_chunk_shift
;
4665 * Rewind the global allocator randomly back to a lower object number
4666 * to force backfilling and reclamation of recently freed dnodes.
4668 mutex_enter(&os
->os_obj_lock
);
4669 object
= ztest_random(os
->os_obj_next_chunk
);
4670 os
->os_obj_next_chunk
= P2ALIGN(object
, dnodes_per_chunk
);
4671 mutex_exit(&os
->os_obj_lock
);
4674 #undef OD_ARRAY_SIZE
4675 #define OD_ARRAY_SIZE 2
4678 * Verify that dmu_{read,write} work as expected.
4681 ztest_dmu_read_write(ztest_ds_t
*zd
, uint64_t id
)
4686 objset_t
*os
= zd
->zd_os
;
4687 size
= sizeof (ztest_od_t
) * OD_ARRAY_SIZE
;
4688 od
= umem_alloc(size
, UMEM_NOFAIL
);
4691 uint64_t i
, n
, s
, txg
;
4692 bufwad_t
*packbuf
, *bigbuf
, *pack
, *bigH
, *bigT
;
4693 uint64_t packobj
, packoff
, packsize
, bigobj
, bigoff
, bigsize
;
4694 uint64_t chunksize
= (1000 + ztest_random(1000)) * sizeof (uint64_t);
4695 uint64_t regions
= 997;
4696 uint64_t stride
= 123456789ULL;
4697 uint64_t width
= 40;
4698 int free_percent
= 5;
4701 * This test uses two objects, packobj and bigobj, that are always
4702 * updated together (i.e. in the same tx) so that their contents are
4703 * in sync and can be compared. Their contents relate to each other
4704 * in a simple way: packobj is a dense array of 'bufwad' structures,
4705 * while bigobj is a sparse array of the same bufwads. Specifically,
4706 * for any index n, there are three bufwads that should be identical:
4708 * packobj, at offset n * sizeof (bufwad_t)
4709 * bigobj, at the head of the nth chunk
4710 * bigobj, at the tail of the nth chunk
4712 * The chunk size is arbitrary. It doesn't have to be a power of two,
4713 * and it doesn't have any relation to the object blocksize.
4714 * The only requirement is that it can hold at least two bufwads.
4716 * Normally, we write the bufwad to each of these locations.
4717 * However, free_percent of the time we instead write zeroes to
4718 * packobj and perform a dmu_free_range() on bigobj. By comparing
4719 * bigobj to packobj, we can verify that the DMU is correctly
4720 * tracking which parts of an object are allocated and free,
4721 * and that the contents of the allocated blocks are correct.
4725 * Read the directory info. If it's the first time, set things up.
4727 ztest_od_init(od
, id
, FTAG
, 0, DMU_OT_UINT64_OTHER
, 0, 0, chunksize
);
4728 ztest_od_init(od
+ 1, id
, FTAG
, 1, DMU_OT_UINT64_OTHER
, 0, 0,
4731 if (ztest_object_init(zd
, od
, size
, B_FALSE
) != 0) {
4732 umem_free(od
, size
);
4736 bigobj
= od
[0].od_object
;
4737 packobj
= od
[1].od_object
;
4738 chunksize
= od
[0].od_gen
;
4739 ASSERT3U(chunksize
, ==, od
[1].od_gen
);
4742 * Prefetch a random chunk of the big object.
4743 * Our aim here is to get some async reads in flight
4744 * for blocks that we may free below; the DMU should
4745 * handle this race correctly.
4747 n
= ztest_random(regions
) * stride
+ ztest_random(width
);
4748 s
= 1 + ztest_random(2 * width
- 1);
4749 dmu_prefetch(os
, bigobj
, 0, n
* chunksize
, s
* chunksize
,
4750 ZIO_PRIORITY_SYNC_READ
);
4753 * Pick a random index and compute the offsets into packobj and bigobj.
4755 n
= ztest_random(regions
) * stride
+ ztest_random(width
);
4756 s
= 1 + ztest_random(width
- 1);
4758 packoff
= n
* sizeof (bufwad_t
);
4759 packsize
= s
* sizeof (bufwad_t
);
4761 bigoff
= n
* chunksize
;
4762 bigsize
= s
* chunksize
;
4764 packbuf
= umem_alloc(packsize
, UMEM_NOFAIL
);
4765 bigbuf
= umem_alloc(bigsize
, UMEM_NOFAIL
);
4768 * free_percent of the time, free a range of bigobj rather than
4771 freeit
= (ztest_random(100) < free_percent
);
4774 * Read the current contents of our objects.
4776 error
= dmu_read(os
, packobj
, packoff
, packsize
, packbuf
,
4779 error
= dmu_read(os
, bigobj
, bigoff
, bigsize
, bigbuf
,
4784 * Get a tx for the mods to both packobj and bigobj.
4786 tx
= dmu_tx_create(os
);
4788 dmu_tx_hold_write(tx
, packobj
, packoff
, packsize
);
4791 dmu_tx_hold_free(tx
, bigobj
, bigoff
, bigsize
);
4793 dmu_tx_hold_write(tx
, bigobj
, bigoff
, bigsize
);
4795 /* This accounts for setting the checksum/compression. */
4796 dmu_tx_hold_bonus(tx
, bigobj
);
4798 txg
= ztest_tx_assign(tx
, TXG_MIGHTWAIT
, FTAG
);
4800 umem_free(packbuf
, packsize
);
4801 umem_free(bigbuf
, bigsize
);
4802 umem_free(od
, size
);
4806 enum zio_checksum cksum
;
4808 cksum
= (enum zio_checksum
)
4809 ztest_random_dsl_prop(ZFS_PROP_CHECKSUM
);
4810 } while (cksum
>= ZIO_CHECKSUM_LEGACY_FUNCTIONS
);
4811 dmu_object_set_checksum(os
, bigobj
, cksum
, tx
);
4813 enum zio_compress comp
;
4815 comp
= (enum zio_compress
)
4816 ztest_random_dsl_prop(ZFS_PROP_COMPRESSION
);
4817 } while (comp
>= ZIO_COMPRESS_LEGACY_FUNCTIONS
);
4818 dmu_object_set_compress(os
, bigobj
, comp
, tx
);
4821 * For each index from n to n + s, verify that the existing bufwad
4822 * in packobj matches the bufwads at the head and tail of the
4823 * corresponding chunk in bigobj. Then update all three bufwads
4824 * with the new values we want to write out.
4826 for (i
= 0; i
< s
; i
++) {
4828 pack
= (bufwad_t
*)((char *)packbuf
+ i
* sizeof (bufwad_t
));
4830 bigH
= (bufwad_t
*)((char *)bigbuf
+ i
* chunksize
);
4832 bigT
= (bufwad_t
*)((char *)bigH
+ chunksize
) - 1;
4834 ASSERT3U((uintptr_t)bigH
- (uintptr_t)bigbuf
, <, bigsize
);
4835 ASSERT3U((uintptr_t)bigT
- (uintptr_t)bigbuf
, <, bigsize
);
4837 if (pack
->bw_txg
> txg
)
4839 "future leak: got %"PRIx64
", open txg is %"PRIx64
"",
4842 if (pack
->bw_data
!= 0 && pack
->bw_index
!= n
+ i
)
4843 fatal(B_FALSE
, "wrong index: "
4844 "got %"PRIx64
", wanted %"PRIx64
"+%"PRIx64
"",
4845 pack
->bw_index
, n
, i
);
4847 if (memcmp(pack
, bigH
, sizeof (bufwad_t
)) != 0)
4848 fatal(B_FALSE
, "pack/bigH mismatch in %p/%p",
4851 if (memcmp(pack
, bigT
, sizeof (bufwad_t
)) != 0)
4852 fatal(B_FALSE
, "pack/bigT mismatch in %p/%p",
4856 memset(pack
, 0, sizeof (bufwad_t
));
4858 pack
->bw_index
= n
+ i
;
4860 pack
->bw_data
= 1 + ztest_random(-2ULL);
4867 * We've verified all the old bufwads, and made new ones.
4868 * Now write them out.
4870 dmu_write(os
, packobj
, packoff
, packsize
, packbuf
, tx
);
4873 if (ztest_opts
.zo_verbose
>= 7) {
4874 (void) printf("freeing offset %"PRIx64
" size %"PRIx64
""
4876 bigoff
, bigsize
, txg
);
4878 VERIFY0(dmu_free_range(os
, bigobj
, bigoff
, bigsize
, tx
));
4880 if (ztest_opts
.zo_verbose
>= 7) {
4881 (void) printf("writing offset %"PRIx64
" size %"PRIx64
""
4883 bigoff
, bigsize
, txg
);
4885 dmu_write(os
, bigobj
, bigoff
, bigsize
, bigbuf
, tx
);
4891 * Sanity check the stuff we just wrote.
4894 void *packcheck
= umem_alloc(packsize
, UMEM_NOFAIL
);
4895 void *bigcheck
= umem_alloc(bigsize
, UMEM_NOFAIL
);
4897 VERIFY0(dmu_read(os
, packobj
, packoff
,
4898 packsize
, packcheck
, DMU_READ_PREFETCH
));
4899 VERIFY0(dmu_read(os
, bigobj
, bigoff
,
4900 bigsize
, bigcheck
, DMU_READ_PREFETCH
));
4902 ASSERT0(memcmp(packbuf
, packcheck
, packsize
));
4903 ASSERT0(memcmp(bigbuf
, bigcheck
, bigsize
));
4905 umem_free(packcheck
, packsize
);
4906 umem_free(bigcheck
, bigsize
);
4909 umem_free(packbuf
, packsize
);
4910 umem_free(bigbuf
, bigsize
);
4911 umem_free(od
, size
);
4915 compare_and_update_pbbufs(uint64_t s
, bufwad_t
*packbuf
, bufwad_t
*bigbuf
,
4916 uint64_t bigsize
, uint64_t n
, uint64_t chunksize
, uint64_t txg
)
4924 * For each index from n to n + s, verify that the existing bufwad
4925 * in packobj matches the bufwads at the head and tail of the
4926 * corresponding chunk in bigobj. Then update all three bufwads
4927 * with the new values we want to write out.
4929 for (i
= 0; i
< s
; i
++) {
4931 pack
= (bufwad_t
*)((char *)packbuf
+ i
* sizeof (bufwad_t
));
4933 bigH
= (bufwad_t
*)((char *)bigbuf
+ i
* chunksize
);
4935 bigT
= (bufwad_t
*)((char *)bigH
+ chunksize
) - 1;
4937 ASSERT3U((uintptr_t)bigH
- (uintptr_t)bigbuf
, <, bigsize
);
4938 ASSERT3U((uintptr_t)bigT
- (uintptr_t)bigbuf
, <, bigsize
);
4940 if (pack
->bw_txg
> txg
)
4942 "future leak: got %"PRIx64
", open txg is %"PRIx64
"",
4945 if (pack
->bw_data
!= 0 && pack
->bw_index
!= n
+ i
)
4946 fatal(B_FALSE
, "wrong index: "
4947 "got %"PRIx64
", wanted %"PRIx64
"+%"PRIx64
"",
4948 pack
->bw_index
, n
, i
);
4950 if (memcmp(pack
, bigH
, sizeof (bufwad_t
)) != 0)
4951 fatal(B_FALSE
, "pack/bigH mismatch in %p/%p",
4954 if (memcmp(pack
, bigT
, sizeof (bufwad_t
)) != 0)
4955 fatal(B_FALSE
, "pack/bigT mismatch in %p/%p",
4958 pack
->bw_index
= n
+ i
;
4960 pack
->bw_data
= 1 + ztest_random(-2ULL);
4967 #undef OD_ARRAY_SIZE
4968 #define OD_ARRAY_SIZE 2
4971 ztest_dmu_read_write_zcopy(ztest_ds_t
*zd
, uint64_t id
)
4973 objset_t
*os
= zd
->zd_os
;
4980 bufwad_t
*packbuf
, *bigbuf
;
4981 uint64_t packobj
, packoff
, packsize
, bigobj
, bigoff
, bigsize
;
4982 uint64_t blocksize
= ztest_random_blocksize();
4983 uint64_t chunksize
= blocksize
;
4984 uint64_t regions
= 997;
4985 uint64_t stride
= 123456789ULL;
4987 dmu_buf_t
*bonus_db
;
4988 arc_buf_t
**bigbuf_arcbufs
;
4989 dmu_object_info_t doi
;
4991 size
= sizeof (ztest_od_t
) * OD_ARRAY_SIZE
;
4992 od
= umem_alloc(size
, UMEM_NOFAIL
);
4995 * This test uses two objects, packobj and bigobj, that are always
4996 * updated together (i.e. in the same tx) so that their contents are
4997 * in sync and can be compared. Their contents relate to each other
4998 * in a simple way: packobj is a dense array of 'bufwad' structures,
4999 * while bigobj is a sparse array of the same bufwads. Specifically,
5000 * for any index n, there are three bufwads that should be identical:
5002 * packobj, at offset n * sizeof (bufwad_t)
5003 * bigobj, at the head of the nth chunk
5004 * bigobj, at the tail of the nth chunk
5006 * The chunk size is set equal to bigobj block size so that
5007 * dmu_assign_arcbuf_by_dbuf() can be tested for object updates.
5011 * Read the directory info. If it's the first time, set things up.
5013 ztest_od_init(od
, id
, FTAG
, 0, DMU_OT_UINT64_OTHER
, blocksize
, 0, 0);
5014 ztest_od_init(od
+ 1, id
, FTAG
, 1, DMU_OT_UINT64_OTHER
, 0, 0,
5018 if (ztest_object_init(zd
, od
, size
, B_FALSE
) != 0) {
5019 umem_free(od
, size
);
5023 bigobj
= od
[0].od_object
;
5024 packobj
= od
[1].od_object
;
5025 blocksize
= od
[0].od_blocksize
;
5026 chunksize
= blocksize
;
5027 ASSERT3U(chunksize
, ==, od
[1].od_gen
);
5029 VERIFY0(dmu_object_info(os
, bigobj
, &doi
));
5030 VERIFY(ISP2(doi
.doi_data_block_size
));
5031 VERIFY3U(chunksize
, ==, doi
.doi_data_block_size
);
5032 VERIFY3U(chunksize
, >=, 2 * sizeof (bufwad_t
));
5035 * Pick a random index and compute the offsets into packobj and bigobj.
5037 n
= ztest_random(regions
) * stride
+ ztest_random(width
);
5038 s
= 1 + ztest_random(width
- 1);
5040 packoff
= n
* sizeof (bufwad_t
);
5041 packsize
= s
* sizeof (bufwad_t
);
5043 bigoff
= n
* chunksize
;
5044 bigsize
= s
* chunksize
;
5046 packbuf
= umem_zalloc(packsize
, UMEM_NOFAIL
);
5047 bigbuf
= umem_zalloc(bigsize
, UMEM_NOFAIL
);
5049 VERIFY0(dmu_bonus_hold(os
, bigobj
, FTAG
, &bonus_db
));
5051 bigbuf_arcbufs
= umem_zalloc(2 * s
* sizeof (arc_buf_t
*), UMEM_NOFAIL
);
5054 * Iteration 0 test zcopy for DB_UNCACHED dbufs.
5055 * Iteration 1 test zcopy to already referenced dbufs.
5056 * Iteration 2 test zcopy to dirty dbuf in the same txg.
5057 * Iteration 3 test zcopy to dbuf dirty in previous txg.
5058 * Iteration 4 test zcopy when dbuf is no longer dirty.
5059 * Iteration 5 test zcopy when it can't be done.
5060 * Iteration 6 one more zcopy write.
5062 for (i
= 0; i
< 7; i
++) {
5067 * In iteration 5 (i == 5) use arcbufs
5068 * that don't match bigobj blksz to test
5069 * dmu_assign_arcbuf_by_dbuf() when it can't directly
5070 * assign an arcbuf to a dbuf.
5072 for (j
= 0; j
< s
; j
++) {
5073 if (i
!= 5 || chunksize
< (SPA_MINBLOCKSIZE
* 2)) {
5075 dmu_request_arcbuf(bonus_db
, chunksize
);
5077 bigbuf_arcbufs
[2 * j
] =
5078 dmu_request_arcbuf(bonus_db
, chunksize
/ 2);
5079 bigbuf_arcbufs
[2 * j
+ 1] =
5080 dmu_request_arcbuf(bonus_db
, chunksize
/ 2);
5085 * Get a tx for the mods to both packobj and bigobj.
5087 tx
= dmu_tx_create(os
);
5089 dmu_tx_hold_write(tx
, packobj
, packoff
, packsize
);
5090 dmu_tx_hold_write(tx
, bigobj
, bigoff
, bigsize
);
5092 txg
= ztest_tx_assign(tx
, TXG_MIGHTWAIT
, FTAG
);
5094 umem_free(packbuf
, packsize
);
5095 umem_free(bigbuf
, bigsize
);
5096 for (j
= 0; j
< s
; j
++) {
5098 chunksize
< (SPA_MINBLOCKSIZE
* 2)) {
5099 dmu_return_arcbuf(bigbuf_arcbufs
[j
]);
5102 bigbuf_arcbufs
[2 * j
]);
5104 bigbuf_arcbufs
[2 * j
+ 1]);
5107 umem_free(bigbuf_arcbufs
, 2 * s
* sizeof (arc_buf_t
*));
5108 umem_free(od
, size
);
5109 dmu_buf_rele(bonus_db
, FTAG
);
5114 * 50% of the time don't read objects in the 1st iteration to
5115 * test dmu_assign_arcbuf_by_dbuf() for the case when there are
5116 * no existing dbufs for the specified offsets.
5118 if (i
!= 0 || ztest_random(2) != 0) {
5119 error
= dmu_read(os
, packobj
, packoff
,
5120 packsize
, packbuf
, DMU_READ_PREFETCH
);
5122 error
= dmu_read(os
, bigobj
, bigoff
, bigsize
,
5123 bigbuf
, DMU_READ_PREFETCH
);
5126 compare_and_update_pbbufs(s
, packbuf
, bigbuf
, bigsize
,
5130 * We've verified all the old bufwads, and made new ones.
5131 * Now write them out.
5133 dmu_write(os
, packobj
, packoff
, packsize
, packbuf
, tx
);
5134 if (ztest_opts
.zo_verbose
>= 7) {
5135 (void) printf("writing offset %"PRIx64
" size %"PRIx64
""
5137 bigoff
, bigsize
, txg
);
5139 for (off
= bigoff
, j
= 0; j
< s
; j
++, off
+= chunksize
) {
5141 if (i
!= 5 || chunksize
< (SPA_MINBLOCKSIZE
* 2)) {
5142 memcpy(bigbuf_arcbufs
[j
]->b_data
,
5143 (caddr_t
)bigbuf
+ (off
- bigoff
),
5146 memcpy(bigbuf_arcbufs
[2 * j
]->b_data
,
5147 (caddr_t
)bigbuf
+ (off
- bigoff
),
5149 memcpy(bigbuf_arcbufs
[2 * j
+ 1]->b_data
,
5150 (caddr_t
)bigbuf
+ (off
- bigoff
) +
5156 VERIFY(dmu_buf_hold(os
, bigobj
, off
,
5157 FTAG
, &dbt
, DMU_READ_NO_PREFETCH
) == 0);
5159 if (i
!= 5 || chunksize
< (SPA_MINBLOCKSIZE
* 2)) {
5160 VERIFY0(dmu_assign_arcbuf_by_dbuf(bonus_db
,
5161 off
, bigbuf_arcbufs
[j
], tx
));
5163 VERIFY0(dmu_assign_arcbuf_by_dbuf(bonus_db
,
5164 off
, bigbuf_arcbufs
[2 * j
], tx
));
5165 VERIFY0(dmu_assign_arcbuf_by_dbuf(bonus_db
,
5166 off
+ chunksize
/ 2,
5167 bigbuf_arcbufs
[2 * j
+ 1], tx
));
5170 dmu_buf_rele(dbt
, FTAG
);
5176 * Sanity check the stuff we just wrote.
5179 void *packcheck
= umem_alloc(packsize
, UMEM_NOFAIL
);
5180 void *bigcheck
= umem_alloc(bigsize
, UMEM_NOFAIL
);
5182 VERIFY0(dmu_read(os
, packobj
, packoff
,
5183 packsize
, packcheck
, DMU_READ_PREFETCH
));
5184 VERIFY0(dmu_read(os
, bigobj
, bigoff
,
5185 bigsize
, bigcheck
, DMU_READ_PREFETCH
));
5187 ASSERT0(memcmp(packbuf
, packcheck
, packsize
));
5188 ASSERT0(memcmp(bigbuf
, bigcheck
, bigsize
));
5190 umem_free(packcheck
, packsize
);
5191 umem_free(bigcheck
, bigsize
);
5194 txg_wait_open(dmu_objset_pool(os
), 0, B_TRUE
);
5195 } else if (i
== 3) {
5196 txg_wait_synced(dmu_objset_pool(os
), 0);
5200 dmu_buf_rele(bonus_db
, FTAG
);
5201 umem_free(packbuf
, packsize
);
5202 umem_free(bigbuf
, bigsize
);
5203 umem_free(bigbuf_arcbufs
, 2 * s
* sizeof (arc_buf_t
*));
5204 umem_free(od
, size
);
5208 ztest_dmu_write_parallel(ztest_ds_t
*zd
, uint64_t id
)
5213 od
= umem_alloc(sizeof (ztest_od_t
), UMEM_NOFAIL
);
5214 uint64_t offset
= (1ULL << (ztest_random(20) + 43)) +
5215 (ztest_random(ZTEST_RANGE_LOCKS
) << SPA_MAXBLOCKSHIFT
);
5218 * Have multiple threads write to large offsets in an object
5219 * to verify that parallel writes to an object -- even to the
5220 * same blocks within the object -- doesn't cause any trouble.
5222 ztest_od_init(od
, ID_PARALLEL
, FTAG
, 0, DMU_OT_UINT64_OTHER
, 0, 0, 0);
5224 if (ztest_object_init(zd
, od
, sizeof (ztest_od_t
), B_FALSE
) != 0)
5227 while (ztest_random(10) != 0)
5228 ztest_io(zd
, od
->od_object
, offset
);
5230 umem_free(od
, sizeof (ztest_od_t
));
5234 ztest_dmu_prealloc(ztest_ds_t
*zd
, uint64_t id
)
5237 uint64_t offset
= (1ULL << (ztest_random(4) + SPA_MAXBLOCKSHIFT
)) +
5238 (ztest_random(ZTEST_RANGE_LOCKS
) << SPA_MAXBLOCKSHIFT
);
5239 uint64_t count
= ztest_random(20) + 1;
5240 uint64_t blocksize
= ztest_random_blocksize();
5243 od
= umem_alloc(sizeof (ztest_od_t
), UMEM_NOFAIL
);
5245 ztest_od_init(od
, id
, FTAG
, 0, DMU_OT_UINT64_OTHER
, blocksize
, 0, 0);
5247 if (ztest_object_init(zd
, od
, sizeof (ztest_od_t
),
5248 !ztest_random(2)) != 0) {
5249 umem_free(od
, sizeof (ztest_od_t
));
5253 if (ztest_truncate(zd
, od
->od_object
, offset
, count
* blocksize
) != 0) {
5254 umem_free(od
, sizeof (ztest_od_t
));
5258 ztest_prealloc(zd
, od
->od_object
, offset
, count
* blocksize
);
5260 data
= umem_zalloc(blocksize
, UMEM_NOFAIL
);
5262 while (ztest_random(count
) != 0) {
5263 uint64_t randoff
= offset
+ (ztest_random(count
) * blocksize
);
5264 if (ztest_write(zd
, od
->od_object
, randoff
, blocksize
,
5267 while (ztest_random(4) != 0)
5268 ztest_io(zd
, od
->od_object
, randoff
);
5271 umem_free(data
, blocksize
);
5272 umem_free(od
, sizeof (ztest_od_t
));
5276 * Verify that zap_{create,destroy,add,remove,update} work as expected.
5278 #define ZTEST_ZAP_MIN_INTS 1
5279 #define ZTEST_ZAP_MAX_INTS 4
5280 #define ZTEST_ZAP_MAX_PROPS 1000
5283 ztest_zap(ztest_ds_t
*zd
, uint64_t id
)
5285 objset_t
*os
= zd
->zd_os
;
5288 uint64_t txg
, last_txg
;
5289 uint64_t value
[ZTEST_ZAP_MAX_INTS
];
5290 uint64_t zl_ints
, zl_intsize
, prop
;
5293 char propname
[100], txgname
[100];
5295 char *hc
[2] = { "s.acl.h", ".s.open.h.hyLZlg" };
5297 od
= umem_alloc(sizeof (ztest_od_t
), UMEM_NOFAIL
);
5298 ztest_od_init(od
, id
, FTAG
, 0, DMU_OT_ZAP_OTHER
, 0, 0, 0);
5300 if (ztest_object_init(zd
, od
, sizeof (ztest_od_t
),
5301 !ztest_random(2)) != 0)
5304 object
= od
->od_object
;
5307 * Generate a known hash collision, and verify that
5308 * we can lookup and remove both entries.
5310 tx
= dmu_tx_create(os
);
5311 dmu_tx_hold_zap(tx
, object
, B_TRUE
, NULL
);
5312 txg
= ztest_tx_assign(tx
, TXG_MIGHTWAIT
, FTAG
);
5315 for (i
= 0; i
< 2; i
++) {
5317 VERIFY0(zap_add(os
, object
, hc
[i
], sizeof (uint64_t),
5320 for (i
= 0; i
< 2; i
++) {
5321 VERIFY3U(EEXIST
, ==, zap_add(os
, object
, hc
[i
],
5322 sizeof (uint64_t), 1, &value
[i
], tx
));
5324 zap_length(os
, object
, hc
[i
], &zl_intsize
, &zl_ints
));
5325 ASSERT3U(zl_intsize
, ==, sizeof (uint64_t));
5326 ASSERT3U(zl_ints
, ==, 1);
5328 for (i
= 0; i
< 2; i
++) {
5329 VERIFY0(zap_remove(os
, object
, hc
[i
], tx
));
5334 * Generate a bunch of random entries.
5336 ints
= MAX(ZTEST_ZAP_MIN_INTS
, object
% ZTEST_ZAP_MAX_INTS
);
5338 prop
= ztest_random(ZTEST_ZAP_MAX_PROPS
);
5339 (void) sprintf(propname
, "prop_%"PRIu64
"", prop
);
5340 (void) sprintf(txgname
, "txg_%"PRIu64
"", prop
);
5341 memset(value
, 0, sizeof (value
));
5345 * If these zap entries already exist, validate their contents.
5347 error
= zap_length(os
, object
, txgname
, &zl_intsize
, &zl_ints
);
5349 ASSERT3U(zl_intsize
, ==, sizeof (uint64_t));
5350 ASSERT3U(zl_ints
, ==, 1);
5352 VERIFY0(zap_lookup(os
, object
, txgname
, zl_intsize
,
5353 zl_ints
, &last_txg
));
5355 VERIFY0(zap_length(os
, object
, propname
, &zl_intsize
,
5358 ASSERT3U(zl_intsize
, ==, sizeof (uint64_t));
5359 ASSERT3U(zl_ints
, ==, ints
);
5361 VERIFY0(zap_lookup(os
, object
, propname
, zl_intsize
,
5364 for (i
= 0; i
< ints
; i
++) {
5365 ASSERT3U(value
[i
], ==, last_txg
+ object
+ i
);
5368 ASSERT3U(error
, ==, ENOENT
);
5372 * Atomically update two entries in our zap object.
5373 * The first is named txg_%llu, and contains the txg
5374 * in which the property was last updated. The second
5375 * is named prop_%llu, and the nth element of its value
5376 * should be txg + object + n.
5378 tx
= dmu_tx_create(os
);
5379 dmu_tx_hold_zap(tx
, object
, B_TRUE
, NULL
);
5380 txg
= ztest_tx_assign(tx
, TXG_MIGHTWAIT
, FTAG
);
5385 fatal(B_FALSE
, "zap future leak: old %"PRIu64
" new %"PRIu64
"",
5388 for (i
= 0; i
< ints
; i
++)
5389 value
[i
] = txg
+ object
+ i
;
5391 VERIFY0(zap_update(os
, object
, txgname
, sizeof (uint64_t),
5393 VERIFY0(zap_update(os
, object
, propname
, sizeof (uint64_t),
5399 * Remove a random pair of entries.
5401 prop
= ztest_random(ZTEST_ZAP_MAX_PROPS
);
5402 (void) sprintf(propname
, "prop_%"PRIu64
"", prop
);
5403 (void) sprintf(txgname
, "txg_%"PRIu64
"", prop
);
5405 error
= zap_length(os
, object
, txgname
, &zl_intsize
, &zl_ints
);
5407 if (error
== ENOENT
)
5412 tx
= dmu_tx_create(os
);
5413 dmu_tx_hold_zap(tx
, object
, B_TRUE
, NULL
);
5414 txg
= ztest_tx_assign(tx
, TXG_MIGHTWAIT
, FTAG
);
5417 VERIFY0(zap_remove(os
, object
, txgname
, tx
));
5418 VERIFY0(zap_remove(os
, object
, propname
, tx
));
5421 umem_free(od
, sizeof (ztest_od_t
));
5425 * Test case to test the upgrading of a microzap to fatzap.
5428 ztest_fzap(ztest_ds_t
*zd
, uint64_t id
)
5430 objset_t
*os
= zd
->zd_os
;
5432 uint64_t object
, txg
, value
;
5434 od
= umem_alloc(sizeof (ztest_od_t
), UMEM_NOFAIL
);
5435 ztest_od_init(od
, id
, FTAG
, 0, DMU_OT_ZAP_OTHER
, 0, 0, 0);
5437 if (ztest_object_init(zd
, od
, sizeof (ztest_od_t
),
5438 !ztest_random(2)) != 0)
5440 object
= od
->od_object
;
5443 * Add entries to this ZAP and make sure it spills over
5444 * and gets upgraded to a fatzap. Also, since we are adding
5445 * 2050 entries we should see ptrtbl growth and leaf-block split.
5447 for (value
= 0; value
< 2050; value
++) {
5448 char name
[ZFS_MAX_DATASET_NAME_LEN
];
5452 (void) snprintf(name
, sizeof (name
), "fzap-%"PRIu64
"-%"PRIu64
"",
5455 tx
= dmu_tx_create(os
);
5456 dmu_tx_hold_zap(tx
, object
, B_TRUE
, name
);
5457 txg
= ztest_tx_assign(tx
, TXG_MIGHTWAIT
, FTAG
);
5460 error
= zap_add(os
, object
, name
, sizeof (uint64_t), 1,
5462 ASSERT(error
== 0 || error
== EEXIST
);
5466 umem_free(od
, sizeof (ztest_od_t
));
5470 ztest_zap_parallel(ztest_ds_t
*zd
, uint64_t id
)
5473 objset_t
*os
= zd
->zd_os
;
5475 uint64_t txg
, object
, count
, wsize
, wc
, zl_wsize
, zl_wc
;
5477 int i
, namelen
, error
;
5478 int micro
= ztest_random(2);
5479 char name
[20], string_value
[20];
5482 od
= umem_alloc(sizeof (ztest_od_t
), UMEM_NOFAIL
);
5483 ztest_od_init(od
, ID_PARALLEL
, FTAG
, micro
, DMU_OT_ZAP_OTHER
, 0, 0, 0);
5485 if (ztest_object_init(zd
, od
, sizeof (ztest_od_t
), B_FALSE
) != 0) {
5486 umem_free(od
, sizeof (ztest_od_t
));
5490 object
= od
->od_object
;
5493 * Generate a random name of the form 'xxx.....' where each
5494 * x is a random printable character and the dots are dots.
5495 * There are 94 such characters, and the name length goes from
5496 * 6 to 20, so there are 94^3 * 15 = 12,458,760 possible names.
5498 namelen
= ztest_random(sizeof (name
) - 5) + 5 + 1;
5500 for (i
= 0; i
< 3; i
++)
5501 name
[i
] = '!' + ztest_random('~' - '!' + 1);
5502 for (; i
< namelen
- 1; i
++)
5506 if ((namelen
& 1) || micro
) {
5507 wsize
= sizeof (txg
);
5513 data
= string_value
;
5517 VERIFY0(zap_count(os
, object
, &count
));
5518 ASSERT3S(count
, !=, -1ULL);
5521 * Select an operation: length, lookup, add, update, remove.
5523 i
= ztest_random(5);
5526 tx
= dmu_tx_create(os
);
5527 dmu_tx_hold_zap(tx
, object
, B_TRUE
, NULL
);
5528 txg
= ztest_tx_assign(tx
, TXG_MIGHTWAIT
, FTAG
);
5530 umem_free(od
, sizeof (ztest_od_t
));
5533 memcpy(string_value
, name
, namelen
);
5537 memset(string_value
, 0, namelen
);
5543 error
= zap_length(os
, object
, name
, &zl_wsize
, &zl_wc
);
5545 ASSERT3U(wsize
, ==, zl_wsize
);
5546 ASSERT3U(wc
, ==, zl_wc
);
5548 ASSERT3U(error
, ==, ENOENT
);
5553 error
= zap_lookup(os
, object
, name
, wsize
, wc
, data
);
5555 if (data
== string_value
&&
5556 memcmp(name
, data
, namelen
) != 0)
5557 fatal(B_FALSE
, "name '%s' != val '%s' len %d",
5558 name
, (char *)data
, namelen
);
5560 ASSERT3U(error
, ==, ENOENT
);
5565 error
= zap_add(os
, object
, name
, wsize
, wc
, data
, tx
);
5566 ASSERT(error
== 0 || error
== EEXIST
);
5570 VERIFY0(zap_update(os
, object
, name
, wsize
, wc
, data
, tx
));
5574 error
= zap_remove(os
, object
, name
, tx
);
5575 ASSERT(error
== 0 || error
== ENOENT
);
5582 umem_free(od
, sizeof (ztest_od_t
));
5586 * Commit callback data.
5588 typedef struct ztest_cb_data
{
5589 list_node_t zcd_node
;
5591 int zcd_expected_err
;
5592 boolean_t zcd_added
;
5593 boolean_t zcd_called
;
5597 /* This is the actual commit callback function */
5599 ztest_commit_callback(void *arg
, int error
)
5601 ztest_cb_data_t
*data
= arg
;
5602 uint64_t synced_txg
;
5604 VERIFY3P(data
, !=, NULL
);
5605 VERIFY3S(data
->zcd_expected_err
, ==, error
);
5606 VERIFY(!data
->zcd_called
);
5608 synced_txg
= spa_last_synced_txg(data
->zcd_spa
);
5609 if (data
->zcd_txg
> synced_txg
)
5611 "commit callback of txg %"PRIu64
" called prematurely, "
5612 "last synced txg = %"PRIu64
"\n",
5613 data
->zcd_txg
, synced_txg
);
5615 data
->zcd_called
= B_TRUE
;
5617 if (error
== ECANCELED
) {
5618 ASSERT0(data
->zcd_txg
);
5619 ASSERT(!data
->zcd_added
);
5622 * The private callback data should be destroyed here, but
5623 * since we are going to check the zcd_called field after
5624 * dmu_tx_abort(), we will destroy it there.
5629 ASSERT(data
->zcd_added
);
5630 ASSERT3U(data
->zcd_txg
, !=, 0);
5632 (void) mutex_enter(&zcl
.zcl_callbacks_lock
);
5634 /* See if this cb was called more quickly */
5635 if ((synced_txg
- data
->zcd_txg
) < zc_min_txg_delay
)
5636 zc_min_txg_delay
= synced_txg
- data
->zcd_txg
;
5638 /* Remove our callback from the list */
5639 list_remove(&zcl
.zcl_callbacks
, data
);
5641 (void) mutex_exit(&zcl
.zcl_callbacks_lock
);
5643 umem_free(data
, sizeof (ztest_cb_data_t
));
5646 /* Allocate and initialize callback data structure */
5647 static ztest_cb_data_t
*
5648 ztest_create_cb_data(objset_t
*os
, uint64_t txg
)
5650 ztest_cb_data_t
*cb_data
;
5652 cb_data
= umem_zalloc(sizeof (ztest_cb_data_t
), UMEM_NOFAIL
);
5654 cb_data
->zcd_txg
= txg
;
5655 cb_data
->zcd_spa
= dmu_objset_spa(os
);
5656 list_link_init(&cb_data
->zcd_node
);
5662 * Commit callback test.
5665 ztest_dmu_commit_callbacks(ztest_ds_t
*zd
, uint64_t id
)
5667 objset_t
*os
= zd
->zd_os
;
5670 ztest_cb_data_t
*cb_data
[3], *tmp_cb
;
5671 uint64_t old_txg
, txg
;
5674 od
= umem_alloc(sizeof (ztest_od_t
), UMEM_NOFAIL
);
5675 ztest_od_init(od
, id
, FTAG
, 0, DMU_OT_UINT64_OTHER
, 0, 0, 0);
5677 if (ztest_object_init(zd
, od
, sizeof (ztest_od_t
), B_FALSE
) != 0) {
5678 umem_free(od
, sizeof (ztest_od_t
));
5682 tx
= dmu_tx_create(os
);
5684 cb_data
[0] = ztest_create_cb_data(os
, 0);
5685 dmu_tx_callback_register(tx
, ztest_commit_callback
, cb_data
[0]);
5687 dmu_tx_hold_write(tx
, od
->od_object
, 0, sizeof (uint64_t));
5689 /* Every once in a while, abort the transaction on purpose */
5690 if (ztest_random(100) == 0)
5694 error
= dmu_tx_assign(tx
, TXG_NOWAIT
);
5696 txg
= error
? 0 : dmu_tx_get_txg(tx
);
5698 cb_data
[0]->zcd_txg
= txg
;
5699 cb_data
[1] = ztest_create_cb_data(os
, txg
);
5700 dmu_tx_callback_register(tx
, ztest_commit_callback
, cb_data
[1]);
5704 * It's not a strict requirement to call the registered
5705 * callbacks from inside dmu_tx_abort(), but that's what
5706 * it's supposed to happen in the current implementation
5707 * so we will check for that.
5709 for (i
= 0; i
< 2; i
++) {
5710 cb_data
[i
]->zcd_expected_err
= ECANCELED
;
5711 VERIFY(!cb_data
[i
]->zcd_called
);
5716 for (i
= 0; i
< 2; i
++) {
5717 VERIFY(cb_data
[i
]->zcd_called
);
5718 umem_free(cb_data
[i
], sizeof (ztest_cb_data_t
));
5721 umem_free(od
, sizeof (ztest_od_t
));
5725 cb_data
[2] = ztest_create_cb_data(os
, txg
);
5726 dmu_tx_callback_register(tx
, ztest_commit_callback
, cb_data
[2]);
5729 * Read existing data to make sure there isn't a future leak.
5731 VERIFY0(dmu_read(os
, od
->od_object
, 0, sizeof (uint64_t),
5732 &old_txg
, DMU_READ_PREFETCH
));
5736 "future leak: got %"PRIu64
", open txg is %"PRIu64
"",
5739 dmu_write(os
, od
->od_object
, 0, sizeof (uint64_t), &txg
, tx
);
5741 (void) mutex_enter(&zcl
.zcl_callbacks_lock
);
5744 * Since commit callbacks don't have any ordering requirement and since
5745 * it is theoretically possible for a commit callback to be called
5746 * after an arbitrary amount of time has elapsed since its txg has been
5747 * synced, it is difficult to reliably determine whether a commit
5748 * callback hasn't been called due to high load or due to a flawed
5751 * In practice, we will assume that if after a certain number of txgs a
5752 * commit callback hasn't been called, then most likely there's an
5753 * implementation bug..
5755 tmp_cb
= list_head(&zcl
.zcl_callbacks
);
5756 if (tmp_cb
!= NULL
&&
5757 tmp_cb
->zcd_txg
+ ZTEST_COMMIT_CB_THRESH
< txg
) {
5759 "Commit callback threshold exceeded, "
5760 "oldest txg: %"PRIu64
", open txg: %"PRIu64
"\n",
5761 tmp_cb
->zcd_txg
, txg
);
5765 * Let's find the place to insert our callbacks.
5767 * Even though the list is ordered by txg, it is possible for the
5768 * insertion point to not be the end because our txg may already be
5769 * quiescing at this point and other callbacks in the open txg
5770 * (from other objsets) may have sneaked in.
5772 tmp_cb
= list_tail(&zcl
.zcl_callbacks
);
5773 while (tmp_cb
!= NULL
&& tmp_cb
->zcd_txg
> txg
)
5774 tmp_cb
= list_prev(&zcl
.zcl_callbacks
, tmp_cb
);
5776 /* Add the 3 callbacks to the list */
5777 for (i
= 0; i
< 3; i
++) {
5779 list_insert_head(&zcl
.zcl_callbacks
, cb_data
[i
]);
5781 list_insert_after(&zcl
.zcl_callbacks
, tmp_cb
,
5784 cb_data
[i
]->zcd_added
= B_TRUE
;
5785 VERIFY(!cb_data
[i
]->zcd_called
);
5787 tmp_cb
= cb_data
[i
];
5792 (void) mutex_exit(&zcl
.zcl_callbacks_lock
);
5796 umem_free(od
, sizeof (ztest_od_t
));
5800 * Visit each object in the dataset. Verify that its properties
5801 * are consistent what was stored in the block tag when it was created,
5802 * and that its unused bonus buffer space has not been overwritten.
5805 ztest_verify_dnode_bt(ztest_ds_t
*zd
, uint64_t id
)
5808 objset_t
*os
= zd
->zd_os
;
5812 for (obj
= 0; err
== 0; err
= dmu_object_next(os
, &obj
, FALSE
, 0)) {
5813 ztest_block_tag_t
*bt
= NULL
;
5814 dmu_object_info_t doi
;
5817 ztest_object_lock(zd
, obj
, RL_READER
);
5818 if (dmu_bonus_hold(os
, obj
, FTAG
, &db
) != 0) {
5819 ztest_object_unlock(zd
, obj
);
5823 dmu_object_info_from_db(db
, &doi
);
5824 if (doi
.doi_bonus_size
>= sizeof (*bt
))
5825 bt
= ztest_bt_bonus(db
);
5827 if (bt
&& bt
->bt_magic
== BT_MAGIC
) {
5828 ztest_bt_verify(bt
, os
, obj
, doi
.doi_dnodesize
,
5829 bt
->bt_offset
, bt
->bt_gen
, bt
->bt_txg
,
5831 ztest_verify_unused_bonus(db
, bt
, obj
, os
, bt
->bt_gen
);
5834 dmu_buf_rele(db
, FTAG
);
5835 ztest_object_unlock(zd
, obj
);
5840 ztest_dsl_prop_get_set(ztest_ds_t
*zd
, uint64_t id
)
5843 zfs_prop_t proplist
[] = {
5845 ZFS_PROP_COMPRESSION
,
5850 (void) pthread_rwlock_rdlock(&ztest_name_lock
);
5852 for (int p
= 0; p
< sizeof (proplist
) / sizeof (proplist
[0]); p
++)
5853 (void) ztest_dsl_prop_set_uint64(zd
->zd_name
, proplist
[p
],
5854 ztest_random_dsl_prop(proplist
[p
]), (int)ztest_random(2));
5856 VERIFY0(ztest_dsl_prop_set_uint64(zd
->zd_name
, ZFS_PROP_RECORDSIZE
,
5857 ztest_random_blocksize(), (int)ztest_random(2)));
5859 (void) pthread_rwlock_unlock(&ztest_name_lock
);
5863 ztest_spa_prop_get_set(ztest_ds_t
*zd
, uint64_t id
)
5865 (void) zd
, (void) id
;
5866 nvlist_t
*props
= NULL
;
5868 (void) pthread_rwlock_rdlock(&ztest_name_lock
);
5870 (void) ztest_spa_prop_set_uint64(ZPOOL_PROP_AUTOTRIM
, ztest_random(2));
5872 VERIFY0(spa_prop_get(ztest_spa
, &props
));
5874 if (ztest_opts
.zo_verbose
>= 6)
5875 dump_nvlist(props
, 4);
5877 fnvlist_free(props
);
5879 (void) pthread_rwlock_unlock(&ztest_name_lock
);
5883 user_release_one(const char *snapname
, const char *holdname
)
5885 nvlist_t
*snaps
, *holds
;
5888 snaps
= fnvlist_alloc();
5889 holds
= fnvlist_alloc();
5890 fnvlist_add_boolean(holds
, holdname
);
5891 fnvlist_add_nvlist(snaps
, snapname
, holds
);
5892 fnvlist_free(holds
);
5893 error
= dsl_dataset_user_release(snaps
, NULL
);
5894 fnvlist_free(snaps
);
5899 * Test snapshot hold/release and deferred destroy.
5902 ztest_dmu_snapshot_hold(ztest_ds_t
*zd
, uint64_t id
)
5905 objset_t
*os
= zd
->zd_os
;
5909 char clonename
[100];
5911 char osname
[ZFS_MAX_DATASET_NAME_LEN
];
5914 (void) pthread_rwlock_rdlock(&ztest_name_lock
);
5916 dmu_objset_name(os
, osname
);
5918 (void) snprintf(snapname
, sizeof (snapname
), "sh1_%"PRIu64
"", id
);
5919 (void) snprintf(fullname
, sizeof (fullname
), "%s@%s", osname
, snapname
);
5920 (void) snprintf(clonename
, sizeof (clonename
), "%s/ch1_%"PRIu64
"",
5922 (void) snprintf(tag
, sizeof (tag
), "tag_%"PRIu64
"", id
);
5925 * Clean up from any previous run.
5927 error
= dsl_destroy_head(clonename
);
5928 if (error
!= ENOENT
)
5930 error
= user_release_one(fullname
, tag
);
5931 if (error
!= ESRCH
&& error
!= ENOENT
)
5933 error
= dsl_destroy_snapshot(fullname
, B_FALSE
);
5934 if (error
!= ENOENT
)
5938 * Create snapshot, clone it, mark snap for deferred destroy,
5939 * destroy clone, verify snap was also destroyed.
5941 error
= dmu_objset_snapshot_one(osname
, snapname
);
5943 if (error
== ENOSPC
) {
5944 ztest_record_enospc("dmu_objset_snapshot");
5947 fatal(B_FALSE
, "dmu_objset_snapshot(%s) = %d", fullname
, error
);
5950 error
= dmu_objset_clone(clonename
, fullname
);
5952 if (error
== ENOSPC
) {
5953 ztest_record_enospc("dmu_objset_clone");
5956 fatal(B_FALSE
, "dmu_objset_clone(%s) = %d", clonename
, error
);
5959 error
= dsl_destroy_snapshot(fullname
, B_TRUE
);
5961 fatal(B_FALSE
, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
5965 error
= dsl_destroy_head(clonename
);
5967 fatal(B_FALSE
, "dsl_destroy_head(%s) = %d", clonename
, error
);
5969 error
= dmu_objset_hold(fullname
, FTAG
, &origin
);
5970 if (error
!= ENOENT
)
5971 fatal(B_FALSE
, "dmu_objset_hold(%s) = %d", fullname
, error
);
5974 * Create snapshot, add temporary hold, verify that we can't
5975 * destroy a held snapshot, mark for deferred destroy,
5976 * release hold, verify snapshot was destroyed.
5978 error
= dmu_objset_snapshot_one(osname
, snapname
);
5980 if (error
== ENOSPC
) {
5981 ztest_record_enospc("dmu_objset_snapshot");
5984 fatal(B_FALSE
, "dmu_objset_snapshot(%s) = %d", fullname
, error
);
5987 holds
= fnvlist_alloc();
5988 fnvlist_add_string(holds
, fullname
, tag
);
5989 error
= dsl_dataset_user_hold(holds
, 0, NULL
);
5990 fnvlist_free(holds
);
5992 if (error
== ENOSPC
) {
5993 ztest_record_enospc("dsl_dataset_user_hold");
5996 fatal(B_FALSE
, "dsl_dataset_user_hold(%s, %s) = %u",
5997 fullname
, tag
, error
);
6000 error
= dsl_destroy_snapshot(fullname
, B_FALSE
);
6001 if (error
!= EBUSY
) {
6002 fatal(B_FALSE
, "dsl_destroy_snapshot(%s, B_FALSE) = %d",
6006 error
= dsl_destroy_snapshot(fullname
, B_TRUE
);
6008 fatal(B_FALSE
, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
6012 error
= user_release_one(fullname
, tag
);
6014 fatal(B_FALSE
, "user_release_one(%s, %s) = %d",
6015 fullname
, tag
, error
);
6017 VERIFY3U(dmu_objset_hold(fullname
, FTAG
, &origin
), ==, ENOENT
);
6020 (void) pthread_rwlock_unlock(&ztest_name_lock
);
6024 * Inject random faults into the on-disk data.
6027 ztest_fault_inject(ztest_ds_t
*zd
, uint64_t id
)
6029 (void) zd
, (void) id
;
6030 ztest_shared_t
*zs
= ztest_shared
;
6031 spa_t
*spa
= ztest_spa
;
6035 uint64_t bad
= 0x1990c0ffeedecadeull
;
6040 int bshift
= SPA_MAXBLOCKSHIFT
+ 2;
6046 boolean_t islog
= B_FALSE
;
6048 path0
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
6049 pathrand
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
6051 mutex_enter(&ztest_vdev_lock
);
6054 * Device removal is in progress, fault injection must be disabled
6055 * until it completes and the pool is scrubbed. The fault injection
6056 * strategy for damaging blocks does not take in to account evacuated
6057 * blocks which may have already been damaged.
6059 if (ztest_device_removal_active
) {
6060 mutex_exit(&ztest_vdev_lock
);
6064 maxfaults
= MAXFAULTS(zs
);
6065 leaves
= MAX(zs
->zs_mirrors
, 1) * ztest_opts
.zo_raid_children
;
6066 mirror_save
= zs
->zs_mirrors
;
6067 mutex_exit(&ztest_vdev_lock
);
6069 ASSERT3U(leaves
, >=, 1);
6072 * While ztest is running the number of leaves will not change. This
6073 * is critical for the fault injection logic as it determines where
6074 * errors can be safely injected such that they are always repairable.
6076 * When restarting ztest a different number of leaves may be requested
6077 * which will shift the regions to be damaged. This is fine as long
6078 * as the pool has been scrubbed prior to using the new mapping.
6079 * Failure to do can result in non-repairable damage being injected.
6081 if (ztest_pool_scrubbed
== B_FALSE
)
6085 * Grab the name lock as reader. There are some operations
6086 * which don't like to have their vdevs changed while
6087 * they are in progress (i.e. spa_change_guid). Those
6088 * operations will have grabbed the name lock as writer.
6090 (void) pthread_rwlock_rdlock(&ztest_name_lock
);
6093 * We need SCL_STATE here because we're going to look at vd0->vdev_tsd.
6095 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
6097 if (ztest_random(2) == 0) {
6099 * Inject errors on a normal data device or slog device.
6101 top
= ztest_random_vdev_top(spa
, B_TRUE
);
6102 leaf
= ztest_random(leaves
) + zs
->zs_splits
;
6105 * Generate paths to the first leaf in this top-level vdev,
6106 * and to the random leaf we selected. We'll induce transient
6107 * write failures and random online/offline activity on leaf 0,
6108 * and we'll write random garbage to the randomly chosen leaf.
6110 (void) snprintf(path0
, MAXPATHLEN
, ztest_dev_template
,
6111 ztest_opts
.zo_dir
, ztest_opts
.zo_pool
,
6112 top
* leaves
+ zs
->zs_splits
);
6113 (void) snprintf(pathrand
, MAXPATHLEN
, ztest_dev_template
,
6114 ztest_opts
.zo_dir
, ztest_opts
.zo_pool
,
6115 top
* leaves
+ leaf
);
6117 vd0
= vdev_lookup_by_path(spa
->spa_root_vdev
, path0
);
6118 if (vd0
!= NULL
&& vd0
->vdev_top
->vdev_islog
)
6122 * If the top-level vdev needs to be resilvered
6123 * then we only allow faults on the device that is
6126 if (vd0
!= NULL
&& maxfaults
!= 1 &&
6127 (!vdev_resilver_needed(vd0
->vdev_top
, NULL
, NULL
) ||
6128 vd0
->vdev_resilver_txg
!= 0)) {
6130 * Make vd0 explicitly claim to be unreadable,
6131 * or unwritable, or reach behind its back
6132 * and close the underlying fd. We can do this if
6133 * maxfaults == 0 because we'll fail and reexecute,
6134 * and we can do it if maxfaults >= 2 because we'll
6135 * have enough redundancy. If maxfaults == 1, the
6136 * combination of this with injection of random data
6137 * corruption below exceeds the pool's fault tolerance.
6139 vdev_file_t
*vf
= vd0
->vdev_tsd
;
6141 zfs_dbgmsg("injecting fault to vdev %llu; maxfaults=%d",
6142 (long long)vd0
->vdev_id
, (int)maxfaults
);
6144 if (vf
!= NULL
&& ztest_random(3) == 0) {
6145 (void) close(vf
->vf_file
->f_fd
);
6146 vf
->vf_file
->f_fd
= -1;
6147 } else if (ztest_random(2) == 0) {
6148 vd0
->vdev_cant_read
= B_TRUE
;
6150 vd0
->vdev_cant_write
= B_TRUE
;
6152 guid0
= vd0
->vdev_guid
;
6156 * Inject errors on an l2cache device.
6158 spa_aux_vdev_t
*sav
= &spa
->spa_l2cache
;
6160 if (sav
->sav_count
== 0) {
6161 spa_config_exit(spa
, SCL_STATE
, FTAG
);
6162 (void) pthread_rwlock_unlock(&ztest_name_lock
);
6165 vd0
= sav
->sav_vdevs
[ztest_random(sav
->sav_count
)];
6166 guid0
= vd0
->vdev_guid
;
6167 (void) strcpy(path0
, vd0
->vdev_path
);
6168 (void) strcpy(pathrand
, vd0
->vdev_path
);
6172 maxfaults
= INT_MAX
; /* no limit on cache devices */
6175 spa_config_exit(spa
, SCL_STATE
, FTAG
);
6176 (void) pthread_rwlock_unlock(&ztest_name_lock
);
6179 * If we can tolerate two or more faults, or we're dealing
6180 * with a slog, randomly online/offline vd0.
6182 if ((maxfaults
>= 2 || islog
) && guid0
!= 0) {
6183 if (ztest_random(10) < 6) {
6184 int flags
= (ztest_random(2) == 0 ?
6185 ZFS_OFFLINE_TEMPORARY
: 0);
6188 * We have to grab the zs_name_lock as writer to
6189 * prevent a race between offlining a slog and
6190 * destroying a dataset. Offlining the slog will
6191 * grab a reference on the dataset which may cause
6192 * dsl_destroy_head() to fail with EBUSY thus
6193 * leaving the dataset in an inconsistent state.
6196 (void) pthread_rwlock_wrlock(&ztest_name_lock
);
6198 VERIFY3U(vdev_offline(spa
, guid0
, flags
), !=, EBUSY
);
6201 (void) pthread_rwlock_unlock(&ztest_name_lock
);
6204 * Ideally we would like to be able to randomly
6205 * call vdev_[on|off]line without holding locks
6206 * to force unpredictable failures but the side
6207 * effects of vdev_[on|off]line prevent us from
6208 * doing so. We grab the ztest_vdev_lock here to
6209 * prevent a race between injection testing and
6212 mutex_enter(&ztest_vdev_lock
);
6213 (void) vdev_online(spa
, guid0
, 0, NULL
);
6214 mutex_exit(&ztest_vdev_lock
);
6222 * We have at least single-fault tolerance, so inject data corruption.
6224 fd
= open(pathrand
, O_RDWR
);
6226 if (fd
== -1) /* we hit a gap in the device namespace */
6229 fsize
= lseek(fd
, 0, SEEK_END
);
6231 while (--iters
!= 0) {
6233 * The offset must be chosen carefully to ensure that
6234 * we do not inject a given logical block with errors
6235 * on two different leaf devices, because ZFS can not
6236 * tolerate that (if maxfaults==1).
6238 * To achieve this we divide each leaf device into
6239 * chunks of size (# leaves * SPA_MAXBLOCKSIZE * 4).
6240 * Each chunk is further divided into error-injection
6241 * ranges (can accept errors) and clear ranges (we do
6242 * not inject errors in those). Each error-injection
6243 * range can accept errors only for a single leaf vdev.
6244 * Error-injection ranges are separated by clear ranges.
6246 * For example, with 3 leaves, each chunk looks like:
6247 * 0 to 32M: injection range for leaf 0
6248 * 32M to 64M: clear range - no injection allowed
6249 * 64M to 96M: injection range for leaf 1
6250 * 96M to 128M: clear range - no injection allowed
6251 * 128M to 160M: injection range for leaf 2
6252 * 160M to 192M: clear range - no injection allowed
6254 * Each clear range must be large enough such that a
6255 * single block cannot straddle it. This way a block
6256 * can't be a target in two different injection ranges
6257 * (on different leaf vdevs).
6259 offset
= ztest_random(fsize
/ (leaves
<< bshift
)) *
6260 (leaves
<< bshift
) + (leaf
<< bshift
) +
6261 (ztest_random(1ULL << (bshift
- 1)) & -8ULL);
6264 * Only allow damage to the labels at one end of the vdev.
6266 * If all labels are damaged, the device will be totally
6267 * inaccessible, which will result in loss of data,
6268 * because we also damage (parts of) the other side of
6271 * Additionally, we will always have both an even and an
6272 * odd label, so that we can handle crashes in the
6273 * middle of vdev_config_sync().
6275 if ((leaf
& 1) == 0 && offset
< VDEV_LABEL_START_SIZE
)
6279 * The two end labels are stored at the "end" of the disk, but
6280 * the end of the disk (vdev_psize) is aligned to
6281 * sizeof (vdev_label_t).
6283 uint64_t psize
= P2ALIGN(fsize
, sizeof (vdev_label_t
));
6284 if ((leaf
& 1) == 1 &&
6285 offset
+ sizeof (bad
) > psize
- VDEV_LABEL_END_SIZE
)
6288 mutex_enter(&ztest_vdev_lock
);
6289 if (mirror_save
!= zs
->zs_mirrors
) {
6290 mutex_exit(&ztest_vdev_lock
);
6295 if (pwrite(fd
, &bad
, sizeof (bad
), offset
) != sizeof (bad
))
6297 "can't inject bad word at 0x%"PRIx64
" in %s",
6300 mutex_exit(&ztest_vdev_lock
);
6302 if (ztest_opts
.zo_verbose
>= 7)
6303 (void) printf("injected bad word into %s,"
6304 " offset 0x%"PRIx64
"\n", pathrand
, offset
);
6309 umem_free(path0
, MAXPATHLEN
);
6310 umem_free(pathrand
, MAXPATHLEN
);
6314 * By design ztest will never inject uncorrectable damage in to the pool.
6315 * Issue a scrub, wait for it to complete, and verify there is never any
6316 * persistent damage.
6318 * Only after a full scrub has been completed is it safe to start injecting
6319 * data corruption. See the comment in zfs_fault_inject().
6322 ztest_scrub_impl(spa_t
*spa
)
6324 int error
= spa_scan(spa
, POOL_SCAN_SCRUB
);
6328 while (dsl_scan_scrubbing(spa_get_dsl(spa
)))
6329 txg_wait_synced(spa_get_dsl(spa
), 0);
6331 if (spa_get_errlog_size(spa
) > 0)
6334 ztest_pool_scrubbed
= B_TRUE
;
6343 ztest_scrub(ztest_ds_t
*zd
, uint64_t id
)
6345 (void) zd
, (void) id
;
6346 spa_t
*spa
= ztest_spa
;
6350 * Scrub in progress by device removal.
6352 if (ztest_device_removal_active
)
6356 * Start a scrub, wait a moment, then force a restart.
6358 (void) spa_scan(spa
, POOL_SCAN_SCRUB
);
6359 (void) poll(NULL
, 0, 100);
6361 error
= ztest_scrub_impl(spa
);
6368 * Change the guid for the pool.
6371 ztest_reguid(ztest_ds_t
*zd
, uint64_t id
)
6373 (void) zd
, (void) id
;
6374 spa_t
*spa
= ztest_spa
;
6375 uint64_t orig
, load
;
6378 if (ztest_opts
.zo_mmp_test
)
6381 orig
= spa_guid(spa
);
6382 load
= spa_load_guid(spa
);
6384 (void) pthread_rwlock_wrlock(&ztest_name_lock
);
6385 error
= spa_change_guid(spa
);
6386 (void) pthread_rwlock_unlock(&ztest_name_lock
);
6391 if (ztest_opts
.zo_verbose
>= 4) {
6392 (void) printf("Changed guid old %"PRIu64
" -> %"PRIu64
"\n",
6393 orig
, spa_guid(spa
));
6396 VERIFY3U(orig
, !=, spa_guid(spa
));
6397 VERIFY3U(load
, ==, spa_load_guid(spa
));
6401 ztest_fletcher(ztest_ds_t
*zd
, uint64_t id
)
6403 (void) zd
, (void) id
;
6404 hrtime_t end
= gethrtime() + NANOSEC
;
6406 while (gethrtime() <= end
) {
6407 int run_count
= 100;
6409 struct abd
*abd_data
, *abd_meta
;
6414 zio_cksum_t zc_ref_byteswap
;
6416 size
= ztest_random_blocksize();
6418 buf
= umem_alloc(size
, UMEM_NOFAIL
);
6419 abd_data
= abd_alloc(size
, B_FALSE
);
6420 abd_meta
= abd_alloc(size
, B_TRUE
);
6422 for (i
= 0, ptr
= buf
; i
< size
/ sizeof (*ptr
); i
++, ptr
++)
6423 *ptr
= ztest_random(UINT_MAX
);
6425 abd_copy_from_buf_off(abd_data
, buf
, 0, size
);
6426 abd_copy_from_buf_off(abd_meta
, buf
, 0, size
);
6428 VERIFY0(fletcher_4_impl_set("scalar"));
6429 fletcher_4_native(buf
, size
, NULL
, &zc_ref
);
6430 fletcher_4_byteswap(buf
, size
, NULL
, &zc_ref_byteswap
);
6432 VERIFY0(fletcher_4_impl_set("cycle"));
6433 while (run_count
-- > 0) {
6435 zio_cksum_t zc_byteswap
;
6437 fletcher_4_byteswap(buf
, size
, NULL
, &zc_byteswap
);
6438 fletcher_4_native(buf
, size
, NULL
, &zc
);
6440 VERIFY0(memcmp(&zc
, &zc_ref
, sizeof (zc
)));
6441 VERIFY0(memcmp(&zc_byteswap
, &zc_ref_byteswap
,
6442 sizeof (zc_byteswap
)));
6444 /* Test ABD - data */
6445 abd_fletcher_4_byteswap(abd_data
, size
, NULL
,
6447 abd_fletcher_4_native(abd_data
, size
, NULL
, &zc
);
6449 VERIFY0(memcmp(&zc
, &zc_ref
, sizeof (zc
)));
6450 VERIFY0(memcmp(&zc_byteswap
, &zc_ref_byteswap
,
6451 sizeof (zc_byteswap
)));
6453 /* Test ABD - metadata */
6454 abd_fletcher_4_byteswap(abd_meta
, size
, NULL
,
6456 abd_fletcher_4_native(abd_meta
, size
, NULL
, &zc
);
6458 VERIFY0(memcmp(&zc
, &zc_ref
, sizeof (zc
)));
6459 VERIFY0(memcmp(&zc_byteswap
, &zc_ref_byteswap
,
6460 sizeof (zc_byteswap
)));
6464 umem_free(buf
, size
);
6471 ztest_fletcher_incr(ztest_ds_t
*zd
, uint64_t id
)
6473 (void) zd
, (void) id
;
6479 zio_cksum_t zc_ref_bswap
;
6481 hrtime_t end
= gethrtime() + NANOSEC
;
6483 while (gethrtime() <= end
) {
6484 int run_count
= 100;
6486 size
= ztest_random_blocksize();
6487 buf
= umem_alloc(size
, UMEM_NOFAIL
);
6489 for (i
= 0, ptr
= buf
; i
< size
/ sizeof (*ptr
); i
++, ptr
++)
6490 *ptr
= ztest_random(UINT_MAX
);
6492 VERIFY0(fletcher_4_impl_set("scalar"));
6493 fletcher_4_native(buf
, size
, NULL
, &zc_ref
);
6494 fletcher_4_byteswap(buf
, size
, NULL
, &zc_ref_bswap
);
6496 VERIFY0(fletcher_4_impl_set("cycle"));
6498 while (run_count
-- > 0) {
6500 zio_cksum_t zc_bswap
;
6503 ZIO_SET_CHECKSUM(&zc
, 0, 0, 0, 0);
6504 ZIO_SET_CHECKSUM(&zc_bswap
, 0, 0, 0, 0);
6506 while (pos
< size
) {
6507 size_t inc
= 64 * ztest_random(size
/ 67);
6508 /* sometimes add few bytes to test non-simd */
6509 if (ztest_random(100) < 10)
6510 inc
+= P2ALIGN(ztest_random(64),
6513 if (inc
> (size
- pos
))
6516 fletcher_4_incremental_native(buf
+ pos
, inc
,
6518 fletcher_4_incremental_byteswap(buf
+ pos
, inc
,
6524 VERIFY3U(pos
, ==, size
);
6526 VERIFY(ZIO_CHECKSUM_EQUAL(zc
, zc_ref
));
6527 VERIFY(ZIO_CHECKSUM_EQUAL(zc_bswap
, zc_ref_bswap
));
6530 * verify if incremental on the whole buffer is
6531 * equivalent to non-incremental version
6533 ZIO_SET_CHECKSUM(&zc
, 0, 0, 0, 0);
6534 ZIO_SET_CHECKSUM(&zc_bswap
, 0, 0, 0, 0);
6536 fletcher_4_incremental_native(buf
, size
, &zc
);
6537 fletcher_4_incremental_byteswap(buf
, size
, &zc_bswap
);
6539 VERIFY(ZIO_CHECKSUM_EQUAL(zc
, zc_ref
));
6540 VERIFY(ZIO_CHECKSUM_EQUAL(zc_bswap
, zc_ref_bswap
));
6543 umem_free(buf
, size
);
6548 ztest_set_global_vars(void)
6550 for (size_t i
= 0; i
< ztest_opts
.zo_gvars_count
; i
++) {
6551 char *kv
= ztest_opts
.zo_gvars
[i
];
6552 VERIFY3U(strlen(kv
), <=, ZO_GVARS_MAX_ARGLEN
);
6553 VERIFY3U(strlen(kv
), >, 0);
6554 int err
= set_global_var(kv
);
6555 if (ztest_opts
.zo_verbose
> 0) {
6556 (void) printf("setting global var %s ... %s\n", kv
,
6557 err
? "failed" : "ok");
6560 (void) fprintf(stderr
,
6561 "failed to set global var '%s'\n", kv
);
6569 ztest_global_vars_to_zdb_args(void)
6571 char **args
= calloc(2*ztest_opts
.zo_gvars_count
+ 1, sizeof (char *));
6573 for (size_t i
= 0; i
< ztest_opts
.zo_gvars_count
; i
++) {
6574 char *kv
= ztest_opts
.zo_gvars
[i
];
6580 ASSERT3P(cur
, ==, &args
[2*ztest_opts
.zo_gvars_count
]);
6585 /* The end of strings is indicated by a NULL element */
6587 join_strings(char **strings
, const char *sep
)
6589 size_t totallen
= 0;
6590 for (char **sp
= strings
; *sp
!= NULL
; sp
++) {
6591 totallen
+= strlen(*sp
);
6592 totallen
+= strlen(sep
);
6595 ASSERT(totallen
>= strlen(sep
));
6596 totallen
-= strlen(sep
);
6599 size_t buflen
= totallen
+ 1;
6600 char *o
= malloc(buflen
); /* trailing 0 byte */
6602 for (char **sp
= strings
; *sp
!= NULL
; sp
++) {
6604 would
= strlcat(o
, *sp
, buflen
);
6605 VERIFY3U(would
, <, buflen
);
6606 if (*(sp
+1) == NULL
) {
6609 would
= strlcat(o
, sep
, buflen
);
6610 VERIFY3U(would
, <, buflen
);
6612 ASSERT3S(strlen(o
), ==, totallen
);
6617 ztest_check_path(char *path
)
6620 /* return true on success */
6621 return (!stat(path
, &s
));
6625 ztest_get_zdb_bin(char *bin
, int len
)
6629 * Try to use ZDB_PATH and in-tree zdb path. If not successful, just
6630 * let popen to search through PATH.
6632 if ((zdb_path
= getenv("ZDB_PATH"))) {
6633 strlcpy(bin
, zdb_path
, len
); /* In env */
6634 if (!ztest_check_path(bin
)) {
6635 ztest_dump_core
= 0;
6636 fatal(B_TRUE
, "invalid ZDB_PATH '%s'", bin
);
6641 VERIFY3P(realpath(getexecname(), bin
), !=, NULL
);
6642 if (strstr(bin
, "/ztest/")) {
6643 strstr(bin
, "/ztest/")[0] = '\0'; /* In-tree */
6644 strcat(bin
, "/zdb/zdb");
6645 if (ztest_check_path(bin
))
6652 ztest_random_concrete_vdev_leaf(vdev_t
*vd
)
6657 if (vd
->vdev_children
== 0)
6660 vdev_t
*eligible
[vd
->vdev_children
];
6661 int eligible_idx
= 0, i
;
6662 for (i
= 0; i
< vd
->vdev_children
; i
++) {
6663 vdev_t
*cvd
= vd
->vdev_child
[i
];
6664 if (cvd
->vdev_top
->vdev_removing
)
6666 if (cvd
->vdev_children
> 0 ||
6667 (vdev_is_concrete(cvd
) && !cvd
->vdev_detached
)) {
6668 eligible
[eligible_idx
++] = cvd
;
6671 VERIFY3S(eligible_idx
, >, 0);
6673 uint64_t child_no
= ztest_random(eligible_idx
);
6674 return (ztest_random_concrete_vdev_leaf(eligible
[child_no
]));
6678 ztest_initialize(ztest_ds_t
*zd
, uint64_t id
)
6680 (void) zd
, (void) id
;
6681 spa_t
*spa
= ztest_spa
;
6684 mutex_enter(&ztest_vdev_lock
);
6686 spa_config_enter(spa
, SCL_VDEV
, FTAG
, RW_READER
);
6688 /* Random leaf vdev */
6689 vdev_t
*rand_vd
= ztest_random_concrete_vdev_leaf(spa
->spa_root_vdev
);
6690 if (rand_vd
== NULL
) {
6691 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
6692 mutex_exit(&ztest_vdev_lock
);
6697 * The random vdev we've selected may change as soon as we
6698 * drop the spa_config_lock. We create local copies of things
6699 * we're interested in.
6701 uint64_t guid
= rand_vd
->vdev_guid
;
6702 char *path
= strdup(rand_vd
->vdev_path
);
6703 boolean_t active
= rand_vd
->vdev_initialize_thread
!= NULL
;
6705 zfs_dbgmsg("vd %px, guid %llu", rand_vd
, (u_longlong_t
)guid
);
6706 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
6708 uint64_t cmd
= ztest_random(POOL_INITIALIZE_FUNCS
);
6710 nvlist_t
*vdev_guids
= fnvlist_alloc();
6711 nvlist_t
*vdev_errlist
= fnvlist_alloc();
6712 fnvlist_add_uint64(vdev_guids
, path
, guid
);
6713 error
= spa_vdev_initialize(spa
, vdev_guids
, cmd
, vdev_errlist
);
6714 fnvlist_free(vdev_guids
);
6715 fnvlist_free(vdev_errlist
);
6718 case POOL_INITIALIZE_CANCEL
:
6719 if (ztest_opts
.zo_verbose
>= 4) {
6720 (void) printf("Cancel initialize %s", path
);
6722 (void) printf(" failed (no initialize active)");
6723 (void) printf("\n");
6726 case POOL_INITIALIZE_START
:
6727 if (ztest_opts
.zo_verbose
>= 4) {
6728 (void) printf("Start initialize %s", path
);
6729 if (active
&& error
== 0)
6730 (void) printf(" failed (already active)");
6731 else if (error
!= 0)
6732 (void) printf(" failed (error %d)", error
);
6733 (void) printf("\n");
6736 case POOL_INITIALIZE_SUSPEND
:
6737 if (ztest_opts
.zo_verbose
>= 4) {
6738 (void) printf("Suspend initialize %s", path
);
6740 (void) printf(" failed (no initialize active)");
6741 (void) printf("\n");
6746 mutex_exit(&ztest_vdev_lock
);
6750 ztest_trim(ztest_ds_t
*zd
, uint64_t id
)
6752 (void) zd
, (void) id
;
6753 spa_t
*spa
= ztest_spa
;
6756 mutex_enter(&ztest_vdev_lock
);
6758 spa_config_enter(spa
, SCL_VDEV
, FTAG
, RW_READER
);
6760 /* Random leaf vdev */
6761 vdev_t
*rand_vd
= ztest_random_concrete_vdev_leaf(spa
->spa_root_vdev
);
6762 if (rand_vd
== NULL
) {
6763 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
6764 mutex_exit(&ztest_vdev_lock
);
6769 * The random vdev we've selected may change as soon as we
6770 * drop the spa_config_lock. We create local copies of things
6771 * we're interested in.
6773 uint64_t guid
= rand_vd
->vdev_guid
;
6774 char *path
= strdup(rand_vd
->vdev_path
);
6775 boolean_t active
= rand_vd
->vdev_trim_thread
!= NULL
;
6777 zfs_dbgmsg("vd %p, guid %llu", rand_vd
, (u_longlong_t
)guid
);
6778 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
6780 uint64_t cmd
= ztest_random(POOL_TRIM_FUNCS
);
6781 uint64_t rate
= 1 << ztest_random(30);
6782 boolean_t partial
= (ztest_random(5) > 0);
6783 boolean_t secure
= (ztest_random(5) > 0);
6785 nvlist_t
*vdev_guids
= fnvlist_alloc();
6786 nvlist_t
*vdev_errlist
= fnvlist_alloc();
6787 fnvlist_add_uint64(vdev_guids
, path
, guid
);
6788 error
= spa_vdev_trim(spa
, vdev_guids
, cmd
, rate
, partial
,
6789 secure
, vdev_errlist
);
6790 fnvlist_free(vdev_guids
);
6791 fnvlist_free(vdev_errlist
);
6794 case POOL_TRIM_CANCEL
:
6795 if (ztest_opts
.zo_verbose
>= 4) {
6796 (void) printf("Cancel TRIM %s", path
);
6798 (void) printf(" failed (no TRIM active)");
6799 (void) printf("\n");
6802 case POOL_TRIM_START
:
6803 if (ztest_opts
.zo_verbose
>= 4) {
6804 (void) printf("Start TRIM %s", path
);
6805 if (active
&& error
== 0)
6806 (void) printf(" failed (already active)");
6807 else if (error
!= 0)
6808 (void) printf(" failed (error %d)", error
);
6809 (void) printf("\n");
6812 case POOL_TRIM_SUSPEND
:
6813 if (ztest_opts
.zo_verbose
>= 4) {
6814 (void) printf("Suspend TRIM %s", path
);
6816 (void) printf(" failed (no TRIM active)");
6817 (void) printf("\n");
6822 mutex_exit(&ztest_vdev_lock
);
6826 * Verify pool integrity by running zdb.
6829 ztest_run_zdb(char *pool
)
6835 const int len
= MAXPATHLEN
+ MAXNAMELEN
+ 20;
6838 bin
= umem_alloc(len
, UMEM_NOFAIL
);
6839 zdb
= umem_alloc(len
, UMEM_NOFAIL
);
6840 zbuf
= umem_alloc(1024, UMEM_NOFAIL
);
6842 ztest_get_zdb_bin(bin
, len
);
6844 char **set_gvars_args
= ztest_global_vars_to_zdb_args();
6845 char *set_gvars_args_joined
= join_strings(set_gvars_args
, " ");
6846 free(set_gvars_args
);
6848 size_t would
= snprintf(zdb
, len
,
6849 "%s -bcc%s%s -G -d -Y -e -y %s -p %s %s",
6851 ztest_opts
.zo_verbose
>= 3 ? "s" : "",
6852 ztest_opts
.zo_verbose
>= 4 ? "v" : "",
6853 set_gvars_args_joined
,
6856 ASSERT3U(would
, <, len
);
6858 free(set_gvars_args_joined
);
6860 if (ztest_opts
.zo_verbose
>= 5)
6861 (void) printf("Executing %s\n", strstr(zdb
, "zdb "));
6863 fp
= popen(zdb
, "r");
6865 while (fgets(zbuf
, 1024, fp
) != NULL
)
6866 if (ztest_opts
.zo_verbose
>= 3)
6867 (void) printf("%s", zbuf
);
6869 status
= pclose(fp
);
6874 ztest_dump_core
= 0;
6875 if (WIFEXITED(status
))
6876 fatal(B_FALSE
, "'%s' exit code %d", zdb
, WEXITSTATUS(status
));
6878 fatal(B_FALSE
, "'%s' died with signal %d",
6879 zdb
, WTERMSIG(status
));
6881 umem_free(bin
, len
);
6882 umem_free(zdb
, len
);
6883 umem_free(zbuf
, 1024);
6887 ztest_walk_pool_directory(char *header
)
6891 if (ztest_opts
.zo_verbose
>= 6)
6892 (void) printf("%s\n", header
);
6894 mutex_enter(&spa_namespace_lock
);
6895 while ((spa
= spa_next(spa
)) != NULL
)
6896 if (ztest_opts
.zo_verbose
>= 6)
6897 (void) printf("\t%s\n", spa_name(spa
));
6898 mutex_exit(&spa_namespace_lock
);
6902 ztest_spa_import_export(char *oldname
, char *newname
)
6904 nvlist_t
*config
, *newconfig
;
6909 if (ztest_opts
.zo_verbose
>= 4) {
6910 (void) printf("import/export: old = %s, new = %s\n",
6915 * Clean up from previous runs.
6917 (void) spa_destroy(newname
);
6920 * Get the pool's configuration and guid.
6922 VERIFY0(spa_open(oldname
, &spa
, FTAG
));
6925 * Kick off a scrub to tickle scrub/export races.
6927 if (ztest_random(2) == 0)
6928 (void) spa_scan(spa
, POOL_SCAN_SCRUB
);
6930 pool_guid
= spa_guid(spa
);
6931 spa_close(spa
, FTAG
);
6933 ztest_walk_pool_directory("pools before export");
6938 VERIFY0(spa_export(oldname
, &config
, B_FALSE
, B_FALSE
));
6940 ztest_walk_pool_directory("pools after export");
6945 newconfig
= spa_tryimport(config
);
6946 ASSERT3P(newconfig
, !=, NULL
);
6947 fnvlist_free(newconfig
);
6950 * Import it under the new name.
6952 error
= spa_import(newname
, config
, NULL
, 0);
6954 dump_nvlist(config
, 0);
6955 fatal(B_FALSE
, "couldn't import pool %s as %s: error %u",
6956 oldname
, newname
, error
);
6959 ztest_walk_pool_directory("pools after import");
6962 * Try to import it again -- should fail with EEXIST.
6964 VERIFY3U(EEXIST
, ==, spa_import(newname
, config
, NULL
, 0));
6967 * Try to import it under a different name -- should fail with EEXIST.
6969 VERIFY3U(EEXIST
, ==, spa_import(oldname
, config
, NULL
, 0));
6972 * Verify that the pool is no longer visible under the old name.
6974 VERIFY3U(ENOENT
, ==, spa_open(oldname
, &spa
, FTAG
));
6977 * Verify that we can open and close the pool using the new name.
6979 VERIFY0(spa_open(newname
, &spa
, FTAG
));
6980 ASSERT3U(pool_guid
, ==, spa_guid(spa
));
6981 spa_close(spa
, FTAG
);
6983 fnvlist_free(config
);
6987 ztest_resume(spa_t
*spa
)
6989 if (spa_suspended(spa
) && ztest_opts
.zo_verbose
>= 6)
6990 (void) printf("resuming from suspended state\n");
6991 spa_vdev_state_enter(spa
, SCL_NONE
);
6992 vdev_clear(spa
, NULL
);
6993 (void) spa_vdev_state_exit(spa
, NULL
, 0);
6994 (void) zio_resume(spa
);
6997 static _Noreturn
void
6998 ztest_resume_thread(void *arg
)
7002 while (!ztest_exiting
) {
7003 if (spa_suspended(spa
))
7005 (void) poll(NULL
, 0, 100);
7008 * Periodically change the zfs_compressed_arc_enabled setting.
7010 if (ztest_random(10) == 0)
7011 zfs_compressed_arc_enabled
= ztest_random(2);
7014 * Periodically change the zfs_abd_scatter_enabled setting.
7016 if (ztest_random(10) == 0)
7017 zfs_abd_scatter_enabled
= ztest_random(2);
7023 static _Noreturn
void
7024 ztest_deadman_thread(void *arg
)
7026 ztest_shared_t
*zs
= arg
;
7027 spa_t
*spa
= ztest_spa
;
7028 hrtime_t delay
, overdue
, last_run
= gethrtime();
7030 delay
= (zs
->zs_thread_stop
- zs
->zs_thread_start
) +
7031 MSEC2NSEC(zfs_deadman_synctime_ms
);
7033 while (!ztest_exiting
) {
7035 * Wait for the delay timer while checking occasionally
7036 * if we should stop.
7038 if (gethrtime() < last_run
+ delay
) {
7039 (void) poll(NULL
, 0, 1000);
7044 * If the pool is suspended then fail immediately. Otherwise,
7045 * check to see if the pool is making any progress. If
7046 * vdev_deadman() discovers that there hasn't been any recent
7047 * I/Os then it will end up aborting the tests.
7049 if (spa_suspended(spa
) || spa
->spa_root_vdev
== NULL
) {
7051 "aborting test after %lu seconds because "
7052 "pool has transitioned to a suspended state.",
7053 zfs_deadman_synctime_ms
/ 1000);
7055 vdev_deadman(spa
->spa_root_vdev
, FTAG
);
7058 * If the process doesn't complete within a grace period of
7059 * zfs_deadman_synctime_ms over the expected finish time,
7060 * then it may be hung and is terminated.
7062 overdue
= zs
->zs_proc_stop
+ MSEC2NSEC(zfs_deadman_synctime_ms
);
7063 if (gethrtime() > overdue
) {
7065 "aborting test after %llu seconds because "
7066 "the process is overdue for termination.",
7067 (gethrtime() - zs
->zs_proc_start
) / NANOSEC
);
7070 (void) printf("ztest has been running for %lld seconds\n",
7071 (gethrtime() - zs
->zs_proc_start
) / NANOSEC
);
7073 last_run
= gethrtime();
7074 delay
= MSEC2NSEC(zfs_deadman_checktime_ms
);
7081 ztest_execute(int test
, ztest_info_t
*zi
, uint64_t id
)
7083 ztest_ds_t
*zd
= &ztest_ds
[id
% ztest_opts
.zo_datasets
];
7084 ztest_shared_callstate_t
*zc
= ZTEST_GET_SHARED_CALLSTATE(test
);
7085 hrtime_t functime
= gethrtime();
7088 for (i
= 0; i
< zi
->zi_iters
; i
++)
7089 zi
->zi_func(zd
, id
);
7091 functime
= gethrtime() - functime
;
7093 atomic_add_64(&zc
->zc_count
, 1);
7094 atomic_add_64(&zc
->zc_time
, functime
);
7096 if (ztest_opts
.zo_verbose
>= 4)
7097 (void) printf("%6.2f sec in %s\n",
7098 (double)functime
/ NANOSEC
, zi
->zi_funcname
);
7101 static _Noreturn
void
7102 ztest_thread(void *arg
)
7105 uint64_t id
= (uintptr_t)arg
;
7106 ztest_shared_t
*zs
= ztest_shared
;
7110 ztest_shared_callstate_t
*zc
;
7112 while ((now
= gethrtime()) < zs
->zs_thread_stop
) {
7114 * See if it's time to force a crash.
7116 if (now
> zs
->zs_thread_kill
)
7120 * If we're getting ENOSPC with some regularity, stop.
7122 if (zs
->zs_enospc_count
> 10)
7126 * Pick a random function to execute.
7128 rand
= ztest_random(ZTEST_FUNCS
);
7129 zi
= &ztest_info
[rand
];
7130 zc
= ZTEST_GET_SHARED_CALLSTATE(rand
);
7131 call_next
= zc
->zc_next
;
7133 if (now
>= call_next
&&
7134 atomic_cas_64(&zc
->zc_next
, call_next
, call_next
+
7135 ztest_random(2 * zi
->zi_interval
[0] + 1)) == call_next
) {
7136 ztest_execute(rand
, zi
, id
);
7144 ztest_dataset_name(char *dsname
, char *pool
, int d
)
7146 (void) snprintf(dsname
, ZFS_MAX_DATASET_NAME_LEN
, "%s/ds_%d", pool
, d
);
7150 ztest_dataset_destroy(int d
)
7152 char name
[ZFS_MAX_DATASET_NAME_LEN
];
7155 ztest_dataset_name(name
, ztest_opts
.zo_pool
, d
);
7157 if (ztest_opts
.zo_verbose
>= 3)
7158 (void) printf("Destroying %s to free up space\n", name
);
7161 * Cleanup any non-standard clones and snapshots. In general,
7162 * ztest thread t operates on dataset (t % zopt_datasets),
7163 * so there may be more than one thing to clean up.
7165 for (t
= d
; t
< ztest_opts
.zo_threads
;
7166 t
+= ztest_opts
.zo_datasets
)
7167 ztest_dsl_dataset_cleanup(name
, t
);
7169 (void) dmu_objset_find(name
, ztest_objset_destroy_cb
, NULL
,
7170 DS_FIND_SNAPSHOTS
| DS_FIND_CHILDREN
);
7174 ztest_dataset_dirobj_verify(ztest_ds_t
*zd
)
7176 uint64_t usedobjs
, dirobjs
, scratch
;
7179 * ZTEST_DIROBJ is the object directory for the entire dataset.
7180 * Therefore, the number of objects in use should equal the
7181 * number of ZTEST_DIROBJ entries, +1 for ZTEST_DIROBJ itself.
7182 * If not, we have an object leak.
7184 * Note that we can only check this in ztest_dataset_open(),
7185 * when the open-context and syncing-context values agree.
7186 * That's because zap_count() returns the open-context value,
7187 * while dmu_objset_space() returns the rootbp fill count.
7189 VERIFY0(zap_count(zd
->zd_os
, ZTEST_DIROBJ
, &dirobjs
));
7190 dmu_objset_space(zd
->zd_os
, &scratch
, &scratch
, &usedobjs
, &scratch
);
7191 ASSERT3U(dirobjs
+ 1, ==, usedobjs
);
7195 ztest_dataset_open(int d
)
7197 ztest_ds_t
*zd
= &ztest_ds
[d
];
7198 uint64_t committed_seq
= ZTEST_GET_SHARED_DS(d
)->zd_seq
;
7201 char name
[ZFS_MAX_DATASET_NAME_LEN
];
7204 ztest_dataset_name(name
, ztest_opts
.zo_pool
, d
);
7206 (void) pthread_rwlock_rdlock(&ztest_name_lock
);
7208 error
= ztest_dataset_create(name
);
7209 if (error
== ENOSPC
) {
7210 (void) pthread_rwlock_unlock(&ztest_name_lock
);
7211 ztest_record_enospc(FTAG
);
7214 ASSERT(error
== 0 || error
== EEXIST
);
7216 VERIFY0(ztest_dmu_objset_own(name
, DMU_OST_OTHER
, B_FALSE
,
7218 (void) pthread_rwlock_unlock(&ztest_name_lock
);
7220 ztest_zd_init(zd
, ZTEST_GET_SHARED_DS(d
), os
);
7222 zilog
= zd
->zd_zilog
;
7224 if (zilog
->zl_header
->zh_claim_lr_seq
!= 0 &&
7225 zilog
->zl_header
->zh_claim_lr_seq
< committed_seq
)
7226 fatal(B_FALSE
, "missing log records: "
7227 "claimed %"PRIu64
" < committed %"PRIu64
"",
7228 zilog
->zl_header
->zh_claim_lr_seq
, committed_seq
);
7230 ztest_dataset_dirobj_verify(zd
);
7232 zil_replay(os
, zd
, ztest_replay_vector
);
7234 ztest_dataset_dirobj_verify(zd
);
7236 if (ztest_opts
.zo_verbose
>= 6)
7237 (void) printf("%s replay %"PRIu64
" blocks, "
7238 "%"PRIu64
" records, seq %"PRIu64
"\n",
7240 zilog
->zl_parse_blk_count
,
7241 zilog
->zl_parse_lr_count
,
7242 zilog
->zl_replaying_seq
);
7244 zilog
= zil_open(os
, ztest_get_data
);
7246 if (zilog
->zl_replaying_seq
!= 0 &&
7247 zilog
->zl_replaying_seq
< committed_seq
)
7248 fatal(B_FALSE
, "missing log records: "
7249 "replayed %"PRIu64
" < committed %"PRIu64
"",
7250 zilog
->zl_replaying_seq
, committed_seq
);
7256 ztest_dataset_close(int d
)
7258 ztest_ds_t
*zd
= &ztest_ds
[d
];
7260 zil_close(zd
->zd_zilog
);
7261 dmu_objset_disown(zd
->zd_os
, B_TRUE
, zd
);
7267 ztest_replay_zil_cb(const char *name
, void *arg
)
7273 VERIFY0(ztest_dmu_objset_own(name
, DMU_OST_ANY
, B_TRUE
,
7274 B_TRUE
, FTAG
, &os
));
7276 zdtmp
= umem_alloc(sizeof (ztest_ds_t
), UMEM_NOFAIL
);
7278 ztest_zd_init(zdtmp
, NULL
, os
);
7279 zil_replay(os
, zdtmp
, ztest_replay_vector
);
7280 ztest_zd_fini(zdtmp
);
7282 if (dmu_objset_zil(os
)->zl_parse_lr_count
!= 0 &&
7283 ztest_opts
.zo_verbose
>= 6) {
7284 zilog_t
*zilog
= dmu_objset_zil(os
);
7286 (void) printf("%s replay %"PRIu64
" blocks, "
7287 "%"PRIu64
" records, seq %"PRIu64
"\n",
7289 zilog
->zl_parse_blk_count
,
7290 zilog
->zl_parse_lr_count
,
7291 zilog
->zl_replaying_seq
);
7294 umem_free(zdtmp
, sizeof (ztest_ds_t
));
7296 dmu_objset_disown(os
, B_TRUE
, FTAG
);
7303 ztest_ds_t
*zd
= &ztest_ds
[0];
7307 if (ztest_opts
.zo_verbose
>= 3)
7308 (void) printf("testing spa_freeze()...\n");
7310 kernel_init(SPA_MODE_READ
| SPA_MODE_WRITE
);
7311 VERIFY0(spa_open(ztest_opts
.zo_pool
, &spa
, FTAG
));
7312 VERIFY0(ztest_dataset_open(0));
7316 * Force the first log block to be transactionally allocated.
7317 * We have to do this before we freeze the pool -- otherwise
7318 * the log chain won't be anchored.
7320 while (BP_IS_HOLE(&zd
->zd_zilog
->zl_header
->zh_log
)) {
7321 ztest_dmu_object_alloc_free(zd
, 0);
7322 zil_commit(zd
->zd_zilog
, 0);
7325 txg_wait_synced(spa_get_dsl(spa
), 0);
7328 * Freeze the pool. This stops spa_sync() from doing anything,
7329 * so that the only way to record changes from now on is the ZIL.
7334 * Because it is hard to predict how much space a write will actually
7335 * require beforehand, we leave ourselves some fudge space to write over
7338 uint64_t capacity
= metaslab_class_get_space(spa_normal_class(spa
)) / 2;
7341 * Run tests that generate log records but don't alter the pool config
7342 * or depend on DSL sync tasks (snapshots, objset create/destroy, etc).
7343 * We do a txg_wait_synced() after each iteration to force the txg
7344 * to increase well beyond the last synced value in the uberblock.
7345 * The ZIL should be OK with that.
7347 * Run a random number of times less than zo_maxloops and ensure we do
7348 * not run out of space on the pool.
7350 while (ztest_random(10) != 0 &&
7351 numloops
++ < ztest_opts
.zo_maxloops
&&
7352 metaslab_class_get_alloc(spa_normal_class(spa
)) < capacity
) {
7354 ztest_od_init(&od
, 0, FTAG
, 0, DMU_OT_UINT64_OTHER
, 0, 0, 0);
7355 VERIFY0(ztest_object_init(zd
, &od
, sizeof (od
), B_FALSE
));
7356 ztest_io(zd
, od
.od_object
,
7357 ztest_random(ZTEST_RANGE_LOCKS
) << SPA_MAXBLOCKSHIFT
);
7358 txg_wait_synced(spa_get_dsl(spa
), 0);
7362 * Commit all of the changes we just generated.
7364 zil_commit(zd
->zd_zilog
, 0);
7365 txg_wait_synced(spa_get_dsl(spa
), 0);
7368 * Close our dataset and close the pool.
7370 ztest_dataset_close(0);
7371 spa_close(spa
, FTAG
);
7375 * Open and close the pool and dataset to induce log replay.
7377 kernel_init(SPA_MODE_READ
| SPA_MODE_WRITE
);
7378 VERIFY0(spa_open(ztest_opts
.zo_pool
, &spa
, FTAG
));
7379 ASSERT3U(spa_freeze_txg(spa
), ==, UINT64_MAX
);
7380 VERIFY0(ztest_dataset_open(0));
7382 txg_wait_synced(spa_get_dsl(spa
), 0);
7383 ztest_dataset_close(0);
7384 ztest_reguid(NULL
, 0);
7386 spa_close(spa
, FTAG
);
7391 ztest_import_impl(void)
7393 importargs_t args
= { 0 };
7394 nvlist_t
*cfg
= NULL
;
7396 char *searchdirs
[nsearch
];
7397 int flags
= ZFS_IMPORT_MISSING_LOG
;
7399 searchdirs
[0] = ztest_opts
.zo_dir
;
7400 args
.paths
= nsearch
;
7401 args
.path
= searchdirs
;
7402 args
.can_be_active
= B_FALSE
;
7404 VERIFY0(zpool_find_config(NULL
, ztest_opts
.zo_pool
, &cfg
, &args
,
7405 &libzpool_config_ops
));
7406 VERIFY0(spa_import(ztest_opts
.zo_pool
, cfg
, NULL
, flags
));
7411 * Import a storage pool with the given name.
7414 ztest_import(ztest_shared_t
*zs
)
7418 mutex_init(&ztest_vdev_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
7419 mutex_init(&ztest_checkpoint_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
7420 VERIFY0(pthread_rwlock_init(&ztest_name_lock
, NULL
));
7422 kernel_init(SPA_MODE_READ
| SPA_MODE_WRITE
);
7424 ztest_import_impl();
7426 VERIFY0(spa_open(ztest_opts
.zo_pool
, &spa
, FTAG
));
7427 zs
->zs_metaslab_sz
=
7428 1ULL << spa
->spa_root_vdev
->vdev_child
[0]->vdev_ms_shift
;
7429 spa_close(spa
, FTAG
);
7433 if (!ztest_opts
.zo_mmp_test
) {
7434 ztest_run_zdb(ztest_opts
.zo_pool
);
7436 ztest_run_zdb(ztest_opts
.zo_pool
);
7439 (void) pthread_rwlock_destroy(&ztest_name_lock
);
7440 mutex_destroy(&ztest_vdev_lock
);
7441 mutex_destroy(&ztest_checkpoint_lock
);
7445 * Kick off threads to run tests on all datasets in parallel.
7448 ztest_run(ztest_shared_t
*zs
)
7452 kthread_t
*resume_thread
, *deadman_thread
;
7453 kthread_t
**run_threads
;
7458 ztest_exiting
= B_FALSE
;
7461 * Initialize parent/child shared state.
7463 mutex_init(&ztest_vdev_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
7464 mutex_init(&ztest_checkpoint_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
7465 VERIFY0(pthread_rwlock_init(&ztest_name_lock
, NULL
));
7467 zs
->zs_thread_start
= gethrtime();
7468 zs
->zs_thread_stop
=
7469 zs
->zs_thread_start
+ ztest_opts
.zo_passtime
* NANOSEC
;
7470 zs
->zs_thread_stop
= MIN(zs
->zs_thread_stop
, zs
->zs_proc_stop
);
7471 zs
->zs_thread_kill
= zs
->zs_thread_stop
;
7472 if (ztest_random(100) < ztest_opts
.zo_killrate
) {
7473 zs
->zs_thread_kill
-=
7474 ztest_random(ztest_opts
.zo_passtime
* NANOSEC
);
7477 mutex_init(&zcl
.zcl_callbacks_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
7479 list_create(&zcl
.zcl_callbacks
, sizeof (ztest_cb_data_t
),
7480 offsetof(ztest_cb_data_t
, zcd_node
));
7483 * Open our pool. It may need to be imported first depending on
7484 * what tests were running when the previous pass was terminated.
7486 kernel_init(SPA_MODE_READ
| SPA_MODE_WRITE
);
7487 error
= spa_open(ztest_opts
.zo_pool
, &spa
, FTAG
);
7489 VERIFY3S(error
, ==, ENOENT
);
7490 ztest_import_impl();
7491 VERIFY0(spa_open(ztest_opts
.zo_pool
, &spa
, FTAG
));
7492 zs
->zs_metaslab_sz
=
7493 1ULL << spa
->spa_root_vdev
->vdev_child
[0]->vdev_ms_shift
;
7496 metaslab_preload_limit
= ztest_random(20) + 1;
7499 VERIFY0(vdev_raidz_impl_set("cycle"));
7501 dmu_objset_stats_t dds
;
7502 VERIFY0(ztest_dmu_objset_own(ztest_opts
.zo_pool
,
7503 DMU_OST_ANY
, B_TRUE
, B_TRUE
, FTAG
, &os
));
7504 dsl_pool_config_enter(dmu_objset_pool(os
), FTAG
);
7505 dmu_objset_fast_stat(os
, &dds
);
7506 dsl_pool_config_exit(dmu_objset_pool(os
), FTAG
);
7507 zs
->zs_guid
= dds
.dds_guid
;
7508 dmu_objset_disown(os
, B_TRUE
, FTAG
);
7511 * Create a thread to periodically resume suspended I/O.
7513 resume_thread
= thread_create(NULL
, 0, ztest_resume_thread
,
7514 spa
, 0, NULL
, TS_RUN
| TS_JOINABLE
, defclsyspri
);
7517 * Create a deadman thread and set to panic if we hang.
7519 deadman_thread
= thread_create(NULL
, 0, ztest_deadman_thread
,
7520 zs
, 0, NULL
, TS_RUN
| TS_JOINABLE
, defclsyspri
);
7522 spa
->spa_deadman_failmode
= ZIO_FAILURE_MODE_PANIC
;
7525 * Verify that we can safely inquire about any object,
7526 * whether it's allocated or not. To make it interesting,
7527 * we probe a 5-wide window around each power of two.
7528 * This hits all edge cases, including zero and the max.
7530 for (t
= 0; t
< 64; t
++) {
7531 for (d
= -5; d
<= 5; d
++) {
7532 error
= dmu_object_info(spa
->spa_meta_objset
,
7533 (1ULL << t
) + d
, NULL
);
7534 ASSERT(error
== 0 || error
== ENOENT
||
7540 * If we got any ENOSPC errors on the previous run, destroy something.
7542 if (zs
->zs_enospc_count
!= 0) {
7543 int d
= ztest_random(ztest_opts
.zo_datasets
);
7544 ztest_dataset_destroy(d
);
7546 zs
->zs_enospc_count
= 0;
7549 * If we were in the middle of ztest_device_removal() and were killed
7550 * we need to ensure the removal and scrub complete before running
7551 * any tests that check ztest_device_removal_active. The removal will
7552 * be restarted automatically when the spa is opened, but we need to
7553 * initiate the scrub manually if it is not already in progress. Note
7554 * that we always run the scrub whenever an indirect vdev exists
7555 * because we have no way of knowing for sure if ztest_device_removal()
7556 * fully completed its scrub before the pool was reimported.
7558 if (spa
->spa_removing_phys
.sr_state
== DSS_SCANNING
||
7559 spa
->spa_removing_phys
.sr_prev_indirect_vdev
!= -1) {
7560 while (spa
->spa_removing_phys
.sr_state
== DSS_SCANNING
)
7561 txg_wait_synced(spa_get_dsl(spa
), 0);
7563 error
= ztest_scrub_impl(spa
);
7569 run_threads
= umem_zalloc(ztest_opts
.zo_threads
* sizeof (kthread_t
*),
7572 if (ztest_opts
.zo_verbose
>= 4)
7573 (void) printf("starting main threads...\n");
7576 * Replay all logs of all datasets in the pool. This is primarily for
7577 * temporary datasets which wouldn't otherwise get replayed, which
7578 * can trigger failures when attempting to offline a SLOG in
7579 * ztest_fault_inject().
7581 (void) dmu_objset_find(ztest_opts
.zo_pool
, ztest_replay_zil_cb
,
7582 NULL
, DS_FIND_CHILDREN
);
7585 * Kick off all the tests that run in parallel.
7587 for (t
= 0; t
< ztest_opts
.zo_threads
; t
++) {
7588 if (t
< ztest_opts
.zo_datasets
&& ztest_dataset_open(t
) != 0) {
7589 umem_free(run_threads
, ztest_opts
.zo_threads
*
7590 sizeof (kthread_t
*));
7594 run_threads
[t
] = thread_create(NULL
, 0, ztest_thread
,
7595 (void *)(uintptr_t)t
, 0, NULL
, TS_RUN
| TS_JOINABLE
,
7600 * Wait for all of the tests to complete.
7602 for (t
= 0; t
< ztest_opts
.zo_threads
; t
++)
7603 VERIFY0(thread_join(run_threads
[t
]));
7606 * Close all datasets. This must be done after all the threads
7607 * are joined so we can be sure none of the datasets are in-use
7608 * by any of the threads.
7610 for (t
= 0; t
< ztest_opts
.zo_threads
; t
++) {
7611 if (t
< ztest_opts
.zo_datasets
)
7612 ztest_dataset_close(t
);
7615 txg_wait_synced(spa_get_dsl(spa
), 0);
7617 zs
->zs_alloc
= metaslab_class_get_alloc(spa_normal_class(spa
));
7618 zs
->zs_space
= metaslab_class_get_space(spa_normal_class(spa
));
7620 umem_free(run_threads
, ztest_opts
.zo_threads
* sizeof (kthread_t
*));
7622 /* Kill the resume and deadman threads */
7623 ztest_exiting
= B_TRUE
;
7624 VERIFY0(thread_join(resume_thread
));
7625 VERIFY0(thread_join(deadman_thread
));
7629 * Right before closing the pool, kick off a bunch of async I/O;
7630 * spa_close() should wait for it to complete.
7632 for (object
= 1; object
< 50; object
++) {
7633 dmu_prefetch(spa
->spa_meta_objset
, object
, 0, 0, 1ULL << 20,
7634 ZIO_PRIORITY_SYNC_READ
);
7637 /* Verify that at least one commit cb was called in a timely fashion */
7638 if (zc_cb_counter
>= ZTEST_COMMIT_CB_MIN_REG
)
7639 VERIFY0(zc_min_txg_delay
);
7641 spa_close(spa
, FTAG
);
7644 * Verify that we can loop over all pools.
7646 mutex_enter(&spa_namespace_lock
);
7647 for (spa
= spa_next(NULL
); spa
!= NULL
; spa
= spa_next(spa
))
7648 if (ztest_opts
.zo_verbose
> 3)
7649 (void) printf("spa_next: found %s\n", spa_name(spa
));
7650 mutex_exit(&spa_namespace_lock
);
7653 * Verify that we can export the pool and reimport it under a
7656 if ((ztest_random(2) == 0) && !ztest_opts
.zo_mmp_test
) {
7657 char name
[ZFS_MAX_DATASET_NAME_LEN
];
7658 (void) snprintf(name
, sizeof (name
), "%s_import",
7659 ztest_opts
.zo_pool
);
7660 ztest_spa_import_export(ztest_opts
.zo_pool
, name
);
7661 ztest_spa_import_export(name
, ztest_opts
.zo_pool
);
7666 list_destroy(&zcl
.zcl_callbacks
);
7667 mutex_destroy(&zcl
.zcl_callbacks_lock
);
7668 (void) pthread_rwlock_destroy(&ztest_name_lock
);
7669 mutex_destroy(&ztest_vdev_lock
);
7670 mutex_destroy(&ztest_checkpoint_lock
);
7674 print_time(hrtime_t t
, char *timebuf
)
7676 hrtime_t s
= t
/ NANOSEC
;
7677 hrtime_t m
= s
/ 60;
7678 hrtime_t h
= m
/ 60;
7679 hrtime_t d
= h
/ 24;
7688 (void) sprintf(timebuf
,
7689 "%llud%02lluh%02llum%02llus", d
, h
, m
, s
);
7691 (void) sprintf(timebuf
, "%lluh%02llum%02llus", h
, m
, s
);
7693 (void) sprintf(timebuf
, "%llum%02llus", m
, s
);
7695 (void) sprintf(timebuf
, "%llus", s
);
7699 make_random_props(void)
7703 props
= fnvlist_alloc();
7705 if (ztest_random(2) == 0)
7708 fnvlist_add_uint64(props
,
7709 zpool_prop_to_name(ZPOOL_PROP_AUTOREPLACE
), 1);
7715 * Create a storage pool with the given name and initial vdev size.
7716 * Then test spa_freeze() functionality.
7719 ztest_init(ztest_shared_t
*zs
)
7722 nvlist_t
*nvroot
, *props
;
7725 mutex_init(&ztest_vdev_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
7726 mutex_init(&ztest_checkpoint_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
7727 VERIFY0(pthread_rwlock_init(&ztest_name_lock
, NULL
));
7729 kernel_init(SPA_MODE_READ
| SPA_MODE_WRITE
);
7732 * Create the storage pool.
7734 (void) spa_destroy(ztest_opts
.zo_pool
);
7735 ztest_shared
->zs_vdev_next_leaf
= 0;
7737 zs
->zs_mirrors
= ztest_opts
.zo_mirrors
;
7738 nvroot
= make_vdev_root(NULL
, NULL
, NULL
, ztest_opts
.zo_vdev_size
, 0,
7739 NULL
, ztest_opts
.zo_raid_children
, zs
->zs_mirrors
, 1);
7740 props
= make_random_props();
7743 * We don't expect the pool to suspend unless maxfaults == 0,
7744 * in which case ztest_fault_inject() temporarily takes away
7745 * the only valid replica.
7747 fnvlist_add_uint64(props
,
7748 zpool_prop_to_name(ZPOOL_PROP_FAILUREMODE
),
7749 MAXFAULTS(zs
) ? ZIO_FAILURE_MODE_PANIC
: ZIO_FAILURE_MODE_WAIT
);
7751 for (i
= 0; i
< SPA_FEATURES
; i
++) {
7754 if (!spa_feature_table
[i
].fi_zfs_mod_supported
)
7758 * 75% chance of using the log space map feature. We want ztest
7759 * to exercise both the code paths that use the log space map
7760 * feature and the ones that don't.
7762 if (i
== SPA_FEATURE_LOG_SPACEMAP
&& ztest_random(4) == 0)
7765 VERIFY3S(-1, !=, asprintf(&buf
, "feature@%s",
7766 spa_feature_table
[i
].fi_uname
));
7767 fnvlist_add_uint64(props
, buf
, 0);
7771 VERIFY0(spa_create(ztest_opts
.zo_pool
, nvroot
, props
, NULL
, NULL
));
7772 fnvlist_free(nvroot
);
7773 fnvlist_free(props
);
7775 VERIFY0(spa_open(ztest_opts
.zo_pool
, &spa
, FTAG
));
7776 zs
->zs_metaslab_sz
=
7777 1ULL << spa
->spa_root_vdev
->vdev_child
[0]->vdev_ms_shift
;
7778 spa_close(spa
, FTAG
);
7782 if (!ztest_opts
.zo_mmp_test
) {
7783 ztest_run_zdb(ztest_opts
.zo_pool
);
7785 ztest_run_zdb(ztest_opts
.zo_pool
);
7788 (void) pthread_rwlock_destroy(&ztest_name_lock
);
7789 mutex_destroy(&ztest_vdev_lock
);
7790 mutex_destroy(&ztest_checkpoint_lock
);
7796 static char ztest_name_data
[] = "/tmp/ztest.data.XXXXXX";
7798 ztest_fd_data
= mkstemp(ztest_name_data
);
7799 ASSERT3S(ztest_fd_data
, >=, 0);
7800 (void) unlink(ztest_name_data
);
7804 shared_data_size(ztest_shared_hdr_t
*hdr
)
7808 size
= hdr
->zh_hdr_size
;
7809 size
+= hdr
->zh_opts_size
;
7810 size
+= hdr
->zh_size
;
7811 size
+= hdr
->zh_stats_size
* hdr
->zh_stats_count
;
7812 size
+= hdr
->zh_ds_size
* hdr
->zh_ds_count
;
7821 ztest_shared_hdr_t
*hdr
;
7823 hdr
= (void *)mmap(0, P2ROUNDUP(sizeof (*hdr
), getpagesize()),
7824 PROT_READ
| PROT_WRITE
, MAP_SHARED
, ztest_fd_data
, 0);
7825 ASSERT3P(hdr
, !=, MAP_FAILED
);
7827 VERIFY0(ftruncate(ztest_fd_data
, sizeof (ztest_shared_hdr_t
)));
7829 hdr
->zh_hdr_size
= sizeof (ztest_shared_hdr_t
);
7830 hdr
->zh_opts_size
= sizeof (ztest_shared_opts_t
);
7831 hdr
->zh_size
= sizeof (ztest_shared_t
);
7832 hdr
->zh_stats_size
= sizeof (ztest_shared_callstate_t
);
7833 hdr
->zh_stats_count
= ZTEST_FUNCS
;
7834 hdr
->zh_ds_size
= sizeof (ztest_shared_ds_t
);
7835 hdr
->zh_ds_count
= ztest_opts
.zo_datasets
;
7837 size
= shared_data_size(hdr
);
7838 VERIFY0(ftruncate(ztest_fd_data
, size
));
7840 (void) munmap((caddr_t
)hdr
, P2ROUNDUP(sizeof (*hdr
), getpagesize()));
7847 ztest_shared_hdr_t
*hdr
;
7850 hdr
= (void *)mmap(0, P2ROUNDUP(sizeof (*hdr
), getpagesize()),
7851 PROT_READ
, MAP_SHARED
, ztest_fd_data
, 0);
7852 ASSERT3P(hdr
, !=, MAP_FAILED
);
7854 size
= shared_data_size(hdr
);
7856 (void) munmap((caddr_t
)hdr
, P2ROUNDUP(sizeof (*hdr
), getpagesize()));
7857 hdr
= ztest_shared_hdr
= (void *)mmap(0, P2ROUNDUP(size
, getpagesize()),
7858 PROT_READ
| PROT_WRITE
, MAP_SHARED
, ztest_fd_data
, 0);
7859 ASSERT3P(hdr
, !=, MAP_FAILED
);
7860 buf
= (uint8_t *)hdr
;
7862 offset
= hdr
->zh_hdr_size
;
7863 ztest_shared_opts
= (void *)&buf
[offset
];
7864 offset
+= hdr
->zh_opts_size
;
7865 ztest_shared
= (void *)&buf
[offset
];
7866 offset
+= hdr
->zh_size
;
7867 ztest_shared_callstate
= (void *)&buf
[offset
];
7868 offset
+= hdr
->zh_stats_size
* hdr
->zh_stats_count
;
7869 ztest_shared_ds
= (void *)&buf
[offset
];
7873 exec_child(char *cmd
, char *libpath
, boolean_t ignorekill
, int *statusp
)
7877 char *cmdbuf
= NULL
;
7882 cmdbuf
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
7883 (void) strlcpy(cmdbuf
, getexecname(), MAXPATHLEN
);
7888 fatal(B_TRUE
, "fork failed");
7890 if (pid
== 0) { /* child */
7891 char *emptyargv
[2] = { cmd
, NULL
};
7892 char fd_data_str
[12];
7894 struct rlimit rl
= { 1024, 1024 };
7895 (void) setrlimit(RLIMIT_NOFILE
, &rl
);
7897 (void) close(ztest_fd_rand
);
7899 snprintf(fd_data_str
, 12, "%d", ztest_fd_data
));
7900 VERIFY0(setenv("ZTEST_FD_DATA", fd_data_str
, 1));
7902 (void) enable_extended_FILE_stdio(-1, -1);
7903 if (libpath
!= NULL
)
7904 VERIFY0(setenv("LD_LIBRARY_PATH", libpath
, 1));
7905 (void) execv(cmd
, emptyargv
);
7906 ztest_dump_core
= B_FALSE
;
7907 fatal(B_TRUE
, "exec failed: %s", cmd
);
7910 if (cmdbuf
!= NULL
) {
7911 umem_free(cmdbuf
, MAXPATHLEN
);
7915 while (waitpid(pid
, &status
, 0) != pid
)
7917 if (statusp
!= NULL
)
7920 if (WIFEXITED(status
)) {
7921 if (WEXITSTATUS(status
) != 0) {
7922 (void) fprintf(stderr
, "child exited with code %d\n",
7923 WEXITSTATUS(status
));
7927 } else if (WIFSIGNALED(status
)) {
7928 if (!ignorekill
|| WTERMSIG(status
) != SIGKILL
) {
7929 (void) fprintf(stderr
, "child died with signal %d\n",
7935 (void) fprintf(stderr
, "something strange happened to child\n");
7941 ztest_run_init(void)
7945 ztest_shared_t
*zs
= ztest_shared
;
7948 * Blow away any existing copy of zpool.cache
7950 (void) remove(spa_config_path
);
7952 if (ztest_opts
.zo_init
== 0) {
7953 if (ztest_opts
.zo_verbose
>= 1)
7954 (void) printf("Importing pool %s\n",
7955 ztest_opts
.zo_pool
);
7961 * Create and initialize our storage pool.
7963 for (i
= 1; i
<= ztest_opts
.zo_init
; i
++) {
7964 memset(zs
, 0, sizeof (*zs
));
7965 if (ztest_opts
.zo_verbose
>= 3 &&
7966 ztest_opts
.zo_init
!= 1) {
7967 (void) printf("ztest_init(), pass %d\n", i
);
7974 main(int argc
, char **argv
)
7982 ztest_shared_callstate_t
*zc
;
7984 char numbuf
[NN_NUMBUF_SZ
];
7988 char *fd_data_str
= getenv("ZTEST_FD_DATA");
7989 struct sigaction action
;
7991 (void) setvbuf(stdout
, NULL
, _IOLBF
, 0);
7993 dprintf_setup(&argc
, argv
);
7994 zfs_deadman_synctime_ms
= 300000;
7995 zfs_deadman_checktime_ms
= 30000;
7997 * As two-word space map entries may not come up often (especially
7998 * if pool and vdev sizes are small) we want to force at least some
7999 * of them so the feature get tested.
8001 zfs_force_some_double_word_sm_entries
= B_TRUE
;
8004 * Verify that even extensively damaged split blocks with many
8005 * segments can be reconstructed in a reasonable amount of time
8006 * when reconstruction is known to be possible.
8008 * Note: the lower this value is, the more damage we inflict, and
8009 * the more time ztest spends in recovering that damage. We chose
8010 * to induce damage 1/100th of the time so recovery is tested but
8011 * not so frequently that ztest doesn't get to test other code paths.
8013 zfs_reconstruct_indirect_damage_fraction
= 100;
8015 action
.sa_handler
= sig_handler
;
8016 sigemptyset(&action
.sa_mask
);
8017 action
.sa_flags
= 0;
8019 if (sigaction(SIGSEGV
, &action
, NULL
) < 0) {
8020 (void) fprintf(stderr
, "ztest: cannot catch SIGSEGV: %s.\n",
8025 if (sigaction(SIGABRT
, &action
, NULL
) < 0) {
8026 (void) fprintf(stderr
, "ztest: cannot catch SIGABRT: %s.\n",
8032 * Force random_get_bytes() to use /dev/urandom in order to prevent
8033 * ztest from needlessly depleting the system entropy pool.
8035 random_path
= "/dev/urandom";
8036 ztest_fd_rand
= open(random_path
, O_RDONLY
);
8037 ASSERT3S(ztest_fd_rand
, >=, 0);
8040 process_options(argc
, argv
);
8045 memcpy(ztest_shared_opts
, &ztest_opts
,
8046 sizeof (*ztest_shared_opts
));
8048 ztest_fd_data
= atoi(fd_data_str
);
8050 memcpy(&ztest_opts
, ztest_shared_opts
, sizeof (ztest_opts
));
8052 ASSERT3U(ztest_opts
.zo_datasets
, ==, ztest_shared_hdr
->zh_ds_count
);
8054 err
= ztest_set_global_vars();
8055 if (err
!= 0 && !fd_data_str
) {
8056 /* error message done by ztest_set_global_vars */
8059 /* children should not be spawned if setting gvars fails */
8060 VERIFY3S(err
, ==, 0);
8063 /* Override location of zpool.cache */
8064 VERIFY3S(asprintf((char **)&spa_config_path
, "%s/zpool.cache",
8065 ztest_opts
.zo_dir
), !=, -1);
8067 ztest_ds
= umem_alloc(ztest_opts
.zo_datasets
* sizeof (ztest_ds_t
),
8072 metaslab_force_ganging
= ztest_opts
.zo_metaslab_force_ganging
;
8073 metaslab_df_alloc_threshold
=
8074 zs
->zs_metaslab_df_alloc_threshold
;
8083 hasalt
= (strlen(ztest_opts
.zo_alt_ztest
) != 0);
8085 if (ztest_opts
.zo_verbose
>= 1) {
8086 (void) printf("%"PRIu64
" vdevs, %d datasets, %d threads,"
8087 "%d %s disks, %"PRIu64
" seconds...\n\n",
8088 ztest_opts
.zo_vdevs
,
8089 ztest_opts
.zo_datasets
,
8090 ztest_opts
.zo_threads
,
8091 ztest_opts
.zo_raid_children
,
8092 ztest_opts
.zo_raid_type
,
8093 ztest_opts
.zo_time
);
8096 cmd
= umem_alloc(MAXNAMELEN
, UMEM_NOFAIL
);
8097 (void) strlcpy(cmd
, getexecname(), MAXNAMELEN
);
8099 zs
->zs_do_init
= B_TRUE
;
8100 if (strlen(ztest_opts
.zo_alt_ztest
) != 0) {
8101 if (ztest_opts
.zo_verbose
>= 1) {
8102 (void) printf("Executing older ztest for "
8103 "initialization: %s\n", ztest_opts
.zo_alt_ztest
);
8105 VERIFY(!exec_child(ztest_opts
.zo_alt_ztest
,
8106 ztest_opts
.zo_alt_libpath
, B_FALSE
, NULL
));
8108 VERIFY(!exec_child(NULL
, NULL
, B_FALSE
, NULL
));
8110 zs
->zs_do_init
= B_FALSE
;
8112 zs
->zs_proc_start
= gethrtime();
8113 zs
->zs_proc_stop
= zs
->zs_proc_start
+ ztest_opts
.zo_time
* NANOSEC
;
8115 for (f
= 0; f
< ZTEST_FUNCS
; f
++) {
8116 zi
= &ztest_info
[f
];
8117 zc
= ZTEST_GET_SHARED_CALLSTATE(f
);
8118 if (zs
->zs_proc_start
+ zi
->zi_interval
[0] > zs
->zs_proc_stop
)
8119 zc
->zc_next
= UINT64_MAX
;
8121 zc
->zc_next
= zs
->zs_proc_start
+
8122 ztest_random(2 * zi
->zi_interval
[0] + 1);
8126 * Run the tests in a loop. These tests include fault injection
8127 * to verify that self-healing data works, and forced crashes
8128 * to verify that we never lose on-disk consistency.
8130 while (gethrtime() < zs
->zs_proc_stop
) {
8135 * Initialize the workload counters for each function.
8137 for (f
= 0; f
< ZTEST_FUNCS
; f
++) {
8138 zc
= ZTEST_GET_SHARED_CALLSTATE(f
);
8143 /* Set the allocation switch size */
8144 zs
->zs_metaslab_df_alloc_threshold
=
8145 ztest_random(zs
->zs_metaslab_sz
/ 4) + 1;
8147 if (!hasalt
|| ztest_random(2) == 0) {
8148 if (hasalt
&& ztest_opts
.zo_verbose
>= 1) {
8149 (void) printf("Executing newer ztest: %s\n",
8153 killed
= exec_child(cmd
, NULL
, B_TRUE
, &status
);
8155 if (hasalt
&& ztest_opts
.zo_verbose
>= 1) {
8156 (void) printf("Executing older ztest: %s\n",
8157 ztest_opts
.zo_alt_ztest
);
8160 killed
= exec_child(ztest_opts
.zo_alt_ztest
,
8161 ztest_opts
.zo_alt_libpath
, B_TRUE
, &status
);
8168 if (ztest_opts
.zo_verbose
>= 1) {
8169 hrtime_t now
= gethrtime();
8171 now
= MIN(now
, zs
->zs_proc_stop
);
8172 print_time(zs
->zs_proc_stop
- now
, timebuf
);
8173 nicenum(zs
->zs_space
, numbuf
, sizeof (numbuf
));
8175 (void) printf("Pass %3d, %8s, %3"PRIu64
" ENOSPC, "
8176 "%4.1f%% of %5s used, %3.0f%% done, %8s to go\n",
8178 WIFEXITED(status
) ? "Complete" : "SIGKILL",
8179 zs
->zs_enospc_count
,
8180 100.0 * zs
->zs_alloc
/ zs
->zs_space
,
8182 100.0 * (now
- zs
->zs_proc_start
) /
8183 (ztest_opts
.zo_time
* NANOSEC
), timebuf
);
8186 if (ztest_opts
.zo_verbose
>= 2) {
8187 (void) printf("\nWorkload summary:\n\n");
8188 (void) printf("%7s %9s %s\n",
8189 "Calls", "Time", "Function");
8190 (void) printf("%7s %9s %s\n",
8191 "-----", "----", "--------");
8192 for (f
= 0; f
< ZTEST_FUNCS
; f
++) {
8193 zi
= &ztest_info
[f
];
8194 zc
= ZTEST_GET_SHARED_CALLSTATE(f
);
8195 print_time(zc
->zc_time
, timebuf
);
8196 (void) printf("%7"PRIu64
" %9s %s\n",
8197 zc
->zc_count
, timebuf
,
8200 (void) printf("\n");
8203 if (!ztest_opts
.zo_mmp_test
)
8204 ztest_run_zdb(ztest_opts
.zo_pool
);
8207 if (ztest_opts
.zo_verbose
>= 1) {
8209 (void) printf("%d runs of older ztest: %s\n", older
,
8210 ztest_opts
.zo_alt_ztest
);
8211 (void) printf("%d runs of newer ztest: %s\n", newer
,
8214 (void) printf("%d killed, %d completed, %.0f%% kill rate\n",
8215 kills
, iters
- kills
, (100.0 * kills
) / MAX(1, iters
));
8218 umem_free(cmd
, MAXNAMELEN
);