4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2011, 2016 by Delphix. All rights reserved.
24 * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
25 * Copyright (c) 2013 Steven Hartland. All rights reserved.
29 * The objective of this program is to provide a DMU/ZAP/SPA stress test
30 * that runs entirely in userland, is easy to use, and easy to extend.
32 * The overall design of the ztest program is as follows:
34 * (1) For each major functional area (e.g. adding vdevs to a pool,
35 * creating and destroying datasets, reading and writing objects, etc)
36 * we have a simple routine to test that functionality. These
37 * individual routines do not have to do anything "stressful".
39 * (2) We turn these simple functionality tests into a stress test by
40 * running them all in parallel, with as many threads as desired,
41 * and spread across as many datasets, objects, and vdevs as desired.
43 * (3) While all this is happening, we inject faults into the pool to
44 * verify that self-healing data really works.
46 * (4) Every time we open a dataset, we change its checksum and compression
47 * functions. Thus even individual objects vary from block to block
48 * in which checksum they use and whether they're compressed.
50 * (5) To verify that we never lose on-disk consistency after a crash,
51 * we run the entire test in a child of the main process.
52 * At random times, the child self-immolates with a SIGKILL.
53 * This is the software equivalent of pulling the power cord.
54 * The parent then runs the test again, using the existing
55 * storage pool, as many times as desired. If backwards compatibility
56 * testing is enabled ztest will sometimes run the "older" version
57 * of ztest after a SIGKILL.
59 * (6) To verify that we don't have future leaks or temporal incursions,
60 * many of the functional tests record the transaction group number
61 * as part of their data. When reading old data, they verify that
62 * the transaction group number is less than the current, open txg.
63 * If you add a new test, please do this if applicable.
65 * (7) Threads are created with a reduced stack size, for sanity checking.
66 * Therefore, it's important not to allocate huge buffers on the stack.
68 * When run with no arguments, ztest runs for about five minutes and
69 * produces no output if successful. To get a little bit of information,
70 * specify -V. To get more information, specify -VV, and so on.
72 * To turn this into an overnight stress test, use -T to specify run time.
74 * You can ask more more vdevs [-v], datasets [-d], or threads [-t]
75 * to increase the pool capacity, fanout, and overall stress level.
77 * Use the -k option to set the desired frequency of kills.
79 * When ztest invokes itself it passes all relevant information through a
80 * temporary file which is mmap-ed in the child process. This allows shared
81 * memory to survive the exec syscall. The ztest_shared_hdr_t struct is always
82 * stored at offset 0 of this file and contains information on the size and
83 * number of shared structures in the file. The information stored in this file
84 * must remain backwards compatible with older versions of ztest so that
85 * ztest can invoke them during backwards compatibility testing (-B).
88 #include <sys/zfs_context.h>
94 #include <sys/dmu_objset.h>
100 #include <sys/resource.h>
103 #include <sys/zil_impl.h>
104 #include <sys/zfs_rlock.h>
105 #include <sys/vdev_impl.h>
106 #include <sys/vdev_file.h>
107 #include <sys/spa_impl.h>
108 #include <sys/metaslab_impl.h>
109 #include <sys/dsl_prop.h>
110 #include <sys/dsl_dataset.h>
111 #include <sys/dsl_destroy.h>
112 #include <sys/dsl_scan.h>
113 #include <sys/zio_checksum.h>
114 #include <sys/refcount.h>
115 #include <sys/zfeature.h>
116 #include <sys/dsl_userhold.h>
119 #include <stdio_ext.h>
126 #include <sys/fs/zfs.h>
127 #include <zfs_fletcher.h>
128 #include <libnvpair.h>
130 #include <execinfo.h> /* for backtrace() */
133 static int ztest_fd_data
= -1;
134 static int ztest_fd_rand
= -1;
136 typedef struct ztest_shared_hdr
{
137 uint64_t zh_hdr_size
;
138 uint64_t zh_opts_size
;
140 uint64_t zh_stats_size
;
141 uint64_t zh_stats_count
;
143 uint64_t zh_ds_count
;
144 } ztest_shared_hdr_t
;
146 static ztest_shared_hdr_t
*ztest_shared_hdr
;
148 typedef struct ztest_shared_opts
{
149 char zo_pool
[ZFS_MAX_DATASET_NAME_LEN
];
150 char zo_dir
[ZFS_MAX_DATASET_NAME_LEN
];
151 char zo_alt_ztest
[MAXNAMELEN
];
152 char zo_alt_libpath
[MAXNAMELEN
];
154 uint64_t zo_vdevtime
;
162 uint64_t zo_passtime
;
163 uint64_t zo_killrate
;
167 uint64_t zo_maxloops
;
168 uint64_t zo_metaslab_gang_bang
;
169 } ztest_shared_opts_t
;
171 static const ztest_shared_opts_t ztest_opts_defaults
= {
172 .zo_pool
= { 'z', 't', 'e', 's', 't', '\0' },
173 .zo_dir
= { '/', 't', 'm', 'p', '\0' },
174 .zo_alt_ztest
= { '\0' },
175 .zo_alt_libpath
= { '\0' },
177 .zo_ashift
= SPA_MINBLOCKSHIFT
,
180 .zo_raidz_parity
= 1,
181 .zo_vdev_size
= SPA_MINDEVSIZE
* 4, /* 256m default size */
184 .zo_passtime
= 60, /* 60 seconds */
185 .zo_killrate
= 70, /* 70% kill rate */
188 .zo_time
= 300, /* 5 minutes */
189 .zo_maxloops
= 50, /* max loops during spa_freeze() */
190 .zo_metaslab_gang_bang
= 32 << 10
193 extern uint64_t metaslab_gang_bang
;
194 extern uint64_t metaslab_df_alloc_threshold
;
195 extern int metaslab_preload_limit
;
196 extern boolean_t zfs_compressed_arc_enabled
;
197 extern int zfs_abd_scatter_enabled
;
199 static ztest_shared_opts_t
*ztest_shared_opts
;
200 static ztest_shared_opts_t ztest_opts
;
202 typedef struct ztest_shared_ds
{
206 static ztest_shared_ds_t
*ztest_shared_ds
;
207 #define ZTEST_GET_SHARED_DS(d) (&ztest_shared_ds[d])
209 #define BT_MAGIC 0x123456789abcdefULL
210 #define MAXFAULTS() \
211 (MAX(zs->zs_mirrors, 1) * (ztest_opts.zo_raidz_parity + 1) - 1)
215 ZTEST_IO_WRITE_PATTERN
,
216 ZTEST_IO_WRITE_ZEROES
,
223 typedef struct ztest_block_tag
{
227 uint64_t bt_dnodesize
;
234 typedef struct bufwad
{
252 #define ZTEST_RANGE_LOCKS 64
253 #define ZTEST_OBJECT_LOCKS 64
256 * Object descriptor. Used as a template for object lookup/create/remove.
258 typedef struct ztest_od
{
261 dmu_object_type_t od_type
;
262 dmu_object_type_t od_crtype
;
263 uint64_t od_blocksize
;
264 uint64_t od_crblocksize
;
265 uint64_t od_crdnodesize
;
268 char od_name
[ZFS_MAX_DATASET_NAME_LEN
];
274 typedef struct ztest_ds
{
275 ztest_shared_ds_t
*zd_shared
;
277 rwlock_t zd_zilog_lock
;
279 ztest_od_t
*zd_od
; /* debugging aid */
280 char zd_name
[ZFS_MAX_DATASET_NAME_LEN
];
281 kmutex_t zd_dirobj_lock
;
282 rll_t zd_object_lock
[ZTEST_OBJECT_LOCKS
];
283 zll_t zd_range_lock
[ZTEST_RANGE_LOCKS
];
287 * Per-iteration state.
289 typedef void ztest_func_t(ztest_ds_t
*zd
, uint64_t id
);
291 typedef struct ztest_info
{
292 ztest_func_t
*zi_func
; /* test function */
293 uint64_t zi_iters
; /* iterations per execution */
294 uint64_t *zi_interval
; /* execute every <interval> seconds */
295 const char *zi_funcname
; /* name of test function */
298 typedef struct ztest_shared_callstate
{
299 uint64_t zc_count
; /* per-pass count */
300 uint64_t zc_time
; /* per-pass time */
301 uint64_t zc_next
; /* next time to call this function */
302 } ztest_shared_callstate_t
;
304 static ztest_shared_callstate_t
*ztest_shared_callstate
;
305 #define ZTEST_GET_SHARED_CALLSTATE(c) (&ztest_shared_callstate[c])
307 ztest_func_t ztest_dmu_read_write
;
308 ztest_func_t ztest_dmu_write_parallel
;
309 ztest_func_t ztest_dmu_object_alloc_free
;
310 ztest_func_t ztest_dmu_commit_callbacks
;
311 ztest_func_t ztest_zap
;
312 ztest_func_t ztest_zap_parallel
;
313 ztest_func_t ztest_zil_commit
;
314 ztest_func_t ztest_zil_remount
;
315 ztest_func_t ztest_dmu_read_write_zcopy
;
316 ztest_func_t ztest_dmu_objset_create_destroy
;
317 ztest_func_t ztest_dmu_prealloc
;
318 ztest_func_t ztest_fzap
;
319 ztest_func_t ztest_dmu_snapshot_create_destroy
;
320 ztest_func_t ztest_dsl_prop_get_set
;
321 ztest_func_t ztest_spa_prop_get_set
;
322 ztest_func_t ztest_spa_create_destroy
;
323 ztest_func_t ztest_fault_inject
;
324 ztest_func_t ztest_ddt_repair
;
325 ztest_func_t ztest_dmu_snapshot_hold
;
326 ztest_func_t ztest_spa_rename
;
327 ztest_func_t ztest_scrub
;
328 ztest_func_t ztest_dsl_dataset_promote_busy
;
329 ztest_func_t ztest_vdev_attach_detach
;
330 ztest_func_t ztest_vdev_LUN_growth
;
331 ztest_func_t ztest_vdev_add_remove
;
332 ztest_func_t ztest_vdev_aux_add_remove
;
333 ztest_func_t ztest_split_pool
;
334 ztest_func_t ztest_reguid
;
335 ztest_func_t ztest_spa_upgrade
;
336 ztest_func_t ztest_fletcher
;
337 ztest_func_t ztest_fletcher_incr
;
338 ztest_func_t ztest_verify_dnode_bt
;
340 uint64_t zopt_always
= 0ULL * NANOSEC
; /* all the time */
341 uint64_t zopt_incessant
= 1ULL * NANOSEC
/ 10; /* every 1/10 second */
342 uint64_t zopt_often
= 1ULL * NANOSEC
; /* every second */
343 uint64_t zopt_sometimes
= 10ULL * NANOSEC
; /* every 10 seconds */
344 uint64_t zopt_rarely
= 60ULL * NANOSEC
; /* every 60 seconds */
346 #define ZTI_INIT(func, iters, interval) \
347 { .zi_func = (func), \
348 .zi_iters = (iters), \
349 .zi_interval = (interval), \
350 .zi_funcname = # func }
352 ztest_info_t ztest_info
[] = {
353 ZTI_INIT(ztest_dmu_read_write
, 1, &zopt_always
),
354 ZTI_INIT(ztest_dmu_write_parallel
, 10, &zopt_always
),
355 ZTI_INIT(ztest_dmu_object_alloc_free
, 1, &zopt_always
),
356 ZTI_INIT(ztest_dmu_commit_callbacks
, 1, &zopt_always
),
357 ZTI_INIT(ztest_zap
, 30, &zopt_always
),
358 ZTI_INIT(ztest_zap_parallel
, 100, &zopt_always
),
359 ZTI_INIT(ztest_split_pool
, 1, &zopt_always
),
360 ZTI_INIT(ztest_zil_commit
, 1, &zopt_incessant
),
361 ZTI_INIT(ztest_zil_remount
, 1, &zopt_sometimes
),
362 ZTI_INIT(ztest_dmu_read_write_zcopy
, 1, &zopt_often
),
363 ZTI_INIT(ztest_dmu_objset_create_destroy
, 1, &zopt_often
),
364 ZTI_INIT(ztest_dsl_prop_get_set
, 1, &zopt_often
),
365 ZTI_INIT(ztest_spa_prop_get_set
, 1, &zopt_sometimes
),
367 ZTI_INIT(ztest_dmu_prealloc
, 1, &zopt_sometimes
),
369 ZTI_INIT(ztest_fzap
, 1, &zopt_sometimes
),
370 ZTI_INIT(ztest_dmu_snapshot_create_destroy
, 1, &zopt_sometimes
),
371 ZTI_INIT(ztest_spa_create_destroy
, 1, &zopt_sometimes
),
372 ZTI_INIT(ztest_fault_inject
, 1, &zopt_sometimes
),
373 ZTI_INIT(ztest_ddt_repair
, 1, &zopt_sometimes
),
374 ZTI_INIT(ztest_dmu_snapshot_hold
, 1, &zopt_sometimes
),
375 ZTI_INIT(ztest_reguid
, 1, &zopt_rarely
),
376 ZTI_INIT(ztest_spa_rename
, 1, &zopt_rarely
),
377 ZTI_INIT(ztest_scrub
, 1, &zopt_rarely
),
378 ZTI_INIT(ztest_spa_upgrade
, 1, &zopt_rarely
),
379 ZTI_INIT(ztest_dsl_dataset_promote_busy
, 1, &zopt_rarely
),
380 ZTI_INIT(ztest_vdev_attach_detach
, 1, &zopt_sometimes
),
381 ZTI_INIT(ztest_vdev_LUN_growth
, 1, &zopt_rarely
),
382 ZTI_INIT(ztest_vdev_add_remove
, 1, &ztest_opts
.zo_vdevtime
),
383 ZTI_INIT(ztest_vdev_aux_add_remove
, 1, &ztest_opts
.zo_vdevtime
),
384 ZTI_INIT(ztest_fletcher
, 1, &zopt_rarely
),
385 ZTI_INIT(ztest_fletcher_incr
, 1, &zopt_rarely
),
386 ZTI_INIT(ztest_verify_dnode_bt
, 1, &zopt_sometimes
),
389 #define ZTEST_FUNCS (sizeof (ztest_info) / sizeof (ztest_info_t))
392 * The following struct is used to hold a list of uncalled commit callbacks.
393 * The callbacks are ordered by txg number.
395 typedef struct ztest_cb_list
{
396 kmutex_t zcl_callbacks_lock
;
397 list_t zcl_callbacks
;
401 * Stuff we need to share writably between parent and child.
403 typedef struct ztest_shared
{
404 boolean_t zs_do_init
;
405 hrtime_t zs_proc_start
;
406 hrtime_t zs_proc_stop
;
407 hrtime_t zs_thread_start
;
408 hrtime_t zs_thread_stop
;
409 hrtime_t zs_thread_kill
;
410 uint64_t zs_enospc_count
;
411 uint64_t zs_vdev_next_leaf
;
412 uint64_t zs_vdev_aux
;
417 uint64_t zs_metaslab_sz
;
418 uint64_t zs_metaslab_df_alloc_threshold
;
422 #define ID_PARALLEL -1ULL
424 static char ztest_dev_template
[] = "%s/%s.%llua";
425 static char ztest_aux_template
[] = "%s/%s.%s.%llu";
426 ztest_shared_t
*ztest_shared
;
428 static spa_t
*ztest_spa
= NULL
;
429 static ztest_ds_t
*ztest_ds
;
431 static kmutex_t ztest_vdev_lock
;
434 * The ztest_name_lock protects the pool and dataset namespace used by
435 * the individual tests. To modify the namespace, consumers must grab
436 * this lock as writer. Grabbing the lock as reader will ensure that the
437 * namespace does not change while the lock is held.
439 static rwlock_t ztest_name_lock
;
441 static boolean_t ztest_dump_core
= B_TRUE
;
442 static boolean_t ztest_exiting
;
444 /* Global commit callback list */
445 static ztest_cb_list_t zcl
;
446 /* Commit cb delay */
447 static uint64_t zc_min_txg_delay
= UINT64_MAX
;
448 static int zc_cb_counter
= 0;
451 * Minimum number of commit callbacks that need to be registered for us to check
452 * whether the minimum txg delay is acceptable.
454 #define ZTEST_COMMIT_CB_MIN_REG 100
457 * If a number of txgs equal to this threshold have been created after a commit
458 * callback has been registered but not called, then we assume there is an
459 * implementation bug.
461 #define ZTEST_COMMIT_CB_THRESH (TXG_CONCURRENT_STATES + 1000)
463 extern uint64_t metaslab_gang_bang
;
464 extern uint64_t metaslab_df_alloc_threshold
;
467 ZTEST_META_DNODE
= 0,
472 static void usage(boolean_t
) __NORETURN
;
475 * These libumem hooks provide a reasonable set of defaults for the allocator's
476 * debugging facilities.
479 _umem_debug_init(void)
481 return ("default,verbose"); /* $UMEM_DEBUG setting */
485 _umem_logging_init(void)
487 return ("fail,contents"); /* $UMEM_LOGGING setting */
490 #define BACKTRACE_SZ 100
492 static void sig_handler(int signo
)
494 struct sigaction action
;
495 #ifdef __GLIBC__ /* backtrace() is a GNU extension */
497 void *buffer
[BACKTRACE_SZ
];
499 nptrs
= backtrace(buffer
, BACKTRACE_SZ
);
500 backtrace_symbols_fd(buffer
, nptrs
, STDERR_FILENO
);
504 * Restore default action and re-raise signal so SIGSEGV and
505 * SIGABRT can trigger a core dump.
507 action
.sa_handler
= SIG_DFL
;
508 sigemptyset(&action
.sa_mask
);
510 (void) sigaction(signo
, &action
, NULL
);
514 #define FATAL_MSG_SZ 1024
519 fatal(int do_perror
, char *message
, ...)
522 int save_errno
= errno
;
525 (void) fflush(stdout
);
526 buf
= umem_alloc(FATAL_MSG_SZ
, UMEM_NOFAIL
);
528 va_start(args
, message
);
529 (void) sprintf(buf
, "ztest: ");
531 (void) vsprintf(buf
+ strlen(buf
), message
, args
);
534 (void) snprintf(buf
+ strlen(buf
), FATAL_MSG_SZ
- strlen(buf
),
535 ": %s", strerror(save_errno
));
537 (void) fprintf(stderr
, "%s\n", buf
);
538 fatal_msg
= buf
; /* to ease debugging */
545 str2shift(const char *buf
)
547 const char *ends
= "BKMGTPEZ";
552 for (i
= 0; i
< strlen(ends
); i
++) {
553 if (toupper(buf
[0]) == ends
[i
])
556 if (i
== strlen(ends
)) {
557 (void) fprintf(stderr
, "ztest: invalid bytes suffix: %s\n",
561 if (buf
[1] == '\0' || (toupper(buf
[1]) == 'B' && buf
[2] == '\0')) {
564 (void) fprintf(stderr
, "ztest: invalid bytes suffix: %s\n", buf
);
570 nicenumtoull(const char *buf
)
575 val
= strtoull(buf
, &end
, 0);
577 (void) fprintf(stderr
, "ztest: bad numeric value: %s\n", buf
);
579 } else if (end
[0] == '.') {
580 double fval
= strtod(buf
, &end
);
581 fval
*= pow(2, str2shift(end
));
582 if (fval
> UINT64_MAX
) {
583 (void) fprintf(stderr
, "ztest: value too large: %s\n",
587 val
= (uint64_t)fval
;
589 int shift
= str2shift(end
);
590 if (shift
>= 64 || (val
<< shift
) >> shift
!= val
) {
591 (void) fprintf(stderr
, "ztest: value too large: %s\n",
601 usage(boolean_t requested
)
603 const ztest_shared_opts_t
*zo
= &ztest_opts_defaults
;
605 char nice_vdev_size
[10];
606 char nice_gang_bang
[10];
607 FILE *fp
= requested
? stdout
: stderr
;
609 nicenum(zo
->zo_vdev_size
, nice_vdev_size
);
610 nicenum(zo
->zo_metaslab_gang_bang
, nice_gang_bang
);
612 (void) fprintf(fp
, "Usage: %s\n"
613 "\t[-v vdevs (default: %llu)]\n"
614 "\t[-s size_of_each_vdev (default: %s)]\n"
615 "\t[-a alignment_shift (default: %d)] use 0 for random\n"
616 "\t[-m mirror_copies (default: %d)]\n"
617 "\t[-r raidz_disks (default: %d)]\n"
618 "\t[-R raidz_parity (default: %d)]\n"
619 "\t[-d datasets (default: %d)]\n"
620 "\t[-t threads (default: %d)]\n"
621 "\t[-g gang_block_threshold (default: %s)]\n"
622 "\t[-i init_count (default: %d)] initialize pool i times\n"
623 "\t[-k kill_percentage (default: %llu%%)]\n"
624 "\t[-p pool_name (default: %s)]\n"
625 "\t[-f dir (default: %s)] file directory for vdev files\n"
626 "\t[-V] verbose (use multiple times for ever more blather)\n"
627 "\t[-E] use existing pool instead of creating new one\n"
628 "\t[-T time (default: %llu sec)] total run time\n"
629 "\t[-F freezeloops (default: %llu)] max loops in spa_freeze()\n"
630 "\t[-P passtime (default: %llu sec)] time per pass\n"
631 "\t[-B alt_ztest (default: <none>)] alternate ztest path\n"
632 "\t[-o variable=value] ... set global variable to an unsigned\n"
633 "\t 32-bit integer value\n"
634 "\t[-h] (print help)\n"
637 (u_longlong_t
)zo
->zo_vdevs
, /* -v */
638 nice_vdev_size
, /* -s */
639 zo
->zo_ashift
, /* -a */
640 zo
->zo_mirrors
, /* -m */
641 zo
->zo_raidz
, /* -r */
642 zo
->zo_raidz_parity
, /* -R */
643 zo
->zo_datasets
, /* -d */
644 zo
->zo_threads
, /* -t */
645 nice_gang_bang
, /* -g */
646 zo
->zo_init
, /* -i */
647 (u_longlong_t
)zo
->zo_killrate
, /* -k */
648 zo
->zo_pool
, /* -p */
650 (u_longlong_t
)zo
->zo_time
, /* -T */
651 (u_longlong_t
)zo
->zo_maxloops
, /* -F */
652 (u_longlong_t
)zo
->zo_passtime
);
653 exit(requested
? 0 : 1);
657 process_options(int argc
, char **argv
)
660 ztest_shared_opts_t
*zo
= &ztest_opts
;
664 char altdir
[MAXNAMELEN
] = { 0 };
666 bcopy(&ztest_opts_defaults
, zo
, sizeof (*zo
));
668 while ((opt
= getopt(argc
, argv
,
669 "v:s:a:m:r:R:d:t:g:i:k:p:f:VET:P:hF:B:o:")) != EOF
) {
686 value
= nicenumtoull(optarg
);
690 zo
->zo_vdevs
= value
;
693 zo
->zo_vdev_size
= MAX(SPA_MINDEVSIZE
, value
);
696 zo
->zo_ashift
= value
;
699 zo
->zo_mirrors
= value
;
702 zo
->zo_raidz
= MAX(1, value
);
705 zo
->zo_raidz_parity
= MIN(MAX(value
, 1), 3);
708 zo
->zo_datasets
= MAX(1, value
);
711 zo
->zo_threads
= MAX(1, value
);
714 zo
->zo_metaslab_gang_bang
= MAX(SPA_MINBLOCKSIZE
<< 1,
721 zo
->zo_killrate
= value
;
724 (void) strlcpy(zo
->zo_pool
, optarg
,
725 sizeof (zo
->zo_pool
));
728 path
= realpath(optarg
, NULL
);
730 (void) fprintf(stderr
, "error: %s: %s\n",
731 optarg
, strerror(errno
));
734 (void) strlcpy(zo
->zo_dir
, path
,
735 sizeof (zo
->zo_dir
));
749 zo
->zo_passtime
= MAX(1, value
);
752 zo
->zo_maxloops
= MAX(1, value
);
755 (void) strlcpy(altdir
, optarg
, sizeof (altdir
));
758 if (set_global_var(optarg
) != 0)
771 zo
->zo_raidz_parity
= MIN(zo
->zo_raidz_parity
, zo
->zo_raidz
- 1);
774 (zo
->zo_vdevs
> 0 ? zo
->zo_time
* NANOSEC
/ zo
->zo_vdevs
:
777 if (strlen(altdir
) > 0) {
785 cmd
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
786 realaltdir
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
788 VERIFY(NULL
!= realpath(getexecname(), cmd
));
789 if (0 != access(altdir
, F_OK
)) {
790 ztest_dump_core
= B_FALSE
;
791 fatal(B_TRUE
, "invalid alternate ztest path: %s",
794 VERIFY(NULL
!= realpath(altdir
, realaltdir
));
797 * 'cmd' should be of the form "<anything>/usr/bin/<isa>/ztest".
798 * We want to extract <isa> to determine if we should use
799 * 32 or 64 bit binaries.
801 bin
= strstr(cmd
, "/usr/bin/");
802 ztest
= strstr(bin
, "/ztest");
804 isalen
= ztest
- isa
;
805 (void) snprintf(zo
->zo_alt_ztest
, sizeof (zo
->zo_alt_ztest
),
806 "%s/usr/bin/%.*s/ztest", realaltdir
, isalen
, isa
);
807 (void) snprintf(zo
->zo_alt_libpath
, sizeof (zo
->zo_alt_libpath
),
808 "%s/usr/lib/%.*s", realaltdir
, isalen
, isa
);
810 if (0 != access(zo
->zo_alt_ztest
, X_OK
)) {
811 ztest_dump_core
= B_FALSE
;
812 fatal(B_TRUE
, "invalid alternate ztest: %s",
814 } else if (0 != access(zo
->zo_alt_libpath
, X_OK
)) {
815 ztest_dump_core
= B_FALSE
;
816 fatal(B_TRUE
, "invalid alternate lib directory %s",
820 umem_free(cmd
, MAXPATHLEN
);
821 umem_free(realaltdir
, MAXPATHLEN
);
826 ztest_kill(ztest_shared_t
*zs
)
828 zs
->zs_alloc
= metaslab_class_get_alloc(spa_normal_class(ztest_spa
));
829 zs
->zs_space
= metaslab_class_get_space(spa_normal_class(ztest_spa
));
832 * Before we kill off ztest, make sure that the config is updated.
833 * See comment above spa_config_sync().
835 mutex_enter(&spa_namespace_lock
);
836 spa_config_sync(ztest_spa
, B_FALSE
, B_FALSE
);
837 mutex_exit(&spa_namespace_lock
);
839 (void) kill(getpid(), SIGKILL
);
843 ztest_random(uint64_t range
)
847 ASSERT3S(ztest_fd_rand
, >=, 0);
852 if (read(ztest_fd_rand
, &r
, sizeof (r
)) != sizeof (r
))
853 fatal(1, "short read from /dev/urandom");
860 ztest_record_enospc(const char *s
)
862 ztest_shared
->zs_enospc_count
++;
866 ztest_get_ashift(void)
868 if (ztest_opts
.zo_ashift
== 0)
869 return (SPA_MINBLOCKSHIFT
+ ztest_random(5));
870 return (ztest_opts
.zo_ashift
);
874 make_vdev_file(char *path
, char *aux
, char *pool
, size_t size
, uint64_t ashift
)
880 pathbuf
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
883 ashift
= ztest_get_ashift();
889 vdev
= ztest_shared
->zs_vdev_aux
;
890 (void) snprintf(path
, MAXPATHLEN
,
891 ztest_aux_template
, ztest_opts
.zo_dir
,
892 pool
== NULL
? ztest_opts
.zo_pool
: pool
,
895 vdev
= ztest_shared
->zs_vdev_next_leaf
++;
896 (void) snprintf(path
, MAXPATHLEN
,
897 ztest_dev_template
, ztest_opts
.zo_dir
,
898 pool
== NULL
? ztest_opts
.zo_pool
: pool
, vdev
);
903 int fd
= open(path
, O_RDWR
| O_CREAT
| O_TRUNC
, 0666);
905 fatal(1, "can't open %s", path
);
906 if (ftruncate(fd
, size
) != 0)
907 fatal(1, "can't ftruncate %s", path
);
911 VERIFY(nvlist_alloc(&file
, NV_UNIQUE_NAME
, 0) == 0);
912 VERIFY(nvlist_add_string(file
, ZPOOL_CONFIG_TYPE
, VDEV_TYPE_FILE
) == 0);
913 VERIFY(nvlist_add_string(file
, ZPOOL_CONFIG_PATH
, path
) == 0);
914 VERIFY(nvlist_add_uint64(file
, ZPOOL_CONFIG_ASHIFT
, ashift
) == 0);
915 umem_free(pathbuf
, MAXPATHLEN
);
921 make_vdev_raidz(char *path
, char *aux
, char *pool
, size_t size
,
922 uint64_t ashift
, int r
)
924 nvlist_t
*raidz
, **child
;
928 return (make_vdev_file(path
, aux
, pool
, size
, ashift
));
929 child
= umem_alloc(r
* sizeof (nvlist_t
*), UMEM_NOFAIL
);
931 for (c
= 0; c
< r
; c
++)
932 child
[c
] = make_vdev_file(path
, aux
, pool
, size
, ashift
);
934 VERIFY(nvlist_alloc(&raidz
, NV_UNIQUE_NAME
, 0) == 0);
935 VERIFY(nvlist_add_string(raidz
, ZPOOL_CONFIG_TYPE
,
936 VDEV_TYPE_RAIDZ
) == 0);
937 VERIFY(nvlist_add_uint64(raidz
, ZPOOL_CONFIG_NPARITY
,
938 ztest_opts
.zo_raidz_parity
) == 0);
939 VERIFY(nvlist_add_nvlist_array(raidz
, ZPOOL_CONFIG_CHILDREN
,
942 for (c
= 0; c
< r
; c
++)
943 nvlist_free(child
[c
]);
945 umem_free(child
, r
* sizeof (nvlist_t
*));
951 make_vdev_mirror(char *path
, char *aux
, char *pool
, size_t size
,
952 uint64_t ashift
, int r
, int m
)
954 nvlist_t
*mirror
, **child
;
958 return (make_vdev_raidz(path
, aux
, pool
, size
, ashift
, r
));
960 child
= umem_alloc(m
* sizeof (nvlist_t
*), UMEM_NOFAIL
);
962 for (c
= 0; c
< m
; c
++)
963 child
[c
] = make_vdev_raidz(path
, aux
, pool
, size
, ashift
, r
);
965 VERIFY(nvlist_alloc(&mirror
, NV_UNIQUE_NAME
, 0) == 0);
966 VERIFY(nvlist_add_string(mirror
, ZPOOL_CONFIG_TYPE
,
967 VDEV_TYPE_MIRROR
) == 0);
968 VERIFY(nvlist_add_nvlist_array(mirror
, ZPOOL_CONFIG_CHILDREN
,
971 for (c
= 0; c
< m
; c
++)
972 nvlist_free(child
[c
]);
974 umem_free(child
, m
* sizeof (nvlist_t
*));
980 make_vdev_root(char *path
, char *aux
, char *pool
, size_t size
, uint64_t ashift
,
981 int log
, int r
, int m
, int t
)
983 nvlist_t
*root
, **child
;
988 child
= umem_alloc(t
* sizeof (nvlist_t
*), UMEM_NOFAIL
);
990 for (c
= 0; c
< t
; c
++) {
991 child
[c
] = make_vdev_mirror(path
, aux
, pool
, size
, ashift
,
993 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_IS_LOG
,
997 VERIFY(nvlist_alloc(&root
, NV_UNIQUE_NAME
, 0) == 0);
998 VERIFY(nvlist_add_string(root
, ZPOOL_CONFIG_TYPE
, VDEV_TYPE_ROOT
) == 0);
999 VERIFY(nvlist_add_nvlist_array(root
, aux
? aux
: ZPOOL_CONFIG_CHILDREN
,
1002 for (c
= 0; c
< t
; c
++)
1003 nvlist_free(child
[c
]);
1005 umem_free(child
, t
* sizeof (nvlist_t
*));
1011 * Find a random spa version. Returns back a random spa version in the
1012 * range [initial_version, SPA_VERSION_FEATURES].
1015 ztest_random_spa_version(uint64_t initial_version
)
1017 uint64_t version
= initial_version
;
1019 if (version
<= SPA_VERSION_BEFORE_FEATURES
) {
1021 ztest_random(SPA_VERSION_BEFORE_FEATURES
- version
+ 1);
1024 if (version
> SPA_VERSION_BEFORE_FEATURES
)
1025 version
= SPA_VERSION_FEATURES
;
1027 ASSERT(SPA_VERSION_IS_SUPPORTED(version
));
1032 ztest_random_blocksize(void)
1035 * Choose a block size >= the ashift.
1036 * If the SPA supports new MAXBLOCKSIZE, test up to 1MB blocks.
1038 int maxbs
= SPA_OLD_MAXBLOCKSHIFT
;
1039 if (spa_maxblocksize(ztest_spa
) == SPA_MAXBLOCKSIZE
)
1041 uint64_t block_shift
=
1042 ztest_random(maxbs
- ztest_spa
->spa_max_ashift
+ 1);
1043 return (1 << (SPA_MINBLOCKSHIFT
+ block_shift
));
1047 ztest_random_dnodesize(void)
1050 int max_slots
= spa_maxdnodesize(ztest_spa
) >> DNODE_SHIFT
;
1052 if (max_slots
== DNODE_MIN_SLOTS
)
1053 return (DNODE_MIN_SIZE
);
1056 * Weight the random distribution more heavily toward smaller
1057 * dnode sizes since that is more likely to reflect real-world
1060 ASSERT3U(max_slots
, >, 4);
1061 switch (ztest_random(10)) {
1063 slots
= 5 + ztest_random(max_slots
- 4);
1066 slots
= 2 + ztest_random(3);
1073 return (slots
<< DNODE_SHIFT
);
1077 ztest_random_ibshift(void)
1079 return (DN_MIN_INDBLKSHIFT
+
1080 ztest_random(DN_MAX_INDBLKSHIFT
- DN_MIN_INDBLKSHIFT
+ 1));
1084 ztest_random_vdev_top(spa_t
*spa
, boolean_t log_ok
)
1087 vdev_t
*rvd
= spa
->spa_root_vdev
;
1090 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_READER
) != 0);
1093 top
= ztest_random(rvd
->vdev_children
);
1094 tvd
= rvd
->vdev_child
[top
];
1095 } while (tvd
->vdev_ishole
|| (tvd
->vdev_islog
&& !log_ok
) ||
1096 tvd
->vdev_mg
== NULL
|| tvd
->vdev_mg
->mg_class
== NULL
);
1102 ztest_random_dsl_prop(zfs_prop_t prop
)
1107 value
= zfs_prop_random_value(prop
, ztest_random(-1ULL));
1108 } while (prop
== ZFS_PROP_CHECKSUM
&& value
== ZIO_CHECKSUM_OFF
);
1114 ztest_dsl_prop_set_uint64(char *osname
, zfs_prop_t prop
, uint64_t value
,
1117 const char *propname
= zfs_prop_to_name(prop
);
1118 const char *valname
;
1123 error
= dsl_prop_set_int(osname
, propname
,
1124 (inherit
? ZPROP_SRC_NONE
: ZPROP_SRC_LOCAL
), value
);
1126 if (error
== ENOSPC
) {
1127 ztest_record_enospc(FTAG
);
1132 setpoint
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
1133 VERIFY0(dsl_prop_get_integer(osname
, propname
, &curval
, setpoint
));
1135 if (ztest_opts
.zo_verbose
>= 6) {
1138 err
= zfs_prop_index_to_string(prop
, curval
, &valname
);
1140 (void) printf("%s %s = %llu at '%s'\n", osname
,
1141 propname
, (unsigned long long)curval
, setpoint
);
1143 (void) printf("%s %s = %s at '%s'\n",
1144 osname
, propname
, valname
, setpoint
);
1146 umem_free(setpoint
, MAXPATHLEN
);
1152 ztest_spa_prop_set_uint64(zpool_prop_t prop
, uint64_t value
)
1154 spa_t
*spa
= ztest_spa
;
1155 nvlist_t
*props
= NULL
;
1158 VERIFY(nvlist_alloc(&props
, NV_UNIQUE_NAME
, 0) == 0);
1159 VERIFY(nvlist_add_uint64(props
, zpool_prop_to_name(prop
), value
) == 0);
1161 error
= spa_prop_set(spa
, props
);
1165 if (error
== ENOSPC
) {
1166 ztest_record_enospc(FTAG
);
1176 * Object and range lock mechanics
1179 list_node_t z_lnode
;
1180 refcount_t z_refcnt
;
1182 zfs_rlock_t z_range_lock
;
1187 ztest_znode_t
*z_ztznode
;
1190 static ztest_znode_t
*
1191 ztest_znode_init(uint64_t object
)
1193 ztest_znode_t
*zp
= umem_alloc(sizeof (*zp
), UMEM_NOFAIL
);
1195 list_link_init(&zp
->z_lnode
);
1196 refcount_create(&zp
->z_refcnt
);
1197 zp
->z_object
= object
;
1198 zfs_rlock_init(&zp
->z_range_lock
);
1204 ztest_znode_fini(ztest_znode_t
*zp
)
1206 ASSERT(refcount_is_zero(&zp
->z_refcnt
));
1207 zfs_rlock_destroy(&zp
->z_range_lock
);
1209 refcount_destroy(&zp
->z_refcnt
);
1210 list_link_init(&zp
->z_lnode
);
1211 umem_free(zp
, sizeof (*zp
));
1215 ztest_zll_init(zll_t
*zll
)
1217 mutex_init(&zll
->z_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
1218 list_create(&zll
->z_list
, sizeof (ztest_znode_t
),
1219 offsetof(ztest_znode_t
, z_lnode
));
1223 ztest_zll_destroy(zll_t
*zll
)
1225 list_destroy(&zll
->z_list
);
1226 mutex_destroy(&zll
->z_lock
);
1229 #define RL_TAG "range_lock"
1230 static ztest_znode_t
*
1231 ztest_znode_get(ztest_ds_t
*zd
, uint64_t object
)
1233 zll_t
*zll
= &zd
->zd_range_lock
[object
& (ZTEST_OBJECT_LOCKS
- 1)];
1234 ztest_znode_t
*zp
= NULL
;
1235 mutex_enter(&zll
->z_lock
);
1236 for (zp
= list_head(&zll
->z_list
); (zp
);
1237 zp
= list_next(&zll
->z_list
, zp
)) {
1238 if (zp
->z_object
== object
) {
1239 refcount_add(&zp
->z_refcnt
, RL_TAG
);
1244 zp
= ztest_znode_init(object
);
1245 refcount_add(&zp
->z_refcnt
, RL_TAG
);
1246 list_insert_head(&zll
->z_list
, zp
);
1248 mutex_exit(&zll
->z_lock
);
1253 ztest_znode_put(ztest_ds_t
*zd
, ztest_znode_t
*zp
)
1256 ASSERT3U(zp
->z_object
, !=, 0);
1257 zll
= &zd
->zd_range_lock
[zp
->z_object
& (ZTEST_OBJECT_LOCKS
- 1)];
1258 mutex_enter(&zll
->z_lock
);
1259 refcount_remove(&zp
->z_refcnt
, RL_TAG
);
1260 if (refcount_is_zero(&zp
->z_refcnt
)) {
1261 list_remove(&zll
->z_list
, zp
);
1262 ztest_znode_fini(zp
);
1264 mutex_exit(&zll
->z_lock
);
1269 ztest_rll_init(rll_t
*rll
)
1271 rll
->rll_writer
= NULL
;
1272 rll
->rll_readers
= 0;
1273 mutex_init(&rll
->rll_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
1274 cv_init(&rll
->rll_cv
, NULL
, CV_DEFAULT
, NULL
);
1278 ztest_rll_destroy(rll_t
*rll
)
1280 ASSERT(rll
->rll_writer
== NULL
);
1281 ASSERT(rll
->rll_readers
== 0);
1282 mutex_destroy(&rll
->rll_lock
);
1283 cv_destroy(&rll
->rll_cv
);
1287 ztest_rll_lock(rll_t
*rll
, rl_type_t type
)
1289 mutex_enter(&rll
->rll_lock
);
1291 if (type
== RL_READER
) {
1292 while (rll
->rll_writer
!= NULL
)
1293 (void) cv_wait(&rll
->rll_cv
, &rll
->rll_lock
);
1296 while (rll
->rll_writer
!= NULL
|| rll
->rll_readers
)
1297 (void) cv_wait(&rll
->rll_cv
, &rll
->rll_lock
);
1298 rll
->rll_writer
= curthread
;
1301 mutex_exit(&rll
->rll_lock
);
1305 ztest_rll_unlock(rll_t
*rll
)
1307 mutex_enter(&rll
->rll_lock
);
1309 if (rll
->rll_writer
) {
1310 ASSERT(rll
->rll_readers
== 0);
1311 rll
->rll_writer
= NULL
;
1313 ASSERT(rll
->rll_readers
!= 0);
1314 ASSERT(rll
->rll_writer
== NULL
);
1318 if (rll
->rll_writer
== NULL
&& rll
->rll_readers
== 0)
1319 cv_broadcast(&rll
->rll_cv
);
1321 mutex_exit(&rll
->rll_lock
);
1325 ztest_object_lock(ztest_ds_t
*zd
, uint64_t object
, rl_type_t type
)
1327 rll_t
*rll
= &zd
->zd_object_lock
[object
& (ZTEST_OBJECT_LOCKS
- 1)];
1329 ztest_rll_lock(rll
, type
);
1333 ztest_object_unlock(ztest_ds_t
*zd
, uint64_t object
)
1335 rll_t
*rll
= &zd
->zd_object_lock
[object
& (ZTEST_OBJECT_LOCKS
- 1)];
1337 ztest_rll_unlock(rll
);
1340 static ztest_zrl_t
*
1341 ztest_zrl_init(rl_t
*rl
, ztest_znode_t
*zp
)
1343 ztest_zrl_t
*zrl
= umem_alloc(sizeof (*zrl
), UMEM_NOFAIL
);
1345 zrl
->z_ztznode
= zp
;
1350 ztest_zrl_fini(ztest_zrl_t
*zrl
)
1352 umem_free(zrl
, sizeof (*zrl
));
1355 static ztest_zrl_t
*
1356 ztest_range_lock(ztest_ds_t
*zd
, uint64_t object
, uint64_t offset
,
1357 uint64_t size
, rl_type_t type
)
1359 ztest_znode_t
*zp
= ztest_znode_get(zd
, object
);
1360 rl_t
*rl
= zfs_range_lock(&zp
->z_range_lock
, offset
,
1362 return (ztest_zrl_init(rl
, zp
));
1366 ztest_range_unlock(ztest_ds_t
*zd
, ztest_zrl_t
*zrl
)
1368 zfs_range_unlock(zrl
->z_rl
);
1369 ztest_znode_put(zd
, zrl
->z_ztznode
);
1370 ztest_zrl_fini(zrl
);
1374 ztest_zd_init(ztest_ds_t
*zd
, ztest_shared_ds_t
*szd
, objset_t
*os
)
1377 zd
->zd_zilog
= dmu_objset_zil(os
);
1378 zd
->zd_shared
= szd
;
1379 dmu_objset_name(os
, zd
->zd_name
);
1382 if (zd
->zd_shared
!= NULL
)
1383 zd
->zd_shared
->zd_seq
= 0;
1385 VERIFY(rwlock_init(&zd
->zd_zilog_lock
, USYNC_THREAD
, NULL
) == 0);
1386 mutex_init(&zd
->zd_dirobj_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
1388 for (l
= 0; l
< ZTEST_OBJECT_LOCKS
; l
++)
1389 ztest_rll_init(&zd
->zd_object_lock
[l
]);
1391 for (l
= 0; l
< ZTEST_RANGE_LOCKS
; l
++)
1392 ztest_zll_init(&zd
->zd_range_lock
[l
]);
1396 ztest_zd_fini(ztest_ds_t
*zd
)
1400 mutex_destroy(&zd
->zd_dirobj_lock
);
1401 (void) rwlock_destroy(&zd
->zd_zilog_lock
);
1403 for (l
= 0; l
< ZTEST_OBJECT_LOCKS
; l
++)
1404 ztest_rll_destroy(&zd
->zd_object_lock
[l
]);
1406 for (l
= 0; l
< ZTEST_RANGE_LOCKS
; l
++)
1407 ztest_zll_destroy(&zd
->zd_range_lock
[l
]);
1410 #define TXG_MIGHTWAIT (ztest_random(10) == 0 ? TXG_NOWAIT : TXG_WAIT)
1413 ztest_tx_assign(dmu_tx_t
*tx
, uint64_t txg_how
, const char *tag
)
1419 * Attempt to assign tx to some transaction group.
1421 error
= dmu_tx_assign(tx
, txg_how
);
1423 if (error
== ERESTART
) {
1424 ASSERT(txg_how
== TXG_NOWAIT
);
1427 ASSERT3U(error
, ==, ENOSPC
);
1428 ztest_record_enospc(tag
);
1433 txg
= dmu_tx_get_txg(tx
);
1439 ztest_pattern_set(void *buf
, uint64_t size
, uint64_t value
)
1442 uint64_t *ip_end
= (uint64_t *)((uintptr_t)buf
+ (uintptr_t)size
);
1450 ztest_pattern_match(void *buf
, uint64_t size
, uint64_t value
)
1453 uint64_t *ip_end
= (uint64_t *)((uintptr_t)buf
+ (uintptr_t)size
);
1457 diff
|= (value
- *ip
++);
1464 ztest_bt_generate(ztest_block_tag_t
*bt
, objset_t
*os
, uint64_t object
,
1465 uint64_t dnodesize
, uint64_t offset
, uint64_t gen
, uint64_t txg
,
1468 bt
->bt_magic
= BT_MAGIC
;
1469 bt
->bt_objset
= dmu_objset_id(os
);
1470 bt
->bt_object
= object
;
1471 bt
->bt_dnodesize
= dnodesize
;
1472 bt
->bt_offset
= offset
;
1475 bt
->bt_crtxg
= crtxg
;
1479 ztest_bt_verify(ztest_block_tag_t
*bt
, objset_t
*os
, uint64_t object
,
1480 uint64_t dnodesize
, uint64_t offset
, uint64_t gen
, uint64_t txg
,
1483 ASSERT3U(bt
->bt_magic
, ==, BT_MAGIC
);
1484 ASSERT3U(bt
->bt_objset
, ==, dmu_objset_id(os
));
1485 ASSERT3U(bt
->bt_object
, ==, object
);
1486 ASSERT3U(bt
->bt_dnodesize
, ==, dnodesize
);
1487 ASSERT3U(bt
->bt_offset
, ==, offset
);
1488 ASSERT3U(bt
->bt_gen
, <=, gen
);
1489 ASSERT3U(bt
->bt_txg
, <=, txg
);
1490 ASSERT3U(bt
->bt_crtxg
, ==, crtxg
);
1493 static ztest_block_tag_t
*
1494 ztest_bt_bonus(dmu_buf_t
*db
)
1496 dmu_object_info_t doi
;
1497 ztest_block_tag_t
*bt
;
1499 dmu_object_info_from_db(db
, &doi
);
1500 ASSERT3U(doi
.doi_bonus_size
, <=, db
->db_size
);
1501 ASSERT3U(doi
.doi_bonus_size
, >=, sizeof (*bt
));
1502 bt
= (void *)((char *)db
->db_data
+ doi
.doi_bonus_size
- sizeof (*bt
));
1508 * Generate a token to fill up unused bonus buffer space. Try to make
1509 * it unique to the object, generation, and offset to verify that data
1510 * is not getting overwritten by data from other dnodes.
1512 #define ZTEST_BONUS_FILL_TOKEN(obj, ds, gen, offset) \
1513 (((ds) << 48) | ((gen) << 32) | ((obj) << 8) | (offset))
1516 * Fill up the unused bonus buffer region before the block tag with a
1517 * verifiable pattern. Filling the whole bonus area with non-zero data
1518 * helps ensure that all dnode traversal code properly skips the
1519 * interior regions of large dnodes.
1522 ztest_fill_unused_bonus(dmu_buf_t
*db
, void *end
, uint64_t obj
,
1523 objset_t
*os
, uint64_t gen
)
1527 ASSERT(IS_P2ALIGNED((char *)end
- (char *)db
->db_data
, 8));
1529 for (bonusp
= db
->db_data
; bonusp
< (uint64_t *)end
; bonusp
++) {
1530 uint64_t token
= ZTEST_BONUS_FILL_TOKEN(obj
, dmu_objset_id(os
),
1531 gen
, bonusp
- (uint64_t *)db
->db_data
);
1537 * Verify that the unused area of a bonus buffer is filled with the
1541 ztest_verify_unused_bonus(dmu_buf_t
*db
, void *end
, uint64_t obj
,
1542 objset_t
*os
, uint64_t gen
)
1546 for (bonusp
= db
->db_data
; bonusp
< (uint64_t *)end
; bonusp
++) {
1547 uint64_t token
= ZTEST_BONUS_FILL_TOKEN(obj
, dmu_objset_id(os
),
1548 gen
, bonusp
- (uint64_t *)db
->db_data
);
1549 VERIFY3U(*bonusp
, ==, token
);
1557 #define lrz_type lr_mode
1558 #define lrz_blocksize lr_uid
1559 #define lrz_ibshift lr_gid
1560 #define lrz_bonustype lr_rdev
1561 #define lrz_dnodesize lr_crtime[1]
1564 ztest_log_create(ztest_ds_t
*zd
, dmu_tx_t
*tx
, lr_create_t
*lr
)
1566 char *name
= (void *)(lr
+ 1); /* name follows lr */
1567 size_t namesize
= strlen(name
) + 1;
1570 if (zil_replaying(zd
->zd_zilog
, tx
))
1573 itx
= zil_itx_create(TX_CREATE
, sizeof (*lr
) + namesize
);
1574 bcopy(&lr
->lr_common
+ 1, &itx
->itx_lr
+ 1,
1575 sizeof (*lr
) + namesize
- sizeof (lr_t
));
1577 zil_itx_assign(zd
->zd_zilog
, itx
, tx
);
1581 ztest_log_remove(ztest_ds_t
*zd
, dmu_tx_t
*tx
, lr_remove_t
*lr
, uint64_t object
)
1583 char *name
= (void *)(lr
+ 1); /* name follows lr */
1584 size_t namesize
= strlen(name
) + 1;
1587 if (zil_replaying(zd
->zd_zilog
, tx
))
1590 itx
= zil_itx_create(TX_REMOVE
, sizeof (*lr
) + namesize
);
1591 bcopy(&lr
->lr_common
+ 1, &itx
->itx_lr
+ 1,
1592 sizeof (*lr
) + namesize
- sizeof (lr_t
));
1594 itx
->itx_oid
= object
;
1595 zil_itx_assign(zd
->zd_zilog
, itx
, tx
);
1599 ztest_log_write(ztest_ds_t
*zd
, dmu_tx_t
*tx
, lr_write_t
*lr
)
1602 itx_wr_state_t write_state
= ztest_random(WR_NUM_STATES
);
1604 if (zil_replaying(zd
->zd_zilog
, tx
))
1607 if (lr
->lr_length
> ZIL_MAX_LOG_DATA
)
1608 write_state
= WR_INDIRECT
;
1610 itx
= zil_itx_create(TX_WRITE
,
1611 sizeof (*lr
) + (write_state
== WR_COPIED
? lr
->lr_length
: 0));
1613 if (write_state
== WR_COPIED
&&
1614 dmu_read(zd
->zd_os
, lr
->lr_foid
, lr
->lr_offset
, lr
->lr_length
,
1615 ((lr_write_t
*)&itx
->itx_lr
) + 1, DMU_READ_NO_PREFETCH
) != 0) {
1616 zil_itx_destroy(itx
);
1617 itx
= zil_itx_create(TX_WRITE
, sizeof (*lr
));
1618 write_state
= WR_NEED_COPY
;
1620 itx
->itx_private
= zd
;
1621 itx
->itx_wr_state
= write_state
;
1622 itx
->itx_sync
= (ztest_random(8) == 0);
1624 bcopy(&lr
->lr_common
+ 1, &itx
->itx_lr
+ 1,
1625 sizeof (*lr
) - sizeof (lr_t
));
1627 zil_itx_assign(zd
->zd_zilog
, itx
, tx
);
1631 ztest_log_truncate(ztest_ds_t
*zd
, dmu_tx_t
*tx
, lr_truncate_t
*lr
)
1635 if (zil_replaying(zd
->zd_zilog
, tx
))
1638 itx
= zil_itx_create(TX_TRUNCATE
, sizeof (*lr
));
1639 bcopy(&lr
->lr_common
+ 1, &itx
->itx_lr
+ 1,
1640 sizeof (*lr
) - sizeof (lr_t
));
1642 itx
->itx_sync
= B_FALSE
;
1643 zil_itx_assign(zd
->zd_zilog
, itx
, tx
);
1647 ztest_log_setattr(ztest_ds_t
*zd
, dmu_tx_t
*tx
, lr_setattr_t
*lr
)
1651 if (zil_replaying(zd
->zd_zilog
, tx
))
1654 itx
= zil_itx_create(TX_SETATTR
, sizeof (*lr
));
1655 bcopy(&lr
->lr_common
+ 1, &itx
->itx_lr
+ 1,
1656 sizeof (*lr
) - sizeof (lr_t
));
1658 itx
->itx_sync
= B_FALSE
;
1659 zil_itx_assign(zd
->zd_zilog
, itx
, tx
);
1666 ztest_replay_create(ztest_ds_t
*zd
, lr_create_t
*lr
, boolean_t byteswap
)
1668 char *name
= (void *)(lr
+ 1); /* name follows lr */
1669 objset_t
*os
= zd
->zd_os
;
1670 ztest_block_tag_t
*bbt
;
1678 byteswap_uint64_array(lr
, sizeof (*lr
));
1680 ASSERT(lr
->lr_doid
== ZTEST_DIROBJ
);
1681 ASSERT(name
[0] != '\0');
1683 tx
= dmu_tx_create(os
);
1685 dmu_tx_hold_zap(tx
, lr
->lr_doid
, B_TRUE
, name
);
1687 if (lr
->lrz_type
== DMU_OT_ZAP_OTHER
) {
1688 dmu_tx_hold_zap(tx
, DMU_NEW_OBJECT
, B_TRUE
, NULL
);
1690 dmu_tx_hold_bonus(tx
, DMU_NEW_OBJECT
);
1693 txg
= ztest_tx_assign(tx
, TXG_WAIT
, FTAG
);
1697 ASSERT(dmu_objset_zil(os
)->zl_replay
== !!lr
->lr_foid
);
1698 bonuslen
= DN_BONUS_SIZE(lr
->lrz_dnodesize
);
1700 if (lr
->lrz_type
== DMU_OT_ZAP_OTHER
) {
1701 if (lr
->lr_foid
== 0) {
1702 lr
->lr_foid
= zap_create_dnsize(os
,
1703 lr
->lrz_type
, lr
->lrz_bonustype
,
1704 bonuslen
, lr
->lrz_dnodesize
, tx
);
1706 error
= zap_create_claim_dnsize(os
, lr
->lr_foid
,
1707 lr
->lrz_type
, lr
->lrz_bonustype
,
1708 bonuslen
, lr
->lrz_dnodesize
, tx
);
1711 if (lr
->lr_foid
== 0) {
1712 lr
->lr_foid
= dmu_object_alloc_dnsize(os
,
1713 lr
->lrz_type
, 0, lr
->lrz_bonustype
,
1714 bonuslen
, lr
->lrz_dnodesize
, tx
);
1716 error
= dmu_object_claim_dnsize(os
, lr
->lr_foid
,
1717 lr
->lrz_type
, 0, lr
->lrz_bonustype
,
1718 bonuslen
, lr
->lrz_dnodesize
, tx
);
1723 ASSERT3U(error
, ==, EEXIST
);
1724 ASSERT(zd
->zd_zilog
->zl_replay
);
1729 ASSERT(lr
->lr_foid
!= 0);
1731 if (lr
->lrz_type
!= DMU_OT_ZAP_OTHER
)
1732 VERIFY3U(0, ==, dmu_object_set_blocksize(os
, lr
->lr_foid
,
1733 lr
->lrz_blocksize
, lr
->lrz_ibshift
, tx
));
1735 VERIFY3U(0, ==, dmu_bonus_hold(os
, lr
->lr_foid
, FTAG
, &db
));
1736 bbt
= ztest_bt_bonus(db
);
1737 dmu_buf_will_dirty(db
, tx
);
1738 ztest_bt_generate(bbt
, os
, lr
->lr_foid
, lr
->lrz_dnodesize
, -1ULL,
1739 lr
->lr_gen
, txg
, txg
);
1740 ztest_fill_unused_bonus(db
, bbt
, lr
->lr_foid
, os
, lr
->lr_gen
);
1741 dmu_buf_rele(db
, FTAG
);
1743 VERIFY3U(0, ==, zap_add(os
, lr
->lr_doid
, name
, sizeof (uint64_t), 1,
1746 (void) ztest_log_create(zd
, tx
, lr
);
1754 ztest_replay_remove(ztest_ds_t
*zd
, lr_remove_t
*lr
, boolean_t byteswap
)
1756 char *name
= (void *)(lr
+ 1); /* name follows lr */
1757 objset_t
*os
= zd
->zd_os
;
1758 dmu_object_info_t doi
;
1760 uint64_t object
, txg
;
1763 byteswap_uint64_array(lr
, sizeof (*lr
));
1765 ASSERT(lr
->lr_doid
== ZTEST_DIROBJ
);
1766 ASSERT(name
[0] != '\0');
1769 zap_lookup(os
, lr
->lr_doid
, name
, sizeof (object
), 1, &object
));
1770 ASSERT(object
!= 0);
1772 ztest_object_lock(zd
, object
, RL_WRITER
);
1774 VERIFY3U(0, ==, dmu_object_info(os
, object
, &doi
));
1776 tx
= dmu_tx_create(os
);
1778 dmu_tx_hold_zap(tx
, lr
->lr_doid
, B_FALSE
, name
);
1779 dmu_tx_hold_free(tx
, object
, 0, DMU_OBJECT_END
);
1781 txg
= ztest_tx_assign(tx
, TXG_WAIT
, FTAG
);
1783 ztest_object_unlock(zd
, object
);
1787 if (doi
.doi_type
== DMU_OT_ZAP_OTHER
) {
1788 VERIFY3U(0, ==, zap_destroy(os
, object
, tx
));
1790 VERIFY3U(0, ==, dmu_object_free(os
, object
, tx
));
1793 VERIFY3U(0, ==, zap_remove(os
, lr
->lr_doid
, name
, tx
));
1795 (void) ztest_log_remove(zd
, tx
, lr
, object
);
1799 ztest_object_unlock(zd
, object
);
1805 ztest_replay_write(ztest_ds_t
*zd
, lr_write_t
*lr
, boolean_t byteswap
)
1807 objset_t
*os
= zd
->zd_os
;
1808 void *data
= lr
+ 1; /* data follows lr */
1809 uint64_t offset
, length
;
1810 ztest_block_tag_t
*bt
= data
;
1811 ztest_block_tag_t
*bbt
;
1812 uint64_t gen
, txg
, lrtxg
, crtxg
;
1813 dmu_object_info_t doi
;
1816 arc_buf_t
*abuf
= NULL
;
1820 byteswap_uint64_array(lr
, sizeof (*lr
));
1822 offset
= lr
->lr_offset
;
1823 length
= lr
->lr_length
;
1825 /* If it's a dmu_sync() block, write the whole block */
1826 if (lr
->lr_common
.lrc_reclen
== sizeof (lr_write_t
)) {
1827 uint64_t blocksize
= BP_GET_LSIZE(&lr
->lr_blkptr
);
1828 if (length
< blocksize
) {
1829 offset
-= offset
% blocksize
;
1834 if (bt
->bt_magic
== BSWAP_64(BT_MAGIC
))
1835 byteswap_uint64_array(bt
, sizeof (*bt
));
1837 if (bt
->bt_magic
!= BT_MAGIC
)
1840 ztest_object_lock(zd
, lr
->lr_foid
, RL_READER
);
1841 rl
= ztest_range_lock(zd
, lr
->lr_foid
, offset
, length
, RL_WRITER
);
1843 VERIFY3U(0, ==, dmu_bonus_hold(os
, lr
->lr_foid
, FTAG
, &db
));
1845 dmu_object_info_from_db(db
, &doi
);
1847 bbt
= ztest_bt_bonus(db
);
1848 ASSERT3U(bbt
->bt_magic
, ==, BT_MAGIC
);
1850 crtxg
= bbt
->bt_crtxg
;
1851 lrtxg
= lr
->lr_common
.lrc_txg
;
1853 tx
= dmu_tx_create(os
);
1855 dmu_tx_hold_write(tx
, lr
->lr_foid
, offset
, length
);
1857 if (ztest_random(8) == 0 && length
== doi
.doi_data_block_size
&&
1858 P2PHASE(offset
, length
) == 0)
1859 abuf
= dmu_request_arcbuf(db
, length
);
1861 txg
= ztest_tx_assign(tx
, TXG_WAIT
, FTAG
);
1864 dmu_return_arcbuf(abuf
);
1865 dmu_buf_rele(db
, FTAG
);
1866 ztest_range_unlock(zd
, rl
);
1867 ztest_object_unlock(zd
, lr
->lr_foid
);
1873 * Usually, verify the old data before writing new data --
1874 * but not always, because we also want to verify correct
1875 * behavior when the data was not recently read into cache.
1877 ASSERT(offset
% doi
.doi_data_block_size
== 0);
1878 if (ztest_random(4) != 0) {
1879 int prefetch
= ztest_random(2) ?
1880 DMU_READ_PREFETCH
: DMU_READ_NO_PREFETCH
;
1881 ztest_block_tag_t rbt
;
1883 VERIFY(dmu_read(os
, lr
->lr_foid
, offset
,
1884 sizeof (rbt
), &rbt
, prefetch
) == 0);
1885 if (rbt
.bt_magic
== BT_MAGIC
) {
1886 ztest_bt_verify(&rbt
, os
, lr
->lr_foid
, 0,
1887 offset
, gen
, txg
, crtxg
);
1892 * Writes can appear to be newer than the bonus buffer because
1893 * the ztest_get_data() callback does a dmu_read() of the
1894 * open-context data, which may be different than the data
1895 * as it was when the write was generated.
1897 if (zd
->zd_zilog
->zl_replay
) {
1898 ztest_bt_verify(bt
, os
, lr
->lr_foid
, 0, offset
,
1899 MAX(gen
, bt
->bt_gen
), MAX(txg
, lrtxg
),
1904 * Set the bt's gen/txg to the bonus buffer's gen/txg
1905 * so that all of the usual ASSERTs will work.
1907 ztest_bt_generate(bt
, os
, lr
->lr_foid
, 0, offset
, gen
, txg
,
1912 dmu_write(os
, lr
->lr_foid
, offset
, length
, data
, tx
);
1914 bcopy(data
, abuf
->b_data
, length
);
1915 dmu_assign_arcbuf(db
, offset
, abuf
, tx
);
1918 (void) ztest_log_write(zd
, tx
, lr
);
1920 dmu_buf_rele(db
, FTAG
);
1924 ztest_range_unlock(zd
, rl
);
1925 ztest_object_unlock(zd
, lr
->lr_foid
);
1931 ztest_replay_truncate(ztest_ds_t
*zd
, lr_truncate_t
*lr
, boolean_t byteswap
)
1933 objset_t
*os
= zd
->zd_os
;
1939 byteswap_uint64_array(lr
, sizeof (*lr
));
1941 ztest_object_lock(zd
, lr
->lr_foid
, RL_READER
);
1942 rl
= ztest_range_lock(zd
, lr
->lr_foid
, lr
->lr_offset
, lr
->lr_length
,
1945 tx
= dmu_tx_create(os
);
1947 dmu_tx_hold_free(tx
, lr
->lr_foid
, lr
->lr_offset
, lr
->lr_length
);
1949 txg
= ztest_tx_assign(tx
, TXG_WAIT
, FTAG
);
1951 ztest_range_unlock(zd
, rl
);
1952 ztest_object_unlock(zd
, lr
->lr_foid
);
1956 VERIFY(dmu_free_range(os
, lr
->lr_foid
, lr
->lr_offset
,
1957 lr
->lr_length
, tx
) == 0);
1959 (void) ztest_log_truncate(zd
, tx
, lr
);
1963 ztest_range_unlock(zd
, rl
);
1964 ztest_object_unlock(zd
, lr
->lr_foid
);
1970 ztest_replay_setattr(ztest_ds_t
*zd
, lr_setattr_t
*lr
, boolean_t byteswap
)
1972 objset_t
*os
= zd
->zd_os
;
1975 ztest_block_tag_t
*bbt
;
1976 uint64_t txg
, lrtxg
, crtxg
, dnodesize
;
1979 byteswap_uint64_array(lr
, sizeof (*lr
));
1981 ztest_object_lock(zd
, lr
->lr_foid
, RL_WRITER
);
1983 VERIFY3U(0, ==, dmu_bonus_hold(os
, lr
->lr_foid
, FTAG
, &db
));
1985 tx
= dmu_tx_create(os
);
1986 dmu_tx_hold_bonus(tx
, lr
->lr_foid
);
1988 txg
= ztest_tx_assign(tx
, TXG_WAIT
, FTAG
);
1990 dmu_buf_rele(db
, FTAG
);
1991 ztest_object_unlock(zd
, lr
->lr_foid
);
1995 bbt
= ztest_bt_bonus(db
);
1996 ASSERT3U(bbt
->bt_magic
, ==, BT_MAGIC
);
1997 crtxg
= bbt
->bt_crtxg
;
1998 lrtxg
= lr
->lr_common
.lrc_txg
;
1999 dnodesize
= bbt
->bt_dnodesize
;
2001 if (zd
->zd_zilog
->zl_replay
) {
2002 ASSERT(lr
->lr_size
!= 0);
2003 ASSERT(lr
->lr_mode
!= 0);
2007 * Randomly change the size and increment the generation.
2009 lr
->lr_size
= (ztest_random(db
->db_size
/ sizeof (*bbt
)) + 1) *
2011 lr
->lr_mode
= bbt
->bt_gen
+ 1;
2016 * Verify that the current bonus buffer is not newer than our txg.
2018 ztest_bt_verify(bbt
, os
, lr
->lr_foid
, dnodesize
, -1ULL, lr
->lr_mode
,
2019 MAX(txg
, lrtxg
), crtxg
);
2021 dmu_buf_will_dirty(db
, tx
);
2023 ASSERT3U(lr
->lr_size
, >=, sizeof (*bbt
));
2024 ASSERT3U(lr
->lr_size
, <=, db
->db_size
);
2025 VERIFY0(dmu_set_bonus(db
, lr
->lr_size
, tx
));
2026 bbt
= ztest_bt_bonus(db
);
2028 ztest_bt_generate(bbt
, os
, lr
->lr_foid
, dnodesize
, -1ULL, lr
->lr_mode
,
2030 ztest_fill_unused_bonus(db
, bbt
, lr
->lr_foid
, os
, bbt
->bt_gen
);
2031 dmu_buf_rele(db
, FTAG
);
2033 (void) ztest_log_setattr(zd
, tx
, lr
);
2037 ztest_object_unlock(zd
, lr
->lr_foid
);
2042 zil_replay_func_t ztest_replay_vector
[TX_MAX_TYPE
] = {
2043 NULL
, /* 0 no such transaction type */
2044 (zil_replay_func_t
)ztest_replay_create
, /* TX_CREATE */
2045 NULL
, /* TX_MKDIR */
2046 NULL
, /* TX_MKXATTR */
2047 NULL
, /* TX_SYMLINK */
2048 (zil_replay_func_t
)ztest_replay_remove
, /* TX_REMOVE */
2049 NULL
, /* TX_RMDIR */
2051 NULL
, /* TX_RENAME */
2052 (zil_replay_func_t
)ztest_replay_write
, /* TX_WRITE */
2053 (zil_replay_func_t
)ztest_replay_truncate
, /* TX_TRUNCATE */
2054 (zil_replay_func_t
)ztest_replay_setattr
, /* TX_SETATTR */
2056 NULL
, /* TX_CREATE_ACL */
2057 NULL
, /* TX_CREATE_ATTR */
2058 NULL
, /* TX_CREATE_ACL_ATTR */
2059 NULL
, /* TX_MKDIR_ACL */
2060 NULL
, /* TX_MKDIR_ATTR */
2061 NULL
, /* TX_MKDIR_ACL_ATTR */
2062 NULL
, /* TX_WRITE2 */
2066 * ZIL get_data callbacks
2068 typedef struct ztest_zgd_private
{
2072 } ztest_zgd_private_t
;
2075 ztest_get_done(zgd_t
*zgd
, int error
)
2077 ztest_zgd_private_t
*zzp
= zgd
->zgd_private
;
2078 ztest_ds_t
*zd
= zzp
->z_zd
;
2079 uint64_t object
= zzp
->z_object
;
2082 dmu_buf_rele(zgd
->zgd_db
, zgd
);
2084 ztest_range_unlock(zd
, zzp
->z_rl
);
2085 ztest_object_unlock(zd
, object
);
2087 if (error
== 0 && zgd
->zgd_bp
)
2088 zil_add_block(zgd
->zgd_zilog
, zgd
->zgd_bp
);
2090 umem_free(zgd
, sizeof (*zgd
));
2091 umem_free(zzp
, sizeof (*zzp
));
2095 ztest_get_data(void *arg
, lr_write_t
*lr
, char *buf
, zio_t
*zio
)
2097 ztest_ds_t
*zd
= arg
;
2098 objset_t
*os
= zd
->zd_os
;
2099 uint64_t object
= lr
->lr_foid
;
2100 uint64_t offset
= lr
->lr_offset
;
2101 uint64_t size
= lr
->lr_length
;
2102 blkptr_t
*bp
= &lr
->lr_blkptr
;
2103 uint64_t txg
= lr
->lr_common
.lrc_txg
;
2105 dmu_object_info_t doi
;
2109 ztest_zgd_private_t
*zgd_private
;
2111 ztest_object_lock(zd
, object
, RL_READER
);
2112 error
= dmu_bonus_hold(os
, object
, FTAG
, &db
);
2114 ztest_object_unlock(zd
, object
);
2118 crtxg
= ztest_bt_bonus(db
)->bt_crtxg
;
2120 if (crtxg
== 0 || crtxg
> txg
) {
2121 dmu_buf_rele(db
, FTAG
);
2122 ztest_object_unlock(zd
, object
);
2126 dmu_object_info_from_db(db
, &doi
);
2127 dmu_buf_rele(db
, FTAG
);
2130 zgd
= umem_zalloc(sizeof (*zgd
), UMEM_NOFAIL
);
2131 zgd
->zgd_zilog
= zd
->zd_zilog
;
2132 zgd_private
= umem_zalloc(sizeof (ztest_zgd_private_t
), UMEM_NOFAIL
);
2133 zgd_private
->z_zd
= zd
;
2134 zgd_private
->z_object
= object
;
2135 zgd
->zgd_private
= zgd_private
;
2137 if (buf
!= NULL
) { /* immediate write */
2138 zgd_private
->z_rl
= ztest_range_lock(zd
, object
, offset
, size
,
2141 error
= dmu_read(os
, object
, offset
, size
, buf
,
2142 DMU_READ_NO_PREFETCH
);
2145 size
= doi
.doi_data_block_size
;
2147 offset
= P2ALIGN(offset
, size
);
2149 ASSERT(offset
< size
);
2153 zgd_private
->z_rl
= ztest_range_lock(zd
, object
, offset
, size
,
2156 error
= dmu_buf_hold(os
, object
, offset
, zgd
, &db
,
2157 DMU_READ_NO_PREFETCH
);
2160 blkptr_t
*obp
= dmu_buf_get_blkptr(db
);
2162 ASSERT(BP_IS_HOLE(bp
));
2169 ASSERT(db
->db_offset
== offset
);
2170 ASSERT(db
->db_size
== size
);
2172 error
= dmu_sync(zio
, lr
->lr_common
.lrc_txg
,
2173 ztest_get_done
, zgd
);
2180 ztest_get_done(zgd
, error
);
2186 ztest_lr_alloc(size_t lrsize
, char *name
)
2189 size_t namesize
= name
? strlen(name
) + 1 : 0;
2191 lr
= umem_zalloc(lrsize
+ namesize
, UMEM_NOFAIL
);
2194 bcopy(name
, lr
+ lrsize
, namesize
);
2200 ztest_lr_free(void *lr
, size_t lrsize
, char *name
)
2202 size_t namesize
= name
? strlen(name
) + 1 : 0;
2204 umem_free(lr
, lrsize
+ namesize
);
2208 * Lookup a bunch of objects. Returns the number of objects not found.
2211 ztest_lookup(ztest_ds_t
*zd
, ztest_od_t
*od
, int count
)
2217 ASSERT(mutex_held(&zd
->zd_dirobj_lock
));
2219 for (i
= 0; i
< count
; i
++, od
++) {
2221 error
= zap_lookup(zd
->zd_os
, od
->od_dir
, od
->od_name
,
2222 sizeof (uint64_t), 1, &od
->od_object
);
2224 ASSERT(error
== ENOENT
);
2225 ASSERT(od
->od_object
== 0);
2229 ztest_block_tag_t
*bbt
;
2230 dmu_object_info_t doi
;
2232 ASSERT(od
->od_object
!= 0);
2233 ASSERT(missing
== 0); /* there should be no gaps */
2235 ztest_object_lock(zd
, od
->od_object
, RL_READER
);
2236 VERIFY3U(0, ==, dmu_bonus_hold(zd
->zd_os
,
2237 od
->od_object
, FTAG
, &db
));
2238 dmu_object_info_from_db(db
, &doi
);
2239 bbt
= ztest_bt_bonus(db
);
2240 ASSERT3U(bbt
->bt_magic
, ==, BT_MAGIC
);
2241 od
->od_type
= doi
.doi_type
;
2242 od
->od_blocksize
= doi
.doi_data_block_size
;
2243 od
->od_gen
= bbt
->bt_gen
;
2244 dmu_buf_rele(db
, FTAG
);
2245 ztest_object_unlock(zd
, od
->od_object
);
2253 ztest_create(ztest_ds_t
*zd
, ztest_od_t
*od
, int count
)
2258 ASSERT(mutex_held(&zd
->zd_dirobj_lock
));
2260 for (i
= 0; i
< count
; i
++, od
++) {
2267 lr_create_t
*lr
= ztest_lr_alloc(sizeof (*lr
), od
->od_name
);
2269 lr
->lr_doid
= od
->od_dir
;
2270 lr
->lr_foid
= 0; /* 0 to allocate, > 0 to claim */
2271 lr
->lrz_type
= od
->od_crtype
;
2272 lr
->lrz_blocksize
= od
->od_crblocksize
;
2273 lr
->lrz_ibshift
= ztest_random_ibshift();
2274 lr
->lrz_bonustype
= DMU_OT_UINT64_OTHER
;
2275 lr
->lrz_dnodesize
= od
->od_crdnodesize
;
2276 lr
->lr_gen
= od
->od_crgen
;
2277 lr
->lr_crtime
[0] = time(NULL
);
2279 if (ztest_replay_create(zd
, lr
, B_FALSE
) != 0) {
2280 ASSERT(missing
== 0);
2284 od
->od_object
= lr
->lr_foid
;
2285 od
->od_type
= od
->od_crtype
;
2286 od
->od_blocksize
= od
->od_crblocksize
;
2287 od
->od_gen
= od
->od_crgen
;
2288 ASSERT(od
->od_object
!= 0);
2291 ztest_lr_free(lr
, sizeof (*lr
), od
->od_name
);
2298 ztest_remove(ztest_ds_t
*zd
, ztest_od_t
*od
, int count
)
2304 ASSERT(mutex_held(&zd
->zd_dirobj_lock
));
2308 for (i
= count
- 1; i
>= 0; i
--, od
--) {
2315 * No object was found.
2317 if (od
->od_object
== 0)
2320 lr_remove_t
*lr
= ztest_lr_alloc(sizeof (*lr
), od
->od_name
);
2322 lr
->lr_doid
= od
->od_dir
;
2324 if ((error
= ztest_replay_remove(zd
, lr
, B_FALSE
)) != 0) {
2325 ASSERT3U(error
, ==, ENOSPC
);
2330 ztest_lr_free(lr
, sizeof (*lr
), od
->od_name
);
2337 ztest_write(ztest_ds_t
*zd
, uint64_t object
, uint64_t offset
, uint64_t size
,
2343 lr
= ztest_lr_alloc(sizeof (*lr
) + size
, NULL
);
2345 lr
->lr_foid
= object
;
2346 lr
->lr_offset
= offset
;
2347 lr
->lr_length
= size
;
2349 BP_ZERO(&lr
->lr_blkptr
);
2351 bcopy(data
, lr
+ 1, size
);
2353 error
= ztest_replay_write(zd
, lr
, B_FALSE
);
2355 ztest_lr_free(lr
, sizeof (*lr
) + size
, NULL
);
2361 ztest_truncate(ztest_ds_t
*zd
, uint64_t object
, uint64_t offset
, uint64_t size
)
2366 lr
= ztest_lr_alloc(sizeof (*lr
), NULL
);
2368 lr
->lr_foid
= object
;
2369 lr
->lr_offset
= offset
;
2370 lr
->lr_length
= size
;
2372 error
= ztest_replay_truncate(zd
, lr
, B_FALSE
);
2374 ztest_lr_free(lr
, sizeof (*lr
), NULL
);
2380 ztest_setattr(ztest_ds_t
*zd
, uint64_t object
)
2385 lr
= ztest_lr_alloc(sizeof (*lr
), NULL
);
2387 lr
->lr_foid
= object
;
2391 error
= ztest_replay_setattr(zd
, lr
, B_FALSE
);
2393 ztest_lr_free(lr
, sizeof (*lr
), NULL
);
2399 ztest_prealloc(ztest_ds_t
*zd
, uint64_t object
, uint64_t offset
, uint64_t size
)
2401 objset_t
*os
= zd
->zd_os
;
2406 txg_wait_synced(dmu_objset_pool(os
), 0);
2408 ztest_object_lock(zd
, object
, RL_READER
);
2409 rl
= ztest_range_lock(zd
, object
, offset
, size
, RL_WRITER
);
2411 tx
= dmu_tx_create(os
);
2413 dmu_tx_hold_write(tx
, object
, offset
, size
);
2415 txg
= ztest_tx_assign(tx
, TXG_WAIT
, FTAG
);
2418 dmu_prealloc(os
, object
, offset
, size
, tx
);
2420 txg_wait_synced(dmu_objset_pool(os
), txg
);
2422 (void) dmu_free_long_range(os
, object
, offset
, size
);
2425 ztest_range_unlock(zd
, rl
);
2426 ztest_object_unlock(zd
, object
);
2430 ztest_io(ztest_ds_t
*zd
, uint64_t object
, uint64_t offset
)
2433 ztest_block_tag_t wbt
;
2434 dmu_object_info_t doi
;
2435 enum ztest_io_type io_type
;
2439 VERIFY(dmu_object_info(zd
->zd_os
, object
, &doi
) == 0);
2440 blocksize
= doi
.doi_data_block_size
;
2441 data
= umem_alloc(blocksize
, UMEM_NOFAIL
);
2444 * Pick an i/o type at random, biased toward writing block tags.
2446 io_type
= ztest_random(ZTEST_IO_TYPES
);
2447 if (ztest_random(2) == 0)
2448 io_type
= ZTEST_IO_WRITE_TAG
;
2450 (void) rw_rdlock(&zd
->zd_zilog_lock
);
2454 case ZTEST_IO_WRITE_TAG
:
2455 ztest_bt_generate(&wbt
, zd
->zd_os
, object
, doi
.doi_dnodesize
,
2457 (void) ztest_write(zd
, object
, offset
, sizeof (wbt
), &wbt
);
2460 case ZTEST_IO_WRITE_PATTERN
:
2461 (void) memset(data
, 'a' + (object
+ offset
) % 5, blocksize
);
2462 if (ztest_random(2) == 0) {
2464 * Induce fletcher2 collisions to ensure that
2465 * zio_ddt_collision() detects and resolves them
2466 * when using fletcher2-verify for deduplication.
2468 ((uint64_t *)data
)[0] ^= 1ULL << 63;
2469 ((uint64_t *)data
)[4] ^= 1ULL << 63;
2471 (void) ztest_write(zd
, object
, offset
, blocksize
, data
);
2474 case ZTEST_IO_WRITE_ZEROES
:
2475 bzero(data
, blocksize
);
2476 (void) ztest_write(zd
, object
, offset
, blocksize
, data
);
2479 case ZTEST_IO_TRUNCATE
:
2480 (void) ztest_truncate(zd
, object
, offset
, blocksize
);
2483 case ZTEST_IO_SETATTR
:
2484 (void) ztest_setattr(zd
, object
);
2489 case ZTEST_IO_REWRITE
:
2490 (void) rw_rdlock(&ztest_name_lock
);
2491 err
= ztest_dsl_prop_set_uint64(zd
->zd_name
,
2492 ZFS_PROP_CHECKSUM
, spa_dedup_checksum(ztest_spa
),
2494 VERIFY(err
== 0 || err
== ENOSPC
);
2495 err
= ztest_dsl_prop_set_uint64(zd
->zd_name
,
2496 ZFS_PROP_COMPRESSION
,
2497 ztest_random_dsl_prop(ZFS_PROP_COMPRESSION
),
2499 VERIFY(err
== 0 || err
== ENOSPC
);
2500 (void) rw_unlock(&ztest_name_lock
);
2502 VERIFY0(dmu_read(zd
->zd_os
, object
, offset
, blocksize
, data
,
2503 DMU_READ_NO_PREFETCH
));
2505 (void) ztest_write(zd
, object
, offset
, blocksize
, data
);
2509 (void) rw_unlock(&zd
->zd_zilog_lock
);
2511 umem_free(data
, blocksize
);
2515 * Initialize an object description template.
2518 ztest_od_init(ztest_od_t
*od
, uint64_t id
, char *tag
, uint64_t index
,
2519 dmu_object_type_t type
, uint64_t blocksize
, uint64_t dnodesize
,
2522 od
->od_dir
= ZTEST_DIROBJ
;
2525 od
->od_crtype
= type
;
2526 od
->od_crblocksize
= blocksize
? blocksize
: ztest_random_blocksize();
2527 od
->od_crdnodesize
= dnodesize
? dnodesize
: ztest_random_dnodesize();
2530 od
->od_type
= DMU_OT_NONE
;
2531 od
->od_blocksize
= 0;
2534 (void) snprintf(od
->od_name
, sizeof (od
->od_name
), "%s(%lld)[%llu]",
2535 tag
, (longlong_t
)id
, (u_longlong_t
)index
);
2539 * Lookup or create the objects for a test using the od template.
2540 * If the objects do not all exist, or if 'remove' is specified,
2541 * remove any existing objects and create new ones. Otherwise,
2542 * use the existing objects.
2545 ztest_object_init(ztest_ds_t
*zd
, ztest_od_t
*od
, size_t size
, boolean_t remove
)
2547 int count
= size
/ sizeof (*od
);
2550 mutex_enter(&zd
->zd_dirobj_lock
);
2551 if ((ztest_lookup(zd
, od
, count
) != 0 || remove
) &&
2552 (ztest_remove(zd
, od
, count
) != 0 ||
2553 ztest_create(zd
, od
, count
) != 0))
2556 mutex_exit(&zd
->zd_dirobj_lock
);
2563 ztest_zil_commit(ztest_ds_t
*zd
, uint64_t id
)
2565 zilog_t
*zilog
= zd
->zd_zilog
;
2567 (void) rw_rdlock(&zd
->zd_zilog_lock
);
2569 zil_commit(zilog
, ztest_random(ZTEST_OBJECTS
));
2572 * Remember the committed values in zd, which is in parent/child
2573 * shared memory. If we die, the next iteration of ztest_run()
2574 * will verify that the log really does contain this record.
2576 mutex_enter(&zilog
->zl_lock
);
2577 ASSERT(zd
->zd_shared
!= NULL
);
2578 ASSERT3U(zd
->zd_shared
->zd_seq
, <=, zilog
->zl_commit_lr_seq
);
2579 zd
->zd_shared
->zd_seq
= zilog
->zl_commit_lr_seq
;
2580 mutex_exit(&zilog
->zl_lock
);
2582 (void) rw_unlock(&zd
->zd_zilog_lock
);
2586 * This function is designed to simulate the operations that occur during a
2587 * mount/unmount operation. We hold the dataset across these operations in an
2588 * attempt to expose any implicit assumptions about ZIL management.
2592 ztest_zil_remount(ztest_ds_t
*zd
, uint64_t id
)
2594 objset_t
*os
= zd
->zd_os
;
2597 * We grab the zd_dirobj_lock to ensure that no other thread is
2598 * updating the zil (i.e. adding in-memory log records) and the
2599 * zd_zilog_lock to block any I/O.
2601 mutex_enter(&zd
->zd_dirobj_lock
);
2602 (void) rw_wrlock(&zd
->zd_zilog_lock
);
2604 /* zfsvfs_teardown() */
2605 zil_close(zd
->zd_zilog
);
2607 /* zfsvfs_setup() */
2608 VERIFY(zil_open(os
, ztest_get_data
) == zd
->zd_zilog
);
2609 zil_replay(os
, zd
, ztest_replay_vector
);
2611 (void) rw_unlock(&zd
->zd_zilog_lock
);
2612 mutex_exit(&zd
->zd_dirobj_lock
);
2616 * Verify that we can't destroy an active pool, create an existing pool,
2617 * or create a pool with a bad vdev spec.
2621 ztest_spa_create_destroy(ztest_ds_t
*zd
, uint64_t id
)
2623 ztest_shared_opts_t
*zo
= &ztest_opts
;
2628 * Attempt to create using a bad file.
2630 nvroot
= make_vdev_root("/dev/bogus", NULL
, NULL
, 0, 0, 0, 0, 0, 1);
2631 VERIFY3U(ENOENT
, ==,
2632 spa_create("ztest_bad_file", nvroot
, NULL
, NULL
));
2633 nvlist_free(nvroot
);
2636 * Attempt to create using a bad mirror.
2638 nvroot
= make_vdev_root("/dev/bogus", NULL
, NULL
, 0, 0, 0, 0, 2, 1);
2639 VERIFY3U(ENOENT
, ==,
2640 spa_create("ztest_bad_mirror", nvroot
, NULL
, NULL
));
2641 nvlist_free(nvroot
);
2644 * Attempt to create an existing pool. It shouldn't matter
2645 * what's in the nvroot; we should fail with EEXIST.
2647 (void) rw_rdlock(&ztest_name_lock
);
2648 nvroot
= make_vdev_root("/dev/bogus", NULL
, NULL
, 0, 0, 0, 0, 0, 1);
2649 VERIFY3U(EEXIST
, ==, spa_create(zo
->zo_pool
, nvroot
, NULL
, NULL
));
2650 nvlist_free(nvroot
);
2651 VERIFY3U(0, ==, spa_open(zo
->zo_pool
, &spa
, FTAG
));
2652 VERIFY3U(EBUSY
, ==, spa_destroy(zo
->zo_pool
));
2653 spa_close(spa
, FTAG
);
2655 (void) rw_unlock(&ztest_name_lock
);
2660 ztest_spa_upgrade(ztest_ds_t
*zd
, uint64_t id
)
2663 uint64_t initial_version
= SPA_VERSION_INITIAL
;
2664 uint64_t version
, newversion
;
2665 nvlist_t
*nvroot
, *props
;
2668 mutex_enter(&ztest_vdev_lock
);
2669 name
= kmem_asprintf("%s_upgrade", ztest_opts
.zo_pool
);
2672 * Clean up from previous runs.
2674 (void) spa_destroy(name
);
2676 nvroot
= make_vdev_root(NULL
, NULL
, name
, ztest_opts
.zo_vdev_size
, 0,
2677 0, ztest_opts
.zo_raidz
, ztest_opts
.zo_mirrors
, 1);
2680 * If we're configuring a RAIDZ device then make sure that the
2681 * the initial version is capable of supporting that feature.
2683 switch (ztest_opts
.zo_raidz_parity
) {
2686 initial_version
= SPA_VERSION_INITIAL
;
2689 initial_version
= SPA_VERSION_RAIDZ2
;
2692 initial_version
= SPA_VERSION_RAIDZ3
;
2697 * Create a pool with a spa version that can be upgraded. Pick
2698 * a value between initial_version and SPA_VERSION_BEFORE_FEATURES.
2701 version
= ztest_random_spa_version(initial_version
);
2702 } while (version
> SPA_VERSION_BEFORE_FEATURES
);
2704 props
= fnvlist_alloc();
2705 fnvlist_add_uint64(props
,
2706 zpool_prop_to_name(ZPOOL_PROP_VERSION
), version
);
2707 VERIFY3S(spa_create(name
, nvroot
, props
, NULL
), ==, 0);
2708 fnvlist_free(nvroot
);
2709 fnvlist_free(props
);
2711 VERIFY3S(spa_open(name
, &spa
, FTAG
), ==, 0);
2712 VERIFY3U(spa_version(spa
), ==, version
);
2713 newversion
= ztest_random_spa_version(version
+ 1);
2715 if (ztest_opts
.zo_verbose
>= 4) {
2716 (void) printf("upgrading spa version from %llu to %llu\n",
2717 (u_longlong_t
)version
, (u_longlong_t
)newversion
);
2720 spa_upgrade(spa
, newversion
);
2721 VERIFY3U(spa_version(spa
), >, version
);
2722 VERIFY3U(spa_version(spa
), ==, fnvlist_lookup_uint64(spa
->spa_config
,
2723 zpool_prop_to_name(ZPOOL_PROP_VERSION
)));
2724 spa_close(spa
, FTAG
);
2727 mutex_exit(&ztest_vdev_lock
);
2731 vdev_lookup_by_path(vdev_t
*vd
, const char *path
)
2736 if (vd
->vdev_path
!= NULL
&& strcmp(path
, vd
->vdev_path
) == 0)
2739 for (c
= 0; c
< vd
->vdev_children
; c
++)
2740 if ((mvd
= vdev_lookup_by_path(vd
->vdev_child
[c
], path
)) !=
2748 * Find the first available hole which can be used as a top-level.
2751 find_vdev_hole(spa_t
*spa
)
2753 vdev_t
*rvd
= spa
->spa_root_vdev
;
2756 ASSERT(spa_config_held(spa
, SCL_VDEV
, RW_READER
) == SCL_VDEV
);
2758 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
2759 vdev_t
*cvd
= rvd
->vdev_child
[c
];
2761 if (cvd
->vdev_ishole
)
2768 * Verify that vdev_add() works as expected.
2772 ztest_vdev_add_remove(ztest_ds_t
*zd
, uint64_t id
)
2774 ztest_shared_t
*zs
= ztest_shared
;
2775 spa_t
*spa
= ztest_spa
;
2781 mutex_enter(&ztest_vdev_lock
);
2782 leaves
= MAX(zs
->zs_mirrors
+ zs
->zs_splits
, 1) * ztest_opts
.zo_raidz
;
2784 spa_config_enter(spa
, SCL_VDEV
, FTAG
, RW_READER
);
2786 ztest_shared
->zs_vdev_next_leaf
= find_vdev_hole(spa
) * leaves
;
2789 * If we have slogs then remove them 1/4 of the time.
2791 if (spa_has_slogs(spa
) && ztest_random(4) == 0) {
2793 * Grab the guid from the head of the log class rotor.
2795 guid
= spa_log_class(spa
)->mc_rotor
->mg_vd
->vdev_guid
;
2797 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
2800 * We have to grab the zs_name_lock as writer to
2801 * prevent a race between removing a slog (dmu_objset_find)
2802 * and destroying a dataset. Removing the slog will
2803 * grab a reference on the dataset which may cause
2804 * dsl_destroy_head() to fail with EBUSY thus
2805 * leaving the dataset in an inconsistent state.
2807 rw_wrlock(&ztest_name_lock
);
2808 error
= spa_vdev_remove(spa
, guid
, B_FALSE
);
2809 rw_unlock(&ztest_name_lock
);
2811 if (error
&& error
!= EEXIST
)
2812 fatal(0, "spa_vdev_remove() = %d", error
);
2814 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
2817 * Make 1/4 of the devices be log devices.
2819 nvroot
= make_vdev_root(NULL
, NULL
, NULL
,
2820 ztest_opts
.zo_vdev_size
, 0,
2821 ztest_random(4) == 0, ztest_opts
.zo_raidz
,
2824 error
= spa_vdev_add(spa
, nvroot
);
2825 nvlist_free(nvroot
);
2827 if (error
== ENOSPC
)
2828 ztest_record_enospc("spa_vdev_add");
2829 else if (error
!= 0)
2830 fatal(0, "spa_vdev_add() = %d", error
);
2833 mutex_exit(&ztest_vdev_lock
);
2837 * Verify that adding/removing aux devices (l2arc, hot spare) works as expected.
2841 ztest_vdev_aux_add_remove(ztest_ds_t
*zd
, uint64_t id
)
2843 ztest_shared_t
*zs
= ztest_shared
;
2844 spa_t
*spa
= ztest_spa
;
2845 vdev_t
*rvd
= spa
->spa_root_vdev
;
2846 spa_aux_vdev_t
*sav
;
2852 path
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
2854 if (ztest_random(2) == 0) {
2855 sav
= &spa
->spa_spares
;
2856 aux
= ZPOOL_CONFIG_SPARES
;
2858 sav
= &spa
->spa_l2cache
;
2859 aux
= ZPOOL_CONFIG_L2CACHE
;
2862 mutex_enter(&ztest_vdev_lock
);
2864 spa_config_enter(spa
, SCL_VDEV
, FTAG
, RW_READER
);
2866 if (sav
->sav_count
!= 0 && ztest_random(4) == 0) {
2868 * Pick a random device to remove.
2870 guid
= sav
->sav_vdevs
[ztest_random(sav
->sav_count
)]->vdev_guid
;
2873 * Find an unused device we can add.
2875 zs
->zs_vdev_aux
= 0;
2878 (void) snprintf(path
, MAXPATHLEN
, ztest_aux_template
,
2879 ztest_opts
.zo_dir
, ztest_opts
.zo_pool
, aux
,
2881 for (c
= 0; c
< sav
->sav_count
; c
++)
2882 if (strcmp(sav
->sav_vdevs
[c
]->vdev_path
,
2885 if (c
== sav
->sav_count
&&
2886 vdev_lookup_by_path(rvd
, path
) == NULL
)
2892 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
2898 nvlist_t
*nvroot
= make_vdev_root(NULL
, aux
, NULL
,
2899 (ztest_opts
.zo_vdev_size
* 5) / 4, 0, 0, 0, 0, 1);
2900 error
= spa_vdev_add(spa
, nvroot
);
2902 fatal(0, "spa_vdev_add(%p) = %d", nvroot
, error
);
2903 nvlist_free(nvroot
);
2906 * Remove an existing device. Sometimes, dirty its
2907 * vdev state first to make sure we handle removal
2908 * of devices that have pending state changes.
2910 if (ztest_random(2) == 0)
2911 (void) vdev_online(spa
, guid
, 0, NULL
);
2913 error
= spa_vdev_remove(spa
, guid
, B_FALSE
);
2914 if (error
!= 0 && error
!= EBUSY
)
2915 fatal(0, "spa_vdev_remove(%llu) = %d", guid
, error
);
2918 mutex_exit(&ztest_vdev_lock
);
2920 umem_free(path
, MAXPATHLEN
);
2924 * split a pool if it has mirror tlvdevs
2928 ztest_split_pool(ztest_ds_t
*zd
, uint64_t id
)
2930 ztest_shared_t
*zs
= ztest_shared
;
2931 spa_t
*spa
= ztest_spa
;
2932 vdev_t
*rvd
= spa
->spa_root_vdev
;
2933 nvlist_t
*tree
, **child
, *config
, *split
, **schild
;
2934 uint_t c
, children
, schildren
= 0, lastlogid
= 0;
2937 mutex_enter(&ztest_vdev_lock
);
2939 /* ensure we have a useable config; mirrors of raidz aren't supported */
2940 if (zs
->zs_mirrors
< 3 || ztest_opts
.zo_raidz
> 1) {
2941 mutex_exit(&ztest_vdev_lock
);
2945 /* clean up the old pool, if any */
2946 (void) spa_destroy("splitp");
2948 spa_config_enter(spa
, SCL_VDEV
, FTAG
, RW_READER
);
2950 /* generate a config from the existing config */
2951 mutex_enter(&spa
->spa_props_lock
);
2952 VERIFY(nvlist_lookup_nvlist(spa
->spa_config
, ZPOOL_CONFIG_VDEV_TREE
,
2954 mutex_exit(&spa
->spa_props_lock
);
2956 VERIFY(nvlist_lookup_nvlist_array(tree
, ZPOOL_CONFIG_CHILDREN
, &child
,
2959 schild
= malloc(rvd
->vdev_children
* sizeof (nvlist_t
*));
2960 for (c
= 0; c
< children
; c
++) {
2961 vdev_t
*tvd
= rvd
->vdev_child
[c
];
2965 if (tvd
->vdev_islog
|| tvd
->vdev_ops
== &vdev_hole_ops
) {
2966 VERIFY(nvlist_alloc(&schild
[schildren
], NV_UNIQUE_NAME
,
2968 VERIFY(nvlist_add_string(schild
[schildren
],
2969 ZPOOL_CONFIG_TYPE
, VDEV_TYPE_HOLE
) == 0);
2970 VERIFY(nvlist_add_uint64(schild
[schildren
],
2971 ZPOOL_CONFIG_IS_HOLE
, 1) == 0);
2973 lastlogid
= schildren
;
2978 VERIFY(nvlist_lookup_nvlist_array(child
[c
],
2979 ZPOOL_CONFIG_CHILDREN
, &mchild
, &mchildren
) == 0);
2980 VERIFY(nvlist_dup(mchild
[0], &schild
[schildren
++], 0) == 0);
2983 /* OK, create a config that can be used to split */
2984 VERIFY(nvlist_alloc(&split
, NV_UNIQUE_NAME
, 0) == 0);
2985 VERIFY(nvlist_add_string(split
, ZPOOL_CONFIG_TYPE
,
2986 VDEV_TYPE_ROOT
) == 0);
2987 VERIFY(nvlist_add_nvlist_array(split
, ZPOOL_CONFIG_CHILDREN
, schild
,
2988 lastlogid
!= 0 ? lastlogid
: schildren
) == 0);
2990 VERIFY(nvlist_alloc(&config
, NV_UNIQUE_NAME
, 0) == 0);
2991 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, split
) == 0);
2993 for (c
= 0; c
< schildren
; c
++)
2994 nvlist_free(schild
[c
]);
2998 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
3000 (void) rw_wrlock(&ztest_name_lock
);
3001 error
= spa_vdev_split_mirror(spa
, "splitp", config
, NULL
, B_FALSE
);
3002 (void) rw_unlock(&ztest_name_lock
);
3004 nvlist_free(config
);
3007 (void) printf("successful split - results:\n");
3008 mutex_enter(&spa_namespace_lock
);
3009 show_pool_stats(spa
);
3010 show_pool_stats(spa_lookup("splitp"));
3011 mutex_exit(&spa_namespace_lock
);
3015 mutex_exit(&ztest_vdev_lock
);
3020 * Verify that we can attach and detach devices.
3024 ztest_vdev_attach_detach(ztest_ds_t
*zd
, uint64_t id
)
3026 ztest_shared_t
*zs
= ztest_shared
;
3027 spa_t
*spa
= ztest_spa
;
3028 spa_aux_vdev_t
*sav
= &spa
->spa_spares
;
3029 vdev_t
*rvd
= spa
->spa_root_vdev
;
3030 vdev_t
*oldvd
, *newvd
, *pvd
;
3034 uint64_t ashift
= ztest_get_ashift();
3035 uint64_t oldguid
, pguid
;
3036 uint64_t oldsize
, newsize
;
3037 char *oldpath
, *newpath
;
3039 int oldvd_has_siblings
= B_FALSE
;
3040 int newvd_is_spare
= B_FALSE
;
3042 int error
, expected_error
;
3044 oldpath
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
3045 newpath
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
3047 mutex_enter(&ztest_vdev_lock
);
3048 leaves
= MAX(zs
->zs_mirrors
, 1) * ztest_opts
.zo_raidz
;
3050 spa_config_enter(spa
, SCL_VDEV
, FTAG
, RW_READER
);
3053 * Decide whether to do an attach or a replace.
3055 replacing
= ztest_random(2);
3058 * Pick a random top-level vdev.
3060 top
= ztest_random_vdev_top(spa
, B_TRUE
);
3063 * Pick a random leaf within it.
3065 leaf
= ztest_random(leaves
);
3070 oldvd
= rvd
->vdev_child
[top
];
3071 if (zs
->zs_mirrors
>= 1) {
3072 ASSERT(oldvd
->vdev_ops
== &vdev_mirror_ops
);
3073 ASSERT(oldvd
->vdev_children
>= zs
->zs_mirrors
);
3074 oldvd
= oldvd
->vdev_child
[leaf
/ ztest_opts
.zo_raidz
];
3076 if (ztest_opts
.zo_raidz
> 1) {
3077 ASSERT(oldvd
->vdev_ops
== &vdev_raidz_ops
);
3078 ASSERT(oldvd
->vdev_children
== ztest_opts
.zo_raidz
);
3079 oldvd
= oldvd
->vdev_child
[leaf
% ztest_opts
.zo_raidz
];
3083 * If we're already doing an attach or replace, oldvd may be a
3084 * mirror vdev -- in which case, pick a random child.
3086 while (oldvd
->vdev_children
!= 0) {
3087 oldvd_has_siblings
= B_TRUE
;
3088 ASSERT(oldvd
->vdev_children
>= 2);
3089 oldvd
= oldvd
->vdev_child
[ztest_random(oldvd
->vdev_children
)];
3092 oldguid
= oldvd
->vdev_guid
;
3093 oldsize
= vdev_get_min_asize(oldvd
);
3094 oldvd_is_log
= oldvd
->vdev_top
->vdev_islog
;
3095 (void) strcpy(oldpath
, oldvd
->vdev_path
);
3096 pvd
= oldvd
->vdev_parent
;
3097 pguid
= pvd
->vdev_guid
;
3100 * If oldvd has siblings, then half of the time, detach it.
3102 if (oldvd_has_siblings
&& ztest_random(2) == 0) {
3103 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
3104 error
= spa_vdev_detach(spa
, oldguid
, pguid
, B_FALSE
);
3105 if (error
!= 0 && error
!= ENODEV
&& error
!= EBUSY
&&
3107 fatal(0, "detach (%s) returned %d", oldpath
, error
);
3112 * For the new vdev, choose with equal probability between the two
3113 * standard paths (ending in either 'a' or 'b') or a random hot spare.
3115 if (sav
->sav_count
!= 0 && ztest_random(3) == 0) {
3116 newvd
= sav
->sav_vdevs
[ztest_random(sav
->sav_count
)];
3117 newvd_is_spare
= B_TRUE
;
3118 (void) strcpy(newpath
, newvd
->vdev_path
);
3120 (void) snprintf(newpath
, MAXPATHLEN
, ztest_dev_template
,
3121 ztest_opts
.zo_dir
, ztest_opts
.zo_pool
,
3122 top
* leaves
+ leaf
);
3123 if (ztest_random(2) == 0)
3124 newpath
[strlen(newpath
) - 1] = 'b';
3125 newvd
= vdev_lookup_by_path(rvd
, newpath
);
3129 newsize
= vdev_get_min_asize(newvd
);
3132 * Make newsize a little bigger or smaller than oldsize.
3133 * If it's smaller, the attach should fail.
3134 * If it's larger, and we're doing a replace,
3135 * we should get dynamic LUN growth when we're done.
3137 newsize
= 10 * oldsize
/ (9 + ztest_random(3));
3141 * If pvd is not a mirror or root, the attach should fail with ENOTSUP,
3142 * unless it's a replace; in that case any non-replacing parent is OK.
3144 * If newvd is already part of the pool, it should fail with EBUSY.
3146 * If newvd is too small, it should fail with EOVERFLOW.
3148 if (pvd
->vdev_ops
!= &vdev_mirror_ops
&&
3149 pvd
->vdev_ops
!= &vdev_root_ops
&& (!replacing
||
3150 pvd
->vdev_ops
== &vdev_replacing_ops
||
3151 pvd
->vdev_ops
== &vdev_spare_ops
))
3152 expected_error
= ENOTSUP
;
3153 else if (newvd_is_spare
&& (!replacing
|| oldvd_is_log
))
3154 expected_error
= ENOTSUP
;
3155 else if (newvd
== oldvd
)
3156 expected_error
= replacing
? 0 : EBUSY
;
3157 else if (vdev_lookup_by_path(rvd
, newpath
) != NULL
)
3158 expected_error
= EBUSY
;
3159 else if (newsize
< oldsize
)
3160 expected_error
= EOVERFLOW
;
3161 else if (ashift
> oldvd
->vdev_top
->vdev_ashift
)
3162 expected_error
= EDOM
;
3166 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
3169 * Build the nvlist describing newpath.
3171 root
= make_vdev_root(newpath
, NULL
, NULL
, newvd
== NULL
? newsize
: 0,
3172 ashift
, 0, 0, 0, 1);
3174 error
= spa_vdev_attach(spa
, oldguid
, root
, replacing
);
3179 * If our parent was the replacing vdev, but the replace completed,
3180 * then instead of failing with ENOTSUP we may either succeed,
3181 * fail with ENODEV, or fail with EOVERFLOW.
3183 if (expected_error
== ENOTSUP
&&
3184 (error
== 0 || error
== ENODEV
|| error
== EOVERFLOW
))
3185 expected_error
= error
;
3188 * If someone grew the LUN, the replacement may be too small.
3190 if (error
== EOVERFLOW
|| error
== EBUSY
)
3191 expected_error
= error
;
3193 /* XXX workaround 6690467 */
3194 if (error
!= expected_error
&& expected_error
!= EBUSY
) {
3195 fatal(0, "attach (%s %llu, %s %llu, %d) "
3196 "returned %d, expected %d",
3197 oldpath
, oldsize
, newpath
,
3198 newsize
, replacing
, error
, expected_error
);
3201 mutex_exit(&ztest_vdev_lock
);
3203 umem_free(oldpath
, MAXPATHLEN
);
3204 umem_free(newpath
, MAXPATHLEN
);
3208 * Callback function which expands the physical size of the vdev.
3211 grow_vdev(vdev_t
*vd
, void *arg
)
3213 ASSERTV(spa_t
*spa
= vd
->vdev_spa
);
3214 size_t *newsize
= arg
;
3218 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_READER
) == SCL_STATE
);
3219 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
3221 if ((fd
= open(vd
->vdev_path
, O_RDWR
)) == -1)
3224 fsize
= lseek(fd
, 0, SEEK_END
);
3225 VERIFY(ftruncate(fd
, *newsize
) == 0);
3227 if (ztest_opts
.zo_verbose
>= 6) {
3228 (void) printf("%s grew from %lu to %lu bytes\n",
3229 vd
->vdev_path
, (ulong_t
)fsize
, (ulong_t
)*newsize
);
3236 * Callback function which expands a given vdev by calling vdev_online().
3240 online_vdev(vdev_t
*vd
, void *arg
)
3242 spa_t
*spa
= vd
->vdev_spa
;
3243 vdev_t
*tvd
= vd
->vdev_top
;
3244 uint64_t guid
= vd
->vdev_guid
;
3245 uint64_t generation
= spa
->spa_config_generation
+ 1;
3246 vdev_state_t newstate
= VDEV_STATE_UNKNOWN
;
3249 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_READER
) == SCL_STATE
);
3250 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
3252 /* Calling vdev_online will initialize the new metaslabs */
3253 spa_config_exit(spa
, SCL_STATE
, spa
);
3254 error
= vdev_online(spa
, guid
, ZFS_ONLINE_EXPAND
, &newstate
);
3255 spa_config_enter(spa
, SCL_STATE
, spa
, RW_READER
);
3258 * If vdev_online returned an error or the underlying vdev_open
3259 * failed then we abort the expand. The only way to know that
3260 * vdev_open fails is by checking the returned newstate.
3262 if (error
|| newstate
!= VDEV_STATE_HEALTHY
) {
3263 if (ztest_opts
.zo_verbose
>= 5) {
3264 (void) printf("Unable to expand vdev, state %llu, "
3265 "error %d\n", (u_longlong_t
)newstate
, error
);
3269 ASSERT3U(newstate
, ==, VDEV_STATE_HEALTHY
);
3272 * Since we dropped the lock we need to ensure that we're
3273 * still talking to the original vdev. It's possible this
3274 * vdev may have been detached/replaced while we were
3275 * trying to online it.
3277 if (generation
!= spa
->spa_config_generation
) {
3278 if (ztest_opts
.zo_verbose
>= 5) {
3279 (void) printf("vdev configuration has changed, "
3280 "guid %llu, state %llu, expected gen %llu, "
3283 (u_longlong_t
)tvd
->vdev_state
,
3284 (u_longlong_t
)generation
,
3285 (u_longlong_t
)spa
->spa_config_generation
);
3293 * Traverse the vdev tree calling the supplied function.
3294 * We continue to walk the tree until we either have walked all
3295 * children or we receive a non-NULL return from the callback.
3296 * If a NULL callback is passed, then we just return back the first
3297 * leaf vdev we encounter.
3300 vdev_walk_tree(vdev_t
*vd
, vdev_t
*(*func
)(vdev_t
*, void *), void *arg
)
3304 if (vd
->vdev_ops
->vdev_op_leaf
) {
3308 return (func(vd
, arg
));
3311 for (c
= 0; c
< vd
->vdev_children
; c
++) {
3312 vdev_t
*cvd
= vd
->vdev_child
[c
];
3313 if ((cvd
= vdev_walk_tree(cvd
, func
, arg
)) != NULL
)
3320 * Verify that dynamic LUN growth works as expected.
3324 ztest_vdev_LUN_growth(ztest_ds_t
*zd
, uint64_t id
)
3326 spa_t
*spa
= ztest_spa
;
3328 metaslab_class_t
*mc
;
3329 metaslab_group_t
*mg
;
3330 size_t psize
, newsize
;
3332 uint64_t old_class_space
, new_class_space
, old_ms_count
, new_ms_count
;
3334 mutex_enter(&ztest_vdev_lock
);
3335 spa_config_enter(spa
, SCL_STATE
, spa
, RW_READER
);
3337 top
= ztest_random_vdev_top(spa
, B_TRUE
);
3339 tvd
= spa
->spa_root_vdev
->vdev_child
[top
];
3342 old_ms_count
= tvd
->vdev_ms_count
;
3343 old_class_space
= metaslab_class_get_space(mc
);
3346 * Determine the size of the first leaf vdev associated with
3347 * our top-level device.
3349 vd
= vdev_walk_tree(tvd
, NULL
, NULL
);
3350 ASSERT3P(vd
, !=, NULL
);
3351 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
3353 psize
= vd
->vdev_psize
;
3356 * We only try to expand the vdev if it's healthy, less than 4x its
3357 * original size, and it has a valid psize.
3359 if (tvd
->vdev_state
!= VDEV_STATE_HEALTHY
||
3360 psize
== 0 || psize
>= 4 * ztest_opts
.zo_vdev_size
) {
3361 spa_config_exit(spa
, SCL_STATE
, spa
);
3362 mutex_exit(&ztest_vdev_lock
);
3366 newsize
= psize
+ psize
/ 8;
3367 ASSERT3U(newsize
, >, psize
);
3369 if (ztest_opts
.zo_verbose
>= 6) {
3370 (void) printf("Expanding LUN %s from %lu to %lu\n",
3371 vd
->vdev_path
, (ulong_t
)psize
, (ulong_t
)newsize
);
3375 * Growing the vdev is a two step process:
3376 * 1). expand the physical size (i.e. relabel)
3377 * 2). online the vdev to create the new metaslabs
3379 if (vdev_walk_tree(tvd
, grow_vdev
, &newsize
) != NULL
||
3380 vdev_walk_tree(tvd
, online_vdev
, NULL
) != NULL
||
3381 tvd
->vdev_state
!= VDEV_STATE_HEALTHY
) {
3382 if (ztest_opts
.zo_verbose
>= 5) {
3383 (void) printf("Could not expand LUN because "
3384 "the vdev configuration changed.\n");
3386 spa_config_exit(spa
, SCL_STATE
, spa
);
3387 mutex_exit(&ztest_vdev_lock
);
3391 spa_config_exit(spa
, SCL_STATE
, spa
);
3394 * Expanding the LUN will update the config asynchronously,
3395 * thus we must wait for the async thread to complete any
3396 * pending tasks before proceeding.
3400 mutex_enter(&spa
->spa_async_lock
);
3401 done
= (spa
->spa_async_thread
== NULL
&& !spa
->spa_async_tasks
);
3402 mutex_exit(&spa
->spa_async_lock
);
3405 txg_wait_synced(spa_get_dsl(spa
), 0);
3406 (void) poll(NULL
, 0, 100);
3409 spa_config_enter(spa
, SCL_STATE
, spa
, RW_READER
);
3411 tvd
= spa
->spa_root_vdev
->vdev_child
[top
];
3412 new_ms_count
= tvd
->vdev_ms_count
;
3413 new_class_space
= metaslab_class_get_space(mc
);
3415 if (tvd
->vdev_mg
!= mg
|| mg
->mg_class
!= mc
) {
3416 if (ztest_opts
.zo_verbose
>= 5) {
3417 (void) printf("Could not verify LUN expansion due to "
3418 "intervening vdev offline or remove.\n");
3420 spa_config_exit(spa
, SCL_STATE
, spa
);
3421 mutex_exit(&ztest_vdev_lock
);
3426 * Make sure we were able to grow the vdev.
3428 if (new_ms_count
<= old_ms_count
)
3429 fatal(0, "LUN expansion failed: ms_count %llu <= %llu\n",
3430 old_ms_count
, new_ms_count
);
3433 * Make sure we were able to grow the pool.
3435 if (new_class_space
<= old_class_space
)
3436 fatal(0, "LUN expansion failed: class_space %llu <= %llu\n",
3437 old_class_space
, new_class_space
);
3439 if (ztest_opts
.zo_verbose
>= 5) {
3440 char oldnumbuf
[6], newnumbuf
[6];
3442 nicenum(old_class_space
, oldnumbuf
);
3443 nicenum(new_class_space
, newnumbuf
);
3444 (void) printf("%s grew from %s to %s\n",
3445 spa
->spa_name
, oldnumbuf
, newnumbuf
);
3448 spa_config_exit(spa
, SCL_STATE
, spa
);
3449 mutex_exit(&ztest_vdev_lock
);
3453 * Verify that dmu_objset_{create,destroy,open,close} work as expected.
3457 ztest_objset_create_cb(objset_t
*os
, void *arg
, cred_t
*cr
, dmu_tx_t
*tx
)
3460 * Create the objects common to all ztest datasets.
3462 VERIFY(zap_create_claim(os
, ZTEST_DIROBJ
,
3463 DMU_OT_ZAP_OTHER
, DMU_OT_NONE
, 0, tx
) == 0);
3467 ztest_dataset_create(char *dsname
)
3469 uint64_t zilset
= ztest_random(100);
3470 int err
= dmu_objset_create(dsname
, DMU_OST_OTHER
, 0,
3471 ztest_objset_create_cb
, NULL
);
3473 if (err
|| zilset
< 80)
3476 if (ztest_opts
.zo_verbose
>= 5)
3477 (void) printf("Setting dataset %s to sync always\n", dsname
);
3478 return (ztest_dsl_prop_set_uint64(dsname
, ZFS_PROP_SYNC
,
3479 ZFS_SYNC_ALWAYS
, B_FALSE
));
3484 ztest_objset_destroy_cb(const char *name
, void *arg
)
3487 dmu_object_info_t doi
;
3491 * Verify that the dataset contains a directory object.
3493 VERIFY0(dmu_objset_own(name
, DMU_OST_OTHER
, B_TRUE
, FTAG
, &os
));
3494 error
= dmu_object_info(os
, ZTEST_DIROBJ
, &doi
);
3495 if (error
!= ENOENT
) {
3496 /* We could have crashed in the middle of destroying it */
3498 ASSERT3U(doi
.doi_type
, ==, DMU_OT_ZAP_OTHER
);
3499 ASSERT3S(doi
.doi_physical_blocks_512
, >=, 0);
3501 dmu_objset_disown(os
, FTAG
);
3504 * Destroy the dataset.
3506 if (strchr(name
, '@') != NULL
) {
3507 VERIFY0(dsl_destroy_snapshot(name
, B_TRUE
));
3509 error
= dsl_destroy_head(name
);
3510 /* There could be a hold on this dataset */
3518 ztest_snapshot_create(char *osname
, uint64_t id
)
3520 char snapname
[ZFS_MAX_DATASET_NAME_LEN
];
3523 (void) snprintf(snapname
, sizeof (snapname
), "%llu", (u_longlong_t
)id
);
3525 error
= dmu_objset_snapshot_one(osname
, snapname
);
3526 if (error
== ENOSPC
) {
3527 ztest_record_enospc(FTAG
);
3530 if (error
!= 0 && error
!= EEXIST
) {
3531 fatal(0, "ztest_snapshot_create(%s@%s) = %d", osname
,
3538 ztest_snapshot_destroy(char *osname
, uint64_t id
)
3540 char snapname
[ZFS_MAX_DATASET_NAME_LEN
];
3543 (void) snprintf(snapname
, sizeof (snapname
), "%s@%llu", osname
,
3546 error
= dsl_destroy_snapshot(snapname
, B_FALSE
);
3547 if (error
!= 0 && error
!= ENOENT
)
3548 fatal(0, "ztest_snapshot_destroy(%s) = %d", snapname
, error
);
3554 ztest_dmu_objset_create_destroy(ztest_ds_t
*zd
, uint64_t id
)
3560 char name
[ZFS_MAX_DATASET_NAME_LEN
];
3564 zdtmp
= umem_alloc(sizeof (ztest_ds_t
), UMEM_NOFAIL
);
3566 (void) rw_rdlock(&ztest_name_lock
);
3568 (void) snprintf(name
, sizeof (name
), "%s/temp_%llu",
3569 ztest_opts
.zo_pool
, (u_longlong_t
)id
);
3572 * If this dataset exists from a previous run, process its replay log
3573 * half of the time. If we don't replay it, then dsl_destroy_head()
3574 * (invoked from ztest_objset_destroy_cb()) should just throw it away.
3576 if (ztest_random(2) == 0 &&
3577 dmu_objset_own(name
, DMU_OST_OTHER
, B_FALSE
, FTAG
, &os
) == 0) {
3578 ztest_zd_init(zdtmp
, NULL
, os
);
3579 zil_replay(os
, zdtmp
, ztest_replay_vector
);
3580 ztest_zd_fini(zdtmp
);
3581 dmu_objset_disown(os
, FTAG
);
3585 * There may be an old instance of the dataset we're about to
3586 * create lying around from a previous run. If so, destroy it
3587 * and all of its snapshots.
3589 (void) dmu_objset_find(name
, ztest_objset_destroy_cb
, NULL
,
3590 DS_FIND_CHILDREN
| DS_FIND_SNAPSHOTS
);
3593 * Verify that the destroyed dataset is no longer in the namespace.
3595 VERIFY3U(ENOENT
, ==, dmu_objset_own(name
, DMU_OST_OTHER
, B_TRUE
,
3599 * Verify that we can create a new dataset.
3601 error
= ztest_dataset_create(name
);
3603 if (error
== ENOSPC
) {
3604 ztest_record_enospc(FTAG
);
3607 fatal(0, "dmu_objset_create(%s) = %d", name
, error
);
3610 VERIFY0(dmu_objset_own(name
, DMU_OST_OTHER
, B_FALSE
, FTAG
, &os
));
3612 ztest_zd_init(zdtmp
, NULL
, os
);
3615 * Open the intent log for it.
3617 zilog
= zil_open(os
, ztest_get_data
);
3620 * Put some objects in there, do a little I/O to them,
3621 * and randomly take a couple of snapshots along the way.
3623 iters
= ztest_random(5);
3624 for (i
= 0; i
< iters
; i
++) {
3625 ztest_dmu_object_alloc_free(zdtmp
, id
);
3626 if (ztest_random(iters
) == 0)
3627 (void) ztest_snapshot_create(name
, i
);
3631 * Verify that we cannot create an existing dataset.
3633 VERIFY3U(EEXIST
, ==,
3634 dmu_objset_create(name
, DMU_OST_OTHER
, 0, NULL
, NULL
));
3637 * Verify that we can hold an objset that is also owned.
3639 VERIFY3U(0, ==, dmu_objset_hold(name
, FTAG
, &os2
));
3640 dmu_objset_rele(os2
, FTAG
);
3643 * Verify that we cannot own an objset that is already owned.
3646 dmu_objset_own(name
, DMU_OST_OTHER
, B_FALSE
, FTAG
, &os2
));
3649 dmu_objset_disown(os
, FTAG
);
3650 ztest_zd_fini(zdtmp
);
3652 (void) rw_unlock(&ztest_name_lock
);
3654 umem_free(zdtmp
, sizeof (ztest_ds_t
));
3658 * Verify that dmu_snapshot_{create,destroy,open,close} work as expected.
3661 ztest_dmu_snapshot_create_destroy(ztest_ds_t
*zd
, uint64_t id
)
3663 (void) rw_rdlock(&ztest_name_lock
);
3664 (void) ztest_snapshot_destroy(zd
->zd_name
, id
);
3665 (void) ztest_snapshot_create(zd
->zd_name
, id
);
3666 (void) rw_unlock(&ztest_name_lock
);
3670 * Cleanup non-standard snapshots and clones.
3673 ztest_dsl_dataset_cleanup(char *osname
, uint64_t id
)
3682 snap1name
= umem_alloc(ZFS_MAX_DATASET_NAME_LEN
, UMEM_NOFAIL
);
3683 clone1name
= umem_alloc(ZFS_MAX_DATASET_NAME_LEN
, UMEM_NOFAIL
);
3684 snap2name
= umem_alloc(ZFS_MAX_DATASET_NAME_LEN
, UMEM_NOFAIL
);
3685 clone2name
= umem_alloc(ZFS_MAX_DATASET_NAME_LEN
, UMEM_NOFAIL
);
3686 snap3name
= umem_alloc(ZFS_MAX_DATASET_NAME_LEN
, UMEM_NOFAIL
);
3688 (void) snprintf(snap1name
, ZFS_MAX_DATASET_NAME_LEN
,
3689 "%s@s1_%llu", osname
, (u_longlong_t
)id
);
3690 (void) snprintf(clone1name
, ZFS_MAX_DATASET_NAME_LEN
,
3691 "%s/c1_%llu", osname
, (u_longlong_t
)id
);
3692 (void) snprintf(snap2name
, ZFS_MAX_DATASET_NAME_LEN
,
3693 "%s@s2_%llu", clone1name
, (u_longlong_t
)id
);
3694 (void) snprintf(clone2name
, ZFS_MAX_DATASET_NAME_LEN
,
3695 "%s/c2_%llu", osname
, (u_longlong_t
)id
);
3696 (void) snprintf(snap3name
, ZFS_MAX_DATASET_NAME_LEN
,
3697 "%s@s3_%llu", clone1name
, (u_longlong_t
)id
);
3699 error
= dsl_destroy_head(clone2name
);
3700 if (error
&& error
!= ENOENT
)
3701 fatal(0, "dsl_destroy_head(%s) = %d", clone2name
, error
);
3702 error
= dsl_destroy_snapshot(snap3name
, B_FALSE
);
3703 if (error
&& error
!= ENOENT
)
3704 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap3name
, error
);
3705 error
= dsl_destroy_snapshot(snap2name
, B_FALSE
);
3706 if (error
&& error
!= ENOENT
)
3707 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap2name
, error
);
3708 error
= dsl_destroy_head(clone1name
);
3709 if (error
&& error
!= ENOENT
)
3710 fatal(0, "dsl_destroy_head(%s) = %d", clone1name
, error
);
3711 error
= dsl_destroy_snapshot(snap1name
, B_FALSE
);
3712 if (error
&& error
!= ENOENT
)
3713 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap1name
, error
);
3715 umem_free(snap1name
, ZFS_MAX_DATASET_NAME_LEN
);
3716 umem_free(clone1name
, ZFS_MAX_DATASET_NAME_LEN
);
3717 umem_free(snap2name
, ZFS_MAX_DATASET_NAME_LEN
);
3718 umem_free(clone2name
, ZFS_MAX_DATASET_NAME_LEN
);
3719 umem_free(snap3name
, ZFS_MAX_DATASET_NAME_LEN
);
3723 * Verify dsl_dataset_promote handles EBUSY
3726 ztest_dsl_dataset_promote_busy(ztest_ds_t
*zd
, uint64_t id
)
3734 char *osname
= zd
->zd_name
;
3737 snap1name
= umem_alloc(ZFS_MAX_DATASET_NAME_LEN
, UMEM_NOFAIL
);
3738 clone1name
= umem_alloc(ZFS_MAX_DATASET_NAME_LEN
, UMEM_NOFAIL
);
3739 snap2name
= umem_alloc(ZFS_MAX_DATASET_NAME_LEN
, UMEM_NOFAIL
);
3740 clone2name
= umem_alloc(ZFS_MAX_DATASET_NAME_LEN
, UMEM_NOFAIL
);
3741 snap3name
= umem_alloc(ZFS_MAX_DATASET_NAME_LEN
, UMEM_NOFAIL
);
3743 (void) rw_rdlock(&ztest_name_lock
);
3745 ztest_dsl_dataset_cleanup(osname
, id
);
3747 (void) snprintf(snap1name
, ZFS_MAX_DATASET_NAME_LEN
,
3748 "%s@s1_%llu", osname
, (u_longlong_t
)id
);
3749 (void) snprintf(clone1name
, ZFS_MAX_DATASET_NAME_LEN
,
3750 "%s/c1_%llu", osname
, (u_longlong_t
)id
);
3751 (void) snprintf(snap2name
, ZFS_MAX_DATASET_NAME_LEN
,
3752 "%s@s2_%llu", clone1name
, (u_longlong_t
)id
);
3753 (void) snprintf(clone2name
, ZFS_MAX_DATASET_NAME_LEN
,
3754 "%s/c2_%llu", osname
, (u_longlong_t
)id
);
3755 (void) snprintf(snap3name
, ZFS_MAX_DATASET_NAME_LEN
,
3756 "%s@s3_%llu", clone1name
, (u_longlong_t
)id
);
3758 error
= dmu_objset_snapshot_one(osname
, strchr(snap1name
, '@') + 1);
3759 if (error
&& error
!= EEXIST
) {
3760 if (error
== ENOSPC
) {
3761 ztest_record_enospc(FTAG
);
3764 fatal(0, "dmu_take_snapshot(%s) = %d", snap1name
, error
);
3767 error
= dmu_objset_clone(clone1name
, snap1name
);
3769 if (error
== ENOSPC
) {
3770 ztest_record_enospc(FTAG
);
3773 fatal(0, "dmu_objset_create(%s) = %d", clone1name
, error
);
3776 error
= dmu_objset_snapshot_one(clone1name
, strchr(snap2name
, '@') + 1);
3777 if (error
&& error
!= EEXIST
) {
3778 if (error
== ENOSPC
) {
3779 ztest_record_enospc(FTAG
);
3782 fatal(0, "dmu_open_snapshot(%s) = %d", snap2name
, error
);
3785 error
= dmu_objset_snapshot_one(clone1name
, strchr(snap3name
, '@') + 1);
3786 if (error
&& error
!= EEXIST
) {
3787 if (error
== ENOSPC
) {
3788 ztest_record_enospc(FTAG
);
3791 fatal(0, "dmu_open_snapshot(%s) = %d", snap3name
, error
);
3794 error
= dmu_objset_clone(clone2name
, snap3name
);
3796 if (error
== ENOSPC
) {
3797 ztest_record_enospc(FTAG
);
3800 fatal(0, "dmu_objset_create(%s) = %d", clone2name
, error
);
3803 error
= dmu_objset_own(snap2name
, DMU_OST_ANY
, B_TRUE
, FTAG
, &os
);
3805 fatal(0, "dmu_objset_own(%s) = %d", snap2name
, error
);
3806 error
= dsl_dataset_promote(clone2name
, NULL
);
3807 if (error
== ENOSPC
) {
3808 dmu_objset_disown(os
, FTAG
);
3809 ztest_record_enospc(FTAG
);
3813 fatal(0, "dsl_dataset_promote(%s), %d, not EBUSY", clone2name
,
3815 dmu_objset_disown(os
, FTAG
);
3818 ztest_dsl_dataset_cleanup(osname
, id
);
3820 (void) rw_unlock(&ztest_name_lock
);
3822 umem_free(snap1name
, ZFS_MAX_DATASET_NAME_LEN
);
3823 umem_free(clone1name
, ZFS_MAX_DATASET_NAME_LEN
);
3824 umem_free(snap2name
, ZFS_MAX_DATASET_NAME_LEN
);
3825 umem_free(clone2name
, ZFS_MAX_DATASET_NAME_LEN
);
3826 umem_free(snap3name
, ZFS_MAX_DATASET_NAME_LEN
);
3829 #undef OD_ARRAY_SIZE
3830 #define OD_ARRAY_SIZE 4
3833 * Verify that dmu_object_{alloc,free} work as expected.
3836 ztest_dmu_object_alloc_free(ztest_ds_t
*zd
, uint64_t id
)
3843 size
= sizeof (ztest_od_t
) * OD_ARRAY_SIZE
;
3844 od
= umem_alloc(size
, UMEM_NOFAIL
);
3845 batchsize
= OD_ARRAY_SIZE
;
3847 for (b
= 0; b
< batchsize
; b
++)
3848 ztest_od_init(od
+ b
, id
, FTAG
, b
, DMU_OT_UINT64_OTHER
,
3852 * Destroy the previous batch of objects, create a new batch,
3853 * and do some I/O on the new objects.
3855 if (ztest_object_init(zd
, od
, size
, B_TRUE
) != 0)
3858 while (ztest_random(4 * batchsize
) != 0)
3859 ztest_io(zd
, od
[ztest_random(batchsize
)].od_object
,
3860 ztest_random(ZTEST_RANGE_LOCKS
) << SPA_MAXBLOCKSHIFT
);
3862 umem_free(od
, size
);
3865 #undef OD_ARRAY_SIZE
3866 #define OD_ARRAY_SIZE 2
3869 * Verify that dmu_{read,write} work as expected.
3872 ztest_dmu_read_write(ztest_ds_t
*zd
, uint64_t id
)
3877 objset_t
*os
= zd
->zd_os
;
3878 size
= sizeof (ztest_od_t
) * OD_ARRAY_SIZE
;
3879 od
= umem_alloc(size
, UMEM_NOFAIL
);
3881 int i
, freeit
, error
;
3883 bufwad_t
*packbuf
, *bigbuf
, *pack
, *bigH
, *bigT
;
3884 uint64_t packobj
, packoff
, packsize
, bigobj
, bigoff
, bigsize
;
3885 uint64_t chunksize
= (1000 + ztest_random(1000)) * sizeof (uint64_t);
3886 uint64_t regions
= 997;
3887 uint64_t stride
= 123456789ULL;
3888 uint64_t width
= 40;
3889 int free_percent
= 5;
3892 * This test uses two objects, packobj and bigobj, that are always
3893 * updated together (i.e. in the same tx) so that their contents are
3894 * in sync and can be compared. Their contents relate to each other
3895 * in a simple way: packobj is a dense array of 'bufwad' structures,
3896 * while bigobj is a sparse array of the same bufwads. Specifically,
3897 * for any index n, there are three bufwads that should be identical:
3899 * packobj, at offset n * sizeof (bufwad_t)
3900 * bigobj, at the head of the nth chunk
3901 * bigobj, at the tail of the nth chunk
3903 * The chunk size is arbitrary. It doesn't have to be a power of two,
3904 * and it doesn't have any relation to the object blocksize.
3905 * The only requirement is that it can hold at least two bufwads.
3907 * Normally, we write the bufwad to each of these locations.
3908 * However, free_percent of the time we instead write zeroes to
3909 * packobj and perform a dmu_free_range() on bigobj. By comparing
3910 * bigobj to packobj, we can verify that the DMU is correctly
3911 * tracking which parts of an object are allocated and free,
3912 * and that the contents of the allocated blocks are correct.
3916 * Read the directory info. If it's the first time, set things up.
3918 ztest_od_init(od
, id
, FTAG
, 0, DMU_OT_UINT64_OTHER
, 0, 0, chunksize
);
3919 ztest_od_init(od
+ 1, id
, FTAG
, 1, DMU_OT_UINT64_OTHER
, 0, 0,
3922 if (ztest_object_init(zd
, od
, size
, B_FALSE
) != 0) {
3923 umem_free(od
, size
);
3927 bigobj
= od
[0].od_object
;
3928 packobj
= od
[1].od_object
;
3929 chunksize
= od
[0].od_gen
;
3930 ASSERT(chunksize
== od
[1].od_gen
);
3933 * Prefetch a random chunk of the big object.
3934 * Our aim here is to get some async reads in flight
3935 * for blocks that we may free below; the DMU should
3936 * handle this race correctly.
3938 n
= ztest_random(regions
) * stride
+ ztest_random(width
);
3939 s
= 1 + ztest_random(2 * width
- 1);
3940 dmu_prefetch(os
, bigobj
, 0, n
* chunksize
, s
* chunksize
,
3941 ZIO_PRIORITY_SYNC_READ
);
3944 * Pick a random index and compute the offsets into packobj and bigobj.
3946 n
= ztest_random(regions
) * stride
+ ztest_random(width
);
3947 s
= 1 + ztest_random(width
- 1);
3949 packoff
= n
* sizeof (bufwad_t
);
3950 packsize
= s
* sizeof (bufwad_t
);
3952 bigoff
= n
* chunksize
;
3953 bigsize
= s
* chunksize
;
3955 packbuf
= umem_alloc(packsize
, UMEM_NOFAIL
);
3956 bigbuf
= umem_alloc(bigsize
, UMEM_NOFAIL
);
3959 * free_percent of the time, free a range of bigobj rather than
3962 freeit
= (ztest_random(100) < free_percent
);
3965 * Read the current contents of our objects.
3967 error
= dmu_read(os
, packobj
, packoff
, packsize
, packbuf
,
3970 error
= dmu_read(os
, bigobj
, bigoff
, bigsize
, bigbuf
,
3975 * Get a tx for the mods to both packobj and bigobj.
3977 tx
= dmu_tx_create(os
);
3979 dmu_tx_hold_write(tx
, packobj
, packoff
, packsize
);
3982 dmu_tx_hold_free(tx
, bigobj
, bigoff
, bigsize
);
3984 dmu_tx_hold_write(tx
, bigobj
, bigoff
, bigsize
);
3986 /* This accounts for setting the checksum/compression. */
3987 dmu_tx_hold_bonus(tx
, bigobj
);
3989 txg
= ztest_tx_assign(tx
, TXG_MIGHTWAIT
, FTAG
);
3991 umem_free(packbuf
, packsize
);
3992 umem_free(bigbuf
, bigsize
);
3993 umem_free(od
, size
);
3997 enum zio_checksum cksum
;
3999 cksum
= (enum zio_checksum
)
4000 ztest_random_dsl_prop(ZFS_PROP_CHECKSUM
);
4001 } while (cksum
>= ZIO_CHECKSUM_LEGACY_FUNCTIONS
);
4002 dmu_object_set_checksum(os
, bigobj
, cksum
, tx
);
4004 enum zio_compress comp
;
4006 comp
= (enum zio_compress
)
4007 ztest_random_dsl_prop(ZFS_PROP_COMPRESSION
);
4008 } while (comp
>= ZIO_COMPRESS_LEGACY_FUNCTIONS
);
4009 dmu_object_set_compress(os
, bigobj
, comp
, tx
);
4012 * For each index from n to n + s, verify that the existing bufwad
4013 * in packobj matches the bufwads at the head and tail of the
4014 * corresponding chunk in bigobj. Then update all three bufwads
4015 * with the new values we want to write out.
4017 for (i
= 0; i
< s
; i
++) {
4019 pack
= (bufwad_t
*)((char *)packbuf
+ i
* sizeof (bufwad_t
));
4021 bigH
= (bufwad_t
*)((char *)bigbuf
+ i
* chunksize
);
4023 bigT
= (bufwad_t
*)((char *)bigH
+ chunksize
) - 1;
4025 ASSERT((uintptr_t)bigH
- (uintptr_t)bigbuf
< bigsize
);
4026 ASSERT((uintptr_t)bigT
- (uintptr_t)bigbuf
< bigsize
);
4028 if (pack
->bw_txg
> txg
)
4029 fatal(0, "future leak: got %llx, open txg is %llx",
4032 if (pack
->bw_data
!= 0 && pack
->bw_index
!= n
+ i
)
4033 fatal(0, "wrong index: got %llx, wanted %llx+%llx",
4034 pack
->bw_index
, n
, i
);
4036 if (bcmp(pack
, bigH
, sizeof (bufwad_t
)) != 0)
4037 fatal(0, "pack/bigH mismatch in %p/%p", pack
, bigH
);
4039 if (bcmp(pack
, bigT
, sizeof (bufwad_t
)) != 0)
4040 fatal(0, "pack/bigT mismatch in %p/%p", pack
, bigT
);
4043 bzero(pack
, sizeof (bufwad_t
));
4045 pack
->bw_index
= n
+ i
;
4047 pack
->bw_data
= 1 + ztest_random(-2ULL);
4054 * We've verified all the old bufwads, and made new ones.
4055 * Now write them out.
4057 dmu_write(os
, packobj
, packoff
, packsize
, packbuf
, tx
);
4060 if (ztest_opts
.zo_verbose
>= 7) {
4061 (void) printf("freeing offset %llx size %llx"
4063 (u_longlong_t
)bigoff
,
4064 (u_longlong_t
)bigsize
,
4067 VERIFY(0 == dmu_free_range(os
, bigobj
, bigoff
, bigsize
, tx
));
4069 if (ztest_opts
.zo_verbose
>= 7) {
4070 (void) printf("writing offset %llx size %llx"
4072 (u_longlong_t
)bigoff
,
4073 (u_longlong_t
)bigsize
,
4076 dmu_write(os
, bigobj
, bigoff
, bigsize
, bigbuf
, tx
);
4082 * Sanity check the stuff we just wrote.
4085 void *packcheck
= umem_alloc(packsize
, UMEM_NOFAIL
);
4086 void *bigcheck
= umem_alloc(bigsize
, UMEM_NOFAIL
);
4088 VERIFY(0 == dmu_read(os
, packobj
, packoff
,
4089 packsize
, packcheck
, DMU_READ_PREFETCH
));
4090 VERIFY(0 == dmu_read(os
, bigobj
, bigoff
,
4091 bigsize
, bigcheck
, DMU_READ_PREFETCH
));
4093 ASSERT(bcmp(packbuf
, packcheck
, packsize
) == 0);
4094 ASSERT(bcmp(bigbuf
, bigcheck
, bigsize
) == 0);
4096 umem_free(packcheck
, packsize
);
4097 umem_free(bigcheck
, bigsize
);
4100 umem_free(packbuf
, packsize
);
4101 umem_free(bigbuf
, bigsize
);
4102 umem_free(od
, size
);
4106 compare_and_update_pbbufs(uint64_t s
, bufwad_t
*packbuf
, bufwad_t
*bigbuf
,
4107 uint64_t bigsize
, uint64_t n
, uint64_t chunksize
, uint64_t txg
)
4115 * For each index from n to n + s, verify that the existing bufwad
4116 * in packobj matches the bufwads at the head and tail of the
4117 * corresponding chunk in bigobj. Then update all three bufwads
4118 * with the new values we want to write out.
4120 for (i
= 0; i
< s
; i
++) {
4122 pack
= (bufwad_t
*)((char *)packbuf
+ i
* sizeof (bufwad_t
));
4124 bigH
= (bufwad_t
*)((char *)bigbuf
+ i
* chunksize
);
4126 bigT
= (bufwad_t
*)((char *)bigH
+ chunksize
) - 1;
4128 ASSERT((uintptr_t)bigH
- (uintptr_t)bigbuf
< bigsize
);
4129 ASSERT((uintptr_t)bigT
- (uintptr_t)bigbuf
< bigsize
);
4131 if (pack
->bw_txg
> txg
)
4132 fatal(0, "future leak: got %llx, open txg is %llx",
4135 if (pack
->bw_data
!= 0 && pack
->bw_index
!= n
+ i
)
4136 fatal(0, "wrong index: got %llx, wanted %llx+%llx",
4137 pack
->bw_index
, n
, i
);
4139 if (bcmp(pack
, bigH
, sizeof (bufwad_t
)) != 0)
4140 fatal(0, "pack/bigH mismatch in %p/%p", pack
, bigH
);
4142 if (bcmp(pack
, bigT
, sizeof (bufwad_t
)) != 0)
4143 fatal(0, "pack/bigT mismatch in %p/%p", pack
, bigT
);
4145 pack
->bw_index
= n
+ i
;
4147 pack
->bw_data
= 1 + ztest_random(-2ULL);
4154 #undef OD_ARRAY_SIZE
4155 #define OD_ARRAY_SIZE 2
4158 ztest_dmu_read_write_zcopy(ztest_ds_t
*zd
, uint64_t id
)
4160 objset_t
*os
= zd
->zd_os
;
4167 bufwad_t
*packbuf
, *bigbuf
;
4168 uint64_t packobj
, packoff
, packsize
, bigobj
, bigoff
, bigsize
;
4169 uint64_t blocksize
= ztest_random_blocksize();
4170 uint64_t chunksize
= blocksize
;
4171 uint64_t regions
= 997;
4172 uint64_t stride
= 123456789ULL;
4174 dmu_buf_t
*bonus_db
;
4175 arc_buf_t
**bigbuf_arcbufs
;
4176 dmu_object_info_t doi
;
4178 size
= sizeof (ztest_od_t
) * OD_ARRAY_SIZE
;
4179 od
= umem_alloc(size
, UMEM_NOFAIL
);
4182 * This test uses two objects, packobj and bigobj, that are always
4183 * updated together (i.e. in the same tx) so that their contents are
4184 * in sync and can be compared. Their contents relate to each other
4185 * in a simple way: packobj is a dense array of 'bufwad' structures,
4186 * while bigobj is a sparse array of the same bufwads. Specifically,
4187 * for any index n, there are three bufwads that should be identical:
4189 * packobj, at offset n * sizeof (bufwad_t)
4190 * bigobj, at the head of the nth chunk
4191 * bigobj, at the tail of the nth chunk
4193 * The chunk size is set equal to bigobj block size so that
4194 * dmu_assign_arcbuf() can be tested for object updates.
4198 * Read the directory info. If it's the first time, set things up.
4200 ztest_od_init(od
, id
, FTAG
, 0, DMU_OT_UINT64_OTHER
, blocksize
, 0, 0);
4201 ztest_od_init(od
+ 1, id
, FTAG
, 1, DMU_OT_UINT64_OTHER
, 0, 0,
4205 if (ztest_object_init(zd
, od
, size
, B_FALSE
) != 0) {
4206 umem_free(od
, size
);
4210 bigobj
= od
[0].od_object
;
4211 packobj
= od
[1].od_object
;
4212 blocksize
= od
[0].od_blocksize
;
4213 chunksize
= blocksize
;
4214 ASSERT(chunksize
== od
[1].od_gen
);
4216 VERIFY(dmu_object_info(os
, bigobj
, &doi
) == 0);
4217 VERIFY(ISP2(doi
.doi_data_block_size
));
4218 VERIFY(chunksize
== doi
.doi_data_block_size
);
4219 VERIFY(chunksize
>= 2 * sizeof (bufwad_t
));
4222 * Pick a random index and compute the offsets into packobj and bigobj.
4224 n
= ztest_random(regions
) * stride
+ ztest_random(width
);
4225 s
= 1 + ztest_random(width
- 1);
4227 packoff
= n
* sizeof (bufwad_t
);
4228 packsize
= s
* sizeof (bufwad_t
);
4230 bigoff
= n
* chunksize
;
4231 bigsize
= s
* chunksize
;
4233 packbuf
= umem_zalloc(packsize
, UMEM_NOFAIL
);
4234 bigbuf
= umem_zalloc(bigsize
, UMEM_NOFAIL
);
4236 VERIFY3U(0, ==, dmu_bonus_hold(os
, bigobj
, FTAG
, &bonus_db
));
4238 bigbuf_arcbufs
= umem_zalloc(2 * s
* sizeof (arc_buf_t
*), UMEM_NOFAIL
);
4241 * Iteration 0 test zcopy for DB_UNCACHED dbufs.
4242 * Iteration 1 test zcopy to already referenced dbufs.
4243 * Iteration 2 test zcopy to dirty dbuf in the same txg.
4244 * Iteration 3 test zcopy to dbuf dirty in previous txg.
4245 * Iteration 4 test zcopy when dbuf is no longer dirty.
4246 * Iteration 5 test zcopy when it can't be done.
4247 * Iteration 6 one more zcopy write.
4249 for (i
= 0; i
< 7; i
++) {
4254 * In iteration 5 (i == 5) use arcbufs
4255 * that don't match bigobj blksz to test
4256 * dmu_assign_arcbuf() when it can't directly
4257 * assign an arcbuf to a dbuf.
4259 for (j
= 0; j
< s
; j
++) {
4260 if (i
!= 5 || chunksize
< (SPA_MINBLOCKSIZE
* 2)) {
4262 dmu_request_arcbuf(bonus_db
, chunksize
);
4264 bigbuf_arcbufs
[2 * j
] =
4265 dmu_request_arcbuf(bonus_db
, chunksize
/ 2);
4266 bigbuf_arcbufs
[2 * j
+ 1] =
4267 dmu_request_arcbuf(bonus_db
, chunksize
/ 2);
4272 * Get a tx for the mods to both packobj and bigobj.
4274 tx
= dmu_tx_create(os
);
4276 dmu_tx_hold_write(tx
, packobj
, packoff
, packsize
);
4277 dmu_tx_hold_write(tx
, bigobj
, bigoff
, bigsize
);
4279 txg
= ztest_tx_assign(tx
, TXG_MIGHTWAIT
, FTAG
);
4281 umem_free(packbuf
, packsize
);
4282 umem_free(bigbuf
, bigsize
);
4283 for (j
= 0; j
< s
; j
++) {
4285 chunksize
< (SPA_MINBLOCKSIZE
* 2)) {
4286 dmu_return_arcbuf(bigbuf_arcbufs
[j
]);
4289 bigbuf_arcbufs
[2 * j
]);
4291 bigbuf_arcbufs
[2 * j
+ 1]);
4294 umem_free(bigbuf_arcbufs
, 2 * s
* sizeof (arc_buf_t
*));
4295 umem_free(od
, size
);
4296 dmu_buf_rele(bonus_db
, FTAG
);
4301 * 50% of the time don't read objects in the 1st iteration to
4302 * test dmu_assign_arcbuf() for the case when there're no
4303 * existing dbufs for the specified offsets.
4305 if (i
!= 0 || ztest_random(2) != 0) {
4306 error
= dmu_read(os
, packobj
, packoff
,
4307 packsize
, packbuf
, DMU_READ_PREFETCH
);
4309 error
= dmu_read(os
, bigobj
, bigoff
, bigsize
,
4310 bigbuf
, DMU_READ_PREFETCH
);
4313 compare_and_update_pbbufs(s
, packbuf
, bigbuf
, bigsize
,
4317 * We've verified all the old bufwads, and made new ones.
4318 * Now write them out.
4320 dmu_write(os
, packobj
, packoff
, packsize
, packbuf
, tx
);
4321 if (ztest_opts
.zo_verbose
>= 7) {
4322 (void) printf("writing offset %llx size %llx"
4324 (u_longlong_t
)bigoff
,
4325 (u_longlong_t
)bigsize
,
4328 for (off
= bigoff
, j
= 0; j
< s
; j
++, off
+= chunksize
) {
4330 if (i
!= 5 || chunksize
< (SPA_MINBLOCKSIZE
* 2)) {
4331 bcopy((caddr_t
)bigbuf
+ (off
- bigoff
),
4332 bigbuf_arcbufs
[j
]->b_data
, chunksize
);
4334 bcopy((caddr_t
)bigbuf
+ (off
- bigoff
),
4335 bigbuf_arcbufs
[2 * j
]->b_data
,
4337 bcopy((caddr_t
)bigbuf
+ (off
- bigoff
) +
4339 bigbuf_arcbufs
[2 * j
+ 1]->b_data
,
4344 VERIFY(dmu_buf_hold(os
, bigobj
, off
,
4345 FTAG
, &dbt
, DMU_READ_NO_PREFETCH
) == 0);
4347 if (i
!= 5 || chunksize
< (SPA_MINBLOCKSIZE
* 2)) {
4348 dmu_assign_arcbuf(bonus_db
, off
,
4349 bigbuf_arcbufs
[j
], tx
);
4351 dmu_assign_arcbuf(bonus_db
, off
,
4352 bigbuf_arcbufs
[2 * j
], tx
);
4353 dmu_assign_arcbuf(bonus_db
,
4354 off
+ chunksize
/ 2,
4355 bigbuf_arcbufs
[2 * j
+ 1], tx
);
4358 dmu_buf_rele(dbt
, FTAG
);
4364 * Sanity check the stuff we just wrote.
4367 void *packcheck
= umem_alloc(packsize
, UMEM_NOFAIL
);
4368 void *bigcheck
= umem_alloc(bigsize
, UMEM_NOFAIL
);
4370 VERIFY(0 == dmu_read(os
, packobj
, packoff
,
4371 packsize
, packcheck
, DMU_READ_PREFETCH
));
4372 VERIFY(0 == dmu_read(os
, bigobj
, bigoff
,
4373 bigsize
, bigcheck
, DMU_READ_PREFETCH
));
4375 ASSERT(bcmp(packbuf
, packcheck
, packsize
) == 0);
4376 ASSERT(bcmp(bigbuf
, bigcheck
, bigsize
) == 0);
4378 umem_free(packcheck
, packsize
);
4379 umem_free(bigcheck
, bigsize
);
4382 txg_wait_open(dmu_objset_pool(os
), 0);
4383 } else if (i
== 3) {
4384 txg_wait_synced(dmu_objset_pool(os
), 0);
4388 dmu_buf_rele(bonus_db
, FTAG
);
4389 umem_free(packbuf
, packsize
);
4390 umem_free(bigbuf
, bigsize
);
4391 umem_free(bigbuf_arcbufs
, 2 * s
* sizeof (arc_buf_t
*));
4392 umem_free(od
, size
);
4397 ztest_dmu_write_parallel(ztest_ds_t
*zd
, uint64_t id
)
4401 od
= umem_alloc(sizeof (ztest_od_t
), UMEM_NOFAIL
);
4402 uint64_t offset
= (1ULL << (ztest_random(20) + 43)) +
4403 (ztest_random(ZTEST_RANGE_LOCKS
) << SPA_MAXBLOCKSHIFT
);
4406 * Have multiple threads write to large offsets in an object
4407 * to verify that parallel writes to an object -- even to the
4408 * same blocks within the object -- doesn't cause any trouble.
4410 ztest_od_init(od
, ID_PARALLEL
, FTAG
, 0, DMU_OT_UINT64_OTHER
, 0, 0, 0);
4412 if (ztest_object_init(zd
, od
, sizeof (ztest_od_t
), B_FALSE
) != 0)
4415 while (ztest_random(10) != 0)
4416 ztest_io(zd
, od
->od_object
, offset
);
4418 umem_free(od
, sizeof (ztest_od_t
));
4422 ztest_dmu_prealloc(ztest_ds_t
*zd
, uint64_t id
)
4425 uint64_t offset
= (1ULL << (ztest_random(4) + SPA_MAXBLOCKSHIFT
)) +
4426 (ztest_random(ZTEST_RANGE_LOCKS
) << SPA_MAXBLOCKSHIFT
);
4427 uint64_t count
= ztest_random(20) + 1;
4428 uint64_t blocksize
= ztest_random_blocksize();
4431 od
= umem_alloc(sizeof (ztest_od_t
), UMEM_NOFAIL
);
4433 ztest_od_init(od
, id
, FTAG
, 0, DMU_OT_UINT64_OTHER
, blocksize
, 0, 0);
4435 if (ztest_object_init(zd
, od
, sizeof (ztest_od_t
),
4436 !ztest_random(2)) != 0) {
4437 umem_free(od
, sizeof (ztest_od_t
));
4441 if (ztest_truncate(zd
, od
->od_object
, offset
, count
* blocksize
) != 0) {
4442 umem_free(od
, sizeof (ztest_od_t
));
4446 ztest_prealloc(zd
, od
->od_object
, offset
, count
* blocksize
);
4448 data
= umem_zalloc(blocksize
, UMEM_NOFAIL
);
4450 while (ztest_random(count
) != 0) {
4451 uint64_t randoff
= offset
+ (ztest_random(count
) * blocksize
);
4452 if (ztest_write(zd
, od
->od_object
, randoff
, blocksize
,
4455 while (ztest_random(4) != 0)
4456 ztest_io(zd
, od
->od_object
, randoff
);
4459 umem_free(data
, blocksize
);
4460 umem_free(od
, sizeof (ztest_od_t
));
4464 * Verify that zap_{create,destroy,add,remove,update} work as expected.
4466 #define ZTEST_ZAP_MIN_INTS 1
4467 #define ZTEST_ZAP_MAX_INTS 4
4468 #define ZTEST_ZAP_MAX_PROPS 1000
4471 ztest_zap(ztest_ds_t
*zd
, uint64_t id
)
4473 objset_t
*os
= zd
->zd_os
;
4476 uint64_t txg
, last_txg
;
4477 uint64_t value
[ZTEST_ZAP_MAX_INTS
];
4478 uint64_t zl_ints
, zl_intsize
, prop
;
4481 char propname
[100], txgname
[100];
4483 char *hc
[2] = { "s.acl.h", ".s.open.h.hyLZlg" };
4485 od
= umem_alloc(sizeof (ztest_od_t
), UMEM_NOFAIL
);
4486 ztest_od_init(od
, id
, FTAG
, 0, DMU_OT_ZAP_OTHER
, 0, 0, 0);
4488 if (ztest_object_init(zd
, od
, sizeof (ztest_od_t
),
4489 !ztest_random(2)) != 0)
4492 object
= od
->od_object
;
4495 * Generate a known hash collision, and verify that
4496 * we can lookup and remove both entries.
4498 tx
= dmu_tx_create(os
);
4499 dmu_tx_hold_zap(tx
, object
, B_TRUE
, NULL
);
4500 txg
= ztest_tx_assign(tx
, TXG_MIGHTWAIT
, FTAG
);
4503 for (i
= 0; i
< 2; i
++) {
4505 VERIFY3U(0, ==, zap_add(os
, object
, hc
[i
], sizeof (uint64_t),
4508 for (i
= 0; i
< 2; i
++) {
4509 VERIFY3U(EEXIST
, ==, zap_add(os
, object
, hc
[i
],
4510 sizeof (uint64_t), 1, &value
[i
], tx
));
4512 zap_length(os
, object
, hc
[i
], &zl_intsize
, &zl_ints
));
4513 ASSERT3U(zl_intsize
, ==, sizeof (uint64_t));
4514 ASSERT3U(zl_ints
, ==, 1);
4516 for (i
= 0; i
< 2; i
++) {
4517 VERIFY3U(0, ==, zap_remove(os
, object
, hc
[i
], tx
));
4522 * Generate a buch of random entries.
4524 ints
= MAX(ZTEST_ZAP_MIN_INTS
, object
% ZTEST_ZAP_MAX_INTS
);
4526 prop
= ztest_random(ZTEST_ZAP_MAX_PROPS
);
4527 (void) sprintf(propname
, "prop_%llu", (u_longlong_t
)prop
);
4528 (void) sprintf(txgname
, "txg_%llu", (u_longlong_t
)prop
);
4529 bzero(value
, sizeof (value
));
4533 * If these zap entries already exist, validate their contents.
4535 error
= zap_length(os
, object
, txgname
, &zl_intsize
, &zl_ints
);
4537 ASSERT3U(zl_intsize
, ==, sizeof (uint64_t));
4538 ASSERT3U(zl_ints
, ==, 1);
4540 VERIFY(zap_lookup(os
, object
, txgname
, zl_intsize
,
4541 zl_ints
, &last_txg
) == 0);
4543 VERIFY(zap_length(os
, object
, propname
, &zl_intsize
,
4546 ASSERT3U(zl_intsize
, ==, sizeof (uint64_t));
4547 ASSERT3U(zl_ints
, ==, ints
);
4549 VERIFY(zap_lookup(os
, object
, propname
, zl_intsize
,
4550 zl_ints
, value
) == 0);
4552 for (i
= 0; i
< ints
; i
++) {
4553 ASSERT3U(value
[i
], ==, last_txg
+ object
+ i
);
4556 ASSERT3U(error
, ==, ENOENT
);
4560 * Atomically update two entries in our zap object.
4561 * The first is named txg_%llu, and contains the txg
4562 * in which the property was last updated. The second
4563 * is named prop_%llu, and the nth element of its value
4564 * should be txg + object + n.
4566 tx
= dmu_tx_create(os
);
4567 dmu_tx_hold_zap(tx
, object
, B_TRUE
, NULL
);
4568 txg
= ztest_tx_assign(tx
, TXG_MIGHTWAIT
, FTAG
);
4573 fatal(0, "zap future leak: old %llu new %llu", last_txg
, txg
);
4575 for (i
= 0; i
< ints
; i
++)
4576 value
[i
] = txg
+ object
+ i
;
4578 VERIFY3U(0, ==, zap_update(os
, object
, txgname
, sizeof (uint64_t),
4580 VERIFY3U(0, ==, zap_update(os
, object
, propname
, sizeof (uint64_t),
4586 * Remove a random pair of entries.
4588 prop
= ztest_random(ZTEST_ZAP_MAX_PROPS
);
4589 (void) sprintf(propname
, "prop_%llu", (u_longlong_t
)prop
);
4590 (void) sprintf(txgname
, "txg_%llu", (u_longlong_t
)prop
);
4592 error
= zap_length(os
, object
, txgname
, &zl_intsize
, &zl_ints
);
4594 if (error
== ENOENT
)
4599 tx
= dmu_tx_create(os
);
4600 dmu_tx_hold_zap(tx
, object
, B_TRUE
, NULL
);
4601 txg
= ztest_tx_assign(tx
, TXG_MIGHTWAIT
, FTAG
);
4604 VERIFY3U(0, ==, zap_remove(os
, object
, txgname
, tx
));
4605 VERIFY3U(0, ==, zap_remove(os
, object
, propname
, tx
));
4608 umem_free(od
, sizeof (ztest_od_t
));
4612 * Testcase to test the upgrading of a microzap to fatzap.
4615 ztest_fzap(ztest_ds_t
*zd
, uint64_t id
)
4617 objset_t
*os
= zd
->zd_os
;
4619 uint64_t object
, txg
;
4622 od
= umem_alloc(sizeof (ztest_od_t
), UMEM_NOFAIL
);
4623 ztest_od_init(od
, id
, FTAG
, 0, DMU_OT_ZAP_OTHER
, 0, 0, 0);
4625 if (ztest_object_init(zd
, od
, sizeof (ztest_od_t
),
4626 !ztest_random(2)) != 0)
4628 object
= od
->od_object
;
4631 * Add entries to this ZAP and make sure it spills over
4632 * and gets upgraded to a fatzap. Also, since we are adding
4633 * 2050 entries we should see ptrtbl growth and leaf-block split.
4635 for (i
= 0; i
< 2050; i
++) {
4636 char name
[ZFS_MAX_DATASET_NAME_LEN
];
4641 (void) snprintf(name
, sizeof (name
), "fzap-%llu-%llu",
4642 (u_longlong_t
)id
, (u_longlong_t
)value
);
4644 tx
= dmu_tx_create(os
);
4645 dmu_tx_hold_zap(tx
, object
, B_TRUE
, name
);
4646 txg
= ztest_tx_assign(tx
, TXG_MIGHTWAIT
, FTAG
);
4649 error
= zap_add(os
, object
, name
, sizeof (uint64_t), 1,
4651 ASSERT(error
== 0 || error
== EEXIST
);
4655 umem_free(od
, sizeof (ztest_od_t
));
4660 ztest_zap_parallel(ztest_ds_t
*zd
, uint64_t id
)
4662 objset_t
*os
= zd
->zd_os
;
4664 uint64_t txg
, object
, count
, wsize
, wc
, zl_wsize
, zl_wc
;
4666 int i
, namelen
, error
;
4667 int micro
= ztest_random(2);
4668 char name
[20], string_value
[20];
4671 od
= umem_alloc(sizeof (ztest_od_t
), UMEM_NOFAIL
);
4672 ztest_od_init(od
, ID_PARALLEL
, FTAG
, micro
, DMU_OT_ZAP_OTHER
, 0, 0, 0);
4674 if (ztest_object_init(zd
, od
, sizeof (ztest_od_t
), B_FALSE
) != 0) {
4675 umem_free(od
, sizeof (ztest_od_t
));
4679 object
= od
->od_object
;
4682 * Generate a random name of the form 'xxx.....' where each
4683 * x is a random printable character and the dots are dots.
4684 * There are 94 such characters, and the name length goes from
4685 * 6 to 20, so there are 94^3 * 15 = 12,458,760 possible names.
4687 namelen
= ztest_random(sizeof (name
) - 5) + 5 + 1;
4689 for (i
= 0; i
< 3; i
++)
4690 name
[i
] = '!' + ztest_random('~' - '!' + 1);
4691 for (; i
< namelen
- 1; i
++)
4695 if ((namelen
& 1) || micro
) {
4696 wsize
= sizeof (txg
);
4702 data
= string_value
;
4706 VERIFY0(zap_count(os
, object
, &count
));
4707 ASSERT(count
!= -1ULL);
4710 * Select an operation: length, lookup, add, update, remove.
4712 i
= ztest_random(5);
4715 tx
= dmu_tx_create(os
);
4716 dmu_tx_hold_zap(tx
, object
, B_TRUE
, NULL
);
4717 txg
= ztest_tx_assign(tx
, TXG_MIGHTWAIT
, FTAG
);
4719 umem_free(od
, sizeof (ztest_od_t
));
4722 bcopy(name
, string_value
, namelen
);
4726 bzero(string_value
, namelen
);
4732 error
= zap_length(os
, object
, name
, &zl_wsize
, &zl_wc
);
4734 ASSERT3U(wsize
, ==, zl_wsize
);
4735 ASSERT3U(wc
, ==, zl_wc
);
4737 ASSERT3U(error
, ==, ENOENT
);
4742 error
= zap_lookup(os
, object
, name
, wsize
, wc
, data
);
4744 if (data
== string_value
&&
4745 bcmp(name
, data
, namelen
) != 0)
4746 fatal(0, "name '%s' != val '%s' len %d",
4747 name
, data
, namelen
);
4749 ASSERT3U(error
, ==, ENOENT
);
4754 error
= zap_add(os
, object
, name
, wsize
, wc
, data
, tx
);
4755 ASSERT(error
== 0 || error
== EEXIST
);
4759 VERIFY(zap_update(os
, object
, name
, wsize
, wc
, data
, tx
) == 0);
4763 error
= zap_remove(os
, object
, name
, tx
);
4764 ASSERT(error
== 0 || error
== ENOENT
);
4771 umem_free(od
, sizeof (ztest_od_t
));
4775 * Commit callback data.
4777 typedef struct ztest_cb_data
{
4778 list_node_t zcd_node
;
4780 int zcd_expected_err
;
4781 boolean_t zcd_added
;
4782 boolean_t zcd_called
;
4786 /* This is the actual commit callback function */
4788 ztest_commit_callback(void *arg
, int error
)
4790 ztest_cb_data_t
*data
= arg
;
4791 uint64_t synced_txg
;
4793 VERIFY(data
!= NULL
);
4794 VERIFY3S(data
->zcd_expected_err
, ==, error
);
4795 VERIFY(!data
->zcd_called
);
4797 synced_txg
= spa_last_synced_txg(data
->zcd_spa
);
4798 if (data
->zcd_txg
> synced_txg
)
4799 fatal(0, "commit callback of txg %" PRIu64
" called prematurely"
4800 ", last synced txg = %" PRIu64
"\n", data
->zcd_txg
,
4803 data
->zcd_called
= B_TRUE
;
4805 if (error
== ECANCELED
) {
4806 ASSERT0(data
->zcd_txg
);
4807 ASSERT(!data
->zcd_added
);
4810 * The private callback data should be destroyed here, but
4811 * since we are going to check the zcd_called field after
4812 * dmu_tx_abort(), we will destroy it there.
4817 ASSERT(data
->zcd_added
);
4818 ASSERT3U(data
->zcd_txg
, !=, 0);
4820 (void) mutex_enter(&zcl
.zcl_callbacks_lock
);
4822 /* See if this cb was called more quickly */
4823 if ((synced_txg
- data
->zcd_txg
) < zc_min_txg_delay
)
4824 zc_min_txg_delay
= synced_txg
- data
->zcd_txg
;
4826 /* Remove our callback from the list */
4827 list_remove(&zcl
.zcl_callbacks
, data
);
4829 (void) mutex_exit(&zcl
.zcl_callbacks_lock
);
4831 umem_free(data
, sizeof (ztest_cb_data_t
));
4834 /* Allocate and initialize callback data structure */
4835 static ztest_cb_data_t
*
4836 ztest_create_cb_data(objset_t
*os
, uint64_t txg
)
4838 ztest_cb_data_t
*cb_data
;
4840 cb_data
= umem_zalloc(sizeof (ztest_cb_data_t
), UMEM_NOFAIL
);
4842 cb_data
->zcd_txg
= txg
;
4843 cb_data
->zcd_spa
= dmu_objset_spa(os
);
4844 list_link_init(&cb_data
->zcd_node
);
4850 * Commit callback test.
4853 ztest_dmu_commit_callbacks(ztest_ds_t
*zd
, uint64_t id
)
4855 objset_t
*os
= zd
->zd_os
;
4858 ztest_cb_data_t
*cb_data
[3], *tmp_cb
;
4859 uint64_t old_txg
, txg
;
4862 od
= umem_alloc(sizeof (ztest_od_t
), UMEM_NOFAIL
);
4863 ztest_od_init(od
, id
, FTAG
, 0, DMU_OT_UINT64_OTHER
, 0, 0, 0);
4865 if (ztest_object_init(zd
, od
, sizeof (ztest_od_t
), B_FALSE
) != 0) {
4866 umem_free(od
, sizeof (ztest_od_t
));
4870 tx
= dmu_tx_create(os
);
4872 cb_data
[0] = ztest_create_cb_data(os
, 0);
4873 dmu_tx_callback_register(tx
, ztest_commit_callback
, cb_data
[0]);
4875 dmu_tx_hold_write(tx
, od
->od_object
, 0, sizeof (uint64_t));
4877 /* Every once in a while, abort the transaction on purpose */
4878 if (ztest_random(100) == 0)
4882 error
= dmu_tx_assign(tx
, TXG_NOWAIT
);
4884 txg
= error
? 0 : dmu_tx_get_txg(tx
);
4886 cb_data
[0]->zcd_txg
= txg
;
4887 cb_data
[1] = ztest_create_cb_data(os
, txg
);
4888 dmu_tx_callback_register(tx
, ztest_commit_callback
, cb_data
[1]);
4892 * It's not a strict requirement to call the registered
4893 * callbacks from inside dmu_tx_abort(), but that's what
4894 * it's supposed to happen in the current implementation
4895 * so we will check for that.
4897 for (i
= 0; i
< 2; i
++) {
4898 cb_data
[i
]->zcd_expected_err
= ECANCELED
;
4899 VERIFY(!cb_data
[i
]->zcd_called
);
4904 for (i
= 0; i
< 2; i
++) {
4905 VERIFY(cb_data
[i
]->zcd_called
);
4906 umem_free(cb_data
[i
], sizeof (ztest_cb_data_t
));
4909 umem_free(od
, sizeof (ztest_od_t
));
4913 cb_data
[2] = ztest_create_cb_data(os
, txg
);
4914 dmu_tx_callback_register(tx
, ztest_commit_callback
, cb_data
[2]);
4917 * Read existing data to make sure there isn't a future leak.
4919 VERIFY(0 == dmu_read(os
, od
->od_object
, 0, sizeof (uint64_t),
4920 &old_txg
, DMU_READ_PREFETCH
));
4923 fatal(0, "future leak: got %" PRIu64
", open txg is %" PRIu64
,
4926 dmu_write(os
, od
->od_object
, 0, sizeof (uint64_t), &txg
, tx
);
4928 (void) mutex_enter(&zcl
.zcl_callbacks_lock
);
4931 * Since commit callbacks don't have any ordering requirement and since
4932 * it is theoretically possible for a commit callback to be called
4933 * after an arbitrary amount of time has elapsed since its txg has been
4934 * synced, it is difficult to reliably determine whether a commit
4935 * callback hasn't been called due to high load or due to a flawed
4938 * In practice, we will assume that if after a certain number of txgs a
4939 * commit callback hasn't been called, then most likely there's an
4940 * implementation bug..
4942 tmp_cb
= list_head(&zcl
.zcl_callbacks
);
4943 if (tmp_cb
!= NULL
&&
4944 tmp_cb
->zcd_txg
+ ZTEST_COMMIT_CB_THRESH
< txg
) {
4945 fatal(0, "Commit callback threshold exceeded, oldest txg: %"
4946 PRIu64
", open txg: %" PRIu64
"\n", tmp_cb
->zcd_txg
, txg
);
4950 * Let's find the place to insert our callbacks.
4952 * Even though the list is ordered by txg, it is possible for the
4953 * insertion point to not be the end because our txg may already be
4954 * quiescing at this point and other callbacks in the open txg
4955 * (from other objsets) may have sneaked in.
4957 tmp_cb
= list_tail(&zcl
.zcl_callbacks
);
4958 while (tmp_cb
!= NULL
&& tmp_cb
->zcd_txg
> txg
)
4959 tmp_cb
= list_prev(&zcl
.zcl_callbacks
, tmp_cb
);
4961 /* Add the 3 callbacks to the list */
4962 for (i
= 0; i
< 3; i
++) {
4964 list_insert_head(&zcl
.zcl_callbacks
, cb_data
[i
]);
4966 list_insert_after(&zcl
.zcl_callbacks
, tmp_cb
,
4969 cb_data
[i
]->zcd_added
= B_TRUE
;
4970 VERIFY(!cb_data
[i
]->zcd_called
);
4972 tmp_cb
= cb_data
[i
];
4977 (void) mutex_exit(&zcl
.zcl_callbacks_lock
);
4981 umem_free(od
, sizeof (ztest_od_t
));
4985 * Visit each object in the dataset. Verify that its properties
4986 * are consistent what was stored in the block tag when it was created,
4987 * and that its unused bonus buffer space has not been overwritten.
4991 ztest_verify_dnode_bt(ztest_ds_t
*zd
, uint64_t id
)
4993 objset_t
*os
= zd
->zd_os
;
4997 for (obj
= 0; err
== 0; err
= dmu_object_next(os
, &obj
, FALSE
, 0)) {
4998 ztest_block_tag_t
*bt
= NULL
;
4999 dmu_object_info_t doi
;
5002 if (dmu_bonus_hold(os
, obj
, FTAG
, &db
) != 0)
5005 dmu_object_info_from_db(db
, &doi
);
5006 if (doi
.doi_bonus_size
>= sizeof (*bt
))
5007 bt
= ztest_bt_bonus(db
);
5009 if (bt
&& bt
->bt_magic
== BT_MAGIC
) {
5010 ztest_bt_verify(bt
, os
, obj
, doi
.doi_dnodesize
,
5011 bt
->bt_offset
, bt
->bt_gen
, bt
->bt_txg
,
5013 ztest_verify_unused_bonus(db
, bt
, obj
, os
, bt
->bt_gen
);
5016 dmu_buf_rele(db
, FTAG
);
5022 ztest_dsl_prop_get_set(ztest_ds_t
*zd
, uint64_t id
)
5024 zfs_prop_t proplist
[] = {
5026 ZFS_PROP_COMPRESSION
,
5032 (void) rw_rdlock(&ztest_name_lock
);
5034 for (p
= 0; p
< sizeof (proplist
) / sizeof (proplist
[0]); p
++)
5035 (void) ztest_dsl_prop_set_uint64(zd
->zd_name
, proplist
[p
],
5036 ztest_random_dsl_prop(proplist
[p
]), (int)ztest_random(2));
5038 VERIFY0(ztest_dsl_prop_set_uint64(zd
->zd_name
, ZFS_PROP_RECORDSIZE
,
5039 ztest_random_blocksize(), (int)ztest_random(2)));
5041 (void) rw_unlock(&ztest_name_lock
);
5046 ztest_spa_prop_get_set(ztest_ds_t
*zd
, uint64_t id
)
5048 nvlist_t
*props
= NULL
;
5050 (void) rw_rdlock(&ztest_name_lock
);
5052 (void) ztest_spa_prop_set_uint64(ZPOOL_PROP_DEDUPDITTO
,
5053 ZIO_DEDUPDITTO_MIN
+ ztest_random(ZIO_DEDUPDITTO_MIN
));
5055 VERIFY0(spa_prop_get(ztest_spa
, &props
));
5057 if (ztest_opts
.zo_verbose
>= 6)
5058 dump_nvlist(props
, 4);
5062 (void) rw_unlock(&ztest_name_lock
);
5066 user_release_one(const char *snapname
, const char *holdname
)
5068 nvlist_t
*snaps
, *holds
;
5071 snaps
= fnvlist_alloc();
5072 holds
= fnvlist_alloc();
5073 fnvlist_add_boolean(holds
, holdname
);
5074 fnvlist_add_nvlist(snaps
, snapname
, holds
);
5075 fnvlist_free(holds
);
5076 error
= dsl_dataset_user_release(snaps
, NULL
);
5077 fnvlist_free(snaps
);
5082 * Test snapshot hold/release and deferred destroy.
5085 ztest_dmu_snapshot_hold(ztest_ds_t
*zd
, uint64_t id
)
5088 objset_t
*os
= zd
->zd_os
;
5092 char clonename
[100];
5094 char osname
[ZFS_MAX_DATASET_NAME_LEN
];
5097 (void) rw_rdlock(&ztest_name_lock
);
5099 dmu_objset_name(os
, osname
);
5101 (void) snprintf(snapname
, sizeof (snapname
), "sh1_%llu",
5103 (void) snprintf(fullname
, sizeof (fullname
), "%s@%s", osname
, snapname
);
5104 (void) snprintf(clonename
, sizeof (clonename
),
5105 "%s/ch1_%llu", osname
, (u_longlong_t
)id
);
5106 (void) snprintf(tag
, sizeof (tag
), "tag_%llu", (u_longlong_t
)id
);
5109 * Clean up from any previous run.
5111 error
= dsl_destroy_head(clonename
);
5112 if (error
!= ENOENT
)
5114 error
= user_release_one(fullname
, tag
);
5115 if (error
!= ESRCH
&& error
!= ENOENT
)
5117 error
= dsl_destroy_snapshot(fullname
, B_FALSE
);
5118 if (error
!= ENOENT
)
5122 * Create snapshot, clone it, mark snap for deferred destroy,
5123 * destroy clone, verify snap was also destroyed.
5125 error
= dmu_objset_snapshot_one(osname
, snapname
);
5127 if (error
== ENOSPC
) {
5128 ztest_record_enospc("dmu_objset_snapshot");
5131 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname
, error
);
5134 error
= dmu_objset_clone(clonename
, fullname
);
5136 if (error
== ENOSPC
) {
5137 ztest_record_enospc("dmu_objset_clone");
5140 fatal(0, "dmu_objset_clone(%s) = %d", clonename
, error
);
5143 error
= dsl_destroy_snapshot(fullname
, B_TRUE
);
5145 fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
5149 error
= dsl_destroy_head(clonename
);
5151 fatal(0, "dsl_destroy_head(%s) = %d", clonename
, error
);
5153 error
= dmu_objset_hold(fullname
, FTAG
, &origin
);
5154 if (error
!= ENOENT
)
5155 fatal(0, "dmu_objset_hold(%s) = %d", fullname
, error
);
5158 * Create snapshot, add temporary hold, verify that we can't
5159 * destroy a held snapshot, mark for deferred destroy,
5160 * release hold, verify snapshot was destroyed.
5162 error
= dmu_objset_snapshot_one(osname
, snapname
);
5164 if (error
== ENOSPC
) {
5165 ztest_record_enospc("dmu_objset_snapshot");
5168 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname
, error
);
5171 holds
= fnvlist_alloc();
5172 fnvlist_add_string(holds
, fullname
, tag
);
5173 error
= dsl_dataset_user_hold(holds
, 0, NULL
);
5174 fnvlist_free(holds
);
5176 if (error
== ENOSPC
) {
5177 ztest_record_enospc("dsl_dataset_user_hold");
5180 fatal(0, "dsl_dataset_user_hold(%s, %s) = %u",
5181 fullname
, tag
, error
);
5184 error
= dsl_destroy_snapshot(fullname
, B_FALSE
);
5185 if (error
!= EBUSY
) {
5186 fatal(0, "dsl_destroy_snapshot(%s, B_FALSE) = %d",
5190 error
= dsl_destroy_snapshot(fullname
, B_TRUE
);
5192 fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
5196 error
= user_release_one(fullname
, tag
);
5198 fatal(0, "user_release_one(%s, %s) = %d", fullname
, tag
, error
);
5200 VERIFY3U(dmu_objset_hold(fullname
, FTAG
, &origin
), ==, ENOENT
);
5203 (void) rw_unlock(&ztest_name_lock
);
5207 * Inject random faults into the on-disk data.
5211 ztest_fault_inject(ztest_ds_t
*zd
, uint64_t id
)
5213 ztest_shared_t
*zs
= ztest_shared
;
5214 spa_t
*spa
= ztest_spa
;
5218 uint64_t bad
= 0x1990c0ffeedecadeull
;
5223 int bshift
= SPA_MAXBLOCKSHIFT
+ 2;
5229 boolean_t islog
= B_FALSE
;
5231 path0
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
5232 pathrand
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
5234 mutex_enter(&ztest_vdev_lock
);
5235 maxfaults
= MAXFAULTS();
5236 leaves
= MAX(zs
->zs_mirrors
, 1) * ztest_opts
.zo_raidz
;
5237 mirror_save
= zs
->zs_mirrors
;
5238 mutex_exit(&ztest_vdev_lock
);
5240 ASSERT(leaves
>= 1);
5243 * Grab the name lock as reader. There are some operations
5244 * which don't like to have their vdevs changed while
5245 * they are in progress (i.e. spa_change_guid). Those
5246 * operations will have grabbed the name lock as writer.
5248 (void) rw_rdlock(&ztest_name_lock
);
5251 * We need SCL_STATE here because we're going to look at vd0->vdev_tsd.
5253 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
5255 if (ztest_random(2) == 0) {
5257 * Inject errors on a normal data device or slog device.
5259 top
= ztest_random_vdev_top(spa
, B_TRUE
);
5260 leaf
= ztest_random(leaves
) + zs
->zs_splits
;
5263 * Generate paths to the first leaf in this top-level vdev,
5264 * and to the random leaf we selected. We'll induce transient
5265 * write failures and random online/offline activity on leaf 0,
5266 * and we'll write random garbage to the randomly chosen leaf.
5268 (void) snprintf(path0
, MAXPATHLEN
, ztest_dev_template
,
5269 ztest_opts
.zo_dir
, ztest_opts
.zo_pool
,
5270 top
* leaves
+ zs
->zs_splits
);
5271 (void) snprintf(pathrand
, MAXPATHLEN
, ztest_dev_template
,
5272 ztest_opts
.zo_dir
, ztest_opts
.zo_pool
,
5273 top
* leaves
+ leaf
);
5275 vd0
= vdev_lookup_by_path(spa
->spa_root_vdev
, path0
);
5276 if (vd0
!= NULL
&& vd0
->vdev_top
->vdev_islog
)
5280 * If the top-level vdev needs to be resilvered
5281 * then we only allow faults on the device that is
5284 if (vd0
!= NULL
&& maxfaults
!= 1 &&
5285 (!vdev_resilver_needed(vd0
->vdev_top
, NULL
, NULL
) ||
5286 vd0
->vdev_resilver_txg
!= 0)) {
5288 * Make vd0 explicitly claim to be unreadable,
5289 * or unwriteable, or reach behind its back
5290 * and close the underlying fd. We can do this if
5291 * maxfaults == 0 because we'll fail and reexecute,
5292 * and we can do it if maxfaults >= 2 because we'll
5293 * have enough redundancy. If maxfaults == 1, the
5294 * combination of this with injection of random data
5295 * corruption below exceeds the pool's fault tolerance.
5297 vdev_file_t
*vf
= vd0
->vdev_tsd
;
5299 if (vf
!= NULL
&& ztest_random(3) == 0) {
5300 (void) close(vf
->vf_vnode
->v_fd
);
5301 vf
->vf_vnode
->v_fd
= -1;
5302 } else if (ztest_random(2) == 0) {
5303 vd0
->vdev_cant_read
= B_TRUE
;
5305 vd0
->vdev_cant_write
= B_TRUE
;
5307 guid0
= vd0
->vdev_guid
;
5311 * Inject errors on an l2cache device.
5313 spa_aux_vdev_t
*sav
= &spa
->spa_l2cache
;
5315 if (sav
->sav_count
== 0) {
5316 spa_config_exit(spa
, SCL_STATE
, FTAG
);
5317 (void) rw_unlock(&ztest_name_lock
);
5320 vd0
= sav
->sav_vdevs
[ztest_random(sav
->sav_count
)];
5321 guid0
= vd0
->vdev_guid
;
5322 (void) strcpy(path0
, vd0
->vdev_path
);
5323 (void) strcpy(pathrand
, vd0
->vdev_path
);
5327 maxfaults
= INT_MAX
; /* no limit on cache devices */
5330 spa_config_exit(spa
, SCL_STATE
, FTAG
);
5331 (void) rw_unlock(&ztest_name_lock
);
5334 * If we can tolerate two or more faults, or we're dealing
5335 * with a slog, randomly online/offline vd0.
5337 if ((maxfaults
>= 2 || islog
) && guid0
!= 0) {
5338 if (ztest_random(10) < 6) {
5339 int flags
= (ztest_random(2) == 0 ?
5340 ZFS_OFFLINE_TEMPORARY
: 0);
5343 * We have to grab the zs_name_lock as writer to
5344 * prevent a race between offlining a slog and
5345 * destroying a dataset. Offlining the slog will
5346 * grab a reference on the dataset which may cause
5347 * dsl_destroy_head() to fail with EBUSY thus
5348 * leaving the dataset in an inconsistent state.
5351 (void) rw_wrlock(&ztest_name_lock
);
5353 VERIFY(vdev_offline(spa
, guid0
, flags
) != EBUSY
);
5356 (void) rw_unlock(&ztest_name_lock
);
5359 * Ideally we would like to be able to randomly
5360 * call vdev_[on|off]line without holding locks
5361 * to force unpredictable failures but the side
5362 * effects of vdev_[on|off]line prevent us from
5363 * doing so. We grab the ztest_vdev_lock here to
5364 * prevent a race between injection testing and
5367 mutex_enter(&ztest_vdev_lock
);
5368 (void) vdev_online(spa
, guid0
, 0, NULL
);
5369 mutex_exit(&ztest_vdev_lock
);
5377 * We have at least single-fault tolerance, so inject data corruption.
5379 fd
= open(pathrand
, O_RDWR
);
5381 if (fd
== -1) /* we hit a gap in the device namespace */
5384 fsize
= lseek(fd
, 0, SEEK_END
);
5386 while (--iters
!= 0) {
5388 * The offset must be chosen carefully to ensure that
5389 * we do not inject a given logical block with errors
5390 * on two different leaf devices, because ZFS can not
5391 * tolerate that (if maxfaults==1).
5393 * We divide each leaf into chunks of size
5394 * (# leaves * SPA_MAXBLOCKSIZE * 4). Within each chunk
5395 * there is a series of ranges to which we can inject errors.
5396 * Each range can accept errors on only a single leaf vdev.
5397 * The error injection ranges are separated by ranges
5398 * which we will not inject errors on any device (DMZs).
5399 * Each DMZ must be large enough such that a single block
5400 * can not straddle it, so that a single block can not be
5401 * a target in two different injection ranges (on different
5404 * For example, with 3 leaves, each chunk looks like:
5405 * 0 to 32M: injection range for leaf 0
5406 * 32M to 64M: DMZ - no injection allowed
5407 * 64M to 96M: injection range for leaf 1
5408 * 96M to 128M: DMZ - no injection allowed
5409 * 128M to 160M: injection range for leaf 2
5410 * 160M to 192M: DMZ - no injection allowed
5412 offset
= ztest_random(fsize
/ (leaves
<< bshift
)) *
5413 (leaves
<< bshift
) + (leaf
<< bshift
) +
5414 (ztest_random(1ULL << (bshift
- 1)) & -8ULL);
5417 * Only allow damage to the labels at one end of the vdev.
5419 * If all labels are damaged, the device will be totally
5420 * inaccessible, which will result in loss of data,
5421 * because we also damage (parts of) the other side of
5424 * Additionally, we will always have both an even and an
5425 * odd label, so that we can handle crashes in the
5426 * middle of vdev_config_sync().
5428 if ((leaf
& 1) == 0 && offset
< VDEV_LABEL_START_SIZE
)
5432 * The two end labels are stored at the "end" of the disk, but
5433 * the end of the disk (vdev_psize) is aligned to
5434 * sizeof (vdev_label_t).
5436 uint64_t psize
= P2ALIGN(fsize
, sizeof (vdev_label_t
));
5437 if ((leaf
& 1) == 1 &&
5438 offset
+ sizeof (bad
) > psize
- VDEV_LABEL_END_SIZE
)
5441 mutex_enter(&ztest_vdev_lock
);
5442 if (mirror_save
!= zs
->zs_mirrors
) {
5443 mutex_exit(&ztest_vdev_lock
);
5448 if (pwrite(fd
, &bad
, sizeof (bad
), offset
) != sizeof (bad
))
5449 fatal(1, "can't inject bad word at 0x%llx in %s",
5452 mutex_exit(&ztest_vdev_lock
);
5454 if (ztest_opts
.zo_verbose
>= 7)
5455 (void) printf("injected bad word into %s,"
5456 " offset 0x%llx\n", pathrand
, (u_longlong_t
)offset
);
5461 umem_free(path0
, MAXPATHLEN
);
5462 umem_free(pathrand
, MAXPATHLEN
);
5466 * Verify that DDT repair works as expected.
5469 ztest_ddt_repair(ztest_ds_t
*zd
, uint64_t id
)
5471 ztest_shared_t
*zs
= ztest_shared
;
5472 spa_t
*spa
= ztest_spa
;
5473 objset_t
*os
= zd
->zd_os
;
5475 uint64_t object
, blocksize
, txg
, pattern
, psize
;
5476 enum zio_checksum checksum
= spa_dedup_checksum(spa
);
5481 int copies
= 2 * ZIO_DEDUPDITTO_MIN
;
5484 blocksize
= ztest_random_blocksize();
5485 blocksize
= MIN(blocksize
, 2048); /* because we write so many */
5487 od
= umem_alloc(sizeof (ztest_od_t
), UMEM_NOFAIL
);
5488 ztest_od_init(od
, id
, FTAG
, 0, DMU_OT_UINT64_OTHER
, blocksize
, 0, 0);
5490 if (ztest_object_init(zd
, od
, sizeof (ztest_od_t
), B_FALSE
) != 0) {
5491 umem_free(od
, sizeof (ztest_od_t
));
5496 * Take the name lock as writer to prevent anyone else from changing
5497 * the pool and dataset properies we need to maintain during this test.
5499 (void) rw_wrlock(&ztest_name_lock
);
5501 if (ztest_dsl_prop_set_uint64(zd
->zd_name
, ZFS_PROP_DEDUP
, checksum
,
5503 ztest_dsl_prop_set_uint64(zd
->zd_name
, ZFS_PROP_COPIES
, 1,
5505 (void) rw_unlock(&ztest_name_lock
);
5506 umem_free(od
, sizeof (ztest_od_t
));
5510 dmu_objset_stats_t dds
;
5511 dsl_pool_config_enter(dmu_objset_pool(os
), FTAG
);
5512 dmu_objset_fast_stat(os
, &dds
);
5513 dsl_pool_config_exit(dmu_objset_pool(os
), FTAG
);
5515 object
= od
[0].od_object
;
5516 blocksize
= od
[0].od_blocksize
;
5517 pattern
= zs
->zs_guid
^ dds
.dds_guid
;
5519 ASSERT(object
!= 0);
5521 tx
= dmu_tx_create(os
);
5522 dmu_tx_hold_write(tx
, object
, 0, copies
* blocksize
);
5523 txg
= ztest_tx_assign(tx
, TXG_WAIT
, FTAG
);
5525 (void) rw_unlock(&ztest_name_lock
);
5526 umem_free(od
, sizeof (ztest_od_t
));
5531 * Write all the copies of our block.
5533 for (i
= 0; i
< copies
; i
++) {
5534 uint64_t offset
= i
* blocksize
;
5535 int error
= dmu_buf_hold(os
, object
, offset
, FTAG
, &db
,
5536 DMU_READ_NO_PREFETCH
);
5538 fatal(B_FALSE
, "dmu_buf_hold(%p, %llu, %llu) = %u",
5539 os
, (long long)object
, (long long) offset
, error
);
5541 ASSERT(db
->db_offset
== offset
);
5542 ASSERT(db
->db_size
== blocksize
);
5543 ASSERT(ztest_pattern_match(db
->db_data
, db
->db_size
, pattern
) ||
5544 ztest_pattern_match(db
->db_data
, db
->db_size
, 0ULL));
5545 dmu_buf_will_fill(db
, tx
);
5546 ztest_pattern_set(db
->db_data
, db
->db_size
, pattern
);
5547 dmu_buf_rele(db
, FTAG
);
5551 txg_wait_synced(spa_get_dsl(spa
), txg
);
5554 * Find out what block we got.
5556 VERIFY0(dmu_buf_hold(os
, object
, 0, FTAG
, &db
,
5557 DMU_READ_NO_PREFETCH
));
5558 blk
= *((dmu_buf_impl_t
*)db
)->db_blkptr
;
5559 dmu_buf_rele(db
, FTAG
);
5562 * Damage the block. Dedup-ditto will save us when we read it later.
5564 psize
= BP_GET_PSIZE(&blk
);
5565 abd
= abd_alloc_linear(psize
, B_TRUE
);
5566 ztest_pattern_set(abd_to_buf(abd
), psize
, ~pattern
);
5568 (void) zio_wait(zio_rewrite(NULL
, spa
, 0, &blk
,
5569 abd
, psize
, NULL
, NULL
, ZIO_PRIORITY_SYNC_WRITE
,
5570 ZIO_FLAG_CANFAIL
| ZIO_FLAG_INDUCE_DAMAGE
, NULL
));
5574 (void) rw_unlock(&ztest_name_lock
);
5575 umem_free(od
, sizeof (ztest_od_t
));
5583 ztest_scrub(ztest_ds_t
*zd
, uint64_t id
)
5585 spa_t
*spa
= ztest_spa
;
5587 (void) spa_scan(spa
, POOL_SCAN_SCRUB
);
5588 (void) poll(NULL
, 0, 100); /* wait a moment, then force a restart */
5589 (void) spa_scan(spa
, POOL_SCAN_SCRUB
);
5593 * Change the guid for the pool.
5597 ztest_reguid(ztest_ds_t
*zd
, uint64_t id
)
5599 spa_t
*spa
= ztest_spa
;
5600 uint64_t orig
, load
;
5603 orig
= spa_guid(spa
);
5604 load
= spa_load_guid(spa
);
5606 (void) rw_wrlock(&ztest_name_lock
);
5607 error
= spa_change_guid(spa
);
5608 (void) rw_unlock(&ztest_name_lock
);
5613 if (ztest_opts
.zo_verbose
>= 4) {
5614 (void) printf("Changed guid old %llu -> %llu\n",
5615 (u_longlong_t
)orig
, (u_longlong_t
)spa_guid(spa
));
5618 VERIFY3U(orig
, !=, spa_guid(spa
));
5619 VERIFY3U(load
, ==, spa_load_guid(spa
));
5623 * Rename the pool to a different name and then rename it back.
5627 ztest_spa_rename(ztest_ds_t
*zd
, uint64_t id
)
5629 char *oldname
, *newname
;
5632 (void) rw_wrlock(&ztest_name_lock
);
5634 oldname
= ztest_opts
.zo_pool
;
5635 newname
= umem_alloc(strlen(oldname
) + 5, UMEM_NOFAIL
);
5636 (void) strcpy(newname
, oldname
);
5637 (void) strcat(newname
, "_tmp");
5642 VERIFY3U(0, ==, spa_rename(oldname
, newname
));
5645 * Try to open it under the old name, which shouldn't exist
5647 VERIFY3U(ENOENT
, ==, spa_open(oldname
, &spa
, FTAG
));
5650 * Open it under the new name and make sure it's still the same spa_t.
5652 VERIFY3U(0, ==, spa_open(newname
, &spa
, FTAG
));
5654 ASSERT(spa
== ztest_spa
);
5655 spa_close(spa
, FTAG
);
5658 * Rename it back to the original
5660 VERIFY3U(0, ==, spa_rename(newname
, oldname
));
5663 * Make sure it can still be opened
5665 VERIFY3U(0, ==, spa_open(oldname
, &spa
, FTAG
));
5667 ASSERT(spa
== ztest_spa
);
5668 spa_close(spa
, FTAG
);
5670 umem_free(newname
, strlen(newname
) + 1);
5672 (void) rw_unlock(&ztest_name_lock
);
5676 ztest_fletcher(ztest_ds_t
*zd
, uint64_t id
)
5678 hrtime_t end
= gethrtime() + NANOSEC
;
5680 while (gethrtime() <= end
) {
5681 int run_count
= 100;
5683 struct abd
*abd_data
, *abd_meta
;
5688 zio_cksum_t zc_ref_byteswap
;
5690 size
= ztest_random_blocksize();
5692 buf
= umem_alloc(size
, UMEM_NOFAIL
);
5693 abd_data
= abd_alloc(size
, B_FALSE
);
5694 abd_meta
= abd_alloc(size
, B_TRUE
);
5696 for (i
= 0, ptr
= buf
; i
< size
/ sizeof (*ptr
); i
++, ptr
++)
5697 *ptr
= ztest_random(UINT_MAX
);
5699 abd_copy_from_buf_off(abd_data
, buf
, 0, size
);
5700 abd_copy_from_buf_off(abd_meta
, buf
, 0, size
);
5702 VERIFY0(fletcher_4_impl_set("scalar"));
5703 fletcher_4_native(buf
, size
, NULL
, &zc_ref
);
5704 fletcher_4_byteswap(buf
, size
, NULL
, &zc_ref_byteswap
);
5706 VERIFY0(fletcher_4_impl_set("cycle"));
5707 while (run_count
-- > 0) {
5709 zio_cksum_t zc_byteswap
;
5711 fletcher_4_byteswap(buf
, size
, NULL
, &zc_byteswap
);
5712 fletcher_4_native(buf
, size
, NULL
, &zc
);
5714 VERIFY0(bcmp(&zc
, &zc_ref
, sizeof (zc
)));
5715 VERIFY0(bcmp(&zc_byteswap
, &zc_ref_byteswap
,
5716 sizeof (zc_byteswap
)));
5718 /* Test ABD - data */
5719 abd_fletcher_4_byteswap(abd_data
, size
, NULL
,
5721 abd_fletcher_4_native(abd_data
, size
, NULL
, &zc
);
5723 VERIFY0(bcmp(&zc
, &zc_ref
, sizeof (zc
)));
5724 VERIFY0(bcmp(&zc_byteswap
, &zc_ref_byteswap
,
5725 sizeof (zc_byteswap
)));
5727 /* Test ABD - metadata */
5728 abd_fletcher_4_byteswap(abd_meta
, size
, NULL
,
5730 abd_fletcher_4_native(abd_meta
, size
, NULL
, &zc
);
5732 VERIFY0(bcmp(&zc
, &zc_ref
, sizeof (zc
)));
5733 VERIFY0(bcmp(&zc_byteswap
, &zc_ref_byteswap
,
5734 sizeof (zc_byteswap
)));
5738 umem_free(buf
, size
);
5745 ztest_fletcher_incr(ztest_ds_t
*zd
, uint64_t id
)
5752 zio_cksum_t zc_ref_bswap
;
5754 hrtime_t end
= gethrtime() + NANOSEC
;
5756 while (gethrtime() <= end
) {
5757 int run_count
= 100;
5759 size
= ztest_random_blocksize();
5760 buf
= umem_alloc(size
, UMEM_NOFAIL
);
5762 for (i
= 0, ptr
= buf
; i
< size
/ sizeof (*ptr
); i
++, ptr
++)
5763 *ptr
= ztest_random(UINT_MAX
);
5765 VERIFY0(fletcher_4_impl_set("scalar"));
5766 fletcher_4_native(buf
, size
, NULL
, &zc_ref
);
5767 fletcher_4_byteswap(buf
, size
, NULL
, &zc_ref_bswap
);
5769 VERIFY0(fletcher_4_impl_set("cycle"));
5771 while (run_count
-- > 0) {
5773 zio_cksum_t zc_bswap
;
5776 ZIO_SET_CHECKSUM(&zc
, 0, 0, 0, 0);
5777 ZIO_SET_CHECKSUM(&zc_bswap
, 0, 0, 0, 0);
5779 while (pos
< size
) {
5780 size_t inc
= 64 * ztest_random(size
/ 67);
5781 /* sometimes add few bytes to test non-simd */
5782 if (ztest_random(100) < 10)
5783 inc
+= P2ALIGN(ztest_random(64),
5786 if (inc
> (size
- pos
))
5789 fletcher_4_incremental_native(buf
+ pos
, inc
,
5791 fletcher_4_incremental_byteswap(buf
+ pos
, inc
,
5797 VERIFY3U(pos
, ==, size
);
5799 VERIFY(ZIO_CHECKSUM_EQUAL(zc
, zc_ref
));
5800 VERIFY(ZIO_CHECKSUM_EQUAL(zc_bswap
, zc_ref_bswap
));
5803 * verify if incremental on the whole buffer is
5804 * equivalent to non-incremental version
5806 ZIO_SET_CHECKSUM(&zc
, 0, 0, 0, 0);
5807 ZIO_SET_CHECKSUM(&zc_bswap
, 0, 0, 0, 0);
5809 fletcher_4_incremental_native(buf
, size
, &zc
);
5810 fletcher_4_incremental_byteswap(buf
, size
, &zc_bswap
);
5812 VERIFY(ZIO_CHECKSUM_EQUAL(zc
, zc_ref
));
5813 VERIFY(ZIO_CHECKSUM_EQUAL(zc_bswap
, zc_ref_bswap
));
5816 umem_free(buf
, size
);
5821 ztest_check_path(char *path
)
5824 /* return true on success */
5825 return (!stat(path
, &s
));
5829 ztest_get_zdb_bin(char *bin
, int len
)
5833 * Try to use ZDB_PATH and in-tree zdb path. If not successful, just
5834 * let popen to search through PATH.
5836 if ((zdb_path
= getenv("ZDB_PATH"))) {
5837 strlcpy(bin
, zdb_path
, len
); /* In env */
5838 if (!ztest_check_path(bin
)) {
5839 ztest_dump_core
= 0;
5840 fatal(1, "invalid ZDB_PATH '%s'", bin
);
5845 VERIFY(realpath(getexecname(), bin
) != NULL
);
5846 if (strstr(bin
, "/ztest/")) {
5847 strstr(bin
, "/ztest/")[0] = '\0'; /* In-tree */
5848 strcat(bin
, "/zdb/zdb");
5849 if (ztest_check_path(bin
))
5856 * Verify pool integrity by running zdb.
5859 ztest_run_zdb(char *pool
)
5865 const int len
= MAXPATHLEN
+ MAXNAMELEN
+ 20;
5868 bin
= umem_alloc(len
, UMEM_NOFAIL
);
5869 zdb
= umem_alloc(len
, UMEM_NOFAIL
);
5870 zbuf
= umem_alloc(1024, UMEM_NOFAIL
);
5872 ztest_get_zdb_bin(bin
, len
);
5875 "%s -bcc%s%s -G -d -U %s %s",
5877 ztest_opts
.zo_verbose
>= 3 ? "s" : "",
5878 ztest_opts
.zo_verbose
>= 4 ? "v" : "",
5882 if (ztest_opts
.zo_verbose
>= 5)
5883 (void) printf("Executing %s\n", strstr(zdb
, "zdb "));
5885 fp
= popen(zdb
, "r");
5887 while (fgets(zbuf
, 1024, fp
) != NULL
)
5888 if (ztest_opts
.zo_verbose
>= 3)
5889 (void) printf("%s", zbuf
);
5891 status
= pclose(fp
);
5896 ztest_dump_core
= 0;
5897 if (WIFEXITED(status
))
5898 fatal(0, "'%s' exit code %d", zdb
, WEXITSTATUS(status
));
5900 fatal(0, "'%s' died with signal %d", zdb
, WTERMSIG(status
));
5902 umem_free(bin
, len
);
5903 umem_free(zdb
, len
);
5904 umem_free(zbuf
, 1024);
5908 ztest_walk_pool_directory(char *header
)
5912 if (ztest_opts
.zo_verbose
>= 6)
5913 (void) printf("%s\n", header
);
5915 mutex_enter(&spa_namespace_lock
);
5916 while ((spa
= spa_next(spa
)) != NULL
)
5917 if (ztest_opts
.zo_verbose
>= 6)
5918 (void) printf("\t%s\n", spa_name(spa
));
5919 mutex_exit(&spa_namespace_lock
);
5923 ztest_spa_import_export(char *oldname
, char *newname
)
5925 nvlist_t
*config
, *newconfig
;
5930 if (ztest_opts
.zo_verbose
>= 4) {
5931 (void) printf("import/export: old = %s, new = %s\n",
5936 * Clean up from previous runs.
5938 (void) spa_destroy(newname
);
5941 * Get the pool's configuration and guid.
5943 VERIFY3U(0, ==, spa_open(oldname
, &spa
, FTAG
));
5946 * Kick off a scrub to tickle scrub/export races.
5948 if (ztest_random(2) == 0)
5949 (void) spa_scan(spa
, POOL_SCAN_SCRUB
);
5951 pool_guid
= spa_guid(spa
);
5952 spa_close(spa
, FTAG
);
5954 ztest_walk_pool_directory("pools before export");
5959 VERIFY3U(0, ==, spa_export(oldname
, &config
, B_FALSE
, B_FALSE
));
5961 ztest_walk_pool_directory("pools after export");
5966 newconfig
= spa_tryimport(config
);
5967 ASSERT(newconfig
!= NULL
);
5968 nvlist_free(newconfig
);
5971 * Import it under the new name.
5973 error
= spa_import(newname
, config
, NULL
, 0);
5975 dump_nvlist(config
, 0);
5976 fatal(B_FALSE
, "couldn't import pool %s as %s: error %u",
5977 oldname
, newname
, error
);
5980 ztest_walk_pool_directory("pools after import");
5983 * Try to import it again -- should fail with EEXIST.
5985 VERIFY3U(EEXIST
, ==, spa_import(newname
, config
, NULL
, 0));
5988 * Try to import it under a different name -- should fail with EEXIST.
5990 VERIFY3U(EEXIST
, ==, spa_import(oldname
, config
, NULL
, 0));
5993 * Verify that the pool is no longer visible under the old name.
5995 VERIFY3U(ENOENT
, ==, spa_open(oldname
, &spa
, FTAG
));
5998 * Verify that we can open and close the pool using the new name.
6000 VERIFY3U(0, ==, spa_open(newname
, &spa
, FTAG
));
6001 ASSERT(pool_guid
== spa_guid(spa
));
6002 spa_close(spa
, FTAG
);
6004 nvlist_free(config
);
6008 ztest_resume(spa_t
*spa
)
6010 if (spa_suspended(spa
) && ztest_opts
.zo_verbose
>= 6)
6011 (void) printf("resuming from suspended state\n");
6012 spa_vdev_state_enter(spa
, SCL_NONE
);
6013 vdev_clear(spa
, NULL
);
6014 (void) spa_vdev_state_exit(spa
, NULL
, 0);
6015 (void) zio_resume(spa
);
6019 ztest_resume_thread(void *arg
)
6023 while (!ztest_exiting
) {
6024 if (spa_suspended(spa
))
6026 (void) poll(NULL
, 0, 100);
6029 * Periodically change the zfs_compressed_arc_enabled setting.
6031 if (ztest_random(10) == 0)
6032 zfs_compressed_arc_enabled
= ztest_random(2);
6035 * Periodically change the zfs_abd_scatter_enabled setting.
6037 if (ztest_random(10) == 0)
6038 zfs_abd_scatter_enabled
= ztest_random(2);
6050 ztest_deadman_alarm(int sig
)
6052 fatal(0, "failed to complete within %d seconds of deadline", GRACE
);
6057 ztest_execute(int test
, ztest_info_t
*zi
, uint64_t id
)
6059 ztest_ds_t
*zd
= &ztest_ds
[id
% ztest_opts
.zo_datasets
];
6060 ztest_shared_callstate_t
*zc
= ZTEST_GET_SHARED_CALLSTATE(test
);
6061 hrtime_t functime
= gethrtime();
6064 for (i
= 0; i
< zi
->zi_iters
; i
++)
6065 zi
->zi_func(zd
, id
);
6067 functime
= gethrtime() - functime
;
6069 atomic_add_64(&zc
->zc_count
, 1);
6070 atomic_add_64(&zc
->zc_time
, functime
);
6072 if (ztest_opts
.zo_verbose
>= 4)
6073 (void) printf("%6.2f sec in %s\n",
6074 (double)functime
/ NANOSEC
, zi
->zi_funcname
);
6078 ztest_thread(void *arg
)
6081 uint64_t id
= (uintptr_t)arg
;
6082 ztest_shared_t
*zs
= ztest_shared
;
6086 ztest_shared_callstate_t
*zc
;
6088 while ((now
= gethrtime()) < zs
->zs_thread_stop
) {
6090 * See if it's time to force a crash.
6092 if (now
> zs
->zs_thread_kill
)
6096 * If we're getting ENOSPC with some regularity, stop.
6098 if (zs
->zs_enospc_count
> 10)
6102 * Pick a random function to execute.
6104 rand
= ztest_random(ZTEST_FUNCS
);
6105 zi
= &ztest_info
[rand
];
6106 zc
= ZTEST_GET_SHARED_CALLSTATE(rand
);
6107 call_next
= zc
->zc_next
;
6109 if (now
>= call_next
&&
6110 atomic_cas_64(&zc
->zc_next
, call_next
, call_next
+
6111 ztest_random(2 * zi
->zi_interval
[0] + 1)) == call_next
) {
6112 ztest_execute(rand
, zi
, id
);
6122 ztest_dataset_name(char *dsname
, char *pool
, int d
)
6124 (void) snprintf(dsname
, ZFS_MAX_DATASET_NAME_LEN
, "%s/ds_%d", pool
, d
);
6128 ztest_dataset_destroy(int d
)
6130 char name
[ZFS_MAX_DATASET_NAME_LEN
];
6133 ztest_dataset_name(name
, ztest_opts
.zo_pool
, d
);
6135 if (ztest_opts
.zo_verbose
>= 3)
6136 (void) printf("Destroying %s to free up space\n", name
);
6139 * Cleanup any non-standard clones and snapshots. In general,
6140 * ztest thread t operates on dataset (t % zopt_datasets),
6141 * so there may be more than one thing to clean up.
6143 for (t
= d
; t
< ztest_opts
.zo_threads
;
6144 t
+= ztest_opts
.zo_datasets
)
6145 ztest_dsl_dataset_cleanup(name
, t
);
6147 (void) dmu_objset_find(name
, ztest_objset_destroy_cb
, NULL
,
6148 DS_FIND_SNAPSHOTS
| DS_FIND_CHILDREN
);
6152 ztest_dataset_dirobj_verify(ztest_ds_t
*zd
)
6154 uint64_t usedobjs
, dirobjs
, scratch
;
6157 * ZTEST_DIROBJ is the object directory for the entire dataset.
6158 * Therefore, the number of objects in use should equal the
6159 * number of ZTEST_DIROBJ entries, +1 for ZTEST_DIROBJ itself.
6160 * If not, we have an object leak.
6162 * Note that we can only check this in ztest_dataset_open(),
6163 * when the open-context and syncing-context values agree.
6164 * That's because zap_count() returns the open-context value,
6165 * while dmu_objset_space() returns the rootbp fill count.
6167 VERIFY3U(0, ==, zap_count(zd
->zd_os
, ZTEST_DIROBJ
, &dirobjs
));
6168 dmu_objset_space(zd
->zd_os
, &scratch
, &scratch
, &usedobjs
, &scratch
);
6169 ASSERT3U(dirobjs
+ 1, ==, usedobjs
);
6173 ztest_dataset_open(int d
)
6175 ztest_ds_t
*zd
= &ztest_ds
[d
];
6176 uint64_t committed_seq
= ZTEST_GET_SHARED_DS(d
)->zd_seq
;
6179 char name
[ZFS_MAX_DATASET_NAME_LEN
];
6182 ztest_dataset_name(name
, ztest_opts
.zo_pool
, d
);
6184 (void) rw_rdlock(&ztest_name_lock
);
6186 error
= ztest_dataset_create(name
);
6187 if (error
== ENOSPC
) {
6188 (void) rw_unlock(&ztest_name_lock
);
6189 ztest_record_enospc(FTAG
);
6192 ASSERT(error
== 0 || error
== EEXIST
);
6194 VERIFY0(dmu_objset_own(name
, DMU_OST_OTHER
, B_FALSE
, zd
, &os
));
6195 (void) rw_unlock(&ztest_name_lock
);
6197 ztest_zd_init(zd
, ZTEST_GET_SHARED_DS(d
), os
);
6199 zilog
= zd
->zd_zilog
;
6201 if (zilog
->zl_header
->zh_claim_lr_seq
!= 0 &&
6202 zilog
->zl_header
->zh_claim_lr_seq
< committed_seq
)
6203 fatal(0, "missing log records: claimed %llu < committed %llu",
6204 zilog
->zl_header
->zh_claim_lr_seq
, committed_seq
);
6206 ztest_dataset_dirobj_verify(zd
);
6208 zil_replay(os
, zd
, ztest_replay_vector
);
6210 ztest_dataset_dirobj_verify(zd
);
6212 if (ztest_opts
.zo_verbose
>= 6)
6213 (void) printf("%s replay %llu blocks, %llu records, seq %llu\n",
6215 (u_longlong_t
)zilog
->zl_parse_blk_count
,
6216 (u_longlong_t
)zilog
->zl_parse_lr_count
,
6217 (u_longlong_t
)zilog
->zl_replaying_seq
);
6219 zilog
= zil_open(os
, ztest_get_data
);
6221 if (zilog
->zl_replaying_seq
!= 0 &&
6222 zilog
->zl_replaying_seq
< committed_seq
)
6223 fatal(0, "missing log records: replayed %llu < committed %llu",
6224 zilog
->zl_replaying_seq
, committed_seq
);
6230 ztest_dataset_close(int d
)
6232 ztest_ds_t
*zd
= &ztest_ds
[d
];
6234 zil_close(zd
->zd_zilog
);
6235 dmu_objset_disown(zd
->zd_os
, zd
);
6241 * Kick off threads to run tests on all datasets in parallel.
6244 ztest_run(ztest_shared_t
*zs
)
6249 kthread_t
*resume_thread
;
6254 ztest_exiting
= B_FALSE
;
6257 * Initialize parent/child shared state.
6259 mutex_init(&ztest_vdev_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
6260 VERIFY(rwlock_init(&ztest_name_lock
, USYNC_THREAD
, NULL
) == 0);
6262 zs
->zs_thread_start
= gethrtime();
6263 zs
->zs_thread_stop
=
6264 zs
->zs_thread_start
+ ztest_opts
.zo_passtime
* NANOSEC
;
6265 zs
->zs_thread_stop
= MIN(zs
->zs_thread_stop
, zs
->zs_proc_stop
);
6266 zs
->zs_thread_kill
= zs
->zs_thread_stop
;
6267 if (ztest_random(100) < ztest_opts
.zo_killrate
) {
6268 zs
->zs_thread_kill
-=
6269 ztest_random(ztest_opts
.zo_passtime
* NANOSEC
);
6272 mutex_init(&zcl
.zcl_callbacks_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
6274 list_create(&zcl
.zcl_callbacks
, sizeof (ztest_cb_data_t
),
6275 offsetof(ztest_cb_data_t
, zcd_node
));
6280 kernel_init(FREAD
| FWRITE
);
6281 VERIFY0(spa_open(ztest_opts
.zo_pool
, &spa
, FTAG
));
6282 spa
->spa_debug
= B_TRUE
;
6283 metaslab_preload_limit
= ztest_random(20) + 1;
6286 dmu_objset_stats_t dds
;
6287 VERIFY0(dmu_objset_own(ztest_opts
.zo_pool
,
6288 DMU_OST_ANY
, B_TRUE
, FTAG
, &os
));
6289 dsl_pool_config_enter(dmu_objset_pool(os
), FTAG
);
6290 dmu_objset_fast_stat(os
, &dds
);
6291 dsl_pool_config_exit(dmu_objset_pool(os
), FTAG
);
6292 zs
->zs_guid
= dds
.dds_guid
;
6293 dmu_objset_disown(os
, FTAG
);
6295 spa
->spa_dedup_ditto
= 2 * ZIO_DEDUPDITTO_MIN
;
6298 * We don't expect the pool to suspend unless maxfaults == 0,
6299 * in which case ztest_fault_inject() temporarily takes away
6300 * the only valid replica.
6302 if (MAXFAULTS() == 0)
6303 spa
->spa_failmode
= ZIO_FAILURE_MODE_WAIT
;
6305 spa
->spa_failmode
= ZIO_FAILURE_MODE_PANIC
;
6308 * Create a thread to periodically resume suspended I/O.
6310 VERIFY3P((resume_thread
= zk_thread_create(NULL
, 0,
6311 (thread_func_t
)ztest_resume_thread
, spa
, 0, NULL
, TS_RUN
, 0,
6312 PTHREAD_CREATE_JOINABLE
)), !=, NULL
);
6316 * Set a deadman alarm to abort() if we hang.
6318 signal(SIGALRM
, ztest_deadman_alarm
);
6319 alarm((zs
->zs_thread_stop
- zs
->zs_thread_start
) / NANOSEC
+ GRACE
);
6323 * Verify that we can safely inquire about about any object,
6324 * whether it's allocated or not. To make it interesting,
6325 * we probe a 5-wide window around each power of two.
6326 * This hits all edge cases, including zero and the max.
6328 for (t
= 0; t
< 64; t
++) {
6329 for (d
= -5; d
<= 5; d
++) {
6330 error
= dmu_object_info(spa
->spa_meta_objset
,
6331 (1ULL << t
) + d
, NULL
);
6332 ASSERT(error
== 0 || error
== ENOENT
||
6338 * If we got any ENOSPC errors on the previous run, destroy something.
6340 if (zs
->zs_enospc_count
!= 0) {
6341 int d
= ztest_random(ztest_opts
.zo_datasets
);
6342 ztest_dataset_destroy(d
);
6344 zs
->zs_enospc_count
= 0;
6346 tid
= umem_zalloc(ztest_opts
.zo_threads
* sizeof (kt_did_t
),
6349 if (ztest_opts
.zo_verbose
>= 4)
6350 (void) printf("starting main threads...\n");
6353 * Kick off all the tests that run in parallel.
6355 for (t
= 0; t
< ztest_opts
.zo_threads
; t
++) {
6358 if (t
< ztest_opts
.zo_datasets
&&
6359 ztest_dataset_open(t
) != 0) {
6361 ztest_opts
.zo_threads
* sizeof (kt_did_t
));
6365 VERIFY3P(thread
= zk_thread_create(NULL
, 0,
6366 (thread_func_t
)ztest_thread
,
6367 (void *)(uintptr_t)t
, 0, NULL
, TS_RUN
, 0,
6368 PTHREAD_CREATE_JOINABLE
), !=, NULL
);
6369 tid
[t
] = thread
->t_tid
;
6373 * Wait for all of the tests to complete. We go in reverse order
6374 * so we don't close datasets while threads are still using them.
6376 for (t
= ztest_opts
.zo_threads
- 1; t
>= 0; t
--) {
6377 thread_join(tid
[t
]);
6378 if (t
< ztest_opts
.zo_datasets
)
6379 ztest_dataset_close(t
);
6382 txg_wait_synced(spa_get_dsl(spa
), 0);
6384 zs
->zs_alloc
= metaslab_class_get_alloc(spa_normal_class(spa
));
6385 zs
->zs_space
= metaslab_class_get_space(spa_normal_class(spa
));
6387 umem_free(tid
, ztest_opts
.zo_threads
* sizeof (kt_did_t
));
6389 /* Kill the resume thread */
6390 ztest_exiting
= B_TRUE
;
6391 thread_join(resume_thread
->t_tid
);
6395 * Right before closing the pool, kick off a bunch of async I/O;
6396 * spa_close() should wait for it to complete.
6398 for (object
= 1; object
< 50; object
++) {
6399 dmu_prefetch(spa
->spa_meta_objset
, object
, 0, 0, 1ULL << 20,
6400 ZIO_PRIORITY_SYNC_READ
);
6403 /* Verify that at least one commit cb was called in a timely fashion */
6404 if (zc_cb_counter
>= ZTEST_COMMIT_CB_MIN_REG
)
6405 VERIFY0(zc_min_txg_delay
);
6407 spa_close(spa
, FTAG
);
6410 * Verify that we can loop over all pools.
6412 mutex_enter(&spa_namespace_lock
);
6413 for (spa
= spa_next(NULL
); spa
!= NULL
; spa
= spa_next(spa
))
6414 if (ztest_opts
.zo_verbose
> 3)
6415 (void) printf("spa_next: found %s\n", spa_name(spa
));
6416 mutex_exit(&spa_namespace_lock
);
6419 * Verify that we can export the pool and reimport it under a
6422 if (ztest_random(2) == 0) {
6423 char name
[ZFS_MAX_DATASET_NAME_LEN
];
6424 (void) snprintf(name
, sizeof (name
), "%s_import",
6425 ztest_opts
.zo_pool
);
6426 ztest_spa_import_export(ztest_opts
.zo_pool
, name
);
6427 ztest_spa_import_export(name
, ztest_opts
.zo_pool
);
6432 list_destroy(&zcl
.zcl_callbacks
);
6433 mutex_destroy(&zcl
.zcl_callbacks_lock
);
6434 (void) rwlock_destroy(&ztest_name_lock
);
6435 mutex_destroy(&ztest_vdev_lock
);
6441 ztest_ds_t
*zd
= &ztest_ds
[0];
6445 if (ztest_opts
.zo_verbose
>= 3)
6446 (void) printf("testing spa_freeze()...\n");
6448 kernel_init(FREAD
| FWRITE
);
6449 VERIFY3U(0, ==, spa_open(ztest_opts
.zo_pool
, &spa
, FTAG
));
6450 VERIFY3U(0, ==, ztest_dataset_open(0));
6451 spa
->spa_debug
= B_TRUE
;
6455 * Force the first log block to be transactionally allocated.
6456 * We have to do this before we freeze the pool -- otherwise
6457 * the log chain won't be anchored.
6459 while (BP_IS_HOLE(&zd
->zd_zilog
->zl_header
->zh_log
)) {
6460 ztest_dmu_object_alloc_free(zd
, 0);
6461 zil_commit(zd
->zd_zilog
, 0);
6464 txg_wait_synced(spa_get_dsl(spa
), 0);
6467 * Freeze the pool. This stops spa_sync() from doing anything,
6468 * so that the only way to record changes from now on is the ZIL.
6473 * Because it is hard to predict how much space a write will actually
6474 * require beforehand, we leave ourselves some fudge space to write over
6477 uint64_t capacity
= metaslab_class_get_space(spa_normal_class(spa
)) / 2;
6480 * Run tests that generate log records but don't alter the pool config
6481 * or depend on DSL sync tasks (snapshots, objset create/destroy, etc).
6482 * We do a txg_wait_synced() after each iteration to force the txg
6483 * to increase well beyond the last synced value in the uberblock.
6484 * The ZIL should be OK with that.
6486 * Run a random number of times less than zo_maxloops and ensure we do
6487 * not run out of space on the pool.
6489 while (ztest_random(10) != 0 &&
6490 numloops
++ < ztest_opts
.zo_maxloops
&&
6491 metaslab_class_get_alloc(spa_normal_class(spa
)) < capacity
) {
6493 ztest_od_init(&od
, 0, FTAG
, 0, DMU_OT_UINT64_OTHER
, 0, 0, 0);
6494 VERIFY0(ztest_object_init(zd
, &od
, sizeof (od
), B_FALSE
));
6495 ztest_io(zd
, od
.od_object
,
6496 ztest_random(ZTEST_RANGE_LOCKS
) << SPA_MAXBLOCKSHIFT
);
6497 txg_wait_synced(spa_get_dsl(spa
), 0);
6501 * Commit all of the changes we just generated.
6503 zil_commit(zd
->zd_zilog
, 0);
6504 txg_wait_synced(spa_get_dsl(spa
), 0);
6507 * Close our dataset and close the pool.
6509 ztest_dataset_close(0);
6510 spa_close(spa
, FTAG
);
6514 * Open and close the pool and dataset to induce log replay.
6516 kernel_init(FREAD
| FWRITE
);
6517 VERIFY3U(0, ==, spa_open(ztest_opts
.zo_pool
, &spa
, FTAG
));
6518 ASSERT(spa_freeze_txg(spa
) == UINT64_MAX
);
6519 VERIFY3U(0, ==, ztest_dataset_open(0));
6520 ztest_dataset_close(0);
6522 spa
->spa_debug
= B_TRUE
;
6524 txg_wait_synced(spa_get_dsl(spa
), 0);
6525 ztest_reguid(NULL
, 0);
6527 spa_close(spa
, FTAG
);
6532 print_time(hrtime_t t
, char *timebuf
)
6534 hrtime_t s
= t
/ NANOSEC
;
6535 hrtime_t m
= s
/ 60;
6536 hrtime_t h
= m
/ 60;
6537 hrtime_t d
= h
/ 24;
6546 (void) sprintf(timebuf
,
6547 "%llud%02lluh%02llum%02llus", d
, h
, m
, s
);
6549 (void) sprintf(timebuf
, "%lluh%02llum%02llus", h
, m
, s
);
6551 (void) sprintf(timebuf
, "%llum%02llus", m
, s
);
6553 (void) sprintf(timebuf
, "%llus", s
);
6557 make_random_props(void)
6561 VERIFY(nvlist_alloc(&props
, NV_UNIQUE_NAME
, 0) == 0);
6562 if (ztest_random(2) == 0)
6564 VERIFY(nvlist_add_uint64(props
, "autoreplace", 1) == 0);
6570 * Create a storage pool with the given name and initial vdev size.
6571 * Then test spa_freeze() functionality.
6574 ztest_init(ztest_shared_t
*zs
)
6577 nvlist_t
*nvroot
, *props
;
6580 mutex_init(&ztest_vdev_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
6581 VERIFY(rwlock_init(&ztest_name_lock
, USYNC_THREAD
, NULL
) == 0);
6583 kernel_init(FREAD
| FWRITE
);
6586 * Create the storage pool.
6588 (void) spa_destroy(ztest_opts
.zo_pool
);
6589 ztest_shared
->zs_vdev_next_leaf
= 0;
6591 zs
->zs_mirrors
= ztest_opts
.zo_mirrors
;
6592 nvroot
= make_vdev_root(NULL
, NULL
, NULL
, ztest_opts
.zo_vdev_size
, 0,
6593 0, ztest_opts
.zo_raidz
, zs
->zs_mirrors
, 1);
6594 props
= make_random_props();
6595 for (i
= 0; i
< SPA_FEATURES
; i
++) {
6597 VERIFY3S(-1, !=, asprintf(&buf
, "feature@%s",
6598 spa_feature_table
[i
].fi_uname
));
6599 VERIFY3U(0, ==, nvlist_add_uint64(props
, buf
, 0));
6602 VERIFY3U(0, ==, spa_create(ztest_opts
.zo_pool
, nvroot
, props
, NULL
));
6603 nvlist_free(nvroot
);
6606 VERIFY3U(0, ==, spa_open(ztest_opts
.zo_pool
, &spa
, FTAG
));
6607 zs
->zs_metaslab_sz
=
6608 1ULL << spa
->spa_root_vdev
->vdev_child
[0]->vdev_ms_shift
;
6609 spa_close(spa
, FTAG
);
6613 ztest_run_zdb(ztest_opts
.zo_pool
);
6617 ztest_run_zdb(ztest_opts
.zo_pool
);
6619 (void) rwlock_destroy(&ztest_name_lock
);
6620 mutex_destroy(&ztest_vdev_lock
);
6626 static char ztest_name_data
[] = "/tmp/ztest.data.XXXXXX";
6628 ztest_fd_data
= mkstemp(ztest_name_data
);
6629 ASSERT3S(ztest_fd_data
, >=, 0);
6630 (void) unlink(ztest_name_data
);
6634 shared_data_size(ztest_shared_hdr_t
*hdr
)
6638 size
= hdr
->zh_hdr_size
;
6639 size
+= hdr
->zh_opts_size
;
6640 size
+= hdr
->zh_size
;
6641 size
+= hdr
->zh_stats_size
* hdr
->zh_stats_count
;
6642 size
+= hdr
->zh_ds_size
* hdr
->zh_ds_count
;
6651 ztest_shared_hdr_t
*hdr
;
6653 hdr
= (void *)mmap(0, P2ROUNDUP(sizeof (*hdr
), getpagesize()),
6654 PROT_READ
| PROT_WRITE
, MAP_SHARED
, ztest_fd_data
, 0);
6655 ASSERT(hdr
!= MAP_FAILED
);
6657 VERIFY3U(0, ==, ftruncate(ztest_fd_data
, sizeof (ztest_shared_hdr_t
)));
6659 hdr
->zh_hdr_size
= sizeof (ztest_shared_hdr_t
);
6660 hdr
->zh_opts_size
= sizeof (ztest_shared_opts_t
);
6661 hdr
->zh_size
= sizeof (ztest_shared_t
);
6662 hdr
->zh_stats_size
= sizeof (ztest_shared_callstate_t
);
6663 hdr
->zh_stats_count
= ZTEST_FUNCS
;
6664 hdr
->zh_ds_size
= sizeof (ztest_shared_ds_t
);
6665 hdr
->zh_ds_count
= ztest_opts
.zo_datasets
;
6667 size
= shared_data_size(hdr
);
6668 VERIFY3U(0, ==, ftruncate(ztest_fd_data
, size
));
6670 (void) munmap((caddr_t
)hdr
, P2ROUNDUP(sizeof (*hdr
), getpagesize()));
6677 ztest_shared_hdr_t
*hdr
;
6680 hdr
= (void *)mmap(0, P2ROUNDUP(sizeof (*hdr
), getpagesize()),
6681 PROT_READ
, MAP_SHARED
, ztest_fd_data
, 0);
6682 ASSERT(hdr
!= MAP_FAILED
);
6684 size
= shared_data_size(hdr
);
6686 (void) munmap((caddr_t
)hdr
, P2ROUNDUP(sizeof (*hdr
), getpagesize()));
6687 hdr
= ztest_shared_hdr
= (void *)mmap(0, P2ROUNDUP(size
, getpagesize()),
6688 PROT_READ
| PROT_WRITE
, MAP_SHARED
, ztest_fd_data
, 0);
6689 ASSERT(hdr
!= MAP_FAILED
);
6690 buf
= (uint8_t *)hdr
;
6692 offset
= hdr
->zh_hdr_size
;
6693 ztest_shared_opts
= (void *)&buf
[offset
];
6694 offset
+= hdr
->zh_opts_size
;
6695 ztest_shared
= (void *)&buf
[offset
];
6696 offset
+= hdr
->zh_size
;
6697 ztest_shared_callstate
= (void *)&buf
[offset
];
6698 offset
+= hdr
->zh_stats_size
* hdr
->zh_stats_count
;
6699 ztest_shared_ds
= (void *)&buf
[offset
];
6703 exec_child(char *cmd
, char *libpath
, boolean_t ignorekill
, int *statusp
)
6707 char *cmdbuf
= NULL
;
6712 cmdbuf
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
6713 (void) strlcpy(cmdbuf
, getexecname(), MAXPATHLEN
);
6718 fatal(1, "fork failed");
6720 if (pid
== 0) { /* child */
6721 char *emptyargv
[2] = { cmd
, NULL
};
6722 char fd_data_str
[12];
6724 struct rlimit rl
= { 1024, 1024 };
6725 (void) setrlimit(RLIMIT_NOFILE
, &rl
);
6727 (void) close(ztest_fd_rand
);
6728 VERIFY(11 >= snprintf(fd_data_str
, 12, "%d", ztest_fd_data
));
6729 VERIFY(0 == setenv("ZTEST_FD_DATA", fd_data_str
, 1));
6731 (void) enable_extended_FILE_stdio(-1, -1);
6732 if (libpath
!= NULL
)
6733 VERIFY(0 == setenv("LD_LIBRARY_PATH", libpath
, 1));
6734 (void) execv(cmd
, emptyargv
);
6735 ztest_dump_core
= B_FALSE
;
6736 fatal(B_TRUE
, "exec failed: %s", cmd
);
6739 if (cmdbuf
!= NULL
) {
6740 umem_free(cmdbuf
, MAXPATHLEN
);
6744 while (waitpid(pid
, &status
, 0) != pid
)
6746 if (statusp
!= NULL
)
6749 if (WIFEXITED(status
)) {
6750 if (WEXITSTATUS(status
) != 0) {
6751 (void) fprintf(stderr
, "child exited with code %d\n",
6752 WEXITSTATUS(status
));
6756 } else if (WIFSIGNALED(status
)) {
6757 if (!ignorekill
|| WTERMSIG(status
) != SIGKILL
) {
6758 (void) fprintf(stderr
, "child died with signal %d\n",
6764 (void) fprintf(stderr
, "something strange happened to child\n");
6771 ztest_run_init(void)
6775 ztest_shared_t
*zs
= ztest_shared
;
6777 ASSERT(ztest_opts
.zo_init
!= 0);
6780 * Blow away any existing copy of zpool.cache
6782 (void) remove(spa_config_path
);
6785 * Create and initialize our storage pool.
6787 for (i
= 1; i
<= ztest_opts
.zo_init
; i
++) {
6788 bzero(zs
, sizeof (ztest_shared_t
));
6789 if (ztest_opts
.zo_verbose
>= 3 &&
6790 ztest_opts
.zo_init
!= 1) {
6791 (void) printf("ztest_init(), pass %d\n", i
);
6798 main(int argc
, char **argv
)
6806 ztest_shared_callstate_t
*zc
;
6813 char *fd_data_str
= getenv("ZTEST_FD_DATA");
6814 struct sigaction action
;
6816 (void) setvbuf(stdout
, NULL
, _IOLBF
, 0);
6818 dprintf_setup(&argc
, argv
);
6820 action
.sa_handler
= sig_handler
;
6821 sigemptyset(&action
.sa_mask
);
6822 action
.sa_flags
= 0;
6824 if (sigaction(SIGSEGV
, &action
, NULL
) < 0) {
6825 (void) fprintf(stderr
, "ztest: cannot catch SIGSEGV: %s.\n",
6830 if (sigaction(SIGABRT
, &action
, NULL
) < 0) {
6831 (void) fprintf(stderr
, "ztest: cannot catch SIGABRT: %s.\n",
6836 ztest_fd_rand
= open("/dev/urandom", O_RDONLY
);
6837 ASSERT3S(ztest_fd_rand
, >=, 0);
6840 process_options(argc
, argv
);
6845 bcopy(&ztest_opts
, ztest_shared_opts
,
6846 sizeof (*ztest_shared_opts
));
6848 ztest_fd_data
= atoi(fd_data_str
);
6850 bcopy(ztest_shared_opts
, &ztest_opts
, sizeof (ztest_opts
));
6852 ASSERT3U(ztest_opts
.zo_datasets
, ==, ztest_shared_hdr
->zh_ds_count
);
6854 /* Override location of zpool.cache */
6855 VERIFY(asprintf((char **)&spa_config_path
, "%s/zpool.cache",
6856 ztest_opts
.zo_dir
) != -1);
6858 ztest_ds
= umem_alloc(ztest_opts
.zo_datasets
* sizeof (ztest_ds_t
),
6863 metaslab_gang_bang
= ztest_opts
.zo_metaslab_gang_bang
;
6864 metaslab_df_alloc_threshold
=
6865 zs
->zs_metaslab_df_alloc_threshold
;
6874 hasalt
= (strlen(ztest_opts
.zo_alt_ztest
) != 0);
6876 if (ztest_opts
.zo_verbose
>= 1) {
6877 (void) printf("%llu vdevs, %d datasets, %d threads,"
6878 " %llu seconds...\n",
6879 (u_longlong_t
)ztest_opts
.zo_vdevs
,
6880 ztest_opts
.zo_datasets
,
6881 ztest_opts
.zo_threads
,
6882 (u_longlong_t
)ztest_opts
.zo_time
);
6885 cmd
= umem_alloc(MAXNAMELEN
, UMEM_NOFAIL
);
6886 (void) strlcpy(cmd
, getexecname(), MAXNAMELEN
);
6888 zs
->zs_do_init
= B_TRUE
;
6889 if (strlen(ztest_opts
.zo_alt_ztest
) != 0) {
6890 if (ztest_opts
.zo_verbose
>= 1) {
6891 (void) printf("Executing older ztest for "
6892 "initialization: %s\n", ztest_opts
.zo_alt_ztest
);
6894 VERIFY(!exec_child(ztest_opts
.zo_alt_ztest
,
6895 ztest_opts
.zo_alt_libpath
, B_FALSE
, NULL
));
6897 VERIFY(!exec_child(NULL
, NULL
, B_FALSE
, NULL
));
6899 zs
->zs_do_init
= B_FALSE
;
6901 zs
->zs_proc_start
= gethrtime();
6902 zs
->zs_proc_stop
= zs
->zs_proc_start
+ ztest_opts
.zo_time
* NANOSEC
;
6904 for (f
= 0; f
< ZTEST_FUNCS
; f
++) {
6905 zi
= &ztest_info
[f
];
6906 zc
= ZTEST_GET_SHARED_CALLSTATE(f
);
6907 if (zs
->zs_proc_start
+ zi
->zi_interval
[0] > zs
->zs_proc_stop
)
6908 zc
->zc_next
= UINT64_MAX
;
6910 zc
->zc_next
= zs
->zs_proc_start
+
6911 ztest_random(2 * zi
->zi_interval
[0] + 1);
6915 * Run the tests in a loop. These tests include fault injection
6916 * to verify that self-healing data works, and forced crashes
6917 * to verify that we never lose on-disk consistency.
6919 while (gethrtime() < zs
->zs_proc_stop
) {
6924 * Initialize the workload counters for each function.
6926 for (f
= 0; f
< ZTEST_FUNCS
; f
++) {
6927 zc
= ZTEST_GET_SHARED_CALLSTATE(f
);
6932 /* Set the allocation switch size */
6933 zs
->zs_metaslab_df_alloc_threshold
=
6934 ztest_random(zs
->zs_metaslab_sz
/ 4) + 1;
6936 if (!hasalt
|| ztest_random(2) == 0) {
6937 if (hasalt
&& ztest_opts
.zo_verbose
>= 1) {
6938 (void) printf("Executing newer ztest: %s\n",
6942 killed
= exec_child(cmd
, NULL
, B_TRUE
, &status
);
6944 if (hasalt
&& ztest_opts
.zo_verbose
>= 1) {
6945 (void) printf("Executing older ztest: %s\n",
6946 ztest_opts
.zo_alt_ztest
);
6949 killed
= exec_child(ztest_opts
.zo_alt_ztest
,
6950 ztest_opts
.zo_alt_libpath
, B_TRUE
, &status
);
6957 if (ztest_opts
.zo_verbose
>= 1) {
6958 hrtime_t now
= gethrtime();
6960 now
= MIN(now
, zs
->zs_proc_stop
);
6961 print_time(zs
->zs_proc_stop
- now
, timebuf
);
6962 nicenum(zs
->zs_space
, numbuf
);
6964 (void) printf("Pass %3d, %8s, %3llu ENOSPC, "
6965 "%4.1f%% of %5s used, %3.0f%% done, %8s to go\n",
6967 WIFEXITED(status
) ? "Complete" : "SIGKILL",
6968 (u_longlong_t
)zs
->zs_enospc_count
,
6969 100.0 * zs
->zs_alloc
/ zs
->zs_space
,
6971 100.0 * (now
- zs
->zs_proc_start
) /
6972 (ztest_opts
.zo_time
* NANOSEC
), timebuf
);
6975 if (ztest_opts
.zo_verbose
>= 2) {
6976 (void) printf("\nWorkload summary:\n\n");
6977 (void) printf("%7s %9s %s\n",
6978 "Calls", "Time", "Function");
6979 (void) printf("%7s %9s %s\n",
6980 "-----", "----", "--------");
6981 for (f
= 0; f
< ZTEST_FUNCS
; f
++) {
6982 zi
= &ztest_info
[f
];
6983 zc
= ZTEST_GET_SHARED_CALLSTATE(f
);
6984 print_time(zc
->zc_time
, timebuf
);
6985 (void) printf("%7llu %9s %s\n",
6986 (u_longlong_t
)zc
->zc_count
, timebuf
,
6989 (void) printf("\n");
6993 * It's possible that we killed a child during a rename test,
6994 * in which case we'll have a 'ztest_tmp' pool lying around
6995 * instead of 'ztest'. Do a blind rename in case this happened.
6998 if (spa_open(ztest_opts
.zo_pool
, &spa
, FTAG
) == 0) {
6999 spa_close(spa
, FTAG
);
7001 char tmpname
[ZFS_MAX_DATASET_NAME_LEN
];
7003 kernel_init(FREAD
| FWRITE
);
7004 (void) snprintf(tmpname
, sizeof (tmpname
), "%s_tmp",
7005 ztest_opts
.zo_pool
);
7006 (void) spa_rename(tmpname
, ztest_opts
.zo_pool
);
7010 ztest_run_zdb(ztest_opts
.zo_pool
);
7013 if (ztest_opts
.zo_verbose
>= 1) {
7015 (void) printf("%d runs of older ztest: %s\n", older
,
7016 ztest_opts
.zo_alt_ztest
);
7017 (void) printf("%d runs of newer ztest: %s\n", newer
,
7020 (void) printf("%d killed, %d completed, %.0f%% kill rate\n",
7021 kills
, iters
- kills
, (100.0 * kills
) / MAX(1, iters
));
7024 umem_free(cmd
, MAXNAMELEN
);