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[-h] (print help)\n"
635 (u_longlong_t
)zo
->zo_vdevs
, /* -v */
636 nice_vdev_size
, /* -s */
637 zo
->zo_ashift
, /* -a */
638 zo
->zo_mirrors
, /* -m */
639 zo
->zo_raidz
, /* -r */
640 zo
->zo_raidz_parity
, /* -R */
641 zo
->zo_datasets
, /* -d */
642 zo
->zo_threads
, /* -t */
643 nice_gang_bang
, /* -g */
644 zo
->zo_init
, /* -i */
645 (u_longlong_t
)zo
->zo_killrate
, /* -k */
646 zo
->zo_pool
, /* -p */
648 (u_longlong_t
)zo
->zo_time
, /* -T */
649 (u_longlong_t
)zo
->zo_maxloops
, /* -F */
650 (u_longlong_t
)zo
->zo_passtime
);
651 exit(requested
? 0 : 1);
655 process_options(int argc
, char **argv
)
658 ztest_shared_opts_t
*zo
= &ztest_opts
;
662 char altdir
[MAXNAMELEN
] = { 0 };
664 bcopy(&ztest_opts_defaults
, zo
, sizeof (*zo
));
666 while ((opt
= getopt(argc
, argv
,
667 "v:s:a:m:r:R:d:t:g:i:k:p:f:VET:P:hF:B:")) != EOF
) {
684 value
= nicenumtoull(optarg
);
688 zo
->zo_vdevs
= value
;
691 zo
->zo_vdev_size
= MAX(SPA_MINDEVSIZE
, value
);
694 zo
->zo_ashift
= value
;
697 zo
->zo_mirrors
= value
;
700 zo
->zo_raidz
= MAX(1, value
);
703 zo
->zo_raidz_parity
= MIN(MAX(value
, 1), 3);
706 zo
->zo_datasets
= MAX(1, value
);
709 zo
->zo_threads
= MAX(1, value
);
712 zo
->zo_metaslab_gang_bang
= MAX(SPA_MINBLOCKSIZE
<< 1,
719 zo
->zo_killrate
= value
;
722 (void) strlcpy(zo
->zo_pool
, optarg
,
723 sizeof (zo
->zo_pool
));
726 path
= realpath(optarg
, NULL
);
728 (void) fprintf(stderr
, "error: %s: %s\n",
729 optarg
, strerror(errno
));
732 (void) strlcpy(zo
->zo_dir
, path
,
733 sizeof (zo
->zo_dir
));
747 zo
->zo_passtime
= MAX(1, value
);
750 zo
->zo_maxloops
= MAX(1, value
);
753 (void) strlcpy(altdir
, optarg
, sizeof (altdir
));
765 zo
->zo_raidz_parity
= MIN(zo
->zo_raidz_parity
, zo
->zo_raidz
- 1);
768 (zo
->zo_vdevs
> 0 ? zo
->zo_time
* NANOSEC
/ zo
->zo_vdevs
:
771 if (strlen(altdir
) > 0) {
779 cmd
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
780 realaltdir
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
782 VERIFY(NULL
!= realpath(getexecname(), cmd
));
783 if (0 != access(altdir
, F_OK
)) {
784 ztest_dump_core
= B_FALSE
;
785 fatal(B_TRUE
, "invalid alternate ztest path: %s",
788 VERIFY(NULL
!= realpath(altdir
, realaltdir
));
791 * 'cmd' should be of the form "<anything>/usr/bin/<isa>/ztest".
792 * We want to extract <isa> to determine if we should use
793 * 32 or 64 bit binaries.
795 bin
= strstr(cmd
, "/usr/bin/");
796 ztest
= strstr(bin
, "/ztest");
798 isalen
= ztest
- isa
;
799 (void) snprintf(zo
->zo_alt_ztest
, sizeof (zo
->zo_alt_ztest
),
800 "%s/usr/bin/%.*s/ztest", realaltdir
, isalen
, isa
);
801 (void) snprintf(zo
->zo_alt_libpath
, sizeof (zo
->zo_alt_libpath
),
802 "%s/usr/lib/%.*s", realaltdir
, isalen
, isa
);
804 if (0 != access(zo
->zo_alt_ztest
, X_OK
)) {
805 ztest_dump_core
= B_FALSE
;
806 fatal(B_TRUE
, "invalid alternate ztest: %s",
808 } else if (0 != access(zo
->zo_alt_libpath
, X_OK
)) {
809 ztest_dump_core
= B_FALSE
;
810 fatal(B_TRUE
, "invalid alternate lib directory %s",
814 umem_free(cmd
, MAXPATHLEN
);
815 umem_free(realaltdir
, MAXPATHLEN
);
820 ztest_kill(ztest_shared_t
*zs
)
822 zs
->zs_alloc
= metaslab_class_get_alloc(spa_normal_class(ztest_spa
));
823 zs
->zs_space
= metaslab_class_get_space(spa_normal_class(ztest_spa
));
826 * Before we kill off ztest, make sure that the config is updated.
827 * See comment above spa_config_sync().
829 mutex_enter(&spa_namespace_lock
);
830 spa_config_sync(ztest_spa
, B_FALSE
, B_FALSE
);
831 mutex_exit(&spa_namespace_lock
);
833 (void) kill(getpid(), SIGKILL
);
837 ztest_random(uint64_t range
)
841 ASSERT3S(ztest_fd_rand
, >=, 0);
846 if (read(ztest_fd_rand
, &r
, sizeof (r
)) != sizeof (r
))
847 fatal(1, "short read from /dev/urandom");
854 ztest_record_enospc(const char *s
)
856 ztest_shared
->zs_enospc_count
++;
860 ztest_get_ashift(void)
862 if (ztest_opts
.zo_ashift
== 0)
863 return (SPA_MINBLOCKSHIFT
+ ztest_random(5));
864 return (ztest_opts
.zo_ashift
);
868 make_vdev_file(char *path
, char *aux
, char *pool
, size_t size
, uint64_t ashift
)
874 pathbuf
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
877 ashift
= ztest_get_ashift();
883 vdev
= ztest_shared
->zs_vdev_aux
;
884 (void) snprintf(path
, MAXPATHLEN
,
885 ztest_aux_template
, ztest_opts
.zo_dir
,
886 pool
== NULL
? ztest_opts
.zo_pool
: pool
,
889 vdev
= ztest_shared
->zs_vdev_next_leaf
++;
890 (void) snprintf(path
, MAXPATHLEN
,
891 ztest_dev_template
, ztest_opts
.zo_dir
,
892 pool
== NULL
? ztest_opts
.zo_pool
: pool
, vdev
);
897 int fd
= open(path
, O_RDWR
| O_CREAT
| O_TRUNC
, 0666);
899 fatal(1, "can't open %s", path
);
900 if (ftruncate(fd
, size
) != 0)
901 fatal(1, "can't ftruncate %s", path
);
905 VERIFY(nvlist_alloc(&file
, NV_UNIQUE_NAME
, 0) == 0);
906 VERIFY(nvlist_add_string(file
, ZPOOL_CONFIG_TYPE
, VDEV_TYPE_FILE
) == 0);
907 VERIFY(nvlist_add_string(file
, ZPOOL_CONFIG_PATH
, path
) == 0);
908 VERIFY(nvlist_add_uint64(file
, ZPOOL_CONFIG_ASHIFT
, ashift
) == 0);
909 umem_free(pathbuf
, MAXPATHLEN
);
915 make_vdev_raidz(char *path
, char *aux
, char *pool
, size_t size
,
916 uint64_t ashift
, int r
)
918 nvlist_t
*raidz
, **child
;
922 return (make_vdev_file(path
, aux
, pool
, size
, ashift
));
923 child
= umem_alloc(r
* sizeof (nvlist_t
*), UMEM_NOFAIL
);
925 for (c
= 0; c
< r
; c
++)
926 child
[c
] = make_vdev_file(path
, aux
, pool
, size
, ashift
);
928 VERIFY(nvlist_alloc(&raidz
, NV_UNIQUE_NAME
, 0) == 0);
929 VERIFY(nvlist_add_string(raidz
, ZPOOL_CONFIG_TYPE
,
930 VDEV_TYPE_RAIDZ
) == 0);
931 VERIFY(nvlist_add_uint64(raidz
, ZPOOL_CONFIG_NPARITY
,
932 ztest_opts
.zo_raidz_parity
) == 0);
933 VERIFY(nvlist_add_nvlist_array(raidz
, ZPOOL_CONFIG_CHILDREN
,
936 for (c
= 0; c
< r
; c
++)
937 nvlist_free(child
[c
]);
939 umem_free(child
, r
* sizeof (nvlist_t
*));
945 make_vdev_mirror(char *path
, char *aux
, char *pool
, size_t size
,
946 uint64_t ashift
, int r
, int m
)
948 nvlist_t
*mirror
, **child
;
952 return (make_vdev_raidz(path
, aux
, pool
, size
, ashift
, r
));
954 child
= umem_alloc(m
* sizeof (nvlist_t
*), UMEM_NOFAIL
);
956 for (c
= 0; c
< m
; c
++)
957 child
[c
] = make_vdev_raidz(path
, aux
, pool
, size
, ashift
, r
);
959 VERIFY(nvlist_alloc(&mirror
, NV_UNIQUE_NAME
, 0) == 0);
960 VERIFY(nvlist_add_string(mirror
, ZPOOL_CONFIG_TYPE
,
961 VDEV_TYPE_MIRROR
) == 0);
962 VERIFY(nvlist_add_nvlist_array(mirror
, ZPOOL_CONFIG_CHILDREN
,
965 for (c
= 0; c
< m
; c
++)
966 nvlist_free(child
[c
]);
968 umem_free(child
, m
* sizeof (nvlist_t
*));
974 make_vdev_root(char *path
, char *aux
, char *pool
, size_t size
, uint64_t ashift
,
975 int log
, int r
, int m
, int t
)
977 nvlist_t
*root
, **child
;
982 child
= umem_alloc(t
* sizeof (nvlist_t
*), UMEM_NOFAIL
);
984 for (c
= 0; c
< t
; c
++) {
985 child
[c
] = make_vdev_mirror(path
, aux
, pool
, size
, ashift
,
987 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_IS_LOG
,
991 VERIFY(nvlist_alloc(&root
, NV_UNIQUE_NAME
, 0) == 0);
992 VERIFY(nvlist_add_string(root
, ZPOOL_CONFIG_TYPE
, VDEV_TYPE_ROOT
) == 0);
993 VERIFY(nvlist_add_nvlist_array(root
, aux
? aux
: ZPOOL_CONFIG_CHILDREN
,
996 for (c
= 0; c
< t
; c
++)
997 nvlist_free(child
[c
]);
999 umem_free(child
, t
* sizeof (nvlist_t
*));
1005 * Find a random spa version. Returns back a random spa version in the
1006 * range [initial_version, SPA_VERSION_FEATURES].
1009 ztest_random_spa_version(uint64_t initial_version
)
1011 uint64_t version
= initial_version
;
1013 if (version
<= SPA_VERSION_BEFORE_FEATURES
) {
1015 ztest_random(SPA_VERSION_BEFORE_FEATURES
- version
+ 1);
1018 if (version
> SPA_VERSION_BEFORE_FEATURES
)
1019 version
= SPA_VERSION_FEATURES
;
1021 ASSERT(SPA_VERSION_IS_SUPPORTED(version
));
1026 ztest_random_blocksize(void)
1029 * Choose a block size >= the ashift.
1030 * If the SPA supports new MAXBLOCKSIZE, test up to 1MB blocks.
1032 int maxbs
= SPA_OLD_MAXBLOCKSHIFT
;
1033 if (spa_maxblocksize(ztest_spa
) == SPA_MAXBLOCKSIZE
)
1035 uint64_t block_shift
=
1036 ztest_random(maxbs
- ztest_spa
->spa_max_ashift
+ 1);
1037 return (1 << (SPA_MINBLOCKSHIFT
+ block_shift
));
1041 ztest_random_dnodesize(void)
1044 int max_slots
= spa_maxdnodesize(ztest_spa
) >> DNODE_SHIFT
;
1046 if (max_slots
== DNODE_MIN_SLOTS
)
1047 return (DNODE_MIN_SIZE
);
1050 * Weight the random distribution more heavily toward smaller
1051 * dnode sizes since that is more likely to reflect real-world
1054 ASSERT3U(max_slots
, >, 4);
1055 switch (ztest_random(10)) {
1057 slots
= 5 + ztest_random(max_slots
- 4);
1060 slots
= 2 + ztest_random(3);
1067 return (slots
<< DNODE_SHIFT
);
1071 ztest_random_ibshift(void)
1073 return (DN_MIN_INDBLKSHIFT
+
1074 ztest_random(DN_MAX_INDBLKSHIFT
- DN_MIN_INDBLKSHIFT
+ 1));
1078 ztest_random_vdev_top(spa_t
*spa
, boolean_t log_ok
)
1081 vdev_t
*rvd
= spa
->spa_root_vdev
;
1084 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_READER
) != 0);
1087 top
= ztest_random(rvd
->vdev_children
);
1088 tvd
= rvd
->vdev_child
[top
];
1089 } while (tvd
->vdev_ishole
|| (tvd
->vdev_islog
&& !log_ok
) ||
1090 tvd
->vdev_mg
== NULL
|| tvd
->vdev_mg
->mg_class
== NULL
);
1096 ztest_random_dsl_prop(zfs_prop_t prop
)
1101 value
= zfs_prop_random_value(prop
, ztest_random(-1ULL));
1102 } while (prop
== ZFS_PROP_CHECKSUM
&& value
== ZIO_CHECKSUM_OFF
);
1108 ztest_dsl_prop_set_uint64(char *osname
, zfs_prop_t prop
, uint64_t value
,
1111 const char *propname
= zfs_prop_to_name(prop
);
1112 const char *valname
;
1117 error
= dsl_prop_set_int(osname
, propname
,
1118 (inherit
? ZPROP_SRC_NONE
: ZPROP_SRC_LOCAL
), value
);
1120 if (error
== ENOSPC
) {
1121 ztest_record_enospc(FTAG
);
1126 setpoint
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
1127 VERIFY0(dsl_prop_get_integer(osname
, propname
, &curval
, setpoint
));
1129 if (ztest_opts
.zo_verbose
>= 6) {
1132 err
= zfs_prop_index_to_string(prop
, curval
, &valname
);
1134 (void) printf("%s %s = %llu at '%s'\n", osname
,
1135 propname
, (unsigned long long)curval
, setpoint
);
1137 (void) printf("%s %s = %s at '%s'\n",
1138 osname
, propname
, valname
, setpoint
);
1140 umem_free(setpoint
, MAXPATHLEN
);
1146 ztest_spa_prop_set_uint64(zpool_prop_t prop
, uint64_t value
)
1148 spa_t
*spa
= ztest_spa
;
1149 nvlist_t
*props
= NULL
;
1152 VERIFY(nvlist_alloc(&props
, NV_UNIQUE_NAME
, 0) == 0);
1153 VERIFY(nvlist_add_uint64(props
, zpool_prop_to_name(prop
), value
) == 0);
1155 error
= spa_prop_set(spa
, props
);
1159 if (error
== ENOSPC
) {
1160 ztest_record_enospc(FTAG
);
1170 * Object and range lock mechanics
1173 list_node_t z_lnode
;
1174 refcount_t z_refcnt
;
1176 zfs_rlock_t z_range_lock
;
1181 ztest_znode_t
*z_ztznode
;
1184 static ztest_znode_t
*
1185 ztest_znode_init(uint64_t object
)
1187 ztest_znode_t
*zp
= umem_alloc(sizeof (*zp
), UMEM_NOFAIL
);
1189 list_link_init(&zp
->z_lnode
);
1190 refcount_create(&zp
->z_refcnt
);
1191 zp
->z_object
= object
;
1192 zfs_rlock_init(&zp
->z_range_lock
);
1198 ztest_znode_fini(ztest_znode_t
*zp
)
1200 ASSERT(refcount_is_zero(&zp
->z_refcnt
));
1201 zfs_rlock_destroy(&zp
->z_range_lock
);
1203 refcount_destroy(&zp
->z_refcnt
);
1204 list_link_init(&zp
->z_lnode
);
1205 umem_free(zp
, sizeof (*zp
));
1209 ztest_zll_init(zll_t
*zll
)
1211 mutex_init(&zll
->z_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
1212 list_create(&zll
->z_list
, sizeof (ztest_znode_t
),
1213 offsetof(ztest_znode_t
, z_lnode
));
1217 ztest_zll_destroy(zll_t
*zll
)
1219 list_destroy(&zll
->z_list
);
1220 mutex_destroy(&zll
->z_lock
);
1223 #define RL_TAG "range_lock"
1224 static ztest_znode_t
*
1225 ztest_znode_get(ztest_ds_t
*zd
, uint64_t object
)
1227 zll_t
*zll
= &zd
->zd_range_lock
[object
& (ZTEST_OBJECT_LOCKS
- 1)];
1228 ztest_znode_t
*zp
= NULL
;
1229 mutex_enter(&zll
->z_lock
);
1230 for (zp
= list_head(&zll
->z_list
); (zp
);
1231 zp
= list_next(&zll
->z_list
, zp
)) {
1232 if (zp
->z_object
== object
) {
1233 refcount_add(&zp
->z_refcnt
, RL_TAG
);
1238 zp
= ztest_znode_init(object
);
1239 refcount_add(&zp
->z_refcnt
, RL_TAG
);
1240 list_insert_head(&zll
->z_list
, zp
);
1242 mutex_exit(&zll
->z_lock
);
1247 ztest_znode_put(ztest_ds_t
*zd
, ztest_znode_t
*zp
)
1250 ASSERT3U(zp
->z_object
, !=, 0);
1251 zll
= &zd
->zd_range_lock
[zp
->z_object
& (ZTEST_OBJECT_LOCKS
- 1)];
1252 mutex_enter(&zll
->z_lock
);
1253 refcount_remove(&zp
->z_refcnt
, RL_TAG
);
1254 if (refcount_is_zero(&zp
->z_refcnt
)) {
1255 list_remove(&zll
->z_list
, zp
);
1256 ztest_znode_fini(zp
);
1258 mutex_exit(&zll
->z_lock
);
1263 ztest_rll_init(rll_t
*rll
)
1265 rll
->rll_writer
= NULL
;
1266 rll
->rll_readers
= 0;
1267 mutex_init(&rll
->rll_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
1268 cv_init(&rll
->rll_cv
, NULL
, CV_DEFAULT
, NULL
);
1272 ztest_rll_destroy(rll_t
*rll
)
1274 ASSERT(rll
->rll_writer
== NULL
);
1275 ASSERT(rll
->rll_readers
== 0);
1276 mutex_destroy(&rll
->rll_lock
);
1277 cv_destroy(&rll
->rll_cv
);
1281 ztest_rll_lock(rll_t
*rll
, rl_type_t type
)
1283 mutex_enter(&rll
->rll_lock
);
1285 if (type
== RL_READER
) {
1286 while (rll
->rll_writer
!= NULL
)
1287 (void) cv_wait(&rll
->rll_cv
, &rll
->rll_lock
);
1290 while (rll
->rll_writer
!= NULL
|| rll
->rll_readers
)
1291 (void) cv_wait(&rll
->rll_cv
, &rll
->rll_lock
);
1292 rll
->rll_writer
= curthread
;
1295 mutex_exit(&rll
->rll_lock
);
1299 ztest_rll_unlock(rll_t
*rll
)
1301 mutex_enter(&rll
->rll_lock
);
1303 if (rll
->rll_writer
) {
1304 ASSERT(rll
->rll_readers
== 0);
1305 rll
->rll_writer
= NULL
;
1307 ASSERT(rll
->rll_readers
!= 0);
1308 ASSERT(rll
->rll_writer
== NULL
);
1312 if (rll
->rll_writer
== NULL
&& rll
->rll_readers
== 0)
1313 cv_broadcast(&rll
->rll_cv
);
1315 mutex_exit(&rll
->rll_lock
);
1319 ztest_object_lock(ztest_ds_t
*zd
, uint64_t object
, rl_type_t type
)
1321 rll_t
*rll
= &zd
->zd_object_lock
[object
& (ZTEST_OBJECT_LOCKS
- 1)];
1323 ztest_rll_lock(rll
, type
);
1327 ztest_object_unlock(ztest_ds_t
*zd
, uint64_t object
)
1329 rll_t
*rll
= &zd
->zd_object_lock
[object
& (ZTEST_OBJECT_LOCKS
- 1)];
1331 ztest_rll_unlock(rll
);
1334 static ztest_zrl_t
*
1335 ztest_zrl_init(rl_t
*rl
, ztest_znode_t
*zp
)
1337 ztest_zrl_t
*zrl
= umem_alloc(sizeof (*zrl
), UMEM_NOFAIL
);
1339 zrl
->z_ztznode
= zp
;
1344 ztest_zrl_fini(ztest_zrl_t
*zrl
)
1346 umem_free(zrl
, sizeof (*zrl
));
1349 static ztest_zrl_t
*
1350 ztest_range_lock(ztest_ds_t
*zd
, uint64_t object
, uint64_t offset
,
1351 uint64_t size
, rl_type_t type
)
1353 ztest_znode_t
*zp
= ztest_znode_get(zd
, object
);
1354 rl_t
*rl
= zfs_range_lock(&zp
->z_range_lock
, offset
,
1356 return (ztest_zrl_init(rl
, zp
));
1360 ztest_range_unlock(ztest_ds_t
*zd
, ztest_zrl_t
*zrl
)
1362 zfs_range_unlock(zrl
->z_rl
);
1363 ztest_znode_put(zd
, zrl
->z_ztznode
);
1364 ztest_zrl_fini(zrl
);
1368 ztest_zd_init(ztest_ds_t
*zd
, ztest_shared_ds_t
*szd
, objset_t
*os
)
1371 zd
->zd_zilog
= dmu_objset_zil(os
);
1372 zd
->zd_shared
= szd
;
1373 dmu_objset_name(os
, zd
->zd_name
);
1376 if (zd
->zd_shared
!= NULL
)
1377 zd
->zd_shared
->zd_seq
= 0;
1379 VERIFY(rwlock_init(&zd
->zd_zilog_lock
, USYNC_THREAD
, NULL
) == 0);
1380 mutex_init(&zd
->zd_dirobj_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
1382 for (l
= 0; l
< ZTEST_OBJECT_LOCKS
; l
++)
1383 ztest_rll_init(&zd
->zd_object_lock
[l
]);
1385 for (l
= 0; l
< ZTEST_RANGE_LOCKS
; l
++)
1386 ztest_zll_init(&zd
->zd_range_lock
[l
]);
1390 ztest_zd_fini(ztest_ds_t
*zd
)
1394 mutex_destroy(&zd
->zd_dirobj_lock
);
1395 (void) rwlock_destroy(&zd
->zd_zilog_lock
);
1397 for (l
= 0; l
< ZTEST_OBJECT_LOCKS
; l
++)
1398 ztest_rll_destroy(&zd
->zd_object_lock
[l
]);
1400 for (l
= 0; l
< ZTEST_RANGE_LOCKS
; l
++)
1401 ztest_zll_destroy(&zd
->zd_range_lock
[l
]);
1404 #define TXG_MIGHTWAIT (ztest_random(10) == 0 ? TXG_NOWAIT : TXG_WAIT)
1407 ztest_tx_assign(dmu_tx_t
*tx
, uint64_t txg_how
, const char *tag
)
1413 * Attempt to assign tx to some transaction group.
1415 error
= dmu_tx_assign(tx
, txg_how
);
1417 if (error
== ERESTART
) {
1418 ASSERT(txg_how
== TXG_NOWAIT
);
1421 ASSERT3U(error
, ==, ENOSPC
);
1422 ztest_record_enospc(tag
);
1427 txg
= dmu_tx_get_txg(tx
);
1433 ztest_pattern_set(void *buf
, uint64_t size
, uint64_t value
)
1436 uint64_t *ip_end
= (uint64_t *)((uintptr_t)buf
+ (uintptr_t)size
);
1444 ztest_pattern_match(void *buf
, uint64_t size
, uint64_t value
)
1447 uint64_t *ip_end
= (uint64_t *)((uintptr_t)buf
+ (uintptr_t)size
);
1451 diff
|= (value
- *ip
++);
1458 ztest_bt_generate(ztest_block_tag_t
*bt
, objset_t
*os
, uint64_t object
,
1459 uint64_t dnodesize
, uint64_t offset
, uint64_t gen
, uint64_t txg
,
1462 bt
->bt_magic
= BT_MAGIC
;
1463 bt
->bt_objset
= dmu_objset_id(os
);
1464 bt
->bt_object
= object
;
1465 bt
->bt_dnodesize
= dnodesize
;
1466 bt
->bt_offset
= offset
;
1469 bt
->bt_crtxg
= crtxg
;
1473 ztest_bt_verify(ztest_block_tag_t
*bt
, objset_t
*os
, uint64_t object
,
1474 uint64_t dnodesize
, uint64_t offset
, uint64_t gen
, uint64_t txg
,
1477 ASSERT3U(bt
->bt_magic
, ==, BT_MAGIC
);
1478 ASSERT3U(bt
->bt_objset
, ==, dmu_objset_id(os
));
1479 ASSERT3U(bt
->bt_object
, ==, object
);
1480 ASSERT3U(bt
->bt_dnodesize
, ==, dnodesize
);
1481 ASSERT3U(bt
->bt_offset
, ==, offset
);
1482 ASSERT3U(bt
->bt_gen
, <=, gen
);
1483 ASSERT3U(bt
->bt_txg
, <=, txg
);
1484 ASSERT3U(bt
->bt_crtxg
, ==, crtxg
);
1487 static ztest_block_tag_t
*
1488 ztest_bt_bonus(dmu_buf_t
*db
)
1490 dmu_object_info_t doi
;
1491 ztest_block_tag_t
*bt
;
1493 dmu_object_info_from_db(db
, &doi
);
1494 ASSERT3U(doi
.doi_bonus_size
, <=, db
->db_size
);
1495 ASSERT3U(doi
.doi_bonus_size
, >=, sizeof (*bt
));
1496 bt
= (void *)((char *)db
->db_data
+ doi
.doi_bonus_size
- sizeof (*bt
));
1502 * Generate a token to fill up unused bonus buffer space. Try to make
1503 * it unique to the object, generation, and offset to verify that data
1504 * is not getting overwritten by data from other dnodes.
1506 #define ZTEST_BONUS_FILL_TOKEN(obj, ds, gen, offset) \
1507 (((ds) << 48) | ((gen) << 32) | ((obj) << 8) | (offset))
1510 * Fill up the unused bonus buffer region before the block tag with a
1511 * verifiable pattern. Filling the whole bonus area with non-zero data
1512 * helps ensure that all dnode traversal code properly skips the
1513 * interior regions of large dnodes.
1516 ztest_fill_unused_bonus(dmu_buf_t
*db
, void *end
, uint64_t obj
,
1517 objset_t
*os
, uint64_t gen
)
1521 ASSERT(IS_P2ALIGNED((char *)end
- (char *)db
->db_data
, 8));
1523 for (bonusp
= db
->db_data
; bonusp
< (uint64_t *)end
; bonusp
++) {
1524 uint64_t token
= ZTEST_BONUS_FILL_TOKEN(obj
, dmu_objset_id(os
),
1525 gen
, bonusp
- (uint64_t *)db
->db_data
);
1531 * Verify that the unused area of a bonus buffer is filled with the
1535 ztest_verify_unused_bonus(dmu_buf_t
*db
, void *end
, uint64_t obj
,
1536 objset_t
*os
, uint64_t gen
)
1540 for (bonusp
= db
->db_data
; bonusp
< (uint64_t *)end
; bonusp
++) {
1541 uint64_t token
= ZTEST_BONUS_FILL_TOKEN(obj
, dmu_objset_id(os
),
1542 gen
, bonusp
- (uint64_t *)db
->db_data
);
1543 VERIFY3U(*bonusp
, ==, token
);
1551 #define lrz_type lr_mode
1552 #define lrz_blocksize lr_uid
1553 #define lrz_ibshift lr_gid
1554 #define lrz_bonustype lr_rdev
1555 #define lrz_dnodesize lr_crtime[1]
1558 ztest_log_create(ztest_ds_t
*zd
, dmu_tx_t
*tx
, lr_create_t
*lr
)
1560 char *name
= (void *)(lr
+ 1); /* name follows lr */
1561 size_t namesize
= strlen(name
) + 1;
1564 if (zil_replaying(zd
->zd_zilog
, tx
))
1567 itx
= zil_itx_create(TX_CREATE
, sizeof (*lr
) + namesize
);
1568 bcopy(&lr
->lr_common
+ 1, &itx
->itx_lr
+ 1,
1569 sizeof (*lr
) + namesize
- sizeof (lr_t
));
1571 zil_itx_assign(zd
->zd_zilog
, itx
, tx
);
1575 ztest_log_remove(ztest_ds_t
*zd
, dmu_tx_t
*tx
, lr_remove_t
*lr
, uint64_t object
)
1577 char *name
= (void *)(lr
+ 1); /* name follows lr */
1578 size_t namesize
= strlen(name
) + 1;
1581 if (zil_replaying(zd
->zd_zilog
, tx
))
1584 itx
= zil_itx_create(TX_REMOVE
, sizeof (*lr
) + namesize
);
1585 bcopy(&lr
->lr_common
+ 1, &itx
->itx_lr
+ 1,
1586 sizeof (*lr
) + namesize
- sizeof (lr_t
));
1588 itx
->itx_oid
= object
;
1589 zil_itx_assign(zd
->zd_zilog
, itx
, tx
);
1593 ztest_log_write(ztest_ds_t
*zd
, dmu_tx_t
*tx
, lr_write_t
*lr
)
1596 itx_wr_state_t write_state
= ztest_random(WR_NUM_STATES
);
1598 if (zil_replaying(zd
->zd_zilog
, tx
))
1601 if (lr
->lr_length
> ZIL_MAX_LOG_DATA
)
1602 write_state
= WR_INDIRECT
;
1604 itx
= zil_itx_create(TX_WRITE
,
1605 sizeof (*lr
) + (write_state
== WR_COPIED
? lr
->lr_length
: 0));
1607 if (write_state
== WR_COPIED
&&
1608 dmu_read(zd
->zd_os
, lr
->lr_foid
, lr
->lr_offset
, lr
->lr_length
,
1609 ((lr_write_t
*)&itx
->itx_lr
) + 1, DMU_READ_NO_PREFETCH
) != 0) {
1610 zil_itx_destroy(itx
);
1611 itx
= zil_itx_create(TX_WRITE
, sizeof (*lr
));
1612 write_state
= WR_NEED_COPY
;
1614 itx
->itx_private
= zd
;
1615 itx
->itx_wr_state
= write_state
;
1616 itx
->itx_sync
= (ztest_random(8) == 0);
1617 itx
->itx_sod
+= (write_state
== WR_NEED_COPY
? lr
->lr_length
: 0);
1619 bcopy(&lr
->lr_common
+ 1, &itx
->itx_lr
+ 1,
1620 sizeof (*lr
) - sizeof (lr_t
));
1622 zil_itx_assign(zd
->zd_zilog
, itx
, tx
);
1626 ztest_log_truncate(ztest_ds_t
*zd
, dmu_tx_t
*tx
, lr_truncate_t
*lr
)
1630 if (zil_replaying(zd
->zd_zilog
, tx
))
1633 itx
= zil_itx_create(TX_TRUNCATE
, sizeof (*lr
));
1634 bcopy(&lr
->lr_common
+ 1, &itx
->itx_lr
+ 1,
1635 sizeof (*lr
) - sizeof (lr_t
));
1637 itx
->itx_sync
= B_FALSE
;
1638 zil_itx_assign(zd
->zd_zilog
, itx
, tx
);
1642 ztest_log_setattr(ztest_ds_t
*zd
, dmu_tx_t
*tx
, lr_setattr_t
*lr
)
1646 if (zil_replaying(zd
->zd_zilog
, tx
))
1649 itx
= zil_itx_create(TX_SETATTR
, sizeof (*lr
));
1650 bcopy(&lr
->lr_common
+ 1, &itx
->itx_lr
+ 1,
1651 sizeof (*lr
) - sizeof (lr_t
));
1653 itx
->itx_sync
= B_FALSE
;
1654 zil_itx_assign(zd
->zd_zilog
, itx
, tx
);
1661 ztest_replay_create(ztest_ds_t
*zd
, lr_create_t
*lr
, boolean_t byteswap
)
1663 char *name
= (void *)(lr
+ 1); /* name follows lr */
1664 objset_t
*os
= zd
->zd_os
;
1665 ztest_block_tag_t
*bbt
;
1673 byteswap_uint64_array(lr
, sizeof (*lr
));
1675 ASSERT(lr
->lr_doid
== ZTEST_DIROBJ
);
1676 ASSERT(name
[0] != '\0');
1678 tx
= dmu_tx_create(os
);
1680 dmu_tx_hold_zap(tx
, lr
->lr_doid
, B_TRUE
, name
);
1682 if (lr
->lrz_type
== DMU_OT_ZAP_OTHER
) {
1683 dmu_tx_hold_zap(tx
, DMU_NEW_OBJECT
, B_TRUE
, NULL
);
1685 dmu_tx_hold_bonus(tx
, DMU_NEW_OBJECT
);
1688 txg
= ztest_tx_assign(tx
, TXG_WAIT
, FTAG
);
1692 ASSERT(dmu_objset_zil(os
)->zl_replay
== !!lr
->lr_foid
);
1693 bonuslen
= DN_BONUS_SIZE(lr
->lrz_dnodesize
);
1695 if (lr
->lrz_type
== DMU_OT_ZAP_OTHER
) {
1696 if (lr
->lr_foid
== 0) {
1697 lr
->lr_foid
= zap_create_dnsize(os
,
1698 lr
->lrz_type
, lr
->lrz_bonustype
,
1699 bonuslen
, lr
->lrz_dnodesize
, tx
);
1701 error
= zap_create_claim_dnsize(os
, lr
->lr_foid
,
1702 lr
->lrz_type
, lr
->lrz_bonustype
,
1703 bonuslen
, lr
->lrz_dnodesize
, tx
);
1706 if (lr
->lr_foid
== 0) {
1707 lr
->lr_foid
= dmu_object_alloc_dnsize(os
,
1708 lr
->lrz_type
, 0, lr
->lrz_bonustype
,
1709 bonuslen
, lr
->lrz_dnodesize
, tx
);
1711 error
= dmu_object_claim_dnsize(os
, lr
->lr_foid
,
1712 lr
->lrz_type
, 0, lr
->lrz_bonustype
,
1713 bonuslen
, lr
->lrz_dnodesize
, tx
);
1718 ASSERT3U(error
, ==, EEXIST
);
1719 ASSERT(zd
->zd_zilog
->zl_replay
);
1724 ASSERT(lr
->lr_foid
!= 0);
1726 if (lr
->lrz_type
!= DMU_OT_ZAP_OTHER
)
1727 VERIFY3U(0, ==, dmu_object_set_blocksize(os
, lr
->lr_foid
,
1728 lr
->lrz_blocksize
, lr
->lrz_ibshift
, tx
));
1730 VERIFY3U(0, ==, dmu_bonus_hold(os
, lr
->lr_foid
, FTAG
, &db
));
1731 bbt
= ztest_bt_bonus(db
);
1732 dmu_buf_will_dirty(db
, tx
);
1733 ztest_bt_generate(bbt
, os
, lr
->lr_foid
, lr
->lrz_dnodesize
, -1ULL,
1734 lr
->lr_gen
, txg
, txg
);
1735 ztest_fill_unused_bonus(db
, bbt
, lr
->lr_foid
, os
, lr
->lr_gen
);
1736 dmu_buf_rele(db
, FTAG
);
1738 VERIFY3U(0, ==, zap_add(os
, lr
->lr_doid
, name
, sizeof (uint64_t), 1,
1741 (void) ztest_log_create(zd
, tx
, lr
);
1749 ztest_replay_remove(ztest_ds_t
*zd
, lr_remove_t
*lr
, boolean_t byteswap
)
1751 char *name
= (void *)(lr
+ 1); /* name follows lr */
1752 objset_t
*os
= zd
->zd_os
;
1753 dmu_object_info_t doi
;
1755 uint64_t object
, txg
;
1758 byteswap_uint64_array(lr
, sizeof (*lr
));
1760 ASSERT(lr
->lr_doid
== ZTEST_DIROBJ
);
1761 ASSERT(name
[0] != '\0');
1764 zap_lookup(os
, lr
->lr_doid
, name
, sizeof (object
), 1, &object
));
1765 ASSERT(object
!= 0);
1767 ztest_object_lock(zd
, object
, RL_WRITER
);
1769 VERIFY3U(0, ==, dmu_object_info(os
, object
, &doi
));
1771 tx
= dmu_tx_create(os
);
1773 dmu_tx_hold_zap(tx
, lr
->lr_doid
, B_FALSE
, name
);
1774 dmu_tx_hold_free(tx
, object
, 0, DMU_OBJECT_END
);
1776 txg
= ztest_tx_assign(tx
, TXG_WAIT
, FTAG
);
1778 ztest_object_unlock(zd
, object
);
1782 if (doi
.doi_type
== DMU_OT_ZAP_OTHER
) {
1783 VERIFY3U(0, ==, zap_destroy(os
, object
, tx
));
1785 VERIFY3U(0, ==, dmu_object_free(os
, object
, tx
));
1788 VERIFY3U(0, ==, zap_remove(os
, lr
->lr_doid
, name
, tx
));
1790 (void) ztest_log_remove(zd
, tx
, lr
, object
);
1794 ztest_object_unlock(zd
, object
);
1800 ztest_replay_write(ztest_ds_t
*zd
, lr_write_t
*lr
, boolean_t byteswap
)
1802 objset_t
*os
= zd
->zd_os
;
1803 void *data
= lr
+ 1; /* data follows lr */
1804 uint64_t offset
, length
;
1805 ztest_block_tag_t
*bt
= data
;
1806 ztest_block_tag_t
*bbt
;
1807 uint64_t gen
, txg
, lrtxg
, crtxg
;
1808 dmu_object_info_t doi
;
1811 arc_buf_t
*abuf
= NULL
;
1815 byteswap_uint64_array(lr
, sizeof (*lr
));
1817 offset
= lr
->lr_offset
;
1818 length
= lr
->lr_length
;
1820 /* If it's a dmu_sync() block, write the whole block */
1821 if (lr
->lr_common
.lrc_reclen
== sizeof (lr_write_t
)) {
1822 uint64_t blocksize
= BP_GET_LSIZE(&lr
->lr_blkptr
);
1823 if (length
< blocksize
) {
1824 offset
-= offset
% blocksize
;
1829 if (bt
->bt_magic
== BSWAP_64(BT_MAGIC
))
1830 byteswap_uint64_array(bt
, sizeof (*bt
));
1832 if (bt
->bt_magic
!= BT_MAGIC
)
1835 ztest_object_lock(zd
, lr
->lr_foid
, RL_READER
);
1836 rl
= ztest_range_lock(zd
, lr
->lr_foid
, offset
, length
, RL_WRITER
);
1838 VERIFY3U(0, ==, dmu_bonus_hold(os
, lr
->lr_foid
, FTAG
, &db
));
1840 dmu_object_info_from_db(db
, &doi
);
1842 bbt
= ztest_bt_bonus(db
);
1843 ASSERT3U(bbt
->bt_magic
, ==, BT_MAGIC
);
1845 crtxg
= bbt
->bt_crtxg
;
1846 lrtxg
= lr
->lr_common
.lrc_txg
;
1848 tx
= dmu_tx_create(os
);
1850 dmu_tx_hold_write(tx
, lr
->lr_foid
, offset
, length
);
1852 if (ztest_random(8) == 0 && length
== doi
.doi_data_block_size
&&
1853 P2PHASE(offset
, length
) == 0)
1854 abuf
= dmu_request_arcbuf(db
, length
);
1856 txg
= ztest_tx_assign(tx
, TXG_WAIT
, FTAG
);
1859 dmu_return_arcbuf(abuf
);
1860 dmu_buf_rele(db
, FTAG
);
1861 ztest_range_unlock(zd
, rl
);
1862 ztest_object_unlock(zd
, lr
->lr_foid
);
1868 * Usually, verify the old data before writing new data --
1869 * but not always, because we also want to verify correct
1870 * behavior when the data was not recently read into cache.
1872 ASSERT(offset
% doi
.doi_data_block_size
== 0);
1873 if (ztest_random(4) != 0) {
1874 int prefetch
= ztest_random(2) ?
1875 DMU_READ_PREFETCH
: DMU_READ_NO_PREFETCH
;
1876 ztest_block_tag_t rbt
;
1878 VERIFY(dmu_read(os
, lr
->lr_foid
, offset
,
1879 sizeof (rbt
), &rbt
, prefetch
) == 0);
1880 if (rbt
.bt_magic
== BT_MAGIC
) {
1881 ztest_bt_verify(&rbt
, os
, lr
->lr_foid
, 0,
1882 offset
, gen
, txg
, crtxg
);
1887 * Writes can appear to be newer than the bonus buffer because
1888 * the ztest_get_data() callback does a dmu_read() of the
1889 * open-context data, which may be different than the data
1890 * as it was when the write was generated.
1892 if (zd
->zd_zilog
->zl_replay
) {
1893 ztest_bt_verify(bt
, os
, lr
->lr_foid
, 0, offset
,
1894 MAX(gen
, bt
->bt_gen
), MAX(txg
, lrtxg
),
1899 * Set the bt's gen/txg to the bonus buffer's gen/txg
1900 * so that all of the usual ASSERTs will work.
1902 ztest_bt_generate(bt
, os
, lr
->lr_foid
, 0, offset
, gen
, txg
,
1907 dmu_write(os
, lr
->lr_foid
, offset
, length
, data
, tx
);
1909 bcopy(data
, abuf
->b_data
, length
);
1910 dmu_assign_arcbuf(db
, offset
, abuf
, tx
);
1913 (void) ztest_log_write(zd
, tx
, lr
);
1915 dmu_buf_rele(db
, FTAG
);
1919 ztest_range_unlock(zd
, rl
);
1920 ztest_object_unlock(zd
, lr
->lr_foid
);
1926 ztest_replay_truncate(ztest_ds_t
*zd
, lr_truncate_t
*lr
, boolean_t byteswap
)
1928 objset_t
*os
= zd
->zd_os
;
1934 byteswap_uint64_array(lr
, sizeof (*lr
));
1936 ztest_object_lock(zd
, lr
->lr_foid
, RL_READER
);
1937 rl
= ztest_range_lock(zd
, lr
->lr_foid
, lr
->lr_offset
, lr
->lr_length
,
1940 tx
= dmu_tx_create(os
);
1942 dmu_tx_hold_free(tx
, lr
->lr_foid
, lr
->lr_offset
, lr
->lr_length
);
1944 txg
= ztest_tx_assign(tx
, TXG_WAIT
, FTAG
);
1946 ztest_range_unlock(zd
, rl
);
1947 ztest_object_unlock(zd
, lr
->lr_foid
);
1951 VERIFY(dmu_free_range(os
, lr
->lr_foid
, lr
->lr_offset
,
1952 lr
->lr_length
, tx
) == 0);
1954 (void) ztest_log_truncate(zd
, tx
, lr
);
1958 ztest_range_unlock(zd
, rl
);
1959 ztest_object_unlock(zd
, lr
->lr_foid
);
1965 ztest_replay_setattr(ztest_ds_t
*zd
, lr_setattr_t
*lr
, boolean_t byteswap
)
1967 objset_t
*os
= zd
->zd_os
;
1970 ztest_block_tag_t
*bbt
;
1971 uint64_t txg
, lrtxg
, crtxg
, dnodesize
;
1974 byteswap_uint64_array(lr
, sizeof (*lr
));
1976 ztest_object_lock(zd
, lr
->lr_foid
, RL_WRITER
);
1978 VERIFY3U(0, ==, dmu_bonus_hold(os
, lr
->lr_foid
, FTAG
, &db
));
1980 tx
= dmu_tx_create(os
);
1981 dmu_tx_hold_bonus(tx
, lr
->lr_foid
);
1983 txg
= ztest_tx_assign(tx
, TXG_WAIT
, FTAG
);
1985 dmu_buf_rele(db
, FTAG
);
1986 ztest_object_unlock(zd
, lr
->lr_foid
);
1990 bbt
= ztest_bt_bonus(db
);
1991 ASSERT3U(bbt
->bt_magic
, ==, BT_MAGIC
);
1992 crtxg
= bbt
->bt_crtxg
;
1993 lrtxg
= lr
->lr_common
.lrc_txg
;
1994 dnodesize
= bbt
->bt_dnodesize
;
1996 if (zd
->zd_zilog
->zl_replay
) {
1997 ASSERT(lr
->lr_size
!= 0);
1998 ASSERT(lr
->lr_mode
!= 0);
2002 * Randomly change the size and increment the generation.
2004 lr
->lr_size
= (ztest_random(db
->db_size
/ sizeof (*bbt
)) + 1) *
2006 lr
->lr_mode
= bbt
->bt_gen
+ 1;
2011 * Verify that the current bonus buffer is not newer than our txg.
2013 ztest_bt_verify(bbt
, os
, lr
->lr_foid
, dnodesize
, -1ULL, lr
->lr_mode
,
2014 MAX(txg
, lrtxg
), crtxg
);
2016 dmu_buf_will_dirty(db
, tx
);
2018 ASSERT3U(lr
->lr_size
, >=, sizeof (*bbt
));
2019 ASSERT3U(lr
->lr_size
, <=, db
->db_size
);
2020 VERIFY0(dmu_set_bonus(db
, lr
->lr_size
, tx
));
2021 bbt
= ztest_bt_bonus(db
);
2023 ztest_bt_generate(bbt
, os
, lr
->lr_foid
, dnodesize
, -1ULL, lr
->lr_mode
,
2025 ztest_fill_unused_bonus(db
, bbt
, lr
->lr_foid
, os
, bbt
->bt_gen
);
2026 dmu_buf_rele(db
, FTAG
);
2028 (void) ztest_log_setattr(zd
, tx
, lr
);
2032 ztest_object_unlock(zd
, lr
->lr_foid
);
2037 zil_replay_func_t ztest_replay_vector
[TX_MAX_TYPE
] = {
2038 NULL
, /* 0 no such transaction type */
2039 (zil_replay_func_t
)ztest_replay_create
, /* TX_CREATE */
2040 NULL
, /* TX_MKDIR */
2041 NULL
, /* TX_MKXATTR */
2042 NULL
, /* TX_SYMLINK */
2043 (zil_replay_func_t
)ztest_replay_remove
, /* TX_REMOVE */
2044 NULL
, /* TX_RMDIR */
2046 NULL
, /* TX_RENAME */
2047 (zil_replay_func_t
)ztest_replay_write
, /* TX_WRITE */
2048 (zil_replay_func_t
)ztest_replay_truncate
, /* TX_TRUNCATE */
2049 (zil_replay_func_t
)ztest_replay_setattr
, /* TX_SETATTR */
2051 NULL
, /* TX_CREATE_ACL */
2052 NULL
, /* TX_CREATE_ATTR */
2053 NULL
, /* TX_CREATE_ACL_ATTR */
2054 NULL
, /* TX_MKDIR_ACL */
2055 NULL
, /* TX_MKDIR_ATTR */
2056 NULL
, /* TX_MKDIR_ACL_ATTR */
2057 NULL
, /* TX_WRITE2 */
2061 * ZIL get_data callbacks
2063 typedef struct ztest_zgd_private
{
2067 } ztest_zgd_private_t
;
2070 ztest_get_done(zgd_t
*zgd
, int error
)
2072 ztest_zgd_private_t
*zzp
= zgd
->zgd_private
;
2073 ztest_ds_t
*zd
= zzp
->z_zd
;
2074 uint64_t object
= zzp
->z_object
;
2077 dmu_buf_rele(zgd
->zgd_db
, zgd
);
2079 ztest_range_unlock(zd
, zzp
->z_rl
);
2080 ztest_object_unlock(zd
, object
);
2082 if (error
== 0 && zgd
->zgd_bp
)
2083 zil_add_block(zgd
->zgd_zilog
, zgd
->zgd_bp
);
2085 umem_free(zgd
, sizeof (*zgd
));
2086 umem_free(zzp
, sizeof (*zzp
));
2090 ztest_get_data(void *arg
, lr_write_t
*lr
, char *buf
, zio_t
*zio
)
2092 ztest_ds_t
*zd
= arg
;
2093 objset_t
*os
= zd
->zd_os
;
2094 uint64_t object
= lr
->lr_foid
;
2095 uint64_t offset
= lr
->lr_offset
;
2096 uint64_t size
= lr
->lr_length
;
2097 blkptr_t
*bp
= &lr
->lr_blkptr
;
2098 uint64_t txg
= lr
->lr_common
.lrc_txg
;
2100 dmu_object_info_t doi
;
2104 ztest_zgd_private_t
*zgd_private
;
2106 ztest_object_lock(zd
, object
, RL_READER
);
2107 error
= dmu_bonus_hold(os
, object
, FTAG
, &db
);
2109 ztest_object_unlock(zd
, object
);
2113 crtxg
= ztest_bt_bonus(db
)->bt_crtxg
;
2115 if (crtxg
== 0 || crtxg
> txg
) {
2116 dmu_buf_rele(db
, FTAG
);
2117 ztest_object_unlock(zd
, object
);
2121 dmu_object_info_from_db(db
, &doi
);
2122 dmu_buf_rele(db
, FTAG
);
2125 zgd
= umem_zalloc(sizeof (*zgd
), UMEM_NOFAIL
);
2126 zgd
->zgd_zilog
= zd
->zd_zilog
;
2127 zgd_private
= umem_zalloc(sizeof (ztest_zgd_private_t
), UMEM_NOFAIL
);
2128 zgd_private
->z_zd
= zd
;
2129 zgd_private
->z_object
= object
;
2130 zgd
->zgd_private
= zgd_private
;
2132 if (buf
!= NULL
) { /* immediate write */
2133 zgd_private
->z_rl
= ztest_range_lock(zd
, object
, offset
, size
,
2136 error
= dmu_read(os
, object
, offset
, size
, buf
,
2137 DMU_READ_NO_PREFETCH
);
2140 size
= doi
.doi_data_block_size
;
2142 offset
= P2ALIGN(offset
, size
);
2144 ASSERT(offset
< size
);
2148 zgd_private
->z_rl
= ztest_range_lock(zd
, object
, offset
, size
,
2151 error
= dmu_buf_hold(os
, object
, offset
, zgd
, &db
,
2152 DMU_READ_NO_PREFETCH
);
2155 blkptr_t
*obp
= dmu_buf_get_blkptr(db
);
2157 ASSERT(BP_IS_HOLE(bp
));
2164 ASSERT(db
->db_offset
== offset
);
2165 ASSERT(db
->db_size
== size
);
2167 error
= dmu_sync(zio
, lr
->lr_common
.lrc_txg
,
2168 ztest_get_done
, zgd
);
2175 ztest_get_done(zgd
, error
);
2181 ztest_lr_alloc(size_t lrsize
, char *name
)
2184 size_t namesize
= name
? strlen(name
) + 1 : 0;
2186 lr
= umem_zalloc(lrsize
+ namesize
, UMEM_NOFAIL
);
2189 bcopy(name
, lr
+ lrsize
, namesize
);
2195 ztest_lr_free(void *lr
, size_t lrsize
, char *name
)
2197 size_t namesize
= name
? strlen(name
) + 1 : 0;
2199 umem_free(lr
, lrsize
+ namesize
);
2203 * Lookup a bunch of objects. Returns the number of objects not found.
2206 ztest_lookup(ztest_ds_t
*zd
, ztest_od_t
*od
, int count
)
2212 ASSERT(mutex_held(&zd
->zd_dirobj_lock
));
2214 for (i
= 0; i
< count
; i
++, od
++) {
2216 error
= zap_lookup(zd
->zd_os
, od
->od_dir
, od
->od_name
,
2217 sizeof (uint64_t), 1, &od
->od_object
);
2219 ASSERT(error
== ENOENT
);
2220 ASSERT(od
->od_object
== 0);
2224 ztest_block_tag_t
*bbt
;
2225 dmu_object_info_t doi
;
2227 ASSERT(od
->od_object
!= 0);
2228 ASSERT(missing
== 0); /* there should be no gaps */
2230 ztest_object_lock(zd
, od
->od_object
, RL_READER
);
2231 VERIFY3U(0, ==, dmu_bonus_hold(zd
->zd_os
,
2232 od
->od_object
, FTAG
, &db
));
2233 dmu_object_info_from_db(db
, &doi
);
2234 bbt
= ztest_bt_bonus(db
);
2235 ASSERT3U(bbt
->bt_magic
, ==, BT_MAGIC
);
2236 od
->od_type
= doi
.doi_type
;
2237 od
->od_blocksize
= doi
.doi_data_block_size
;
2238 od
->od_gen
= bbt
->bt_gen
;
2239 dmu_buf_rele(db
, FTAG
);
2240 ztest_object_unlock(zd
, od
->od_object
);
2248 ztest_create(ztest_ds_t
*zd
, ztest_od_t
*od
, int count
)
2253 ASSERT(mutex_held(&zd
->zd_dirobj_lock
));
2255 for (i
= 0; i
< count
; i
++, od
++) {
2262 lr_create_t
*lr
= ztest_lr_alloc(sizeof (*lr
), od
->od_name
);
2264 lr
->lr_doid
= od
->od_dir
;
2265 lr
->lr_foid
= 0; /* 0 to allocate, > 0 to claim */
2266 lr
->lrz_type
= od
->od_crtype
;
2267 lr
->lrz_blocksize
= od
->od_crblocksize
;
2268 lr
->lrz_ibshift
= ztest_random_ibshift();
2269 lr
->lrz_bonustype
= DMU_OT_UINT64_OTHER
;
2270 lr
->lrz_dnodesize
= od
->od_crdnodesize
;
2271 lr
->lr_gen
= od
->od_crgen
;
2272 lr
->lr_crtime
[0] = time(NULL
);
2274 if (ztest_replay_create(zd
, lr
, B_FALSE
) != 0) {
2275 ASSERT(missing
== 0);
2279 od
->od_object
= lr
->lr_foid
;
2280 od
->od_type
= od
->od_crtype
;
2281 od
->od_blocksize
= od
->od_crblocksize
;
2282 od
->od_gen
= od
->od_crgen
;
2283 ASSERT(od
->od_object
!= 0);
2286 ztest_lr_free(lr
, sizeof (*lr
), od
->od_name
);
2293 ztest_remove(ztest_ds_t
*zd
, ztest_od_t
*od
, int count
)
2299 ASSERT(mutex_held(&zd
->zd_dirobj_lock
));
2303 for (i
= count
- 1; i
>= 0; i
--, od
--) {
2310 * No object was found.
2312 if (od
->od_object
== 0)
2315 lr_remove_t
*lr
= ztest_lr_alloc(sizeof (*lr
), od
->od_name
);
2317 lr
->lr_doid
= od
->od_dir
;
2319 if ((error
= ztest_replay_remove(zd
, lr
, B_FALSE
)) != 0) {
2320 ASSERT3U(error
, ==, ENOSPC
);
2325 ztest_lr_free(lr
, sizeof (*lr
), od
->od_name
);
2332 ztest_write(ztest_ds_t
*zd
, uint64_t object
, uint64_t offset
, uint64_t size
,
2338 lr
= ztest_lr_alloc(sizeof (*lr
) + size
, NULL
);
2340 lr
->lr_foid
= object
;
2341 lr
->lr_offset
= offset
;
2342 lr
->lr_length
= size
;
2344 BP_ZERO(&lr
->lr_blkptr
);
2346 bcopy(data
, lr
+ 1, size
);
2348 error
= ztest_replay_write(zd
, lr
, B_FALSE
);
2350 ztest_lr_free(lr
, sizeof (*lr
) + size
, NULL
);
2356 ztest_truncate(ztest_ds_t
*zd
, uint64_t object
, uint64_t offset
, uint64_t size
)
2361 lr
= ztest_lr_alloc(sizeof (*lr
), NULL
);
2363 lr
->lr_foid
= object
;
2364 lr
->lr_offset
= offset
;
2365 lr
->lr_length
= size
;
2367 error
= ztest_replay_truncate(zd
, lr
, B_FALSE
);
2369 ztest_lr_free(lr
, sizeof (*lr
), NULL
);
2375 ztest_setattr(ztest_ds_t
*zd
, uint64_t object
)
2380 lr
= ztest_lr_alloc(sizeof (*lr
), NULL
);
2382 lr
->lr_foid
= object
;
2386 error
= ztest_replay_setattr(zd
, lr
, B_FALSE
);
2388 ztest_lr_free(lr
, sizeof (*lr
), NULL
);
2394 ztest_prealloc(ztest_ds_t
*zd
, uint64_t object
, uint64_t offset
, uint64_t size
)
2396 objset_t
*os
= zd
->zd_os
;
2401 txg_wait_synced(dmu_objset_pool(os
), 0);
2403 ztest_object_lock(zd
, object
, RL_READER
);
2404 rl
= ztest_range_lock(zd
, object
, offset
, size
, RL_WRITER
);
2406 tx
= dmu_tx_create(os
);
2408 dmu_tx_hold_write(tx
, object
, offset
, size
);
2410 txg
= ztest_tx_assign(tx
, TXG_WAIT
, FTAG
);
2413 dmu_prealloc(os
, object
, offset
, size
, tx
);
2415 txg_wait_synced(dmu_objset_pool(os
), txg
);
2417 (void) dmu_free_long_range(os
, object
, offset
, size
);
2420 ztest_range_unlock(zd
, rl
);
2421 ztest_object_unlock(zd
, object
);
2425 ztest_io(ztest_ds_t
*zd
, uint64_t object
, uint64_t offset
)
2428 ztest_block_tag_t wbt
;
2429 dmu_object_info_t doi
;
2430 enum ztest_io_type io_type
;
2434 VERIFY(dmu_object_info(zd
->zd_os
, object
, &doi
) == 0);
2435 blocksize
= doi
.doi_data_block_size
;
2436 data
= umem_alloc(blocksize
, UMEM_NOFAIL
);
2439 * Pick an i/o type at random, biased toward writing block tags.
2441 io_type
= ztest_random(ZTEST_IO_TYPES
);
2442 if (ztest_random(2) == 0)
2443 io_type
= ZTEST_IO_WRITE_TAG
;
2445 (void) rw_rdlock(&zd
->zd_zilog_lock
);
2449 case ZTEST_IO_WRITE_TAG
:
2450 ztest_bt_generate(&wbt
, zd
->zd_os
, object
, doi
.doi_dnodesize
,
2452 (void) ztest_write(zd
, object
, offset
, sizeof (wbt
), &wbt
);
2455 case ZTEST_IO_WRITE_PATTERN
:
2456 (void) memset(data
, 'a' + (object
+ offset
) % 5, blocksize
);
2457 if (ztest_random(2) == 0) {
2459 * Induce fletcher2 collisions to ensure that
2460 * zio_ddt_collision() detects and resolves them
2461 * when using fletcher2-verify for deduplication.
2463 ((uint64_t *)data
)[0] ^= 1ULL << 63;
2464 ((uint64_t *)data
)[4] ^= 1ULL << 63;
2466 (void) ztest_write(zd
, object
, offset
, blocksize
, data
);
2469 case ZTEST_IO_WRITE_ZEROES
:
2470 bzero(data
, blocksize
);
2471 (void) ztest_write(zd
, object
, offset
, blocksize
, data
);
2474 case ZTEST_IO_TRUNCATE
:
2475 (void) ztest_truncate(zd
, object
, offset
, blocksize
);
2478 case ZTEST_IO_SETATTR
:
2479 (void) ztest_setattr(zd
, object
);
2484 case ZTEST_IO_REWRITE
:
2485 (void) rw_rdlock(&ztest_name_lock
);
2486 err
= ztest_dsl_prop_set_uint64(zd
->zd_name
,
2487 ZFS_PROP_CHECKSUM
, spa_dedup_checksum(ztest_spa
),
2489 VERIFY(err
== 0 || err
== ENOSPC
);
2490 err
= ztest_dsl_prop_set_uint64(zd
->zd_name
,
2491 ZFS_PROP_COMPRESSION
,
2492 ztest_random_dsl_prop(ZFS_PROP_COMPRESSION
),
2494 VERIFY(err
== 0 || err
== ENOSPC
);
2495 (void) rw_unlock(&ztest_name_lock
);
2497 VERIFY0(dmu_read(zd
->zd_os
, object
, offset
, blocksize
, data
,
2498 DMU_READ_NO_PREFETCH
));
2500 (void) ztest_write(zd
, object
, offset
, blocksize
, data
);
2504 (void) rw_unlock(&zd
->zd_zilog_lock
);
2506 umem_free(data
, blocksize
);
2510 * Initialize an object description template.
2513 ztest_od_init(ztest_od_t
*od
, uint64_t id
, char *tag
, uint64_t index
,
2514 dmu_object_type_t type
, uint64_t blocksize
, uint64_t dnodesize
,
2517 od
->od_dir
= ZTEST_DIROBJ
;
2520 od
->od_crtype
= type
;
2521 od
->od_crblocksize
= blocksize
? blocksize
: ztest_random_blocksize();
2522 od
->od_crdnodesize
= dnodesize
? dnodesize
: ztest_random_dnodesize();
2525 od
->od_type
= DMU_OT_NONE
;
2526 od
->od_blocksize
= 0;
2529 (void) snprintf(od
->od_name
, sizeof (od
->od_name
), "%s(%lld)[%llu]",
2530 tag
, (longlong_t
)id
, (u_longlong_t
)index
);
2534 * Lookup or create the objects for a test using the od template.
2535 * If the objects do not all exist, or if 'remove' is specified,
2536 * remove any existing objects and create new ones. Otherwise,
2537 * use the existing objects.
2540 ztest_object_init(ztest_ds_t
*zd
, ztest_od_t
*od
, size_t size
, boolean_t remove
)
2542 int count
= size
/ sizeof (*od
);
2545 mutex_enter(&zd
->zd_dirobj_lock
);
2546 if ((ztest_lookup(zd
, od
, count
) != 0 || remove
) &&
2547 (ztest_remove(zd
, od
, count
) != 0 ||
2548 ztest_create(zd
, od
, count
) != 0))
2551 mutex_exit(&zd
->zd_dirobj_lock
);
2558 ztest_zil_commit(ztest_ds_t
*zd
, uint64_t id
)
2560 zilog_t
*zilog
= zd
->zd_zilog
;
2562 (void) rw_rdlock(&zd
->zd_zilog_lock
);
2564 zil_commit(zilog
, ztest_random(ZTEST_OBJECTS
));
2567 * Remember the committed values in zd, which is in parent/child
2568 * shared memory. If we die, the next iteration of ztest_run()
2569 * will verify that the log really does contain this record.
2571 mutex_enter(&zilog
->zl_lock
);
2572 ASSERT(zd
->zd_shared
!= NULL
);
2573 ASSERT3U(zd
->zd_shared
->zd_seq
, <=, zilog
->zl_commit_lr_seq
);
2574 zd
->zd_shared
->zd_seq
= zilog
->zl_commit_lr_seq
;
2575 mutex_exit(&zilog
->zl_lock
);
2577 (void) rw_unlock(&zd
->zd_zilog_lock
);
2581 * This function is designed to simulate the operations that occur during a
2582 * mount/unmount operation. We hold the dataset across these operations in an
2583 * attempt to expose any implicit assumptions about ZIL management.
2587 ztest_zil_remount(ztest_ds_t
*zd
, uint64_t id
)
2589 objset_t
*os
= zd
->zd_os
;
2592 * We grab the zd_dirobj_lock to ensure that no other thread is
2593 * updating the zil (i.e. adding in-memory log records) and the
2594 * zd_zilog_lock to block any I/O.
2596 mutex_enter(&zd
->zd_dirobj_lock
);
2597 (void) rw_wrlock(&zd
->zd_zilog_lock
);
2599 /* zfs_sb_teardown() */
2600 zil_close(zd
->zd_zilog
);
2602 /* zfsvfs_setup() */
2603 VERIFY(zil_open(os
, ztest_get_data
) == zd
->zd_zilog
);
2604 zil_replay(os
, zd
, ztest_replay_vector
);
2606 (void) rw_unlock(&zd
->zd_zilog_lock
);
2607 mutex_exit(&zd
->zd_dirobj_lock
);
2611 * Verify that we can't destroy an active pool, create an existing pool,
2612 * or create a pool with a bad vdev spec.
2616 ztest_spa_create_destroy(ztest_ds_t
*zd
, uint64_t id
)
2618 ztest_shared_opts_t
*zo
= &ztest_opts
;
2623 * Attempt to create using a bad file.
2625 nvroot
= make_vdev_root("/dev/bogus", NULL
, NULL
, 0, 0, 0, 0, 0, 1);
2626 VERIFY3U(ENOENT
, ==,
2627 spa_create("ztest_bad_file", nvroot
, NULL
, NULL
));
2628 nvlist_free(nvroot
);
2631 * Attempt to create using a bad mirror.
2633 nvroot
= make_vdev_root("/dev/bogus", NULL
, NULL
, 0, 0, 0, 0, 2, 1);
2634 VERIFY3U(ENOENT
, ==,
2635 spa_create("ztest_bad_mirror", nvroot
, NULL
, NULL
));
2636 nvlist_free(nvroot
);
2639 * Attempt to create an existing pool. It shouldn't matter
2640 * what's in the nvroot; we should fail with EEXIST.
2642 (void) rw_rdlock(&ztest_name_lock
);
2643 nvroot
= make_vdev_root("/dev/bogus", NULL
, NULL
, 0, 0, 0, 0, 0, 1);
2644 VERIFY3U(EEXIST
, ==, spa_create(zo
->zo_pool
, nvroot
, NULL
, NULL
));
2645 nvlist_free(nvroot
);
2646 VERIFY3U(0, ==, spa_open(zo
->zo_pool
, &spa
, FTAG
));
2647 VERIFY3U(EBUSY
, ==, spa_destroy(zo
->zo_pool
));
2648 spa_close(spa
, FTAG
);
2650 (void) rw_unlock(&ztest_name_lock
);
2655 ztest_spa_upgrade(ztest_ds_t
*zd
, uint64_t id
)
2658 uint64_t initial_version
= SPA_VERSION_INITIAL
;
2659 uint64_t version
, newversion
;
2660 nvlist_t
*nvroot
, *props
;
2663 mutex_enter(&ztest_vdev_lock
);
2664 name
= kmem_asprintf("%s_upgrade", ztest_opts
.zo_pool
);
2667 * Clean up from previous runs.
2669 (void) spa_destroy(name
);
2671 nvroot
= make_vdev_root(NULL
, NULL
, name
, ztest_opts
.zo_vdev_size
, 0,
2672 0, ztest_opts
.zo_raidz
, ztest_opts
.zo_mirrors
, 1);
2675 * If we're configuring a RAIDZ device then make sure that the
2676 * the initial version is capable of supporting that feature.
2678 switch (ztest_opts
.zo_raidz_parity
) {
2681 initial_version
= SPA_VERSION_INITIAL
;
2684 initial_version
= SPA_VERSION_RAIDZ2
;
2687 initial_version
= SPA_VERSION_RAIDZ3
;
2692 * Create a pool with a spa version that can be upgraded. Pick
2693 * a value between initial_version and SPA_VERSION_BEFORE_FEATURES.
2696 version
= ztest_random_spa_version(initial_version
);
2697 } while (version
> SPA_VERSION_BEFORE_FEATURES
);
2699 props
= fnvlist_alloc();
2700 fnvlist_add_uint64(props
,
2701 zpool_prop_to_name(ZPOOL_PROP_VERSION
), version
);
2702 VERIFY3S(spa_create(name
, nvroot
, props
, NULL
), ==, 0);
2703 fnvlist_free(nvroot
);
2704 fnvlist_free(props
);
2706 VERIFY3S(spa_open(name
, &spa
, FTAG
), ==, 0);
2707 VERIFY3U(spa_version(spa
), ==, version
);
2708 newversion
= ztest_random_spa_version(version
+ 1);
2710 if (ztest_opts
.zo_verbose
>= 4) {
2711 (void) printf("upgrading spa version from %llu to %llu\n",
2712 (u_longlong_t
)version
, (u_longlong_t
)newversion
);
2715 spa_upgrade(spa
, newversion
);
2716 VERIFY3U(spa_version(spa
), >, version
);
2717 VERIFY3U(spa_version(spa
), ==, fnvlist_lookup_uint64(spa
->spa_config
,
2718 zpool_prop_to_name(ZPOOL_PROP_VERSION
)));
2719 spa_close(spa
, FTAG
);
2722 mutex_exit(&ztest_vdev_lock
);
2726 vdev_lookup_by_path(vdev_t
*vd
, const char *path
)
2731 if (vd
->vdev_path
!= NULL
&& strcmp(path
, vd
->vdev_path
) == 0)
2734 for (c
= 0; c
< vd
->vdev_children
; c
++)
2735 if ((mvd
= vdev_lookup_by_path(vd
->vdev_child
[c
], path
)) !=
2743 * Find the first available hole which can be used as a top-level.
2746 find_vdev_hole(spa_t
*spa
)
2748 vdev_t
*rvd
= spa
->spa_root_vdev
;
2751 ASSERT(spa_config_held(spa
, SCL_VDEV
, RW_READER
) == SCL_VDEV
);
2753 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
2754 vdev_t
*cvd
= rvd
->vdev_child
[c
];
2756 if (cvd
->vdev_ishole
)
2763 * Verify that vdev_add() works as expected.
2767 ztest_vdev_add_remove(ztest_ds_t
*zd
, uint64_t id
)
2769 ztest_shared_t
*zs
= ztest_shared
;
2770 spa_t
*spa
= ztest_spa
;
2776 mutex_enter(&ztest_vdev_lock
);
2777 leaves
= MAX(zs
->zs_mirrors
+ zs
->zs_splits
, 1) * ztest_opts
.zo_raidz
;
2779 spa_config_enter(spa
, SCL_VDEV
, FTAG
, RW_READER
);
2781 ztest_shared
->zs_vdev_next_leaf
= find_vdev_hole(spa
) * leaves
;
2784 * If we have slogs then remove them 1/4 of the time.
2786 if (spa_has_slogs(spa
) && ztest_random(4) == 0) {
2788 * Grab the guid from the head of the log class rotor.
2790 guid
= spa_log_class(spa
)->mc_rotor
->mg_vd
->vdev_guid
;
2792 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
2795 * We have to grab the zs_name_lock as writer to
2796 * prevent a race between removing a slog (dmu_objset_find)
2797 * and destroying a dataset. Removing the slog will
2798 * grab a reference on the dataset which may cause
2799 * dsl_destroy_head() to fail with EBUSY thus
2800 * leaving the dataset in an inconsistent state.
2802 rw_wrlock(&ztest_name_lock
);
2803 error
= spa_vdev_remove(spa
, guid
, B_FALSE
);
2804 rw_unlock(&ztest_name_lock
);
2806 if (error
&& error
!= EEXIST
)
2807 fatal(0, "spa_vdev_remove() = %d", error
);
2809 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
2812 * Make 1/4 of the devices be log devices.
2814 nvroot
= make_vdev_root(NULL
, NULL
, NULL
,
2815 ztest_opts
.zo_vdev_size
, 0,
2816 ztest_random(4) == 0, ztest_opts
.zo_raidz
,
2819 error
= spa_vdev_add(spa
, nvroot
);
2820 nvlist_free(nvroot
);
2822 if (error
== ENOSPC
)
2823 ztest_record_enospc("spa_vdev_add");
2824 else if (error
!= 0)
2825 fatal(0, "spa_vdev_add() = %d", error
);
2828 mutex_exit(&ztest_vdev_lock
);
2832 * Verify that adding/removing aux devices (l2arc, hot spare) works as expected.
2836 ztest_vdev_aux_add_remove(ztest_ds_t
*zd
, uint64_t id
)
2838 ztest_shared_t
*zs
= ztest_shared
;
2839 spa_t
*spa
= ztest_spa
;
2840 vdev_t
*rvd
= spa
->spa_root_vdev
;
2841 spa_aux_vdev_t
*sav
;
2847 path
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
2849 if (ztest_random(2) == 0) {
2850 sav
= &spa
->spa_spares
;
2851 aux
= ZPOOL_CONFIG_SPARES
;
2853 sav
= &spa
->spa_l2cache
;
2854 aux
= ZPOOL_CONFIG_L2CACHE
;
2857 mutex_enter(&ztest_vdev_lock
);
2859 spa_config_enter(spa
, SCL_VDEV
, FTAG
, RW_READER
);
2861 if (sav
->sav_count
!= 0 && ztest_random(4) == 0) {
2863 * Pick a random device to remove.
2865 guid
= sav
->sav_vdevs
[ztest_random(sav
->sav_count
)]->vdev_guid
;
2868 * Find an unused device we can add.
2870 zs
->zs_vdev_aux
= 0;
2873 (void) snprintf(path
, MAXPATHLEN
, ztest_aux_template
,
2874 ztest_opts
.zo_dir
, ztest_opts
.zo_pool
, aux
,
2876 for (c
= 0; c
< sav
->sav_count
; c
++)
2877 if (strcmp(sav
->sav_vdevs
[c
]->vdev_path
,
2880 if (c
== sav
->sav_count
&&
2881 vdev_lookup_by_path(rvd
, path
) == NULL
)
2887 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
2893 nvlist_t
*nvroot
= make_vdev_root(NULL
, aux
, NULL
,
2894 (ztest_opts
.zo_vdev_size
* 5) / 4, 0, 0, 0, 0, 1);
2895 error
= spa_vdev_add(spa
, nvroot
);
2897 fatal(0, "spa_vdev_add(%p) = %d", nvroot
, error
);
2898 nvlist_free(nvroot
);
2901 * Remove an existing device. Sometimes, dirty its
2902 * vdev state first to make sure we handle removal
2903 * of devices that have pending state changes.
2905 if (ztest_random(2) == 0)
2906 (void) vdev_online(spa
, guid
, 0, NULL
);
2908 error
= spa_vdev_remove(spa
, guid
, B_FALSE
);
2909 if (error
!= 0 && error
!= EBUSY
)
2910 fatal(0, "spa_vdev_remove(%llu) = %d", guid
, error
);
2913 mutex_exit(&ztest_vdev_lock
);
2915 umem_free(path
, MAXPATHLEN
);
2919 * split a pool if it has mirror tlvdevs
2923 ztest_split_pool(ztest_ds_t
*zd
, uint64_t id
)
2925 ztest_shared_t
*zs
= ztest_shared
;
2926 spa_t
*spa
= ztest_spa
;
2927 vdev_t
*rvd
= spa
->spa_root_vdev
;
2928 nvlist_t
*tree
, **child
, *config
, *split
, **schild
;
2929 uint_t c
, children
, schildren
= 0, lastlogid
= 0;
2932 mutex_enter(&ztest_vdev_lock
);
2934 /* ensure we have a useable config; mirrors of raidz aren't supported */
2935 if (zs
->zs_mirrors
< 3 || ztest_opts
.zo_raidz
> 1) {
2936 mutex_exit(&ztest_vdev_lock
);
2940 /* clean up the old pool, if any */
2941 (void) spa_destroy("splitp");
2943 spa_config_enter(spa
, SCL_VDEV
, FTAG
, RW_READER
);
2945 /* generate a config from the existing config */
2946 mutex_enter(&spa
->spa_props_lock
);
2947 VERIFY(nvlist_lookup_nvlist(spa
->spa_config
, ZPOOL_CONFIG_VDEV_TREE
,
2949 mutex_exit(&spa
->spa_props_lock
);
2951 VERIFY(nvlist_lookup_nvlist_array(tree
, ZPOOL_CONFIG_CHILDREN
, &child
,
2954 schild
= malloc(rvd
->vdev_children
* sizeof (nvlist_t
*));
2955 for (c
= 0; c
< children
; c
++) {
2956 vdev_t
*tvd
= rvd
->vdev_child
[c
];
2960 if (tvd
->vdev_islog
|| tvd
->vdev_ops
== &vdev_hole_ops
) {
2961 VERIFY(nvlist_alloc(&schild
[schildren
], NV_UNIQUE_NAME
,
2963 VERIFY(nvlist_add_string(schild
[schildren
],
2964 ZPOOL_CONFIG_TYPE
, VDEV_TYPE_HOLE
) == 0);
2965 VERIFY(nvlist_add_uint64(schild
[schildren
],
2966 ZPOOL_CONFIG_IS_HOLE
, 1) == 0);
2968 lastlogid
= schildren
;
2973 VERIFY(nvlist_lookup_nvlist_array(child
[c
],
2974 ZPOOL_CONFIG_CHILDREN
, &mchild
, &mchildren
) == 0);
2975 VERIFY(nvlist_dup(mchild
[0], &schild
[schildren
++], 0) == 0);
2978 /* OK, create a config that can be used to split */
2979 VERIFY(nvlist_alloc(&split
, NV_UNIQUE_NAME
, 0) == 0);
2980 VERIFY(nvlist_add_string(split
, ZPOOL_CONFIG_TYPE
,
2981 VDEV_TYPE_ROOT
) == 0);
2982 VERIFY(nvlist_add_nvlist_array(split
, ZPOOL_CONFIG_CHILDREN
, schild
,
2983 lastlogid
!= 0 ? lastlogid
: schildren
) == 0);
2985 VERIFY(nvlist_alloc(&config
, NV_UNIQUE_NAME
, 0) == 0);
2986 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, split
) == 0);
2988 for (c
= 0; c
< schildren
; c
++)
2989 nvlist_free(schild
[c
]);
2993 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
2995 (void) rw_wrlock(&ztest_name_lock
);
2996 error
= spa_vdev_split_mirror(spa
, "splitp", config
, NULL
, B_FALSE
);
2997 (void) rw_unlock(&ztest_name_lock
);
2999 nvlist_free(config
);
3002 (void) printf("successful split - results:\n");
3003 mutex_enter(&spa_namespace_lock
);
3004 show_pool_stats(spa
);
3005 show_pool_stats(spa_lookup("splitp"));
3006 mutex_exit(&spa_namespace_lock
);
3010 mutex_exit(&ztest_vdev_lock
);
3015 * Verify that we can attach and detach devices.
3019 ztest_vdev_attach_detach(ztest_ds_t
*zd
, uint64_t id
)
3021 ztest_shared_t
*zs
= ztest_shared
;
3022 spa_t
*spa
= ztest_spa
;
3023 spa_aux_vdev_t
*sav
= &spa
->spa_spares
;
3024 vdev_t
*rvd
= spa
->spa_root_vdev
;
3025 vdev_t
*oldvd
, *newvd
, *pvd
;
3029 uint64_t ashift
= ztest_get_ashift();
3030 uint64_t oldguid
, pguid
;
3031 uint64_t oldsize
, newsize
;
3032 char *oldpath
, *newpath
;
3034 int oldvd_has_siblings
= B_FALSE
;
3035 int newvd_is_spare
= B_FALSE
;
3037 int error
, expected_error
;
3039 oldpath
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
3040 newpath
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
3042 mutex_enter(&ztest_vdev_lock
);
3043 leaves
= MAX(zs
->zs_mirrors
, 1) * ztest_opts
.zo_raidz
;
3045 spa_config_enter(spa
, SCL_VDEV
, FTAG
, RW_READER
);
3048 * Decide whether to do an attach or a replace.
3050 replacing
= ztest_random(2);
3053 * Pick a random top-level vdev.
3055 top
= ztest_random_vdev_top(spa
, B_TRUE
);
3058 * Pick a random leaf within it.
3060 leaf
= ztest_random(leaves
);
3065 oldvd
= rvd
->vdev_child
[top
];
3066 if (zs
->zs_mirrors
>= 1) {
3067 ASSERT(oldvd
->vdev_ops
== &vdev_mirror_ops
);
3068 ASSERT(oldvd
->vdev_children
>= zs
->zs_mirrors
);
3069 oldvd
= oldvd
->vdev_child
[leaf
/ ztest_opts
.zo_raidz
];
3071 if (ztest_opts
.zo_raidz
> 1) {
3072 ASSERT(oldvd
->vdev_ops
== &vdev_raidz_ops
);
3073 ASSERT(oldvd
->vdev_children
== ztest_opts
.zo_raidz
);
3074 oldvd
= oldvd
->vdev_child
[leaf
% ztest_opts
.zo_raidz
];
3078 * If we're already doing an attach or replace, oldvd may be a
3079 * mirror vdev -- in which case, pick a random child.
3081 while (oldvd
->vdev_children
!= 0) {
3082 oldvd_has_siblings
= B_TRUE
;
3083 ASSERT(oldvd
->vdev_children
>= 2);
3084 oldvd
= oldvd
->vdev_child
[ztest_random(oldvd
->vdev_children
)];
3087 oldguid
= oldvd
->vdev_guid
;
3088 oldsize
= vdev_get_min_asize(oldvd
);
3089 oldvd_is_log
= oldvd
->vdev_top
->vdev_islog
;
3090 (void) strcpy(oldpath
, oldvd
->vdev_path
);
3091 pvd
= oldvd
->vdev_parent
;
3092 pguid
= pvd
->vdev_guid
;
3095 * If oldvd has siblings, then half of the time, detach it.
3097 if (oldvd_has_siblings
&& ztest_random(2) == 0) {
3098 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
3099 error
= spa_vdev_detach(spa
, oldguid
, pguid
, B_FALSE
);
3100 if (error
!= 0 && error
!= ENODEV
&& error
!= EBUSY
&&
3102 fatal(0, "detach (%s) returned %d", oldpath
, error
);
3107 * For the new vdev, choose with equal probability between the two
3108 * standard paths (ending in either 'a' or 'b') or a random hot spare.
3110 if (sav
->sav_count
!= 0 && ztest_random(3) == 0) {
3111 newvd
= sav
->sav_vdevs
[ztest_random(sav
->sav_count
)];
3112 newvd_is_spare
= B_TRUE
;
3113 (void) strcpy(newpath
, newvd
->vdev_path
);
3115 (void) snprintf(newpath
, MAXPATHLEN
, ztest_dev_template
,
3116 ztest_opts
.zo_dir
, ztest_opts
.zo_pool
,
3117 top
* leaves
+ leaf
);
3118 if (ztest_random(2) == 0)
3119 newpath
[strlen(newpath
) - 1] = 'b';
3120 newvd
= vdev_lookup_by_path(rvd
, newpath
);
3124 newsize
= vdev_get_min_asize(newvd
);
3127 * Make newsize a little bigger or smaller than oldsize.
3128 * If it's smaller, the attach should fail.
3129 * If it's larger, and we're doing a replace,
3130 * we should get dynamic LUN growth when we're done.
3132 newsize
= 10 * oldsize
/ (9 + ztest_random(3));
3136 * If pvd is not a mirror or root, the attach should fail with ENOTSUP,
3137 * unless it's a replace; in that case any non-replacing parent is OK.
3139 * If newvd is already part of the pool, it should fail with EBUSY.
3141 * If newvd is too small, it should fail with EOVERFLOW.
3143 if (pvd
->vdev_ops
!= &vdev_mirror_ops
&&
3144 pvd
->vdev_ops
!= &vdev_root_ops
&& (!replacing
||
3145 pvd
->vdev_ops
== &vdev_replacing_ops
||
3146 pvd
->vdev_ops
== &vdev_spare_ops
))
3147 expected_error
= ENOTSUP
;
3148 else if (newvd_is_spare
&& (!replacing
|| oldvd_is_log
))
3149 expected_error
= ENOTSUP
;
3150 else if (newvd
== oldvd
)
3151 expected_error
= replacing
? 0 : EBUSY
;
3152 else if (vdev_lookup_by_path(rvd
, newpath
) != NULL
)
3153 expected_error
= EBUSY
;
3154 else if (newsize
< oldsize
)
3155 expected_error
= EOVERFLOW
;
3156 else if (ashift
> oldvd
->vdev_top
->vdev_ashift
)
3157 expected_error
= EDOM
;
3161 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
3164 * Build the nvlist describing newpath.
3166 root
= make_vdev_root(newpath
, NULL
, NULL
, newvd
== NULL
? newsize
: 0,
3167 ashift
, 0, 0, 0, 1);
3169 error
= spa_vdev_attach(spa
, oldguid
, root
, replacing
);
3174 * If our parent was the replacing vdev, but the replace completed,
3175 * then instead of failing with ENOTSUP we may either succeed,
3176 * fail with ENODEV, or fail with EOVERFLOW.
3178 if (expected_error
== ENOTSUP
&&
3179 (error
== 0 || error
== ENODEV
|| error
== EOVERFLOW
))
3180 expected_error
= error
;
3183 * If someone grew the LUN, the replacement may be too small.
3185 if (error
== EOVERFLOW
|| error
== EBUSY
)
3186 expected_error
= error
;
3188 /* XXX workaround 6690467 */
3189 if (error
!= expected_error
&& expected_error
!= EBUSY
) {
3190 fatal(0, "attach (%s %llu, %s %llu, %d) "
3191 "returned %d, expected %d",
3192 oldpath
, oldsize
, newpath
,
3193 newsize
, replacing
, error
, expected_error
);
3196 mutex_exit(&ztest_vdev_lock
);
3198 umem_free(oldpath
, MAXPATHLEN
);
3199 umem_free(newpath
, MAXPATHLEN
);
3203 * Callback function which expands the physical size of the vdev.
3206 grow_vdev(vdev_t
*vd
, void *arg
)
3208 ASSERTV(spa_t
*spa
= vd
->vdev_spa
);
3209 size_t *newsize
= arg
;
3213 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_READER
) == SCL_STATE
);
3214 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
3216 if ((fd
= open(vd
->vdev_path
, O_RDWR
)) == -1)
3219 fsize
= lseek(fd
, 0, SEEK_END
);
3220 VERIFY(ftruncate(fd
, *newsize
) == 0);
3222 if (ztest_opts
.zo_verbose
>= 6) {
3223 (void) printf("%s grew from %lu to %lu bytes\n",
3224 vd
->vdev_path
, (ulong_t
)fsize
, (ulong_t
)*newsize
);
3231 * Callback function which expands a given vdev by calling vdev_online().
3235 online_vdev(vdev_t
*vd
, void *arg
)
3237 spa_t
*spa
= vd
->vdev_spa
;
3238 vdev_t
*tvd
= vd
->vdev_top
;
3239 uint64_t guid
= vd
->vdev_guid
;
3240 uint64_t generation
= spa
->spa_config_generation
+ 1;
3241 vdev_state_t newstate
= VDEV_STATE_UNKNOWN
;
3244 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_READER
) == SCL_STATE
);
3245 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
3247 /* Calling vdev_online will initialize the new metaslabs */
3248 spa_config_exit(spa
, SCL_STATE
, spa
);
3249 error
= vdev_online(spa
, guid
, ZFS_ONLINE_EXPAND
, &newstate
);
3250 spa_config_enter(spa
, SCL_STATE
, spa
, RW_READER
);
3253 * If vdev_online returned an error or the underlying vdev_open
3254 * failed then we abort the expand. The only way to know that
3255 * vdev_open fails is by checking the returned newstate.
3257 if (error
|| newstate
!= VDEV_STATE_HEALTHY
) {
3258 if (ztest_opts
.zo_verbose
>= 5) {
3259 (void) printf("Unable to expand vdev, state %llu, "
3260 "error %d\n", (u_longlong_t
)newstate
, error
);
3264 ASSERT3U(newstate
, ==, VDEV_STATE_HEALTHY
);
3267 * Since we dropped the lock we need to ensure that we're
3268 * still talking to the original vdev. It's possible this
3269 * vdev may have been detached/replaced while we were
3270 * trying to online it.
3272 if (generation
!= spa
->spa_config_generation
) {
3273 if (ztest_opts
.zo_verbose
>= 5) {
3274 (void) printf("vdev configuration has changed, "
3275 "guid %llu, state %llu, expected gen %llu, "
3278 (u_longlong_t
)tvd
->vdev_state
,
3279 (u_longlong_t
)generation
,
3280 (u_longlong_t
)spa
->spa_config_generation
);
3288 * Traverse the vdev tree calling the supplied function.
3289 * We continue to walk the tree until we either have walked all
3290 * children or we receive a non-NULL return from the callback.
3291 * If a NULL callback is passed, then we just return back the first
3292 * leaf vdev we encounter.
3295 vdev_walk_tree(vdev_t
*vd
, vdev_t
*(*func
)(vdev_t
*, void *), void *arg
)
3299 if (vd
->vdev_ops
->vdev_op_leaf
) {
3303 return (func(vd
, arg
));
3306 for (c
= 0; c
< vd
->vdev_children
; c
++) {
3307 vdev_t
*cvd
= vd
->vdev_child
[c
];
3308 if ((cvd
= vdev_walk_tree(cvd
, func
, arg
)) != NULL
)
3315 * Verify that dynamic LUN growth works as expected.
3319 ztest_vdev_LUN_growth(ztest_ds_t
*zd
, uint64_t id
)
3321 spa_t
*spa
= ztest_spa
;
3323 metaslab_class_t
*mc
;
3324 metaslab_group_t
*mg
;
3325 size_t psize
, newsize
;
3327 uint64_t old_class_space
, new_class_space
, old_ms_count
, new_ms_count
;
3329 mutex_enter(&ztest_vdev_lock
);
3330 spa_config_enter(spa
, SCL_STATE
, spa
, RW_READER
);
3332 top
= ztest_random_vdev_top(spa
, B_TRUE
);
3334 tvd
= spa
->spa_root_vdev
->vdev_child
[top
];
3337 old_ms_count
= tvd
->vdev_ms_count
;
3338 old_class_space
= metaslab_class_get_space(mc
);
3341 * Determine the size of the first leaf vdev associated with
3342 * our top-level device.
3344 vd
= vdev_walk_tree(tvd
, NULL
, NULL
);
3345 ASSERT3P(vd
, !=, NULL
);
3346 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
3348 psize
= vd
->vdev_psize
;
3351 * We only try to expand the vdev if it's healthy, less than 4x its
3352 * original size, and it has a valid psize.
3354 if (tvd
->vdev_state
!= VDEV_STATE_HEALTHY
||
3355 psize
== 0 || psize
>= 4 * ztest_opts
.zo_vdev_size
) {
3356 spa_config_exit(spa
, SCL_STATE
, spa
);
3357 mutex_exit(&ztest_vdev_lock
);
3361 newsize
= psize
+ psize
/ 8;
3362 ASSERT3U(newsize
, >, psize
);
3364 if (ztest_opts
.zo_verbose
>= 6) {
3365 (void) printf("Expanding LUN %s from %lu to %lu\n",
3366 vd
->vdev_path
, (ulong_t
)psize
, (ulong_t
)newsize
);
3370 * Growing the vdev is a two step process:
3371 * 1). expand the physical size (i.e. relabel)
3372 * 2). online the vdev to create the new metaslabs
3374 if (vdev_walk_tree(tvd
, grow_vdev
, &newsize
) != NULL
||
3375 vdev_walk_tree(tvd
, online_vdev
, NULL
) != NULL
||
3376 tvd
->vdev_state
!= VDEV_STATE_HEALTHY
) {
3377 if (ztest_opts
.zo_verbose
>= 5) {
3378 (void) printf("Could not expand LUN because "
3379 "the vdev configuration changed.\n");
3381 spa_config_exit(spa
, SCL_STATE
, spa
);
3382 mutex_exit(&ztest_vdev_lock
);
3386 spa_config_exit(spa
, SCL_STATE
, spa
);
3389 * Expanding the LUN will update the config asynchronously,
3390 * thus we must wait for the async thread to complete any
3391 * pending tasks before proceeding.
3395 mutex_enter(&spa
->spa_async_lock
);
3396 done
= (spa
->spa_async_thread
== NULL
&& !spa
->spa_async_tasks
);
3397 mutex_exit(&spa
->spa_async_lock
);
3400 txg_wait_synced(spa_get_dsl(spa
), 0);
3401 (void) poll(NULL
, 0, 100);
3404 spa_config_enter(spa
, SCL_STATE
, spa
, RW_READER
);
3406 tvd
= spa
->spa_root_vdev
->vdev_child
[top
];
3407 new_ms_count
= tvd
->vdev_ms_count
;
3408 new_class_space
= metaslab_class_get_space(mc
);
3410 if (tvd
->vdev_mg
!= mg
|| mg
->mg_class
!= mc
) {
3411 if (ztest_opts
.zo_verbose
>= 5) {
3412 (void) printf("Could not verify LUN expansion due to "
3413 "intervening vdev offline or remove.\n");
3415 spa_config_exit(spa
, SCL_STATE
, spa
);
3416 mutex_exit(&ztest_vdev_lock
);
3421 * Make sure we were able to grow the vdev.
3423 if (new_ms_count
<= old_ms_count
)
3424 fatal(0, "LUN expansion failed: ms_count %llu <= %llu\n",
3425 old_ms_count
, new_ms_count
);
3428 * Make sure we were able to grow the pool.
3430 if (new_class_space
<= old_class_space
)
3431 fatal(0, "LUN expansion failed: class_space %llu <= %llu\n",
3432 old_class_space
, new_class_space
);
3434 if (ztest_opts
.zo_verbose
>= 5) {
3435 char oldnumbuf
[6], newnumbuf
[6];
3437 nicenum(old_class_space
, oldnumbuf
);
3438 nicenum(new_class_space
, newnumbuf
);
3439 (void) printf("%s grew from %s to %s\n",
3440 spa
->spa_name
, oldnumbuf
, newnumbuf
);
3443 spa_config_exit(spa
, SCL_STATE
, spa
);
3444 mutex_exit(&ztest_vdev_lock
);
3448 * Verify that dmu_objset_{create,destroy,open,close} work as expected.
3452 ztest_objset_create_cb(objset_t
*os
, void *arg
, cred_t
*cr
, dmu_tx_t
*tx
)
3455 * Create the objects common to all ztest datasets.
3457 VERIFY(zap_create_claim(os
, ZTEST_DIROBJ
,
3458 DMU_OT_ZAP_OTHER
, DMU_OT_NONE
, 0, tx
) == 0);
3462 ztest_dataset_create(char *dsname
)
3464 uint64_t zilset
= ztest_random(100);
3465 int err
= dmu_objset_create(dsname
, DMU_OST_OTHER
, 0,
3466 ztest_objset_create_cb
, NULL
);
3468 if (err
|| zilset
< 80)
3471 if (ztest_opts
.zo_verbose
>= 5)
3472 (void) printf("Setting dataset %s to sync always\n", dsname
);
3473 return (ztest_dsl_prop_set_uint64(dsname
, ZFS_PROP_SYNC
,
3474 ZFS_SYNC_ALWAYS
, B_FALSE
));
3479 ztest_objset_destroy_cb(const char *name
, void *arg
)
3482 dmu_object_info_t doi
;
3486 * Verify that the dataset contains a directory object.
3488 VERIFY0(dmu_objset_own(name
, DMU_OST_OTHER
, B_TRUE
, FTAG
, &os
));
3489 error
= dmu_object_info(os
, ZTEST_DIROBJ
, &doi
);
3490 if (error
!= ENOENT
) {
3491 /* We could have crashed in the middle of destroying it */
3493 ASSERT3U(doi
.doi_type
, ==, DMU_OT_ZAP_OTHER
);
3494 ASSERT3S(doi
.doi_physical_blocks_512
, >=, 0);
3496 dmu_objset_disown(os
, FTAG
);
3499 * Destroy the dataset.
3501 if (strchr(name
, '@') != NULL
) {
3502 VERIFY0(dsl_destroy_snapshot(name
, B_TRUE
));
3504 error
= dsl_destroy_head(name
);
3505 /* There could be a hold on this dataset */
3513 ztest_snapshot_create(char *osname
, uint64_t id
)
3515 char snapname
[ZFS_MAX_DATASET_NAME_LEN
];
3518 (void) snprintf(snapname
, sizeof (snapname
), "%llu", (u_longlong_t
)id
);
3520 error
= dmu_objset_snapshot_one(osname
, snapname
);
3521 if (error
== ENOSPC
) {
3522 ztest_record_enospc(FTAG
);
3525 if (error
!= 0 && error
!= EEXIST
) {
3526 fatal(0, "ztest_snapshot_create(%s@%s) = %d", osname
,
3533 ztest_snapshot_destroy(char *osname
, uint64_t id
)
3535 char snapname
[ZFS_MAX_DATASET_NAME_LEN
];
3538 (void) snprintf(snapname
, sizeof (snapname
), "%s@%llu", osname
,
3541 error
= dsl_destroy_snapshot(snapname
, B_FALSE
);
3542 if (error
!= 0 && error
!= ENOENT
)
3543 fatal(0, "ztest_snapshot_destroy(%s) = %d", snapname
, error
);
3549 ztest_dmu_objset_create_destroy(ztest_ds_t
*zd
, uint64_t id
)
3555 char name
[ZFS_MAX_DATASET_NAME_LEN
];
3559 zdtmp
= umem_alloc(sizeof (ztest_ds_t
), UMEM_NOFAIL
);
3561 (void) rw_rdlock(&ztest_name_lock
);
3563 (void) snprintf(name
, sizeof (name
), "%s/temp_%llu",
3564 ztest_opts
.zo_pool
, (u_longlong_t
)id
);
3567 * If this dataset exists from a previous run, process its replay log
3568 * half of the time. If we don't replay it, then dsl_destroy_head()
3569 * (invoked from ztest_objset_destroy_cb()) should just throw it away.
3571 if (ztest_random(2) == 0 &&
3572 dmu_objset_own(name
, DMU_OST_OTHER
, B_FALSE
, FTAG
, &os
) == 0) {
3573 ztest_zd_init(zdtmp
, NULL
, os
);
3574 zil_replay(os
, zdtmp
, ztest_replay_vector
);
3575 ztest_zd_fini(zdtmp
);
3576 dmu_objset_disown(os
, FTAG
);
3580 * There may be an old instance of the dataset we're about to
3581 * create lying around from a previous run. If so, destroy it
3582 * and all of its snapshots.
3584 (void) dmu_objset_find(name
, ztest_objset_destroy_cb
, NULL
,
3585 DS_FIND_CHILDREN
| DS_FIND_SNAPSHOTS
);
3588 * Verify that the destroyed dataset is no longer in the namespace.
3590 VERIFY3U(ENOENT
, ==, dmu_objset_own(name
, DMU_OST_OTHER
, B_TRUE
,
3594 * Verify that we can create a new dataset.
3596 error
= ztest_dataset_create(name
);
3598 if (error
== ENOSPC
) {
3599 ztest_record_enospc(FTAG
);
3602 fatal(0, "dmu_objset_create(%s) = %d", name
, error
);
3605 VERIFY0(dmu_objset_own(name
, DMU_OST_OTHER
, B_FALSE
, FTAG
, &os
));
3607 ztest_zd_init(zdtmp
, NULL
, os
);
3610 * Open the intent log for it.
3612 zilog
= zil_open(os
, ztest_get_data
);
3615 * Put some objects in there, do a little I/O to them,
3616 * and randomly take a couple of snapshots along the way.
3618 iters
= ztest_random(5);
3619 for (i
= 0; i
< iters
; i
++) {
3620 ztest_dmu_object_alloc_free(zdtmp
, id
);
3621 if (ztest_random(iters
) == 0)
3622 (void) ztest_snapshot_create(name
, i
);
3626 * Verify that we cannot create an existing dataset.
3628 VERIFY3U(EEXIST
, ==,
3629 dmu_objset_create(name
, DMU_OST_OTHER
, 0, NULL
, NULL
));
3632 * Verify that we can hold an objset that is also owned.
3634 VERIFY3U(0, ==, dmu_objset_hold(name
, FTAG
, &os2
));
3635 dmu_objset_rele(os2
, FTAG
);
3638 * Verify that we cannot own an objset that is already owned.
3641 dmu_objset_own(name
, DMU_OST_OTHER
, B_FALSE
, FTAG
, &os2
));
3644 dmu_objset_disown(os
, FTAG
);
3645 ztest_zd_fini(zdtmp
);
3647 (void) rw_unlock(&ztest_name_lock
);
3649 umem_free(zdtmp
, sizeof (ztest_ds_t
));
3653 * Verify that dmu_snapshot_{create,destroy,open,close} work as expected.
3656 ztest_dmu_snapshot_create_destroy(ztest_ds_t
*zd
, uint64_t id
)
3658 (void) rw_rdlock(&ztest_name_lock
);
3659 (void) ztest_snapshot_destroy(zd
->zd_name
, id
);
3660 (void) ztest_snapshot_create(zd
->zd_name
, id
);
3661 (void) rw_unlock(&ztest_name_lock
);
3665 * Cleanup non-standard snapshots and clones.
3668 ztest_dsl_dataset_cleanup(char *osname
, uint64_t id
)
3677 snap1name
= umem_alloc(ZFS_MAX_DATASET_NAME_LEN
, UMEM_NOFAIL
);
3678 clone1name
= umem_alloc(ZFS_MAX_DATASET_NAME_LEN
, UMEM_NOFAIL
);
3679 snap2name
= umem_alloc(ZFS_MAX_DATASET_NAME_LEN
, UMEM_NOFAIL
);
3680 clone2name
= umem_alloc(ZFS_MAX_DATASET_NAME_LEN
, UMEM_NOFAIL
);
3681 snap3name
= umem_alloc(ZFS_MAX_DATASET_NAME_LEN
, UMEM_NOFAIL
);
3683 (void) snprintf(snap1name
, ZFS_MAX_DATASET_NAME_LEN
,
3684 "%s@s1_%llu", osname
, (u_longlong_t
)id
);
3685 (void) snprintf(clone1name
, ZFS_MAX_DATASET_NAME_LEN
,
3686 "%s/c1_%llu", osname
, (u_longlong_t
)id
);
3687 (void) snprintf(snap2name
, ZFS_MAX_DATASET_NAME_LEN
,
3688 "%s@s2_%llu", clone1name
, (u_longlong_t
)id
);
3689 (void) snprintf(clone2name
, ZFS_MAX_DATASET_NAME_LEN
,
3690 "%s/c2_%llu", osname
, (u_longlong_t
)id
);
3691 (void) snprintf(snap3name
, ZFS_MAX_DATASET_NAME_LEN
,
3692 "%s@s3_%llu", clone1name
, (u_longlong_t
)id
);
3694 error
= dsl_destroy_head(clone2name
);
3695 if (error
&& error
!= ENOENT
)
3696 fatal(0, "dsl_destroy_head(%s) = %d", clone2name
, error
);
3697 error
= dsl_destroy_snapshot(snap3name
, B_FALSE
);
3698 if (error
&& error
!= ENOENT
)
3699 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap3name
, error
);
3700 error
= dsl_destroy_snapshot(snap2name
, B_FALSE
);
3701 if (error
&& error
!= ENOENT
)
3702 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap2name
, error
);
3703 error
= dsl_destroy_head(clone1name
);
3704 if (error
&& error
!= ENOENT
)
3705 fatal(0, "dsl_destroy_head(%s) = %d", clone1name
, error
);
3706 error
= dsl_destroy_snapshot(snap1name
, B_FALSE
);
3707 if (error
&& error
!= ENOENT
)
3708 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap1name
, error
);
3710 umem_free(snap1name
, ZFS_MAX_DATASET_NAME_LEN
);
3711 umem_free(clone1name
, ZFS_MAX_DATASET_NAME_LEN
);
3712 umem_free(snap2name
, ZFS_MAX_DATASET_NAME_LEN
);
3713 umem_free(clone2name
, ZFS_MAX_DATASET_NAME_LEN
);
3714 umem_free(snap3name
, ZFS_MAX_DATASET_NAME_LEN
);
3718 * Verify dsl_dataset_promote handles EBUSY
3721 ztest_dsl_dataset_promote_busy(ztest_ds_t
*zd
, uint64_t id
)
3729 char *osname
= zd
->zd_name
;
3732 snap1name
= umem_alloc(ZFS_MAX_DATASET_NAME_LEN
, UMEM_NOFAIL
);
3733 clone1name
= umem_alloc(ZFS_MAX_DATASET_NAME_LEN
, UMEM_NOFAIL
);
3734 snap2name
= umem_alloc(ZFS_MAX_DATASET_NAME_LEN
, UMEM_NOFAIL
);
3735 clone2name
= umem_alloc(ZFS_MAX_DATASET_NAME_LEN
, UMEM_NOFAIL
);
3736 snap3name
= umem_alloc(ZFS_MAX_DATASET_NAME_LEN
, UMEM_NOFAIL
);
3738 (void) rw_rdlock(&ztest_name_lock
);
3740 ztest_dsl_dataset_cleanup(osname
, id
);
3742 (void) snprintf(snap1name
, ZFS_MAX_DATASET_NAME_LEN
,
3743 "%s@s1_%llu", osname
, (u_longlong_t
)id
);
3744 (void) snprintf(clone1name
, ZFS_MAX_DATASET_NAME_LEN
,
3745 "%s/c1_%llu", osname
, (u_longlong_t
)id
);
3746 (void) snprintf(snap2name
, ZFS_MAX_DATASET_NAME_LEN
,
3747 "%s@s2_%llu", clone1name
, (u_longlong_t
)id
);
3748 (void) snprintf(clone2name
, ZFS_MAX_DATASET_NAME_LEN
,
3749 "%s/c2_%llu", osname
, (u_longlong_t
)id
);
3750 (void) snprintf(snap3name
, ZFS_MAX_DATASET_NAME_LEN
,
3751 "%s@s3_%llu", clone1name
, (u_longlong_t
)id
);
3753 error
= dmu_objset_snapshot_one(osname
, strchr(snap1name
, '@') + 1);
3754 if (error
&& error
!= EEXIST
) {
3755 if (error
== ENOSPC
) {
3756 ztest_record_enospc(FTAG
);
3759 fatal(0, "dmu_take_snapshot(%s) = %d", snap1name
, error
);
3762 error
= dmu_objset_clone(clone1name
, snap1name
);
3764 if (error
== ENOSPC
) {
3765 ztest_record_enospc(FTAG
);
3768 fatal(0, "dmu_objset_create(%s) = %d", clone1name
, error
);
3771 error
= dmu_objset_snapshot_one(clone1name
, strchr(snap2name
, '@') + 1);
3772 if (error
&& error
!= EEXIST
) {
3773 if (error
== ENOSPC
) {
3774 ztest_record_enospc(FTAG
);
3777 fatal(0, "dmu_open_snapshot(%s) = %d", snap2name
, error
);
3780 error
= dmu_objset_snapshot_one(clone1name
, strchr(snap3name
, '@') + 1);
3781 if (error
&& error
!= EEXIST
) {
3782 if (error
== ENOSPC
) {
3783 ztest_record_enospc(FTAG
);
3786 fatal(0, "dmu_open_snapshot(%s) = %d", snap3name
, error
);
3789 error
= dmu_objset_clone(clone2name
, snap3name
);
3791 if (error
== ENOSPC
) {
3792 ztest_record_enospc(FTAG
);
3795 fatal(0, "dmu_objset_create(%s) = %d", clone2name
, error
);
3798 error
= dmu_objset_own(snap2name
, DMU_OST_ANY
, B_TRUE
, FTAG
, &os
);
3800 fatal(0, "dmu_objset_own(%s) = %d", snap2name
, error
);
3801 error
= dsl_dataset_promote(clone2name
, NULL
);
3802 if (error
== ENOSPC
) {
3803 dmu_objset_disown(os
, FTAG
);
3804 ztest_record_enospc(FTAG
);
3808 fatal(0, "dsl_dataset_promote(%s), %d, not EBUSY", clone2name
,
3810 dmu_objset_disown(os
, FTAG
);
3813 ztest_dsl_dataset_cleanup(osname
, id
);
3815 (void) rw_unlock(&ztest_name_lock
);
3817 umem_free(snap1name
, ZFS_MAX_DATASET_NAME_LEN
);
3818 umem_free(clone1name
, ZFS_MAX_DATASET_NAME_LEN
);
3819 umem_free(snap2name
, ZFS_MAX_DATASET_NAME_LEN
);
3820 umem_free(clone2name
, ZFS_MAX_DATASET_NAME_LEN
);
3821 umem_free(snap3name
, ZFS_MAX_DATASET_NAME_LEN
);
3824 #undef OD_ARRAY_SIZE
3825 #define OD_ARRAY_SIZE 4
3828 * Verify that dmu_object_{alloc,free} work as expected.
3831 ztest_dmu_object_alloc_free(ztest_ds_t
*zd
, uint64_t id
)
3838 size
= sizeof (ztest_od_t
) * OD_ARRAY_SIZE
;
3839 od
= umem_alloc(size
, UMEM_NOFAIL
);
3840 batchsize
= OD_ARRAY_SIZE
;
3842 for (b
= 0; b
< batchsize
; b
++)
3843 ztest_od_init(od
+ b
, id
, FTAG
, b
, DMU_OT_UINT64_OTHER
,
3847 * Destroy the previous batch of objects, create a new batch,
3848 * and do some I/O on the new objects.
3850 if (ztest_object_init(zd
, od
, size
, B_TRUE
) != 0)
3853 while (ztest_random(4 * batchsize
) != 0)
3854 ztest_io(zd
, od
[ztest_random(batchsize
)].od_object
,
3855 ztest_random(ZTEST_RANGE_LOCKS
) << SPA_MAXBLOCKSHIFT
);
3857 umem_free(od
, size
);
3860 #undef OD_ARRAY_SIZE
3861 #define OD_ARRAY_SIZE 2
3864 * Verify that dmu_{read,write} work as expected.
3867 ztest_dmu_read_write(ztest_ds_t
*zd
, uint64_t id
)
3872 objset_t
*os
= zd
->zd_os
;
3873 size
= sizeof (ztest_od_t
) * OD_ARRAY_SIZE
;
3874 od
= umem_alloc(size
, UMEM_NOFAIL
);
3876 int i
, freeit
, error
;
3878 bufwad_t
*packbuf
, *bigbuf
, *pack
, *bigH
, *bigT
;
3879 uint64_t packobj
, packoff
, packsize
, bigobj
, bigoff
, bigsize
;
3880 uint64_t chunksize
= (1000 + ztest_random(1000)) * sizeof (uint64_t);
3881 uint64_t regions
= 997;
3882 uint64_t stride
= 123456789ULL;
3883 uint64_t width
= 40;
3884 int free_percent
= 5;
3887 * This test uses two objects, packobj and bigobj, that are always
3888 * updated together (i.e. in the same tx) so that their contents are
3889 * in sync and can be compared. Their contents relate to each other
3890 * in a simple way: packobj is a dense array of 'bufwad' structures,
3891 * while bigobj is a sparse array of the same bufwads. Specifically,
3892 * for any index n, there are three bufwads that should be identical:
3894 * packobj, at offset n * sizeof (bufwad_t)
3895 * bigobj, at the head of the nth chunk
3896 * bigobj, at the tail of the nth chunk
3898 * The chunk size is arbitrary. It doesn't have to be a power of two,
3899 * and it doesn't have any relation to the object blocksize.
3900 * The only requirement is that it can hold at least two bufwads.
3902 * Normally, we write the bufwad to each of these locations.
3903 * However, free_percent of the time we instead write zeroes to
3904 * packobj and perform a dmu_free_range() on bigobj. By comparing
3905 * bigobj to packobj, we can verify that the DMU is correctly
3906 * tracking which parts of an object are allocated and free,
3907 * and that the contents of the allocated blocks are correct.
3911 * Read the directory info. If it's the first time, set things up.
3913 ztest_od_init(od
, id
, FTAG
, 0, DMU_OT_UINT64_OTHER
, 0, 0, chunksize
);
3914 ztest_od_init(od
+ 1, id
, FTAG
, 1, DMU_OT_UINT64_OTHER
, 0, 0,
3917 if (ztest_object_init(zd
, od
, size
, B_FALSE
) != 0) {
3918 umem_free(od
, size
);
3922 bigobj
= od
[0].od_object
;
3923 packobj
= od
[1].od_object
;
3924 chunksize
= od
[0].od_gen
;
3925 ASSERT(chunksize
== od
[1].od_gen
);
3928 * Prefetch a random chunk of the big object.
3929 * Our aim here is to get some async reads in flight
3930 * for blocks that we may free below; the DMU should
3931 * handle this race correctly.
3933 n
= ztest_random(regions
) * stride
+ ztest_random(width
);
3934 s
= 1 + ztest_random(2 * width
- 1);
3935 dmu_prefetch(os
, bigobj
, 0, n
* chunksize
, s
* chunksize
,
3936 ZIO_PRIORITY_SYNC_READ
);
3939 * Pick a random index and compute the offsets into packobj and bigobj.
3941 n
= ztest_random(regions
) * stride
+ ztest_random(width
);
3942 s
= 1 + ztest_random(width
- 1);
3944 packoff
= n
* sizeof (bufwad_t
);
3945 packsize
= s
* sizeof (bufwad_t
);
3947 bigoff
= n
* chunksize
;
3948 bigsize
= s
* chunksize
;
3950 packbuf
= umem_alloc(packsize
, UMEM_NOFAIL
);
3951 bigbuf
= umem_alloc(bigsize
, UMEM_NOFAIL
);
3954 * free_percent of the time, free a range of bigobj rather than
3957 freeit
= (ztest_random(100) < free_percent
);
3960 * Read the current contents of our objects.
3962 error
= dmu_read(os
, packobj
, packoff
, packsize
, packbuf
,
3965 error
= dmu_read(os
, bigobj
, bigoff
, bigsize
, bigbuf
,
3970 * Get a tx for the mods to both packobj and bigobj.
3972 tx
= dmu_tx_create(os
);
3974 dmu_tx_hold_write(tx
, packobj
, packoff
, packsize
);
3977 dmu_tx_hold_free(tx
, bigobj
, bigoff
, bigsize
);
3979 dmu_tx_hold_write(tx
, bigobj
, bigoff
, bigsize
);
3981 /* This accounts for setting the checksum/compression. */
3982 dmu_tx_hold_bonus(tx
, bigobj
);
3984 txg
= ztest_tx_assign(tx
, TXG_MIGHTWAIT
, FTAG
);
3986 umem_free(packbuf
, packsize
);
3987 umem_free(bigbuf
, bigsize
);
3988 umem_free(od
, size
);
3992 enum zio_checksum cksum
;
3994 cksum
= (enum zio_checksum
)
3995 ztest_random_dsl_prop(ZFS_PROP_CHECKSUM
);
3996 } while (cksum
>= ZIO_CHECKSUM_LEGACY_FUNCTIONS
);
3997 dmu_object_set_checksum(os
, bigobj
, cksum
, tx
);
3999 enum zio_compress comp
;
4001 comp
= (enum zio_compress
)
4002 ztest_random_dsl_prop(ZFS_PROP_COMPRESSION
);
4003 } while (comp
>= ZIO_COMPRESS_LEGACY_FUNCTIONS
);
4004 dmu_object_set_compress(os
, bigobj
, comp
, tx
);
4007 * For each index from n to n + s, verify that the existing bufwad
4008 * in packobj matches the bufwads at the head and tail of the
4009 * corresponding chunk in bigobj. Then update all three bufwads
4010 * with the new values we want to write out.
4012 for (i
= 0; i
< s
; i
++) {
4014 pack
= (bufwad_t
*)((char *)packbuf
+ i
* sizeof (bufwad_t
));
4016 bigH
= (bufwad_t
*)((char *)bigbuf
+ i
* chunksize
);
4018 bigT
= (bufwad_t
*)((char *)bigH
+ chunksize
) - 1;
4020 ASSERT((uintptr_t)bigH
- (uintptr_t)bigbuf
< bigsize
);
4021 ASSERT((uintptr_t)bigT
- (uintptr_t)bigbuf
< bigsize
);
4023 if (pack
->bw_txg
> txg
)
4024 fatal(0, "future leak: got %llx, open txg is %llx",
4027 if (pack
->bw_data
!= 0 && pack
->bw_index
!= n
+ i
)
4028 fatal(0, "wrong index: got %llx, wanted %llx+%llx",
4029 pack
->bw_index
, n
, i
);
4031 if (bcmp(pack
, bigH
, sizeof (bufwad_t
)) != 0)
4032 fatal(0, "pack/bigH mismatch in %p/%p", pack
, bigH
);
4034 if (bcmp(pack
, bigT
, sizeof (bufwad_t
)) != 0)
4035 fatal(0, "pack/bigT mismatch in %p/%p", pack
, bigT
);
4038 bzero(pack
, sizeof (bufwad_t
));
4040 pack
->bw_index
= n
+ i
;
4042 pack
->bw_data
= 1 + ztest_random(-2ULL);
4049 * We've verified all the old bufwads, and made new ones.
4050 * Now write them out.
4052 dmu_write(os
, packobj
, packoff
, packsize
, packbuf
, tx
);
4055 if (ztest_opts
.zo_verbose
>= 7) {
4056 (void) printf("freeing offset %llx size %llx"
4058 (u_longlong_t
)bigoff
,
4059 (u_longlong_t
)bigsize
,
4062 VERIFY(0 == dmu_free_range(os
, bigobj
, bigoff
, bigsize
, tx
));
4064 if (ztest_opts
.zo_verbose
>= 7) {
4065 (void) printf("writing offset %llx size %llx"
4067 (u_longlong_t
)bigoff
,
4068 (u_longlong_t
)bigsize
,
4071 dmu_write(os
, bigobj
, bigoff
, bigsize
, bigbuf
, tx
);
4077 * Sanity check the stuff we just wrote.
4080 void *packcheck
= umem_alloc(packsize
, UMEM_NOFAIL
);
4081 void *bigcheck
= umem_alloc(bigsize
, UMEM_NOFAIL
);
4083 VERIFY(0 == dmu_read(os
, packobj
, packoff
,
4084 packsize
, packcheck
, DMU_READ_PREFETCH
));
4085 VERIFY(0 == dmu_read(os
, bigobj
, bigoff
,
4086 bigsize
, bigcheck
, DMU_READ_PREFETCH
));
4088 ASSERT(bcmp(packbuf
, packcheck
, packsize
) == 0);
4089 ASSERT(bcmp(bigbuf
, bigcheck
, bigsize
) == 0);
4091 umem_free(packcheck
, packsize
);
4092 umem_free(bigcheck
, bigsize
);
4095 umem_free(packbuf
, packsize
);
4096 umem_free(bigbuf
, bigsize
);
4097 umem_free(od
, size
);
4101 compare_and_update_pbbufs(uint64_t s
, bufwad_t
*packbuf
, bufwad_t
*bigbuf
,
4102 uint64_t bigsize
, uint64_t n
, uint64_t chunksize
, uint64_t txg
)
4110 * For each index from n to n + s, verify that the existing bufwad
4111 * in packobj matches the bufwads at the head and tail of the
4112 * corresponding chunk in bigobj. Then update all three bufwads
4113 * with the new values we want to write out.
4115 for (i
= 0; i
< s
; i
++) {
4117 pack
= (bufwad_t
*)((char *)packbuf
+ i
* sizeof (bufwad_t
));
4119 bigH
= (bufwad_t
*)((char *)bigbuf
+ i
* chunksize
);
4121 bigT
= (bufwad_t
*)((char *)bigH
+ chunksize
) - 1;
4123 ASSERT((uintptr_t)bigH
- (uintptr_t)bigbuf
< bigsize
);
4124 ASSERT((uintptr_t)bigT
- (uintptr_t)bigbuf
< bigsize
);
4126 if (pack
->bw_txg
> txg
)
4127 fatal(0, "future leak: got %llx, open txg is %llx",
4130 if (pack
->bw_data
!= 0 && pack
->bw_index
!= n
+ i
)
4131 fatal(0, "wrong index: got %llx, wanted %llx+%llx",
4132 pack
->bw_index
, n
, i
);
4134 if (bcmp(pack
, bigH
, sizeof (bufwad_t
)) != 0)
4135 fatal(0, "pack/bigH mismatch in %p/%p", pack
, bigH
);
4137 if (bcmp(pack
, bigT
, sizeof (bufwad_t
)) != 0)
4138 fatal(0, "pack/bigT mismatch in %p/%p", pack
, bigT
);
4140 pack
->bw_index
= n
+ i
;
4142 pack
->bw_data
= 1 + ztest_random(-2ULL);
4149 #undef OD_ARRAY_SIZE
4150 #define OD_ARRAY_SIZE 2
4153 ztest_dmu_read_write_zcopy(ztest_ds_t
*zd
, uint64_t id
)
4155 objset_t
*os
= zd
->zd_os
;
4162 bufwad_t
*packbuf
, *bigbuf
;
4163 uint64_t packobj
, packoff
, packsize
, bigobj
, bigoff
, bigsize
;
4164 uint64_t blocksize
= ztest_random_blocksize();
4165 uint64_t chunksize
= blocksize
;
4166 uint64_t regions
= 997;
4167 uint64_t stride
= 123456789ULL;
4169 dmu_buf_t
*bonus_db
;
4170 arc_buf_t
**bigbuf_arcbufs
;
4171 dmu_object_info_t doi
;
4173 size
= sizeof (ztest_od_t
) * OD_ARRAY_SIZE
;
4174 od
= umem_alloc(size
, UMEM_NOFAIL
);
4177 * This test uses two objects, packobj and bigobj, that are always
4178 * updated together (i.e. in the same tx) so that their contents are
4179 * in sync and can be compared. Their contents relate to each other
4180 * in a simple way: packobj is a dense array of 'bufwad' structures,
4181 * while bigobj is a sparse array of the same bufwads. Specifically,
4182 * for any index n, there are three bufwads that should be identical:
4184 * packobj, at offset n * sizeof (bufwad_t)
4185 * bigobj, at the head of the nth chunk
4186 * bigobj, at the tail of the nth chunk
4188 * The chunk size is set equal to bigobj block size so that
4189 * dmu_assign_arcbuf() can be tested for object updates.
4193 * Read the directory info. If it's the first time, set things up.
4195 ztest_od_init(od
, id
, FTAG
, 0, DMU_OT_UINT64_OTHER
, blocksize
, 0, 0);
4196 ztest_od_init(od
+ 1, id
, FTAG
, 1, DMU_OT_UINT64_OTHER
, 0, 0,
4200 if (ztest_object_init(zd
, od
, size
, B_FALSE
) != 0) {
4201 umem_free(od
, size
);
4205 bigobj
= od
[0].od_object
;
4206 packobj
= od
[1].od_object
;
4207 blocksize
= od
[0].od_blocksize
;
4208 chunksize
= blocksize
;
4209 ASSERT(chunksize
== od
[1].od_gen
);
4211 VERIFY(dmu_object_info(os
, bigobj
, &doi
) == 0);
4212 VERIFY(ISP2(doi
.doi_data_block_size
));
4213 VERIFY(chunksize
== doi
.doi_data_block_size
);
4214 VERIFY(chunksize
>= 2 * sizeof (bufwad_t
));
4217 * Pick a random index and compute the offsets into packobj and bigobj.
4219 n
= ztest_random(regions
) * stride
+ ztest_random(width
);
4220 s
= 1 + ztest_random(width
- 1);
4222 packoff
= n
* sizeof (bufwad_t
);
4223 packsize
= s
* sizeof (bufwad_t
);
4225 bigoff
= n
* chunksize
;
4226 bigsize
= s
* chunksize
;
4228 packbuf
= umem_zalloc(packsize
, UMEM_NOFAIL
);
4229 bigbuf
= umem_zalloc(bigsize
, UMEM_NOFAIL
);
4231 VERIFY3U(0, ==, dmu_bonus_hold(os
, bigobj
, FTAG
, &bonus_db
));
4233 bigbuf_arcbufs
= umem_zalloc(2 * s
* sizeof (arc_buf_t
*), UMEM_NOFAIL
);
4236 * Iteration 0 test zcopy for DB_UNCACHED dbufs.
4237 * Iteration 1 test zcopy to already referenced dbufs.
4238 * Iteration 2 test zcopy to dirty dbuf in the same txg.
4239 * Iteration 3 test zcopy to dbuf dirty in previous txg.
4240 * Iteration 4 test zcopy when dbuf is no longer dirty.
4241 * Iteration 5 test zcopy when it can't be done.
4242 * Iteration 6 one more zcopy write.
4244 for (i
= 0; i
< 7; i
++) {
4249 * In iteration 5 (i == 5) use arcbufs
4250 * that don't match bigobj blksz to test
4251 * dmu_assign_arcbuf() when it can't directly
4252 * assign an arcbuf to a dbuf.
4254 for (j
= 0; j
< s
; j
++) {
4255 if (i
!= 5 || chunksize
< (SPA_MINBLOCKSIZE
* 2)) {
4257 dmu_request_arcbuf(bonus_db
, chunksize
);
4259 bigbuf_arcbufs
[2 * j
] =
4260 dmu_request_arcbuf(bonus_db
, chunksize
/ 2);
4261 bigbuf_arcbufs
[2 * j
+ 1] =
4262 dmu_request_arcbuf(bonus_db
, chunksize
/ 2);
4267 * Get a tx for the mods to both packobj and bigobj.
4269 tx
= dmu_tx_create(os
);
4271 dmu_tx_hold_write(tx
, packobj
, packoff
, packsize
);
4272 dmu_tx_hold_write(tx
, bigobj
, bigoff
, bigsize
);
4274 txg
= ztest_tx_assign(tx
, TXG_MIGHTWAIT
, FTAG
);
4276 umem_free(packbuf
, packsize
);
4277 umem_free(bigbuf
, bigsize
);
4278 for (j
= 0; j
< s
; j
++) {
4280 chunksize
< (SPA_MINBLOCKSIZE
* 2)) {
4281 dmu_return_arcbuf(bigbuf_arcbufs
[j
]);
4284 bigbuf_arcbufs
[2 * j
]);
4286 bigbuf_arcbufs
[2 * j
+ 1]);
4289 umem_free(bigbuf_arcbufs
, 2 * s
* sizeof (arc_buf_t
*));
4290 umem_free(od
, size
);
4291 dmu_buf_rele(bonus_db
, FTAG
);
4296 * 50% of the time don't read objects in the 1st iteration to
4297 * test dmu_assign_arcbuf() for the case when there're no
4298 * existing dbufs for the specified offsets.
4300 if (i
!= 0 || ztest_random(2) != 0) {
4301 error
= dmu_read(os
, packobj
, packoff
,
4302 packsize
, packbuf
, DMU_READ_PREFETCH
);
4304 error
= dmu_read(os
, bigobj
, bigoff
, bigsize
,
4305 bigbuf
, DMU_READ_PREFETCH
);
4308 compare_and_update_pbbufs(s
, packbuf
, bigbuf
, bigsize
,
4312 * We've verified all the old bufwads, and made new ones.
4313 * Now write them out.
4315 dmu_write(os
, packobj
, packoff
, packsize
, packbuf
, tx
);
4316 if (ztest_opts
.zo_verbose
>= 7) {
4317 (void) printf("writing offset %llx size %llx"
4319 (u_longlong_t
)bigoff
,
4320 (u_longlong_t
)bigsize
,
4323 for (off
= bigoff
, j
= 0; j
< s
; j
++, off
+= chunksize
) {
4325 if (i
!= 5 || chunksize
< (SPA_MINBLOCKSIZE
* 2)) {
4326 bcopy((caddr_t
)bigbuf
+ (off
- bigoff
),
4327 bigbuf_arcbufs
[j
]->b_data
, chunksize
);
4329 bcopy((caddr_t
)bigbuf
+ (off
- bigoff
),
4330 bigbuf_arcbufs
[2 * j
]->b_data
,
4332 bcopy((caddr_t
)bigbuf
+ (off
- bigoff
) +
4334 bigbuf_arcbufs
[2 * j
+ 1]->b_data
,
4339 VERIFY(dmu_buf_hold(os
, bigobj
, off
,
4340 FTAG
, &dbt
, DMU_READ_NO_PREFETCH
) == 0);
4342 if (i
!= 5 || chunksize
< (SPA_MINBLOCKSIZE
* 2)) {
4343 dmu_assign_arcbuf(bonus_db
, off
,
4344 bigbuf_arcbufs
[j
], tx
);
4346 dmu_assign_arcbuf(bonus_db
, off
,
4347 bigbuf_arcbufs
[2 * j
], tx
);
4348 dmu_assign_arcbuf(bonus_db
,
4349 off
+ chunksize
/ 2,
4350 bigbuf_arcbufs
[2 * j
+ 1], tx
);
4353 dmu_buf_rele(dbt
, FTAG
);
4359 * Sanity check the stuff we just wrote.
4362 void *packcheck
= umem_alloc(packsize
, UMEM_NOFAIL
);
4363 void *bigcheck
= umem_alloc(bigsize
, UMEM_NOFAIL
);
4365 VERIFY(0 == dmu_read(os
, packobj
, packoff
,
4366 packsize
, packcheck
, DMU_READ_PREFETCH
));
4367 VERIFY(0 == dmu_read(os
, bigobj
, bigoff
,
4368 bigsize
, bigcheck
, DMU_READ_PREFETCH
));
4370 ASSERT(bcmp(packbuf
, packcheck
, packsize
) == 0);
4371 ASSERT(bcmp(bigbuf
, bigcheck
, bigsize
) == 0);
4373 umem_free(packcheck
, packsize
);
4374 umem_free(bigcheck
, bigsize
);
4377 txg_wait_open(dmu_objset_pool(os
), 0);
4378 } else if (i
== 3) {
4379 txg_wait_synced(dmu_objset_pool(os
), 0);
4383 dmu_buf_rele(bonus_db
, FTAG
);
4384 umem_free(packbuf
, packsize
);
4385 umem_free(bigbuf
, bigsize
);
4386 umem_free(bigbuf_arcbufs
, 2 * s
* sizeof (arc_buf_t
*));
4387 umem_free(od
, size
);
4392 ztest_dmu_write_parallel(ztest_ds_t
*zd
, uint64_t id
)
4396 od
= umem_alloc(sizeof (ztest_od_t
), UMEM_NOFAIL
);
4397 uint64_t offset
= (1ULL << (ztest_random(20) + 43)) +
4398 (ztest_random(ZTEST_RANGE_LOCKS
) << SPA_MAXBLOCKSHIFT
);
4401 * Have multiple threads write to large offsets in an object
4402 * to verify that parallel writes to an object -- even to the
4403 * same blocks within the object -- doesn't cause any trouble.
4405 ztest_od_init(od
, ID_PARALLEL
, FTAG
, 0, DMU_OT_UINT64_OTHER
, 0, 0, 0);
4407 if (ztest_object_init(zd
, od
, sizeof (ztest_od_t
), B_FALSE
) != 0)
4410 while (ztest_random(10) != 0)
4411 ztest_io(zd
, od
->od_object
, offset
);
4413 umem_free(od
, sizeof (ztest_od_t
));
4417 ztest_dmu_prealloc(ztest_ds_t
*zd
, uint64_t id
)
4420 uint64_t offset
= (1ULL << (ztest_random(4) + SPA_MAXBLOCKSHIFT
)) +
4421 (ztest_random(ZTEST_RANGE_LOCKS
) << SPA_MAXBLOCKSHIFT
);
4422 uint64_t count
= ztest_random(20) + 1;
4423 uint64_t blocksize
= ztest_random_blocksize();
4426 od
= umem_alloc(sizeof (ztest_od_t
), UMEM_NOFAIL
);
4428 ztest_od_init(od
, id
, FTAG
, 0, DMU_OT_UINT64_OTHER
, blocksize
, 0, 0);
4430 if (ztest_object_init(zd
, od
, sizeof (ztest_od_t
),
4431 !ztest_random(2)) != 0) {
4432 umem_free(od
, sizeof (ztest_od_t
));
4436 if (ztest_truncate(zd
, od
->od_object
, offset
, count
* blocksize
) != 0) {
4437 umem_free(od
, sizeof (ztest_od_t
));
4441 ztest_prealloc(zd
, od
->od_object
, offset
, count
* blocksize
);
4443 data
= umem_zalloc(blocksize
, UMEM_NOFAIL
);
4445 while (ztest_random(count
) != 0) {
4446 uint64_t randoff
= offset
+ (ztest_random(count
) * blocksize
);
4447 if (ztest_write(zd
, od
->od_object
, randoff
, blocksize
,
4450 while (ztest_random(4) != 0)
4451 ztest_io(zd
, od
->od_object
, randoff
);
4454 umem_free(data
, blocksize
);
4455 umem_free(od
, sizeof (ztest_od_t
));
4459 * Verify that zap_{create,destroy,add,remove,update} work as expected.
4461 #define ZTEST_ZAP_MIN_INTS 1
4462 #define ZTEST_ZAP_MAX_INTS 4
4463 #define ZTEST_ZAP_MAX_PROPS 1000
4466 ztest_zap(ztest_ds_t
*zd
, uint64_t id
)
4468 objset_t
*os
= zd
->zd_os
;
4471 uint64_t txg
, last_txg
;
4472 uint64_t value
[ZTEST_ZAP_MAX_INTS
];
4473 uint64_t zl_ints
, zl_intsize
, prop
;
4476 char propname
[100], txgname
[100];
4478 char *hc
[2] = { "s.acl.h", ".s.open.h.hyLZlg" };
4480 od
= umem_alloc(sizeof (ztest_od_t
), UMEM_NOFAIL
);
4481 ztest_od_init(od
, id
, FTAG
, 0, DMU_OT_ZAP_OTHER
, 0, 0, 0);
4483 if (ztest_object_init(zd
, od
, sizeof (ztest_od_t
),
4484 !ztest_random(2)) != 0)
4487 object
= od
->od_object
;
4490 * Generate a known hash collision, and verify that
4491 * we can lookup and remove both entries.
4493 tx
= dmu_tx_create(os
);
4494 dmu_tx_hold_zap(tx
, object
, B_TRUE
, NULL
);
4495 txg
= ztest_tx_assign(tx
, TXG_MIGHTWAIT
, FTAG
);
4498 for (i
= 0; i
< 2; i
++) {
4500 VERIFY3U(0, ==, zap_add(os
, object
, hc
[i
], sizeof (uint64_t),
4503 for (i
= 0; i
< 2; i
++) {
4504 VERIFY3U(EEXIST
, ==, zap_add(os
, object
, hc
[i
],
4505 sizeof (uint64_t), 1, &value
[i
], tx
));
4507 zap_length(os
, object
, hc
[i
], &zl_intsize
, &zl_ints
));
4508 ASSERT3U(zl_intsize
, ==, sizeof (uint64_t));
4509 ASSERT3U(zl_ints
, ==, 1);
4511 for (i
= 0; i
< 2; i
++) {
4512 VERIFY3U(0, ==, zap_remove(os
, object
, hc
[i
], tx
));
4517 * Generate a buch of random entries.
4519 ints
= MAX(ZTEST_ZAP_MIN_INTS
, object
% ZTEST_ZAP_MAX_INTS
);
4521 prop
= ztest_random(ZTEST_ZAP_MAX_PROPS
);
4522 (void) sprintf(propname
, "prop_%llu", (u_longlong_t
)prop
);
4523 (void) sprintf(txgname
, "txg_%llu", (u_longlong_t
)prop
);
4524 bzero(value
, sizeof (value
));
4528 * If these zap entries already exist, validate their contents.
4530 error
= zap_length(os
, object
, txgname
, &zl_intsize
, &zl_ints
);
4532 ASSERT3U(zl_intsize
, ==, sizeof (uint64_t));
4533 ASSERT3U(zl_ints
, ==, 1);
4535 VERIFY(zap_lookup(os
, object
, txgname
, zl_intsize
,
4536 zl_ints
, &last_txg
) == 0);
4538 VERIFY(zap_length(os
, object
, propname
, &zl_intsize
,
4541 ASSERT3U(zl_intsize
, ==, sizeof (uint64_t));
4542 ASSERT3U(zl_ints
, ==, ints
);
4544 VERIFY(zap_lookup(os
, object
, propname
, zl_intsize
,
4545 zl_ints
, value
) == 0);
4547 for (i
= 0; i
< ints
; i
++) {
4548 ASSERT3U(value
[i
], ==, last_txg
+ object
+ i
);
4551 ASSERT3U(error
, ==, ENOENT
);
4555 * Atomically update two entries in our zap object.
4556 * The first is named txg_%llu, and contains the txg
4557 * in which the property was last updated. The second
4558 * is named prop_%llu, and the nth element of its value
4559 * should be txg + object + n.
4561 tx
= dmu_tx_create(os
);
4562 dmu_tx_hold_zap(tx
, object
, B_TRUE
, NULL
);
4563 txg
= ztest_tx_assign(tx
, TXG_MIGHTWAIT
, FTAG
);
4568 fatal(0, "zap future leak: old %llu new %llu", last_txg
, txg
);
4570 for (i
= 0; i
< ints
; i
++)
4571 value
[i
] = txg
+ object
+ i
;
4573 VERIFY3U(0, ==, zap_update(os
, object
, txgname
, sizeof (uint64_t),
4575 VERIFY3U(0, ==, zap_update(os
, object
, propname
, sizeof (uint64_t),
4581 * Remove a random pair of entries.
4583 prop
= ztest_random(ZTEST_ZAP_MAX_PROPS
);
4584 (void) sprintf(propname
, "prop_%llu", (u_longlong_t
)prop
);
4585 (void) sprintf(txgname
, "txg_%llu", (u_longlong_t
)prop
);
4587 error
= zap_length(os
, object
, txgname
, &zl_intsize
, &zl_ints
);
4589 if (error
== ENOENT
)
4594 tx
= dmu_tx_create(os
);
4595 dmu_tx_hold_zap(tx
, object
, B_TRUE
, NULL
);
4596 txg
= ztest_tx_assign(tx
, TXG_MIGHTWAIT
, FTAG
);
4599 VERIFY3U(0, ==, zap_remove(os
, object
, txgname
, tx
));
4600 VERIFY3U(0, ==, zap_remove(os
, object
, propname
, tx
));
4603 umem_free(od
, sizeof (ztest_od_t
));
4607 * Testcase to test the upgrading of a microzap to fatzap.
4610 ztest_fzap(ztest_ds_t
*zd
, uint64_t id
)
4612 objset_t
*os
= zd
->zd_os
;
4614 uint64_t object
, txg
;
4617 od
= umem_alloc(sizeof (ztest_od_t
), UMEM_NOFAIL
);
4618 ztest_od_init(od
, id
, FTAG
, 0, DMU_OT_ZAP_OTHER
, 0, 0, 0);
4620 if (ztest_object_init(zd
, od
, sizeof (ztest_od_t
),
4621 !ztest_random(2)) != 0)
4623 object
= od
->od_object
;
4626 * Add entries to this ZAP and make sure it spills over
4627 * and gets upgraded to a fatzap. Also, since we are adding
4628 * 2050 entries we should see ptrtbl growth and leaf-block split.
4630 for (i
= 0; i
< 2050; i
++) {
4631 char name
[ZFS_MAX_DATASET_NAME_LEN
];
4636 (void) snprintf(name
, sizeof (name
), "fzap-%llu-%llu",
4637 (u_longlong_t
)id
, (u_longlong_t
)value
);
4639 tx
= dmu_tx_create(os
);
4640 dmu_tx_hold_zap(tx
, object
, B_TRUE
, name
);
4641 txg
= ztest_tx_assign(tx
, TXG_MIGHTWAIT
, FTAG
);
4644 error
= zap_add(os
, object
, name
, sizeof (uint64_t), 1,
4646 ASSERT(error
== 0 || error
== EEXIST
);
4650 umem_free(od
, sizeof (ztest_od_t
));
4655 ztest_zap_parallel(ztest_ds_t
*zd
, uint64_t id
)
4657 objset_t
*os
= zd
->zd_os
;
4659 uint64_t txg
, object
, count
, wsize
, wc
, zl_wsize
, zl_wc
;
4661 int i
, namelen
, error
;
4662 int micro
= ztest_random(2);
4663 char name
[20], string_value
[20];
4666 od
= umem_alloc(sizeof (ztest_od_t
), UMEM_NOFAIL
);
4667 ztest_od_init(od
, ID_PARALLEL
, FTAG
, micro
, DMU_OT_ZAP_OTHER
, 0, 0, 0);
4669 if (ztest_object_init(zd
, od
, sizeof (ztest_od_t
), B_FALSE
) != 0) {
4670 umem_free(od
, sizeof (ztest_od_t
));
4674 object
= od
->od_object
;
4677 * Generate a random name of the form 'xxx.....' where each
4678 * x is a random printable character and the dots are dots.
4679 * There are 94 such characters, and the name length goes from
4680 * 6 to 20, so there are 94^3 * 15 = 12,458,760 possible names.
4682 namelen
= ztest_random(sizeof (name
) - 5) + 5 + 1;
4684 for (i
= 0; i
< 3; i
++)
4685 name
[i
] = '!' + ztest_random('~' - '!' + 1);
4686 for (; i
< namelen
- 1; i
++)
4690 if ((namelen
& 1) || micro
) {
4691 wsize
= sizeof (txg
);
4697 data
= string_value
;
4701 VERIFY0(zap_count(os
, object
, &count
));
4702 ASSERT(count
!= -1ULL);
4705 * Select an operation: length, lookup, add, update, remove.
4707 i
= ztest_random(5);
4710 tx
= dmu_tx_create(os
);
4711 dmu_tx_hold_zap(tx
, object
, B_TRUE
, NULL
);
4712 txg
= ztest_tx_assign(tx
, TXG_MIGHTWAIT
, FTAG
);
4714 umem_free(od
, sizeof (ztest_od_t
));
4717 bcopy(name
, string_value
, namelen
);
4721 bzero(string_value
, namelen
);
4727 error
= zap_length(os
, object
, name
, &zl_wsize
, &zl_wc
);
4729 ASSERT3U(wsize
, ==, zl_wsize
);
4730 ASSERT3U(wc
, ==, zl_wc
);
4732 ASSERT3U(error
, ==, ENOENT
);
4737 error
= zap_lookup(os
, object
, name
, wsize
, wc
, data
);
4739 if (data
== string_value
&&
4740 bcmp(name
, data
, namelen
) != 0)
4741 fatal(0, "name '%s' != val '%s' len %d",
4742 name
, data
, namelen
);
4744 ASSERT3U(error
, ==, ENOENT
);
4749 error
= zap_add(os
, object
, name
, wsize
, wc
, data
, tx
);
4750 ASSERT(error
== 0 || error
== EEXIST
);
4754 VERIFY(zap_update(os
, object
, name
, wsize
, wc
, data
, tx
) == 0);
4758 error
= zap_remove(os
, object
, name
, tx
);
4759 ASSERT(error
== 0 || error
== ENOENT
);
4766 umem_free(od
, sizeof (ztest_od_t
));
4770 * Commit callback data.
4772 typedef struct ztest_cb_data
{
4773 list_node_t zcd_node
;
4775 int zcd_expected_err
;
4776 boolean_t zcd_added
;
4777 boolean_t zcd_called
;
4781 /* This is the actual commit callback function */
4783 ztest_commit_callback(void *arg
, int error
)
4785 ztest_cb_data_t
*data
= arg
;
4786 uint64_t synced_txg
;
4788 VERIFY(data
!= NULL
);
4789 VERIFY3S(data
->zcd_expected_err
, ==, error
);
4790 VERIFY(!data
->zcd_called
);
4792 synced_txg
= spa_last_synced_txg(data
->zcd_spa
);
4793 if (data
->zcd_txg
> synced_txg
)
4794 fatal(0, "commit callback of txg %" PRIu64
" called prematurely"
4795 ", last synced txg = %" PRIu64
"\n", data
->zcd_txg
,
4798 data
->zcd_called
= B_TRUE
;
4800 if (error
== ECANCELED
) {
4801 ASSERT0(data
->zcd_txg
);
4802 ASSERT(!data
->zcd_added
);
4805 * The private callback data should be destroyed here, but
4806 * since we are going to check the zcd_called field after
4807 * dmu_tx_abort(), we will destroy it there.
4812 ASSERT(data
->zcd_added
);
4813 ASSERT3U(data
->zcd_txg
, !=, 0);
4815 (void) mutex_enter(&zcl
.zcl_callbacks_lock
);
4817 /* See if this cb was called more quickly */
4818 if ((synced_txg
- data
->zcd_txg
) < zc_min_txg_delay
)
4819 zc_min_txg_delay
= synced_txg
- data
->zcd_txg
;
4821 /* Remove our callback from the list */
4822 list_remove(&zcl
.zcl_callbacks
, data
);
4824 (void) mutex_exit(&zcl
.zcl_callbacks_lock
);
4826 umem_free(data
, sizeof (ztest_cb_data_t
));
4829 /* Allocate and initialize callback data structure */
4830 static ztest_cb_data_t
*
4831 ztest_create_cb_data(objset_t
*os
, uint64_t txg
)
4833 ztest_cb_data_t
*cb_data
;
4835 cb_data
= umem_zalloc(sizeof (ztest_cb_data_t
), UMEM_NOFAIL
);
4837 cb_data
->zcd_txg
= txg
;
4838 cb_data
->zcd_spa
= dmu_objset_spa(os
);
4839 list_link_init(&cb_data
->zcd_node
);
4845 * Commit callback test.
4848 ztest_dmu_commit_callbacks(ztest_ds_t
*zd
, uint64_t id
)
4850 objset_t
*os
= zd
->zd_os
;
4853 ztest_cb_data_t
*cb_data
[3], *tmp_cb
;
4854 uint64_t old_txg
, txg
;
4857 od
= umem_alloc(sizeof (ztest_od_t
), UMEM_NOFAIL
);
4858 ztest_od_init(od
, id
, FTAG
, 0, DMU_OT_UINT64_OTHER
, 0, 0, 0);
4860 if (ztest_object_init(zd
, od
, sizeof (ztest_od_t
), B_FALSE
) != 0) {
4861 umem_free(od
, sizeof (ztest_od_t
));
4865 tx
= dmu_tx_create(os
);
4867 cb_data
[0] = ztest_create_cb_data(os
, 0);
4868 dmu_tx_callback_register(tx
, ztest_commit_callback
, cb_data
[0]);
4870 dmu_tx_hold_write(tx
, od
->od_object
, 0, sizeof (uint64_t));
4872 /* Every once in a while, abort the transaction on purpose */
4873 if (ztest_random(100) == 0)
4877 error
= dmu_tx_assign(tx
, TXG_NOWAIT
);
4879 txg
= error
? 0 : dmu_tx_get_txg(tx
);
4881 cb_data
[0]->zcd_txg
= txg
;
4882 cb_data
[1] = ztest_create_cb_data(os
, txg
);
4883 dmu_tx_callback_register(tx
, ztest_commit_callback
, cb_data
[1]);
4887 * It's not a strict requirement to call the registered
4888 * callbacks from inside dmu_tx_abort(), but that's what
4889 * it's supposed to happen in the current implementation
4890 * so we will check for that.
4892 for (i
= 0; i
< 2; i
++) {
4893 cb_data
[i
]->zcd_expected_err
= ECANCELED
;
4894 VERIFY(!cb_data
[i
]->zcd_called
);
4899 for (i
= 0; i
< 2; i
++) {
4900 VERIFY(cb_data
[i
]->zcd_called
);
4901 umem_free(cb_data
[i
], sizeof (ztest_cb_data_t
));
4904 umem_free(od
, sizeof (ztest_od_t
));
4908 cb_data
[2] = ztest_create_cb_data(os
, txg
);
4909 dmu_tx_callback_register(tx
, ztest_commit_callback
, cb_data
[2]);
4912 * Read existing data to make sure there isn't a future leak.
4914 VERIFY(0 == dmu_read(os
, od
->od_object
, 0, sizeof (uint64_t),
4915 &old_txg
, DMU_READ_PREFETCH
));
4918 fatal(0, "future leak: got %" PRIu64
", open txg is %" PRIu64
,
4921 dmu_write(os
, od
->od_object
, 0, sizeof (uint64_t), &txg
, tx
);
4923 (void) mutex_enter(&zcl
.zcl_callbacks_lock
);
4926 * Since commit callbacks don't have any ordering requirement and since
4927 * it is theoretically possible for a commit callback to be called
4928 * after an arbitrary amount of time has elapsed since its txg has been
4929 * synced, it is difficult to reliably determine whether a commit
4930 * callback hasn't been called due to high load or due to a flawed
4933 * In practice, we will assume that if after a certain number of txgs a
4934 * commit callback hasn't been called, then most likely there's an
4935 * implementation bug..
4937 tmp_cb
= list_head(&zcl
.zcl_callbacks
);
4938 if (tmp_cb
!= NULL
&&
4939 tmp_cb
->zcd_txg
+ ZTEST_COMMIT_CB_THRESH
< txg
) {
4940 fatal(0, "Commit callback threshold exceeded, oldest txg: %"
4941 PRIu64
", open txg: %" PRIu64
"\n", tmp_cb
->zcd_txg
, txg
);
4945 * Let's find the place to insert our callbacks.
4947 * Even though the list is ordered by txg, it is possible for the
4948 * insertion point to not be the end because our txg may already be
4949 * quiescing at this point and other callbacks in the open txg
4950 * (from other objsets) may have sneaked in.
4952 tmp_cb
= list_tail(&zcl
.zcl_callbacks
);
4953 while (tmp_cb
!= NULL
&& tmp_cb
->zcd_txg
> txg
)
4954 tmp_cb
= list_prev(&zcl
.zcl_callbacks
, tmp_cb
);
4956 /* Add the 3 callbacks to the list */
4957 for (i
= 0; i
< 3; i
++) {
4959 list_insert_head(&zcl
.zcl_callbacks
, cb_data
[i
]);
4961 list_insert_after(&zcl
.zcl_callbacks
, tmp_cb
,
4964 cb_data
[i
]->zcd_added
= B_TRUE
;
4965 VERIFY(!cb_data
[i
]->zcd_called
);
4967 tmp_cb
= cb_data
[i
];
4972 (void) mutex_exit(&zcl
.zcl_callbacks_lock
);
4976 umem_free(od
, sizeof (ztest_od_t
));
4980 * Visit each object in the dataset. Verify that its properties
4981 * are consistent what was stored in the block tag when it was created,
4982 * and that its unused bonus buffer space has not been overwritten.
4985 ztest_verify_dnode_bt(ztest_ds_t
*zd
, uint64_t id
)
4987 objset_t
*os
= zd
->zd_os
;
4991 for (obj
= 0; err
== 0; err
= dmu_object_next(os
, &obj
, FALSE
, 0)) {
4992 ztest_block_tag_t
*bt
= NULL
;
4993 dmu_object_info_t doi
;
4996 if (dmu_bonus_hold(os
, obj
, FTAG
, &db
) != 0)
4999 dmu_object_info_from_db(db
, &doi
);
5000 if (doi
.doi_bonus_size
>= sizeof (*bt
))
5001 bt
= ztest_bt_bonus(db
);
5003 if (bt
&& bt
->bt_magic
== BT_MAGIC
) {
5004 ztest_bt_verify(bt
, os
, obj
, doi
.doi_dnodesize
,
5005 bt
->bt_offset
, bt
->bt_gen
, bt
->bt_txg
,
5007 ztest_verify_unused_bonus(db
, bt
, obj
, os
, bt
->bt_gen
);
5010 dmu_buf_rele(db
, FTAG
);
5016 ztest_dsl_prop_get_set(ztest_ds_t
*zd
, uint64_t id
)
5018 zfs_prop_t proplist
[] = {
5020 ZFS_PROP_COMPRESSION
,
5026 (void) rw_rdlock(&ztest_name_lock
);
5028 for (p
= 0; p
< sizeof (proplist
) / sizeof (proplist
[0]); p
++)
5029 (void) ztest_dsl_prop_set_uint64(zd
->zd_name
, proplist
[p
],
5030 ztest_random_dsl_prop(proplist
[p
]), (int)ztest_random(2));
5032 VERIFY0(ztest_dsl_prop_set_uint64(zd
->zd_name
, ZFS_PROP_RECORDSIZE
,
5033 ztest_random_blocksize(), (int)ztest_random(2)));
5035 (void) rw_unlock(&ztest_name_lock
);
5040 ztest_spa_prop_get_set(ztest_ds_t
*zd
, uint64_t id
)
5042 nvlist_t
*props
= NULL
;
5044 (void) rw_rdlock(&ztest_name_lock
);
5046 (void) ztest_spa_prop_set_uint64(ZPOOL_PROP_DEDUPDITTO
,
5047 ZIO_DEDUPDITTO_MIN
+ ztest_random(ZIO_DEDUPDITTO_MIN
));
5049 VERIFY0(spa_prop_get(ztest_spa
, &props
));
5051 if (ztest_opts
.zo_verbose
>= 6)
5052 dump_nvlist(props
, 4);
5056 (void) rw_unlock(&ztest_name_lock
);
5060 user_release_one(const char *snapname
, const char *holdname
)
5062 nvlist_t
*snaps
, *holds
;
5065 snaps
= fnvlist_alloc();
5066 holds
= fnvlist_alloc();
5067 fnvlist_add_boolean(holds
, holdname
);
5068 fnvlist_add_nvlist(snaps
, snapname
, holds
);
5069 fnvlist_free(holds
);
5070 error
= dsl_dataset_user_release(snaps
, NULL
);
5071 fnvlist_free(snaps
);
5076 * Test snapshot hold/release and deferred destroy.
5079 ztest_dmu_snapshot_hold(ztest_ds_t
*zd
, uint64_t id
)
5082 objset_t
*os
= zd
->zd_os
;
5086 char clonename
[100];
5088 char osname
[ZFS_MAX_DATASET_NAME_LEN
];
5091 (void) rw_rdlock(&ztest_name_lock
);
5093 dmu_objset_name(os
, osname
);
5095 (void) snprintf(snapname
, sizeof (snapname
), "sh1_%llu",
5097 (void) snprintf(fullname
, sizeof (fullname
), "%s@%s", osname
, snapname
);
5098 (void) snprintf(clonename
, sizeof (clonename
),
5099 "%s/ch1_%llu", osname
, (u_longlong_t
)id
);
5100 (void) snprintf(tag
, sizeof (tag
), "tag_%llu", (u_longlong_t
)id
);
5103 * Clean up from any previous run.
5105 error
= dsl_destroy_head(clonename
);
5106 if (error
!= ENOENT
)
5108 error
= user_release_one(fullname
, tag
);
5109 if (error
!= ESRCH
&& error
!= ENOENT
)
5111 error
= dsl_destroy_snapshot(fullname
, B_FALSE
);
5112 if (error
!= ENOENT
)
5116 * Create snapshot, clone it, mark snap for deferred destroy,
5117 * destroy clone, verify snap was also destroyed.
5119 error
= dmu_objset_snapshot_one(osname
, snapname
);
5121 if (error
== ENOSPC
) {
5122 ztest_record_enospc("dmu_objset_snapshot");
5125 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname
, error
);
5128 error
= dmu_objset_clone(clonename
, fullname
);
5130 if (error
== ENOSPC
) {
5131 ztest_record_enospc("dmu_objset_clone");
5134 fatal(0, "dmu_objset_clone(%s) = %d", clonename
, error
);
5137 error
= dsl_destroy_snapshot(fullname
, B_TRUE
);
5139 fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
5143 error
= dsl_destroy_head(clonename
);
5145 fatal(0, "dsl_destroy_head(%s) = %d", clonename
, error
);
5147 error
= dmu_objset_hold(fullname
, FTAG
, &origin
);
5148 if (error
!= ENOENT
)
5149 fatal(0, "dmu_objset_hold(%s) = %d", fullname
, error
);
5152 * Create snapshot, add temporary hold, verify that we can't
5153 * destroy a held snapshot, mark for deferred destroy,
5154 * release hold, verify snapshot was destroyed.
5156 error
= dmu_objset_snapshot_one(osname
, snapname
);
5158 if (error
== ENOSPC
) {
5159 ztest_record_enospc("dmu_objset_snapshot");
5162 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname
, error
);
5165 holds
= fnvlist_alloc();
5166 fnvlist_add_string(holds
, fullname
, tag
);
5167 error
= dsl_dataset_user_hold(holds
, 0, NULL
);
5168 fnvlist_free(holds
);
5170 if (error
== ENOSPC
) {
5171 ztest_record_enospc("dsl_dataset_user_hold");
5174 fatal(0, "dsl_dataset_user_hold(%s, %s) = %u",
5175 fullname
, tag
, error
);
5178 error
= dsl_destroy_snapshot(fullname
, B_FALSE
);
5179 if (error
!= EBUSY
) {
5180 fatal(0, "dsl_destroy_snapshot(%s, B_FALSE) = %d",
5184 error
= dsl_destroy_snapshot(fullname
, B_TRUE
);
5186 fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
5190 error
= user_release_one(fullname
, tag
);
5192 fatal(0, "user_release_one(%s, %s) = %d", fullname
, tag
, error
);
5194 VERIFY3U(dmu_objset_hold(fullname
, FTAG
, &origin
), ==, ENOENT
);
5197 (void) rw_unlock(&ztest_name_lock
);
5201 * Inject random faults into the on-disk data.
5205 ztest_fault_inject(ztest_ds_t
*zd
, uint64_t id
)
5207 ztest_shared_t
*zs
= ztest_shared
;
5208 spa_t
*spa
= ztest_spa
;
5212 uint64_t bad
= 0x1990c0ffeedecadeull
;
5217 int bshift
= SPA_MAXBLOCKSHIFT
+ 2; /* don't scrog all labels */
5223 boolean_t islog
= B_FALSE
;
5225 path0
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
5226 pathrand
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
5228 mutex_enter(&ztest_vdev_lock
);
5229 maxfaults
= MAXFAULTS();
5230 leaves
= MAX(zs
->zs_mirrors
, 1) * ztest_opts
.zo_raidz
;
5231 mirror_save
= zs
->zs_mirrors
;
5232 mutex_exit(&ztest_vdev_lock
);
5234 ASSERT(leaves
>= 1);
5237 * Grab the name lock as reader. There are some operations
5238 * which don't like to have their vdevs changed while
5239 * they are in progress (i.e. spa_change_guid). Those
5240 * operations will have grabbed the name lock as writer.
5242 (void) rw_rdlock(&ztest_name_lock
);
5245 * We need SCL_STATE here because we're going to look at vd0->vdev_tsd.
5247 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
5249 if (ztest_random(2) == 0) {
5251 * Inject errors on a normal data device or slog device.
5253 top
= ztest_random_vdev_top(spa
, B_TRUE
);
5254 leaf
= ztest_random(leaves
) + zs
->zs_splits
;
5257 * Generate paths to the first leaf in this top-level vdev,
5258 * and to the random leaf we selected. We'll induce transient
5259 * write failures and random online/offline activity on leaf 0,
5260 * and we'll write random garbage to the randomly chosen leaf.
5262 (void) snprintf(path0
, MAXPATHLEN
, ztest_dev_template
,
5263 ztest_opts
.zo_dir
, ztest_opts
.zo_pool
,
5264 top
* leaves
+ zs
->zs_splits
);
5265 (void) snprintf(pathrand
, MAXPATHLEN
, ztest_dev_template
,
5266 ztest_opts
.zo_dir
, ztest_opts
.zo_pool
,
5267 top
* leaves
+ leaf
);
5269 vd0
= vdev_lookup_by_path(spa
->spa_root_vdev
, path0
);
5270 if (vd0
!= NULL
&& vd0
->vdev_top
->vdev_islog
)
5274 * If the top-level vdev needs to be resilvered
5275 * then we only allow faults on the device that is
5278 if (vd0
!= NULL
&& maxfaults
!= 1 &&
5279 (!vdev_resilver_needed(vd0
->vdev_top
, NULL
, NULL
) ||
5280 vd0
->vdev_resilver_txg
!= 0)) {
5282 * Make vd0 explicitly claim to be unreadable,
5283 * or unwriteable, or reach behind its back
5284 * and close the underlying fd. We can do this if
5285 * maxfaults == 0 because we'll fail and reexecute,
5286 * and we can do it if maxfaults >= 2 because we'll
5287 * have enough redundancy. If maxfaults == 1, the
5288 * combination of this with injection of random data
5289 * corruption below exceeds the pool's fault tolerance.
5291 vdev_file_t
*vf
= vd0
->vdev_tsd
;
5293 if (vf
!= NULL
&& ztest_random(3) == 0) {
5294 (void) close(vf
->vf_vnode
->v_fd
);
5295 vf
->vf_vnode
->v_fd
= -1;
5296 } else if (ztest_random(2) == 0) {
5297 vd0
->vdev_cant_read
= B_TRUE
;
5299 vd0
->vdev_cant_write
= B_TRUE
;
5301 guid0
= vd0
->vdev_guid
;
5305 * Inject errors on an l2cache device.
5307 spa_aux_vdev_t
*sav
= &spa
->spa_l2cache
;
5309 if (sav
->sav_count
== 0) {
5310 spa_config_exit(spa
, SCL_STATE
, FTAG
);
5311 (void) rw_unlock(&ztest_name_lock
);
5314 vd0
= sav
->sav_vdevs
[ztest_random(sav
->sav_count
)];
5315 guid0
= vd0
->vdev_guid
;
5316 (void) strcpy(path0
, vd0
->vdev_path
);
5317 (void) strcpy(pathrand
, vd0
->vdev_path
);
5321 maxfaults
= INT_MAX
; /* no limit on cache devices */
5324 spa_config_exit(spa
, SCL_STATE
, FTAG
);
5325 (void) rw_unlock(&ztest_name_lock
);
5328 * If we can tolerate two or more faults, or we're dealing
5329 * with a slog, randomly online/offline vd0.
5331 if ((maxfaults
>= 2 || islog
) && guid0
!= 0) {
5332 if (ztest_random(10) < 6) {
5333 int flags
= (ztest_random(2) == 0 ?
5334 ZFS_OFFLINE_TEMPORARY
: 0);
5337 * We have to grab the zs_name_lock as writer to
5338 * prevent a race between offlining a slog and
5339 * destroying a dataset. Offlining the slog will
5340 * grab a reference on the dataset which may cause
5341 * dsl_destroy_head() to fail with EBUSY thus
5342 * leaving the dataset in an inconsistent state.
5345 (void) rw_wrlock(&ztest_name_lock
);
5347 VERIFY(vdev_offline(spa
, guid0
, flags
) != EBUSY
);
5350 (void) rw_unlock(&ztest_name_lock
);
5353 * Ideally we would like to be able to randomly
5354 * call vdev_[on|off]line without holding locks
5355 * to force unpredictable failures but the side
5356 * effects of vdev_[on|off]line prevent us from
5357 * doing so. We grab the ztest_vdev_lock here to
5358 * prevent a race between injection testing and
5361 mutex_enter(&ztest_vdev_lock
);
5362 (void) vdev_online(spa
, guid0
, 0, NULL
);
5363 mutex_exit(&ztest_vdev_lock
);
5371 * We have at least single-fault tolerance, so inject data corruption.
5373 fd
= open(pathrand
, O_RDWR
);
5375 if (fd
== -1) /* we hit a gap in the device namespace */
5378 fsize
= lseek(fd
, 0, SEEK_END
);
5380 while (--iters
!= 0) {
5382 * The offset must be chosen carefully to ensure that
5383 * we do not inject a given logical block with errors
5384 * on two different leaf devices, because ZFS can not
5385 * tolerate that (if maxfaults==1).
5387 * We divide each leaf into chunks of size
5388 * (# leaves * SPA_MAXBLOCKSIZE * 4). Within each chunk
5389 * there is a series of ranges to which we can inject errors.
5390 * Each range can accept errors on only a single leaf vdev.
5391 * The error injection ranges are separated by ranges
5392 * which we will not inject errors on any device (DMZs).
5393 * Each DMZ must be large enough such that a single block
5394 * can not straddle it, so that a single block can not be
5395 * a target in two different injection ranges (on different
5398 * For example, with 3 leaves, each chunk looks like:
5399 * 0 to 32M: injection range for leaf 0
5400 * 32M to 64M: DMZ - no injection allowed
5401 * 64M to 96M: injection range for leaf 1
5402 * 96M to 128M: DMZ - no injection allowed
5403 * 128M to 160M: injection range for leaf 2
5404 * 160M to 192M: DMZ - no injection allowed
5406 offset
= ztest_random(fsize
/ (leaves
<< bshift
)) *
5407 (leaves
<< bshift
) + (leaf
<< bshift
) +
5408 (ztest_random(1ULL << (bshift
- 1)) & -8ULL);
5410 if (offset
>= fsize
)
5413 mutex_enter(&ztest_vdev_lock
);
5414 if (mirror_save
!= zs
->zs_mirrors
) {
5415 mutex_exit(&ztest_vdev_lock
);
5420 if (pwrite(fd
, &bad
, sizeof (bad
), offset
) != sizeof (bad
))
5421 fatal(1, "can't inject bad word at 0x%llx in %s",
5424 mutex_exit(&ztest_vdev_lock
);
5426 if (ztest_opts
.zo_verbose
>= 7)
5427 (void) printf("injected bad word into %s,"
5428 " offset 0x%llx\n", pathrand
, (u_longlong_t
)offset
);
5433 umem_free(path0
, MAXPATHLEN
);
5434 umem_free(pathrand
, MAXPATHLEN
);
5438 * Verify that DDT repair works as expected.
5441 ztest_ddt_repair(ztest_ds_t
*zd
, uint64_t id
)
5443 ztest_shared_t
*zs
= ztest_shared
;
5444 spa_t
*spa
= ztest_spa
;
5445 objset_t
*os
= zd
->zd_os
;
5447 uint64_t object
, blocksize
, txg
, pattern
, psize
;
5448 enum zio_checksum checksum
= spa_dedup_checksum(spa
);
5453 int copies
= 2 * ZIO_DEDUPDITTO_MIN
;
5456 blocksize
= ztest_random_blocksize();
5457 blocksize
= MIN(blocksize
, 2048); /* because we write so many */
5459 od
= umem_alloc(sizeof (ztest_od_t
), UMEM_NOFAIL
);
5460 ztest_od_init(od
, id
, FTAG
, 0, DMU_OT_UINT64_OTHER
, blocksize
, 0, 0);
5462 if (ztest_object_init(zd
, od
, sizeof (ztest_od_t
), B_FALSE
) != 0) {
5463 umem_free(od
, sizeof (ztest_od_t
));
5468 * Take the name lock as writer to prevent anyone else from changing
5469 * the pool and dataset properies we need to maintain during this test.
5471 (void) rw_wrlock(&ztest_name_lock
);
5473 if (ztest_dsl_prop_set_uint64(zd
->zd_name
, ZFS_PROP_DEDUP
, checksum
,
5475 ztest_dsl_prop_set_uint64(zd
->zd_name
, ZFS_PROP_COPIES
, 1,
5477 (void) rw_unlock(&ztest_name_lock
);
5478 umem_free(od
, sizeof (ztest_od_t
));
5482 dmu_objset_stats_t dds
;
5483 dsl_pool_config_enter(dmu_objset_pool(os
), FTAG
);
5484 dmu_objset_fast_stat(os
, &dds
);
5485 dsl_pool_config_exit(dmu_objset_pool(os
), FTAG
);
5487 object
= od
[0].od_object
;
5488 blocksize
= od
[0].od_blocksize
;
5489 pattern
= zs
->zs_guid
^ dds
.dds_guid
;
5491 ASSERT(object
!= 0);
5493 tx
= dmu_tx_create(os
);
5494 dmu_tx_hold_write(tx
, object
, 0, copies
* blocksize
);
5495 txg
= ztest_tx_assign(tx
, TXG_WAIT
, FTAG
);
5497 (void) rw_unlock(&ztest_name_lock
);
5498 umem_free(od
, sizeof (ztest_od_t
));
5503 * Write all the copies of our block.
5505 for (i
= 0; i
< copies
; i
++) {
5506 uint64_t offset
= i
* blocksize
;
5507 int error
= dmu_buf_hold(os
, object
, offset
, FTAG
, &db
,
5508 DMU_READ_NO_PREFETCH
);
5510 fatal(B_FALSE
, "dmu_buf_hold(%p, %llu, %llu) = %u",
5511 os
, (long long)object
, (long long) offset
, error
);
5513 ASSERT(db
->db_offset
== offset
);
5514 ASSERT(db
->db_size
== blocksize
);
5515 ASSERT(ztest_pattern_match(db
->db_data
, db
->db_size
, pattern
) ||
5516 ztest_pattern_match(db
->db_data
, db
->db_size
, 0ULL));
5517 dmu_buf_will_fill(db
, tx
);
5518 ztest_pattern_set(db
->db_data
, db
->db_size
, pattern
);
5519 dmu_buf_rele(db
, FTAG
);
5523 txg_wait_synced(spa_get_dsl(spa
), txg
);
5526 * Find out what block we got.
5528 VERIFY0(dmu_buf_hold(os
, object
, 0, FTAG
, &db
,
5529 DMU_READ_NO_PREFETCH
));
5530 blk
= *((dmu_buf_impl_t
*)db
)->db_blkptr
;
5531 dmu_buf_rele(db
, FTAG
);
5534 * Damage the block. Dedup-ditto will save us when we read it later.
5536 psize
= BP_GET_PSIZE(&blk
);
5537 abd
= abd_alloc_linear(psize
, B_TRUE
);
5538 ztest_pattern_set(abd_to_buf(abd
), psize
, ~pattern
);
5540 (void) zio_wait(zio_rewrite(NULL
, spa
, 0, &blk
,
5541 abd
, psize
, NULL
, NULL
, ZIO_PRIORITY_SYNC_WRITE
,
5542 ZIO_FLAG_CANFAIL
| ZIO_FLAG_INDUCE_DAMAGE
, NULL
));
5546 (void) rw_unlock(&ztest_name_lock
);
5547 umem_free(od
, sizeof (ztest_od_t
));
5555 ztest_scrub(ztest_ds_t
*zd
, uint64_t id
)
5557 spa_t
*spa
= ztest_spa
;
5559 (void) spa_scan(spa
, POOL_SCAN_SCRUB
);
5560 (void) poll(NULL
, 0, 100); /* wait a moment, then force a restart */
5561 (void) spa_scan(spa
, POOL_SCAN_SCRUB
);
5565 * Change the guid for the pool.
5569 ztest_reguid(ztest_ds_t
*zd
, uint64_t id
)
5571 spa_t
*spa
= ztest_spa
;
5572 uint64_t orig
, load
;
5575 orig
= spa_guid(spa
);
5576 load
= spa_load_guid(spa
);
5578 (void) rw_wrlock(&ztest_name_lock
);
5579 error
= spa_change_guid(spa
);
5580 (void) rw_unlock(&ztest_name_lock
);
5585 if (ztest_opts
.zo_verbose
>= 4) {
5586 (void) printf("Changed guid old %llu -> %llu\n",
5587 (u_longlong_t
)orig
, (u_longlong_t
)spa_guid(spa
));
5590 VERIFY3U(orig
, !=, spa_guid(spa
));
5591 VERIFY3U(load
, ==, spa_load_guid(spa
));
5595 * Rename the pool to a different name and then rename it back.
5599 ztest_spa_rename(ztest_ds_t
*zd
, uint64_t id
)
5601 char *oldname
, *newname
;
5604 (void) rw_wrlock(&ztest_name_lock
);
5606 oldname
= ztest_opts
.zo_pool
;
5607 newname
= umem_alloc(strlen(oldname
) + 5, UMEM_NOFAIL
);
5608 (void) strcpy(newname
, oldname
);
5609 (void) strcat(newname
, "_tmp");
5614 VERIFY3U(0, ==, spa_rename(oldname
, newname
));
5617 * Try to open it under the old name, which shouldn't exist
5619 VERIFY3U(ENOENT
, ==, spa_open(oldname
, &spa
, FTAG
));
5622 * Open it under the new name and make sure it's still the same spa_t.
5624 VERIFY3U(0, ==, spa_open(newname
, &spa
, FTAG
));
5626 ASSERT(spa
== ztest_spa
);
5627 spa_close(spa
, FTAG
);
5630 * Rename it back to the original
5632 VERIFY3U(0, ==, spa_rename(newname
, oldname
));
5635 * Make sure it can still be opened
5637 VERIFY3U(0, ==, spa_open(oldname
, &spa
, FTAG
));
5639 ASSERT(spa
== ztest_spa
);
5640 spa_close(spa
, FTAG
);
5642 umem_free(newname
, strlen(newname
) + 1);
5644 (void) rw_unlock(&ztest_name_lock
);
5648 ztest_fletcher(ztest_ds_t
*zd
, uint64_t id
)
5650 hrtime_t end
= gethrtime() + NANOSEC
;
5652 while (gethrtime() <= end
) {
5653 int run_count
= 100;
5659 zio_cksum_t zc_ref_byteswap
;
5661 size
= ztest_random_blocksize();
5662 buf
= umem_alloc(size
, UMEM_NOFAIL
);
5664 for (i
= 0, ptr
= buf
; i
< size
/ sizeof (*ptr
); i
++, ptr
++)
5665 *ptr
= ztest_random(UINT_MAX
);
5667 VERIFY0(fletcher_4_impl_set("scalar"));
5668 fletcher_4_native(buf
, size
, NULL
, &zc_ref
);
5669 fletcher_4_byteswap(buf
, size
, NULL
, &zc_ref_byteswap
);
5671 VERIFY0(fletcher_4_impl_set("cycle"));
5672 while (run_count
-- > 0) {
5674 zio_cksum_t zc_byteswap
;
5676 fletcher_4_byteswap(buf
, size
, NULL
, &zc_byteswap
);
5677 fletcher_4_native(buf
, size
, NULL
, &zc
);
5679 VERIFY0(bcmp(&zc
, &zc_ref
, sizeof (zc
)));
5680 VERIFY0(bcmp(&zc_byteswap
, &zc_ref_byteswap
,
5681 sizeof (zc_byteswap
)));
5684 umem_free(buf
, size
);
5689 ztest_fletcher_incr(ztest_ds_t
*zd
, uint64_t id
)
5696 zio_cksum_t zc_ref_bswap
;
5698 hrtime_t end
= gethrtime() + NANOSEC
;
5700 while (gethrtime() <= end
) {
5701 int run_count
= 100;
5703 size
= ztest_random_blocksize();
5704 buf
= umem_alloc(size
, UMEM_NOFAIL
);
5706 for (i
= 0, ptr
= buf
; i
< size
/ sizeof (*ptr
); i
++, ptr
++)
5707 *ptr
= ztest_random(UINT_MAX
);
5709 VERIFY0(fletcher_4_impl_set("scalar"));
5710 fletcher_4_native(buf
, size
, NULL
, &zc_ref
);
5711 fletcher_4_byteswap(buf
, size
, NULL
, &zc_ref_bswap
);
5713 VERIFY0(fletcher_4_impl_set("cycle"));
5715 while (run_count
-- > 0) {
5717 zio_cksum_t zc_bswap
;
5720 ZIO_SET_CHECKSUM(&zc
, 0, 0, 0, 0);
5721 ZIO_SET_CHECKSUM(&zc_bswap
, 0, 0, 0, 0);
5723 while (pos
< size
) {
5724 size_t inc
= 64 * ztest_random(size
/ 67);
5725 /* sometimes add few bytes to test non-simd */
5726 if (ztest_random(100) < 10)
5727 inc
+= P2ALIGN(ztest_random(64),
5730 if (inc
> (size
- pos
))
5733 fletcher_4_incremental_native(buf
+ pos
, inc
,
5735 fletcher_4_incremental_byteswap(buf
+ pos
, inc
,
5741 VERIFY3U(pos
, ==, size
);
5743 VERIFY(ZIO_CHECKSUM_EQUAL(zc
, zc_ref
));
5744 VERIFY(ZIO_CHECKSUM_EQUAL(zc_bswap
, zc_ref_bswap
));
5747 * verify if incremental on the whole buffer is
5748 * equivalent to non-incremental version
5750 ZIO_SET_CHECKSUM(&zc
, 0, 0, 0, 0);
5751 ZIO_SET_CHECKSUM(&zc_bswap
, 0, 0, 0, 0);
5753 fletcher_4_incremental_native(buf
, size
, &zc
);
5754 fletcher_4_incremental_byteswap(buf
, size
, &zc_bswap
);
5756 VERIFY(ZIO_CHECKSUM_EQUAL(zc
, zc_ref
));
5757 VERIFY(ZIO_CHECKSUM_EQUAL(zc_bswap
, zc_ref_bswap
));
5760 umem_free(buf
, size
);
5765 ztest_check_path(char *path
)
5768 /* return true on success */
5769 return (!stat(path
, &s
));
5773 ztest_get_zdb_bin(char *bin
, int len
)
5777 * Try to use ZDB_PATH and in-tree zdb path. If not successful, just
5778 * let popen to search through PATH.
5780 if ((zdb_path
= getenv("ZDB_PATH"))) {
5781 strlcpy(bin
, zdb_path
, len
); /* In env */
5782 if (!ztest_check_path(bin
)) {
5783 ztest_dump_core
= 0;
5784 fatal(1, "invalid ZDB_PATH '%s'", bin
);
5789 VERIFY(realpath(getexecname(), bin
) != NULL
);
5790 if (strstr(bin
, "/ztest/")) {
5791 strstr(bin
, "/ztest/")[0] = '\0'; /* In-tree */
5792 strcat(bin
, "/zdb/zdb");
5793 if (ztest_check_path(bin
))
5800 * Verify pool integrity by running zdb.
5803 ztest_run_zdb(char *pool
)
5809 const int len
= MAXPATHLEN
+ MAXNAMELEN
+ 20;
5812 bin
= umem_alloc(len
, UMEM_NOFAIL
);
5813 zdb
= umem_alloc(len
, UMEM_NOFAIL
);
5814 zbuf
= umem_alloc(1024, UMEM_NOFAIL
);
5816 ztest_get_zdb_bin(bin
, len
);
5819 "%s -bcc%s%s -d -U %s %s",
5821 ztest_opts
.zo_verbose
>= 3 ? "s" : "",
5822 ztest_opts
.zo_verbose
>= 4 ? "v" : "",
5826 if (ztest_opts
.zo_verbose
>= 5)
5827 (void) printf("Executing %s\n", strstr(zdb
, "zdb "));
5829 fp
= popen(zdb
, "r");
5831 while (fgets(zbuf
, 1024, fp
) != NULL
)
5832 if (ztest_opts
.zo_verbose
>= 3)
5833 (void) printf("%s", zbuf
);
5835 status
= pclose(fp
);
5840 ztest_dump_core
= 0;
5841 if (WIFEXITED(status
))
5842 fatal(0, "'%s' exit code %d", zdb
, WEXITSTATUS(status
));
5844 fatal(0, "'%s' died with signal %d", zdb
, WTERMSIG(status
));
5846 umem_free(bin
, len
);
5847 umem_free(zdb
, len
);
5848 umem_free(zbuf
, 1024);
5852 ztest_walk_pool_directory(char *header
)
5856 if (ztest_opts
.zo_verbose
>= 6)
5857 (void) printf("%s\n", header
);
5859 mutex_enter(&spa_namespace_lock
);
5860 while ((spa
= spa_next(spa
)) != NULL
)
5861 if (ztest_opts
.zo_verbose
>= 6)
5862 (void) printf("\t%s\n", spa_name(spa
));
5863 mutex_exit(&spa_namespace_lock
);
5867 ztest_spa_import_export(char *oldname
, char *newname
)
5869 nvlist_t
*config
, *newconfig
;
5874 if (ztest_opts
.zo_verbose
>= 4) {
5875 (void) printf("import/export: old = %s, new = %s\n",
5880 * Clean up from previous runs.
5882 (void) spa_destroy(newname
);
5885 * Get the pool's configuration and guid.
5887 VERIFY3U(0, ==, spa_open(oldname
, &spa
, FTAG
));
5890 * Kick off a scrub to tickle scrub/export races.
5892 if (ztest_random(2) == 0)
5893 (void) spa_scan(spa
, POOL_SCAN_SCRUB
);
5895 pool_guid
= spa_guid(spa
);
5896 spa_close(spa
, FTAG
);
5898 ztest_walk_pool_directory("pools before export");
5903 VERIFY3U(0, ==, spa_export(oldname
, &config
, B_FALSE
, B_FALSE
));
5905 ztest_walk_pool_directory("pools after export");
5910 newconfig
= spa_tryimport(config
);
5911 ASSERT(newconfig
!= NULL
);
5912 nvlist_free(newconfig
);
5915 * Import it under the new name.
5917 error
= spa_import(newname
, config
, NULL
, 0);
5919 dump_nvlist(config
, 0);
5920 fatal(B_FALSE
, "couldn't import pool %s as %s: error %u",
5921 oldname
, newname
, error
);
5924 ztest_walk_pool_directory("pools after import");
5927 * Try to import it again -- should fail with EEXIST.
5929 VERIFY3U(EEXIST
, ==, spa_import(newname
, config
, NULL
, 0));
5932 * Try to import it under a different name -- should fail with EEXIST.
5934 VERIFY3U(EEXIST
, ==, spa_import(oldname
, config
, NULL
, 0));
5937 * Verify that the pool is no longer visible under the old name.
5939 VERIFY3U(ENOENT
, ==, spa_open(oldname
, &spa
, FTAG
));
5942 * Verify that we can open and close the pool using the new name.
5944 VERIFY3U(0, ==, spa_open(newname
, &spa
, FTAG
));
5945 ASSERT(pool_guid
== spa_guid(spa
));
5946 spa_close(spa
, FTAG
);
5948 nvlist_free(config
);
5952 ztest_resume(spa_t
*spa
)
5954 if (spa_suspended(spa
) && ztest_opts
.zo_verbose
>= 6)
5955 (void) printf("resuming from suspended state\n");
5956 spa_vdev_state_enter(spa
, SCL_NONE
);
5957 vdev_clear(spa
, NULL
);
5958 (void) spa_vdev_state_exit(spa
, NULL
, 0);
5959 (void) zio_resume(spa
);
5963 ztest_resume_thread(void *arg
)
5967 while (!ztest_exiting
) {
5968 if (spa_suspended(spa
))
5970 (void) poll(NULL
, 0, 100);
5973 * Periodically change the zfs_compressed_arc_enabled setting.
5975 if (ztest_random(10) == 0)
5976 zfs_compressed_arc_enabled
= ztest_random(2);
5979 * Periodically change the zfs_abd_scatter_enabled setting.
5981 if (ztest_random(10) == 0)
5982 zfs_abd_scatter_enabled
= ztest_random(2);
5994 ztest_deadman_alarm(int sig
)
5996 fatal(0, "failed to complete within %d seconds of deadline", GRACE
);
6001 ztest_execute(int test
, ztest_info_t
*zi
, uint64_t id
)
6003 ztest_ds_t
*zd
= &ztest_ds
[id
% ztest_opts
.zo_datasets
];
6004 ztest_shared_callstate_t
*zc
= ZTEST_GET_SHARED_CALLSTATE(test
);
6005 hrtime_t functime
= gethrtime();
6008 for (i
= 0; i
< zi
->zi_iters
; i
++)
6009 zi
->zi_func(zd
, id
);
6011 functime
= gethrtime() - functime
;
6013 atomic_add_64(&zc
->zc_count
, 1);
6014 atomic_add_64(&zc
->zc_time
, functime
);
6016 if (ztest_opts
.zo_verbose
>= 4)
6017 (void) printf("%6.2f sec in %s\n",
6018 (double)functime
/ NANOSEC
, zi
->zi_funcname
);
6022 ztest_thread(void *arg
)
6025 uint64_t id
= (uintptr_t)arg
;
6026 ztest_shared_t
*zs
= ztest_shared
;
6030 ztest_shared_callstate_t
*zc
;
6032 while ((now
= gethrtime()) < zs
->zs_thread_stop
) {
6034 * See if it's time to force a crash.
6036 if (now
> zs
->zs_thread_kill
)
6040 * If we're getting ENOSPC with some regularity, stop.
6042 if (zs
->zs_enospc_count
> 10)
6046 * Pick a random function to execute.
6048 rand
= ztest_random(ZTEST_FUNCS
);
6049 zi
= &ztest_info
[rand
];
6050 zc
= ZTEST_GET_SHARED_CALLSTATE(rand
);
6051 call_next
= zc
->zc_next
;
6053 if (now
>= call_next
&&
6054 atomic_cas_64(&zc
->zc_next
, call_next
, call_next
+
6055 ztest_random(2 * zi
->zi_interval
[0] + 1)) == call_next
) {
6056 ztest_execute(rand
, zi
, id
);
6066 ztest_dataset_name(char *dsname
, char *pool
, int d
)
6068 (void) snprintf(dsname
, ZFS_MAX_DATASET_NAME_LEN
, "%s/ds_%d", pool
, d
);
6072 ztest_dataset_destroy(int d
)
6074 char name
[ZFS_MAX_DATASET_NAME_LEN
];
6077 ztest_dataset_name(name
, ztest_opts
.zo_pool
, d
);
6079 if (ztest_opts
.zo_verbose
>= 3)
6080 (void) printf("Destroying %s to free up space\n", name
);
6083 * Cleanup any non-standard clones and snapshots. In general,
6084 * ztest thread t operates on dataset (t % zopt_datasets),
6085 * so there may be more than one thing to clean up.
6087 for (t
= d
; t
< ztest_opts
.zo_threads
;
6088 t
+= ztest_opts
.zo_datasets
)
6089 ztest_dsl_dataset_cleanup(name
, t
);
6091 (void) dmu_objset_find(name
, ztest_objset_destroy_cb
, NULL
,
6092 DS_FIND_SNAPSHOTS
| DS_FIND_CHILDREN
);
6096 ztest_dataset_dirobj_verify(ztest_ds_t
*zd
)
6098 uint64_t usedobjs
, dirobjs
, scratch
;
6101 * ZTEST_DIROBJ is the object directory for the entire dataset.
6102 * Therefore, the number of objects in use should equal the
6103 * number of ZTEST_DIROBJ entries, +1 for ZTEST_DIROBJ itself.
6104 * If not, we have an object leak.
6106 * Note that we can only check this in ztest_dataset_open(),
6107 * when the open-context and syncing-context values agree.
6108 * That's because zap_count() returns the open-context value,
6109 * while dmu_objset_space() returns the rootbp fill count.
6111 VERIFY3U(0, ==, zap_count(zd
->zd_os
, ZTEST_DIROBJ
, &dirobjs
));
6112 dmu_objset_space(zd
->zd_os
, &scratch
, &scratch
, &usedobjs
, &scratch
);
6113 ASSERT3U(dirobjs
+ 1, ==, usedobjs
);
6117 ztest_dataset_open(int d
)
6119 ztest_ds_t
*zd
= &ztest_ds
[d
];
6120 uint64_t committed_seq
= ZTEST_GET_SHARED_DS(d
)->zd_seq
;
6123 char name
[ZFS_MAX_DATASET_NAME_LEN
];
6126 ztest_dataset_name(name
, ztest_opts
.zo_pool
, d
);
6128 (void) rw_rdlock(&ztest_name_lock
);
6130 error
= ztest_dataset_create(name
);
6131 if (error
== ENOSPC
) {
6132 (void) rw_unlock(&ztest_name_lock
);
6133 ztest_record_enospc(FTAG
);
6136 ASSERT(error
== 0 || error
== EEXIST
);
6138 VERIFY0(dmu_objset_own(name
, DMU_OST_OTHER
, B_FALSE
, zd
, &os
));
6139 (void) rw_unlock(&ztest_name_lock
);
6141 ztest_zd_init(zd
, ZTEST_GET_SHARED_DS(d
), os
);
6143 zilog
= zd
->zd_zilog
;
6145 if (zilog
->zl_header
->zh_claim_lr_seq
!= 0 &&
6146 zilog
->zl_header
->zh_claim_lr_seq
< committed_seq
)
6147 fatal(0, "missing log records: claimed %llu < committed %llu",
6148 zilog
->zl_header
->zh_claim_lr_seq
, committed_seq
);
6150 ztest_dataset_dirobj_verify(zd
);
6152 zil_replay(os
, zd
, ztest_replay_vector
);
6154 ztest_dataset_dirobj_verify(zd
);
6156 if (ztest_opts
.zo_verbose
>= 6)
6157 (void) printf("%s replay %llu blocks, %llu records, seq %llu\n",
6159 (u_longlong_t
)zilog
->zl_parse_blk_count
,
6160 (u_longlong_t
)zilog
->zl_parse_lr_count
,
6161 (u_longlong_t
)zilog
->zl_replaying_seq
);
6163 zilog
= zil_open(os
, ztest_get_data
);
6165 if (zilog
->zl_replaying_seq
!= 0 &&
6166 zilog
->zl_replaying_seq
< committed_seq
)
6167 fatal(0, "missing log records: replayed %llu < committed %llu",
6168 zilog
->zl_replaying_seq
, committed_seq
);
6174 ztest_dataset_close(int d
)
6176 ztest_ds_t
*zd
= &ztest_ds
[d
];
6178 zil_close(zd
->zd_zilog
);
6179 dmu_objset_disown(zd
->zd_os
, zd
);
6185 * Kick off threads to run tests on all datasets in parallel.
6188 ztest_run(ztest_shared_t
*zs
)
6193 kthread_t
*resume_thread
;
6198 ztest_exiting
= B_FALSE
;
6201 * Initialize parent/child shared state.
6203 mutex_init(&ztest_vdev_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
6204 VERIFY(rwlock_init(&ztest_name_lock
, USYNC_THREAD
, NULL
) == 0);
6206 zs
->zs_thread_start
= gethrtime();
6207 zs
->zs_thread_stop
=
6208 zs
->zs_thread_start
+ ztest_opts
.zo_passtime
* NANOSEC
;
6209 zs
->zs_thread_stop
= MIN(zs
->zs_thread_stop
, zs
->zs_proc_stop
);
6210 zs
->zs_thread_kill
= zs
->zs_thread_stop
;
6211 if (ztest_random(100) < ztest_opts
.zo_killrate
) {
6212 zs
->zs_thread_kill
-=
6213 ztest_random(ztest_opts
.zo_passtime
* NANOSEC
);
6216 mutex_init(&zcl
.zcl_callbacks_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
6218 list_create(&zcl
.zcl_callbacks
, sizeof (ztest_cb_data_t
),
6219 offsetof(ztest_cb_data_t
, zcd_node
));
6224 kernel_init(FREAD
| FWRITE
);
6225 VERIFY0(spa_open(ztest_opts
.zo_pool
, &spa
, FTAG
));
6226 spa
->spa_debug
= B_TRUE
;
6227 metaslab_preload_limit
= ztest_random(20) + 1;
6230 dmu_objset_stats_t dds
;
6231 VERIFY0(dmu_objset_own(ztest_opts
.zo_pool
,
6232 DMU_OST_ANY
, B_TRUE
, FTAG
, &os
));
6233 dsl_pool_config_enter(dmu_objset_pool(os
), FTAG
);
6234 dmu_objset_fast_stat(os
, &dds
);
6235 dsl_pool_config_exit(dmu_objset_pool(os
), FTAG
);
6236 zs
->zs_guid
= dds
.dds_guid
;
6237 dmu_objset_disown(os
, FTAG
);
6239 spa
->spa_dedup_ditto
= 2 * ZIO_DEDUPDITTO_MIN
;
6242 * We don't expect the pool to suspend unless maxfaults == 0,
6243 * in which case ztest_fault_inject() temporarily takes away
6244 * the only valid replica.
6246 if (MAXFAULTS() == 0)
6247 spa
->spa_failmode
= ZIO_FAILURE_MODE_WAIT
;
6249 spa
->spa_failmode
= ZIO_FAILURE_MODE_PANIC
;
6252 * Create a thread to periodically resume suspended I/O.
6254 VERIFY3P((resume_thread
= zk_thread_create(NULL
, 0,
6255 (thread_func_t
)ztest_resume_thread
, spa
, 0, NULL
, TS_RUN
, 0,
6256 PTHREAD_CREATE_JOINABLE
)), !=, NULL
);
6260 * Set a deadman alarm to abort() if we hang.
6262 signal(SIGALRM
, ztest_deadman_alarm
);
6263 alarm((zs
->zs_thread_stop
- zs
->zs_thread_start
) / NANOSEC
+ GRACE
);
6267 * Verify that we can safely inquire about about any object,
6268 * whether it's allocated or not. To make it interesting,
6269 * we probe a 5-wide window around each power of two.
6270 * This hits all edge cases, including zero and the max.
6272 for (t
= 0; t
< 64; t
++) {
6273 for (d
= -5; d
<= 5; d
++) {
6274 error
= dmu_object_info(spa
->spa_meta_objset
,
6275 (1ULL << t
) + d
, NULL
);
6276 ASSERT(error
== 0 || error
== ENOENT
||
6282 * If we got any ENOSPC errors on the previous run, destroy something.
6284 if (zs
->zs_enospc_count
!= 0) {
6285 int d
= ztest_random(ztest_opts
.zo_datasets
);
6286 ztest_dataset_destroy(d
);
6288 zs
->zs_enospc_count
= 0;
6290 tid
= umem_zalloc(ztest_opts
.zo_threads
* sizeof (kt_did_t
),
6293 if (ztest_opts
.zo_verbose
>= 4)
6294 (void) printf("starting main threads...\n");
6297 * Kick off all the tests that run in parallel.
6299 for (t
= 0; t
< ztest_opts
.zo_threads
; t
++) {
6302 if (t
< ztest_opts
.zo_datasets
&&
6303 ztest_dataset_open(t
) != 0) {
6305 ztest_opts
.zo_threads
* sizeof (kt_did_t
));
6309 VERIFY3P(thread
= zk_thread_create(NULL
, 0,
6310 (thread_func_t
)ztest_thread
,
6311 (void *)(uintptr_t)t
, 0, NULL
, TS_RUN
, 0,
6312 PTHREAD_CREATE_JOINABLE
), !=, NULL
);
6313 tid
[t
] = thread
->t_tid
;
6317 * Wait for all of the tests to complete. We go in reverse order
6318 * so we don't close datasets while threads are still using them.
6320 for (t
= ztest_opts
.zo_threads
- 1; t
>= 0; t
--) {
6321 thread_join(tid
[t
]);
6322 if (t
< ztest_opts
.zo_datasets
)
6323 ztest_dataset_close(t
);
6326 txg_wait_synced(spa_get_dsl(spa
), 0);
6328 zs
->zs_alloc
= metaslab_class_get_alloc(spa_normal_class(spa
));
6329 zs
->zs_space
= metaslab_class_get_space(spa_normal_class(spa
));
6331 umem_free(tid
, ztest_opts
.zo_threads
* sizeof (kt_did_t
));
6333 /* Kill the resume thread */
6334 ztest_exiting
= B_TRUE
;
6335 thread_join(resume_thread
->t_tid
);
6339 * Right before closing the pool, kick off a bunch of async I/O;
6340 * spa_close() should wait for it to complete.
6342 for (object
= 1; object
< 50; object
++) {
6343 dmu_prefetch(spa
->spa_meta_objset
, object
, 0, 0, 1ULL << 20,
6344 ZIO_PRIORITY_SYNC_READ
);
6347 /* Verify that at least one commit cb was called in a timely fashion */
6348 if (zc_cb_counter
>= ZTEST_COMMIT_CB_MIN_REG
)
6349 VERIFY0(zc_min_txg_delay
);
6351 spa_close(spa
, FTAG
);
6354 * Verify that we can loop over all pools.
6356 mutex_enter(&spa_namespace_lock
);
6357 for (spa
= spa_next(NULL
); spa
!= NULL
; spa
= spa_next(spa
))
6358 if (ztest_opts
.zo_verbose
> 3)
6359 (void) printf("spa_next: found %s\n", spa_name(spa
));
6360 mutex_exit(&spa_namespace_lock
);
6363 * Verify that we can export the pool and reimport it under a
6366 if (ztest_random(2) == 0) {
6367 char name
[ZFS_MAX_DATASET_NAME_LEN
];
6368 (void) snprintf(name
, sizeof (name
), "%s_import",
6369 ztest_opts
.zo_pool
);
6370 ztest_spa_import_export(ztest_opts
.zo_pool
, name
);
6371 ztest_spa_import_export(name
, ztest_opts
.zo_pool
);
6376 list_destroy(&zcl
.zcl_callbacks
);
6377 mutex_destroy(&zcl
.zcl_callbacks_lock
);
6378 (void) rwlock_destroy(&ztest_name_lock
);
6379 mutex_destroy(&ztest_vdev_lock
);
6385 ztest_ds_t
*zd
= &ztest_ds
[0];
6389 if (ztest_opts
.zo_verbose
>= 3)
6390 (void) printf("testing spa_freeze()...\n");
6392 kernel_init(FREAD
| FWRITE
);
6393 VERIFY3U(0, ==, spa_open(ztest_opts
.zo_pool
, &spa
, FTAG
));
6394 VERIFY3U(0, ==, ztest_dataset_open(0));
6395 spa
->spa_debug
= B_TRUE
;
6399 * Force the first log block to be transactionally allocated.
6400 * We have to do this before we freeze the pool -- otherwise
6401 * the log chain won't be anchored.
6403 while (BP_IS_HOLE(&zd
->zd_zilog
->zl_header
->zh_log
)) {
6404 ztest_dmu_object_alloc_free(zd
, 0);
6405 zil_commit(zd
->zd_zilog
, 0);
6408 txg_wait_synced(spa_get_dsl(spa
), 0);
6411 * Freeze the pool. This stops spa_sync() from doing anything,
6412 * so that the only way to record changes from now on is the ZIL.
6417 * Because it is hard to predict how much space a write will actually
6418 * require beforehand, we leave ourselves some fudge space to write over
6421 uint64_t capacity
= metaslab_class_get_space(spa_normal_class(spa
)) / 2;
6424 * Run tests that generate log records but don't alter the pool config
6425 * or depend on DSL sync tasks (snapshots, objset create/destroy, etc).
6426 * We do a txg_wait_synced() after each iteration to force the txg
6427 * to increase well beyond the last synced value in the uberblock.
6428 * The ZIL should be OK with that.
6430 * Run a random number of times less than zo_maxloops and ensure we do
6431 * not run out of space on the pool.
6433 while (ztest_random(10) != 0 &&
6434 numloops
++ < ztest_opts
.zo_maxloops
&&
6435 metaslab_class_get_alloc(spa_normal_class(spa
)) < capacity
) {
6437 ztest_od_init(&od
, 0, FTAG
, 0, DMU_OT_UINT64_OTHER
, 0, 0, 0);
6438 VERIFY0(ztest_object_init(zd
, &od
, sizeof (od
), B_FALSE
));
6439 ztest_io(zd
, od
.od_object
,
6440 ztest_random(ZTEST_RANGE_LOCKS
) << SPA_MAXBLOCKSHIFT
);
6441 txg_wait_synced(spa_get_dsl(spa
), 0);
6445 * Commit all of the changes we just generated.
6447 zil_commit(zd
->zd_zilog
, 0);
6448 txg_wait_synced(spa_get_dsl(spa
), 0);
6451 * Close our dataset and close the pool.
6453 ztest_dataset_close(0);
6454 spa_close(spa
, FTAG
);
6458 * Open and close the pool and dataset to induce log replay.
6460 kernel_init(FREAD
| FWRITE
);
6461 VERIFY3U(0, ==, spa_open(ztest_opts
.zo_pool
, &spa
, FTAG
));
6462 ASSERT(spa_freeze_txg(spa
) == UINT64_MAX
);
6463 VERIFY3U(0, ==, ztest_dataset_open(0));
6464 ztest_dataset_close(0);
6466 spa
->spa_debug
= B_TRUE
;
6468 txg_wait_synced(spa_get_dsl(spa
), 0);
6469 ztest_reguid(NULL
, 0);
6471 spa_close(spa
, FTAG
);
6476 print_time(hrtime_t t
, char *timebuf
)
6478 hrtime_t s
= t
/ NANOSEC
;
6479 hrtime_t m
= s
/ 60;
6480 hrtime_t h
= m
/ 60;
6481 hrtime_t d
= h
/ 24;
6490 (void) sprintf(timebuf
,
6491 "%llud%02lluh%02llum%02llus", d
, h
, m
, s
);
6493 (void) sprintf(timebuf
, "%lluh%02llum%02llus", h
, m
, s
);
6495 (void) sprintf(timebuf
, "%llum%02llus", m
, s
);
6497 (void) sprintf(timebuf
, "%llus", s
);
6501 make_random_props(void)
6505 VERIFY(nvlist_alloc(&props
, NV_UNIQUE_NAME
, 0) == 0);
6506 if (ztest_random(2) == 0)
6508 VERIFY(nvlist_add_uint64(props
, "autoreplace", 1) == 0);
6514 * Create a storage pool with the given name and initial vdev size.
6515 * Then test spa_freeze() functionality.
6518 ztest_init(ztest_shared_t
*zs
)
6521 nvlist_t
*nvroot
, *props
;
6524 mutex_init(&ztest_vdev_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
6525 VERIFY(rwlock_init(&ztest_name_lock
, USYNC_THREAD
, NULL
) == 0);
6527 kernel_init(FREAD
| FWRITE
);
6530 * Create the storage pool.
6532 (void) spa_destroy(ztest_opts
.zo_pool
);
6533 ztest_shared
->zs_vdev_next_leaf
= 0;
6535 zs
->zs_mirrors
= ztest_opts
.zo_mirrors
;
6536 nvroot
= make_vdev_root(NULL
, NULL
, NULL
, ztest_opts
.zo_vdev_size
, 0,
6537 0, ztest_opts
.zo_raidz
, zs
->zs_mirrors
, 1);
6538 props
= make_random_props();
6539 for (i
= 0; i
< SPA_FEATURES
; i
++) {
6541 VERIFY3S(-1, !=, asprintf(&buf
, "feature@%s",
6542 spa_feature_table
[i
].fi_uname
));
6543 VERIFY3U(0, ==, nvlist_add_uint64(props
, buf
, 0));
6546 VERIFY3U(0, ==, spa_create(ztest_opts
.zo_pool
, nvroot
, props
, NULL
));
6547 nvlist_free(nvroot
);
6550 VERIFY3U(0, ==, spa_open(ztest_opts
.zo_pool
, &spa
, FTAG
));
6551 zs
->zs_metaslab_sz
=
6552 1ULL << spa
->spa_root_vdev
->vdev_child
[0]->vdev_ms_shift
;
6553 spa_close(spa
, FTAG
);
6557 ztest_run_zdb(ztest_opts
.zo_pool
);
6561 ztest_run_zdb(ztest_opts
.zo_pool
);
6563 (void) rwlock_destroy(&ztest_name_lock
);
6564 mutex_destroy(&ztest_vdev_lock
);
6570 static char ztest_name_data
[] = "/tmp/ztest.data.XXXXXX";
6572 ztest_fd_data
= mkstemp(ztest_name_data
);
6573 ASSERT3S(ztest_fd_data
, >=, 0);
6574 (void) unlink(ztest_name_data
);
6578 shared_data_size(ztest_shared_hdr_t
*hdr
)
6582 size
= hdr
->zh_hdr_size
;
6583 size
+= hdr
->zh_opts_size
;
6584 size
+= hdr
->zh_size
;
6585 size
+= hdr
->zh_stats_size
* hdr
->zh_stats_count
;
6586 size
+= hdr
->zh_ds_size
* hdr
->zh_ds_count
;
6595 ztest_shared_hdr_t
*hdr
;
6597 hdr
= (void *)mmap(0, P2ROUNDUP(sizeof (*hdr
), getpagesize()),
6598 PROT_READ
| PROT_WRITE
, MAP_SHARED
, ztest_fd_data
, 0);
6599 ASSERT(hdr
!= MAP_FAILED
);
6601 VERIFY3U(0, ==, ftruncate(ztest_fd_data
, sizeof (ztest_shared_hdr_t
)));
6603 hdr
->zh_hdr_size
= sizeof (ztest_shared_hdr_t
);
6604 hdr
->zh_opts_size
= sizeof (ztest_shared_opts_t
);
6605 hdr
->zh_size
= sizeof (ztest_shared_t
);
6606 hdr
->zh_stats_size
= sizeof (ztest_shared_callstate_t
);
6607 hdr
->zh_stats_count
= ZTEST_FUNCS
;
6608 hdr
->zh_ds_size
= sizeof (ztest_shared_ds_t
);
6609 hdr
->zh_ds_count
= ztest_opts
.zo_datasets
;
6611 size
= shared_data_size(hdr
);
6612 VERIFY3U(0, ==, ftruncate(ztest_fd_data
, size
));
6614 (void) munmap((caddr_t
)hdr
, P2ROUNDUP(sizeof (*hdr
), getpagesize()));
6621 ztest_shared_hdr_t
*hdr
;
6624 hdr
= (void *)mmap(0, P2ROUNDUP(sizeof (*hdr
), getpagesize()),
6625 PROT_READ
, MAP_SHARED
, ztest_fd_data
, 0);
6626 ASSERT(hdr
!= MAP_FAILED
);
6628 size
= shared_data_size(hdr
);
6630 (void) munmap((caddr_t
)hdr
, P2ROUNDUP(sizeof (*hdr
), getpagesize()));
6631 hdr
= ztest_shared_hdr
= (void *)mmap(0, P2ROUNDUP(size
, getpagesize()),
6632 PROT_READ
| PROT_WRITE
, MAP_SHARED
, ztest_fd_data
, 0);
6633 ASSERT(hdr
!= MAP_FAILED
);
6634 buf
= (uint8_t *)hdr
;
6636 offset
= hdr
->zh_hdr_size
;
6637 ztest_shared_opts
= (void *)&buf
[offset
];
6638 offset
+= hdr
->zh_opts_size
;
6639 ztest_shared
= (void *)&buf
[offset
];
6640 offset
+= hdr
->zh_size
;
6641 ztest_shared_callstate
= (void *)&buf
[offset
];
6642 offset
+= hdr
->zh_stats_size
* hdr
->zh_stats_count
;
6643 ztest_shared_ds
= (void *)&buf
[offset
];
6647 exec_child(char *cmd
, char *libpath
, boolean_t ignorekill
, int *statusp
)
6651 char *cmdbuf
= NULL
;
6656 cmdbuf
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
6657 (void) strlcpy(cmdbuf
, getexecname(), MAXPATHLEN
);
6662 fatal(1, "fork failed");
6664 if (pid
== 0) { /* child */
6665 char *emptyargv
[2] = { cmd
, NULL
};
6666 char fd_data_str
[12];
6668 struct rlimit rl
= { 1024, 1024 };
6669 (void) setrlimit(RLIMIT_NOFILE
, &rl
);
6671 (void) close(ztest_fd_rand
);
6672 VERIFY(11 >= snprintf(fd_data_str
, 12, "%d", ztest_fd_data
));
6673 VERIFY(0 == setenv("ZTEST_FD_DATA", fd_data_str
, 1));
6675 (void) enable_extended_FILE_stdio(-1, -1);
6676 if (libpath
!= NULL
)
6677 VERIFY(0 == setenv("LD_LIBRARY_PATH", libpath
, 1));
6678 (void) execv(cmd
, emptyargv
);
6679 ztest_dump_core
= B_FALSE
;
6680 fatal(B_TRUE
, "exec failed: %s", cmd
);
6683 if (cmdbuf
!= NULL
) {
6684 umem_free(cmdbuf
, MAXPATHLEN
);
6688 while (waitpid(pid
, &status
, 0) != pid
)
6690 if (statusp
!= NULL
)
6693 if (WIFEXITED(status
)) {
6694 if (WEXITSTATUS(status
) != 0) {
6695 (void) fprintf(stderr
, "child exited with code %d\n",
6696 WEXITSTATUS(status
));
6700 } else if (WIFSIGNALED(status
)) {
6701 if (!ignorekill
|| WTERMSIG(status
) != SIGKILL
) {
6702 (void) fprintf(stderr
, "child died with signal %d\n",
6708 (void) fprintf(stderr
, "something strange happened to child\n");
6715 ztest_run_init(void)
6719 ztest_shared_t
*zs
= ztest_shared
;
6721 ASSERT(ztest_opts
.zo_init
!= 0);
6724 * Blow away any existing copy of zpool.cache
6726 (void) remove(spa_config_path
);
6729 * Create and initialize our storage pool.
6731 for (i
= 1; i
<= ztest_opts
.zo_init
; i
++) {
6732 bzero(zs
, sizeof (ztest_shared_t
));
6733 if (ztest_opts
.zo_verbose
>= 3 &&
6734 ztest_opts
.zo_init
!= 1) {
6735 (void) printf("ztest_init(), pass %d\n", i
);
6742 main(int argc
, char **argv
)
6750 ztest_shared_callstate_t
*zc
;
6757 char *fd_data_str
= getenv("ZTEST_FD_DATA");
6758 struct sigaction action
;
6760 (void) setvbuf(stdout
, NULL
, _IOLBF
, 0);
6762 dprintf_setup(&argc
, argv
);
6764 action
.sa_handler
= sig_handler
;
6765 sigemptyset(&action
.sa_mask
);
6766 action
.sa_flags
= 0;
6768 if (sigaction(SIGSEGV
, &action
, NULL
) < 0) {
6769 (void) fprintf(stderr
, "ztest: cannot catch SIGSEGV: %s.\n",
6774 if (sigaction(SIGABRT
, &action
, NULL
) < 0) {
6775 (void) fprintf(stderr
, "ztest: cannot catch SIGABRT: %s.\n",
6780 ztest_fd_rand
= open("/dev/urandom", O_RDONLY
);
6781 ASSERT3S(ztest_fd_rand
, >=, 0);
6784 process_options(argc
, argv
);
6789 bcopy(&ztest_opts
, ztest_shared_opts
,
6790 sizeof (*ztest_shared_opts
));
6792 ztest_fd_data
= atoi(fd_data_str
);
6794 bcopy(ztest_shared_opts
, &ztest_opts
, sizeof (ztest_opts
));
6796 ASSERT3U(ztest_opts
.zo_datasets
, ==, ztest_shared_hdr
->zh_ds_count
);
6798 /* Override location of zpool.cache */
6799 VERIFY(asprintf((char **)&spa_config_path
, "%s/zpool.cache",
6800 ztest_opts
.zo_dir
) != -1);
6802 ztest_ds
= umem_alloc(ztest_opts
.zo_datasets
* sizeof (ztest_ds_t
),
6807 metaslab_gang_bang
= ztest_opts
.zo_metaslab_gang_bang
;
6808 metaslab_df_alloc_threshold
=
6809 zs
->zs_metaslab_df_alloc_threshold
;
6818 hasalt
= (strlen(ztest_opts
.zo_alt_ztest
) != 0);
6820 if (ztest_opts
.zo_verbose
>= 1) {
6821 (void) printf("%llu vdevs, %d datasets, %d threads,"
6822 " %llu seconds...\n",
6823 (u_longlong_t
)ztest_opts
.zo_vdevs
,
6824 ztest_opts
.zo_datasets
,
6825 ztest_opts
.zo_threads
,
6826 (u_longlong_t
)ztest_opts
.zo_time
);
6829 cmd
= umem_alloc(MAXNAMELEN
, UMEM_NOFAIL
);
6830 (void) strlcpy(cmd
, getexecname(), MAXNAMELEN
);
6832 zs
->zs_do_init
= B_TRUE
;
6833 if (strlen(ztest_opts
.zo_alt_ztest
) != 0) {
6834 if (ztest_opts
.zo_verbose
>= 1) {
6835 (void) printf("Executing older ztest for "
6836 "initialization: %s\n", ztest_opts
.zo_alt_ztest
);
6838 VERIFY(!exec_child(ztest_opts
.zo_alt_ztest
,
6839 ztest_opts
.zo_alt_libpath
, B_FALSE
, NULL
));
6841 VERIFY(!exec_child(NULL
, NULL
, B_FALSE
, NULL
));
6843 zs
->zs_do_init
= B_FALSE
;
6845 zs
->zs_proc_start
= gethrtime();
6846 zs
->zs_proc_stop
= zs
->zs_proc_start
+ ztest_opts
.zo_time
* NANOSEC
;
6848 for (f
= 0; f
< ZTEST_FUNCS
; f
++) {
6849 zi
= &ztest_info
[f
];
6850 zc
= ZTEST_GET_SHARED_CALLSTATE(f
);
6851 if (zs
->zs_proc_start
+ zi
->zi_interval
[0] > zs
->zs_proc_stop
)
6852 zc
->zc_next
= UINT64_MAX
;
6854 zc
->zc_next
= zs
->zs_proc_start
+
6855 ztest_random(2 * zi
->zi_interval
[0] + 1);
6859 * Run the tests in a loop. These tests include fault injection
6860 * to verify that self-healing data works, and forced crashes
6861 * to verify that we never lose on-disk consistency.
6863 while (gethrtime() < zs
->zs_proc_stop
) {
6868 * Initialize the workload counters for each function.
6870 for (f
= 0; f
< ZTEST_FUNCS
; f
++) {
6871 zc
= ZTEST_GET_SHARED_CALLSTATE(f
);
6876 /* Set the allocation switch size */
6877 zs
->zs_metaslab_df_alloc_threshold
=
6878 ztest_random(zs
->zs_metaslab_sz
/ 4) + 1;
6880 if (!hasalt
|| ztest_random(2) == 0) {
6881 if (hasalt
&& ztest_opts
.zo_verbose
>= 1) {
6882 (void) printf("Executing newer ztest: %s\n",
6886 killed
= exec_child(cmd
, NULL
, B_TRUE
, &status
);
6888 if (hasalt
&& ztest_opts
.zo_verbose
>= 1) {
6889 (void) printf("Executing older ztest: %s\n",
6890 ztest_opts
.zo_alt_ztest
);
6893 killed
= exec_child(ztest_opts
.zo_alt_ztest
,
6894 ztest_opts
.zo_alt_libpath
, B_TRUE
, &status
);
6901 if (ztest_opts
.zo_verbose
>= 1) {
6902 hrtime_t now
= gethrtime();
6904 now
= MIN(now
, zs
->zs_proc_stop
);
6905 print_time(zs
->zs_proc_stop
- now
, timebuf
);
6906 nicenum(zs
->zs_space
, numbuf
);
6908 (void) printf("Pass %3d, %8s, %3llu ENOSPC, "
6909 "%4.1f%% of %5s used, %3.0f%% done, %8s to go\n",
6911 WIFEXITED(status
) ? "Complete" : "SIGKILL",
6912 (u_longlong_t
)zs
->zs_enospc_count
,
6913 100.0 * zs
->zs_alloc
/ zs
->zs_space
,
6915 100.0 * (now
- zs
->zs_proc_start
) /
6916 (ztest_opts
.zo_time
* NANOSEC
), timebuf
);
6919 if (ztest_opts
.zo_verbose
>= 2) {
6920 (void) printf("\nWorkload summary:\n\n");
6921 (void) printf("%7s %9s %s\n",
6922 "Calls", "Time", "Function");
6923 (void) printf("%7s %9s %s\n",
6924 "-----", "----", "--------");
6925 for (f
= 0; f
< ZTEST_FUNCS
; f
++) {
6926 zi
= &ztest_info
[f
];
6927 zc
= ZTEST_GET_SHARED_CALLSTATE(f
);
6928 print_time(zc
->zc_time
, timebuf
);
6929 (void) printf("%7llu %9s %s\n",
6930 (u_longlong_t
)zc
->zc_count
, timebuf
,
6933 (void) printf("\n");
6937 * It's possible that we killed a child during a rename test,
6938 * in which case we'll have a 'ztest_tmp' pool lying around
6939 * instead of 'ztest'. Do a blind rename in case this happened.
6942 if (spa_open(ztest_opts
.zo_pool
, &spa
, FTAG
) == 0) {
6943 spa_close(spa
, FTAG
);
6945 char tmpname
[ZFS_MAX_DATASET_NAME_LEN
];
6947 kernel_init(FREAD
| FWRITE
);
6948 (void) snprintf(tmpname
, sizeof (tmpname
), "%s_tmp",
6949 ztest_opts
.zo_pool
);
6950 (void) spa_rename(tmpname
, ztest_opts
.zo_pool
);
6954 ztest_run_zdb(ztest_opts
.zo_pool
);
6957 if (ztest_opts
.zo_verbose
>= 1) {
6959 (void) printf("%d runs of older ztest: %s\n", older
,
6960 ztest_opts
.zo_alt_ztest
);
6961 (void) printf("%d runs of newer ztest: %s\n", newer
,
6964 (void) printf("%d killed, %d completed, %.0f%% kill rate\n",
6965 kills
, iters
- kills
, (100.0 * kills
) / MAX(1, iters
));
6968 umem_free(cmd
, MAXNAMELEN
);