4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2011, 2016 by Delphix. All rights reserved.
24 * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
25 * Copyright (c) 2013 Steven Hartland. All rights reserved.
26 * Copyright (c) 2014 Integros [integros.com]
27 * Copyright 2017 Joyent, Inc.
31 * The objective of this program is to provide a DMU/ZAP/SPA stress test
32 * that runs entirely in userland, is easy to use, and easy to extend.
34 * The overall design of the ztest program is as follows:
36 * (1) For each major functional area (e.g. adding vdevs to a pool,
37 * creating and destroying datasets, reading and writing objects, etc)
38 * we have a simple routine to test that functionality. These
39 * individual routines do not have to do anything "stressful".
41 * (2) We turn these simple functionality tests into a stress test by
42 * running them all in parallel, with as many threads as desired,
43 * and spread across as many datasets, objects, and vdevs as desired.
45 * (3) While all this is happening, we inject faults into the pool to
46 * verify that self-healing data really works.
48 * (4) Every time we open a dataset, we change its checksum and compression
49 * functions. Thus even individual objects vary from block to block
50 * in which checksum they use and whether they're compressed.
52 * (5) To verify that we never lose on-disk consistency after a crash,
53 * we run the entire test in a child of the main process.
54 * At random times, the child self-immolates with a SIGKILL.
55 * This is the software equivalent of pulling the power cord.
56 * The parent then runs the test again, using the existing
57 * storage pool, as many times as desired. If backwards compatibility
58 * testing is enabled ztest will sometimes run the "older" version
59 * of ztest after a SIGKILL.
61 * (6) To verify that we don't have future leaks or temporal incursions,
62 * many of the functional tests record the transaction group number
63 * as part of their data. When reading old data, they verify that
64 * the transaction group number is less than the current, open txg.
65 * If you add a new test, please do this if applicable.
67 * (7) Threads are created with a reduced stack size, for sanity checking.
68 * Therefore, it's important not to allocate huge buffers on the stack.
70 * When run with no arguments, ztest runs for about five minutes and
71 * produces no output if successful. To get a little bit of information,
72 * specify -V. To get more information, specify -VV, and so on.
74 * To turn this into an overnight stress test, use -T to specify run time.
76 * You can ask more more vdevs [-v], datasets [-d], or threads [-t]
77 * to increase the pool capacity, fanout, and overall stress level.
79 * Use the -k option to set the desired frequency of kills.
81 * When ztest invokes itself it passes all relevant information through a
82 * temporary file which is mmap-ed in the child process. This allows shared
83 * memory to survive the exec syscall. The ztest_shared_hdr_t struct is always
84 * stored at offset 0 of this file and contains information on the size and
85 * number of shared structures in the file. The information stored in this file
86 * must remain backwards compatible with older versions of ztest so that
87 * ztest can invoke them during backwards compatibility testing (-B).
90 #include <sys/zfs_context.h>
96 #include <sys/dmu_objset.h>
100 #include <sys/wait.h>
101 #include <sys/mman.h>
102 #include <sys/resource.h>
105 #include <sys/zil_impl.h>
106 #include <sys/zfs_rlock.h>
107 #include <sys/vdev_impl.h>
108 #include <sys/vdev_file.h>
109 #include <sys/spa_impl.h>
110 #include <sys/metaslab_impl.h>
111 #include <sys/dsl_prop.h>
112 #include <sys/dsl_dataset.h>
113 #include <sys/dsl_destroy.h>
114 #include <sys/dsl_scan.h>
115 #include <sys/zio_checksum.h>
116 #include <sys/refcount.h>
117 #include <sys/zfeature.h>
118 #include <sys/dsl_userhold.h>
121 #include <stdio_ext.h>
128 #include <sys/fs/zfs.h>
129 #include <zfs_fletcher.h>
130 #include <libnvpair.h>
133 #include <execinfo.h> /* for backtrace() */
136 static int ztest_fd_data
= -1;
137 static int ztest_fd_rand
= -1;
139 typedef struct ztest_shared_hdr
{
140 uint64_t zh_hdr_size
;
141 uint64_t zh_opts_size
;
143 uint64_t zh_stats_size
;
144 uint64_t zh_stats_count
;
146 uint64_t zh_ds_count
;
147 } ztest_shared_hdr_t
;
149 static ztest_shared_hdr_t
*ztest_shared_hdr
;
151 typedef struct ztest_shared_opts
{
152 char zo_pool
[ZFS_MAX_DATASET_NAME_LEN
];
153 char zo_dir
[ZFS_MAX_DATASET_NAME_LEN
];
154 char zo_alt_ztest
[MAXNAMELEN
];
155 char zo_alt_libpath
[MAXNAMELEN
];
157 uint64_t zo_vdevtime
;
165 uint64_t zo_passtime
;
166 uint64_t zo_killrate
;
170 uint64_t zo_maxloops
;
171 uint64_t zo_metaslab_gang_bang
;
173 } ztest_shared_opts_t
;
175 static const ztest_shared_opts_t ztest_opts_defaults
= {
176 .zo_pool
= { 'z', 't', 'e', 's', 't', '\0' },
177 .zo_dir
= { '/', 't', 'm', 'p', '\0' },
178 .zo_alt_ztest
= { '\0' },
179 .zo_alt_libpath
= { '\0' },
181 .zo_ashift
= SPA_MINBLOCKSHIFT
,
184 .zo_raidz_parity
= 1,
185 .zo_vdev_size
= SPA_MINDEVSIZE
* 4, /* 256m default size */
188 .zo_passtime
= 60, /* 60 seconds */
189 .zo_killrate
= 70, /* 70% kill rate */
193 .zo_time
= 300, /* 5 minutes */
194 .zo_maxloops
= 50, /* max loops during spa_freeze() */
195 .zo_metaslab_gang_bang
= 32 << 10
198 extern uint64_t metaslab_gang_bang
;
199 extern uint64_t metaslab_df_alloc_threshold
;
200 extern int metaslab_preload_limit
;
201 extern boolean_t zfs_compressed_arc_enabled
;
202 extern int zfs_abd_scatter_enabled
;
204 static ztest_shared_opts_t
*ztest_shared_opts
;
205 static ztest_shared_opts_t ztest_opts
;
206 static char *ztest_wkeydata
= "abcdefghijklmnopqrstuvwxyz012345";
208 typedef struct ztest_shared_ds
{
212 static ztest_shared_ds_t
*ztest_shared_ds
;
213 #define ZTEST_GET_SHARED_DS(d) (&ztest_shared_ds[d])
215 #define BT_MAGIC 0x123456789abcdefULL
216 #define MAXFAULTS() \
217 (MAX(zs->zs_mirrors, 1) * (ztest_opts.zo_raidz_parity + 1) - 1)
221 ZTEST_IO_WRITE_PATTERN
,
222 ZTEST_IO_WRITE_ZEROES
,
229 typedef struct ztest_block_tag
{
233 uint64_t bt_dnodesize
;
240 typedef struct bufwad
{
258 #define ZTEST_RANGE_LOCKS 64
259 #define ZTEST_OBJECT_LOCKS 64
262 * Object descriptor. Used as a template for object lookup/create/remove.
264 typedef struct ztest_od
{
267 dmu_object_type_t od_type
;
268 dmu_object_type_t od_crtype
;
269 uint64_t od_blocksize
;
270 uint64_t od_crblocksize
;
271 uint64_t od_crdnodesize
;
274 char od_name
[ZFS_MAX_DATASET_NAME_LEN
];
280 typedef struct ztest_ds
{
281 ztest_shared_ds_t
*zd_shared
;
283 rwlock_t zd_zilog_lock
;
285 ztest_od_t
*zd_od
; /* debugging aid */
286 char zd_name
[ZFS_MAX_DATASET_NAME_LEN
];
287 kmutex_t zd_dirobj_lock
;
288 rll_t zd_object_lock
[ZTEST_OBJECT_LOCKS
];
289 zll_t zd_range_lock
[ZTEST_RANGE_LOCKS
];
293 * Per-iteration state.
295 typedef void ztest_func_t(ztest_ds_t
*zd
, uint64_t id
);
297 typedef struct ztest_info
{
298 ztest_func_t
*zi_func
; /* test function */
299 uint64_t zi_iters
; /* iterations per execution */
300 uint64_t *zi_interval
; /* execute every <interval> seconds */
301 const char *zi_funcname
; /* name of test function */
304 typedef struct ztest_shared_callstate
{
305 uint64_t zc_count
; /* per-pass count */
306 uint64_t zc_time
; /* per-pass time */
307 uint64_t zc_next
; /* next time to call this function */
308 } ztest_shared_callstate_t
;
310 static ztest_shared_callstate_t
*ztest_shared_callstate
;
311 #define ZTEST_GET_SHARED_CALLSTATE(c) (&ztest_shared_callstate[c])
313 ztest_func_t ztest_dmu_read_write
;
314 ztest_func_t ztest_dmu_write_parallel
;
315 ztest_func_t ztest_dmu_object_alloc_free
;
316 ztest_func_t ztest_dmu_commit_callbacks
;
317 ztest_func_t ztest_zap
;
318 ztest_func_t ztest_zap_parallel
;
319 ztest_func_t ztest_zil_commit
;
320 ztest_func_t ztest_zil_remount
;
321 ztest_func_t ztest_dmu_read_write_zcopy
;
322 ztest_func_t ztest_dmu_objset_create_destroy
;
323 ztest_func_t ztest_dmu_prealloc
;
324 ztest_func_t ztest_fzap
;
325 ztest_func_t ztest_dmu_snapshot_create_destroy
;
326 ztest_func_t ztest_dsl_prop_get_set
;
327 ztest_func_t ztest_spa_prop_get_set
;
328 ztest_func_t ztest_spa_create_destroy
;
329 ztest_func_t ztest_fault_inject
;
330 ztest_func_t ztest_ddt_repair
;
331 ztest_func_t ztest_dmu_snapshot_hold
;
332 ztest_func_t ztest_mmp_enable_disable
;
333 ztest_func_t ztest_spa_rename
;
334 ztest_func_t ztest_scrub
;
335 ztest_func_t ztest_dsl_dataset_promote_busy
;
336 ztest_func_t ztest_vdev_attach_detach
;
337 ztest_func_t ztest_vdev_LUN_growth
;
338 ztest_func_t ztest_vdev_add_remove
;
339 ztest_func_t ztest_vdev_aux_add_remove
;
340 ztest_func_t ztest_split_pool
;
341 ztest_func_t ztest_reguid
;
342 ztest_func_t ztest_spa_upgrade
;
343 ztest_func_t ztest_fletcher
;
344 ztest_func_t ztest_fletcher_incr
;
345 ztest_func_t ztest_verify_dnode_bt
;
347 uint64_t zopt_always
= 0ULL * NANOSEC
; /* all the time */
348 uint64_t zopt_incessant
= 1ULL * NANOSEC
/ 10; /* every 1/10 second */
349 uint64_t zopt_often
= 1ULL * NANOSEC
; /* every second */
350 uint64_t zopt_sometimes
= 10ULL * NANOSEC
; /* every 10 seconds */
351 uint64_t zopt_rarely
= 60ULL * NANOSEC
; /* every 60 seconds */
353 #define ZTI_INIT(func, iters, interval) \
354 { .zi_func = (func), \
355 .zi_iters = (iters), \
356 .zi_interval = (interval), \
357 .zi_funcname = # func }
359 ztest_info_t ztest_info
[] = {
360 ZTI_INIT(ztest_dmu_read_write
, 1, &zopt_always
),
361 ZTI_INIT(ztest_dmu_write_parallel
, 10, &zopt_always
),
362 ZTI_INIT(ztest_dmu_object_alloc_free
, 1, &zopt_always
),
363 ZTI_INIT(ztest_dmu_commit_callbacks
, 1, &zopt_always
),
364 ZTI_INIT(ztest_zap
, 30, &zopt_always
),
365 ZTI_INIT(ztest_zap_parallel
, 100, &zopt_always
),
366 ZTI_INIT(ztest_split_pool
, 1, &zopt_always
),
367 ZTI_INIT(ztest_zil_commit
, 1, &zopt_incessant
),
368 ZTI_INIT(ztest_zil_remount
, 1, &zopt_sometimes
),
369 ZTI_INIT(ztest_dmu_read_write_zcopy
, 1, &zopt_often
),
370 ZTI_INIT(ztest_dmu_objset_create_destroy
, 1, &zopt_often
),
371 ZTI_INIT(ztest_dsl_prop_get_set
, 1, &zopt_often
),
372 ZTI_INIT(ztest_spa_prop_get_set
, 1, &zopt_sometimes
),
374 ZTI_INIT(ztest_dmu_prealloc
, 1, &zopt_sometimes
),
376 ZTI_INIT(ztest_fzap
, 1, &zopt_sometimes
),
377 ZTI_INIT(ztest_dmu_snapshot_create_destroy
, 1, &zopt_sometimes
),
378 ZTI_INIT(ztest_spa_create_destroy
, 1, &zopt_sometimes
),
379 ZTI_INIT(ztest_fault_inject
, 1, &zopt_sometimes
),
380 ZTI_INIT(ztest_ddt_repair
, 1, &zopt_sometimes
),
381 ZTI_INIT(ztest_dmu_snapshot_hold
, 1, &zopt_sometimes
),
382 ZTI_INIT(ztest_mmp_enable_disable
, 1, &zopt_sometimes
),
383 ZTI_INIT(ztest_reguid
, 1, &zopt_rarely
),
384 ZTI_INIT(ztest_spa_rename
, 1, &zopt_rarely
),
385 ZTI_INIT(ztest_scrub
, 1, &zopt_rarely
),
386 ZTI_INIT(ztest_spa_upgrade
, 1, &zopt_rarely
),
387 ZTI_INIT(ztest_dsl_dataset_promote_busy
, 1, &zopt_rarely
),
388 ZTI_INIT(ztest_vdev_attach_detach
, 1, &zopt_sometimes
),
389 ZTI_INIT(ztest_vdev_LUN_growth
, 1, &zopt_rarely
),
390 ZTI_INIT(ztest_vdev_add_remove
, 1, &ztest_opts
.zo_vdevtime
),
391 ZTI_INIT(ztest_vdev_aux_add_remove
, 1, &ztest_opts
.zo_vdevtime
),
392 ZTI_INIT(ztest_fletcher
, 1, &zopt_rarely
),
393 ZTI_INIT(ztest_fletcher_incr
, 1, &zopt_rarely
),
394 ZTI_INIT(ztest_verify_dnode_bt
, 1, &zopt_sometimes
),
397 #define ZTEST_FUNCS (sizeof (ztest_info) / sizeof (ztest_info_t))
400 * The following struct is used to hold a list of uncalled commit callbacks.
401 * The callbacks are ordered by txg number.
403 typedef struct ztest_cb_list
{
404 kmutex_t zcl_callbacks_lock
;
405 list_t zcl_callbacks
;
409 * Stuff we need to share writably between parent and child.
411 typedef struct ztest_shared
{
412 boolean_t zs_do_init
;
413 hrtime_t zs_proc_start
;
414 hrtime_t zs_proc_stop
;
415 hrtime_t zs_thread_start
;
416 hrtime_t zs_thread_stop
;
417 hrtime_t zs_thread_kill
;
418 uint64_t zs_enospc_count
;
419 uint64_t zs_vdev_next_leaf
;
420 uint64_t zs_vdev_aux
;
425 uint64_t zs_metaslab_sz
;
426 uint64_t zs_metaslab_df_alloc_threshold
;
430 #define ID_PARALLEL -1ULL
432 static char ztest_dev_template
[] = "%s/%s.%llua";
433 static char ztest_aux_template
[] = "%s/%s.%s.%llu";
434 ztest_shared_t
*ztest_shared
;
436 static spa_t
*ztest_spa
= NULL
;
437 static ztest_ds_t
*ztest_ds
;
439 static kmutex_t ztest_vdev_lock
;
442 * The ztest_name_lock protects the pool and dataset namespace used by
443 * the individual tests. To modify the namespace, consumers must grab
444 * this lock as writer. Grabbing the lock as reader will ensure that the
445 * namespace does not change while the lock is held.
447 static rwlock_t ztest_name_lock
;
449 static boolean_t ztest_dump_core
= B_TRUE
;
450 static boolean_t ztest_exiting
;
452 /* Global commit callback list */
453 static ztest_cb_list_t zcl
;
454 /* Commit cb delay */
455 static uint64_t zc_min_txg_delay
= UINT64_MAX
;
456 static int zc_cb_counter
= 0;
459 * Minimum number of commit callbacks that need to be registered for us to check
460 * whether the minimum txg delay is acceptable.
462 #define ZTEST_COMMIT_CB_MIN_REG 100
465 * If a number of txgs equal to this threshold have been created after a commit
466 * callback has been registered but not called, then we assume there is an
467 * implementation bug.
469 #define ZTEST_COMMIT_CB_THRESH (TXG_CONCURRENT_STATES + 1000)
471 extern uint64_t metaslab_gang_bang
;
472 extern uint64_t metaslab_df_alloc_threshold
;
475 ZTEST_META_DNODE
= 0,
480 static void usage(boolean_t
) __NORETURN
;
483 * These libumem hooks provide a reasonable set of defaults for the allocator's
484 * debugging facilities.
487 _umem_debug_init(void)
489 return ("default,verbose"); /* $UMEM_DEBUG setting */
493 _umem_logging_init(void)
495 return ("fail,contents"); /* $UMEM_LOGGING setting */
498 #define BACKTRACE_SZ 100
500 static void sig_handler(int signo
)
502 struct sigaction action
;
503 #ifdef __GLIBC__ /* backtrace() is a GNU extension */
505 void *buffer
[BACKTRACE_SZ
];
507 nptrs
= backtrace(buffer
, BACKTRACE_SZ
);
508 backtrace_symbols_fd(buffer
, nptrs
, STDERR_FILENO
);
512 * Restore default action and re-raise signal so SIGSEGV and
513 * SIGABRT can trigger a core dump.
515 action
.sa_handler
= SIG_DFL
;
516 sigemptyset(&action
.sa_mask
);
518 (void) sigaction(signo
, &action
, NULL
);
522 #define FATAL_MSG_SZ 1024
527 fatal(int do_perror
, char *message
, ...)
530 int save_errno
= errno
;
533 (void) fflush(stdout
);
534 buf
= umem_alloc(FATAL_MSG_SZ
, UMEM_NOFAIL
);
536 va_start(args
, message
);
537 (void) sprintf(buf
, "ztest: ");
539 (void) vsprintf(buf
+ strlen(buf
), message
, args
);
542 (void) snprintf(buf
+ strlen(buf
), FATAL_MSG_SZ
- strlen(buf
),
543 ": %s", strerror(save_errno
));
545 (void) fprintf(stderr
, "%s\n", buf
);
546 fatal_msg
= buf
; /* to ease debugging */
553 str2shift(const char *buf
)
555 const char *ends
= "BKMGTPEZ";
560 for (i
= 0; i
< strlen(ends
); i
++) {
561 if (toupper(buf
[0]) == ends
[i
])
564 if (i
== strlen(ends
)) {
565 (void) fprintf(stderr
, "ztest: invalid bytes suffix: %s\n",
569 if (buf
[1] == '\0' || (toupper(buf
[1]) == 'B' && buf
[2] == '\0')) {
572 (void) fprintf(stderr
, "ztest: invalid bytes suffix: %s\n", buf
);
578 nicenumtoull(const char *buf
)
583 val
= strtoull(buf
, &end
, 0);
585 (void) fprintf(stderr
, "ztest: bad numeric value: %s\n", buf
);
587 } else if (end
[0] == '.') {
588 double fval
= strtod(buf
, &end
);
589 fval
*= pow(2, str2shift(end
));
590 if (fval
> UINT64_MAX
) {
591 (void) fprintf(stderr
, "ztest: value too large: %s\n",
595 val
= (uint64_t)fval
;
597 int shift
= str2shift(end
);
598 if (shift
>= 64 || (val
<< shift
) >> shift
!= val
) {
599 (void) fprintf(stderr
, "ztest: value too large: %s\n",
609 usage(boolean_t requested
)
611 const ztest_shared_opts_t
*zo
= &ztest_opts_defaults
;
613 char nice_vdev_size
[NN_NUMBUF_SZ
];
614 char nice_gang_bang
[NN_NUMBUF_SZ
];
615 FILE *fp
= requested
? stdout
: stderr
;
617 nicenum(zo
->zo_vdev_size
, nice_vdev_size
, sizeof (nice_vdev_size
));
618 nicenum(zo
->zo_metaslab_gang_bang
, nice_gang_bang
,
619 sizeof (nice_gang_bang
));
621 (void) fprintf(fp
, "Usage: %s\n"
622 "\t[-v vdevs (default: %llu)]\n"
623 "\t[-s size_of_each_vdev (default: %s)]\n"
624 "\t[-a alignment_shift (default: %d)] use 0 for random\n"
625 "\t[-m mirror_copies (default: %d)]\n"
626 "\t[-r raidz_disks (default: %d)]\n"
627 "\t[-R raidz_parity (default: %d)]\n"
628 "\t[-d datasets (default: %d)]\n"
629 "\t[-t threads (default: %d)]\n"
630 "\t[-g gang_block_threshold (default: %s)]\n"
631 "\t[-i init_count (default: %d)] initialize pool i times\n"
632 "\t[-k kill_percentage (default: %llu%%)]\n"
633 "\t[-p pool_name (default: %s)]\n"
634 "\t[-f dir (default: %s)] file directory for vdev files\n"
635 "\t[-M] Multi-host simulate pool imported on remote host\n"
636 "\t[-V] verbose (use multiple times for ever more blather)\n"
637 "\t[-E] use existing pool instead of creating new one\n"
638 "\t[-T time (default: %llu sec)] total run time\n"
639 "\t[-F freezeloops (default: %llu)] max loops in spa_freeze()\n"
640 "\t[-P passtime (default: %llu sec)] time per pass\n"
641 "\t[-B alt_ztest (default: <none>)] alternate ztest path\n"
642 "\t[-o variable=value] ... set global variable to an unsigned\n"
643 "\t 32-bit integer value\n"
644 "\t[-h] (print help)\n"
647 (u_longlong_t
)zo
->zo_vdevs
, /* -v */
648 nice_vdev_size
, /* -s */
649 zo
->zo_ashift
, /* -a */
650 zo
->zo_mirrors
, /* -m */
651 zo
->zo_raidz
, /* -r */
652 zo
->zo_raidz_parity
, /* -R */
653 zo
->zo_datasets
, /* -d */
654 zo
->zo_threads
, /* -t */
655 nice_gang_bang
, /* -g */
656 zo
->zo_init
, /* -i */
657 (u_longlong_t
)zo
->zo_killrate
, /* -k */
658 zo
->zo_pool
, /* -p */
660 (u_longlong_t
)zo
->zo_time
, /* -T */
661 (u_longlong_t
)zo
->zo_maxloops
, /* -F */
662 (u_longlong_t
)zo
->zo_passtime
);
663 exit(requested
? 0 : 1);
667 process_options(int argc
, char **argv
)
670 ztest_shared_opts_t
*zo
= &ztest_opts
;
674 char altdir
[MAXNAMELEN
] = { 0 };
676 bcopy(&ztest_opts_defaults
, zo
, sizeof (*zo
));
678 while ((opt
= getopt(argc
, argv
,
679 "v:s:a:m:r:R:d:t:g:i:k:p:f:MVET:P:hF:B:o:")) != EOF
) {
696 value
= nicenumtoull(optarg
);
700 zo
->zo_vdevs
= value
;
703 zo
->zo_vdev_size
= MAX(SPA_MINDEVSIZE
, value
);
706 zo
->zo_ashift
= value
;
709 zo
->zo_mirrors
= value
;
712 zo
->zo_raidz
= MAX(1, value
);
715 zo
->zo_raidz_parity
= MIN(MAX(value
, 1), 3);
718 zo
->zo_datasets
= MAX(1, value
);
721 zo
->zo_threads
= MAX(1, value
);
724 zo
->zo_metaslab_gang_bang
= MAX(SPA_MINBLOCKSIZE
<< 1,
731 zo
->zo_killrate
= value
;
734 (void) strlcpy(zo
->zo_pool
, optarg
,
735 sizeof (zo
->zo_pool
));
738 path
= realpath(optarg
, NULL
);
740 (void) fprintf(stderr
, "error: %s: %s\n",
741 optarg
, strerror(errno
));
744 (void) strlcpy(zo
->zo_dir
, path
,
745 sizeof (zo
->zo_dir
));
762 zo
->zo_passtime
= MAX(1, value
);
765 zo
->zo_maxloops
= MAX(1, value
);
768 (void) strlcpy(altdir
, optarg
, sizeof (altdir
));
771 if (set_global_var(optarg
) != 0)
784 zo
->zo_raidz_parity
= MIN(zo
->zo_raidz_parity
, zo
->zo_raidz
- 1);
787 (zo
->zo_vdevs
> 0 ? zo
->zo_time
* NANOSEC
/ zo
->zo_vdevs
:
790 if (strlen(altdir
) > 0) {
798 cmd
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
799 realaltdir
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
801 VERIFY(NULL
!= realpath(getexecname(), cmd
));
802 if (0 != access(altdir
, F_OK
)) {
803 ztest_dump_core
= B_FALSE
;
804 fatal(B_TRUE
, "invalid alternate ztest path: %s",
807 VERIFY(NULL
!= realpath(altdir
, realaltdir
));
810 * 'cmd' should be of the form "<anything>/usr/bin/<isa>/ztest".
811 * We want to extract <isa> to determine if we should use
812 * 32 or 64 bit binaries.
814 bin
= strstr(cmd
, "/usr/bin/");
815 ztest
= strstr(bin
, "/ztest");
817 isalen
= ztest
- isa
;
818 (void) snprintf(zo
->zo_alt_ztest
, sizeof (zo
->zo_alt_ztest
),
819 "%s/usr/bin/%.*s/ztest", realaltdir
, isalen
, isa
);
820 (void) snprintf(zo
->zo_alt_libpath
, sizeof (zo
->zo_alt_libpath
),
821 "%s/usr/lib/%.*s", realaltdir
, isalen
, isa
);
823 if (0 != access(zo
->zo_alt_ztest
, X_OK
)) {
824 ztest_dump_core
= B_FALSE
;
825 fatal(B_TRUE
, "invalid alternate ztest: %s",
827 } else if (0 != access(zo
->zo_alt_libpath
, X_OK
)) {
828 ztest_dump_core
= B_FALSE
;
829 fatal(B_TRUE
, "invalid alternate lib directory %s",
833 umem_free(cmd
, MAXPATHLEN
);
834 umem_free(realaltdir
, MAXPATHLEN
);
839 ztest_kill(ztest_shared_t
*zs
)
841 zs
->zs_alloc
= metaslab_class_get_alloc(spa_normal_class(ztest_spa
));
842 zs
->zs_space
= metaslab_class_get_space(spa_normal_class(ztest_spa
));
845 * Before we kill off ztest, make sure that the config is updated.
846 * See comment above spa_config_sync().
848 mutex_enter(&spa_namespace_lock
);
849 spa_config_sync(ztest_spa
, B_FALSE
, B_FALSE
);
850 mutex_exit(&spa_namespace_lock
);
852 (void) kill(getpid(), SIGKILL
);
856 ztest_random(uint64_t range
)
860 ASSERT3S(ztest_fd_rand
, >=, 0);
865 if (read(ztest_fd_rand
, &r
, sizeof (r
)) != sizeof (r
))
866 fatal(1, "short read from /dev/urandom");
873 ztest_record_enospc(const char *s
)
875 ztest_shared
->zs_enospc_count
++;
879 ztest_get_ashift(void)
881 if (ztest_opts
.zo_ashift
== 0)
882 return (SPA_MINBLOCKSHIFT
+ ztest_random(5));
883 return (ztest_opts
.zo_ashift
);
887 make_vdev_file(char *path
, char *aux
, char *pool
, size_t size
, uint64_t ashift
)
893 pathbuf
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
896 ashift
= ztest_get_ashift();
902 vdev
= ztest_shared
->zs_vdev_aux
;
903 (void) snprintf(path
, MAXPATHLEN
,
904 ztest_aux_template
, ztest_opts
.zo_dir
,
905 pool
== NULL
? ztest_opts
.zo_pool
: pool
,
908 vdev
= ztest_shared
->zs_vdev_next_leaf
++;
909 (void) snprintf(path
, MAXPATHLEN
,
910 ztest_dev_template
, ztest_opts
.zo_dir
,
911 pool
== NULL
? ztest_opts
.zo_pool
: pool
, vdev
);
916 int fd
= open(path
, O_RDWR
| O_CREAT
| O_TRUNC
, 0666);
918 fatal(1, "can't open %s", path
);
919 if (ftruncate(fd
, size
) != 0)
920 fatal(1, "can't ftruncate %s", path
);
924 VERIFY(nvlist_alloc(&file
, NV_UNIQUE_NAME
, 0) == 0);
925 VERIFY(nvlist_add_string(file
, ZPOOL_CONFIG_TYPE
, VDEV_TYPE_FILE
) == 0);
926 VERIFY(nvlist_add_string(file
, ZPOOL_CONFIG_PATH
, path
) == 0);
927 VERIFY(nvlist_add_uint64(file
, ZPOOL_CONFIG_ASHIFT
, ashift
) == 0);
928 umem_free(pathbuf
, MAXPATHLEN
);
934 make_vdev_raidz(char *path
, char *aux
, char *pool
, size_t size
,
935 uint64_t ashift
, int r
)
937 nvlist_t
*raidz
, **child
;
941 return (make_vdev_file(path
, aux
, pool
, size
, ashift
));
942 child
= umem_alloc(r
* sizeof (nvlist_t
*), UMEM_NOFAIL
);
944 for (c
= 0; c
< r
; c
++)
945 child
[c
] = make_vdev_file(path
, aux
, pool
, size
, ashift
);
947 VERIFY(nvlist_alloc(&raidz
, NV_UNIQUE_NAME
, 0) == 0);
948 VERIFY(nvlist_add_string(raidz
, ZPOOL_CONFIG_TYPE
,
949 VDEV_TYPE_RAIDZ
) == 0);
950 VERIFY(nvlist_add_uint64(raidz
, ZPOOL_CONFIG_NPARITY
,
951 ztest_opts
.zo_raidz_parity
) == 0);
952 VERIFY(nvlist_add_nvlist_array(raidz
, ZPOOL_CONFIG_CHILDREN
,
955 for (c
= 0; c
< r
; c
++)
956 nvlist_free(child
[c
]);
958 umem_free(child
, r
* sizeof (nvlist_t
*));
964 make_vdev_mirror(char *path
, char *aux
, char *pool
, size_t size
,
965 uint64_t ashift
, int r
, int m
)
967 nvlist_t
*mirror
, **child
;
971 return (make_vdev_raidz(path
, aux
, pool
, size
, ashift
, r
));
973 child
= umem_alloc(m
* sizeof (nvlist_t
*), UMEM_NOFAIL
);
975 for (c
= 0; c
< m
; c
++)
976 child
[c
] = make_vdev_raidz(path
, aux
, pool
, size
, ashift
, r
);
978 VERIFY(nvlist_alloc(&mirror
, NV_UNIQUE_NAME
, 0) == 0);
979 VERIFY(nvlist_add_string(mirror
, ZPOOL_CONFIG_TYPE
,
980 VDEV_TYPE_MIRROR
) == 0);
981 VERIFY(nvlist_add_nvlist_array(mirror
, ZPOOL_CONFIG_CHILDREN
,
984 for (c
= 0; c
< m
; c
++)
985 nvlist_free(child
[c
]);
987 umem_free(child
, m
* sizeof (nvlist_t
*));
993 make_vdev_root(char *path
, char *aux
, char *pool
, size_t size
, uint64_t ashift
,
994 int log
, int r
, int m
, int t
)
996 nvlist_t
*root
, **child
;
1001 child
= umem_alloc(t
* sizeof (nvlist_t
*), UMEM_NOFAIL
);
1003 for (c
= 0; c
< t
; c
++) {
1004 child
[c
] = make_vdev_mirror(path
, aux
, pool
, size
, ashift
,
1006 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_IS_LOG
,
1010 VERIFY(nvlist_alloc(&root
, NV_UNIQUE_NAME
, 0) == 0);
1011 VERIFY(nvlist_add_string(root
, ZPOOL_CONFIG_TYPE
, VDEV_TYPE_ROOT
) == 0);
1012 VERIFY(nvlist_add_nvlist_array(root
, aux
? aux
: ZPOOL_CONFIG_CHILDREN
,
1015 for (c
= 0; c
< t
; c
++)
1016 nvlist_free(child
[c
]);
1018 umem_free(child
, t
* sizeof (nvlist_t
*));
1024 * Find a random spa version. Returns back a random spa version in the
1025 * range [initial_version, SPA_VERSION_FEATURES].
1028 ztest_random_spa_version(uint64_t initial_version
)
1030 uint64_t version
= initial_version
;
1032 if (version
<= SPA_VERSION_BEFORE_FEATURES
) {
1034 ztest_random(SPA_VERSION_BEFORE_FEATURES
- version
+ 1);
1037 if (version
> SPA_VERSION_BEFORE_FEATURES
)
1038 version
= SPA_VERSION_FEATURES
;
1040 ASSERT(SPA_VERSION_IS_SUPPORTED(version
));
1045 ztest_random_blocksize(void)
1048 * Choose a block size >= the ashift.
1049 * If the SPA supports new MAXBLOCKSIZE, test up to 1MB blocks.
1051 int maxbs
= SPA_OLD_MAXBLOCKSHIFT
;
1052 if (spa_maxblocksize(ztest_spa
) == SPA_MAXBLOCKSIZE
)
1054 uint64_t block_shift
=
1055 ztest_random(maxbs
- ztest_spa
->spa_max_ashift
+ 1);
1056 return (1 << (SPA_MINBLOCKSHIFT
+ block_shift
));
1060 ztest_random_dnodesize(void)
1063 int max_slots
= spa_maxdnodesize(ztest_spa
) >> DNODE_SHIFT
;
1065 if (max_slots
== DNODE_MIN_SLOTS
)
1066 return (DNODE_MIN_SIZE
);
1069 * Weight the random distribution more heavily toward smaller
1070 * dnode sizes since that is more likely to reflect real-world
1073 ASSERT3U(max_slots
, >, 4);
1074 switch (ztest_random(10)) {
1076 slots
= 5 + ztest_random(max_slots
- 4);
1079 slots
= 2 + ztest_random(3);
1086 return (slots
<< DNODE_SHIFT
);
1090 ztest_random_ibshift(void)
1092 return (DN_MIN_INDBLKSHIFT
+
1093 ztest_random(DN_MAX_INDBLKSHIFT
- DN_MIN_INDBLKSHIFT
+ 1));
1097 ztest_random_vdev_top(spa_t
*spa
, boolean_t log_ok
)
1100 vdev_t
*rvd
= spa
->spa_root_vdev
;
1103 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_READER
) != 0);
1106 top
= ztest_random(rvd
->vdev_children
);
1107 tvd
= rvd
->vdev_child
[top
];
1108 } while (tvd
->vdev_ishole
|| (tvd
->vdev_islog
&& !log_ok
) ||
1109 tvd
->vdev_mg
== NULL
|| tvd
->vdev_mg
->mg_class
== NULL
);
1115 ztest_random_dsl_prop(zfs_prop_t prop
)
1120 value
= zfs_prop_random_value(prop
, ztest_random(-1ULL));
1121 } while (prop
== ZFS_PROP_CHECKSUM
&& value
== ZIO_CHECKSUM_OFF
);
1127 ztest_dsl_prop_set_uint64(char *osname
, zfs_prop_t prop
, uint64_t value
,
1130 const char *propname
= zfs_prop_to_name(prop
);
1131 const char *valname
;
1136 error
= dsl_prop_set_int(osname
, propname
,
1137 (inherit
? ZPROP_SRC_NONE
: ZPROP_SRC_LOCAL
), value
);
1139 if (error
== ENOSPC
) {
1140 ztest_record_enospc(FTAG
);
1145 setpoint
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
1146 VERIFY0(dsl_prop_get_integer(osname
, propname
, &curval
, setpoint
));
1148 if (ztest_opts
.zo_verbose
>= 6) {
1151 err
= zfs_prop_index_to_string(prop
, curval
, &valname
);
1153 (void) printf("%s %s = %llu at '%s'\n", osname
,
1154 propname
, (unsigned long long)curval
, setpoint
);
1156 (void) printf("%s %s = %s at '%s'\n",
1157 osname
, propname
, valname
, setpoint
);
1159 umem_free(setpoint
, MAXPATHLEN
);
1165 ztest_spa_prop_set_uint64(zpool_prop_t prop
, uint64_t value
)
1167 spa_t
*spa
= ztest_spa
;
1168 nvlist_t
*props
= NULL
;
1171 VERIFY(nvlist_alloc(&props
, NV_UNIQUE_NAME
, 0) == 0);
1172 VERIFY(nvlist_add_uint64(props
, zpool_prop_to_name(prop
), value
) == 0);
1174 error
= spa_prop_set(spa
, props
);
1178 if (error
== ENOSPC
) {
1179 ztest_record_enospc(FTAG
);
1188 ztest_dmu_objset_own(const char *name
, dmu_objset_type_t type
,
1189 boolean_t readonly
, boolean_t decrypt
, void *tag
, objset_t
**osp
)
1193 err
= dmu_objset_own(name
, type
, readonly
, decrypt
, tag
, osp
);
1194 if (decrypt
&& err
== EACCES
) {
1195 char ddname
[ZFS_MAX_DATASET_NAME_LEN
];
1196 dsl_crypto_params_t
*dcp
;
1197 nvlist_t
*crypto_args
= fnvlist_alloc();
1200 /* spa_keystore_load_wkey() expects a dsl dir name */
1201 strcpy(ddname
, name
);
1202 cp
= strchr(ddname
, '@');
1206 fnvlist_add_uint8_array(crypto_args
, "wkeydata",
1207 (uint8_t *)ztest_wkeydata
, WRAPPING_KEY_LEN
);
1208 VERIFY0(dsl_crypto_params_create_nvlist(DCP_CMD_NONE
, NULL
,
1209 crypto_args
, &dcp
));
1210 err
= spa_keystore_load_wkey(ddname
, dcp
, B_FALSE
);
1211 dsl_crypto_params_free(dcp
, B_FALSE
);
1212 fnvlist_free(crypto_args
);
1217 err
= dmu_objset_own(name
, type
, readonly
, decrypt
, tag
, osp
);
1225 * Object and range lock mechanics
1228 list_node_t z_lnode
;
1229 refcount_t z_refcnt
;
1231 zfs_rlock_t z_range_lock
;
1236 ztest_znode_t
*z_ztznode
;
1239 static ztest_znode_t
*
1240 ztest_znode_init(uint64_t object
)
1242 ztest_znode_t
*zp
= umem_alloc(sizeof (*zp
), UMEM_NOFAIL
);
1244 list_link_init(&zp
->z_lnode
);
1245 refcount_create(&zp
->z_refcnt
);
1246 zp
->z_object
= object
;
1247 zfs_rlock_init(&zp
->z_range_lock
);
1253 ztest_znode_fini(ztest_znode_t
*zp
)
1255 ASSERT(refcount_is_zero(&zp
->z_refcnt
));
1256 zfs_rlock_destroy(&zp
->z_range_lock
);
1258 refcount_destroy(&zp
->z_refcnt
);
1259 list_link_init(&zp
->z_lnode
);
1260 umem_free(zp
, sizeof (*zp
));
1264 ztest_zll_init(zll_t
*zll
)
1266 mutex_init(&zll
->z_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
1267 list_create(&zll
->z_list
, sizeof (ztest_znode_t
),
1268 offsetof(ztest_znode_t
, z_lnode
));
1272 ztest_zll_destroy(zll_t
*zll
)
1274 list_destroy(&zll
->z_list
);
1275 mutex_destroy(&zll
->z_lock
);
1278 #define RL_TAG "range_lock"
1279 static ztest_znode_t
*
1280 ztest_znode_get(ztest_ds_t
*zd
, uint64_t object
)
1282 zll_t
*zll
= &zd
->zd_range_lock
[object
& (ZTEST_OBJECT_LOCKS
- 1)];
1283 ztest_znode_t
*zp
= NULL
;
1284 mutex_enter(&zll
->z_lock
);
1285 for (zp
= list_head(&zll
->z_list
); (zp
);
1286 zp
= list_next(&zll
->z_list
, zp
)) {
1287 if (zp
->z_object
== object
) {
1288 refcount_add(&zp
->z_refcnt
, RL_TAG
);
1293 zp
= ztest_znode_init(object
);
1294 refcount_add(&zp
->z_refcnt
, RL_TAG
);
1295 list_insert_head(&zll
->z_list
, zp
);
1297 mutex_exit(&zll
->z_lock
);
1302 ztest_znode_put(ztest_ds_t
*zd
, ztest_znode_t
*zp
)
1305 ASSERT3U(zp
->z_object
, !=, 0);
1306 zll
= &zd
->zd_range_lock
[zp
->z_object
& (ZTEST_OBJECT_LOCKS
- 1)];
1307 mutex_enter(&zll
->z_lock
);
1308 refcount_remove(&zp
->z_refcnt
, RL_TAG
);
1309 if (refcount_is_zero(&zp
->z_refcnt
)) {
1310 list_remove(&zll
->z_list
, zp
);
1311 ztest_znode_fini(zp
);
1313 mutex_exit(&zll
->z_lock
);
1318 ztest_rll_init(rll_t
*rll
)
1320 rll
->rll_writer
= NULL
;
1321 rll
->rll_readers
= 0;
1322 mutex_init(&rll
->rll_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
1323 cv_init(&rll
->rll_cv
, NULL
, CV_DEFAULT
, NULL
);
1327 ztest_rll_destroy(rll_t
*rll
)
1329 ASSERT(rll
->rll_writer
== NULL
);
1330 ASSERT(rll
->rll_readers
== 0);
1331 mutex_destroy(&rll
->rll_lock
);
1332 cv_destroy(&rll
->rll_cv
);
1336 ztest_rll_lock(rll_t
*rll
, rl_type_t type
)
1338 mutex_enter(&rll
->rll_lock
);
1340 if (type
== RL_READER
) {
1341 while (rll
->rll_writer
!= NULL
)
1342 (void) cv_wait(&rll
->rll_cv
, &rll
->rll_lock
);
1345 while (rll
->rll_writer
!= NULL
|| rll
->rll_readers
)
1346 (void) cv_wait(&rll
->rll_cv
, &rll
->rll_lock
);
1347 rll
->rll_writer
= curthread
;
1350 mutex_exit(&rll
->rll_lock
);
1354 ztest_rll_unlock(rll_t
*rll
)
1356 mutex_enter(&rll
->rll_lock
);
1358 if (rll
->rll_writer
) {
1359 ASSERT(rll
->rll_readers
== 0);
1360 rll
->rll_writer
= NULL
;
1362 ASSERT(rll
->rll_readers
!= 0);
1363 ASSERT(rll
->rll_writer
== NULL
);
1367 if (rll
->rll_writer
== NULL
&& rll
->rll_readers
== 0)
1368 cv_broadcast(&rll
->rll_cv
);
1370 mutex_exit(&rll
->rll_lock
);
1374 ztest_object_lock(ztest_ds_t
*zd
, uint64_t object
, rl_type_t type
)
1376 rll_t
*rll
= &zd
->zd_object_lock
[object
& (ZTEST_OBJECT_LOCKS
- 1)];
1378 ztest_rll_lock(rll
, type
);
1382 ztest_object_unlock(ztest_ds_t
*zd
, uint64_t object
)
1384 rll_t
*rll
= &zd
->zd_object_lock
[object
& (ZTEST_OBJECT_LOCKS
- 1)];
1386 ztest_rll_unlock(rll
);
1389 static ztest_zrl_t
*
1390 ztest_zrl_init(rl_t
*rl
, ztest_znode_t
*zp
)
1392 ztest_zrl_t
*zrl
= umem_alloc(sizeof (*zrl
), UMEM_NOFAIL
);
1394 zrl
->z_ztznode
= zp
;
1399 ztest_zrl_fini(ztest_zrl_t
*zrl
)
1401 umem_free(zrl
, sizeof (*zrl
));
1404 static ztest_zrl_t
*
1405 ztest_range_lock(ztest_ds_t
*zd
, uint64_t object
, uint64_t offset
,
1406 uint64_t size
, rl_type_t type
)
1408 ztest_znode_t
*zp
= ztest_znode_get(zd
, object
);
1409 rl_t
*rl
= zfs_range_lock(&zp
->z_range_lock
, offset
,
1411 return (ztest_zrl_init(rl
, zp
));
1415 ztest_range_unlock(ztest_ds_t
*zd
, ztest_zrl_t
*zrl
)
1417 zfs_range_unlock(zrl
->z_rl
);
1418 ztest_znode_put(zd
, zrl
->z_ztznode
);
1419 ztest_zrl_fini(zrl
);
1423 ztest_zd_init(ztest_ds_t
*zd
, ztest_shared_ds_t
*szd
, objset_t
*os
)
1426 zd
->zd_zilog
= dmu_objset_zil(os
);
1427 zd
->zd_shared
= szd
;
1428 dmu_objset_name(os
, zd
->zd_name
);
1431 if (zd
->zd_shared
!= NULL
)
1432 zd
->zd_shared
->zd_seq
= 0;
1434 VERIFY(rwlock_init(&zd
->zd_zilog_lock
, USYNC_THREAD
, NULL
) == 0);
1435 mutex_init(&zd
->zd_dirobj_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
1437 for (l
= 0; l
< ZTEST_OBJECT_LOCKS
; l
++)
1438 ztest_rll_init(&zd
->zd_object_lock
[l
]);
1440 for (l
= 0; l
< ZTEST_RANGE_LOCKS
; l
++)
1441 ztest_zll_init(&zd
->zd_range_lock
[l
]);
1445 ztest_zd_fini(ztest_ds_t
*zd
)
1449 mutex_destroy(&zd
->zd_dirobj_lock
);
1450 (void) rwlock_destroy(&zd
->zd_zilog_lock
);
1452 for (l
= 0; l
< ZTEST_OBJECT_LOCKS
; l
++)
1453 ztest_rll_destroy(&zd
->zd_object_lock
[l
]);
1455 for (l
= 0; l
< ZTEST_RANGE_LOCKS
; l
++)
1456 ztest_zll_destroy(&zd
->zd_range_lock
[l
]);
1459 #define TXG_MIGHTWAIT (ztest_random(10) == 0 ? TXG_NOWAIT : TXG_WAIT)
1462 ztest_tx_assign(dmu_tx_t
*tx
, uint64_t txg_how
, const char *tag
)
1468 * Attempt to assign tx to some transaction group.
1470 error
= dmu_tx_assign(tx
, txg_how
);
1472 if (error
== ERESTART
) {
1473 ASSERT(txg_how
== TXG_NOWAIT
);
1476 ASSERT3U(error
, ==, ENOSPC
);
1477 ztest_record_enospc(tag
);
1482 txg
= dmu_tx_get_txg(tx
);
1488 ztest_pattern_set(void *buf
, uint64_t size
, uint64_t value
)
1491 uint64_t *ip_end
= (uint64_t *)((uintptr_t)buf
+ (uintptr_t)size
);
1499 ztest_pattern_match(void *buf
, uint64_t size
, uint64_t value
)
1502 uint64_t *ip_end
= (uint64_t *)((uintptr_t)buf
+ (uintptr_t)size
);
1506 diff
|= (value
- *ip
++);
1513 ztest_bt_generate(ztest_block_tag_t
*bt
, objset_t
*os
, uint64_t object
,
1514 uint64_t dnodesize
, uint64_t offset
, uint64_t gen
, uint64_t txg
,
1517 bt
->bt_magic
= BT_MAGIC
;
1518 bt
->bt_objset
= dmu_objset_id(os
);
1519 bt
->bt_object
= object
;
1520 bt
->bt_dnodesize
= dnodesize
;
1521 bt
->bt_offset
= offset
;
1524 bt
->bt_crtxg
= crtxg
;
1528 ztest_bt_verify(ztest_block_tag_t
*bt
, objset_t
*os
, uint64_t object
,
1529 uint64_t dnodesize
, uint64_t offset
, uint64_t gen
, uint64_t txg
,
1532 ASSERT3U(bt
->bt_magic
, ==, BT_MAGIC
);
1533 ASSERT3U(bt
->bt_objset
, ==, dmu_objset_id(os
));
1534 ASSERT3U(bt
->bt_object
, ==, object
);
1535 ASSERT3U(bt
->bt_dnodesize
, ==, dnodesize
);
1536 ASSERT3U(bt
->bt_offset
, ==, offset
);
1537 ASSERT3U(bt
->bt_gen
, <=, gen
);
1538 ASSERT3U(bt
->bt_txg
, <=, txg
);
1539 ASSERT3U(bt
->bt_crtxg
, ==, crtxg
);
1542 static ztest_block_tag_t
*
1543 ztest_bt_bonus(dmu_buf_t
*db
)
1545 dmu_object_info_t doi
;
1546 ztest_block_tag_t
*bt
;
1548 dmu_object_info_from_db(db
, &doi
);
1549 ASSERT3U(doi
.doi_bonus_size
, <=, db
->db_size
);
1550 ASSERT3U(doi
.doi_bonus_size
, >=, sizeof (*bt
));
1551 bt
= (void *)((char *)db
->db_data
+ doi
.doi_bonus_size
- sizeof (*bt
));
1557 * Generate a token to fill up unused bonus buffer space. Try to make
1558 * it unique to the object, generation, and offset to verify that data
1559 * is not getting overwritten by data from other dnodes.
1561 #define ZTEST_BONUS_FILL_TOKEN(obj, ds, gen, offset) \
1562 (((ds) << 48) | ((gen) << 32) | ((obj) << 8) | (offset))
1565 * Fill up the unused bonus buffer region before the block tag with a
1566 * verifiable pattern. Filling the whole bonus area with non-zero data
1567 * helps ensure that all dnode traversal code properly skips the
1568 * interior regions of large dnodes.
1571 ztest_fill_unused_bonus(dmu_buf_t
*db
, void *end
, uint64_t obj
,
1572 objset_t
*os
, uint64_t gen
)
1576 ASSERT(IS_P2ALIGNED((char *)end
- (char *)db
->db_data
, 8));
1578 for (bonusp
= db
->db_data
; bonusp
< (uint64_t *)end
; bonusp
++) {
1579 uint64_t token
= ZTEST_BONUS_FILL_TOKEN(obj
, dmu_objset_id(os
),
1580 gen
, bonusp
- (uint64_t *)db
->db_data
);
1586 * Verify that the unused area of a bonus buffer is filled with the
1590 ztest_verify_unused_bonus(dmu_buf_t
*db
, void *end
, uint64_t obj
,
1591 objset_t
*os
, uint64_t gen
)
1595 for (bonusp
= db
->db_data
; bonusp
< (uint64_t *)end
; bonusp
++) {
1596 uint64_t token
= ZTEST_BONUS_FILL_TOKEN(obj
, dmu_objset_id(os
),
1597 gen
, bonusp
- (uint64_t *)db
->db_data
);
1598 VERIFY3U(*bonusp
, ==, token
);
1606 #define lrz_type lr_mode
1607 #define lrz_blocksize lr_uid
1608 #define lrz_ibshift lr_gid
1609 #define lrz_bonustype lr_rdev
1610 #define lrz_dnodesize lr_crtime[1]
1613 ztest_log_create(ztest_ds_t
*zd
, dmu_tx_t
*tx
, lr_create_t
*lr
)
1615 char *name
= (void *)(lr
+ 1); /* name follows lr */
1616 size_t namesize
= strlen(name
) + 1;
1619 if (zil_replaying(zd
->zd_zilog
, tx
))
1622 itx
= zil_itx_create(TX_CREATE
, sizeof (*lr
) + namesize
);
1623 bcopy(&lr
->lr_common
+ 1, &itx
->itx_lr
+ 1,
1624 sizeof (*lr
) + namesize
- sizeof (lr_t
));
1626 zil_itx_assign(zd
->zd_zilog
, itx
, tx
);
1630 ztest_log_remove(ztest_ds_t
*zd
, dmu_tx_t
*tx
, lr_remove_t
*lr
, uint64_t object
)
1632 char *name
= (void *)(lr
+ 1); /* name follows lr */
1633 size_t namesize
= strlen(name
) + 1;
1636 if (zil_replaying(zd
->zd_zilog
, tx
))
1639 itx
= zil_itx_create(TX_REMOVE
, sizeof (*lr
) + namesize
);
1640 bcopy(&lr
->lr_common
+ 1, &itx
->itx_lr
+ 1,
1641 sizeof (*lr
) + namesize
- sizeof (lr_t
));
1643 itx
->itx_oid
= object
;
1644 zil_itx_assign(zd
->zd_zilog
, itx
, tx
);
1648 ztest_log_write(ztest_ds_t
*zd
, dmu_tx_t
*tx
, lr_write_t
*lr
)
1651 itx_wr_state_t write_state
= ztest_random(WR_NUM_STATES
);
1653 if (zil_replaying(zd
->zd_zilog
, tx
))
1656 if (lr
->lr_length
> ZIL_MAX_LOG_DATA
)
1657 write_state
= WR_INDIRECT
;
1659 itx
= zil_itx_create(TX_WRITE
,
1660 sizeof (*lr
) + (write_state
== WR_COPIED
? lr
->lr_length
: 0));
1662 if (write_state
== WR_COPIED
&&
1663 dmu_read(zd
->zd_os
, lr
->lr_foid
, lr
->lr_offset
, lr
->lr_length
,
1664 ((lr_write_t
*)&itx
->itx_lr
) + 1, DMU_READ_NO_PREFETCH
) != 0) {
1665 zil_itx_destroy(itx
);
1666 itx
= zil_itx_create(TX_WRITE
, sizeof (*lr
));
1667 write_state
= WR_NEED_COPY
;
1669 itx
->itx_private
= zd
;
1670 itx
->itx_wr_state
= write_state
;
1671 itx
->itx_sync
= (ztest_random(8) == 0);
1673 bcopy(&lr
->lr_common
+ 1, &itx
->itx_lr
+ 1,
1674 sizeof (*lr
) - sizeof (lr_t
));
1676 zil_itx_assign(zd
->zd_zilog
, itx
, tx
);
1680 ztest_log_truncate(ztest_ds_t
*zd
, dmu_tx_t
*tx
, lr_truncate_t
*lr
)
1684 if (zil_replaying(zd
->zd_zilog
, tx
))
1687 itx
= zil_itx_create(TX_TRUNCATE
, sizeof (*lr
));
1688 bcopy(&lr
->lr_common
+ 1, &itx
->itx_lr
+ 1,
1689 sizeof (*lr
) - sizeof (lr_t
));
1691 itx
->itx_sync
= B_FALSE
;
1692 zil_itx_assign(zd
->zd_zilog
, itx
, tx
);
1696 ztest_log_setattr(ztest_ds_t
*zd
, dmu_tx_t
*tx
, lr_setattr_t
*lr
)
1700 if (zil_replaying(zd
->zd_zilog
, tx
))
1703 itx
= zil_itx_create(TX_SETATTR
, sizeof (*lr
));
1704 bcopy(&lr
->lr_common
+ 1, &itx
->itx_lr
+ 1,
1705 sizeof (*lr
) - sizeof (lr_t
));
1707 itx
->itx_sync
= B_FALSE
;
1708 zil_itx_assign(zd
->zd_zilog
, itx
, tx
);
1715 ztest_replay_create(void *arg1
, void *arg2
, boolean_t byteswap
)
1717 ztest_ds_t
*zd
= arg1
;
1718 lr_create_t
*lr
= arg2
;
1719 char *name
= (void *)(lr
+ 1); /* name follows lr */
1720 objset_t
*os
= zd
->zd_os
;
1721 ztest_block_tag_t
*bbt
;
1729 byteswap_uint64_array(lr
, sizeof (*lr
));
1731 ASSERT(lr
->lr_doid
== ZTEST_DIROBJ
);
1732 ASSERT(name
[0] != '\0');
1734 tx
= dmu_tx_create(os
);
1736 dmu_tx_hold_zap(tx
, lr
->lr_doid
, B_TRUE
, name
);
1738 if (lr
->lrz_type
== DMU_OT_ZAP_OTHER
) {
1739 dmu_tx_hold_zap(tx
, DMU_NEW_OBJECT
, B_TRUE
, NULL
);
1741 dmu_tx_hold_bonus(tx
, DMU_NEW_OBJECT
);
1744 txg
= ztest_tx_assign(tx
, TXG_WAIT
, FTAG
);
1748 ASSERT(dmu_objset_zil(os
)->zl_replay
== !!lr
->lr_foid
);
1749 bonuslen
= DN_BONUS_SIZE(lr
->lrz_dnodesize
);
1751 if (lr
->lrz_type
== DMU_OT_ZAP_OTHER
) {
1752 if (lr
->lr_foid
== 0) {
1753 lr
->lr_foid
= zap_create_dnsize(os
,
1754 lr
->lrz_type
, lr
->lrz_bonustype
,
1755 bonuslen
, lr
->lrz_dnodesize
, tx
);
1757 error
= zap_create_claim_dnsize(os
, lr
->lr_foid
,
1758 lr
->lrz_type
, lr
->lrz_bonustype
,
1759 bonuslen
, lr
->lrz_dnodesize
, tx
);
1762 if (lr
->lr_foid
== 0) {
1763 lr
->lr_foid
= dmu_object_alloc_dnsize(os
,
1764 lr
->lrz_type
, 0, lr
->lrz_bonustype
,
1765 bonuslen
, lr
->lrz_dnodesize
, tx
);
1767 error
= dmu_object_claim_dnsize(os
, lr
->lr_foid
,
1768 lr
->lrz_type
, 0, lr
->lrz_bonustype
,
1769 bonuslen
, lr
->lrz_dnodesize
, tx
);
1774 ASSERT3U(error
, ==, EEXIST
);
1775 ASSERT(zd
->zd_zilog
->zl_replay
);
1780 ASSERT(lr
->lr_foid
!= 0);
1782 if (lr
->lrz_type
!= DMU_OT_ZAP_OTHER
)
1783 VERIFY3U(0, ==, dmu_object_set_blocksize(os
, lr
->lr_foid
,
1784 lr
->lrz_blocksize
, lr
->lrz_ibshift
, tx
));
1786 VERIFY3U(0, ==, dmu_bonus_hold(os
, lr
->lr_foid
, FTAG
, &db
));
1787 bbt
= ztest_bt_bonus(db
);
1788 dmu_buf_will_dirty(db
, tx
);
1789 ztest_bt_generate(bbt
, os
, lr
->lr_foid
, lr
->lrz_dnodesize
, -1ULL,
1790 lr
->lr_gen
, txg
, txg
);
1791 ztest_fill_unused_bonus(db
, bbt
, lr
->lr_foid
, os
, lr
->lr_gen
);
1792 dmu_buf_rele(db
, FTAG
);
1794 VERIFY3U(0, ==, zap_add(os
, lr
->lr_doid
, name
, sizeof (uint64_t), 1,
1797 (void) ztest_log_create(zd
, tx
, lr
);
1805 ztest_replay_remove(void *arg1
, void *arg2
, boolean_t byteswap
)
1807 ztest_ds_t
*zd
= arg1
;
1808 lr_remove_t
*lr
= arg2
;
1809 char *name
= (void *)(lr
+ 1); /* name follows lr */
1810 objset_t
*os
= zd
->zd_os
;
1811 dmu_object_info_t doi
;
1813 uint64_t object
, txg
;
1816 byteswap_uint64_array(lr
, sizeof (*lr
));
1818 ASSERT(lr
->lr_doid
== ZTEST_DIROBJ
);
1819 ASSERT(name
[0] != '\0');
1822 zap_lookup(os
, lr
->lr_doid
, name
, sizeof (object
), 1, &object
));
1823 ASSERT(object
!= 0);
1825 ztest_object_lock(zd
, object
, RL_WRITER
);
1827 VERIFY3U(0, ==, dmu_object_info(os
, object
, &doi
));
1829 tx
= dmu_tx_create(os
);
1831 dmu_tx_hold_zap(tx
, lr
->lr_doid
, B_FALSE
, name
);
1832 dmu_tx_hold_free(tx
, object
, 0, DMU_OBJECT_END
);
1834 txg
= ztest_tx_assign(tx
, TXG_WAIT
, FTAG
);
1836 ztest_object_unlock(zd
, object
);
1840 if (doi
.doi_type
== DMU_OT_ZAP_OTHER
) {
1841 VERIFY3U(0, ==, zap_destroy(os
, object
, tx
));
1843 VERIFY3U(0, ==, dmu_object_free(os
, object
, tx
));
1846 VERIFY3U(0, ==, zap_remove(os
, lr
->lr_doid
, name
, tx
));
1848 (void) ztest_log_remove(zd
, tx
, lr
, object
);
1852 ztest_object_unlock(zd
, object
);
1858 ztest_replay_write(void *arg1
, void *arg2
, boolean_t byteswap
)
1860 ztest_ds_t
*zd
= arg1
;
1861 lr_write_t
*lr
= arg2
;
1862 objset_t
*os
= zd
->zd_os
;
1863 void *data
= lr
+ 1; /* data follows lr */
1864 uint64_t offset
, length
;
1865 ztest_block_tag_t
*bt
= data
;
1866 ztest_block_tag_t
*bbt
;
1867 uint64_t gen
, txg
, lrtxg
, crtxg
;
1868 dmu_object_info_t doi
;
1871 arc_buf_t
*abuf
= NULL
;
1875 byteswap_uint64_array(lr
, sizeof (*lr
));
1877 offset
= lr
->lr_offset
;
1878 length
= lr
->lr_length
;
1880 /* If it's a dmu_sync() block, write the whole block */
1881 if (lr
->lr_common
.lrc_reclen
== sizeof (lr_write_t
)) {
1882 uint64_t blocksize
= BP_GET_LSIZE(&lr
->lr_blkptr
);
1883 if (length
< blocksize
) {
1884 offset
-= offset
% blocksize
;
1889 if (bt
->bt_magic
== BSWAP_64(BT_MAGIC
))
1890 byteswap_uint64_array(bt
, sizeof (*bt
));
1892 if (bt
->bt_magic
!= BT_MAGIC
)
1895 ztest_object_lock(zd
, lr
->lr_foid
, RL_READER
);
1896 rl
= ztest_range_lock(zd
, lr
->lr_foid
, offset
, length
, RL_WRITER
);
1898 VERIFY3U(0, ==, dmu_bonus_hold(os
, lr
->lr_foid
, FTAG
, &db
));
1900 dmu_object_info_from_db(db
, &doi
);
1902 bbt
= ztest_bt_bonus(db
);
1903 ASSERT3U(bbt
->bt_magic
, ==, BT_MAGIC
);
1905 crtxg
= bbt
->bt_crtxg
;
1906 lrtxg
= lr
->lr_common
.lrc_txg
;
1908 tx
= dmu_tx_create(os
);
1910 dmu_tx_hold_write(tx
, lr
->lr_foid
, offset
, length
);
1912 if (ztest_random(8) == 0 && length
== doi
.doi_data_block_size
&&
1913 P2PHASE(offset
, length
) == 0)
1914 abuf
= dmu_request_arcbuf(db
, length
);
1916 txg
= ztest_tx_assign(tx
, TXG_WAIT
, FTAG
);
1919 dmu_return_arcbuf(abuf
);
1920 dmu_buf_rele(db
, FTAG
);
1921 ztest_range_unlock(zd
, rl
);
1922 ztest_object_unlock(zd
, lr
->lr_foid
);
1928 * Usually, verify the old data before writing new data --
1929 * but not always, because we also want to verify correct
1930 * behavior when the data was not recently read into cache.
1932 ASSERT(offset
% doi
.doi_data_block_size
== 0);
1933 if (ztest_random(4) != 0) {
1934 int prefetch
= ztest_random(2) ?
1935 DMU_READ_PREFETCH
: DMU_READ_NO_PREFETCH
;
1936 ztest_block_tag_t rbt
;
1938 VERIFY(dmu_read(os
, lr
->lr_foid
, offset
,
1939 sizeof (rbt
), &rbt
, prefetch
) == 0);
1940 if (rbt
.bt_magic
== BT_MAGIC
) {
1941 ztest_bt_verify(&rbt
, os
, lr
->lr_foid
, 0,
1942 offset
, gen
, txg
, crtxg
);
1947 * Writes can appear to be newer than the bonus buffer because
1948 * the ztest_get_data() callback does a dmu_read() of the
1949 * open-context data, which may be different than the data
1950 * as it was when the write was generated.
1952 if (zd
->zd_zilog
->zl_replay
) {
1953 ztest_bt_verify(bt
, os
, lr
->lr_foid
, 0, offset
,
1954 MAX(gen
, bt
->bt_gen
), MAX(txg
, lrtxg
),
1959 * Set the bt's gen/txg to the bonus buffer's gen/txg
1960 * so that all of the usual ASSERTs will work.
1962 ztest_bt_generate(bt
, os
, lr
->lr_foid
, 0, offset
, gen
, txg
,
1967 dmu_write(os
, lr
->lr_foid
, offset
, length
, data
, tx
);
1969 bcopy(data
, abuf
->b_data
, length
);
1970 dmu_assign_arcbuf_by_dbuf(db
, offset
, abuf
, tx
);
1973 (void) ztest_log_write(zd
, tx
, lr
);
1975 dmu_buf_rele(db
, FTAG
);
1979 ztest_range_unlock(zd
, rl
);
1980 ztest_object_unlock(zd
, lr
->lr_foid
);
1986 ztest_replay_truncate(void *arg1
, void *arg2
, boolean_t byteswap
)
1988 ztest_ds_t
*zd
= arg1
;
1989 lr_truncate_t
*lr
= arg2
;
1990 objset_t
*os
= zd
->zd_os
;
1996 byteswap_uint64_array(lr
, sizeof (*lr
));
1998 ztest_object_lock(zd
, lr
->lr_foid
, RL_READER
);
1999 rl
= ztest_range_lock(zd
, lr
->lr_foid
, lr
->lr_offset
, lr
->lr_length
,
2002 tx
= dmu_tx_create(os
);
2004 dmu_tx_hold_free(tx
, lr
->lr_foid
, lr
->lr_offset
, lr
->lr_length
);
2006 txg
= ztest_tx_assign(tx
, TXG_WAIT
, FTAG
);
2008 ztest_range_unlock(zd
, rl
);
2009 ztest_object_unlock(zd
, lr
->lr_foid
);
2013 VERIFY(dmu_free_range(os
, lr
->lr_foid
, lr
->lr_offset
,
2014 lr
->lr_length
, tx
) == 0);
2016 (void) ztest_log_truncate(zd
, tx
, lr
);
2020 ztest_range_unlock(zd
, rl
);
2021 ztest_object_unlock(zd
, lr
->lr_foid
);
2027 ztest_replay_setattr(void *arg1
, void *arg2
, boolean_t byteswap
)
2029 ztest_ds_t
*zd
= arg1
;
2030 lr_setattr_t
*lr
= arg2
;
2031 objset_t
*os
= zd
->zd_os
;
2034 ztest_block_tag_t
*bbt
;
2035 uint64_t txg
, lrtxg
, crtxg
, dnodesize
;
2038 byteswap_uint64_array(lr
, sizeof (*lr
));
2040 ztest_object_lock(zd
, lr
->lr_foid
, RL_WRITER
);
2042 VERIFY3U(0, ==, dmu_bonus_hold(os
, lr
->lr_foid
, FTAG
, &db
));
2044 tx
= dmu_tx_create(os
);
2045 dmu_tx_hold_bonus(tx
, lr
->lr_foid
);
2047 txg
= ztest_tx_assign(tx
, TXG_WAIT
, FTAG
);
2049 dmu_buf_rele(db
, FTAG
);
2050 ztest_object_unlock(zd
, lr
->lr_foid
);
2054 bbt
= ztest_bt_bonus(db
);
2055 ASSERT3U(bbt
->bt_magic
, ==, BT_MAGIC
);
2056 crtxg
= bbt
->bt_crtxg
;
2057 lrtxg
= lr
->lr_common
.lrc_txg
;
2058 dnodesize
= bbt
->bt_dnodesize
;
2060 if (zd
->zd_zilog
->zl_replay
) {
2061 ASSERT(lr
->lr_size
!= 0);
2062 ASSERT(lr
->lr_mode
!= 0);
2066 * Randomly change the size and increment the generation.
2068 lr
->lr_size
= (ztest_random(db
->db_size
/ sizeof (*bbt
)) + 1) *
2070 lr
->lr_mode
= bbt
->bt_gen
+ 1;
2075 * Verify that the current bonus buffer is not newer than our txg.
2077 ztest_bt_verify(bbt
, os
, lr
->lr_foid
, dnodesize
, -1ULL, lr
->lr_mode
,
2078 MAX(txg
, lrtxg
), crtxg
);
2080 dmu_buf_will_dirty(db
, tx
);
2082 ASSERT3U(lr
->lr_size
, >=, sizeof (*bbt
));
2083 ASSERT3U(lr
->lr_size
, <=, db
->db_size
);
2084 VERIFY0(dmu_set_bonus(db
, lr
->lr_size
, tx
));
2085 bbt
= ztest_bt_bonus(db
);
2087 ztest_bt_generate(bbt
, os
, lr
->lr_foid
, dnodesize
, -1ULL, lr
->lr_mode
,
2089 ztest_fill_unused_bonus(db
, bbt
, lr
->lr_foid
, os
, bbt
->bt_gen
);
2090 dmu_buf_rele(db
, FTAG
);
2092 (void) ztest_log_setattr(zd
, tx
, lr
);
2096 ztest_object_unlock(zd
, lr
->lr_foid
);
2101 zil_replay_func_t
*ztest_replay_vector
[TX_MAX_TYPE
] = {
2102 NULL
, /* 0 no such transaction type */
2103 ztest_replay_create
, /* TX_CREATE */
2104 NULL
, /* TX_MKDIR */
2105 NULL
, /* TX_MKXATTR */
2106 NULL
, /* TX_SYMLINK */
2107 ztest_replay_remove
, /* TX_REMOVE */
2108 NULL
, /* TX_RMDIR */
2110 NULL
, /* TX_RENAME */
2111 ztest_replay_write
, /* TX_WRITE */
2112 ztest_replay_truncate
, /* TX_TRUNCATE */
2113 ztest_replay_setattr
, /* TX_SETATTR */
2115 NULL
, /* TX_CREATE_ACL */
2116 NULL
, /* TX_CREATE_ATTR */
2117 NULL
, /* TX_CREATE_ACL_ATTR */
2118 NULL
, /* TX_MKDIR_ACL */
2119 NULL
, /* TX_MKDIR_ATTR */
2120 NULL
, /* TX_MKDIR_ACL_ATTR */
2121 NULL
, /* TX_WRITE2 */
2125 * ZIL get_data callbacks
2127 typedef struct ztest_zgd_private
{
2131 } ztest_zgd_private_t
;
2134 ztest_get_done(zgd_t
*zgd
, int error
)
2136 ztest_zgd_private_t
*zzp
= zgd
->zgd_private
;
2137 ztest_ds_t
*zd
= zzp
->z_zd
;
2138 uint64_t object
= zzp
->z_object
;
2141 dmu_buf_rele(zgd
->zgd_db
, zgd
);
2143 ztest_range_unlock(zd
, zzp
->z_rl
);
2144 ztest_object_unlock(zd
, object
);
2146 if (error
== 0 && zgd
->zgd_bp
)
2147 zil_add_block(zgd
->zgd_zilog
, zgd
->zgd_bp
);
2149 umem_free(zgd
, sizeof (*zgd
));
2150 umem_free(zzp
, sizeof (*zzp
));
2154 ztest_get_data(void *arg
, lr_write_t
*lr
, char *buf
, zio_t
*zio
)
2156 ztest_ds_t
*zd
= arg
;
2157 objset_t
*os
= zd
->zd_os
;
2158 uint64_t object
= lr
->lr_foid
;
2159 uint64_t offset
= lr
->lr_offset
;
2160 uint64_t size
= lr
->lr_length
;
2161 uint64_t txg
= lr
->lr_common
.lrc_txg
;
2163 dmu_object_info_t doi
;
2167 ztest_zgd_private_t
*zgd_private
;
2169 ztest_object_lock(zd
, object
, RL_READER
);
2170 error
= dmu_bonus_hold(os
, object
, FTAG
, &db
);
2172 ztest_object_unlock(zd
, object
);
2176 crtxg
= ztest_bt_bonus(db
)->bt_crtxg
;
2178 if (crtxg
== 0 || crtxg
> txg
) {
2179 dmu_buf_rele(db
, FTAG
);
2180 ztest_object_unlock(zd
, object
);
2184 dmu_object_info_from_db(db
, &doi
);
2185 dmu_buf_rele(db
, FTAG
);
2188 zgd
= umem_zalloc(sizeof (*zgd
), UMEM_NOFAIL
);
2189 zgd
->zgd_zilog
= zd
->zd_zilog
;
2190 zgd_private
= umem_zalloc(sizeof (ztest_zgd_private_t
), UMEM_NOFAIL
);
2191 zgd_private
->z_zd
= zd
;
2192 zgd_private
->z_object
= object
;
2193 zgd
->zgd_private
= zgd_private
;
2195 if (buf
!= NULL
) { /* immediate write */
2196 zgd_private
->z_rl
= ztest_range_lock(zd
, object
, offset
, size
,
2198 zgd
->zgd_rl
= zgd_private
->z_rl
->z_rl
;
2200 error
= dmu_read(os
, object
, offset
, size
, buf
,
2201 DMU_READ_NO_PREFETCH
);
2204 size
= doi
.doi_data_block_size
;
2206 offset
= P2ALIGN(offset
, size
);
2208 ASSERT(offset
< size
);
2212 zgd_private
->z_rl
= ztest_range_lock(zd
, object
, offset
, size
,
2214 zgd
->zgd_rl
= zgd_private
->z_rl
->z_rl
;
2216 error
= dmu_buf_hold(os
, object
, offset
, zgd
, &db
,
2217 DMU_READ_NO_PREFETCH
);
2220 blkptr_t
*bp
= &lr
->lr_blkptr
;
2225 ASSERT(db
->db_offset
== offset
);
2226 ASSERT(db
->db_size
== size
);
2228 error
= dmu_sync(zio
, lr
->lr_common
.lrc_txg
,
2229 ztest_get_done
, zgd
);
2236 ztest_get_done(zgd
, error
);
2242 ztest_lr_alloc(size_t lrsize
, char *name
)
2245 size_t namesize
= name
? strlen(name
) + 1 : 0;
2247 lr
= umem_zalloc(lrsize
+ namesize
, UMEM_NOFAIL
);
2250 bcopy(name
, lr
+ lrsize
, namesize
);
2256 ztest_lr_free(void *lr
, size_t lrsize
, char *name
)
2258 size_t namesize
= name
? strlen(name
) + 1 : 0;
2260 umem_free(lr
, lrsize
+ namesize
);
2264 * Lookup a bunch of objects. Returns the number of objects not found.
2267 ztest_lookup(ztest_ds_t
*zd
, ztest_od_t
*od
, int count
)
2273 ASSERT(MUTEX_HELD(&zd
->zd_dirobj_lock
));
2275 for (i
= 0; i
< count
; i
++, od
++) {
2277 error
= zap_lookup(zd
->zd_os
, od
->od_dir
, od
->od_name
,
2278 sizeof (uint64_t), 1, &od
->od_object
);
2280 ASSERT(error
== ENOENT
);
2281 ASSERT(od
->od_object
== 0);
2285 ztest_block_tag_t
*bbt
;
2286 dmu_object_info_t doi
;
2288 ASSERT(od
->od_object
!= 0);
2289 ASSERT(missing
== 0); /* there should be no gaps */
2291 ztest_object_lock(zd
, od
->od_object
, RL_READER
);
2292 VERIFY3U(0, ==, dmu_bonus_hold(zd
->zd_os
,
2293 od
->od_object
, FTAG
, &db
));
2294 dmu_object_info_from_db(db
, &doi
);
2295 bbt
= ztest_bt_bonus(db
);
2296 ASSERT3U(bbt
->bt_magic
, ==, BT_MAGIC
);
2297 od
->od_type
= doi
.doi_type
;
2298 od
->od_blocksize
= doi
.doi_data_block_size
;
2299 od
->od_gen
= bbt
->bt_gen
;
2300 dmu_buf_rele(db
, FTAG
);
2301 ztest_object_unlock(zd
, od
->od_object
);
2309 ztest_create(ztest_ds_t
*zd
, ztest_od_t
*od
, int count
)
2314 ASSERT(MUTEX_HELD(&zd
->zd_dirobj_lock
));
2316 for (i
= 0; i
< count
; i
++, od
++) {
2323 lr_create_t
*lr
= ztest_lr_alloc(sizeof (*lr
), od
->od_name
);
2325 lr
->lr_doid
= od
->od_dir
;
2326 lr
->lr_foid
= 0; /* 0 to allocate, > 0 to claim */
2327 lr
->lrz_type
= od
->od_crtype
;
2328 lr
->lrz_blocksize
= od
->od_crblocksize
;
2329 lr
->lrz_ibshift
= ztest_random_ibshift();
2330 lr
->lrz_bonustype
= DMU_OT_UINT64_OTHER
;
2331 lr
->lrz_dnodesize
= od
->od_crdnodesize
;
2332 lr
->lr_gen
= od
->od_crgen
;
2333 lr
->lr_crtime
[0] = time(NULL
);
2335 if (ztest_replay_create(zd
, lr
, B_FALSE
) != 0) {
2336 ASSERT(missing
== 0);
2340 od
->od_object
= lr
->lr_foid
;
2341 od
->od_type
= od
->od_crtype
;
2342 od
->od_blocksize
= od
->od_crblocksize
;
2343 od
->od_gen
= od
->od_crgen
;
2344 ASSERT(od
->od_object
!= 0);
2347 ztest_lr_free(lr
, sizeof (*lr
), od
->od_name
);
2354 ztest_remove(ztest_ds_t
*zd
, ztest_od_t
*od
, int count
)
2360 ASSERT(MUTEX_HELD(&zd
->zd_dirobj_lock
));
2364 for (i
= count
- 1; i
>= 0; i
--, od
--) {
2371 * No object was found.
2373 if (od
->od_object
== 0)
2376 lr_remove_t
*lr
= ztest_lr_alloc(sizeof (*lr
), od
->od_name
);
2378 lr
->lr_doid
= od
->od_dir
;
2380 if ((error
= ztest_replay_remove(zd
, lr
, B_FALSE
)) != 0) {
2381 ASSERT3U(error
, ==, ENOSPC
);
2386 ztest_lr_free(lr
, sizeof (*lr
), od
->od_name
);
2393 ztest_write(ztest_ds_t
*zd
, uint64_t object
, uint64_t offset
, uint64_t size
,
2399 lr
= ztest_lr_alloc(sizeof (*lr
) + size
, NULL
);
2401 lr
->lr_foid
= object
;
2402 lr
->lr_offset
= offset
;
2403 lr
->lr_length
= size
;
2405 BP_ZERO(&lr
->lr_blkptr
);
2407 bcopy(data
, lr
+ 1, size
);
2409 error
= ztest_replay_write(zd
, lr
, B_FALSE
);
2411 ztest_lr_free(lr
, sizeof (*lr
) + size
, NULL
);
2417 ztest_truncate(ztest_ds_t
*zd
, uint64_t object
, uint64_t offset
, uint64_t size
)
2422 lr
= ztest_lr_alloc(sizeof (*lr
), NULL
);
2424 lr
->lr_foid
= object
;
2425 lr
->lr_offset
= offset
;
2426 lr
->lr_length
= size
;
2428 error
= ztest_replay_truncate(zd
, lr
, B_FALSE
);
2430 ztest_lr_free(lr
, sizeof (*lr
), NULL
);
2436 ztest_setattr(ztest_ds_t
*zd
, uint64_t object
)
2441 lr
= ztest_lr_alloc(sizeof (*lr
), NULL
);
2443 lr
->lr_foid
= object
;
2447 error
= ztest_replay_setattr(zd
, lr
, B_FALSE
);
2449 ztest_lr_free(lr
, sizeof (*lr
), NULL
);
2455 ztest_prealloc(ztest_ds_t
*zd
, uint64_t object
, uint64_t offset
, uint64_t size
)
2457 objset_t
*os
= zd
->zd_os
;
2462 txg_wait_synced(dmu_objset_pool(os
), 0);
2464 ztest_object_lock(zd
, object
, RL_READER
);
2465 rl
= ztest_range_lock(zd
, object
, offset
, size
, RL_WRITER
);
2467 tx
= dmu_tx_create(os
);
2469 dmu_tx_hold_write(tx
, object
, offset
, size
);
2471 txg
= ztest_tx_assign(tx
, TXG_WAIT
, FTAG
);
2474 dmu_prealloc(os
, object
, offset
, size
, tx
);
2476 txg_wait_synced(dmu_objset_pool(os
), txg
);
2478 (void) dmu_free_long_range(os
, object
, offset
, size
);
2481 ztest_range_unlock(zd
, rl
);
2482 ztest_object_unlock(zd
, object
);
2486 ztest_io(ztest_ds_t
*zd
, uint64_t object
, uint64_t offset
)
2489 ztest_block_tag_t wbt
;
2490 dmu_object_info_t doi
;
2491 enum ztest_io_type io_type
;
2495 VERIFY(dmu_object_info(zd
->zd_os
, object
, &doi
) == 0);
2496 blocksize
= doi
.doi_data_block_size
;
2497 data
= umem_alloc(blocksize
, UMEM_NOFAIL
);
2500 * Pick an i/o type at random, biased toward writing block tags.
2502 io_type
= ztest_random(ZTEST_IO_TYPES
);
2503 if (ztest_random(2) == 0)
2504 io_type
= ZTEST_IO_WRITE_TAG
;
2506 (void) rw_rdlock(&zd
->zd_zilog_lock
);
2510 case ZTEST_IO_WRITE_TAG
:
2511 ztest_bt_generate(&wbt
, zd
->zd_os
, object
, doi
.doi_dnodesize
,
2513 (void) ztest_write(zd
, object
, offset
, sizeof (wbt
), &wbt
);
2516 case ZTEST_IO_WRITE_PATTERN
:
2517 (void) memset(data
, 'a' + (object
+ offset
) % 5, blocksize
);
2518 if (ztest_random(2) == 0) {
2520 * Induce fletcher2 collisions to ensure that
2521 * zio_ddt_collision() detects and resolves them
2522 * when using fletcher2-verify for deduplication.
2524 ((uint64_t *)data
)[0] ^= 1ULL << 63;
2525 ((uint64_t *)data
)[4] ^= 1ULL << 63;
2527 (void) ztest_write(zd
, object
, offset
, blocksize
, data
);
2530 case ZTEST_IO_WRITE_ZEROES
:
2531 bzero(data
, blocksize
);
2532 (void) ztest_write(zd
, object
, offset
, blocksize
, data
);
2535 case ZTEST_IO_TRUNCATE
:
2536 (void) ztest_truncate(zd
, object
, offset
, blocksize
);
2539 case ZTEST_IO_SETATTR
:
2540 (void) ztest_setattr(zd
, object
);
2545 case ZTEST_IO_REWRITE
:
2546 (void) rw_rdlock(&ztest_name_lock
);
2547 err
= ztest_dsl_prop_set_uint64(zd
->zd_name
,
2548 ZFS_PROP_CHECKSUM
, spa_dedup_checksum(ztest_spa
),
2550 VERIFY(err
== 0 || err
== ENOSPC
);
2551 err
= ztest_dsl_prop_set_uint64(zd
->zd_name
,
2552 ZFS_PROP_COMPRESSION
,
2553 ztest_random_dsl_prop(ZFS_PROP_COMPRESSION
),
2555 VERIFY(err
== 0 || err
== ENOSPC
);
2556 (void) rw_unlock(&ztest_name_lock
);
2558 VERIFY0(dmu_read(zd
->zd_os
, object
, offset
, blocksize
, data
,
2559 DMU_READ_NO_PREFETCH
));
2561 (void) ztest_write(zd
, object
, offset
, blocksize
, data
);
2565 (void) rw_unlock(&zd
->zd_zilog_lock
);
2567 umem_free(data
, blocksize
);
2571 * Initialize an object description template.
2574 ztest_od_init(ztest_od_t
*od
, uint64_t id
, char *tag
, uint64_t index
,
2575 dmu_object_type_t type
, uint64_t blocksize
, uint64_t dnodesize
,
2578 od
->od_dir
= ZTEST_DIROBJ
;
2581 od
->od_crtype
= type
;
2582 od
->od_crblocksize
= blocksize
? blocksize
: ztest_random_blocksize();
2583 od
->od_crdnodesize
= dnodesize
? dnodesize
: ztest_random_dnodesize();
2586 od
->od_type
= DMU_OT_NONE
;
2587 od
->od_blocksize
= 0;
2590 (void) snprintf(od
->od_name
, sizeof (od
->od_name
), "%s(%lld)[%llu]",
2591 tag
, (longlong_t
)id
, (u_longlong_t
)index
);
2595 * Lookup or create the objects for a test using the od template.
2596 * If the objects do not all exist, or if 'remove' is specified,
2597 * remove any existing objects and create new ones. Otherwise,
2598 * use the existing objects.
2601 ztest_object_init(ztest_ds_t
*zd
, ztest_od_t
*od
, size_t size
, boolean_t remove
)
2603 int count
= size
/ sizeof (*od
);
2606 mutex_enter(&zd
->zd_dirobj_lock
);
2607 if ((ztest_lookup(zd
, od
, count
) != 0 || remove
) &&
2608 (ztest_remove(zd
, od
, count
) != 0 ||
2609 ztest_create(zd
, od
, count
) != 0))
2612 mutex_exit(&zd
->zd_dirobj_lock
);
2619 ztest_zil_commit(ztest_ds_t
*zd
, uint64_t id
)
2621 zilog_t
*zilog
= zd
->zd_zilog
;
2623 (void) rw_rdlock(&zd
->zd_zilog_lock
);
2625 zil_commit(zilog
, ztest_random(ZTEST_OBJECTS
));
2628 * Remember the committed values in zd, which is in parent/child
2629 * shared memory. If we die, the next iteration of ztest_run()
2630 * will verify that the log really does contain this record.
2632 mutex_enter(&zilog
->zl_lock
);
2633 ASSERT(zd
->zd_shared
!= NULL
);
2634 ASSERT3U(zd
->zd_shared
->zd_seq
, <=, zilog
->zl_commit_lr_seq
);
2635 zd
->zd_shared
->zd_seq
= zilog
->zl_commit_lr_seq
;
2636 mutex_exit(&zilog
->zl_lock
);
2638 (void) rw_unlock(&zd
->zd_zilog_lock
);
2642 * This function is designed to simulate the operations that occur during a
2643 * mount/unmount operation. We hold the dataset across these operations in an
2644 * attempt to expose any implicit assumptions about ZIL management.
2648 ztest_zil_remount(ztest_ds_t
*zd
, uint64_t id
)
2650 objset_t
*os
= zd
->zd_os
;
2653 * We grab the zd_dirobj_lock to ensure that no other thread is
2654 * updating the zil (i.e. adding in-memory log records) and the
2655 * zd_zilog_lock to block any I/O.
2657 mutex_enter(&zd
->zd_dirobj_lock
);
2658 (void) rw_wrlock(&zd
->zd_zilog_lock
);
2660 /* zfsvfs_teardown() */
2661 zil_close(zd
->zd_zilog
);
2663 /* zfsvfs_setup() */
2664 VERIFY(zil_open(os
, ztest_get_data
) == zd
->zd_zilog
);
2665 zil_replay(os
, zd
, ztest_replay_vector
);
2667 (void) rw_unlock(&zd
->zd_zilog_lock
);
2668 mutex_exit(&zd
->zd_dirobj_lock
);
2672 * Verify that we can't destroy an active pool, create an existing pool,
2673 * or create a pool with a bad vdev spec.
2677 ztest_spa_create_destroy(ztest_ds_t
*zd
, uint64_t id
)
2679 ztest_shared_opts_t
*zo
= &ztest_opts
;
2683 if (zo
->zo_mmp_test
)
2687 * Attempt to create using a bad file.
2689 nvroot
= make_vdev_root("/dev/bogus", NULL
, NULL
, 0, 0, 0, 0, 0, 1);
2690 VERIFY3U(ENOENT
, ==,
2691 spa_create("ztest_bad_file", nvroot
, NULL
, NULL
, NULL
));
2692 nvlist_free(nvroot
);
2695 * Attempt to create using a bad mirror.
2697 nvroot
= make_vdev_root("/dev/bogus", NULL
, NULL
, 0, 0, 0, 0, 2, 1);
2698 VERIFY3U(ENOENT
, ==,
2699 spa_create("ztest_bad_mirror", nvroot
, NULL
, NULL
, NULL
));
2700 nvlist_free(nvroot
);
2703 * Attempt to create an existing pool. It shouldn't matter
2704 * what's in the nvroot; we should fail with EEXIST.
2706 (void) rw_rdlock(&ztest_name_lock
);
2707 nvroot
= make_vdev_root("/dev/bogus", NULL
, NULL
, 0, 0, 0, 0, 0, 1);
2708 VERIFY3U(EEXIST
, ==,
2709 spa_create(zo
->zo_pool
, nvroot
, NULL
, NULL
, NULL
));
2710 nvlist_free(nvroot
);
2711 VERIFY3U(0, ==, spa_open(zo
->zo_pool
, &spa
, FTAG
));
2712 VERIFY3U(EBUSY
, ==, spa_destroy(zo
->zo_pool
));
2713 spa_close(spa
, FTAG
);
2715 (void) rw_unlock(&ztest_name_lock
);
2719 * Start and then stop the MMP threads to ensure the startup and shutdown code
2720 * works properly. Actual protection and property-related code tested via ZTS.
2724 ztest_mmp_enable_disable(ztest_ds_t
*zd
, uint64_t id
)
2726 ztest_shared_opts_t
*zo
= &ztest_opts
;
2727 spa_t
*spa
= ztest_spa
;
2729 if (zo
->zo_mmp_test
)
2732 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
2733 mutex_enter(&spa
->spa_props_lock
);
2735 if (!spa_multihost(spa
)) {
2736 spa
->spa_multihost
= B_TRUE
;
2737 mmp_thread_start(spa
);
2740 mutex_exit(&spa
->spa_props_lock
);
2741 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
2743 txg_wait_synced(spa_get_dsl(spa
), 0);
2744 mmp_signal_all_threads();
2745 txg_wait_synced(spa_get_dsl(spa
), 0);
2747 spa_config_enter(spa
, SCL_CONFIG
, FTAG
, RW_READER
);
2748 mutex_enter(&spa
->spa_props_lock
);
2750 if (spa_multihost(spa
)) {
2751 mmp_thread_stop(spa
);
2752 spa
->spa_multihost
= B_FALSE
;
2755 mutex_exit(&spa
->spa_props_lock
);
2756 spa_config_exit(spa
, SCL_CONFIG
, FTAG
);
2761 ztest_spa_upgrade(ztest_ds_t
*zd
, uint64_t id
)
2764 uint64_t initial_version
= SPA_VERSION_INITIAL
;
2765 uint64_t version
, newversion
;
2766 nvlist_t
*nvroot
, *props
;
2769 if (ztest_opts
.zo_mmp_test
)
2772 mutex_enter(&ztest_vdev_lock
);
2773 name
= kmem_asprintf("%s_upgrade", ztest_opts
.zo_pool
);
2776 * Clean up from previous runs.
2778 (void) spa_destroy(name
);
2780 nvroot
= make_vdev_root(NULL
, NULL
, name
, ztest_opts
.zo_vdev_size
, 0,
2781 0, ztest_opts
.zo_raidz
, ztest_opts
.zo_mirrors
, 1);
2784 * If we're configuring a RAIDZ device then make sure that the
2785 * the initial version is capable of supporting that feature.
2787 switch (ztest_opts
.zo_raidz_parity
) {
2790 initial_version
= SPA_VERSION_INITIAL
;
2793 initial_version
= SPA_VERSION_RAIDZ2
;
2796 initial_version
= SPA_VERSION_RAIDZ3
;
2801 * Create a pool with a spa version that can be upgraded. Pick
2802 * a value between initial_version and SPA_VERSION_BEFORE_FEATURES.
2805 version
= ztest_random_spa_version(initial_version
);
2806 } while (version
> SPA_VERSION_BEFORE_FEATURES
);
2808 props
= fnvlist_alloc();
2809 fnvlist_add_uint64(props
,
2810 zpool_prop_to_name(ZPOOL_PROP_VERSION
), version
);
2811 VERIFY3S(spa_create(name
, nvroot
, props
, NULL
, NULL
), ==, 0);
2812 fnvlist_free(nvroot
);
2813 fnvlist_free(props
);
2815 VERIFY3S(spa_open(name
, &spa
, FTAG
), ==, 0);
2816 VERIFY3U(spa_version(spa
), ==, version
);
2817 newversion
= ztest_random_spa_version(version
+ 1);
2819 if (ztest_opts
.zo_verbose
>= 4) {
2820 (void) printf("upgrading spa version from %llu to %llu\n",
2821 (u_longlong_t
)version
, (u_longlong_t
)newversion
);
2824 spa_upgrade(spa
, newversion
);
2825 VERIFY3U(spa_version(spa
), >, version
);
2826 VERIFY3U(spa_version(spa
), ==, fnvlist_lookup_uint64(spa
->spa_config
,
2827 zpool_prop_to_name(ZPOOL_PROP_VERSION
)));
2828 spa_close(spa
, FTAG
);
2831 mutex_exit(&ztest_vdev_lock
);
2835 vdev_lookup_by_path(vdev_t
*vd
, const char *path
)
2840 if (vd
->vdev_path
!= NULL
&& strcmp(path
, vd
->vdev_path
) == 0)
2843 for (c
= 0; c
< vd
->vdev_children
; c
++)
2844 if ((mvd
= vdev_lookup_by_path(vd
->vdev_child
[c
], path
)) !=
2852 * Find the first available hole which can be used as a top-level.
2855 find_vdev_hole(spa_t
*spa
)
2857 vdev_t
*rvd
= spa
->spa_root_vdev
;
2860 ASSERT(spa_config_held(spa
, SCL_VDEV
, RW_READER
) == SCL_VDEV
);
2862 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
2863 vdev_t
*cvd
= rvd
->vdev_child
[c
];
2865 if (cvd
->vdev_ishole
)
2872 * Verify that vdev_add() works as expected.
2876 ztest_vdev_add_remove(ztest_ds_t
*zd
, uint64_t id
)
2878 ztest_shared_t
*zs
= ztest_shared
;
2879 spa_t
*spa
= ztest_spa
;
2885 if (ztest_opts
.zo_mmp_test
)
2888 mutex_enter(&ztest_vdev_lock
);
2889 leaves
= MAX(zs
->zs_mirrors
+ zs
->zs_splits
, 1) * ztest_opts
.zo_raidz
;
2891 spa_config_enter(spa
, SCL_VDEV
, FTAG
, RW_READER
);
2893 ztest_shared
->zs_vdev_next_leaf
= find_vdev_hole(spa
) * leaves
;
2896 * If we have slogs then remove them 1/4 of the time.
2898 if (spa_has_slogs(spa
) && ztest_random(4) == 0) {
2900 * Grab the guid from the head of the log class rotor.
2902 guid
= spa_log_class(spa
)->mc_rotor
->mg_vd
->vdev_guid
;
2904 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
2907 * We have to grab the zs_name_lock as writer to
2908 * prevent a race between removing a slog (dmu_objset_find)
2909 * and destroying a dataset. Removing the slog will
2910 * grab a reference on the dataset which may cause
2911 * dsl_destroy_head() to fail with EBUSY thus
2912 * leaving the dataset in an inconsistent state.
2914 rw_wrlock(&ztest_name_lock
);
2915 error
= spa_vdev_remove(spa
, guid
, B_FALSE
);
2916 rw_unlock(&ztest_name_lock
);
2918 if (error
&& error
!= EEXIST
)
2919 fatal(0, "spa_vdev_remove() = %d", error
);
2921 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
2924 * Make 1/4 of the devices be log devices.
2926 nvroot
= make_vdev_root(NULL
, NULL
, NULL
,
2927 ztest_opts
.zo_vdev_size
, 0,
2928 ztest_random(4) == 0, ztest_opts
.zo_raidz
,
2931 error
= spa_vdev_add(spa
, nvroot
);
2932 nvlist_free(nvroot
);
2934 if (error
== ENOSPC
)
2935 ztest_record_enospc("spa_vdev_add");
2936 else if (error
!= 0)
2937 fatal(0, "spa_vdev_add() = %d", error
);
2940 mutex_exit(&ztest_vdev_lock
);
2944 * Verify that adding/removing aux devices (l2arc, hot spare) works as expected.
2948 ztest_vdev_aux_add_remove(ztest_ds_t
*zd
, uint64_t id
)
2950 ztest_shared_t
*zs
= ztest_shared
;
2951 spa_t
*spa
= ztest_spa
;
2952 vdev_t
*rvd
= spa
->spa_root_vdev
;
2953 spa_aux_vdev_t
*sav
;
2959 if (ztest_opts
.zo_mmp_test
)
2962 path
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
2964 if (ztest_random(2) == 0) {
2965 sav
= &spa
->spa_spares
;
2966 aux
= ZPOOL_CONFIG_SPARES
;
2968 sav
= &spa
->spa_l2cache
;
2969 aux
= ZPOOL_CONFIG_L2CACHE
;
2972 mutex_enter(&ztest_vdev_lock
);
2974 spa_config_enter(spa
, SCL_VDEV
, FTAG
, RW_READER
);
2976 if (sav
->sav_count
!= 0 && ztest_random(4) == 0) {
2978 * Pick a random device to remove.
2980 guid
= sav
->sav_vdevs
[ztest_random(sav
->sav_count
)]->vdev_guid
;
2983 * Find an unused device we can add.
2985 zs
->zs_vdev_aux
= 0;
2988 (void) snprintf(path
, MAXPATHLEN
, ztest_aux_template
,
2989 ztest_opts
.zo_dir
, ztest_opts
.zo_pool
, aux
,
2991 for (c
= 0; c
< sav
->sav_count
; c
++)
2992 if (strcmp(sav
->sav_vdevs
[c
]->vdev_path
,
2995 if (c
== sav
->sav_count
&&
2996 vdev_lookup_by_path(rvd
, path
) == NULL
)
3002 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
3008 nvlist_t
*nvroot
= make_vdev_root(NULL
, aux
, NULL
,
3009 (ztest_opts
.zo_vdev_size
* 5) / 4, 0, 0, 0, 0, 1);
3010 error
= spa_vdev_add(spa
, nvroot
);
3012 fatal(0, "spa_vdev_add(%p) = %d", nvroot
, error
);
3013 nvlist_free(nvroot
);
3016 * Remove an existing device. Sometimes, dirty its
3017 * vdev state first to make sure we handle removal
3018 * of devices that have pending state changes.
3020 if (ztest_random(2) == 0)
3021 (void) vdev_online(spa
, guid
, 0, NULL
);
3023 error
= spa_vdev_remove(spa
, guid
, B_FALSE
);
3024 if (error
!= 0 && error
!= EBUSY
)
3025 fatal(0, "spa_vdev_remove(%llu) = %d", guid
, error
);
3028 mutex_exit(&ztest_vdev_lock
);
3030 umem_free(path
, MAXPATHLEN
);
3034 * split a pool if it has mirror tlvdevs
3038 ztest_split_pool(ztest_ds_t
*zd
, uint64_t id
)
3040 ztest_shared_t
*zs
= ztest_shared
;
3041 spa_t
*spa
= ztest_spa
;
3042 vdev_t
*rvd
= spa
->spa_root_vdev
;
3043 nvlist_t
*tree
, **child
, *config
, *split
, **schild
;
3044 uint_t c
, children
, schildren
= 0, lastlogid
= 0;
3047 if (ztest_opts
.zo_mmp_test
)
3050 mutex_enter(&ztest_vdev_lock
);
3052 /* ensure we have a useable config; mirrors of raidz aren't supported */
3053 if (zs
->zs_mirrors
< 3 || ztest_opts
.zo_raidz
> 1) {
3054 mutex_exit(&ztest_vdev_lock
);
3058 /* clean up the old pool, if any */
3059 (void) spa_destroy("splitp");
3061 spa_config_enter(spa
, SCL_VDEV
, FTAG
, RW_READER
);
3063 /* generate a config from the existing config */
3064 mutex_enter(&spa
->spa_props_lock
);
3065 VERIFY(nvlist_lookup_nvlist(spa
->spa_config
, ZPOOL_CONFIG_VDEV_TREE
,
3067 mutex_exit(&spa
->spa_props_lock
);
3069 VERIFY(nvlist_lookup_nvlist_array(tree
, ZPOOL_CONFIG_CHILDREN
, &child
,
3072 schild
= malloc(rvd
->vdev_children
* sizeof (nvlist_t
*));
3073 for (c
= 0; c
< children
; c
++) {
3074 vdev_t
*tvd
= rvd
->vdev_child
[c
];
3078 if (tvd
->vdev_islog
|| tvd
->vdev_ops
== &vdev_hole_ops
) {
3079 VERIFY(nvlist_alloc(&schild
[schildren
], NV_UNIQUE_NAME
,
3081 VERIFY(nvlist_add_string(schild
[schildren
],
3082 ZPOOL_CONFIG_TYPE
, VDEV_TYPE_HOLE
) == 0);
3083 VERIFY(nvlist_add_uint64(schild
[schildren
],
3084 ZPOOL_CONFIG_IS_HOLE
, 1) == 0);
3086 lastlogid
= schildren
;
3091 VERIFY(nvlist_lookup_nvlist_array(child
[c
],
3092 ZPOOL_CONFIG_CHILDREN
, &mchild
, &mchildren
) == 0);
3093 VERIFY(nvlist_dup(mchild
[0], &schild
[schildren
++], 0) == 0);
3096 /* OK, create a config that can be used to split */
3097 VERIFY(nvlist_alloc(&split
, NV_UNIQUE_NAME
, 0) == 0);
3098 VERIFY(nvlist_add_string(split
, ZPOOL_CONFIG_TYPE
,
3099 VDEV_TYPE_ROOT
) == 0);
3100 VERIFY(nvlist_add_nvlist_array(split
, ZPOOL_CONFIG_CHILDREN
, schild
,
3101 lastlogid
!= 0 ? lastlogid
: schildren
) == 0);
3103 VERIFY(nvlist_alloc(&config
, NV_UNIQUE_NAME
, 0) == 0);
3104 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, split
) == 0);
3106 for (c
= 0; c
< schildren
; c
++)
3107 nvlist_free(schild
[c
]);
3111 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
3113 (void) rw_wrlock(&ztest_name_lock
);
3114 error
= spa_vdev_split_mirror(spa
, "splitp", config
, NULL
, B_FALSE
);
3115 (void) rw_unlock(&ztest_name_lock
);
3117 nvlist_free(config
);
3120 (void) printf("successful split - results:\n");
3121 mutex_enter(&spa_namespace_lock
);
3122 show_pool_stats(spa
);
3123 show_pool_stats(spa_lookup("splitp"));
3124 mutex_exit(&spa_namespace_lock
);
3128 mutex_exit(&ztest_vdev_lock
);
3133 * Verify that we can attach and detach devices.
3137 ztest_vdev_attach_detach(ztest_ds_t
*zd
, uint64_t id
)
3139 ztest_shared_t
*zs
= ztest_shared
;
3140 spa_t
*spa
= ztest_spa
;
3141 spa_aux_vdev_t
*sav
= &spa
->spa_spares
;
3142 vdev_t
*rvd
= spa
->spa_root_vdev
;
3143 vdev_t
*oldvd
, *newvd
, *pvd
;
3147 uint64_t ashift
= ztest_get_ashift();
3148 uint64_t oldguid
, pguid
;
3149 uint64_t oldsize
, newsize
;
3150 char *oldpath
, *newpath
;
3152 int oldvd_has_siblings
= B_FALSE
;
3153 int newvd_is_spare
= B_FALSE
;
3155 int error
, expected_error
;
3157 if (ztest_opts
.zo_mmp_test
)
3160 oldpath
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
3161 newpath
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
3163 mutex_enter(&ztest_vdev_lock
);
3164 leaves
= MAX(zs
->zs_mirrors
, 1) * ztest_opts
.zo_raidz
;
3166 spa_config_enter(spa
, SCL_VDEV
, FTAG
, RW_READER
);
3169 * Decide whether to do an attach or a replace.
3171 replacing
= ztest_random(2);
3174 * Pick a random top-level vdev.
3176 top
= ztest_random_vdev_top(spa
, B_TRUE
);
3179 * Pick a random leaf within it.
3181 leaf
= ztest_random(leaves
);
3186 oldvd
= rvd
->vdev_child
[top
];
3187 if (zs
->zs_mirrors
>= 1) {
3188 ASSERT(oldvd
->vdev_ops
== &vdev_mirror_ops
);
3189 ASSERT(oldvd
->vdev_children
>= zs
->zs_mirrors
);
3190 oldvd
= oldvd
->vdev_child
[leaf
/ ztest_opts
.zo_raidz
];
3192 if (ztest_opts
.zo_raidz
> 1) {
3193 ASSERT(oldvd
->vdev_ops
== &vdev_raidz_ops
);
3194 ASSERT(oldvd
->vdev_children
== ztest_opts
.zo_raidz
);
3195 oldvd
= oldvd
->vdev_child
[leaf
% ztest_opts
.zo_raidz
];
3199 * If we're already doing an attach or replace, oldvd may be a
3200 * mirror vdev -- in which case, pick a random child.
3202 while (oldvd
->vdev_children
!= 0) {
3203 oldvd_has_siblings
= B_TRUE
;
3204 ASSERT(oldvd
->vdev_children
>= 2);
3205 oldvd
= oldvd
->vdev_child
[ztest_random(oldvd
->vdev_children
)];
3208 oldguid
= oldvd
->vdev_guid
;
3209 oldsize
= vdev_get_min_asize(oldvd
);
3210 oldvd_is_log
= oldvd
->vdev_top
->vdev_islog
;
3211 (void) strcpy(oldpath
, oldvd
->vdev_path
);
3212 pvd
= oldvd
->vdev_parent
;
3213 pguid
= pvd
->vdev_guid
;
3216 * If oldvd has siblings, then half of the time, detach it.
3218 if (oldvd_has_siblings
&& ztest_random(2) == 0) {
3219 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
3220 error
= spa_vdev_detach(spa
, oldguid
, pguid
, B_FALSE
);
3221 if (error
!= 0 && error
!= ENODEV
&& error
!= EBUSY
&&
3223 fatal(0, "detach (%s) returned %d", oldpath
, error
);
3228 * For the new vdev, choose with equal probability between the two
3229 * standard paths (ending in either 'a' or 'b') or a random hot spare.
3231 if (sav
->sav_count
!= 0 && ztest_random(3) == 0) {
3232 newvd
= sav
->sav_vdevs
[ztest_random(sav
->sav_count
)];
3233 newvd_is_spare
= B_TRUE
;
3234 (void) strcpy(newpath
, newvd
->vdev_path
);
3236 (void) snprintf(newpath
, MAXPATHLEN
, ztest_dev_template
,
3237 ztest_opts
.zo_dir
, ztest_opts
.zo_pool
,
3238 top
* leaves
+ leaf
);
3239 if (ztest_random(2) == 0)
3240 newpath
[strlen(newpath
) - 1] = 'b';
3241 newvd
= vdev_lookup_by_path(rvd
, newpath
);
3245 newsize
= vdev_get_min_asize(newvd
);
3248 * Make newsize a little bigger or smaller than oldsize.
3249 * If it's smaller, the attach should fail.
3250 * If it's larger, and we're doing a replace,
3251 * we should get dynamic LUN growth when we're done.
3253 newsize
= 10 * oldsize
/ (9 + ztest_random(3));
3257 * If pvd is not a mirror or root, the attach should fail with ENOTSUP,
3258 * unless it's a replace; in that case any non-replacing parent is OK.
3260 * If newvd is already part of the pool, it should fail with EBUSY.
3262 * If newvd is too small, it should fail with EOVERFLOW.
3264 if (pvd
->vdev_ops
!= &vdev_mirror_ops
&&
3265 pvd
->vdev_ops
!= &vdev_root_ops
&& (!replacing
||
3266 pvd
->vdev_ops
== &vdev_replacing_ops
||
3267 pvd
->vdev_ops
== &vdev_spare_ops
))
3268 expected_error
= ENOTSUP
;
3269 else if (newvd_is_spare
&& (!replacing
|| oldvd_is_log
))
3270 expected_error
= ENOTSUP
;
3271 else if (newvd
== oldvd
)
3272 expected_error
= replacing
? 0 : EBUSY
;
3273 else if (vdev_lookup_by_path(rvd
, newpath
) != NULL
)
3274 expected_error
= EBUSY
;
3275 else if (newsize
< oldsize
)
3276 expected_error
= EOVERFLOW
;
3277 else if (ashift
> oldvd
->vdev_top
->vdev_ashift
)
3278 expected_error
= EDOM
;
3282 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
3285 * Build the nvlist describing newpath.
3287 root
= make_vdev_root(newpath
, NULL
, NULL
, newvd
== NULL
? newsize
: 0,
3288 ashift
, 0, 0, 0, 1);
3290 error
= spa_vdev_attach(spa
, oldguid
, root
, replacing
);
3295 * If our parent was the replacing vdev, but the replace completed,
3296 * then instead of failing with ENOTSUP we may either succeed,
3297 * fail with ENODEV, or fail with EOVERFLOW.
3299 if (expected_error
== ENOTSUP
&&
3300 (error
== 0 || error
== ENODEV
|| error
== EOVERFLOW
))
3301 expected_error
= error
;
3304 * If someone grew the LUN, the replacement may be too small.
3306 if (error
== EOVERFLOW
|| error
== EBUSY
)
3307 expected_error
= error
;
3309 /* XXX workaround 6690467 */
3310 if (error
!= expected_error
&& expected_error
!= EBUSY
) {
3311 fatal(0, "attach (%s %llu, %s %llu, %d) "
3312 "returned %d, expected %d",
3313 oldpath
, oldsize
, newpath
,
3314 newsize
, replacing
, error
, expected_error
);
3317 mutex_exit(&ztest_vdev_lock
);
3319 umem_free(oldpath
, MAXPATHLEN
);
3320 umem_free(newpath
, MAXPATHLEN
);
3324 * Callback function which expands the physical size of the vdev.
3327 grow_vdev(vdev_t
*vd
, void *arg
)
3329 ASSERTV(spa_t
*spa
= vd
->vdev_spa
);
3330 size_t *newsize
= arg
;
3334 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_READER
) == SCL_STATE
);
3335 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
3337 if ((fd
= open(vd
->vdev_path
, O_RDWR
)) == -1)
3340 fsize
= lseek(fd
, 0, SEEK_END
);
3341 VERIFY(ftruncate(fd
, *newsize
) == 0);
3343 if (ztest_opts
.zo_verbose
>= 6) {
3344 (void) printf("%s grew from %lu to %lu bytes\n",
3345 vd
->vdev_path
, (ulong_t
)fsize
, (ulong_t
)*newsize
);
3352 * Callback function which expands a given vdev by calling vdev_online().
3356 online_vdev(vdev_t
*vd
, void *arg
)
3358 spa_t
*spa
= vd
->vdev_spa
;
3359 vdev_t
*tvd
= vd
->vdev_top
;
3360 uint64_t guid
= vd
->vdev_guid
;
3361 uint64_t generation
= spa
->spa_config_generation
+ 1;
3362 vdev_state_t newstate
= VDEV_STATE_UNKNOWN
;
3365 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_READER
) == SCL_STATE
);
3366 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
3368 /* Calling vdev_online will initialize the new metaslabs */
3369 spa_config_exit(spa
, SCL_STATE
, spa
);
3370 error
= vdev_online(spa
, guid
, ZFS_ONLINE_EXPAND
, &newstate
);
3371 spa_config_enter(spa
, SCL_STATE
, spa
, RW_READER
);
3374 * If vdev_online returned an error or the underlying vdev_open
3375 * failed then we abort the expand. The only way to know that
3376 * vdev_open fails is by checking the returned newstate.
3378 if (error
|| newstate
!= VDEV_STATE_HEALTHY
) {
3379 if (ztest_opts
.zo_verbose
>= 5) {
3380 (void) printf("Unable to expand vdev, state %llu, "
3381 "error %d\n", (u_longlong_t
)newstate
, error
);
3385 ASSERT3U(newstate
, ==, VDEV_STATE_HEALTHY
);
3388 * Since we dropped the lock we need to ensure that we're
3389 * still talking to the original vdev. It's possible this
3390 * vdev may have been detached/replaced while we were
3391 * trying to online it.
3393 if (generation
!= spa
->spa_config_generation
) {
3394 if (ztest_opts
.zo_verbose
>= 5) {
3395 (void) printf("vdev configuration has changed, "
3396 "guid %llu, state %llu, expected gen %llu, "
3399 (u_longlong_t
)tvd
->vdev_state
,
3400 (u_longlong_t
)generation
,
3401 (u_longlong_t
)spa
->spa_config_generation
);
3409 * Traverse the vdev tree calling the supplied function.
3410 * We continue to walk the tree until we either have walked all
3411 * children or we receive a non-NULL return from the callback.
3412 * If a NULL callback is passed, then we just return back the first
3413 * leaf vdev we encounter.
3416 vdev_walk_tree(vdev_t
*vd
, vdev_t
*(*func
)(vdev_t
*, void *), void *arg
)
3420 if (vd
->vdev_ops
->vdev_op_leaf
) {
3424 return (func(vd
, arg
));
3427 for (c
= 0; c
< vd
->vdev_children
; c
++) {
3428 vdev_t
*cvd
= vd
->vdev_child
[c
];
3429 if ((cvd
= vdev_walk_tree(cvd
, func
, arg
)) != NULL
)
3436 * Verify that dynamic LUN growth works as expected.
3440 ztest_vdev_LUN_growth(ztest_ds_t
*zd
, uint64_t id
)
3442 spa_t
*spa
= ztest_spa
;
3444 metaslab_class_t
*mc
;
3445 metaslab_group_t
*mg
;
3446 size_t psize
, newsize
;
3448 uint64_t old_class_space
, new_class_space
, old_ms_count
, new_ms_count
;
3450 mutex_enter(&ztest_vdev_lock
);
3451 spa_config_enter(spa
, SCL_STATE
, spa
, RW_READER
);
3453 top
= ztest_random_vdev_top(spa
, B_TRUE
);
3455 tvd
= spa
->spa_root_vdev
->vdev_child
[top
];
3458 old_ms_count
= tvd
->vdev_ms_count
;
3459 old_class_space
= metaslab_class_get_space(mc
);
3462 * Determine the size of the first leaf vdev associated with
3463 * our top-level device.
3465 vd
= vdev_walk_tree(tvd
, NULL
, NULL
);
3466 ASSERT3P(vd
, !=, NULL
);
3467 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
3469 psize
= vd
->vdev_psize
;
3472 * We only try to expand the vdev if it's healthy, less than 4x its
3473 * original size, and it has a valid psize.
3475 if (tvd
->vdev_state
!= VDEV_STATE_HEALTHY
||
3476 psize
== 0 || psize
>= 4 * ztest_opts
.zo_vdev_size
) {
3477 spa_config_exit(spa
, SCL_STATE
, spa
);
3478 mutex_exit(&ztest_vdev_lock
);
3482 newsize
= psize
+ psize
/ 8;
3483 ASSERT3U(newsize
, >, psize
);
3485 if (ztest_opts
.zo_verbose
>= 6) {
3486 (void) printf("Expanding LUN %s from %lu to %lu\n",
3487 vd
->vdev_path
, (ulong_t
)psize
, (ulong_t
)newsize
);
3491 * Growing the vdev is a two step process:
3492 * 1). expand the physical size (i.e. relabel)
3493 * 2). online the vdev to create the new metaslabs
3495 if (vdev_walk_tree(tvd
, grow_vdev
, &newsize
) != NULL
||
3496 vdev_walk_tree(tvd
, online_vdev
, NULL
) != NULL
||
3497 tvd
->vdev_state
!= VDEV_STATE_HEALTHY
) {
3498 if (ztest_opts
.zo_verbose
>= 5) {
3499 (void) printf("Could not expand LUN because "
3500 "the vdev configuration changed.\n");
3502 spa_config_exit(spa
, SCL_STATE
, spa
);
3503 mutex_exit(&ztest_vdev_lock
);
3507 spa_config_exit(spa
, SCL_STATE
, spa
);
3510 * Expanding the LUN will update the config asynchronously,
3511 * thus we must wait for the async thread to complete any
3512 * pending tasks before proceeding.
3516 mutex_enter(&spa
->spa_async_lock
);
3517 done
= (spa
->spa_async_thread
== NULL
&& !spa
->spa_async_tasks
);
3518 mutex_exit(&spa
->spa_async_lock
);
3521 txg_wait_synced(spa_get_dsl(spa
), 0);
3522 (void) poll(NULL
, 0, 100);
3525 spa_config_enter(spa
, SCL_STATE
, spa
, RW_READER
);
3527 tvd
= spa
->spa_root_vdev
->vdev_child
[top
];
3528 new_ms_count
= tvd
->vdev_ms_count
;
3529 new_class_space
= metaslab_class_get_space(mc
);
3531 if (tvd
->vdev_mg
!= mg
|| mg
->mg_class
!= mc
) {
3532 if (ztest_opts
.zo_verbose
>= 5) {
3533 (void) printf("Could not verify LUN expansion due to "
3534 "intervening vdev offline or remove.\n");
3536 spa_config_exit(spa
, SCL_STATE
, spa
);
3537 mutex_exit(&ztest_vdev_lock
);
3542 * Make sure we were able to grow the vdev.
3544 if (new_ms_count
<= old_ms_count
)
3545 fatal(0, "LUN expansion failed: ms_count %llu <= %llu\n",
3546 old_ms_count
, new_ms_count
);
3549 * Make sure we were able to grow the pool.
3551 if (new_class_space
<= old_class_space
)
3552 fatal(0, "LUN expansion failed: class_space %llu <= %llu\n",
3553 old_class_space
, new_class_space
);
3555 if (ztest_opts
.zo_verbose
>= 5) {
3556 char oldnumbuf
[NN_NUMBUF_SZ
], newnumbuf
[NN_NUMBUF_SZ
];
3558 nicenum(old_class_space
, oldnumbuf
, sizeof (oldnumbuf
));
3559 nicenum(new_class_space
, newnumbuf
, sizeof (newnumbuf
));
3560 (void) printf("%s grew from %s to %s\n",
3561 spa
->spa_name
, oldnumbuf
, newnumbuf
);
3564 spa_config_exit(spa
, SCL_STATE
, spa
);
3565 mutex_exit(&ztest_vdev_lock
);
3569 * Verify that dmu_objset_{create,destroy,open,close} work as expected.
3573 ztest_objset_create_cb(objset_t
*os
, void *arg
, cred_t
*cr
, dmu_tx_t
*tx
)
3576 * Create the objects common to all ztest datasets.
3578 VERIFY(zap_create_claim(os
, ZTEST_DIROBJ
,
3579 DMU_OT_ZAP_OTHER
, DMU_OT_NONE
, 0, tx
) == 0);
3583 ztest_dataset_create(char *dsname
)
3587 dsl_crypto_params_t
*dcp
= NULL
;
3590 * 50% of the time, we create encrypted datasets
3591 * using a random cipher suite and a hard-coded
3594 rand
= ztest_random(2);
3596 nvlist_t
*crypto_args
= fnvlist_alloc();
3597 nvlist_t
*props
= fnvlist_alloc();
3599 /* slight bias towards the default cipher suite */
3600 rand
= ztest_random(ZIO_CRYPT_FUNCTIONS
);
3601 if (rand
< ZIO_CRYPT_AES_128_CCM
)
3602 rand
= ZIO_CRYPT_ON
;
3604 fnvlist_add_uint64(props
,
3605 zfs_prop_to_name(ZFS_PROP_ENCRYPTION
), rand
);
3606 fnvlist_add_uint8_array(crypto_args
, "wkeydata",
3607 (uint8_t *)ztest_wkeydata
, WRAPPING_KEY_LEN
);
3610 * These parameters aren't really used by the kernel. They
3611 * are simply stored so that userspace knows how to load
3614 fnvlist_add_uint64(props
,
3615 zfs_prop_to_name(ZFS_PROP_KEYFORMAT
), ZFS_KEYFORMAT_RAW
);
3616 fnvlist_add_string(props
,
3617 zfs_prop_to_name(ZFS_PROP_KEYLOCATION
), "prompt");
3618 fnvlist_add_uint64(props
,
3619 zfs_prop_to_name(ZFS_PROP_PBKDF2_SALT
), 0ULL);
3620 fnvlist_add_uint64(props
,
3621 zfs_prop_to_name(ZFS_PROP_PBKDF2_ITERS
), 0ULL);
3623 VERIFY0(dsl_crypto_params_create_nvlist(DCP_CMD_NONE
, props
,
3624 crypto_args
, &dcp
));
3626 fnvlist_free(crypto_args
);
3627 fnvlist_free(props
);
3630 err
= dmu_objset_create(dsname
, DMU_OST_OTHER
, 0, dcp
,
3631 ztest_objset_create_cb
, NULL
);
3632 dsl_crypto_params_free(dcp
, !!err
);
3634 rand
= ztest_random(100);
3635 if (err
|| rand
< 80)
3638 if (ztest_opts
.zo_verbose
>= 5)
3639 (void) printf("Setting dataset %s to sync always\n", dsname
);
3640 return (ztest_dsl_prop_set_uint64(dsname
, ZFS_PROP_SYNC
,
3641 ZFS_SYNC_ALWAYS
, B_FALSE
));
3646 ztest_objset_destroy_cb(const char *name
, void *arg
)
3649 dmu_object_info_t doi
;
3653 * Verify that the dataset contains a directory object.
3655 VERIFY0(ztest_dmu_objset_own(name
, DMU_OST_OTHER
, B_TRUE
,
3656 B_TRUE
, FTAG
, &os
));
3657 error
= dmu_object_info(os
, ZTEST_DIROBJ
, &doi
);
3658 if (error
!= ENOENT
) {
3659 /* We could have crashed in the middle of destroying it */
3661 ASSERT3U(doi
.doi_type
, ==, DMU_OT_ZAP_OTHER
);
3662 ASSERT3S(doi
.doi_physical_blocks_512
, >=, 0);
3664 dmu_objset_disown(os
, B_TRUE
, FTAG
);
3667 * Destroy the dataset.
3669 if (strchr(name
, '@') != NULL
) {
3670 VERIFY0(dsl_destroy_snapshot(name
, B_TRUE
));
3672 error
= dsl_destroy_head(name
);
3673 /* There could be a hold on this dataset */
3681 ztest_snapshot_create(char *osname
, uint64_t id
)
3683 char snapname
[ZFS_MAX_DATASET_NAME_LEN
];
3686 (void) snprintf(snapname
, sizeof (snapname
), "%llu", (u_longlong_t
)id
);
3688 error
= dmu_objset_snapshot_one(osname
, snapname
);
3689 if (error
== ENOSPC
) {
3690 ztest_record_enospc(FTAG
);
3693 if (error
!= 0 && error
!= EEXIST
) {
3694 fatal(0, "ztest_snapshot_create(%s@%s) = %d", osname
,
3701 ztest_snapshot_destroy(char *osname
, uint64_t id
)
3703 char snapname
[ZFS_MAX_DATASET_NAME_LEN
];
3706 (void) snprintf(snapname
, sizeof (snapname
), "%s@%llu", osname
,
3709 error
= dsl_destroy_snapshot(snapname
, B_FALSE
);
3710 if (error
!= 0 && error
!= ENOENT
)
3711 fatal(0, "ztest_snapshot_destroy(%s) = %d", snapname
, error
);
3717 ztest_dmu_objset_create_destroy(ztest_ds_t
*zd
, uint64_t id
)
3723 char name
[ZFS_MAX_DATASET_NAME_LEN
];
3727 zdtmp
= umem_alloc(sizeof (ztest_ds_t
), UMEM_NOFAIL
);
3729 (void) rw_rdlock(&ztest_name_lock
);
3731 (void) snprintf(name
, sizeof (name
), "%s/temp_%llu",
3732 ztest_opts
.zo_pool
, (u_longlong_t
)id
);
3735 * If this dataset exists from a previous run, process its replay log
3736 * half of the time. If we don't replay it, then dsl_destroy_head()
3737 * (invoked from ztest_objset_destroy_cb()) should just throw it away.
3739 if (ztest_random(2) == 0 &&
3740 ztest_dmu_objset_own(name
, DMU_OST_OTHER
, B_FALSE
,
3741 B_TRUE
, FTAG
, &os
) == 0) {
3742 ztest_zd_init(zdtmp
, NULL
, os
);
3743 zil_replay(os
, zdtmp
, ztest_replay_vector
);
3744 ztest_zd_fini(zdtmp
);
3745 txg_wait_synced(dmu_objset_pool(os
), 0);
3746 dmu_objset_disown(os
, B_TRUE
, FTAG
);
3750 * There may be an old instance of the dataset we're about to
3751 * create lying around from a previous run. If so, destroy it
3752 * and all of its snapshots.
3754 (void) dmu_objset_find(name
, ztest_objset_destroy_cb
, NULL
,
3755 DS_FIND_CHILDREN
| DS_FIND_SNAPSHOTS
);
3758 * Verify that the destroyed dataset is no longer in the namespace.
3760 VERIFY3U(ENOENT
, ==, ztest_dmu_objset_own(name
, DMU_OST_OTHER
, B_TRUE
,
3761 B_TRUE
, FTAG
, &os
));
3764 * Verify that we can create a new dataset.
3766 error
= ztest_dataset_create(name
);
3768 if (error
== ENOSPC
) {
3769 ztest_record_enospc(FTAG
);
3772 fatal(0, "dmu_objset_create(%s) = %d", name
, error
);
3775 VERIFY0(ztest_dmu_objset_own(name
, DMU_OST_OTHER
, B_FALSE
, B_TRUE
,
3778 ztest_zd_init(zdtmp
, NULL
, os
);
3781 * Open the intent log for it.
3783 zilog
= zil_open(os
, ztest_get_data
);
3786 * Put some objects in there, do a little I/O to them,
3787 * and randomly take a couple of snapshots along the way.
3789 iters
= ztest_random(5);
3790 for (i
= 0; i
< iters
; i
++) {
3791 ztest_dmu_object_alloc_free(zdtmp
, id
);
3792 if (ztest_random(iters
) == 0)
3793 (void) ztest_snapshot_create(name
, i
);
3797 * Verify that we cannot create an existing dataset.
3799 VERIFY3U(EEXIST
, ==,
3800 dmu_objset_create(name
, DMU_OST_OTHER
, 0, NULL
, NULL
, NULL
));
3803 * Verify that we can hold an objset that is also owned.
3805 VERIFY3U(0, ==, dmu_objset_hold(name
, FTAG
, &os2
));
3806 dmu_objset_rele(os2
, FTAG
);
3809 * Verify that we cannot own an objset that is already owned.
3811 VERIFY3U(EBUSY
, ==, ztest_dmu_objset_own(name
, DMU_OST_OTHER
,
3812 B_FALSE
, B_TRUE
, FTAG
, &os2
));
3815 txg_wait_synced(spa_get_dsl(os
->os_spa
), 0);
3816 dmu_objset_disown(os
, B_TRUE
, FTAG
);
3817 ztest_zd_fini(zdtmp
);
3819 (void) rw_unlock(&ztest_name_lock
);
3821 umem_free(zdtmp
, sizeof (ztest_ds_t
));
3825 * Verify that dmu_snapshot_{create,destroy,open,close} work as expected.
3828 ztest_dmu_snapshot_create_destroy(ztest_ds_t
*zd
, uint64_t id
)
3830 (void) rw_rdlock(&ztest_name_lock
);
3831 (void) ztest_snapshot_destroy(zd
->zd_name
, id
);
3832 (void) ztest_snapshot_create(zd
->zd_name
, id
);
3833 (void) rw_unlock(&ztest_name_lock
);
3837 * Cleanup non-standard snapshots and clones.
3840 ztest_dsl_dataset_cleanup(char *osname
, uint64_t id
)
3849 snap1name
= umem_alloc(ZFS_MAX_DATASET_NAME_LEN
, UMEM_NOFAIL
);
3850 clone1name
= umem_alloc(ZFS_MAX_DATASET_NAME_LEN
, UMEM_NOFAIL
);
3851 snap2name
= umem_alloc(ZFS_MAX_DATASET_NAME_LEN
, UMEM_NOFAIL
);
3852 clone2name
= umem_alloc(ZFS_MAX_DATASET_NAME_LEN
, UMEM_NOFAIL
);
3853 snap3name
= umem_alloc(ZFS_MAX_DATASET_NAME_LEN
, UMEM_NOFAIL
);
3855 (void) snprintf(snap1name
, ZFS_MAX_DATASET_NAME_LEN
,
3856 "%s@s1_%llu", osname
, (u_longlong_t
)id
);
3857 (void) snprintf(clone1name
, ZFS_MAX_DATASET_NAME_LEN
,
3858 "%s/c1_%llu", osname
, (u_longlong_t
)id
);
3859 (void) snprintf(snap2name
, ZFS_MAX_DATASET_NAME_LEN
,
3860 "%s@s2_%llu", clone1name
, (u_longlong_t
)id
);
3861 (void) snprintf(clone2name
, ZFS_MAX_DATASET_NAME_LEN
,
3862 "%s/c2_%llu", osname
, (u_longlong_t
)id
);
3863 (void) snprintf(snap3name
, ZFS_MAX_DATASET_NAME_LEN
,
3864 "%s@s3_%llu", clone1name
, (u_longlong_t
)id
);
3866 error
= dsl_destroy_head(clone2name
);
3867 if (error
&& error
!= ENOENT
)
3868 fatal(0, "dsl_destroy_head(%s) = %d", clone2name
, error
);
3869 error
= dsl_destroy_snapshot(snap3name
, B_FALSE
);
3870 if (error
&& error
!= ENOENT
)
3871 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap3name
, error
);
3872 error
= dsl_destroy_snapshot(snap2name
, B_FALSE
);
3873 if (error
&& error
!= ENOENT
)
3874 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap2name
, error
);
3875 error
= dsl_destroy_head(clone1name
);
3876 if (error
&& error
!= ENOENT
)
3877 fatal(0, "dsl_destroy_head(%s) = %d", clone1name
, error
);
3878 error
= dsl_destroy_snapshot(snap1name
, B_FALSE
);
3879 if (error
&& error
!= ENOENT
)
3880 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap1name
, error
);
3882 umem_free(snap1name
, ZFS_MAX_DATASET_NAME_LEN
);
3883 umem_free(clone1name
, ZFS_MAX_DATASET_NAME_LEN
);
3884 umem_free(snap2name
, ZFS_MAX_DATASET_NAME_LEN
);
3885 umem_free(clone2name
, ZFS_MAX_DATASET_NAME_LEN
);
3886 umem_free(snap3name
, ZFS_MAX_DATASET_NAME_LEN
);
3890 * Verify dsl_dataset_promote handles EBUSY
3893 ztest_dsl_dataset_promote_busy(ztest_ds_t
*zd
, uint64_t id
)
3901 char *osname
= zd
->zd_name
;
3904 snap1name
= umem_alloc(ZFS_MAX_DATASET_NAME_LEN
, UMEM_NOFAIL
);
3905 clone1name
= umem_alloc(ZFS_MAX_DATASET_NAME_LEN
, UMEM_NOFAIL
);
3906 snap2name
= umem_alloc(ZFS_MAX_DATASET_NAME_LEN
, UMEM_NOFAIL
);
3907 clone2name
= umem_alloc(ZFS_MAX_DATASET_NAME_LEN
, UMEM_NOFAIL
);
3908 snap3name
= umem_alloc(ZFS_MAX_DATASET_NAME_LEN
, UMEM_NOFAIL
);
3910 (void) rw_rdlock(&ztest_name_lock
);
3912 ztest_dsl_dataset_cleanup(osname
, id
);
3914 (void) snprintf(snap1name
, ZFS_MAX_DATASET_NAME_LEN
,
3915 "%s@s1_%llu", osname
, (u_longlong_t
)id
);
3916 (void) snprintf(clone1name
, ZFS_MAX_DATASET_NAME_LEN
,
3917 "%s/c1_%llu", osname
, (u_longlong_t
)id
);
3918 (void) snprintf(snap2name
, ZFS_MAX_DATASET_NAME_LEN
,
3919 "%s@s2_%llu", clone1name
, (u_longlong_t
)id
);
3920 (void) snprintf(clone2name
, ZFS_MAX_DATASET_NAME_LEN
,
3921 "%s/c2_%llu", osname
, (u_longlong_t
)id
);
3922 (void) snprintf(snap3name
, ZFS_MAX_DATASET_NAME_LEN
,
3923 "%s@s3_%llu", clone1name
, (u_longlong_t
)id
);
3925 error
= dmu_objset_snapshot_one(osname
, strchr(snap1name
, '@') + 1);
3926 if (error
&& error
!= EEXIST
) {
3927 if (error
== ENOSPC
) {
3928 ztest_record_enospc(FTAG
);
3931 fatal(0, "dmu_take_snapshot(%s) = %d", snap1name
, error
);
3934 error
= dmu_objset_clone(clone1name
, snap1name
);
3936 if (error
== ENOSPC
) {
3937 ztest_record_enospc(FTAG
);
3940 fatal(0, "dmu_objset_create(%s) = %d", clone1name
, error
);
3943 error
= dmu_objset_snapshot_one(clone1name
, strchr(snap2name
, '@') + 1);
3944 if (error
&& error
!= EEXIST
) {
3945 if (error
== ENOSPC
) {
3946 ztest_record_enospc(FTAG
);
3949 fatal(0, "dmu_open_snapshot(%s) = %d", snap2name
, error
);
3952 error
= dmu_objset_snapshot_one(clone1name
, strchr(snap3name
, '@') + 1);
3953 if (error
&& error
!= EEXIST
) {
3954 if (error
== ENOSPC
) {
3955 ztest_record_enospc(FTAG
);
3958 fatal(0, "dmu_open_snapshot(%s) = %d", snap3name
, error
);
3961 error
= dmu_objset_clone(clone2name
, snap3name
);
3963 if (error
== ENOSPC
) {
3964 ztest_record_enospc(FTAG
);
3967 fatal(0, "dmu_objset_create(%s) = %d", clone2name
, error
);
3970 error
= ztest_dmu_objset_own(snap2name
, DMU_OST_ANY
, B_TRUE
, B_TRUE
,
3973 fatal(0, "dmu_objset_own(%s) = %d", snap2name
, error
);
3974 error
= dsl_dataset_promote(clone2name
, NULL
);
3975 if (error
== ENOSPC
) {
3976 dmu_objset_disown(os
, B_TRUE
, FTAG
);
3977 ztest_record_enospc(FTAG
);
3981 fatal(0, "dsl_dataset_promote(%s), %d, not EBUSY", clone2name
,
3983 dmu_objset_disown(os
, B_TRUE
, FTAG
);
3986 ztest_dsl_dataset_cleanup(osname
, id
);
3988 (void) rw_unlock(&ztest_name_lock
);
3990 umem_free(snap1name
, ZFS_MAX_DATASET_NAME_LEN
);
3991 umem_free(clone1name
, ZFS_MAX_DATASET_NAME_LEN
);
3992 umem_free(snap2name
, ZFS_MAX_DATASET_NAME_LEN
);
3993 umem_free(clone2name
, ZFS_MAX_DATASET_NAME_LEN
);
3994 umem_free(snap3name
, ZFS_MAX_DATASET_NAME_LEN
);
3997 #undef OD_ARRAY_SIZE
3998 #define OD_ARRAY_SIZE 4
4001 * Verify that dmu_object_{alloc,free} work as expected.
4004 ztest_dmu_object_alloc_free(ztest_ds_t
*zd
, uint64_t id
)
4011 size
= sizeof (ztest_od_t
) * OD_ARRAY_SIZE
;
4012 od
= umem_alloc(size
, UMEM_NOFAIL
);
4013 batchsize
= OD_ARRAY_SIZE
;
4015 for (b
= 0; b
< batchsize
; b
++)
4016 ztest_od_init(od
+ b
, id
, FTAG
, b
, DMU_OT_UINT64_OTHER
,
4020 * Destroy the previous batch of objects, create a new batch,
4021 * and do some I/O on the new objects.
4023 if (ztest_object_init(zd
, od
, size
, B_TRUE
) != 0)
4026 while (ztest_random(4 * batchsize
) != 0)
4027 ztest_io(zd
, od
[ztest_random(batchsize
)].od_object
,
4028 ztest_random(ZTEST_RANGE_LOCKS
) << SPA_MAXBLOCKSHIFT
);
4030 umem_free(od
, size
);
4033 #undef OD_ARRAY_SIZE
4034 #define OD_ARRAY_SIZE 2
4037 * Verify that dmu_{read,write} work as expected.
4040 ztest_dmu_read_write(ztest_ds_t
*zd
, uint64_t id
)
4045 objset_t
*os
= zd
->zd_os
;
4046 size
= sizeof (ztest_od_t
) * OD_ARRAY_SIZE
;
4047 od
= umem_alloc(size
, UMEM_NOFAIL
);
4049 int i
, freeit
, error
;
4051 bufwad_t
*packbuf
, *bigbuf
, *pack
, *bigH
, *bigT
;
4052 uint64_t packobj
, packoff
, packsize
, bigobj
, bigoff
, bigsize
;
4053 uint64_t chunksize
= (1000 + ztest_random(1000)) * sizeof (uint64_t);
4054 uint64_t regions
= 997;
4055 uint64_t stride
= 123456789ULL;
4056 uint64_t width
= 40;
4057 int free_percent
= 5;
4060 * This test uses two objects, packobj and bigobj, that are always
4061 * updated together (i.e. in the same tx) so that their contents are
4062 * in sync and can be compared. Their contents relate to each other
4063 * in a simple way: packobj is a dense array of 'bufwad' structures,
4064 * while bigobj is a sparse array of the same bufwads. Specifically,
4065 * for any index n, there are three bufwads that should be identical:
4067 * packobj, at offset n * sizeof (bufwad_t)
4068 * bigobj, at the head of the nth chunk
4069 * bigobj, at the tail of the nth chunk
4071 * The chunk size is arbitrary. It doesn't have to be a power of two,
4072 * and it doesn't have any relation to the object blocksize.
4073 * The only requirement is that it can hold at least two bufwads.
4075 * Normally, we write the bufwad to each of these locations.
4076 * However, free_percent of the time we instead write zeroes to
4077 * packobj and perform a dmu_free_range() on bigobj. By comparing
4078 * bigobj to packobj, we can verify that the DMU is correctly
4079 * tracking which parts of an object are allocated and free,
4080 * and that the contents of the allocated blocks are correct.
4084 * Read the directory info. If it's the first time, set things up.
4086 ztest_od_init(od
, id
, FTAG
, 0, DMU_OT_UINT64_OTHER
, 0, 0, chunksize
);
4087 ztest_od_init(od
+ 1, id
, FTAG
, 1, DMU_OT_UINT64_OTHER
, 0, 0,
4090 if (ztest_object_init(zd
, od
, size
, B_FALSE
) != 0) {
4091 umem_free(od
, size
);
4095 bigobj
= od
[0].od_object
;
4096 packobj
= od
[1].od_object
;
4097 chunksize
= od
[0].od_gen
;
4098 ASSERT(chunksize
== od
[1].od_gen
);
4101 * Prefetch a random chunk of the big object.
4102 * Our aim here is to get some async reads in flight
4103 * for blocks that we may free below; the DMU should
4104 * handle this race correctly.
4106 n
= ztest_random(regions
) * stride
+ ztest_random(width
);
4107 s
= 1 + ztest_random(2 * width
- 1);
4108 dmu_prefetch(os
, bigobj
, 0, n
* chunksize
, s
* chunksize
,
4109 ZIO_PRIORITY_SYNC_READ
);
4112 * Pick a random index and compute the offsets into packobj and bigobj.
4114 n
= ztest_random(regions
) * stride
+ ztest_random(width
);
4115 s
= 1 + ztest_random(width
- 1);
4117 packoff
= n
* sizeof (bufwad_t
);
4118 packsize
= s
* sizeof (bufwad_t
);
4120 bigoff
= n
* chunksize
;
4121 bigsize
= s
* chunksize
;
4123 packbuf
= umem_alloc(packsize
, UMEM_NOFAIL
);
4124 bigbuf
= umem_alloc(bigsize
, UMEM_NOFAIL
);
4127 * free_percent of the time, free a range of bigobj rather than
4130 freeit
= (ztest_random(100) < free_percent
);
4133 * Read the current contents of our objects.
4135 error
= dmu_read(os
, packobj
, packoff
, packsize
, packbuf
,
4138 error
= dmu_read(os
, bigobj
, bigoff
, bigsize
, bigbuf
,
4143 * Get a tx for the mods to both packobj and bigobj.
4145 tx
= dmu_tx_create(os
);
4147 dmu_tx_hold_write(tx
, packobj
, packoff
, packsize
);
4150 dmu_tx_hold_free(tx
, bigobj
, bigoff
, bigsize
);
4152 dmu_tx_hold_write(tx
, bigobj
, bigoff
, bigsize
);
4154 /* This accounts for setting the checksum/compression. */
4155 dmu_tx_hold_bonus(tx
, bigobj
);
4157 txg
= ztest_tx_assign(tx
, TXG_MIGHTWAIT
, FTAG
);
4159 umem_free(packbuf
, packsize
);
4160 umem_free(bigbuf
, bigsize
);
4161 umem_free(od
, size
);
4165 enum zio_checksum cksum
;
4167 cksum
= (enum zio_checksum
)
4168 ztest_random_dsl_prop(ZFS_PROP_CHECKSUM
);
4169 } while (cksum
>= ZIO_CHECKSUM_LEGACY_FUNCTIONS
);
4170 dmu_object_set_checksum(os
, bigobj
, cksum
, tx
);
4172 enum zio_compress comp
;
4174 comp
= (enum zio_compress
)
4175 ztest_random_dsl_prop(ZFS_PROP_COMPRESSION
);
4176 } while (comp
>= ZIO_COMPRESS_LEGACY_FUNCTIONS
);
4177 dmu_object_set_compress(os
, bigobj
, comp
, tx
);
4180 * For each index from n to n + s, verify that the existing bufwad
4181 * in packobj matches the bufwads at the head and tail of the
4182 * corresponding chunk in bigobj. Then update all three bufwads
4183 * with the new values we want to write out.
4185 for (i
= 0; i
< s
; i
++) {
4187 pack
= (bufwad_t
*)((char *)packbuf
+ i
* sizeof (bufwad_t
));
4189 bigH
= (bufwad_t
*)((char *)bigbuf
+ i
* chunksize
);
4191 bigT
= (bufwad_t
*)((char *)bigH
+ chunksize
) - 1;
4193 ASSERT((uintptr_t)bigH
- (uintptr_t)bigbuf
< bigsize
);
4194 ASSERT((uintptr_t)bigT
- (uintptr_t)bigbuf
< bigsize
);
4196 if (pack
->bw_txg
> txg
)
4197 fatal(0, "future leak: got %llx, open txg is %llx",
4200 if (pack
->bw_data
!= 0 && pack
->bw_index
!= n
+ i
)
4201 fatal(0, "wrong index: got %llx, wanted %llx+%llx",
4202 pack
->bw_index
, n
, i
);
4204 if (bcmp(pack
, bigH
, sizeof (bufwad_t
)) != 0)
4205 fatal(0, "pack/bigH mismatch in %p/%p", pack
, bigH
);
4207 if (bcmp(pack
, bigT
, sizeof (bufwad_t
)) != 0)
4208 fatal(0, "pack/bigT mismatch in %p/%p", pack
, bigT
);
4211 bzero(pack
, sizeof (bufwad_t
));
4213 pack
->bw_index
= n
+ i
;
4215 pack
->bw_data
= 1 + ztest_random(-2ULL);
4222 * We've verified all the old bufwads, and made new ones.
4223 * Now write them out.
4225 dmu_write(os
, packobj
, packoff
, packsize
, packbuf
, tx
);
4228 if (ztest_opts
.zo_verbose
>= 7) {
4229 (void) printf("freeing offset %llx size %llx"
4231 (u_longlong_t
)bigoff
,
4232 (u_longlong_t
)bigsize
,
4235 VERIFY(0 == dmu_free_range(os
, bigobj
, bigoff
, bigsize
, tx
));
4237 if (ztest_opts
.zo_verbose
>= 7) {
4238 (void) printf("writing offset %llx size %llx"
4240 (u_longlong_t
)bigoff
,
4241 (u_longlong_t
)bigsize
,
4244 dmu_write(os
, bigobj
, bigoff
, bigsize
, bigbuf
, tx
);
4250 * Sanity check the stuff we just wrote.
4253 void *packcheck
= umem_alloc(packsize
, UMEM_NOFAIL
);
4254 void *bigcheck
= umem_alloc(bigsize
, UMEM_NOFAIL
);
4256 VERIFY(0 == dmu_read(os
, packobj
, packoff
,
4257 packsize
, packcheck
, DMU_READ_PREFETCH
));
4258 VERIFY(0 == dmu_read(os
, bigobj
, bigoff
,
4259 bigsize
, bigcheck
, DMU_READ_PREFETCH
));
4261 ASSERT(bcmp(packbuf
, packcheck
, packsize
) == 0);
4262 ASSERT(bcmp(bigbuf
, bigcheck
, bigsize
) == 0);
4264 umem_free(packcheck
, packsize
);
4265 umem_free(bigcheck
, bigsize
);
4268 umem_free(packbuf
, packsize
);
4269 umem_free(bigbuf
, bigsize
);
4270 umem_free(od
, size
);
4274 compare_and_update_pbbufs(uint64_t s
, bufwad_t
*packbuf
, bufwad_t
*bigbuf
,
4275 uint64_t bigsize
, uint64_t n
, uint64_t chunksize
, uint64_t txg
)
4283 * For each index from n to n + s, verify that the existing bufwad
4284 * in packobj matches the bufwads at the head and tail of the
4285 * corresponding chunk in bigobj. Then update all three bufwads
4286 * with the new values we want to write out.
4288 for (i
= 0; i
< s
; i
++) {
4290 pack
= (bufwad_t
*)((char *)packbuf
+ i
* sizeof (bufwad_t
));
4292 bigH
= (bufwad_t
*)((char *)bigbuf
+ i
* chunksize
);
4294 bigT
= (bufwad_t
*)((char *)bigH
+ chunksize
) - 1;
4296 ASSERT((uintptr_t)bigH
- (uintptr_t)bigbuf
< bigsize
);
4297 ASSERT((uintptr_t)bigT
- (uintptr_t)bigbuf
< bigsize
);
4299 if (pack
->bw_txg
> txg
)
4300 fatal(0, "future leak: got %llx, open txg is %llx",
4303 if (pack
->bw_data
!= 0 && pack
->bw_index
!= n
+ i
)
4304 fatal(0, "wrong index: got %llx, wanted %llx+%llx",
4305 pack
->bw_index
, n
, i
);
4307 if (bcmp(pack
, bigH
, sizeof (bufwad_t
)) != 0)
4308 fatal(0, "pack/bigH mismatch in %p/%p", pack
, bigH
);
4310 if (bcmp(pack
, bigT
, sizeof (bufwad_t
)) != 0)
4311 fatal(0, "pack/bigT mismatch in %p/%p", pack
, bigT
);
4313 pack
->bw_index
= n
+ i
;
4315 pack
->bw_data
= 1 + ztest_random(-2ULL);
4322 #undef OD_ARRAY_SIZE
4323 #define OD_ARRAY_SIZE 2
4326 ztest_dmu_read_write_zcopy(ztest_ds_t
*zd
, uint64_t id
)
4328 objset_t
*os
= zd
->zd_os
;
4335 bufwad_t
*packbuf
, *bigbuf
;
4336 uint64_t packobj
, packoff
, packsize
, bigobj
, bigoff
, bigsize
;
4337 uint64_t blocksize
= ztest_random_blocksize();
4338 uint64_t chunksize
= blocksize
;
4339 uint64_t regions
= 997;
4340 uint64_t stride
= 123456789ULL;
4342 dmu_buf_t
*bonus_db
;
4343 arc_buf_t
**bigbuf_arcbufs
;
4344 dmu_object_info_t doi
;
4346 size
= sizeof (ztest_od_t
) * OD_ARRAY_SIZE
;
4347 od
= umem_alloc(size
, UMEM_NOFAIL
);
4350 * This test uses two objects, packobj and bigobj, that are always
4351 * updated together (i.e. in the same tx) so that their contents are
4352 * in sync and can be compared. Their contents relate to each other
4353 * in a simple way: packobj is a dense array of 'bufwad' structures,
4354 * while bigobj is a sparse array of the same bufwads. Specifically,
4355 * for any index n, there are three bufwads that should be identical:
4357 * packobj, at offset n * sizeof (bufwad_t)
4358 * bigobj, at the head of the nth chunk
4359 * bigobj, at the tail of the nth chunk
4361 * The chunk size is set equal to bigobj block size so that
4362 * dmu_assign_arcbuf_by_dbuf() can be tested for object updates.
4366 * Read the directory info. If it's the first time, set things up.
4368 ztest_od_init(od
, id
, FTAG
, 0, DMU_OT_UINT64_OTHER
, blocksize
, 0, 0);
4369 ztest_od_init(od
+ 1, id
, FTAG
, 1, DMU_OT_UINT64_OTHER
, 0, 0,
4373 if (ztest_object_init(zd
, od
, size
, B_FALSE
) != 0) {
4374 umem_free(od
, size
);
4378 bigobj
= od
[0].od_object
;
4379 packobj
= od
[1].od_object
;
4380 blocksize
= od
[0].od_blocksize
;
4381 chunksize
= blocksize
;
4382 ASSERT(chunksize
== od
[1].od_gen
);
4384 VERIFY(dmu_object_info(os
, bigobj
, &doi
) == 0);
4385 VERIFY(ISP2(doi
.doi_data_block_size
));
4386 VERIFY(chunksize
== doi
.doi_data_block_size
);
4387 VERIFY(chunksize
>= 2 * sizeof (bufwad_t
));
4390 * Pick a random index and compute the offsets into packobj and bigobj.
4392 n
= ztest_random(regions
) * stride
+ ztest_random(width
);
4393 s
= 1 + ztest_random(width
- 1);
4395 packoff
= n
* sizeof (bufwad_t
);
4396 packsize
= s
* sizeof (bufwad_t
);
4398 bigoff
= n
* chunksize
;
4399 bigsize
= s
* chunksize
;
4401 packbuf
= umem_zalloc(packsize
, UMEM_NOFAIL
);
4402 bigbuf
= umem_zalloc(bigsize
, UMEM_NOFAIL
);
4404 VERIFY3U(0, ==, dmu_bonus_hold(os
, bigobj
, FTAG
, &bonus_db
));
4406 bigbuf_arcbufs
= umem_zalloc(2 * s
* sizeof (arc_buf_t
*), UMEM_NOFAIL
);
4409 * Iteration 0 test zcopy for DB_UNCACHED dbufs.
4410 * Iteration 1 test zcopy to already referenced dbufs.
4411 * Iteration 2 test zcopy to dirty dbuf in the same txg.
4412 * Iteration 3 test zcopy to dbuf dirty in previous txg.
4413 * Iteration 4 test zcopy when dbuf is no longer dirty.
4414 * Iteration 5 test zcopy when it can't be done.
4415 * Iteration 6 one more zcopy write.
4417 for (i
= 0; i
< 7; i
++) {
4422 * In iteration 5 (i == 5) use arcbufs
4423 * that don't match bigobj blksz to test
4424 * dmu_assign_arcbuf_by_dbuf() when it can't directly
4425 * assign an arcbuf to a dbuf.
4427 for (j
= 0; j
< s
; j
++) {
4428 if (i
!= 5 || chunksize
< (SPA_MINBLOCKSIZE
* 2)) {
4430 dmu_request_arcbuf(bonus_db
, chunksize
);
4432 bigbuf_arcbufs
[2 * j
] =
4433 dmu_request_arcbuf(bonus_db
, chunksize
/ 2);
4434 bigbuf_arcbufs
[2 * j
+ 1] =
4435 dmu_request_arcbuf(bonus_db
, chunksize
/ 2);
4440 * Get a tx for the mods to both packobj and bigobj.
4442 tx
= dmu_tx_create(os
);
4444 dmu_tx_hold_write(tx
, packobj
, packoff
, packsize
);
4445 dmu_tx_hold_write(tx
, bigobj
, bigoff
, bigsize
);
4447 txg
= ztest_tx_assign(tx
, TXG_MIGHTWAIT
, FTAG
);
4449 umem_free(packbuf
, packsize
);
4450 umem_free(bigbuf
, bigsize
);
4451 for (j
= 0; j
< s
; j
++) {
4453 chunksize
< (SPA_MINBLOCKSIZE
* 2)) {
4454 dmu_return_arcbuf(bigbuf_arcbufs
[j
]);
4457 bigbuf_arcbufs
[2 * j
]);
4459 bigbuf_arcbufs
[2 * j
+ 1]);
4462 umem_free(bigbuf_arcbufs
, 2 * s
* sizeof (arc_buf_t
*));
4463 umem_free(od
, size
);
4464 dmu_buf_rele(bonus_db
, FTAG
);
4469 * 50% of the time don't read objects in the 1st iteration to
4470 * test dmu_assign_arcbuf_by_dbuf() for the case when there are
4471 * no existing dbufs for the specified offsets.
4473 if (i
!= 0 || ztest_random(2) != 0) {
4474 error
= dmu_read(os
, packobj
, packoff
,
4475 packsize
, packbuf
, DMU_READ_PREFETCH
);
4477 error
= dmu_read(os
, bigobj
, bigoff
, bigsize
,
4478 bigbuf
, DMU_READ_PREFETCH
);
4481 compare_and_update_pbbufs(s
, packbuf
, bigbuf
, bigsize
,
4485 * We've verified all the old bufwads, and made new ones.
4486 * Now write them out.
4488 dmu_write(os
, packobj
, packoff
, packsize
, packbuf
, tx
);
4489 if (ztest_opts
.zo_verbose
>= 7) {
4490 (void) printf("writing offset %llx size %llx"
4492 (u_longlong_t
)bigoff
,
4493 (u_longlong_t
)bigsize
,
4496 for (off
= bigoff
, j
= 0; j
< s
; j
++, off
+= chunksize
) {
4498 if (i
!= 5 || chunksize
< (SPA_MINBLOCKSIZE
* 2)) {
4499 bcopy((caddr_t
)bigbuf
+ (off
- bigoff
),
4500 bigbuf_arcbufs
[j
]->b_data
, chunksize
);
4502 bcopy((caddr_t
)bigbuf
+ (off
- bigoff
),
4503 bigbuf_arcbufs
[2 * j
]->b_data
,
4505 bcopy((caddr_t
)bigbuf
+ (off
- bigoff
) +
4507 bigbuf_arcbufs
[2 * j
+ 1]->b_data
,
4512 VERIFY(dmu_buf_hold(os
, bigobj
, off
,
4513 FTAG
, &dbt
, DMU_READ_NO_PREFETCH
) == 0);
4515 if (i
!= 5 || chunksize
< (SPA_MINBLOCKSIZE
* 2)) {
4516 dmu_assign_arcbuf_by_dbuf(bonus_db
, off
,
4517 bigbuf_arcbufs
[j
], tx
);
4519 dmu_assign_arcbuf_by_dbuf(bonus_db
, off
,
4520 bigbuf_arcbufs
[2 * j
], tx
);
4521 dmu_assign_arcbuf_by_dbuf(bonus_db
,
4522 off
+ chunksize
/ 2,
4523 bigbuf_arcbufs
[2 * j
+ 1], tx
);
4526 dmu_buf_rele(dbt
, FTAG
);
4532 * Sanity check the stuff we just wrote.
4535 void *packcheck
= umem_alloc(packsize
, UMEM_NOFAIL
);
4536 void *bigcheck
= umem_alloc(bigsize
, UMEM_NOFAIL
);
4538 VERIFY(0 == dmu_read(os
, packobj
, packoff
,
4539 packsize
, packcheck
, DMU_READ_PREFETCH
));
4540 VERIFY(0 == dmu_read(os
, bigobj
, bigoff
,
4541 bigsize
, bigcheck
, DMU_READ_PREFETCH
));
4543 ASSERT(bcmp(packbuf
, packcheck
, packsize
) == 0);
4544 ASSERT(bcmp(bigbuf
, bigcheck
, bigsize
) == 0);
4546 umem_free(packcheck
, packsize
);
4547 umem_free(bigcheck
, bigsize
);
4550 txg_wait_open(dmu_objset_pool(os
), 0);
4551 } else if (i
== 3) {
4552 txg_wait_synced(dmu_objset_pool(os
), 0);
4556 dmu_buf_rele(bonus_db
, FTAG
);
4557 umem_free(packbuf
, packsize
);
4558 umem_free(bigbuf
, bigsize
);
4559 umem_free(bigbuf_arcbufs
, 2 * s
* sizeof (arc_buf_t
*));
4560 umem_free(od
, size
);
4565 ztest_dmu_write_parallel(ztest_ds_t
*zd
, uint64_t id
)
4569 od
= umem_alloc(sizeof (ztest_od_t
), UMEM_NOFAIL
);
4570 uint64_t offset
= (1ULL << (ztest_random(20) + 43)) +
4571 (ztest_random(ZTEST_RANGE_LOCKS
) << SPA_MAXBLOCKSHIFT
);
4574 * Have multiple threads write to large offsets in an object
4575 * to verify that parallel writes to an object -- even to the
4576 * same blocks within the object -- doesn't cause any trouble.
4578 ztest_od_init(od
, ID_PARALLEL
, FTAG
, 0, DMU_OT_UINT64_OTHER
, 0, 0, 0);
4580 if (ztest_object_init(zd
, od
, sizeof (ztest_od_t
), B_FALSE
) != 0)
4583 while (ztest_random(10) != 0)
4584 ztest_io(zd
, od
->od_object
, offset
);
4586 umem_free(od
, sizeof (ztest_od_t
));
4590 ztest_dmu_prealloc(ztest_ds_t
*zd
, uint64_t id
)
4593 uint64_t offset
= (1ULL << (ztest_random(4) + SPA_MAXBLOCKSHIFT
)) +
4594 (ztest_random(ZTEST_RANGE_LOCKS
) << SPA_MAXBLOCKSHIFT
);
4595 uint64_t count
= ztest_random(20) + 1;
4596 uint64_t blocksize
= ztest_random_blocksize();
4599 od
= umem_alloc(sizeof (ztest_od_t
), UMEM_NOFAIL
);
4601 ztest_od_init(od
, id
, FTAG
, 0, DMU_OT_UINT64_OTHER
, blocksize
, 0, 0);
4603 if (ztest_object_init(zd
, od
, sizeof (ztest_od_t
),
4604 !ztest_random(2)) != 0) {
4605 umem_free(od
, sizeof (ztest_od_t
));
4609 if (ztest_truncate(zd
, od
->od_object
, offset
, count
* blocksize
) != 0) {
4610 umem_free(od
, sizeof (ztest_od_t
));
4614 ztest_prealloc(zd
, od
->od_object
, offset
, count
* blocksize
);
4616 data
= umem_zalloc(blocksize
, UMEM_NOFAIL
);
4618 while (ztest_random(count
) != 0) {
4619 uint64_t randoff
= offset
+ (ztest_random(count
) * blocksize
);
4620 if (ztest_write(zd
, od
->od_object
, randoff
, blocksize
,
4623 while (ztest_random(4) != 0)
4624 ztest_io(zd
, od
->od_object
, randoff
);
4627 umem_free(data
, blocksize
);
4628 umem_free(od
, sizeof (ztest_od_t
));
4632 * Verify that zap_{create,destroy,add,remove,update} work as expected.
4634 #define ZTEST_ZAP_MIN_INTS 1
4635 #define ZTEST_ZAP_MAX_INTS 4
4636 #define ZTEST_ZAP_MAX_PROPS 1000
4639 ztest_zap(ztest_ds_t
*zd
, uint64_t id
)
4641 objset_t
*os
= zd
->zd_os
;
4644 uint64_t txg
, last_txg
;
4645 uint64_t value
[ZTEST_ZAP_MAX_INTS
];
4646 uint64_t zl_ints
, zl_intsize
, prop
;
4649 char propname
[100], txgname
[100];
4651 char *hc
[2] = { "s.acl.h", ".s.open.h.hyLZlg" };
4653 od
= umem_alloc(sizeof (ztest_od_t
), UMEM_NOFAIL
);
4654 ztest_od_init(od
, id
, FTAG
, 0, DMU_OT_ZAP_OTHER
, 0, 0, 0);
4656 if (ztest_object_init(zd
, od
, sizeof (ztest_od_t
),
4657 !ztest_random(2)) != 0)
4660 object
= od
->od_object
;
4663 * Generate a known hash collision, and verify that
4664 * we can lookup and remove both entries.
4666 tx
= dmu_tx_create(os
);
4667 dmu_tx_hold_zap(tx
, object
, B_TRUE
, NULL
);
4668 txg
= ztest_tx_assign(tx
, TXG_MIGHTWAIT
, FTAG
);
4671 for (i
= 0; i
< 2; i
++) {
4673 VERIFY3U(0, ==, zap_add(os
, object
, hc
[i
], sizeof (uint64_t),
4676 for (i
= 0; i
< 2; i
++) {
4677 VERIFY3U(EEXIST
, ==, zap_add(os
, object
, hc
[i
],
4678 sizeof (uint64_t), 1, &value
[i
], tx
));
4680 zap_length(os
, object
, hc
[i
], &zl_intsize
, &zl_ints
));
4681 ASSERT3U(zl_intsize
, ==, sizeof (uint64_t));
4682 ASSERT3U(zl_ints
, ==, 1);
4684 for (i
= 0; i
< 2; i
++) {
4685 VERIFY3U(0, ==, zap_remove(os
, object
, hc
[i
], tx
));
4690 * Generate a buch of random entries.
4692 ints
= MAX(ZTEST_ZAP_MIN_INTS
, object
% ZTEST_ZAP_MAX_INTS
);
4694 prop
= ztest_random(ZTEST_ZAP_MAX_PROPS
);
4695 (void) sprintf(propname
, "prop_%llu", (u_longlong_t
)prop
);
4696 (void) sprintf(txgname
, "txg_%llu", (u_longlong_t
)prop
);
4697 bzero(value
, sizeof (value
));
4701 * If these zap entries already exist, validate their contents.
4703 error
= zap_length(os
, object
, txgname
, &zl_intsize
, &zl_ints
);
4705 ASSERT3U(zl_intsize
, ==, sizeof (uint64_t));
4706 ASSERT3U(zl_ints
, ==, 1);
4708 VERIFY(zap_lookup(os
, object
, txgname
, zl_intsize
,
4709 zl_ints
, &last_txg
) == 0);
4711 VERIFY(zap_length(os
, object
, propname
, &zl_intsize
,
4714 ASSERT3U(zl_intsize
, ==, sizeof (uint64_t));
4715 ASSERT3U(zl_ints
, ==, ints
);
4717 VERIFY(zap_lookup(os
, object
, propname
, zl_intsize
,
4718 zl_ints
, value
) == 0);
4720 for (i
= 0; i
< ints
; i
++) {
4721 ASSERT3U(value
[i
], ==, last_txg
+ object
+ i
);
4724 ASSERT3U(error
, ==, ENOENT
);
4728 * Atomically update two entries in our zap object.
4729 * The first is named txg_%llu, and contains the txg
4730 * in which the property was last updated. The second
4731 * is named prop_%llu, and the nth element of its value
4732 * should be txg + object + n.
4734 tx
= dmu_tx_create(os
);
4735 dmu_tx_hold_zap(tx
, object
, B_TRUE
, NULL
);
4736 txg
= ztest_tx_assign(tx
, TXG_MIGHTWAIT
, FTAG
);
4741 fatal(0, "zap future leak: old %llu new %llu", last_txg
, txg
);
4743 for (i
= 0; i
< ints
; i
++)
4744 value
[i
] = txg
+ object
+ i
;
4746 VERIFY3U(0, ==, zap_update(os
, object
, txgname
, sizeof (uint64_t),
4748 VERIFY3U(0, ==, zap_update(os
, object
, propname
, sizeof (uint64_t),
4754 * Remove a random pair of entries.
4756 prop
= ztest_random(ZTEST_ZAP_MAX_PROPS
);
4757 (void) sprintf(propname
, "prop_%llu", (u_longlong_t
)prop
);
4758 (void) sprintf(txgname
, "txg_%llu", (u_longlong_t
)prop
);
4760 error
= zap_length(os
, object
, txgname
, &zl_intsize
, &zl_ints
);
4762 if (error
== ENOENT
)
4767 tx
= dmu_tx_create(os
);
4768 dmu_tx_hold_zap(tx
, object
, B_TRUE
, NULL
);
4769 txg
= ztest_tx_assign(tx
, TXG_MIGHTWAIT
, FTAG
);
4772 VERIFY3U(0, ==, zap_remove(os
, object
, txgname
, tx
));
4773 VERIFY3U(0, ==, zap_remove(os
, object
, propname
, tx
));
4776 umem_free(od
, sizeof (ztest_od_t
));
4780 * Testcase to test the upgrading of a microzap to fatzap.
4783 ztest_fzap(ztest_ds_t
*zd
, uint64_t id
)
4785 objset_t
*os
= zd
->zd_os
;
4787 uint64_t object
, txg
;
4790 od
= umem_alloc(sizeof (ztest_od_t
), UMEM_NOFAIL
);
4791 ztest_od_init(od
, id
, FTAG
, 0, DMU_OT_ZAP_OTHER
, 0, 0, 0);
4793 if (ztest_object_init(zd
, od
, sizeof (ztest_od_t
),
4794 !ztest_random(2)) != 0)
4796 object
= od
->od_object
;
4799 * Add entries to this ZAP and make sure it spills over
4800 * and gets upgraded to a fatzap. Also, since we are adding
4801 * 2050 entries we should see ptrtbl growth and leaf-block split.
4803 for (i
= 0; i
< 2050; i
++) {
4804 char name
[ZFS_MAX_DATASET_NAME_LEN
];
4809 (void) snprintf(name
, sizeof (name
), "fzap-%llu-%llu",
4810 (u_longlong_t
)id
, (u_longlong_t
)value
);
4812 tx
= dmu_tx_create(os
);
4813 dmu_tx_hold_zap(tx
, object
, B_TRUE
, name
);
4814 txg
= ztest_tx_assign(tx
, TXG_MIGHTWAIT
, FTAG
);
4817 error
= zap_add(os
, object
, name
, sizeof (uint64_t), 1,
4819 ASSERT(error
== 0 || error
== EEXIST
);
4823 umem_free(od
, sizeof (ztest_od_t
));
4828 ztest_zap_parallel(ztest_ds_t
*zd
, uint64_t id
)
4830 objset_t
*os
= zd
->zd_os
;
4832 uint64_t txg
, object
, count
, wsize
, wc
, zl_wsize
, zl_wc
;
4834 int i
, namelen
, error
;
4835 int micro
= ztest_random(2);
4836 char name
[20], string_value
[20];
4839 od
= umem_alloc(sizeof (ztest_od_t
), UMEM_NOFAIL
);
4840 ztest_od_init(od
, ID_PARALLEL
, FTAG
, micro
, DMU_OT_ZAP_OTHER
, 0, 0, 0);
4842 if (ztest_object_init(zd
, od
, sizeof (ztest_od_t
), B_FALSE
) != 0) {
4843 umem_free(od
, sizeof (ztest_od_t
));
4847 object
= od
->od_object
;
4850 * Generate a random name of the form 'xxx.....' where each
4851 * x is a random printable character and the dots are dots.
4852 * There are 94 such characters, and the name length goes from
4853 * 6 to 20, so there are 94^3 * 15 = 12,458,760 possible names.
4855 namelen
= ztest_random(sizeof (name
) - 5) + 5 + 1;
4857 for (i
= 0; i
< 3; i
++)
4858 name
[i
] = '!' + ztest_random('~' - '!' + 1);
4859 for (; i
< namelen
- 1; i
++)
4863 if ((namelen
& 1) || micro
) {
4864 wsize
= sizeof (txg
);
4870 data
= string_value
;
4874 VERIFY0(zap_count(os
, object
, &count
));
4875 ASSERT(count
!= -1ULL);
4878 * Select an operation: length, lookup, add, update, remove.
4880 i
= ztest_random(5);
4883 tx
= dmu_tx_create(os
);
4884 dmu_tx_hold_zap(tx
, object
, B_TRUE
, NULL
);
4885 txg
= ztest_tx_assign(tx
, TXG_MIGHTWAIT
, FTAG
);
4887 umem_free(od
, sizeof (ztest_od_t
));
4890 bcopy(name
, string_value
, namelen
);
4894 bzero(string_value
, namelen
);
4900 error
= zap_length(os
, object
, name
, &zl_wsize
, &zl_wc
);
4902 ASSERT3U(wsize
, ==, zl_wsize
);
4903 ASSERT3U(wc
, ==, zl_wc
);
4905 ASSERT3U(error
, ==, ENOENT
);
4910 error
= zap_lookup(os
, object
, name
, wsize
, wc
, data
);
4912 if (data
== string_value
&&
4913 bcmp(name
, data
, namelen
) != 0)
4914 fatal(0, "name '%s' != val '%s' len %d",
4915 name
, data
, namelen
);
4917 ASSERT3U(error
, ==, ENOENT
);
4922 error
= zap_add(os
, object
, name
, wsize
, wc
, data
, tx
);
4923 ASSERT(error
== 0 || error
== EEXIST
);
4927 VERIFY(zap_update(os
, object
, name
, wsize
, wc
, data
, tx
) == 0);
4931 error
= zap_remove(os
, object
, name
, tx
);
4932 ASSERT(error
== 0 || error
== ENOENT
);
4939 umem_free(od
, sizeof (ztest_od_t
));
4943 * Commit callback data.
4945 typedef struct ztest_cb_data
{
4946 list_node_t zcd_node
;
4948 int zcd_expected_err
;
4949 boolean_t zcd_added
;
4950 boolean_t zcd_called
;
4954 /* This is the actual commit callback function */
4956 ztest_commit_callback(void *arg
, int error
)
4958 ztest_cb_data_t
*data
= arg
;
4959 uint64_t synced_txg
;
4961 VERIFY(data
!= NULL
);
4962 VERIFY3S(data
->zcd_expected_err
, ==, error
);
4963 VERIFY(!data
->zcd_called
);
4965 synced_txg
= spa_last_synced_txg(data
->zcd_spa
);
4966 if (data
->zcd_txg
> synced_txg
)
4967 fatal(0, "commit callback of txg %" PRIu64
" called prematurely"
4968 ", last synced txg = %" PRIu64
"\n", data
->zcd_txg
,
4971 data
->zcd_called
= B_TRUE
;
4973 if (error
== ECANCELED
) {
4974 ASSERT0(data
->zcd_txg
);
4975 ASSERT(!data
->zcd_added
);
4978 * The private callback data should be destroyed here, but
4979 * since we are going to check the zcd_called field after
4980 * dmu_tx_abort(), we will destroy it there.
4985 ASSERT(data
->zcd_added
);
4986 ASSERT3U(data
->zcd_txg
, !=, 0);
4988 (void) mutex_enter(&zcl
.zcl_callbacks_lock
);
4990 /* See if this cb was called more quickly */
4991 if ((synced_txg
- data
->zcd_txg
) < zc_min_txg_delay
)
4992 zc_min_txg_delay
= synced_txg
- data
->zcd_txg
;
4994 /* Remove our callback from the list */
4995 list_remove(&zcl
.zcl_callbacks
, data
);
4997 (void) mutex_exit(&zcl
.zcl_callbacks_lock
);
4999 umem_free(data
, sizeof (ztest_cb_data_t
));
5002 /* Allocate and initialize callback data structure */
5003 static ztest_cb_data_t
*
5004 ztest_create_cb_data(objset_t
*os
, uint64_t txg
)
5006 ztest_cb_data_t
*cb_data
;
5008 cb_data
= umem_zalloc(sizeof (ztest_cb_data_t
), UMEM_NOFAIL
);
5010 cb_data
->zcd_txg
= txg
;
5011 cb_data
->zcd_spa
= dmu_objset_spa(os
);
5012 list_link_init(&cb_data
->zcd_node
);
5018 * Commit callback test.
5021 ztest_dmu_commit_callbacks(ztest_ds_t
*zd
, uint64_t id
)
5023 objset_t
*os
= zd
->zd_os
;
5026 ztest_cb_data_t
*cb_data
[3], *tmp_cb
;
5027 uint64_t old_txg
, txg
;
5030 od
= umem_alloc(sizeof (ztest_od_t
), UMEM_NOFAIL
);
5031 ztest_od_init(od
, id
, FTAG
, 0, DMU_OT_UINT64_OTHER
, 0, 0, 0);
5033 if (ztest_object_init(zd
, od
, sizeof (ztest_od_t
), B_FALSE
) != 0) {
5034 umem_free(od
, sizeof (ztest_od_t
));
5038 tx
= dmu_tx_create(os
);
5040 cb_data
[0] = ztest_create_cb_data(os
, 0);
5041 dmu_tx_callback_register(tx
, ztest_commit_callback
, cb_data
[0]);
5043 dmu_tx_hold_write(tx
, od
->od_object
, 0, sizeof (uint64_t));
5045 /* Every once in a while, abort the transaction on purpose */
5046 if (ztest_random(100) == 0)
5050 error
= dmu_tx_assign(tx
, TXG_NOWAIT
);
5052 txg
= error
? 0 : dmu_tx_get_txg(tx
);
5054 cb_data
[0]->zcd_txg
= txg
;
5055 cb_data
[1] = ztest_create_cb_data(os
, txg
);
5056 dmu_tx_callback_register(tx
, ztest_commit_callback
, cb_data
[1]);
5060 * It's not a strict requirement to call the registered
5061 * callbacks from inside dmu_tx_abort(), but that's what
5062 * it's supposed to happen in the current implementation
5063 * so we will check for that.
5065 for (i
= 0; i
< 2; i
++) {
5066 cb_data
[i
]->zcd_expected_err
= ECANCELED
;
5067 VERIFY(!cb_data
[i
]->zcd_called
);
5072 for (i
= 0; i
< 2; i
++) {
5073 VERIFY(cb_data
[i
]->zcd_called
);
5074 umem_free(cb_data
[i
], sizeof (ztest_cb_data_t
));
5077 umem_free(od
, sizeof (ztest_od_t
));
5081 cb_data
[2] = ztest_create_cb_data(os
, txg
);
5082 dmu_tx_callback_register(tx
, ztest_commit_callback
, cb_data
[2]);
5085 * Read existing data to make sure there isn't a future leak.
5087 VERIFY(0 == dmu_read(os
, od
->od_object
, 0, sizeof (uint64_t),
5088 &old_txg
, DMU_READ_PREFETCH
));
5091 fatal(0, "future leak: got %" PRIu64
", open txg is %" PRIu64
,
5094 dmu_write(os
, od
->od_object
, 0, sizeof (uint64_t), &txg
, tx
);
5096 (void) mutex_enter(&zcl
.zcl_callbacks_lock
);
5099 * Since commit callbacks don't have any ordering requirement and since
5100 * it is theoretically possible for a commit callback to be called
5101 * after an arbitrary amount of time has elapsed since its txg has been
5102 * synced, it is difficult to reliably determine whether a commit
5103 * callback hasn't been called due to high load or due to a flawed
5106 * In practice, we will assume that if after a certain number of txgs a
5107 * commit callback hasn't been called, then most likely there's an
5108 * implementation bug..
5110 tmp_cb
= list_head(&zcl
.zcl_callbacks
);
5111 if (tmp_cb
!= NULL
&&
5112 tmp_cb
->zcd_txg
+ ZTEST_COMMIT_CB_THRESH
< txg
) {
5113 fatal(0, "Commit callback threshold exceeded, oldest txg: %"
5114 PRIu64
", open txg: %" PRIu64
"\n", tmp_cb
->zcd_txg
, txg
);
5118 * Let's find the place to insert our callbacks.
5120 * Even though the list is ordered by txg, it is possible for the
5121 * insertion point to not be the end because our txg may already be
5122 * quiescing at this point and other callbacks in the open txg
5123 * (from other objsets) may have sneaked in.
5125 tmp_cb
= list_tail(&zcl
.zcl_callbacks
);
5126 while (tmp_cb
!= NULL
&& tmp_cb
->zcd_txg
> txg
)
5127 tmp_cb
= list_prev(&zcl
.zcl_callbacks
, tmp_cb
);
5129 /* Add the 3 callbacks to the list */
5130 for (i
= 0; i
< 3; i
++) {
5132 list_insert_head(&zcl
.zcl_callbacks
, cb_data
[i
]);
5134 list_insert_after(&zcl
.zcl_callbacks
, tmp_cb
,
5137 cb_data
[i
]->zcd_added
= B_TRUE
;
5138 VERIFY(!cb_data
[i
]->zcd_called
);
5140 tmp_cb
= cb_data
[i
];
5145 (void) mutex_exit(&zcl
.zcl_callbacks_lock
);
5149 umem_free(od
, sizeof (ztest_od_t
));
5153 * Visit each object in the dataset. Verify that its properties
5154 * are consistent what was stored in the block tag when it was created,
5155 * and that its unused bonus buffer space has not been overwritten.
5159 ztest_verify_dnode_bt(ztest_ds_t
*zd
, uint64_t id
)
5161 objset_t
*os
= zd
->zd_os
;
5165 for (obj
= 0; err
== 0; err
= dmu_object_next(os
, &obj
, FALSE
, 0)) {
5166 ztest_block_tag_t
*bt
= NULL
;
5167 dmu_object_info_t doi
;
5170 if (dmu_bonus_hold(os
, obj
, FTAG
, &db
) != 0)
5173 dmu_object_info_from_db(db
, &doi
);
5174 if (doi
.doi_bonus_size
>= sizeof (*bt
))
5175 bt
= ztest_bt_bonus(db
);
5177 if (bt
&& bt
->bt_magic
== BT_MAGIC
) {
5178 ztest_bt_verify(bt
, os
, obj
, doi
.doi_dnodesize
,
5179 bt
->bt_offset
, bt
->bt_gen
, bt
->bt_txg
,
5181 ztest_verify_unused_bonus(db
, bt
, obj
, os
, bt
->bt_gen
);
5184 dmu_buf_rele(db
, FTAG
);
5190 ztest_dsl_prop_get_set(ztest_ds_t
*zd
, uint64_t id
)
5192 zfs_prop_t proplist
[] = {
5194 ZFS_PROP_COMPRESSION
,
5200 (void) rw_rdlock(&ztest_name_lock
);
5202 for (p
= 0; p
< sizeof (proplist
) / sizeof (proplist
[0]); p
++)
5203 (void) ztest_dsl_prop_set_uint64(zd
->zd_name
, proplist
[p
],
5204 ztest_random_dsl_prop(proplist
[p
]), (int)ztest_random(2));
5206 VERIFY0(ztest_dsl_prop_set_uint64(zd
->zd_name
, ZFS_PROP_RECORDSIZE
,
5207 ztest_random_blocksize(), (int)ztest_random(2)));
5209 (void) rw_unlock(&ztest_name_lock
);
5214 ztest_spa_prop_get_set(ztest_ds_t
*zd
, uint64_t id
)
5216 nvlist_t
*props
= NULL
;
5218 (void) rw_rdlock(&ztest_name_lock
);
5220 (void) ztest_spa_prop_set_uint64(ZPOOL_PROP_DEDUPDITTO
,
5221 ZIO_DEDUPDITTO_MIN
+ ztest_random(ZIO_DEDUPDITTO_MIN
));
5223 VERIFY0(spa_prop_get(ztest_spa
, &props
));
5225 if (ztest_opts
.zo_verbose
>= 6)
5226 dump_nvlist(props
, 4);
5230 (void) rw_unlock(&ztest_name_lock
);
5234 user_release_one(const char *snapname
, const char *holdname
)
5236 nvlist_t
*snaps
, *holds
;
5239 snaps
= fnvlist_alloc();
5240 holds
= fnvlist_alloc();
5241 fnvlist_add_boolean(holds
, holdname
);
5242 fnvlist_add_nvlist(snaps
, snapname
, holds
);
5243 fnvlist_free(holds
);
5244 error
= dsl_dataset_user_release(snaps
, NULL
);
5245 fnvlist_free(snaps
);
5250 * Test snapshot hold/release and deferred destroy.
5253 ztest_dmu_snapshot_hold(ztest_ds_t
*zd
, uint64_t id
)
5256 objset_t
*os
= zd
->zd_os
;
5260 char clonename
[100];
5262 char osname
[ZFS_MAX_DATASET_NAME_LEN
];
5265 (void) rw_rdlock(&ztest_name_lock
);
5267 dmu_objset_name(os
, osname
);
5269 (void) snprintf(snapname
, sizeof (snapname
), "sh1_%llu",
5271 (void) snprintf(fullname
, sizeof (fullname
), "%s@%s", osname
, snapname
);
5272 (void) snprintf(clonename
, sizeof (clonename
),
5273 "%s/ch1_%llu", osname
, (u_longlong_t
)id
);
5274 (void) snprintf(tag
, sizeof (tag
), "tag_%llu", (u_longlong_t
)id
);
5277 * Clean up from any previous run.
5279 error
= dsl_destroy_head(clonename
);
5280 if (error
!= ENOENT
)
5282 error
= user_release_one(fullname
, tag
);
5283 if (error
!= ESRCH
&& error
!= ENOENT
)
5285 error
= dsl_destroy_snapshot(fullname
, B_FALSE
);
5286 if (error
!= ENOENT
)
5290 * Create snapshot, clone it, mark snap for deferred destroy,
5291 * destroy clone, verify snap was also destroyed.
5293 error
= dmu_objset_snapshot_one(osname
, snapname
);
5295 if (error
== ENOSPC
) {
5296 ztest_record_enospc("dmu_objset_snapshot");
5299 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname
, error
);
5302 error
= dmu_objset_clone(clonename
, fullname
);
5304 if (error
== ENOSPC
) {
5305 ztest_record_enospc("dmu_objset_clone");
5308 fatal(0, "dmu_objset_clone(%s) = %d", clonename
, error
);
5311 error
= dsl_destroy_snapshot(fullname
, B_TRUE
);
5313 fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
5317 error
= dsl_destroy_head(clonename
);
5319 fatal(0, "dsl_destroy_head(%s) = %d", clonename
, error
);
5321 error
= dmu_objset_hold(fullname
, FTAG
, &origin
);
5322 if (error
!= ENOENT
)
5323 fatal(0, "dmu_objset_hold(%s) = %d", fullname
, error
);
5326 * Create snapshot, add temporary hold, verify that we can't
5327 * destroy a held snapshot, mark for deferred destroy,
5328 * release hold, verify snapshot was destroyed.
5330 error
= dmu_objset_snapshot_one(osname
, snapname
);
5332 if (error
== ENOSPC
) {
5333 ztest_record_enospc("dmu_objset_snapshot");
5336 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname
, error
);
5339 holds
= fnvlist_alloc();
5340 fnvlist_add_string(holds
, fullname
, tag
);
5341 error
= dsl_dataset_user_hold(holds
, 0, NULL
);
5342 fnvlist_free(holds
);
5344 if (error
== ENOSPC
) {
5345 ztest_record_enospc("dsl_dataset_user_hold");
5348 fatal(0, "dsl_dataset_user_hold(%s, %s) = %u",
5349 fullname
, tag
, error
);
5352 error
= dsl_destroy_snapshot(fullname
, B_FALSE
);
5353 if (error
!= EBUSY
) {
5354 fatal(0, "dsl_destroy_snapshot(%s, B_FALSE) = %d",
5358 error
= dsl_destroy_snapshot(fullname
, B_TRUE
);
5360 fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
5364 error
= user_release_one(fullname
, tag
);
5366 fatal(0, "user_release_one(%s, %s) = %d", fullname
, tag
, error
);
5368 VERIFY3U(dmu_objset_hold(fullname
, FTAG
, &origin
), ==, ENOENT
);
5371 (void) rw_unlock(&ztest_name_lock
);
5375 * Inject random faults into the on-disk data.
5379 ztest_fault_inject(ztest_ds_t
*zd
, uint64_t id
)
5381 ztest_shared_t
*zs
= ztest_shared
;
5382 spa_t
*spa
= ztest_spa
;
5386 uint64_t bad
= 0x1990c0ffeedecadeull
;
5391 int bshift
= SPA_MAXBLOCKSHIFT
+ 2;
5397 boolean_t islog
= B_FALSE
;
5399 path0
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
5400 pathrand
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
5402 mutex_enter(&ztest_vdev_lock
);
5403 maxfaults
= MAXFAULTS();
5404 leaves
= MAX(zs
->zs_mirrors
, 1) * ztest_opts
.zo_raidz
;
5405 mirror_save
= zs
->zs_mirrors
;
5406 mutex_exit(&ztest_vdev_lock
);
5408 ASSERT(leaves
>= 1);
5411 * Grab the name lock as reader. There are some operations
5412 * which don't like to have their vdevs changed while
5413 * they are in progress (i.e. spa_change_guid). Those
5414 * operations will have grabbed the name lock as writer.
5416 (void) rw_rdlock(&ztest_name_lock
);
5419 * We need SCL_STATE here because we're going to look at vd0->vdev_tsd.
5421 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
5423 if (ztest_random(2) == 0) {
5425 * Inject errors on a normal data device or slog device.
5427 top
= ztest_random_vdev_top(spa
, B_TRUE
);
5428 leaf
= ztest_random(leaves
) + zs
->zs_splits
;
5431 * Generate paths to the first leaf in this top-level vdev,
5432 * and to the random leaf we selected. We'll induce transient
5433 * write failures and random online/offline activity on leaf 0,
5434 * and we'll write random garbage to the randomly chosen leaf.
5436 (void) snprintf(path0
, MAXPATHLEN
, ztest_dev_template
,
5437 ztest_opts
.zo_dir
, ztest_opts
.zo_pool
,
5438 top
* leaves
+ zs
->zs_splits
);
5439 (void) snprintf(pathrand
, MAXPATHLEN
, ztest_dev_template
,
5440 ztest_opts
.zo_dir
, ztest_opts
.zo_pool
,
5441 top
* leaves
+ leaf
);
5443 vd0
= vdev_lookup_by_path(spa
->spa_root_vdev
, path0
);
5444 if (vd0
!= NULL
&& vd0
->vdev_top
->vdev_islog
)
5448 * If the top-level vdev needs to be resilvered
5449 * then we only allow faults on the device that is
5452 if (vd0
!= NULL
&& maxfaults
!= 1 &&
5453 (!vdev_resilver_needed(vd0
->vdev_top
, NULL
, NULL
) ||
5454 vd0
->vdev_resilver_txg
!= 0)) {
5456 * Make vd0 explicitly claim to be unreadable,
5457 * or unwriteable, or reach behind its back
5458 * and close the underlying fd. We can do this if
5459 * maxfaults == 0 because we'll fail and reexecute,
5460 * and we can do it if maxfaults >= 2 because we'll
5461 * have enough redundancy. If maxfaults == 1, the
5462 * combination of this with injection of random data
5463 * corruption below exceeds the pool's fault tolerance.
5465 vdev_file_t
*vf
= vd0
->vdev_tsd
;
5467 if (vf
!= NULL
&& ztest_random(3) == 0) {
5468 (void) close(vf
->vf_vnode
->v_fd
);
5469 vf
->vf_vnode
->v_fd
= -1;
5470 } else if (ztest_random(2) == 0) {
5471 vd0
->vdev_cant_read
= B_TRUE
;
5473 vd0
->vdev_cant_write
= B_TRUE
;
5475 guid0
= vd0
->vdev_guid
;
5479 * Inject errors on an l2cache device.
5481 spa_aux_vdev_t
*sav
= &spa
->spa_l2cache
;
5483 if (sav
->sav_count
== 0) {
5484 spa_config_exit(spa
, SCL_STATE
, FTAG
);
5485 (void) rw_unlock(&ztest_name_lock
);
5488 vd0
= sav
->sav_vdevs
[ztest_random(sav
->sav_count
)];
5489 guid0
= vd0
->vdev_guid
;
5490 (void) strcpy(path0
, vd0
->vdev_path
);
5491 (void) strcpy(pathrand
, vd0
->vdev_path
);
5495 maxfaults
= INT_MAX
; /* no limit on cache devices */
5498 spa_config_exit(spa
, SCL_STATE
, FTAG
);
5499 (void) rw_unlock(&ztest_name_lock
);
5502 * If we can tolerate two or more faults, or we're dealing
5503 * with a slog, randomly online/offline vd0.
5505 if ((maxfaults
>= 2 || islog
) && guid0
!= 0) {
5506 if (ztest_random(10) < 6) {
5507 int flags
= (ztest_random(2) == 0 ?
5508 ZFS_OFFLINE_TEMPORARY
: 0);
5511 * We have to grab the zs_name_lock as writer to
5512 * prevent a race between offlining a slog and
5513 * destroying a dataset. Offlining the slog will
5514 * grab a reference on the dataset which may cause
5515 * dsl_destroy_head() to fail with EBUSY thus
5516 * leaving the dataset in an inconsistent state.
5519 (void) rw_wrlock(&ztest_name_lock
);
5521 VERIFY(vdev_offline(spa
, guid0
, flags
) != EBUSY
);
5524 (void) rw_unlock(&ztest_name_lock
);
5527 * Ideally we would like to be able to randomly
5528 * call vdev_[on|off]line without holding locks
5529 * to force unpredictable failures but the side
5530 * effects of vdev_[on|off]line prevent us from
5531 * doing so. We grab the ztest_vdev_lock here to
5532 * prevent a race between injection testing and
5535 mutex_enter(&ztest_vdev_lock
);
5536 (void) vdev_online(spa
, guid0
, 0, NULL
);
5537 mutex_exit(&ztest_vdev_lock
);
5545 * We have at least single-fault tolerance, so inject data corruption.
5547 fd
= open(pathrand
, O_RDWR
);
5549 if (fd
== -1) /* we hit a gap in the device namespace */
5552 fsize
= lseek(fd
, 0, SEEK_END
);
5554 while (--iters
!= 0) {
5556 * The offset must be chosen carefully to ensure that
5557 * we do not inject a given logical block with errors
5558 * on two different leaf devices, because ZFS can not
5559 * tolerate that (if maxfaults==1).
5561 * We divide each leaf into chunks of size
5562 * (# leaves * SPA_MAXBLOCKSIZE * 4). Within each chunk
5563 * there is a series of ranges to which we can inject errors.
5564 * Each range can accept errors on only a single leaf vdev.
5565 * The error injection ranges are separated by ranges
5566 * which we will not inject errors on any device (DMZs).
5567 * Each DMZ must be large enough such that a single block
5568 * can not straddle it, so that a single block can not be
5569 * a target in two different injection ranges (on different
5572 * For example, with 3 leaves, each chunk looks like:
5573 * 0 to 32M: injection range for leaf 0
5574 * 32M to 64M: DMZ - no injection allowed
5575 * 64M to 96M: injection range for leaf 1
5576 * 96M to 128M: DMZ - no injection allowed
5577 * 128M to 160M: injection range for leaf 2
5578 * 160M to 192M: DMZ - no injection allowed
5580 offset
= ztest_random(fsize
/ (leaves
<< bshift
)) *
5581 (leaves
<< bshift
) + (leaf
<< bshift
) +
5582 (ztest_random(1ULL << (bshift
- 1)) & -8ULL);
5585 * Only allow damage to the labels at one end of the vdev.
5587 * If all labels are damaged, the device will be totally
5588 * inaccessible, which will result in loss of data,
5589 * because we also damage (parts of) the other side of
5592 * Additionally, we will always have both an even and an
5593 * odd label, so that we can handle crashes in the
5594 * middle of vdev_config_sync().
5596 if ((leaf
& 1) == 0 && offset
< VDEV_LABEL_START_SIZE
)
5600 * The two end labels are stored at the "end" of the disk, but
5601 * the end of the disk (vdev_psize) is aligned to
5602 * sizeof (vdev_label_t).
5604 uint64_t psize
= P2ALIGN(fsize
, sizeof (vdev_label_t
));
5605 if ((leaf
& 1) == 1 &&
5606 offset
+ sizeof (bad
) > psize
- VDEV_LABEL_END_SIZE
)
5609 mutex_enter(&ztest_vdev_lock
);
5610 if (mirror_save
!= zs
->zs_mirrors
) {
5611 mutex_exit(&ztest_vdev_lock
);
5616 if (pwrite(fd
, &bad
, sizeof (bad
), offset
) != sizeof (bad
))
5617 fatal(1, "can't inject bad word at 0x%llx in %s",
5620 mutex_exit(&ztest_vdev_lock
);
5622 if (ztest_opts
.zo_verbose
>= 7)
5623 (void) printf("injected bad word into %s,"
5624 " offset 0x%llx\n", pathrand
, (u_longlong_t
)offset
);
5629 umem_free(path0
, MAXPATHLEN
);
5630 umem_free(pathrand
, MAXPATHLEN
);
5634 * Verify that DDT repair works as expected.
5637 ztest_ddt_repair(ztest_ds_t
*zd
, uint64_t id
)
5639 ztest_shared_t
*zs
= ztest_shared
;
5640 spa_t
*spa
= ztest_spa
;
5641 objset_t
*os
= zd
->zd_os
;
5643 uint64_t object
, blocksize
, txg
, pattern
, psize
;
5644 enum zio_checksum checksum
= spa_dedup_checksum(spa
);
5649 int copies
= 2 * ZIO_DEDUPDITTO_MIN
;
5652 blocksize
= ztest_random_blocksize();
5653 blocksize
= MIN(blocksize
, 2048); /* because we write so many */
5655 od
= umem_alloc(sizeof (ztest_od_t
), UMEM_NOFAIL
);
5656 ztest_od_init(od
, id
, FTAG
, 0, DMU_OT_UINT64_OTHER
, blocksize
, 0, 0);
5658 if (ztest_object_init(zd
, od
, sizeof (ztest_od_t
), B_FALSE
) != 0) {
5659 umem_free(od
, sizeof (ztest_od_t
));
5664 * Take the name lock as writer to prevent anyone else from changing
5665 * the pool and dataset properies we need to maintain during this test.
5667 (void) rw_wrlock(&ztest_name_lock
);
5669 if (ztest_dsl_prop_set_uint64(zd
->zd_name
, ZFS_PROP_DEDUP
, checksum
,
5671 ztest_dsl_prop_set_uint64(zd
->zd_name
, ZFS_PROP_COPIES
, 1,
5673 (void) rw_unlock(&ztest_name_lock
);
5674 umem_free(od
, sizeof (ztest_od_t
));
5678 dmu_objset_stats_t dds
;
5679 dsl_pool_config_enter(dmu_objset_pool(os
), FTAG
);
5680 dmu_objset_fast_stat(os
, &dds
);
5681 dsl_pool_config_exit(dmu_objset_pool(os
), FTAG
);
5683 object
= od
[0].od_object
;
5684 blocksize
= od
[0].od_blocksize
;
5685 pattern
= zs
->zs_guid
^ dds
.dds_guid
;
5687 ASSERT(object
!= 0);
5689 tx
= dmu_tx_create(os
);
5690 dmu_tx_hold_write(tx
, object
, 0, copies
* blocksize
);
5691 txg
= ztest_tx_assign(tx
, TXG_WAIT
, FTAG
);
5693 (void) rw_unlock(&ztest_name_lock
);
5694 umem_free(od
, sizeof (ztest_od_t
));
5699 * Write all the copies of our block.
5701 for (i
= 0; i
< copies
; i
++) {
5702 uint64_t offset
= i
* blocksize
;
5703 int error
= dmu_buf_hold(os
, object
, offset
, FTAG
, &db
,
5704 DMU_READ_NO_PREFETCH
);
5706 fatal(B_FALSE
, "dmu_buf_hold(%p, %llu, %llu) = %u",
5707 os
, (long long)object
, (long long) offset
, error
);
5709 ASSERT(db
->db_offset
== offset
);
5710 ASSERT(db
->db_size
== blocksize
);
5711 ASSERT(ztest_pattern_match(db
->db_data
, db
->db_size
, pattern
) ||
5712 ztest_pattern_match(db
->db_data
, db
->db_size
, 0ULL));
5713 dmu_buf_will_fill(db
, tx
);
5714 ztest_pattern_set(db
->db_data
, db
->db_size
, pattern
);
5715 dmu_buf_rele(db
, FTAG
);
5719 txg_wait_synced(spa_get_dsl(spa
), txg
);
5722 * Find out what block we got.
5724 VERIFY0(dmu_buf_hold(os
, object
, 0, FTAG
, &db
,
5725 DMU_READ_NO_PREFETCH
));
5726 blk
= *((dmu_buf_impl_t
*)db
)->db_blkptr
;
5727 dmu_buf_rele(db
, FTAG
);
5730 * Damage the block. Dedup-ditto will save us when we read it later.
5732 psize
= BP_GET_PSIZE(&blk
);
5733 abd
= abd_alloc_linear(psize
, B_TRUE
);
5734 ztest_pattern_set(abd_to_buf(abd
), psize
, ~pattern
);
5736 (void) zio_wait(zio_rewrite(NULL
, spa
, 0, &blk
,
5737 abd
, psize
, NULL
, NULL
, ZIO_PRIORITY_SYNC_WRITE
,
5738 ZIO_FLAG_CANFAIL
| ZIO_FLAG_INDUCE_DAMAGE
, NULL
));
5742 (void) rw_unlock(&ztest_name_lock
);
5743 umem_free(od
, sizeof (ztest_od_t
));
5751 ztest_scrub(ztest_ds_t
*zd
, uint64_t id
)
5753 spa_t
*spa
= ztest_spa
;
5755 (void) spa_scan(spa
, POOL_SCAN_SCRUB
);
5756 (void) poll(NULL
, 0, 100); /* wait a moment, then force a restart */
5757 (void) spa_scan(spa
, POOL_SCAN_SCRUB
);
5761 * Change the guid for the pool.
5765 ztest_reguid(ztest_ds_t
*zd
, uint64_t id
)
5767 spa_t
*spa
= ztest_spa
;
5768 uint64_t orig
, load
;
5771 if (ztest_opts
.zo_mmp_test
)
5774 orig
= spa_guid(spa
);
5775 load
= spa_load_guid(spa
);
5777 (void) rw_wrlock(&ztest_name_lock
);
5778 error
= spa_change_guid(spa
);
5779 (void) rw_unlock(&ztest_name_lock
);
5784 if (ztest_opts
.zo_verbose
>= 4) {
5785 (void) printf("Changed guid old %llu -> %llu\n",
5786 (u_longlong_t
)orig
, (u_longlong_t
)spa_guid(spa
));
5789 VERIFY3U(orig
, !=, spa_guid(spa
));
5790 VERIFY3U(load
, ==, spa_load_guid(spa
));
5794 * Rename the pool to a different name and then rename it back.
5798 ztest_spa_rename(ztest_ds_t
*zd
, uint64_t id
)
5800 char *oldname
, *newname
;
5803 if (ztest_opts
.zo_mmp_test
)
5806 (void) rw_wrlock(&ztest_name_lock
);
5808 oldname
= ztest_opts
.zo_pool
;
5809 newname
= umem_alloc(strlen(oldname
) + 5, UMEM_NOFAIL
);
5810 (void) strcpy(newname
, oldname
);
5811 (void) strcat(newname
, "_tmp");
5816 VERIFY3U(0, ==, spa_rename(oldname
, newname
));
5819 * Try to open it under the old name, which shouldn't exist
5821 VERIFY3U(ENOENT
, ==, spa_open(oldname
, &spa
, FTAG
));
5824 * Open it under the new name and make sure it's still the same spa_t.
5826 VERIFY3U(0, ==, spa_open(newname
, &spa
, FTAG
));
5828 ASSERT(spa
== ztest_spa
);
5829 spa_close(spa
, FTAG
);
5832 * Rename it back to the original
5834 VERIFY3U(0, ==, spa_rename(newname
, oldname
));
5837 * Make sure it can still be opened
5839 VERIFY3U(0, ==, spa_open(oldname
, &spa
, FTAG
));
5841 ASSERT(spa
== ztest_spa
);
5842 spa_close(spa
, FTAG
);
5844 umem_free(newname
, strlen(newname
) + 1);
5846 (void) rw_unlock(&ztest_name_lock
);
5850 ztest_fletcher(ztest_ds_t
*zd
, uint64_t id
)
5852 hrtime_t end
= gethrtime() + NANOSEC
;
5854 while (gethrtime() <= end
) {
5855 int run_count
= 100;
5857 struct abd
*abd_data
, *abd_meta
;
5862 zio_cksum_t zc_ref_byteswap
;
5864 size
= ztest_random_blocksize();
5866 buf
= umem_alloc(size
, UMEM_NOFAIL
);
5867 abd_data
= abd_alloc(size
, B_FALSE
);
5868 abd_meta
= abd_alloc(size
, B_TRUE
);
5870 for (i
= 0, ptr
= buf
; i
< size
/ sizeof (*ptr
); i
++, ptr
++)
5871 *ptr
= ztest_random(UINT_MAX
);
5873 abd_copy_from_buf_off(abd_data
, buf
, 0, size
);
5874 abd_copy_from_buf_off(abd_meta
, buf
, 0, size
);
5876 VERIFY0(fletcher_4_impl_set("scalar"));
5877 fletcher_4_native(buf
, size
, NULL
, &zc_ref
);
5878 fletcher_4_byteswap(buf
, size
, NULL
, &zc_ref_byteswap
);
5880 VERIFY0(fletcher_4_impl_set("cycle"));
5881 while (run_count
-- > 0) {
5883 zio_cksum_t zc_byteswap
;
5885 fletcher_4_byteswap(buf
, size
, NULL
, &zc_byteswap
);
5886 fletcher_4_native(buf
, size
, NULL
, &zc
);
5888 VERIFY0(bcmp(&zc
, &zc_ref
, sizeof (zc
)));
5889 VERIFY0(bcmp(&zc_byteswap
, &zc_ref_byteswap
,
5890 sizeof (zc_byteswap
)));
5892 /* Test ABD - data */
5893 abd_fletcher_4_byteswap(abd_data
, size
, NULL
,
5895 abd_fletcher_4_native(abd_data
, size
, NULL
, &zc
);
5897 VERIFY0(bcmp(&zc
, &zc_ref
, sizeof (zc
)));
5898 VERIFY0(bcmp(&zc_byteswap
, &zc_ref_byteswap
,
5899 sizeof (zc_byteswap
)));
5901 /* Test ABD - metadata */
5902 abd_fletcher_4_byteswap(abd_meta
, size
, NULL
,
5904 abd_fletcher_4_native(abd_meta
, size
, NULL
, &zc
);
5906 VERIFY0(bcmp(&zc
, &zc_ref
, sizeof (zc
)));
5907 VERIFY0(bcmp(&zc_byteswap
, &zc_ref_byteswap
,
5908 sizeof (zc_byteswap
)));
5912 umem_free(buf
, size
);
5919 ztest_fletcher_incr(ztest_ds_t
*zd
, uint64_t id
)
5926 zio_cksum_t zc_ref_bswap
;
5928 hrtime_t end
= gethrtime() + NANOSEC
;
5930 while (gethrtime() <= end
) {
5931 int run_count
= 100;
5933 size
= ztest_random_blocksize();
5934 buf
= umem_alloc(size
, UMEM_NOFAIL
);
5936 for (i
= 0, ptr
= buf
; i
< size
/ sizeof (*ptr
); i
++, ptr
++)
5937 *ptr
= ztest_random(UINT_MAX
);
5939 VERIFY0(fletcher_4_impl_set("scalar"));
5940 fletcher_4_native(buf
, size
, NULL
, &zc_ref
);
5941 fletcher_4_byteswap(buf
, size
, NULL
, &zc_ref_bswap
);
5943 VERIFY0(fletcher_4_impl_set("cycle"));
5945 while (run_count
-- > 0) {
5947 zio_cksum_t zc_bswap
;
5950 ZIO_SET_CHECKSUM(&zc
, 0, 0, 0, 0);
5951 ZIO_SET_CHECKSUM(&zc_bswap
, 0, 0, 0, 0);
5953 while (pos
< size
) {
5954 size_t inc
= 64 * ztest_random(size
/ 67);
5955 /* sometimes add few bytes to test non-simd */
5956 if (ztest_random(100) < 10)
5957 inc
+= P2ALIGN(ztest_random(64),
5960 if (inc
> (size
- pos
))
5963 fletcher_4_incremental_native(buf
+ pos
, inc
,
5965 fletcher_4_incremental_byteswap(buf
+ pos
, inc
,
5971 VERIFY3U(pos
, ==, size
);
5973 VERIFY(ZIO_CHECKSUM_EQUAL(zc
, zc_ref
));
5974 VERIFY(ZIO_CHECKSUM_EQUAL(zc_bswap
, zc_ref_bswap
));
5977 * verify if incremental on the whole buffer is
5978 * equivalent to non-incremental version
5980 ZIO_SET_CHECKSUM(&zc
, 0, 0, 0, 0);
5981 ZIO_SET_CHECKSUM(&zc_bswap
, 0, 0, 0, 0);
5983 fletcher_4_incremental_native(buf
, size
, &zc
);
5984 fletcher_4_incremental_byteswap(buf
, size
, &zc_bswap
);
5986 VERIFY(ZIO_CHECKSUM_EQUAL(zc
, zc_ref
));
5987 VERIFY(ZIO_CHECKSUM_EQUAL(zc_bswap
, zc_ref_bswap
));
5990 umem_free(buf
, size
);
5995 ztest_check_path(char *path
)
5998 /* return true on success */
5999 return (!stat(path
, &s
));
6003 ztest_get_zdb_bin(char *bin
, int len
)
6007 * Try to use ZDB_PATH and in-tree zdb path. If not successful, just
6008 * let popen to search through PATH.
6010 if ((zdb_path
= getenv("ZDB_PATH"))) {
6011 strlcpy(bin
, zdb_path
, len
); /* In env */
6012 if (!ztest_check_path(bin
)) {
6013 ztest_dump_core
= 0;
6014 fatal(1, "invalid ZDB_PATH '%s'", bin
);
6019 VERIFY(realpath(getexecname(), bin
) != NULL
);
6020 if (strstr(bin
, "/ztest/")) {
6021 strstr(bin
, "/ztest/")[0] = '\0'; /* In-tree */
6022 strcat(bin
, "/zdb/zdb");
6023 if (ztest_check_path(bin
))
6030 * Verify pool integrity by running zdb.
6033 ztest_run_zdb(char *pool
)
6039 const int len
= MAXPATHLEN
+ MAXNAMELEN
+ 20;
6042 bin
= umem_alloc(len
, UMEM_NOFAIL
);
6043 zdb
= umem_alloc(len
, UMEM_NOFAIL
);
6044 zbuf
= umem_alloc(1024, UMEM_NOFAIL
);
6046 ztest_get_zdb_bin(bin
, len
);
6049 "%s -bcc%s%s -G -d -U %s %s",
6051 ztest_opts
.zo_verbose
>= 3 ? "s" : "",
6052 ztest_opts
.zo_verbose
>= 4 ? "v" : "",
6056 if (ztest_opts
.zo_verbose
>= 5)
6057 (void) printf("Executing %s\n", strstr(zdb
, "zdb "));
6059 fp
= popen(zdb
, "r");
6061 while (fgets(zbuf
, 1024, fp
) != NULL
)
6062 if (ztest_opts
.zo_verbose
>= 3)
6063 (void) printf("%s", zbuf
);
6065 status
= pclose(fp
);
6070 ztest_dump_core
= 0;
6071 if (WIFEXITED(status
))
6072 fatal(0, "'%s' exit code %d", zdb
, WEXITSTATUS(status
));
6074 fatal(0, "'%s' died with signal %d", zdb
, WTERMSIG(status
));
6076 umem_free(bin
, len
);
6077 umem_free(zdb
, len
);
6078 umem_free(zbuf
, 1024);
6082 ztest_walk_pool_directory(char *header
)
6086 if (ztest_opts
.zo_verbose
>= 6)
6087 (void) printf("%s\n", header
);
6089 mutex_enter(&spa_namespace_lock
);
6090 while ((spa
= spa_next(spa
)) != NULL
)
6091 if (ztest_opts
.zo_verbose
>= 6)
6092 (void) printf("\t%s\n", spa_name(spa
));
6093 mutex_exit(&spa_namespace_lock
);
6097 ztest_spa_import_export(char *oldname
, char *newname
)
6099 nvlist_t
*config
, *newconfig
;
6104 if (ztest_opts
.zo_verbose
>= 4) {
6105 (void) printf("import/export: old = %s, new = %s\n",
6110 * Clean up from previous runs.
6112 (void) spa_destroy(newname
);
6115 * Get the pool's configuration and guid.
6117 VERIFY3U(0, ==, spa_open(oldname
, &spa
, FTAG
));
6120 * Kick off a scrub to tickle scrub/export races.
6122 if (ztest_random(2) == 0)
6123 (void) spa_scan(spa
, POOL_SCAN_SCRUB
);
6125 pool_guid
= spa_guid(spa
);
6126 spa_close(spa
, FTAG
);
6128 ztest_walk_pool_directory("pools before export");
6133 VERIFY3U(0, ==, spa_export(oldname
, &config
, B_FALSE
, B_FALSE
));
6135 ztest_walk_pool_directory("pools after export");
6140 newconfig
= spa_tryimport(config
);
6141 ASSERT(newconfig
!= NULL
);
6142 nvlist_free(newconfig
);
6145 * Import it under the new name.
6147 error
= spa_import(newname
, config
, NULL
, 0);
6149 dump_nvlist(config
, 0);
6150 fatal(B_FALSE
, "couldn't import pool %s as %s: error %u",
6151 oldname
, newname
, error
);
6154 ztest_walk_pool_directory("pools after import");
6157 * Try to import it again -- should fail with EEXIST.
6159 VERIFY3U(EEXIST
, ==, spa_import(newname
, config
, NULL
, 0));
6162 * Try to import it under a different name -- should fail with EEXIST.
6164 VERIFY3U(EEXIST
, ==, spa_import(oldname
, config
, NULL
, 0));
6167 * Verify that the pool is no longer visible under the old name.
6169 VERIFY3U(ENOENT
, ==, spa_open(oldname
, &spa
, FTAG
));
6172 * Verify that we can open and close the pool using the new name.
6174 VERIFY3U(0, ==, spa_open(newname
, &spa
, FTAG
));
6175 ASSERT(pool_guid
== spa_guid(spa
));
6176 spa_close(spa
, FTAG
);
6178 nvlist_free(config
);
6182 ztest_resume(spa_t
*spa
)
6184 if (spa_suspended(spa
) && ztest_opts
.zo_verbose
>= 6)
6185 (void) printf("resuming from suspended state\n");
6186 spa_vdev_state_enter(spa
, SCL_NONE
);
6187 vdev_clear(spa
, NULL
);
6188 (void) spa_vdev_state_exit(spa
, NULL
, 0);
6189 (void) zio_resume(spa
);
6193 ztest_resume_thread(void *arg
)
6197 while (!ztest_exiting
) {
6198 if (spa_suspended(spa
))
6200 (void) poll(NULL
, 0, 100);
6203 * Periodically change the zfs_compressed_arc_enabled setting.
6205 if (ztest_random(10) == 0)
6206 zfs_compressed_arc_enabled
= ztest_random(2);
6209 * Periodically change the zfs_abd_scatter_enabled setting.
6211 if (ztest_random(10) == 0)
6212 zfs_abd_scatter_enabled
= ztest_random(2);
6222 ztest_deadman_alarm(int sig
)
6224 fatal(0, "failed to complete within %d seconds of deadline", GRACE
);
6229 ztest_execute(int test
, ztest_info_t
*zi
, uint64_t id
)
6231 ztest_ds_t
*zd
= &ztest_ds
[id
% ztest_opts
.zo_datasets
];
6232 ztest_shared_callstate_t
*zc
= ZTEST_GET_SHARED_CALLSTATE(test
);
6233 hrtime_t functime
= gethrtime();
6236 for (i
= 0; i
< zi
->zi_iters
; i
++)
6237 zi
->zi_func(zd
, id
);
6239 functime
= gethrtime() - functime
;
6241 atomic_add_64(&zc
->zc_count
, 1);
6242 atomic_add_64(&zc
->zc_time
, functime
);
6244 if (ztest_opts
.zo_verbose
>= 4)
6245 (void) printf("%6.2f sec in %s\n",
6246 (double)functime
/ NANOSEC
, zi
->zi_funcname
);
6250 ztest_thread(void *arg
)
6253 uint64_t id
= (uintptr_t)arg
;
6254 ztest_shared_t
*zs
= ztest_shared
;
6258 ztest_shared_callstate_t
*zc
;
6260 while ((now
= gethrtime()) < zs
->zs_thread_stop
) {
6262 * See if it's time to force a crash.
6264 if (now
> zs
->zs_thread_kill
)
6268 * If we're getting ENOSPC with some regularity, stop.
6270 if (zs
->zs_enospc_count
> 10)
6274 * Pick a random function to execute.
6276 rand
= ztest_random(ZTEST_FUNCS
);
6277 zi
= &ztest_info
[rand
];
6278 zc
= ZTEST_GET_SHARED_CALLSTATE(rand
);
6279 call_next
= zc
->zc_next
;
6281 if (now
>= call_next
&&
6282 atomic_cas_64(&zc
->zc_next
, call_next
, call_next
+
6283 ztest_random(2 * zi
->zi_interval
[0] + 1)) == call_next
) {
6284 ztest_execute(rand
, zi
, id
);
6292 ztest_dataset_name(char *dsname
, char *pool
, int d
)
6294 (void) snprintf(dsname
, ZFS_MAX_DATASET_NAME_LEN
, "%s/ds_%d", pool
, d
);
6298 ztest_dataset_destroy(int d
)
6300 char name
[ZFS_MAX_DATASET_NAME_LEN
];
6303 ztest_dataset_name(name
, ztest_opts
.zo_pool
, d
);
6305 if (ztest_opts
.zo_verbose
>= 3)
6306 (void) printf("Destroying %s to free up space\n", name
);
6309 * Cleanup any non-standard clones and snapshots. In general,
6310 * ztest thread t operates on dataset (t % zopt_datasets),
6311 * so there may be more than one thing to clean up.
6313 for (t
= d
; t
< ztest_opts
.zo_threads
;
6314 t
+= ztest_opts
.zo_datasets
)
6315 ztest_dsl_dataset_cleanup(name
, t
);
6317 (void) dmu_objset_find(name
, ztest_objset_destroy_cb
, NULL
,
6318 DS_FIND_SNAPSHOTS
| DS_FIND_CHILDREN
);
6322 ztest_dataset_dirobj_verify(ztest_ds_t
*zd
)
6324 uint64_t usedobjs
, dirobjs
, scratch
;
6327 * ZTEST_DIROBJ is the object directory for the entire dataset.
6328 * Therefore, the number of objects in use should equal the
6329 * number of ZTEST_DIROBJ entries, +1 for ZTEST_DIROBJ itself.
6330 * If not, we have an object leak.
6332 * Note that we can only check this in ztest_dataset_open(),
6333 * when the open-context and syncing-context values agree.
6334 * That's because zap_count() returns the open-context value,
6335 * while dmu_objset_space() returns the rootbp fill count.
6337 VERIFY3U(0, ==, zap_count(zd
->zd_os
, ZTEST_DIROBJ
, &dirobjs
));
6338 dmu_objset_space(zd
->zd_os
, &scratch
, &scratch
, &usedobjs
, &scratch
);
6339 ASSERT3U(dirobjs
+ 1, ==, usedobjs
);
6343 ztest_dataset_open(int d
)
6345 ztest_ds_t
*zd
= &ztest_ds
[d
];
6346 uint64_t committed_seq
= ZTEST_GET_SHARED_DS(d
)->zd_seq
;
6349 char name
[ZFS_MAX_DATASET_NAME_LEN
];
6352 ztest_dataset_name(name
, ztest_opts
.zo_pool
, d
);
6354 (void) rw_rdlock(&ztest_name_lock
);
6356 error
= ztest_dataset_create(name
);
6357 if (error
== ENOSPC
) {
6358 (void) rw_unlock(&ztest_name_lock
);
6359 ztest_record_enospc(FTAG
);
6362 ASSERT(error
== 0 || error
== EEXIST
);
6364 VERIFY0(ztest_dmu_objset_own(name
, DMU_OST_OTHER
, B_FALSE
,
6366 (void) rw_unlock(&ztest_name_lock
);
6368 ztest_zd_init(zd
, ZTEST_GET_SHARED_DS(d
), os
);
6370 zilog
= zd
->zd_zilog
;
6372 if (zilog
->zl_header
->zh_claim_lr_seq
!= 0 &&
6373 zilog
->zl_header
->zh_claim_lr_seq
< committed_seq
)
6374 fatal(0, "missing log records: claimed %llu < committed %llu",
6375 zilog
->zl_header
->zh_claim_lr_seq
, committed_seq
);
6377 ztest_dataset_dirobj_verify(zd
);
6379 zil_replay(os
, zd
, ztest_replay_vector
);
6381 ztest_dataset_dirobj_verify(zd
);
6383 if (ztest_opts
.zo_verbose
>= 6)
6384 (void) printf("%s replay %llu blocks, %llu records, seq %llu\n",
6386 (u_longlong_t
)zilog
->zl_parse_blk_count
,
6387 (u_longlong_t
)zilog
->zl_parse_lr_count
,
6388 (u_longlong_t
)zilog
->zl_replaying_seq
);
6390 zilog
= zil_open(os
, ztest_get_data
);
6392 if (zilog
->zl_replaying_seq
!= 0 &&
6393 zilog
->zl_replaying_seq
< committed_seq
)
6394 fatal(0, "missing log records: replayed %llu < committed %llu",
6395 zilog
->zl_replaying_seq
, committed_seq
);
6401 ztest_dataset_close(int d
)
6403 ztest_ds_t
*zd
= &ztest_ds
[d
];
6405 zil_close(zd
->zd_zilog
);
6406 txg_wait_synced(spa_get_dsl(zd
->zd_os
->os_spa
), 0);
6407 dmu_objset_disown(zd
->zd_os
, B_TRUE
, zd
);
6413 * Kick off threads to run tests on all datasets in parallel.
6416 ztest_run(ztest_shared_t
*zs
)
6420 kthread_t
*resume_thread
;
6421 kthread_t
**run_threads
;
6426 ztest_exiting
= B_FALSE
;
6429 * Initialize parent/child shared state.
6431 mutex_init(&ztest_vdev_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
6432 VERIFY(rwlock_init(&ztest_name_lock
, USYNC_THREAD
, NULL
) == 0);
6434 zs
->zs_thread_start
= gethrtime();
6435 zs
->zs_thread_stop
=
6436 zs
->zs_thread_start
+ ztest_opts
.zo_passtime
* NANOSEC
;
6437 zs
->zs_thread_stop
= MIN(zs
->zs_thread_stop
, zs
->zs_proc_stop
);
6438 zs
->zs_thread_kill
= zs
->zs_thread_stop
;
6439 if (ztest_random(100) < ztest_opts
.zo_killrate
) {
6440 zs
->zs_thread_kill
-=
6441 ztest_random(ztest_opts
.zo_passtime
* NANOSEC
);
6444 mutex_init(&zcl
.zcl_callbacks_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
6446 list_create(&zcl
.zcl_callbacks
, sizeof (ztest_cb_data_t
),
6447 offsetof(ztest_cb_data_t
, zcd_node
));
6452 kernel_init(FREAD
| FWRITE
);
6453 VERIFY0(spa_open(ztest_opts
.zo_pool
, &spa
, FTAG
));
6454 spa
->spa_debug
= B_TRUE
;
6455 metaslab_preload_limit
= ztest_random(20) + 1;
6458 dmu_objset_stats_t dds
;
6459 VERIFY0(ztest_dmu_objset_own(ztest_opts
.zo_pool
,
6460 DMU_OST_ANY
, B_TRUE
, B_TRUE
, FTAG
, &os
));
6461 dsl_pool_config_enter(dmu_objset_pool(os
), FTAG
);
6462 dmu_objset_fast_stat(os
, &dds
);
6463 dsl_pool_config_exit(dmu_objset_pool(os
), FTAG
);
6464 zs
->zs_guid
= dds
.dds_guid
;
6465 dmu_objset_disown(os
, B_TRUE
, FTAG
);
6467 spa
->spa_dedup_ditto
= 2 * ZIO_DEDUPDITTO_MIN
;
6470 * We don't expect the pool to suspend unless maxfaults == 0,
6471 * in which case ztest_fault_inject() temporarily takes away
6472 * the only valid replica.
6474 if (MAXFAULTS() == 0)
6475 spa
->spa_failmode
= ZIO_FAILURE_MODE_WAIT
;
6477 spa
->spa_failmode
= ZIO_FAILURE_MODE_PANIC
;
6480 * Create a thread to periodically resume suspended I/O.
6482 resume_thread
= thread_create(NULL
, 0, ztest_resume_thread
,
6483 spa
, 0, NULL
, TS_RUN
| TS_JOINABLE
, defclsyspri
);
6487 * Set a deadman alarm to abort() if we hang.
6489 signal(SIGALRM
, ztest_deadman_alarm
);
6490 alarm((zs
->zs_thread_stop
- zs
->zs_thread_start
) / NANOSEC
+ GRACE
);
6494 * Verify that we can safely inquire about about any object,
6495 * whether it's allocated or not. To make it interesting,
6496 * we probe a 5-wide window around each power of two.
6497 * This hits all edge cases, including zero and the max.
6499 for (t
= 0; t
< 64; t
++) {
6500 for (d
= -5; d
<= 5; d
++) {
6501 error
= dmu_object_info(spa
->spa_meta_objset
,
6502 (1ULL << t
) + d
, NULL
);
6503 ASSERT(error
== 0 || error
== ENOENT
||
6509 * If we got any ENOSPC errors on the previous run, destroy something.
6511 if (zs
->zs_enospc_count
!= 0) {
6512 int d
= ztest_random(ztest_opts
.zo_datasets
);
6513 ztest_dataset_destroy(d
);
6515 zs
->zs_enospc_count
= 0;
6517 run_threads
= umem_zalloc(ztest_opts
.zo_threads
* sizeof (kthread_t
*),
6520 if (ztest_opts
.zo_verbose
>= 4)
6521 (void) printf("starting main threads...\n");
6524 * Kick off all the tests that run in parallel.
6526 for (t
= 0; t
< ztest_opts
.zo_threads
; t
++) {
6527 if (t
< ztest_opts
.zo_datasets
&& ztest_dataset_open(t
) != 0) {
6528 umem_free(run_threads
, ztest_opts
.zo_threads
*
6529 sizeof (kthread_t
*));
6533 run_threads
[t
] = thread_create(NULL
, 0, ztest_thread
,
6534 (void *)(uintptr_t)t
, 0, NULL
, TS_RUN
| TS_JOINABLE
,
6539 * Wait for all of the tests to complete. We go in reverse order
6540 * so we don't close datasets while threads are still using them.
6542 for (t
= ztest_opts
.zo_threads
- 1; t
>= 0; t
--) {
6543 VERIFY0(thread_join(run_threads
[t
]));
6544 if (t
< ztest_opts
.zo_datasets
)
6545 ztest_dataset_close(t
);
6548 txg_wait_synced(spa_get_dsl(spa
), 0);
6550 zs
->zs_alloc
= metaslab_class_get_alloc(spa_normal_class(spa
));
6551 zs
->zs_space
= metaslab_class_get_space(spa_normal_class(spa
));
6553 umem_free(run_threads
, ztest_opts
.zo_threads
* sizeof (kthread_t
*));
6555 /* Kill the resume thread */
6556 ztest_exiting
= B_TRUE
;
6557 VERIFY0(thread_join(resume_thread
));
6561 * Right before closing the pool, kick off a bunch of async I/O;
6562 * spa_close() should wait for it to complete.
6564 for (object
= 1; object
< 50; object
++) {
6565 dmu_prefetch(spa
->spa_meta_objset
, object
, 0, 0, 1ULL << 20,
6566 ZIO_PRIORITY_SYNC_READ
);
6569 /* Verify that at least one commit cb was called in a timely fashion */
6570 if (zc_cb_counter
>= ZTEST_COMMIT_CB_MIN_REG
)
6571 VERIFY0(zc_min_txg_delay
);
6573 spa_close(spa
, FTAG
);
6576 * Verify that we can loop over all pools.
6578 mutex_enter(&spa_namespace_lock
);
6579 for (spa
= spa_next(NULL
); spa
!= NULL
; spa
= spa_next(spa
))
6580 if (ztest_opts
.zo_verbose
> 3)
6581 (void) printf("spa_next: found %s\n", spa_name(spa
));
6582 mutex_exit(&spa_namespace_lock
);
6585 * Verify that we can export the pool and reimport it under a
6588 if ((ztest_random(2) == 0) && !ztest_opts
.zo_mmp_test
) {
6589 char name
[ZFS_MAX_DATASET_NAME_LEN
];
6590 (void) snprintf(name
, sizeof (name
), "%s_import",
6591 ztest_opts
.zo_pool
);
6592 ztest_spa_import_export(ztest_opts
.zo_pool
, name
);
6593 ztest_spa_import_export(name
, ztest_opts
.zo_pool
);
6598 list_destroy(&zcl
.zcl_callbacks
);
6599 mutex_destroy(&zcl
.zcl_callbacks_lock
);
6600 (void) rwlock_destroy(&ztest_name_lock
);
6601 mutex_destroy(&ztest_vdev_lock
);
6607 ztest_ds_t
*zd
= &ztest_ds
[0];
6611 if (ztest_opts
.zo_verbose
>= 3)
6612 (void) printf("testing spa_freeze()...\n");
6614 kernel_init(FREAD
| FWRITE
);
6615 VERIFY3U(0, ==, spa_open(ztest_opts
.zo_pool
, &spa
, FTAG
));
6616 VERIFY3U(0, ==, ztest_dataset_open(0));
6617 spa
->spa_debug
= B_TRUE
;
6621 * Force the first log block to be transactionally allocated.
6622 * We have to do this before we freeze the pool -- otherwise
6623 * the log chain won't be anchored.
6625 while (BP_IS_HOLE(&zd
->zd_zilog
->zl_header
->zh_log
)) {
6626 ztest_dmu_object_alloc_free(zd
, 0);
6627 zil_commit(zd
->zd_zilog
, 0);
6630 txg_wait_synced(spa_get_dsl(spa
), 0);
6633 * Freeze the pool. This stops spa_sync() from doing anything,
6634 * so that the only way to record changes from now on is the ZIL.
6639 * Because it is hard to predict how much space a write will actually
6640 * require beforehand, we leave ourselves some fudge space to write over
6643 uint64_t capacity
= metaslab_class_get_space(spa_normal_class(spa
)) / 2;
6646 * Run tests that generate log records but don't alter the pool config
6647 * or depend on DSL sync tasks (snapshots, objset create/destroy, etc).
6648 * We do a txg_wait_synced() after each iteration to force the txg
6649 * to increase well beyond the last synced value in the uberblock.
6650 * The ZIL should be OK with that.
6652 * Run a random number of times less than zo_maxloops and ensure we do
6653 * not run out of space on the pool.
6655 while (ztest_random(10) != 0 &&
6656 numloops
++ < ztest_opts
.zo_maxloops
&&
6657 metaslab_class_get_alloc(spa_normal_class(spa
)) < capacity
) {
6659 ztest_od_init(&od
, 0, FTAG
, 0, DMU_OT_UINT64_OTHER
, 0, 0, 0);
6660 VERIFY0(ztest_object_init(zd
, &od
, sizeof (od
), B_FALSE
));
6661 ztest_io(zd
, od
.od_object
,
6662 ztest_random(ZTEST_RANGE_LOCKS
) << SPA_MAXBLOCKSHIFT
);
6663 txg_wait_synced(spa_get_dsl(spa
), 0);
6667 * Commit all of the changes we just generated.
6669 zil_commit(zd
->zd_zilog
, 0);
6670 txg_wait_synced(spa_get_dsl(spa
), 0);
6673 * Close our dataset and close the pool.
6675 ztest_dataset_close(0);
6676 spa_close(spa
, FTAG
);
6680 * Open and close the pool and dataset to induce log replay.
6682 kernel_init(FREAD
| FWRITE
);
6683 VERIFY3U(0, ==, spa_open(ztest_opts
.zo_pool
, &spa
, FTAG
));
6684 ASSERT(spa_freeze_txg(spa
) == UINT64_MAX
);
6685 VERIFY3U(0, ==, ztest_dataset_open(0));
6686 spa
->spa_debug
= B_TRUE
;
6688 txg_wait_synced(spa_get_dsl(spa
), 0);
6689 ztest_dataset_close(0);
6690 ztest_reguid(NULL
, 0);
6692 spa_close(spa
, FTAG
);
6697 print_time(hrtime_t t
, char *timebuf
)
6699 hrtime_t s
= t
/ NANOSEC
;
6700 hrtime_t m
= s
/ 60;
6701 hrtime_t h
= m
/ 60;
6702 hrtime_t d
= h
/ 24;
6711 (void) sprintf(timebuf
,
6712 "%llud%02lluh%02llum%02llus", d
, h
, m
, s
);
6714 (void) sprintf(timebuf
, "%lluh%02llum%02llus", h
, m
, s
);
6716 (void) sprintf(timebuf
, "%llum%02llus", m
, s
);
6718 (void) sprintf(timebuf
, "%llus", s
);
6722 make_random_props(void)
6726 VERIFY(nvlist_alloc(&props
, NV_UNIQUE_NAME
, 0) == 0);
6727 if (ztest_random(2) == 0)
6729 VERIFY(nvlist_add_uint64(props
, "autoreplace", 1) == 0);
6735 * Import a storage pool with the given name.
6738 ztest_import(ztest_shared_t
*zs
)
6740 libzfs_handle_t
*hdl
;
6741 importargs_t args
= { 0 };
6743 nvlist_t
*cfg
= NULL
;
6745 char *searchdirs
[nsearch
];
6746 char *name
= ztest_opts
.zo_pool
;
6747 int flags
= ZFS_IMPORT_MISSING_LOG
;
6750 mutex_init(&ztest_vdev_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
6751 VERIFY(rwlock_init(&ztest_name_lock
, USYNC_THREAD
, NULL
) == 0);
6753 kernel_init(FREAD
| FWRITE
);
6754 hdl
= libzfs_init();
6756 searchdirs
[0] = ztest_opts
.zo_dir
;
6757 args
.paths
= nsearch
;
6758 args
.path
= searchdirs
;
6759 args
.can_be_active
= B_FALSE
;
6761 error
= zpool_tryimport(hdl
, name
, &cfg
, &args
);
6763 (void) fatal(0, "No pools found\n");
6765 VERIFY0(spa_import(name
, cfg
, NULL
, flags
));
6766 VERIFY0(spa_open(name
, &spa
, FTAG
));
6767 zs
->zs_metaslab_sz
=
6768 1ULL << spa
->spa_root_vdev
->vdev_child
[0]->vdev_ms_shift
;
6769 spa_close(spa
, FTAG
);
6774 if (!ztest_opts
.zo_mmp_test
) {
6775 ztest_run_zdb(ztest_opts
.zo_pool
);
6777 ztest_run_zdb(ztest_opts
.zo_pool
);
6780 (void) rwlock_destroy(&ztest_name_lock
);
6781 mutex_destroy(&ztest_vdev_lock
);
6785 * Create a storage pool with the given name and initial vdev size.
6786 * Then test spa_freeze() functionality.
6789 ztest_init(ztest_shared_t
*zs
)
6792 nvlist_t
*nvroot
, *props
;
6795 mutex_init(&ztest_vdev_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
6796 VERIFY(rwlock_init(&ztest_name_lock
, USYNC_THREAD
, NULL
) == 0);
6798 kernel_init(FREAD
| FWRITE
);
6801 * Create the storage pool.
6803 (void) spa_destroy(ztest_opts
.zo_pool
);
6804 ztest_shared
->zs_vdev_next_leaf
= 0;
6806 zs
->zs_mirrors
= ztest_opts
.zo_mirrors
;
6807 nvroot
= make_vdev_root(NULL
, NULL
, NULL
, ztest_opts
.zo_vdev_size
, 0,
6808 0, ztest_opts
.zo_raidz
, zs
->zs_mirrors
, 1);
6809 props
= make_random_props();
6810 for (i
= 0; i
< SPA_FEATURES
; i
++) {
6812 VERIFY3S(-1, !=, asprintf(&buf
, "feature@%s",
6813 spa_feature_table
[i
].fi_uname
));
6814 VERIFY3U(0, ==, nvlist_add_uint64(props
, buf
, 0));
6818 spa_create(ztest_opts
.zo_pool
, nvroot
, props
, NULL
, NULL
));
6819 nvlist_free(nvroot
);
6822 VERIFY3U(0, ==, spa_open(ztest_opts
.zo_pool
, &spa
, FTAG
));
6823 zs
->zs_metaslab_sz
=
6824 1ULL << spa
->spa_root_vdev
->vdev_child
[0]->vdev_ms_shift
;
6825 spa_close(spa
, FTAG
);
6829 if (!ztest_opts
.zo_mmp_test
) {
6830 ztest_run_zdb(ztest_opts
.zo_pool
);
6832 ztest_run_zdb(ztest_opts
.zo_pool
);
6835 (void) rwlock_destroy(&ztest_name_lock
);
6836 mutex_destroy(&ztest_vdev_lock
);
6842 static char ztest_name_data
[] = "/tmp/ztest.data.XXXXXX";
6844 ztest_fd_data
= mkstemp(ztest_name_data
);
6845 ASSERT3S(ztest_fd_data
, >=, 0);
6846 (void) unlink(ztest_name_data
);
6850 shared_data_size(ztest_shared_hdr_t
*hdr
)
6854 size
= hdr
->zh_hdr_size
;
6855 size
+= hdr
->zh_opts_size
;
6856 size
+= hdr
->zh_size
;
6857 size
+= hdr
->zh_stats_size
* hdr
->zh_stats_count
;
6858 size
+= hdr
->zh_ds_size
* hdr
->zh_ds_count
;
6867 ztest_shared_hdr_t
*hdr
;
6869 hdr
= (void *)mmap(0, P2ROUNDUP(sizeof (*hdr
), getpagesize()),
6870 PROT_READ
| PROT_WRITE
, MAP_SHARED
, ztest_fd_data
, 0);
6871 ASSERT(hdr
!= MAP_FAILED
);
6873 VERIFY3U(0, ==, ftruncate(ztest_fd_data
, sizeof (ztest_shared_hdr_t
)));
6875 hdr
->zh_hdr_size
= sizeof (ztest_shared_hdr_t
);
6876 hdr
->zh_opts_size
= sizeof (ztest_shared_opts_t
);
6877 hdr
->zh_size
= sizeof (ztest_shared_t
);
6878 hdr
->zh_stats_size
= sizeof (ztest_shared_callstate_t
);
6879 hdr
->zh_stats_count
= ZTEST_FUNCS
;
6880 hdr
->zh_ds_size
= sizeof (ztest_shared_ds_t
);
6881 hdr
->zh_ds_count
= ztest_opts
.zo_datasets
;
6883 size
= shared_data_size(hdr
);
6884 VERIFY3U(0, ==, ftruncate(ztest_fd_data
, size
));
6886 (void) munmap((caddr_t
)hdr
, P2ROUNDUP(sizeof (*hdr
), getpagesize()));
6893 ztest_shared_hdr_t
*hdr
;
6896 hdr
= (void *)mmap(0, P2ROUNDUP(sizeof (*hdr
), getpagesize()),
6897 PROT_READ
, MAP_SHARED
, ztest_fd_data
, 0);
6898 ASSERT(hdr
!= MAP_FAILED
);
6900 size
= shared_data_size(hdr
);
6902 (void) munmap((caddr_t
)hdr
, P2ROUNDUP(sizeof (*hdr
), getpagesize()));
6903 hdr
= ztest_shared_hdr
= (void *)mmap(0, P2ROUNDUP(size
, getpagesize()),
6904 PROT_READ
| PROT_WRITE
, MAP_SHARED
, ztest_fd_data
, 0);
6905 ASSERT(hdr
!= MAP_FAILED
);
6906 buf
= (uint8_t *)hdr
;
6908 offset
= hdr
->zh_hdr_size
;
6909 ztest_shared_opts
= (void *)&buf
[offset
];
6910 offset
+= hdr
->zh_opts_size
;
6911 ztest_shared
= (void *)&buf
[offset
];
6912 offset
+= hdr
->zh_size
;
6913 ztest_shared_callstate
= (void *)&buf
[offset
];
6914 offset
+= hdr
->zh_stats_size
* hdr
->zh_stats_count
;
6915 ztest_shared_ds
= (void *)&buf
[offset
];
6919 exec_child(char *cmd
, char *libpath
, boolean_t ignorekill
, int *statusp
)
6923 char *cmdbuf
= NULL
;
6928 cmdbuf
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
6929 (void) strlcpy(cmdbuf
, getexecname(), MAXPATHLEN
);
6934 fatal(1, "fork failed");
6936 if (pid
== 0) { /* child */
6937 char *emptyargv
[2] = { cmd
, NULL
};
6938 char fd_data_str
[12];
6940 struct rlimit rl
= { 1024, 1024 };
6941 (void) setrlimit(RLIMIT_NOFILE
, &rl
);
6943 (void) close(ztest_fd_rand
);
6944 VERIFY(11 >= snprintf(fd_data_str
, 12, "%d", ztest_fd_data
));
6945 VERIFY(0 == setenv("ZTEST_FD_DATA", fd_data_str
, 1));
6947 (void) enable_extended_FILE_stdio(-1, -1);
6948 if (libpath
!= NULL
)
6949 VERIFY(0 == setenv("LD_LIBRARY_PATH", libpath
, 1));
6950 (void) execv(cmd
, emptyargv
);
6951 ztest_dump_core
= B_FALSE
;
6952 fatal(B_TRUE
, "exec failed: %s", cmd
);
6955 if (cmdbuf
!= NULL
) {
6956 umem_free(cmdbuf
, MAXPATHLEN
);
6960 while (waitpid(pid
, &status
, 0) != pid
)
6962 if (statusp
!= NULL
)
6965 if (WIFEXITED(status
)) {
6966 if (WEXITSTATUS(status
) != 0) {
6967 (void) fprintf(stderr
, "child exited with code %d\n",
6968 WEXITSTATUS(status
));
6972 } else if (WIFSIGNALED(status
)) {
6973 if (!ignorekill
|| WTERMSIG(status
) != SIGKILL
) {
6974 (void) fprintf(stderr
, "child died with signal %d\n",
6980 (void) fprintf(stderr
, "something strange happened to child\n");
6987 ztest_run_init(void)
6991 ztest_shared_t
*zs
= ztest_shared
;
6994 * Blow away any existing copy of zpool.cache
6996 (void) remove(spa_config_path
);
6998 if (ztest_opts
.zo_init
== 0) {
6999 if (ztest_opts
.zo_verbose
>= 1)
7000 (void) printf("Importing pool %s\n",
7001 ztest_opts
.zo_pool
);
7007 * Create and initialize our storage pool.
7009 for (i
= 1; i
<= ztest_opts
.zo_init
; i
++) {
7010 bzero(zs
, sizeof (ztest_shared_t
));
7011 if (ztest_opts
.zo_verbose
>= 3 &&
7012 ztest_opts
.zo_init
!= 1) {
7013 (void) printf("ztest_init(), pass %d\n", i
);
7020 main(int argc
, char **argv
)
7028 ztest_shared_callstate_t
*zc
;
7030 char numbuf
[NN_NUMBUF_SZ
];
7035 char *fd_data_str
= getenv("ZTEST_FD_DATA");
7036 struct sigaction action
;
7038 (void) setvbuf(stdout
, NULL
, _IOLBF
, 0);
7040 dprintf_setup(&argc
, argv
);
7042 action
.sa_handler
= sig_handler
;
7043 sigemptyset(&action
.sa_mask
);
7044 action
.sa_flags
= 0;
7046 if (sigaction(SIGSEGV
, &action
, NULL
) < 0) {
7047 (void) fprintf(stderr
, "ztest: cannot catch SIGSEGV: %s.\n",
7052 if (sigaction(SIGABRT
, &action
, NULL
) < 0) {
7053 (void) fprintf(stderr
, "ztest: cannot catch SIGABRT: %s.\n",
7058 ztest_fd_rand
= open("/dev/urandom", O_RDONLY
);
7059 ASSERT3S(ztest_fd_rand
, >=, 0);
7062 process_options(argc
, argv
);
7067 bcopy(&ztest_opts
, ztest_shared_opts
,
7068 sizeof (*ztest_shared_opts
));
7070 ztest_fd_data
= atoi(fd_data_str
);
7072 bcopy(ztest_shared_opts
, &ztest_opts
, sizeof (ztest_opts
));
7074 ASSERT3U(ztest_opts
.zo_datasets
, ==, ztest_shared_hdr
->zh_ds_count
);
7076 /* Override location of zpool.cache */
7077 VERIFY(asprintf((char **)&spa_config_path
, "%s/zpool.cache",
7078 ztest_opts
.zo_dir
) != -1);
7080 ztest_ds
= umem_alloc(ztest_opts
.zo_datasets
* sizeof (ztest_ds_t
),
7085 metaslab_gang_bang
= ztest_opts
.zo_metaslab_gang_bang
;
7086 metaslab_df_alloc_threshold
=
7087 zs
->zs_metaslab_df_alloc_threshold
;
7096 hasalt
= (strlen(ztest_opts
.zo_alt_ztest
) != 0);
7098 if (ztest_opts
.zo_verbose
>= 1) {
7099 (void) printf("%llu vdevs, %d datasets, %d threads,"
7100 " %llu seconds...\n",
7101 (u_longlong_t
)ztest_opts
.zo_vdevs
,
7102 ztest_opts
.zo_datasets
,
7103 ztest_opts
.zo_threads
,
7104 (u_longlong_t
)ztest_opts
.zo_time
);
7107 cmd
= umem_alloc(MAXNAMELEN
, UMEM_NOFAIL
);
7108 (void) strlcpy(cmd
, getexecname(), MAXNAMELEN
);
7110 zs
->zs_do_init
= B_TRUE
;
7111 if (strlen(ztest_opts
.zo_alt_ztest
) != 0) {
7112 if (ztest_opts
.zo_verbose
>= 1) {
7113 (void) printf("Executing older ztest for "
7114 "initialization: %s\n", ztest_opts
.zo_alt_ztest
);
7116 VERIFY(!exec_child(ztest_opts
.zo_alt_ztest
,
7117 ztest_opts
.zo_alt_libpath
, B_FALSE
, NULL
));
7119 VERIFY(!exec_child(NULL
, NULL
, B_FALSE
, NULL
));
7121 zs
->zs_do_init
= B_FALSE
;
7123 zs
->zs_proc_start
= gethrtime();
7124 zs
->zs_proc_stop
= zs
->zs_proc_start
+ ztest_opts
.zo_time
* NANOSEC
;
7126 for (f
= 0; f
< ZTEST_FUNCS
; f
++) {
7127 zi
= &ztest_info
[f
];
7128 zc
= ZTEST_GET_SHARED_CALLSTATE(f
);
7129 if (zs
->zs_proc_start
+ zi
->zi_interval
[0] > zs
->zs_proc_stop
)
7130 zc
->zc_next
= UINT64_MAX
;
7132 zc
->zc_next
= zs
->zs_proc_start
+
7133 ztest_random(2 * zi
->zi_interval
[0] + 1);
7137 * Run the tests in a loop. These tests include fault injection
7138 * to verify that self-healing data works, and forced crashes
7139 * to verify that we never lose on-disk consistency.
7141 while (gethrtime() < zs
->zs_proc_stop
) {
7146 * Initialize the workload counters for each function.
7148 for (f
= 0; f
< ZTEST_FUNCS
; f
++) {
7149 zc
= ZTEST_GET_SHARED_CALLSTATE(f
);
7154 /* Set the allocation switch size */
7155 zs
->zs_metaslab_df_alloc_threshold
=
7156 ztest_random(zs
->zs_metaslab_sz
/ 4) + 1;
7158 if (!hasalt
|| ztest_random(2) == 0) {
7159 if (hasalt
&& ztest_opts
.zo_verbose
>= 1) {
7160 (void) printf("Executing newer ztest: %s\n",
7164 killed
= exec_child(cmd
, NULL
, B_TRUE
, &status
);
7166 if (hasalt
&& ztest_opts
.zo_verbose
>= 1) {
7167 (void) printf("Executing older ztest: %s\n",
7168 ztest_opts
.zo_alt_ztest
);
7171 killed
= exec_child(ztest_opts
.zo_alt_ztest
,
7172 ztest_opts
.zo_alt_libpath
, B_TRUE
, &status
);
7179 if (ztest_opts
.zo_verbose
>= 1) {
7180 hrtime_t now
= gethrtime();
7182 now
= MIN(now
, zs
->zs_proc_stop
);
7183 print_time(zs
->zs_proc_stop
- now
, timebuf
);
7184 nicenum(zs
->zs_space
, numbuf
, sizeof (numbuf
));
7186 (void) printf("Pass %3d, %8s, %3llu ENOSPC, "
7187 "%4.1f%% of %5s used, %3.0f%% done, %8s to go\n",
7189 WIFEXITED(status
) ? "Complete" : "SIGKILL",
7190 (u_longlong_t
)zs
->zs_enospc_count
,
7191 100.0 * zs
->zs_alloc
/ zs
->zs_space
,
7193 100.0 * (now
- zs
->zs_proc_start
) /
7194 (ztest_opts
.zo_time
* NANOSEC
), timebuf
);
7197 if (ztest_opts
.zo_verbose
>= 2) {
7198 (void) printf("\nWorkload summary:\n\n");
7199 (void) printf("%7s %9s %s\n",
7200 "Calls", "Time", "Function");
7201 (void) printf("%7s %9s %s\n",
7202 "-----", "----", "--------");
7203 for (f
= 0; f
< ZTEST_FUNCS
; f
++) {
7204 zi
= &ztest_info
[f
];
7205 zc
= ZTEST_GET_SHARED_CALLSTATE(f
);
7206 print_time(zc
->zc_time
, timebuf
);
7207 (void) printf("%7llu %9s %s\n",
7208 (u_longlong_t
)zc
->zc_count
, timebuf
,
7211 (void) printf("\n");
7215 * It's possible that we killed a child during a rename test,
7216 * in which case we'll have a 'ztest_tmp' pool lying around
7217 * instead of 'ztest'. Do a blind rename in case this happened.
7220 if (spa_open(ztest_opts
.zo_pool
, &spa
, FTAG
) == 0) {
7221 spa_close(spa
, FTAG
);
7223 char tmpname
[ZFS_MAX_DATASET_NAME_LEN
];
7225 kernel_init(FREAD
| FWRITE
);
7226 (void) snprintf(tmpname
, sizeof (tmpname
), "%s_tmp",
7227 ztest_opts
.zo_pool
);
7228 (void) spa_rename(tmpname
, ztest_opts
.zo_pool
);
7232 if (!ztest_opts
.zo_mmp_test
)
7233 ztest_run_zdb(ztest_opts
.zo_pool
);
7236 if (ztest_opts
.zo_verbose
>= 1) {
7238 (void) printf("%d runs of older ztest: %s\n", older
,
7239 ztest_opts
.zo_alt_ztest
);
7240 (void) printf("%d runs of newer ztest: %s\n", newer
,
7243 (void) printf("%d killed, %d completed, %.0f%% kill rate\n",
7244 kills
, iters
- kills
, (100.0 * kills
) / MAX(1, iters
));
7247 umem_free(cmd
, MAXNAMELEN
);