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) 2012 by Delphix. All rights reserved.
24 * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
28 * The objective of this program is to provide a DMU/ZAP/SPA stress test
29 * that runs entirely in userland, is easy to use, and easy to extend.
31 * The overall design of the ztest program is as follows:
33 * (1) For each major functional area (e.g. adding vdevs to a pool,
34 * creating and destroying datasets, reading and writing objects, etc)
35 * we have a simple routine to test that functionality. These
36 * individual routines do not have to do anything "stressful".
38 * (2) We turn these simple functionality tests into a stress test by
39 * running them all in parallel, with as many threads as desired,
40 * and spread across as many datasets, objects, and vdevs as desired.
42 * (3) While all this is happening, we inject faults into the pool to
43 * verify that self-healing data really works.
45 * (4) Every time we open a dataset, we change its checksum and compression
46 * functions. Thus even individual objects vary from block to block
47 * in which checksum they use and whether they're compressed.
49 * (5) To verify that we never lose on-disk consistency after a crash,
50 * we run the entire test in a child of the main process.
51 * At random times, the child self-immolates with a SIGKILL.
52 * This is the software equivalent of pulling the power cord.
53 * The parent then runs the test again, using the existing
54 * storage pool, as many times as desired. If backwards compatability
55 * testing is enabled ztest will sometimes run the "older" version
56 * of ztest after a SIGKILL.
58 * (6) To verify that we don't have future leaks or temporal incursions,
59 * many of the functional tests record the transaction group number
60 * as part of their data. When reading old data, they verify that
61 * the transaction group number is less than the current, open txg.
62 * If you add a new test, please do this if applicable.
64 * (7) Threads are created with a reduced stack size, for sanity checking.
65 * Therefore, it's important not to allocate huge buffers on the stack.
67 * When run with no arguments, ztest runs for about five minutes and
68 * produces no output if successful. To get a little bit of information,
69 * specify -V. To get more information, specify -VV, and so on.
71 * To turn this into an overnight stress test, use -T to specify run time.
73 * You can ask more more vdevs [-v], datasets [-d], or threads [-t]
74 * to increase the pool capacity, fanout, and overall stress level.
76 * Use the -k option to set the desired frequency of kills.
78 * When ztest invokes itself it passes all relevant information through a
79 * temporary file which is mmap-ed in the child process. This allows shared
80 * memory to survive the exec syscall. The ztest_shared_hdr_t struct is always
81 * stored at offset 0 of this file and contains information on the size and
82 * number of shared structures in the file. The information stored in this file
83 * must remain backwards compatible with older versions of ztest so that
84 * ztest can invoke them during backwards compatibility testing (-B).
87 #include <sys/zfs_context.h>
93 #include <sys/dmu_objset.h>
99 #include <sys/resource.h>
102 #include <sys/zil_impl.h>
103 #include <sys/vdev_impl.h>
104 #include <sys/vdev_file.h>
105 #include <sys/spa_impl.h>
106 #include <sys/metaslab_impl.h>
107 #include <sys/dsl_prop.h>
108 #include <sys/dsl_dataset.h>
109 #include <sys/dsl_scan.h>
110 #include <sys/zio_checksum.h>
111 #include <sys/refcount.h>
112 #include <sys/zfeature.h>
114 #include <stdio_ext.h>
122 #include <sys/fs/zfs.h>
123 #include <libnvpair.h>
125 static int ztest_fd_data
= -1;
126 static int ztest_fd_rand
= -1;
128 typedef struct ztest_shared_hdr
{
129 uint64_t zh_hdr_size
;
130 uint64_t zh_opts_size
;
132 uint64_t zh_stats_size
;
133 uint64_t zh_stats_count
;
135 uint64_t zh_ds_count
;
136 } ztest_shared_hdr_t
;
138 static ztest_shared_hdr_t
*ztest_shared_hdr
;
140 typedef struct ztest_shared_opts
{
141 char zo_pool
[MAXNAMELEN
];
142 char zo_dir
[MAXNAMELEN
];
143 char zo_alt_ztest
[MAXNAMELEN
];
144 char zo_alt_libpath
[MAXNAMELEN
];
146 uint64_t zo_vdevtime
;
154 uint64_t zo_passtime
;
155 uint64_t zo_killrate
;
159 uint64_t zo_maxloops
;
160 uint64_t zo_metaslab_gang_bang
;
161 } ztest_shared_opts_t
;
163 static const ztest_shared_opts_t ztest_opts_defaults
= {
164 .zo_pool
= { 'z', 't', 'e', 's', 't', '\0' },
165 .zo_dir
= { '/', 't', 'm', 'p', '\0' },
166 .zo_alt_ztest
= { '\0' },
167 .zo_alt_libpath
= { '\0' },
169 .zo_ashift
= SPA_MINBLOCKSHIFT
,
172 .zo_raidz_parity
= 1,
173 .zo_vdev_size
= SPA_MINDEVSIZE
,
176 .zo_passtime
= 60, /* 60 seconds */
177 .zo_killrate
= 70, /* 70% kill rate */
180 .zo_time
= 300, /* 5 minutes */
181 .zo_maxloops
= 50, /* max loops during spa_freeze() */
182 .zo_metaslab_gang_bang
= 32 << 10
185 extern uint64_t metaslab_gang_bang
;
186 extern uint64_t metaslab_df_alloc_threshold
;
188 static ztest_shared_opts_t
*ztest_shared_opts
;
189 static ztest_shared_opts_t ztest_opts
;
191 typedef struct ztest_shared_ds
{
195 static ztest_shared_ds_t
*ztest_shared_ds
;
196 #define ZTEST_GET_SHARED_DS(d) (&ztest_shared_ds[d])
198 #define BT_MAGIC 0x123456789abcdefULL
199 #define MAXFAULTS() \
200 (MAX(zs->zs_mirrors, 1) * (ztest_opts.zo_raidz_parity + 1) - 1)
204 ZTEST_IO_WRITE_PATTERN
,
205 ZTEST_IO_WRITE_ZEROES
,
211 typedef struct ztest_block_tag
{
221 typedef struct bufwad
{
228 * XXX -- fix zfs range locks to be generic so we can use them here.
250 #define ZTEST_RANGE_LOCKS 64
251 #define ZTEST_OBJECT_LOCKS 64
254 * Object descriptor. Used as a template for object lookup/create/remove.
256 typedef struct ztest_od
{
259 dmu_object_type_t od_type
;
260 dmu_object_type_t od_crtype
;
261 uint64_t od_blocksize
;
262 uint64_t od_crblocksize
;
265 char od_name
[MAXNAMELEN
];
271 typedef struct ztest_ds
{
272 ztest_shared_ds_t
*zd_shared
;
274 krwlock_t zd_zilog_lock
;
276 ztest_od_t
*zd_od
; /* debugging aid */
277 char zd_name
[MAXNAMELEN
];
278 kmutex_t zd_dirobj_lock
;
279 rll_t zd_object_lock
[ZTEST_OBJECT_LOCKS
];
280 rll_t zd_range_lock
[ZTEST_RANGE_LOCKS
];
284 * Per-iteration state.
286 typedef void ztest_func_t(ztest_ds_t
*zd
, uint64_t id
);
288 typedef struct ztest_info
{
289 ztest_func_t
*zi_func
; /* test function */
290 uint64_t zi_iters
; /* iterations per execution */
291 uint64_t *zi_interval
; /* execute every <interval> seconds */
294 typedef struct ztest_shared_callstate
{
295 uint64_t zc_count
; /* per-pass count */
296 uint64_t zc_time
; /* per-pass time */
297 uint64_t zc_next
; /* next time to call this function */
298 } ztest_shared_callstate_t
;
300 static ztest_shared_callstate_t
*ztest_shared_callstate
;
301 #define ZTEST_GET_SHARED_CALLSTATE(c) (&ztest_shared_callstate[c])
304 * Note: these aren't static because we want dladdr() to work.
306 ztest_func_t ztest_dmu_read_write
;
307 ztest_func_t ztest_dmu_write_parallel
;
308 ztest_func_t ztest_dmu_object_alloc_free
;
309 ztest_func_t ztest_dmu_commit_callbacks
;
310 ztest_func_t ztest_zap
;
311 ztest_func_t ztest_zap_parallel
;
312 ztest_func_t ztest_zil_commit
;
313 ztest_func_t ztest_zil_remount
;
314 ztest_func_t ztest_dmu_read_write_zcopy
;
315 ztest_func_t ztest_dmu_objset_create_destroy
;
316 ztest_func_t ztest_dmu_prealloc
;
317 ztest_func_t ztest_fzap
;
318 ztest_func_t ztest_dmu_snapshot_create_destroy
;
319 ztest_func_t ztest_dsl_prop_get_set
;
320 ztest_func_t ztest_spa_prop_get_set
;
321 ztest_func_t ztest_spa_create_destroy
;
322 ztest_func_t ztest_fault_inject
;
323 ztest_func_t ztest_ddt_repair
;
324 ztest_func_t ztest_dmu_snapshot_hold
;
325 ztest_func_t ztest_spa_rename
;
326 ztest_func_t ztest_scrub
;
327 ztest_func_t ztest_dsl_dataset_promote_busy
;
328 ztest_func_t ztest_vdev_attach_detach
;
329 ztest_func_t ztest_vdev_LUN_growth
;
330 ztest_func_t ztest_vdev_add_remove
;
331 ztest_func_t ztest_vdev_aux_add_remove
;
332 ztest_func_t ztest_split_pool
;
333 ztest_func_t ztest_reguid
;
335 uint64_t zopt_always
= 0ULL * NANOSEC
; /* all the time */
336 uint64_t zopt_incessant
= 1ULL * NANOSEC
/ 10; /* every 1/10 second */
337 uint64_t zopt_often
= 1ULL * NANOSEC
; /* every second */
338 uint64_t zopt_sometimes
= 10ULL * NANOSEC
; /* every 10 seconds */
339 uint64_t zopt_rarely
= 60ULL * NANOSEC
; /* every 60 seconds */
341 ztest_info_t ztest_info
[] = {
342 { ztest_dmu_read_write
, 1, &zopt_always
},
343 { ztest_dmu_write_parallel
, 10, &zopt_always
},
344 { ztest_dmu_object_alloc_free
, 1, &zopt_always
},
345 { ztest_dmu_commit_callbacks
, 1, &zopt_always
},
346 { ztest_zap
, 30, &zopt_always
},
347 { ztest_zap_parallel
, 100, &zopt_always
},
348 { ztest_split_pool
, 1, &zopt_always
},
349 { ztest_zil_commit
, 1, &zopt_incessant
},
350 { ztest_zil_remount
, 1, &zopt_sometimes
},
351 { ztest_dmu_read_write_zcopy
, 1, &zopt_often
},
352 { ztest_dmu_objset_create_destroy
, 1, &zopt_often
},
353 { ztest_dsl_prop_get_set
, 1, &zopt_often
},
354 { ztest_spa_prop_get_set
, 1, &zopt_sometimes
},
356 { ztest_dmu_prealloc
, 1, &zopt_sometimes
},
358 { ztest_fzap
, 1, &zopt_sometimes
},
359 { ztest_dmu_snapshot_create_destroy
, 1, &zopt_sometimes
},
360 { ztest_spa_create_destroy
, 1, &zopt_sometimes
},
361 { ztest_fault_inject
, 1, &zopt_sometimes
},
362 { ztest_ddt_repair
, 1, &zopt_sometimes
},
363 { ztest_dmu_snapshot_hold
, 1, &zopt_sometimes
},
364 { ztest_reguid
, 1, &zopt_sometimes
},
365 { ztest_spa_rename
, 1, &zopt_rarely
},
366 { ztest_scrub
, 1, &zopt_rarely
},
367 { ztest_dsl_dataset_promote_busy
, 1, &zopt_rarely
},
368 { ztest_vdev_attach_detach
, 1, &zopt_rarely
},
369 { ztest_vdev_LUN_growth
, 1, &zopt_rarely
},
370 { ztest_vdev_add_remove
, 1,
371 &ztest_opts
.zo_vdevtime
},
372 { ztest_vdev_aux_add_remove
, 1,
373 &ztest_opts
.zo_vdevtime
},
376 #define ZTEST_FUNCS (sizeof (ztest_info) / sizeof (ztest_info_t))
379 * The following struct is used to hold a list of uncalled commit callbacks.
380 * The callbacks are ordered by txg number.
382 typedef struct ztest_cb_list
{
383 kmutex_t zcl_callbacks_lock
;
384 list_t zcl_callbacks
;
388 * Stuff we need to share writably between parent and child.
390 typedef struct ztest_shared
{
391 boolean_t zs_do_init
;
392 hrtime_t zs_proc_start
;
393 hrtime_t zs_proc_stop
;
394 hrtime_t zs_thread_start
;
395 hrtime_t zs_thread_stop
;
396 hrtime_t zs_thread_kill
;
397 uint64_t zs_enospc_count
;
398 uint64_t zs_vdev_next_leaf
;
399 uint64_t zs_vdev_aux
;
404 uint64_t zs_metaslab_sz
;
405 uint64_t zs_metaslab_df_alloc_threshold
;
409 #define ID_PARALLEL -1ULL
411 static char ztest_dev_template
[] = "%s/%s.%llua";
412 static char ztest_aux_template
[] = "%s/%s.%s.%llu";
413 ztest_shared_t
*ztest_shared
;
415 static spa_t
*ztest_spa
= NULL
;
416 static ztest_ds_t
*ztest_ds
;
418 static kmutex_t ztest_vdev_lock
;
421 * The ztest_name_lock protects the pool and dataset namespace used by
422 * the individual tests. To modify the namespace, consumers must grab
423 * this lock as writer. Grabbing the lock as reader will ensure that the
424 * namespace does not change while the lock is held.
426 static krwlock_t ztest_name_lock
;
428 static boolean_t ztest_dump_core
= B_TRUE
;
429 static boolean_t ztest_exiting
;
431 /* Global commit callback list */
432 static ztest_cb_list_t zcl
;
433 /* Commit cb delay */
434 static uint64_t zc_min_txg_delay
= UINT64_MAX
;
435 static int zc_cb_counter
= 0;
438 * Minimum number of commit callbacks that need to be registered for us to check
439 * whether the minimum txg delay is acceptable.
441 #define ZTEST_COMMIT_CB_MIN_REG 100
444 * If a number of txgs equal to this threshold have been created after a commit
445 * callback has been registered but not called, then we assume there is an
446 * implementation bug.
448 #define ZTEST_COMMIT_CB_THRESH (TXG_CONCURRENT_STATES + 1000)
450 extern uint64_t metaslab_gang_bang
;
451 extern uint64_t metaslab_df_alloc_threshold
;
454 ZTEST_META_DNODE
= 0,
459 static void usage(boolean_t
) __NORETURN
;
462 * These libumem hooks provide a reasonable set of defaults for the allocator's
463 * debugging facilities.
466 _umem_debug_init(void)
468 return ("default,verbose"); /* $UMEM_DEBUG setting */
472 _umem_logging_init(void)
474 return ("fail,contents"); /* $UMEM_LOGGING setting */
477 #define FATAL_MSG_SZ 1024
482 fatal(int do_perror
, char *message
, ...)
485 int save_errno
= errno
;
488 (void) fflush(stdout
);
489 buf
= umem_alloc(FATAL_MSG_SZ
, UMEM_NOFAIL
);
491 va_start(args
, message
);
492 (void) sprintf(buf
, "ztest: ");
494 (void) vsprintf(buf
+ strlen(buf
), message
, args
);
497 (void) snprintf(buf
+ strlen(buf
), FATAL_MSG_SZ
- strlen(buf
),
498 ": %s", strerror(save_errno
));
500 (void) fprintf(stderr
, "%s\n", buf
);
501 fatal_msg
= buf
; /* to ease debugging */
508 str2shift(const char *buf
)
510 const char *ends
= "BKMGTPEZ";
515 for (i
= 0; i
< strlen(ends
); i
++) {
516 if (toupper(buf
[0]) == ends
[i
])
519 if (i
== strlen(ends
)) {
520 (void) fprintf(stderr
, "ztest: invalid bytes suffix: %s\n",
524 if (buf
[1] == '\0' || (toupper(buf
[1]) == 'B' && buf
[2] == '\0')) {
527 (void) fprintf(stderr
, "ztest: invalid bytes suffix: %s\n", buf
);
533 nicenumtoull(const char *buf
)
538 val
= strtoull(buf
, &end
, 0);
540 (void) fprintf(stderr
, "ztest: bad numeric value: %s\n", buf
);
542 } else if (end
[0] == '.') {
543 double fval
= strtod(buf
, &end
);
544 fval
*= pow(2, str2shift(end
));
545 if (fval
> UINT64_MAX
) {
546 (void) fprintf(stderr
, "ztest: value too large: %s\n",
550 val
= (uint64_t)fval
;
552 int shift
= str2shift(end
);
553 if (shift
>= 64 || (val
<< shift
) >> shift
!= val
) {
554 (void) fprintf(stderr
, "ztest: value too large: %s\n",
564 usage(boolean_t requested
)
566 const ztest_shared_opts_t
*zo
= &ztest_opts_defaults
;
568 char nice_vdev_size
[10];
569 char nice_gang_bang
[10];
570 FILE *fp
= requested
? stdout
: stderr
;
572 nicenum(zo
->zo_vdev_size
, nice_vdev_size
);
573 nicenum(zo
->zo_metaslab_gang_bang
, nice_gang_bang
);
575 (void) fprintf(fp
, "Usage: %s\n"
576 "\t[-v vdevs (default: %llu)]\n"
577 "\t[-s size_of_each_vdev (default: %s)]\n"
578 "\t[-a alignment_shift (default: %d)] use 0 for random\n"
579 "\t[-m mirror_copies (default: %d)]\n"
580 "\t[-r raidz_disks (default: %d)]\n"
581 "\t[-R raidz_parity (default: %d)]\n"
582 "\t[-d datasets (default: %d)]\n"
583 "\t[-t threads (default: %d)]\n"
584 "\t[-g gang_block_threshold (default: %s)]\n"
585 "\t[-i init_count (default: %d)] initialize pool i times\n"
586 "\t[-k kill_percentage (default: %llu%%)]\n"
587 "\t[-p pool_name (default: %s)]\n"
588 "\t[-f dir (default: %s)] file directory for vdev files\n"
589 "\t[-V] verbose (use multiple times for ever more blather)\n"
590 "\t[-E] use existing pool instead of creating new one\n"
591 "\t[-T time (default: %llu sec)] total run time\n"
592 "\t[-F freezeloops (default: %llu)] max loops in spa_freeze()\n"
593 "\t[-P passtime (default: %llu sec)] time per pass\n"
594 "\t[-B alt_ztest (default: <none>)] alternate ztest path\n"
595 "\t[-h] (print help)\n"
598 (u_longlong_t
)zo
->zo_vdevs
, /* -v */
599 nice_vdev_size
, /* -s */
600 zo
->zo_ashift
, /* -a */
601 zo
->zo_mirrors
, /* -m */
602 zo
->zo_raidz
, /* -r */
603 zo
->zo_raidz_parity
, /* -R */
604 zo
->zo_datasets
, /* -d */
605 zo
->zo_threads
, /* -t */
606 nice_gang_bang
, /* -g */
607 zo
->zo_init
, /* -i */
608 (u_longlong_t
)zo
->zo_killrate
, /* -k */
609 zo
->zo_pool
, /* -p */
611 (u_longlong_t
)zo
->zo_time
, /* -T */
612 (u_longlong_t
)zo
->zo_maxloops
, /* -F */
613 (u_longlong_t
)zo
->zo_passtime
);
614 exit(requested
? 0 : 1);
618 process_options(int argc
, char **argv
)
621 ztest_shared_opts_t
*zo
= &ztest_opts
;
625 char altdir
[MAXNAMELEN
] = { 0 };
627 bcopy(&ztest_opts_defaults
, zo
, sizeof (*zo
));
629 while ((opt
= getopt(argc
, argv
,
630 "v:s:a:m:r:R:d:t:g:i:k:p:f:VET:P:hF:B:")) != EOF
) {
647 value
= nicenumtoull(optarg
);
651 zo
->zo_vdevs
= value
;
654 zo
->zo_vdev_size
= MAX(SPA_MINDEVSIZE
, value
);
657 zo
->zo_ashift
= value
;
660 zo
->zo_mirrors
= value
;
663 zo
->zo_raidz
= MAX(1, value
);
666 zo
->zo_raidz_parity
= MIN(MAX(value
, 1), 3);
669 zo
->zo_datasets
= MAX(1, value
);
672 zo
->zo_threads
= MAX(1, value
);
675 zo
->zo_metaslab_gang_bang
= MAX(SPA_MINBLOCKSIZE
<< 1,
682 zo
->zo_killrate
= value
;
685 (void) strlcpy(zo
->zo_pool
, optarg
,
686 sizeof (zo
->zo_pool
));
689 path
= realpath(optarg
, NULL
);
691 (void) fprintf(stderr
, "error: %s: %s\n",
692 optarg
, strerror(errno
));
695 (void) strlcpy(zo
->zo_dir
, path
,
696 sizeof (zo
->zo_dir
));
709 zo
->zo_passtime
= MAX(1, value
);
712 zo
->zo_maxloops
= MAX(1, value
);
715 (void) strlcpy(altdir
, optarg
, sizeof (altdir
));
727 zo
->zo_raidz_parity
= MIN(zo
->zo_raidz_parity
, zo
->zo_raidz
- 1);
730 (zo
->zo_vdevs
> 0 ? zo
->zo_time
* NANOSEC
/ zo
->zo_vdevs
:
733 if (strlen(altdir
) > 0) {
741 cmd
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
742 realaltdir
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
744 VERIFY(NULL
!= realpath(getexecname(), cmd
));
745 if (0 != access(altdir
, F_OK
)) {
746 ztest_dump_core
= B_FALSE
;
747 fatal(B_TRUE
, "invalid alternate ztest path: %s",
750 VERIFY(NULL
!= realpath(altdir
, realaltdir
));
753 * 'cmd' should be of the form "<anything>/usr/bin/<isa>/ztest".
754 * We want to extract <isa> to determine if we should use
755 * 32 or 64 bit binaries.
757 bin
= strstr(cmd
, "/usr/bin/");
758 ztest
= strstr(bin
, "/ztest");
760 isalen
= ztest
- isa
;
761 (void) snprintf(zo
->zo_alt_ztest
, sizeof (zo
->zo_alt_ztest
),
762 "%s/usr/bin/%.*s/ztest", realaltdir
, isalen
, isa
);
763 (void) snprintf(zo
->zo_alt_libpath
, sizeof (zo
->zo_alt_libpath
),
764 "%s/usr/lib/%.*s", realaltdir
, isalen
, isa
);
766 if (0 != access(zo
->zo_alt_ztest
, X_OK
)) {
767 ztest_dump_core
= B_FALSE
;
768 fatal(B_TRUE
, "invalid alternate ztest: %s",
770 } else if (0 != access(zo
->zo_alt_libpath
, X_OK
)) {
771 ztest_dump_core
= B_FALSE
;
772 fatal(B_TRUE
, "invalid alternate lib directory %s",
776 umem_free(cmd
, MAXPATHLEN
);
777 umem_free(realaltdir
, MAXPATHLEN
);
782 ztest_kill(ztest_shared_t
*zs
)
784 zs
->zs_alloc
= metaslab_class_get_alloc(spa_normal_class(ztest_spa
));
785 zs
->zs_space
= metaslab_class_get_space(spa_normal_class(ztest_spa
));
786 (void) kill(getpid(), SIGKILL
);
790 ztest_random(uint64_t range
)
794 ASSERT3S(ztest_fd_rand
, >=, 0);
799 if (read(ztest_fd_rand
, &r
, sizeof (r
)) != sizeof (r
))
800 fatal(1, "short read from /dev/urandom");
807 ztest_record_enospc(const char *s
)
809 ztest_shared
->zs_enospc_count
++;
813 ztest_get_ashift(void)
815 if (ztest_opts
.zo_ashift
== 0)
816 return (SPA_MINBLOCKSHIFT
+ ztest_random(3));
817 return (ztest_opts
.zo_ashift
);
821 make_vdev_file(char *path
, char *aux
, size_t size
, uint64_t ashift
)
827 pathbuf
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
830 ashift
= ztest_get_ashift();
836 vdev
= ztest_shared
->zs_vdev_aux
;
837 (void) snprintf(path
, MAXPATHLEN
,
838 ztest_aux_template
, ztest_opts
.zo_dir
,
839 ztest_opts
.zo_pool
, aux
, vdev
);
841 vdev
= ztest_shared
->zs_vdev_next_leaf
++;
842 (void) snprintf(path
, MAXPATHLEN
,
843 ztest_dev_template
, ztest_opts
.zo_dir
,
844 ztest_opts
.zo_pool
, vdev
);
849 int fd
= open(path
, O_RDWR
| O_CREAT
| O_TRUNC
, 0666);
851 fatal(1, "can't open %s", path
);
852 if (ftruncate(fd
, size
) != 0)
853 fatal(1, "can't ftruncate %s", path
);
857 VERIFY(nvlist_alloc(&file
, NV_UNIQUE_NAME
, 0) == 0);
858 VERIFY(nvlist_add_string(file
, ZPOOL_CONFIG_TYPE
, VDEV_TYPE_FILE
) == 0);
859 VERIFY(nvlist_add_string(file
, ZPOOL_CONFIG_PATH
, path
) == 0);
860 VERIFY(nvlist_add_uint64(file
, ZPOOL_CONFIG_ASHIFT
, ashift
) == 0);
861 umem_free(pathbuf
, MAXPATHLEN
);
867 make_vdev_raidz(char *path
, char *aux
, size_t size
, uint64_t ashift
, int r
)
869 nvlist_t
*raidz
, **child
;
873 return (make_vdev_file(path
, aux
, size
, ashift
));
874 child
= umem_alloc(r
* sizeof (nvlist_t
*), UMEM_NOFAIL
);
876 for (c
= 0; c
< r
; c
++)
877 child
[c
] = make_vdev_file(path
, aux
, size
, ashift
);
879 VERIFY(nvlist_alloc(&raidz
, NV_UNIQUE_NAME
, 0) == 0);
880 VERIFY(nvlist_add_string(raidz
, ZPOOL_CONFIG_TYPE
,
881 VDEV_TYPE_RAIDZ
) == 0);
882 VERIFY(nvlist_add_uint64(raidz
, ZPOOL_CONFIG_NPARITY
,
883 ztest_opts
.zo_raidz_parity
) == 0);
884 VERIFY(nvlist_add_nvlist_array(raidz
, ZPOOL_CONFIG_CHILDREN
,
887 for (c
= 0; c
< r
; c
++)
888 nvlist_free(child
[c
]);
890 umem_free(child
, r
* sizeof (nvlist_t
*));
896 make_vdev_mirror(char *path
, char *aux
, size_t size
, uint64_t ashift
,
899 nvlist_t
*mirror
, **child
;
903 return (make_vdev_raidz(path
, aux
, size
, ashift
, r
));
905 child
= umem_alloc(m
* sizeof (nvlist_t
*), UMEM_NOFAIL
);
907 for (c
= 0; c
< m
; c
++)
908 child
[c
] = make_vdev_raidz(path
, aux
, size
, ashift
, r
);
910 VERIFY(nvlist_alloc(&mirror
, NV_UNIQUE_NAME
, 0) == 0);
911 VERIFY(nvlist_add_string(mirror
, ZPOOL_CONFIG_TYPE
,
912 VDEV_TYPE_MIRROR
) == 0);
913 VERIFY(nvlist_add_nvlist_array(mirror
, ZPOOL_CONFIG_CHILDREN
,
916 for (c
= 0; c
< m
; c
++)
917 nvlist_free(child
[c
]);
919 umem_free(child
, m
* sizeof (nvlist_t
*));
925 make_vdev_root(char *path
, char *aux
, size_t size
, uint64_t ashift
,
926 int log
, int r
, int m
, int t
)
928 nvlist_t
*root
, **child
;
933 child
= umem_alloc(t
* sizeof (nvlist_t
*), UMEM_NOFAIL
);
935 for (c
= 0; c
< t
; c
++) {
936 child
[c
] = make_vdev_mirror(path
, aux
, size
, ashift
, r
, m
);
937 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_IS_LOG
,
941 VERIFY(nvlist_alloc(&root
, NV_UNIQUE_NAME
, 0) == 0);
942 VERIFY(nvlist_add_string(root
, ZPOOL_CONFIG_TYPE
, VDEV_TYPE_ROOT
) == 0);
943 VERIFY(nvlist_add_nvlist_array(root
, aux
? aux
: ZPOOL_CONFIG_CHILDREN
,
946 for (c
= 0; c
< t
; c
++)
947 nvlist_free(child
[c
]);
949 umem_free(child
, t
* sizeof (nvlist_t
*));
955 ztest_random_blocksize(void)
957 return (1 << (SPA_MINBLOCKSHIFT
+
958 ztest_random(SPA_MAXBLOCKSHIFT
- SPA_MINBLOCKSHIFT
+ 1)));
962 ztest_random_ibshift(void)
964 return (DN_MIN_INDBLKSHIFT
+
965 ztest_random(DN_MAX_INDBLKSHIFT
- DN_MIN_INDBLKSHIFT
+ 1));
969 ztest_random_vdev_top(spa_t
*spa
, boolean_t log_ok
)
972 vdev_t
*rvd
= spa
->spa_root_vdev
;
975 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_READER
) != 0);
978 top
= ztest_random(rvd
->vdev_children
);
979 tvd
= rvd
->vdev_child
[top
];
980 } while (tvd
->vdev_ishole
|| (tvd
->vdev_islog
&& !log_ok
) ||
981 tvd
->vdev_mg
== NULL
|| tvd
->vdev_mg
->mg_class
== NULL
);
987 ztest_random_dsl_prop(zfs_prop_t prop
)
992 value
= zfs_prop_random_value(prop
, ztest_random(-1ULL));
993 } while (prop
== ZFS_PROP_CHECKSUM
&& value
== ZIO_CHECKSUM_OFF
);
999 ztest_dsl_prop_set_uint64(char *osname
, zfs_prop_t prop
, uint64_t value
,
1002 const char *propname
= zfs_prop_to_name(prop
);
1003 const char *valname
;
1008 error
= dsl_prop_set(osname
, propname
,
1009 (inherit
? ZPROP_SRC_NONE
: ZPROP_SRC_LOCAL
),
1010 sizeof (value
), 1, &value
);
1012 if (error
== ENOSPC
) {
1013 ztest_record_enospc(FTAG
);
1016 ASSERT3U(error
, ==, 0);
1018 setpoint
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
1019 VERIFY3U(dsl_prop_get(osname
, propname
, sizeof (curval
),
1020 1, &curval
, setpoint
), ==, 0);
1022 if (ztest_opts
.zo_verbose
>= 6) {
1023 VERIFY(zfs_prop_index_to_string(prop
, curval
, &valname
) == 0);
1024 (void) printf("%s %s = %s at '%s'\n",
1025 osname
, propname
, valname
, setpoint
);
1027 umem_free(setpoint
, MAXPATHLEN
);
1033 ztest_spa_prop_set_uint64(zpool_prop_t prop
, uint64_t value
)
1035 spa_t
*spa
= ztest_spa
;
1036 nvlist_t
*props
= NULL
;
1039 VERIFY(nvlist_alloc(&props
, NV_UNIQUE_NAME
, 0) == 0);
1040 VERIFY(nvlist_add_uint64(props
, zpool_prop_to_name(prop
), value
) == 0);
1042 error
= spa_prop_set(spa
, props
);
1046 if (error
== ENOSPC
) {
1047 ztest_record_enospc(FTAG
);
1050 ASSERT3U(error
, ==, 0);
1056 ztest_rll_init(rll_t
*rll
)
1058 rll
->rll_writer
= NULL
;
1059 rll
->rll_readers
= 0;
1060 mutex_init(&rll
->rll_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
1061 cv_init(&rll
->rll_cv
, NULL
, CV_DEFAULT
, NULL
);
1065 ztest_rll_destroy(rll_t
*rll
)
1067 ASSERT(rll
->rll_writer
== NULL
);
1068 ASSERT(rll
->rll_readers
== 0);
1069 mutex_destroy(&rll
->rll_lock
);
1070 cv_destroy(&rll
->rll_cv
);
1074 ztest_rll_lock(rll_t
*rll
, rl_type_t type
)
1076 mutex_enter(&rll
->rll_lock
);
1078 if (type
== RL_READER
) {
1079 while (rll
->rll_writer
!= NULL
)
1080 (void) cv_wait(&rll
->rll_cv
, &rll
->rll_lock
);
1083 while (rll
->rll_writer
!= NULL
|| rll
->rll_readers
)
1084 (void) cv_wait(&rll
->rll_cv
, &rll
->rll_lock
);
1085 rll
->rll_writer
= curthread
;
1088 mutex_exit(&rll
->rll_lock
);
1092 ztest_rll_unlock(rll_t
*rll
)
1094 mutex_enter(&rll
->rll_lock
);
1096 if (rll
->rll_writer
) {
1097 ASSERT(rll
->rll_readers
== 0);
1098 rll
->rll_writer
= NULL
;
1100 ASSERT(rll
->rll_readers
!= 0);
1101 ASSERT(rll
->rll_writer
== NULL
);
1105 if (rll
->rll_writer
== NULL
&& rll
->rll_readers
== 0)
1106 cv_broadcast(&rll
->rll_cv
);
1108 mutex_exit(&rll
->rll_lock
);
1112 ztest_object_lock(ztest_ds_t
*zd
, uint64_t object
, rl_type_t type
)
1114 rll_t
*rll
= &zd
->zd_object_lock
[object
& (ZTEST_OBJECT_LOCKS
- 1)];
1116 ztest_rll_lock(rll
, type
);
1120 ztest_object_unlock(ztest_ds_t
*zd
, uint64_t object
)
1122 rll_t
*rll
= &zd
->zd_object_lock
[object
& (ZTEST_OBJECT_LOCKS
- 1)];
1124 ztest_rll_unlock(rll
);
1128 ztest_range_lock(ztest_ds_t
*zd
, uint64_t object
, uint64_t offset
,
1129 uint64_t size
, rl_type_t type
)
1131 uint64_t hash
= object
^ (offset
% (ZTEST_RANGE_LOCKS
+ 1));
1132 rll_t
*rll
= &zd
->zd_range_lock
[hash
& (ZTEST_RANGE_LOCKS
- 1)];
1135 rl
= umem_alloc(sizeof (*rl
), UMEM_NOFAIL
);
1136 rl
->rl_object
= object
;
1137 rl
->rl_offset
= offset
;
1141 ztest_rll_lock(rll
, type
);
1147 ztest_range_unlock(rl_t
*rl
)
1149 rll_t
*rll
= rl
->rl_lock
;
1151 ztest_rll_unlock(rll
);
1153 umem_free(rl
, sizeof (*rl
));
1157 ztest_zd_init(ztest_ds_t
*zd
, ztest_shared_ds_t
*szd
, objset_t
*os
)
1160 zd
->zd_zilog
= dmu_objset_zil(os
);
1161 zd
->zd_shared
= szd
;
1162 dmu_objset_name(os
, zd
->zd_name
);
1165 if (zd
->zd_shared
!= NULL
)
1166 zd
->zd_shared
->zd_seq
= 0;
1168 rw_init(&zd
->zd_zilog_lock
, NULL
, RW_DEFAULT
, NULL
);
1169 mutex_init(&zd
->zd_dirobj_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
1171 for (l
= 0; l
< ZTEST_OBJECT_LOCKS
; l
++)
1172 ztest_rll_init(&zd
->zd_object_lock
[l
]);
1174 for (l
= 0; l
< ZTEST_RANGE_LOCKS
; l
++)
1175 ztest_rll_init(&zd
->zd_range_lock
[l
]);
1179 ztest_zd_fini(ztest_ds_t
*zd
)
1183 mutex_destroy(&zd
->zd_dirobj_lock
);
1184 rw_destroy(&zd
->zd_zilog_lock
);
1186 for (l
= 0; l
< ZTEST_OBJECT_LOCKS
; l
++)
1187 ztest_rll_destroy(&zd
->zd_object_lock
[l
]);
1189 for (l
= 0; l
< ZTEST_RANGE_LOCKS
; l
++)
1190 ztest_rll_destroy(&zd
->zd_range_lock
[l
]);
1193 #define TXG_MIGHTWAIT (ztest_random(10) == 0 ? TXG_NOWAIT : TXG_WAIT)
1196 ztest_tx_assign(dmu_tx_t
*tx
, uint64_t txg_how
, const char *tag
)
1202 * Attempt to assign tx to some transaction group.
1204 error
= dmu_tx_assign(tx
, txg_how
);
1206 if (error
== ERESTART
) {
1207 ASSERT(txg_how
== TXG_NOWAIT
);
1210 ASSERT3U(error
, ==, ENOSPC
);
1211 ztest_record_enospc(tag
);
1216 txg
= dmu_tx_get_txg(tx
);
1222 ztest_pattern_set(void *buf
, uint64_t size
, uint64_t value
)
1225 uint64_t *ip_end
= (uint64_t *)((uintptr_t)buf
+ (uintptr_t)size
);
1233 ztest_pattern_match(void *buf
, uint64_t size
, uint64_t value
)
1236 uint64_t *ip_end
= (uint64_t *)((uintptr_t)buf
+ (uintptr_t)size
);
1240 diff
|= (value
- *ip
++);
1247 ztest_bt_generate(ztest_block_tag_t
*bt
, objset_t
*os
, uint64_t object
,
1248 uint64_t offset
, uint64_t gen
, uint64_t txg
, uint64_t crtxg
)
1250 bt
->bt_magic
= BT_MAGIC
;
1251 bt
->bt_objset
= dmu_objset_id(os
);
1252 bt
->bt_object
= object
;
1253 bt
->bt_offset
= offset
;
1256 bt
->bt_crtxg
= crtxg
;
1260 ztest_bt_verify(ztest_block_tag_t
*bt
, objset_t
*os
, uint64_t object
,
1261 uint64_t offset
, uint64_t gen
, uint64_t txg
, uint64_t crtxg
)
1263 ASSERT(bt
->bt_magic
== BT_MAGIC
);
1264 ASSERT(bt
->bt_objset
== dmu_objset_id(os
));
1265 ASSERT(bt
->bt_object
== object
);
1266 ASSERT(bt
->bt_offset
== offset
);
1267 ASSERT(bt
->bt_gen
<= gen
);
1268 ASSERT(bt
->bt_txg
<= txg
);
1269 ASSERT(bt
->bt_crtxg
== crtxg
);
1272 static ztest_block_tag_t
*
1273 ztest_bt_bonus(dmu_buf_t
*db
)
1275 dmu_object_info_t doi
;
1276 ztest_block_tag_t
*bt
;
1278 dmu_object_info_from_db(db
, &doi
);
1279 ASSERT3U(doi
.doi_bonus_size
, <=, db
->db_size
);
1280 ASSERT3U(doi
.doi_bonus_size
, >=, sizeof (*bt
));
1281 bt
= (void *)((char *)db
->db_data
+ doi
.doi_bonus_size
- sizeof (*bt
));
1290 #define lrz_type lr_mode
1291 #define lrz_blocksize lr_uid
1292 #define lrz_ibshift lr_gid
1293 #define lrz_bonustype lr_rdev
1294 #define lrz_bonuslen lr_crtime[1]
1297 ztest_log_create(ztest_ds_t
*zd
, dmu_tx_t
*tx
, lr_create_t
*lr
)
1299 char *name
= (void *)(lr
+ 1); /* name follows lr */
1300 size_t namesize
= strlen(name
) + 1;
1303 if (zil_replaying(zd
->zd_zilog
, tx
))
1306 itx
= zil_itx_create(TX_CREATE
, sizeof (*lr
) + namesize
);
1307 bcopy(&lr
->lr_common
+ 1, &itx
->itx_lr
+ 1,
1308 sizeof (*lr
) + namesize
- sizeof (lr_t
));
1310 zil_itx_assign(zd
->zd_zilog
, itx
, tx
);
1314 ztest_log_remove(ztest_ds_t
*zd
, dmu_tx_t
*tx
, lr_remove_t
*lr
, uint64_t object
)
1316 char *name
= (void *)(lr
+ 1); /* name follows lr */
1317 size_t namesize
= strlen(name
) + 1;
1320 if (zil_replaying(zd
->zd_zilog
, tx
))
1323 itx
= zil_itx_create(TX_REMOVE
, sizeof (*lr
) + namesize
);
1324 bcopy(&lr
->lr_common
+ 1, &itx
->itx_lr
+ 1,
1325 sizeof (*lr
) + namesize
- sizeof (lr_t
));
1327 itx
->itx_oid
= object
;
1328 zil_itx_assign(zd
->zd_zilog
, itx
, tx
);
1332 ztest_log_write(ztest_ds_t
*zd
, dmu_tx_t
*tx
, lr_write_t
*lr
)
1335 itx_wr_state_t write_state
= ztest_random(WR_NUM_STATES
);
1337 if (zil_replaying(zd
->zd_zilog
, tx
))
1340 if (lr
->lr_length
> ZIL_MAX_LOG_DATA
)
1341 write_state
= WR_INDIRECT
;
1343 itx
= zil_itx_create(TX_WRITE
,
1344 sizeof (*lr
) + (write_state
== WR_COPIED
? lr
->lr_length
: 0));
1346 if (write_state
== WR_COPIED
&&
1347 dmu_read(zd
->zd_os
, lr
->lr_foid
, lr
->lr_offset
, lr
->lr_length
,
1348 ((lr_write_t
*)&itx
->itx_lr
) + 1, DMU_READ_NO_PREFETCH
) != 0) {
1349 zil_itx_destroy(itx
);
1350 itx
= zil_itx_create(TX_WRITE
, sizeof (*lr
));
1351 write_state
= WR_NEED_COPY
;
1353 itx
->itx_private
= zd
;
1354 itx
->itx_wr_state
= write_state
;
1355 itx
->itx_sync
= (ztest_random(8) == 0);
1356 itx
->itx_sod
+= (write_state
== WR_NEED_COPY
? lr
->lr_length
: 0);
1358 bcopy(&lr
->lr_common
+ 1, &itx
->itx_lr
+ 1,
1359 sizeof (*lr
) - sizeof (lr_t
));
1361 zil_itx_assign(zd
->zd_zilog
, itx
, tx
);
1365 ztest_log_truncate(ztest_ds_t
*zd
, dmu_tx_t
*tx
, lr_truncate_t
*lr
)
1369 if (zil_replaying(zd
->zd_zilog
, tx
))
1372 itx
= zil_itx_create(TX_TRUNCATE
, sizeof (*lr
));
1373 bcopy(&lr
->lr_common
+ 1, &itx
->itx_lr
+ 1,
1374 sizeof (*lr
) - sizeof (lr_t
));
1376 itx
->itx_sync
= B_FALSE
;
1377 zil_itx_assign(zd
->zd_zilog
, itx
, tx
);
1381 ztest_log_setattr(ztest_ds_t
*zd
, dmu_tx_t
*tx
, lr_setattr_t
*lr
)
1385 if (zil_replaying(zd
->zd_zilog
, tx
))
1388 itx
= zil_itx_create(TX_SETATTR
, sizeof (*lr
));
1389 bcopy(&lr
->lr_common
+ 1, &itx
->itx_lr
+ 1,
1390 sizeof (*lr
) - sizeof (lr_t
));
1392 itx
->itx_sync
= B_FALSE
;
1393 zil_itx_assign(zd
->zd_zilog
, itx
, tx
);
1400 ztest_replay_create(ztest_ds_t
*zd
, lr_create_t
*lr
, boolean_t byteswap
)
1402 char *name
= (void *)(lr
+ 1); /* name follows lr */
1403 objset_t
*os
= zd
->zd_os
;
1404 ztest_block_tag_t
*bbt
;
1411 byteswap_uint64_array(lr
, sizeof (*lr
));
1413 ASSERT(lr
->lr_doid
== ZTEST_DIROBJ
);
1414 ASSERT(name
[0] != '\0');
1416 tx
= dmu_tx_create(os
);
1418 dmu_tx_hold_zap(tx
, lr
->lr_doid
, B_TRUE
, name
);
1420 if (lr
->lrz_type
== DMU_OT_ZAP_OTHER
) {
1421 dmu_tx_hold_zap(tx
, DMU_NEW_OBJECT
, B_TRUE
, NULL
);
1423 dmu_tx_hold_bonus(tx
, DMU_NEW_OBJECT
);
1426 txg
= ztest_tx_assign(tx
, TXG_WAIT
, FTAG
);
1430 ASSERT(dmu_objset_zil(os
)->zl_replay
== !!lr
->lr_foid
);
1432 if (lr
->lrz_type
== DMU_OT_ZAP_OTHER
) {
1433 if (lr
->lr_foid
== 0) {
1434 lr
->lr_foid
= zap_create(os
,
1435 lr
->lrz_type
, lr
->lrz_bonustype
,
1436 lr
->lrz_bonuslen
, tx
);
1438 error
= zap_create_claim(os
, lr
->lr_foid
,
1439 lr
->lrz_type
, lr
->lrz_bonustype
,
1440 lr
->lrz_bonuslen
, tx
);
1443 if (lr
->lr_foid
== 0) {
1444 lr
->lr_foid
= dmu_object_alloc(os
,
1445 lr
->lrz_type
, 0, lr
->lrz_bonustype
,
1446 lr
->lrz_bonuslen
, tx
);
1448 error
= dmu_object_claim(os
, lr
->lr_foid
,
1449 lr
->lrz_type
, 0, lr
->lrz_bonustype
,
1450 lr
->lrz_bonuslen
, tx
);
1455 ASSERT3U(error
, ==, EEXIST
);
1456 ASSERT(zd
->zd_zilog
->zl_replay
);
1461 ASSERT(lr
->lr_foid
!= 0);
1463 if (lr
->lrz_type
!= DMU_OT_ZAP_OTHER
)
1464 VERIFY3U(0, ==, dmu_object_set_blocksize(os
, lr
->lr_foid
,
1465 lr
->lrz_blocksize
, lr
->lrz_ibshift
, tx
));
1467 VERIFY3U(0, ==, dmu_bonus_hold(os
, lr
->lr_foid
, FTAG
, &db
));
1468 bbt
= ztest_bt_bonus(db
);
1469 dmu_buf_will_dirty(db
, tx
);
1470 ztest_bt_generate(bbt
, os
, lr
->lr_foid
, -1ULL, lr
->lr_gen
, txg
, txg
);
1471 dmu_buf_rele(db
, FTAG
);
1473 VERIFY3U(0, ==, zap_add(os
, lr
->lr_doid
, name
, sizeof (uint64_t), 1,
1476 (void) ztest_log_create(zd
, tx
, lr
);
1484 ztest_replay_remove(ztest_ds_t
*zd
, lr_remove_t
*lr
, boolean_t byteswap
)
1486 char *name
= (void *)(lr
+ 1); /* name follows lr */
1487 objset_t
*os
= zd
->zd_os
;
1488 dmu_object_info_t doi
;
1490 uint64_t object
, txg
;
1493 byteswap_uint64_array(lr
, sizeof (*lr
));
1495 ASSERT(lr
->lr_doid
== ZTEST_DIROBJ
);
1496 ASSERT(name
[0] != '\0');
1499 zap_lookup(os
, lr
->lr_doid
, name
, sizeof (object
), 1, &object
));
1500 ASSERT(object
!= 0);
1502 ztest_object_lock(zd
, object
, RL_WRITER
);
1504 VERIFY3U(0, ==, dmu_object_info(os
, object
, &doi
));
1506 tx
= dmu_tx_create(os
);
1508 dmu_tx_hold_zap(tx
, lr
->lr_doid
, B_FALSE
, name
);
1509 dmu_tx_hold_free(tx
, object
, 0, DMU_OBJECT_END
);
1511 txg
= ztest_tx_assign(tx
, TXG_WAIT
, FTAG
);
1513 ztest_object_unlock(zd
, object
);
1517 if (doi
.doi_type
== DMU_OT_ZAP_OTHER
) {
1518 VERIFY3U(0, ==, zap_destroy(os
, object
, tx
));
1520 VERIFY3U(0, ==, dmu_object_free(os
, object
, tx
));
1523 VERIFY3U(0, ==, zap_remove(os
, lr
->lr_doid
, name
, tx
));
1525 (void) ztest_log_remove(zd
, tx
, lr
, object
);
1529 ztest_object_unlock(zd
, object
);
1535 ztest_replay_write(ztest_ds_t
*zd
, lr_write_t
*lr
, boolean_t byteswap
)
1537 objset_t
*os
= zd
->zd_os
;
1538 void *data
= lr
+ 1; /* data follows lr */
1539 uint64_t offset
, length
;
1540 ztest_block_tag_t
*bt
= data
;
1541 ztest_block_tag_t
*bbt
;
1542 uint64_t gen
, txg
, lrtxg
, crtxg
;
1543 dmu_object_info_t doi
;
1546 arc_buf_t
*abuf
= NULL
;
1550 byteswap_uint64_array(lr
, sizeof (*lr
));
1552 offset
= lr
->lr_offset
;
1553 length
= lr
->lr_length
;
1555 /* If it's a dmu_sync() block, write the whole block */
1556 if (lr
->lr_common
.lrc_reclen
== sizeof (lr_write_t
)) {
1557 uint64_t blocksize
= BP_GET_LSIZE(&lr
->lr_blkptr
);
1558 if (length
< blocksize
) {
1559 offset
-= offset
% blocksize
;
1564 if (bt
->bt_magic
== BSWAP_64(BT_MAGIC
))
1565 byteswap_uint64_array(bt
, sizeof (*bt
));
1567 if (bt
->bt_magic
!= BT_MAGIC
)
1570 ztest_object_lock(zd
, lr
->lr_foid
, RL_READER
);
1571 rl
= ztest_range_lock(zd
, lr
->lr_foid
, offset
, length
, RL_WRITER
);
1573 VERIFY3U(0, ==, dmu_bonus_hold(os
, lr
->lr_foid
, FTAG
, &db
));
1575 dmu_object_info_from_db(db
, &doi
);
1577 bbt
= ztest_bt_bonus(db
);
1578 ASSERT3U(bbt
->bt_magic
, ==, BT_MAGIC
);
1580 crtxg
= bbt
->bt_crtxg
;
1581 lrtxg
= lr
->lr_common
.lrc_txg
;
1583 tx
= dmu_tx_create(os
);
1585 dmu_tx_hold_write(tx
, lr
->lr_foid
, offset
, length
);
1587 if (ztest_random(8) == 0 && length
== doi
.doi_data_block_size
&&
1588 P2PHASE(offset
, length
) == 0)
1589 abuf
= dmu_request_arcbuf(db
, length
);
1591 txg
= ztest_tx_assign(tx
, TXG_WAIT
, FTAG
);
1594 dmu_return_arcbuf(abuf
);
1595 dmu_buf_rele(db
, FTAG
);
1596 ztest_range_unlock(rl
);
1597 ztest_object_unlock(zd
, lr
->lr_foid
);
1603 * Usually, verify the old data before writing new data --
1604 * but not always, because we also want to verify correct
1605 * behavior when the data was not recently read into cache.
1607 ASSERT(offset
% doi
.doi_data_block_size
== 0);
1608 if (ztest_random(4) != 0) {
1609 int prefetch
= ztest_random(2) ?
1610 DMU_READ_PREFETCH
: DMU_READ_NO_PREFETCH
;
1611 ztest_block_tag_t rbt
;
1613 VERIFY(dmu_read(os
, lr
->lr_foid
, offset
,
1614 sizeof (rbt
), &rbt
, prefetch
) == 0);
1615 if (rbt
.bt_magic
== BT_MAGIC
) {
1616 ztest_bt_verify(&rbt
, os
, lr
->lr_foid
,
1617 offset
, gen
, txg
, crtxg
);
1622 * Writes can appear to be newer than the bonus buffer because
1623 * the ztest_get_data() callback does a dmu_read() of the
1624 * open-context data, which may be different than the data
1625 * as it was when the write was generated.
1627 if (zd
->zd_zilog
->zl_replay
) {
1628 ztest_bt_verify(bt
, os
, lr
->lr_foid
, offset
,
1629 MAX(gen
, bt
->bt_gen
), MAX(txg
, lrtxg
),
1634 * Set the bt's gen/txg to the bonus buffer's gen/txg
1635 * so that all of the usual ASSERTs will work.
1637 ztest_bt_generate(bt
, os
, lr
->lr_foid
, offset
, gen
, txg
, crtxg
);
1641 dmu_write(os
, lr
->lr_foid
, offset
, length
, data
, tx
);
1643 bcopy(data
, abuf
->b_data
, length
);
1644 dmu_assign_arcbuf(db
, offset
, abuf
, tx
);
1647 (void) ztest_log_write(zd
, tx
, lr
);
1649 dmu_buf_rele(db
, FTAG
);
1653 ztest_range_unlock(rl
);
1654 ztest_object_unlock(zd
, lr
->lr_foid
);
1660 ztest_replay_truncate(ztest_ds_t
*zd
, lr_truncate_t
*lr
, boolean_t byteswap
)
1662 objset_t
*os
= zd
->zd_os
;
1668 byteswap_uint64_array(lr
, sizeof (*lr
));
1670 ztest_object_lock(zd
, lr
->lr_foid
, RL_READER
);
1671 rl
= ztest_range_lock(zd
, lr
->lr_foid
, lr
->lr_offset
, lr
->lr_length
,
1674 tx
= dmu_tx_create(os
);
1676 dmu_tx_hold_free(tx
, lr
->lr_foid
, lr
->lr_offset
, lr
->lr_length
);
1678 txg
= ztest_tx_assign(tx
, TXG_WAIT
, FTAG
);
1680 ztest_range_unlock(rl
);
1681 ztest_object_unlock(zd
, lr
->lr_foid
);
1685 VERIFY(dmu_free_range(os
, lr
->lr_foid
, lr
->lr_offset
,
1686 lr
->lr_length
, tx
) == 0);
1688 (void) ztest_log_truncate(zd
, tx
, lr
);
1692 ztest_range_unlock(rl
);
1693 ztest_object_unlock(zd
, lr
->lr_foid
);
1699 ztest_replay_setattr(ztest_ds_t
*zd
, lr_setattr_t
*lr
, boolean_t byteswap
)
1701 objset_t
*os
= zd
->zd_os
;
1704 ztest_block_tag_t
*bbt
;
1705 uint64_t txg
, lrtxg
, crtxg
;
1708 byteswap_uint64_array(lr
, sizeof (*lr
));
1710 ztest_object_lock(zd
, lr
->lr_foid
, RL_WRITER
);
1712 VERIFY3U(0, ==, dmu_bonus_hold(os
, lr
->lr_foid
, FTAG
, &db
));
1714 tx
= dmu_tx_create(os
);
1715 dmu_tx_hold_bonus(tx
, lr
->lr_foid
);
1717 txg
= ztest_tx_assign(tx
, TXG_WAIT
, FTAG
);
1719 dmu_buf_rele(db
, FTAG
);
1720 ztest_object_unlock(zd
, lr
->lr_foid
);
1724 bbt
= ztest_bt_bonus(db
);
1725 ASSERT3U(bbt
->bt_magic
, ==, BT_MAGIC
);
1726 crtxg
= bbt
->bt_crtxg
;
1727 lrtxg
= lr
->lr_common
.lrc_txg
;
1729 if (zd
->zd_zilog
->zl_replay
) {
1730 ASSERT(lr
->lr_size
!= 0);
1731 ASSERT(lr
->lr_mode
!= 0);
1735 * Randomly change the size and increment the generation.
1737 lr
->lr_size
= (ztest_random(db
->db_size
/ sizeof (*bbt
)) + 1) *
1739 lr
->lr_mode
= bbt
->bt_gen
+ 1;
1744 * Verify that the current bonus buffer is not newer than our txg.
1746 ztest_bt_verify(bbt
, os
, lr
->lr_foid
, -1ULL, lr
->lr_mode
,
1747 MAX(txg
, lrtxg
), crtxg
);
1749 dmu_buf_will_dirty(db
, tx
);
1751 ASSERT3U(lr
->lr_size
, >=, sizeof (*bbt
));
1752 ASSERT3U(lr
->lr_size
, <=, db
->db_size
);
1753 VERIFY3U(dmu_set_bonus(db
, lr
->lr_size
, tx
), ==, 0);
1754 bbt
= ztest_bt_bonus(db
);
1756 ztest_bt_generate(bbt
, os
, lr
->lr_foid
, -1ULL, lr
->lr_mode
, txg
, crtxg
);
1758 dmu_buf_rele(db
, FTAG
);
1760 (void) ztest_log_setattr(zd
, tx
, lr
);
1764 ztest_object_unlock(zd
, lr
->lr_foid
);
1769 zil_replay_func_t
*ztest_replay_vector
[TX_MAX_TYPE
] = {
1770 NULL
, /* 0 no such transaction type */
1771 (zil_replay_func_t
*)ztest_replay_create
, /* TX_CREATE */
1772 NULL
, /* TX_MKDIR */
1773 NULL
, /* TX_MKXATTR */
1774 NULL
, /* TX_SYMLINK */
1775 (zil_replay_func_t
*)ztest_replay_remove
, /* TX_REMOVE */
1776 NULL
, /* TX_RMDIR */
1778 NULL
, /* TX_RENAME */
1779 (zil_replay_func_t
*)ztest_replay_write
, /* TX_WRITE */
1780 (zil_replay_func_t
*)ztest_replay_truncate
, /* TX_TRUNCATE */
1781 (zil_replay_func_t
*)ztest_replay_setattr
, /* TX_SETATTR */
1783 NULL
, /* TX_CREATE_ACL */
1784 NULL
, /* TX_CREATE_ATTR */
1785 NULL
, /* TX_CREATE_ACL_ATTR */
1786 NULL
, /* TX_MKDIR_ACL */
1787 NULL
, /* TX_MKDIR_ATTR */
1788 NULL
, /* TX_MKDIR_ACL_ATTR */
1789 NULL
, /* TX_WRITE2 */
1793 * ZIL get_data callbacks
1797 ztest_get_done(zgd_t
*zgd
, int error
)
1799 ztest_ds_t
*zd
= zgd
->zgd_private
;
1800 uint64_t object
= zgd
->zgd_rl
->rl_object
;
1803 dmu_buf_rele(zgd
->zgd_db
, zgd
);
1805 ztest_range_unlock(zgd
->zgd_rl
);
1806 ztest_object_unlock(zd
, object
);
1808 if (error
== 0 && zgd
->zgd_bp
)
1809 zil_add_block(zgd
->zgd_zilog
, zgd
->zgd_bp
);
1811 umem_free(zgd
, sizeof (*zgd
));
1815 ztest_get_data(void *arg
, lr_write_t
*lr
, char *buf
, zio_t
*zio
)
1817 ztest_ds_t
*zd
= arg
;
1818 objset_t
*os
= zd
->zd_os
;
1819 uint64_t object
= lr
->lr_foid
;
1820 uint64_t offset
= lr
->lr_offset
;
1821 uint64_t size
= lr
->lr_length
;
1822 blkptr_t
*bp
= &lr
->lr_blkptr
;
1823 uint64_t txg
= lr
->lr_common
.lrc_txg
;
1825 dmu_object_info_t doi
;
1830 ztest_object_lock(zd
, object
, RL_READER
);
1831 error
= dmu_bonus_hold(os
, object
, FTAG
, &db
);
1833 ztest_object_unlock(zd
, object
);
1837 crtxg
= ztest_bt_bonus(db
)->bt_crtxg
;
1839 if (crtxg
== 0 || crtxg
> txg
) {
1840 dmu_buf_rele(db
, FTAG
);
1841 ztest_object_unlock(zd
, object
);
1845 dmu_object_info_from_db(db
, &doi
);
1846 dmu_buf_rele(db
, FTAG
);
1849 zgd
= umem_zalloc(sizeof (*zgd
), UMEM_NOFAIL
);
1850 zgd
->zgd_zilog
= zd
->zd_zilog
;
1851 zgd
->zgd_private
= zd
;
1853 if (buf
!= NULL
) { /* immediate write */
1854 zgd
->zgd_rl
= ztest_range_lock(zd
, object
, offset
, size
,
1857 error
= dmu_read(os
, object
, offset
, size
, buf
,
1858 DMU_READ_NO_PREFETCH
);
1861 size
= doi
.doi_data_block_size
;
1863 offset
= P2ALIGN(offset
, size
);
1865 ASSERT(offset
< size
);
1869 zgd
->zgd_rl
= ztest_range_lock(zd
, object
, offset
, size
,
1872 error
= dmu_buf_hold(os
, object
, offset
, zgd
, &db
,
1873 DMU_READ_NO_PREFETCH
);
1879 ASSERT(db
->db_offset
== offset
);
1880 ASSERT(db
->db_size
== size
);
1882 error
= dmu_sync(zio
, lr
->lr_common
.lrc_txg
,
1883 ztest_get_done
, zgd
);
1890 ztest_get_done(zgd
, error
);
1896 ztest_lr_alloc(size_t lrsize
, char *name
)
1899 size_t namesize
= name
? strlen(name
) + 1 : 0;
1901 lr
= umem_zalloc(lrsize
+ namesize
, UMEM_NOFAIL
);
1904 bcopy(name
, lr
+ lrsize
, namesize
);
1910 ztest_lr_free(void *lr
, size_t lrsize
, char *name
)
1912 size_t namesize
= name
? strlen(name
) + 1 : 0;
1914 umem_free(lr
, lrsize
+ namesize
);
1918 * Lookup a bunch of objects. Returns the number of objects not found.
1921 ztest_lookup(ztest_ds_t
*zd
, ztest_od_t
*od
, int count
)
1927 ASSERT(mutex_held(&zd
->zd_dirobj_lock
));
1929 for (i
= 0; i
< count
; i
++, od
++) {
1931 error
= zap_lookup(zd
->zd_os
, od
->od_dir
, od
->od_name
,
1932 sizeof (uint64_t), 1, &od
->od_object
);
1934 ASSERT(error
== ENOENT
);
1935 ASSERT(od
->od_object
== 0);
1939 ztest_block_tag_t
*bbt
;
1940 dmu_object_info_t doi
;
1942 ASSERT(od
->od_object
!= 0);
1943 ASSERT(missing
== 0); /* there should be no gaps */
1945 ztest_object_lock(zd
, od
->od_object
, RL_READER
);
1946 VERIFY3U(0, ==, dmu_bonus_hold(zd
->zd_os
,
1947 od
->od_object
, FTAG
, &db
));
1948 dmu_object_info_from_db(db
, &doi
);
1949 bbt
= ztest_bt_bonus(db
);
1950 ASSERT3U(bbt
->bt_magic
, ==, BT_MAGIC
);
1951 od
->od_type
= doi
.doi_type
;
1952 od
->od_blocksize
= doi
.doi_data_block_size
;
1953 od
->od_gen
= bbt
->bt_gen
;
1954 dmu_buf_rele(db
, FTAG
);
1955 ztest_object_unlock(zd
, od
->od_object
);
1963 ztest_create(ztest_ds_t
*zd
, ztest_od_t
*od
, int count
)
1968 ASSERT(mutex_held(&zd
->zd_dirobj_lock
));
1970 for (i
= 0; i
< count
; i
++, od
++) {
1977 lr_create_t
*lr
= ztest_lr_alloc(sizeof (*lr
), od
->od_name
);
1979 lr
->lr_doid
= od
->od_dir
;
1980 lr
->lr_foid
= 0; /* 0 to allocate, > 0 to claim */
1981 lr
->lrz_type
= od
->od_crtype
;
1982 lr
->lrz_blocksize
= od
->od_crblocksize
;
1983 lr
->lrz_ibshift
= ztest_random_ibshift();
1984 lr
->lrz_bonustype
= DMU_OT_UINT64_OTHER
;
1985 lr
->lrz_bonuslen
= dmu_bonus_max();
1986 lr
->lr_gen
= od
->od_crgen
;
1987 lr
->lr_crtime
[0] = time(NULL
);
1989 if (ztest_replay_create(zd
, lr
, B_FALSE
) != 0) {
1990 ASSERT(missing
== 0);
1994 od
->od_object
= lr
->lr_foid
;
1995 od
->od_type
= od
->od_crtype
;
1996 od
->od_blocksize
= od
->od_crblocksize
;
1997 od
->od_gen
= od
->od_crgen
;
1998 ASSERT(od
->od_object
!= 0);
2001 ztest_lr_free(lr
, sizeof (*lr
), od
->od_name
);
2008 ztest_remove(ztest_ds_t
*zd
, ztest_od_t
*od
, int count
)
2014 ASSERT(mutex_held(&zd
->zd_dirobj_lock
));
2018 for (i
= count
- 1; i
>= 0; i
--, od
--) {
2024 if (od
->od_object
== 0)
2027 lr_remove_t
*lr
= ztest_lr_alloc(sizeof (*lr
), od
->od_name
);
2029 lr
->lr_doid
= od
->od_dir
;
2031 if ((error
= ztest_replay_remove(zd
, lr
, B_FALSE
)) != 0) {
2032 ASSERT3U(error
, ==, ENOSPC
);
2037 ztest_lr_free(lr
, sizeof (*lr
), od
->od_name
);
2044 ztest_write(ztest_ds_t
*zd
, uint64_t object
, uint64_t offset
, uint64_t size
,
2050 lr
= ztest_lr_alloc(sizeof (*lr
) + size
, NULL
);
2052 lr
->lr_foid
= object
;
2053 lr
->lr_offset
= offset
;
2054 lr
->lr_length
= size
;
2056 BP_ZERO(&lr
->lr_blkptr
);
2058 bcopy(data
, lr
+ 1, size
);
2060 error
= ztest_replay_write(zd
, lr
, B_FALSE
);
2062 ztest_lr_free(lr
, sizeof (*lr
) + size
, NULL
);
2068 ztest_truncate(ztest_ds_t
*zd
, uint64_t object
, uint64_t offset
, uint64_t size
)
2073 lr
= ztest_lr_alloc(sizeof (*lr
), NULL
);
2075 lr
->lr_foid
= object
;
2076 lr
->lr_offset
= offset
;
2077 lr
->lr_length
= size
;
2079 error
= ztest_replay_truncate(zd
, lr
, B_FALSE
);
2081 ztest_lr_free(lr
, sizeof (*lr
), NULL
);
2087 ztest_setattr(ztest_ds_t
*zd
, uint64_t object
)
2092 lr
= ztest_lr_alloc(sizeof (*lr
), NULL
);
2094 lr
->lr_foid
= object
;
2098 error
= ztest_replay_setattr(zd
, lr
, B_FALSE
);
2100 ztest_lr_free(lr
, sizeof (*lr
), NULL
);
2106 ztest_prealloc(ztest_ds_t
*zd
, uint64_t object
, uint64_t offset
, uint64_t size
)
2108 objset_t
*os
= zd
->zd_os
;
2113 txg_wait_synced(dmu_objset_pool(os
), 0);
2115 ztest_object_lock(zd
, object
, RL_READER
);
2116 rl
= ztest_range_lock(zd
, object
, offset
, size
, RL_WRITER
);
2118 tx
= dmu_tx_create(os
);
2120 dmu_tx_hold_write(tx
, object
, offset
, size
);
2122 txg
= ztest_tx_assign(tx
, TXG_WAIT
, FTAG
);
2125 dmu_prealloc(os
, object
, offset
, size
, tx
);
2127 txg_wait_synced(dmu_objset_pool(os
), txg
);
2129 (void) dmu_free_long_range(os
, object
, offset
, size
);
2132 ztest_range_unlock(rl
);
2133 ztest_object_unlock(zd
, object
);
2137 ztest_io(ztest_ds_t
*zd
, uint64_t object
, uint64_t offset
)
2139 ztest_block_tag_t wbt
;
2140 dmu_object_info_t doi
;
2141 enum ztest_io_type io_type
;
2145 VERIFY(dmu_object_info(zd
->zd_os
, object
, &doi
) == 0);
2146 blocksize
= doi
.doi_data_block_size
;
2147 data
= umem_alloc(blocksize
, UMEM_NOFAIL
);
2150 * Pick an i/o type at random, biased toward writing block tags.
2152 io_type
= ztest_random(ZTEST_IO_TYPES
);
2153 if (ztest_random(2) == 0)
2154 io_type
= ZTEST_IO_WRITE_TAG
;
2156 (void) rw_enter(&zd
->zd_zilog_lock
, RW_READER
);
2160 case ZTEST_IO_WRITE_TAG
:
2161 ztest_bt_generate(&wbt
, zd
->zd_os
, object
, offset
, 0, 0, 0);
2162 (void) ztest_write(zd
, object
, offset
, sizeof (wbt
), &wbt
);
2165 case ZTEST_IO_WRITE_PATTERN
:
2166 (void) memset(data
, 'a' + (object
+ offset
) % 5, blocksize
);
2167 if (ztest_random(2) == 0) {
2169 * Induce fletcher2 collisions to ensure that
2170 * zio_ddt_collision() detects and resolves them
2171 * when using fletcher2-verify for deduplication.
2173 ((uint64_t *)data
)[0] ^= 1ULL << 63;
2174 ((uint64_t *)data
)[4] ^= 1ULL << 63;
2176 (void) ztest_write(zd
, object
, offset
, blocksize
, data
);
2179 case ZTEST_IO_WRITE_ZEROES
:
2180 bzero(data
, blocksize
);
2181 (void) ztest_write(zd
, object
, offset
, blocksize
, data
);
2184 case ZTEST_IO_TRUNCATE
:
2185 (void) ztest_truncate(zd
, object
, offset
, blocksize
);
2188 case ZTEST_IO_SETATTR
:
2189 (void) ztest_setattr(zd
, object
);
2195 (void) rw_exit(&zd
->zd_zilog_lock
);
2197 umem_free(data
, blocksize
);
2201 * Initialize an object description template.
2204 ztest_od_init(ztest_od_t
*od
, uint64_t id
, char *tag
, uint64_t index
,
2205 dmu_object_type_t type
, uint64_t blocksize
, uint64_t gen
)
2207 od
->od_dir
= ZTEST_DIROBJ
;
2210 od
->od_crtype
= type
;
2211 od
->od_crblocksize
= blocksize
? blocksize
: ztest_random_blocksize();
2214 od
->od_type
= DMU_OT_NONE
;
2215 od
->od_blocksize
= 0;
2218 (void) snprintf(od
->od_name
, sizeof (od
->od_name
), "%s(%lld)[%llu]",
2219 tag
, (longlong_t
)id
, (u_longlong_t
)index
);
2223 * Lookup or create the objects for a test using the od template.
2224 * If the objects do not all exist, or if 'remove' is specified,
2225 * remove any existing objects and create new ones. Otherwise,
2226 * use the existing objects.
2229 ztest_object_init(ztest_ds_t
*zd
, ztest_od_t
*od
, size_t size
, boolean_t remove
)
2231 int count
= size
/ sizeof (*od
);
2234 mutex_enter(&zd
->zd_dirobj_lock
);
2235 if ((ztest_lookup(zd
, od
, count
) != 0 || remove
) &&
2236 (ztest_remove(zd
, od
, count
) != 0 ||
2237 ztest_create(zd
, od
, count
) != 0))
2240 mutex_exit(&zd
->zd_dirobj_lock
);
2247 ztest_zil_commit(ztest_ds_t
*zd
, uint64_t id
)
2249 zilog_t
*zilog
= zd
->zd_zilog
;
2251 (void) rw_enter(&zd
->zd_zilog_lock
, RW_READER
);
2253 zil_commit(zilog
, ztest_random(ZTEST_OBJECTS
));
2256 * Remember the committed values in zd, which is in parent/child
2257 * shared memory. If we die, the next iteration of ztest_run()
2258 * will verify that the log really does contain this record.
2260 mutex_enter(&zilog
->zl_lock
);
2261 ASSERT(zd
->zd_shared
!= NULL
);
2262 ASSERT3U(zd
->zd_shared
->zd_seq
, <=, zilog
->zl_commit_lr_seq
);
2263 zd
->zd_shared
->zd_seq
= zilog
->zl_commit_lr_seq
;
2264 mutex_exit(&zilog
->zl_lock
);
2266 (void) rw_exit(&zd
->zd_zilog_lock
);
2270 * This function is designed to simulate the operations that occur during a
2271 * mount/unmount operation. We hold the dataset across these operations in an
2272 * attempt to expose any implicit assumptions about ZIL management.
2276 ztest_zil_remount(ztest_ds_t
*zd
, uint64_t id
)
2278 objset_t
*os
= zd
->zd_os
;
2280 (void) rw_enter(&zd
->zd_zilog_lock
, RW_WRITER
);
2282 /* zfs_sb_teardown() */
2283 zil_close(zd
->zd_zilog
);
2285 /* zfsvfs_setup() */
2286 VERIFY(zil_open(os
, ztest_get_data
) == zd
->zd_zilog
);
2287 zil_replay(os
, zd
, ztest_replay_vector
);
2289 (void) rw_exit(&zd
->zd_zilog_lock
);
2293 * Verify that we can't destroy an active pool, create an existing pool,
2294 * or create a pool with a bad vdev spec.
2298 ztest_spa_create_destroy(ztest_ds_t
*zd
, uint64_t id
)
2300 ztest_shared_opts_t
*zo
= &ztest_opts
;
2305 * Attempt to create using a bad file.
2307 nvroot
= make_vdev_root("/dev/bogus", NULL
, 0, 0, 0, 0, 0, 1);
2308 VERIFY3U(ENOENT
, ==,
2309 spa_create("ztest_bad_file", nvroot
, NULL
, NULL
, NULL
));
2310 nvlist_free(nvroot
);
2313 * Attempt to create using a bad mirror.
2315 nvroot
= make_vdev_root("/dev/bogus", NULL
, 0, 0, 0, 0, 2, 1);
2316 VERIFY3U(ENOENT
, ==,
2317 spa_create("ztest_bad_mirror", nvroot
, NULL
, NULL
, NULL
));
2318 nvlist_free(nvroot
);
2321 * Attempt to create an existing pool. It shouldn't matter
2322 * what's in the nvroot; we should fail with EEXIST.
2324 (void) rw_enter(&ztest_name_lock
, RW_READER
);
2325 nvroot
= make_vdev_root("/dev/bogus", NULL
, 0, 0, 0, 0, 0, 1);
2326 VERIFY3U(EEXIST
, ==, spa_create(zo
->zo_pool
, nvroot
, NULL
, NULL
, NULL
));
2327 nvlist_free(nvroot
);
2328 VERIFY3U(0, ==, spa_open(zo
->zo_pool
, &spa
, FTAG
));
2329 VERIFY3U(EBUSY
, ==, spa_destroy(zo
->zo_pool
));
2330 spa_close(spa
, FTAG
);
2332 (void) rw_exit(&ztest_name_lock
);
2336 vdev_lookup_by_path(vdev_t
*vd
, const char *path
)
2341 if (vd
->vdev_path
!= NULL
&& strcmp(path
, vd
->vdev_path
) == 0)
2344 for (c
= 0; c
< vd
->vdev_children
; c
++)
2345 if ((mvd
= vdev_lookup_by_path(vd
->vdev_child
[c
], path
)) !=
2353 * Find the first available hole which can be used as a top-level.
2356 find_vdev_hole(spa_t
*spa
)
2358 vdev_t
*rvd
= spa
->spa_root_vdev
;
2361 ASSERT(spa_config_held(spa
, SCL_VDEV
, RW_READER
) == SCL_VDEV
);
2363 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
2364 vdev_t
*cvd
= rvd
->vdev_child
[c
];
2366 if (cvd
->vdev_ishole
)
2373 * Verify that vdev_add() works as expected.
2377 ztest_vdev_add_remove(ztest_ds_t
*zd
, uint64_t id
)
2379 ztest_shared_t
*zs
= ztest_shared
;
2380 spa_t
*spa
= ztest_spa
;
2386 mutex_enter(&ztest_vdev_lock
);
2388 MAX(zs
->zs_mirrors
+ zs
->zs_splits
, 1) * ztest_opts
.zo_raidz
;
2390 spa_config_enter(spa
, SCL_VDEV
, FTAG
, RW_READER
);
2392 ztest_shared
->zs_vdev_next_leaf
= find_vdev_hole(spa
) * leaves
;
2395 * If we have slogs then remove them 1/4 of the time.
2397 if (spa_has_slogs(spa
) && ztest_random(4) == 0) {
2399 * Grab the guid from the head of the log class rotor.
2401 guid
= spa_log_class(spa
)->mc_rotor
->mg_vd
->vdev_guid
;
2403 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
2406 * We have to grab the zs_name_lock as writer to
2407 * prevent a race between removing a slog (dmu_objset_find)
2408 * and destroying a dataset. Removing the slog will
2409 * grab a reference on the dataset which may cause
2410 * dmu_objset_destroy() to fail with EBUSY thus
2411 * leaving the dataset in an inconsistent state.
2413 rw_enter(&ztest_name_lock
, RW_WRITER
);
2414 error
= spa_vdev_remove(spa
, guid
, B_FALSE
);
2415 rw_exit(&ztest_name_lock
);
2417 if (error
&& error
!= EEXIST
)
2418 fatal(0, "spa_vdev_remove() = %d", error
);
2420 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
2423 * Make 1/4 of the devices be log devices.
2425 nvroot
= make_vdev_root(NULL
, NULL
,
2426 ztest_opts
.zo_vdev_size
, 0,
2427 ztest_random(4) == 0, ztest_opts
.zo_raidz
,
2430 error
= spa_vdev_add(spa
, nvroot
);
2431 nvlist_free(nvroot
);
2433 if (error
== ENOSPC
)
2434 ztest_record_enospc("spa_vdev_add");
2435 else if (error
!= 0)
2436 fatal(0, "spa_vdev_add() = %d", error
);
2439 mutex_exit(&ztest_vdev_lock
);
2443 * Verify that adding/removing aux devices (l2arc, hot spare) works as expected.
2447 ztest_vdev_aux_add_remove(ztest_ds_t
*zd
, uint64_t id
)
2449 ztest_shared_t
*zs
= ztest_shared
;
2450 spa_t
*spa
= ztest_spa
;
2451 vdev_t
*rvd
= spa
->spa_root_vdev
;
2452 spa_aux_vdev_t
*sav
;
2458 path
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
2460 if (ztest_random(2) == 0) {
2461 sav
= &spa
->spa_spares
;
2462 aux
= ZPOOL_CONFIG_SPARES
;
2464 sav
= &spa
->spa_l2cache
;
2465 aux
= ZPOOL_CONFIG_L2CACHE
;
2468 mutex_enter(&ztest_vdev_lock
);
2470 spa_config_enter(spa
, SCL_VDEV
, FTAG
, RW_READER
);
2472 if (sav
->sav_count
!= 0 && ztest_random(4) == 0) {
2474 * Pick a random device to remove.
2476 guid
= sav
->sav_vdevs
[ztest_random(sav
->sav_count
)]->vdev_guid
;
2479 * Find an unused device we can add.
2481 zs
->zs_vdev_aux
= 0;
2484 (void) snprintf(path
, sizeof (path
), ztest_aux_template
,
2485 ztest_opts
.zo_dir
, ztest_opts
.zo_pool
, aux
,
2487 for (c
= 0; c
< sav
->sav_count
; c
++)
2488 if (strcmp(sav
->sav_vdevs
[c
]->vdev_path
,
2491 if (c
== sav
->sav_count
&&
2492 vdev_lookup_by_path(rvd
, path
) == NULL
)
2498 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
2504 nvlist_t
*nvroot
= make_vdev_root(NULL
, aux
,
2505 (ztest_opts
.zo_vdev_size
* 5) / 4, 0, 0, 0, 0, 1);
2506 error
= spa_vdev_add(spa
, nvroot
);
2508 fatal(0, "spa_vdev_add(%p) = %d", nvroot
, error
);
2509 nvlist_free(nvroot
);
2512 * Remove an existing device. Sometimes, dirty its
2513 * vdev state first to make sure we handle removal
2514 * of devices that have pending state changes.
2516 if (ztest_random(2) == 0)
2517 (void) vdev_online(spa
, guid
, 0, NULL
);
2519 error
= spa_vdev_remove(spa
, guid
, B_FALSE
);
2520 if (error
!= 0 && error
!= EBUSY
)
2521 fatal(0, "spa_vdev_remove(%llu) = %d", guid
, error
);
2524 mutex_exit(&ztest_vdev_lock
);
2526 umem_free(path
, MAXPATHLEN
);
2530 * split a pool if it has mirror tlvdevs
2534 ztest_split_pool(ztest_ds_t
*zd
, uint64_t id
)
2536 ztest_shared_t
*zs
= ztest_shared
;
2537 spa_t
*spa
= ztest_spa
;
2538 vdev_t
*rvd
= spa
->spa_root_vdev
;
2539 nvlist_t
*tree
, **child
, *config
, *split
, **schild
;
2540 uint_t c
, children
, schildren
= 0, lastlogid
= 0;
2543 mutex_enter(&ztest_vdev_lock
);
2545 /* ensure we have a useable config; mirrors of raidz aren't supported */
2546 if (zs
->zs_mirrors
< 3 || ztest_opts
.zo_raidz
> 1) {
2547 mutex_exit(&ztest_vdev_lock
);
2551 /* clean up the old pool, if any */
2552 (void) spa_destroy("splitp");
2554 spa_config_enter(spa
, SCL_VDEV
, FTAG
, RW_READER
);
2556 /* generate a config from the existing config */
2557 mutex_enter(&spa
->spa_props_lock
);
2558 VERIFY(nvlist_lookup_nvlist(spa
->spa_config
, ZPOOL_CONFIG_VDEV_TREE
,
2560 mutex_exit(&spa
->spa_props_lock
);
2562 VERIFY(nvlist_lookup_nvlist_array(tree
, ZPOOL_CONFIG_CHILDREN
, &child
,
2565 schild
= malloc(rvd
->vdev_children
* sizeof (nvlist_t
*));
2566 for (c
= 0; c
< children
; c
++) {
2567 vdev_t
*tvd
= rvd
->vdev_child
[c
];
2571 if (tvd
->vdev_islog
|| tvd
->vdev_ops
== &vdev_hole_ops
) {
2572 VERIFY(nvlist_alloc(&schild
[schildren
], NV_UNIQUE_NAME
,
2574 VERIFY(nvlist_add_string(schild
[schildren
],
2575 ZPOOL_CONFIG_TYPE
, VDEV_TYPE_HOLE
) == 0);
2576 VERIFY(nvlist_add_uint64(schild
[schildren
],
2577 ZPOOL_CONFIG_IS_HOLE
, 1) == 0);
2579 lastlogid
= schildren
;
2584 VERIFY(nvlist_lookup_nvlist_array(child
[c
],
2585 ZPOOL_CONFIG_CHILDREN
, &mchild
, &mchildren
) == 0);
2586 VERIFY(nvlist_dup(mchild
[0], &schild
[schildren
++], 0) == 0);
2589 /* OK, create a config that can be used to split */
2590 VERIFY(nvlist_alloc(&split
, NV_UNIQUE_NAME
, 0) == 0);
2591 VERIFY(nvlist_add_string(split
, ZPOOL_CONFIG_TYPE
,
2592 VDEV_TYPE_ROOT
) == 0);
2593 VERIFY(nvlist_add_nvlist_array(split
, ZPOOL_CONFIG_CHILDREN
, schild
,
2594 lastlogid
!= 0 ? lastlogid
: schildren
) == 0);
2596 VERIFY(nvlist_alloc(&config
, NV_UNIQUE_NAME
, 0) == 0);
2597 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, split
) == 0);
2599 for (c
= 0; c
< schildren
; c
++)
2600 nvlist_free(schild
[c
]);
2604 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
2606 (void) rw_enter(&ztest_name_lock
, RW_WRITER
);
2607 error
= spa_vdev_split_mirror(spa
, "splitp", config
, NULL
, B_FALSE
);
2608 (void) rw_exit(&ztest_name_lock
);
2610 nvlist_free(config
);
2613 (void) printf("successful split - results:\n");
2614 mutex_enter(&spa_namespace_lock
);
2615 show_pool_stats(spa
);
2616 show_pool_stats(spa_lookup("splitp"));
2617 mutex_exit(&spa_namespace_lock
);
2621 mutex_exit(&ztest_vdev_lock
);
2626 * Verify that we can attach and detach devices.
2630 ztest_vdev_attach_detach(ztest_ds_t
*zd
, uint64_t id
)
2632 ztest_shared_t
*zs
= ztest_shared
;
2633 spa_t
*spa
= ztest_spa
;
2634 spa_aux_vdev_t
*sav
= &spa
->spa_spares
;
2635 vdev_t
*rvd
= spa
->spa_root_vdev
;
2636 vdev_t
*oldvd
, *newvd
, *pvd
;
2640 uint64_t ashift
= ztest_get_ashift();
2641 uint64_t oldguid
, pguid
;
2642 size_t oldsize
, newsize
;
2643 char *oldpath
, *newpath
;
2645 int oldvd_has_siblings
= B_FALSE
;
2646 int newvd_is_spare
= B_FALSE
;
2648 int error
, expected_error
;
2650 oldpath
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
2651 newpath
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
2653 mutex_enter(&ztest_vdev_lock
);
2654 leaves
= MAX(zs
->zs_mirrors
, 1) * ztest_opts
.zo_raidz
;
2656 spa_config_enter(spa
, SCL_VDEV
, FTAG
, RW_READER
);
2659 * Decide whether to do an attach or a replace.
2661 replacing
= ztest_random(2);
2664 * Pick a random top-level vdev.
2666 top
= ztest_random_vdev_top(spa
, B_TRUE
);
2669 * Pick a random leaf within it.
2671 leaf
= ztest_random(leaves
);
2676 oldvd
= rvd
->vdev_child
[top
];
2677 if (zs
->zs_mirrors
>= 1) {
2678 ASSERT(oldvd
->vdev_ops
== &vdev_mirror_ops
);
2679 ASSERT(oldvd
->vdev_children
>= zs
->zs_mirrors
);
2680 oldvd
= oldvd
->vdev_child
[leaf
/ ztest_opts
.zo_raidz
];
2682 if (ztest_opts
.zo_raidz
> 1) {
2683 ASSERT(oldvd
->vdev_ops
== &vdev_raidz_ops
);
2684 ASSERT(oldvd
->vdev_children
== ztest_opts
.zo_raidz
);
2685 oldvd
= oldvd
->vdev_child
[leaf
% ztest_opts
.zo_raidz
];
2689 * If we're already doing an attach or replace, oldvd may be a
2690 * mirror vdev -- in which case, pick a random child.
2692 while (oldvd
->vdev_children
!= 0) {
2693 oldvd_has_siblings
= B_TRUE
;
2694 ASSERT(oldvd
->vdev_children
>= 2);
2695 oldvd
= oldvd
->vdev_child
[ztest_random(oldvd
->vdev_children
)];
2698 oldguid
= oldvd
->vdev_guid
;
2699 oldsize
= vdev_get_min_asize(oldvd
);
2700 oldvd_is_log
= oldvd
->vdev_top
->vdev_islog
;
2701 (void) strcpy(oldpath
, oldvd
->vdev_path
);
2702 pvd
= oldvd
->vdev_parent
;
2703 pguid
= pvd
->vdev_guid
;
2706 * If oldvd has siblings, then half of the time, detach it.
2708 if (oldvd_has_siblings
&& ztest_random(2) == 0) {
2709 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
2710 error
= spa_vdev_detach(spa
, oldguid
, pguid
, B_FALSE
);
2711 if (error
!= 0 && error
!= ENODEV
&& error
!= EBUSY
&&
2713 fatal(0, "detach (%s) returned %d", oldpath
, error
);
2718 * For the new vdev, choose with equal probability between the two
2719 * standard paths (ending in either 'a' or 'b') or a random hot spare.
2721 if (sav
->sav_count
!= 0 && ztest_random(3) == 0) {
2722 newvd
= sav
->sav_vdevs
[ztest_random(sav
->sav_count
)];
2723 newvd_is_spare
= B_TRUE
;
2724 (void) strcpy(newpath
, newvd
->vdev_path
);
2726 (void) snprintf(newpath
, MAXPATHLEN
, ztest_dev_template
,
2727 ztest_opts
.zo_dir
, ztest_opts
.zo_pool
,
2728 top
* leaves
+ leaf
);
2729 if (ztest_random(2) == 0)
2730 newpath
[strlen(newpath
) - 1] = 'b';
2731 newvd
= vdev_lookup_by_path(rvd
, newpath
);
2735 newsize
= vdev_get_min_asize(newvd
);
2738 * Make newsize a little bigger or smaller than oldsize.
2739 * If it's smaller, the attach should fail.
2740 * If it's larger, and we're doing a replace,
2741 * we should get dynamic LUN growth when we're done.
2743 newsize
= 10 * oldsize
/ (9 + ztest_random(3));
2747 * If pvd is not a mirror or root, the attach should fail with ENOTSUP,
2748 * unless it's a replace; in that case any non-replacing parent is OK.
2750 * If newvd is already part of the pool, it should fail with EBUSY.
2752 * If newvd is too small, it should fail with EOVERFLOW.
2754 if (pvd
->vdev_ops
!= &vdev_mirror_ops
&&
2755 pvd
->vdev_ops
!= &vdev_root_ops
&& (!replacing
||
2756 pvd
->vdev_ops
== &vdev_replacing_ops
||
2757 pvd
->vdev_ops
== &vdev_spare_ops
))
2758 expected_error
= ENOTSUP
;
2759 else if (newvd_is_spare
&& (!replacing
|| oldvd_is_log
))
2760 expected_error
= ENOTSUP
;
2761 else if (newvd
== oldvd
)
2762 expected_error
= replacing
? 0 : EBUSY
;
2763 else if (vdev_lookup_by_path(rvd
, newpath
) != NULL
)
2764 expected_error
= EBUSY
;
2765 else if (newsize
< oldsize
)
2766 expected_error
= EOVERFLOW
;
2767 else if (ashift
> oldvd
->vdev_top
->vdev_ashift
)
2768 expected_error
= EDOM
;
2772 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
2775 * Build the nvlist describing newpath.
2777 root
= make_vdev_root(newpath
, NULL
, newvd
== NULL
? newsize
: 0,
2778 ashift
, 0, 0, 0, 1);
2780 error
= spa_vdev_attach(spa
, oldguid
, root
, replacing
);
2785 * If our parent was the replacing vdev, but the replace completed,
2786 * then instead of failing with ENOTSUP we may either succeed,
2787 * fail with ENODEV, or fail with EOVERFLOW.
2789 if (expected_error
== ENOTSUP
&&
2790 (error
== 0 || error
== ENODEV
|| error
== EOVERFLOW
))
2791 expected_error
= error
;
2794 * If someone grew the LUN, the replacement may be too small.
2796 if (error
== EOVERFLOW
|| error
== EBUSY
)
2797 expected_error
= error
;
2799 /* XXX workaround 6690467 */
2800 if (error
!= expected_error
&& expected_error
!= EBUSY
) {
2801 fatal(0, "attach (%s %llu, %s %llu, %d) "
2802 "returned %d, expected %d",
2803 oldpath
, (longlong_t
)oldsize
, newpath
,
2804 (longlong_t
)newsize
, replacing
, error
, expected_error
);
2807 mutex_exit(&ztest_vdev_lock
);
2809 umem_free(oldpath
, MAXPATHLEN
);
2810 umem_free(newpath
, MAXPATHLEN
);
2814 * Callback function which expands the physical size of the vdev.
2817 grow_vdev(vdev_t
*vd
, void *arg
)
2819 ASSERTV(spa_t
*spa
= vd
->vdev_spa
);
2820 size_t *newsize
= arg
;
2824 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_READER
) == SCL_STATE
);
2825 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
2827 if ((fd
= open(vd
->vdev_path
, O_RDWR
)) == -1)
2830 fsize
= lseek(fd
, 0, SEEK_END
);
2831 VERIFY(ftruncate(fd
, *newsize
) == 0);
2833 if (ztest_opts
.zo_verbose
>= 6) {
2834 (void) printf("%s grew from %lu to %lu bytes\n",
2835 vd
->vdev_path
, (ulong_t
)fsize
, (ulong_t
)*newsize
);
2842 * Callback function which expands a given vdev by calling vdev_online().
2846 online_vdev(vdev_t
*vd
, void *arg
)
2848 spa_t
*spa
= vd
->vdev_spa
;
2849 vdev_t
*tvd
= vd
->vdev_top
;
2850 uint64_t guid
= vd
->vdev_guid
;
2851 uint64_t generation
= spa
->spa_config_generation
+ 1;
2852 vdev_state_t newstate
= VDEV_STATE_UNKNOWN
;
2855 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_READER
) == SCL_STATE
);
2856 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
2858 /* Calling vdev_online will initialize the new metaslabs */
2859 spa_config_exit(spa
, SCL_STATE
, spa
);
2860 error
= vdev_online(spa
, guid
, ZFS_ONLINE_EXPAND
, &newstate
);
2861 spa_config_enter(spa
, SCL_STATE
, spa
, RW_READER
);
2864 * If vdev_online returned an error or the underlying vdev_open
2865 * failed then we abort the expand. The only way to know that
2866 * vdev_open fails is by checking the returned newstate.
2868 if (error
|| newstate
!= VDEV_STATE_HEALTHY
) {
2869 if (ztest_opts
.zo_verbose
>= 5) {
2870 (void) printf("Unable to expand vdev, state %llu, "
2871 "error %d\n", (u_longlong_t
)newstate
, error
);
2875 ASSERT3U(newstate
, ==, VDEV_STATE_HEALTHY
);
2878 * Since we dropped the lock we need to ensure that we're
2879 * still talking to the original vdev. It's possible this
2880 * vdev may have been detached/replaced while we were
2881 * trying to online it.
2883 if (generation
!= spa
->spa_config_generation
) {
2884 if (ztest_opts
.zo_verbose
>= 5) {
2885 (void) printf("vdev configuration has changed, "
2886 "guid %llu, state %llu, expected gen %llu, "
2889 (u_longlong_t
)tvd
->vdev_state
,
2890 (u_longlong_t
)generation
,
2891 (u_longlong_t
)spa
->spa_config_generation
);
2899 * Traverse the vdev tree calling the supplied function.
2900 * We continue to walk the tree until we either have walked all
2901 * children or we receive a non-NULL return from the callback.
2902 * If a NULL callback is passed, then we just return back the first
2903 * leaf vdev we encounter.
2906 vdev_walk_tree(vdev_t
*vd
, vdev_t
*(*func
)(vdev_t
*, void *), void *arg
)
2910 if (vd
->vdev_ops
->vdev_op_leaf
) {
2914 return (func(vd
, arg
));
2917 for (c
= 0; c
< vd
->vdev_children
; c
++) {
2918 vdev_t
*cvd
= vd
->vdev_child
[c
];
2919 if ((cvd
= vdev_walk_tree(cvd
, func
, arg
)) != NULL
)
2926 * Verify that dynamic LUN growth works as expected.
2930 ztest_vdev_LUN_growth(ztest_ds_t
*zd
, uint64_t id
)
2932 spa_t
*spa
= ztest_spa
;
2934 metaslab_class_t
*mc
;
2935 metaslab_group_t
*mg
;
2936 size_t psize
, newsize
;
2938 uint64_t old_class_space
, new_class_space
, old_ms_count
, new_ms_count
;
2940 mutex_enter(&ztest_vdev_lock
);
2941 spa_config_enter(spa
, SCL_STATE
, spa
, RW_READER
);
2943 top
= ztest_random_vdev_top(spa
, B_TRUE
);
2945 tvd
= spa
->spa_root_vdev
->vdev_child
[top
];
2948 old_ms_count
= tvd
->vdev_ms_count
;
2949 old_class_space
= metaslab_class_get_space(mc
);
2952 * Determine the size of the first leaf vdev associated with
2953 * our top-level device.
2955 vd
= vdev_walk_tree(tvd
, NULL
, NULL
);
2956 ASSERT3P(vd
, !=, NULL
);
2957 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
2959 psize
= vd
->vdev_psize
;
2962 * We only try to expand the vdev if it's healthy, less than 4x its
2963 * original size, and it has a valid psize.
2965 if (tvd
->vdev_state
!= VDEV_STATE_HEALTHY
||
2966 psize
== 0 || psize
>= 4 * ztest_opts
.zo_vdev_size
) {
2967 spa_config_exit(spa
, SCL_STATE
, spa
);
2968 mutex_exit(&ztest_vdev_lock
);
2972 newsize
= psize
+ psize
/ 8;
2973 ASSERT3U(newsize
, >, psize
);
2975 if (ztest_opts
.zo_verbose
>= 6) {
2976 (void) printf("Expanding LUN %s from %lu to %lu\n",
2977 vd
->vdev_path
, (ulong_t
)psize
, (ulong_t
)newsize
);
2981 * Growing the vdev is a two step process:
2982 * 1). expand the physical size (i.e. relabel)
2983 * 2). online the vdev to create the new metaslabs
2985 if (vdev_walk_tree(tvd
, grow_vdev
, &newsize
) != NULL
||
2986 vdev_walk_tree(tvd
, online_vdev
, NULL
) != NULL
||
2987 tvd
->vdev_state
!= VDEV_STATE_HEALTHY
) {
2988 if (ztest_opts
.zo_verbose
>= 5) {
2989 (void) printf("Could not expand LUN because "
2990 "the vdev configuration changed.\n");
2992 spa_config_exit(spa
, SCL_STATE
, spa
);
2993 mutex_exit(&ztest_vdev_lock
);
2997 spa_config_exit(spa
, SCL_STATE
, spa
);
3000 * Expanding the LUN will update the config asynchronously,
3001 * thus we must wait for the async thread to complete any
3002 * pending tasks before proceeding.
3006 mutex_enter(&spa
->spa_async_lock
);
3007 done
= (spa
->spa_async_thread
== NULL
&& !spa
->spa_async_tasks
);
3008 mutex_exit(&spa
->spa_async_lock
);
3011 txg_wait_synced(spa_get_dsl(spa
), 0);
3012 (void) poll(NULL
, 0, 100);
3015 spa_config_enter(spa
, SCL_STATE
, spa
, RW_READER
);
3017 tvd
= spa
->spa_root_vdev
->vdev_child
[top
];
3018 new_ms_count
= tvd
->vdev_ms_count
;
3019 new_class_space
= metaslab_class_get_space(mc
);
3021 if (tvd
->vdev_mg
!= mg
|| mg
->mg_class
!= mc
) {
3022 if (ztest_opts
.zo_verbose
>= 5) {
3023 (void) printf("Could not verify LUN expansion due to "
3024 "intervening vdev offline or remove.\n");
3026 spa_config_exit(spa
, SCL_STATE
, spa
);
3027 mutex_exit(&ztest_vdev_lock
);
3032 * Make sure we were able to grow the vdev.
3034 if (new_ms_count
<= old_ms_count
)
3035 fatal(0, "LUN expansion failed: ms_count %llu <= %llu\n",
3036 old_ms_count
, new_ms_count
);
3039 * Make sure we were able to grow the pool.
3041 if (new_class_space
<= old_class_space
)
3042 fatal(0, "LUN expansion failed: class_space %llu <= %llu\n",
3043 old_class_space
, new_class_space
);
3045 if (ztest_opts
.zo_verbose
>= 5) {
3046 char oldnumbuf
[6], newnumbuf
[6];
3048 nicenum(old_class_space
, oldnumbuf
);
3049 nicenum(new_class_space
, newnumbuf
);
3050 (void) printf("%s grew from %s to %s\n",
3051 spa
->spa_name
, oldnumbuf
, newnumbuf
);
3054 spa_config_exit(spa
, SCL_STATE
, spa
);
3055 mutex_exit(&ztest_vdev_lock
);
3059 * Verify that dmu_objset_{create,destroy,open,close} work as expected.
3063 ztest_objset_create_cb(objset_t
*os
, void *arg
, cred_t
*cr
, dmu_tx_t
*tx
)
3066 * Create the objects common to all ztest datasets.
3068 VERIFY(zap_create_claim(os
, ZTEST_DIROBJ
,
3069 DMU_OT_ZAP_OTHER
, DMU_OT_NONE
, 0, tx
) == 0);
3073 ztest_dataset_create(char *dsname
)
3075 uint64_t zilset
= ztest_random(100);
3076 int err
= dmu_objset_create(dsname
, DMU_OST_OTHER
, 0,
3077 ztest_objset_create_cb
, NULL
);
3079 if (err
|| zilset
< 80)
3082 if (ztest_opts
.zo_verbose
>= 5)
3083 (void) printf("Setting dataset %s to sync always\n", dsname
);
3084 return (ztest_dsl_prop_set_uint64(dsname
, ZFS_PROP_SYNC
,
3085 ZFS_SYNC_ALWAYS
, B_FALSE
));
3090 ztest_objset_destroy_cb(const char *name
, void *arg
)
3093 dmu_object_info_t doi
;
3097 * Verify that the dataset contains a directory object.
3099 VERIFY3U(0, ==, dmu_objset_hold(name
, FTAG
, &os
));
3100 error
= dmu_object_info(os
, ZTEST_DIROBJ
, &doi
);
3101 if (error
!= ENOENT
) {
3102 /* We could have crashed in the middle of destroying it */
3103 ASSERT3U(error
, ==, 0);
3104 ASSERT3U(doi
.doi_type
, ==, DMU_OT_ZAP_OTHER
);
3105 ASSERT3S(doi
.doi_physical_blocks_512
, >=, 0);
3107 dmu_objset_rele(os
, FTAG
);
3110 * Destroy the dataset.
3112 VERIFY3U(0, ==, dmu_objset_destroy(name
, B_FALSE
));
3117 ztest_snapshot_create(char *osname
, uint64_t id
)
3119 char snapname
[MAXNAMELEN
];
3122 (void) snprintf(snapname
, MAXNAMELEN
, "%s@%llu", osname
,
3125 error
= dmu_objset_snapshot(osname
, strchr(snapname
, '@') + 1,
3126 NULL
, NULL
, B_FALSE
, B_FALSE
, -1);
3127 if (error
== ENOSPC
) {
3128 ztest_record_enospc(FTAG
);
3131 if (error
!= 0 && error
!= EEXIST
)
3132 fatal(0, "ztest_snapshot_create(%s) = %d", snapname
, error
);
3137 ztest_snapshot_destroy(char *osname
, uint64_t id
)
3139 char snapname
[MAXNAMELEN
];
3142 (void) snprintf(snapname
, MAXNAMELEN
, "%s@%llu", osname
,
3145 error
= dmu_objset_destroy(snapname
, B_FALSE
);
3146 if (error
!= 0 && error
!= ENOENT
)
3147 fatal(0, "ztest_snapshot_destroy(%s) = %d", snapname
, error
);
3153 ztest_dmu_objset_create_destroy(ztest_ds_t
*zd
, uint64_t id
)
3163 zdtmp
= umem_alloc(sizeof (ztest_ds_t
), UMEM_NOFAIL
);
3164 name
= umem_alloc(MAXNAMELEN
, UMEM_NOFAIL
);
3166 (void) rw_enter(&ztest_name_lock
, RW_READER
);
3168 (void) snprintf(name
, MAXNAMELEN
, "%s/temp_%llu",
3169 ztest_opts
.zo_pool
, (u_longlong_t
)id
);
3172 * If this dataset exists from a previous run, process its replay log
3173 * half of the time. If we don't replay it, then dmu_objset_destroy()
3174 * (invoked from ztest_objset_destroy_cb()) should just throw it away.
3176 if (ztest_random(2) == 0 &&
3177 dmu_objset_own(name
, DMU_OST_OTHER
, B_FALSE
, FTAG
, &os
) == 0) {
3178 ztest_zd_init(zdtmp
, NULL
, os
);
3179 zil_replay(os
, zdtmp
, ztest_replay_vector
);
3180 ztest_zd_fini(zdtmp
);
3181 dmu_objset_disown(os
, FTAG
);
3185 * There may be an old instance of the dataset we're about to
3186 * create lying around from a previous run. If so, destroy it
3187 * and all of its snapshots.
3189 (void) dmu_objset_find(name
, ztest_objset_destroy_cb
, NULL
,
3190 DS_FIND_CHILDREN
| DS_FIND_SNAPSHOTS
);
3193 * Verify that the destroyed dataset is no longer in the namespace.
3195 VERIFY3U(ENOENT
, ==, dmu_objset_hold(name
, FTAG
, &os
));
3198 * Verify that we can create a new dataset.
3200 error
= ztest_dataset_create(name
);
3202 if (error
== ENOSPC
) {
3203 ztest_record_enospc(FTAG
);
3206 fatal(0, "dmu_objset_create(%s) = %d", name
, error
);
3210 dmu_objset_own(name
, DMU_OST_OTHER
, B_FALSE
, FTAG
, &os
));
3212 ztest_zd_init(zdtmp
, NULL
, os
);
3215 * Open the intent log for it.
3217 zilog
= zil_open(os
, ztest_get_data
);
3220 * Put some objects in there, do a little I/O to them,
3221 * and randomly take a couple of snapshots along the way.
3223 iters
= ztest_random(5);
3224 for (i
= 0; i
< iters
; i
++) {
3225 ztest_dmu_object_alloc_free(zdtmp
, id
);
3226 if (ztest_random(iters
) == 0)
3227 (void) ztest_snapshot_create(name
, i
);
3231 * Verify that we cannot create an existing dataset.
3233 VERIFY3U(EEXIST
, ==,
3234 dmu_objset_create(name
, DMU_OST_OTHER
, 0, NULL
, NULL
));
3237 * Verify that we can hold an objset that is also owned.
3239 VERIFY3U(0, ==, dmu_objset_hold(name
, FTAG
, &os2
));
3240 dmu_objset_rele(os2
, FTAG
);
3243 * Verify that we cannot own an objset that is already owned.
3246 dmu_objset_own(name
, DMU_OST_OTHER
, B_FALSE
, FTAG
, &os2
));
3249 dmu_objset_disown(os
, FTAG
);
3250 ztest_zd_fini(zdtmp
);
3252 (void) rw_exit(&ztest_name_lock
);
3254 umem_free(name
, MAXNAMELEN
);
3255 umem_free(zdtmp
, sizeof (ztest_ds_t
));
3259 * Verify that dmu_snapshot_{create,destroy,open,close} work as expected.
3262 ztest_dmu_snapshot_create_destroy(ztest_ds_t
*zd
, uint64_t id
)
3264 (void) rw_enter(&ztest_name_lock
, RW_READER
);
3265 (void) ztest_snapshot_destroy(zd
->zd_name
, id
);
3266 (void) ztest_snapshot_create(zd
->zd_name
, id
);
3267 (void) rw_exit(&ztest_name_lock
);
3271 * Cleanup non-standard snapshots and clones.
3274 ztest_dsl_dataset_cleanup(char *osname
, uint64_t id
)
3283 snap1name
= umem_alloc(MAXNAMELEN
, UMEM_NOFAIL
);
3284 clone1name
= umem_alloc(MAXNAMELEN
, UMEM_NOFAIL
);
3285 snap2name
= umem_alloc(MAXNAMELEN
, UMEM_NOFAIL
);
3286 clone2name
= umem_alloc(MAXNAMELEN
, UMEM_NOFAIL
);
3287 snap3name
= umem_alloc(MAXNAMELEN
, UMEM_NOFAIL
);
3289 (void) snprintf(snap1name
, MAXNAMELEN
, "%s@s1_%llu",
3290 osname
, (u_longlong_t
)id
);
3291 (void) snprintf(clone1name
, MAXNAMELEN
, "%s/c1_%llu",
3292 osname
, (u_longlong_t
)id
);
3293 (void) snprintf(snap2name
, MAXNAMELEN
, "%s@s2_%llu",
3294 clone1name
, (u_longlong_t
)id
);
3295 (void) snprintf(clone2name
, MAXNAMELEN
, "%s/c2_%llu",
3296 osname
, (u_longlong_t
)id
);
3297 (void) snprintf(snap3name
, MAXNAMELEN
, "%s@s3_%llu",
3298 clone1name
, (u_longlong_t
)id
);
3300 error
= dmu_objset_destroy(clone2name
, B_FALSE
);
3301 if (error
&& error
!= ENOENT
)
3302 fatal(0, "dmu_objset_destroy(%s) = %d", clone2name
, error
);
3303 error
= dmu_objset_destroy(snap3name
, B_FALSE
);
3304 if (error
&& error
!= ENOENT
)
3305 fatal(0, "dmu_objset_destroy(%s) = %d", snap3name
, error
);
3306 error
= dmu_objset_destroy(snap2name
, B_FALSE
);
3307 if (error
&& error
!= ENOENT
)
3308 fatal(0, "dmu_objset_destroy(%s) = %d", snap2name
, error
);
3309 error
= dmu_objset_destroy(clone1name
, B_FALSE
);
3310 if (error
&& error
!= ENOENT
)
3311 fatal(0, "dmu_objset_destroy(%s) = %d", clone1name
, error
);
3312 error
= dmu_objset_destroy(snap1name
, B_FALSE
);
3313 if (error
&& error
!= ENOENT
)
3314 fatal(0, "dmu_objset_destroy(%s) = %d", snap1name
, error
);
3316 umem_free(snap1name
, MAXNAMELEN
);
3317 umem_free(clone1name
, MAXNAMELEN
);
3318 umem_free(snap2name
, MAXNAMELEN
);
3319 umem_free(clone2name
, MAXNAMELEN
);
3320 umem_free(snap3name
, MAXNAMELEN
);
3324 * Verify dsl_dataset_promote handles EBUSY
3327 ztest_dsl_dataset_promote_busy(ztest_ds_t
*zd
, uint64_t id
)
3336 char *osname
= zd
->zd_name
;
3339 snap1name
= umem_alloc(MAXNAMELEN
, UMEM_NOFAIL
);
3340 clone1name
= umem_alloc(MAXNAMELEN
, UMEM_NOFAIL
);
3341 snap2name
= umem_alloc(MAXNAMELEN
, UMEM_NOFAIL
);
3342 clone2name
= umem_alloc(MAXNAMELEN
, UMEM_NOFAIL
);
3343 snap3name
= umem_alloc(MAXNAMELEN
, UMEM_NOFAIL
);
3345 (void) rw_enter(&ztest_name_lock
, RW_READER
);
3347 ztest_dsl_dataset_cleanup(osname
, id
);
3349 (void) snprintf(snap1name
, MAXNAMELEN
, "%s@s1_%llu",
3350 osname
, (u_longlong_t
)id
);
3351 (void) snprintf(clone1name
, MAXNAMELEN
, "%s/c1_%llu",
3352 osname
, (u_longlong_t
)id
);
3353 (void) snprintf(snap2name
, MAXNAMELEN
, "%s@s2_%llu",
3354 clone1name
, (u_longlong_t
)id
);
3355 (void) snprintf(clone2name
, MAXNAMELEN
, "%s/c2_%llu",
3356 osname
, (u_longlong_t
)id
);
3357 (void) snprintf(snap3name
, MAXNAMELEN
, "%s@s3_%llu",
3358 clone1name
, (u_longlong_t
)id
);
3360 error
= dmu_objset_snapshot(osname
, strchr(snap1name
, '@')+1,
3361 NULL
, NULL
, B_FALSE
, B_FALSE
, -1);
3362 if (error
&& error
!= EEXIST
) {
3363 if (error
== ENOSPC
) {
3364 ztest_record_enospc(FTAG
);
3367 fatal(0, "dmu_take_snapshot(%s) = %d", snap1name
, error
);
3370 error
= dmu_objset_hold(snap1name
, FTAG
, &clone
);
3372 fatal(0, "dmu_open_snapshot(%s) = %d", snap1name
, error
);
3374 error
= dmu_objset_clone(clone1name
, dmu_objset_ds(clone
), 0);
3375 dmu_objset_rele(clone
, FTAG
);
3377 if (error
== ENOSPC
) {
3378 ztest_record_enospc(FTAG
);
3381 fatal(0, "dmu_objset_create(%s) = %d", clone1name
, error
);
3384 error
= dmu_objset_snapshot(clone1name
, strchr(snap2name
, '@')+1,
3385 NULL
, NULL
, B_FALSE
, B_FALSE
, -1);
3386 if (error
&& error
!= EEXIST
) {
3387 if (error
== ENOSPC
) {
3388 ztest_record_enospc(FTAG
);
3391 fatal(0, "dmu_open_snapshot(%s) = %d", snap2name
, error
);
3394 error
= dmu_objset_snapshot(clone1name
, strchr(snap3name
, '@')+1,
3395 NULL
, NULL
, B_FALSE
, B_FALSE
, -1);
3396 if (error
&& error
!= EEXIST
) {
3397 if (error
== ENOSPC
) {
3398 ztest_record_enospc(FTAG
);
3401 fatal(0, "dmu_open_snapshot(%s) = %d", snap3name
, error
);
3404 error
= dmu_objset_hold(snap3name
, FTAG
, &clone
);
3406 fatal(0, "dmu_open_snapshot(%s) = %d", snap3name
, error
);
3408 error
= dmu_objset_clone(clone2name
, dmu_objset_ds(clone
), 0);
3409 dmu_objset_rele(clone
, FTAG
);
3411 if (error
== ENOSPC
) {
3412 ztest_record_enospc(FTAG
);
3415 fatal(0, "dmu_objset_create(%s) = %d", clone2name
, error
);
3418 error
= dsl_dataset_own(snap2name
, B_FALSE
, FTAG
, &ds
);
3420 fatal(0, "dsl_dataset_own(%s) = %d", snap2name
, error
);
3421 error
= dsl_dataset_promote(clone2name
, NULL
);
3423 fatal(0, "dsl_dataset_promote(%s), %d, not EBUSY", clone2name
,
3425 dsl_dataset_disown(ds
, FTAG
);
3428 ztest_dsl_dataset_cleanup(osname
, id
);
3430 (void) rw_exit(&ztest_name_lock
);
3432 umem_free(snap1name
, MAXNAMELEN
);
3433 umem_free(clone1name
, MAXNAMELEN
);
3434 umem_free(snap2name
, MAXNAMELEN
);
3435 umem_free(clone2name
, MAXNAMELEN
);
3436 umem_free(snap3name
, MAXNAMELEN
);
3439 #undef OD_ARRAY_SIZE
3440 #define OD_ARRAY_SIZE 4
3443 * Verify that dmu_object_{alloc,free} work as expected.
3446 ztest_dmu_object_alloc_free(ztest_ds_t
*zd
, uint64_t id
)
3453 size
= sizeof(ztest_od_t
) * OD_ARRAY_SIZE
;
3454 od
= umem_alloc(size
, UMEM_NOFAIL
);
3455 batchsize
= OD_ARRAY_SIZE
;
3457 for (b
= 0; b
< batchsize
; b
++)
3458 ztest_od_init(od
+ b
, id
, FTAG
, b
, DMU_OT_UINT64_OTHER
, 0, 0);
3461 * Destroy the previous batch of objects, create a new batch,
3462 * and do some I/O on the new objects.
3464 if (ztest_object_init(zd
, od
, size
, B_TRUE
) != 0)
3467 while (ztest_random(4 * batchsize
) != 0)
3468 ztest_io(zd
, od
[ztest_random(batchsize
)].od_object
,
3469 ztest_random(ZTEST_RANGE_LOCKS
) << SPA_MAXBLOCKSHIFT
);
3471 umem_free(od
, size
);
3474 #undef OD_ARRAY_SIZE
3475 #define OD_ARRAY_SIZE 2
3478 * Verify that dmu_{read,write} work as expected.
3481 ztest_dmu_read_write(ztest_ds_t
*zd
, uint64_t id
)
3486 objset_t
*os
= zd
->zd_os
;
3487 size
= sizeof(ztest_od_t
) * OD_ARRAY_SIZE
;
3488 od
= umem_alloc(size
, UMEM_NOFAIL
);
3490 int i
, freeit
, error
;
3492 bufwad_t
*packbuf
, *bigbuf
, *pack
, *bigH
, *bigT
;
3493 uint64_t packobj
, packoff
, packsize
, bigobj
, bigoff
, bigsize
;
3494 uint64_t chunksize
= (1000 + ztest_random(1000)) * sizeof (uint64_t);
3495 uint64_t regions
= 997;
3496 uint64_t stride
= 123456789ULL;
3497 uint64_t width
= 40;
3498 int free_percent
= 5;
3501 * This test uses two objects, packobj and bigobj, that are always
3502 * updated together (i.e. in the same tx) so that their contents are
3503 * in sync and can be compared. Their contents relate to each other
3504 * in a simple way: packobj is a dense array of 'bufwad' structures,
3505 * while bigobj is a sparse array of the same bufwads. Specifically,
3506 * for any index n, there are three bufwads that should be identical:
3508 * packobj, at offset n * sizeof (bufwad_t)
3509 * bigobj, at the head of the nth chunk
3510 * bigobj, at the tail of the nth chunk
3512 * The chunk size is arbitrary. It doesn't have to be a power of two,
3513 * and it doesn't have any relation to the object blocksize.
3514 * The only requirement is that it can hold at least two bufwads.
3516 * Normally, we write the bufwad to each of these locations.
3517 * However, free_percent of the time we instead write zeroes to
3518 * packobj and perform a dmu_free_range() on bigobj. By comparing
3519 * bigobj to packobj, we can verify that the DMU is correctly
3520 * tracking which parts of an object are allocated and free,
3521 * and that the contents of the allocated blocks are correct.
3525 * Read the directory info. If it's the first time, set things up.
3527 ztest_od_init(od
, id
, FTAG
, 0, DMU_OT_UINT64_OTHER
, 0, chunksize
);
3528 ztest_od_init(od
+ 1, id
, FTAG
, 1, DMU_OT_UINT64_OTHER
, 0, chunksize
);
3530 if (ztest_object_init(zd
, od
, size
, B_FALSE
) != 0) {
3531 umem_free(od
, size
);
3535 bigobj
= od
[0].od_object
;
3536 packobj
= od
[1].od_object
;
3537 chunksize
= od
[0].od_gen
;
3538 ASSERT(chunksize
== od
[1].od_gen
);
3541 * Prefetch a random chunk of the big object.
3542 * Our aim here is to get some async reads in flight
3543 * for blocks that we may free below; the DMU should
3544 * handle this race correctly.
3546 n
= ztest_random(regions
) * stride
+ ztest_random(width
);
3547 s
= 1 + ztest_random(2 * width
- 1);
3548 dmu_prefetch(os
, bigobj
, n
* chunksize
, s
* chunksize
);
3551 * Pick a random index and compute the offsets into packobj and bigobj.
3553 n
= ztest_random(regions
) * stride
+ ztest_random(width
);
3554 s
= 1 + ztest_random(width
- 1);
3556 packoff
= n
* sizeof (bufwad_t
);
3557 packsize
= s
* sizeof (bufwad_t
);
3559 bigoff
= n
* chunksize
;
3560 bigsize
= s
* chunksize
;
3562 packbuf
= umem_alloc(packsize
, UMEM_NOFAIL
);
3563 bigbuf
= umem_alloc(bigsize
, UMEM_NOFAIL
);
3566 * free_percent of the time, free a range of bigobj rather than
3569 freeit
= (ztest_random(100) < free_percent
);
3572 * Read the current contents of our objects.
3574 error
= dmu_read(os
, packobj
, packoff
, packsize
, packbuf
,
3576 ASSERT3U(error
, ==, 0);
3577 error
= dmu_read(os
, bigobj
, bigoff
, bigsize
, bigbuf
,
3579 ASSERT3U(error
, ==, 0);
3582 * Get a tx for the mods to both packobj and bigobj.
3584 tx
= dmu_tx_create(os
);
3586 dmu_tx_hold_write(tx
, packobj
, packoff
, packsize
);
3589 dmu_tx_hold_free(tx
, bigobj
, bigoff
, bigsize
);
3591 dmu_tx_hold_write(tx
, bigobj
, bigoff
, bigsize
);
3593 txg
= ztest_tx_assign(tx
, TXG_MIGHTWAIT
, FTAG
);
3595 umem_free(packbuf
, packsize
);
3596 umem_free(bigbuf
, bigsize
);
3597 umem_free(od
, size
);
3601 dmu_object_set_checksum(os
, bigobj
,
3602 (enum zio_checksum
)ztest_random_dsl_prop(ZFS_PROP_CHECKSUM
), tx
);
3604 dmu_object_set_compress(os
, bigobj
,
3605 (enum zio_compress
)ztest_random_dsl_prop(ZFS_PROP_COMPRESSION
), tx
);
3608 * For each index from n to n + s, verify that the existing bufwad
3609 * in packobj matches the bufwads at the head and tail of the
3610 * corresponding chunk in bigobj. Then update all three bufwads
3611 * with the new values we want to write out.
3613 for (i
= 0; i
< s
; i
++) {
3615 pack
= (bufwad_t
*)((char *)packbuf
+ i
* sizeof (bufwad_t
));
3617 bigH
= (bufwad_t
*)((char *)bigbuf
+ i
* chunksize
);
3619 bigT
= (bufwad_t
*)((char *)bigH
+ chunksize
) - 1;
3621 ASSERT((uintptr_t)bigH
- (uintptr_t)bigbuf
< bigsize
);
3622 ASSERT((uintptr_t)bigT
- (uintptr_t)bigbuf
< bigsize
);
3624 if (pack
->bw_txg
> txg
)
3625 fatal(0, "future leak: got %llx, open txg is %llx",
3628 if (pack
->bw_data
!= 0 && pack
->bw_index
!= n
+ i
)
3629 fatal(0, "wrong index: got %llx, wanted %llx+%llx",
3630 pack
->bw_index
, n
, i
);
3632 if (bcmp(pack
, bigH
, sizeof (bufwad_t
)) != 0)
3633 fatal(0, "pack/bigH mismatch in %p/%p", pack
, bigH
);
3635 if (bcmp(pack
, bigT
, sizeof (bufwad_t
)) != 0)
3636 fatal(0, "pack/bigT mismatch in %p/%p", pack
, bigT
);
3639 bzero(pack
, sizeof (bufwad_t
));
3641 pack
->bw_index
= n
+ i
;
3643 pack
->bw_data
= 1 + ztest_random(-2ULL);
3650 * We've verified all the old bufwads, and made new ones.
3651 * Now write them out.
3653 dmu_write(os
, packobj
, packoff
, packsize
, packbuf
, tx
);
3656 if (ztest_opts
.zo_verbose
>= 7) {
3657 (void) printf("freeing offset %llx size %llx"
3659 (u_longlong_t
)bigoff
,
3660 (u_longlong_t
)bigsize
,
3663 VERIFY(0 == dmu_free_range(os
, bigobj
, bigoff
, bigsize
, tx
));
3665 if (ztest_opts
.zo_verbose
>= 7) {
3666 (void) printf("writing offset %llx size %llx"
3668 (u_longlong_t
)bigoff
,
3669 (u_longlong_t
)bigsize
,
3672 dmu_write(os
, bigobj
, bigoff
, bigsize
, bigbuf
, tx
);
3678 * Sanity check the stuff we just wrote.
3681 void *packcheck
= umem_alloc(packsize
, UMEM_NOFAIL
);
3682 void *bigcheck
= umem_alloc(bigsize
, UMEM_NOFAIL
);
3684 VERIFY(0 == dmu_read(os
, packobj
, packoff
,
3685 packsize
, packcheck
, DMU_READ_PREFETCH
));
3686 VERIFY(0 == dmu_read(os
, bigobj
, bigoff
,
3687 bigsize
, bigcheck
, DMU_READ_PREFETCH
));
3689 ASSERT(bcmp(packbuf
, packcheck
, packsize
) == 0);
3690 ASSERT(bcmp(bigbuf
, bigcheck
, bigsize
) == 0);
3692 umem_free(packcheck
, packsize
);
3693 umem_free(bigcheck
, bigsize
);
3696 umem_free(packbuf
, packsize
);
3697 umem_free(bigbuf
, bigsize
);
3698 umem_free(od
, size
);
3702 compare_and_update_pbbufs(uint64_t s
, bufwad_t
*packbuf
, bufwad_t
*bigbuf
,
3703 uint64_t bigsize
, uint64_t n
, uint64_t chunksize
, uint64_t txg
)
3711 * For each index from n to n + s, verify that the existing bufwad
3712 * in packobj matches the bufwads at the head and tail of the
3713 * corresponding chunk in bigobj. Then update all three bufwads
3714 * with the new values we want to write out.
3716 for (i
= 0; i
< s
; i
++) {
3718 pack
= (bufwad_t
*)((char *)packbuf
+ i
* sizeof (bufwad_t
));
3720 bigH
= (bufwad_t
*)((char *)bigbuf
+ i
* chunksize
);
3722 bigT
= (bufwad_t
*)((char *)bigH
+ chunksize
) - 1;
3724 ASSERT((uintptr_t)bigH
- (uintptr_t)bigbuf
< bigsize
);
3725 ASSERT((uintptr_t)bigT
- (uintptr_t)bigbuf
< bigsize
);
3727 if (pack
->bw_txg
> txg
)
3728 fatal(0, "future leak: got %llx, open txg is %llx",
3731 if (pack
->bw_data
!= 0 && pack
->bw_index
!= n
+ i
)
3732 fatal(0, "wrong index: got %llx, wanted %llx+%llx",
3733 pack
->bw_index
, n
, i
);
3735 if (bcmp(pack
, bigH
, sizeof (bufwad_t
)) != 0)
3736 fatal(0, "pack/bigH mismatch in %p/%p", pack
, bigH
);
3738 if (bcmp(pack
, bigT
, sizeof (bufwad_t
)) != 0)
3739 fatal(0, "pack/bigT mismatch in %p/%p", pack
, bigT
);
3741 pack
->bw_index
= n
+ i
;
3743 pack
->bw_data
= 1 + ztest_random(-2ULL);
3750 #undef OD_ARRAY_SIZE
3751 #define OD_ARRAY_SIZE 2
3754 ztest_dmu_read_write_zcopy(ztest_ds_t
*zd
, uint64_t id
)
3756 objset_t
*os
= zd
->zd_os
;
3763 bufwad_t
*packbuf
, *bigbuf
;
3764 uint64_t packobj
, packoff
, packsize
, bigobj
, bigoff
, bigsize
;
3765 uint64_t blocksize
= ztest_random_blocksize();
3766 uint64_t chunksize
= blocksize
;
3767 uint64_t regions
= 997;
3768 uint64_t stride
= 123456789ULL;
3770 dmu_buf_t
*bonus_db
;
3771 arc_buf_t
**bigbuf_arcbufs
;
3772 dmu_object_info_t doi
;
3774 size
= sizeof(ztest_od_t
) * OD_ARRAY_SIZE
;
3775 od
= umem_alloc(size
, UMEM_NOFAIL
);
3778 * This test uses two objects, packobj and bigobj, that are always
3779 * updated together (i.e. in the same tx) so that their contents are
3780 * in sync and can be compared. Their contents relate to each other
3781 * in a simple way: packobj is a dense array of 'bufwad' structures,
3782 * while bigobj is a sparse array of the same bufwads. Specifically,
3783 * for any index n, there are three bufwads that should be identical:
3785 * packobj, at offset n * sizeof (bufwad_t)
3786 * bigobj, at the head of the nth chunk
3787 * bigobj, at the tail of the nth chunk
3789 * The chunk size is set equal to bigobj block size so that
3790 * dmu_assign_arcbuf() can be tested for object updates.
3794 * Read the directory info. If it's the first time, set things up.
3796 ztest_od_init(od
, id
, FTAG
, 0, DMU_OT_UINT64_OTHER
, blocksize
, 0);
3797 ztest_od_init(od
+ 1, id
, FTAG
, 1, DMU_OT_UINT64_OTHER
, 0, chunksize
);
3800 if (ztest_object_init(zd
, od
, size
, B_FALSE
) != 0) {
3801 umem_free(od
, size
);
3805 bigobj
= od
[0].od_object
;
3806 packobj
= od
[1].od_object
;
3807 blocksize
= od
[0].od_blocksize
;
3808 chunksize
= blocksize
;
3809 ASSERT(chunksize
== od
[1].od_gen
);
3811 VERIFY(dmu_object_info(os
, bigobj
, &doi
) == 0);
3812 VERIFY(ISP2(doi
.doi_data_block_size
));
3813 VERIFY(chunksize
== doi
.doi_data_block_size
);
3814 VERIFY(chunksize
>= 2 * sizeof (bufwad_t
));
3817 * Pick a random index and compute the offsets into packobj and bigobj.
3819 n
= ztest_random(regions
) * stride
+ ztest_random(width
);
3820 s
= 1 + ztest_random(width
- 1);
3822 packoff
= n
* sizeof (bufwad_t
);
3823 packsize
= s
* sizeof (bufwad_t
);
3825 bigoff
= n
* chunksize
;
3826 bigsize
= s
* chunksize
;
3828 packbuf
= umem_zalloc(packsize
, UMEM_NOFAIL
);
3829 bigbuf
= umem_zalloc(bigsize
, UMEM_NOFAIL
);
3831 VERIFY3U(0, ==, dmu_bonus_hold(os
, bigobj
, FTAG
, &bonus_db
));
3833 bigbuf_arcbufs
= umem_zalloc(2 * s
* sizeof (arc_buf_t
*), UMEM_NOFAIL
);
3836 * Iteration 0 test zcopy for DB_UNCACHED dbufs.
3837 * Iteration 1 test zcopy to already referenced dbufs.
3838 * Iteration 2 test zcopy to dirty dbuf in the same txg.
3839 * Iteration 3 test zcopy to dbuf dirty in previous txg.
3840 * Iteration 4 test zcopy when dbuf is no longer dirty.
3841 * Iteration 5 test zcopy when it can't be done.
3842 * Iteration 6 one more zcopy write.
3844 for (i
= 0; i
< 7; i
++) {
3849 * In iteration 5 (i == 5) use arcbufs
3850 * that don't match bigobj blksz to test
3851 * dmu_assign_arcbuf() when it can't directly
3852 * assign an arcbuf to a dbuf.
3854 for (j
= 0; j
< s
; j
++) {
3857 dmu_request_arcbuf(bonus_db
, chunksize
);
3859 bigbuf_arcbufs
[2 * j
] =
3860 dmu_request_arcbuf(bonus_db
, chunksize
/ 2);
3861 bigbuf_arcbufs
[2 * j
+ 1] =
3862 dmu_request_arcbuf(bonus_db
, chunksize
/ 2);
3867 * Get a tx for the mods to both packobj and bigobj.
3869 tx
= dmu_tx_create(os
);
3871 dmu_tx_hold_write(tx
, packobj
, packoff
, packsize
);
3872 dmu_tx_hold_write(tx
, bigobj
, bigoff
, bigsize
);
3874 txg
= ztest_tx_assign(tx
, TXG_MIGHTWAIT
, FTAG
);
3876 umem_free(packbuf
, packsize
);
3877 umem_free(bigbuf
, bigsize
);
3878 for (j
= 0; j
< s
; j
++) {
3880 dmu_return_arcbuf(bigbuf_arcbufs
[j
]);
3883 bigbuf_arcbufs
[2 * j
]);
3885 bigbuf_arcbufs
[2 * j
+ 1]);
3888 umem_free(bigbuf_arcbufs
, 2 * s
* sizeof (arc_buf_t
*));
3889 umem_free(od
, size
);
3890 dmu_buf_rele(bonus_db
, FTAG
);
3895 * 50% of the time don't read objects in the 1st iteration to
3896 * test dmu_assign_arcbuf() for the case when there're no
3897 * existing dbufs for the specified offsets.
3899 if (i
!= 0 || ztest_random(2) != 0) {
3900 error
= dmu_read(os
, packobj
, packoff
,
3901 packsize
, packbuf
, DMU_READ_PREFETCH
);
3902 ASSERT3U(error
, ==, 0);
3903 error
= dmu_read(os
, bigobj
, bigoff
, bigsize
,
3904 bigbuf
, DMU_READ_PREFETCH
);
3905 ASSERT3U(error
, ==, 0);
3907 compare_and_update_pbbufs(s
, packbuf
, bigbuf
, bigsize
,
3911 * We've verified all the old bufwads, and made new ones.
3912 * Now write them out.
3914 dmu_write(os
, packobj
, packoff
, packsize
, packbuf
, tx
);
3915 if (ztest_opts
.zo_verbose
>= 7) {
3916 (void) printf("writing offset %llx size %llx"
3918 (u_longlong_t
)bigoff
,
3919 (u_longlong_t
)bigsize
,
3922 for (off
= bigoff
, j
= 0; j
< s
; j
++, off
+= chunksize
) {
3925 bcopy((caddr_t
)bigbuf
+ (off
- bigoff
),
3926 bigbuf_arcbufs
[j
]->b_data
, chunksize
);
3928 bcopy((caddr_t
)bigbuf
+ (off
- bigoff
),
3929 bigbuf_arcbufs
[2 * j
]->b_data
,
3931 bcopy((caddr_t
)bigbuf
+ (off
- bigoff
) +
3933 bigbuf_arcbufs
[2 * j
+ 1]->b_data
,
3938 VERIFY(dmu_buf_hold(os
, bigobj
, off
,
3939 FTAG
, &dbt
, DMU_READ_NO_PREFETCH
) == 0);
3942 dmu_assign_arcbuf(bonus_db
, off
,
3943 bigbuf_arcbufs
[j
], tx
);
3945 dmu_assign_arcbuf(bonus_db
, off
,
3946 bigbuf_arcbufs
[2 * j
], tx
);
3947 dmu_assign_arcbuf(bonus_db
,
3948 off
+ chunksize
/ 2,
3949 bigbuf_arcbufs
[2 * j
+ 1], tx
);
3952 dmu_buf_rele(dbt
, FTAG
);
3958 * Sanity check the stuff we just wrote.
3961 void *packcheck
= umem_alloc(packsize
, UMEM_NOFAIL
);
3962 void *bigcheck
= umem_alloc(bigsize
, UMEM_NOFAIL
);
3964 VERIFY(0 == dmu_read(os
, packobj
, packoff
,
3965 packsize
, packcheck
, DMU_READ_PREFETCH
));
3966 VERIFY(0 == dmu_read(os
, bigobj
, bigoff
,
3967 bigsize
, bigcheck
, DMU_READ_PREFETCH
));
3969 ASSERT(bcmp(packbuf
, packcheck
, packsize
) == 0);
3970 ASSERT(bcmp(bigbuf
, bigcheck
, bigsize
) == 0);
3972 umem_free(packcheck
, packsize
);
3973 umem_free(bigcheck
, bigsize
);
3976 txg_wait_open(dmu_objset_pool(os
), 0);
3977 } else if (i
== 3) {
3978 txg_wait_synced(dmu_objset_pool(os
), 0);
3982 dmu_buf_rele(bonus_db
, FTAG
);
3983 umem_free(packbuf
, packsize
);
3984 umem_free(bigbuf
, bigsize
);
3985 umem_free(bigbuf_arcbufs
, 2 * s
* sizeof (arc_buf_t
*));
3986 umem_free(od
, size
);
3991 ztest_dmu_write_parallel(ztest_ds_t
*zd
, uint64_t id
)
3995 od
= umem_alloc(sizeof(ztest_od_t
), UMEM_NOFAIL
);
3996 uint64_t offset
= (1ULL << (ztest_random(20) + 43)) +
3997 (ztest_random(ZTEST_RANGE_LOCKS
) << SPA_MAXBLOCKSHIFT
);
4000 * Have multiple threads write to large offsets in an object
4001 * to verify that parallel writes to an object -- even to the
4002 * same blocks within the object -- doesn't cause any trouble.
4004 ztest_od_init(od
, ID_PARALLEL
, FTAG
, 0, DMU_OT_UINT64_OTHER
, 0, 0);
4006 if (ztest_object_init(zd
, od
, sizeof (ztest_od_t
), B_FALSE
) != 0)
4009 while (ztest_random(10) != 0)
4010 ztest_io(zd
, od
->od_object
, offset
);
4012 umem_free(od
, sizeof(ztest_od_t
));
4016 ztest_dmu_prealloc(ztest_ds_t
*zd
, uint64_t id
)
4019 uint64_t offset
= (1ULL << (ztest_random(4) + SPA_MAXBLOCKSHIFT
)) +
4020 (ztest_random(ZTEST_RANGE_LOCKS
) << SPA_MAXBLOCKSHIFT
);
4021 uint64_t count
= ztest_random(20) + 1;
4022 uint64_t blocksize
= ztest_random_blocksize();
4025 od
= umem_alloc(sizeof(ztest_od_t
), UMEM_NOFAIL
);
4027 ztest_od_init(od
, id
, FTAG
, 0, DMU_OT_UINT64_OTHER
, blocksize
, 0);
4029 if (ztest_object_init(zd
, od
, sizeof (ztest_od_t
), !ztest_random(2)) != 0) {
4030 umem_free(od
, sizeof(ztest_od_t
));
4034 if (ztest_truncate(zd
, od
->od_object
, offset
, count
* blocksize
) != 0) {
4035 umem_free(od
, sizeof(ztest_od_t
));
4039 ztest_prealloc(zd
, od
->od_object
, offset
, count
* blocksize
);
4041 data
= umem_zalloc(blocksize
, UMEM_NOFAIL
);
4043 while (ztest_random(count
) != 0) {
4044 uint64_t randoff
= offset
+ (ztest_random(count
) * blocksize
);
4045 if (ztest_write(zd
, od
->od_object
, randoff
, blocksize
,
4048 while (ztest_random(4) != 0)
4049 ztest_io(zd
, od
->od_object
, randoff
);
4052 umem_free(data
, blocksize
);
4053 umem_free(od
, sizeof(ztest_od_t
));
4057 * Verify that zap_{create,destroy,add,remove,update} work as expected.
4059 #define ZTEST_ZAP_MIN_INTS 1
4060 #define ZTEST_ZAP_MAX_INTS 4
4061 #define ZTEST_ZAP_MAX_PROPS 1000
4064 ztest_zap(ztest_ds_t
*zd
, uint64_t id
)
4066 objset_t
*os
= zd
->zd_os
;
4069 uint64_t txg
, last_txg
;
4070 uint64_t value
[ZTEST_ZAP_MAX_INTS
];
4071 uint64_t zl_ints
, zl_intsize
, prop
;
4074 char propname
[100], txgname
[100];
4076 char *hc
[2] = { "s.acl.h", ".s.open.h.hyLZlg" };
4078 od
= umem_alloc(sizeof(ztest_od_t
), UMEM_NOFAIL
);
4079 ztest_od_init(od
, id
, FTAG
, 0, DMU_OT_ZAP_OTHER
, 0, 0);
4081 if (ztest_object_init(zd
, od
, sizeof (ztest_od_t
),
4082 !ztest_random(2)) != 0)
4085 object
= od
->od_object
;
4088 * Generate a known hash collision, and verify that
4089 * we can lookup and remove both entries.
4091 tx
= dmu_tx_create(os
);
4092 dmu_tx_hold_zap(tx
, object
, B_TRUE
, NULL
);
4093 txg
= ztest_tx_assign(tx
, TXG_MIGHTWAIT
, FTAG
);
4096 for (i
= 0; i
< 2; i
++) {
4098 VERIFY3U(0, ==, zap_add(os
, object
, hc
[i
], sizeof (uint64_t),
4101 for (i
= 0; i
< 2; i
++) {
4102 VERIFY3U(EEXIST
, ==, zap_add(os
, object
, hc
[i
],
4103 sizeof (uint64_t), 1, &value
[i
], tx
));
4105 zap_length(os
, object
, hc
[i
], &zl_intsize
, &zl_ints
));
4106 ASSERT3U(zl_intsize
, ==, sizeof (uint64_t));
4107 ASSERT3U(zl_ints
, ==, 1);
4109 for (i
= 0; i
< 2; i
++) {
4110 VERIFY3U(0, ==, zap_remove(os
, object
, hc
[i
], tx
));
4115 * Generate a buch of random entries.
4117 ints
= MAX(ZTEST_ZAP_MIN_INTS
, object
% ZTEST_ZAP_MAX_INTS
);
4119 prop
= ztest_random(ZTEST_ZAP_MAX_PROPS
);
4120 (void) sprintf(propname
, "prop_%llu", (u_longlong_t
)prop
);
4121 (void) sprintf(txgname
, "txg_%llu", (u_longlong_t
)prop
);
4122 bzero(value
, sizeof (value
));
4126 * If these zap entries already exist, validate their contents.
4128 error
= zap_length(os
, object
, txgname
, &zl_intsize
, &zl_ints
);
4130 ASSERT3U(zl_intsize
, ==, sizeof (uint64_t));
4131 ASSERT3U(zl_ints
, ==, 1);
4133 VERIFY(zap_lookup(os
, object
, txgname
, zl_intsize
,
4134 zl_ints
, &last_txg
) == 0);
4136 VERIFY(zap_length(os
, object
, propname
, &zl_intsize
,
4139 ASSERT3U(zl_intsize
, ==, sizeof (uint64_t));
4140 ASSERT3U(zl_ints
, ==, ints
);
4142 VERIFY(zap_lookup(os
, object
, propname
, zl_intsize
,
4143 zl_ints
, value
) == 0);
4145 for (i
= 0; i
< ints
; i
++) {
4146 ASSERT3U(value
[i
], ==, last_txg
+ object
+ i
);
4149 ASSERT3U(error
, ==, ENOENT
);
4153 * Atomically update two entries in our zap object.
4154 * The first is named txg_%llu, and contains the txg
4155 * in which the property was last updated. The second
4156 * is named prop_%llu, and the nth element of its value
4157 * should be txg + object + n.
4159 tx
= dmu_tx_create(os
);
4160 dmu_tx_hold_zap(tx
, object
, B_TRUE
, NULL
);
4161 txg
= ztest_tx_assign(tx
, TXG_MIGHTWAIT
, FTAG
);
4166 fatal(0, "zap future leak: old %llu new %llu", last_txg
, txg
);
4168 for (i
= 0; i
< ints
; i
++)
4169 value
[i
] = txg
+ object
+ i
;
4171 VERIFY3U(0, ==, zap_update(os
, object
, txgname
, sizeof (uint64_t),
4173 VERIFY3U(0, ==, zap_update(os
, object
, propname
, sizeof (uint64_t),
4179 * Remove a random pair of entries.
4181 prop
= ztest_random(ZTEST_ZAP_MAX_PROPS
);
4182 (void) sprintf(propname
, "prop_%llu", (u_longlong_t
)prop
);
4183 (void) sprintf(txgname
, "txg_%llu", (u_longlong_t
)prop
);
4185 error
= zap_length(os
, object
, txgname
, &zl_intsize
, &zl_ints
);
4187 if (error
== ENOENT
)
4190 ASSERT3U(error
, ==, 0);
4192 tx
= dmu_tx_create(os
);
4193 dmu_tx_hold_zap(tx
, object
, B_TRUE
, NULL
);
4194 txg
= ztest_tx_assign(tx
, TXG_MIGHTWAIT
, FTAG
);
4197 VERIFY3U(0, ==, zap_remove(os
, object
, txgname
, tx
));
4198 VERIFY3U(0, ==, zap_remove(os
, object
, propname
, tx
));
4201 umem_free(od
, sizeof(ztest_od_t
));
4205 * Testcase to test the upgrading of a microzap to fatzap.
4208 ztest_fzap(ztest_ds_t
*zd
, uint64_t id
)
4210 objset_t
*os
= zd
->zd_os
;
4212 uint64_t object
, txg
;
4215 od
= umem_alloc(sizeof(ztest_od_t
), UMEM_NOFAIL
);
4216 ztest_od_init(od
, id
, FTAG
, 0, DMU_OT_ZAP_OTHER
, 0, 0);
4218 if (ztest_object_init(zd
, od
, sizeof (ztest_od_t
),
4219 !ztest_random(2)) != 0)
4221 object
= od
->od_object
;
4224 * Add entries to this ZAP and make sure it spills over
4225 * and gets upgraded to a fatzap. Also, since we are adding
4226 * 2050 entries we should see ptrtbl growth and leaf-block split.
4228 for (i
= 0; i
< 2050; i
++) {
4229 char name
[MAXNAMELEN
];
4234 (void) snprintf(name
, sizeof (name
), "fzap-%llu-%llu",
4235 (u_longlong_t
)id
, (u_longlong_t
)value
);
4237 tx
= dmu_tx_create(os
);
4238 dmu_tx_hold_zap(tx
, object
, B_TRUE
, name
);
4239 txg
= ztest_tx_assign(tx
, TXG_MIGHTWAIT
, FTAG
);
4242 error
= zap_add(os
, object
, name
, sizeof (uint64_t), 1,
4244 ASSERT(error
== 0 || error
== EEXIST
);
4248 umem_free(od
, sizeof(ztest_od_t
));
4253 ztest_zap_parallel(ztest_ds_t
*zd
, uint64_t id
)
4255 objset_t
*os
= zd
->zd_os
;
4257 uint64_t txg
, object
, count
, wsize
, wc
, zl_wsize
, zl_wc
;
4259 int i
, namelen
, error
;
4260 int micro
= ztest_random(2);
4261 char name
[20], string_value
[20];
4264 od
= umem_alloc(sizeof(ztest_od_t
), UMEM_NOFAIL
);
4265 ztest_od_init(od
, ID_PARALLEL
, FTAG
, micro
, DMU_OT_ZAP_OTHER
, 0, 0);
4267 if (ztest_object_init(zd
, od
, sizeof (ztest_od_t
), B_FALSE
) != 0) {
4268 umem_free(od
, sizeof(ztest_od_t
));
4272 object
= od
->od_object
;
4275 * Generate a random name of the form 'xxx.....' where each
4276 * x is a random printable character and the dots are dots.
4277 * There are 94 such characters, and the name length goes from
4278 * 6 to 20, so there are 94^3 * 15 = 12,458,760 possible names.
4280 namelen
= ztest_random(sizeof (name
) - 5) + 5 + 1;
4282 for (i
= 0; i
< 3; i
++)
4283 name
[i
] = '!' + ztest_random('~' - '!' + 1);
4284 for (; i
< namelen
- 1; i
++)
4288 if ((namelen
& 1) || micro
) {
4289 wsize
= sizeof (txg
);
4295 data
= string_value
;
4299 VERIFY(zap_count(os
, object
, &count
) == 0);
4300 ASSERT(count
!= -1ULL);
4303 * Select an operation: length, lookup, add, update, remove.
4305 i
= ztest_random(5);
4308 tx
= dmu_tx_create(os
);
4309 dmu_tx_hold_zap(tx
, object
, B_TRUE
, NULL
);
4310 txg
= ztest_tx_assign(tx
, TXG_MIGHTWAIT
, FTAG
);
4313 bcopy(name
, string_value
, namelen
);
4317 bzero(string_value
, namelen
);
4323 error
= zap_length(os
, object
, name
, &zl_wsize
, &zl_wc
);
4325 ASSERT3U(wsize
, ==, zl_wsize
);
4326 ASSERT3U(wc
, ==, zl_wc
);
4328 ASSERT3U(error
, ==, ENOENT
);
4333 error
= zap_lookup(os
, object
, name
, wsize
, wc
, data
);
4335 if (data
== string_value
&&
4336 bcmp(name
, data
, namelen
) != 0)
4337 fatal(0, "name '%s' != val '%s' len %d",
4338 name
, data
, namelen
);
4340 ASSERT3U(error
, ==, ENOENT
);
4345 error
= zap_add(os
, object
, name
, wsize
, wc
, data
, tx
);
4346 ASSERT(error
== 0 || error
== EEXIST
);
4350 VERIFY(zap_update(os
, object
, name
, wsize
, wc
, data
, tx
) == 0);
4354 error
= zap_remove(os
, object
, name
, tx
);
4355 ASSERT(error
== 0 || error
== ENOENT
);
4362 umem_free(od
, sizeof(ztest_od_t
));
4366 * Commit callback data.
4368 typedef struct ztest_cb_data
{
4369 list_node_t zcd_node
;
4371 int zcd_expected_err
;
4372 boolean_t zcd_added
;
4373 boolean_t zcd_called
;
4377 /* This is the actual commit callback function */
4379 ztest_commit_callback(void *arg
, int error
)
4381 ztest_cb_data_t
*data
= arg
;
4382 uint64_t synced_txg
;
4384 VERIFY(data
!= NULL
);
4385 VERIFY3S(data
->zcd_expected_err
, ==, error
);
4386 VERIFY(!data
->zcd_called
);
4388 synced_txg
= spa_last_synced_txg(data
->zcd_spa
);
4389 if (data
->zcd_txg
> synced_txg
)
4390 fatal(0, "commit callback of txg %" PRIu64
" called prematurely"
4391 ", last synced txg = %" PRIu64
"\n", data
->zcd_txg
,
4394 data
->zcd_called
= B_TRUE
;
4396 if (error
== ECANCELED
) {
4397 ASSERT3U(data
->zcd_txg
, ==, 0);
4398 ASSERT(!data
->zcd_added
);
4401 * The private callback data should be destroyed here, but
4402 * since we are going to check the zcd_called field after
4403 * dmu_tx_abort(), we will destroy it there.
4408 ASSERT(data
->zcd_added
);
4409 ASSERT3U(data
->zcd_txg
, !=, 0);
4411 (void) mutex_enter(&zcl
.zcl_callbacks_lock
);
4413 /* See if this cb was called more quickly */
4414 if ((synced_txg
- data
->zcd_txg
) < zc_min_txg_delay
)
4415 zc_min_txg_delay
= synced_txg
- data
->zcd_txg
;
4417 /* Remove our callback from the list */
4418 list_remove(&zcl
.zcl_callbacks
, data
);
4420 (void) mutex_exit(&zcl
.zcl_callbacks_lock
);
4422 umem_free(data
, sizeof (ztest_cb_data_t
));
4425 /* Allocate and initialize callback data structure */
4426 static ztest_cb_data_t
*
4427 ztest_create_cb_data(objset_t
*os
, uint64_t txg
)
4429 ztest_cb_data_t
*cb_data
;
4431 cb_data
= umem_zalloc(sizeof (ztest_cb_data_t
), UMEM_NOFAIL
);
4433 cb_data
->zcd_txg
= txg
;
4434 cb_data
->zcd_spa
= dmu_objset_spa(os
);
4435 list_link_init(&cb_data
->zcd_node
);
4441 * Commit callback test.
4444 ztest_dmu_commit_callbacks(ztest_ds_t
*zd
, uint64_t id
)
4446 objset_t
*os
= zd
->zd_os
;
4449 ztest_cb_data_t
*cb_data
[3], *tmp_cb
;
4450 uint64_t old_txg
, txg
;
4453 od
= umem_alloc(sizeof(ztest_od_t
), UMEM_NOFAIL
);
4454 ztest_od_init(od
, id
, FTAG
, 0, DMU_OT_UINT64_OTHER
, 0, 0);
4456 if (ztest_object_init(zd
, od
, sizeof (ztest_od_t
), B_FALSE
) != 0) {
4457 umem_free(od
, sizeof(ztest_od_t
));
4461 tx
= dmu_tx_create(os
);
4463 cb_data
[0] = ztest_create_cb_data(os
, 0);
4464 dmu_tx_callback_register(tx
, ztest_commit_callback
, cb_data
[0]);
4466 dmu_tx_hold_write(tx
, od
->od_object
, 0, sizeof (uint64_t));
4468 /* Every once in a while, abort the transaction on purpose */
4469 if (ztest_random(100) == 0)
4473 error
= dmu_tx_assign(tx
, TXG_NOWAIT
);
4475 txg
= error
? 0 : dmu_tx_get_txg(tx
);
4477 cb_data
[0]->zcd_txg
= txg
;
4478 cb_data
[1] = ztest_create_cb_data(os
, txg
);
4479 dmu_tx_callback_register(tx
, ztest_commit_callback
, cb_data
[1]);
4483 * It's not a strict requirement to call the registered
4484 * callbacks from inside dmu_tx_abort(), but that's what
4485 * it's supposed to happen in the current implementation
4486 * so we will check for that.
4488 for (i
= 0; i
< 2; i
++) {
4489 cb_data
[i
]->zcd_expected_err
= ECANCELED
;
4490 VERIFY(!cb_data
[i
]->zcd_called
);
4495 for (i
= 0; i
< 2; i
++) {
4496 VERIFY(cb_data
[i
]->zcd_called
);
4497 umem_free(cb_data
[i
], sizeof (ztest_cb_data_t
));
4500 umem_free(od
, sizeof(ztest_od_t
));
4504 cb_data
[2] = ztest_create_cb_data(os
, txg
);
4505 dmu_tx_callback_register(tx
, ztest_commit_callback
, cb_data
[2]);
4508 * Read existing data to make sure there isn't a future leak.
4510 VERIFY(0 == dmu_read(os
, od
->od_object
, 0, sizeof (uint64_t),
4511 &old_txg
, DMU_READ_PREFETCH
));
4514 fatal(0, "future leak: got %" PRIu64
", open txg is %" PRIu64
,
4517 dmu_write(os
, od
->od_object
, 0, sizeof (uint64_t), &txg
, tx
);
4519 (void) mutex_enter(&zcl
.zcl_callbacks_lock
);
4522 * Since commit callbacks don't have any ordering requirement and since
4523 * it is theoretically possible for a commit callback to be called
4524 * after an arbitrary amount of time has elapsed since its txg has been
4525 * synced, it is difficult to reliably determine whether a commit
4526 * callback hasn't been called due to high load or due to a flawed
4529 * In practice, we will assume that if after a certain number of txgs a
4530 * commit callback hasn't been called, then most likely there's an
4531 * implementation bug..
4533 tmp_cb
= list_head(&zcl
.zcl_callbacks
);
4534 if (tmp_cb
!= NULL
&&
4535 tmp_cb
->zcd_txg
+ ZTEST_COMMIT_CB_THRESH
< txg
) {
4536 fatal(0, "Commit callback threshold exceeded, oldest txg: %"
4537 PRIu64
", open txg: %" PRIu64
"\n", tmp_cb
->zcd_txg
, txg
);
4541 * Let's find the place to insert our callbacks.
4543 * Even though the list is ordered by txg, it is possible for the
4544 * insertion point to not be the end because our txg may already be
4545 * quiescing at this point and other callbacks in the open txg
4546 * (from other objsets) may have sneaked in.
4548 tmp_cb
= list_tail(&zcl
.zcl_callbacks
);
4549 while (tmp_cb
!= NULL
&& tmp_cb
->zcd_txg
> txg
)
4550 tmp_cb
= list_prev(&zcl
.zcl_callbacks
, tmp_cb
);
4552 /* Add the 3 callbacks to the list */
4553 for (i
= 0; i
< 3; i
++) {
4555 list_insert_head(&zcl
.zcl_callbacks
, cb_data
[i
]);
4557 list_insert_after(&zcl
.zcl_callbacks
, tmp_cb
,
4560 cb_data
[i
]->zcd_added
= B_TRUE
;
4561 VERIFY(!cb_data
[i
]->zcd_called
);
4563 tmp_cb
= cb_data
[i
];
4568 (void) mutex_exit(&zcl
.zcl_callbacks_lock
);
4572 umem_free(od
, sizeof(ztest_od_t
));
4577 ztest_dsl_prop_get_set(ztest_ds_t
*zd
, uint64_t id
)
4579 zfs_prop_t proplist
[] = {
4581 ZFS_PROP_COMPRESSION
,
4587 (void) rw_enter(&ztest_name_lock
, RW_READER
);
4589 for (p
= 0; p
< sizeof (proplist
) / sizeof (proplist
[0]); p
++)
4590 (void) ztest_dsl_prop_set_uint64(zd
->zd_name
, proplist
[p
],
4591 ztest_random_dsl_prop(proplist
[p
]), (int)ztest_random(2));
4593 (void) rw_exit(&ztest_name_lock
);
4598 ztest_spa_prop_get_set(ztest_ds_t
*zd
, uint64_t id
)
4600 nvlist_t
*props
= NULL
;
4602 (void) rw_enter(&ztest_name_lock
, RW_READER
);
4604 (void) ztest_spa_prop_set_uint64(ZPOOL_PROP_DEDUPDITTO
,
4605 ZIO_DEDUPDITTO_MIN
+ ztest_random(ZIO_DEDUPDITTO_MIN
));
4607 VERIFY3U(spa_prop_get(ztest_spa
, &props
), ==, 0);
4609 if (ztest_opts
.zo_verbose
>= 6)
4610 dump_nvlist(props
, 4);
4614 (void) rw_exit(&ztest_name_lock
);
4618 * Test snapshot hold/release and deferred destroy.
4621 ztest_dmu_snapshot_hold(ztest_ds_t
*zd
, uint64_t id
)
4624 objset_t
*os
= zd
->zd_os
;
4628 char clonename
[100];
4630 char osname
[MAXNAMELEN
];
4632 (void) rw_enter(&ztest_name_lock
, RW_READER
);
4634 dmu_objset_name(os
, osname
);
4636 (void) snprintf(snapname
, 100, "sh1_%llu", (u_longlong_t
)id
);
4637 (void) snprintf(fullname
, 100, "%s@%s", osname
, snapname
);
4638 (void) snprintf(clonename
, 100, "%s/ch1_%llu",osname
,(u_longlong_t
)id
);
4639 (void) snprintf(tag
, 100, "tag_%llu", (u_longlong_t
)id
);
4642 * Clean up from any previous run.
4644 (void) dmu_objset_destroy(clonename
, B_FALSE
);
4645 (void) dsl_dataset_user_release(osname
, snapname
, tag
, B_FALSE
);
4646 (void) dmu_objset_destroy(fullname
, B_FALSE
);
4649 * Create snapshot, clone it, mark snap for deferred destroy,
4650 * destroy clone, verify snap was also destroyed.
4652 error
= dmu_objset_snapshot(osname
, snapname
, NULL
, NULL
, FALSE
,
4655 if (error
== ENOSPC
) {
4656 ztest_record_enospc("dmu_objset_snapshot");
4659 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname
, error
);
4662 error
= dmu_objset_hold(fullname
, FTAG
, &origin
);
4664 fatal(0, "dmu_objset_hold(%s) = %d", fullname
, error
);
4666 error
= dmu_objset_clone(clonename
, dmu_objset_ds(origin
), 0);
4667 dmu_objset_rele(origin
, FTAG
);
4669 if (error
== ENOSPC
) {
4670 ztest_record_enospc("dmu_objset_clone");
4673 fatal(0, "dmu_objset_clone(%s) = %d", clonename
, error
);
4676 error
= dmu_objset_destroy(fullname
, B_TRUE
);
4678 fatal(0, "dmu_objset_destroy(%s, B_TRUE) = %d",
4682 error
= dmu_objset_destroy(clonename
, B_FALSE
);
4684 fatal(0, "dmu_objset_destroy(%s) = %d", clonename
, error
);
4686 error
= dmu_objset_hold(fullname
, FTAG
, &origin
);
4687 if (error
!= ENOENT
)
4688 fatal(0, "dmu_objset_hold(%s) = %d", fullname
, error
);
4691 * Create snapshot, add temporary hold, verify that we can't
4692 * destroy a held snapshot, mark for deferred destroy,
4693 * release hold, verify snapshot was destroyed.
4695 error
= dmu_objset_snapshot(osname
, snapname
, NULL
, NULL
, FALSE
,
4698 if (error
== ENOSPC
) {
4699 ztest_record_enospc("dmu_objset_snapshot");
4702 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname
, error
);
4705 error
= dsl_dataset_user_hold(osname
, snapname
, tag
, B_FALSE
,
4708 fatal(0, "dsl_dataset_user_hold(%s)", fullname
, tag
);
4710 error
= dmu_objset_destroy(fullname
, B_FALSE
);
4711 if (error
!= EBUSY
) {
4712 fatal(0, "dmu_objset_destroy(%s, B_FALSE) = %d",
4716 error
= dmu_objset_destroy(fullname
, B_TRUE
);
4718 fatal(0, "dmu_objset_destroy(%s, B_TRUE) = %d",
4722 error
= dsl_dataset_user_release(osname
, snapname
, tag
, B_FALSE
);
4724 fatal(0, "dsl_dataset_user_release(%s)", fullname
, tag
);
4726 VERIFY(dmu_objset_hold(fullname
, FTAG
, &origin
) == ENOENT
);
4729 (void) rw_exit(&ztest_name_lock
);
4733 * Inject random faults into the on-disk data.
4737 ztest_fault_inject(ztest_ds_t
*zd
, uint64_t id
)
4739 ztest_shared_t
*zs
= ztest_shared
;
4740 spa_t
*spa
= ztest_spa
;
4744 uint64_t bad
= 0x1990c0ffeedecadeull
;
4749 int bshift
= SPA_MAXBLOCKSHIFT
+ 2; /* don't scrog all labels */
4755 boolean_t islog
= B_FALSE
;
4757 path0
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
4758 pathrand
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
4760 mutex_enter(&ztest_vdev_lock
);
4761 maxfaults
= MAXFAULTS();
4762 leaves
= MAX(zs
->zs_mirrors
, 1) * ztest_opts
.zo_raidz
;
4763 mirror_save
= zs
->zs_mirrors
;
4764 mutex_exit(&ztest_vdev_lock
);
4766 ASSERT(leaves
>= 1);
4769 * We need SCL_STATE here because we're going to look at vd0->vdev_tsd.
4771 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
4773 if (ztest_random(2) == 0) {
4775 * Inject errors on a normal data device or slog device.
4777 top
= ztest_random_vdev_top(spa
, B_TRUE
);
4778 leaf
= ztest_random(leaves
) + zs
->zs_splits
;
4781 * Generate paths to the first leaf in this top-level vdev,
4782 * and to the random leaf we selected. We'll induce transient
4783 * write failures and random online/offline activity on leaf 0,
4784 * and we'll write random garbage to the randomly chosen leaf.
4786 (void) snprintf(path0
, MAXPATHLEN
, ztest_dev_template
,
4787 ztest_opts
.zo_dir
, ztest_opts
.zo_pool
,
4788 top
* leaves
+ zs
->zs_splits
);
4789 (void) snprintf(pathrand
, MAXPATHLEN
, ztest_dev_template
,
4790 ztest_opts
.zo_dir
, ztest_opts
.zo_pool
,
4791 top
* leaves
+ leaf
);
4793 vd0
= vdev_lookup_by_path(spa
->spa_root_vdev
, path0
);
4794 if (vd0
!= NULL
&& vd0
->vdev_top
->vdev_islog
)
4797 if (vd0
!= NULL
&& maxfaults
!= 1) {
4799 * Make vd0 explicitly claim to be unreadable,
4800 * or unwriteable, or reach behind its back
4801 * and close the underlying fd. We can do this if
4802 * maxfaults == 0 because we'll fail and reexecute,
4803 * and we can do it if maxfaults >= 2 because we'll
4804 * have enough redundancy. If maxfaults == 1, the
4805 * combination of this with injection of random data
4806 * corruption below exceeds the pool's fault tolerance.
4808 vdev_file_t
*vf
= vd0
->vdev_tsd
;
4810 if (vf
!= NULL
&& ztest_random(3) == 0) {
4811 (void) close(vf
->vf_vnode
->v_fd
);
4812 vf
->vf_vnode
->v_fd
= -1;
4813 } else if (ztest_random(2) == 0) {
4814 vd0
->vdev_cant_read
= B_TRUE
;
4816 vd0
->vdev_cant_write
= B_TRUE
;
4818 guid0
= vd0
->vdev_guid
;
4822 * Inject errors on an l2cache device.
4824 spa_aux_vdev_t
*sav
= &spa
->spa_l2cache
;
4826 if (sav
->sav_count
== 0) {
4827 spa_config_exit(spa
, SCL_STATE
, FTAG
);
4830 vd0
= sav
->sav_vdevs
[ztest_random(sav
->sav_count
)];
4831 guid0
= vd0
->vdev_guid
;
4832 (void) strcpy(path0
, vd0
->vdev_path
);
4833 (void) strcpy(pathrand
, vd0
->vdev_path
);
4837 maxfaults
= INT_MAX
; /* no limit on cache devices */
4840 spa_config_exit(spa
, SCL_STATE
, FTAG
);
4843 * If we can tolerate two or more faults, or we're dealing
4844 * with a slog, randomly online/offline vd0.
4846 if ((maxfaults
>= 2 || islog
) && guid0
!= 0) {
4847 if (ztest_random(10) < 6) {
4848 int flags
= (ztest_random(2) == 0 ?
4849 ZFS_OFFLINE_TEMPORARY
: 0);
4852 * We have to grab the zs_name_lock as writer to
4853 * prevent a race between offlining a slog and
4854 * destroying a dataset. Offlining the slog will
4855 * grab a reference on the dataset which may cause
4856 * dmu_objset_destroy() to fail with EBUSY thus
4857 * leaving the dataset in an inconsistent state.
4860 (void) rw_enter(&ztest_name_lock
,
4863 VERIFY(vdev_offline(spa
, guid0
, flags
) != EBUSY
);
4866 (void) rw_exit(&ztest_name_lock
);
4868 (void) vdev_online(spa
, guid0
, 0, NULL
);
4876 * We have at least single-fault tolerance, so inject data corruption.
4878 fd
= open(pathrand
, O_RDWR
);
4880 if (fd
== -1) /* we hit a gap in the device namespace */
4883 fsize
= lseek(fd
, 0, SEEK_END
);
4885 while (--iters
!= 0) {
4886 offset
= ztest_random(fsize
/ (leaves
<< bshift
)) *
4887 (leaves
<< bshift
) + (leaf
<< bshift
) +
4888 (ztest_random(1ULL << (bshift
- 1)) & -8ULL);
4890 if (offset
>= fsize
)
4893 mutex_enter(&ztest_vdev_lock
);
4894 if (mirror_save
!= zs
->zs_mirrors
) {
4895 mutex_exit(&ztest_vdev_lock
);
4900 if (pwrite(fd
, &bad
, sizeof (bad
), offset
) != sizeof (bad
))
4901 fatal(1, "can't inject bad word at 0x%llx in %s",
4904 mutex_exit(&ztest_vdev_lock
);
4906 if (ztest_opts
.zo_verbose
>= 7)
4907 (void) printf("injected bad word into %s,"
4908 " offset 0x%llx\n", pathrand
, (u_longlong_t
)offset
);
4913 umem_free(path0
, MAXPATHLEN
);
4914 umem_free(pathrand
, MAXPATHLEN
);
4918 * Verify that DDT repair works as expected.
4921 ztest_ddt_repair(ztest_ds_t
*zd
, uint64_t id
)
4923 ztest_shared_t
*zs
= ztest_shared
;
4924 spa_t
*spa
= ztest_spa
;
4925 objset_t
*os
= zd
->zd_os
;
4927 uint64_t object
, blocksize
, txg
, pattern
, psize
;
4928 enum zio_checksum checksum
= spa_dedup_checksum(spa
);
4933 int copies
= 2 * ZIO_DEDUPDITTO_MIN
;
4936 blocksize
= ztest_random_blocksize();
4937 blocksize
= MIN(blocksize
, 2048); /* because we write so many */
4939 od
= umem_alloc(sizeof(ztest_od_t
), UMEM_NOFAIL
);
4940 ztest_od_init(od
, id
, FTAG
, 0, DMU_OT_UINT64_OTHER
, blocksize
, 0);
4942 if (ztest_object_init(zd
, od
, sizeof (ztest_od_t
), B_FALSE
) != 0) {
4943 umem_free(od
, sizeof(ztest_od_t
));
4948 * Take the name lock as writer to prevent anyone else from changing
4949 * the pool and dataset properies we need to maintain during this test.
4951 (void) rw_enter(&ztest_name_lock
, RW_WRITER
);
4953 if (ztest_dsl_prop_set_uint64(zd
->zd_name
, ZFS_PROP_DEDUP
, checksum
,
4955 ztest_dsl_prop_set_uint64(zd
->zd_name
, ZFS_PROP_COPIES
, 1,
4957 (void) rw_exit(&ztest_name_lock
);
4958 umem_free(od
, sizeof(ztest_od_t
));
4962 object
= od
[0].od_object
;
4963 blocksize
= od
[0].od_blocksize
;
4964 pattern
= zs
->zs_guid
^ dmu_objset_fsid_guid(os
);
4966 ASSERT(object
!= 0);
4968 tx
= dmu_tx_create(os
);
4969 dmu_tx_hold_write(tx
, object
, 0, copies
* blocksize
);
4970 txg
= ztest_tx_assign(tx
, TXG_WAIT
, FTAG
);
4972 (void) rw_exit(&ztest_name_lock
);
4973 umem_free(od
, sizeof(ztest_od_t
));
4978 * Write all the copies of our block.
4980 for (i
= 0; i
< copies
; i
++) {
4981 uint64_t offset
= i
* blocksize
;
4982 VERIFY(dmu_buf_hold(os
, object
, offset
, FTAG
, &db
,
4983 DMU_READ_NO_PREFETCH
) == 0);
4984 ASSERT(db
->db_offset
== offset
);
4985 ASSERT(db
->db_size
== blocksize
);
4986 ASSERT(ztest_pattern_match(db
->db_data
, db
->db_size
, pattern
) ||
4987 ztest_pattern_match(db
->db_data
, db
->db_size
, 0ULL));
4988 dmu_buf_will_fill(db
, tx
);
4989 ztest_pattern_set(db
->db_data
, db
->db_size
, pattern
);
4990 dmu_buf_rele(db
, FTAG
);
4994 txg_wait_synced(spa_get_dsl(spa
), txg
);
4997 * Find out what block we got.
4999 VERIFY(dmu_buf_hold(os
, object
, 0, FTAG
, &db
,
5000 DMU_READ_NO_PREFETCH
) == 0);
5001 blk
= *((dmu_buf_impl_t
*)db
)->db_blkptr
;
5002 dmu_buf_rele(db
, FTAG
);
5005 * Damage the block. Dedup-ditto will save us when we read it later.
5007 psize
= BP_GET_PSIZE(&blk
);
5008 buf
= zio_buf_alloc(psize
);
5009 ztest_pattern_set(buf
, psize
, ~pattern
);
5011 (void) zio_wait(zio_rewrite(NULL
, spa
, 0, &blk
,
5012 buf
, psize
, NULL
, NULL
, ZIO_PRIORITY_SYNC_WRITE
,
5013 ZIO_FLAG_CANFAIL
| ZIO_FLAG_INDUCE_DAMAGE
, NULL
));
5015 zio_buf_free(buf
, psize
);
5017 (void) rw_exit(&ztest_name_lock
);
5018 umem_free(od
, sizeof(ztest_od_t
));
5026 ztest_scrub(ztest_ds_t
*zd
, uint64_t id
)
5028 spa_t
*spa
= ztest_spa
;
5030 (void) spa_scan(spa
, POOL_SCAN_SCRUB
);
5031 (void) poll(NULL
, 0, 100); /* wait a moment, then force a restart */
5032 (void) spa_scan(spa
, POOL_SCAN_SCRUB
);
5036 * Change the guid for the pool.
5040 ztest_reguid(ztest_ds_t
*zd
, uint64_t id
)
5042 spa_t
*spa
= ztest_spa
;
5043 uint64_t orig
, load
;
5046 orig
= spa_guid(spa
);
5047 load
= spa_load_guid(spa
);
5049 (void) rw_enter(&ztest_name_lock
, RW_WRITER
);
5050 error
= spa_change_guid(spa
);
5051 (void) rw_exit(&ztest_name_lock
);
5056 if (ztest_opts
.zo_verbose
>= 3) {
5057 (void) printf("Changed guid old %llu -> %llu\n",
5058 (u_longlong_t
)orig
, (u_longlong_t
)spa_guid(spa
));
5061 VERIFY3U(orig
, !=, spa_guid(spa
));
5062 VERIFY3U(load
, ==, spa_load_guid(spa
));
5066 * Rename the pool to a different name and then rename it back.
5070 ztest_spa_rename(ztest_ds_t
*zd
, uint64_t id
)
5072 char *oldname
, *newname
;
5075 (void) rw_enter(&ztest_name_lock
, RW_WRITER
);
5077 oldname
= ztest_opts
.zo_pool
;
5078 newname
= umem_alloc(strlen(oldname
) + 5, UMEM_NOFAIL
);
5079 (void) strcpy(newname
, oldname
);
5080 (void) strcat(newname
, "_tmp");
5085 VERIFY3U(0, ==, spa_rename(oldname
, newname
));
5088 * Try to open it under the old name, which shouldn't exist
5090 VERIFY3U(ENOENT
, ==, spa_open(oldname
, &spa
, FTAG
));
5093 * Open it under the new name and make sure it's still the same spa_t.
5095 VERIFY3U(0, ==, spa_open(newname
, &spa
, FTAG
));
5097 ASSERT(spa
== ztest_spa
);
5098 spa_close(spa
, FTAG
);
5101 * Rename it back to the original
5103 VERIFY3U(0, ==, spa_rename(newname
, oldname
));
5106 * Make sure it can still be opened
5108 VERIFY3U(0, ==, spa_open(oldname
, &spa
, FTAG
));
5110 ASSERT(spa
== ztest_spa
);
5111 spa_close(spa
, FTAG
);
5113 umem_free(newname
, strlen(newname
) + 1);
5115 (void) rw_exit(&ztest_name_lock
);
5119 * Verify pool integrity by running zdb.
5122 ztest_run_zdb(char *pool
)
5130 bin
= umem_alloc(MAXPATHLEN
+ MAXNAMELEN
+ 20, UMEM_NOFAIL
);
5131 zdb
= umem_alloc(MAXPATHLEN
+ MAXNAMELEN
+ 20, UMEM_NOFAIL
);
5132 zbuf
= umem_alloc(1024, UMEM_NOFAIL
);
5134 VERIFY(realpath(getexecname(), bin
) != NULL
);
5135 if (strncmp(bin
, "/usr/sbin/ztest", 15) == 0) {
5136 strcpy(bin
, "/usr/sbin/zdb"); /* Installed */
5137 } else if (strncmp(bin
, "/sbin/ztest", 11) == 0) {
5138 strcpy(bin
, "/sbin/zdb"); /* Installed */
5140 strstr(bin
, "/ztest/")[0] = '\0'; /* In-tree */
5141 strcat(bin
, "/zdb/zdb");
5145 "%s -bcc%s%s -U %s %s",
5147 ztest_opts
.zo_verbose
>= 3 ? "s" : "",
5148 ztest_opts
.zo_verbose
>= 4 ? "v" : "",
5152 if (ztest_opts
.zo_verbose
>= 5)
5153 (void) printf("Executing %s\n", strstr(zdb
, "zdb "));
5155 fp
= popen(zdb
, "r");
5157 while (fgets(zbuf
, 1024, fp
) != NULL
)
5158 if (ztest_opts
.zo_verbose
>= 3)
5159 (void) printf("%s", zbuf
);
5161 status
= pclose(fp
);
5166 ztest_dump_core
= 0;
5167 if (WIFEXITED(status
))
5168 fatal(0, "'%s' exit code %d", zdb
, WEXITSTATUS(status
));
5170 fatal(0, "'%s' died with signal %d", zdb
, WTERMSIG(status
));
5172 umem_free(bin
, MAXPATHLEN
+ MAXNAMELEN
+ 20);
5173 umem_free(zdb
, MAXPATHLEN
+ MAXNAMELEN
+ 20);
5174 umem_free(zbuf
, 1024);
5178 ztest_walk_pool_directory(char *header
)
5182 if (ztest_opts
.zo_verbose
>= 6)
5183 (void) printf("%s\n", header
);
5185 mutex_enter(&spa_namespace_lock
);
5186 while ((spa
= spa_next(spa
)) != NULL
)
5187 if (ztest_opts
.zo_verbose
>= 6)
5188 (void) printf("\t%s\n", spa_name(spa
));
5189 mutex_exit(&spa_namespace_lock
);
5193 ztest_spa_import_export(char *oldname
, char *newname
)
5195 nvlist_t
*config
, *newconfig
;
5199 if (ztest_opts
.zo_verbose
>= 4) {
5200 (void) printf("import/export: old = %s, new = %s\n",
5205 * Clean up from previous runs.
5207 (void) spa_destroy(newname
);
5210 * Get the pool's configuration and guid.
5212 VERIFY3U(0, ==, spa_open(oldname
, &spa
, FTAG
));
5215 * Kick off a scrub to tickle scrub/export races.
5217 if (ztest_random(2) == 0)
5218 (void) spa_scan(spa
, POOL_SCAN_SCRUB
);
5220 pool_guid
= spa_guid(spa
);
5221 spa_close(spa
, FTAG
);
5223 ztest_walk_pool_directory("pools before export");
5228 VERIFY3U(0, ==, spa_export(oldname
, &config
, B_FALSE
, B_FALSE
));
5230 ztest_walk_pool_directory("pools after export");
5235 newconfig
= spa_tryimport(config
);
5236 ASSERT(newconfig
!= NULL
);
5237 nvlist_free(newconfig
);
5240 * Import it under the new name.
5242 VERIFY3U(0, ==, spa_import(newname
, config
, NULL
, 0));
5244 ztest_walk_pool_directory("pools after import");
5247 * Try to import it again -- should fail with EEXIST.
5249 VERIFY3U(EEXIST
, ==, spa_import(newname
, config
, NULL
, 0));
5252 * Try to import it under a different name -- should fail with EEXIST.
5254 VERIFY3U(EEXIST
, ==, spa_import(oldname
, config
, NULL
, 0));
5257 * Verify that the pool is no longer visible under the old name.
5259 VERIFY3U(ENOENT
, ==, spa_open(oldname
, &spa
, FTAG
));
5262 * Verify that we can open and close the pool using the new name.
5264 VERIFY3U(0, ==, spa_open(newname
, &spa
, FTAG
));
5265 ASSERT(pool_guid
== spa_guid(spa
));
5266 spa_close(spa
, FTAG
);
5268 nvlist_free(config
);
5272 ztest_resume(spa_t
*spa
)
5274 if (spa_suspended(spa
) && ztest_opts
.zo_verbose
>= 6)
5275 (void) printf("resuming from suspended state\n");
5276 spa_vdev_state_enter(spa
, SCL_NONE
);
5277 vdev_clear(spa
, NULL
);
5278 (void) spa_vdev_state_exit(spa
, NULL
, 0);
5279 (void) zio_resume(spa
);
5283 ztest_resume_thread(void *arg
)
5287 while (!ztest_exiting
) {
5288 if (spa_suspended(spa
))
5290 (void) poll(NULL
, 0, 100);
5302 ztest_deadman_alarm(int sig
)
5304 fatal(0, "failed to complete within %d seconds of deadline", GRACE
);
5309 ztest_execute(int test
, ztest_info_t
*zi
, uint64_t id
)
5311 ztest_ds_t
*zd
= &ztest_ds
[id
% ztest_opts
.zo_datasets
];
5312 ztest_shared_callstate_t
*zc
= ZTEST_GET_SHARED_CALLSTATE(test
);
5313 hrtime_t functime
= gethrtime();
5316 for (i
= 0; i
< zi
->zi_iters
; i
++)
5317 zi
->zi_func(zd
, id
);
5319 functime
= gethrtime() - functime
;
5321 atomic_add_64(&zc
->zc_count
, 1);
5322 atomic_add_64(&zc
->zc_time
, functime
);
5324 if (ztest_opts
.zo_verbose
>= 4) {
5326 (void) dladdr((void *)zi
->zi_func
, &dli
);
5327 (void) printf("%6.2f sec in %s\n",
5328 (double)functime
/ NANOSEC
, dli
.dli_sname
);
5333 ztest_thread(void *arg
)
5336 uint64_t id
= (uintptr_t)arg
;
5337 ztest_shared_t
*zs
= ztest_shared
;
5341 ztest_shared_callstate_t
*zc
;
5343 while ((now
= gethrtime()) < zs
->zs_thread_stop
) {
5345 * See if it's time to force a crash.
5347 if (now
> zs
->zs_thread_kill
)
5351 * If we're getting ENOSPC with some regularity, stop.
5353 if (zs
->zs_enospc_count
> 10)
5357 * Pick a random function to execute.
5359 rand
= ztest_random(ZTEST_FUNCS
);
5360 zi
= &ztest_info
[rand
];
5361 zc
= ZTEST_GET_SHARED_CALLSTATE(rand
);
5362 call_next
= zc
->zc_next
;
5364 if (now
>= call_next
&&
5365 atomic_cas_64(&zc
->zc_next
, call_next
, call_next
+
5366 ztest_random(2 * zi
->zi_interval
[0] + 1)) == call_next
) {
5367 ztest_execute(rand
, zi
, id
);
5377 ztest_dataset_name(char *dsname
, char *pool
, int d
)
5379 (void) snprintf(dsname
, MAXNAMELEN
, "%s/ds_%d", pool
, d
);
5383 ztest_dataset_destroy(int d
)
5385 char name
[MAXNAMELEN
];
5388 ztest_dataset_name(name
, ztest_opts
.zo_pool
, d
);
5390 if (ztest_opts
.zo_verbose
>= 3)
5391 (void) printf("Destroying %s to free up space\n", name
);
5394 * Cleanup any non-standard clones and snapshots. In general,
5395 * ztest thread t operates on dataset (t % zopt_datasets),
5396 * so there may be more than one thing to clean up.
5398 for (t
= d
; t
< ztest_opts
.zo_threads
;
5399 t
+= ztest_opts
.zo_datasets
)
5400 ztest_dsl_dataset_cleanup(name
, t
);
5402 (void) dmu_objset_find(name
, ztest_objset_destroy_cb
, NULL
,
5403 DS_FIND_SNAPSHOTS
| DS_FIND_CHILDREN
);
5407 ztest_dataset_dirobj_verify(ztest_ds_t
*zd
)
5409 uint64_t usedobjs
, dirobjs
, scratch
;
5412 * ZTEST_DIROBJ is the object directory for the entire dataset.
5413 * Therefore, the number of objects in use should equal the
5414 * number of ZTEST_DIROBJ entries, +1 for ZTEST_DIROBJ itself.
5415 * If not, we have an object leak.
5417 * Note that we can only check this in ztest_dataset_open(),
5418 * when the open-context and syncing-context values agree.
5419 * That's because zap_count() returns the open-context value,
5420 * while dmu_objset_space() returns the rootbp fill count.
5422 VERIFY3U(0, ==, zap_count(zd
->zd_os
, ZTEST_DIROBJ
, &dirobjs
));
5423 dmu_objset_space(zd
->zd_os
, &scratch
, &scratch
, &usedobjs
, &scratch
);
5424 ASSERT3U(dirobjs
+ 1, ==, usedobjs
);
5428 ztest_dataset_open(int d
)
5430 ztest_ds_t
*zd
= &ztest_ds
[d
];
5431 uint64_t committed_seq
= ZTEST_GET_SHARED_DS(d
)->zd_seq
;
5434 char name
[MAXNAMELEN
];
5437 ztest_dataset_name(name
, ztest_opts
.zo_pool
, d
);
5439 (void) rw_enter(&ztest_name_lock
, RW_READER
);
5441 error
= ztest_dataset_create(name
);
5442 if (error
== ENOSPC
) {
5443 (void) rw_exit(&ztest_name_lock
);
5444 ztest_record_enospc(FTAG
);
5447 ASSERT(error
== 0 || error
== EEXIST
);
5449 VERIFY3U(dmu_objset_hold(name
, zd
, &os
), ==, 0);
5450 (void) rw_exit(&ztest_name_lock
);
5452 ztest_zd_init(zd
, ZTEST_GET_SHARED_DS(d
), os
);
5454 zilog
= zd
->zd_zilog
;
5456 if (zilog
->zl_header
->zh_claim_lr_seq
!= 0 &&
5457 zilog
->zl_header
->zh_claim_lr_seq
< committed_seq
)
5458 fatal(0, "missing log records: claimed %llu < committed %llu",
5459 zilog
->zl_header
->zh_claim_lr_seq
, committed_seq
);
5461 ztest_dataset_dirobj_verify(zd
);
5463 zil_replay(os
, zd
, ztest_replay_vector
);
5465 ztest_dataset_dirobj_verify(zd
);
5467 if (ztest_opts
.zo_verbose
>= 6)
5468 (void) printf("%s replay %llu blocks, %llu records, seq %llu\n",
5470 (u_longlong_t
)zilog
->zl_parse_blk_count
,
5471 (u_longlong_t
)zilog
->zl_parse_lr_count
,
5472 (u_longlong_t
)zilog
->zl_replaying_seq
);
5474 zilog
= zil_open(os
, ztest_get_data
);
5476 if (zilog
->zl_replaying_seq
!= 0 &&
5477 zilog
->zl_replaying_seq
< committed_seq
)
5478 fatal(0, "missing log records: replayed %llu < committed %llu",
5479 zilog
->zl_replaying_seq
, committed_seq
);
5485 ztest_dataset_close(int d
)
5487 ztest_ds_t
*zd
= &ztest_ds
[d
];
5489 zil_close(zd
->zd_zilog
);
5490 dmu_objset_rele(zd
->zd_os
, zd
);
5496 * Kick off threads to run tests on all datasets in parallel.
5499 ztest_run(ztest_shared_t
*zs
)
5504 kthread_t
*resume_thread
;
5509 ztest_exiting
= B_FALSE
;
5512 * Initialize parent/child shared state.
5514 mutex_init(&ztest_vdev_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
5515 rw_init(&ztest_name_lock
, NULL
, RW_DEFAULT
, NULL
);
5517 zs
->zs_thread_start
= gethrtime();
5518 zs
->zs_thread_stop
=
5519 zs
->zs_thread_start
+ ztest_opts
.zo_passtime
* NANOSEC
;
5520 zs
->zs_thread_stop
= MIN(zs
->zs_thread_stop
, zs
->zs_proc_stop
);
5521 zs
->zs_thread_kill
= zs
->zs_thread_stop
;
5522 if (ztest_random(100) < ztest_opts
.zo_killrate
) {
5523 zs
->zs_thread_kill
-=
5524 ztest_random(ztest_opts
.zo_passtime
* NANOSEC
);
5527 mutex_init(&zcl
.zcl_callbacks_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
5529 list_create(&zcl
.zcl_callbacks
, sizeof (ztest_cb_data_t
),
5530 offsetof(ztest_cb_data_t
, zcd_node
));
5535 kernel_init(FREAD
| FWRITE
);
5536 VERIFY(spa_open(ztest_opts
.zo_pool
, &spa
, FTAG
) == 0);
5537 spa
->spa_debug
= B_TRUE
;
5540 VERIFY3U(0, ==, dmu_objset_hold(ztest_opts
.zo_pool
, FTAG
, &os
));
5541 zs
->zs_guid
= dmu_objset_fsid_guid(os
);
5542 dmu_objset_rele(os
, FTAG
);
5544 spa
->spa_dedup_ditto
= 2 * ZIO_DEDUPDITTO_MIN
;
5547 * We don't expect the pool to suspend unless maxfaults == 0,
5548 * in which case ztest_fault_inject() temporarily takes away
5549 * the only valid replica.
5551 if (MAXFAULTS() == 0)
5552 spa
->spa_failmode
= ZIO_FAILURE_MODE_WAIT
;
5554 spa
->spa_failmode
= ZIO_FAILURE_MODE_PANIC
;
5557 * Create a thread to periodically resume suspended I/O.
5559 VERIFY3P((resume_thread
= zk_thread_create(NULL
, 0,
5560 (thread_func_t
)ztest_resume_thread
, spa
, TS_RUN
, NULL
, 0, 0,
5561 PTHREAD_CREATE_JOINABLE
)), !=, NULL
);
5565 * Set a deadman alarm to abort() if we hang.
5567 signal(SIGALRM
, ztest_deadman_alarm
);
5568 alarm((zs
->zs_thread_stop
- zs
->zs_thread_start
) / NANOSEC
+ GRACE
);
5572 * Verify that we can safely inquire about about any object,
5573 * whether it's allocated or not. To make it interesting,
5574 * we probe a 5-wide window around each power of two.
5575 * This hits all edge cases, including zero and the max.
5577 for (t
= 0; t
< 64; t
++) {
5578 for (d
= -5; d
<= 5; d
++) {
5579 error
= dmu_object_info(spa
->spa_meta_objset
,
5580 (1ULL << t
) + d
, NULL
);
5581 ASSERT(error
== 0 || error
== ENOENT
||
5587 * If we got any ENOSPC errors on the previous run, destroy something.
5589 if (zs
->zs_enospc_count
!= 0) {
5590 int d
= ztest_random(ztest_opts
.zo_datasets
);
5591 ztest_dataset_destroy(d
);
5593 zs
->zs_enospc_count
= 0;
5595 tid
= umem_zalloc(ztest_opts
.zo_threads
* sizeof (kt_did_t
),
5598 if (ztest_opts
.zo_verbose
>= 4)
5599 (void) printf("starting main threads...\n");
5602 * Kick off all the tests that run in parallel.
5604 for (t
= 0; t
< ztest_opts
.zo_threads
; t
++) {
5607 if (t
< ztest_opts
.zo_datasets
&&
5608 ztest_dataset_open(t
) != 0)
5611 VERIFY3P(thread
= zk_thread_create(NULL
, 0,
5612 (thread_func_t
)ztest_thread
,
5613 (void *)(uintptr_t)t
, TS_RUN
, NULL
, 0, 0,
5614 PTHREAD_CREATE_JOINABLE
), !=, NULL
);
5615 tid
[t
] = thread
->t_tid
;
5619 * Wait for all of the tests to complete. We go in reverse order
5620 * so we don't close datasets while threads are still using them.
5622 for (t
= ztest_opts
.zo_threads
- 1; t
>= 0; t
--) {
5623 thread_join(tid
[t
]);
5624 if (t
< ztest_opts
.zo_datasets
)
5625 ztest_dataset_close(t
);
5628 txg_wait_synced(spa_get_dsl(spa
), 0);
5630 zs
->zs_alloc
= metaslab_class_get_alloc(spa_normal_class(spa
));
5631 zs
->zs_space
= metaslab_class_get_space(spa_normal_class(spa
));
5633 umem_free(tid
, ztest_opts
.zo_threads
* sizeof (kt_did_t
));
5635 /* Kill the resume thread */
5636 ztest_exiting
= B_TRUE
;
5637 thread_join(resume_thread
->t_tid
);
5641 * Right before closing the pool, kick off a bunch of async I/O;
5642 * spa_close() should wait for it to complete.
5644 for (object
= 1; object
< 50; object
++)
5645 dmu_prefetch(spa
->spa_meta_objset
, object
, 0, 1ULL << 20);
5647 /* Verify that at least one commit cb was called in a timely fashion */
5648 if (zc_cb_counter
>= ZTEST_COMMIT_CB_MIN_REG
)
5649 VERIFY3U(zc_min_txg_delay
, ==, 0);
5651 spa_close(spa
, FTAG
);
5654 * Verify that we can loop over all pools.
5656 mutex_enter(&spa_namespace_lock
);
5657 for (spa
= spa_next(NULL
); spa
!= NULL
; spa
= spa_next(spa
))
5658 if (ztest_opts
.zo_verbose
> 3)
5659 (void) printf("spa_next: found %s\n", spa_name(spa
));
5660 mutex_exit(&spa_namespace_lock
);
5663 * Verify that we can export the pool and reimport it under a
5666 if (ztest_random(2) == 0) {
5667 char name
[MAXNAMELEN
];
5668 (void) snprintf(name
, MAXNAMELEN
, "%s_import",
5669 ztest_opts
.zo_pool
);
5670 ztest_spa_import_export(ztest_opts
.zo_pool
, name
);
5671 ztest_spa_import_export(name
, ztest_opts
.zo_pool
);
5676 list_destroy(&zcl
.zcl_callbacks
);
5677 mutex_destroy(&zcl
.zcl_callbacks_lock
);
5678 rw_destroy(&ztest_name_lock
);
5679 mutex_destroy(&ztest_vdev_lock
);
5685 ztest_ds_t
*zd
= &ztest_ds
[0];
5689 if (ztest_opts
.zo_verbose
>= 3)
5690 (void) printf("testing spa_freeze()...\n");
5692 kernel_init(FREAD
| FWRITE
);
5693 VERIFY3U(0, ==, spa_open(ztest_opts
.zo_pool
, &spa
, FTAG
));
5694 VERIFY3U(0, ==, ztest_dataset_open(0));
5697 * Force the first log block to be transactionally allocated.
5698 * We have to do this before we freeze the pool -- otherwise
5699 * the log chain won't be anchored.
5701 while (BP_IS_HOLE(&zd
->zd_zilog
->zl_header
->zh_log
)) {
5702 ztest_dmu_object_alloc_free(zd
, 0);
5703 zil_commit(zd
->zd_zilog
, 0);
5706 txg_wait_synced(spa_get_dsl(spa
), 0);
5709 * Freeze the pool. This stops spa_sync() from doing anything,
5710 * so that the only way to record changes from now on is the ZIL.
5715 * Run tests that generate log records but don't alter the pool config
5716 * or depend on DSL sync tasks (snapshots, objset create/destroy, etc).
5717 * We do a txg_wait_synced() after each iteration to force the txg
5718 * to increase well beyond the last synced value in the uberblock.
5719 * The ZIL should be OK with that.
5721 while (ztest_random(10) != 0 &&
5722 numloops
++ < ztest_opts
.zo_maxloops
) {
5723 ztest_dmu_write_parallel(zd
, 0);
5724 ztest_dmu_object_alloc_free(zd
, 0);
5725 txg_wait_synced(spa_get_dsl(spa
), 0);
5729 * Commit all of the changes we just generated.
5731 zil_commit(zd
->zd_zilog
, 0);
5732 txg_wait_synced(spa_get_dsl(spa
), 0);
5735 * Close our dataset and close the pool.
5737 ztest_dataset_close(0);
5738 spa_close(spa
, FTAG
);
5742 * Open and close the pool and dataset to induce log replay.
5744 kernel_init(FREAD
| FWRITE
);
5745 VERIFY3U(0, ==, spa_open(ztest_opts
.zo_pool
, &spa
, FTAG
));
5746 VERIFY3U(0, ==, ztest_dataset_open(0));
5747 ztest_dataset_close(0);
5749 spa
->spa_debug
= B_TRUE
;
5751 txg_wait_synced(spa_get_dsl(spa
), 0);
5752 ztest_reguid(NULL
, 0);
5754 spa_close(spa
, FTAG
);
5759 print_time(hrtime_t t
, char *timebuf
)
5761 hrtime_t s
= t
/ NANOSEC
;
5762 hrtime_t m
= s
/ 60;
5763 hrtime_t h
= m
/ 60;
5764 hrtime_t d
= h
/ 24;
5773 (void) sprintf(timebuf
,
5774 "%llud%02lluh%02llum%02llus", d
, h
, m
, s
);
5776 (void) sprintf(timebuf
, "%lluh%02llum%02llus", h
, m
, s
);
5778 (void) sprintf(timebuf
, "%llum%02llus", m
, s
);
5780 (void) sprintf(timebuf
, "%llus", s
);
5784 make_random_props(void)
5788 VERIFY(nvlist_alloc(&props
, NV_UNIQUE_NAME
, 0) == 0);
5789 if (ztest_random(2) == 0)
5791 VERIFY(nvlist_add_uint64(props
, "autoreplace", 1) == 0);
5797 * Create a storage pool with the given name and initial vdev size.
5798 * Then test spa_freeze() functionality.
5801 ztest_init(ztest_shared_t
*zs
)
5804 nvlist_t
*nvroot
, *props
;
5807 mutex_init(&ztest_vdev_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
5808 rw_init(&ztest_name_lock
, NULL
, RW_DEFAULT
, NULL
);
5810 kernel_init(FREAD
| FWRITE
);
5813 * Create the storage pool.
5815 (void) spa_destroy(ztest_opts
.zo_pool
);
5816 ztest_shared
->zs_vdev_next_leaf
= 0;
5818 zs
->zs_mirrors
= ztest_opts
.zo_mirrors
;
5819 nvroot
= make_vdev_root(NULL
, NULL
, ztest_opts
.zo_vdev_size
, 0,
5820 0, ztest_opts
.zo_raidz
, zs
->zs_mirrors
, 1);
5821 props
= make_random_props();
5822 for (i
= 0; i
< SPA_FEATURES
; i
++) {
5824 VERIFY3S(-1, !=, asprintf(&buf
, "feature@%s",
5825 spa_feature_table
[i
].fi_uname
));
5826 VERIFY3U(0, ==, nvlist_add_uint64(props
, buf
, 0));
5829 VERIFY3U(0, ==, spa_create(ztest_opts
.zo_pool
, nvroot
, props
,
5831 nvlist_free(nvroot
);
5833 VERIFY3U(0, ==, spa_open(ztest_opts
.zo_pool
, &spa
, FTAG
));
5834 zs
->zs_metaslab_sz
=
5835 1ULL << spa
->spa_root_vdev
->vdev_child
[0]->vdev_ms_shift
;
5836 spa_close(spa
, FTAG
);
5840 ztest_run_zdb(ztest_opts
.zo_pool
);
5844 ztest_run_zdb(ztest_opts
.zo_pool
);
5846 rw_destroy(&ztest_name_lock
);
5847 mutex_destroy(&ztest_vdev_lock
);
5853 static char ztest_name_data
[] = "/tmp/ztest.data.XXXXXX";
5855 ztest_fd_data
= mkstemp(ztest_name_data
);
5856 ASSERT3S(ztest_fd_data
, >=, 0);
5857 (void) unlink(ztest_name_data
);
5861 shared_data_size(ztest_shared_hdr_t
*hdr
)
5865 size
= hdr
->zh_hdr_size
;
5866 size
+= hdr
->zh_opts_size
;
5867 size
+= hdr
->zh_size
;
5868 size
+= hdr
->zh_stats_size
* hdr
->zh_stats_count
;
5869 size
+= hdr
->zh_ds_size
* hdr
->zh_ds_count
;
5878 ztest_shared_hdr_t
*hdr
;
5880 hdr
= (void *)mmap(0, P2ROUNDUP(sizeof (*hdr
), getpagesize()),
5881 PROT_READ
| PROT_WRITE
, MAP_SHARED
, ztest_fd_data
, 0);
5882 ASSERT(hdr
!= MAP_FAILED
);
5884 VERIFY3U(0, ==, ftruncate(ztest_fd_data
, sizeof (ztest_shared_hdr_t
)));
5886 hdr
->zh_hdr_size
= sizeof (ztest_shared_hdr_t
);
5887 hdr
->zh_opts_size
= sizeof (ztest_shared_opts_t
);
5888 hdr
->zh_size
= sizeof (ztest_shared_t
);
5889 hdr
->zh_stats_size
= sizeof (ztest_shared_callstate_t
);
5890 hdr
->zh_stats_count
= ZTEST_FUNCS
;
5891 hdr
->zh_ds_size
= sizeof (ztest_shared_ds_t
);
5892 hdr
->zh_ds_count
= ztest_opts
.zo_datasets
;
5894 size
= shared_data_size(hdr
);
5895 VERIFY3U(0, ==, ftruncate(ztest_fd_data
, size
));
5897 (void) munmap((caddr_t
)hdr
, P2ROUNDUP(sizeof (*hdr
), getpagesize()));
5904 ztest_shared_hdr_t
*hdr
;
5907 hdr
= (void *)mmap(0, P2ROUNDUP(sizeof (*hdr
), getpagesize()),
5908 PROT_READ
, MAP_SHARED
, ztest_fd_data
, 0);
5909 ASSERT(hdr
!= MAP_FAILED
);
5911 size
= shared_data_size(hdr
);
5913 (void) munmap((caddr_t
)hdr
, P2ROUNDUP(sizeof (*hdr
), getpagesize()));
5914 hdr
= ztest_shared_hdr
= (void *)mmap(0, P2ROUNDUP(size
, getpagesize()),
5915 PROT_READ
| PROT_WRITE
, MAP_SHARED
, ztest_fd_data
, 0);
5916 ASSERT(hdr
!= MAP_FAILED
);
5917 buf
= (uint8_t *)hdr
;
5919 offset
= hdr
->zh_hdr_size
;
5920 ztest_shared_opts
= (void *)&buf
[offset
];
5921 offset
+= hdr
->zh_opts_size
;
5922 ztest_shared
= (void *)&buf
[offset
];
5923 offset
+= hdr
->zh_size
;
5924 ztest_shared_callstate
= (void *)&buf
[offset
];
5925 offset
+= hdr
->zh_stats_size
* hdr
->zh_stats_count
;
5926 ztest_shared_ds
= (void *)&buf
[offset
];
5930 exec_child(char *cmd
, char *libpath
, boolean_t ignorekill
, int *statusp
)
5934 char *cmdbuf
= NULL
;
5939 cmdbuf
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
5940 (void) strlcpy(cmdbuf
, getexecname(), MAXPATHLEN
);
5945 fatal(1, "fork failed");
5947 if (pid
== 0) { /* child */
5948 char *emptyargv
[2] = { cmd
, NULL
};
5949 char fd_data_str
[12];
5951 struct rlimit rl
= { 1024, 1024 };
5952 (void) setrlimit(RLIMIT_NOFILE
, &rl
);
5954 (void) close(ztest_fd_rand
);
5955 VERIFY(11 >= snprintf(fd_data_str
, 12, "%d", ztest_fd_data
));
5956 VERIFY(0 == setenv("ZTEST_FD_DATA", fd_data_str
, 1));
5958 (void) enable_extended_FILE_stdio(-1, -1);
5959 if (libpath
!= NULL
)
5960 VERIFY(0 == setenv("LD_LIBRARY_PATH", libpath
, 1));
5961 (void) execv(cmd
, emptyargv
);
5962 ztest_dump_core
= B_FALSE
;
5963 fatal(B_TRUE
, "exec failed: %s", cmd
);
5966 if (cmdbuf
!= NULL
) {
5967 umem_free(cmdbuf
, MAXPATHLEN
);
5971 while (waitpid(pid
, &status
, 0) != pid
)
5973 if (statusp
!= NULL
)
5976 if (WIFEXITED(status
)) {
5977 if (WEXITSTATUS(status
) != 0) {
5978 (void) fprintf(stderr
, "child exited with code %d\n",
5979 WEXITSTATUS(status
));
5983 } else if (WIFSIGNALED(status
)) {
5984 if (!ignorekill
|| WTERMSIG(status
) != SIGKILL
) {
5985 (void) fprintf(stderr
, "child died with signal %d\n",
5991 (void) fprintf(stderr
, "something strange happened to child\n");
5998 ztest_run_init(void)
6002 ztest_shared_t
*zs
= ztest_shared
;
6004 ASSERT(ztest_opts
.zo_init
!= 0);
6007 * Blow away any existing copy of zpool.cache
6009 (void) remove(spa_config_path
);
6012 * Create and initialize our storage pool.
6014 for (i
= 1; i
<= ztest_opts
.zo_init
; i
++) {
6015 bzero(zs
, sizeof (ztest_shared_t
));
6016 if (ztest_opts
.zo_verbose
>= 3 &&
6017 ztest_opts
.zo_init
!= 1) {
6018 (void) printf("ztest_init(), pass %d\n", i
);
6025 main(int argc
, char **argv
)
6033 ztest_shared_callstate_t
*zc
;
6040 char *fd_data_str
= getenv("ZTEST_FD_DATA");
6042 (void) setvbuf(stdout
, NULL
, _IOLBF
, 0);
6044 ztest_fd_rand
= open("/dev/urandom", O_RDONLY
);
6045 ASSERT3S(ztest_fd_rand
, >=, 0);
6048 dprintf_setup(&argc
, argv
);
6049 process_options(argc
, argv
);
6054 bcopy(&ztest_opts
, ztest_shared_opts
,
6055 sizeof (*ztest_shared_opts
));
6057 ztest_fd_data
= atoi(fd_data_str
);
6059 bcopy(ztest_shared_opts
, &ztest_opts
, sizeof (ztest_opts
));
6061 ASSERT3U(ztest_opts
.zo_datasets
, ==, ztest_shared_hdr
->zh_ds_count
);
6063 /* Override location of zpool.cache */
6064 VERIFY(asprintf((char **)&spa_config_path
, "%s/zpool.cache",
6065 ztest_opts
.zo_dir
) != -1);
6067 ztest_ds
= umem_alloc(ztest_opts
.zo_datasets
* sizeof (ztest_ds_t
),
6072 metaslab_gang_bang
= ztest_opts
.zo_metaslab_gang_bang
;
6073 metaslab_df_alloc_threshold
=
6074 zs
->zs_metaslab_df_alloc_threshold
;
6083 hasalt
= (strlen(ztest_opts
.zo_alt_ztest
) != 0);
6085 if (ztest_opts
.zo_verbose
>= 1) {
6086 (void) printf("%llu vdevs, %d datasets, %d threads,"
6087 " %llu seconds...\n",
6088 (u_longlong_t
)ztest_opts
.zo_vdevs
,
6089 ztest_opts
.zo_datasets
,
6090 ztest_opts
.zo_threads
,
6091 (u_longlong_t
)ztest_opts
.zo_time
);
6094 cmd
= umem_alloc(MAXNAMELEN
, UMEM_NOFAIL
);
6095 (void) strlcpy(cmd
, getexecname(), MAXNAMELEN
);
6097 zs
->zs_do_init
= B_TRUE
;
6098 if (strlen(ztest_opts
.zo_alt_ztest
) != 0) {
6099 if (ztest_opts
.zo_verbose
>= 1) {
6100 (void) printf("Executing older ztest for "
6101 "initialization: %s\n", ztest_opts
.zo_alt_ztest
);
6103 VERIFY(!exec_child(ztest_opts
.zo_alt_ztest
,
6104 ztest_opts
.zo_alt_libpath
, B_FALSE
, NULL
));
6106 VERIFY(!exec_child(NULL
, NULL
, B_FALSE
, NULL
));
6108 zs
->zs_do_init
= B_FALSE
;
6110 zs
->zs_proc_start
= gethrtime();
6111 zs
->zs_proc_stop
= zs
->zs_proc_start
+ ztest_opts
.zo_time
* NANOSEC
;
6113 for (f
= 0; f
< ZTEST_FUNCS
; f
++) {
6114 zi
= &ztest_info
[f
];
6115 zc
= ZTEST_GET_SHARED_CALLSTATE(f
);
6116 if (zs
->zs_proc_start
+ zi
->zi_interval
[0] > zs
->zs_proc_stop
)
6117 zc
->zc_next
= UINT64_MAX
;
6119 zc
->zc_next
= zs
->zs_proc_start
+
6120 ztest_random(2 * zi
->zi_interval
[0] + 1);
6124 * Run the tests in a loop. These tests include fault injection
6125 * to verify that self-healing data works, and forced crashes
6126 * to verify that we never lose on-disk consistency.
6128 while (gethrtime() < zs
->zs_proc_stop
) {
6133 * Initialize the workload counters for each function.
6135 for (f
= 0; f
< ZTEST_FUNCS
; f
++) {
6136 zc
= ZTEST_GET_SHARED_CALLSTATE(f
);
6141 /* Set the allocation switch size */
6142 zs
->zs_metaslab_df_alloc_threshold
=
6143 ztest_random(zs
->zs_metaslab_sz
/ 4) + 1;
6145 if (!hasalt
|| ztest_random(2) == 0) {
6146 if (hasalt
&& ztest_opts
.zo_verbose
>= 1) {
6147 (void) printf("Executing newer ztest: %s\n",
6151 killed
= exec_child(cmd
, NULL
, B_TRUE
, &status
);
6153 if (hasalt
&& ztest_opts
.zo_verbose
>= 1) {
6154 (void) printf("Executing older ztest: %s\n",
6155 ztest_opts
.zo_alt_ztest
);
6158 killed
= exec_child(ztest_opts
.zo_alt_ztest
,
6159 ztest_opts
.zo_alt_libpath
, B_TRUE
, &status
);
6166 if (ztest_opts
.zo_verbose
>= 1) {
6167 hrtime_t now
= gethrtime();
6169 now
= MIN(now
, zs
->zs_proc_stop
);
6170 print_time(zs
->zs_proc_stop
- now
, timebuf
);
6171 nicenum(zs
->zs_space
, numbuf
);
6173 (void) printf("Pass %3d, %8s, %3llu ENOSPC, "
6174 "%4.1f%% of %5s used, %3.0f%% done, %8s to go\n",
6176 WIFEXITED(status
) ? "Complete" : "SIGKILL",
6177 (u_longlong_t
)zs
->zs_enospc_count
,
6178 100.0 * zs
->zs_alloc
/ zs
->zs_space
,
6180 100.0 * (now
- zs
->zs_proc_start
) /
6181 (ztest_opts
.zo_time
* NANOSEC
), timebuf
);
6184 if (ztest_opts
.zo_verbose
>= 2) {
6185 (void) printf("\nWorkload summary:\n\n");
6186 (void) printf("%7s %9s %s\n",
6187 "Calls", "Time", "Function");
6188 (void) printf("%7s %9s %s\n",
6189 "-----", "----", "--------");
6190 for (f
= 0; f
< ZTEST_FUNCS
; f
++) {
6193 zi
= &ztest_info
[f
];
6194 zc
= ZTEST_GET_SHARED_CALLSTATE(f
);
6195 print_time(zc
->zc_time
, timebuf
);
6196 (void) dladdr((void *)zi
->zi_func
, &dli
);
6197 (void) printf("%7llu %9s %s\n",
6198 (u_longlong_t
)zc
->zc_count
, timebuf
,
6201 (void) printf("\n");
6205 * It's possible that we killed a child during a rename test,
6206 * in which case we'll have a 'ztest_tmp' pool lying around
6207 * instead of 'ztest'. Do a blind rename in case this happened.
6210 if (spa_open(ztest_opts
.zo_pool
, &spa
, FTAG
) == 0) {
6211 spa_close(spa
, FTAG
);
6213 char tmpname
[MAXNAMELEN
];
6215 kernel_init(FREAD
| FWRITE
);
6216 (void) snprintf(tmpname
, sizeof (tmpname
), "%s_tmp",
6217 ztest_opts
.zo_pool
);
6218 (void) spa_rename(tmpname
, ztest_opts
.zo_pool
);
6222 ztest_run_zdb(ztest_opts
.zo_pool
);
6225 if (ztest_opts
.zo_verbose
>= 1) {
6227 (void) printf("%d runs of older ztest: %s\n", older
,
6228 ztest_opts
.zo_alt_ztest
);
6229 (void) printf("%d runs of newer ztest: %s\n", newer
,
6232 (void) printf("%d killed, %d completed, %.0f%% kill rate\n",
6233 kills
, iters
- kills
, (100.0 * kills
) / MAX(1, iters
));
6236 umem_free(cmd
, MAXNAMELEN
);