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
;
334 ztest_func_t ztest_spa_upgrade
;
336 uint64_t zopt_always
= 0ULL * NANOSEC
; /* all the time */
337 uint64_t zopt_incessant
= 1ULL * NANOSEC
/ 10; /* every 1/10 second */
338 uint64_t zopt_often
= 1ULL * NANOSEC
; /* every second */
339 uint64_t zopt_sometimes
= 10ULL * NANOSEC
; /* every 10 seconds */
340 uint64_t zopt_rarely
= 60ULL * NANOSEC
; /* every 60 seconds */
342 ztest_info_t ztest_info
[] = {
343 { ztest_dmu_read_write
, 1, &zopt_always
},
344 { ztest_dmu_write_parallel
, 10, &zopt_always
},
345 { ztest_dmu_object_alloc_free
, 1, &zopt_always
},
346 { ztest_dmu_commit_callbacks
, 1, &zopt_always
},
347 { ztest_zap
, 30, &zopt_always
},
348 { ztest_zap_parallel
, 100, &zopt_always
},
349 { ztest_split_pool
, 1, &zopt_always
},
350 { ztest_zil_commit
, 1, &zopt_incessant
},
351 { ztest_zil_remount
, 1, &zopt_sometimes
},
352 { ztest_dmu_read_write_zcopy
, 1, &zopt_often
},
353 { ztest_dmu_objset_create_destroy
, 1, &zopt_often
},
354 { ztest_dsl_prop_get_set
, 1, &zopt_often
},
355 { ztest_spa_prop_get_set
, 1, &zopt_sometimes
},
357 { ztest_dmu_prealloc
, 1, &zopt_sometimes
},
359 { ztest_fzap
, 1, &zopt_sometimes
},
360 { ztest_dmu_snapshot_create_destroy
, 1, &zopt_sometimes
},
361 { ztest_spa_create_destroy
, 1, &zopt_sometimes
},
362 { ztest_fault_inject
, 1, &zopt_sometimes
},
363 { ztest_ddt_repair
, 1, &zopt_sometimes
},
364 { ztest_dmu_snapshot_hold
, 1, &zopt_sometimes
},
365 { ztest_reguid
, 1, &zopt_sometimes
},
366 { ztest_spa_rename
, 1, &zopt_rarely
},
367 { ztest_scrub
, 1, &zopt_rarely
},
368 { ztest_spa_upgrade
, 1, &zopt_rarely
},
369 { ztest_dsl_dataset_promote_busy
, 1, &zopt_rarely
},
370 { ztest_vdev_attach_detach
, 1, &zopt_rarely
},
371 { ztest_vdev_LUN_growth
, 1, &zopt_rarely
},
372 { ztest_vdev_add_remove
, 1,
373 &ztest_opts
.zo_vdevtime
},
374 { ztest_vdev_aux_add_remove
, 1,
375 &ztest_opts
.zo_vdevtime
},
378 #define ZTEST_FUNCS (sizeof (ztest_info) / sizeof (ztest_info_t))
381 * The following struct is used to hold a list of uncalled commit callbacks.
382 * The callbacks are ordered by txg number.
384 typedef struct ztest_cb_list
{
385 kmutex_t zcl_callbacks_lock
;
386 list_t zcl_callbacks
;
390 * Stuff we need to share writably between parent and child.
392 typedef struct ztest_shared
{
393 boolean_t zs_do_init
;
394 hrtime_t zs_proc_start
;
395 hrtime_t zs_proc_stop
;
396 hrtime_t zs_thread_start
;
397 hrtime_t zs_thread_stop
;
398 hrtime_t zs_thread_kill
;
399 uint64_t zs_enospc_count
;
400 uint64_t zs_vdev_next_leaf
;
401 uint64_t zs_vdev_aux
;
406 uint64_t zs_metaslab_sz
;
407 uint64_t zs_metaslab_df_alloc_threshold
;
411 #define ID_PARALLEL -1ULL
413 static char ztest_dev_template
[] = "%s/%s.%llua";
414 static char ztest_aux_template
[] = "%s/%s.%s.%llu";
415 ztest_shared_t
*ztest_shared
;
417 static spa_t
*ztest_spa
= NULL
;
418 static ztest_ds_t
*ztest_ds
;
420 static kmutex_t ztest_vdev_lock
;
423 * The ztest_name_lock protects the pool and dataset namespace used by
424 * the individual tests. To modify the namespace, consumers must grab
425 * this lock as writer. Grabbing the lock as reader will ensure that the
426 * namespace does not change while the lock is held.
428 static krwlock_t ztest_name_lock
;
430 static boolean_t ztest_dump_core
= B_TRUE
;
431 static boolean_t ztest_exiting
;
433 /* Global commit callback list */
434 static ztest_cb_list_t zcl
;
435 /* Commit cb delay */
436 static uint64_t zc_min_txg_delay
= UINT64_MAX
;
437 static int zc_cb_counter
= 0;
440 * Minimum number of commit callbacks that need to be registered for us to check
441 * whether the minimum txg delay is acceptable.
443 #define ZTEST_COMMIT_CB_MIN_REG 100
446 * If a number of txgs equal to this threshold have been created after a commit
447 * callback has been registered but not called, then we assume there is an
448 * implementation bug.
450 #define ZTEST_COMMIT_CB_THRESH (TXG_CONCURRENT_STATES + 1000)
452 extern uint64_t metaslab_gang_bang
;
453 extern uint64_t metaslab_df_alloc_threshold
;
456 ZTEST_META_DNODE
= 0,
461 static void usage(boolean_t
) __NORETURN
;
464 * These libumem hooks provide a reasonable set of defaults for the allocator's
465 * debugging facilities.
468 _umem_debug_init(void)
470 return ("default,verbose"); /* $UMEM_DEBUG setting */
474 _umem_logging_init(void)
476 return ("fail,contents"); /* $UMEM_LOGGING setting */
479 #define FATAL_MSG_SZ 1024
484 fatal(int do_perror
, char *message
, ...)
487 int save_errno
= errno
;
490 (void) fflush(stdout
);
491 buf
= umem_alloc(FATAL_MSG_SZ
, UMEM_NOFAIL
);
493 va_start(args
, message
);
494 (void) sprintf(buf
, "ztest: ");
496 (void) vsprintf(buf
+ strlen(buf
), message
, args
);
499 (void) snprintf(buf
+ strlen(buf
), FATAL_MSG_SZ
- strlen(buf
),
500 ": %s", strerror(save_errno
));
502 (void) fprintf(stderr
, "%s\n", buf
);
503 fatal_msg
= buf
; /* to ease debugging */
510 str2shift(const char *buf
)
512 const char *ends
= "BKMGTPEZ";
517 for (i
= 0; i
< strlen(ends
); i
++) {
518 if (toupper(buf
[0]) == ends
[i
])
521 if (i
== strlen(ends
)) {
522 (void) fprintf(stderr
, "ztest: invalid bytes suffix: %s\n",
526 if (buf
[1] == '\0' || (toupper(buf
[1]) == 'B' && buf
[2] == '\0')) {
529 (void) fprintf(stderr
, "ztest: invalid bytes suffix: %s\n", buf
);
535 nicenumtoull(const char *buf
)
540 val
= strtoull(buf
, &end
, 0);
542 (void) fprintf(stderr
, "ztest: bad numeric value: %s\n", buf
);
544 } else if (end
[0] == '.') {
545 double fval
= strtod(buf
, &end
);
546 fval
*= pow(2, str2shift(end
));
547 if (fval
> UINT64_MAX
) {
548 (void) fprintf(stderr
, "ztest: value too large: %s\n",
552 val
= (uint64_t)fval
;
554 int shift
= str2shift(end
);
555 if (shift
>= 64 || (val
<< shift
) >> shift
!= val
) {
556 (void) fprintf(stderr
, "ztest: value too large: %s\n",
566 usage(boolean_t requested
)
568 const ztest_shared_opts_t
*zo
= &ztest_opts_defaults
;
570 char nice_vdev_size
[10];
571 char nice_gang_bang
[10];
572 FILE *fp
= requested
? stdout
: stderr
;
574 nicenum(zo
->zo_vdev_size
, nice_vdev_size
);
575 nicenum(zo
->zo_metaslab_gang_bang
, nice_gang_bang
);
577 (void) fprintf(fp
, "Usage: %s\n"
578 "\t[-v vdevs (default: %llu)]\n"
579 "\t[-s size_of_each_vdev (default: %s)]\n"
580 "\t[-a alignment_shift (default: %d)] use 0 for random\n"
581 "\t[-m mirror_copies (default: %d)]\n"
582 "\t[-r raidz_disks (default: %d)]\n"
583 "\t[-R raidz_parity (default: %d)]\n"
584 "\t[-d datasets (default: %d)]\n"
585 "\t[-t threads (default: %d)]\n"
586 "\t[-g gang_block_threshold (default: %s)]\n"
587 "\t[-i init_count (default: %d)] initialize pool i times\n"
588 "\t[-k kill_percentage (default: %llu%%)]\n"
589 "\t[-p pool_name (default: %s)]\n"
590 "\t[-f dir (default: %s)] file directory for vdev files\n"
591 "\t[-V] verbose (use multiple times for ever more blather)\n"
592 "\t[-E] use existing pool instead of creating new one\n"
593 "\t[-T time (default: %llu sec)] total run time\n"
594 "\t[-F freezeloops (default: %llu)] max loops in spa_freeze()\n"
595 "\t[-P passtime (default: %llu sec)] time per pass\n"
596 "\t[-B alt_ztest (default: <none>)] alternate ztest path\n"
597 "\t[-h] (print help)\n"
600 (u_longlong_t
)zo
->zo_vdevs
, /* -v */
601 nice_vdev_size
, /* -s */
602 zo
->zo_ashift
, /* -a */
603 zo
->zo_mirrors
, /* -m */
604 zo
->zo_raidz
, /* -r */
605 zo
->zo_raidz_parity
, /* -R */
606 zo
->zo_datasets
, /* -d */
607 zo
->zo_threads
, /* -t */
608 nice_gang_bang
, /* -g */
609 zo
->zo_init
, /* -i */
610 (u_longlong_t
)zo
->zo_killrate
, /* -k */
611 zo
->zo_pool
, /* -p */
613 (u_longlong_t
)zo
->zo_time
, /* -T */
614 (u_longlong_t
)zo
->zo_maxloops
, /* -F */
615 (u_longlong_t
)zo
->zo_passtime
);
616 exit(requested
? 0 : 1);
620 process_options(int argc
, char **argv
)
623 ztest_shared_opts_t
*zo
= &ztest_opts
;
627 char altdir
[MAXNAMELEN
] = { 0 };
629 bcopy(&ztest_opts_defaults
, zo
, sizeof (*zo
));
631 while ((opt
= getopt(argc
, argv
,
632 "v:s:a:m:r:R:d:t:g:i:k:p:f:VET:P:hF:B:")) != EOF
) {
649 value
= nicenumtoull(optarg
);
653 zo
->zo_vdevs
= value
;
656 zo
->zo_vdev_size
= MAX(SPA_MINDEVSIZE
, value
);
659 zo
->zo_ashift
= value
;
662 zo
->zo_mirrors
= value
;
665 zo
->zo_raidz
= MAX(1, value
);
668 zo
->zo_raidz_parity
= MIN(MAX(value
, 1), 3);
671 zo
->zo_datasets
= MAX(1, value
);
674 zo
->zo_threads
= MAX(1, value
);
677 zo
->zo_metaslab_gang_bang
= MAX(SPA_MINBLOCKSIZE
<< 1,
684 zo
->zo_killrate
= value
;
687 (void) strlcpy(zo
->zo_pool
, optarg
,
688 sizeof (zo
->zo_pool
));
691 path
= realpath(optarg
, NULL
);
693 (void) fprintf(stderr
, "error: %s: %s\n",
694 optarg
, strerror(errno
));
697 (void) strlcpy(zo
->zo_dir
, path
,
698 sizeof (zo
->zo_dir
));
711 zo
->zo_passtime
= MAX(1, value
);
714 zo
->zo_maxloops
= MAX(1, value
);
717 (void) strlcpy(altdir
, optarg
, sizeof (altdir
));
729 zo
->zo_raidz_parity
= MIN(zo
->zo_raidz_parity
, zo
->zo_raidz
- 1);
732 (zo
->zo_vdevs
> 0 ? zo
->zo_time
* NANOSEC
/ zo
->zo_vdevs
:
735 if (strlen(altdir
) > 0) {
743 cmd
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
744 realaltdir
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
746 VERIFY(NULL
!= realpath(getexecname(), cmd
));
747 if (0 != access(altdir
, F_OK
)) {
748 ztest_dump_core
= B_FALSE
;
749 fatal(B_TRUE
, "invalid alternate ztest path: %s",
752 VERIFY(NULL
!= realpath(altdir
, realaltdir
));
755 * 'cmd' should be of the form "<anything>/usr/bin/<isa>/ztest".
756 * We want to extract <isa> to determine if we should use
757 * 32 or 64 bit binaries.
759 bin
= strstr(cmd
, "/usr/bin/");
760 ztest
= strstr(bin
, "/ztest");
762 isalen
= ztest
- isa
;
763 (void) snprintf(zo
->zo_alt_ztest
, sizeof (zo
->zo_alt_ztest
),
764 "%s/usr/bin/%.*s/ztest", realaltdir
, isalen
, isa
);
765 (void) snprintf(zo
->zo_alt_libpath
, sizeof (zo
->zo_alt_libpath
),
766 "%s/usr/lib/%.*s", realaltdir
, isalen
, isa
);
768 if (0 != access(zo
->zo_alt_ztest
, X_OK
)) {
769 ztest_dump_core
= B_FALSE
;
770 fatal(B_TRUE
, "invalid alternate ztest: %s",
772 } else if (0 != access(zo
->zo_alt_libpath
, X_OK
)) {
773 ztest_dump_core
= B_FALSE
;
774 fatal(B_TRUE
, "invalid alternate lib directory %s",
778 umem_free(cmd
, MAXPATHLEN
);
779 umem_free(realaltdir
, MAXPATHLEN
);
784 ztest_kill(ztest_shared_t
*zs
)
786 zs
->zs_alloc
= metaslab_class_get_alloc(spa_normal_class(ztest_spa
));
787 zs
->zs_space
= metaslab_class_get_space(spa_normal_class(ztest_spa
));
788 (void) kill(getpid(), SIGKILL
);
792 ztest_random(uint64_t range
)
796 ASSERT3S(ztest_fd_rand
, >=, 0);
801 if (read(ztest_fd_rand
, &r
, sizeof (r
)) != sizeof (r
))
802 fatal(1, "short read from /dev/urandom");
809 ztest_record_enospc(const char *s
)
811 ztest_shared
->zs_enospc_count
++;
815 ztest_get_ashift(void)
817 if (ztest_opts
.zo_ashift
== 0)
818 return (SPA_MINBLOCKSHIFT
+ ztest_random(3));
819 return (ztest_opts
.zo_ashift
);
823 make_vdev_file(char *path
, char *aux
, char *pool
, size_t size
, uint64_t ashift
)
829 pathbuf
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
832 ashift
= ztest_get_ashift();
838 vdev
= ztest_shared
->zs_vdev_aux
;
839 (void) snprintf(path
, MAXPATHLEN
,
840 ztest_aux_template
, ztest_opts
.zo_dir
,
841 pool
== NULL
? ztest_opts
.zo_pool
: pool
,
844 vdev
= ztest_shared
->zs_vdev_next_leaf
++;
845 (void) snprintf(path
, MAXPATHLEN
,
846 ztest_dev_template
, ztest_opts
.zo_dir
,
847 pool
== NULL
? ztest_opts
.zo_pool
: pool
, vdev
);
852 int fd
= open(path
, O_RDWR
| O_CREAT
| O_TRUNC
, 0666);
854 fatal(1, "can't open %s", path
);
855 if (ftruncate(fd
, size
) != 0)
856 fatal(1, "can't ftruncate %s", path
);
860 VERIFY(nvlist_alloc(&file
, NV_UNIQUE_NAME
, 0) == 0);
861 VERIFY(nvlist_add_string(file
, ZPOOL_CONFIG_TYPE
, VDEV_TYPE_FILE
) == 0);
862 VERIFY(nvlist_add_string(file
, ZPOOL_CONFIG_PATH
, path
) == 0);
863 VERIFY(nvlist_add_uint64(file
, ZPOOL_CONFIG_ASHIFT
, ashift
) == 0);
864 umem_free(pathbuf
, MAXPATHLEN
);
870 make_vdev_raidz(char *path
, char *aux
, char *pool
, size_t size
,
871 uint64_t ashift
, int r
)
873 nvlist_t
*raidz
, **child
;
877 return (make_vdev_file(path
, aux
, pool
, size
, ashift
));
878 child
= umem_alloc(r
* sizeof (nvlist_t
*), UMEM_NOFAIL
);
880 for (c
= 0; c
< r
; c
++)
881 child
[c
] = make_vdev_file(path
, aux
, pool
, size
, ashift
);
883 VERIFY(nvlist_alloc(&raidz
, NV_UNIQUE_NAME
, 0) == 0);
884 VERIFY(nvlist_add_string(raidz
, ZPOOL_CONFIG_TYPE
,
885 VDEV_TYPE_RAIDZ
) == 0);
886 VERIFY(nvlist_add_uint64(raidz
, ZPOOL_CONFIG_NPARITY
,
887 ztest_opts
.zo_raidz_parity
) == 0);
888 VERIFY(nvlist_add_nvlist_array(raidz
, ZPOOL_CONFIG_CHILDREN
,
891 for (c
= 0; c
< r
; c
++)
892 nvlist_free(child
[c
]);
894 umem_free(child
, r
* sizeof (nvlist_t
*));
900 make_vdev_mirror(char *path
, char *aux
, char *pool
, size_t size
,
901 uint64_t ashift
, int r
, int m
)
903 nvlist_t
*mirror
, **child
;
907 return (make_vdev_raidz(path
, aux
, pool
, size
, ashift
, r
));
909 child
= umem_alloc(m
* sizeof (nvlist_t
*), UMEM_NOFAIL
);
911 for (c
= 0; c
< m
; c
++)
912 child
[c
] = make_vdev_raidz(path
, aux
, pool
, size
, ashift
, r
);
914 VERIFY(nvlist_alloc(&mirror
, NV_UNIQUE_NAME
, 0) == 0);
915 VERIFY(nvlist_add_string(mirror
, ZPOOL_CONFIG_TYPE
,
916 VDEV_TYPE_MIRROR
) == 0);
917 VERIFY(nvlist_add_nvlist_array(mirror
, ZPOOL_CONFIG_CHILDREN
,
920 for (c
= 0; c
< m
; c
++)
921 nvlist_free(child
[c
]);
923 umem_free(child
, m
* sizeof (nvlist_t
*));
929 make_vdev_root(char *path
, char *aux
, char *pool
, size_t size
, uint64_t ashift
,
930 int log
, int r
, int m
, int t
)
932 nvlist_t
*root
, **child
;
937 child
= umem_alloc(t
* sizeof (nvlist_t
*), UMEM_NOFAIL
);
939 for (c
= 0; c
< t
; c
++) {
940 child
[c
] = make_vdev_mirror(path
, aux
, pool
, size
, ashift
,
942 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_IS_LOG
,
946 VERIFY(nvlist_alloc(&root
, NV_UNIQUE_NAME
, 0) == 0);
947 VERIFY(nvlist_add_string(root
, ZPOOL_CONFIG_TYPE
, VDEV_TYPE_ROOT
) == 0);
948 VERIFY(nvlist_add_nvlist_array(root
, aux
? aux
: ZPOOL_CONFIG_CHILDREN
,
951 for (c
= 0; c
< t
; c
++)
952 nvlist_free(child
[c
]);
954 umem_free(child
, t
* sizeof (nvlist_t
*));
960 * Find a random spa version. Returns back a random spa version in the
961 * range [initial_version, SPA_VERSION_FEATURES].
964 ztest_random_spa_version(uint64_t initial_version
)
966 uint64_t version
= initial_version
;
968 if (version
<= SPA_VERSION_BEFORE_FEATURES
) {
970 ztest_random(SPA_VERSION_BEFORE_FEATURES
- version
+ 1);
973 if (version
> SPA_VERSION_BEFORE_FEATURES
)
974 version
= SPA_VERSION_FEATURES
;
976 ASSERT(SPA_VERSION_IS_SUPPORTED(version
));
981 ztest_random_blocksize(void)
983 return (1 << (SPA_MINBLOCKSHIFT
+
984 ztest_random(SPA_MAXBLOCKSHIFT
- SPA_MINBLOCKSHIFT
+ 1)));
988 ztest_random_ibshift(void)
990 return (DN_MIN_INDBLKSHIFT
+
991 ztest_random(DN_MAX_INDBLKSHIFT
- DN_MIN_INDBLKSHIFT
+ 1));
995 ztest_random_vdev_top(spa_t
*spa
, boolean_t log_ok
)
998 vdev_t
*rvd
= spa
->spa_root_vdev
;
1001 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_READER
) != 0);
1004 top
= ztest_random(rvd
->vdev_children
);
1005 tvd
= rvd
->vdev_child
[top
];
1006 } while (tvd
->vdev_ishole
|| (tvd
->vdev_islog
&& !log_ok
) ||
1007 tvd
->vdev_mg
== NULL
|| tvd
->vdev_mg
->mg_class
== NULL
);
1013 ztest_random_dsl_prop(zfs_prop_t prop
)
1018 value
= zfs_prop_random_value(prop
, ztest_random(-1ULL));
1019 } while (prop
== ZFS_PROP_CHECKSUM
&& value
== ZIO_CHECKSUM_OFF
);
1025 ztest_dsl_prop_set_uint64(char *osname
, zfs_prop_t prop
, uint64_t value
,
1028 const char *propname
= zfs_prop_to_name(prop
);
1029 const char *valname
;
1034 error
= dsl_prop_set(osname
, propname
,
1035 (inherit
? ZPROP_SRC_NONE
: ZPROP_SRC_LOCAL
),
1036 sizeof (value
), 1, &value
);
1038 if (error
== ENOSPC
) {
1039 ztest_record_enospc(FTAG
);
1044 setpoint
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
1045 VERIFY3U(dsl_prop_get(osname
, propname
, sizeof (curval
),
1046 1, &curval
, setpoint
), ==, 0);
1048 if (ztest_opts
.zo_verbose
>= 6) {
1049 VERIFY(zfs_prop_index_to_string(prop
, curval
, &valname
) == 0);
1050 (void) printf("%s %s = %s at '%s'\n",
1051 osname
, propname
, valname
, setpoint
);
1053 umem_free(setpoint
, MAXPATHLEN
);
1059 ztest_spa_prop_set_uint64(zpool_prop_t prop
, uint64_t value
)
1061 spa_t
*spa
= ztest_spa
;
1062 nvlist_t
*props
= NULL
;
1065 VERIFY(nvlist_alloc(&props
, NV_UNIQUE_NAME
, 0) == 0);
1066 VERIFY(nvlist_add_uint64(props
, zpool_prop_to_name(prop
), value
) == 0);
1068 error
= spa_prop_set(spa
, props
);
1072 if (error
== ENOSPC
) {
1073 ztest_record_enospc(FTAG
);
1082 ztest_rll_init(rll_t
*rll
)
1084 rll
->rll_writer
= NULL
;
1085 rll
->rll_readers
= 0;
1086 mutex_init(&rll
->rll_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
1087 cv_init(&rll
->rll_cv
, NULL
, CV_DEFAULT
, NULL
);
1091 ztest_rll_destroy(rll_t
*rll
)
1093 ASSERT(rll
->rll_writer
== NULL
);
1094 ASSERT(rll
->rll_readers
== 0);
1095 mutex_destroy(&rll
->rll_lock
);
1096 cv_destroy(&rll
->rll_cv
);
1100 ztest_rll_lock(rll_t
*rll
, rl_type_t type
)
1102 mutex_enter(&rll
->rll_lock
);
1104 if (type
== RL_READER
) {
1105 while (rll
->rll_writer
!= NULL
)
1106 (void) cv_wait(&rll
->rll_cv
, &rll
->rll_lock
);
1109 while (rll
->rll_writer
!= NULL
|| rll
->rll_readers
)
1110 (void) cv_wait(&rll
->rll_cv
, &rll
->rll_lock
);
1111 rll
->rll_writer
= curthread
;
1114 mutex_exit(&rll
->rll_lock
);
1118 ztest_rll_unlock(rll_t
*rll
)
1120 mutex_enter(&rll
->rll_lock
);
1122 if (rll
->rll_writer
) {
1123 ASSERT(rll
->rll_readers
== 0);
1124 rll
->rll_writer
= NULL
;
1126 ASSERT(rll
->rll_readers
!= 0);
1127 ASSERT(rll
->rll_writer
== NULL
);
1131 if (rll
->rll_writer
== NULL
&& rll
->rll_readers
== 0)
1132 cv_broadcast(&rll
->rll_cv
);
1134 mutex_exit(&rll
->rll_lock
);
1138 ztest_object_lock(ztest_ds_t
*zd
, uint64_t object
, rl_type_t type
)
1140 rll_t
*rll
= &zd
->zd_object_lock
[object
& (ZTEST_OBJECT_LOCKS
- 1)];
1142 ztest_rll_lock(rll
, type
);
1146 ztest_object_unlock(ztest_ds_t
*zd
, uint64_t object
)
1148 rll_t
*rll
= &zd
->zd_object_lock
[object
& (ZTEST_OBJECT_LOCKS
- 1)];
1150 ztest_rll_unlock(rll
);
1154 ztest_range_lock(ztest_ds_t
*zd
, uint64_t object
, uint64_t offset
,
1155 uint64_t size
, rl_type_t type
)
1157 uint64_t hash
= object
^ (offset
% (ZTEST_RANGE_LOCKS
+ 1));
1158 rll_t
*rll
= &zd
->zd_range_lock
[hash
& (ZTEST_RANGE_LOCKS
- 1)];
1161 rl
= umem_alloc(sizeof (*rl
), UMEM_NOFAIL
);
1162 rl
->rl_object
= object
;
1163 rl
->rl_offset
= offset
;
1167 ztest_rll_lock(rll
, type
);
1173 ztest_range_unlock(rl_t
*rl
)
1175 rll_t
*rll
= rl
->rl_lock
;
1177 ztest_rll_unlock(rll
);
1179 umem_free(rl
, sizeof (*rl
));
1183 ztest_zd_init(ztest_ds_t
*zd
, ztest_shared_ds_t
*szd
, objset_t
*os
)
1186 zd
->zd_zilog
= dmu_objset_zil(os
);
1187 zd
->zd_shared
= szd
;
1188 dmu_objset_name(os
, zd
->zd_name
);
1191 if (zd
->zd_shared
!= NULL
)
1192 zd
->zd_shared
->zd_seq
= 0;
1194 rw_init(&zd
->zd_zilog_lock
, NULL
, RW_DEFAULT
, NULL
);
1195 mutex_init(&zd
->zd_dirobj_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
1197 for (l
= 0; l
< ZTEST_OBJECT_LOCKS
; l
++)
1198 ztest_rll_init(&zd
->zd_object_lock
[l
]);
1200 for (l
= 0; l
< ZTEST_RANGE_LOCKS
; l
++)
1201 ztest_rll_init(&zd
->zd_range_lock
[l
]);
1205 ztest_zd_fini(ztest_ds_t
*zd
)
1209 mutex_destroy(&zd
->zd_dirobj_lock
);
1210 rw_destroy(&zd
->zd_zilog_lock
);
1212 for (l
= 0; l
< ZTEST_OBJECT_LOCKS
; l
++)
1213 ztest_rll_destroy(&zd
->zd_object_lock
[l
]);
1215 for (l
= 0; l
< ZTEST_RANGE_LOCKS
; l
++)
1216 ztest_rll_destroy(&zd
->zd_range_lock
[l
]);
1219 #define TXG_MIGHTWAIT (ztest_random(10) == 0 ? TXG_NOWAIT : TXG_WAIT)
1222 ztest_tx_assign(dmu_tx_t
*tx
, uint64_t txg_how
, const char *tag
)
1228 * Attempt to assign tx to some transaction group.
1230 error
= dmu_tx_assign(tx
, txg_how
);
1232 if (error
== ERESTART
) {
1233 ASSERT(txg_how
== TXG_NOWAIT
);
1236 ASSERT3U(error
, ==, ENOSPC
);
1237 ztest_record_enospc(tag
);
1242 txg
= dmu_tx_get_txg(tx
);
1248 ztest_pattern_set(void *buf
, uint64_t size
, uint64_t value
)
1251 uint64_t *ip_end
= (uint64_t *)((uintptr_t)buf
+ (uintptr_t)size
);
1259 ztest_pattern_match(void *buf
, uint64_t size
, uint64_t value
)
1262 uint64_t *ip_end
= (uint64_t *)((uintptr_t)buf
+ (uintptr_t)size
);
1266 diff
|= (value
- *ip
++);
1273 ztest_bt_generate(ztest_block_tag_t
*bt
, objset_t
*os
, uint64_t object
,
1274 uint64_t offset
, uint64_t gen
, uint64_t txg
, uint64_t crtxg
)
1276 bt
->bt_magic
= BT_MAGIC
;
1277 bt
->bt_objset
= dmu_objset_id(os
);
1278 bt
->bt_object
= object
;
1279 bt
->bt_offset
= offset
;
1282 bt
->bt_crtxg
= crtxg
;
1286 ztest_bt_verify(ztest_block_tag_t
*bt
, objset_t
*os
, uint64_t object
,
1287 uint64_t offset
, uint64_t gen
, uint64_t txg
, uint64_t crtxg
)
1289 ASSERT(bt
->bt_magic
== BT_MAGIC
);
1290 ASSERT(bt
->bt_objset
== dmu_objset_id(os
));
1291 ASSERT(bt
->bt_object
== object
);
1292 ASSERT(bt
->bt_offset
== offset
);
1293 ASSERT(bt
->bt_gen
<= gen
);
1294 ASSERT(bt
->bt_txg
<= txg
);
1295 ASSERT(bt
->bt_crtxg
== crtxg
);
1298 static ztest_block_tag_t
*
1299 ztest_bt_bonus(dmu_buf_t
*db
)
1301 dmu_object_info_t doi
;
1302 ztest_block_tag_t
*bt
;
1304 dmu_object_info_from_db(db
, &doi
);
1305 ASSERT3U(doi
.doi_bonus_size
, <=, db
->db_size
);
1306 ASSERT3U(doi
.doi_bonus_size
, >=, sizeof (*bt
));
1307 bt
= (void *)((char *)db
->db_data
+ doi
.doi_bonus_size
- sizeof (*bt
));
1316 #define lrz_type lr_mode
1317 #define lrz_blocksize lr_uid
1318 #define lrz_ibshift lr_gid
1319 #define lrz_bonustype lr_rdev
1320 #define lrz_bonuslen lr_crtime[1]
1323 ztest_log_create(ztest_ds_t
*zd
, dmu_tx_t
*tx
, lr_create_t
*lr
)
1325 char *name
= (void *)(lr
+ 1); /* name follows lr */
1326 size_t namesize
= strlen(name
) + 1;
1329 if (zil_replaying(zd
->zd_zilog
, tx
))
1332 itx
= zil_itx_create(TX_CREATE
, sizeof (*lr
) + namesize
);
1333 bcopy(&lr
->lr_common
+ 1, &itx
->itx_lr
+ 1,
1334 sizeof (*lr
) + namesize
- sizeof (lr_t
));
1336 zil_itx_assign(zd
->zd_zilog
, itx
, tx
);
1340 ztest_log_remove(ztest_ds_t
*zd
, dmu_tx_t
*tx
, lr_remove_t
*lr
, uint64_t object
)
1342 char *name
= (void *)(lr
+ 1); /* name follows lr */
1343 size_t namesize
= strlen(name
) + 1;
1346 if (zil_replaying(zd
->zd_zilog
, tx
))
1349 itx
= zil_itx_create(TX_REMOVE
, sizeof (*lr
) + namesize
);
1350 bcopy(&lr
->lr_common
+ 1, &itx
->itx_lr
+ 1,
1351 sizeof (*lr
) + namesize
- sizeof (lr_t
));
1353 itx
->itx_oid
= object
;
1354 zil_itx_assign(zd
->zd_zilog
, itx
, tx
);
1358 ztest_log_write(ztest_ds_t
*zd
, dmu_tx_t
*tx
, lr_write_t
*lr
)
1361 itx_wr_state_t write_state
= ztest_random(WR_NUM_STATES
);
1363 if (zil_replaying(zd
->zd_zilog
, tx
))
1366 if (lr
->lr_length
> ZIL_MAX_LOG_DATA
)
1367 write_state
= WR_INDIRECT
;
1369 itx
= zil_itx_create(TX_WRITE
,
1370 sizeof (*lr
) + (write_state
== WR_COPIED
? lr
->lr_length
: 0));
1372 if (write_state
== WR_COPIED
&&
1373 dmu_read(zd
->zd_os
, lr
->lr_foid
, lr
->lr_offset
, lr
->lr_length
,
1374 ((lr_write_t
*)&itx
->itx_lr
) + 1, DMU_READ_NO_PREFETCH
) != 0) {
1375 zil_itx_destroy(itx
);
1376 itx
= zil_itx_create(TX_WRITE
, sizeof (*lr
));
1377 write_state
= WR_NEED_COPY
;
1379 itx
->itx_private
= zd
;
1380 itx
->itx_wr_state
= write_state
;
1381 itx
->itx_sync
= (ztest_random(8) == 0);
1382 itx
->itx_sod
+= (write_state
== WR_NEED_COPY
? lr
->lr_length
: 0);
1384 bcopy(&lr
->lr_common
+ 1, &itx
->itx_lr
+ 1,
1385 sizeof (*lr
) - sizeof (lr_t
));
1387 zil_itx_assign(zd
->zd_zilog
, itx
, tx
);
1391 ztest_log_truncate(ztest_ds_t
*zd
, dmu_tx_t
*tx
, lr_truncate_t
*lr
)
1395 if (zil_replaying(zd
->zd_zilog
, tx
))
1398 itx
= zil_itx_create(TX_TRUNCATE
, sizeof (*lr
));
1399 bcopy(&lr
->lr_common
+ 1, &itx
->itx_lr
+ 1,
1400 sizeof (*lr
) - sizeof (lr_t
));
1402 itx
->itx_sync
= B_FALSE
;
1403 zil_itx_assign(zd
->zd_zilog
, itx
, tx
);
1407 ztest_log_setattr(ztest_ds_t
*zd
, dmu_tx_t
*tx
, lr_setattr_t
*lr
)
1411 if (zil_replaying(zd
->zd_zilog
, tx
))
1414 itx
= zil_itx_create(TX_SETATTR
, sizeof (*lr
));
1415 bcopy(&lr
->lr_common
+ 1, &itx
->itx_lr
+ 1,
1416 sizeof (*lr
) - sizeof (lr_t
));
1418 itx
->itx_sync
= B_FALSE
;
1419 zil_itx_assign(zd
->zd_zilog
, itx
, tx
);
1426 ztest_replay_create(ztest_ds_t
*zd
, lr_create_t
*lr
, boolean_t byteswap
)
1428 char *name
= (void *)(lr
+ 1); /* name follows lr */
1429 objset_t
*os
= zd
->zd_os
;
1430 ztest_block_tag_t
*bbt
;
1437 byteswap_uint64_array(lr
, sizeof (*lr
));
1439 ASSERT(lr
->lr_doid
== ZTEST_DIROBJ
);
1440 ASSERT(name
[0] != '\0');
1442 tx
= dmu_tx_create(os
);
1444 dmu_tx_hold_zap(tx
, lr
->lr_doid
, B_TRUE
, name
);
1446 if (lr
->lrz_type
== DMU_OT_ZAP_OTHER
) {
1447 dmu_tx_hold_zap(tx
, DMU_NEW_OBJECT
, B_TRUE
, NULL
);
1449 dmu_tx_hold_bonus(tx
, DMU_NEW_OBJECT
);
1452 txg
= ztest_tx_assign(tx
, TXG_WAIT
, FTAG
);
1456 ASSERT(dmu_objset_zil(os
)->zl_replay
== !!lr
->lr_foid
);
1458 if (lr
->lrz_type
== DMU_OT_ZAP_OTHER
) {
1459 if (lr
->lr_foid
== 0) {
1460 lr
->lr_foid
= zap_create(os
,
1461 lr
->lrz_type
, lr
->lrz_bonustype
,
1462 lr
->lrz_bonuslen
, tx
);
1464 error
= zap_create_claim(os
, lr
->lr_foid
,
1465 lr
->lrz_type
, lr
->lrz_bonustype
,
1466 lr
->lrz_bonuslen
, tx
);
1469 if (lr
->lr_foid
== 0) {
1470 lr
->lr_foid
= dmu_object_alloc(os
,
1471 lr
->lrz_type
, 0, lr
->lrz_bonustype
,
1472 lr
->lrz_bonuslen
, tx
);
1474 error
= dmu_object_claim(os
, lr
->lr_foid
,
1475 lr
->lrz_type
, 0, lr
->lrz_bonustype
,
1476 lr
->lrz_bonuslen
, tx
);
1481 ASSERT3U(error
, ==, EEXIST
);
1482 ASSERT(zd
->zd_zilog
->zl_replay
);
1487 ASSERT(lr
->lr_foid
!= 0);
1489 if (lr
->lrz_type
!= DMU_OT_ZAP_OTHER
)
1490 VERIFY3U(0, ==, dmu_object_set_blocksize(os
, lr
->lr_foid
,
1491 lr
->lrz_blocksize
, lr
->lrz_ibshift
, tx
));
1493 VERIFY3U(0, ==, dmu_bonus_hold(os
, lr
->lr_foid
, FTAG
, &db
));
1494 bbt
= ztest_bt_bonus(db
);
1495 dmu_buf_will_dirty(db
, tx
);
1496 ztest_bt_generate(bbt
, os
, lr
->lr_foid
, -1ULL, lr
->lr_gen
, txg
, txg
);
1497 dmu_buf_rele(db
, FTAG
);
1499 VERIFY3U(0, ==, zap_add(os
, lr
->lr_doid
, name
, sizeof (uint64_t), 1,
1502 (void) ztest_log_create(zd
, tx
, lr
);
1510 ztest_replay_remove(ztest_ds_t
*zd
, lr_remove_t
*lr
, boolean_t byteswap
)
1512 char *name
= (void *)(lr
+ 1); /* name follows lr */
1513 objset_t
*os
= zd
->zd_os
;
1514 dmu_object_info_t doi
;
1516 uint64_t object
, txg
;
1519 byteswap_uint64_array(lr
, sizeof (*lr
));
1521 ASSERT(lr
->lr_doid
== ZTEST_DIROBJ
);
1522 ASSERT(name
[0] != '\0');
1525 zap_lookup(os
, lr
->lr_doid
, name
, sizeof (object
), 1, &object
));
1526 ASSERT(object
!= 0);
1528 ztest_object_lock(zd
, object
, RL_WRITER
);
1530 VERIFY3U(0, ==, dmu_object_info(os
, object
, &doi
));
1532 tx
= dmu_tx_create(os
);
1534 dmu_tx_hold_zap(tx
, lr
->lr_doid
, B_FALSE
, name
);
1535 dmu_tx_hold_free(tx
, object
, 0, DMU_OBJECT_END
);
1537 txg
= ztest_tx_assign(tx
, TXG_WAIT
, FTAG
);
1539 ztest_object_unlock(zd
, object
);
1543 if (doi
.doi_type
== DMU_OT_ZAP_OTHER
) {
1544 VERIFY3U(0, ==, zap_destroy(os
, object
, tx
));
1546 VERIFY3U(0, ==, dmu_object_free(os
, object
, tx
));
1549 VERIFY3U(0, ==, zap_remove(os
, lr
->lr_doid
, name
, tx
));
1551 (void) ztest_log_remove(zd
, tx
, lr
, object
);
1555 ztest_object_unlock(zd
, object
);
1561 ztest_replay_write(ztest_ds_t
*zd
, lr_write_t
*lr
, boolean_t byteswap
)
1563 objset_t
*os
= zd
->zd_os
;
1564 void *data
= lr
+ 1; /* data follows lr */
1565 uint64_t offset
, length
;
1566 ztest_block_tag_t
*bt
= data
;
1567 ztest_block_tag_t
*bbt
;
1568 uint64_t gen
, txg
, lrtxg
, crtxg
;
1569 dmu_object_info_t doi
;
1572 arc_buf_t
*abuf
= NULL
;
1576 byteswap_uint64_array(lr
, sizeof (*lr
));
1578 offset
= lr
->lr_offset
;
1579 length
= lr
->lr_length
;
1581 /* If it's a dmu_sync() block, write the whole block */
1582 if (lr
->lr_common
.lrc_reclen
== sizeof (lr_write_t
)) {
1583 uint64_t blocksize
= BP_GET_LSIZE(&lr
->lr_blkptr
);
1584 if (length
< blocksize
) {
1585 offset
-= offset
% blocksize
;
1590 if (bt
->bt_magic
== BSWAP_64(BT_MAGIC
))
1591 byteswap_uint64_array(bt
, sizeof (*bt
));
1593 if (bt
->bt_magic
!= BT_MAGIC
)
1596 ztest_object_lock(zd
, lr
->lr_foid
, RL_READER
);
1597 rl
= ztest_range_lock(zd
, lr
->lr_foid
, offset
, length
, RL_WRITER
);
1599 VERIFY3U(0, ==, dmu_bonus_hold(os
, lr
->lr_foid
, FTAG
, &db
));
1601 dmu_object_info_from_db(db
, &doi
);
1603 bbt
= ztest_bt_bonus(db
);
1604 ASSERT3U(bbt
->bt_magic
, ==, BT_MAGIC
);
1606 crtxg
= bbt
->bt_crtxg
;
1607 lrtxg
= lr
->lr_common
.lrc_txg
;
1609 tx
= dmu_tx_create(os
);
1611 dmu_tx_hold_write(tx
, lr
->lr_foid
, offset
, length
);
1613 if (ztest_random(8) == 0 && length
== doi
.doi_data_block_size
&&
1614 P2PHASE(offset
, length
) == 0)
1615 abuf
= dmu_request_arcbuf(db
, length
);
1617 txg
= ztest_tx_assign(tx
, TXG_WAIT
, FTAG
);
1620 dmu_return_arcbuf(abuf
);
1621 dmu_buf_rele(db
, FTAG
);
1622 ztest_range_unlock(rl
);
1623 ztest_object_unlock(zd
, lr
->lr_foid
);
1629 * Usually, verify the old data before writing new data --
1630 * but not always, because we also want to verify correct
1631 * behavior when the data was not recently read into cache.
1633 ASSERT(offset
% doi
.doi_data_block_size
== 0);
1634 if (ztest_random(4) != 0) {
1635 int prefetch
= ztest_random(2) ?
1636 DMU_READ_PREFETCH
: DMU_READ_NO_PREFETCH
;
1637 ztest_block_tag_t rbt
;
1639 VERIFY(dmu_read(os
, lr
->lr_foid
, offset
,
1640 sizeof (rbt
), &rbt
, prefetch
) == 0);
1641 if (rbt
.bt_magic
== BT_MAGIC
) {
1642 ztest_bt_verify(&rbt
, os
, lr
->lr_foid
,
1643 offset
, gen
, txg
, crtxg
);
1648 * Writes can appear to be newer than the bonus buffer because
1649 * the ztest_get_data() callback does a dmu_read() of the
1650 * open-context data, which may be different than the data
1651 * as it was when the write was generated.
1653 if (zd
->zd_zilog
->zl_replay
) {
1654 ztest_bt_verify(bt
, os
, lr
->lr_foid
, offset
,
1655 MAX(gen
, bt
->bt_gen
), MAX(txg
, lrtxg
),
1660 * Set the bt's gen/txg to the bonus buffer's gen/txg
1661 * so that all of the usual ASSERTs will work.
1663 ztest_bt_generate(bt
, os
, lr
->lr_foid
, offset
, gen
, txg
, crtxg
);
1667 dmu_write(os
, lr
->lr_foid
, offset
, length
, data
, tx
);
1669 bcopy(data
, abuf
->b_data
, length
);
1670 dmu_assign_arcbuf(db
, offset
, abuf
, tx
);
1673 (void) ztest_log_write(zd
, tx
, lr
);
1675 dmu_buf_rele(db
, FTAG
);
1679 ztest_range_unlock(rl
);
1680 ztest_object_unlock(zd
, lr
->lr_foid
);
1686 ztest_replay_truncate(ztest_ds_t
*zd
, lr_truncate_t
*lr
, boolean_t byteswap
)
1688 objset_t
*os
= zd
->zd_os
;
1694 byteswap_uint64_array(lr
, sizeof (*lr
));
1696 ztest_object_lock(zd
, lr
->lr_foid
, RL_READER
);
1697 rl
= ztest_range_lock(zd
, lr
->lr_foid
, lr
->lr_offset
, lr
->lr_length
,
1700 tx
= dmu_tx_create(os
);
1702 dmu_tx_hold_free(tx
, lr
->lr_foid
, lr
->lr_offset
, lr
->lr_length
);
1704 txg
= ztest_tx_assign(tx
, TXG_WAIT
, FTAG
);
1706 ztest_range_unlock(rl
);
1707 ztest_object_unlock(zd
, lr
->lr_foid
);
1711 VERIFY(dmu_free_range(os
, lr
->lr_foid
, lr
->lr_offset
,
1712 lr
->lr_length
, tx
) == 0);
1714 (void) ztest_log_truncate(zd
, tx
, lr
);
1718 ztest_range_unlock(rl
);
1719 ztest_object_unlock(zd
, lr
->lr_foid
);
1725 ztest_replay_setattr(ztest_ds_t
*zd
, lr_setattr_t
*lr
, boolean_t byteswap
)
1727 objset_t
*os
= zd
->zd_os
;
1730 ztest_block_tag_t
*bbt
;
1731 uint64_t txg
, lrtxg
, crtxg
;
1734 byteswap_uint64_array(lr
, sizeof (*lr
));
1736 ztest_object_lock(zd
, lr
->lr_foid
, RL_WRITER
);
1738 VERIFY3U(0, ==, dmu_bonus_hold(os
, lr
->lr_foid
, FTAG
, &db
));
1740 tx
= dmu_tx_create(os
);
1741 dmu_tx_hold_bonus(tx
, lr
->lr_foid
);
1743 txg
= ztest_tx_assign(tx
, TXG_WAIT
, FTAG
);
1745 dmu_buf_rele(db
, FTAG
);
1746 ztest_object_unlock(zd
, lr
->lr_foid
);
1750 bbt
= ztest_bt_bonus(db
);
1751 ASSERT3U(bbt
->bt_magic
, ==, BT_MAGIC
);
1752 crtxg
= bbt
->bt_crtxg
;
1753 lrtxg
= lr
->lr_common
.lrc_txg
;
1755 if (zd
->zd_zilog
->zl_replay
) {
1756 ASSERT(lr
->lr_size
!= 0);
1757 ASSERT(lr
->lr_mode
!= 0);
1761 * Randomly change the size and increment the generation.
1763 lr
->lr_size
= (ztest_random(db
->db_size
/ sizeof (*bbt
)) + 1) *
1765 lr
->lr_mode
= bbt
->bt_gen
+ 1;
1770 * Verify that the current bonus buffer is not newer than our txg.
1772 ztest_bt_verify(bbt
, os
, lr
->lr_foid
, -1ULL, lr
->lr_mode
,
1773 MAX(txg
, lrtxg
), crtxg
);
1775 dmu_buf_will_dirty(db
, tx
);
1777 ASSERT3U(lr
->lr_size
, >=, sizeof (*bbt
));
1778 ASSERT3U(lr
->lr_size
, <=, db
->db_size
);
1779 VERIFY0(dmu_set_bonus(db
, lr
->lr_size
, tx
));
1780 bbt
= ztest_bt_bonus(db
);
1782 ztest_bt_generate(bbt
, os
, lr
->lr_foid
, -1ULL, lr
->lr_mode
, txg
, crtxg
);
1784 dmu_buf_rele(db
, FTAG
);
1786 (void) ztest_log_setattr(zd
, tx
, lr
);
1790 ztest_object_unlock(zd
, lr
->lr_foid
);
1795 zil_replay_func_t ztest_replay_vector
[TX_MAX_TYPE
] = {
1796 NULL
, /* 0 no such transaction type */
1797 (zil_replay_func_t
)ztest_replay_create
, /* TX_CREATE */
1798 NULL
, /* TX_MKDIR */
1799 NULL
, /* TX_MKXATTR */
1800 NULL
, /* TX_SYMLINK */
1801 (zil_replay_func_t
)ztest_replay_remove
, /* TX_REMOVE */
1802 NULL
, /* TX_RMDIR */
1804 NULL
, /* TX_RENAME */
1805 (zil_replay_func_t
)ztest_replay_write
, /* TX_WRITE */
1806 (zil_replay_func_t
)ztest_replay_truncate
, /* TX_TRUNCATE */
1807 (zil_replay_func_t
)ztest_replay_setattr
, /* TX_SETATTR */
1809 NULL
, /* TX_CREATE_ACL */
1810 NULL
, /* TX_CREATE_ATTR */
1811 NULL
, /* TX_CREATE_ACL_ATTR */
1812 NULL
, /* TX_MKDIR_ACL */
1813 NULL
, /* TX_MKDIR_ATTR */
1814 NULL
, /* TX_MKDIR_ACL_ATTR */
1815 NULL
, /* TX_WRITE2 */
1819 * ZIL get_data callbacks
1823 ztest_get_done(zgd_t
*zgd
, int error
)
1825 ztest_ds_t
*zd
= zgd
->zgd_private
;
1826 uint64_t object
= zgd
->zgd_rl
->rl_object
;
1829 dmu_buf_rele(zgd
->zgd_db
, zgd
);
1831 ztest_range_unlock(zgd
->zgd_rl
);
1832 ztest_object_unlock(zd
, object
);
1834 if (error
== 0 && zgd
->zgd_bp
)
1835 zil_add_block(zgd
->zgd_zilog
, zgd
->zgd_bp
);
1837 umem_free(zgd
, sizeof (*zgd
));
1841 ztest_get_data(void *arg
, lr_write_t
*lr
, char *buf
, zio_t
*zio
)
1843 ztest_ds_t
*zd
= arg
;
1844 objset_t
*os
= zd
->zd_os
;
1845 uint64_t object
= lr
->lr_foid
;
1846 uint64_t offset
= lr
->lr_offset
;
1847 uint64_t size
= lr
->lr_length
;
1848 blkptr_t
*bp
= &lr
->lr_blkptr
;
1849 uint64_t txg
= lr
->lr_common
.lrc_txg
;
1851 dmu_object_info_t doi
;
1856 ztest_object_lock(zd
, object
, RL_READER
);
1857 error
= dmu_bonus_hold(os
, object
, FTAG
, &db
);
1859 ztest_object_unlock(zd
, object
);
1863 crtxg
= ztest_bt_bonus(db
)->bt_crtxg
;
1865 if (crtxg
== 0 || crtxg
> txg
) {
1866 dmu_buf_rele(db
, FTAG
);
1867 ztest_object_unlock(zd
, object
);
1871 dmu_object_info_from_db(db
, &doi
);
1872 dmu_buf_rele(db
, FTAG
);
1875 zgd
= umem_zalloc(sizeof (*zgd
), UMEM_NOFAIL
);
1876 zgd
->zgd_zilog
= zd
->zd_zilog
;
1877 zgd
->zgd_private
= zd
;
1879 if (buf
!= NULL
) { /* immediate write */
1880 zgd
->zgd_rl
= ztest_range_lock(zd
, object
, offset
, size
,
1883 error
= dmu_read(os
, object
, offset
, size
, buf
,
1884 DMU_READ_NO_PREFETCH
);
1887 size
= doi
.doi_data_block_size
;
1889 offset
= P2ALIGN(offset
, size
);
1891 ASSERT(offset
< size
);
1895 zgd
->zgd_rl
= ztest_range_lock(zd
, object
, offset
, size
,
1898 error
= dmu_buf_hold(os
, object
, offset
, zgd
, &db
,
1899 DMU_READ_NO_PREFETCH
);
1905 ASSERT(db
->db_offset
== offset
);
1906 ASSERT(db
->db_size
== size
);
1908 error
= dmu_sync(zio
, lr
->lr_common
.lrc_txg
,
1909 ztest_get_done
, zgd
);
1916 ztest_get_done(zgd
, error
);
1922 ztest_lr_alloc(size_t lrsize
, char *name
)
1925 size_t namesize
= name
? strlen(name
) + 1 : 0;
1927 lr
= umem_zalloc(lrsize
+ namesize
, UMEM_NOFAIL
);
1930 bcopy(name
, lr
+ lrsize
, namesize
);
1936 ztest_lr_free(void *lr
, size_t lrsize
, char *name
)
1938 size_t namesize
= name
? strlen(name
) + 1 : 0;
1940 umem_free(lr
, lrsize
+ namesize
);
1944 * Lookup a bunch of objects. Returns the number of objects not found.
1947 ztest_lookup(ztest_ds_t
*zd
, ztest_od_t
*od
, int count
)
1953 ASSERT(mutex_held(&zd
->zd_dirobj_lock
));
1955 for (i
= 0; i
< count
; i
++, od
++) {
1957 error
= zap_lookup(zd
->zd_os
, od
->od_dir
, od
->od_name
,
1958 sizeof (uint64_t), 1, &od
->od_object
);
1960 ASSERT(error
== ENOENT
);
1961 ASSERT(od
->od_object
== 0);
1965 ztest_block_tag_t
*bbt
;
1966 dmu_object_info_t doi
;
1968 ASSERT(od
->od_object
!= 0);
1969 ASSERT(missing
== 0); /* there should be no gaps */
1971 ztest_object_lock(zd
, od
->od_object
, RL_READER
);
1972 VERIFY3U(0, ==, dmu_bonus_hold(zd
->zd_os
,
1973 od
->od_object
, FTAG
, &db
));
1974 dmu_object_info_from_db(db
, &doi
);
1975 bbt
= ztest_bt_bonus(db
);
1976 ASSERT3U(bbt
->bt_magic
, ==, BT_MAGIC
);
1977 od
->od_type
= doi
.doi_type
;
1978 od
->od_blocksize
= doi
.doi_data_block_size
;
1979 od
->od_gen
= bbt
->bt_gen
;
1980 dmu_buf_rele(db
, FTAG
);
1981 ztest_object_unlock(zd
, od
->od_object
);
1989 ztest_create(ztest_ds_t
*zd
, ztest_od_t
*od
, int count
)
1994 ASSERT(mutex_held(&zd
->zd_dirobj_lock
));
1996 for (i
= 0; i
< count
; i
++, od
++) {
2003 lr_create_t
*lr
= ztest_lr_alloc(sizeof (*lr
), od
->od_name
);
2005 lr
->lr_doid
= od
->od_dir
;
2006 lr
->lr_foid
= 0; /* 0 to allocate, > 0 to claim */
2007 lr
->lrz_type
= od
->od_crtype
;
2008 lr
->lrz_blocksize
= od
->od_crblocksize
;
2009 lr
->lrz_ibshift
= ztest_random_ibshift();
2010 lr
->lrz_bonustype
= DMU_OT_UINT64_OTHER
;
2011 lr
->lrz_bonuslen
= dmu_bonus_max();
2012 lr
->lr_gen
= od
->od_crgen
;
2013 lr
->lr_crtime
[0] = time(NULL
);
2015 if (ztest_replay_create(zd
, lr
, B_FALSE
) != 0) {
2016 ASSERT(missing
== 0);
2020 od
->od_object
= lr
->lr_foid
;
2021 od
->od_type
= od
->od_crtype
;
2022 od
->od_blocksize
= od
->od_crblocksize
;
2023 od
->od_gen
= od
->od_crgen
;
2024 ASSERT(od
->od_object
!= 0);
2027 ztest_lr_free(lr
, sizeof (*lr
), od
->od_name
);
2034 ztest_remove(ztest_ds_t
*zd
, ztest_od_t
*od
, int count
)
2040 ASSERT(mutex_held(&zd
->zd_dirobj_lock
));
2044 for (i
= count
- 1; i
>= 0; i
--, od
--) {
2050 if (od
->od_object
== 0)
2053 lr_remove_t
*lr
= ztest_lr_alloc(sizeof (*lr
), od
->od_name
);
2055 lr
->lr_doid
= od
->od_dir
;
2057 if ((error
= ztest_replay_remove(zd
, lr
, B_FALSE
)) != 0) {
2058 ASSERT3U(error
, ==, ENOSPC
);
2063 ztest_lr_free(lr
, sizeof (*lr
), od
->od_name
);
2070 ztest_write(ztest_ds_t
*zd
, uint64_t object
, uint64_t offset
, uint64_t size
,
2076 lr
= ztest_lr_alloc(sizeof (*lr
) + size
, NULL
);
2078 lr
->lr_foid
= object
;
2079 lr
->lr_offset
= offset
;
2080 lr
->lr_length
= size
;
2082 BP_ZERO(&lr
->lr_blkptr
);
2084 bcopy(data
, lr
+ 1, size
);
2086 error
= ztest_replay_write(zd
, lr
, B_FALSE
);
2088 ztest_lr_free(lr
, sizeof (*lr
) + size
, NULL
);
2094 ztest_truncate(ztest_ds_t
*zd
, uint64_t object
, uint64_t offset
, uint64_t size
)
2099 lr
= ztest_lr_alloc(sizeof (*lr
), NULL
);
2101 lr
->lr_foid
= object
;
2102 lr
->lr_offset
= offset
;
2103 lr
->lr_length
= size
;
2105 error
= ztest_replay_truncate(zd
, lr
, B_FALSE
);
2107 ztest_lr_free(lr
, sizeof (*lr
), NULL
);
2113 ztest_setattr(ztest_ds_t
*zd
, uint64_t object
)
2118 lr
= ztest_lr_alloc(sizeof (*lr
), NULL
);
2120 lr
->lr_foid
= object
;
2124 error
= ztest_replay_setattr(zd
, lr
, B_FALSE
);
2126 ztest_lr_free(lr
, sizeof (*lr
), NULL
);
2132 ztest_prealloc(ztest_ds_t
*zd
, uint64_t object
, uint64_t offset
, uint64_t size
)
2134 objset_t
*os
= zd
->zd_os
;
2139 txg_wait_synced(dmu_objset_pool(os
), 0);
2141 ztest_object_lock(zd
, object
, RL_READER
);
2142 rl
= ztest_range_lock(zd
, object
, offset
, size
, RL_WRITER
);
2144 tx
= dmu_tx_create(os
);
2146 dmu_tx_hold_write(tx
, object
, offset
, size
);
2148 txg
= ztest_tx_assign(tx
, TXG_WAIT
, FTAG
);
2151 dmu_prealloc(os
, object
, offset
, size
, tx
);
2153 txg_wait_synced(dmu_objset_pool(os
), txg
);
2155 (void) dmu_free_long_range(os
, object
, offset
, size
);
2158 ztest_range_unlock(rl
);
2159 ztest_object_unlock(zd
, object
);
2163 ztest_io(ztest_ds_t
*zd
, uint64_t object
, uint64_t offset
)
2165 ztest_block_tag_t wbt
;
2166 dmu_object_info_t doi
;
2167 enum ztest_io_type io_type
;
2171 VERIFY(dmu_object_info(zd
->zd_os
, object
, &doi
) == 0);
2172 blocksize
= doi
.doi_data_block_size
;
2173 data
= umem_alloc(blocksize
, UMEM_NOFAIL
);
2176 * Pick an i/o type at random, biased toward writing block tags.
2178 io_type
= ztest_random(ZTEST_IO_TYPES
);
2179 if (ztest_random(2) == 0)
2180 io_type
= ZTEST_IO_WRITE_TAG
;
2182 (void) rw_enter(&zd
->zd_zilog_lock
, RW_READER
);
2186 case ZTEST_IO_WRITE_TAG
:
2187 ztest_bt_generate(&wbt
, zd
->zd_os
, object
, offset
, 0, 0, 0);
2188 (void) ztest_write(zd
, object
, offset
, sizeof (wbt
), &wbt
);
2191 case ZTEST_IO_WRITE_PATTERN
:
2192 (void) memset(data
, 'a' + (object
+ offset
) % 5, blocksize
);
2193 if (ztest_random(2) == 0) {
2195 * Induce fletcher2 collisions to ensure that
2196 * zio_ddt_collision() detects and resolves them
2197 * when using fletcher2-verify for deduplication.
2199 ((uint64_t *)data
)[0] ^= 1ULL << 63;
2200 ((uint64_t *)data
)[4] ^= 1ULL << 63;
2202 (void) ztest_write(zd
, object
, offset
, blocksize
, data
);
2205 case ZTEST_IO_WRITE_ZEROES
:
2206 bzero(data
, blocksize
);
2207 (void) ztest_write(zd
, object
, offset
, blocksize
, data
);
2210 case ZTEST_IO_TRUNCATE
:
2211 (void) ztest_truncate(zd
, object
, offset
, blocksize
);
2214 case ZTEST_IO_SETATTR
:
2215 (void) ztest_setattr(zd
, object
);
2221 (void) rw_exit(&zd
->zd_zilog_lock
);
2223 umem_free(data
, blocksize
);
2227 * Initialize an object description template.
2230 ztest_od_init(ztest_od_t
*od
, uint64_t id
, char *tag
, uint64_t index
,
2231 dmu_object_type_t type
, uint64_t blocksize
, uint64_t gen
)
2233 od
->od_dir
= ZTEST_DIROBJ
;
2236 od
->od_crtype
= type
;
2237 od
->od_crblocksize
= blocksize
? blocksize
: ztest_random_blocksize();
2240 od
->od_type
= DMU_OT_NONE
;
2241 od
->od_blocksize
= 0;
2244 (void) snprintf(od
->od_name
, sizeof (od
->od_name
), "%s(%lld)[%llu]",
2245 tag
, (longlong_t
)id
, (u_longlong_t
)index
);
2249 * Lookup or create the objects for a test using the od template.
2250 * If the objects do not all exist, or if 'remove' is specified,
2251 * remove any existing objects and create new ones. Otherwise,
2252 * use the existing objects.
2255 ztest_object_init(ztest_ds_t
*zd
, ztest_od_t
*od
, size_t size
, boolean_t remove
)
2257 int count
= size
/ sizeof (*od
);
2260 mutex_enter(&zd
->zd_dirobj_lock
);
2261 if ((ztest_lookup(zd
, od
, count
) != 0 || remove
) &&
2262 (ztest_remove(zd
, od
, count
) != 0 ||
2263 ztest_create(zd
, od
, count
) != 0))
2266 mutex_exit(&zd
->zd_dirobj_lock
);
2273 ztest_zil_commit(ztest_ds_t
*zd
, uint64_t id
)
2275 zilog_t
*zilog
= zd
->zd_zilog
;
2277 (void) rw_enter(&zd
->zd_zilog_lock
, RW_READER
);
2279 zil_commit(zilog
, ztest_random(ZTEST_OBJECTS
));
2282 * Remember the committed values in zd, which is in parent/child
2283 * shared memory. If we die, the next iteration of ztest_run()
2284 * will verify that the log really does contain this record.
2286 mutex_enter(&zilog
->zl_lock
);
2287 ASSERT(zd
->zd_shared
!= NULL
);
2288 ASSERT3U(zd
->zd_shared
->zd_seq
, <=, zilog
->zl_commit_lr_seq
);
2289 zd
->zd_shared
->zd_seq
= zilog
->zl_commit_lr_seq
;
2290 mutex_exit(&zilog
->zl_lock
);
2292 (void) rw_exit(&zd
->zd_zilog_lock
);
2296 * This function is designed to simulate the operations that occur during a
2297 * mount/unmount operation. We hold the dataset across these operations in an
2298 * attempt to expose any implicit assumptions about ZIL management.
2302 ztest_zil_remount(ztest_ds_t
*zd
, uint64_t id
)
2304 objset_t
*os
= zd
->zd_os
;
2306 mutex_enter(&zd
->zd_dirobj_lock
);
2307 (void) rw_enter(&zd
->zd_zilog_lock
, RW_WRITER
);
2309 /* zfs_sb_teardown() */
2310 zil_close(zd
->zd_zilog
);
2312 /* zfsvfs_setup() */
2313 VERIFY(zil_open(os
, ztest_get_data
) == zd
->zd_zilog
);
2314 zil_replay(os
, zd
, ztest_replay_vector
);
2316 (void) rw_exit(&zd
->zd_zilog_lock
);
2317 mutex_exit(&zd
->zd_dirobj_lock
);
2321 * Verify that we can't destroy an active pool, create an existing pool,
2322 * or create a pool with a bad vdev spec.
2326 ztest_spa_create_destroy(ztest_ds_t
*zd
, uint64_t id
)
2328 ztest_shared_opts_t
*zo
= &ztest_opts
;
2333 * Attempt to create using a bad file.
2335 nvroot
= make_vdev_root("/dev/bogus", NULL
, NULL
, 0, 0, 0, 0, 0, 1);
2336 VERIFY3U(ENOENT
, ==,
2337 spa_create("ztest_bad_file", nvroot
, NULL
, NULL
, NULL
));
2338 nvlist_free(nvroot
);
2341 * Attempt to create using a bad mirror.
2343 nvroot
= make_vdev_root("/dev/bogus", NULL
, NULL
, 0, 0, 0, 0, 2, 1);
2344 VERIFY3U(ENOENT
, ==,
2345 spa_create("ztest_bad_mirror", nvroot
, NULL
, NULL
, NULL
));
2346 nvlist_free(nvroot
);
2349 * Attempt to create an existing pool. It shouldn't matter
2350 * what's in the nvroot; we should fail with EEXIST.
2352 (void) rw_enter(&ztest_name_lock
, RW_READER
);
2353 nvroot
= make_vdev_root("/dev/bogus", NULL
, NULL
, 0, 0, 0, 0, 0, 1);
2354 VERIFY3U(EEXIST
, ==, spa_create(zo
->zo_pool
, nvroot
, NULL
, NULL
, NULL
));
2355 nvlist_free(nvroot
);
2356 VERIFY3U(0, ==, spa_open(zo
->zo_pool
, &spa
, FTAG
));
2357 VERIFY3U(EBUSY
, ==, spa_destroy(zo
->zo_pool
));
2358 spa_close(spa
, FTAG
);
2360 (void) rw_exit(&ztest_name_lock
);
2365 ztest_spa_upgrade(ztest_ds_t
*zd
, uint64_t id
)
2368 uint64_t initial_version
= SPA_VERSION_INITIAL
;
2369 uint64_t version
, newversion
;
2370 nvlist_t
*nvroot
, *props
;
2373 mutex_enter(&ztest_vdev_lock
);
2374 name
= kmem_asprintf("%s_upgrade", ztest_opts
.zo_pool
);
2377 * Clean up from previous runs.
2379 (void) spa_destroy(name
);
2381 nvroot
= make_vdev_root(NULL
, NULL
, name
, ztest_opts
.zo_vdev_size
, 0,
2382 0, ztest_opts
.zo_raidz
, ztest_opts
.zo_mirrors
, 1);
2385 * If we're configuring a RAIDZ device then make sure that the
2386 * the initial version is capable of supporting that feature.
2388 switch (ztest_opts
.zo_raidz_parity
) {
2391 initial_version
= SPA_VERSION_INITIAL
;
2394 initial_version
= SPA_VERSION_RAIDZ2
;
2397 initial_version
= SPA_VERSION_RAIDZ3
;
2402 * Create a pool with a spa version that can be upgraded. Pick
2403 * a value between initial_version and SPA_VERSION_BEFORE_FEATURES.
2406 version
= ztest_random_spa_version(initial_version
);
2407 } while (version
> SPA_VERSION_BEFORE_FEATURES
);
2409 props
= fnvlist_alloc();
2410 fnvlist_add_uint64(props
,
2411 zpool_prop_to_name(ZPOOL_PROP_VERSION
), version
);
2412 VERIFY3S(spa_create(name
, nvroot
, props
, NULL
, NULL
), ==, 0);
2413 fnvlist_free(nvroot
);
2414 fnvlist_free(props
);
2416 VERIFY3S(spa_open(name
, &spa
, FTAG
), ==, 0);
2417 VERIFY3U(spa_version(spa
), ==, version
);
2418 newversion
= ztest_random_spa_version(version
+ 1);
2420 if (ztest_opts
.zo_verbose
>= 4) {
2421 (void) printf("upgrading spa version from %llu to %llu\n",
2422 (u_longlong_t
)version
, (u_longlong_t
)newversion
);
2425 spa_upgrade(spa
, newversion
);
2426 VERIFY3U(spa_version(spa
), >, version
);
2427 VERIFY3U(spa_version(spa
), ==, fnvlist_lookup_uint64(spa
->spa_config
,
2428 zpool_prop_to_name(ZPOOL_PROP_VERSION
)));
2429 spa_close(spa
, FTAG
);
2432 mutex_exit(&ztest_vdev_lock
);
2436 vdev_lookup_by_path(vdev_t
*vd
, const char *path
)
2441 if (vd
->vdev_path
!= NULL
&& strcmp(path
, vd
->vdev_path
) == 0)
2444 for (c
= 0; c
< vd
->vdev_children
; c
++)
2445 if ((mvd
= vdev_lookup_by_path(vd
->vdev_child
[c
], path
)) !=
2453 * Find the first available hole which can be used as a top-level.
2456 find_vdev_hole(spa_t
*spa
)
2458 vdev_t
*rvd
= spa
->spa_root_vdev
;
2461 ASSERT(spa_config_held(spa
, SCL_VDEV
, RW_READER
) == SCL_VDEV
);
2463 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
2464 vdev_t
*cvd
= rvd
->vdev_child
[c
];
2466 if (cvd
->vdev_ishole
)
2473 * Verify that vdev_add() works as expected.
2477 ztest_vdev_add_remove(ztest_ds_t
*zd
, uint64_t id
)
2479 ztest_shared_t
*zs
= ztest_shared
;
2480 spa_t
*spa
= ztest_spa
;
2486 mutex_enter(&ztest_vdev_lock
);
2488 MAX(zs
->zs_mirrors
+ zs
->zs_splits
, 1) * ztest_opts
.zo_raidz
;
2490 spa_config_enter(spa
, SCL_VDEV
, FTAG
, RW_READER
);
2492 ztest_shared
->zs_vdev_next_leaf
= find_vdev_hole(spa
) * leaves
;
2495 * If we have slogs then remove them 1/4 of the time.
2497 if (spa_has_slogs(spa
) && ztest_random(4) == 0) {
2499 * Grab the guid from the head of the log class rotor.
2501 guid
= spa_log_class(spa
)->mc_rotor
->mg_vd
->vdev_guid
;
2503 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
2506 * We have to grab the zs_name_lock as writer to
2507 * prevent a race between removing a slog (dmu_objset_find)
2508 * and destroying a dataset. Removing the slog will
2509 * grab a reference on the dataset which may cause
2510 * dmu_objset_destroy() to fail with EBUSY thus
2511 * leaving the dataset in an inconsistent state.
2513 rw_enter(&ztest_name_lock
, RW_WRITER
);
2514 error
= spa_vdev_remove(spa
, guid
, B_FALSE
);
2515 rw_exit(&ztest_name_lock
);
2517 if (error
&& error
!= EEXIST
)
2518 fatal(0, "spa_vdev_remove() = %d", error
);
2520 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
2523 * Make 1/4 of the devices be log devices.
2525 nvroot
= make_vdev_root(NULL
, NULL
, NULL
,
2526 ztest_opts
.zo_vdev_size
, 0,
2527 ztest_random(4) == 0, ztest_opts
.zo_raidz
,
2530 error
= spa_vdev_add(spa
, nvroot
);
2531 nvlist_free(nvroot
);
2533 if (error
== ENOSPC
)
2534 ztest_record_enospc("spa_vdev_add");
2535 else if (error
!= 0)
2536 fatal(0, "spa_vdev_add() = %d", error
);
2539 mutex_exit(&ztest_vdev_lock
);
2543 * Verify that adding/removing aux devices (l2arc, hot spare) works as expected.
2547 ztest_vdev_aux_add_remove(ztest_ds_t
*zd
, uint64_t id
)
2549 ztest_shared_t
*zs
= ztest_shared
;
2550 spa_t
*spa
= ztest_spa
;
2551 vdev_t
*rvd
= spa
->spa_root_vdev
;
2552 spa_aux_vdev_t
*sav
;
2558 path
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
2560 if (ztest_random(2) == 0) {
2561 sav
= &spa
->spa_spares
;
2562 aux
= ZPOOL_CONFIG_SPARES
;
2564 sav
= &spa
->spa_l2cache
;
2565 aux
= ZPOOL_CONFIG_L2CACHE
;
2568 mutex_enter(&ztest_vdev_lock
);
2570 spa_config_enter(spa
, SCL_VDEV
, FTAG
, RW_READER
);
2572 if (sav
->sav_count
!= 0 && ztest_random(4) == 0) {
2574 * Pick a random device to remove.
2576 guid
= sav
->sav_vdevs
[ztest_random(sav
->sav_count
)]->vdev_guid
;
2579 * Find an unused device we can add.
2581 zs
->zs_vdev_aux
= 0;
2584 (void) snprintf(path
, MAXPATHLEN
, ztest_aux_template
,
2585 ztest_opts
.zo_dir
, ztest_opts
.zo_pool
, aux
,
2587 for (c
= 0; c
< sav
->sav_count
; c
++)
2588 if (strcmp(sav
->sav_vdevs
[c
]->vdev_path
,
2591 if (c
== sav
->sav_count
&&
2592 vdev_lookup_by_path(rvd
, path
) == NULL
)
2598 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
2604 nvlist_t
*nvroot
= make_vdev_root(NULL
, aux
, NULL
,
2605 (ztest_opts
.zo_vdev_size
* 5) / 4, 0, 0, 0, 0, 1);
2606 error
= spa_vdev_add(spa
, nvroot
);
2608 fatal(0, "spa_vdev_add(%p) = %d", nvroot
, error
);
2609 nvlist_free(nvroot
);
2612 * Remove an existing device. Sometimes, dirty its
2613 * vdev state first to make sure we handle removal
2614 * of devices that have pending state changes.
2616 if (ztest_random(2) == 0)
2617 (void) vdev_online(spa
, guid
, 0, NULL
);
2619 error
= spa_vdev_remove(spa
, guid
, B_FALSE
);
2620 if (error
!= 0 && error
!= EBUSY
)
2621 fatal(0, "spa_vdev_remove(%llu) = %d", guid
, error
);
2624 mutex_exit(&ztest_vdev_lock
);
2626 umem_free(path
, MAXPATHLEN
);
2630 * split a pool if it has mirror tlvdevs
2634 ztest_split_pool(ztest_ds_t
*zd
, uint64_t id
)
2636 ztest_shared_t
*zs
= ztest_shared
;
2637 spa_t
*spa
= ztest_spa
;
2638 vdev_t
*rvd
= spa
->spa_root_vdev
;
2639 nvlist_t
*tree
, **child
, *config
, *split
, **schild
;
2640 uint_t c
, children
, schildren
= 0, lastlogid
= 0;
2643 mutex_enter(&ztest_vdev_lock
);
2645 /* ensure we have a useable config; mirrors of raidz aren't supported */
2646 if (zs
->zs_mirrors
< 3 || ztest_opts
.zo_raidz
> 1) {
2647 mutex_exit(&ztest_vdev_lock
);
2651 /* clean up the old pool, if any */
2652 (void) spa_destroy("splitp");
2654 spa_config_enter(spa
, SCL_VDEV
, FTAG
, RW_READER
);
2656 /* generate a config from the existing config */
2657 mutex_enter(&spa
->spa_props_lock
);
2658 VERIFY(nvlist_lookup_nvlist(spa
->spa_config
, ZPOOL_CONFIG_VDEV_TREE
,
2660 mutex_exit(&spa
->spa_props_lock
);
2662 VERIFY(nvlist_lookup_nvlist_array(tree
, ZPOOL_CONFIG_CHILDREN
, &child
,
2665 schild
= malloc(rvd
->vdev_children
* sizeof (nvlist_t
*));
2666 for (c
= 0; c
< children
; c
++) {
2667 vdev_t
*tvd
= rvd
->vdev_child
[c
];
2671 if (tvd
->vdev_islog
|| tvd
->vdev_ops
== &vdev_hole_ops
) {
2672 VERIFY(nvlist_alloc(&schild
[schildren
], NV_UNIQUE_NAME
,
2674 VERIFY(nvlist_add_string(schild
[schildren
],
2675 ZPOOL_CONFIG_TYPE
, VDEV_TYPE_HOLE
) == 0);
2676 VERIFY(nvlist_add_uint64(schild
[schildren
],
2677 ZPOOL_CONFIG_IS_HOLE
, 1) == 0);
2679 lastlogid
= schildren
;
2684 VERIFY(nvlist_lookup_nvlist_array(child
[c
],
2685 ZPOOL_CONFIG_CHILDREN
, &mchild
, &mchildren
) == 0);
2686 VERIFY(nvlist_dup(mchild
[0], &schild
[schildren
++], 0) == 0);
2689 /* OK, create a config that can be used to split */
2690 VERIFY(nvlist_alloc(&split
, NV_UNIQUE_NAME
, 0) == 0);
2691 VERIFY(nvlist_add_string(split
, ZPOOL_CONFIG_TYPE
,
2692 VDEV_TYPE_ROOT
) == 0);
2693 VERIFY(nvlist_add_nvlist_array(split
, ZPOOL_CONFIG_CHILDREN
, schild
,
2694 lastlogid
!= 0 ? lastlogid
: schildren
) == 0);
2696 VERIFY(nvlist_alloc(&config
, NV_UNIQUE_NAME
, 0) == 0);
2697 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, split
) == 0);
2699 for (c
= 0; c
< schildren
; c
++)
2700 nvlist_free(schild
[c
]);
2704 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
2706 (void) rw_enter(&ztest_name_lock
, RW_WRITER
);
2707 error
= spa_vdev_split_mirror(spa
, "splitp", config
, NULL
, B_FALSE
);
2708 (void) rw_exit(&ztest_name_lock
);
2710 nvlist_free(config
);
2713 (void) printf("successful split - results:\n");
2714 mutex_enter(&spa_namespace_lock
);
2715 show_pool_stats(spa
);
2716 show_pool_stats(spa_lookup("splitp"));
2717 mutex_exit(&spa_namespace_lock
);
2721 mutex_exit(&ztest_vdev_lock
);
2726 * Verify that we can attach and detach devices.
2730 ztest_vdev_attach_detach(ztest_ds_t
*zd
, uint64_t id
)
2732 ztest_shared_t
*zs
= ztest_shared
;
2733 spa_t
*spa
= ztest_spa
;
2734 spa_aux_vdev_t
*sav
= &spa
->spa_spares
;
2735 vdev_t
*rvd
= spa
->spa_root_vdev
;
2736 vdev_t
*oldvd
, *newvd
, *pvd
;
2740 uint64_t ashift
= ztest_get_ashift();
2741 uint64_t oldguid
, pguid
;
2742 size_t oldsize
, newsize
;
2743 char *oldpath
, *newpath
;
2745 int oldvd_has_siblings
= B_FALSE
;
2746 int newvd_is_spare
= B_FALSE
;
2748 int error
, expected_error
;
2750 oldpath
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
2751 newpath
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
2753 mutex_enter(&ztest_vdev_lock
);
2754 leaves
= MAX(zs
->zs_mirrors
, 1) * ztest_opts
.zo_raidz
;
2756 spa_config_enter(spa
, SCL_VDEV
, FTAG
, RW_READER
);
2759 * Decide whether to do an attach or a replace.
2761 replacing
= ztest_random(2);
2764 * Pick a random top-level vdev.
2766 top
= ztest_random_vdev_top(spa
, B_TRUE
);
2769 * Pick a random leaf within it.
2771 leaf
= ztest_random(leaves
);
2776 oldvd
= rvd
->vdev_child
[top
];
2777 if (zs
->zs_mirrors
>= 1) {
2778 ASSERT(oldvd
->vdev_ops
== &vdev_mirror_ops
);
2779 ASSERT(oldvd
->vdev_children
>= zs
->zs_mirrors
);
2780 oldvd
= oldvd
->vdev_child
[leaf
/ ztest_opts
.zo_raidz
];
2782 if (ztest_opts
.zo_raidz
> 1) {
2783 ASSERT(oldvd
->vdev_ops
== &vdev_raidz_ops
);
2784 ASSERT(oldvd
->vdev_children
== ztest_opts
.zo_raidz
);
2785 oldvd
= oldvd
->vdev_child
[leaf
% ztest_opts
.zo_raidz
];
2789 * If we're already doing an attach or replace, oldvd may be a
2790 * mirror vdev -- in which case, pick a random child.
2792 while (oldvd
->vdev_children
!= 0) {
2793 oldvd_has_siblings
= B_TRUE
;
2794 ASSERT(oldvd
->vdev_children
>= 2);
2795 oldvd
= oldvd
->vdev_child
[ztest_random(oldvd
->vdev_children
)];
2798 oldguid
= oldvd
->vdev_guid
;
2799 oldsize
= vdev_get_min_asize(oldvd
);
2800 oldvd_is_log
= oldvd
->vdev_top
->vdev_islog
;
2801 (void) strcpy(oldpath
, oldvd
->vdev_path
);
2802 pvd
= oldvd
->vdev_parent
;
2803 pguid
= pvd
->vdev_guid
;
2806 * If oldvd has siblings, then half of the time, detach it.
2808 if (oldvd_has_siblings
&& ztest_random(2) == 0) {
2809 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
2810 error
= spa_vdev_detach(spa
, oldguid
, pguid
, B_FALSE
);
2811 if (error
!= 0 && error
!= ENODEV
&& error
!= EBUSY
&&
2813 fatal(0, "detach (%s) returned %d", oldpath
, error
);
2818 * For the new vdev, choose with equal probability between the two
2819 * standard paths (ending in either 'a' or 'b') or a random hot spare.
2821 if (sav
->sav_count
!= 0 && ztest_random(3) == 0) {
2822 newvd
= sav
->sav_vdevs
[ztest_random(sav
->sav_count
)];
2823 newvd_is_spare
= B_TRUE
;
2824 (void) strcpy(newpath
, newvd
->vdev_path
);
2826 (void) snprintf(newpath
, MAXPATHLEN
, ztest_dev_template
,
2827 ztest_opts
.zo_dir
, ztest_opts
.zo_pool
,
2828 top
* leaves
+ leaf
);
2829 if (ztest_random(2) == 0)
2830 newpath
[strlen(newpath
) - 1] = 'b';
2831 newvd
= vdev_lookup_by_path(rvd
, newpath
);
2835 newsize
= vdev_get_min_asize(newvd
);
2838 * Make newsize a little bigger or smaller than oldsize.
2839 * If it's smaller, the attach should fail.
2840 * If it's larger, and we're doing a replace,
2841 * we should get dynamic LUN growth when we're done.
2843 newsize
= 10 * oldsize
/ (9 + ztest_random(3));
2847 * If pvd is not a mirror or root, the attach should fail with ENOTSUP,
2848 * unless it's a replace; in that case any non-replacing parent is OK.
2850 * If newvd is already part of the pool, it should fail with EBUSY.
2852 * If newvd is too small, it should fail with EOVERFLOW.
2854 if (pvd
->vdev_ops
!= &vdev_mirror_ops
&&
2855 pvd
->vdev_ops
!= &vdev_root_ops
&& (!replacing
||
2856 pvd
->vdev_ops
== &vdev_replacing_ops
||
2857 pvd
->vdev_ops
== &vdev_spare_ops
))
2858 expected_error
= ENOTSUP
;
2859 else if (newvd_is_spare
&& (!replacing
|| oldvd_is_log
))
2860 expected_error
= ENOTSUP
;
2861 else if (newvd
== oldvd
)
2862 expected_error
= replacing
? 0 : EBUSY
;
2863 else if (vdev_lookup_by_path(rvd
, newpath
) != NULL
)
2864 expected_error
= EBUSY
;
2865 else if (newsize
< oldsize
)
2866 expected_error
= EOVERFLOW
;
2867 else if (ashift
> oldvd
->vdev_top
->vdev_ashift
)
2868 expected_error
= EDOM
;
2872 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
2875 * Build the nvlist describing newpath.
2877 root
= make_vdev_root(newpath
, NULL
, NULL
, newvd
== NULL
? newsize
: 0,
2878 ashift
, 0, 0, 0, 1);
2880 error
= spa_vdev_attach(spa
, oldguid
, root
, replacing
);
2885 * If our parent was the replacing vdev, but the replace completed,
2886 * then instead of failing with ENOTSUP we may either succeed,
2887 * fail with ENODEV, or fail with EOVERFLOW.
2889 if (expected_error
== ENOTSUP
&&
2890 (error
== 0 || error
== ENODEV
|| error
== EOVERFLOW
))
2891 expected_error
= error
;
2894 * If someone grew the LUN, the replacement may be too small.
2896 if (error
== EOVERFLOW
|| error
== EBUSY
)
2897 expected_error
= error
;
2899 /* XXX workaround 6690467 */
2900 if (error
!= expected_error
&& expected_error
!= EBUSY
) {
2901 fatal(0, "attach (%s %llu, %s %llu, %d) "
2902 "returned %d, expected %d",
2903 oldpath
, (longlong_t
)oldsize
, newpath
,
2904 (longlong_t
)newsize
, replacing
, error
, expected_error
);
2907 mutex_exit(&ztest_vdev_lock
);
2909 umem_free(oldpath
, MAXPATHLEN
);
2910 umem_free(newpath
, MAXPATHLEN
);
2914 * Callback function which expands the physical size of the vdev.
2917 grow_vdev(vdev_t
*vd
, void *arg
)
2919 ASSERTV(spa_t
*spa
= vd
->vdev_spa
);
2920 size_t *newsize
= arg
;
2924 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_READER
) == SCL_STATE
);
2925 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
2927 if ((fd
= open(vd
->vdev_path
, O_RDWR
)) == -1)
2930 fsize
= lseek(fd
, 0, SEEK_END
);
2931 VERIFY(ftruncate(fd
, *newsize
) == 0);
2933 if (ztest_opts
.zo_verbose
>= 6) {
2934 (void) printf("%s grew from %lu to %lu bytes\n",
2935 vd
->vdev_path
, (ulong_t
)fsize
, (ulong_t
)*newsize
);
2942 * Callback function which expands a given vdev by calling vdev_online().
2946 online_vdev(vdev_t
*vd
, void *arg
)
2948 spa_t
*spa
= vd
->vdev_spa
;
2949 vdev_t
*tvd
= vd
->vdev_top
;
2950 uint64_t guid
= vd
->vdev_guid
;
2951 uint64_t generation
= spa
->spa_config_generation
+ 1;
2952 vdev_state_t newstate
= VDEV_STATE_UNKNOWN
;
2955 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_READER
) == SCL_STATE
);
2956 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
2958 /* Calling vdev_online will initialize the new metaslabs */
2959 spa_config_exit(spa
, SCL_STATE
, spa
);
2960 error
= vdev_online(spa
, guid
, ZFS_ONLINE_EXPAND
, &newstate
);
2961 spa_config_enter(spa
, SCL_STATE
, spa
, RW_READER
);
2964 * If vdev_online returned an error or the underlying vdev_open
2965 * failed then we abort the expand. The only way to know that
2966 * vdev_open fails is by checking the returned newstate.
2968 if (error
|| newstate
!= VDEV_STATE_HEALTHY
) {
2969 if (ztest_opts
.zo_verbose
>= 5) {
2970 (void) printf("Unable to expand vdev, state %llu, "
2971 "error %d\n", (u_longlong_t
)newstate
, error
);
2975 ASSERT3U(newstate
, ==, VDEV_STATE_HEALTHY
);
2978 * Since we dropped the lock we need to ensure that we're
2979 * still talking to the original vdev. It's possible this
2980 * vdev may have been detached/replaced while we were
2981 * trying to online it.
2983 if (generation
!= spa
->spa_config_generation
) {
2984 if (ztest_opts
.zo_verbose
>= 5) {
2985 (void) printf("vdev configuration has changed, "
2986 "guid %llu, state %llu, expected gen %llu, "
2989 (u_longlong_t
)tvd
->vdev_state
,
2990 (u_longlong_t
)generation
,
2991 (u_longlong_t
)spa
->spa_config_generation
);
2999 * Traverse the vdev tree calling the supplied function.
3000 * We continue to walk the tree until we either have walked all
3001 * children or we receive a non-NULL return from the callback.
3002 * If a NULL callback is passed, then we just return back the first
3003 * leaf vdev we encounter.
3006 vdev_walk_tree(vdev_t
*vd
, vdev_t
*(*func
)(vdev_t
*, void *), void *arg
)
3010 if (vd
->vdev_ops
->vdev_op_leaf
) {
3014 return (func(vd
, arg
));
3017 for (c
= 0; c
< vd
->vdev_children
; c
++) {
3018 vdev_t
*cvd
= vd
->vdev_child
[c
];
3019 if ((cvd
= vdev_walk_tree(cvd
, func
, arg
)) != NULL
)
3026 * Verify that dynamic LUN growth works as expected.
3030 ztest_vdev_LUN_growth(ztest_ds_t
*zd
, uint64_t id
)
3032 spa_t
*spa
= ztest_spa
;
3034 metaslab_class_t
*mc
;
3035 metaslab_group_t
*mg
;
3036 size_t psize
, newsize
;
3038 uint64_t old_class_space
, new_class_space
, old_ms_count
, new_ms_count
;
3040 mutex_enter(&ztest_vdev_lock
);
3041 spa_config_enter(spa
, SCL_STATE
, spa
, RW_READER
);
3043 top
= ztest_random_vdev_top(spa
, B_TRUE
);
3045 tvd
= spa
->spa_root_vdev
->vdev_child
[top
];
3048 old_ms_count
= tvd
->vdev_ms_count
;
3049 old_class_space
= metaslab_class_get_space(mc
);
3052 * Determine the size of the first leaf vdev associated with
3053 * our top-level device.
3055 vd
= vdev_walk_tree(tvd
, NULL
, NULL
);
3056 ASSERT3P(vd
, !=, NULL
);
3057 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
3059 psize
= vd
->vdev_psize
;
3062 * We only try to expand the vdev if it's healthy, less than 4x its
3063 * original size, and it has a valid psize.
3065 if (tvd
->vdev_state
!= VDEV_STATE_HEALTHY
||
3066 psize
== 0 || psize
>= 4 * ztest_opts
.zo_vdev_size
) {
3067 spa_config_exit(spa
, SCL_STATE
, spa
);
3068 mutex_exit(&ztest_vdev_lock
);
3072 newsize
= psize
+ psize
/ 8;
3073 ASSERT3U(newsize
, >, psize
);
3075 if (ztest_opts
.zo_verbose
>= 6) {
3076 (void) printf("Expanding LUN %s from %lu to %lu\n",
3077 vd
->vdev_path
, (ulong_t
)psize
, (ulong_t
)newsize
);
3081 * Growing the vdev is a two step process:
3082 * 1). expand the physical size (i.e. relabel)
3083 * 2). online the vdev to create the new metaslabs
3085 if (vdev_walk_tree(tvd
, grow_vdev
, &newsize
) != NULL
||
3086 vdev_walk_tree(tvd
, online_vdev
, NULL
) != NULL
||
3087 tvd
->vdev_state
!= VDEV_STATE_HEALTHY
) {
3088 if (ztest_opts
.zo_verbose
>= 5) {
3089 (void) printf("Could not expand LUN because "
3090 "the vdev configuration changed.\n");
3092 spa_config_exit(spa
, SCL_STATE
, spa
);
3093 mutex_exit(&ztest_vdev_lock
);
3097 spa_config_exit(spa
, SCL_STATE
, spa
);
3100 * Expanding the LUN will update the config asynchronously,
3101 * thus we must wait for the async thread to complete any
3102 * pending tasks before proceeding.
3106 mutex_enter(&spa
->spa_async_lock
);
3107 done
= (spa
->spa_async_thread
== NULL
&& !spa
->spa_async_tasks
);
3108 mutex_exit(&spa
->spa_async_lock
);
3111 txg_wait_synced(spa_get_dsl(spa
), 0);
3112 (void) poll(NULL
, 0, 100);
3115 spa_config_enter(spa
, SCL_STATE
, spa
, RW_READER
);
3117 tvd
= spa
->spa_root_vdev
->vdev_child
[top
];
3118 new_ms_count
= tvd
->vdev_ms_count
;
3119 new_class_space
= metaslab_class_get_space(mc
);
3121 if (tvd
->vdev_mg
!= mg
|| mg
->mg_class
!= mc
) {
3122 if (ztest_opts
.zo_verbose
>= 5) {
3123 (void) printf("Could not verify LUN expansion due to "
3124 "intervening vdev offline or remove.\n");
3126 spa_config_exit(spa
, SCL_STATE
, spa
);
3127 mutex_exit(&ztest_vdev_lock
);
3132 * Make sure we were able to grow the vdev.
3134 if (new_ms_count
<= old_ms_count
)
3135 fatal(0, "LUN expansion failed: ms_count %llu <= %llu\n",
3136 old_ms_count
, new_ms_count
);
3139 * Make sure we were able to grow the pool.
3141 if (new_class_space
<= old_class_space
)
3142 fatal(0, "LUN expansion failed: class_space %llu <= %llu\n",
3143 old_class_space
, new_class_space
);
3145 if (ztest_opts
.zo_verbose
>= 5) {
3146 char oldnumbuf
[6], newnumbuf
[6];
3148 nicenum(old_class_space
, oldnumbuf
);
3149 nicenum(new_class_space
, newnumbuf
);
3150 (void) printf("%s grew from %s to %s\n",
3151 spa
->spa_name
, oldnumbuf
, newnumbuf
);
3154 spa_config_exit(spa
, SCL_STATE
, spa
);
3155 mutex_exit(&ztest_vdev_lock
);
3159 * Verify that dmu_objset_{create,destroy,open,close} work as expected.
3163 ztest_objset_create_cb(objset_t
*os
, void *arg
, cred_t
*cr
, dmu_tx_t
*tx
)
3166 * Create the objects common to all ztest datasets.
3168 VERIFY(zap_create_claim(os
, ZTEST_DIROBJ
,
3169 DMU_OT_ZAP_OTHER
, DMU_OT_NONE
, 0, tx
) == 0);
3173 ztest_dataset_create(char *dsname
)
3175 uint64_t zilset
= ztest_random(100);
3176 int err
= dmu_objset_create(dsname
, DMU_OST_OTHER
, 0,
3177 ztest_objset_create_cb
, NULL
);
3179 if (err
|| zilset
< 80)
3182 if (ztest_opts
.zo_verbose
>= 5)
3183 (void) printf("Setting dataset %s to sync always\n", dsname
);
3184 return (ztest_dsl_prop_set_uint64(dsname
, ZFS_PROP_SYNC
,
3185 ZFS_SYNC_ALWAYS
, B_FALSE
));
3190 ztest_objset_destroy_cb(const char *name
, void *arg
)
3193 dmu_object_info_t doi
;
3197 * Verify that the dataset contains a directory object.
3199 VERIFY3U(0, ==, dmu_objset_hold(name
, FTAG
, &os
));
3200 error
= dmu_object_info(os
, ZTEST_DIROBJ
, &doi
);
3201 if (error
!= ENOENT
) {
3202 /* We could have crashed in the middle of destroying it */
3204 ASSERT3U(doi
.doi_type
, ==, DMU_OT_ZAP_OTHER
);
3205 ASSERT3S(doi
.doi_physical_blocks_512
, >=, 0);
3207 dmu_objset_rele(os
, FTAG
);
3210 * Destroy the dataset.
3212 VERIFY3U(0, ==, dmu_objset_destroy(name
, B_FALSE
));
3217 ztest_snapshot_create(char *osname
, uint64_t id
)
3219 char snapname
[MAXNAMELEN
];
3222 (void) snprintf(snapname
, MAXNAMELEN
, "%s@%llu", osname
,
3225 error
= dmu_objset_snapshot(osname
, strchr(snapname
, '@') + 1,
3226 NULL
, NULL
, B_FALSE
, B_FALSE
, -1);
3227 if (error
== ENOSPC
) {
3228 ztest_record_enospc(FTAG
);
3231 if (error
!= 0 && error
!= EEXIST
)
3232 fatal(0, "ztest_snapshot_create(%s) = %d", snapname
, error
);
3237 ztest_snapshot_destroy(char *osname
, uint64_t id
)
3239 char snapname
[MAXNAMELEN
];
3242 (void) snprintf(snapname
, MAXNAMELEN
, "%s@%llu", osname
,
3245 error
= dmu_objset_destroy(snapname
, B_FALSE
);
3246 if (error
!= 0 && error
!= ENOENT
)
3247 fatal(0, "ztest_snapshot_destroy(%s) = %d", snapname
, error
);
3253 ztest_dmu_objset_create_destroy(ztest_ds_t
*zd
, uint64_t id
)
3263 zdtmp
= umem_alloc(sizeof (ztest_ds_t
), UMEM_NOFAIL
);
3264 name
= umem_alloc(MAXNAMELEN
, UMEM_NOFAIL
);
3266 (void) rw_enter(&ztest_name_lock
, RW_READER
);
3268 (void) snprintf(name
, MAXNAMELEN
, "%s/temp_%llu",
3269 ztest_opts
.zo_pool
, (u_longlong_t
)id
);
3272 * If this dataset exists from a previous run, process its replay log
3273 * half of the time. If we don't replay it, then dmu_objset_destroy()
3274 * (invoked from ztest_objset_destroy_cb()) should just throw it away.
3276 if (ztest_random(2) == 0 &&
3277 dmu_objset_own(name
, DMU_OST_OTHER
, B_FALSE
, FTAG
, &os
) == 0) {
3278 ztest_zd_init(zdtmp
, NULL
, os
);
3279 zil_replay(os
, zdtmp
, ztest_replay_vector
);
3280 ztest_zd_fini(zdtmp
);
3281 dmu_objset_disown(os
, FTAG
);
3285 * There may be an old instance of the dataset we're about to
3286 * create lying around from a previous run. If so, destroy it
3287 * and all of its snapshots.
3289 (void) dmu_objset_find(name
, ztest_objset_destroy_cb
, NULL
,
3290 DS_FIND_CHILDREN
| DS_FIND_SNAPSHOTS
);
3293 * Verify that the destroyed dataset is no longer in the namespace.
3295 VERIFY3U(ENOENT
, ==, dmu_objset_hold(name
, FTAG
, &os
));
3298 * Verify that we can create a new dataset.
3300 error
= ztest_dataset_create(name
);
3302 if (error
== ENOSPC
) {
3303 ztest_record_enospc(FTAG
);
3306 fatal(0, "dmu_objset_create(%s) = %d", name
, error
);
3310 dmu_objset_own(name
, DMU_OST_OTHER
, B_FALSE
, FTAG
, &os
));
3312 ztest_zd_init(zdtmp
, NULL
, os
);
3315 * Open the intent log for it.
3317 zilog
= zil_open(os
, ztest_get_data
);
3320 * Put some objects in there, do a little I/O to them,
3321 * and randomly take a couple of snapshots along the way.
3323 iters
= ztest_random(5);
3324 for (i
= 0; i
< iters
; i
++) {
3325 ztest_dmu_object_alloc_free(zdtmp
, id
);
3326 if (ztest_random(iters
) == 0)
3327 (void) ztest_snapshot_create(name
, i
);
3331 * Verify that we cannot create an existing dataset.
3333 VERIFY3U(EEXIST
, ==,
3334 dmu_objset_create(name
, DMU_OST_OTHER
, 0, NULL
, NULL
));
3337 * Verify that we can hold an objset that is also owned.
3339 VERIFY3U(0, ==, dmu_objset_hold(name
, FTAG
, &os2
));
3340 dmu_objset_rele(os2
, FTAG
);
3343 * Verify that we cannot own an objset that is already owned.
3346 dmu_objset_own(name
, DMU_OST_OTHER
, B_FALSE
, FTAG
, &os2
));
3349 dmu_objset_disown(os
, FTAG
);
3350 ztest_zd_fini(zdtmp
);
3352 (void) rw_exit(&ztest_name_lock
);
3354 umem_free(name
, MAXNAMELEN
);
3355 umem_free(zdtmp
, sizeof (ztest_ds_t
));
3359 * Verify that dmu_snapshot_{create,destroy,open,close} work as expected.
3362 ztest_dmu_snapshot_create_destroy(ztest_ds_t
*zd
, uint64_t id
)
3364 (void) rw_enter(&ztest_name_lock
, RW_READER
);
3365 (void) ztest_snapshot_destroy(zd
->zd_name
, id
);
3366 (void) ztest_snapshot_create(zd
->zd_name
, id
);
3367 (void) rw_exit(&ztest_name_lock
);
3371 * Cleanup non-standard snapshots and clones.
3374 ztest_dsl_dataset_cleanup(char *osname
, uint64_t id
)
3383 snap1name
= umem_alloc(MAXNAMELEN
, UMEM_NOFAIL
);
3384 clone1name
= umem_alloc(MAXNAMELEN
, UMEM_NOFAIL
);
3385 snap2name
= umem_alloc(MAXNAMELEN
, UMEM_NOFAIL
);
3386 clone2name
= umem_alloc(MAXNAMELEN
, UMEM_NOFAIL
);
3387 snap3name
= umem_alloc(MAXNAMELEN
, UMEM_NOFAIL
);
3389 (void) snprintf(snap1name
, MAXNAMELEN
, "%s@s1_%llu",
3390 osname
, (u_longlong_t
)id
);
3391 (void) snprintf(clone1name
, MAXNAMELEN
, "%s/c1_%llu",
3392 osname
, (u_longlong_t
)id
);
3393 (void) snprintf(snap2name
, MAXNAMELEN
, "%s@s2_%llu",
3394 clone1name
, (u_longlong_t
)id
);
3395 (void) snprintf(clone2name
, MAXNAMELEN
, "%s/c2_%llu",
3396 osname
, (u_longlong_t
)id
);
3397 (void) snprintf(snap3name
, MAXNAMELEN
, "%s@s3_%llu",
3398 clone1name
, (u_longlong_t
)id
);
3400 error
= dmu_objset_destroy(clone2name
, B_FALSE
);
3401 if (error
&& error
!= ENOENT
)
3402 fatal(0, "dmu_objset_destroy(%s) = %d", clone2name
, error
);
3403 error
= dmu_objset_destroy(snap3name
, B_FALSE
);
3404 if (error
&& error
!= ENOENT
)
3405 fatal(0, "dmu_objset_destroy(%s) = %d", snap3name
, error
);
3406 error
= dmu_objset_destroy(snap2name
, B_FALSE
);
3407 if (error
&& error
!= ENOENT
)
3408 fatal(0, "dmu_objset_destroy(%s) = %d", snap2name
, error
);
3409 error
= dmu_objset_destroy(clone1name
, B_FALSE
);
3410 if (error
&& error
!= ENOENT
)
3411 fatal(0, "dmu_objset_destroy(%s) = %d", clone1name
, error
);
3412 error
= dmu_objset_destroy(snap1name
, B_FALSE
);
3413 if (error
&& error
!= ENOENT
)
3414 fatal(0, "dmu_objset_destroy(%s) = %d", snap1name
, error
);
3416 umem_free(snap1name
, MAXNAMELEN
);
3417 umem_free(clone1name
, MAXNAMELEN
);
3418 umem_free(snap2name
, MAXNAMELEN
);
3419 umem_free(clone2name
, MAXNAMELEN
);
3420 umem_free(snap3name
, MAXNAMELEN
);
3424 * Verify dsl_dataset_promote handles EBUSY
3427 ztest_dsl_dataset_promote_busy(ztest_ds_t
*zd
, uint64_t id
)
3436 char *osname
= zd
->zd_name
;
3439 snap1name
= umem_alloc(MAXNAMELEN
, UMEM_NOFAIL
);
3440 clone1name
= umem_alloc(MAXNAMELEN
, UMEM_NOFAIL
);
3441 snap2name
= umem_alloc(MAXNAMELEN
, UMEM_NOFAIL
);
3442 clone2name
= umem_alloc(MAXNAMELEN
, UMEM_NOFAIL
);
3443 snap3name
= umem_alloc(MAXNAMELEN
, UMEM_NOFAIL
);
3445 (void) rw_enter(&ztest_name_lock
, RW_READER
);
3447 ztest_dsl_dataset_cleanup(osname
, id
);
3449 (void) snprintf(snap1name
, MAXNAMELEN
, "%s@s1_%llu",
3450 osname
, (u_longlong_t
)id
);
3451 (void) snprintf(clone1name
, MAXNAMELEN
, "%s/c1_%llu",
3452 osname
, (u_longlong_t
)id
);
3453 (void) snprintf(snap2name
, MAXNAMELEN
, "%s@s2_%llu",
3454 clone1name
, (u_longlong_t
)id
);
3455 (void) snprintf(clone2name
, MAXNAMELEN
, "%s/c2_%llu",
3456 osname
, (u_longlong_t
)id
);
3457 (void) snprintf(snap3name
, MAXNAMELEN
, "%s@s3_%llu",
3458 clone1name
, (u_longlong_t
)id
);
3460 error
= dmu_objset_snapshot(osname
, strchr(snap1name
, '@')+1,
3461 NULL
, NULL
, B_FALSE
, B_FALSE
, -1);
3462 if (error
&& error
!= EEXIST
) {
3463 if (error
== ENOSPC
) {
3464 ztest_record_enospc(FTAG
);
3467 fatal(0, "dmu_take_snapshot(%s) = %d", snap1name
, error
);
3470 error
= dmu_objset_hold(snap1name
, FTAG
, &clone
);
3472 fatal(0, "dmu_open_snapshot(%s) = %d", snap1name
, error
);
3474 error
= dmu_objset_clone(clone1name
, dmu_objset_ds(clone
), 0);
3475 dmu_objset_rele(clone
, FTAG
);
3477 if (error
== ENOSPC
) {
3478 ztest_record_enospc(FTAG
);
3481 fatal(0, "dmu_objset_create(%s) = %d", clone1name
, error
);
3484 error
= dmu_objset_snapshot(clone1name
, strchr(snap2name
, '@')+1,
3485 NULL
, NULL
, B_FALSE
, B_FALSE
, -1);
3486 if (error
&& error
!= EEXIST
) {
3487 if (error
== ENOSPC
) {
3488 ztest_record_enospc(FTAG
);
3491 fatal(0, "dmu_open_snapshot(%s) = %d", snap2name
, error
);
3494 error
= dmu_objset_snapshot(clone1name
, strchr(snap3name
, '@')+1,
3495 NULL
, NULL
, B_FALSE
, B_FALSE
, -1);
3496 if (error
&& error
!= EEXIST
) {
3497 if (error
== ENOSPC
) {
3498 ztest_record_enospc(FTAG
);
3501 fatal(0, "dmu_open_snapshot(%s) = %d", snap3name
, error
);
3504 error
= dmu_objset_hold(snap3name
, FTAG
, &clone
);
3506 fatal(0, "dmu_open_snapshot(%s) = %d", snap3name
, error
);
3508 error
= dmu_objset_clone(clone2name
, dmu_objset_ds(clone
), 0);
3509 dmu_objset_rele(clone
, FTAG
);
3511 if (error
== ENOSPC
) {
3512 ztest_record_enospc(FTAG
);
3515 fatal(0, "dmu_objset_create(%s) = %d", clone2name
, error
);
3518 error
= dsl_dataset_own(snap2name
, B_FALSE
, FTAG
, &ds
);
3520 fatal(0, "dsl_dataset_own(%s) = %d", snap2name
, error
);
3521 error
= dsl_dataset_promote(clone2name
, NULL
);
3523 fatal(0, "dsl_dataset_promote(%s), %d, not EBUSY", clone2name
,
3525 dsl_dataset_disown(ds
, FTAG
);
3528 ztest_dsl_dataset_cleanup(osname
, id
);
3530 (void) rw_exit(&ztest_name_lock
);
3532 umem_free(snap1name
, MAXNAMELEN
);
3533 umem_free(clone1name
, MAXNAMELEN
);
3534 umem_free(snap2name
, MAXNAMELEN
);
3535 umem_free(clone2name
, MAXNAMELEN
);
3536 umem_free(snap3name
, MAXNAMELEN
);
3539 #undef OD_ARRAY_SIZE
3540 #define OD_ARRAY_SIZE 4
3543 * Verify that dmu_object_{alloc,free} work as expected.
3546 ztest_dmu_object_alloc_free(ztest_ds_t
*zd
, uint64_t id
)
3553 size
= sizeof(ztest_od_t
) * OD_ARRAY_SIZE
;
3554 od
= umem_alloc(size
, UMEM_NOFAIL
);
3555 batchsize
= OD_ARRAY_SIZE
;
3557 for (b
= 0; b
< batchsize
; b
++)
3558 ztest_od_init(od
+ b
, id
, FTAG
, b
, DMU_OT_UINT64_OTHER
, 0, 0);
3561 * Destroy the previous batch of objects, create a new batch,
3562 * and do some I/O on the new objects.
3564 if (ztest_object_init(zd
, od
, size
, B_TRUE
) != 0)
3567 while (ztest_random(4 * batchsize
) != 0)
3568 ztest_io(zd
, od
[ztest_random(batchsize
)].od_object
,
3569 ztest_random(ZTEST_RANGE_LOCKS
) << SPA_MAXBLOCKSHIFT
);
3571 umem_free(od
, size
);
3574 #undef OD_ARRAY_SIZE
3575 #define OD_ARRAY_SIZE 2
3578 * Verify that dmu_{read,write} work as expected.
3581 ztest_dmu_read_write(ztest_ds_t
*zd
, uint64_t id
)
3586 objset_t
*os
= zd
->zd_os
;
3587 size
= sizeof(ztest_od_t
) * OD_ARRAY_SIZE
;
3588 od
= umem_alloc(size
, UMEM_NOFAIL
);
3590 int i
, freeit
, error
;
3592 bufwad_t
*packbuf
, *bigbuf
, *pack
, *bigH
, *bigT
;
3593 uint64_t packobj
, packoff
, packsize
, bigobj
, bigoff
, bigsize
;
3594 uint64_t chunksize
= (1000 + ztest_random(1000)) * sizeof (uint64_t);
3595 uint64_t regions
= 997;
3596 uint64_t stride
= 123456789ULL;
3597 uint64_t width
= 40;
3598 int free_percent
= 5;
3601 * This test uses two objects, packobj and bigobj, that are always
3602 * updated together (i.e. in the same tx) so that their contents are
3603 * in sync and can be compared. Their contents relate to each other
3604 * in a simple way: packobj is a dense array of 'bufwad' structures,
3605 * while bigobj is a sparse array of the same bufwads. Specifically,
3606 * for any index n, there are three bufwads that should be identical:
3608 * packobj, at offset n * sizeof (bufwad_t)
3609 * bigobj, at the head of the nth chunk
3610 * bigobj, at the tail of the nth chunk
3612 * The chunk size is arbitrary. It doesn't have to be a power of two,
3613 * and it doesn't have any relation to the object blocksize.
3614 * The only requirement is that it can hold at least two bufwads.
3616 * Normally, we write the bufwad to each of these locations.
3617 * However, free_percent of the time we instead write zeroes to
3618 * packobj and perform a dmu_free_range() on bigobj. By comparing
3619 * bigobj to packobj, we can verify that the DMU is correctly
3620 * tracking which parts of an object are allocated and free,
3621 * and that the contents of the allocated blocks are correct.
3625 * Read the directory info. If it's the first time, set things up.
3627 ztest_od_init(od
, id
, FTAG
, 0, DMU_OT_UINT64_OTHER
, 0, chunksize
);
3628 ztest_od_init(od
+ 1, id
, FTAG
, 1, DMU_OT_UINT64_OTHER
, 0, chunksize
);
3630 if (ztest_object_init(zd
, od
, size
, B_FALSE
) != 0) {
3631 umem_free(od
, size
);
3635 bigobj
= od
[0].od_object
;
3636 packobj
= od
[1].od_object
;
3637 chunksize
= od
[0].od_gen
;
3638 ASSERT(chunksize
== od
[1].od_gen
);
3641 * Prefetch a random chunk of the big object.
3642 * Our aim here is to get some async reads in flight
3643 * for blocks that we may free below; the DMU should
3644 * handle this race correctly.
3646 n
= ztest_random(regions
) * stride
+ ztest_random(width
);
3647 s
= 1 + ztest_random(2 * width
- 1);
3648 dmu_prefetch(os
, bigobj
, n
* chunksize
, s
* chunksize
);
3651 * Pick a random index and compute the offsets into packobj and bigobj.
3653 n
= ztest_random(regions
) * stride
+ ztest_random(width
);
3654 s
= 1 + ztest_random(width
- 1);
3656 packoff
= n
* sizeof (bufwad_t
);
3657 packsize
= s
* sizeof (bufwad_t
);
3659 bigoff
= n
* chunksize
;
3660 bigsize
= s
* chunksize
;
3662 packbuf
= umem_alloc(packsize
, UMEM_NOFAIL
);
3663 bigbuf
= umem_alloc(bigsize
, UMEM_NOFAIL
);
3666 * free_percent of the time, free a range of bigobj rather than
3669 freeit
= (ztest_random(100) < free_percent
);
3672 * Read the current contents of our objects.
3674 error
= dmu_read(os
, packobj
, packoff
, packsize
, packbuf
,
3677 error
= dmu_read(os
, bigobj
, bigoff
, bigsize
, bigbuf
,
3682 * Get a tx for the mods to both packobj and bigobj.
3684 tx
= dmu_tx_create(os
);
3686 dmu_tx_hold_write(tx
, packobj
, packoff
, packsize
);
3689 dmu_tx_hold_free(tx
, bigobj
, bigoff
, bigsize
);
3691 dmu_tx_hold_write(tx
, bigobj
, bigoff
, bigsize
);
3693 txg
= ztest_tx_assign(tx
, TXG_MIGHTWAIT
, FTAG
);
3695 umem_free(packbuf
, packsize
);
3696 umem_free(bigbuf
, bigsize
);
3697 umem_free(od
, size
);
3701 dmu_object_set_checksum(os
, bigobj
,
3702 (enum zio_checksum
)ztest_random_dsl_prop(ZFS_PROP_CHECKSUM
), tx
);
3704 dmu_object_set_compress(os
, bigobj
,
3705 (enum zio_compress
)ztest_random_dsl_prop(ZFS_PROP_COMPRESSION
), tx
);
3708 * For each index from n to n + s, verify that the existing bufwad
3709 * in packobj matches the bufwads at the head and tail of the
3710 * corresponding chunk in bigobj. Then update all three bufwads
3711 * with the new values we want to write out.
3713 for (i
= 0; i
< s
; i
++) {
3715 pack
= (bufwad_t
*)((char *)packbuf
+ i
* sizeof (bufwad_t
));
3717 bigH
= (bufwad_t
*)((char *)bigbuf
+ i
* chunksize
);
3719 bigT
= (bufwad_t
*)((char *)bigH
+ chunksize
) - 1;
3721 ASSERT((uintptr_t)bigH
- (uintptr_t)bigbuf
< bigsize
);
3722 ASSERT((uintptr_t)bigT
- (uintptr_t)bigbuf
< bigsize
);
3724 if (pack
->bw_txg
> txg
)
3725 fatal(0, "future leak: got %llx, open txg is %llx",
3728 if (pack
->bw_data
!= 0 && pack
->bw_index
!= n
+ i
)
3729 fatal(0, "wrong index: got %llx, wanted %llx+%llx",
3730 pack
->bw_index
, n
, i
);
3732 if (bcmp(pack
, bigH
, sizeof (bufwad_t
)) != 0)
3733 fatal(0, "pack/bigH mismatch in %p/%p", pack
, bigH
);
3735 if (bcmp(pack
, bigT
, sizeof (bufwad_t
)) != 0)
3736 fatal(0, "pack/bigT mismatch in %p/%p", pack
, bigT
);
3739 bzero(pack
, sizeof (bufwad_t
));
3741 pack
->bw_index
= n
+ i
;
3743 pack
->bw_data
= 1 + ztest_random(-2ULL);
3750 * We've verified all the old bufwads, and made new ones.
3751 * Now write them out.
3753 dmu_write(os
, packobj
, packoff
, packsize
, packbuf
, tx
);
3756 if (ztest_opts
.zo_verbose
>= 7) {
3757 (void) printf("freeing offset %llx size %llx"
3759 (u_longlong_t
)bigoff
,
3760 (u_longlong_t
)bigsize
,
3763 VERIFY(0 == dmu_free_range(os
, bigobj
, bigoff
, bigsize
, tx
));
3765 if (ztest_opts
.zo_verbose
>= 7) {
3766 (void) printf("writing offset %llx size %llx"
3768 (u_longlong_t
)bigoff
,
3769 (u_longlong_t
)bigsize
,
3772 dmu_write(os
, bigobj
, bigoff
, bigsize
, bigbuf
, tx
);
3778 * Sanity check the stuff we just wrote.
3781 void *packcheck
= umem_alloc(packsize
, UMEM_NOFAIL
);
3782 void *bigcheck
= umem_alloc(bigsize
, UMEM_NOFAIL
);
3784 VERIFY(0 == dmu_read(os
, packobj
, packoff
,
3785 packsize
, packcheck
, DMU_READ_PREFETCH
));
3786 VERIFY(0 == dmu_read(os
, bigobj
, bigoff
,
3787 bigsize
, bigcheck
, DMU_READ_PREFETCH
));
3789 ASSERT(bcmp(packbuf
, packcheck
, packsize
) == 0);
3790 ASSERT(bcmp(bigbuf
, bigcheck
, bigsize
) == 0);
3792 umem_free(packcheck
, packsize
);
3793 umem_free(bigcheck
, bigsize
);
3796 umem_free(packbuf
, packsize
);
3797 umem_free(bigbuf
, bigsize
);
3798 umem_free(od
, size
);
3802 compare_and_update_pbbufs(uint64_t s
, bufwad_t
*packbuf
, bufwad_t
*bigbuf
,
3803 uint64_t bigsize
, uint64_t n
, uint64_t chunksize
, uint64_t txg
)
3811 * For each index from n to n + s, verify that the existing bufwad
3812 * in packobj matches the bufwads at the head and tail of the
3813 * corresponding chunk in bigobj. Then update all three bufwads
3814 * with the new values we want to write out.
3816 for (i
= 0; i
< s
; i
++) {
3818 pack
= (bufwad_t
*)((char *)packbuf
+ i
* sizeof (bufwad_t
));
3820 bigH
= (bufwad_t
*)((char *)bigbuf
+ i
* chunksize
);
3822 bigT
= (bufwad_t
*)((char *)bigH
+ chunksize
) - 1;
3824 ASSERT((uintptr_t)bigH
- (uintptr_t)bigbuf
< bigsize
);
3825 ASSERT((uintptr_t)bigT
- (uintptr_t)bigbuf
< bigsize
);
3827 if (pack
->bw_txg
> txg
)
3828 fatal(0, "future leak: got %llx, open txg is %llx",
3831 if (pack
->bw_data
!= 0 && pack
->bw_index
!= n
+ i
)
3832 fatal(0, "wrong index: got %llx, wanted %llx+%llx",
3833 pack
->bw_index
, n
, i
);
3835 if (bcmp(pack
, bigH
, sizeof (bufwad_t
)) != 0)
3836 fatal(0, "pack/bigH mismatch in %p/%p", pack
, bigH
);
3838 if (bcmp(pack
, bigT
, sizeof (bufwad_t
)) != 0)
3839 fatal(0, "pack/bigT mismatch in %p/%p", pack
, bigT
);
3841 pack
->bw_index
= n
+ i
;
3843 pack
->bw_data
= 1 + ztest_random(-2ULL);
3850 #undef OD_ARRAY_SIZE
3851 #define OD_ARRAY_SIZE 2
3854 ztest_dmu_read_write_zcopy(ztest_ds_t
*zd
, uint64_t id
)
3856 objset_t
*os
= zd
->zd_os
;
3863 bufwad_t
*packbuf
, *bigbuf
;
3864 uint64_t packobj
, packoff
, packsize
, bigobj
, bigoff
, bigsize
;
3865 uint64_t blocksize
= ztest_random_blocksize();
3866 uint64_t chunksize
= blocksize
;
3867 uint64_t regions
= 997;
3868 uint64_t stride
= 123456789ULL;
3870 dmu_buf_t
*bonus_db
;
3871 arc_buf_t
**bigbuf_arcbufs
;
3872 dmu_object_info_t doi
;
3874 size
= sizeof(ztest_od_t
) * OD_ARRAY_SIZE
;
3875 od
= umem_alloc(size
, UMEM_NOFAIL
);
3878 * This test uses two objects, packobj and bigobj, that are always
3879 * updated together (i.e. in the same tx) so that their contents are
3880 * in sync and can be compared. Their contents relate to each other
3881 * in a simple way: packobj is a dense array of 'bufwad' structures,
3882 * while bigobj is a sparse array of the same bufwads. Specifically,
3883 * for any index n, there are three bufwads that should be identical:
3885 * packobj, at offset n * sizeof (bufwad_t)
3886 * bigobj, at the head of the nth chunk
3887 * bigobj, at the tail of the nth chunk
3889 * The chunk size is set equal to bigobj block size so that
3890 * dmu_assign_arcbuf() can be tested for object updates.
3894 * Read the directory info. If it's the first time, set things up.
3896 ztest_od_init(od
, id
, FTAG
, 0, DMU_OT_UINT64_OTHER
, blocksize
, 0);
3897 ztest_od_init(od
+ 1, id
, FTAG
, 1, DMU_OT_UINT64_OTHER
, 0, chunksize
);
3900 if (ztest_object_init(zd
, od
, size
, B_FALSE
) != 0) {
3901 umem_free(od
, size
);
3905 bigobj
= od
[0].od_object
;
3906 packobj
= od
[1].od_object
;
3907 blocksize
= od
[0].od_blocksize
;
3908 chunksize
= blocksize
;
3909 ASSERT(chunksize
== od
[1].od_gen
);
3911 VERIFY(dmu_object_info(os
, bigobj
, &doi
) == 0);
3912 VERIFY(ISP2(doi
.doi_data_block_size
));
3913 VERIFY(chunksize
== doi
.doi_data_block_size
);
3914 VERIFY(chunksize
>= 2 * sizeof (bufwad_t
));
3917 * Pick a random index and compute the offsets into packobj and bigobj.
3919 n
= ztest_random(regions
) * stride
+ ztest_random(width
);
3920 s
= 1 + ztest_random(width
- 1);
3922 packoff
= n
* sizeof (bufwad_t
);
3923 packsize
= s
* sizeof (bufwad_t
);
3925 bigoff
= n
* chunksize
;
3926 bigsize
= s
* chunksize
;
3928 packbuf
= umem_zalloc(packsize
, UMEM_NOFAIL
);
3929 bigbuf
= umem_zalloc(bigsize
, UMEM_NOFAIL
);
3931 VERIFY3U(0, ==, dmu_bonus_hold(os
, bigobj
, FTAG
, &bonus_db
));
3933 bigbuf_arcbufs
= umem_zalloc(2 * s
* sizeof (arc_buf_t
*), UMEM_NOFAIL
);
3936 * Iteration 0 test zcopy for DB_UNCACHED dbufs.
3937 * Iteration 1 test zcopy to already referenced dbufs.
3938 * Iteration 2 test zcopy to dirty dbuf in the same txg.
3939 * Iteration 3 test zcopy to dbuf dirty in previous txg.
3940 * Iteration 4 test zcopy when dbuf is no longer dirty.
3941 * Iteration 5 test zcopy when it can't be done.
3942 * Iteration 6 one more zcopy write.
3944 for (i
= 0; i
< 7; i
++) {
3949 * In iteration 5 (i == 5) use arcbufs
3950 * that don't match bigobj blksz to test
3951 * dmu_assign_arcbuf() when it can't directly
3952 * assign an arcbuf to a dbuf.
3954 for (j
= 0; j
< s
; j
++) {
3957 dmu_request_arcbuf(bonus_db
, chunksize
);
3959 bigbuf_arcbufs
[2 * j
] =
3960 dmu_request_arcbuf(bonus_db
, chunksize
/ 2);
3961 bigbuf_arcbufs
[2 * j
+ 1] =
3962 dmu_request_arcbuf(bonus_db
, chunksize
/ 2);
3967 * Get a tx for the mods to both packobj and bigobj.
3969 tx
= dmu_tx_create(os
);
3971 dmu_tx_hold_write(tx
, packobj
, packoff
, packsize
);
3972 dmu_tx_hold_write(tx
, bigobj
, bigoff
, bigsize
);
3974 txg
= ztest_tx_assign(tx
, TXG_MIGHTWAIT
, FTAG
);
3976 umem_free(packbuf
, packsize
);
3977 umem_free(bigbuf
, bigsize
);
3978 for (j
= 0; j
< s
; j
++) {
3980 dmu_return_arcbuf(bigbuf_arcbufs
[j
]);
3983 bigbuf_arcbufs
[2 * j
]);
3985 bigbuf_arcbufs
[2 * j
+ 1]);
3988 umem_free(bigbuf_arcbufs
, 2 * s
* sizeof (arc_buf_t
*));
3989 umem_free(od
, size
);
3990 dmu_buf_rele(bonus_db
, FTAG
);
3995 * 50% of the time don't read objects in the 1st iteration to
3996 * test dmu_assign_arcbuf() for the case when there're no
3997 * existing dbufs for the specified offsets.
3999 if (i
!= 0 || ztest_random(2) != 0) {
4000 error
= dmu_read(os
, packobj
, packoff
,
4001 packsize
, packbuf
, DMU_READ_PREFETCH
);
4003 error
= dmu_read(os
, bigobj
, bigoff
, bigsize
,
4004 bigbuf
, DMU_READ_PREFETCH
);
4007 compare_and_update_pbbufs(s
, packbuf
, bigbuf
, bigsize
,
4011 * We've verified all the old bufwads, and made new ones.
4012 * Now write them out.
4014 dmu_write(os
, packobj
, packoff
, packsize
, packbuf
, tx
);
4015 if (ztest_opts
.zo_verbose
>= 7) {
4016 (void) printf("writing offset %llx size %llx"
4018 (u_longlong_t
)bigoff
,
4019 (u_longlong_t
)bigsize
,
4022 for (off
= bigoff
, j
= 0; j
< s
; j
++, off
+= chunksize
) {
4025 bcopy((caddr_t
)bigbuf
+ (off
- bigoff
),
4026 bigbuf_arcbufs
[j
]->b_data
, chunksize
);
4028 bcopy((caddr_t
)bigbuf
+ (off
- bigoff
),
4029 bigbuf_arcbufs
[2 * j
]->b_data
,
4031 bcopy((caddr_t
)bigbuf
+ (off
- bigoff
) +
4033 bigbuf_arcbufs
[2 * j
+ 1]->b_data
,
4038 VERIFY(dmu_buf_hold(os
, bigobj
, off
,
4039 FTAG
, &dbt
, DMU_READ_NO_PREFETCH
) == 0);
4042 dmu_assign_arcbuf(bonus_db
, off
,
4043 bigbuf_arcbufs
[j
], tx
);
4045 dmu_assign_arcbuf(bonus_db
, off
,
4046 bigbuf_arcbufs
[2 * j
], tx
);
4047 dmu_assign_arcbuf(bonus_db
,
4048 off
+ chunksize
/ 2,
4049 bigbuf_arcbufs
[2 * j
+ 1], tx
);
4052 dmu_buf_rele(dbt
, FTAG
);
4058 * Sanity check the stuff we just wrote.
4061 void *packcheck
= umem_alloc(packsize
, UMEM_NOFAIL
);
4062 void *bigcheck
= umem_alloc(bigsize
, UMEM_NOFAIL
);
4064 VERIFY(0 == dmu_read(os
, packobj
, packoff
,
4065 packsize
, packcheck
, DMU_READ_PREFETCH
));
4066 VERIFY(0 == dmu_read(os
, bigobj
, bigoff
,
4067 bigsize
, bigcheck
, DMU_READ_PREFETCH
));
4069 ASSERT(bcmp(packbuf
, packcheck
, packsize
) == 0);
4070 ASSERT(bcmp(bigbuf
, bigcheck
, bigsize
) == 0);
4072 umem_free(packcheck
, packsize
);
4073 umem_free(bigcheck
, bigsize
);
4076 txg_wait_open(dmu_objset_pool(os
), 0);
4077 } else if (i
== 3) {
4078 txg_wait_synced(dmu_objset_pool(os
), 0);
4082 dmu_buf_rele(bonus_db
, FTAG
);
4083 umem_free(packbuf
, packsize
);
4084 umem_free(bigbuf
, bigsize
);
4085 umem_free(bigbuf_arcbufs
, 2 * s
* sizeof (arc_buf_t
*));
4086 umem_free(od
, size
);
4091 ztest_dmu_write_parallel(ztest_ds_t
*zd
, uint64_t id
)
4095 od
= umem_alloc(sizeof(ztest_od_t
), UMEM_NOFAIL
);
4096 uint64_t offset
= (1ULL << (ztest_random(20) + 43)) +
4097 (ztest_random(ZTEST_RANGE_LOCKS
) << SPA_MAXBLOCKSHIFT
);
4100 * Have multiple threads write to large offsets in an object
4101 * to verify that parallel writes to an object -- even to the
4102 * same blocks within the object -- doesn't cause any trouble.
4104 ztest_od_init(od
, ID_PARALLEL
, FTAG
, 0, DMU_OT_UINT64_OTHER
, 0, 0);
4106 if (ztest_object_init(zd
, od
, sizeof (ztest_od_t
), B_FALSE
) != 0)
4109 while (ztest_random(10) != 0)
4110 ztest_io(zd
, od
->od_object
, offset
);
4112 umem_free(od
, sizeof(ztest_od_t
));
4116 ztest_dmu_prealloc(ztest_ds_t
*zd
, uint64_t id
)
4119 uint64_t offset
= (1ULL << (ztest_random(4) + SPA_MAXBLOCKSHIFT
)) +
4120 (ztest_random(ZTEST_RANGE_LOCKS
) << SPA_MAXBLOCKSHIFT
);
4121 uint64_t count
= ztest_random(20) + 1;
4122 uint64_t blocksize
= ztest_random_blocksize();
4125 od
= umem_alloc(sizeof(ztest_od_t
), UMEM_NOFAIL
);
4127 ztest_od_init(od
, id
, FTAG
, 0, DMU_OT_UINT64_OTHER
, blocksize
, 0);
4129 if (ztest_object_init(zd
, od
, sizeof (ztest_od_t
), !ztest_random(2)) != 0) {
4130 umem_free(od
, sizeof(ztest_od_t
));
4134 if (ztest_truncate(zd
, od
->od_object
, offset
, count
* blocksize
) != 0) {
4135 umem_free(od
, sizeof(ztest_od_t
));
4139 ztest_prealloc(zd
, od
->od_object
, offset
, count
* blocksize
);
4141 data
= umem_zalloc(blocksize
, UMEM_NOFAIL
);
4143 while (ztest_random(count
) != 0) {
4144 uint64_t randoff
= offset
+ (ztest_random(count
) * blocksize
);
4145 if (ztest_write(zd
, od
->od_object
, randoff
, blocksize
,
4148 while (ztest_random(4) != 0)
4149 ztest_io(zd
, od
->od_object
, randoff
);
4152 umem_free(data
, blocksize
);
4153 umem_free(od
, sizeof(ztest_od_t
));
4157 * Verify that zap_{create,destroy,add,remove,update} work as expected.
4159 #define ZTEST_ZAP_MIN_INTS 1
4160 #define ZTEST_ZAP_MAX_INTS 4
4161 #define ZTEST_ZAP_MAX_PROPS 1000
4164 ztest_zap(ztest_ds_t
*zd
, uint64_t id
)
4166 objset_t
*os
= zd
->zd_os
;
4169 uint64_t txg
, last_txg
;
4170 uint64_t value
[ZTEST_ZAP_MAX_INTS
];
4171 uint64_t zl_ints
, zl_intsize
, prop
;
4174 char propname
[100], txgname
[100];
4176 char *hc
[2] = { "s.acl.h", ".s.open.h.hyLZlg" };
4178 od
= umem_alloc(sizeof(ztest_od_t
), UMEM_NOFAIL
);
4179 ztest_od_init(od
, id
, FTAG
, 0, DMU_OT_ZAP_OTHER
, 0, 0);
4181 if (ztest_object_init(zd
, od
, sizeof (ztest_od_t
),
4182 !ztest_random(2)) != 0)
4185 object
= od
->od_object
;
4188 * Generate a known hash collision, and verify that
4189 * we can lookup and remove both entries.
4191 tx
= dmu_tx_create(os
);
4192 dmu_tx_hold_zap(tx
, object
, B_TRUE
, NULL
);
4193 txg
= ztest_tx_assign(tx
, TXG_MIGHTWAIT
, FTAG
);
4196 for (i
= 0; i
< 2; i
++) {
4198 VERIFY3U(0, ==, zap_add(os
, object
, hc
[i
], sizeof (uint64_t),
4201 for (i
= 0; i
< 2; i
++) {
4202 VERIFY3U(EEXIST
, ==, zap_add(os
, object
, hc
[i
],
4203 sizeof (uint64_t), 1, &value
[i
], tx
));
4205 zap_length(os
, object
, hc
[i
], &zl_intsize
, &zl_ints
));
4206 ASSERT3U(zl_intsize
, ==, sizeof (uint64_t));
4207 ASSERT3U(zl_ints
, ==, 1);
4209 for (i
= 0; i
< 2; i
++) {
4210 VERIFY3U(0, ==, zap_remove(os
, object
, hc
[i
], tx
));
4215 * Generate a buch of random entries.
4217 ints
= MAX(ZTEST_ZAP_MIN_INTS
, object
% ZTEST_ZAP_MAX_INTS
);
4219 prop
= ztest_random(ZTEST_ZAP_MAX_PROPS
);
4220 (void) sprintf(propname
, "prop_%llu", (u_longlong_t
)prop
);
4221 (void) sprintf(txgname
, "txg_%llu", (u_longlong_t
)prop
);
4222 bzero(value
, sizeof (value
));
4226 * If these zap entries already exist, validate their contents.
4228 error
= zap_length(os
, object
, txgname
, &zl_intsize
, &zl_ints
);
4230 ASSERT3U(zl_intsize
, ==, sizeof (uint64_t));
4231 ASSERT3U(zl_ints
, ==, 1);
4233 VERIFY(zap_lookup(os
, object
, txgname
, zl_intsize
,
4234 zl_ints
, &last_txg
) == 0);
4236 VERIFY(zap_length(os
, object
, propname
, &zl_intsize
,
4239 ASSERT3U(zl_intsize
, ==, sizeof (uint64_t));
4240 ASSERT3U(zl_ints
, ==, ints
);
4242 VERIFY(zap_lookup(os
, object
, propname
, zl_intsize
,
4243 zl_ints
, value
) == 0);
4245 for (i
= 0; i
< ints
; i
++) {
4246 ASSERT3U(value
[i
], ==, last_txg
+ object
+ i
);
4249 ASSERT3U(error
, ==, ENOENT
);
4253 * Atomically update two entries in our zap object.
4254 * The first is named txg_%llu, and contains the txg
4255 * in which the property was last updated. The second
4256 * is named prop_%llu, and the nth element of its value
4257 * should be txg + object + n.
4259 tx
= dmu_tx_create(os
);
4260 dmu_tx_hold_zap(tx
, object
, B_TRUE
, NULL
);
4261 txg
= ztest_tx_assign(tx
, TXG_MIGHTWAIT
, FTAG
);
4266 fatal(0, "zap future leak: old %llu new %llu", last_txg
, txg
);
4268 for (i
= 0; i
< ints
; i
++)
4269 value
[i
] = txg
+ object
+ i
;
4271 VERIFY3U(0, ==, zap_update(os
, object
, txgname
, sizeof (uint64_t),
4273 VERIFY3U(0, ==, zap_update(os
, object
, propname
, sizeof (uint64_t),
4279 * Remove a random pair of entries.
4281 prop
= ztest_random(ZTEST_ZAP_MAX_PROPS
);
4282 (void) sprintf(propname
, "prop_%llu", (u_longlong_t
)prop
);
4283 (void) sprintf(txgname
, "txg_%llu", (u_longlong_t
)prop
);
4285 error
= zap_length(os
, object
, txgname
, &zl_intsize
, &zl_ints
);
4287 if (error
== ENOENT
)
4292 tx
= dmu_tx_create(os
);
4293 dmu_tx_hold_zap(tx
, object
, B_TRUE
, NULL
);
4294 txg
= ztest_tx_assign(tx
, TXG_MIGHTWAIT
, FTAG
);
4297 VERIFY3U(0, ==, zap_remove(os
, object
, txgname
, tx
));
4298 VERIFY3U(0, ==, zap_remove(os
, object
, propname
, tx
));
4301 umem_free(od
, sizeof(ztest_od_t
));
4305 * Testcase to test the upgrading of a microzap to fatzap.
4308 ztest_fzap(ztest_ds_t
*zd
, uint64_t id
)
4310 objset_t
*os
= zd
->zd_os
;
4312 uint64_t object
, txg
;
4315 od
= umem_alloc(sizeof(ztest_od_t
), UMEM_NOFAIL
);
4316 ztest_od_init(od
, id
, FTAG
, 0, DMU_OT_ZAP_OTHER
, 0, 0);
4318 if (ztest_object_init(zd
, od
, sizeof (ztest_od_t
),
4319 !ztest_random(2)) != 0)
4321 object
= od
->od_object
;
4324 * Add entries to this ZAP and make sure it spills over
4325 * and gets upgraded to a fatzap. Also, since we are adding
4326 * 2050 entries we should see ptrtbl growth and leaf-block split.
4328 for (i
= 0; i
< 2050; i
++) {
4329 char name
[MAXNAMELEN
];
4334 (void) snprintf(name
, sizeof (name
), "fzap-%llu-%llu",
4335 (u_longlong_t
)id
, (u_longlong_t
)value
);
4337 tx
= dmu_tx_create(os
);
4338 dmu_tx_hold_zap(tx
, object
, B_TRUE
, name
);
4339 txg
= ztest_tx_assign(tx
, TXG_MIGHTWAIT
, FTAG
);
4342 error
= zap_add(os
, object
, name
, sizeof (uint64_t), 1,
4344 ASSERT(error
== 0 || error
== EEXIST
);
4348 umem_free(od
, sizeof(ztest_od_t
));
4353 ztest_zap_parallel(ztest_ds_t
*zd
, uint64_t id
)
4355 objset_t
*os
= zd
->zd_os
;
4357 uint64_t txg
, object
, count
, wsize
, wc
, zl_wsize
, zl_wc
;
4359 int i
, namelen
, error
;
4360 int micro
= ztest_random(2);
4361 char name
[20], string_value
[20];
4364 od
= umem_alloc(sizeof(ztest_od_t
), UMEM_NOFAIL
);
4365 ztest_od_init(od
, ID_PARALLEL
, FTAG
, micro
, DMU_OT_ZAP_OTHER
, 0, 0);
4367 if (ztest_object_init(zd
, od
, sizeof (ztest_od_t
), B_FALSE
) != 0) {
4368 umem_free(od
, sizeof(ztest_od_t
));
4372 object
= od
->od_object
;
4375 * Generate a random name of the form 'xxx.....' where each
4376 * x is a random printable character and the dots are dots.
4377 * There are 94 such characters, and the name length goes from
4378 * 6 to 20, so there are 94^3 * 15 = 12,458,760 possible names.
4380 namelen
= ztest_random(sizeof (name
) - 5) + 5 + 1;
4382 for (i
= 0; i
< 3; i
++)
4383 name
[i
] = '!' + ztest_random('~' - '!' + 1);
4384 for (; i
< namelen
- 1; i
++)
4388 if ((namelen
& 1) || micro
) {
4389 wsize
= sizeof (txg
);
4395 data
= string_value
;
4399 VERIFY(zap_count(os
, object
, &count
) == 0);
4400 ASSERT(count
!= -1ULL);
4403 * Select an operation: length, lookup, add, update, remove.
4405 i
= ztest_random(5);
4408 tx
= dmu_tx_create(os
);
4409 dmu_tx_hold_zap(tx
, object
, B_TRUE
, NULL
);
4410 txg
= ztest_tx_assign(tx
, TXG_MIGHTWAIT
, FTAG
);
4413 bcopy(name
, string_value
, namelen
);
4417 bzero(string_value
, namelen
);
4423 error
= zap_length(os
, object
, name
, &zl_wsize
, &zl_wc
);
4425 ASSERT3U(wsize
, ==, zl_wsize
);
4426 ASSERT3U(wc
, ==, zl_wc
);
4428 ASSERT3U(error
, ==, ENOENT
);
4433 error
= zap_lookup(os
, object
, name
, wsize
, wc
, data
);
4435 if (data
== string_value
&&
4436 bcmp(name
, data
, namelen
) != 0)
4437 fatal(0, "name '%s' != val '%s' len %d",
4438 name
, data
, namelen
);
4440 ASSERT3U(error
, ==, ENOENT
);
4445 error
= zap_add(os
, object
, name
, wsize
, wc
, data
, tx
);
4446 ASSERT(error
== 0 || error
== EEXIST
);
4450 VERIFY(zap_update(os
, object
, name
, wsize
, wc
, data
, tx
) == 0);
4454 error
= zap_remove(os
, object
, name
, tx
);
4455 ASSERT(error
== 0 || error
== ENOENT
);
4462 umem_free(od
, sizeof(ztest_od_t
));
4466 * Commit callback data.
4468 typedef struct ztest_cb_data
{
4469 list_node_t zcd_node
;
4471 int zcd_expected_err
;
4472 boolean_t zcd_added
;
4473 boolean_t zcd_called
;
4477 /* This is the actual commit callback function */
4479 ztest_commit_callback(void *arg
, int error
)
4481 ztest_cb_data_t
*data
= arg
;
4482 uint64_t synced_txg
;
4484 VERIFY(data
!= NULL
);
4485 VERIFY3S(data
->zcd_expected_err
, ==, error
);
4486 VERIFY(!data
->zcd_called
);
4488 synced_txg
= spa_last_synced_txg(data
->zcd_spa
);
4489 if (data
->zcd_txg
> synced_txg
)
4490 fatal(0, "commit callback of txg %" PRIu64
" called prematurely"
4491 ", last synced txg = %" PRIu64
"\n", data
->zcd_txg
,
4494 data
->zcd_called
= B_TRUE
;
4496 if (error
== ECANCELED
) {
4497 ASSERT0(data
->zcd_txg
);
4498 ASSERT(!data
->zcd_added
);
4501 * The private callback data should be destroyed here, but
4502 * since we are going to check the zcd_called field after
4503 * dmu_tx_abort(), we will destroy it there.
4508 ASSERT(data
->zcd_added
);
4509 ASSERT3U(data
->zcd_txg
, !=, 0);
4511 (void) mutex_enter(&zcl
.zcl_callbacks_lock
);
4513 /* See if this cb was called more quickly */
4514 if ((synced_txg
- data
->zcd_txg
) < zc_min_txg_delay
)
4515 zc_min_txg_delay
= synced_txg
- data
->zcd_txg
;
4517 /* Remove our callback from the list */
4518 list_remove(&zcl
.zcl_callbacks
, data
);
4520 (void) mutex_exit(&zcl
.zcl_callbacks_lock
);
4522 umem_free(data
, sizeof (ztest_cb_data_t
));
4525 /* Allocate and initialize callback data structure */
4526 static ztest_cb_data_t
*
4527 ztest_create_cb_data(objset_t
*os
, uint64_t txg
)
4529 ztest_cb_data_t
*cb_data
;
4531 cb_data
= umem_zalloc(sizeof (ztest_cb_data_t
), UMEM_NOFAIL
);
4533 cb_data
->zcd_txg
= txg
;
4534 cb_data
->zcd_spa
= dmu_objset_spa(os
);
4535 list_link_init(&cb_data
->zcd_node
);
4541 * Commit callback test.
4544 ztest_dmu_commit_callbacks(ztest_ds_t
*zd
, uint64_t id
)
4546 objset_t
*os
= zd
->zd_os
;
4549 ztest_cb_data_t
*cb_data
[3], *tmp_cb
;
4550 uint64_t old_txg
, txg
;
4553 od
= umem_alloc(sizeof(ztest_od_t
), UMEM_NOFAIL
);
4554 ztest_od_init(od
, id
, FTAG
, 0, DMU_OT_UINT64_OTHER
, 0, 0);
4556 if (ztest_object_init(zd
, od
, sizeof (ztest_od_t
), B_FALSE
) != 0) {
4557 umem_free(od
, sizeof(ztest_od_t
));
4561 tx
= dmu_tx_create(os
);
4563 cb_data
[0] = ztest_create_cb_data(os
, 0);
4564 dmu_tx_callback_register(tx
, ztest_commit_callback
, cb_data
[0]);
4566 dmu_tx_hold_write(tx
, od
->od_object
, 0, sizeof (uint64_t));
4568 /* Every once in a while, abort the transaction on purpose */
4569 if (ztest_random(100) == 0)
4573 error
= dmu_tx_assign(tx
, TXG_NOWAIT
);
4575 txg
= error
? 0 : dmu_tx_get_txg(tx
);
4577 cb_data
[0]->zcd_txg
= txg
;
4578 cb_data
[1] = ztest_create_cb_data(os
, txg
);
4579 dmu_tx_callback_register(tx
, ztest_commit_callback
, cb_data
[1]);
4583 * It's not a strict requirement to call the registered
4584 * callbacks from inside dmu_tx_abort(), but that's what
4585 * it's supposed to happen in the current implementation
4586 * so we will check for that.
4588 for (i
= 0; i
< 2; i
++) {
4589 cb_data
[i
]->zcd_expected_err
= ECANCELED
;
4590 VERIFY(!cb_data
[i
]->zcd_called
);
4595 for (i
= 0; i
< 2; i
++) {
4596 VERIFY(cb_data
[i
]->zcd_called
);
4597 umem_free(cb_data
[i
], sizeof (ztest_cb_data_t
));
4600 umem_free(od
, sizeof(ztest_od_t
));
4604 cb_data
[2] = ztest_create_cb_data(os
, txg
);
4605 dmu_tx_callback_register(tx
, ztest_commit_callback
, cb_data
[2]);
4608 * Read existing data to make sure there isn't a future leak.
4610 VERIFY(0 == dmu_read(os
, od
->od_object
, 0, sizeof (uint64_t),
4611 &old_txg
, DMU_READ_PREFETCH
));
4614 fatal(0, "future leak: got %" PRIu64
", open txg is %" PRIu64
,
4617 dmu_write(os
, od
->od_object
, 0, sizeof (uint64_t), &txg
, tx
);
4619 (void) mutex_enter(&zcl
.zcl_callbacks_lock
);
4622 * Since commit callbacks don't have any ordering requirement and since
4623 * it is theoretically possible for a commit callback to be called
4624 * after an arbitrary amount of time has elapsed since its txg has been
4625 * synced, it is difficult to reliably determine whether a commit
4626 * callback hasn't been called due to high load or due to a flawed
4629 * In practice, we will assume that if after a certain number of txgs a
4630 * commit callback hasn't been called, then most likely there's an
4631 * implementation bug..
4633 tmp_cb
= list_head(&zcl
.zcl_callbacks
);
4634 if (tmp_cb
!= NULL
&&
4635 tmp_cb
->zcd_txg
+ ZTEST_COMMIT_CB_THRESH
< txg
) {
4636 fatal(0, "Commit callback threshold exceeded, oldest txg: %"
4637 PRIu64
", open txg: %" PRIu64
"\n", tmp_cb
->zcd_txg
, txg
);
4641 * Let's find the place to insert our callbacks.
4643 * Even though the list is ordered by txg, it is possible for the
4644 * insertion point to not be the end because our txg may already be
4645 * quiescing at this point and other callbacks in the open txg
4646 * (from other objsets) may have sneaked in.
4648 tmp_cb
= list_tail(&zcl
.zcl_callbacks
);
4649 while (tmp_cb
!= NULL
&& tmp_cb
->zcd_txg
> txg
)
4650 tmp_cb
= list_prev(&zcl
.zcl_callbacks
, tmp_cb
);
4652 /* Add the 3 callbacks to the list */
4653 for (i
= 0; i
< 3; i
++) {
4655 list_insert_head(&zcl
.zcl_callbacks
, cb_data
[i
]);
4657 list_insert_after(&zcl
.zcl_callbacks
, tmp_cb
,
4660 cb_data
[i
]->zcd_added
= B_TRUE
;
4661 VERIFY(!cb_data
[i
]->zcd_called
);
4663 tmp_cb
= cb_data
[i
];
4668 (void) mutex_exit(&zcl
.zcl_callbacks_lock
);
4672 umem_free(od
, sizeof(ztest_od_t
));
4677 ztest_dsl_prop_get_set(ztest_ds_t
*zd
, uint64_t id
)
4679 zfs_prop_t proplist
[] = {
4681 ZFS_PROP_COMPRESSION
,
4687 (void) rw_enter(&ztest_name_lock
, RW_READER
);
4689 for (p
= 0; p
< sizeof (proplist
) / sizeof (proplist
[0]); p
++)
4690 (void) ztest_dsl_prop_set_uint64(zd
->zd_name
, proplist
[p
],
4691 ztest_random_dsl_prop(proplist
[p
]), (int)ztest_random(2));
4693 (void) rw_exit(&ztest_name_lock
);
4698 ztest_spa_prop_get_set(ztest_ds_t
*zd
, uint64_t id
)
4700 nvlist_t
*props
= NULL
;
4702 (void) rw_enter(&ztest_name_lock
, RW_READER
);
4704 (void) ztest_spa_prop_set_uint64(ZPOOL_PROP_DEDUPDITTO
,
4705 ZIO_DEDUPDITTO_MIN
+ ztest_random(ZIO_DEDUPDITTO_MIN
));
4707 VERIFY0(spa_prop_get(ztest_spa
, &props
));
4709 if (ztest_opts
.zo_verbose
>= 6)
4710 dump_nvlist(props
, 4);
4714 (void) rw_exit(&ztest_name_lock
);
4718 * Test snapshot hold/release and deferred destroy.
4721 ztest_dmu_snapshot_hold(ztest_ds_t
*zd
, uint64_t id
)
4724 objset_t
*os
= zd
->zd_os
;
4728 char clonename
[100];
4730 char osname
[MAXNAMELEN
];
4732 (void) rw_enter(&ztest_name_lock
, RW_READER
);
4734 dmu_objset_name(os
, osname
);
4736 (void) snprintf(snapname
, 100, "sh1_%llu", (u_longlong_t
)id
);
4737 (void) snprintf(fullname
, 100, "%s@%s", osname
, snapname
);
4738 (void) snprintf(clonename
, 100, "%s/ch1_%llu",osname
,(u_longlong_t
)id
);
4739 (void) snprintf(tag
, 100, "tag_%llu", (u_longlong_t
)id
);
4742 * Clean up from any previous run.
4744 (void) dmu_objset_destroy(clonename
, B_FALSE
);
4745 (void) dsl_dataset_user_release(osname
, snapname
, tag
, B_FALSE
);
4746 (void) dmu_objset_destroy(fullname
, B_FALSE
);
4749 * Create snapshot, clone it, mark snap for deferred destroy,
4750 * destroy clone, verify snap was also destroyed.
4752 error
= dmu_objset_snapshot(osname
, snapname
, NULL
, NULL
, FALSE
,
4755 if (error
== ENOSPC
) {
4756 ztest_record_enospc("dmu_objset_snapshot");
4759 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname
, error
);
4762 error
= dmu_objset_hold(fullname
, FTAG
, &origin
);
4764 fatal(0, "dmu_objset_hold(%s) = %d", fullname
, error
);
4766 error
= dmu_objset_clone(clonename
, dmu_objset_ds(origin
), 0);
4767 dmu_objset_rele(origin
, FTAG
);
4769 if (error
== ENOSPC
) {
4770 ztest_record_enospc("dmu_objset_clone");
4773 fatal(0, "dmu_objset_clone(%s) = %d", clonename
, error
);
4776 error
= dmu_objset_destroy(fullname
, B_TRUE
);
4778 fatal(0, "dmu_objset_destroy(%s, B_TRUE) = %d",
4782 error
= dmu_objset_destroy(clonename
, B_FALSE
);
4784 fatal(0, "dmu_objset_destroy(%s) = %d", clonename
, error
);
4786 error
= dmu_objset_hold(fullname
, FTAG
, &origin
);
4787 if (error
!= ENOENT
)
4788 fatal(0, "dmu_objset_hold(%s) = %d", fullname
, error
);
4791 * Create snapshot, add temporary hold, verify that we can't
4792 * destroy a held snapshot, mark for deferred destroy,
4793 * release hold, verify snapshot was destroyed.
4795 error
= dmu_objset_snapshot(osname
, snapname
, NULL
, NULL
, FALSE
,
4798 if (error
== ENOSPC
) {
4799 ztest_record_enospc("dmu_objset_snapshot");
4802 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname
, error
);
4805 error
= dsl_dataset_user_hold(osname
, snapname
, tag
, B_FALSE
,
4808 fatal(0, "dsl_dataset_user_hold(%s)", fullname
, tag
);
4810 error
= dmu_objset_destroy(fullname
, B_FALSE
);
4811 if (error
!= EBUSY
) {
4812 fatal(0, "dmu_objset_destroy(%s, B_FALSE) = %d",
4816 error
= dmu_objset_destroy(fullname
, B_TRUE
);
4818 fatal(0, "dmu_objset_destroy(%s, B_TRUE) = %d",
4822 error
= dsl_dataset_user_release(osname
, snapname
, tag
, B_FALSE
);
4824 fatal(0, "dsl_dataset_user_release(%s)", fullname
, tag
);
4826 VERIFY(dmu_objset_hold(fullname
, FTAG
, &origin
) == ENOENT
);
4829 (void) rw_exit(&ztest_name_lock
);
4833 * Inject random faults into the on-disk data.
4837 ztest_fault_inject(ztest_ds_t
*zd
, uint64_t id
)
4839 ztest_shared_t
*zs
= ztest_shared
;
4840 spa_t
*spa
= ztest_spa
;
4844 uint64_t bad
= 0x1990c0ffeedecadeull
;
4849 int bshift
= SPA_MAXBLOCKSHIFT
+ 2; /* don't scrog all labels */
4855 boolean_t islog
= B_FALSE
;
4857 path0
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
4858 pathrand
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
4860 mutex_enter(&ztest_vdev_lock
);
4861 maxfaults
= MAXFAULTS();
4862 leaves
= MAX(zs
->zs_mirrors
, 1) * ztest_opts
.zo_raidz
;
4863 mirror_save
= zs
->zs_mirrors
;
4864 mutex_exit(&ztest_vdev_lock
);
4866 ASSERT(leaves
>= 1);
4869 * We need SCL_STATE here because we're going to look at vd0->vdev_tsd.
4871 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
4873 if (ztest_random(2) == 0) {
4875 * Inject errors on a normal data device or slog device.
4877 top
= ztest_random_vdev_top(spa
, B_TRUE
);
4878 leaf
= ztest_random(leaves
) + zs
->zs_splits
;
4881 * Generate paths to the first leaf in this top-level vdev,
4882 * and to the random leaf we selected. We'll induce transient
4883 * write failures and random online/offline activity on leaf 0,
4884 * and we'll write random garbage to the randomly chosen leaf.
4886 (void) snprintf(path0
, MAXPATHLEN
, ztest_dev_template
,
4887 ztest_opts
.zo_dir
, ztest_opts
.zo_pool
,
4888 top
* leaves
+ zs
->zs_splits
);
4889 (void) snprintf(pathrand
, MAXPATHLEN
, ztest_dev_template
,
4890 ztest_opts
.zo_dir
, ztest_opts
.zo_pool
,
4891 top
* leaves
+ leaf
);
4893 vd0
= vdev_lookup_by_path(spa
->spa_root_vdev
, path0
);
4894 if (vd0
!= NULL
&& vd0
->vdev_top
->vdev_islog
)
4897 if (vd0
!= NULL
&& maxfaults
!= 1) {
4899 * Make vd0 explicitly claim to be unreadable,
4900 * or unwriteable, or reach behind its back
4901 * and close the underlying fd. We can do this if
4902 * maxfaults == 0 because we'll fail and reexecute,
4903 * and we can do it if maxfaults >= 2 because we'll
4904 * have enough redundancy. If maxfaults == 1, the
4905 * combination of this with injection of random data
4906 * corruption below exceeds the pool's fault tolerance.
4908 vdev_file_t
*vf
= vd0
->vdev_tsd
;
4910 if (vf
!= NULL
&& ztest_random(3) == 0) {
4911 (void) close(vf
->vf_vnode
->v_fd
);
4912 vf
->vf_vnode
->v_fd
= -1;
4913 } else if (ztest_random(2) == 0) {
4914 vd0
->vdev_cant_read
= B_TRUE
;
4916 vd0
->vdev_cant_write
= B_TRUE
;
4918 guid0
= vd0
->vdev_guid
;
4922 * Inject errors on an l2cache device.
4924 spa_aux_vdev_t
*sav
= &spa
->spa_l2cache
;
4926 if (sav
->sav_count
== 0) {
4927 spa_config_exit(spa
, SCL_STATE
, FTAG
);
4930 vd0
= sav
->sav_vdevs
[ztest_random(sav
->sav_count
)];
4931 guid0
= vd0
->vdev_guid
;
4932 (void) strcpy(path0
, vd0
->vdev_path
);
4933 (void) strcpy(pathrand
, vd0
->vdev_path
);
4937 maxfaults
= INT_MAX
; /* no limit on cache devices */
4940 spa_config_exit(spa
, SCL_STATE
, FTAG
);
4943 * If we can tolerate two or more faults, or we're dealing
4944 * with a slog, randomly online/offline vd0.
4946 if ((maxfaults
>= 2 || islog
) && guid0
!= 0) {
4947 if (ztest_random(10) < 6) {
4948 int flags
= (ztest_random(2) == 0 ?
4949 ZFS_OFFLINE_TEMPORARY
: 0);
4952 * We have to grab the zs_name_lock as writer to
4953 * prevent a race between offlining a slog and
4954 * destroying a dataset. Offlining the slog will
4955 * grab a reference on the dataset which may cause
4956 * dmu_objset_destroy() to fail with EBUSY thus
4957 * leaving the dataset in an inconsistent state.
4960 (void) rw_enter(&ztest_name_lock
,
4963 VERIFY(vdev_offline(spa
, guid0
, flags
) != EBUSY
);
4966 (void) rw_exit(&ztest_name_lock
);
4969 * Ideally we would like to be able to randomly
4970 * call vdev_[on|off]line without holding locks
4971 * to force unpredictable failures but the side
4972 * effects of vdev_[on|off]line prevent us from
4973 * doing so. We grab the ztest_vdev_lock here to
4974 * prevent a race between injection testing and
4977 mutex_enter(&ztest_vdev_lock
);
4978 (void) vdev_online(spa
, guid0
, 0, NULL
);
4979 mutex_exit(&ztest_vdev_lock
);
4987 * We have at least single-fault tolerance, so inject data corruption.
4989 fd
= open(pathrand
, O_RDWR
);
4991 if (fd
== -1) /* we hit a gap in the device namespace */
4994 fsize
= lseek(fd
, 0, SEEK_END
);
4996 while (--iters
!= 0) {
4997 offset
= ztest_random(fsize
/ (leaves
<< bshift
)) *
4998 (leaves
<< bshift
) + (leaf
<< bshift
) +
4999 (ztest_random(1ULL << (bshift
- 1)) & -8ULL);
5001 if (offset
>= fsize
)
5004 mutex_enter(&ztest_vdev_lock
);
5005 if (mirror_save
!= zs
->zs_mirrors
) {
5006 mutex_exit(&ztest_vdev_lock
);
5011 if (pwrite(fd
, &bad
, sizeof (bad
), offset
) != sizeof (bad
))
5012 fatal(1, "can't inject bad word at 0x%llx in %s",
5015 mutex_exit(&ztest_vdev_lock
);
5017 if (ztest_opts
.zo_verbose
>= 7)
5018 (void) printf("injected bad word into %s,"
5019 " offset 0x%llx\n", pathrand
, (u_longlong_t
)offset
);
5024 umem_free(path0
, MAXPATHLEN
);
5025 umem_free(pathrand
, MAXPATHLEN
);
5029 * Verify that DDT repair works as expected.
5032 ztest_ddt_repair(ztest_ds_t
*zd
, uint64_t id
)
5034 ztest_shared_t
*zs
= ztest_shared
;
5035 spa_t
*spa
= ztest_spa
;
5036 objset_t
*os
= zd
->zd_os
;
5038 uint64_t object
, blocksize
, txg
, pattern
, psize
;
5039 enum zio_checksum checksum
= spa_dedup_checksum(spa
);
5044 int copies
= 2 * ZIO_DEDUPDITTO_MIN
;
5047 blocksize
= ztest_random_blocksize();
5048 blocksize
= MIN(blocksize
, 2048); /* because we write so many */
5050 od
= umem_alloc(sizeof(ztest_od_t
), UMEM_NOFAIL
);
5051 ztest_od_init(od
, id
, FTAG
, 0, DMU_OT_UINT64_OTHER
, blocksize
, 0);
5053 if (ztest_object_init(zd
, od
, sizeof (ztest_od_t
), B_FALSE
) != 0) {
5054 umem_free(od
, sizeof(ztest_od_t
));
5059 * Take the name lock as writer to prevent anyone else from changing
5060 * the pool and dataset properies we need to maintain during this test.
5062 (void) rw_enter(&ztest_name_lock
, RW_WRITER
);
5064 if (ztest_dsl_prop_set_uint64(zd
->zd_name
, ZFS_PROP_DEDUP
, checksum
,
5066 ztest_dsl_prop_set_uint64(zd
->zd_name
, ZFS_PROP_COPIES
, 1,
5068 (void) rw_exit(&ztest_name_lock
);
5069 umem_free(od
, sizeof(ztest_od_t
));
5073 object
= od
[0].od_object
;
5074 blocksize
= od
[0].od_blocksize
;
5075 pattern
= zs
->zs_guid
^ dmu_objset_fsid_guid(os
);
5077 ASSERT(object
!= 0);
5079 tx
= dmu_tx_create(os
);
5080 dmu_tx_hold_write(tx
, object
, 0, copies
* blocksize
);
5081 txg
= ztest_tx_assign(tx
, TXG_WAIT
, FTAG
);
5083 (void) rw_exit(&ztest_name_lock
);
5084 umem_free(od
, sizeof(ztest_od_t
));
5089 * Write all the copies of our block.
5091 for (i
= 0; i
< copies
; i
++) {
5092 uint64_t offset
= i
* blocksize
;
5093 VERIFY(dmu_buf_hold(os
, object
, offset
, FTAG
, &db
,
5094 DMU_READ_NO_PREFETCH
) == 0);
5095 ASSERT(db
->db_offset
== offset
);
5096 ASSERT(db
->db_size
== blocksize
);
5097 ASSERT(ztest_pattern_match(db
->db_data
, db
->db_size
, pattern
) ||
5098 ztest_pattern_match(db
->db_data
, db
->db_size
, 0ULL));
5099 dmu_buf_will_fill(db
, tx
);
5100 ztest_pattern_set(db
->db_data
, db
->db_size
, pattern
);
5101 dmu_buf_rele(db
, FTAG
);
5105 txg_wait_synced(spa_get_dsl(spa
), txg
);
5108 * Find out what block we got.
5110 VERIFY(dmu_buf_hold(os
, object
, 0, FTAG
, &db
,
5111 DMU_READ_NO_PREFETCH
) == 0);
5112 blk
= *((dmu_buf_impl_t
*)db
)->db_blkptr
;
5113 dmu_buf_rele(db
, FTAG
);
5116 * Damage the block. Dedup-ditto will save us when we read it later.
5118 psize
= BP_GET_PSIZE(&blk
);
5119 buf
= zio_buf_alloc(psize
);
5120 ztest_pattern_set(buf
, psize
, ~pattern
);
5122 (void) zio_wait(zio_rewrite(NULL
, spa
, 0, &blk
,
5123 buf
, psize
, NULL
, NULL
, ZIO_PRIORITY_SYNC_WRITE
,
5124 ZIO_FLAG_CANFAIL
| ZIO_FLAG_INDUCE_DAMAGE
, NULL
));
5126 zio_buf_free(buf
, psize
);
5128 (void) rw_exit(&ztest_name_lock
);
5129 umem_free(od
, sizeof(ztest_od_t
));
5137 ztest_scrub(ztest_ds_t
*zd
, uint64_t id
)
5139 spa_t
*spa
= ztest_spa
;
5141 (void) spa_scan(spa
, POOL_SCAN_SCRUB
);
5142 (void) poll(NULL
, 0, 100); /* wait a moment, then force a restart */
5143 (void) spa_scan(spa
, POOL_SCAN_SCRUB
);
5147 * Change the guid for the pool.
5151 ztest_reguid(ztest_ds_t
*zd
, uint64_t id
)
5153 spa_t
*spa
= ztest_spa
;
5154 uint64_t orig
, load
;
5157 orig
= spa_guid(spa
);
5158 load
= spa_load_guid(spa
);
5160 (void) rw_enter(&ztest_name_lock
, RW_WRITER
);
5161 error
= spa_change_guid(spa
);
5162 (void) rw_exit(&ztest_name_lock
);
5167 if (ztest_opts
.zo_verbose
>= 4) {
5168 (void) printf("Changed guid old %llu -> %llu\n",
5169 (u_longlong_t
)orig
, (u_longlong_t
)spa_guid(spa
));
5172 VERIFY3U(orig
, !=, spa_guid(spa
));
5173 VERIFY3U(load
, ==, spa_load_guid(spa
));
5177 * Rename the pool to a different name and then rename it back.
5181 ztest_spa_rename(ztest_ds_t
*zd
, uint64_t id
)
5183 char *oldname
, *newname
;
5186 (void) rw_enter(&ztest_name_lock
, RW_WRITER
);
5188 oldname
= ztest_opts
.zo_pool
;
5189 newname
= umem_alloc(strlen(oldname
) + 5, UMEM_NOFAIL
);
5190 (void) strcpy(newname
, oldname
);
5191 (void) strcat(newname
, "_tmp");
5196 VERIFY3U(0, ==, spa_rename(oldname
, newname
));
5199 * Try to open it under the old name, which shouldn't exist
5201 VERIFY3U(ENOENT
, ==, spa_open(oldname
, &spa
, FTAG
));
5204 * Open it under the new name and make sure it's still the same spa_t.
5206 VERIFY3U(0, ==, spa_open(newname
, &spa
, FTAG
));
5208 ASSERT(spa
== ztest_spa
);
5209 spa_close(spa
, FTAG
);
5212 * Rename it back to the original
5214 VERIFY3U(0, ==, spa_rename(newname
, oldname
));
5217 * Make sure it can still be opened
5219 VERIFY3U(0, ==, spa_open(oldname
, &spa
, FTAG
));
5221 ASSERT(spa
== ztest_spa
);
5222 spa_close(spa
, FTAG
);
5224 umem_free(newname
, strlen(newname
) + 1);
5226 (void) rw_exit(&ztest_name_lock
);
5230 * Verify pool integrity by running zdb.
5233 ztest_run_zdb(char *pool
)
5241 bin
= umem_alloc(MAXPATHLEN
+ MAXNAMELEN
+ 20, UMEM_NOFAIL
);
5242 zdb
= umem_alloc(MAXPATHLEN
+ MAXNAMELEN
+ 20, UMEM_NOFAIL
);
5243 zbuf
= umem_alloc(1024, UMEM_NOFAIL
);
5245 VERIFY(realpath(getexecname(), bin
) != NULL
);
5246 if (strncmp(bin
, "/usr/sbin/ztest", 15) == 0) {
5247 strcpy(bin
, "/usr/sbin/zdb"); /* Installed */
5248 } else if (strncmp(bin
, "/sbin/ztest", 11) == 0) {
5249 strcpy(bin
, "/sbin/zdb"); /* Installed */
5251 strstr(bin
, "/ztest/")[0] = '\0'; /* In-tree */
5252 strcat(bin
, "/zdb/zdb");
5256 "%s -bcc%s%s -U %s %s",
5258 ztest_opts
.zo_verbose
>= 3 ? "s" : "",
5259 ztest_opts
.zo_verbose
>= 4 ? "v" : "",
5263 if (ztest_opts
.zo_verbose
>= 5)
5264 (void) printf("Executing %s\n", strstr(zdb
, "zdb "));
5266 fp
= popen(zdb
, "r");
5268 while (fgets(zbuf
, 1024, fp
) != NULL
)
5269 if (ztest_opts
.zo_verbose
>= 3)
5270 (void) printf("%s", zbuf
);
5272 status
= pclose(fp
);
5277 ztest_dump_core
= 0;
5278 if (WIFEXITED(status
))
5279 fatal(0, "'%s' exit code %d", zdb
, WEXITSTATUS(status
));
5281 fatal(0, "'%s' died with signal %d", zdb
, WTERMSIG(status
));
5283 umem_free(bin
, MAXPATHLEN
+ MAXNAMELEN
+ 20);
5284 umem_free(zdb
, MAXPATHLEN
+ MAXNAMELEN
+ 20);
5285 umem_free(zbuf
, 1024);
5289 ztest_walk_pool_directory(char *header
)
5293 if (ztest_opts
.zo_verbose
>= 6)
5294 (void) printf("%s\n", header
);
5296 mutex_enter(&spa_namespace_lock
);
5297 while ((spa
= spa_next(spa
)) != NULL
)
5298 if (ztest_opts
.zo_verbose
>= 6)
5299 (void) printf("\t%s\n", spa_name(spa
));
5300 mutex_exit(&spa_namespace_lock
);
5304 ztest_spa_import_export(char *oldname
, char *newname
)
5306 nvlist_t
*config
, *newconfig
;
5310 if (ztest_opts
.zo_verbose
>= 4) {
5311 (void) printf("import/export: old = %s, new = %s\n",
5316 * Clean up from previous runs.
5318 (void) spa_destroy(newname
);
5321 * Get the pool's configuration and guid.
5323 VERIFY3U(0, ==, spa_open(oldname
, &spa
, FTAG
));
5326 * Kick off a scrub to tickle scrub/export races.
5328 if (ztest_random(2) == 0)
5329 (void) spa_scan(spa
, POOL_SCAN_SCRUB
);
5331 pool_guid
= spa_guid(spa
);
5332 spa_close(spa
, FTAG
);
5334 ztest_walk_pool_directory("pools before export");
5339 VERIFY3U(0, ==, spa_export(oldname
, &config
, B_FALSE
, B_FALSE
));
5341 ztest_walk_pool_directory("pools after export");
5346 newconfig
= spa_tryimport(config
);
5347 ASSERT(newconfig
!= NULL
);
5348 nvlist_free(newconfig
);
5351 * Import it under the new name.
5353 VERIFY3U(0, ==, spa_import(newname
, config
, NULL
, 0));
5355 ztest_walk_pool_directory("pools after import");
5358 * Try to import it again -- should fail with EEXIST.
5360 VERIFY3U(EEXIST
, ==, spa_import(newname
, config
, NULL
, 0));
5363 * Try to import it under a different name -- should fail with EEXIST.
5365 VERIFY3U(EEXIST
, ==, spa_import(oldname
, config
, NULL
, 0));
5368 * Verify that the pool is no longer visible under the old name.
5370 VERIFY3U(ENOENT
, ==, spa_open(oldname
, &spa
, FTAG
));
5373 * Verify that we can open and close the pool using the new name.
5375 VERIFY3U(0, ==, spa_open(newname
, &spa
, FTAG
));
5376 ASSERT(pool_guid
== spa_guid(spa
));
5377 spa_close(spa
, FTAG
);
5379 nvlist_free(config
);
5383 ztest_resume(spa_t
*spa
)
5385 if (spa_suspended(spa
) && ztest_opts
.zo_verbose
>= 6)
5386 (void) printf("resuming from suspended state\n");
5387 spa_vdev_state_enter(spa
, SCL_NONE
);
5388 vdev_clear(spa
, NULL
);
5389 (void) spa_vdev_state_exit(spa
, NULL
, 0);
5390 (void) zio_resume(spa
);
5394 ztest_resume_thread(void *arg
)
5398 while (!ztest_exiting
) {
5399 if (spa_suspended(spa
))
5401 (void) poll(NULL
, 0, 100);
5413 ztest_deadman_alarm(int sig
)
5415 fatal(0, "failed to complete within %d seconds of deadline", GRACE
);
5420 ztest_execute(int test
, ztest_info_t
*zi
, uint64_t id
)
5422 ztest_ds_t
*zd
= &ztest_ds
[id
% ztest_opts
.zo_datasets
];
5423 ztest_shared_callstate_t
*zc
= ZTEST_GET_SHARED_CALLSTATE(test
);
5424 hrtime_t functime
= gethrtime();
5427 for (i
= 0; i
< zi
->zi_iters
; i
++)
5428 zi
->zi_func(zd
, id
);
5430 functime
= gethrtime() - functime
;
5432 atomic_add_64(&zc
->zc_count
, 1);
5433 atomic_add_64(&zc
->zc_time
, functime
);
5435 if (ztest_opts
.zo_verbose
>= 4) {
5437 (void) dladdr((void *)zi
->zi_func
, &dli
);
5438 (void) printf("%6.2f sec in %s\n",
5439 (double)functime
/ NANOSEC
, dli
.dli_sname
);
5444 ztest_thread(void *arg
)
5447 uint64_t id
= (uintptr_t)arg
;
5448 ztest_shared_t
*zs
= ztest_shared
;
5452 ztest_shared_callstate_t
*zc
;
5454 while ((now
= gethrtime()) < zs
->zs_thread_stop
) {
5456 * See if it's time to force a crash.
5458 if (now
> zs
->zs_thread_kill
)
5462 * If we're getting ENOSPC with some regularity, stop.
5464 if (zs
->zs_enospc_count
> 10)
5468 * Pick a random function to execute.
5470 rand
= ztest_random(ZTEST_FUNCS
);
5471 zi
= &ztest_info
[rand
];
5472 zc
= ZTEST_GET_SHARED_CALLSTATE(rand
);
5473 call_next
= zc
->zc_next
;
5475 if (now
>= call_next
&&
5476 atomic_cas_64(&zc
->zc_next
, call_next
, call_next
+
5477 ztest_random(2 * zi
->zi_interval
[0] + 1)) == call_next
) {
5478 ztest_execute(rand
, zi
, id
);
5488 ztest_dataset_name(char *dsname
, char *pool
, int d
)
5490 (void) snprintf(dsname
, MAXNAMELEN
, "%s/ds_%d", pool
, d
);
5494 ztest_dataset_destroy(int d
)
5496 char name
[MAXNAMELEN
];
5499 ztest_dataset_name(name
, ztest_opts
.zo_pool
, d
);
5501 if (ztest_opts
.zo_verbose
>= 3)
5502 (void) printf("Destroying %s to free up space\n", name
);
5505 * Cleanup any non-standard clones and snapshots. In general,
5506 * ztest thread t operates on dataset (t % zopt_datasets),
5507 * so there may be more than one thing to clean up.
5509 for (t
= d
; t
< ztest_opts
.zo_threads
;
5510 t
+= ztest_opts
.zo_datasets
)
5511 ztest_dsl_dataset_cleanup(name
, t
);
5513 (void) dmu_objset_find(name
, ztest_objset_destroy_cb
, NULL
,
5514 DS_FIND_SNAPSHOTS
| DS_FIND_CHILDREN
);
5518 ztest_dataset_dirobj_verify(ztest_ds_t
*zd
)
5520 uint64_t usedobjs
, dirobjs
, scratch
;
5523 * ZTEST_DIROBJ is the object directory for the entire dataset.
5524 * Therefore, the number of objects in use should equal the
5525 * number of ZTEST_DIROBJ entries, +1 for ZTEST_DIROBJ itself.
5526 * If not, we have an object leak.
5528 * Note that we can only check this in ztest_dataset_open(),
5529 * when the open-context and syncing-context values agree.
5530 * That's because zap_count() returns the open-context value,
5531 * while dmu_objset_space() returns the rootbp fill count.
5533 VERIFY3U(0, ==, zap_count(zd
->zd_os
, ZTEST_DIROBJ
, &dirobjs
));
5534 dmu_objset_space(zd
->zd_os
, &scratch
, &scratch
, &usedobjs
, &scratch
);
5535 ASSERT3U(dirobjs
+ 1, ==, usedobjs
);
5539 ztest_dataset_open(int d
)
5541 ztest_ds_t
*zd
= &ztest_ds
[d
];
5542 uint64_t committed_seq
= ZTEST_GET_SHARED_DS(d
)->zd_seq
;
5545 char name
[MAXNAMELEN
];
5548 ztest_dataset_name(name
, ztest_opts
.zo_pool
, d
);
5550 (void) rw_enter(&ztest_name_lock
, RW_READER
);
5552 error
= ztest_dataset_create(name
);
5553 if (error
== ENOSPC
) {
5554 (void) rw_exit(&ztest_name_lock
);
5555 ztest_record_enospc(FTAG
);
5558 ASSERT(error
== 0 || error
== EEXIST
);
5560 VERIFY0(dmu_objset_hold(name
, zd
, &os
));
5561 (void) rw_exit(&ztest_name_lock
);
5563 ztest_zd_init(zd
, ZTEST_GET_SHARED_DS(d
), os
);
5565 zilog
= zd
->zd_zilog
;
5567 if (zilog
->zl_header
->zh_claim_lr_seq
!= 0 &&
5568 zilog
->zl_header
->zh_claim_lr_seq
< committed_seq
)
5569 fatal(0, "missing log records: claimed %llu < committed %llu",
5570 zilog
->zl_header
->zh_claim_lr_seq
, committed_seq
);
5572 ztest_dataset_dirobj_verify(zd
);
5574 zil_replay(os
, zd
, ztest_replay_vector
);
5576 ztest_dataset_dirobj_verify(zd
);
5578 if (ztest_opts
.zo_verbose
>= 6)
5579 (void) printf("%s replay %llu blocks, %llu records, seq %llu\n",
5581 (u_longlong_t
)zilog
->zl_parse_blk_count
,
5582 (u_longlong_t
)zilog
->zl_parse_lr_count
,
5583 (u_longlong_t
)zilog
->zl_replaying_seq
);
5585 zilog
= zil_open(os
, ztest_get_data
);
5587 if (zilog
->zl_replaying_seq
!= 0 &&
5588 zilog
->zl_replaying_seq
< committed_seq
)
5589 fatal(0, "missing log records: replayed %llu < committed %llu",
5590 zilog
->zl_replaying_seq
, committed_seq
);
5596 ztest_dataset_close(int d
)
5598 ztest_ds_t
*zd
= &ztest_ds
[d
];
5600 zil_close(zd
->zd_zilog
);
5601 dmu_objset_rele(zd
->zd_os
, zd
);
5607 * Kick off threads to run tests on all datasets in parallel.
5610 ztest_run(ztest_shared_t
*zs
)
5615 kthread_t
*resume_thread
;
5620 ztest_exiting
= B_FALSE
;
5623 * Initialize parent/child shared state.
5625 mutex_init(&ztest_vdev_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
5626 rw_init(&ztest_name_lock
, NULL
, RW_DEFAULT
, NULL
);
5628 zs
->zs_thread_start
= gethrtime();
5629 zs
->zs_thread_stop
=
5630 zs
->zs_thread_start
+ ztest_opts
.zo_passtime
* NANOSEC
;
5631 zs
->zs_thread_stop
= MIN(zs
->zs_thread_stop
, zs
->zs_proc_stop
);
5632 zs
->zs_thread_kill
= zs
->zs_thread_stop
;
5633 if (ztest_random(100) < ztest_opts
.zo_killrate
) {
5634 zs
->zs_thread_kill
-=
5635 ztest_random(ztest_opts
.zo_passtime
* NANOSEC
);
5638 mutex_init(&zcl
.zcl_callbacks_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
5640 list_create(&zcl
.zcl_callbacks
, sizeof (ztest_cb_data_t
),
5641 offsetof(ztest_cb_data_t
, zcd_node
));
5646 kernel_init(FREAD
| FWRITE
);
5647 VERIFY(spa_open(ztest_opts
.zo_pool
, &spa
, FTAG
) == 0);
5648 spa
->spa_debug
= B_TRUE
;
5651 VERIFY3U(0, ==, dmu_objset_hold(ztest_opts
.zo_pool
, FTAG
, &os
));
5652 zs
->zs_guid
= dmu_objset_fsid_guid(os
);
5653 dmu_objset_rele(os
, FTAG
);
5655 spa
->spa_dedup_ditto
= 2 * ZIO_DEDUPDITTO_MIN
;
5658 * We don't expect the pool to suspend unless maxfaults == 0,
5659 * in which case ztest_fault_inject() temporarily takes away
5660 * the only valid replica.
5662 if (MAXFAULTS() == 0)
5663 spa
->spa_failmode
= ZIO_FAILURE_MODE_WAIT
;
5665 spa
->spa_failmode
= ZIO_FAILURE_MODE_PANIC
;
5668 * Create a thread to periodically resume suspended I/O.
5670 VERIFY3P((resume_thread
= zk_thread_create(NULL
, 0,
5671 (thread_func_t
)ztest_resume_thread
, spa
, TS_RUN
, NULL
, 0, 0,
5672 PTHREAD_CREATE_JOINABLE
)), !=, NULL
);
5676 * Set a deadman alarm to abort() if we hang.
5678 signal(SIGALRM
, ztest_deadman_alarm
);
5679 alarm((zs
->zs_thread_stop
- zs
->zs_thread_start
) / NANOSEC
+ GRACE
);
5683 * Verify that we can safely inquire about about any object,
5684 * whether it's allocated or not. To make it interesting,
5685 * we probe a 5-wide window around each power of two.
5686 * This hits all edge cases, including zero and the max.
5688 for (t
= 0; t
< 64; t
++) {
5689 for (d
= -5; d
<= 5; d
++) {
5690 error
= dmu_object_info(spa
->spa_meta_objset
,
5691 (1ULL << t
) + d
, NULL
);
5692 ASSERT(error
== 0 || error
== ENOENT
||
5698 * If we got any ENOSPC errors on the previous run, destroy something.
5700 if (zs
->zs_enospc_count
!= 0) {
5701 int d
= ztest_random(ztest_opts
.zo_datasets
);
5702 ztest_dataset_destroy(d
);
5704 zs
->zs_enospc_count
= 0;
5706 tid
= umem_zalloc(ztest_opts
.zo_threads
* sizeof (kt_did_t
),
5709 if (ztest_opts
.zo_verbose
>= 4)
5710 (void) printf("starting main threads...\n");
5713 * Kick off all the tests that run in parallel.
5715 for (t
= 0; t
< ztest_opts
.zo_threads
; t
++) {
5718 if (t
< ztest_opts
.zo_datasets
&&
5719 ztest_dataset_open(t
) != 0)
5722 VERIFY3P(thread
= zk_thread_create(NULL
, 0,
5723 (thread_func_t
)ztest_thread
,
5724 (void *)(uintptr_t)t
, TS_RUN
, NULL
, 0, 0,
5725 PTHREAD_CREATE_JOINABLE
), !=, NULL
);
5726 tid
[t
] = thread
->t_tid
;
5730 * Wait for all of the tests to complete. We go in reverse order
5731 * so we don't close datasets while threads are still using them.
5733 for (t
= ztest_opts
.zo_threads
- 1; t
>= 0; t
--) {
5734 thread_join(tid
[t
]);
5735 if (t
< ztest_opts
.zo_datasets
)
5736 ztest_dataset_close(t
);
5739 txg_wait_synced(spa_get_dsl(spa
), 0);
5741 zs
->zs_alloc
= metaslab_class_get_alloc(spa_normal_class(spa
));
5742 zs
->zs_space
= metaslab_class_get_space(spa_normal_class(spa
));
5744 umem_free(tid
, ztest_opts
.zo_threads
* sizeof (kt_did_t
));
5746 /* Kill the resume thread */
5747 ztest_exiting
= B_TRUE
;
5748 thread_join(resume_thread
->t_tid
);
5752 * Right before closing the pool, kick off a bunch of async I/O;
5753 * spa_close() should wait for it to complete.
5755 for (object
= 1; object
< 50; object
++)
5756 dmu_prefetch(spa
->spa_meta_objset
, object
, 0, 1ULL << 20);
5758 /* Verify that at least one commit cb was called in a timely fashion */
5759 if (zc_cb_counter
>= ZTEST_COMMIT_CB_MIN_REG
)
5760 VERIFY0(zc_min_txg_delay
);
5762 spa_close(spa
, FTAG
);
5765 * Verify that we can loop over all pools.
5767 mutex_enter(&spa_namespace_lock
);
5768 for (spa
= spa_next(NULL
); spa
!= NULL
; spa
= spa_next(spa
))
5769 if (ztest_opts
.zo_verbose
> 3)
5770 (void) printf("spa_next: found %s\n", spa_name(spa
));
5771 mutex_exit(&spa_namespace_lock
);
5774 * Verify that we can export the pool and reimport it under a
5777 if (ztest_random(2) == 0) {
5778 char name
[MAXNAMELEN
];
5779 (void) snprintf(name
, MAXNAMELEN
, "%s_import",
5780 ztest_opts
.zo_pool
);
5781 ztest_spa_import_export(ztest_opts
.zo_pool
, name
);
5782 ztest_spa_import_export(name
, ztest_opts
.zo_pool
);
5787 list_destroy(&zcl
.zcl_callbacks
);
5788 mutex_destroy(&zcl
.zcl_callbacks_lock
);
5789 rw_destroy(&ztest_name_lock
);
5790 mutex_destroy(&ztest_vdev_lock
);
5796 ztest_ds_t
*zd
= &ztest_ds
[0];
5800 if (ztest_opts
.zo_verbose
>= 3)
5801 (void) printf("testing spa_freeze()...\n");
5803 kernel_init(FREAD
| FWRITE
);
5804 VERIFY3U(0, ==, spa_open(ztest_opts
.zo_pool
, &spa
, FTAG
));
5805 VERIFY3U(0, ==, ztest_dataset_open(0));
5808 * Force the first log block to be transactionally allocated.
5809 * We have to do this before we freeze the pool -- otherwise
5810 * the log chain won't be anchored.
5812 while (BP_IS_HOLE(&zd
->zd_zilog
->zl_header
->zh_log
)) {
5813 ztest_dmu_object_alloc_free(zd
, 0);
5814 zil_commit(zd
->zd_zilog
, 0);
5817 txg_wait_synced(spa_get_dsl(spa
), 0);
5820 * Freeze the pool. This stops spa_sync() from doing anything,
5821 * so that the only way to record changes from now on is the ZIL.
5826 * Run tests that generate log records but don't alter the pool config
5827 * or depend on DSL sync tasks (snapshots, objset create/destroy, etc).
5828 * We do a txg_wait_synced() after each iteration to force the txg
5829 * to increase well beyond the last synced value in the uberblock.
5830 * The ZIL should be OK with that.
5832 while (ztest_random(10) != 0 &&
5833 numloops
++ < ztest_opts
.zo_maxloops
) {
5834 ztest_dmu_write_parallel(zd
, 0);
5835 ztest_dmu_object_alloc_free(zd
, 0);
5836 txg_wait_synced(spa_get_dsl(spa
), 0);
5840 * Commit all of the changes we just generated.
5842 zil_commit(zd
->zd_zilog
, 0);
5843 txg_wait_synced(spa_get_dsl(spa
), 0);
5846 * Close our dataset and close the pool.
5848 ztest_dataset_close(0);
5849 spa_close(spa
, FTAG
);
5853 * Open and close the pool and dataset to induce log replay.
5855 kernel_init(FREAD
| FWRITE
);
5856 VERIFY3U(0, ==, spa_open(ztest_opts
.zo_pool
, &spa
, FTAG
));
5857 ASSERT(spa_freeze_txg(spa
) == UINT64_MAX
);
5858 VERIFY3U(0, ==, ztest_dataset_open(0));
5859 ztest_dataset_close(0);
5861 spa
->spa_debug
= B_TRUE
;
5863 txg_wait_synced(spa_get_dsl(spa
), 0);
5864 ztest_reguid(NULL
, 0);
5866 spa_close(spa
, FTAG
);
5871 print_time(hrtime_t t
, char *timebuf
)
5873 hrtime_t s
= t
/ NANOSEC
;
5874 hrtime_t m
= s
/ 60;
5875 hrtime_t h
= m
/ 60;
5876 hrtime_t d
= h
/ 24;
5885 (void) sprintf(timebuf
,
5886 "%llud%02lluh%02llum%02llus", d
, h
, m
, s
);
5888 (void) sprintf(timebuf
, "%lluh%02llum%02llus", h
, m
, s
);
5890 (void) sprintf(timebuf
, "%llum%02llus", m
, s
);
5892 (void) sprintf(timebuf
, "%llus", s
);
5896 make_random_props(void)
5900 VERIFY(nvlist_alloc(&props
, NV_UNIQUE_NAME
, 0) == 0);
5901 if (ztest_random(2) == 0)
5903 VERIFY(nvlist_add_uint64(props
, "autoreplace", 1) == 0);
5909 * Create a storage pool with the given name and initial vdev size.
5910 * Then test spa_freeze() functionality.
5913 ztest_init(ztest_shared_t
*zs
)
5916 nvlist_t
*nvroot
, *props
;
5919 mutex_init(&ztest_vdev_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
5920 rw_init(&ztest_name_lock
, NULL
, RW_DEFAULT
, NULL
);
5922 kernel_init(FREAD
| FWRITE
);
5925 * Create the storage pool.
5927 (void) spa_destroy(ztest_opts
.zo_pool
);
5928 ztest_shared
->zs_vdev_next_leaf
= 0;
5930 zs
->zs_mirrors
= ztest_opts
.zo_mirrors
;
5931 nvroot
= make_vdev_root(NULL
, NULL
, NULL
, ztest_opts
.zo_vdev_size
, 0,
5932 0, ztest_opts
.zo_raidz
, zs
->zs_mirrors
, 1);
5933 props
= make_random_props();
5934 for (i
= 0; i
< SPA_FEATURES
; i
++) {
5936 VERIFY3S(-1, !=, asprintf(&buf
, "feature@%s",
5937 spa_feature_table
[i
].fi_uname
));
5938 VERIFY3U(0, ==, nvlist_add_uint64(props
, buf
, 0));
5941 VERIFY3U(0, ==, spa_create(ztest_opts
.zo_pool
, nvroot
, props
,
5943 nvlist_free(nvroot
);
5945 VERIFY3U(0, ==, spa_open(ztest_opts
.zo_pool
, &spa
, FTAG
));
5946 zs
->zs_metaslab_sz
=
5947 1ULL << spa
->spa_root_vdev
->vdev_child
[0]->vdev_ms_shift
;
5948 spa_close(spa
, FTAG
);
5952 ztest_run_zdb(ztest_opts
.zo_pool
);
5956 ztest_run_zdb(ztest_opts
.zo_pool
);
5958 rw_destroy(&ztest_name_lock
);
5959 mutex_destroy(&ztest_vdev_lock
);
5965 static char ztest_name_data
[] = "/tmp/ztest.data.XXXXXX";
5967 ztest_fd_data
= mkstemp(ztest_name_data
);
5968 ASSERT3S(ztest_fd_data
, >=, 0);
5969 (void) unlink(ztest_name_data
);
5973 shared_data_size(ztest_shared_hdr_t
*hdr
)
5977 size
= hdr
->zh_hdr_size
;
5978 size
+= hdr
->zh_opts_size
;
5979 size
+= hdr
->zh_size
;
5980 size
+= hdr
->zh_stats_size
* hdr
->zh_stats_count
;
5981 size
+= hdr
->zh_ds_size
* hdr
->zh_ds_count
;
5990 ztest_shared_hdr_t
*hdr
;
5992 hdr
= (void *)mmap(0, P2ROUNDUP(sizeof (*hdr
), getpagesize()),
5993 PROT_READ
| PROT_WRITE
, MAP_SHARED
, ztest_fd_data
, 0);
5994 ASSERT(hdr
!= MAP_FAILED
);
5996 VERIFY3U(0, ==, ftruncate(ztest_fd_data
, sizeof (ztest_shared_hdr_t
)));
5998 hdr
->zh_hdr_size
= sizeof (ztest_shared_hdr_t
);
5999 hdr
->zh_opts_size
= sizeof (ztest_shared_opts_t
);
6000 hdr
->zh_size
= sizeof (ztest_shared_t
);
6001 hdr
->zh_stats_size
= sizeof (ztest_shared_callstate_t
);
6002 hdr
->zh_stats_count
= ZTEST_FUNCS
;
6003 hdr
->zh_ds_size
= sizeof (ztest_shared_ds_t
);
6004 hdr
->zh_ds_count
= ztest_opts
.zo_datasets
;
6006 size
= shared_data_size(hdr
);
6007 VERIFY3U(0, ==, ftruncate(ztest_fd_data
, size
));
6009 (void) munmap((caddr_t
)hdr
, P2ROUNDUP(sizeof (*hdr
), getpagesize()));
6016 ztest_shared_hdr_t
*hdr
;
6019 hdr
= (void *)mmap(0, P2ROUNDUP(sizeof (*hdr
), getpagesize()),
6020 PROT_READ
, MAP_SHARED
, ztest_fd_data
, 0);
6021 ASSERT(hdr
!= MAP_FAILED
);
6023 size
= shared_data_size(hdr
);
6025 (void) munmap((caddr_t
)hdr
, P2ROUNDUP(sizeof (*hdr
), getpagesize()));
6026 hdr
= ztest_shared_hdr
= (void *)mmap(0, P2ROUNDUP(size
, getpagesize()),
6027 PROT_READ
| PROT_WRITE
, MAP_SHARED
, ztest_fd_data
, 0);
6028 ASSERT(hdr
!= MAP_FAILED
);
6029 buf
= (uint8_t *)hdr
;
6031 offset
= hdr
->zh_hdr_size
;
6032 ztest_shared_opts
= (void *)&buf
[offset
];
6033 offset
+= hdr
->zh_opts_size
;
6034 ztest_shared
= (void *)&buf
[offset
];
6035 offset
+= hdr
->zh_size
;
6036 ztest_shared_callstate
= (void *)&buf
[offset
];
6037 offset
+= hdr
->zh_stats_size
* hdr
->zh_stats_count
;
6038 ztest_shared_ds
= (void *)&buf
[offset
];
6042 exec_child(char *cmd
, char *libpath
, boolean_t ignorekill
, int *statusp
)
6046 char *cmdbuf
= NULL
;
6051 cmdbuf
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
6052 (void) strlcpy(cmdbuf
, getexecname(), MAXPATHLEN
);
6057 fatal(1, "fork failed");
6059 if (pid
== 0) { /* child */
6060 char *emptyargv
[2] = { cmd
, NULL
};
6061 char fd_data_str
[12];
6063 struct rlimit rl
= { 1024, 1024 };
6064 (void) setrlimit(RLIMIT_NOFILE
, &rl
);
6066 (void) close(ztest_fd_rand
);
6067 VERIFY(11 >= snprintf(fd_data_str
, 12, "%d", ztest_fd_data
));
6068 VERIFY(0 == setenv("ZTEST_FD_DATA", fd_data_str
, 1));
6070 (void) enable_extended_FILE_stdio(-1, -1);
6071 if (libpath
!= NULL
)
6072 VERIFY(0 == setenv("LD_LIBRARY_PATH", libpath
, 1));
6073 (void) execv(cmd
, emptyargv
);
6074 ztest_dump_core
= B_FALSE
;
6075 fatal(B_TRUE
, "exec failed: %s", cmd
);
6078 if (cmdbuf
!= NULL
) {
6079 umem_free(cmdbuf
, MAXPATHLEN
);
6083 while (waitpid(pid
, &status
, 0) != pid
)
6085 if (statusp
!= NULL
)
6088 if (WIFEXITED(status
)) {
6089 if (WEXITSTATUS(status
) != 0) {
6090 (void) fprintf(stderr
, "child exited with code %d\n",
6091 WEXITSTATUS(status
));
6095 } else if (WIFSIGNALED(status
)) {
6096 if (!ignorekill
|| WTERMSIG(status
) != SIGKILL
) {
6097 (void) fprintf(stderr
, "child died with signal %d\n",
6103 (void) fprintf(stderr
, "something strange happened to child\n");
6110 ztest_run_init(void)
6114 ztest_shared_t
*zs
= ztest_shared
;
6116 ASSERT(ztest_opts
.zo_init
!= 0);
6119 * Blow away any existing copy of zpool.cache
6121 (void) remove(spa_config_path
);
6124 * Create and initialize our storage pool.
6126 for (i
= 1; i
<= ztest_opts
.zo_init
; i
++) {
6127 bzero(zs
, sizeof (ztest_shared_t
));
6128 if (ztest_opts
.zo_verbose
>= 3 &&
6129 ztest_opts
.zo_init
!= 1) {
6130 (void) printf("ztest_init(), pass %d\n", i
);
6137 main(int argc
, char **argv
)
6145 ztest_shared_callstate_t
*zc
;
6152 char *fd_data_str
= getenv("ZTEST_FD_DATA");
6154 (void) setvbuf(stdout
, NULL
, _IOLBF
, 0);
6156 ztest_fd_rand
= open("/dev/urandom", O_RDONLY
);
6157 ASSERT3S(ztest_fd_rand
, >=, 0);
6160 dprintf_setup(&argc
, argv
);
6161 process_options(argc
, argv
);
6166 bcopy(&ztest_opts
, ztest_shared_opts
,
6167 sizeof (*ztest_shared_opts
));
6169 ztest_fd_data
= atoi(fd_data_str
);
6171 bcopy(ztest_shared_opts
, &ztest_opts
, sizeof (ztest_opts
));
6173 ASSERT3U(ztest_opts
.zo_datasets
, ==, ztest_shared_hdr
->zh_ds_count
);
6175 /* Override location of zpool.cache */
6176 VERIFY(asprintf((char **)&spa_config_path
, "%s/zpool.cache",
6177 ztest_opts
.zo_dir
) != -1);
6179 ztest_ds
= umem_alloc(ztest_opts
.zo_datasets
* sizeof (ztest_ds_t
),
6184 metaslab_gang_bang
= ztest_opts
.zo_metaslab_gang_bang
;
6185 metaslab_df_alloc_threshold
=
6186 zs
->zs_metaslab_df_alloc_threshold
;
6195 hasalt
= (strlen(ztest_opts
.zo_alt_ztest
) != 0);
6197 if (ztest_opts
.zo_verbose
>= 1) {
6198 (void) printf("%llu vdevs, %d datasets, %d threads,"
6199 " %llu seconds...\n",
6200 (u_longlong_t
)ztest_opts
.zo_vdevs
,
6201 ztest_opts
.zo_datasets
,
6202 ztest_opts
.zo_threads
,
6203 (u_longlong_t
)ztest_opts
.zo_time
);
6206 cmd
= umem_alloc(MAXNAMELEN
, UMEM_NOFAIL
);
6207 (void) strlcpy(cmd
, getexecname(), MAXNAMELEN
);
6209 zs
->zs_do_init
= B_TRUE
;
6210 if (strlen(ztest_opts
.zo_alt_ztest
) != 0) {
6211 if (ztest_opts
.zo_verbose
>= 1) {
6212 (void) printf("Executing older ztest for "
6213 "initialization: %s\n", ztest_opts
.zo_alt_ztest
);
6215 VERIFY(!exec_child(ztest_opts
.zo_alt_ztest
,
6216 ztest_opts
.zo_alt_libpath
, B_FALSE
, NULL
));
6218 VERIFY(!exec_child(NULL
, NULL
, B_FALSE
, NULL
));
6220 zs
->zs_do_init
= B_FALSE
;
6222 zs
->zs_proc_start
= gethrtime();
6223 zs
->zs_proc_stop
= zs
->zs_proc_start
+ ztest_opts
.zo_time
* NANOSEC
;
6225 for (f
= 0; f
< ZTEST_FUNCS
; f
++) {
6226 zi
= &ztest_info
[f
];
6227 zc
= ZTEST_GET_SHARED_CALLSTATE(f
);
6228 if (zs
->zs_proc_start
+ zi
->zi_interval
[0] > zs
->zs_proc_stop
)
6229 zc
->zc_next
= UINT64_MAX
;
6231 zc
->zc_next
= zs
->zs_proc_start
+
6232 ztest_random(2 * zi
->zi_interval
[0] + 1);
6236 * Run the tests in a loop. These tests include fault injection
6237 * to verify that self-healing data works, and forced crashes
6238 * to verify that we never lose on-disk consistency.
6240 while (gethrtime() < zs
->zs_proc_stop
) {
6245 * Initialize the workload counters for each function.
6247 for (f
= 0; f
< ZTEST_FUNCS
; f
++) {
6248 zc
= ZTEST_GET_SHARED_CALLSTATE(f
);
6253 /* Set the allocation switch size */
6254 zs
->zs_metaslab_df_alloc_threshold
=
6255 ztest_random(zs
->zs_metaslab_sz
/ 4) + 1;
6257 if (!hasalt
|| ztest_random(2) == 0) {
6258 if (hasalt
&& ztest_opts
.zo_verbose
>= 1) {
6259 (void) printf("Executing newer ztest: %s\n",
6263 killed
= exec_child(cmd
, NULL
, B_TRUE
, &status
);
6265 if (hasalt
&& ztest_opts
.zo_verbose
>= 1) {
6266 (void) printf("Executing older ztest: %s\n",
6267 ztest_opts
.zo_alt_ztest
);
6270 killed
= exec_child(ztest_opts
.zo_alt_ztest
,
6271 ztest_opts
.zo_alt_libpath
, B_TRUE
, &status
);
6278 if (ztest_opts
.zo_verbose
>= 1) {
6279 hrtime_t now
= gethrtime();
6281 now
= MIN(now
, zs
->zs_proc_stop
);
6282 print_time(zs
->zs_proc_stop
- now
, timebuf
);
6283 nicenum(zs
->zs_space
, numbuf
);
6285 (void) printf("Pass %3d, %8s, %3llu ENOSPC, "
6286 "%4.1f%% of %5s used, %3.0f%% done, %8s to go\n",
6288 WIFEXITED(status
) ? "Complete" : "SIGKILL",
6289 (u_longlong_t
)zs
->zs_enospc_count
,
6290 100.0 * zs
->zs_alloc
/ zs
->zs_space
,
6292 100.0 * (now
- zs
->zs_proc_start
) /
6293 (ztest_opts
.zo_time
* NANOSEC
), timebuf
);
6296 if (ztest_opts
.zo_verbose
>= 2) {
6297 (void) printf("\nWorkload summary:\n\n");
6298 (void) printf("%7s %9s %s\n",
6299 "Calls", "Time", "Function");
6300 (void) printf("%7s %9s %s\n",
6301 "-----", "----", "--------");
6302 for (f
= 0; f
< ZTEST_FUNCS
; f
++) {
6305 zi
= &ztest_info
[f
];
6306 zc
= ZTEST_GET_SHARED_CALLSTATE(f
);
6307 print_time(zc
->zc_time
, timebuf
);
6308 (void) dladdr((void *)zi
->zi_func
, &dli
);
6309 (void) printf("%7llu %9s %s\n",
6310 (u_longlong_t
)zc
->zc_count
, timebuf
,
6313 (void) printf("\n");
6317 * It's possible that we killed a child during a rename test,
6318 * in which case we'll have a 'ztest_tmp' pool lying around
6319 * instead of 'ztest'. Do a blind rename in case this happened.
6322 if (spa_open(ztest_opts
.zo_pool
, &spa
, FTAG
) == 0) {
6323 spa_close(spa
, FTAG
);
6325 char tmpname
[MAXNAMELEN
];
6327 kernel_init(FREAD
| FWRITE
);
6328 (void) snprintf(tmpname
, sizeof (tmpname
), "%s_tmp",
6329 ztest_opts
.zo_pool
);
6330 (void) spa_rename(tmpname
, ztest_opts
.zo_pool
);
6334 ztest_run_zdb(ztest_opts
.zo_pool
);
6337 if (ztest_opts
.zo_verbose
>= 1) {
6339 (void) printf("%d runs of older ztest: %s\n", older
,
6340 ztest_opts
.zo_alt_ztest
);
6341 (void) printf("%d runs of newer ztest: %s\n", newer
,
6344 (void) printf("%d killed, %d completed, %.0f%% kill rate\n",
6345 kills
, iters
- kills
, (100.0 * kills
) / MAX(1, iters
));
6348 umem_free(cmd
, MAXNAMELEN
);