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_destroy.h>
110 #include <sys/dsl_scan.h>
111 #include <sys/zio_checksum.h>
112 #include <sys/refcount.h>
113 #include <sys/zfeature.h>
114 #include <sys/dsl_userhold.h>
116 #include <stdio_ext.h>
124 #include <sys/fs/zfs.h>
125 #include <libnvpair.h>
127 static int ztest_fd_data
= -1;
128 static int ztest_fd_rand
= -1;
130 typedef struct ztest_shared_hdr
{
131 uint64_t zh_hdr_size
;
132 uint64_t zh_opts_size
;
134 uint64_t zh_stats_size
;
135 uint64_t zh_stats_count
;
137 uint64_t zh_ds_count
;
138 } ztest_shared_hdr_t
;
140 static ztest_shared_hdr_t
*ztest_shared_hdr
;
142 typedef struct ztest_shared_opts
{
143 char zo_pool
[MAXNAMELEN
];
144 char zo_dir
[MAXNAMELEN
];
145 char zo_alt_ztest
[MAXNAMELEN
];
146 char zo_alt_libpath
[MAXNAMELEN
];
148 uint64_t zo_vdevtime
;
156 uint64_t zo_passtime
;
157 uint64_t zo_killrate
;
161 uint64_t zo_maxloops
;
162 uint64_t zo_metaslab_gang_bang
;
163 } ztest_shared_opts_t
;
165 static const ztest_shared_opts_t ztest_opts_defaults
= {
166 .zo_pool
= { 'z', 't', 'e', 's', 't', '\0' },
167 .zo_dir
= { '/', 't', 'm', 'p', '\0' },
168 .zo_alt_ztest
= { '\0' },
169 .zo_alt_libpath
= { '\0' },
171 .zo_ashift
= SPA_MINBLOCKSHIFT
,
174 .zo_raidz_parity
= 1,
175 .zo_vdev_size
= SPA_MINDEVSIZE
,
178 .zo_passtime
= 60, /* 60 seconds */
179 .zo_killrate
= 70, /* 70% kill rate */
182 .zo_time
= 300, /* 5 minutes */
183 .zo_maxloops
= 50, /* max loops during spa_freeze() */
184 .zo_metaslab_gang_bang
= 32 << 10
187 extern uint64_t metaslab_gang_bang
;
188 extern uint64_t metaslab_df_alloc_threshold
;
190 static ztest_shared_opts_t
*ztest_shared_opts
;
191 static ztest_shared_opts_t ztest_opts
;
193 typedef struct ztest_shared_ds
{
197 static ztest_shared_ds_t
*ztest_shared_ds
;
198 #define ZTEST_GET_SHARED_DS(d) (&ztest_shared_ds[d])
200 #define BT_MAGIC 0x123456789abcdefULL
201 #define MAXFAULTS() \
202 (MAX(zs->zs_mirrors, 1) * (ztest_opts.zo_raidz_parity + 1) - 1)
206 ZTEST_IO_WRITE_PATTERN
,
207 ZTEST_IO_WRITE_ZEROES
,
213 typedef struct ztest_block_tag
{
223 typedef struct bufwad
{
230 * XXX -- fix zfs range locks to be generic so we can use them here.
252 #define ZTEST_RANGE_LOCKS 64
253 #define ZTEST_OBJECT_LOCKS 64
256 * Object descriptor. Used as a template for object lookup/create/remove.
258 typedef struct ztest_od
{
261 dmu_object_type_t od_type
;
262 dmu_object_type_t od_crtype
;
263 uint64_t od_blocksize
;
264 uint64_t od_crblocksize
;
267 char od_name
[MAXNAMELEN
];
273 typedef struct ztest_ds
{
274 ztest_shared_ds_t
*zd_shared
;
276 krwlock_t zd_zilog_lock
;
278 ztest_od_t
*zd_od
; /* debugging aid */
279 char zd_name
[MAXNAMELEN
];
280 kmutex_t zd_dirobj_lock
;
281 rll_t zd_object_lock
[ZTEST_OBJECT_LOCKS
];
282 rll_t zd_range_lock
[ZTEST_RANGE_LOCKS
];
286 * Per-iteration state.
288 typedef void ztest_func_t(ztest_ds_t
*zd
, uint64_t id
);
290 typedef struct ztest_info
{
291 ztest_func_t
*zi_func
; /* test function */
292 uint64_t zi_iters
; /* iterations per execution */
293 uint64_t *zi_interval
; /* execute every <interval> seconds */
296 typedef struct ztest_shared_callstate
{
297 uint64_t zc_count
; /* per-pass count */
298 uint64_t zc_time
; /* per-pass time */
299 uint64_t zc_next
; /* next time to call this function */
300 } ztest_shared_callstate_t
;
302 static ztest_shared_callstate_t
*ztest_shared_callstate
;
303 #define ZTEST_GET_SHARED_CALLSTATE(c) (&ztest_shared_callstate[c])
306 * Note: these aren't static because we want dladdr() to work.
308 ztest_func_t ztest_dmu_read_write
;
309 ztest_func_t ztest_dmu_write_parallel
;
310 ztest_func_t ztest_dmu_object_alloc_free
;
311 ztest_func_t ztest_dmu_commit_callbacks
;
312 ztest_func_t ztest_zap
;
313 ztest_func_t ztest_zap_parallel
;
314 ztest_func_t ztest_zil_commit
;
315 ztest_func_t ztest_zil_remount
;
316 ztest_func_t ztest_dmu_read_write_zcopy
;
317 ztest_func_t ztest_dmu_objset_create_destroy
;
318 ztest_func_t ztest_dmu_prealloc
;
319 ztest_func_t ztest_fzap
;
320 ztest_func_t ztest_dmu_snapshot_create_destroy
;
321 ztest_func_t ztest_dsl_prop_get_set
;
322 ztest_func_t ztest_spa_prop_get_set
;
323 ztest_func_t ztest_spa_create_destroy
;
324 ztest_func_t ztest_fault_inject
;
325 ztest_func_t ztest_ddt_repair
;
326 ztest_func_t ztest_dmu_snapshot_hold
;
327 ztest_func_t ztest_spa_rename
;
328 ztest_func_t ztest_scrub
;
329 ztest_func_t ztest_dsl_dataset_promote_busy
;
330 ztest_func_t ztest_vdev_attach_detach
;
331 ztest_func_t ztest_vdev_LUN_growth
;
332 ztest_func_t ztest_vdev_add_remove
;
333 ztest_func_t ztest_vdev_aux_add_remove
;
334 ztest_func_t ztest_split_pool
;
335 ztest_func_t ztest_reguid
;
336 ztest_func_t ztest_spa_upgrade
;
338 uint64_t zopt_always
= 0ULL * NANOSEC
; /* all the time */
339 uint64_t zopt_incessant
= 1ULL * NANOSEC
/ 10; /* every 1/10 second */
340 uint64_t zopt_often
= 1ULL * NANOSEC
; /* every second */
341 uint64_t zopt_sometimes
= 10ULL * NANOSEC
; /* every 10 seconds */
342 uint64_t zopt_rarely
= 60ULL * NANOSEC
; /* every 60 seconds */
344 ztest_info_t ztest_info
[] = {
345 { ztest_dmu_read_write
, 1, &zopt_always
},
346 { ztest_dmu_write_parallel
, 10, &zopt_always
},
347 { ztest_dmu_object_alloc_free
, 1, &zopt_always
},
348 { ztest_dmu_commit_callbacks
, 1, &zopt_always
},
349 { ztest_zap
, 30, &zopt_always
},
350 { ztest_zap_parallel
, 100, &zopt_always
},
351 { ztest_split_pool
, 1, &zopt_always
},
352 { ztest_zil_commit
, 1, &zopt_incessant
},
353 { ztest_zil_remount
, 1, &zopt_sometimes
},
354 { ztest_dmu_read_write_zcopy
, 1, &zopt_often
},
355 { ztest_dmu_objset_create_destroy
, 1, &zopt_often
},
356 { ztest_dsl_prop_get_set
, 1, &zopt_often
},
357 { ztest_spa_prop_get_set
, 1, &zopt_sometimes
},
359 { ztest_dmu_prealloc
, 1, &zopt_sometimes
},
361 { ztest_fzap
, 1, &zopt_sometimes
},
362 { ztest_dmu_snapshot_create_destroy
, 1, &zopt_sometimes
},
363 { ztest_spa_create_destroy
, 1, &zopt_sometimes
},
364 { ztest_fault_inject
, 1, &zopt_sometimes
},
365 { ztest_ddt_repair
, 1, &zopt_sometimes
},
366 { ztest_dmu_snapshot_hold
, 1, &zopt_sometimes
},
367 { ztest_reguid
, 1, &zopt_sometimes
},
368 { ztest_spa_rename
, 1, &zopt_rarely
},
369 { ztest_scrub
, 1, &zopt_rarely
},
370 { ztest_spa_upgrade
, 1, &zopt_rarely
},
371 { ztest_dsl_dataset_promote_busy
, 1, &zopt_rarely
},
372 { ztest_vdev_attach_detach
, 1, &zopt_sometimes
},
373 { ztest_vdev_LUN_growth
, 1, &zopt_rarely
},
374 { ztest_vdev_add_remove
, 1,
375 &ztest_opts
.zo_vdevtime
},
376 { ztest_vdev_aux_add_remove
, 1,
377 &ztest_opts
.zo_vdevtime
},
380 #define ZTEST_FUNCS (sizeof (ztest_info) / sizeof (ztest_info_t))
383 * The following struct is used to hold a list of uncalled commit callbacks.
384 * The callbacks are ordered by txg number.
386 typedef struct ztest_cb_list
{
387 kmutex_t zcl_callbacks_lock
;
388 list_t zcl_callbacks
;
392 * Stuff we need to share writably between parent and child.
394 typedef struct ztest_shared
{
395 boolean_t zs_do_init
;
396 hrtime_t zs_proc_start
;
397 hrtime_t zs_proc_stop
;
398 hrtime_t zs_thread_start
;
399 hrtime_t zs_thread_stop
;
400 hrtime_t zs_thread_kill
;
401 uint64_t zs_enospc_count
;
402 uint64_t zs_vdev_next_leaf
;
403 uint64_t zs_vdev_aux
;
408 uint64_t zs_metaslab_sz
;
409 uint64_t zs_metaslab_df_alloc_threshold
;
413 #define ID_PARALLEL -1ULL
415 static char ztest_dev_template
[] = "%s/%s.%llua";
416 static char ztest_aux_template
[] = "%s/%s.%s.%llu";
417 ztest_shared_t
*ztest_shared
;
419 static spa_t
*ztest_spa
= NULL
;
420 static ztest_ds_t
*ztest_ds
;
422 static kmutex_t ztest_vdev_lock
;
425 * The ztest_name_lock protects the pool and dataset namespace used by
426 * the individual tests. To modify the namespace, consumers must grab
427 * this lock as writer. Grabbing the lock as reader will ensure that the
428 * namespace does not change while the lock is held.
430 static krwlock_t ztest_name_lock
;
432 static boolean_t ztest_dump_core
= B_TRUE
;
433 static boolean_t ztest_exiting
;
435 /* Global commit callback list */
436 static ztest_cb_list_t zcl
;
437 /* Commit cb delay */
438 static uint64_t zc_min_txg_delay
= UINT64_MAX
;
439 static int zc_cb_counter
= 0;
442 * Minimum number of commit callbacks that need to be registered for us to check
443 * whether the minimum txg delay is acceptable.
445 #define ZTEST_COMMIT_CB_MIN_REG 100
448 * If a number of txgs equal to this threshold have been created after a commit
449 * callback has been registered but not called, then we assume there is an
450 * implementation bug.
452 #define ZTEST_COMMIT_CB_THRESH (TXG_CONCURRENT_STATES + 1000)
454 extern uint64_t metaslab_gang_bang
;
455 extern uint64_t metaslab_df_alloc_threshold
;
458 ZTEST_META_DNODE
= 0,
463 static void usage(boolean_t
) __NORETURN
;
466 * These libumem hooks provide a reasonable set of defaults for the allocator's
467 * debugging facilities.
470 _umem_debug_init(void)
472 return ("default,verbose"); /* $UMEM_DEBUG setting */
476 _umem_logging_init(void)
478 return ("fail,contents"); /* $UMEM_LOGGING setting */
481 #define FATAL_MSG_SZ 1024
486 fatal(int do_perror
, char *message
, ...)
489 int save_errno
= errno
;
492 (void) fflush(stdout
);
493 buf
= umem_alloc(FATAL_MSG_SZ
, UMEM_NOFAIL
);
495 va_start(args
, message
);
496 (void) sprintf(buf
, "ztest: ");
498 (void) vsprintf(buf
+ strlen(buf
), message
, args
);
501 (void) snprintf(buf
+ strlen(buf
), FATAL_MSG_SZ
- strlen(buf
),
502 ": %s", strerror(save_errno
));
504 (void) fprintf(stderr
, "%s\n", buf
);
505 fatal_msg
= buf
; /* to ease debugging */
512 str2shift(const char *buf
)
514 const char *ends
= "BKMGTPEZ";
519 for (i
= 0; i
< strlen(ends
); i
++) {
520 if (toupper(buf
[0]) == ends
[i
])
523 if (i
== strlen(ends
)) {
524 (void) fprintf(stderr
, "ztest: invalid bytes suffix: %s\n",
528 if (buf
[1] == '\0' || (toupper(buf
[1]) == 'B' && buf
[2] == '\0')) {
531 (void) fprintf(stderr
, "ztest: invalid bytes suffix: %s\n", buf
);
537 nicenumtoull(const char *buf
)
542 val
= strtoull(buf
, &end
, 0);
544 (void) fprintf(stderr
, "ztest: bad numeric value: %s\n", buf
);
546 } else if (end
[0] == '.') {
547 double fval
= strtod(buf
, &end
);
548 fval
*= pow(2, str2shift(end
));
549 if (fval
> UINT64_MAX
) {
550 (void) fprintf(stderr
, "ztest: value too large: %s\n",
554 val
= (uint64_t)fval
;
556 int shift
= str2shift(end
);
557 if (shift
>= 64 || (val
<< shift
) >> shift
!= val
) {
558 (void) fprintf(stderr
, "ztest: value too large: %s\n",
568 usage(boolean_t requested
)
570 const ztest_shared_opts_t
*zo
= &ztest_opts_defaults
;
572 char nice_vdev_size
[10];
573 char nice_gang_bang
[10];
574 FILE *fp
= requested
? stdout
: stderr
;
576 nicenum(zo
->zo_vdev_size
, nice_vdev_size
);
577 nicenum(zo
->zo_metaslab_gang_bang
, nice_gang_bang
);
579 (void) fprintf(fp
, "Usage: %s\n"
580 "\t[-v vdevs (default: %llu)]\n"
581 "\t[-s size_of_each_vdev (default: %s)]\n"
582 "\t[-a alignment_shift (default: %d)] use 0 for random\n"
583 "\t[-m mirror_copies (default: %d)]\n"
584 "\t[-r raidz_disks (default: %d)]\n"
585 "\t[-R raidz_parity (default: %d)]\n"
586 "\t[-d datasets (default: %d)]\n"
587 "\t[-t threads (default: %d)]\n"
588 "\t[-g gang_block_threshold (default: %s)]\n"
589 "\t[-i init_count (default: %d)] initialize pool i times\n"
590 "\t[-k kill_percentage (default: %llu%%)]\n"
591 "\t[-p pool_name (default: %s)]\n"
592 "\t[-f dir (default: %s)] file directory for vdev files\n"
593 "\t[-V] verbose (use multiple times for ever more blather)\n"
594 "\t[-E] use existing pool instead of creating new one\n"
595 "\t[-T time (default: %llu sec)] total run time\n"
596 "\t[-F freezeloops (default: %llu)] max loops in spa_freeze()\n"
597 "\t[-P passtime (default: %llu sec)] time per pass\n"
598 "\t[-B alt_ztest (default: <none>)] alternate ztest path\n"
599 "\t[-h] (print help)\n"
602 (u_longlong_t
)zo
->zo_vdevs
, /* -v */
603 nice_vdev_size
, /* -s */
604 zo
->zo_ashift
, /* -a */
605 zo
->zo_mirrors
, /* -m */
606 zo
->zo_raidz
, /* -r */
607 zo
->zo_raidz_parity
, /* -R */
608 zo
->zo_datasets
, /* -d */
609 zo
->zo_threads
, /* -t */
610 nice_gang_bang
, /* -g */
611 zo
->zo_init
, /* -i */
612 (u_longlong_t
)zo
->zo_killrate
, /* -k */
613 zo
->zo_pool
, /* -p */
615 (u_longlong_t
)zo
->zo_time
, /* -T */
616 (u_longlong_t
)zo
->zo_maxloops
, /* -F */
617 (u_longlong_t
)zo
->zo_passtime
);
618 exit(requested
? 0 : 1);
622 process_options(int argc
, char **argv
)
625 ztest_shared_opts_t
*zo
= &ztest_opts
;
629 char altdir
[MAXNAMELEN
] = { 0 };
631 bcopy(&ztest_opts_defaults
, zo
, sizeof (*zo
));
633 while ((opt
= getopt(argc
, argv
,
634 "v:s:a:m:r:R:d:t:g:i:k:p:f:VET:P:hF:B:")) != EOF
) {
651 value
= nicenumtoull(optarg
);
655 zo
->zo_vdevs
= value
;
658 zo
->zo_vdev_size
= MAX(SPA_MINDEVSIZE
, value
);
661 zo
->zo_ashift
= value
;
664 zo
->zo_mirrors
= value
;
667 zo
->zo_raidz
= MAX(1, value
);
670 zo
->zo_raidz_parity
= MIN(MAX(value
, 1), 3);
673 zo
->zo_datasets
= MAX(1, value
);
676 zo
->zo_threads
= MAX(1, value
);
679 zo
->zo_metaslab_gang_bang
= MAX(SPA_MINBLOCKSIZE
<< 1,
686 zo
->zo_killrate
= value
;
689 (void) strlcpy(zo
->zo_pool
, optarg
,
690 sizeof (zo
->zo_pool
));
693 path
= realpath(optarg
, NULL
);
695 (void) fprintf(stderr
, "error: %s: %s\n",
696 optarg
, strerror(errno
));
699 (void) strlcpy(zo
->zo_dir
, path
,
700 sizeof (zo
->zo_dir
));
713 zo
->zo_passtime
= MAX(1, value
);
716 zo
->zo_maxloops
= MAX(1, value
);
719 (void) strlcpy(altdir
, optarg
, sizeof (altdir
));
731 zo
->zo_raidz_parity
= MIN(zo
->zo_raidz_parity
, zo
->zo_raidz
- 1);
734 (zo
->zo_vdevs
> 0 ? zo
->zo_time
* NANOSEC
/ zo
->zo_vdevs
:
737 if (strlen(altdir
) > 0) {
745 cmd
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
746 realaltdir
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
748 VERIFY(NULL
!= realpath(getexecname(), cmd
));
749 if (0 != access(altdir
, F_OK
)) {
750 ztest_dump_core
= B_FALSE
;
751 fatal(B_TRUE
, "invalid alternate ztest path: %s",
754 VERIFY(NULL
!= realpath(altdir
, realaltdir
));
757 * 'cmd' should be of the form "<anything>/usr/bin/<isa>/ztest".
758 * We want to extract <isa> to determine if we should use
759 * 32 or 64 bit binaries.
761 bin
= strstr(cmd
, "/usr/bin/");
762 ztest
= strstr(bin
, "/ztest");
764 isalen
= ztest
- isa
;
765 (void) snprintf(zo
->zo_alt_ztest
, sizeof (zo
->zo_alt_ztest
),
766 "%s/usr/bin/%.*s/ztest", realaltdir
, isalen
, isa
);
767 (void) snprintf(zo
->zo_alt_libpath
, sizeof (zo
->zo_alt_libpath
),
768 "%s/usr/lib/%.*s", realaltdir
, isalen
, isa
);
770 if (0 != access(zo
->zo_alt_ztest
, X_OK
)) {
771 ztest_dump_core
= B_FALSE
;
772 fatal(B_TRUE
, "invalid alternate ztest: %s",
774 } else if (0 != access(zo
->zo_alt_libpath
, X_OK
)) {
775 ztest_dump_core
= B_FALSE
;
776 fatal(B_TRUE
, "invalid alternate lib directory %s",
780 umem_free(cmd
, MAXPATHLEN
);
781 umem_free(realaltdir
, MAXPATHLEN
);
786 ztest_kill(ztest_shared_t
*zs
)
788 zs
->zs_alloc
= metaslab_class_get_alloc(spa_normal_class(ztest_spa
));
789 zs
->zs_space
= metaslab_class_get_space(spa_normal_class(ztest_spa
));
790 (void) kill(getpid(), SIGKILL
);
794 ztest_random(uint64_t range
)
798 ASSERT3S(ztest_fd_rand
, >=, 0);
803 if (read(ztest_fd_rand
, &r
, sizeof (r
)) != sizeof (r
))
804 fatal(1, "short read from /dev/urandom");
811 ztest_record_enospc(const char *s
)
813 ztest_shared
->zs_enospc_count
++;
817 ztest_get_ashift(void)
819 if (ztest_opts
.zo_ashift
== 0)
820 return (SPA_MINBLOCKSHIFT
+ ztest_random(3));
821 return (ztest_opts
.zo_ashift
);
825 make_vdev_file(char *path
, char *aux
, char *pool
, size_t size
, uint64_t ashift
)
831 pathbuf
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
834 ashift
= ztest_get_ashift();
840 vdev
= ztest_shared
->zs_vdev_aux
;
841 (void) snprintf(path
, MAXPATHLEN
,
842 ztest_aux_template
, ztest_opts
.zo_dir
,
843 pool
== NULL
? ztest_opts
.zo_pool
: pool
,
846 vdev
= ztest_shared
->zs_vdev_next_leaf
++;
847 (void) snprintf(path
, MAXPATHLEN
,
848 ztest_dev_template
, ztest_opts
.zo_dir
,
849 pool
== NULL
? ztest_opts
.zo_pool
: pool
, vdev
);
854 int fd
= open(path
, O_RDWR
| O_CREAT
| O_TRUNC
, 0666);
856 fatal(1, "can't open %s", path
);
857 if (ftruncate(fd
, size
) != 0)
858 fatal(1, "can't ftruncate %s", path
);
862 VERIFY(nvlist_alloc(&file
, NV_UNIQUE_NAME
, 0) == 0);
863 VERIFY(nvlist_add_string(file
, ZPOOL_CONFIG_TYPE
, VDEV_TYPE_FILE
) == 0);
864 VERIFY(nvlist_add_string(file
, ZPOOL_CONFIG_PATH
, path
) == 0);
865 VERIFY(nvlist_add_uint64(file
, ZPOOL_CONFIG_ASHIFT
, ashift
) == 0);
866 umem_free(pathbuf
, MAXPATHLEN
);
872 make_vdev_raidz(char *path
, char *aux
, char *pool
, size_t size
,
873 uint64_t ashift
, int r
)
875 nvlist_t
*raidz
, **child
;
879 return (make_vdev_file(path
, aux
, pool
, size
, ashift
));
880 child
= umem_alloc(r
* sizeof (nvlist_t
*), UMEM_NOFAIL
);
882 for (c
= 0; c
< r
; c
++)
883 child
[c
] = make_vdev_file(path
, aux
, pool
, size
, ashift
);
885 VERIFY(nvlist_alloc(&raidz
, NV_UNIQUE_NAME
, 0) == 0);
886 VERIFY(nvlist_add_string(raidz
, ZPOOL_CONFIG_TYPE
,
887 VDEV_TYPE_RAIDZ
) == 0);
888 VERIFY(nvlist_add_uint64(raidz
, ZPOOL_CONFIG_NPARITY
,
889 ztest_opts
.zo_raidz_parity
) == 0);
890 VERIFY(nvlist_add_nvlist_array(raidz
, ZPOOL_CONFIG_CHILDREN
,
893 for (c
= 0; c
< r
; c
++)
894 nvlist_free(child
[c
]);
896 umem_free(child
, r
* sizeof (nvlist_t
*));
902 make_vdev_mirror(char *path
, char *aux
, char *pool
, size_t size
,
903 uint64_t ashift
, int r
, int m
)
905 nvlist_t
*mirror
, **child
;
909 return (make_vdev_raidz(path
, aux
, pool
, size
, ashift
, r
));
911 child
= umem_alloc(m
* sizeof (nvlist_t
*), UMEM_NOFAIL
);
913 for (c
= 0; c
< m
; c
++)
914 child
[c
] = make_vdev_raidz(path
, aux
, pool
, size
, ashift
, r
);
916 VERIFY(nvlist_alloc(&mirror
, NV_UNIQUE_NAME
, 0) == 0);
917 VERIFY(nvlist_add_string(mirror
, ZPOOL_CONFIG_TYPE
,
918 VDEV_TYPE_MIRROR
) == 0);
919 VERIFY(nvlist_add_nvlist_array(mirror
, ZPOOL_CONFIG_CHILDREN
,
922 for (c
= 0; c
< m
; c
++)
923 nvlist_free(child
[c
]);
925 umem_free(child
, m
* sizeof (nvlist_t
*));
931 make_vdev_root(char *path
, char *aux
, char *pool
, size_t size
, uint64_t ashift
,
932 int log
, int r
, int m
, int t
)
934 nvlist_t
*root
, **child
;
939 child
= umem_alloc(t
* sizeof (nvlist_t
*), UMEM_NOFAIL
);
941 for (c
= 0; c
< t
; c
++) {
942 child
[c
] = make_vdev_mirror(path
, aux
, pool
, size
, ashift
,
944 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_IS_LOG
,
948 VERIFY(nvlist_alloc(&root
, NV_UNIQUE_NAME
, 0) == 0);
949 VERIFY(nvlist_add_string(root
, ZPOOL_CONFIG_TYPE
, VDEV_TYPE_ROOT
) == 0);
950 VERIFY(nvlist_add_nvlist_array(root
, aux
? aux
: ZPOOL_CONFIG_CHILDREN
,
953 for (c
= 0; c
< t
; c
++)
954 nvlist_free(child
[c
]);
956 umem_free(child
, t
* sizeof (nvlist_t
*));
962 * Find a random spa version. Returns back a random spa version in the
963 * range [initial_version, SPA_VERSION_FEATURES].
966 ztest_random_spa_version(uint64_t initial_version
)
968 uint64_t version
= initial_version
;
970 if (version
<= SPA_VERSION_BEFORE_FEATURES
) {
972 ztest_random(SPA_VERSION_BEFORE_FEATURES
- version
+ 1);
975 if (version
> SPA_VERSION_BEFORE_FEATURES
)
976 version
= SPA_VERSION_FEATURES
;
978 ASSERT(SPA_VERSION_IS_SUPPORTED(version
));
983 ztest_random_blocksize(void)
985 return (1 << (SPA_MINBLOCKSHIFT
+
986 ztest_random(SPA_MAXBLOCKSHIFT
- SPA_MINBLOCKSHIFT
+ 1)));
990 ztest_random_ibshift(void)
992 return (DN_MIN_INDBLKSHIFT
+
993 ztest_random(DN_MAX_INDBLKSHIFT
- DN_MIN_INDBLKSHIFT
+ 1));
997 ztest_random_vdev_top(spa_t
*spa
, boolean_t log_ok
)
1000 vdev_t
*rvd
= spa
->spa_root_vdev
;
1003 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_READER
) != 0);
1006 top
= ztest_random(rvd
->vdev_children
);
1007 tvd
= rvd
->vdev_child
[top
];
1008 } while (tvd
->vdev_ishole
|| (tvd
->vdev_islog
&& !log_ok
) ||
1009 tvd
->vdev_mg
== NULL
|| tvd
->vdev_mg
->mg_class
== NULL
);
1015 ztest_random_dsl_prop(zfs_prop_t prop
)
1020 value
= zfs_prop_random_value(prop
, ztest_random(-1ULL));
1021 } while (prop
== ZFS_PROP_CHECKSUM
&& value
== ZIO_CHECKSUM_OFF
);
1027 ztest_dsl_prop_set_uint64(char *osname
, zfs_prop_t prop
, uint64_t value
,
1030 const char *propname
= zfs_prop_to_name(prop
);
1031 const char *valname
;
1036 error
= dsl_prop_set_int(osname
, propname
,
1037 (inherit
? ZPROP_SRC_NONE
: ZPROP_SRC_LOCAL
), value
);
1039 if (error
== ENOSPC
) {
1040 ztest_record_enospc(FTAG
);
1045 setpoint
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
1046 VERIFY0(dsl_prop_get_integer(osname
, propname
, &curval
, setpoint
));
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
));
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
));
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
));
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
), ==, 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
);
2487 leaves
= MAX(zs
->zs_mirrors
+ zs
->zs_splits
, 1) * ztest_opts
.zo_raidz
;
2489 spa_config_enter(spa
, SCL_VDEV
, FTAG
, RW_READER
);
2491 ztest_shared
->zs_vdev_next_leaf
= find_vdev_hole(spa
) * leaves
;
2494 * If we have slogs then remove them 1/4 of the time.
2496 if (spa_has_slogs(spa
) && ztest_random(4) == 0) {
2498 * Grab the guid from the head of the log class rotor.
2500 guid
= spa_log_class(spa
)->mc_rotor
->mg_vd
->vdev_guid
;
2502 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
2505 * We have to grab the zs_name_lock as writer to
2506 * prevent a race between removing a slog (dmu_objset_find)
2507 * and destroying a dataset. Removing the slog will
2508 * grab a reference on the dataset which may cause
2509 * dsl_destroy_head() to fail with EBUSY thus
2510 * leaving the dataset in an inconsistent state.
2512 rw_enter(&ztest_name_lock
, RW_WRITER
);
2513 error
= spa_vdev_remove(spa
, guid
, B_FALSE
);
2514 rw_exit(&ztest_name_lock
);
2516 if (error
&& error
!= EEXIST
)
2517 fatal(0, "spa_vdev_remove() = %d", error
);
2519 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
2522 * Make 1/4 of the devices be log devices.
2524 nvroot
= make_vdev_root(NULL
, NULL
, NULL
,
2525 ztest_opts
.zo_vdev_size
, 0,
2526 ztest_random(4) == 0, ztest_opts
.zo_raidz
,
2529 error
= spa_vdev_add(spa
, nvroot
);
2530 nvlist_free(nvroot
);
2532 if (error
== ENOSPC
)
2533 ztest_record_enospc("spa_vdev_add");
2534 else if (error
!= 0)
2535 fatal(0, "spa_vdev_add() = %d", error
);
2538 mutex_exit(&ztest_vdev_lock
);
2542 * Verify that adding/removing aux devices (l2arc, hot spare) works as expected.
2546 ztest_vdev_aux_add_remove(ztest_ds_t
*zd
, uint64_t id
)
2548 ztest_shared_t
*zs
= ztest_shared
;
2549 spa_t
*spa
= ztest_spa
;
2550 vdev_t
*rvd
= spa
->spa_root_vdev
;
2551 spa_aux_vdev_t
*sav
;
2557 path
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
2559 if (ztest_random(2) == 0) {
2560 sav
= &spa
->spa_spares
;
2561 aux
= ZPOOL_CONFIG_SPARES
;
2563 sav
= &spa
->spa_l2cache
;
2564 aux
= ZPOOL_CONFIG_L2CACHE
;
2567 mutex_enter(&ztest_vdev_lock
);
2569 spa_config_enter(spa
, SCL_VDEV
, FTAG
, RW_READER
);
2571 if (sav
->sav_count
!= 0 && ztest_random(4) == 0) {
2573 * Pick a random device to remove.
2575 guid
= sav
->sav_vdevs
[ztest_random(sav
->sav_count
)]->vdev_guid
;
2578 * Find an unused device we can add.
2580 zs
->zs_vdev_aux
= 0;
2583 (void) snprintf(path
, MAXPATHLEN
, ztest_aux_template
,
2584 ztest_opts
.zo_dir
, ztest_opts
.zo_pool
, aux
,
2586 for (c
= 0; c
< sav
->sav_count
; c
++)
2587 if (strcmp(sav
->sav_vdevs
[c
]->vdev_path
,
2590 if (c
== sav
->sav_count
&&
2591 vdev_lookup_by_path(rvd
, path
) == NULL
)
2597 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
2603 nvlist_t
*nvroot
= make_vdev_root(NULL
, aux
, NULL
,
2604 (ztest_opts
.zo_vdev_size
* 5) / 4, 0, 0, 0, 0, 1);
2605 error
= spa_vdev_add(spa
, nvroot
);
2607 fatal(0, "spa_vdev_add(%p) = %d", nvroot
, error
);
2608 nvlist_free(nvroot
);
2611 * Remove an existing device. Sometimes, dirty its
2612 * vdev state first to make sure we handle removal
2613 * of devices that have pending state changes.
2615 if (ztest_random(2) == 0)
2616 (void) vdev_online(spa
, guid
, 0, NULL
);
2618 error
= spa_vdev_remove(spa
, guid
, B_FALSE
);
2619 if (error
!= 0 && error
!= EBUSY
)
2620 fatal(0, "spa_vdev_remove(%llu) = %d", guid
, error
);
2623 mutex_exit(&ztest_vdev_lock
);
2625 umem_free(path
, MAXPATHLEN
);
2629 * split a pool if it has mirror tlvdevs
2633 ztest_split_pool(ztest_ds_t
*zd
, uint64_t id
)
2635 ztest_shared_t
*zs
= ztest_shared
;
2636 spa_t
*spa
= ztest_spa
;
2637 vdev_t
*rvd
= spa
->spa_root_vdev
;
2638 nvlist_t
*tree
, **child
, *config
, *split
, **schild
;
2639 uint_t c
, children
, schildren
= 0, lastlogid
= 0;
2642 mutex_enter(&ztest_vdev_lock
);
2644 /* ensure we have a useable config; mirrors of raidz aren't supported */
2645 if (zs
->zs_mirrors
< 3 || ztest_opts
.zo_raidz
> 1) {
2646 mutex_exit(&ztest_vdev_lock
);
2650 /* clean up the old pool, if any */
2651 (void) spa_destroy("splitp");
2653 spa_config_enter(spa
, SCL_VDEV
, FTAG
, RW_READER
);
2655 /* generate a config from the existing config */
2656 mutex_enter(&spa
->spa_props_lock
);
2657 VERIFY(nvlist_lookup_nvlist(spa
->spa_config
, ZPOOL_CONFIG_VDEV_TREE
,
2659 mutex_exit(&spa
->spa_props_lock
);
2661 VERIFY(nvlist_lookup_nvlist_array(tree
, ZPOOL_CONFIG_CHILDREN
, &child
,
2664 schild
= malloc(rvd
->vdev_children
* sizeof (nvlist_t
*));
2665 for (c
= 0; c
< children
; c
++) {
2666 vdev_t
*tvd
= rvd
->vdev_child
[c
];
2670 if (tvd
->vdev_islog
|| tvd
->vdev_ops
== &vdev_hole_ops
) {
2671 VERIFY(nvlist_alloc(&schild
[schildren
], NV_UNIQUE_NAME
,
2673 VERIFY(nvlist_add_string(schild
[schildren
],
2674 ZPOOL_CONFIG_TYPE
, VDEV_TYPE_HOLE
) == 0);
2675 VERIFY(nvlist_add_uint64(schild
[schildren
],
2676 ZPOOL_CONFIG_IS_HOLE
, 1) == 0);
2678 lastlogid
= schildren
;
2683 VERIFY(nvlist_lookup_nvlist_array(child
[c
],
2684 ZPOOL_CONFIG_CHILDREN
, &mchild
, &mchildren
) == 0);
2685 VERIFY(nvlist_dup(mchild
[0], &schild
[schildren
++], 0) == 0);
2688 /* OK, create a config that can be used to split */
2689 VERIFY(nvlist_alloc(&split
, NV_UNIQUE_NAME
, 0) == 0);
2690 VERIFY(nvlist_add_string(split
, ZPOOL_CONFIG_TYPE
,
2691 VDEV_TYPE_ROOT
) == 0);
2692 VERIFY(nvlist_add_nvlist_array(split
, ZPOOL_CONFIG_CHILDREN
, schild
,
2693 lastlogid
!= 0 ? lastlogid
: schildren
) == 0);
2695 VERIFY(nvlist_alloc(&config
, NV_UNIQUE_NAME
, 0) == 0);
2696 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, split
) == 0);
2698 for (c
= 0; c
< schildren
; c
++)
2699 nvlist_free(schild
[c
]);
2703 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
2705 (void) rw_enter(&ztest_name_lock
, RW_WRITER
);
2706 error
= spa_vdev_split_mirror(spa
, "splitp", config
, NULL
, B_FALSE
);
2707 (void) rw_exit(&ztest_name_lock
);
2709 nvlist_free(config
);
2712 (void) printf("successful split - results:\n");
2713 mutex_enter(&spa_namespace_lock
);
2714 show_pool_stats(spa
);
2715 show_pool_stats(spa_lookup("splitp"));
2716 mutex_exit(&spa_namespace_lock
);
2720 mutex_exit(&ztest_vdev_lock
);
2725 * Verify that we can attach and detach devices.
2729 ztest_vdev_attach_detach(ztest_ds_t
*zd
, uint64_t id
)
2731 ztest_shared_t
*zs
= ztest_shared
;
2732 spa_t
*spa
= ztest_spa
;
2733 spa_aux_vdev_t
*sav
= &spa
->spa_spares
;
2734 vdev_t
*rvd
= spa
->spa_root_vdev
;
2735 vdev_t
*oldvd
, *newvd
, *pvd
;
2739 uint64_t ashift
= ztest_get_ashift();
2740 uint64_t oldguid
, pguid
;
2741 size_t oldsize
, newsize
;
2742 char *oldpath
, *newpath
;
2744 int oldvd_has_siblings
= B_FALSE
;
2745 int newvd_is_spare
= B_FALSE
;
2747 int error
, expected_error
;
2749 oldpath
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
2750 newpath
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
2752 mutex_enter(&ztest_vdev_lock
);
2753 leaves
= MAX(zs
->zs_mirrors
, 1) * ztest_opts
.zo_raidz
;
2755 spa_config_enter(spa
, SCL_VDEV
, FTAG
, RW_READER
);
2758 * Decide whether to do an attach or a replace.
2760 replacing
= ztest_random(2);
2763 * Pick a random top-level vdev.
2765 top
= ztest_random_vdev_top(spa
, B_TRUE
);
2768 * Pick a random leaf within it.
2770 leaf
= ztest_random(leaves
);
2775 oldvd
= rvd
->vdev_child
[top
];
2776 if (zs
->zs_mirrors
>= 1) {
2777 ASSERT(oldvd
->vdev_ops
== &vdev_mirror_ops
);
2778 ASSERT(oldvd
->vdev_children
>= zs
->zs_mirrors
);
2779 oldvd
= oldvd
->vdev_child
[leaf
/ ztest_opts
.zo_raidz
];
2781 if (ztest_opts
.zo_raidz
> 1) {
2782 ASSERT(oldvd
->vdev_ops
== &vdev_raidz_ops
);
2783 ASSERT(oldvd
->vdev_children
== ztest_opts
.zo_raidz
);
2784 oldvd
= oldvd
->vdev_child
[leaf
% ztest_opts
.zo_raidz
];
2788 * If we're already doing an attach or replace, oldvd may be a
2789 * mirror vdev -- in which case, pick a random child.
2791 while (oldvd
->vdev_children
!= 0) {
2792 oldvd_has_siblings
= B_TRUE
;
2793 ASSERT(oldvd
->vdev_children
>= 2);
2794 oldvd
= oldvd
->vdev_child
[ztest_random(oldvd
->vdev_children
)];
2797 oldguid
= oldvd
->vdev_guid
;
2798 oldsize
= vdev_get_min_asize(oldvd
);
2799 oldvd_is_log
= oldvd
->vdev_top
->vdev_islog
;
2800 (void) strcpy(oldpath
, oldvd
->vdev_path
);
2801 pvd
= oldvd
->vdev_parent
;
2802 pguid
= pvd
->vdev_guid
;
2805 * If oldvd has siblings, then half of the time, detach it.
2807 if (oldvd_has_siblings
&& ztest_random(2) == 0) {
2808 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
2809 error
= spa_vdev_detach(spa
, oldguid
, pguid
, B_FALSE
);
2810 if (error
!= 0 && error
!= ENODEV
&& error
!= EBUSY
&&
2812 fatal(0, "detach (%s) returned %d", oldpath
, error
);
2817 * For the new vdev, choose with equal probability between the two
2818 * standard paths (ending in either 'a' or 'b') or a random hot spare.
2820 if (sav
->sav_count
!= 0 && ztest_random(3) == 0) {
2821 newvd
= sav
->sav_vdevs
[ztest_random(sav
->sav_count
)];
2822 newvd_is_spare
= B_TRUE
;
2823 (void) strcpy(newpath
, newvd
->vdev_path
);
2825 (void) snprintf(newpath
, MAXPATHLEN
, ztest_dev_template
,
2826 ztest_opts
.zo_dir
, ztest_opts
.zo_pool
,
2827 top
* leaves
+ leaf
);
2828 if (ztest_random(2) == 0)
2829 newpath
[strlen(newpath
) - 1] = 'b';
2830 newvd
= vdev_lookup_by_path(rvd
, newpath
);
2834 newsize
= vdev_get_min_asize(newvd
);
2837 * Make newsize a little bigger or smaller than oldsize.
2838 * If it's smaller, the attach should fail.
2839 * If it's larger, and we're doing a replace,
2840 * we should get dynamic LUN growth when we're done.
2842 newsize
= 10 * oldsize
/ (9 + ztest_random(3));
2846 * If pvd is not a mirror or root, the attach should fail with ENOTSUP,
2847 * unless it's a replace; in that case any non-replacing parent is OK.
2849 * If newvd is already part of the pool, it should fail with EBUSY.
2851 * If newvd is too small, it should fail with EOVERFLOW.
2853 if (pvd
->vdev_ops
!= &vdev_mirror_ops
&&
2854 pvd
->vdev_ops
!= &vdev_root_ops
&& (!replacing
||
2855 pvd
->vdev_ops
== &vdev_replacing_ops
||
2856 pvd
->vdev_ops
== &vdev_spare_ops
))
2857 expected_error
= ENOTSUP
;
2858 else if (newvd_is_spare
&& (!replacing
|| oldvd_is_log
))
2859 expected_error
= ENOTSUP
;
2860 else if (newvd
== oldvd
)
2861 expected_error
= replacing
? 0 : EBUSY
;
2862 else if (vdev_lookup_by_path(rvd
, newpath
) != NULL
)
2863 expected_error
= EBUSY
;
2864 else if (newsize
< oldsize
)
2865 expected_error
= EOVERFLOW
;
2866 else if (ashift
> oldvd
->vdev_top
->vdev_ashift
)
2867 expected_error
= EDOM
;
2871 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
2874 * Build the nvlist describing newpath.
2876 root
= make_vdev_root(newpath
, NULL
, NULL
, newvd
== NULL
? newsize
: 0,
2877 ashift
, 0, 0, 0, 1);
2879 error
= spa_vdev_attach(spa
, oldguid
, root
, replacing
);
2884 * If our parent was the replacing vdev, but the replace completed,
2885 * then instead of failing with ENOTSUP we may either succeed,
2886 * fail with ENODEV, or fail with EOVERFLOW.
2888 if (expected_error
== ENOTSUP
&&
2889 (error
== 0 || error
== ENODEV
|| error
== EOVERFLOW
))
2890 expected_error
= error
;
2893 * If someone grew the LUN, the replacement may be too small.
2895 if (error
== EOVERFLOW
|| error
== EBUSY
)
2896 expected_error
= error
;
2898 /* XXX workaround 6690467 */
2899 if (error
!= expected_error
&& expected_error
!= EBUSY
) {
2900 fatal(0, "attach (%s %llu, %s %llu, %d) "
2901 "returned %d, expected %d",
2902 oldpath
, (longlong_t
)oldsize
, newpath
,
2903 (longlong_t
)newsize
, replacing
, error
, expected_error
);
2906 mutex_exit(&ztest_vdev_lock
);
2908 umem_free(oldpath
, MAXPATHLEN
);
2909 umem_free(newpath
, MAXPATHLEN
);
2913 * Callback function which expands the physical size of the vdev.
2916 grow_vdev(vdev_t
*vd
, void *arg
)
2918 ASSERTV(spa_t
*spa
= vd
->vdev_spa
);
2919 size_t *newsize
= arg
;
2923 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_READER
) == SCL_STATE
);
2924 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
2926 if ((fd
= open(vd
->vdev_path
, O_RDWR
)) == -1)
2929 fsize
= lseek(fd
, 0, SEEK_END
);
2930 VERIFY(ftruncate(fd
, *newsize
) == 0);
2932 if (ztest_opts
.zo_verbose
>= 6) {
2933 (void) printf("%s grew from %lu to %lu bytes\n",
2934 vd
->vdev_path
, (ulong_t
)fsize
, (ulong_t
)*newsize
);
2941 * Callback function which expands a given vdev by calling vdev_online().
2945 online_vdev(vdev_t
*vd
, void *arg
)
2947 spa_t
*spa
= vd
->vdev_spa
;
2948 vdev_t
*tvd
= vd
->vdev_top
;
2949 uint64_t guid
= vd
->vdev_guid
;
2950 uint64_t generation
= spa
->spa_config_generation
+ 1;
2951 vdev_state_t newstate
= VDEV_STATE_UNKNOWN
;
2954 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_READER
) == SCL_STATE
);
2955 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
2957 /* Calling vdev_online will initialize the new metaslabs */
2958 spa_config_exit(spa
, SCL_STATE
, spa
);
2959 error
= vdev_online(spa
, guid
, ZFS_ONLINE_EXPAND
, &newstate
);
2960 spa_config_enter(spa
, SCL_STATE
, spa
, RW_READER
);
2963 * If vdev_online returned an error or the underlying vdev_open
2964 * failed then we abort the expand. The only way to know that
2965 * vdev_open fails is by checking the returned newstate.
2967 if (error
|| newstate
!= VDEV_STATE_HEALTHY
) {
2968 if (ztest_opts
.zo_verbose
>= 5) {
2969 (void) printf("Unable to expand vdev, state %llu, "
2970 "error %d\n", (u_longlong_t
)newstate
, error
);
2974 ASSERT3U(newstate
, ==, VDEV_STATE_HEALTHY
);
2977 * Since we dropped the lock we need to ensure that we're
2978 * still talking to the original vdev. It's possible this
2979 * vdev may have been detached/replaced while we were
2980 * trying to online it.
2982 if (generation
!= spa
->spa_config_generation
) {
2983 if (ztest_opts
.zo_verbose
>= 5) {
2984 (void) printf("vdev configuration has changed, "
2985 "guid %llu, state %llu, expected gen %llu, "
2988 (u_longlong_t
)tvd
->vdev_state
,
2989 (u_longlong_t
)generation
,
2990 (u_longlong_t
)spa
->spa_config_generation
);
2998 * Traverse the vdev tree calling the supplied function.
2999 * We continue to walk the tree until we either have walked all
3000 * children or we receive a non-NULL return from the callback.
3001 * If a NULL callback is passed, then we just return back the first
3002 * leaf vdev we encounter.
3005 vdev_walk_tree(vdev_t
*vd
, vdev_t
*(*func
)(vdev_t
*, void *), void *arg
)
3009 if (vd
->vdev_ops
->vdev_op_leaf
) {
3013 return (func(vd
, arg
));
3016 for (c
= 0; c
< vd
->vdev_children
; c
++) {
3017 vdev_t
*cvd
= vd
->vdev_child
[c
];
3018 if ((cvd
= vdev_walk_tree(cvd
, func
, arg
)) != NULL
)
3025 * Verify that dynamic LUN growth works as expected.
3029 ztest_vdev_LUN_growth(ztest_ds_t
*zd
, uint64_t id
)
3031 spa_t
*spa
= ztest_spa
;
3033 metaslab_class_t
*mc
;
3034 metaslab_group_t
*mg
;
3035 size_t psize
, newsize
;
3037 uint64_t old_class_space
, new_class_space
, old_ms_count
, new_ms_count
;
3039 mutex_enter(&ztest_vdev_lock
);
3040 spa_config_enter(spa
, SCL_STATE
, spa
, RW_READER
);
3042 top
= ztest_random_vdev_top(spa
, B_TRUE
);
3044 tvd
= spa
->spa_root_vdev
->vdev_child
[top
];
3047 old_ms_count
= tvd
->vdev_ms_count
;
3048 old_class_space
= metaslab_class_get_space(mc
);
3051 * Determine the size of the first leaf vdev associated with
3052 * our top-level device.
3054 vd
= vdev_walk_tree(tvd
, NULL
, NULL
);
3055 ASSERT3P(vd
, !=, NULL
);
3056 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
3058 psize
= vd
->vdev_psize
;
3061 * We only try to expand the vdev if it's healthy, less than 4x its
3062 * original size, and it has a valid psize.
3064 if (tvd
->vdev_state
!= VDEV_STATE_HEALTHY
||
3065 psize
== 0 || psize
>= 4 * ztest_opts
.zo_vdev_size
) {
3066 spa_config_exit(spa
, SCL_STATE
, spa
);
3067 mutex_exit(&ztest_vdev_lock
);
3071 newsize
= psize
+ psize
/ 8;
3072 ASSERT3U(newsize
, >, psize
);
3074 if (ztest_opts
.zo_verbose
>= 6) {
3075 (void) printf("Expanding LUN %s from %lu to %lu\n",
3076 vd
->vdev_path
, (ulong_t
)psize
, (ulong_t
)newsize
);
3080 * Growing the vdev is a two step process:
3081 * 1). expand the physical size (i.e. relabel)
3082 * 2). online the vdev to create the new metaslabs
3084 if (vdev_walk_tree(tvd
, grow_vdev
, &newsize
) != NULL
||
3085 vdev_walk_tree(tvd
, online_vdev
, NULL
) != NULL
||
3086 tvd
->vdev_state
!= VDEV_STATE_HEALTHY
) {
3087 if (ztest_opts
.zo_verbose
>= 5) {
3088 (void) printf("Could not expand LUN because "
3089 "the vdev configuration changed.\n");
3091 spa_config_exit(spa
, SCL_STATE
, spa
);
3092 mutex_exit(&ztest_vdev_lock
);
3096 spa_config_exit(spa
, SCL_STATE
, spa
);
3099 * Expanding the LUN will update the config asynchronously,
3100 * thus we must wait for the async thread to complete any
3101 * pending tasks before proceeding.
3105 mutex_enter(&spa
->spa_async_lock
);
3106 done
= (spa
->spa_async_thread
== NULL
&& !spa
->spa_async_tasks
);
3107 mutex_exit(&spa
->spa_async_lock
);
3110 txg_wait_synced(spa_get_dsl(spa
), 0);
3111 (void) poll(NULL
, 0, 100);
3114 spa_config_enter(spa
, SCL_STATE
, spa
, RW_READER
);
3116 tvd
= spa
->spa_root_vdev
->vdev_child
[top
];
3117 new_ms_count
= tvd
->vdev_ms_count
;
3118 new_class_space
= metaslab_class_get_space(mc
);
3120 if (tvd
->vdev_mg
!= mg
|| mg
->mg_class
!= mc
) {
3121 if (ztest_opts
.zo_verbose
>= 5) {
3122 (void) printf("Could not verify LUN expansion due to "
3123 "intervening vdev offline or remove.\n");
3125 spa_config_exit(spa
, SCL_STATE
, spa
);
3126 mutex_exit(&ztest_vdev_lock
);
3131 * Make sure we were able to grow the vdev.
3133 if (new_ms_count
<= old_ms_count
)
3134 fatal(0, "LUN expansion failed: ms_count %llu <= %llu\n",
3135 old_ms_count
, new_ms_count
);
3138 * Make sure we were able to grow the pool.
3140 if (new_class_space
<= old_class_space
)
3141 fatal(0, "LUN expansion failed: class_space %llu <= %llu\n",
3142 old_class_space
, new_class_space
);
3144 if (ztest_opts
.zo_verbose
>= 5) {
3145 char oldnumbuf
[6], newnumbuf
[6];
3147 nicenum(old_class_space
, oldnumbuf
);
3148 nicenum(new_class_space
, newnumbuf
);
3149 (void) printf("%s grew from %s to %s\n",
3150 spa
->spa_name
, oldnumbuf
, newnumbuf
);
3153 spa_config_exit(spa
, SCL_STATE
, spa
);
3154 mutex_exit(&ztest_vdev_lock
);
3158 * Verify that dmu_objset_{create,destroy,open,close} work as expected.
3162 ztest_objset_create_cb(objset_t
*os
, void *arg
, cred_t
*cr
, dmu_tx_t
*tx
)
3165 * Create the objects common to all ztest datasets.
3167 VERIFY(zap_create_claim(os
, ZTEST_DIROBJ
,
3168 DMU_OT_ZAP_OTHER
, DMU_OT_NONE
, 0, tx
) == 0);
3172 ztest_dataset_create(char *dsname
)
3174 uint64_t zilset
= ztest_random(100);
3175 int err
= dmu_objset_create(dsname
, DMU_OST_OTHER
, 0,
3176 ztest_objset_create_cb
, NULL
);
3178 if (err
|| zilset
< 80)
3181 if (ztest_opts
.zo_verbose
>= 5)
3182 (void) printf("Setting dataset %s to sync always\n", dsname
);
3183 return (ztest_dsl_prop_set_uint64(dsname
, ZFS_PROP_SYNC
,
3184 ZFS_SYNC_ALWAYS
, B_FALSE
));
3189 ztest_objset_destroy_cb(const char *name
, void *arg
)
3192 dmu_object_info_t doi
;
3196 * Verify that the dataset contains a directory object.
3198 VERIFY0(dmu_objset_own(name
, DMU_OST_OTHER
, B_TRUE
, FTAG
, &os
));
3199 error
= dmu_object_info(os
, ZTEST_DIROBJ
, &doi
);
3200 if (error
!= ENOENT
) {
3201 /* We could have crashed in the middle of destroying it */
3203 ASSERT3U(doi
.doi_type
, ==, DMU_OT_ZAP_OTHER
);
3204 ASSERT3S(doi
.doi_physical_blocks_512
, >=, 0);
3206 dmu_objset_disown(os
, FTAG
);
3209 * Destroy the dataset.
3211 if (strchr(name
, '@') != NULL
) {
3212 VERIFY0(dsl_destroy_snapshot(name
, B_FALSE
));
3214 VERIFY0(dsl_destroy_head(name
));
3220 ztest_snapshot_create(char *osname
, uint64_t id
)
3222 char snapname
[MAXNAMELEN
];
3225 (void) snprintf(snapname
, sizeof (snapname
), "%llu", (u_longlong_t
)id
);
3227 error
= dmu_objset_snapshot_one(osname
, snapname
);
3228 if (error
== ENOSPC
) {
3229 ztest_record_enospc(FTAG
);
3232 if (error
!= 0 && error
!= EEXIST
) {
3233 fatal(0, "ztest_snapshot_create(%s@%s) = %d", osname
,
3240 ztest_snapshot_destroy(char *osname
, uint64_t id
)
3242 char snapname
[MAXNAMELEN
];
3245 (void) snprintf(snapname
, MAXNAMELEN
, "%s@%llu", osname
,
3248 error
= dsl_destroy_snapshot(snapname
, B_FALSE
);
3249 if (error
!= 0 && error
!= ENOENT
)
3250 fatal(0, "ztest_snapshot_destroy(%s) = %d", snapname
, error
);
3256 ztest_dmu_objset_create_destroy(ztest_ds_t
*zd
, uint64_t id
)
3266 zdtmp
= umem_alloc(sizeof (ztest_ds_t
), UMEM_NOFAIL
);
3267 name
= umem_alloc(MAXNAMELEN
, UMEM_NOFAIL
);
3269 (void) rw_enter(&ztest_name_lock
, RW_READER
);
3271 (void) snprintf(name
, MAXNAMELEN
, "%s/temp_%llu",
3272 ztest_opts
.zo_pool
, (u_longlong_t
)id
);
3275 * If this dataset exists from a previous run, process its replay log
3276 * half of the time. If we don't replay it, then dsl_destroy_head()
3277 * (invoked from ztest_objset_destroy_cb()) should just throw it away.
3279 if (ztest_random(2) == 0 &&
3280 dmu_objset_own(name
, DMU_OST_OTHER
, B_FALSE
, FTAG
, &os
) == 0) {
3281 ztest_zd_init(zdtmp
, NULL
, os
);
3282 zil_replay(os
, zdtmp
, ztest_replay_vector
);
3283 ztest_zd_fini(zdtmp
);
3284 dmu_objset_disown(os
, FTAG
);
3288 * There may be an old instance of the dataset we're about to
3289 * create lying around from a previous run. If so, destroy it
3290 * and all of its snapshots.
3292 (void) dmu_objset_find(name
, ztest_objset_destroy_cb
, NULL
,
3293 DS_FIND_CHILDREN
| DS_FIND_SNAPSHOTS
);
3296 * Verify that the destroyed dataset is no longer in the namespace.
3298 VERIFY3U(ENOENT
, ==, dmu_objset_own(name
, DMU_OST_OTHER
, B_TRUE
,
3302 * Verify that we can create a new dataset.
3304 error
= ztest_dataset_create(name
);
3306 if (error
== ENOSPC
) {
3307 ztest_record_enospc(FTAG
);
3310 fatal(0, "dmu_objset_create(%s) = %d", name
, error
);
3313 VERIFY0(dmu_objset_own(name
, DMU_OST_OTHER
, B_FALSE
, FTAG
, &os
));
3315 ztest_zd_init(zdtmp
, NULL
, os
);
3318 * Open the intent log for it.
3320 zilog
= zil_open(os
, ztest_get_data
);
3323 * Put some objects in there, do a little I/O to them,
3324 * and randomly take a couple of snapshots along the way.
3326 iters
= ztest_random(5);
3327 for (i
= 0; i
< iters
; i
++) {
3328 ztest_dmu_object_alloc_free(zdtmp
, id
);
3329 if (ztest_random(iters
) == 0)
3330 (void) ztest_snapshot_create(name
, i
);
3334 * Verify that we cannot create an existing dataset.
3336 VERIFY3U(EEXIST
, ==,
3337 dmu_objset_create(name
, DMU_OST_OTHER
, 0, NULL
, NULL
));
3340 * Verify that we can hold an objset that is also owned.
3342 VERIFY3U(0, ==, dmu_objset_hold(name
, FTAG
, &os2
));
3343 dmu_objset_rele(os2
, FTAG
);
3346 * Verify that we cannot own an objset that is already owned.
3349 dmu_objset_own(name
, DMU_OST_OTHER
, B_FALSE
, FTAG
, &os2
));
3352 dmu_objset_disown(os
, FTAG
);
3353 ztest_zd_fini(zdtmp
);
3355 (void) rw_exit(&ztest_name_lock
);
3357 umem_free(name
, MAXNAMELEN
);
3358 umem_free(zdtmp
, sizeof (ztest_ds_t
));
3362 * Verify that dmu_snapshot_{create,destroy,open,close} work as expected.
3365 ztest_dmu_snapshot_create_destroy(ztest_ds_t
*zd
, uint64_t id
)
3367 (void) rw_enter(&ztest_name_lock
, RW_READER
);
3368 (void) ztest_snapshot_destroy(zd
->zd_name
, id
);
3369 (void) ztest_snapshot_create(zd
->zd_name
, id
);
3370 (void) rw_exit(&ztest_name_lock
);
3374 * Cleanup non-standard snapshots and clones.
3377 ztest_dsl_dataset_cleanup(char *osname
, uint64_t id
)
3386 snap1name
= umem_alloc(MAXNAMELEN
, UMEM_NOFAIL
);
3387 clone1name
= umem_alloc(MAXNAMELEN
, UMEM_NOFAIL
);
3388 snap2name
= umem_alloc(MAXNAMELEN
, UMEM_NOFAIL
);
3389 clone2name
= umem_alloc(MAXNAMELEN
, UMEM_NOFAIL
);
3390 snap3name
= umem_alloc(MAXNAMELEN
, UMEM_NOFAIL
);
3392 (void) snprintf(snap1name
, MAXNAMELEN
, "%s@s1_%llu",
3393 osname
, (u_longlong_t
)id
);
3394 (void) snprintf(clone1name
, MAXNAMELEN
, "%s/c1_%llu",
3395 osname
, (u_longlong_t
)id
);
3396 (void) snprintf(snap2name
, MAXNAMELEN
, "%s@s2_%llu",
3397 clone1name
, (u_longlong_t
)id
);
3398 (void) snprintf(clone2name
, MAXNAMELEN
, "%s/c2_%llu",
3399 osname
, (u_longlong_t
)id
);
3400 (void) snprintf(snap3name
, MAXNAMELEN
, "%s@s3_%llu",
3401 clone1name
, (u_longlong_t
)id
);
3403 error
= dsl_destroy_head(clone2name
);
3404 if (error
&& error
!= ENOENT
)
3405 fatal(0, "dsl_destroy_head(%s) = %d", clone2name
, error
);
3406 error
= dsl_destroy_snapshot(snap3name
, B_FALSE
);
3407 if (error
&& error
!= ENOENT
)
3408 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap3name
, error
);
3409 error
= dsl_destroy_snapshot(snap2name
, B_FALSE
);
3410 if (error
&& error
!= ENOENT
)
3411 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap2name
, error
);
3412 error
= dsl_destroy_head(clone1name
);
3413 if (error
&& error
!= ENOENT
)
3414 fatal(0, "dsl_destroy_head(%s) = %d", clone1name
, error
);
3415 error
= dsl_destroy_snapshot(snap1name
, B_FALSE
);
3416 if (error
&& error
!= ENOENT
)
3417 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap1name
, error
);
3419 umem_free(snap1name
, MAXNAMELEN
);
3420 umem_free(clone1name
, MAXNAMELEN
);
3421 umem_free(snap2name
, MAXNAMELEN
);
3422 umem_free(clone2name
, MAXNAMELEN
);
3423 umem_free(snap3name
, MAXNAMELEN
);
3427 * Verify dsl_dataset_promote handles EBUSY
3430 ztest_dsl_dataset_promote_busy(ztest_ds_t
*zd
, uint64_t id
)
3438 char *osname
= zd
->zd_name
;
3441 snap1name
= umem_alloc(MAXNAMELEN
, UMEM_NOFAIL
);
3442 clone1name
= umem_alloc(MAXNAMELEN
, UMEM_NOFAIL
);
3443 snap2name
= umem_alloc(MAXNAMELEN
, UMEM_NOFAIL
);
3444 clone2name
= umem_alloc(MAXNAMELEN
, UMEM_NOFAIL
);
3445 snap3name
= umem_alloc(MAXNAMELEN
, UMEM_NOFAIL
);
3447 (void) rw_enter(&ztest_name_lock
, RW_READER
);
3449 ztest_dsl_dataset_cleanup(osname
, id
);
3451 (void) snprintf(snap1name
, MAXNAMELEN
, "%s@s1_%llu",
3452 osname
, (u_longlong_t
)id
);
3453 (void) snprintf(clone1name
, MAXNAMELEN
, "%s/c1_%llu",
3454 osname
, (u_longlong_t
)id
);
3455 (void) snprintf(snap2name
, MAXNAMELEN
, "%s@s2_%llu",
3456 clone1name
, (u_longlong_t
)id
);
3457 (void) snprintf(clone2name
, MAXNAMELEN
, "%s/c2_%llu",
3458 osname
, (u_longlong_t
)id
);
3459 (void) snprintf(snap3name
, MAXNAMELEN
, "%s@s3_%llu",
3460 clone1name
, (u_longlong_t
)id
);
3462 error
= dmu_objset_snapshot_one(osname
, strchr(snap1name
, '@') + 1);
3463 if (error
&& error
!= EEXIST
) {
3464 if (error
== ENOSPC
) {
3465 ztest_record_enospc(FTAG
);
3468 fatal(0, "dmu_take_snapshot(%s) = %d", snap1name
, error
);
3471 error
= dmu_objset_clone(clone1name
, snap1name
);
3473 if (error
== ENOSPC
) {
3474 ztest_record_enospc(FTAG
);
3477 fatal(0, "dmu_objset_create(%s) = %d", clone1name
, error
);
3480 error
= dmu_objset_snapshot_one(clone1name
, strchr(snap2name
, '@') + 1);
3481 if (error
&& error
!= EEXIST
) {
3482 if (error
== ENOSPC
) {
3483 ztest_record_enospc(FTAG
);
3486 fatal(0, "dmu_open_snapshot(%s) = %d", snap2name
, error
);
3489 error
= dmu_objset_snapshot_one(clone1name
, strchr(snap3name
, '@') + 1);
3490 if (error
&& error
!= EEXIST
) {
3491 if (error
== ENOSPC
) {
3492 ztest_record_enospc(FTAG
);
3495 fatal(0, "dmu_open_snapshot(%s) = %d", snap3name
, error
);
3498 error
= dmu_objset_clone(clone2name
, snap3name
);
3500 if (error
== ENOSPC
) {
3501 ztest_record_enospc(FTAG
);
3504 fatal(0, "dmu_objset_create(%s) = %d", clone2name
, error
);
3507 error
= dmu_objset_own(snap2name
, DMU_OST_ANY
, B_TRUE
, FTAG
, &os
);
3509 fatal(0, "dmu_objset_own(%s) = %d", snap2name
, error
);
3510 error
= dsl_dataset_promote(clone2name
, NULL
);
3512 fatal(0, "dsl_dataset_promote(%s), %d, not EBUSY", clone2name
,
3514 dmu_objset_disown(os
, FTAG
);
3517 ztest_dsl_dataset_cleanup(osname
, id
);
3519 (void) rw_exit(&ztest_name_lock
);
3521 umem_free(snap1name
, MAXNAMELEN
);
3522 umem_free(clone1name
, MAXNAMELEN
);
3523 umem_free(snap2name
, MAXNAMELEN
);
3524 umem_free(clone2name
, MAXNAMELEN
);
3525 umem_free(snap3name
, MAXNAMELEN
);
3528 #undef OD_ARRAY_SIZE
3529 #define OD_ARRAY_SIZE 4
3532 * Verify that dmu_object_{alloc,free} work as expected.
3535 ztest_dmu_object_alloc_free(ztest_ds_t
*zd
, uint64_t id
)
3542 size
= sizeof(ztest_od_t
) * OD_ARRAY_SIZE
;
3543 od
= umem_alloc(size
, UMEM_NOFAIL
);
3544 batchsize
= OD_ARRAY_SIZE
;
3546 for (b
= 0; b
< batchsize
; b
++)
3547 ztest_od_init(od
+ b
, id
, FTAG
, b
, DMU_OT_UINT64_OTHER
, 0, 0);
3550 * Destroy the previous batch of objects, create a new batch,
3551 * and do some I/O on the new objects.
3553 if (ztest_object_init(zd
, od
, size
, B_TRUE
) != 0)
3556 while (ztest_random(4 * batchsize
) != 0)
3557 ztest_io(zd
, od
[ztest_random(batchsize
)].od_object
,
3558 ztest_random(ZTEST_RANGE_LOCKS
) << SPA_MAXBLOCKSHIFT
);
3560 umem_free(od
, size
);
3563 #undef OD_ARRAY_SIZE
3564 #define OD_ARRAY_SIZE 2
3567 * Verify that dmu_{read,write} work as expected.
3570 ztest_dmu_read_write(ztest_ds_t
*zd
, uint64_t id
)
3575 objset_t
*os
= zd
->zd_os
;
3576 size
= sizeof(ztest_od_t
) * OD_ARRAY_SIZE
;
3577 od
= umem_alloc(size
, UMEM_NOFAIL
);
3579 int i
, freeit
, error
;
3581 bufwad_t
*packbuf
, *bigbuf
, *pack
, *bigH
, *bigT
;
3582 uint64_t packobj
, packoff
, packsize
, bigobj
, bigoff
, bigsize
;
3583 uint64_t chunksize
= (1000 + ztest_random(1000)) * sizeof (uint64_t);
3584 uint64_t regions
= 997;
3585 uint64_t stride
= 123456789ULL;
3586 uint64_t width
= 40;
3587 int free_percent
= 5;
3590 * This test uses two objects, packobj and bigobj, that are always
3591 * updated together (i.e. in the same tx) so that their contents are
3592 * in sync and can be compared. Their contents relate to each other
3593 * in a simple way: packobj is a dense array of 'bufwad' structures,
3594 * while bigobj is a sparse array of the same bufwads. Specifically,
3595 * for any index n, there are three bufwads that should be identical:
3597 * packobj, at offset n * sizeof (bufwad_t)
3598 * bigobj, at the head of the nth chunk
3599 * bigobj, at the tail of the nth chunk
3601 * The chunk size is arbitrary. It doesn't have to be a power of two,
3602 * and it doesn't have any relation to the object blocksize.
3603 * The only requirement is that it can hold at least two bufwads.
3605 * Normally, we write the bufwad to each of these locations.
3606 * However, free_percent of the time we instead write zeroes to
3607 * packobj and perform a dmu_free_range() on bigobj. By comparing
3608 * bigobj to packobj, we can verify that the DMU is correctly
3609 * tracking which parts of an object are allocated and free,
3610 * and that the contents of the allocated blocks are correct.
3614 * Read the directory info. If it's the first time, set things up.
3616 ztest_od_init(od
, id
, FTAG
, 0, DMU_OT_UINT64_OTHER
, 0, chunksize
);
3617 ztest_od_init(od
+ 1, id
, FTAG
, 1, DMU_OT_UINT64_OTHER
, 0, chunksize
);
3619 if (ztest_object_init(zd
, od
, size
, B_FALSE
) != 0) {
3620 umem_free(od
, size
);
3624 bigobj
= od
[0].od_object
;
3625 packobj
= od
[1].od_object
;
3626 chunksize
= od
[0].od_gen
;
3627 ASSERT(chunksize
== od
[1].od_gen
);
3630 * Prefetch a random chunk of the big object.
3631 * Our aim here is to get some async reads in flight
3632 * for blocks that we may free below; the DMU should
3633 * handle this race correctly.
3635 n
= ztest_random(regions
) * stride
+ ztest_random(width
);
3636 s
= 1 + ztest_random(2 * width
- 1);
3637 dmu_prefetch(os
, bigobj
, n
* chunksize
, s
* chunksize
);
3640 * Pick a random index and compute the offsets into packobj and bigobj.
3642 n
= ztest_random(regions
) * stride
+ ztest_random(width
);
3643 s
= 1 + ztest_random(width
- 1);
3645 packoff
= n
* sizeof (bufwad_t
);
3646 packsize
= s
* sizeof (bufwad_t
);
3648 bigoff
= n
* chunksize
;
3649 bigsize
= s
* chunksize
;
3651 packbuf
= umem_alloc(packsize
, UMEM_NOFAIL
);
3652 bigbuf
= umem_alloc(bigsize
, UMEM_NOFAIL
);
3655 * free_percent of the time, free a range of bigobj rather than
3658 freeit
= (ztest_random(100) < free_percent
);
3661 * Read the current contents of our objects.
3663 error
= dmu_read(os
, packobj
, packoff
, packsize
, packbuf
,
3666 error
= dmu_read(os
, bigobj
, bigoff
, bigsize
, bigbuf
,
3671 * Get a tx for the mods to both packobj and bigobj.
3673 tx
= dmu_tx_create(os
);
3675 dmu_tx_hold_write(tx
, packobj
, packoff
, packsize
);
3678 dmu_tx_hold_free(tx
, bigobj
, bigoff
, bigsize
);
3680 dmu_tx_hold_write(tx
, bigobj
, bigoff
, bigsize
);
3682 txg
= ztest_tx_assign(tx
, TXG_MIGHTWAIT
, FTAG
);
3684 umem_free(packbuf
, packsize
);
3685 umem_free(bigbuf
, bigsize
);
3686 umem_free(od
, size
);
3690 dmu_object_set_checksum(os
, bigobj
,
3691 (enum zio_checksum
)ztest_random_dsl_prop(ZFS_PROP_CHECKSUM
), tx
);
3693 dmu_object_set_compress(os
, bigobj
,
3694 (enum zio_compress
)ztest_random_dsl_prop(ZFS_PROP_COMPRESSION
), tx
);
3697 * For each index from n to n + s, verify that the existing bufwad
3698 * in packobj matches the bufwads at the head and tail of the
3699 * corresponding chunk in bigobj. Then update all three bufwads
3700 * with the new values we want to write out.
3702 for (i
= 0; i
< s
; i
++) {
3704 pack
= (bufwad_t
*)((char *)packbuf
+ i
* sizeof (bufwad_t
));
3706 bigH
= (bufwad_t
*)((char *)bigbuf
+ i
* chunksize
);
3708 bigT
= (bufwad_t
*)((char *)bigH
+ chunksize
) - 1;
3710 ASSERT((uintptr_t)bigH
- (uintptr_t)bigbuf
< bigsize
);
3711 ASSERT((uintptr_t)bigT
- (uintptr_t)bigbuf
< bigsize
);
3713 if (pack
->bw_txg
> txg
)
3714 fatal(0, "future leak: got %llx, open txg is %llx",
3717 if (pack
->bw_data
!= 0 && pack
->bw_index
!= n
+ i
)
3718 fatal(0, "wrong index: got %llx, wanted %llx+%llx",
3719 pack
->bw_index
, n
, i
);
3721 if (bcmp(pack
, bigH
, sizeof (bufwad_t
)) != 0)
3722 fatal(0, "pack/bigH mismatch in %p/%p", pack
, bigH
);
3724 if (bcmp(pack
, bigT
, sizeof (bufwad_t
)) != 0)
3725 fatal(0, "pack/bigT mismatch in %p/%p", pack
, bigT
);
3728 bzero(pack
, sizeof (bufwad_t
));
3730 pack
->bw_index
= n
+ i
;
3732 pack
->bw_data
= 1 + ztest_random(-2ULL);
3739 * We've verified all the old bufwads, and made new ones.
3740 * Now write them out.
3742 dmu_write(os
, packobj
, packoff
, packsize
, packbuf
, tx
);
3745 if (ztest_opts
.zo_verbose
>= 7) {
3746 (void) printf("freeing offset %llx size %llx"
3748 (u_longlong_t
)bigoff
,
3749 (u_longlong_t
)bigsize
,
3752 VERIFY(0 == dmu_free_range(os
, bigobj
, bigoff
, bigsize
, tx
));
3754 if (ztest_opts
.zo_verbose
>= 7) {
3755 (void) printf("writing offset %llx size %llx"
3757 (u_longlong_t
)bigoff
,
3758 (u_longlong_t
)bigsize
,
3761 dmu_write(os
, bigobj
, bigoff
, bigsize
, bigbuf
, tx
);
3767 * Sanity check the stuff we just wrote.
3770 void *packcheck
= umem_alloc(packsize
, UMEM_NOFAIL
);
3771 void *bigcheck
= umem_alloc(bigsize
, UMEM_NOFAIL
);
3773 VERIFY(0 == dmu_read(os
, packobj
, packoff
,
3774 packsize
, packcheck
, DMU_READ_PREFETCH
));
3775 VERIFY(0 == dmu_read(os
, bigobj
, bigoff
,
3776 bigsize
, bigcheck
, DMU_READ_PREFETCH
));
3778 ASSERT(bcmp(packbuf
, packcheck
, packsize
) == 0);
3779 ASSERT(bcmp(bigbuf
, bigcheck
, bigsize
) == 0);
3781 umem_free(packcheck
, packsize
);
3782 umem_free(bigcheck
, bigsize
);
3785 umem_free(packbuf
, packsize
);
3786 umem_free(bigbuf
, bigsize
);
3787 umem_free(od
, size
);
3791 compare_and_update_pbbufs(uint64_t s
, bufwad_t
*packbuf
, bufwad_t
*bigbuf
,
3792 uint64_t bigsize
, uint64_t n
, uint64_t chunksize
, uint64_t txg
)
3800 * For each index from n to n + s, verify that the existing bufwad
3801 * in packobj matches the bufwads at the head and tail of the
3802 * corresponding chunk in bigobj. Then update all three bufwads
3803 * with the new values we want to write out.
3805 for (i
= 0; i
< s
; i
++) {
3807 pack
= (bufwad_t
*)((char *)packbuf
+ i
* sizeof (bufwad_t
));
3809 bigH
= (bufwad_t
*)((char *)bigbuf
+ i
* chunksize
);
3811 bigT
= (bufwad_t
*)((char *)bigH
+ chunksize
) - 1;
3813 ASSERT((uintptr_t)bigH
- (uintptr_t)bigbuf
< bigsize
);
3814 ASSERT((uintptr_t)bigT
- (uintptr_t)bigbuf
< bigsize
);
3816 if (pack
->bw_txg
> txg
)
3817 fatal(0, "future leak: got %llx, open txg is %llx",
3820 if (pack
->bw_data
!= 0 && pack
->bw_index
!= n
+ i
)
3821 fatal(0, "wrong index: got %llx, wanted %llx+%llx",
3822 pack
->bw_index
, n
, i
);
3824 if (bcmp(pack
, bigH
, sizeof (bufwad_t
)) != 0)
3825 fatal(0, "pack/bigH mismatch in %p/%p", pack
, bigH
);
3827 if (bcmp(pack
, bigT
, sizeof (bufwad_t
)) != 0)
3828 fatal(0, "pack/bigT mismatch in %p/%p", pack
, bigT
);
3830 pack
->bw_index
= n
+ i
;
3832 pack
->bw_data
= 1 + ztest_random(-2ULL);
3839 #undef OD_ARRAY_SIZE
3840 #define OD_ARRAY_SIZE 2
3843 ztest_dmu_read_write_zcopy(ztest_ds_t
*zd
, uint64_t id
)
3845 objset_t
*os
= zd
->zd_os
;
3852 bufwad_t
*packbuf
, *bigbuf
;
3853 uint64_t packobj
, packoff
, packsize
, bigobj
, bigoff
, bigsize
;
3854 uint64_t blocksize
= ztest_random_blocksize();
3855 uint64_t chunksize
= blocksize
;
3856 uint64_t regions
= 997;
3857 uint64_t stride
= 123456789ULL;
3859 dmu_buf_t
*bonus_db
;
3860 arc_buf_t
**bigbuf_arcbufs
;
3861 dmu_object_info_t doi
;
3863 size
= sizeof(ztest_od_t
) * OD_ARRAY_SIZE
;
3864 od
= umem_alloc(size
, UMEM_NOFAIL
);
3867 * This test uses two objects, packobj and bigobj, that are always
3868 * updated together (i.e. in the same tx) so that their contents are
3869 * in sync and can be compared. Their contents relate to each other
3870 * in a simple way: packobj is a dense array of 'bufwad' structures,
3871 * while bigobj is a sparse array of the same bufwads. Specifically,
3872 * for any index n, there are three bufwads that should be identical:
3874 * packobj, at offset n * sizeof (bufwad_t)
3875 * bigobj, at the head of the nth chunk
3876 * bigobj, at the tail of the nth chunk
3878 * The chunk size is set equal to bigobj block size so that
3879 * dmu_assign_arcbuf() can be tested for object updates.
3883 * Read the directory info. If it's the first time, set things up.
3885 ztest_od_init(od
, id
, FTAG
, 0, DMU_OT_UINT64_OTHER
, blocksize
, 0);
3886 ztest_od_init(od
+ 1, id
, FTAG
, 1, DMU_OT_UINT64_OTHER
, 0, chunksize
);
3889 if (ztest_object_init(zd
, od
, size
, B_FALSE
) != 0) {
3890 umem_free(od
, size
);
3894 bigobj
= od
[0].od_object
;
3895 packobj
= od
[1].od_object
;
3896 blocksize
= od
[0].od_blocksize
;
3897 chunksize
= blocksize
;
3898 ASSERT(chunksize
== od
[1].od_gen
);
3900 VERIFY(dmu_object_info(os
, bigobj
, &doi
) == 0);
3901 VERIFY(ISP2(doi
.doi_data_block_size
));
3902 VERIFY(chunksize
== doi
.doi_data_block_size
);
3903 VERIFY(chunksize
>= 2 * sizeof (bufwad_t
));
3906 * Pick a random index and compute the offsets into packobj and bigobj.
3908 n
= ztest_random(regions
) * stride
+ ztest_random(width
);
3909 s
= 1 + ztest_random(width
- 1);
3911 packoff
= n
* sizeof (bufwad_t
);
3912 packsize
= s
* sizeof (bufwad_t
);
3914 bigoff
= n
* chunksize
;
3915 bigsize
= s
* chunksize
;
3917 packbuf
= umem_zalloc(packsize
, UMEM_NOFAIL
);
3918 bigbuf
= umem_zalloc(bigsize
, UMEM_NOFAIL
);
3920 VERIFY3U(0, ==, dmu_bonus_hold(os
, bigobj
, FTAG
, &bonus_db
));
3922 bigbuf_arcbufs
= umem_zalloc(2 * s
* sizeof (arc_buf_t
*), UMEM_NOFAIL
);
3925 * Iteration 0 test zcopy for DB_UNCACHED dbufs.
3926 * Iteration 1 test zcopy to already referenced dbufs.
3927 * Iteration 2 test zcopy to dirty dbuf in the same txg.
3928 * Iteration 3 test zcopy to dbuf dirty in previous txg.
3929 * Iteration 4 test zcopy when dbuf is no longer dirty.
3930 * Iteration 5 test zcopy when it can't be done.
3931 * Iteration 6 one more zcopy write.
3933 for (i
= 0; i
< 7; i
++) {
3938 * In iteration 5 (i == 5) use arcbufs
3939 * that don't match bigobj blksz to test
3940 * dmu_assign_arcbuf() when it can't directly
3941 * assign an arcbuf to a dbuf.
3943 for (j
= 0; j
< s
; j
++) {
3946 dmu_request_arcbuf(bonus_db
, chunksize
);
3948 bigbuf_arcbufs
[2 * j
] =
3949 dmu_request_arcbuf(bonus_db
, chunksize
/ 2);
3950 bigbuf_arcbufs
[2 * j
+ 1] =
3951 dmu_request_arcbuf(bonus_db
, chunksize
/ 2);
3956 * Get a tx for the mods to both packobj and bigobj.
3958 tx
= dmu_tx_create(os
);
3960 dmu_tx_hold_write(tx
, packobj
, packoff
, packsize
);
3961 dmu_tx_hold_write(tx
, bigobj
, bigoff
, bigsize
);
3963 txg
= ztest_tx_assign(tx
, TXG_MIGHTWAIT
, FTAG
);
3965 umem_free(packbuf
, packsize
);
3966 umem_free(bigbuf
, bigsize
);
3967 for (j
= 0; j
< s
; j
++) {
3969 dmu_return_arcbuf(bigbuf_arcbufs
[j
]);
3972 bigbuf_arcbufs
[2 * j
]);
3974 bigbuf_arcbufs
[2 * j
+ 1]);
3977 umem_free(bigbuf_arcbufs
, 2 * s
* sizeof (arc_buf_t
*));
3978 umem_free(od
, size
);
3979 dmu_buf_rele(bonus_db
, FTAG
);
3984 * 50% of the time don't read objects in the 1st iteration to
3985 * test dmu_assign_arcbuf() for the case when there're no
3986 * existing dbufs for the specified offsets.
3988 if (i
!= 0 || ztest_random(2) != 0) {
3989 error
= dmu_read(os
, packobj
, packoff
,
3990 packsize
, packbuf
, DMU_READ_PREFETCH
);
3992 error
= dmu_read(os
, bigobj
, bigoff
, bigsize
,
3993 bigbuf
, DMU_READ_PREFETCH
);
3996 compare_and_update_pbbufs(s
, packbuf
, bigbuf
, bigsize
,
4000 * We've verified all the old bufwads, and made new ones.
4001 * Now write them out.
4003 dmu_write(os
, packobj
, packoff
, packsize
, packbuf
, tx
);
4004 if (ztest_opts
.zo_verbose
>= 7) {
4005 (void) printf("writing offset %llx size %llx"
4007 (u_longlong_t
)bigoff
,
4008 (u_longlong_t
)bigsize
,
4011 for (off
= bigoff
, j
= 0; j
< s
; j
++, off
+= chunksize
) {
4014 bcopy((caddr_t
)bigbuf
+ (off
- bigoff
),
4015 bigbuf_arcbufs
[j
]->b_data
, chunksize
);
4017 bcopy((caddr_t
)bigbuf
+ (off
- bigoff
),
4018 bigbuf_arcbufs
[2 * j
]->b_data
,
4020 bcopy((caddr_t
)bigbuf
+ (off
- bigoff
) +
4022 bigbuf_arcbufs
[2 * j
+ 1]->b_data
,
4027 VERIFY(dmu_buf_hold(os
, bigobj
, off
,
4028 FTAG
, &dbt
, DMU_READ_NO_PREFETCH
) == 0);
4031 dmu_assign_arcbuf(bonus_db
, off
,
4032 bigbuf_arcbufs
[j
], tx
);
4034 dmu_assign_arcbuf(bonus_db
, off
,
4035 bigbuf_arcbufs
[2 * j
], tx
);
4036 dmu_assign_arcbuf(bonus_db
,
4037 off
+ chunksize
/ 2,
4038 bigbuf_arcbufs
[2 * j
+ 1], tx
);
4041 dmu_buf_rele(dbt
, FTAG
);
4047 * Sanity check the stuff we just wrote.
4050 void *packcheck
= umem_alloc(packsize
, UMEM_NOFAIL
);
4051 void *bigcheck
= umem_alloc(bigsize
, UMEM_NOFAIL
);
4053 VERIFY(0 == dmu_read(os
, packobj
, packoff
,
4054 packsize
, packcheck
, DMU_READ_PREFETCH
));
4055 VERIFY(0 == dmu_read(os
, bigobj
, bigoff
,
4056 bigsize
, bigcheck
, DMU_READ_PREFETCH
));
4058 ASSERT(bcmp(packbuf
, packcheck
, packsize
) == 0);
4059 ASSERT(bcmp(bigbuf
, bigcheck
, bigsize
) == 0);
4061 umem_free(packcheck
, packsize
);
4062 umem_free(bigcheck
, bigsize
);
4065 txg_wait_open(dmu_objset_pool(os
), 0);
4066 } else if (i
== 3) {
4067 txg_wait_synced(dmu_objset_pool(os
), 0);
4071 dmu_buf_rele(bonus_db
, FTAG
);
4072 umem_free(packbuf
, packsize
);
4073 umem_free(bigbuf
, bigsize
);
4074 umem_free(bigbuf_arcbufs
, 2 * s
* sizeof (arc_buf_t
*));
4075 umem_free(od
, size
);
4080 ztest_dmu_write_parallel(ztest_ds_t
*zd
, uint64_t id
)
4084 od
= umem_alloc(sizeof(ztest_od_t
), UMEM_NOFAIL
);
4085 uint64_t offset
= (1ULL << (ztest_random(20) + 43)) +
4086 (ztest_random(ZTEST_RANGE_LOCKS
) << SPA_MAXBLOCKSHIFT
);
4089 * Have multiple threads write to large offsets in an object
4090 * to verify that parallel writes to an object -- even to the
4091 * same blocks within the object -- doesn't cause any trouble.
4093 ztest_od_init(od
, ID_PARALLEL
, FTAG
, 0, DMU_OT_UINT64_OTHER
, 0, 0);
4095 if (ztest_object_init(zd
, od
, sizeof (ztest_od_t
), B_FALSE
) != 0)
4098 while (ztest_random(10) != 0)
4099 ztest_io(zd
, od
->od_object
, offset
);
4101 umem_free(od
, sizeof(ztest_od_t
));
4105 ztest_dmu_prealloc(ztest_ds_t
*zd
, uint64_t id
)
4108 uint64_t offset
= (1ULL << (ztest_random(4) + SPA_MAXBLOCKSHIFT
)) +
4109 (ztest_random(ZTEST_RANGE_LOCKS
) << SPA_MAXBLOCKSHIFT
);
4110 uint64_t count
= ztest_random(20) + 1;
4111 uint64_t blocksize
= ztest_random_blocksize();
4114 od
= umem_alloc(sizeof(ztest_od_t
), UMEM_NOFAIL
);
4116 ztest_od_init(od
, id
, FTAG
, 0, DMU_OT_UINT64_OTHER
, blocksize
, 0);
4118 if (ztest_object_init(zd
, od
, sizeof (ztest_od_t
), !ztest_random(2)) != 0) {
4119 umem_free(od
, sizeof(ztest_od_t
));
4123 if (ztest_truncate(zd
, od
->od_object
, offset
, count
* blocksize
) != 0) {
4124 umem_free(od
, sizeof(ztest_od_t
));
4128 ztest_prealloc(zd
, od
->od_object
, offset
, count
* blocksize
);
4130 data
= umem_zalloc(blocksize
, UMEM_NOFAIL
);
4132 while (ztest_random(count
) != 0) {
4133 uint64_t randoff
= offset
+ (ztest_random(count
) * blocksize
);
4134 if (ztest_write(zd
, od
->od_object
, randoff
, blocksize
,
4137 while (ztest_random(4) != 0)
4138 ztest_io(zd
, od
->od_object
, randoff
);
4141 umem_free(data
, blocksize
);
4142 umem_free(od
, sizeof(ztest_od_t
));
4146 * Verify that zap_{create,destroy,add,remove,update} work as expected.
4148 #define ZTEST_ZAP_MIN_INTS 1
4149 #define ZTEST_ZAP_MAX_INTS 4
4150 #define ZTEST_ZAP_MAX_PROPS 1000
4153 ztest_zap(ztest_ds_t
*zd
, uint64_t id
)
4155 objset_t
*os
= zd
->zd_os
;
4158 uint64_t txg
, last_txg
;
4159 uint64_t value
[ZTEST_ZAP_MAX_INTS
];
4160 uint64_t zl_ints
, zl_intsize
, prop
;
4163 char propname
[100], txgname
[100];
4165 char *hc
[2] = { "s.acl.h", ".s.open.h.hyLZlg" };
4167 od
= umem_alloc(sizeof(ztest_od_t
), UMEM_NOFAIL
);
4168 ztest_od_init(od
, id
, FTAG
, 0, DMU_OT_ZAP_OTHER
, 0, 0);
4170 if (ztest_object_init(zd
, od
, sizeof (ztest_od_t
),
4171 !ztest_random(2)) != 0)
4174 object
= od
->od_object
;
4177 * Generate a known hash collision, and verify that
4178 * we can lookup and remove both entries.
4180 tx
= dmu_tx_create(os
);
4181 dmu_tx_hold_zap(tx
, object
, B_TRUE
, NULL
);
4182 txg
= ztest_tx_assign(tx
, TXG_MIGHTWAIT
, FTAG
);
4185 for (i
= 0; i
< 2; i
++) {
4187 VERIFY3U(0, ==, zap_add(os
, object
, hc
[i
], sizeof (uint64_t),
4190 for (i
= 0; i
< 2; i
++) {
4191 VERIFY3U(EEXIST
, ==, zap_add(os
, object
, hc
[i
],
4192 sizeof (uint64_t), 1, &value
[i
], tx
));
4194 zap_length(os
, object
, hc
[i
], &zl_intsize
, &zl_ints
));
4195 ASSERT3U(zl_intsize
, ==, sizeof (uint64_t));
4196 ASSERT3U(zl_ints
, ==, 1);
4198 for (i
= 0; i
< 2; i
++) {
4199 VERIFY3U(0, ==, zap_remove(os
, object
, hc
[i
], tx
));
4204 * Generate a buch of random entries.
4206 ints
= MAX(ZTEST_ZAP_MIN_INTS
, object
% ZTEST_ZAP_MAX_INTS
);
4208 prop
= ztest_random(ZTEST_ZAP_MAX_PROPS
);
4209 (void) sprintf(propname
, "prop_%llu", (u_longlong_t
)prop
);
4210 (void) sprintf(txgname
, "txg_%llu", (u_longlong_t
)prop
);
4211 bzero(value
, sizeof (value
));
4215 * If these zap entries already exist, validate their contents.
4217 error
= zap_length(os
, object
, txgname
, &zl_intsize
, &zl_ints
);
4219 ASSERT3U(zl_intsize
, ==, sizeof (uint64_t));
4220 ASSERT3U(zl_ints
, ==, 1);
4222 VERIFY(zap_lookup(os
, object
, txgname
, zl_intsize
,
4223 zl_ints
, &last_txg
) == 0);
4225 VERIFY(zap_length(os
, object
, propname
, &zl_intsize
,
4228 ASSERT3U(zl_intsize
, ==, sizeof (uint64_t));
4229 ASSERT3U(zl_ints
, ==, ints
);
4231 VERIFY(zap_lookup(os
, object
, propname
, zl_intsize
,
4232 zl_ints
, value
) == 0);
4234 for (i
= 0; i
< ints
; i
++) {
4235 ASSERT3U(value
[i
], ==, last_txg
+ object
+ i
);
4238 ASSERT3U(error
, ==, ENOENT
);
4242 * Atomically update two entries in our zap object.
4243 * The first is named txg_%llu, and contains the txg
4244 * in which the property was last updated. The second
4245 * is named prop_%llu, and the nth element of its value
4246 * should be txg + object + n.
4248 tx
= dmu_tx_create(os
);
4249 dmu_tx_hold_zap(tx
, object
, B_TRUE
, NULL
);
4250 txg
= ztest_tx_assign(tx
, TXG_MIGHTWAIT
, FTAG
);
4255 fatal(0, "zap future leak: old %llu new %llu", last_txg
, txg
);
4257 for (i
= 0; i
< ints
; i
++)
4258 value
[i
] = txg
+ object
+ i
;
4260 VERIFY3U(0, ==, zap_update(os
, object
, txgname
, sizeof (uint64_t),
4262 VERIFY3U(0, ==, zap_update(os
, object
, propname
, sizeof (uint64_t),
4268 * Remove a random pair of entries.
4270 prop
= ztest_random(ZTEST_ZAP_MAX_PROPS
);
4271 (void) sprintf(propname
, "prop_%llu", (u_longlong_t
)prop
);
4272 (void) sprintf(txgname
, "txg_%llu", (u_longlong_t
)prop
);
4274 error
= zap_length(os
, object
, txgname
, &zl_intsize
, &zl_ints
);
4276 if (error
== ENOENT
)
4281 tx
= dmu_tx_create(os
);
4282 dmu_tx_hold_zap(tx
, object
, B_TRUE
, NULL
);
4283 txg
= ztest_tx_assign(tx
, TXG_MIGHTWAIT
, FTAG
);
4286 VERIFY3U(0, ==, zap_remove(os
, object
, txgname
, tx
));
4287 VERIFY3U(0, ==, zap_remove(os
, object
, propname
, tx
));
4290 umem_free(od
, sizeof(ztest_od_t
));
4294 * Testcase to test the upgrading of a microzap to fatzap.
4297 ztest_fzap(ztest_ds_t
*zd
, uint64_t id
)
4299 objset_t
*os
= zd
->zd_os
;
4301 uint64_t object
, txg
;
4304 od
= umem_alloc(sizeof(ztest_od_t
), UMEM_NOFAIL
);
4305 ztest_od_init(od
, id
, FTAG
, 0, DMU_OT_ZAP_OTHER
, 0, 0);
4307 if (ztest_object_init(zd
, od
, sizeof (ztest_od_t
),
4308 !ztest_random(2)) != 0)
4310 object
= od
->od_object
;
4313 * Add entries to this ZAP and make sure it spills over
4314 * and gets upgraded to a fatzap. Also, since we are adding
4315 * 2050 entries we should see ptrtbl growth and leaf-block split.
4317 for (i
= 0; i
< 2050; i
++) {
4318 char name
[MAXNAMELEN
];
4323 (void) snprintf(name
, sizeof (name
), "fzap-%llu-%llu",
4324 (u_longlong_t
)id
, (u_longlong_t
)value
);
4326 tx
= dmu_tx_create(os
);
4327 dmu_tx_hold_zap(tx
, object
, B_TRUE
, name
);
4328 txg
= ztest_tx_assign(tx
, TXG_MIGHTWAIT
, FTAG
);
4331 error
= zap_add(os
, object
, name
, sizeof (uint64_t), 1,
4333 ASSERT(error
== 0 || error
== EEXIST
);
4337 umem_free(od
, sizeof(ztest_od_t
));
4342 ztest_zap_parallel(ztest_ds_t
*zd
, uint64_t id
)
4344 objset_t
*os
= zd
->zd_os
;
4346 uint64_t txg
, object
, count
, wsize
, wc
, zl_wsize
, zl_wc
;
4348 int i
, namelen
, error
;
4349 int micro
= ztest_random(2);
4350 char name
[20], string_value
[20];
4353 od
= umem_alloc(sizeof(ztest_od_t
), UMEM_NOFAIL
);
4354 ztest_od_init(od
, ID_PARALLEL
, FTAG
, micro
, DMU_OT_ZAP_OTHER
, 0, 0);
4356 if (ztest_object_init(zd
, od
, sizeof (ztest_od_t
), B_FALSE
) != 0) {
4357 umem_free(od
, sizeof(ztest_od_t
));
4361 object
= od
->od_object
;
4364 * Generate a random name of the form 'xxx.....' where each
4365 * x is a random printable character and the dots are dots.
4366 * There are 94 such characters, and the name length goes from
4367 * 6 to 20, so there are 94^3 * 15 = 12,458,760 possible names.
4369 namelen
= ztest_random(sizeof (name
) - 5) + 5 + 1;
4371 for (i
= 0; i
< 3; i
++)
4372 name
[i
] = '!' + ztest_random('~' - '!' + 1);
4373 for (; i
< namelen
- 1; i
++)
4377 if ((namelen
& 1) || micro
) {
4378 wsize
= sizeof (txg
);
4384 data
= string_value
;
4388 VERIFY0(zap_count(os
, object
, &count
));
4389 ASSERT(count
!= -1ULL);
4392 * Select an operation: length, lookup, add, update, remove.
4394 i
= ztest_random(5);
4397 tx
= dmu_tx_create(os
);
4398 dmu_tx_hold_zap(tx
, object
, B_TRUE
, NULL
);
4399 txg
= ztest_tx_assign(tx
, TXG_MIGHTWAIT
, FTAG
);
4402 bcopy(name
, string_value
, namelen
);
4406 bzero(string_value
, namelen
);
4412 error
= zap_length(os
, object
, name
, &zl_wsize
, &zl_wc
);
4414 ASSERT3U(wsize
, ==, zl_wsize
);
4415 ASSERT3U(wc
, ==, zl_wc
);
4417 ASSERT3U(error
, ==, ENOENT
);
4422 error
= zap_lookup(os
, object
, name
, wsize
, wc
, data
);
4424 if (data
== string_value
&&
4425 bcmp(name
, data
, namelen
) != 0)
4426 fatal(0, "name '%s' != val '%s' len %d",
4427 name
, data
, namelen
);
4429 ASSERT3U(error
, ==, ENOENT
);
4434 error
= zap_add(os
, object
, name
, wsize
, wc
, data
, tx
);
4435 ASSERT(error
== 0 || error
== EEXIST
);
4439 VERIFY(zap_update(os
, object
, name
, wsize
, wc
, data
, tx
) == 0);
4443 error
= zap_remove(os
, object
, name
, tx
);
4444 ASSERT(error
== 0 || error
== ENOENT
);
4451 umem_free(od
, sizeof(ztest_od_t
));
4455 * Commit callback data.
4457 typedef struct ztest_cb_data
{
4458 list_node_t zcd_node
;
4460 int zcd_expected_err
;
4461 boolean_t zcd_added
;
4462 boolean_t zcd_called
;
4466 /* This is the actual commit callback function */
4468 ztest_commit_callback(void *arg
, int error
)
4470 ztest_cb_data_t
*data
= arg
;
4471 uint64_t synced_txg
;
4473 VERIFY(data
!= NULL
);
4474 VERIFY3S(data
->zcd_expected_err
, ==, error
);
4475 VERIFY(!data
->zcd_called
);
4477 synced_txg
= spa_last_synced_txg(data
->zcd_spa
);
4478 if (data
->zcd_txg
> synced_txg
)
4479 fatal(0, "commit callback of txg %" PRIu64
" called prematurely"
4480 ", last synced txg = %" PRIu64
"\n", data
->zcd_txg
,
4483 data
->zcd_called
= B_TRUE
;
4485 if (error
== ECANCELED
) {
4486 ASSERT0(data
->zcd_txg
);
4487 ASSERT(!data
->zcd_added
);
4490 * The private callback data should be destroyed here, but
4491 * since we are going to check the zcd_called field after
4492 * dmu_tx_abort(), we will destroy it there.
4497 ASSERT(data
->zcd_added
);
4498 ASSERT3U(data
->zcd_txg
, !=, 0);
4500 (void) mutex_enter(&zcl
.zcl_callbacks_lock
);
4502 /* See if this cb was called more quickly */
4503 if ((synced_txg
- data
->zcd_txg
) < zc_min_txg_delay
)
4504 zc_min_txg_delay
= synced_txg
- data
->zcd_txg
;
4506 /* Remove our callback from the list */
4507 list_remove(&zcl
.zcl_callbacks
, data
);
4509 (void) mutex_exit(&zcl
.zcl_callbacks_lock
);
4511 umem_free(data
, sizeof (ztest_cb_data_t
));
4514 /* Allocate and initialize callback data structure */
4515 static ztest_cb_data_t
*
4516 ztest_create_cb_data(objset_t
*os
, uint64_t txg
)
4518 ztest_cb_data_t
*cb_data
;
4520 cb_data
= umem_zalloc(sizeof (ztest_cb_data_t
), UMEM_NOFAIL
);
4522 cb_data
->zcd_txg
= txg
;
4523 cb_data
->zcd_spa
= dmu_objset_spa(os
);
4524 list_link_init(&cb_data
->zcd_node
);
4530 * Commit callback test.
4533 ztest_dmu_commit_callbacks(ztest_ds_t
*zd
, uint64_t id
)
4535 objset_t
*os
= zd
->zd_os
;
4538 ztest_cb_data_t
*cb_data
[3], *tmp_cb
;
4539 uint64_t old_txg
, txg
;
4542 od
= umem_alloc(sizeof(ztest_od_t
), UMEM_NOFAIL
);
4543 ztest_od_init(od
, id
, FTAG
, 0, DMU_OT_UINT64_OTHER
, 0, 0);
4545 if (ztest_object_init(zd
, od
, sizeof (ztest_od_t
), B_FALSE
) != 0) {
4546 umem_free(od
, sizeof(ztest_od_t
));
4550 tx
= dmu_tx_create(os
);
4552 cb_data
[0] = ztest_create_cb_data(os
, 0);
4553 dmu_tx_callback_register(tx
, ztest_commit_callback
, cb_data
[0]);
4555 dmu_tx_hold_write(tx
, od
->od_object
, 0, sizeof (uint64_t));
4557 /* Every once in a while, abort the transaction on purpose */
4558 if (ztest_random(100) == 0)
4562 error
= dmu_tx_assign(tx
, TXG_NOWAIT
);
4564 txg
= error
? 0 : dmu_tx_get_txg(tx
);
4566 cb_data
[0]->zcd_txg
= txg
;
4567 cb_data
[1] = ztest_create_cb_data(os
, txg
);
4568 dmu_tx_callback_register(tx
, ztest_commit_callback
, cb_data
[1]);
4572 * It's not a strict requirement to call the registered
4573 * callbacks from inside dmu_tx_abort(), but that's what
4574 * it's supposed to happen in the current implementation
4575 * so we will check for that.
4577 for (i
= 0; i
< 2; i
++) {
4578 cb_data
[i
]->zcd_expected_err
= ECANCELED
;
4579 VERIFY(!cb_data
[i
]->zcd_called
);
4584 for (i
= 0; i
< 2; i
++) {
4585 VERIFY(cb_data
[i
]->zcd_called
);
4586 umem_free(cb_data
[i
], sizeof (ztest_cb_data_t
));
4589 umem_free(od
, sizeof(ztest_od_t
));
4593 cb_data
[2] = ztest_create_cb_data(os
, txg
);
4594 dmu_tx_callback_register(tx
, ztest_commit_callback
, cb_data
[2]);
4597 * Read existing data to make sure there isn't a future leak.
4599 VERIFY(0 == dmu_read(os
, od
->od_object
, 0, sizeof (uint64_t),
4600 &old_txg
, DMU_READ_PREFETCH
));
4603 fatal(0, "future leak: got %" PRIu64
", open txg is %" PRIu64
,
4606 dmu_write(os
, od
->od_object
, 0, sizeof (uint64_t), &txg
, tx
);
4608 (void) mutex_enter(&zcl
.zcl_callbacks_lock
);
4611 * Since commit callbacks don't have any ordering requirement and since
4612 * it is theoretically possible for a commit callback to be called
4613 * after an arbitrary amount of time has elapsed since its txg has been
4614 * synced, it is difficult to reliably determine whether a commit
4615 * callback hasn't been called due to high load or due to a flawed
4618 * In practice, we will assume that if after a certain number of txgs a
4619 * commit callback hasn't been called, then most likely there's an
4620 * implementation bug..
4622 tmp_cb
= list_head(&zcl
.zcl_callbacks
);
4623 if (tmp_cb
!= NULL
&&
4624 tmp_cb
->zcd_txg
+ ZTEST_COMMIT_CB_THRESH
< txg
) {
4625 fatal(0, "Commit callback threshold exceeded, oldest txg: %"
4626 PRIu64
", open txg: %" PRIu64
"\n", tmp_cb
->zcd_txg
, txg
);
4630 * Let's find the place to insert our callbacks.
4632 * Even though the list is ordered by txg, it is possible for the
4633 * insertion point to not be the end because our txg may already be
4634 * quiescing at this point and other callbacks in the open txg
4635 * (from other objsets) may have sneaked in.
4637 tmp_cb
= list_tail(&zcl
.zcl_callbacks
);
4638 while (tmp_cb
!= NULL
&& tmp_cb
->zcd_txg
> txg
)
4639 tmp_cb
= list_prev(&zcl
.zcl_callbacks
, tmp_cb
);
4641 /* Add the 3 callbacks to the list */
4642 for (i
= 0; i
< 3; i
++) {
4644 list_insert_head(&zcl
.zcl_callbacks
, cb_data
[i
]);
4646 list_insert_after(&zcl
.zcl_callbacks
, tmp_cb
,
4649 cb_data
[i
]->zcd_added
= B_TRUE
;
4650 VERIFY(!cb_data
[i
]->zcd_called
);
4652 tmp_cb
= cb_data
[i
];
4657 (void) mutex_exit(&zcl
.zcl_callbacks_lock
);
4661 umem_free(od
, sizeof(ztest_od_t
));
4666 ztest_dsl_prop_get_set(ztest_ds_t
*zd
, uint64_t id
)
4668 zfs_prop_t proplist
[] = {
4670 ZFS_PROP_COMPRESSION
,
4676 (void) rw_enter(&ztest_name_lock
, RW_READER
);
4678 for (p
= 0; p
< sizeof (proplist
) / sizeof (proplist
[0]); p
++)
4679 (void) ztest_dsl_prop_set_uint64(zd
->zd_name
, proplist
[p
],
4680 ztest_random_dsl_prop(proplist
[p
]), (int)ztest_random(2));
4682 (void) rw_exit(&ztest_name_lock
);
4687 ztest_spa_prop_get_set(ztest_ds_t
*zd
, uint64_t id
)
4689 nvlist_t
*props
= NULL
;
4691 (void) rw_enter(&ztest_name_lock
, RW_READER
);
4693 (void) ztest_spa_prop_set_uint64(ZPOOL_PROP_DEDUPDITTO
,
4694 ZIO_DEDUPDITTO_MIN
+ ztest_random(ZIO_DEDUPDITTO_MIN
));
4696 VERIFY0(spa_prop_get(ztest_spa
, &props
));
4698 if (ztest_opts
.zo_verbose
>= 6)
4699 dump_nvlist(props
, 4);
4703 (void) rw_exit(&ztest_name_lock
);
4707 user_release_one(const char *snapname
, const char *holdname
)
4709 nvlist_t
*snaps
, *holds
;
4712 snaps
= fnvlist_alloc();
4713 holds
= fnvlist_alloc();
4714 fnvlist_add_boolean(holds
, holdname
);
4715 fnvlist_add_nvlist(snaps
, snapname
, holds
);
4716 fnvlist_free(holds
);
4717 error
= dsl_dataset_user_release(snaps
, NULL
);
4718 fnvlist_free(snaps
);
4723 * Test snapshot hold/release and deferred destroy.
4726 ztest_dmu_snapshot_hold(ztest_ds_t
*zd
, uint64_t id
)
4729 objset_t
*os
= zd
->zd_os
;
4733 char clonename
[100];
4735 char osname
[MAXNAMELEN
];
4738 (void) rw_enter(&ztest_name_lock
, RW_READER
);
4740 dmu_objset_name(os
, osname
);
4742 (void) snprintf(snapname
, sizeof (snapname
), "sh1_%llu", (long long unsigned int)id
);
4743 (void) snprintf(fullname
, sizeof (fullname
), "%s@%s", osname
, snapname
);
4744 (void) snprintf(clonename
, sizeof (clonename
),
4745 "%s/ch1_%llu", osname
, (long long unsigned int)id
);
4746 (void) snprintf(tag
, sizeof (tag
), "tag_%llu", (long long unsigned int)id
);
4749 * Clean up from any previous run.
4751 error
= dsl_destroy_head(clonename
);
4752 if (error
!= ENOENT
)
4754 error
= user_release_one(fullname
, tag
);
4755 if (error
!= ESRCH
&& error
!= ENOENT
)
4757 error
= dsl_destroy_snapshot(fullname
, B_FALSE
);
4758 if (error
!= ENOENT
)
4762 * Create snapshot, clone it, mark snap for deferred destroy,
4763 * destroy clone, verify snap was also destroyed.
4765 error
= dmu_objset_snapshot_one(osname
, snapname
);
4767 if (error
== ENOSPC
) {
4768 ztest_record_enospc("dmu_objset_snapshot");
4771 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname
, error
);
4774 error
= dmu_objset_clone(clonename
, fullname
);
4776 if (error
== ENOSPC
) {
4777 ztest_record_enospc("dmu_objset_clone");
4780 fatal(0, "dmu_objset_clone(%s) = %d", clonename
, error
);
4783 error
= dsl_destroy_snapshot(fullname
, B_TRUE
);
4785 fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
4789 error
= dsl_destroy_head(clonename
);
4791 fatal(0, "dsl_destroy_head(%s) = %d", clonename
, error
);
4793 error
= dmu_objset_hold(fullname
, FTAG
, &origin
);
4794 if (error
!= ENOENT
)
4795 fatal(0, "dmu_objset_hold(%s) = %d", fullname
, error
);
4798 * Create snapshot, add temporary hold, verify that we can't
4799 * destroy a held snapshot, mark for deferred destroy,
4800 * release hold, verify snapshot was destroyed.
4802 error
= dmu_objset_snapshot_one(osname
, snapname
);
4804 if (error
== ENOSPC
) {
4805 ztest_record_enospc("dmu_objset_snapshot");
4808 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname
, error
);
4811 holds
= fnvlist_alloc();
4812 fnvlist_add_string(holds
, fullname
, tag
);
4813 error
= dsl_dataset_user_hold(holds
, 0, NULL
);
4814 fnvlist_free(holds
);
4817 fatal(0, "dsl_dataset_user_hold(%s)", fullname
, tag
);
4819 error
= dsl_destroy_snapshot(fullname
, B_FALSE
);
4820 if (error
!= EBUSY
) {
4821 fatal(0, "dsl_destroy_snapshot(%s, B_FALSE) = %d",
4825 error
= dsl_destroy_snapshot(fullname
, B_TRUE
);
4827 fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
4831 error
= user_release_one(fullname
, tag
);
4833 fatal(0, "user_release_one(%s)", fullname
, tag
);
4835 VERIFY3U(dmu_objset_hold(fullname
, FTAG
, &origin
), ==, ENOENT
);
4838 (void) rw_exit(&ztest_name_lock
);
4842 * Inject random faults into the on-disk data.
4846 ztest_fault_inject(ztest_ds_t
*zd
, uint64_t id
)
4848 ztest_shared_t
*zs
= ztest_shared
;
4849 spa_t
*spa
= ztest_spa
;
4853 uint64_t bad
= 0x1990c0ffeedecadeull
;
4858 int bshift
= SPA_MAXBLOCKSHIFT
+ 2; /* don't scrog all labels */
4864 boolean_t islog
= B_FALSE
;
4866 path0
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
4867 pathrand
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
4869 mutex_enter(&ztest_vdev_lock
);
4870 maxfaults
= MAXFAULTS();
4871 leaves
= MAX(zs
->zs_mirrors
, 1) * ztest_opts
.zo_raidz
;
4872 mirror_save
= zs
->zs_mirrors
;
4873 mutex_exit(&ztest_vdev_lock
);
4875 ASSERT(leaves
>= 1);
4878 * We need SCL_STATE here because we're going to look at vd0->vdev_tsd.
4880 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
4882 if (ztest_random(2) == 0) {
4884 * Inject errors on a normal data device or slog device.
4886 top
= ztest_random_vdev_top(spa
, B_TRUE
);
4887 leaf
= ztest_random(leaves
) + zs
->zs_splits
;
4890 * Generate paths to the first leaf in this top-level vdev,
4891 * and to the random leaf we selected. We'll induce transient
4892 * write failures and random online/offline activity on leaf 0,
4893 * and we'll write random garbage to the randomly chosen leaf.
4895 (void) snprintf(path0
, MAXPATHLEN
, ztest_dev_template
,
4896 ztest_opts
.zo_dir
, ztest_opts
.zo_pool
,
4897 top
* leaves
+ zs
->zs_splits
);
4898 (void) snprintf(pathrand
, MAXPATHLEN
, ztest_dev_template
,
4899 ztest_opts
.zo_dir
, ztest_opts
.zo_pool
,
4900 top
* leaves
+ leaf
);
4902 vd0
= vdev_lookup_by_path(spa
->spa_root_vdev
, path0
);
4903 if (vd0
!= NULL
&& vd0
->vdev_top
->vdev_islog
)
4906 if (vd0
!= NULL
&& maxfaults
!= 1) {
4908 * Make vd0 explicitly claim to be unreadable,
4909 * or unwriteable, or reach behind its back
4910 * and close the underlying fd. We can do this if
4911 * maxfaults == 0 because we'll fail and reexecute,
4912 * and we can do it if maxfaults >= 2 because we'll
4913 * have enough redundancy. If maxfaults == 1, the
4914 * combination of this with injection of random data
4915 * corruption below exceeds the pool's fault tolerance.
4917 vdev_file_t
*vf
= vd0
->vdev_tsd
;
4919 if (vf
!= NULL
&& ztest_random(3) == 0) {
4920 (void) close(vf
->vf_vnode
->v_fd
);
4921 vf
->vf_vnode
->v_fd
= -1;
4922 } else if (ztest_random(2) == 0) {
4923 vd0
->vdev_cant_read
= B_TRUE
;
4925 vd0
->vdev_cant_write
= B_TRUE
;
4927 guid0
= vd0
->vdev_guid
;
4931 * Inject errors on an l2cache device.
4933 spa_aux_vdev_t
*sav
= &spa
->spa_l2cache
;
4935 if (sav
->sav_count
== 0) {
4936 spa_config_exit(spa
, SCL_STATE
, FTAG
);
4939 vd0
= sav
->sav_vdevs
[ztest_random(sav
->sav_count
)];
4940 guid0
= vd0
->vdev_guid
;
4941 (void) strcpy(path0
, vd0
->vdev_path
);
4942 (void) strcpy(pathrand
, vd0
->vdev_path
);
4946 maxfaults
= INT_MAX
; /* no limit on cache devices */
4949 spa_config_exit(spa
, SCL_STATE
, FTAG
);
4952 * If we can tolerate two or more faults, or we're dealing
4953 * with a slog, randomly online/offline vd0.
4955 if ((maxfaults
>= 2 || islog
) && guid0
!= 0) {
4956 if (ztest_random(10) < 6) {
4957 int flags
= (ztest_random(2) == 0 ?
4958 ZFS_OFFLINE_TEMPORARY
: 0);
4961 * We have to grab the zs_name_lock as writer to
4962 * prevent a race between offlining a slog and
4963 * destroying a dataset. Offlining the slog will
4964 * grab a reference on the dataset which may cause
4965 * dsl_destroy_head() to fail with EBUSY thus
4966 * leaving the dataset in an inconsistent state.
4969 (void) rw_enter(&ztest_name_lock
,
4972 VERIFY(vdev_offline(spa
, guid0
, flags
) != EBUSY
);
4975 (void) rw_exit(&ztest_name_lock
);
4978 * Ideally we would like to be able to randomly
4979 * call vdev_[on|off]line without holding locks
4980 * to force unpredictable failures but the side
4981 * effects of vdev_[on|off]line prevent us from
4982 * doing so. We grab the ztest_vdev_lock here to
4983 * prevent a race between injection testing and
4986 mutex_enter(&ztest_vdev_lock
);
4987 (void) vdev_online(spa
, guid0
, 0, NULL
);
4988 mutex_exit(&ztest_vdev_lock
);
4996 * We have at least single-fault tolerance, so inject data corruption.
4998 fd
= open(pathrand
, O_RDWR
);
5000 if (fd
== -1) /* we hit a gap in the device namespace */
5003 fsize
= lseek(fd
, 0, SEEK_END
);
5005 while (--iters
!= 0) {
5006 offset
= ztest_random(fsize
/ (leaves
<< bshift
)) *
5007 (leaves
<< bshift
) + (leaf
<< bshift
) +
5008 (ztest_random(1ULL << (bshift
- 1)) & -8ULL);
5010 if (offset
>= fsize
)
5013 mutex_enter(&ztest_vdev_lock
);
5014 if (mirror_save
!= zs
->zs_mirrors
) {
5015 mutex_exit(&ztest_vdev_lock
);
5020 if (pwrite(fd
, &bad
, sizeof (bad
), offset
) != sizeof (bad
))
5021 fatal(1, "can't inject bad word at 0x%llx in %s",
5024 mutex_exit(&ztest_vdev_lock
);
5026 if (ztest_opts
.zo_verbose
>= 7)
5027 (void) printf("injected bad word into %s,"
5028 " offset 0x%llx\n", pathrand
, (u_longlong_t
)offset
);
5033 umem_free(path0
, MAXPATHLEN
);
5034 umem_free(pathrand
, MAXPATHLEN
);
5038 * Verify that DDT repair works as expected.
5041 ztest_ddt_repair(ztest_ds_t
*zd
, uint64_t id
)
5043 ztest_shared_t
*zs
= ztest_shared
;
5044 spa_t
*spa
= ztest_spa
;
5045 objset_t
*os
= zd
->zd_os
;
5047 uint64_t object
, blocksize
, txg
, pattern
, psize
;
5048 enum zio_checksum checksum
= spa_dedup_checksum(spa
);
5053 int copies
= 2 * ZIO_DEDUPDITTO_MIN
;
5056 blocksize
= ztest_random_blocksize();
5057 blocksize
= MIN(blocksize
, 2048); /* because we write so many */
5059 od
= umem_alloc(sizeof(ztest_od_t
), UMEM_NOFAIL
);
5060 ztest_od_init(od
, id
, FTAG
, 0, DMU_OT_UINT64_OTHER
, blocksize
, 0);
5062 if (ztest_object_init(zd
, od
, sizeof (ztest_od_t
), B_FALSE
) != 0) {
5063 umem_free(od
, sizeof(ztest_od_t
));
5068 * Take the name lock as writer to prevent anyone else from changing
5069 * the pool and dataset properies we need to maintain during this test.
5071 (void) rw_enter(&ztest_name_lock
, RW_WRITER
);
5073 if (ztest_dsl_prop_set_uint64(zd
->zd_name
, ZFS_PROP_DEDUP
, checksum
,
5075 ztest_dsl_prop_set_uint64(zd
->zd_name
, ZFS_PROP_COPIES
, 1,
5077 (void) rw_exit(&ztest_name_lock
);
5078 umem_free(od
, sizeof(ztest_od_t
));
5082 object
= od
[0].od_object
;
5083 blocksize
= od
[0].od_blocksize
;
5084 pattern
= zs
->zs_guid
^ dmu_objset_fsid_guid(os
);
5086 ASSERT(object
!= 0);
5088 tx
= dmu_tx_create(os
);
5089 dmu_tx_hold_write(tx
, object
, 0, copies
* blocksize
);
5090 txg
= ztest_tx_assign(tx
, TXG_WAIT
, FTAG
);
5092 (void) rw_exit(&ztest_name_lock
);
5093 umem_free(od
, sizeof(ztest_od_t
));
5098 * Write all the copies of our block.
5100 for (i
= 0; i
< copies
; i
++) {
5101 uint64_t offset
= i
* blocksize
;
5102 int error
= dmu_buf_hold(os
, object
, offset
, FTAG
, &db
,
5103 DMU_READ_NO_PREFETCH
);
5105 fatal(B_FALSE
, "dmu_buf_hold(%p, %llu, %llu) = %u",
5106 os
, (long long)object
, (long long) offset
, error
);
5108 ASSERT(db
->db_offset
== offset
);
5109 ASSERT(db
->db_size
== blocksize
);
5110 ASSERT(ztest_pattern_match(db
->db_data
, db
->db_size
, pattern
) ||
5111 ztest_pattern_match(db
->db_data
, db
->db_size
, 0ULL));
5112 dmu_buf_will_fill(db
, tx
);
5113 ztest_pattern_set(db
->db_data
, db
->db_size
, pattern
);
5114 dmu_buf_rele(db
, FTAG
);
5118 txg_wait_synced(spa_get_dsl(spa
), txg
);
5121 * Find out what block we got.
5123 VERIFY(dmu_buf_hold(os
, object
, 0, FTAG
, &db
,
5124 DMU_READ_NO_PREFETCH
) == 0);
5125 blk
= *((dmu_buf_impl_t
*)db
)->db_blkptr
;
5126 dmu_buf_rele(db
, FTAG
);
5129 * Damage the block. Dedup-ditto will save us when we read it later.
5131 psize
= BP_GET_PSIZE(&blk
);
5132 buf
= zio_buf_alloc(psize
);
5133 ztest_pattern_set(buf
, psize
, ~pattern
);
5135 (void) zio_wait(zio_rewrite(NULL
, spa
, 0, &blk
,
5136 buf
, psize
, NULL
, NULL
, ZIO_PRIORITY_SYNC_WRITE
,
5137 ZIO_FLAG_CANFAIL
| ZIO_FLAG_INDUCE_DAMAGE
, NULL
));
5139 zio_buf_free(buf
, psize
);
5141 (void) rw_exit(&ztest_name_lock
);
5142 umem_free(od
, sizeof(ztest_od_t
));
5150 ztest_scrub(ztest_ds_t
*zd
, uint64_t id
)
5152 spa_t
*spa
= ztest_spa
;
5154 (void) spa_scan(spa
, POOL_SCAN_SCRUB
);
5155 (void) poll(NULL
, 0, 100); /* wait a moment, then force a restart */
5156 (void) spa_scan(spa
, POOL_SCAN_SCRUB
);
5160 * Change the guid for the pool.
5164 ztest_reguid(ztest_ds_t
*zd
, uint64_t id
)
5166 spa_t
*spa
= ztest_spa
;
5167 uint64_t orig
, load
;
5170 orig
= spa_guid(spa
);
5171 load
= spa_load_guid(spa
);
5173 (void) rw_enter(&ztest_name_lock
, RW_WRITER
);
5174 error
= spa_change_guid(spa
);
5175 (void) rw_exit(&ztest_name_lock
);
5180 if (ztest_opts
.zo_verbose
>= 4) {
5181 (void) printf("Changed guid old %llu -> %llu\n",
5182 (u_longlong_t
)orig
, (u_longlong_t
)spa_guid(spa
));
5185 VERIFY3U(orig
, !=, spa_guid(spa
));
5186 VERIFY3U(load
, ==, spa_load_guid(spa
));
5190 * Rename the pool to a different name and then rename it back.
5194 ztest_spa_rename(ztest_ds_t
*zd
, uint64_t id
)
5196 char *oldname
, *newname
;
5199 (void) rw_enter(&ztest_name_lock
, RW_WRITER
);
5201 oldname
= ztest_opts
.zo_pool
;
5202 newname
= umem_alloc(strlen(oldname
) + 5, UMEM_NOFAIL
);
5203 (void) strcpy(newname
, oldname
);
5204 (void) strcat(newname
, "_tmp");
5209 VERIFY3U(0, ==, spa_rename(oldname
, newname
));
5212 * Try to open it under the old name, which shouldn't exist
5214 VERIFY3U(ENOENT
, ==, spa_open(oldname
, &spa
, FTAG
));
5217 * Open it under the new name and make sure it's still the same spa_t.
5219 VERIFY3U(0, ==, spa_open(newname
, &spa
, FTAG
));
5221 ASSERT(spa
== ztest_spa
);
5222 spa_close(spa
, FTAG
);
5225 * Rename it back to the original
5227 VERIFY3U(0, ==, spa_rename(newname
, oldname
));
5230 * Make sure it can still be opened
5232 VERIFY3U(0, ==, spa_open(oldname
, &spa
, FTAG
));
5234 ASSERT(spa
== ztest_spa
);
5235 spa_close(spa
, FTAG
);
5237 umem_free(newname
, strlen(newname
) + 1);
5239 (void) rw_exit(&ztest_name_lock
);
5243 * Verify pool integrity by running zdb.
5246 ztest_run_zdb(char *pool
)
5254 bin
= umem_alloc(MAXPATHLEN
+ MAXNAMELEN
+ 20, UMEM_NOFAIL
);
5255 zdb
= umem_alloc(MAXPATHLEN
+ MAXNAMELEN
+ 20, UMEM_NOFAIL
);
5256 zbuf
= umem_alloc(1024, UMEM_NOFAIL
);
5258 VERIFY(realpath(getexecname(), bin
) != NULL
);
5259 if (strncmp(bin
, "/usr/sbin/ztest", 15) == 0) {
5260 strcpy(bin
, "/usr/sbin/zdb"); /* Installed */
5261 } else if (strncmp(bin
, "/sbin/ztest", 11) == 0) {
5262 strcpy(bin
, "/sbin/zdb"); /* Installed */
5264 strstr(bin
, "/ztest/")[0] = '\0'; /* In-tree */
5265 strcat(bin
, "/zdb/zdb");
5269 "%s -bcc%s%s -U %s %s",
5271 ztest_opts
.zo_verbose
>= 3 ? "s" : "",
5272 ztest_opts
.zo_verbose
>= 4 ? "v" : "",
5276 if (ztest_opts
.zo_verbose
>= 5)
5277 (void) printf("Executing %s\n", strstr(zdb
, "zdb "));
5279 fp
= popen(zdb
, "r");
5281 while (fgets(zbuf
, 1024, fp
) != NULL
)
5282 if (ztest_opts
.zo_verbose
>= 3)
5283 (void) printf("%s", zbuf
);
5285 status
= pclose(fp
);
5290 ztest_dump_core
= 0;
5291 if (WIFEXITED(status
))
5292 fatal(0, "'%s' exit code %d", zdb
, WEXITSTATUS(status
));
5294 fatal(0, "'%s' died with signal %d", zdb
, WTERMSIG(status
));
5296 umem_free(bin
, MAXPATHLEN
+ MAXNAMELEN
+ 20);
5297 umem_free(zdb
, MAXPATHLEN
+ MAXNAMELEN
+ 20);
5298 umem_free(zbuf
, 1024);
5302 ztest_walk_pool_directory(char *header
)
5306 if (ztest_opts
.zo_verbose
>= 6)
5307 (void) printf("%s\n", header
);
5309 mutex_enter(&spa_namespace_lock
);
5310 while ((spa
= spa_next(spa
)) != NULL
)
5311 if (ztest_opts
.zo_verbose
>= 6)
5312 (void) printf("\t%s\n", spa_name(spa
));
5313 mutex_exit(&spa_namespace_lock
);
5317 ztest_spa_import_export(char *oldname
, char *newname
)
5319 nvlist_t
*config
, *newconfig
;
5324 if (ztest_opts
.zo_verbose
>= 4) {
5325 (void) printf("import/export: old = %s, new = %s\n",
5330 * Clean up from previous runs.
5332 (void) spa_destroy(newname
);
5335 * Get the pool's configuration and guid.
5337 VERIFY3U(0, ==, spa_open(oldname
, &spa
, FTAG
));
5340 * Kick off a scrub to tickle scrub/export races.
5342 if (ztest_random(2) == 0)
5343 (void) spa_scan(spa
, POOL_SCAN_SCRUB
);
5345 pool_guid
= spa_guid(spa
);
5346 spa_close(spa
, FTAG
);
5348 ztest_walk_pool_directory("pools before export");
5353 VERIFY3U(0, ==, spa_export(oldname
, &config
, B_FALSE
, B_FALSE
));
5355 ztest_walk_pool_directory("pools after export");
5360 newconfig
= spa_tryimport(config
);
5361 ASSERT(newconfig
!= NULL
);
5362 nvlist_free(newconfig
);
5365 * Import it under the new name.
5367 error
= spa_import(newname
, config
, NULL
, 0);
5369 dump_nvlist(config
, 0);
5370 fatal(B_FALSE
, "couldn't import pool %s as %s: error %u",
5371 oldname
, newname
, error
);
5374 ztest_walk_pool_directory("pools after import");
5377 * Try to import it again -- should fail with EEXIST.
5379 VERIFY3U(EEXIST
, ==, spa_import(newname
, config
, NULL
, 0));
5382 * Try to import it under a different name -- should fail with EEXIST.
5384 VERIFY3U(EEXIST
, ==, spa_import(oldname
, config
, NULL
, 0));
5387 * Verify that the pool is no longer visible under the old name.
5389 VERIFY3U(ENOENT
, ==, spa_open(oldname
, &spa
, FTAG
));
5392 * Verify that we can open and close the pool using the new name.
5394 VERIFY3U(0, ==, spa_open(newname
, &spa
, FTAG
));
5395 ASSERT(pool_guid
== spa_guid(spa
));
5396 spa_close(spa
, FTAG
);
5398 nvlist_free(config
);
5402 ztest_resume(spa_t
*spa
)
5404 if (spa_suspended(spa
) && ztest_opts
.zo_verbose
>= 6)
5405 (void) printf("resuming from suspended state\n");
5406 spa_vdev_state_enter(spa
, SCL_NONE
);
5407 vdev_clear(spa
, NULL
);
5408 (void) spa_vdev_state_exit(spa
, NULL
, 0);
5409 (void) zio_resume(spa
);
5413 ztest_resume_thread(void *arg
)
5417 while (!ztest_exiting
) {
5418 if (spa_suspended(spa
))
5420 (void) poll(NULL
, 0, 100);
5432 ztest_deadman_alarm(int sig
)
5434 fatal(0, "failed to complete within %d seconds of deadline", GRACE
);
5439 ztest_execute(int test
, ztest_info_t
*zi
, uint64_t id
)
5441 ztest_ds_t
*zd
= &ztest_ds
[id
% ztest_opts
.zo_datasets
];
5442 ztest_shared_callstate_t
*zc
= ZTEST_GET_SHARED_CALLSTATE(test
);
5443 hrtime_t functime
= gethrtime();
5446 for (i
= 0; i
< zi
->zi_iters
; i
++)
5447 zi
->zi_func(zd
, id
);
5449 functime
= gethrtime() - functime
;
5451 atomic_add_64(&zc
->zc_count
, 1);
5452 atomic_add_64(&zc
->zc_time
, functime
);
5454 if (ztest_opts
.zo_verbose
>= 4) {
5456 (void) dladdr((void *)zi
->zi_func
, &dli
);
5457 (void) printf("%6.2f sec in %s\n",
5458 (double)functime
/ NANOSEC
, dli
.dli_sname
);
5463 ztest_thread(void *arg
)
5466 uint64_t id
= (uintptr_t)arg
;
5467 ztest_shared_t
*zs
= ztest_shared
;
5471 ztest_shared_callstate_t
*zc
;
5473 while ((now
= gethrtime()) < zs
->zs_thread_stop
) {
5475 * See if it's time to force a crash.
5477 if (now
> zs
->zs_thread_kill
)
5481 * If we're getting ENOSPC with some regularity, stop.
5483 if (zs
->zs_enospc_count
> 10)
5487 * Pick a random function to execute.
5489 rand
= ztest_random(ZTEST_FUNCS
);
5490 zi
= &ztest_info
[rand
];
5491 zc
= ZTEST_GET_SHARED_CALLSTATE(rand
);
5492 call_next
= zc
->zc_next
;
5494 if (now
>= call_next
&&
5495 atomic_cas_64(&zc
->zc_next
, call_next
, call_next
+
5496 ztest_random(2 * zi
->zi_interval
[0] + 1)) == call_next
) {
5497 ztest_execute(rand
, zi
, id
);
5507 ztest_dataset_name(char *dsname
, char *pool
, int d
)
5509 (void) snprintf(dsname
, MAXNAMELEN
, "%s/ds_%d", pool
, d
);
5513 ztest_dataset_destroy(int d
)
5515 char name
[MAXNAMELEN
];
5518 ztest_dataset_name(name
, ztest_opts
.zo_pool
, d
);
5520 if (ztest_opts
.zo_verbose
>= 3)
5521 (void) printf("Destroying %s to free up space\n", name
);
5524 * Cleanup any non-standard clones and snapshots. In general,
5525 * ztest thread t operates on dataset (t % zopt_datasets),
5526 * so there may be more than one thing to clean up.
5528 for (t
= d
; t
< ztest_opts
.zo_threads
;
5529 t
+= ztest_opts
.zo_datasets
)
5530 ztest_dsl_dataset_cleanup(name
, t
);
5532 (void) dmu_objset_find(name
, ztest_objset_destroy_cb
, NULL
,
5533 DS_FIND_SNAPSHOTS
| DS_FIND_CHILDREN
);
5537 ztest_dataset_dirobj_verify(ztest_ds_t
*zd
)
5539 uint64_t usedobjs
, dirobjs
, scratch
;
5542 * ZTEST_DIROBJ is the object directory for the entire dataset.
5543 * Therefore, the number of objects in use should equal the
5544 * number of ZTEST_DIROBJ entries, +1 for ZTEST_DIROBJ itself.
5545 * If not, we have an object leak.
5547 * Note that we can only check this in ztest_dataset_open(),
5548 * when the open-context and syncing-context values agree.
5549 * That's because zap_count() returns the open-context value,
5550 * while dmu_objset_space() returns the rootbp fill count.
5552 VERIFY3U(0, ==, zap_count(zd
->zd_os
, ZTEST_DIROBJ
, &dirobjs
));
5553 dmu_objset_space(zd
->zd_os
, &scratch
, &scratch
, &usedobjs
, &scratch
);
5554 ASSERT3U(dirobjs
+ 1, ==, usedobjs
);
5558 ztest_dataset_open(int d
)
5560 ztest_ds_t
*zd
= &ztest_ds
[d
];
5561 uint64_t committed_seq
= ZTEST_GET_SHARED_DS(d
)->zd_seq
;
5564 char name
[MAXNAMELEN
];
5567 ztest_dataset_name(name
, ztest_opts
.zo_pool
, d
);
5569 (void) rw_enter(&ztest_name_lock
, RW_READER
);
5571 error
= ztest_dataset_create(name
);
5572 if (error
== ENOSPC
) {
5573 (void) rw_exit(&ztest_name_lock
);
5574 ztest_record_enospc(FTAG
);
5577 ASSERT(error
== 0 || error
== EEXIST
);
5579 VERIFY0(dmu_objset_own(name
, DMU_OST_OTHER
, B_FALSE
, zd
, &os
));
5580 (void) rw_exit(&ztest_name_lock
);
5582 ztest_zd_init(zd
, ZTEST_GET_SHARED_DS(d
), os
);
5584 zilog
= zd
->zd_zilog
;
5586 if (zilog
->zl_header
->zh_claim_lr_seq
!= 0 &&
5587 zilog
->zl_header
->zh_claim_lr_seq
< committed_seq
)
5588 fatal(0, "missing log records: claimed %llu < committed %llu",
5589 zilog
->zl_header
->zh_claim_lr_seq
, committed_seq
);
5591 ztest_dataset_dirobj_verify(zd
);
5593 zil_replay(os
, zd
, ztest_replay_vector
);
5595 ztest_dataset_dirobj_verify(zd
);
5597 if (ztest_opts
.zo_verbose
>= 6)
5598 (void) printf("%s replay %llu blocks, %llu records, seq %llu\n",
5600 (u_longlong_t
)zilog
->zl_parse_blk_count
,
5601 (u_longlong_t
)zilog
->zl_parse_lr_count
,
5602 (u_longlong_t
)zilog
->zl_replaying_seq
);
5604 zilog
= zil_open(os
, ztest_get_data
);
5606 if (zilog
->zl_replaying_seq
!= 0 &&
5607 zilog
->zl_replaying_seq
< committed_seq
)
5608 fatal(0, "missing log records: replayed %llu < committed %llu",
5609 zilog
->zl_replaying_seq
, committed_seq
);
5615 ztest_dataset_close(int d
)
5617 ztest_ds_t
*zd
= &ztest_ds
[d
];
5619 zil_close(zd
->zd_zilog
);
5620 dmu_objset_disown(zd
->zd_os
, zd
);
5626 * Kick off threads to run tests on all datasets in parallel.
5629 ztest_run(ztest_shared_t
*zs
)
5634 kthread_t
*resume_thread
;
5639 ztest_exiting
= B_FALSE
;
5642 * Initialize parent/child shared state.
5644 mutex_init(&ztest_vdev_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
5645 rw_init(&ztest_name_lock
, NULL
, RW_DEFAULT
, NULL
);
5647 zs
->zs_thread_start
= gethrtime();
5648 zs
->zs_thread_stop
=
5649 zs
->zs_thread_start
+ ztest_opts
.zo_passtime
* NANOSEC
;
5650 zs
->zs_thread_stop
= MIN(zs
->zs_thread_stop
, zs
->zs_proc_stop
);
5651 zs
->zs_thread_kill
= zs
->zs_thread_stop
;
5652 if (ztest_random(100) < ztest_opts
.zo_killrate
) {
5653 zs
->zs_thread_kill
-=
5654 ztest_random(ztest_opts
.zo_passtime
* NANOSEC
);
5657 mutex_init(&zcl
.zcl_callbacks_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
5659 list_create(&zcl
.zcl_callbacks
, sizeof (ztest_cb_data_t
),
5660 offsetof(ztest_cb_data_t
, zcd_node
));
5665 kernel_init(FREAD
| FWRITE
);
5666 VERIFY0(spa_open(ztest_opts
.zo_pool
, &spa
, FTAG
));
5667 spa
->spa_debug
= B_TRUE
;
5670 VERIFY0(dmu_objset_own(ztest_opts
.zo_pool
,
5671 DMU_OST_ANY
, B_TRUE
, FTAG
, &os
));
5672 zs
->zs_guid
= dmu_objset_fsid_guid(os
);
5673 dmu_objset_disown(os
, FTAG
);
5675 spa
->spa_dedup_ditto
= 2 * ZIO_DEDUPDITTO_MIN
;
5678 * We don't expect the pool to suspend unless maxfaults == 0,
5679 * in which case ztest_fault_inject() temporarily takes away
5680 * the only valid replica.
5682 if (MAXFAULTS() == 0)
5683 spa
->spa_failmode
= ZIO_FAILURE_MODE_WAIT
;
5685 spa
->spa_failmode
= ZIO_FAILURE_MODE_PANIC
;
5688 * Create a thread to periodically resume suspended I/O.
5690 VERIFY3P((resume_thread
= zk_thread_create(NULL
, 0,
5691 (thread_func_t
)ztest_resume_thread
, spa
, TS_RUN
, NULL
, 0, 0,
5692 PTHREAD_CREATE_JOINABLE
)), !=, NULL
);
5696 * Set a deadman alarm to abort() if we hang.
5698 signal(SIGALRM
, ztest_deadman_alarm
);
5699 alarm((zs
->zs_thread_stop
- zs
->zs_thread_start
) / NANOSEC
+ GRACE
);
5703 * Verify that we can safely inquire about about any object,
5704 * whether it's allocated or not. To make it interesting,
5705 * we probe a 5-wide window around each power of two.
5706 * This hits all edge cases, including zero and the max.
5708 for (t
= 0; t
< 64; t
++) {
5709 for (d
= -5; d
<= 5; d
++) {
5710 error
= dmu_object_info(spa
->spa_meta_objset
,
5711 (1ULL << t
) + d
, NULL
);
5712 ASSERT(error
== 0 || error
== ENOENT
||
5718 * If we got any ENOSPC errors on the previous run, destroy something.
5720 if (zs
->zs_enospc_count
!= 0) {
5721 int d
= ztest_random(ztest_opts
.zo_datasets
);
5722 ztest_dataset_destroy(d
);
5724 zs
->zs_enospc_count
= 0;
5726 tid
= umem_zalloc(ztest_opts
.zo_threads
* sizeof (kt_did_t
),
5729 if (ztest_opts
.zo_verbose
>= 4)
5730 (void) printf("starting main threads...\n");
5733 * Kick off all the tests that run in parallel.
5735 for (t
= 0; t
< ztest_opts
.zo_threads
; t
++) {
5738 if (t
< ztest_opts
.zo_datasets
&&
5739 ztest_dataset_open(t
) != 0)
5742 VERIFY3P(thread
= zk_thread_create(NULL
, 0,
5743 (thread_func_t
)ztest_thread
,
5744 (void *)(uintptr_t)t
, TS_RUN
, NULL
, 0, 0,
5745 PTHREAD_CREATE_JOINABLE
), !=, NULL
);
5746 tid
[t
] = thread
->t_tid
;
5750 * Wait for all of the tests to complete. We go in reverse order
5751 * so we don't close datasets while threads are still using them.
5753 for (t
= ztest_opts
.zo_threads
- 1; t
>= 0; t
--) {
5754 thread_join(tid
[t
]);
5755 if (t
< ztest_opts
.zo_datasets
)
5756 ztest_dataset_close(t
);
5759 txg_wait_synced(spa_get_dsl(spa
), 0);
5761 zs
->zs_alloc
= metaslab_class_get_alloc(spa_normal_class(spa
));
5762 zs
->zs_space
= metaslab_class_get_space(spa_normal_class(spa
));
5764 umem_free(tid
, ztest_opts
.zo_threads
* sizeof (kt_did_t
));
5766 /* Kill the resume thread */
5767 ztest_exiting
= B_TRUE
;
5768 thread_join(resume_thread
->t_tid
);
5772 * Right before closing the pool, kick off a bunch of async I/O;
5773 * spa_close() should wait for it to complete.
5775 for (object
= 1; object
< 50; object
++)
5776 dmu_prefetch(spa
->spa_meta_objset
, object
, 0, 1ULL << 20);
5778 /* Verify that at least one commit cb was called in a timely fashion */
5779 if (zc_cb_counter
>= ZTEST_COMMIT_CB_MIN_REG
)
5780 VERIFY0(zc_min_txg_delay
);
5782 spa_close(spa
, FTAG
);
5785 * Verify that we can loop over all pools.
5787 mutex_enter(&spa_namespace_lock
);
5788 for (spa
= spa_next(NULL
); spa
!= NULL
; spa
= spa_next(spa
))
5789 if (ztest_opts
.zo_verbose
> 3)
5790 (void) printf("spa_next: found %s\n", spa_name(spa
));
5791 mutex_exit(&spa_namespace_lock
);
5794 * Verify that we can export the pool and reimport it under a
5797 if (ztest_random(2) == 0) {
5798 char name
[MAXNAMELEN
];
5799 (void) snprintf(name
, MAXNAMELEN
, "%s_import",
5800 ztest_opts
.zo_pool
);
5801 ztest_spa_import_export(ztest_opts
.zo_pool
, name
);
5802 ztest_spa_import_export(name
, ztest_opts
.zo_pool
);
5807 list_destroy(&zcl
.zcl_callbacks
);
5808 mutex_destroy(&zcl
.zcl_callbacks_lock
);
5809 rw_destroy(&ztest_name_lock
);
5810 mutex_destroy(&ztest_vdev_lock
);
5816 ztest_ds_t
*zd
= &ztest_ds
[0];
5820 if (ztest_opts
.zo_verbose
>= 3)
5821 (void) printf("testing spa_freeze()...\n");
5823 kernel_init(FREAD
| FWRITE
);
5824 VERIFY3U(0, ==, spa_open(ztest_opts
.zo_pool
, &spa
, FTAG
));
5825 VERIFY3U(0, ==, ztest_dataset_open(0));
5828 * Force the first log block to be transactionally allocated.
5829 * We have to do this before we freeze the pool -- otherwise
5830 * the log chain won't be anchored.
5832 while (BP_IS_HOLE(&zd
->zd_zilog
->zl_header
->zh_log
)) {
5833 ztest_dmu_object_alloc_free(zd
, 0);
5834 zil_commit(zd
->zd_zilog
, 0);
5837 txg_wait_synced(spa_get_dsl(spa
), 0);
5840 * Freeze the pool. This stops spa_sync() from doing anything,
5841 * so that the only way to record changes from now on is the ZIL.
5846 * Run tests that generate log records but don't alter the pool config
5847 * or depend on DSL sync tasks (snapshots, objset create/destroy, etc).
5848 * We do a txg_wait_synced() after each iteration to force the txg
5849 * to increase well beyond the last synced value in the uberblock.
5850 * The ZIL should be OK with that.
5852 while (ztest_random(10) != 0 &&
5853 numloops
++ < ztest_opts
.zo_maxloops
) {
5854 ztest_dmu_write_parallel(zd
, 0);
5855 ztest_dmu_object_alloc_free(zd
, 0);
5856 txg_wait_synced(spa_get_dsl(spa
), 0);
5860 * Commit all of the changes we just generated.
5862 zil_commit(zd
->zd_zilog
, 0);
5863 txg_wait_synced(spa_get_dsl(spa
), 0);
5866 * Close our dataset and close the pool.
5868 ztest_dataset_close(0);
5869 spa_close(spa
, FTAG
);
5873 * Open and close the pool and dataset to induce log replay.
5875 kernel_init(FREAD
| FWRITE
);
5876 VERIFY3U(0, ==, spa_open(ztest_opts
.zo_pool
, &spa
, FTAG
));
5877 ASSERT(spa_freeze_txg(spa
) == UINT64_MAX
);
5878 VERIFY3U(0, ==, ztest_dataset_open(0));
5879 ztest_dataset_close(0);
5881 spa
->spa_debug
= B_TRUE
;
5883 txg_wait_synced(spa_get_dsl(spa
), 0);
5884 ztest_reguid(NULL
, 0);
5886 spa_close(spa
, FTAG
);
5891 print_time(hrtime_t t
, char *timebuf
)
5893 hrtime_t s
= t
/ NANOSEC
;
5894 hrtime_t m
= s
/ 60;
5895 hrtime_t h
= m
/ 60;
5896 hrtime_t d
= h
/ 24;
5905 (void) sprintf(timebuf
,
5906 "%llud%02lluh%02llum%02llus", d
, h
, m
, s
);
5908 (void) sprintf(timebuf
, "%lluh%02llum%02llus", h
, m
, s
);
5910 (void) sprintf(timebuf
, "%llum%02llus", m
, s
);
5912 (void) sprintf(timebuf
, "%llus", s
);
5916 make_random_props(void)
5920 VERIFY(nvlist_alloc(&props
, NV_UNIQUE_NAME
, 0) == 0);
5921 if (ztest_random(2) == 0)
5923 VERIFY(nvlist_add_uint64(props
, "autoreplace", 1) == 0);
5929 * Create a storage pool with the given name and initial vdev size.
5930 * Then test spa_freeze() functionality.
5933 ztest_init(ztest_shared_t
*zs
)
5936 nvlist_t
*nvroot
, *props
;
5939 mutex_init(&ztest_vdev_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
5940 rw_init(&ztest_name_lock
, NULL
, RW_DEFAULT
, NULL
);
5942 kernel_init(FREAD
| FWRITE
);
5945 * Create the storage pool.
5947 (void) spa_destroy(ztest_opts
.zo_pool
);
5948 ztest_shared
->zs_vdev_next_leaf
= 0;
5950 zs
->zs_mirrors
= ztest_opts
.zo_mirrors
;
5951 nvroot
= make_vdev_root(NULL
, NULL
, NULL
, ztest_opts
.zo_vdev_size
, 0,
5952 0, ztest_opts
.zo_raidz
, zs
->zs_mirrors
, 1);
5953 props
= make_random_props();
5954 for (i
= 0; i
< SPA_FEATURES
; i
++) {
5956 VERIFY3S(-1, !=, asprintf(&buf
, "feature@%s",
5957 spa_feature_table
[i
].fi_uname
));
5958 VERIFY3U(0, ==, nvlist_add_uint64(props
, buf
, 0));
5961 VERIFY3U(0, ==, spa_create(ztest_opts
.zo_pool
, nvroot
, props
, NULL
));
5962 nvlist_free(nvroot
);
5964 VERIFY3U(0, ==, spa_open(ztest_opts
.zo_pool
, &spa
, FTAG
));
5965 zs
->zs_metaslab_sz
=
5966 1ULL << spa
->spa_root_vdev
->vdev_child
[0]->vdev_ms_shift
;
5967 spa_close(spa
, FTAG
);
5971 ztest_run_zdb(ztest_opts
.zo_pool
);
5975 ztest_run_zdb(ztest_opts
.zo_pool
);
5977 rw_destroy(&ztest_name_lock
);
5978 mutex_destroy(&ztest_vdev_lock
);
5984 static char ztest_name_data
[] = "/tmp/ztest.data.XXXXXX";
5986 ztest_fd_data
= mkstemp(ztest_name_data
);
5987 ASSERT3S(ztest_fd_data
, >=, 0);
5988 (void) unlink(ztest_name_data
);
5992 shared_data_size(ztest_shared_hdr_t
*hdr
)
5996 size
= hdr
->zh_hdr_size
;
5997 size
+= hdr
->zh_opts_size
;
5998 size
+= hdr
->zh_size
;
5999 size
+= hdr
->zh_stats_size
* hdr
->zh_stats_count
;
6000 size
+= hdr
->zh_ds_size
* hdr
->zh_ds_count
;
6009 ztest_shared_hdr_t
*hdr
;
6011 hdr
= (void *)mmap(0, P2ROUNDUP(sizeof (*hdr
), getpagesize()),
6012 PROT_READ
| PROT_WRITE
, MAP_SHARED
, ztest_fd_data
, 0);
6013 ASSERT(hdr
!= MAP_FAILED
);
6015 VERIFY3U(0, ==, ftruncate(ztest_fd_data
, sizeof (ztest_shared_hdr_t
)));
6017 hdr
->zh_hdr_size
= sizeof (ztest_shared_hdr_t
);
6018 hdr
->zh_opts_size
= sizeof (ztest_shared_opts_t
);
6019 hdr
->zh_size
= sizeof (ztest_shared_t
);
6020 hdr
->zh_stats_size
= sizeof (ztest_shared_callstate_t
);
6021 hdr
->zh_stats_count
= ZTEST_FUNCS
;
6022 hdr
->zh_ds_size
= sizeof (ztest_shared_ds_t
);
6023 hdr
->zh_ds_count
= ztest_opts
.zo_datasets
;
6025 size
= shared_data_size(hdr
);
6026 VERIFY3U(0, ==, ftruncate(ztest_fd_data
, size
));
6028 (void) munmap((caddr_t
)hdr
, P2ROUNDUP(sizeof (*hdr
), getpagesize()));
6035 ztest_shared_hdr_t
*hdr
;
6038 hdr
= (void *)mmap(0, P2ROUNDUP(sizeof (*hdr
), getpagesize()),
6039 PROT_READ
, MAP_SHARED
, ztest_fd_data
, 0);
6040 ASSERT(hdr
!= MAP_FAILED
);
6042 size
= shared_data_size(hdr
);
6044 (void) munmap((caddr_t
)hdr
, P2ROUNDUP(sizeof (*hdr
), getpagesize()));
6045 hdr
= ztest_shared_hdr
= (void *)mmap(0, P2ROUNDUP(size
, getpagesize()),
6046 PROT_READ
| PROT_WRITE
, MAP_SHARED
, ztest_fd_data
, 0);
6047 ASSERT(hdr
!= MAP_FAILED
);
6048 buf
= (uint8_t *)hdr
;
6050 offset
= hdr
->zh_hdr_size
;
6051 ztest_shared_opts
= (void *)&buf
[offset
];
6052 offset
+= hdr
->zh_opts_size
;
6053 ztest_shared
= (void *)&buf
[offset
];
6054 offset
+= hdr
->zh_size
;
6055 ztest_shared_callstate
= (void *)&buf
[offset
];
6056 offset
+= hdr
->zh_stats_size
* hdr
->zh_stats_count
;
6057 ztest_shared_ds
= (void *)&buf
[offset
];
6061 exec_child(char *cmd
, char *libpath
, boolean_t ignorekill
, int *statusp
)
6065 char *cmdbuf
= NULL
;
6070 cmdbuf
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
6071 (void) strlcpy(cmdbuf
, getexecname(), MAXPATHLEN
);
6076 fatal(1, "fork failed");
6078 if (pid
== 0) { /* child */
6079 char *emptyargv
[2] = { cmd
, NULL
};
6080 char fd_data_str
[12];
6082 struct rlimit rl
= { 1024, 1024 };
6083 (void) setrlimit(RLIMIT_NOFILE
, &rl
);
6085 (void) close(ztest_fd_rand
);
6086 VERIFY(11 >= snprintf(fd_data_str
, 12, "%d", ztest_fd_data
));
6087 VERIFY(0 == setenv("ZTEST_FD_DATA", fd_data_str
, 1));
6089 (void) enable_extended_FILE_stdio(-1, -1);
6090 if (libpath
!= NULL
)
6091 VERIFY(0 == setenv("LD_LIBRARY_PATH", libpath
, 1));
6092 (void) execv(cmd
, emptyargv
);
6093 ztest_dump_core
= B_FALSE
;
6094 fatal(B_TRUE
, "exec failed: %s", cmd
);
6097 if (cmdbuf
!= NULL
) {
6098 umem_free(cmdbuf
, MAXPATHLEN
);
6102 while (waitpid(pid
, &status
, 0) != pid
)
6104 if (statusp
!= NULL
)
6107 if (WIFEXITED(status
)) {
6108 if (WEXITSTATUS(status
) != 0) {
6109 (void) fprintf(stderr
, "child exited with code %d\n",
6110 WEXITSTATUS(status
));
6114 } else if (WIFSIGNALED(status
)) {
6115 if (!ignorekill
|| WTERMSIG(status
) != SIGKILL
) {
6116 (void) fprintf(stderr
, "child died with signal %d\n",
6122 (void) fprintf(stderr
, "something strange happened to child\n");
6129 ztest_run_init(void)
6133 ztest_shared_t
*zs
= ztest_shared
;
6135 ASSERT(ztest_opts
.zo_init
!= 0);
6138 * Blow away any existing copy of zpool.cache
6140 (void) remove(spa_config_path
);
6143 * Create and initialize our storage pool.
6145 for (i
= 1; i
<= ztest_opts
.zo_init
; i
++) {
6146 bzero(zs
, sizeof (ztest_shared_t
));
6147 if (ztest_opts
.zo_verbose
>= 3 &&
6148 ztest_opts
.zo_init
!= 1) {
6149 (void) printf("ztest_init(), pass %d\n", i
);
6156 main(int argc
, char **argv
)
6164 ztest_shared_callstate_t
*zc
;
6171 char *fd_data_str
= getenv("ZTEST_FD_DATA");
6173 (void) setvbuf(stdout
, NULL
, _IOLBF
, 0);
6175 ztest_fd_rand
= open("/dev/urandom", O_RDONLY
);
6176 ASSERT3S(ztest_fd_rand
, >=, 0);
6179 dprintf_setup(&argc
, argv
);
6180 process_options(argc
, argv
);
6185 bcopy(&ztest_opts
, ztest_shared_opts
,
6186 sizeof (*ztest_shared_opts
));
6188 ztest_fd_data
= atoi(fd_data_str
);
6190 bcopy(ztest_shared_opts
, &ztest_opts
, sizeof (ztest_opts
));
6192 ASSERT3U(ztest_opts
.zo_datasets
, ==, ztest_shared_hdr
->zh_ds_count
);
6194 /* Override location of zpool.cache */
6195 VERIFY(asprintf((char **)&spa_config_path
, "%s/zpool.cache",
6196 ztest_opts
.zo_dir
) != -1);
6198 ztest_ds
= umem_alloc(ztest_opts
.zo_datasets
* sizeof (ztest_ds_t
),
6203 metaslab_gang_bang
= ztest_opts
.zo_metaslab_gang_bang
;
6204 metaslab_df_alloc_threshold
=
6205 zs
->zs_metaslab_df_alloc_threshold
;
6214 hasalt
= (strlen(ztest_opts
.zo_alt_ztest
) != 0);
6216 if (ztest_opts
.zo_verbose
>= 1) {
6217 (void) printf("%llu vdevs, %d datasets, %d threads,"
6218 " %llu seconds...\n",
6219 (u_longlong_t
)ztest_opts
.zo_vdevs
,
6220 ztest_opts
.zo_datasets
,
6221 ztest_opts
.zo_threads
,
6222 (u_longlong_t
)ztest_opts
.zo_time
);
6225 cmd
= umem_alloc(MAXNAMELEN
, UMEM_NOFAIL
);
6226 (void) strlcpy(cmd
, getexecname(), MAXNAMELEN
);
6228 zs
->zs_do_init
= B_TRUE
;
6229 if (strlen(ztest_opts
.zo_alt_ztest
) != 0) {
6230 if (ztest_opts
.zo_verbose
>= 1) {
6231 (void) printf("Executing older ztest for "
6232 "initialization: %s\n", ztest_opts
.zo_alt_ztest
);
6234 VERIFY(!exec_child(ztest_opts
.zo_alt_ztest
,
6235 ztest_opts
.zo_alt_libpath
, B_FALSE
, NULL
));
6237 VERIFY(!exec_child(NULL
, NULL
, B_FALSE
, NULL
));
6239 zs
->zs_do_init
= B_FALSE
;
6241 zs
->zs_proc_start
= gethrtime();
6242 zs
->zs_proc_stop
= zs
->zs_proc_start
+ ztest_opts
.zo_time
* NANOSEC
;
6244 for (f
= 0; f
< ZTEST_FUNCS
; f
++) {
6245 zi
= &ztest_info
[f
];
6246 zc
= ZTEST_GET_SHARED_CALLSTATE(f
);
6247 if (zs
->zs_proc_start
+ zi
->zi_interval
[0] > zs
->zs_proc_stop
)
6248 zc
->zc_next
= UINT64_MAX
;
6250 zc
->zc_next
= zs
->zs_proc_start
+
6251 ztest_random(2 * zi
->zi_interval
[0] + 1);
6255 * Run the tests in a loop. These tests include fault injection
6256 * to verify that self-healing data works, and forced crashes
6257 * to verify that we never lose on-disk consistency.
6259 while (gethrtime() < zs
->zs_proc_stop
) {
6264 * Initialize the workload counters for each function.
6266 for (f
= 0; f
< ZTEST_FUNCS
; f
++) {
6267 zc
= ZTEST_GET_SHARED_CALLSTATE(f
);
6272 /* Set the allocation switch size */
6273 zs
->zs_metaslab_df_alloc_threshold
=
6274 ztest_random(zs
->zs_metaslab_sz
/ 4) + 1;
6276 if (!hasalt
|| ztest_random(2) == 0) {
6277 if (hasalt
&& ztest_opts
.zo_verbose
>= 1) {
6278 (void) printf("Executing newer ztest: %s\n",
6282 killed
= exec_child(cmd
, NULL
, B_TRUE
, &status
);
6284 if (hasalt
&& ztest_opts
.zo_verbose
>= 1) {
6285 (void) printf("Executing older ztest: %s\n",
6286 ztest_opts
.zo_alt_ztest
);
6289 killed
= exec_child(ztest_opts
.zo_alt_ztest
,
6290 ztest_opts
.zo_alt_libpath
, B_TRUE
, &status
);
6297 if (ztest_opts
.zo_verbose
>= 1) {
6298 hrtime_t now
= gethrtime();
6300 now
= MIN(now
, zs
->zs_proc_stop
);
6301 print_time(zs
->zs_proc_stop
- now
, timebuf
);
6302 nicenum(zs
->zs_space
, numbuf
);
6304 (void) printf("Pass %3d, %8s, %3llu ENOSPC, "
6305 "%4.1f%% of %5s used, %3.0f%% done, %8s to go\n",
6307 WIFEXITED(status
) ? "Complete" : "SIGKILL",
6308 (u_longlong_t
)zs
->zs_enospc_count
,
6309 100.0 * zs
->zs_alloc
/ zs
->zs_space
,
6311 100.0 * (now
- zs
->zs_proc_start
) /
6312 (ztest_opts
.zo_time
* NANOSEC
), timebuf
);
6315 if (ztest_opts
.zo_verbose
>= 2) {
6316 (void) printf("\nWorkload summary:\n\n");
6317 (void) printf("%7s %9s %s\n",
6318 "Calls", "Time", "Function");
6319 (void) printf("%7s %9s %s\n",
6320 "-----", "----", "--------");
6321 for (f
= 0; f
< ZTEST_FUNCS
; f
++) {
6324 zi
= &ztest_info
[f
];
6325 zc
= ZTEST_GET_SHARED_CALLSTATE(f
);
6326 print_time(zc
->zc_time
, timebuf
);
6327 (void) dladdr((void *)zi
->zi_func
, &dli
);
6328 (void) printf("%7llu %9s %s\n",
6329 (u_longlong_t
)zc
->zc_count
, timebuf
,
6332 (void) printf("\n");
6336 * It's possible that we killed a child during a rename test,
6337 * in which case we'll have a 'ztest_tmp' pool lying around
6338 * instead of 'ztest'. Do a blind rename in case this happened.
6341 if (spa_open(ztest_opts
.zo_pool
, &spa
, FTAG
) == 0) {
6342 spa_close(spa
, FTAG
);
6344 char tmpname
[MAXNAMELEN
];
6346 kernel_init(FREAD
| FWRITE
);
6347 (void) snprintf(tmpname
, sizeof (tmpname
), "%s_tmp",
6348 ztest_opts
.zo_pool
);
6349 (void) spa_rename(tmpname
, ztest_opts
.zo_pool
);
6353 ztest_run_zdb(ztest_opts
.zo_pool
);
6356 if (ztest_opts
.zo_verbose
>= 1) {
6358 (void) printf("%d runs of older ztest: %s\n", older
,
6359 ztest_opts
.zo_alt_ztest
);
6360 (void) printf("%d runs of newer ztest: %s\n", newer
,
6363 (void) printf("%d killed, %d completed, %.0f%% kill rate\n",
6364 kills
, iters
- kills
, (100.0 * kills
) / MAX(1, iters
));
6367 umem_free(cmd
, MAXNAMELEN
);