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.
25 * Copyright (c) 2013 Steven Hartland. All rights reserved.
29 * The objective of this program is to provide a DMU/ZAP/SPA stress test
30 * that runs entirely in userland, is easy to use, and easy to extend.
32 * The overall design of the ztest program is as follows:
34 * (1) For each major functional area (e.g. adding vdevs to a pool,
35 * creating and destroying datasets, reading and writing objects, etc)
36 * we have a simple routine to test that functionality. These
37 * individual routines do not have to do anything "stressful".
39 * (2) We turn these simple functionality tests into a stress test by
40 * running them all in parallel, with as many threads as desired,
41 * and spread across as many datasets, objects, and vdevs as desired.
43 * (3) While all this is happening, we inject faults into the pool to
44 * verify that self-healing data really works.
46 * (4) Every time we open a dataset, we change its checksum and compression
47 * functions. Thus even individual objects vary from block to block
48 * in which checksum they use and whether they're compressed.
50 * (5) To verify that we never lose on-disk consistency after a crash,
51 * we run the entire test in a child of the main process.
52 * At random times, the child self-immolates with a SIGKILL.
53 * This is the software equivalent of pulling the power cord.
54 * The parent then runs the test again, using the existing
55 * storage pool, as many times as desired. If backwards compatability
56 * testing is enabled ztest will sometimes run the "older" version
57 * of ztest after a SIGKILL.
59 * (6) To verify that we don't have future leaks or temporal incursions,
60 * many of the functional tests record the transaction group number
61 * as part of their data. When reading old data, they verify that
62 * the transaction group number is less than the current, open txg.
63 * If you add a new test, please do this if applicable.
65 * (7) Threads are created with a reduced stack size, for sanity checking.
66 * Therefore, it's important not to allocate huge buffers on the stack.
68 * When run with no arguments, ztest runs for about five minutes and
69 * produces no output if successful. To get a little bit of information,
70 * specify -V. To get more information, specify -VV, and so on.
72 * To turn this into an overnight stress test, use -T to specify run time.
74 * You can ask more more vdevs [-v], datasets [-d], or threads [-t]
75 * to increase the pool capacity, fanout, and overall stress level.
77 * Use the -k option to set the desired frequency of kills.
79 * When ztest invokes itself it passes all relevant information through a
80 * temporary file which is mmap-ed in the child process. This allows shared
81 * memory to survive the exec syscall. The ztest_shared_hdr_t struct is always
82 * stored at offset 0 of this file and contains information on the size and
83 * number of shared structures in the file. The information stored in this file
84 * must remain backwards compatible with older versions of ztest so that
85 * ztest can invoke them during backwards compatibility testing (-B).
88 #include <sys/zfs_context.h>
94 #include <sys/dmu_objset.h>
100 #include <sys/resource.h>
103 #include <sys/zil_impl.h>
104 #include <sys/vdev_impl.h>
105 #include <sys/vdev_file.h>
106 #include <sys/spa_impl.h>
107 #include <sys/metaslab_impl.h>
108 #include <sys/dsl_prop.h>
109 #include <sys/dsl_dataset.h>
110 #include <sys/dsl_destroy.h>
111 #include <sys/dsl_scan.h>
112 #include <sys/zio_checksum.h>
113 #include <sys/refcount.h>
114 #include <sys/zfeature.h>
115 #include <sys/dsl_userhold.h>
117 #include <stdio_ext.h>
125 #include <sys/fs/zfs.h>
126 #include <libnvpair.h>
128 static int ztest_fd_data
= -1;
129 static int ztest_fd_rand
= -1;
131 typedef struct ztest_shared_hdr
{
132 uint64_t zh_hdr_size
;
133 uint64_t zh_opts_size
;
135 uint64_t zh_stats_size
;
136 uint64_t zh_stats_count
;
138 uint64_t zh_ds_count
;
139 } ztest_shared_hdr_t
;
141 static ztest_shared_hdr_t
*ztest_shared_hdr
;
143 typedef struct ztest_shared_opts
{
144 char zo_pool
[MAXNAMELEN
];
145 char zo_dir
[MAXNAMELEN
];
146 char zo_alt_ztest
[MAXNAMELEN
];
147 char zo_alt_libpath
[MAXNAMELEN
];
149 uint64_t zo_vdevtime
;
157 uint64_t zo_passtime
;
158 uint64_t zo_killrate
;
162 uint64_t zo_maxloops
;
163 uint64_t zo_metaslab_gang_bang
;
164 } ztest_shared_opts_t
;
166 static const ztest_shared_opts_t ztest_opts_defaults
= {
167 .zo_pool
= { 'z', 't', 'e', 's', 't', '\0' },
168 .zo_dir
= { '/', 't', 'm', 'p', '\0' },
169 .zo_alt_ztest
= { '\0' },
170 .zo_alt_libpath
= { '\0' },
172 .zo_ashift
= SPA_MINBLOCKSHIFT
,
175 .zo_raidz_parity
= 1,
176 .zo_vdev_size
= SPA_MINDEVSIZE
,
179 .zo_passtime
= 60, /* 60 seconds */
180 .zo_killrate
= 70, /* 70% kill rate */
183 .zo_time
= 300, /* 5 minutes */
184 .zo_maxloops
= 50, /* max loops during spa_freeze() */
185 .zo_metaslab_gang_bang
= 32 << 10
188 extern uint64_t metaslab_gang_bang
;
189 extern uint64_t metaslab_df_alloc_threshold
;
191 static ztest_shared_opts_t
*ztest_shared_opts
;
192 static ztest_shared_opts_t ztest_opts
;
194 typedef struct ztest_shared_ds
{
198 static ztest_shared_ds_t
*ztest_shared_ds
;
199 #define ZTEST_GET_SHARED_DS(d) (&ztest_shared_ds[d])
201 #define BT_MAGIC 0x123456789abcdefULL
202 #define MAXFAULTS() \
203 (MAX(zs->zs_mirrors, 1) * (ztest_opts.zo_raidz_parity + 1) - 1)
207 ZTEST_IO_WRITE_PATTERN
,
208 ZTEST_IO_WRITE_ZEROES
,
215 typedef struct ztest_block_tag
{
225 typedef struct bufwad
{
232 * XXX -- fix zfs range locks to be generic so we can use them here.
254 #define ZTEST_RANGE_LOCKS 64
255 #define ZTEST_OBJECT_LOCKS 64
258 * Object descriptor. Used as a template for object lookup/create/remove.
260 typedef struct ztest_od
{
263 dmu_object_type_t od_type
;
264 dmu_object_type_t od_crtype
;
265 uint64_t od_blocksize
;
266 uint64_t od_crblocksize
;
269 char od_name
[MAXNAMELEN
];
275 typedef struct ztest_ds
{
276 ztest_shared_ds_t
*zd_shared
;
278 krwlock_t zd_zilog_lock
;
280 ztest_od_t
*zd_od
; /* debugging aid */
281 char zd_name
[MAXNAMELEN
];
282 kmutex_t zd_dirobj_lock
;
283 rll_t zd_object_lock
[ZTEST_OBJECT_LOCKS
];
284 rll_t zd_range_lock
[ZTEST_RANGE_LOCKS
];
288 * Per-iteration state.
290 typedef void ztest_func_t(ztest_ds_t
*zd
, uint64_t id
);
292 typedef struct ztest_info
{
293 ztest_func_t
*zi_func
; /* test function */
294 uint64_t zi_iters
; /* iterations per execution */
295 uint64_t *zi_interval
; /* execute every <interval> seconds */
298 typedef struct ztest_shared_callstate
{
299 uint64_t zc_count
; /* per-pass count */
300 uint64_t zc_time
; /* per-pass time */
301 uint64_t zc_next
; /* next time to call this function */
302 } ztest_shared_callstate_t
;
304 static ztest_shared_callstate_t
*ztest_shared_callstate
;
305 #define ZTEST_GET_SHARED_CALLSTATE(c) (&ztest_shared_callstate[c])
308 * Note: these aren't static because we want dladdr() to work.
310 ztest_func_t ztest_dmu_read_write
;
311 ztest_func_t ztest_dmu_write_parallel
;
312 ztest_func_t ztest_dmu_object_alloc_free
;
313 ztest_func_t ztest_dmu_commit_callbacks
;
314 ztest_func_t ztest_zap
;
315 ztest_func_t ztest_zap_parallel
;
316 ztest_func_t ztest_zil_commit
;
317 ztest_func_t ztest_zil_remount
;
318 ztest_func_t ztest_dmu_read_write_zcopy
;
319 ztest_func_t ztest_dmu_objset_create_destroy
;
320 ztest_func_t ztest_dmu_prealloc
;
321 ztest_func_t ztest_fzap
;
322 ztest_func_t ztest_dmu_snapshot_create_destroy
;
323 ztest_func_t ztest_dsl_prop_get_set
;
324 ztest_func_t ztest_spa_prop_get_set
;
325 ztest_func_t ztest_spa_create_destroy
;
326 ztest_func_t ztest_fault_inject
;
327 ztest_func_t ztest_ddt_repair
;
328 ztest_func_t ztest_dmu_snapshot_hold
;
329 ztest_func_t ztest_spa_rename
;
330 ztest_func_t ztest_scrub
;
331 ztest_func_t ztest_dsl_dataset_promote_busy
;
332 ztest_func_t ztest_vdev_attach_detach
;
333 ztest_func_t ztest_vdev_LUN_growth
;
334 ztest_func_t ztest_vdev_add_remove
;
335 ztest_func_t ztest_vdev_aux_add_remove
;
336 ztest_func_t ztest_split_pool
;
337 ztest_func_t ztest_reguid
;
338 ztest_func_t ztest_spa_upgrade
;
340 uint64_t zopt_always
= 0ULL * NANOSEC
; /* all the time */
341 uint64_t zopt_incessant
= 1ULL * NANOSEC
/ 10; /* every 1/10 second */
342 uint64_t zopt_often
= 1ULL * NANOSEC
; /* every second */
343 uint64_t zopt_sometimes
= 10ULL * NANOSEC
; /* every 10 seconds */
344 uint64_t zopt_rarely
= 60ULL * NANOSEC
; /* every 60 seconds */
346 ztest_info_t ztest_info
[] = {
347 { ztest_dmu_read_write
, 1, &zopt_always
},
348 { ztest_dmu_write_parallel
, 10, &zopt_always
},
349 { ztest_dmu_object_alloc_free
, 1, &zopt_always
},
350 { ztest_dmu_commit_callbacks
, 1, &zopt_always
},
351 { ztest_zap
, 30, &zopt_always
},
352 { ztest_zap_parallel
, 100, &zopt_always
},
353 { ztest_split_pool
, 1, &zopt_always
},
354 { ztest_zil_commit
, 1, &zopt_incessant
},
355 { ztest_zil_remount
, 1, &zopt_sometimes
},
356 { ztest_dmu_read_write_zcopy
, 1, &zopt_often
},
357 { ztest_dmu_objset_create_destroy
, 1, &zopt_often
},
358 { ztest_dsl_prop_get_set
, 1, &zopt_often
},
359 { ztest_spa_prop_get_set
, 1, &zopt_sometimes
},
361 { ztest_dmu_prealloc
, 1, &zopt_sometimes
},
363 { ztest_fzap
, 1, &zopt_sometimes
},
364 { ztest_dmu_snapshot_create_destroy
, 1, &zopt_sometimes
},
365 { ztest_spa_create_destroy
, 1, &zopt_sometimes
},
366 { ztest_fault_inject
, 1, &zopt_sometimes
},
367 { ztest_ddt_repair
, 1, &zopt_sometimes
},
368 { ztest_dmu_snapshot_hold
, 1, &zopt_sometimes
},
369 { ztest_reguid
, 1, &zopt_sometimes
},
370 { ztest_spa_rename
, 1, &zopt_rarely
},
371 { ztest_scrub
, 1, &zopt_rarely
},
372 { ztest_spa_upgrade
, 1, &zopt_rarely
},
373 { ztest_dsl_dataset_promote_busy
, 1, &zopt_rarely
},
374 { ztest_vdev_attach_detach
, 1, &zopt_sometimes
},
375 { ztest_vdev_LUN_growth
, 1, &zopt_rarely
},
376 { ztest_vdev_add_remove
, 1,
377 &ztest_opts
.zo_vdevtime
},
378 { ztest_vdev_aux_add_remove
, 1,
379 &ztest_opts
.zo_vdevtime
},
382 #define ZTEST_FUNCS (sizeof (ztest_info) / sizeof (ztest_info_t))
385 * The following struct is used to hold a list of uncalled commit callbacks.
386 * The callbacks are ordered by txg number.
388 typedef struct ztest_cb_list
{
389 kmutex_t zcl_callbacks_lock
;
390 list_t zcl_callbacks
;
394 * Stuff we need to share writably between parent and child.
396 typedef struct ztest_shared
{
397 boolean_t zs_do_init
;
398 hrtime_t zs_proc_start
;
399 hrtime_t zs_proc_stop
;
400 hrtime_t zs_thread_start
;
401 hrtime_t zs_thread_stop
;
402 hrtime_t zs_thread_kill
;
403 uint64_t zs_enospc_count
;
404 uint64_t zs_vdev_next_leaf
;
405 uint64_t zs_vdev_aux
;
410 uint64_t zs_metaslab_sz
;
411 uint64_t zs_metaslab_df_alloc_threshold
;
415 #define ID_PARALLEL -1ULL
417 static char ztest_dev_template
[] = "%s/%s.%llua";
418 static char ztest_aux_template
[] = "%s/%s.%s.%llu";
419 ztest_shared_t
*ztest_shared
;
421 static spa_t
*ztest_spa
= NULL
;
422 static ztest_ds_t
*ztest_ds
;
424 static kmutex_t ztest_vdev_lock
;
427 * The ztest_name_lock protects the pool and dataset namespace used by
428 * the individual tests. To modify the namespace, consumers must grab
429 * this lock as writer. Grabbing the lock as reader will ensure that the
430 * namespace does not change while the lock is held.
432 static krwlock_t ztest_name_lock
;
434 static boolean_t ztest_dump_core
= B_TRUE
;
435 static boolean_t ztest_exiting
;
437 /* Global commit callback list */
438 static ztest_cb_list_t zcl
;
439 /* Commit cb delay */
440 static uint64_t zc_min_txg_delay
= UINT64_MAX
;
441 static int zc_cb_counter
= 0;
444 * Minimum number of commit callbacks that need to be registered for us to check
445 * whether the minimum txg delay is acceptable.
447 #define ZTEST_COMMIT_CB_MIN_REG 100
450 * If a number of txgs equal to this threshold have been created after a commit
451 * callback has been registered but not called, then we assume there is an
452 * implementation bug.
454 #define ZTEST_COMMIT_CB_THRESH (TXG_CONCURRENT_STATES + 1000)
456 extern uint64_t metaslab_gang_bang
;
457 extern uint64_t metaslab_df_alloc_threshold
;
460 ZTEST_META_DNODE
= 0,
465 static void usage(boolean_t
) __NORETURN
;
468 * These libumem hooks provide a reasonable set of defaults for the allocator's
469 * debugging facilities.
472 _umem_debug_init(void)
474 return ("default,verbose"); /* $UMEM_DEBUG setting */
478 _umem_logging_init(void)
480 return ("fail,contents"); /* $UMEM_LOGGING setting */
483 #define FATAL_MSG_SZ 1024
488 fatal(int do_perror
, char *message
, ...)
491 int save_errno
= errno
;
494 (void) fflush(stdout
);
495 buf
= umem_alloc(FATAL_MSG_SZ
, UMEM_NOFAIL
);
497 va_start(args
, message
);
498 (void) sprintf(buf
, "ztest: ");
500 (void) vsprintf(buf
+ strlen(buf
), message
, args
);
503 (void) snprintf(buf
+ strlen(buf
), FATAL_MSG_SZ
- strlen(buf
),
504 ": %s", strerror(save_errno
));
506 (void) fprintf(stderr
, "%s\n", buf
);
507 fatal_msg
= buf
; /* to ease debugging */
514 str2shift(const char *buf
)
516 const char *ends
= "BKMGTPEZ";
521 for (i
= 0; i
< strlen(ends
); i
++) {
522 if (toupper(buf
[0]) == ends
[i
])
525 if (i
== strlen(ends
)) {
526 (void) fprintf(stderr
, "ztest: invalid bytes suffix: %s\n",
530 if (buf
[1] == '\0' || (toupper(buf
[1]) == 'B' && buf
[2] == '\0')) {
533 (void) fprintf(stderr
, "ztest: invalid bytes suffix: %s\n", buf
);
539 nicenumtoull(const char *buf
)
544 val
= strtoull(buf
, &end
, 0);
546 (void) fprintf(stderr
, "ztest: bad numeric value: %s\n", buf
);
548 } else if (end
[0] == '.') {
549 double fval
= strtod(buf
, &end
);
550 fval
*= pow(2, str2shift(end
));
551 if (fval
> UINT64_MAX
) {
552 (void) fprintf(stderr
, "ztest: value too large: %s\n",
556 val
= (uint64_t)fval
;
558 int shift
= str2shift(end
);
559 if (shift
>= 64 || (val
<< shift
) >> shift
!= val
) {
560 (void) fprintf(stderr
, "ztest: value too large: %s\n",
570 usage(boolean_t requested
)
572 const ztest_shared_opts_t
*zo
= &ztest_opts_defaults
;
574 char nice_vdev_size
[10];
575 char nice_gang_bang
[10];
576 FILE *fp
= requested
? stdout
: stderr
;
578 nicenum(zo
->zo_vdev_size
, nice_vdev_size
);
579 nicenum(zo
->zo_metaslab_gang_bang
, nice_gang_bang
);
581 (void) fprintf(fp
, "Usage: %s\n"
582 "\t[-v vdevs (default: %llu)]\n"
583 "\t[-s size_of_each_vdev (default: %s)]\n"
584 "\t[-a alignment_shift (default: %d)] use 0 for random\n"
585 "\t[-m mirror_copies (default: %d)]\n"
586 "\t[-r raidz_disks (default: %d)]\n"
587 "\t[-R raidz_parity (default: %d)]\n"
588 "\t[-d datasets (default: %d)]\n"
589 "\t[-t threads (default: %d)]\n"
590 "\t[-g gang_block_threshold (default: %s)]\n"
591 "\t[-i init_count (default: %d)] initialize pool i times\n"
592 "\t[-k kill_percentage (default: %llu%%)]\n"
593 "\t[-p pool_name (default: %s)]\n"
594 "\t[-f dir (default: %s)] file directory for vdev files\n"
595 "\t[-V] verbose (use multiple times for ever more blather)\n"
596 "\t[-E] use existing pool instead of creating new one\n"
597 "\t[-T time (default: %llu sec)] total run time\n"
598 "\t[-F freezeloops (default: %llu)] max loops in spa_freeze()\n"
599 "\t[-P passtime (default: %llu sec)] time per pass\n"
600 "\t[-B alt_ztest (default: <none>)] alternate ztest path\n"
601 "\t[-h] (print help)\n"
604 (u_longlong_t
)zo
->zo_vdevs
, /* -v */
605 nice_vdev_size
, /* -s */
606 zo
->zo_ashift
, /* -a */
607 zo
->zo_mirrors
, /* -m */
608 zo
->zo_raidz
, /* -r */
609 zo
->zo_raidz_parity
, /* -R */
610 zo
->zo_datasets
, /* -d */
611 zo
->zo_threads
, /* -t */
612 nice_gang_bang
, /* -g */
613 zo
->zo_init
, /* -i */
614 (u_longlong_t
)zo
->zo_killrate
, /* -k */
615 zo
->zo_pool
, /* -p */
617 (u_longlong_t
)zo
->zo_time
, /* -T */
618 (u_longlong_t
)zo
->zo_maxloops
, /* -F */
619 (u_longlong_t
)zo
->zo_passtime
);
620 exit(requested
? 0 : 1);
624 process_options(int argc
, char **argv
)
627 ztest_shared_opts_t
*zo
= &ztest_opts
;
631 char altdir
[MAXNAMELEN
] = { 0 };
633 bcopy(&ztest_opts_defaults
, zo
, sizeof (*zo
));
635 while ((opt
= getopt(argc
, argv
,
636 "v:s:a:m:r:R:d:t:g:i:k:p:f:VET:P:hF:B:")) != EOF
) {
653 value
= nicenumtoull(optarg
);
657 zo
->zo_vdevs
= value
;
660 zo
->zo_vdev_size
= MAX(SPA_MINDEVSIZE
, value
);
663 zo
->zo_ashift
= value
;
666 zo
->zo_mirrors
= value
;
669 zo
->zo_raidz
= MAX(1, value
);
672 zo
->zo_raidz_parity
= MIN(MAX(value
, 1), 3);
675 zo
->zo_datasets
= MAX(1, value
);
678 zo
->zo_threads
= MAX(1, value
);
681 zo
->zo_metaslab_gang_bang
= MAX(SPA_MINBLOCKSIZE
<< 1,
688 zo
->zo_killrate
= value
;
691 (void) strlcpy(zo
->zo_pool
, optarg
,
692 sizeof (zo
->zo_pool
));
695 path
= realpath(optarg
, NULL
);
697 (void) fprintf(stderr
, "error: %s: %s\n",
698 optarg
, strerror(errno
));
701 (void) strlcpy(zo
->zo_dir
, path
,
702 sizeof (zo
->zo_dir
));
715 zo
->zo_passtime
= MAX(1, value
);
718 zo
->zo_maxloops
= MAX(1, value
);
721 (void) strlcpy(altdir
, optarg
, sizeof (altdir
));
733 zo
->zo_raidz_parity
= MIN(zo
->zo_raidz_parity
, zo
->zo_raidz
- 1);
736 (zo
->zo_vdevs
> 0 ? zo
->zo_time
* NANOSEC
/ zo
->zo_vdevs
:
739 if (strlen(altdir
) > 0) {
747 cmd
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
748 realaltdir
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
750 VERIFY(NULL
!= realpath(getexecname(), cmd
));
751 if (0 != access(altdir
, F_OK
)) {
752 ztest_dump_core
= B_FALSE
;
753 fatal(B_TRUE
, "invalid alternate ztest path: %s",
756 VERIFY(NULL
!= realpath(altdir
, realaltdir
));
759 * 'cmd' should be of the form "<anything>/usr/bin/<isa>/ztest".
760 * We want to extract <isa> to determine if we should use
761 * 32 or 64 bit binaries.
763 bin
= strstr(cmd
, "/usr/bin/");
764 ztest
= strstr(bin
, "/ztest");
766 isalen
= ztest
- isa
;
767 (void) snprintf(zo
->zo_alt_ztest
, sizeof (zo
->zo_alt_ztest
),
768 "%s/usr/bin/%.*s/ztest", realaltdir
, isalen
, isa
);
769 (void) snprintf(zo
->zo_alt_libpath
, sizeof (zo
->zo_alt_libpath
),
770 "%s/usr/lib/%.*s", realaltdir
, isalen
, isa
);
772 if (0 != access(zo
->zo_alt_ztest
, X_OK
)) {
773 ztest_dump_core
= B_FALSE
;
774 fatal(B_TRUE
, "invalid alternate ztest: %s",
776 } else if (0 != access(zo
->zo_alt_libpath
, X_OK
)) {
777 ztest_dump_core
= B_FALSE
;
778 fatal(B_TRUE
, "invalid alternate lib directory %s",
782 umem_free(cmd
, MAXPATHLEN
);
783 umem_free(realaltdir
, MAXPATHLEN
);
788 ztest_kill(ztest_shared_t
*zs
)
790 zs
->zs_alloc
= metaslab_class_get_alloc(spa_normal_class(ztest_spa
));
791 zs
->zs_space
= metaslab_class_get_space(spa_normal_class(ztest_spa
));
792 (void) kill(getpid(), SIGKILL
);
796 ztest_random(uint64_t range
)
800 ASSERT3S(ztest_fd_rand
, >=, 0);
805 if (read(ztest_fd_rand
, &r
, sizeof (r
)) != sizeof (r
))
806 fatal(1, "short read from /dev/urandom");
813 ztest_record_enospc(const char *s
)
815 ztest_shared
->zs_enospc_count
++;
819 ztest_get_ashift(void)
821 if (ztest_opts
.zo_ashift
== 0)
822 return (SPA_MINBLOCKSHIFT
+ ztest_random(3));
823 return (ztest_opts
.zo_ashift
);
827 make_vdev_file(char *path
, char *aux
, char *pool
, size_t size
, uint64_t ashift
)
833 pathbuf
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
836 ashift
= ztest_get_ashift();
842 vdev
= ztest_shared
->zs_vdev_aux
;
843 (void) snprintf(path
, MAXPATHLEN
,
844 ztest_aux_template
, ztest_opts
.zo_dir
,
845 pool
== NULL
? ztest_opts
.zo_pool
: pool
,
848 vdev
= ztest_shared
->zs_vdev_next_leaf
++;
849 (void) snprintf(path
, MAXPATHLEN
,
850 ztest_dev_template
, ztest_opts
.zo_dir
,
851 pool
== NULL
? ztest_opts
.zo_pool
: pool
, vdev
);
856 int fd
= open(path
, O_RDWR
| O_CREAT
| O_TRUNC
, 0666);
858 fatal(1, "can't open %s", path
);
859 if (ftruncate(fd
, size
) != 0)
860 fatal(1, "can't ftruncate %s", path
);
864 VERIFY(nvlist_alloc(&file
, NV_UNIQUE_NAME
, 0) == 0);
865 VERIFY(nvlist_add_string(file
, ZPOOL_CONFIG_TYPE
, VDEV_TYPE_FILE
) == 0);
866 VERIFY(nvlist_add_string(file
, ZPOOL_CONFIG_PATH
, path
) == 0);
867 VERIFY(nvlist_add_uint64(file
, ZPOOL_CONFIG_ASHIFT
, ashift
) == 0);
868 umem_free(pathbuf
, MAXPATHLEN
);
874 make_vdev_raidz(char *path
, char *aux
, char *pool
, size_t size
,
875 uint64_t ashift
, int r
)
877 nvlist_t
*raidz
, **child
;
881 return (make_vdev_file(path
, aux
, pool
, size
, ashift
));
882 child
= umem_alloc(r
* sizeof (nvlist_t
*), UMEM_NOFAIL
);
884 for (c
= 0; c
< r
; c
++)
885 child
[c
] = make_vdev_file(path
, aux
, pool
, size
, ashift
);
887 VERIFY(nvlist_alloc(&raidz
, NV_UNIQUE_NAME
, 0) == 0);
888 VERIFY(nvlist_add_string(raidz
, ZPOOL_CONFIG_TYPE
,
889 VDEV_TYPE_RAIDZ
) == 0);
890 VERIFY(nvlist_add_uint64(raidz
, ZPOOL_CONFIG_NPARITY
,
891 ztest_opts
.zo_raidz_parity
) == 0);
892 VERIFY(nvlist_add_nvlist_array(raidz
, ZPOOL_CONFIG_CHILDREN
,
895 for (c
= 0; c
< r
; c
++)
896 nvlist_free(child
[c
]);
898 umem_free(child
, r
* sizeof (nvlist_t
*));
904 make_vdev_mirror(char *path
, char *aux
, char *pool
, size_t size
,
905 uint64_t ashift
, int r
, int m
)
907 nvlist_t
*mirror
, **child
;
911 return (make_vdev_raidz(path
, aux
, pool
, size
, ashift
, r
));
913 child
= umem_alloc(m
* sizeof (nvlist_t
*), UMEM_NOFAIL
);
915 for (c
= 0; c
< m
; c
++)
916 child
[c
] = make_vdev_raidz(path
, aux
, pool
, size
, ashift
, r
);
918 VERIFY(nvlist_alloc(&mirror
, NV_UNIQUE_NAME
, 0) == 0);
919 VERIFY(nvlist_add_string(mirror
, ZPOOL_CONFIG_TYPE
,
920 VDEV_TYPE_MIRROR
) == 0);
921 VERIFY(nvlist_add_nvlist_array(mirror
, ZPOOL_CONFIG_CHILDREN
,
924 for (c
= 0; c
< m
; c
++)
925 nvlist_free(child
[c
]);
927 umem_free(child
, m
* sizeof (nvlist_t
*));
933 make_vdev_root(char *path
, char *aux
, char *pool
, size_t size
, uint64_t ashift
,
934 int log
, int r
, int m
, int t
)
936 nvlist_t
*root
, **child
;
941 child
= umem_alloc(t
* sizeof (nvlist_t
*), UMEM_NOFAIL
);
943 for (c
= 0; c
< t
; c
++) {
944 child
[c
] = make_vdev_mirror(path
, aux
, pool
, size
, ashift
,
946 VERIFY(nvlist_add_uint64(child
[c
], ZPOOL_CONFIG_IS_LOG
,
950 VERIFY(nvlist_alloc(&root
, NV_UNIQUE_NAME
, 0) == 0);
951 VERIFY(nvlist_add_string(root
, ZPOOL_CONFIG_TYPE
, VDEV_TYPE_ROOT
) == 0);
952 VERIFY(nvlist_add_nvlist_array(root
, aux
? aux
: ZPOOL_CONFIG_CHILDREN
,
955 for (c
= 0; c
< t
; c
++)
956 nvlist_free(child
[c
]);
958 umem_free(child
, t
* sizeof (nvlist_t
*));
964 * Find a random spa version. Returns back a random spa version in the
965 * range [initial_version, SPA_VERSION_FEATURES].
968 ztest_random_spa_version(uint64_t initial_version
)
970 uint64_t version
= initial_version
;
972 if (version
<= SPA_VERSION_BEFORE_FEATURES
) {
974 ztest_random(SPA_VERSION_BEFORE_FEATURES
- version
+ 1);
977 if (version
> SPA_VERSION_BEFORE_FEATURES
)
978 version
= SPA_VERSION_FEATURES
;
980 ASSERT(SPA_VERSION_IS_SUPPORTED(version
));
985 ztest_random_blocksize(void)
987 return (1 << (SPA_MINBLOCKSHIFT
+
988 ztest_random(SPA_MAXBLOCKSHIFT
- SPA_MINBLOCKSHIFT
+ 1)));
992 ztest_random_ibshift(void)
994 return (DN_MIN_INDBLKSHIFT
+
995 ztest_random(DN_MAX_INDBLKSHIFT
- DN_MIN_INDBLKSHIFT
+ 1));
999 ztest_random_vdev_top(spa_t
*spa
, boolean_t log_ok
)
1002 vdev_t
*rvd
= spa
->spa_root_vdev
;
1005 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_READER
) != 0);
1008 top
= ztest_random(rvd
->vdev_children
);
1009 tvd
= rvd
->vdev_child
[top
];
1010 } while (tvd
->vdev_ishole
|| (tvd
->vdev_islog
&& !log_ok
) ||
1011 tvd
->vdev_mg
== NULL
|| tvd
->vdev_mg
->mg_class
== NULL
);
1017 ztest_random_dsl_prop(zfs_prop_t prop
)
1022 value
= zfs_prop_random_value(prop
, ztest_random(-1ULL));
1023 } while (prop
== ZFS_PROP_CHECKSUM
&& value
== ZIO_CHECKSUM_OFF
);
1029 ztest_dsl_prop_set_uint64(char *osname
, zfs_prop_t prop
, uint64_t value
,
1032 const char *propname
= zfs_prop_to_name(prop
);
1033 const char *valname
;
1038 error
= dsl_prop_set_int(osname
, propname
,
1039 (inherit
? ZPROP_SRC_NONE
: ZPROP_SRC_LOCAL
), value
);
1041 if (error
== ENOSPC
) {
1042 ztest_record_enospc(FTAG
);
1047 setpoint
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
1048 VERIFY0(dsl_prop_get_integer(osname
, propname
, &curval
, setpoint
));
1050 if (ztest_opts
.zo_verbose
>= 6) {
1051 VERIFY(zfs_prop_index_to_string(prop
, curval
, &valname
) == 0);
1052 (void) printf("%s %s = %s at '%s'\n",
1053 osname
, propname
, valname
, setpoint
);
1055 umem_free(setpoint
, MAXPATHLEN
);
1061 ztest_spa_prop_set_uint64(zpool_prop_t prop
, uint64_t value
)
1063 spa_t
*spa
= ztest_spa
;
1064 nvlist_t
*props
= NULL
;
1067 VERIFY(nvlist_alloc(&props
, NV_UNIQUE_NAME
, 0) == 0);
1068 VERIFY(nvlist_add_uint64(props
, zpool_prop_to_name(prop
), value
) == 0);
1070 error
= spa_prop_set(spa
, props
);
1074 if (error
== ENOSPC
) {
1075 ztest_record_enospc(FTAG
);
1084 ztest_rll_init(rll_t
*rll
)
1086 rll
->rll_writer
= NULL
;
1087 rll
->rll_readers
= 0;
1088 mutex_init(&rll
->rll_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
1089 cv_init(&rll
->rll_cv
, NULL
, CV_DEFAULT
, NULL
);
1093 ztest_rll_destroy(rll_t
*rll
)
1095 ASSERT(rll
->rll_writer
== NULL
);
1096 ASSERT(rll
->rll_readers
== 0);
1097 mutex_destroy(&rll
->rll_lock
);
1098 cv_destroy(&rll
->rll_cv
);
1102 ztest_rll_lock(rll_t
*rll
, rl_type_t type
)
1104 mutex_enter(&rll
->rll_lock
);
1106 if (type
== RL_READER
) {
1107 while (rll
->rll_writer
!= NULL
)
1108 (void) cv_wait(&rll
->rll_cv
, &rll
->rll_lock
);
1111 while (rll
->rll_writer
!= NULL
|| rll
->rll_readers
)
1112 (void) cv_wait(&rll
->rll_cv
, &rll
->rll_lock
);
1113 rll
->rll_writer
= curthread
;
1116 mutex_exit(&rll
->rll_lock
);
1120 ztest_rll_unlock(rll_t
*rll
)
1122 mutex_enter(&rll
->rll_lock
);
1124 if (rll
->rll_writer
) {
1125 ASSERT(rll
->rll_readers
== 0);
1126 rll
->rll_writer
= NULL
;
1128 ASSERT(rll
->rll_readers
!= 0);
1129 ASSERT(rll
->rll_writer
== NULL
);
1133 if (rll
->rll_writer
== NULL
&& rll
->rll_readers
== 0)
1134 cv_broadcast(&rll
->rll_cv
);
1136 mutex_exit(&rll
->rll_lock
);
1140 ztest_object_lock(ztest_ds_t
*zd
, uint64_t object
, rl_type_t type
)
1142 rll_t
*rll
= &zd
->zd_object_lock
[object
& (ZTEST_OBJECT_LOCKS
- 1)];
1144 ztest_rll_lock(rll
, type
);
1148 ztest_object_unlock(ztest_ds_t
*zd
, uint64_t object
)
1150 rll_t
*rll
= &zd
->zd_object_lock
[object
& (ZTEST_OBJECT_LOCKS
- 1)];
1152 ztest_rll_unlock(rll
);
1156 ztest_range_lock(ztest_ds_t
*zd
, uint64_t object
, uint64_t offset
,
1157 uint64_t size
, rl_type_t type
)
1159 uint64_t hash
= object
^ (offset
% (ZTEST_RANGE_LOCKS
+ 1));
1160 rll_t
*rll
= &zd
->zd_range_lock
[hash
& (ZTEST_RANGE_LOCKS
- 1)];
1163 rl
= umem_alloc(sizeof (*rl
), UMEM_NOFAIL
);
1164 rl
->rl_object
= object
;
1165 rl
->rl_offset
= offset
;
1169 ztest_rll_lock(rll
, type
);
1175 ztest_range_unlock(rl_t
*rl
)
1177 rll_t
*rll
= rl
->rl_lock
;
1179 ztest_rll_unlock(rll
);
1181 umem_free(rl
, sizeof (*rl
));
1185 ztest_zd_init(ztest_ds_t
*zd
, ztest_shared_ds_t
*szd
, objset_t
*os
)
1188 zd
->zd_zilog
= dmu_objset_zil(os
);
1189 zd
->zd_shared
= szd
;
1190 dmu_objset_name(os
, zd
->zd_name
);
1193 if (zd
->zd_shared
!= NULL
)
1194 zd
->zd_shared
->zd_seq
= 0;
1196 rw_init(&zd
->zd_zilog_lock
, NULL
, RW_DEFAULT
, NULL
);
1197 mutex_init(&zd
->zd_dirobj_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
1199 for (l
= 0; l
< ZTEST_OBJECT_LOCKS
; l
++)
1200 ztest_rll_init(&zd
->zd_object_lock
[l
]);
1202 for (l
= 0; l
< ZTEST_RANGE_LOCKS
; l
++)
1203 ztest_rll_init(&zd
->zd_range_lock
[l
]);
1207 ztest_zd_fini(ztest_ds_t
*zd
)
1211 mutex_destroy(&zd
->zd_dirobj_lock
);
1212 rw_destroy(&zd
->zd_zilog_lock
);
1214 for (l
= 0; l
< ZTEST_OBJECT_LOCKS
; l
++)
1215 ztest_rll_destroy(&zd
->zd_object_lock
[l
]);
1217 for (l
= 0; l
< ZTEST_RANGE_LOCKS
; l
++)
1218 ztest_rll_destroy(&zd
->zd_range_lock
[l
]);
1221 #define TXG_MIGHTWAIT (ztest_random(10) == 0 ? TXG_NOWAIT : TXG_WAIT)
1224 ztest_tx_assign(dmu_tx_t
*tx
, uint64_t txg_how
, const char *tag
)
1230 * Attempt to assign tx to some transaction group.
1232 error
= dmu_tx_assign(tx
, txg_how
);
1234 if (error
== ERESTART
) {
1235 ASSERT(txg_how
== TXG_NOWAIT
);
1238 ASSERT3U(error
, ==, ENOSPC
);
1239 ztest_record_enospc(tag
);
1244 txg
= dmu_tx_get_txg(tx
);
1250 ztest_pattern_set(void *buf
, uint64_t size
, uint64_t value
)
1253 uint64_t *ip_end
= (uint64_t *)((uintptr_t)buf
+ (uintptr_t)size
);
1261 ztest_pattern_match(void *buf
, uint64_t size
, uint64_t value
)
1264 uint64_t *ip_end
= (uint64_t *)((uintptr_t)buf
+ (uintptr_t)size
);
1268 diff
|= (value
- *ip
++);
1275 ztest_bt_generate(ztest_block_tag_t
*bt
, objset_t
*os
, uint64_t object
,
1276 uint64_t offset
, uint64_t gen
, uint64_t txg
, uint64_t crtxg
)
1278 bt
->bt_magic
= BT_MAGIC
;
1279 bt
->bt_objset
= dmu_objset_id(os
);
1280 bt
->bt_object
= object
;
1281 bt
->bt_offset
= offset
;
1284 bt
->bt_crtxg
= crtxg
;
1288 ztest_bt_verify(ztest_block_tag_t
*bt
, objset_t
*os
, uint64_t object
,
1289 uint64_t offset
, uint64_t gen
, uint64_t txg
, uint64_t crtxg
)
1291 ASSERT(bt
->bt_magic
== BT_MAGIC
);
1292 ASSERT(bt
->bt_objset
== dmu_objset_id(os
));
1293 ASSERT(bt
->bt_object
== object
);
1294 ASSERT(bt
->bt_offset
== offset
);
1295 ASSERT(bt
->bt_gen
<= gen
);
1296 ASSERT(bt
->bt_txg
<= txg
);
1297 ASSERT(bt
->bt_crtxg
== crtxg
);
1300 static ztest_block_tag_t
*
1301 ztest_bt_bonus(dmu_buf_t
*db
)
1303 dmu_object_info_t doi
;
1304 ztest_block_tag_t
*bt
;
1306 dmu_object_info_from_db(db
, &doi
);
1307 ASSERT3U(doi
.doi_bonus_size
, <=, db
->db_size
);
1308 ASSERT3U(doi
.doi_bonus_size
, >=, sizeof (*bt
));
1309 bt
= (void *)((char *)db
->db_data
+ doi
.doi_bonus_size
- sizeof (*bt
));
1318 #define lrz_type lr_mode
1319 #define lrz_blocksize lr_uid
1320 #define lrz_ibshift lr_gid
1321 #define lrz_bonustype lr_rdev
1322 #define lrz_bonuslen lr_crtime[1]
1325 ztest_log_create(ztest_ds_t
*zd
, dmu_tx_t
*tx
, lr_create_t
*lr
)
1327 char *name
= (void *)(lr
+ 1); /* name follows lr */
1328 size_t namesize
= strlen(name
) + 1;
1331 if (zil_replaying(zd
->zd_zilog
, tx
))
1334 itx
= zil_itx_create(TX_CREATE
, sizeof (*lr
) + namesize
);
1335 bcopy(&lr
->lr_common
+ 1, &itx
->itx_lr
+ 1,
1336 sizeof (*lr
) + namesize
- sizeof (lr_t
));
1338 zil_itx_assign(zd
->zd_zilog
, itx
, tx
);
1342 ztest_log_remove(ztest_ds_t
*zd
, dmu_tx_t
*tx
, lr_remove_t
*lr
, uint64_t object
)
1344 char *name
= (void *)(lr
+ 1); /* name follows lr */
1345 size_t namesize
= strlen(name
) + 1;
1348 if (zil_replaying(zd
->zd_zilog
, tx
))
1351 itx
= zil_itx_create(TX_REMOVE
, sizeof (*lr
) + namesize
);
1352 bcopy(&lr
->lr_common
+ 1, &itx
->itx_lr
+ 1,
1353 sizeof (*lr
) + namesize
- sizeof (lr_t
));
1355 itx
->itx_oid
= object
;
1356 zil_itx_assign(zd
->zd_zilog
, itx
, tx
);
1360 ztest_log_write(ztest_ds_t
*zd
, dmu_tx_t
*tx
, lr_write_t
*lr
)
1363 itx_wr_state_t write_state
= ztest_random(WR_NUM_STATES
);
1365 if (zil_replaying(zd
->zd_zilog
, tx
))
1368 if (lr
->lr_length
> ZIL_MAX_LOG_DATA
)
1369 write_state
= WR_INDIRECT
;
1371 itx
= zil_itx_create(TX_WRITE
,
1372 sizeof (*lr
) + (write_state
== WR_COPIED
? lr
->lr_length
: 0));
1374 if (write_state
== WR_COPIED
&&
1375 dmu_read(zd
->zd_os
, lr
->lr_foid
, lr
->lr_offset
, lr
->lr_length
,
1376 ((lr_write_t
*)&itx
->itx_lr
) + 1, DMU_READ_NO_PREFETCH
) != 0) {
1377 zil_itx_destroy(itx
);
1378 itx
= zil_itx_create(TX_WRITE
, sizeof (*lr
));
1379 write_state
= WR_NEED_COPY
;
1381 itx
->itx_private
= zd
;
1382 itx
->itx_wr_state
= write_state
;
1383 itx
->itx_sync
= (ztest_random(8) == 0);
1384 itx
->itx_sod
+= (write_state
== WR_NEED_COPY
? lr
->lr_length
: 0);
1386 bcopy(&lr
->lr_common
+ 1, &itx
->itx_lr
+ 1,
1387 sizeof (*lr
) - sizeof (lr_t
));
1389 zil_itx_assign(zd
->zd_zilog
, itx
, tx
);
1393 ztest_log_truncate(ztest_ds_t
*zd
, dmu_tx_t
*tx
, lr_truncate_t
*lr
)
1397 if (zil_replaying(zd
->zd_zilog
, tx
))
1400 itx
= zil_itx_create(TX_TRUNCATE
, sizeof (*lr
));
1401 bcopy(&lr
->lr_common
+ 1, &itx
->itx_lr
+ 1,
1402 sizeof (*lr
) - sizeof (lr_t
));
1404 itx
->itx_sync
= B_FALSE
;
1405 zil_itx_assign(zd
->zd_zilog
, itx
, tx
);
1409 ztest_log_setattr(ztest_ds_t
*zd
, dmu_tx_t
*tx
, lr_setattr_t
*lr
)
1413 if (zil_replaying(zd
->zd_zilog
, tx
))
1416 itx
= zil_itx_create(TX_SETATTR
, sizeof (*lr
));
1417 bcopy(&lr
->lr_common
+ 1, &itx
->itx_lr
+ 1,
1418 sizeof (*lr
) - sizeof (lr_t
));
1420 itx
->itx_sync
= B_FALSE
;
1421 zil_itx_assign(zd
->zd_zilog
, itx
, tx
);
1428 ztest_replay_create(ztest_ds_t
*zd
, lr_create_t
*lr
, boolean_t byteswap
)
1430 char *name
= (void *)(lr
+ 1); /* name follows lr */
1431 objset_t
*os
= zd
->zd_os
;
1432 ztest_block_tag_t
*bbt
;
1439 byteswap_uint64_array(lr
, sizeof (*lr
));
1441 ASSERT(lr
->lr_doid
== ZTEST_DIROBJ
);
1442 ASSERT(name
[0] != '\0');
1444 tx
= dmu_tx_create(os
);
1446 dmu_tx_hold_zap(tx
, lr
->lr_doid
, B_TRUE
, name
);
1448 if (lr
->lrz_type
== DMU_OT_ZAP_OTHER
) {
1449 dmu_tx_hold_zap(tx
, DMU_NEW_OBJECT
, B_TRUE
, NULL
);
1451 dmu_tx_hold_bonus(tx
, DMU_NEW_OBJECT
);
1454 txg
= ztest_tx_assign(tx
, TXG_WAIT
, FTAG
);
1458 ASSERT(dmu_objset_zil(os
)->zl_replay
== !!lr
->lr_foid
);
1460 if (lr
->lrz_type
== DMU_OT_ZAP_OTHER
) {
1461 if (lr
->lr_foid
== 0) {
1462 lr
->lr_foid
= zap_create(os
,
1463 lr
->lrz_type
, lr
->lrz_bonustype
,
1464 lr
->lrz_bonuslen
, tx
);
1466 error
= zap_create_claim(os
, lr
->lr_foid
,
1467 lr
->lrz_type
, lr
->lrz_bonustype
,
1468 lr
->lrz_bonuslen
, tx
);
1471 if (lr
->lr_foid
== 0) {
1472 lr
->lr_foid
= dmu_object_alloc(os
,
1473 lr
->lrz_type
, 0, lr
->lrz_bonustype
,
1474 lr
->lrz_bonuslen
, tx
);
1476 error
= dmu_object_claim(os
, lr
->lr_foid
,
1477 lr
->lrz_type
, 0, lr
->lrz_bonustype
,
1478 lr
->lrz_bonuslen
, tx
);
1483 ASSERT3U(error
, ==, EEXIST
);
1484 ASSERT(zd
->zd_zilog
->zl_replay
);
1489 ASSERT(lr
->lr_foid
!= 0);
1491 if (lr
->lrz_type
!= DMU_OT_ZAP_OTHER
)
1492 VERIFY3U(0, ==, dmu_object_set_blocksize(os
, lr
->lr_foid
,
1493 lr
->lrz_blocksize
, lr
->lrz_ibshift
, tx
));
1495 VERIFY3U(0, ==, dmu_bonus_hold(os
, lr
->lr_foid
, FTAG
, &db
));
1496 bbt
= ztest_bt_bonus(db
);
1497 dmu_buf_will_dirty(db
, tx
);
1498 ztest_bt_generate(bbt
, os
, lr
->lr_foid
, -1ULL, lr
->lr_gen
, txg
, txg
);
1499 dmu_buf_rele(db
, FTAG
);
1501 VERIFY3U(0, ==, zap_add(os
, lr
->lr_doid
, name
, sizeof (uint64_t), 1,
1504 (void) ztest_log_create(zd
, tx
, lr
);
1512 ztest_replay_remove(ztest_ds_t
*zd
, lr_remove_t
*lr
, boolean_t byteswap
)
1514 char *name
= (void *)(lr
+ 1); /* name follows lr */
1515 objset_t
*os
= zd
->zd_os
;
1516 dmu_object_info_t doi
;
1518 uint64_t object
, txg
;
1521 byteswap_uint64_array(lr
, sizeof (*lr
));
1523 ASSERT(lr
->lr_doid
== ZTEST_DIROBJ
);
1524 ASSERT(name
[0] != '\0');
1527 zap_lookup(os
, lr
->lr_doid
, name
, sizeof (object
), 1, &object
));
1528 ASSERT(object
!= 0);
1530 ztest_object_lock(zd
, object
, RL_WRITER
);
1532 VERIFY3U(0, ==, dmu_object_info(os
, object
, &doi
));
1534 tx
= dmu_tx_create(os
);
1536 dmu_tx_hold_zap(tx
, lr
->lr_doid
, B_FALSE
, name
);
1537 dmu_tx_hold_free(tx
, object
, 0, DMU_OBJECT_END
);
1539 txg
= ztest_tx_assign(tx
, TXG_WAIT
, FTAG
);
1541 ztest_object_unlock(zd
, object
);
1545 if (doi
.doi_type
== DMU_OT_ZAP_OTHER
) {
1546 VERIFY3U(0, ==, zap_destroy(os
, object
, tx
));
1548 VERIFY3U(0, ==, dmu_object_free(os
, object
, tx
));
1551 VERIFY3U(0, ==, zap_remove(os
, lr
->lr_doid
, name
, tx
));
1553 (void) ztest_log_remove(zd
, tx
, lr
, object
);
1557 ztest_object_unlock(zd
, object
);
1563 ztest_replay_write(ztest_ds_t
*zd
, lr_write_t
*lr
, boolean_t byteswap
)
1565 objset_t
*os
= zd
->zd_os
;
1566 void *data
= lr
+ 1; /* data follows lr */
1567 uint64_t offset
, length
;
1568 ztest_block_tag_t
*bt
= data
;
1569 ztest_block_tag_t
*bbt
;
1570 uint64_t gen
, txg
, lrtxg
, crtxg
;
1571 dmu_object_info_t doi
;
1574 arc_buf_t
*abuf
= NULL
;
1578 byteswap_uint64_array(lr
, sizeof (*lr
));
1580 offset
= lr
->lr_offset
;
1581 length
= lr
->lr_length
;
1583 /* If it's a dmu_sync() block, write the whole block */
1584 if (lr
->lr_common
.lrc_reclen
== sizeof (lr_write_t
)) {
1585 uint64_t blocksize
= BP_GET_LSIZE(&lr
->lr_blkptr
);
1586 if (length
< blocksize
) {
1587 offset
-= offset
% blocksize
;
1592 if (bt
->bt_magic
== BSWAP_64(BT_MAGIC
))
1593 byteswap_uint64_array(bt
, sizeof (*bt
));
1595 if (bt
->bt_magic
!= BT_MAGIC
)
1598 ztest_object_lock(zd
, lr
->lr_foid
, RL_READER
);
1599 rl
= ztest_range_lock(zd
, lr
->lr_foid
, offset
, length
, RL_WRITER
);
1601 VERIFY3U(0, ==, dmu_bonus_hold(os
, lr
->lr_foid
, FTAG
, &db
));
1603 dmu_object_info_from_db(db
, &doi
);
1605 bbt
= ztest_bt_bonus(db
);
1606 ASSERT3U(bbt
->bt_magic
, ==, BT_MAGIC
);
1608 crtxg
= bbt
->bt_crtxg
;
1609 lrtxg
= lr
->lr_common
.lrc_txg
;
1611 tx
= dmu_tx_create(os
);
1613 dmu_tx_hold_write(tx
, lr
->lr_foid
, offset
, length
);
1615 if (ztest_random(8) == 0 && length
== doi
.doi_data_block_size
&&
1616 P2PHASE(offset
, length
) == 0)
1617 abuf
= dmu_request_arcbuf(db
, length
);
1619 txg
= ztest_tx_assign(tx
, TXG_WAIT
, FTAG
);
1622 dmu_return_arcbuf(abuf
);
1623 dmu_buf_rele(db
, FTAG
);
1624 ztest_range_unlock(rl
);
1625 ztest_object_unlock(zd
, lr
->lr_foid
);
1631 * Usually, verify the old data before writing new data --
1632 * but not always, because we also want to verify correct
1633 * behavior when the data was not recently read into cache.
1635 ASSERT(offset
% doi
.doi_data_block_size
== 0);
1636 if (ztest_random(4) != 0) {
1637 int prefetch
= ztest_random(2) ?
1638 DMU_READ_PREFETCH
: DMU_READ_NO_PREFETCH
;
1639 ztest_block_tag_t rbt
;
1641 VERIFY(dmu_read(os
, lr
->lr_foid
, offset
,
1642 sizeof (rbt
), &rbt
, prefetch
) == 0);
1643 if (rbt
.bt_magic
== BT_MAGIC
) {
1644 ztest_bt_verify(&rbt
, os
, lr
->lr_foid
,
1645 offset
, gen
, txg
, crtxg
);
1650 * Writes can appear to be newer than the bonus buffer because
1651 * the ztest_get_data() callback does a dmu_read() of the
1652 * open-context data, which may be different than the data
1653 * as it was when the write was generated.
1655 if (zd
->zd_zilog
->zl_replay
) {
1656 ztest_bt_verify(bt
, os
, lr
->lr_foid
, offset
,
1657 MAX(gen
, bt
->bt_gen
), MAX(txg
, lrtxg
),
1662 * Set the bt's gen/txg to the bonus buffer's gen/txg
1663 * so that all of the usual ASSERTs will work.
1665 ztest_bt_generate(bt
, os
, lr
->lr_foid
, offset
, gen
, txg
, crtxg
);
1669 dmu_write(os
, lr
->lr_foid
, offset
, length
, data
, tx
);
1671 bcopy(data
, abuf
->b_data
, length
);
1672 dmu_assign_arcbuf(db
, offset
, abuf
, tx
);
1675 (void) ztest_log_write(zd
, tx
, lr
);
1677 dmu_buf_rele(db
, FTAG
);
1681 ztest_range_unlock(rl
);
1682 ztest_object_unlock(zd
, lr
->lr_foid
);
1688 ztest_replay_truncate(ztest_ds_t
*zd
, lr_truncate_t
*lr
, boolean_t byteswap
)
1690 objset_t
*os
= zd
->zd_os
;
1696 byteswap_uint64_array(lr
, sizeof (*lr
));
1698 ztest_object_lock(zd
, lr
->lr_foid
, RL_READER
);
1699 rl
= ztest_range_lock(zd
, lr
->lr_foid
, lr
->lr_offset
, lr
->lr_length
,
1702 tx
= dmu_tx_create(os
);
1704 dmu_tx_hold_free(tx
, lr
->lr_foid
, lr
->lr_offset
, lr
->lr_length
);
1706 txg
= ztest_tx_assign(tx
, TXG_WAIT
, FTAG
);
1708 ztest_range_unlock(rl
);
1709 ztest_object_unlock(zd
, lr
->lr_foid
);
1713 VERIFY(dmu_free_range(os
, lr
->lr_foid
, lr
->lr_offset
,
1714 lr
->lr_length
, tx
) == 0);
1716 (void) ztest_log_truncate(zd
, tx
, lr
);
1720 ztest_range_unlock(rl
);
1721 ztest_object_unlock(zd
, lr
->lr_foid
);
1727 ztest_replay_setattr(ztest_ds_t
*zd
, lr_setattr_t
*lr
, boolean_t byteswap
)
1729 objset_t
*os
= zd
->zd_os
;
1732 ztest_block_tag_t
*bbt
;
1733 uint64_t txg
, lrtxg
, crtxg
;
1736 byteswap_uint64_array(lr
, sizeof (*lr
));
1738 ztest_object_lock(zd
, lr
->lr_foid
, RL_WRITER
);
1740 VERIFY3U(0, ==, dmu_bonus_hold(os
, lr
->lr_foid
, FTAG
, &db
));
1742 tx
= dmu_tx_create(os
);
1743 dmu_tx_hold_bonus(tx
, lr
->lr_foid
);
1745 txg
= ztest_tx_assign(tx
, TXG_WAIT
, FTAG
);
1747 dmu_buf_rele(db
, FTAG
);
1748 ztest_object_unlock(zd
, lr
->lr_foid
);
1752 bbt
= ztest_bt_bonus(db
);
1753 ASSERT3U(bbt
->bt_magic
, ==, BT_MAGIC
);
1754 crtxg
= bbt
->bt_crtxg
;
1755 lrtxg
= lr
->lr_common
.lrc_txg
;
1757 if (zd
->zd_zilog
->zl_replay
) {
1758 ASSERT(lr
->lr_size
!= 0);
1759 ASSERT(lr
->lr_mode
!= 0);
1763 * Randomly change the size and increment the generation.
1765 lr
->lr_size
= (ztest_random(db
->db_size
/ sizeof (*bbt
)) + 1) *
1767 lr
->lr_mode
= bbt
->bt_gen
+ 1;
1772 * Verify that the current bonus buffer is not newer than our txg.
1774 ztest_bt_verify(bbt
, os
, lr
->lr_foid
, -1ULL, lr
->lr_mode
,
1775 MAX(txg
, lrtxg
), crtxg
);
1777 dmu_buf_will_dirty(db
, tx
);
1779 ASSERT3U(lr
->lr_size
, >=, sizeof (*bbt
));
1780 ASSERT3U(lr
->lr_size
, <=, db
->db_size
);
1781 VERIFY0(dmu_set_bonus(db
, lr
->lr_size
, tx
));
1782 bbt
= ztest_bt_bonus(db
);
1784 ztest_bt_generate(bbt
, os
, lr
->lr_foid
, -1ULL, lr
->lr_mode
, txg
, crtxg
);
1786 dmu_buf_rele(db
, FTAG
);
1788 (void) ztest_log_setattr(zd
, tx
, lr
);
1792 ztest_object_unlock(zd
, lr
->lr_foid
);
1797 zil_replay_func_t ztest_replay_vector
[TX_MAX_TYPE
] = {
1798 NULL
, /* 0 no such transaction type */
1799 (zil_replay_func_t
)ztest_replay_create
, /* TX_CREATE */
1800 NULL
, /* TX_MKDIR */
1801 NULL
, /* TX_MKXATTR */
1802 NULL
, /* TX_SYMLINK */
1803 (zil_replay_func_t
)ztest_replay_remove
, /* TX_REMOVE */
1804 NULL
, /* TX_RMDIR */
1806 NULL
, /* TX_RENAME */
1807 (zil_replay_func_t
)ztest_replay_write
, /* TX_WRITE */
1808 (zil_replay_func_t
)ztest_replay_truncate
, /* TX_TRUNCATE */
1809 (zil_replay_func_t
)ztest_replay_setattr
, /* TX_SETATTR */
1811 NULL
, /* TX_CREATE_ACL */
1812 NULL
, /* TX_CREATE_ATTR */
1813 NULL
, /* TX_CREATE_ACL_ATTR */
1814 NULL
, /* TX_MKDIR_ACL */
1815 NULL
, /* TX_MKDIR_ATTR */
1816 NULL
, /* TX_MKDIR_ACL_ATTR */
1817 NULL
, /* TX_WRITE2 */
1821 * ZIL get_data callbacks
1825 ztest_get_done(zgd_t
*zgd
, int error
)
1827 ztest_ds_t
*zd
= zgd
->zgd_private
;
1828 uint64_t object
= zgd
->zgd_rl
->rl_object
;
1831 dmu_buf_rele(zgd
->zgd_db
, zgd
);
1833 ztest_range_unlock(zgd
->zgd_rl
);
1834 ztest_object_unlock(zd
, object
);
1836 if (error
== 0 && zgd
->zgd_bp
)
1837 zil_add_block(zgd
->zgd_zilog
, zgd
->zgd_bp
);
1839 umem_free(zgd
, sizeof (*zgd
));
1843 ztest_get_data(void *arg
, lr_write_t
*lr
, char *buf
, zio_t
*zio
)
1845 ztest_ds_t
*zd
= arg
;
1846 objset_t
*os
= zd
->zd_os
;
1847 uint64_t object
= lr
->lr_foid
;
1848 uint64_t offset
= lr
->lr_offset
;
1849 uint64_t size
= lr
->lr_length
;
1850 blkptr_t
*bp
= &lr
->lr_blkptr
;
1851 uint64_t txg
= lr
->lr_common
.lrc_txg
;
1853 dmu_object_info_t doi
;
1858 ztest_object_lock(zd
, object
, RL_READER
);
1859 error
= dmu_bonus_hold(os
, object
, FTAG
, &db
);
1861 ztest_object_unlock(zd
, object
);
1865 crtxg
= ztest_bt_bonus(db
)->bt_crtxg
;
1867 if (crtxg
== 0 || crtxg
> txg
) {
1868 dmu_buf_rele(db
, FTAG
);
1869 ztest_object_unlock(zd
, object
);
1873 dmu_object_info_from_db(db
, &doi
);
1874 dmu_buf_rele(db
, FTAG
);
1877 zgd
= umem_zalloc(sizeof (*zgd
), UMEM_NOFAIL
);
1878 zgd
->zgd_zilog
= zd
->zd_zilog
;
1879 zgd
->zgd_private
= zd
;
1881 if (buf
!= NULL
) { /* immediate write */
1882 zgd
->zgd_rl
= ztest_range_lock(zd
, object
, offset
, size
,
1885 error
= dmu_read(os
, object
, offset
, size
, buf
,
1886 DMU_READ_NO_PREFETCH
);
1889 size
= doi
.doi_data_block_size
;
1891 offset
= P2ALIGN(offset
, size
);
1893 ASSERT(offset
< size
);
1897 zgd
->zgd_rl
= ztest_range_lock(zd
, object
, offset
, size
,
1900 error
= dmu_buf_hold(os
, object
, offset
, zgd
, &db
,
1901 DMU_READ_NO_PREFETCH
);
1904 blkptr_t
*obp
= dmu_buf_get_blkptr(db
);
1906 ASSERT(BP_IS_HOLE(bp
));
1913 ASSERT(db
->db_offset
== offset
);
1914 ASSERT(db
->db_size
== size
);
1916 error
= dmu_sync(zio
, lr
->lr_common
.lrc_txg
,
1917 ztest_get_done
, zgd
);
1924 ztest_get_done(zgd
, error
);
1930 ztest_lr_alloc(size_t lrsize
, char *name
)
1933 size_t namesize
= name
? strlen(name
) + 1 : 0;
1935 lr
= umem_zalloc(lrsize
+ namesize
, UMEM_NOFAIL
);
1938 bcopy(name
, lr
+ lrsize
, namesize
);
1944 ztest_lr_free(void *lr
, size_t lrsize
, char *name
)
1946 size_t namesize
= name
? strlen(name
) + 1 : 0;
1948 umem_free(lr
, lrsize
+ namesize
);
1952 * Lookup a bunch of objects. Returns the number of objects not found.
1955 ztest_lookup(ztest_ds_t
*zd
, ztest_od_t
*od
, int count
)
1961 ASSERT(mutex_held(&zd
->zd_dirobj_lock
));
1963 for (i
= 0; i
< count
; i
++, od
++) {
1965 error
= zap_lookup(zd
->zd_os
, od
->od_dir
, od
->od_name
,
1966 sizeof (uint64_t), 1, &od
->od_object
);
1968 ASSERT(error
== ENOENT
);
1969 ASSERT(od
->od_object
== 0);
1973 ztest_block_tag_t
*bbt
;
1974 dmu_object_info_t doi
;
1976 ASSERT(od
->od_object
!= 0);
1977 ASSERT(missing
== 0); /* there should be no gaps */
1979 ztest_object_lock(zd
, od
->od_object
, RL_READER
);
1980 VERIFY3U(0, ==, dmu_bonus_hold(zd
->zd_os
,
1981 od
->od_object
, FTAG
, &db
));
1982 dmu_object_info_from_db(db
, &doi
);
1983 bbt
= ztest_bt_bonus(db
);
1984 ASSERT3U(bbt
->bt_magic
, ==, BT_MAGIC
);
1985 od
->od_type
= doi
.doi_type
;
1986 od
->od_blocksize
= doi
.doi_data_block_size
;
1987 od
->od_gen
= bbt
->bt_gen
;
1988 dmu_buf_rele(db
, FTAG
);
1989 ztest_object_unlock(zd
, od
->od_object
);
1997 ztest_create(ztest_ds_t
*zd
, ztest_od_t
*od
, int count
)
2002 ASSERT(mutex_held(&zd
->zd_dirobj_lock
));
2004 for (i
= 0; i
< count
; i
++, od
++) {
2011 lr_create_t
*lr
= ztest_lr_alloc(sizeof (*lr
), od
->od_name
);
2013 lr
->lr_doid
= od
->od_dir
;
2014 lr
->lr_foid
= 0; /* 0 to allocate, > 0 to claim */
2015 lr
->lrz_type
= od
->od_crtype
;
2016 lr
->lrz_blocksize
= od
->od_crblocksize
;
2017 lr
->lrz_ibshift
= ztest_random_ibshift();
2018 lr
->lrz_bonustype
= DMU_OT_UINT64_OTHER
;
2019 lr
->lrz_bonuslen
= dmu_bonus_max();
2020 lr
->lr_gen
= od
->od_crgen
;
2021 lr
->lr_crtime
[0] = time(NULL
);
2023 if (ztest_replay_create(zd
, lr
, B_FALSE
) != 0) {
2024 ASSERT(missing
== 0);
2028 od
->od_object
= lr
->lr_foid
;
2029 od
->od_type
= od
->od_crtype
;
2030 od
->od_blocksize
= od
->od_crblocksize
;
2031 od
->od_gen
= od
->od_crgen
;
2032 ASSERT(od
->od_object
!= 0);
2035 ztest_lr_free(lr
, sizeof (*lr
), od
->od_name
);
2042 ztest_remove(ztest_ds_t
*zd
, ztest_od_t
*od
, int count
)
2048 ASSERT(mutex_held(&zd
->zd_dirobj_lock
));
2052 for (i
= count
- 1; i
>= 0; i
--, od
--) {
2059 * No object was found.
2061 if (od
->od_object
== 0)
2064 lr_remove_t
*lr
= ztest_lr_alloc(sizeof (*lr
), od
->od_name
);
2066 lr
->lr_doid
= od
->od_dir
;
2068 if ((error
= ztest_replay_remove(zd
, lr
, B_FALSE
)) != 0) {
2069 ASSERT3U(error
, ==, ENOSPC
);
2074 ztest_lr_free(lr
, sizeof (*lr
), od
->od_name
);
2081 ztest_write(ztest_ds_t
*zd
, uint64_t object
, uint64_t offset
, uint64_t size
,
2087 lr
= ztest_lr_alloc(sizeof (*lr
) + size
, NULL
);
2089 lr
->lr_foid
= object
;
2090 lr
->lr_offset
= offset
;
2091 lr
->lr_length
= size
;
2093 BP_ZERO(&lr
->lr_blkptr
);
2095 bcopy(data
, lr
+ 1, size
);
2097 error
= ztest_replay_write(zd
, lr
, B_FALSE
);
2099 ztest_lr_free(lr
, sizeof (*lr
) + size
, NULL
);
2105 ztest_truncate(ztest_ds_t
*zd
, uint64_t object
, uint64_t offset
, uint64_t size
)
2110 lr
= ztest_lr_alloc(sizeof (*lr
), NULL
);
2112 lr
->lr_foid
= object
;
2113 lr
->lr_offset
= offset
;
2114 lr
->lr_length
= size
;
2116 error
= ztest_replay_truncate(zd
, lr
, B_FALSE
);
2118 ztest_lr_free(lr
, sizeof (*lr
), NULL
);
2124 ztest_setattr(ztest_ds_t
*zd
, uint64_t object
)
2129 lr
= ztest_lr_alloc(sizeof (*lr
), NULL
);
2131 lr
->lr_foid
= object
;
2135 error
= ztest_replay_setattr(zd
, lr
, B_FALSE
);
2137 ztest_lr_free(lr
, sizeof (*lr
), NULL
);
2143 ztest_prealloc(ztest_ds_t
*zd
, uint64_t object
, uint64_t offset
, uint64_t size
)
2145 objset_t
*os
= zd
->zd_os
;
2150 txg_wait_synced(dmu_objset_pool(os
), 0);
2152 ztest_object_lock(zd
, object
, RL_READER
);
2153 rl
= ztest_range_lock(zd
, object
, offset
, size
, RL_WRITER
);
2155 tx
= dmu_tx_create(os
);
2157 dmu_tx_hold_write(tx
, object
, offset
, size
);
2159 txg
= ztest_tx_assign(tx
, TXG_WAIT
, FTAG
);
2162 dmu_prealloc(os
, object
, offset
, size
, tx
);
2164 txg_wait_synced(dmu_objset_pool(os
), txg
);
2166 (void) dmu_free_long_range(os
, object
, offset
, size
);
2169 ztest_range_unlock(rl
);
2170 ztest_object_unlock(zd
, object
);
2174 ztest_io(ztest_ds_t
*zd
, uint64_t object
, uint64_t offset
)
2177 ztest_block_tag_t wbt
;
2178 dmu_object_info_t doi
;
2179 enum ztest_io_type io_type
;
2183 VERIFY(dmu_object_info(zd
->zd_os
, object
, &doi
) == 0);
2184 blocksize
= doi
.doi_data_block_size
;
2185 data
= umem_alloc(blocksize
, UMEM_NOFAIL
);
2188 * Pick an i/o type at random, biased toward writing block tags.
2190 io_type
= ztest_random(ZTEST_IO_TYPES
);
2191 if (ztest_random(2) == 0)
2192 io_type
= ZTEST_IO_WRITE_TAG
;
2194 (void) rw_enter(&zd
->zd_zilog_lock
, RW_READER
);
2198 case ZTEST_IO_WRITE_TAG
:
2199 ztest_bt_generate(&wbt
, zd
->zd_os
, object
, offset
, 0, 0, 0);
2200 (void) ztest_write(zd
, object
, offset
, sizeof (wbt
), &wbt
);
2203 case ZTEST_IO_WRITE_PATTERN
:
2204 (void) memset(data
, 'a' + (object
+ offset
) % 5, blocksize
);
2205 if (ztest_random(2) == 0) {
2207 * Induce fletcher2 collisions to ensure that
2208 * zio_ddt_collision() detects and resolves them
2209 * when using fletcher2-verify for deduplication.
2211 ((uint64_t *)data
)[0] ^= 1ULL << 63;
2212 ((uint64_t *)data
)[4] ^= 1ULL << 63;
2214 (void) ztest_write(zd
, object
, offset
, blocksize
, data
);
2217 case ZTEST_IO_WRITE_ZEROES
:
2218 bzero(data
, blocksize
);
2219 (void) ztest_write(zd
, object
, offset
, blocksize
, data
);
2222 case ZTEST_IO_TRUNCATE
:
2223 (void) ztest_truncate(zd
, object
, offset
, blocksize
);
2226 case ZTEST_IO_SETATTR
:
2227 (void) ztest_setattr(zd
, object
);
2232 case ZTEST_IO_REWRITE
:
2233 (void) rw_enter(&ztest_name_lock
, RW_READER
);
2234 err
= ztest_dsl_prop_set_uint64(zd
->zd_name
,
2235 ZFS_PROP_CHECKSUM
, spa_dedup_checksum(ztest_spa
),
2237 VERIFY(err
== 0 || err
== ENOSPC
);
2238 err
= ztest_dsl_prop_set_uint64(zd
->zd_name
,
2239 ZFS_PROP_COMPRESSION
,
2240 ztest_random_dsl_prop(ZFS_PROP_COMPRESSION
),
2242 VERIFY(err
== 0 || err
== ENOSPC
);
2243 (void) rw_exit(&ztest_name_lock
);
2245 VERIFY0(dmu_read(zd
->zd_os
, object
, offset
, blocksize
, data
,
2246 DMU_READ_NO_PREFETCH
));
2248 (void) ztest_write(zd
, object
, offset
, blocksize
, data
);
2252 (void) rw_exit(&zd
->zd_zilog_lock
);
2254 umem_free(data
, blocksize
);
2258 * Initialize an object description template.
2261 ztest_od_init(ztest_od_t
*od
, uint64_t id
, char *tag
, uint64_t index
,
2262 dmu_object_type_t type
, uint64_t blocksize
, uint64_t gen
)
2264 od
->od_dir
= ZTEST_DIROBJ
;
2267 od
->od_crtype
= type
;
2268 od
->od_crblocksize
= blocksize
? blocksize
: ztest_random_blocksize();
2271 od
->od_type
= DMU_OT_NONE
;
2272 od
->od_blocksize
= 0;
2275 (void) snprintf(od
->od_name
, sizeof (od
->od_name
), "%s(%lld)[%llu]",
2276 tag
, (longlong_t
)id
, (u_longlong_t
)index
);
2280 * Lookup or create the objects for a test using the od template.
2281 * If the objects do not all exist, or if 'remove' is specified,
2282 * remove any existing objects and create new ones. Otherwise,
2283 * use the existing objects.
2286 ztest_object_init(ztest_ds_t
*zd
, ztest_od_t
*od
, size_t size
, boolean_t remove
)
2288 int count
= size
/ sizeof (*od
);
2291 mutex_enter(&zd
->zd_dirobj_lock
);
2292 if ((ztest_lookup(zd
, od
, count
) != 0 || remove
) &&
2293 (ztest_remove(zd
, od
, count
) != 0 ||
2294 ztest_create(zd
, od
, count
) != 0))
2297 mutex_exit(&zd
->zd_dirobj_lock
);
2304 ztest_zil_commit(ztest_ds_t
*zd
, uint64_t id
)
2306 zilog_t
*zilog
= zd
->zd_zilog
;
2308 (void) rw_enter(&zd
->zd_zilog_lock
, RW_READER
);
2310 zil_commit(zilog
, ztest_random(ZTEST_OBJECTS
));
2313 * Remember the committed values in zd, which is in parent/child
2314 * shared memory. If we die, the next iteration of ztest_run()
2315 * will verify that the log really does contain this record.
2317 mutex_enter(&zilog
->zl_lock
);
2318 ASSERT(zd
->zd_shared
!= NULL
);
2319 ASSERT3U(zd
->zd_shared
->zd_seq
, <=, zilog
->zl_commit_lr_seq
);
2320 zd
->zd_shared
->zd_seq
= zilog
->zl_commit_lr_seq
;
2321 mutex_exit(&zilog
->zl_lock
);
2323 (void) rw_exit(&zd
->zd_zilog_lock
);
2327 * This function is designed to simulate the operations that occur during a
2328 * mount/unmount operation. We hold the dataset across these operations in an
2329 * attempt to expose any implicit assumptions about ZIL management.
2333 ztest_zil_remount(ztest_ds_t
*zd
, uint64_t id
)
2335 objset_t
*os
= zd
->zd_os
;
2338 * We grab the zd_dirobj_lock to ensure that no other thread is
2339 * updating the zil (i.e. adding in-memory log records) and the
2340 * zd_zilog_lock to block any I/O.
2342 mutex_enter(&zd
->zd_dirobj_lock
);
2343 (void) rw_enter(&zd
->zd_zilog_lock
, RW_WRITER
);
2345 /* zfs_sb_teardown() */
2346 zil_close(zd
->zd_zilog
);
2348 /* zfsvfs_setup() */
2349 VERIFY(zil_open(os
, ztest_get_data
) == zd
->zd_zilog
);
2350 zil_replay(os
, zd
, ztest_replay_vector
);
2352 (void) rw_exit(&zd
->zd_zilog_lock
);
2353 mutex_exit(&zd
->zd_dirobj_lock
);
2357 * Verify that we can't destroy an active pool, create an existing pool,
2358 * or create a pool with a bad vdev spec.
2362 ztest_spa_create_destroy(ztest_ds_t
*zd
, uint64_t id
)
2364 ztest_shared_opts_t
*zo
= &ztest_opts
;
2369 * Attempt to create using a bad file.
2371 nvroot
= make_vdev_root("/dev/bogus", NULL
, NULL
, 0, 0, 0, 0, 0, 1);
2372 VERIFY3U(ENOENT
, ==,
2373 spa_create("ztest_bad_file", nvroot
, NULL
, NULL
));
2374 nvlist_free(nvroot
);
2377 * Attempt to create using a bad mirror.
2379 nvroot
= make_vdev_root("/dev/bogus", NULL
, NULL
, 0, 0, 0, 0, 2, 1);
2380 VERIFY3U(ENOENT
, ==,
2381 spa_create("ztest_bad_mirror", nvroot
, NULL
, NULL
));
2382 nvlist_free(nvroot
);
2385 * Attempt to create an existing pool. It shouldn't matter
2386 * what's in the nvroot; we should fail with EEXIST.
2388 (void) rw_enter(&ztest_name_lock
, RW_READER
);
2389 nvroot
= make_vdev_root("/dev/bogus", NULL
, NULL
, 0, 0, 0, 0, 0, 1);
2390 VERIFY3U(EEXIST
, ==, spa_create(zo
->zo_pool
, nvroot
, NULL
, NULL
));
2391 nvlist_free(nvroot
);
2392 VERIFY3U(0, ==, spa_open(zo
->zo_pool
, &spa
, FTAG
));
2393 VERIFY3U(EBUSY
, ==, spa_destroy(zo
->zo_pool
));
2394 spa_close(spa
, FTAG
);
2396 (void) rw_exit(&ztest_name_lock
);
2401 ztest_spa_upgrade(ztest_ds_t
*zd
, uint64_t id
)
2404 uint64_t initial_version
= SPA_VERSION_INITIAL
;
2405 uint64_t version
, newversion
;
2406 nvlist_t
*nvroot
, *props
;
2409 mutex_enter(&ztest_vdev_lock
);
2410 name
= kmem_asprintf("%s_upgrade", ztest_opts
.zo_pool
);
2413 * Clean up from previous runs.
2415 (void) spa_destroy(name
);
2417 nvroot
= make_vdev_root(NULL
, NULL
, name
, ztest_opts
.zo_vdev_size
, 0,
2418 0, ztest_opts
.zo_raidz
, ztest_opts
.zo_mirrors
, 1);
2421 * If we're configuring a RAIDZ device then make sure that the
2422 * the initial version is capable of supporting that feature.
2424 switch (ztest_opts
.zo_raidz_parity
) {
2427 initial_version
= SPA_VERSION_INITIAL
;
2430 initial_version
= SPA_VERSION_RAIDZ2
;
2433 initial_version
= SPA_VERSION_RAIDZ3
;
2438 * Create a pool with a spa version that can be upgraded. Pick
2439 * a value between initial_version and SPA_VERSION_BEFORE_FEATURES.
2442 version
= ztest_random_spa_version(initial_version
);
2443 } while (version
> SPA_VERSION_BEFORE_FEATURES
);
2445 props
= fnvlist_alloc();
2446 fnvlist_add_uint64(props
,
2447 zpool_prop_to_name(ZPOOL_PROP_VERSION
), version
);
2448 VERIFY3S(spa_create(name
, nvroot
, props
, NULL
), ==, 0);
2449 fnvlist_free(nvroot
);
2450 fnvlist_free(props
);
2452 VERIFY3S(spa_open(name
, &spa
, FTAG
), ==, 0);
2453 VERIFY3U(spa_version(spa
), ==, version
);
2454 newversion
= ztest_random_spa_version(version
+ 1);
2456 if (ztest_opts
.zo_verbose
>= 4) {
2457 (void) printf("upgrading spa version from %llu to %llu\n",
2458 (u_longlong_t
)version
, (u_longlong_t
)newversion
);
2461 spa_upgrade(spa
, newversion
);
2462 VERIFY3U(spa_version(spa
), >, version
);
2463 VERIFY3U(spa_version(spa
), ==, fnvlist_lookup_uint64(spa
->spa_config
,
2464 zpool_prop_to_name(ZPOOL_PROP_VERSION
)));
2465 spa_close(spa
, FTAG
);
2468 mutex_exit(&ztest_vdev_lock
);
2472 vdev_lookup_by_path(vdev_t
*vd
, const char *path
)
2477 if (vd
->vdev_path
!= NULL
&& strcmp(path
, vd
->vdev_path
) == 0)
2480 for (c
= 0; c
< vd
->vdev_children
; c
++)
2481 if ((mvd
= vdev_lookup_by_path(vd
->vdev_child
[c
], path
)) !=
2489 * Find the first available hole which can be used as a top-level.
2492 find_vdev_hole(spa_t
*spa
)
2494 vdev_t
*rvd
= spa
->spa_root_vdev
;
2497 ASSERT(spa_config_held(spa
, SCL_VDEV
, RW_READER
) == SCL_VDEV
);
2499 for (c
= 0; c
< rvd
->vdev_children
; c
++) {
2500 vdev_t
*cvd
= rvd
->vdev_child
[c
];
2502 if (cvd
->vdev_ishole
)
2509 * Verify that vdev_add() works as expected.
2513 ztest_vdev_add_remove(ztest_ds_t
*zd
, uint64_t id
)
2515 ztest_shared_t
*zs
= ztest_shared
;
2516 spa_t
*spa
= ztest_spa
;
2522 mutex_enter(&ztest_vdev_lock
);
2523 leaves
= MAX(zs
->zs_mirrors
+ zs
->zs_splits
, 1) * ztest_opts
.zo_raidz
;
2525 spa_config_enter(spa
, SCL_VDEV
, FTAG
, RW_READER
);
2527 ztest_shared
->zs_vdev_next_leaf
= find_vdev_hole(spa
) * leaves
;
2530 * If we have slogs then remove them 1/4 of the time.
2532 if (spa_has_slogs(spa
) && ztest_random(4) == 0) {
2534 * Grab the guid from the head of the log class rotor.
2536 guid
= spa_log_class(spa
)->mc_rotor
->mg_vd
->vdev_guid
;
2538 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
2541 * We have to grab the zs_name_lock as writer to
2542 * prevent a race between removing a slog (dmu_objset_find)
2543 * and destroying a dataset. Removing the slog will
2544 * grab a reference on the dataset which may cause
2545 * dsl_destroy_head() to fail with EBUSY thus
2546 * leaving the dataset in an inconsistent state.
2548 rw_enter(&ztest_name_lock
, RW_WRITER
);
2549 error
= spa_vdev_remove(spa
, guid
, B_FALSE
);
2550 rw_exit(&ztest_name_lock
);
2552 if (error
&& error
!= EEXIST
)
2553 fatal(0, "spa_vdev_remove() = %d", error
);
2555 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
2558 * Make 1/4 of the devices be log devices.
2560 nvroot
= make_vdev_root(NULL
, NULL
, NULL
,
2561 ztest_opts
.zo_vdev_size
, 0,
2562 ztest_random(4) == 0, ztest_opts
.zo_raidz
,
2565 error
= spa_vdev_add(spa
, nvroot
);
2566 nvlist_free(nvroot
);
2568 if (error
== ENOSPC
)
2569 ztest_record_enospc("spa_vdev_add");
2570 else if (error
!= 0)
2571 fatal(0, "spa_vdev_add() = %d", error
);
2574 mutex_exit(&ztest_vdev_lock
);
2578 * Verify that adding/removing aux devices (l2arc, hot spare) works as expected.
2582 ztest_vdev_aux_add_remove(ztest_ds_t
*zd
, uint64_t id
)
2584 ztest_shared_t
*zs
= ztest_shared
;
2585 spa_t
*spa
= ztest_spa
;
2586 vdev_t
*rvd
= spa
->spa_root_vdev
;
2587 spa_aux_vdev_t
*sav
;
2593 path
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
2595 if (ztest_random(2) == 0) {
2596 sav
= &spa
->spa_spares
;
2597 aux
= ZPOOL_CONFIG_SPARES
;
2599 sav
= &spa
->spa_l2cache
;
2600 aux
= ZPOOL_CONFIG_L2CACHE
;
2603 mutex_enter(&ztest_vdev_lock
);
2605 spa_config_enter(spa
, SCL_VDEV
, FTAG
, RW_READER
);
2607 if (sav
->sav_count
!= 0 && ztest_random(4) == 0) {
2609 * Pick a random device to remove.
2611 guid
= sav
->sav_vdevs
[ztest_random(sav
->sav_count
)]->vdev_guid
;
2614 * Find an unused device we can add.
2616 zs
->zs_vdev_aux
= 0;
2619 (void) snprintf(path
, MAXPATHLEN
, ztest_aux_template
,
2620 ztest_opts
.zo_dir
, ztest_opts
.zo_pool
, aux
,
2622 for (c
= 0; c
< sav
->sav_count
; c
++)
2623 if (strcmp(sav
->sav_vdevs
[c
]->vdev_path
,
2626 if (c
== sav
->sav_count
&&
2627 vdev_lookup_by_path(rvd
, path
) == NULL
)
2633 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
2639 nvlist_t
*nvroot
= make_vdev_root(NULL
, aux
, NULL
,
2640 (ztest_opts
.zo_vdev_size
* 5) / 4, 0, 0, 0, 0, 1);
2641 error
= spa_vdev_add(spa
, nvroot
);
2643 fatal(0, "spa_vdev_add(%p) = %d", nvroot
, error
);
2644 nvlist_free(nvroot
);
2647 * Remove an existing device. Sometimes, dirty its
2648 * vdev state first to make sure we handle removal
2649 * of devices that have pending state changes.
2651 if (ztest_random(2) == 0)
2652 (void) vdev_online(spa
, guid
, 0, NULL
);
2654 error
= spa_vdev_remove(spa
, guid
, B_FALSE
);
2655 if (error
!= 0 && error
!= EBUSY
)
2656 fatal(0, "spa_vdev_remove(%llu) = %d", guid
, error
);
2659 mutex_exit(&ztest_vdev_lock
);
2661 umem_free(path
, MAXPATHLEN
);
2665 * split a pool if it has mirror tlvdevs
2669 ztest_split_pool(ztest_ds_t
*zd
, uint64_t id
)
2671 ztest_shared_t
*zs
= ztest_shared
;
2672 spa_t
*spa
= ztest_spa
;
2673 vdev_t
*rvd
= spa
->spa_root_vdev
;
2674 nvlist_t
*tree
, **child
, *config
, *split
, **schild
;
2675 uint_t c
, children
, schildren
= 0, lastlogid
= 0;
2678 mutex_enter(&ztest_vdev_lock
);
2680 /* ensure we have a useable config; mirrors of raidz aren't supported */
2681 if (zs
->zs_mirrors
< 3 || ztest_opts
.zo_raidz
> 1) {
2682 mutex_exit(&ztest_vdev_lock
);
2686 /* clean up the old pool, if any */
2687 (void) spa_destroy("splitp");
2689 spa_config_enter(spa
, SCL_VDEV
, FTAG
, RW_READER
);
2691 /* generate a config from the existing config */
2692 mutex_enter(&spa
->spa_props_lock
);
2693 VERIFY(nvlist_lookup_nvlist(spa
->spa_config
, ZPOOL_CONFIG_VDEV_TREE
,
2695 mutex_exit(&spa
->spa_props_lock
);
2697 VERIFY(nvlist_lookup_nvlist_array(tree
, ZPOOL_CONFIG_CHILDREN
, &child
,
2700 schild
= malloc(rvd
->vdev_children
* sizeof (nvlist_t
*));
2701 for (c
= 0; c
< children
; c
++) {
2702 vdev_t
*tvd
= rvd
->vdev_child
[c
];
2706 if (tvd
->vdev_islog
|| tvd
->vdev_ops
== &vdev_hole_ops
) {
2707 VERIFY(nvlist_alloc(&schild
[schildren
], NV_UNIQUE_NAME
,
2709 VERIFY(nvlist_add_string(schild
[schildren
],
2710 ZPOOL_CONFIG_TYPE
, VDEV_TYPE_HOLE
) == 0);
2711 VERIFY(nvlist_add_uint64(schild
[schildren
],
2712 ZPOOL_CONFIG_IS_HOLE
, 1) == 0);
2714 lastlogid
= schildren
;
2719 VERIFY(nvlist_lookup_nvlist_array(child
[c
],
2720 ZPOOL_CONFIG_CHILDREN
, &mchild
, &mchildren
) == 0);
2721 VERIFY(nvlist_dup(mchild
[0], &schild
[schildren
++], 0) == 0);
2724 /* OK, create a config that can be used to split */
2725 VERIFY(nvlist_alloc(&split
, NV_UNIQUE_NAME
, 0) == 0);
2726 VERIFY(nvlist_add_string(split
, ZPOOL_CONFIG_TYPE
,
2727 VDEV_TYPE_ROOT
) == 0);
2728 VERIFY(nvlist_add_nvlist_array(split
, ZPOOL_CONFIG_CHILDREN
, schild
,
2729 lastlogid
!= 0 ? lastlogid
: schildren
) == 0);
2731 VERIFY(nvlist_alloc(&config
, NV_UNIQUE_NAME
, 0) == 0);
2732 VERIFY(nvlist_add_nvlist(config
, ZPOOL_CONFIG_VDEV_TREE
, split
) == 0);
2734 for (c
= 0; c
< schildren
; c
++)
2735 nvlist_free(schild
[c
]);
2739 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
2741 (void) rw_enter(&ztest_name_lock
, RW_WRITER
);
2742 error
= spa_vdev_split_mirror(spa
, "splitp", config
, NULL
, B_FALSE
);
2743 (void) rw_exit(&ztest_name_lock
);
2745 nvlist_free(config
);
2748 (void) printf("successful split - results:\n");
2749 mutex_enter(&spa_namespace_lock
);
2750 show_pool_stats(spa
);
2751 show_pool_stats(spa_lookup("splitp"));
2752 mutex_exit(&spa_namespace_lock
);
2756 mutex_exit(&ztest_vdev_lock
);
2761 * Verify that we can attach and detach devices.
2765 ztest_vdev_attach_detach(ztest_ds_t
*zd
, uint64_t id
)
2767 ztest_shared_t
*zs
= ztest_shared
;
2768 spa_t
*spa
= ztest_spa
;
2769 spa_aux_vdev_t
*sav
= &spa
->spa_spares
;
2770 vdev_t
*rvd
= spa
->spa_root_vdev
;
2771 vdev_t
*oldvd
, *newvd
, *pvd
;
2775 uint64_t ashift
= ztest_get_ashift();
2776 uint64_t oldguid
, pguid
;
2777 size_t oldsize
, newsize
;
2778 char *oldpath
, *newpath
;
2780 int oldvd_has_siblings
= B_FALSE
;
2781 int newvd_is_spare
= B_FALSE
;
2783 int error
, expected_error
;
2785 oldpath
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
2786 newpath
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
2788 mutex_enter(&ztest_vdev_lock
);
2789 leaves
= MAX(zs
->zs_mirrors
, 1) * ztest_opts
.zo_raidz
;
2791 spa_config_enter(spa
, SCL_VDEV
, FTAG
, RW_READER
);
2794 * Decide whether to do an attach or a replace.
2796 replacing
= ztest_random(2);
2799 * Pick a random top-level vdev.
2801 top
= ztest_random_vdev_top(spa
, B_TRUE
);
2804 * Pick a random leaf within it.
2806 leaf
= ztest_random(leaves
);
2811 oldvd
= rvd
->vdev_child
[top
];
2812 if (zs
->zs_mirrors
>= 1) {
2813 ASSERT(oldvd
->vdev_ops
== &vdev_mirror_ops
);
2814 ASSERT(oldvd
->vdev_children
>= zs
->zs_mirrors
);
2815 oldvd
= oldvd
->vdev_child
[leaf
/ ztest_opts
.zo_raidz
];
2817 if (ztest_opts
.zo_raidz
> 1) {
2818 ASSERT(oldvd
->vdev_ops
== &vdev_raidz_ops
);
2819 ASSERT(oldvd
->vdev_children
== ztest_opts
.zo_raidz
);
2820 oldvd
= oldvd
->vdev_child
[leaf
% ztest_opts
.zo_raidz
];
2824 * If we're already doing an attach or replace, oldvd may be a
2825 * mirror vdev -- in which case, pick a random child.
2827 while (oldvd
->vdev_children
!= 0) {
2828 oldvd_has_siblings
= B_TRUE
;
2829 ASSERT(oldvd
->vdev_children
>= 2);
2830 oldvd
= oldvd
->vdev_child
[ztest_random(oldvd
->vdev_children
)];
2833 oldguid
= oldvd
->vdev_guid
;
2834 oldsize
= vdev_get_min_asize(oldvd
);
2835 oldvd_is_log
= oldvd
->vdev_top
->vdev_islog
;
2836 (void) strcpy(oldpath
, oldvd
->vdev_path
);
2837 pvd
= oldvd
->vdev_parent
;
2838 pguid
= pvd
->vdev_guid
;
2841 * If oldvd has siblings, then half of the time, detach it.
2843 if (oldvd_has_siblings
&& ztest_random(2) == 0) {
2844 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
2845 error
= spa_vdev_detach(spa
, oldguid
, pguid
, B_FALSE
);
2846 if (error
!= 0 && error
!= ENODEV
&& error
!= EBUSY
&&
2848 fatal(0, "detach (%s) returned %d", oldpath
, error
);
2853 * For the new vdev, choose with equal probability between the two
2854 * standard paths (ending in either 'a' or 'b') or a random hot spare.
2856 if (sav
->sav_count
!= 0 && ztest_random(3) == 0) {
2857 newvd
= sav
->sav_vdevs
[ztest_random(sav
->sav_count
)];
2858 newvd_is_spare
= B_TRUE
;
2859 (void) strcpy(newpath
, newvd
->vdev_path
);
2861 (void) snprintf(newpath
, MAXPATHLEN
, ztest_dev_template
,
2862 ztest_opts
.zo_dir
, ztest_opts
.zo_pool
,
2863 top
* leaves
+ leaf
);
2864 if (ztest_random(2) == 0)
2865 newpath
[strlen(newpath
) - 1] = 'b';
2866 newvd
= vdev_lookup_by_path(rvd
, newpath
);
2870 newsize
= vdev_get_min_asize(newvd
);
2873 * Make newsize a little bigger or smaller than oldsize.
2874 * If it's smaller, the attach should fail.
2875 * If it's larger, and we're doing a replace,
2876 * we should get dynamic LUN growth when we're done.
2878 newsize
= 10 * oldsize
/ (9 + ztest_random(3));
2882 * If pvd is not a mirror or root, the attach should fail with ENOTSUP,
2883 * unless it's a replace; in that case any non-replacing parent is OK.
2885 * If newvd is already part of the pool, it should fail with EBUSY.
2887 * If newvd is too small, it should fail with EOVERFLOW.
2889 if (pvd
->vdev_ops
!= &vdev_mirror_ops
&&
2890 pvd
->vdev_ops
!= &vdev_root_ops
&& (!replacing
||
2891 pvd
->vdev_ops
== &vdev_replacing_ops
||
2892 pvd
->vdev_ops
== &vdev_spare_ops
))
2893 expected_error
= ENOTSUP
;
2894 else if (newvd_is_spare
&& (!replacing
|| oldvd_is_log
))
2895 expected_error
= ENOTSUP
;
2896 else if (newvd
== oldvd
)
2897 expected_error
= replacing
? 0 : EBUSY
;
2898 else if (vdev_lookup_by_path(rvd
, newpath
) != NULL
)
2899 expected_error
= EBUSY
;
2900 else if (newsize
< oldsize
)
2901 expected_error
= EOVERFLOW
;
2902 else if (ashift
> oldvd
->vdev_top
->vdev_ashift
)
2903 expected_error
= EDOM
;
2907 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
2910 * Build the nvlist describing newpath.
2912 root
= make_vdev_root(newpath
, NULL
, NULL
, newvd
== NULL
? newsize
: 0,
2913 ashift
, 0, 0, 0, 1);
2915 error
= spa_vdev_attach(spa
, oldguid
, root
, replacing
);
2920 * If our parent was the replacing vdev, but the replace completed,
2921 * then instead of failing with ENOTSUP we may either succeed,
2922 * fail with ENODEV, or fail with EOVERFLOW.
2924 if (expected_error
== ENOTSUP
&&
2925 (error
== 0 || error
== ENODEV
|| error
== EOVERFLOW
))
2926 expected_error
= error
;
2929 * If someone grew the LUN, the replacement may be too small.
2931 if (error
== EOVERFLOW
|| error
== EBUSY
)
2932 expected_error
= error
;
2934 /* XXX workaround 6690467 */
2935 if (error
!= expected_error
&& expected_error
!= EBUSY
) {
2936 fatal(0, "attach (%s %llu, %s %llu, %d) "
2937 "returned %d, expected %d",
2938 oldpath
, (longlong_t
)oldsize
, newpath
,
2939 (longlong_t
)newsize
, replacing
, error
, expected_error
);
2942 mutex_exit(&ztest_vdev_lock
);
2944 umem_free(oldpath
, MAXPATHLEN
);
2945 umem_free(newpath
, MAXPATHLEN
);
2949 * Callback function which expands the physical size of the vdev.
2952 grow_vdev(vdev_t
*vd
, void *arg
)
2954 ASSERTV(spa_t
*spa
= vd
->vdev_spa
);
2955 size_t *newsize
= arg
;
2959 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_READER
) == SCL_STATE
);
2960 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
2962 if ((fd
= open(vd
->vdev_path
, O_RDWR
)) == -1)
2965 fsize
= lseek(fd
, 0, SEEK_END
);
2966 VERIFY(ftruncate(fd
, *newsize
) == 0);
2968 if (ztest_opts
.zo_verbose
>= 6) {
2969 (void) printf("%s grew from %lu to %lu bytes\n",
2970 vd
->vdev_path
, (ulong_t
)fsize
, (ulong_t
)*newsize
);
2977 * Callback function which expands a given vdev by calling vdev_online().
2981 online_vdev(vdev_t
*vd
, void *arg
)
2983 spa_t
*spa
= vd
->vdev_spa
;
2984 vdev_t
*tvd
= vd
->vdev_top
;
2985 uint64_t guid
= vd
->vdev_guid
;
2986 uint64_t generation
= spa
->spa_config_generation
+ 1;
2987 vdev_state_t newstate
= VDEV_STATE_UNKNOWN
;
2990 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_READER
) == SCL_STATE
);
2991 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
2993 /* Calling vdev_online will initialize the new metaslabs */
2994 spa_config_exit(spa
, SCL_STATE
, spa
);
2995 error
= vdev_online(spa
, guid
, ZFS_ONLINE_EXPAND
, &newstate
);
2996 spa_config_enter(spa
, SCL_STATE
, spa
, RW_READER
);
2999 * If vdev_online returned an error or the underlying vdev_open
3000 * failed then we abort the expand. The only way to know that
3001 * vdev_open fails is by checking the returned newstate.
3003 if (error
|| newstate
!= VDEV_STATE_HEALTHY
) {
3004 if (ztest_opts
.zo_verbose
>= 5) {
3005 (void) printf("Unable to expand vdev, state %llu, "
3006 "error %d\n", (u_longlong_t
)newstate
, error
);
3010 ASSERT3U(newstate
, ==, VDEV_STATE_HEALTHY
);
3013 * Since we dropped the lock we need to ensure that we're
3014 * still talking to the original vdev. It's possible this
3015 * vdev may have been detached/replaced while we were
3016 * trying to online it.
3018 if (generation
!= spa
->spa_config_generation
) {
3019 if (ztest_opts
.zo_verbose
>= 5) {
3020 (void) printf("vdev configuration has changed, "
3021 "guid %llu, state %llu, expected gen %llu, "
3024 (u_longlong_t
)tvd
->vdev_state
,
3025 (u_longlong_t
)generation
,
3026 (u_longlong_t
)spa
->spa_config_generation
);
3034 * Traverse the vdev tree calling the supplied function.
3035 * We continue to walk the tree until we either have walked all
3036 * children or we receive a non-NULL return from the callback.
3037 * If a NULL callback is passed, then we just return back the first
3038 * leaf vdev we encounter.
3041 vdev_walk_tree(vdev_t
*vd
, vdev_t
*(*func
)(vdev_t
*, void *), void *arg
)
3045 if (vd
->vdev_ops
->vdev_op_leaf
) {
3049 return (func(vd
, arg
));
3052 for (c
= 0; c
< vd
->vdev_children
; c
++) {
3053 vdev_t
*cvd
= vd
->vdev_child
[c
];
3054 if ((cvd
= vdev_walk_tree(cvd
, func
, arg
)) != NULL
)
3061 * Verify that dynamic LUN growth works as expected.
3065 ztest_vdev_LUN_growth(ztest_ds_t
*zd
, uint64_t id
)
3067 spa_t
*spa
= ztest_spa
;
3069 metaslab_class_t
*mc
;
3070 metaslab_group_t
*mg
;
3071 size_t psize
, newsize
;
3073 uint64_t old_class_space
, new_class_space
, old_ms_count
, new_ms_count
;
3075 mutex_enter(&ztest_vdev_lock
);
3076 spa_config_enter(spa
, SCL_STATE
, spa
, RW_READER
);
3078 top
= ztest_random_vdev_top(spa
, B_TRUE
);
3080 tvd
= spa
->spa_root_vdev
->vdev_child
[top
];
3083 old_ms_count
= tvd
->vdev_ms_count
;
3084 old_class_space
= metaslab_class_get_space(mc
);
3087 * Determine the size of the first leaf vdev associated with
3088 * our top-level device.
3090 vd
= vdev_walk_tree(tvd
, NULL
, NULL
);
3091 ASSERT3P(vd
, !=, NULL
);
3092 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
3094 psize
= vd
->vdev_psize
;
3097 * We only try to expand the vdev if it's healthy, less than 4x its
3098 * original size, and it has a valid psize.
3100 if (tvd
->vdev_state
!= VDEV_STATE_HEALTHY
||
3101 psize
== 0 || psize
>= 4 * ztest_opts
.zo_vdev_size
) {
3102 spa_config_exit(spa
, SCL_STATE
, spa
);
3103 mutex_exit(&ztest_vdev_lock
);
3107 newsize
= psize
+ psize
/ 8;
3108 ASSERT3U(newsize
, >, psize
);
3110 if (ztest_opts
.zo_verbose
>= 6) {
3111 (void) printf("Expanding LUN %s from %lu to %lu\n",
3112 vd
->vdev_path
, (ulong_t
)psize
, (ulong_t
)newsize
);
3116 * Growing the vdev is a two step process:
3117 * 1). expand the physical size (i.e. relabel)
3118 * 2). online the vdev to create the new metaslabs
3120 if (vdev_walk_tree(tvd
, grow_vdev
, &newsize
) != NULL
||
3121 vdev_walk_tree(tvd
, online_vdev
, NULL
) != NULL
||
3122 tvd
->vdev_state
!= VDEV_STATE_HEALTHY
) {
3123 if (ztest_opts
.zo_verbose
>= 5) {
3124 (void) printf("Could not expand LUN because "
3125 "the vdev configuration changed.\n");
3127 spa_config_exit(spa
, SCL_STATE
, spa
);
3128 mutex_exit(&ztest_vdev_lock
);
3132 spa_config_exit(spa
, SCL_STATE
, spa
);
3135 * Expanding the LUN will update the config asynchronously,
3136 * thus we must wait for the async thread to complete any
3137 * pending tasks before proceeding.
3141 mutex_enter(&spa
->spa_async_lock
);
3142 done
= (spa
->spa_async_thread
== NULL
&& !spa
->spa_async_tasks
);
3143 mutex_exit(&spa
->spa_async_lock
);
3146 txg_wait_synced(spa_get_dsl(spa
), 0);
3147 (void) poll(NULL
, 0, 100);
3150 spa_config_enter(spa
, SCL_STATE
, spa
, RW_READER
);
3152 tvd
= spa
->spa_root_vdev
->vdev_child
[top
];
3153 new_ms_count
= tvd
->vdev_ms_count
;
3154 new_class_space
= metaslab_class_get_space(mc
);
3156 if (tvd
->vdev_mg
!= mg
|| mg
->mg_class
!= mc
) {
3157 if (ztest_opts
.zo_verbose
>= 5) {
3158 (void) printf("Could not verify LUN expansion due to "
3159 "intervening vdev offline or remove.\n");
3161 spa_config_exit(spa
, SCL_STATE
, spa
);
3162 mutex_exit(&ztest_vdev_lock
);
3167 * Make sure we were able to grow the vdev.
3169 if (new_ms_count
<= old_ms_count
)
3170 fatal(0, "LUN expansion failed: ms_count %llu <= %llu\n",
3171 old_ms_count
, new_ms_count
);
3174 * Make sure we were able to grow the pool.
3176 if (new_class_space
<= old_class_space
)
3177 fatal(0, "LUN expansion failed: class_space %llu <= %llu\n",
3178 old_class_space
, new_class_space
);
3180 if (ztest_opts
.zo_verbose
>= 5) {
3181 char oldnumbuf
[6], newnumbuf
[6];
3183 nicenum(old_class_space
, oldnumbuf
);
3184 nicenum(new_class_space
, newnumbuf
);
3185 (void) printf("%s grew from %s to %s\n",
3186 spa
->spa_name
, oldnumbuf
, newnumbuf
);
3189 spa_config_exit(spa
, SCL_STATE
, spa
);
3190 mutex_exit(&ztest_vdev_lock
);
3194 * Verify that dmu_objset_{create,destroy,open,close} work as expected.
3198 ztest_objset_create_cb(objset_t
*os
, void *arg
, cred_t
*cr
, dmu_tx_t
*tx
)
3201 * Create the objects common to all ztest datasets.
3203 VERIFY(zap_create_claim(os
, ZTEST_DIROBJ
,
3204 DMU_OT_ZAP_OTHER
, DMU_OT_NONE
, 0, tx
) == 0);
3208 ztest_dataset_create(char *dsname
)
3210 uint64_t zilset
= ztest_random(100);
3211 int err
= dmu_objset_create(dsname
, DMU_OST_OTHER
, 0,
3212 ztest_objset_create_cb
, NULL
);
3214 if (err
|| zilset
< 80)
3217 if (ztest_opts
.zo_verbose
>= 5)
3218 (void) printf("Setting dataset %s to sync always\n", dsname
);
3219 return (ztest_dsl_prop_set_uint64(dsname
, ZFS_PROP_SYNC
,
3220 ZFS_SYNC_ALWAYS
, B_FALSE
));
3225 ztest_objset_destroy_cb(const char *name
, void *arg
)
3228 dmu_object_info_t doi
;
3232 * Verify that the dataset contains a directory object.
3234 VERIFY0(dmu_objset_own(name
, DMU_OST_OTHER
, B_TRUE
, FTAG
, &os
));
3235 error
= dmu_object_info(os
, ZTEST_DIROBJ
, &doi
);
3236 if (error
!= ENOENT
) {
3237 /* We could have crashed in the middle of destroying it */
3239 ASSERT3U(doi
.doi_type
, ==, DMU_OT_ZAP_OTHER
);
3240 ASSERT3S(doi
.doi_physical_blocks_512
, >=, 0);
3242 dmu_objset_disown(os
, FTAG
);
3245 * Destroy the dataset.
3247 if (strchr(name
, '@') != NULL
) {
3248 VERIFY0(dsl_destroy_snapshot(name
, B_FALSE
));
3250 VERIFY0(dsl_destroy_head(name
));
3256 ztest_snapshot_create(char *osname
, uint64_t id
)
3258 char snapname
[MAXNAMELEN
];
3261 (void) snprintf(snapname
, sizeof (snapname
), "%llu", (u_longlong_t
)id
);
3263 error
= dmu_objset_snapshot_one(osname
, snapname
);
3264 if (error
== ENOSPC
) {
3265 ztest_record_enospc(FTAG
);
3268 if (error
!= 0 && error
!= EEXIST
) {
3269 fatal(0, "ztest_snapshot_create(%s@%s) = %d", osname
,
3276 ztest_snapshot_destroy(char *osname
, uint64_t id
)
3278 char snapname
[MAXNAMELEN
];
3281 (void) snprintf(snapname
, MAXNAMELEN
, "%s@%llu", osname
,
3284 error
= dsl_destroy_snapshot(snapname
, B_FALSE
);
3285 if (error
!= 0 && error
!= ENOENT
)
3286 fatal(0, "ztest_snapshot_destroy(%s) = %d", snapname
, error
);
3292 ztest_dmu_objset_create_destroy(ztest_ds_t
*zd
, uint64_t id
)
3302 zdtmp
= umem_alloc(sizeof (ztest_ds_t
), UMEM_NOFAIL
);
3303 name
= umem_alloc(MAXNAMELEN
, UMEM_NOFAIL
);
3305 (void) rw_enter(&ztest_name_lock
, RW_READER
);
3307 (void) snprintf(name
, MAXNAMELEN
, "%s/temp_%llu",
3308 ztest_opts
.zo_pool
, (u_longlong_t
)id
);
3311 * If this dataset exists from a previous run, process its replay log
3312 * half of the time. If we don't replay it, then dsl_destroy_head()
3313 * (invoked from ztest_objset_destroy_cb()) should just throw it away.
3315 if (ztest_random(2) == 0 &&
3316 dmu_objset_own(name
, DMU_OST_OTHER
, B_FALSE
, FTAG
, &os
) == 0) {
3317 ztest_zd_init(zdtmp
, NULL
, os
);
3318 zil_replay(os
, zdtmp
, ztest_replay_vector
);
3319 ztest_zd_fini(zdtmp
);
3320 dmu_objset_disown(os
, FTAG
);
3324 * There may be an old instance of the dataset we're about to
3325 * create lying around from a previous run. If so, destroy it
3326 * and all of its snapshots.
3328 (void) dmu_objset_find(name
, ztest_objset_destroy_cb
, NULL
,
3329 DS_FIND_CHILDREN
| DS_FIND_SNAPSHOTS
);
3332 * Verify that the destroyed dataset is no longer in the namespace.
3334 VERIFY3U(ENOENT
, ==, dmu_objset_own(name
, DMU_OST_OTHER
, B_TRUE
,
3338 * Verify that we can create a new dataset.
3340 error
= ztest_dataset_create(name
);
3342 if (error
== ENOSPC
) {
3343 ztest_record_enospc(FTAG
);
3346 fatal(0, "dmu_objset_create(%s) = %d", name
, error
);
3349 VERIFY0(dmu_objset_own(name
, DMU_OST_OTHER
, B_FALSE
, FTAG
, &os
));
3351 ztest_zd_init(zdtmp
, NULL
, os
);
3354 * Open the intent log for it.
3356 zilog
= zil_open(os
, ztest_get_data
);
3359 * Put some objects in there, do a little I/O to them,
3360 * and randomly take a couple of snapshots along the way.
3362 iters
= ztest_random(5);
3363 for (i
= 0; i
< iters
; i
++) {
3364 ztest_dmu_object_alloc_free(zdtmp
, id
);
3365 if (ztest_random(iters
) == 0)
3366 (void) ztest_snapshot_create(name
, i
);
3370 * Verify that we cannot create an existing dataset.
3372 VERIFY3U(EEXIST
, ==,
3373 dmu_objset_create(name
, DMU_OST_OTHER
, 0, NULL
, NULL
));
3376 * Verify that we can hold an objset that is also owned.
3378 VERIFY3U(0, ==, dmu_objset_hold(name
, FTAG
, &os2
));
3379 dmu_objset_rele(os2
, FTAG
);
3382 * Verify that we cannot own an objset that is already owned.
3385 dmu_objset_own(name
, DMU_OST_OTHER
, B_FALSE
, FTAG
, &os2
));
3388 dmu_objset_disown(os
, FTAG
);
3389 ztest_zd_fini(zdtmp
);
3391 (void) rw_exit(&ztest_name_lock
);
3393 umem_free(name
, MAXNAMELEN
);
3394 umem_free(zdtmp
, sizeof (ztest_ds_t
));
3398 * Verify that dmu_snapshot_{create,destroy,open,close} work as expected.
3401 ztest_dmu_snapshot_create_destroy(ztest_ds_t
*zd
, uint64_t id
)
3403 (void) rw_enter(&ztest_name_lock
, RW_READER
);
3404 (void) ztest_snapshot_destroy(zd
->zd_name
, id
);
3405 (void) ztest_snapshot_create(zd
->zd_name
, id
);
3406 (void) rw_exit(&ztest_name_lock
);
3410 * Cleanup non-standard snapshots and clones.
3413 ztest_dsl_dataset_cleanup(char *osname
, uint64_t id
)
3422 snap1name
= umem_alloc(MAXNAMELEN
, UMEM_NOFAIL
);
3423 clone1name
= umem_alloc(MAXNAMELEN
, UMEM_NOFAIL
);
3424 snap2name
= umem_alloc(MAXNAMELEN
, UMEM_NOFAIL
);
3425 clone2name
= umem_alloc(MAXNAMELEN
, UMEM_NOFAIL
);
3426 snap3name
= umem_alloc(MAXNAMELEN
, UMEM_NOFAIL
);
3428 (void) snprintf(snap1name
, MAXNAMELEN
, "%s@s1_%llu",
3429 osname
, (u_longlong_t
)id
);
3430 (void) snprintf(clone1name
, MAXNAMELEN
, "%s/c1_%llu",
3431 osname
, (u_longlong_t
)id
);
3432 (void) snprintf(snap2name
, MAXNAMELEN
, "%s@s2_%llu",
3433 clone1name
, (u_longlong_t
)id
);
3434 (void) snprintf(clone2name
, MAXNAMELEN
, "%s/c2_%llu",
3435 osname
, (u_longlong_t
)id
);
3436 (void) snprintf(snap3name
, MAXNAMELEN
, "%s@s3_%llu",
3437 clone1name
, (u_longlong_t
)id
);
3439 error
= dsl_destroy_head(clone2name
);
3440 if (error
&& error
!= ENOENT
)
3441 fatal(0, "dsl_destroy_head(%s) = %d", clone2name
, error
);
3442 error
= dsl_destroy_snapshot(snap3name
, B_FALSE
);
3443 if (error
&& error
!= ENOENT
)
3444 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap3name
, error
);
3445 error
= dsl_destroy_snapshot(snap2name
, B_FALSE
);
3446 if (error
&& error
!= ENOENT
)
3447 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap2name
, error
);
3448 error
= dsl_destroy_head(clone1name
);
3449 if (error
&& error
!= ENOENT
)
3450 fatal(0, "dsl_destroy_head(%s) = %d", clone1name
, error
);
3451 error
= dsl_destroy_snapshot(snap1name
, B_FALSE
);
3452 if (error
&& error
!= ENOENT
)
3453 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap1name
, error
);
3455 umem_free(snap1name
, MAXNAMELEN
);
3456 umem_free(clone1name
, MAXNAMELEN
);
3457 umem_free(snap2name
, MAXNAMELEN
);
3458 umem_free(clone2name
, MAXNAMELEN
);
3459 umem_free(snap3name
, MAXNAMELEN
);
3463 * Verify dsl_dataset_promote handles EBUSY
3466 ztest_dsl_dataset_promote_busy(ztest_ds_t
*zd
, uint64_t id
)
3474 char *osname
= zd
->zd_name
;
3477 snap1name
= umem_alloc(MAXNAMELEN
, UMEM_NOFAIL
);
3478 clone1name
= umem_alloc(MAXNAMELEN
, UMEM_NOFAIL
);
3479 snap2name
= umem_alloc(MAXNAMELEN
, UMEM_NOFAIL
);
3480 clone2name
= umem_alloc(MAXNAMELEN
, UMEM_NOFAIL
);
3481 snap3name
= umem_alloc(MAXNAMELEN
, UMEM_NOFAIL
);
3483 (void) rw_enter(&ztest_name_lock
, RW_READER
);
3485 ztest_dsl_dataset_cleanup(osname
, id
);
3487 (void) snprintf(snap1name
, MAXNAMELEN
, "%s@s1_%llu",
3488 osname
, (u_longlong_t
)id
);
3489 (void) snprintf(clone1name
, MAXNAMELEN
, "%s/c1_%llu",
3490 osname
, (u_longlong_t
)id
);
3491 (void) snprintf(snap2name
, MAXNAMELEN
, "%s@s2_%llu",
3492 clone1name
, (u_longlong_t
)id
);
3493 (void) snprintf(clone2name
, MAXNAMELEN
, "%s/c2_%llu",
3494 osname
, (u_longlong_t
)id
);
3495 (void) snprintf(snap3name
, MAXNAMELEN
, "%s@s3_%llu",
3496 clone1name
, (u_longlong_t
)id
);
3498 error
= dmu_objset_snapshot_one(osname
, strchr(snap1name
, '@') + 1);
3499 if (error
&& error
!= EEXIST
) {
3500 if (error
== ENOSPC
) {
3501 ztest_record_enospc(FTAG
);
3504 fatal(0, "dmu_take_snapshot(%s) = %d", snap1name
, error
);
3507 error
= dmu_objset_clone(clone1name
, snap1name
);
3509 if (error
== ENOSPC
) {
3510 ztest_record_enospc(FTAG
);
3513 fatal(0, "dmu_objset_create(%s) = %d", clone1name
, error
);
3516 error
= dmu_objset_snapshot_one(clone1name
, strchr(snap2name
, '@') + 1);
3517 if (error
&& error
!= EEXIST
) {
3518 if (error
== ENOSPC
) {
3519 ztest_record_enospc(FTAG
);
3522 fatal(0, "dmu_open_snapshot(%s) = %d", snap2name
, error
);
3525 error
= dmu_objset_snapshot_one(clone1name
, strchr(snap3name
, '@') + 1);
3526 if (error
&& error
!= EEXIST
) {
3527 if (error
== ENOSPC
) {
3528 ztest_record_enospc(FTAG
);
3531 fatal(0, "dmu_open_snapshot(%s) = %d", snap3name
, error
);
3534 error
= dmu_objset_clone(clone2name
, snap3name
);
3536 if (error
== ENOSPC
) {
3537 ztest_record_enospc(FTAG
);
3540 fatal(0, "dmu_objset_create(%s) = %d", clone2name
, error
);
3543 error
= dmu_objset_own(snap2name
, DMU_OST_ANY
, B_TRUE
, FTAG
, &os
);
3545 fatal(0, "dmu_objset_own(%s) = %d", snap2name
, error
);
3546 error
= dsl_dataset_promote(clone2name
, NULL
);
3548 fatal(0, "dsl_dataset_promote(%s), %d, not EBUSY", clone2name
,
3550 dmu_objset_disown(os
, FTAG
);
3553 ztest_dsl_dataset_cleanup(osname
, id
);
3555 (void) rw_exit(&ztest_name_lock
);
3557 umem_free(snap1name
, MAXNAMELEN
);
3558 umem_free(clone1name
, MAXNAMELEN
);
3559 umem_free(snap2name
, MAXNAMELEN
);
3560 umem_free(clone2name
, MAXNAMELEN
);
3561 umem_free(snap3name
, MAXNAMELEN
);
3564 #undef OD_ARRAY_SIZE
3565 #define OD_ARRAY_SIZE 4
3568 * Verify that dmu_object_{alloc,free} work as expected.
3571 ztest_dmu_object_alloc_free(ztest_ds_t
*zd
, uint64_t id
)
3578 size
= sizeof(ztest_od_t
) * OD_ARRAY_SIZE
;
3579 od
= umem_alloc(size
, UMEM_NOFAIL
);
3580 batchsize
= OD_ARRAY_SIZE
;
3582 for (b
= 0; b
< batchsize
; b
++)
3583 ztest_od_init(od
+ b
, id
, FTAG
, b
, DMU_OT_UINT64_OTHER
, 0, 0);
3586 * Destroy the previous batch of objects, create a new batch,
3587 * and do some I/O on the new objects.
3589 if (ztest_object_init(zd
, od
, size
, B_TRUE
) != 0)
3592 while (ztest_random(4 * batchsize
) != 0)
3593 ztest_io(zd
, od
[ztest_random(batchsize
)].od_object
,
3594 ztest_random(ZTEST_RANGE_LOCKS
) << SPA_MAXBLOCKSHIFT
);
3596 umem_free(od
, size
);
3599 #undef OD_ARRAY_SIZE
3600 #define OD_ARRAY_SIZE 2
3603 * Verify that dmu_{read,write} work as expected.
3606 ztest_dmu_read_write(ztest_ds_t
*zd
, uint64_t id
)
3611 objset_t
*os
= zd
->zd_os
;
3612 size
= sizeof(ztest_od_t
) * OD_ARRAY_SIZE
;
3613 od
= umem_alloc(size
, UMEM_NOFAIL
);
3615 int i
, freeit
, error
;
3617 bufwad_t
*packbuf
, *bigbuf
, *pack
, *bigH
, *bigT
;
3618 uint64_t packobj
, packoff
, packsize
, bigobj
, bigoff
, bigsize
;
3619 uint64_t chunksize
= (1000 + ztest_random(1000)) * sizeof (uint64_t);
3620 uint64_t regions
= 997;
3621 uint64_t stride
= 123456789ULL;
3622 uint64_t width
= 40;
3623 int free_percent
= 5;
3626 * This test uses two objects, packobj and bigobj, that are always
3627 * updated together (i.e. in the same tx) so that their contents are
3628 * in sync and can be compared. Their contents relate to each other
3629 * in a simple way: packobj is a dense array of 'bufwad' structures,
3630 * while bigobj is a sparse array of the same bufwads. Specifically,
3631 * for any index n, there are three bufwads that should be identical:
3633 * packobj, at offset n * sizeof (bufwad_t)
3634 * bigobj, at the head of the nth chunk
3635 * bigobj, at the tail of the nth chunk
3637 * The chunk size is arbitrary. It doesn't have to be a power of two,
3638 * and it doesn't have any relation to the object blocksize.
3639 * The only requirement is that it can hold at least two bufwads.
3641 * Normally, we write the bufwad to each of these locations.
3642 * However, free_percent of the time we instead write zeroes to
3643 * packobj and perform a dmu_free_range() on bigobj. By comparing
3644 * bigobj to packobj, we can verify that the DMU is correctly
3645 * tracking which parts of an object are allocated and free,
3646 * and that the contents of the allocated blocks are correct.
3650 * Read the directory info. If it's the first time, set things up.
3652 ztest_od_init(od
, id
, FTAG
, 0, DMU_OT_UINT64_OTHER
, 0, chunksize
);
3653 ztest_od_init(od
+ 1, id
, FTAG
, 1, DMU_OT_UINT64_OTHER
, 0, chunksize
);
3655 if (ztest_object_init(zd
, od
, size
, B_FALSE
) != 0) {
3656 umem_free(od
, size
);
3660 bigobj
= od
[0].od_object
;
3661 packobj
= od
[1].od_object
;
3662 chunksize
= od
[0].od_gen
;
3663 ASSERT(chunksize
== od
[1].od_gen
);
3666 * Prefetch a random chunk of the big object.
3667 * Our aim here is to get some async reads in flight
3668 * for blocks that we may free below; the DMU should
3669 * handle this race correctly.
3671 n
= ztest_random(regions
) * stride
+ ztest_random(width
);
3672 s
= 1 + ztest_random(2 * width
- 1);
3673 dmu_prefetch(os
, bigobj
, n
* chunksize
, s
* chunksize
);
3676 * Pick a random index and compute the offsets into packobj and bigobj.
3678 n
= ztest_random(regions
) * stride
+ ztest_random(width
);
3679 s
= 1 + ztest_random(width
- 1);
3681 packoff
= n
* sizeof (bufwad_t
);
3682 packsize
= s
* sizeof (bufwad_t
);
3684 bigoff
= n
* chunksize
;
3685 bigsize
= s
* chunksize
;
3687 packbuf
= umem_alloc(packsize
, UMEM_NOFAIL
);
3688 bigbuf
= umem_alloc(bigsize
, UMEM_NOFAIL
);
3691 * free_percent of the time, free a range of bigobj rather than
3694 freeit
= (ztest_random(100) < free_percent
);
3697 * Read the current contents of our objects.
3699 error
= dmu_read(os
, packobj
, packoff
, packsize
, packbuf
,
3702 error
= dmu_read(os
, bigobj
, bigoff
, bigsize
, bigbuf
,
3707 * Get a tx for the mods to both packobj and bigobj.
3709 tx
= dmu_tx_create(os
);
3711 dmu_tx_hold_write(tx
, packobj
, packoff
, packsize
);
3714 dmu_tx_hold_free(tx
, bigobj
, bigoff
, bigsize
);
3716 dmu_tx_hold_write(tx
, bigobj
, bigoff
, bigsize
);
3718 txg
= ztest_tx_assign(tx
, TXG_MIGHTWAIT
, FTAG
);
3720 umem_free(packbuf
, packsize
);
3721 umem_free(bigbuf
, bigsize
);
3722 umem_free(od
, size
);
3726 dmu_object_set_checksum(os
, bigobj
,
3727 (enum zio_checksum
)ztest_random_dsl_prop(ZFS_PROP_CHECKSUM
), tx
);
3729 dmu_object_set_compress(os
, bigobj
,
3730 (enum zio_compress
)ztest_random_dsl_prop(ZFS_PROP_COMPRESSION
), tx
);
3733 * For each index from n to n + s, verify that the existing bufwad
3734 * in packobj matches the bufwads at the head and tail of the
3735 * corresponding chunk in bigobj. Then update all three bufwads
3736 * with the new values we want to write out.
3738 for (i
= 0; i
< s
; i
++) {
3740 pack
= (bufwad_t
*)((char *)packbuf
+ i
* sizeof (bufwad_t
));
3742 bigH
= (bufwad_t
*)((char *)bigbuf
+ i
* chunksize
);
3744 bigT
= (bufwad_t
*)((char *)bigH
+ chunksize
) - 1;
3746 ASSERT((uintptr_t)bigH
- (uintptr_t)bigbuf
< bigsize
);
3747 ASSERT((uintptr_t)bigT
- (uintptr_t)bigbuf
< bigsize
);
3749 if (pack
->bw_txg
> txg
)
3750 fatal(0, "future leak: got %llx, open txg is %llx",
3753 if (pack
->bw_data
!= 0 && pack
->bw_index
!= n
+ i
)
3754 fatal(0, "wrong index: got %llx, wanted %llx+%llx",
3755 pack
->bw_index
, n
, i
);
3757 if (bcmp(pack
, bigH
, sizeof (bufwad_t
)) != 0)
3758 fatal(0, "pack/bigH mismatch in %p/%p", pack
, bigH
);
3760 if (bcmp(pack
, bigT
, sizeof (bufwad_t
)) != 0)
3761 fatal(0, "pack/bigT mismatch in %p/%p", pack
, bigT
);
3764 bzero(pack
, sizeof (bufwad_t
));
3766 pack
->bw_index
= n
+ i
;
3768 pack
->bw_data
= 1 + ztest_random(-2ULL);
3775 * We've verified all the old bufwads, and made new ones.
3776 * Now write them out.
3778 dmu_write(os
, packobj
, packoff
, packsize
, packbuf
, tx
);
3781 if (ztest_opts
.zo_verbose
>= 7) {
3782 (void) printf("freeing offset %llx size %llx"
3784 (u_longlong_t
)bigoff
,
3785 (u_longlong_t
)bigsize
,
3788 VERIFY(0 == dmu_free_range(os
, bigobj
, bigoff
, bigsize
, tx
));
3790 if (ztest_opts
.zo_verbose
>= 7) {
3791 (void) printf("writing offset %llx size %llx"
3793 (u_longlong_t
)bigoff
,
3794 (u_longlong_t
)bigsize
,
3797 dmu_write(os
, bigobj
, bigoff
, bigsize
, bigbuf
, tx
);
3803 * Sanity check the stuff we just wrote.
3806 void *packcheck
= umem_alloc(packsize
, UMEM_NOFAIL
);
3807 void *bigcheck
= umem_alloc(bigsize
, UMEM_NOFAIL
);
3809 VERIFY(0 == dmu_read(os
, packobj
, packoff
,
3810 packsize
, packcheck
, DMU_READ_PREFETCH
));
3811 VERIFY(0 == dmu_read(os
, bigobj
, bigoff
,
3812 bigsize
, bigcheck
, DMU_READ_PREFETCH
));
3814 ASSERT(bcmp(packbuf
, packcheck
, packsize
) == 0);
3815 ASSERT(bcmp(bigbuf
, bigcheck
, bigsize
) == 0);
3817 umem_free(packcheck
, packsize
);
3818 umem_free(bigcheck
, bigsize
);
3821 umem_free(packbuf
, packsize
);
3822 umem_free(bigbuf
, bigsize
);
3823 umem_free(od
, size
);
3827 compare_and_update_pbbufs(uint64_t s
, bufwad_t
*packbuf
, bufwad_t
*bigbuf
,
3828 uint64_t bigsize
, uint64_t n
, uint64_t chunksize
, uint64_t txg
)
3836 * For each index from n to n + s, verify that the existing bufwad
3837 * in packobj matches the bufwads at the head and tail of the
3838 * corresponding chunk in bigobj. Then update all three bufwads
3839 * with the new values we want to write out.
3841 for (i
= 0; i
< s
; i
++) {
3843 pack
= (bufwad_t
*)((char *)packbuf
+ i
* sizeof (bufwad_t
));
3845 bigH
= (bufwad_t
*)((char *)bigbuf
+ i
* chunksize
);
3847 bigT
= (bufwad_t
*)((char *)bigH
+ chunksize
) - 1;
3849 ASSERT((uintptr_t)bigH
- (uintptr_t)bigbuf
< bigsize
);
3850 ASSERT((uintptr_t)bigT
- (uintptr_t)bigbuf
< bigsize
);
3852 if (pack
->bw_txg
> txg
)
3853 fatal(0, "future leak: got %llx, open txg is %llx",
3856 if (pack
->bw_data
!= 0 && pack
->bw_index
!= n
+ i
)
3857 fatal(0, "wrong index: got %llx, wanted %llx+%llx",
3858 pack
->bw_index
, n
, i
);
3860 if (bcmp(pack
, bigH
, sizeof (bufwad_t
)) != 0)
3861 fatal(0, "pack/bigH mismatch in %p/%p", pack
, bigH
);
3863 if (bcmp(pack
, bigT
, sizeof (bufwad_t
)) != 0)
3864 fatal(0, "pack/bigT mismatch in %p/%p", pack
, bigT
);
3866 pack
->bw_index
= n
+ i
;
3868 pack
->bw_data
= 1 + ztest_random(-2ULL);
3875 #undef OD_ARRAY_SIZE
3876 #define OD_ARRAY_SIZE 2
3879 ztest_dmu_read_write_zcopy(ztest_ds_t
*zd
, uint64_t id
)
3881 objset_t
*os
= zd
->zd_os
;
3888 bufwad_t
*packbuf
, *bigbuf
;
3889 uint64_t packobj
, packoff
, packsize
, bigobj
, bigoff
, bigsize
;
3890 uint64_t blocksize
= ztest_random_blocksize();
3891 uint64_t chunksize
= blocksize
;
3892 uint64_t regions
= 997;
3893 uint64_t stride
= 123456789ULL;
3895 dmu_buf_t
*bonus_db
;
3896 arc_buf_t
**bigbuf_arcbufs
;
3897 dmu_object_info_t doi
;
3899 size
= sizeof(ztest_od_t
) * OD_ARRAY_SIZE
;
3900 od
= umem_alloc(size
, UMEM_NOFAIL
);
3903 * This test uses two objects, packobj and bigobj, that are always
3904 * updated together (i.e. in the same tx) so that their contents are
3905 * in sync and can be compared. Their contents relate to each other
3906 * in a simple way: packobj is a dense array of 'bufwad' structures,
3907 * while bigobj is a sparse array of the same bufwads. Specifically,
3908 * for any index n, there are three bufwads that should be identical:
3910 * packobj, at offset n * sizeof (bufwad_t)
3911 * bigobj, at the head of the nth chunk
3912 * bigobj, at the tail of the nth chunk
3914 * The chunk size is set equal to bigobj block size so that
3915 * dmu_assign_arcbuf() can be tested for object updates.
3919 * Read the directory info. If it's the first time, set things up.
3921 ztest_od_init(od
, id
, FTAG
, 0, DMU_OT_UINT64_OTHER
, blocksize
, 0);
3922 ztest_od_init(od
+ 1, id
, FTAG
, 1, DMU_OT_UINT64_OTHER
, 0, chunksize
);
3925 if (ztest_object_init(zd
, od
, size
, B_FALSE
) != 0) {
3926 umem_free(od
, size
);
3930 bigobj
= od
[0].od_object
;
3931 packobj
= od
[1].od_object
;
3932 blocksize
= od
[0].od_blocksize
;
3933 chunksize
= blocksize
;
3934 ASSERT(chunksize
== od
[1].od_gen
);
3936 VERIFY(dmu_object_info(os
, bigobj
, &doi
) == 0);
3937 VERIFY(ISP2(doi
.doi_data_block_size
));
3938 VERIFY(chunksize
== doi
.doi_data_block_size
);
3939 VERIFY(chunksize
>= 2 * sizeof (bufwad_t
));
3942 * Pick a random index and compute the offsets into packobj and bigobj.
3944 n
= ztest_random(regions
) * stride
+ ztest_random(width
);
3945 s
= 1 + ztest_random(width
- 1);
3947 packoff
= n
* sizeof (bufwad_t
);
3948 packsize
= s
* sizeof (bufwad_t
);
3950 bigoff
= n
* chunksize
;
3951 bigsize
= s
* chunksize
;
3953 packbuf
= umem_zalloc(packsize
, UMEM_NOFAIL
);
3954 bigbuf
= umem_zalloc(bigsize
, UMEM_NOFAIL
);
3956 VERIFY3U(0, ==, dmu_bonus_hold(os
, bigobj
, FTAG
, &bonus_db
));
3958 bigbuf_arcbufs
= umem_zalloc(2 * s
* sizeof (arc_buf_t
*), UMEM_NOFAIL
);
3961 * Iteration 0 test zcopy for DB_UNCACHED dbufs.
3962 * Iteration 1 test zcopy to already referenced dbufs.
3963 * Iteration 2 test zcopy to dirty dbuf in the same txg.
3964 * Iteration 3 test zcopy to dbuf dirty in previous txg.
3965 * Iteration 4 test zcopy when dbuf is no longer dirty.
3966 * Iteration 5 test zcopy when it can't be done.
3967 * Iteration 6 one more zcopy write.
3969 for (i
= 0; i
< 7; i
++) {
3974 * In iteration 5 (i == 5) use arcbufs
3975 * that don't match bigobj blksz to test
3976 * dmu_assign_arcbuf() when it can't directly
3977 * assign an arcbuf to a dbuf.
3979 for (j
= 0; j
< s
; j
++) {
3982 dmu_request_arcbuf(bonus_db
, chunksize
);
3984 bigbuf_arcbufs
[2 * j
] =
3985 dmu_request_arcbuf(bonus_db
, chunksize
/ 2);
3986 bigbuf_arcbufs
[2 * j
+ 1] =
3987 dmu_request_arcbuf(bonus_db
, chunksize
/ 2);
3992 * Get a tx for the mods to both packobj and bigobj.
3994 tx
= dmu_tx_create(os
);
3996 dmu_tx_hold_write(tx
, packobj
, packoff
, packsize
);
3997 dmu_tx_hold_write(tx
, bigobj
, bigoff
, bigsize
);
3999 txg
= ztest_tx_assign(tx
, TXG_MIGHTWAIT
, FTAG
);
4001 umem_free(packbuf
, packsize
);
4002 umem_free(bigbuf
, bigsize
);
4003 for (j
= 0; j
< s
; j
++) {
4005 dmu_return_arcbuf(bigbuf_arcbufs
[j
]);
4008 bigbuf_arcbufs
[2 * j
]);
4010 bigbuf_arcbufs
[2 * j
+ 1]);
4013 umem_free(bigbuf_arcbufs
, 2 * s
* sizeof (arc_buf_t
*));
4014 umem_free(od
, size
);
4015 dmu_buf_rele(bonus_db
, FTAG
);
4020 * 50% of the time don't read objects in the 1st iteration to
4021 * test dmu_assign_arcbuf() for the case when there're no
4022 * existing dbufs for the specified offsets.
4024 if (i
!= 0 || ztest_random(2) != 0) {
4025 error
= dmu_read(os
, packobj
, packoff
,
4026 packsize
, packbuf
, DMU_READ_PREFETCH
);
4028 error
= dmu_read(os
, bigobj
, bigoff
, bigsize
,
4029 bigbuf
, DMU_READ_PREFETCH
);
4032 compare_and_update_pbbufs(s
, packbuf
, bigbuf
, bigsize
,
4036 * We've verified all the old bufwads, and made new ones.
4037 * Now write them out.
4039 dmu_write(os
, packobj
, packoff
, packsize
, packbuf
, tx
);
4040 if (ztest_opts
.zo_verbose
>= 7) {
4041 (void) printf("writing offset %llx size %llx"
4043 (u_longlong_t
)bigoff
,
4044 (u_longlong_t
)bigsize
,
4047 for (off
= bigoff
, j
= 0; j
< s
; j
++, off
+= chunksize
) {
4050 bcopy((caddr_t
)bigbuf
+ (off
- bigoff
),
4051 bigbuf_arcbufs
[j
]->b_data
, chunksize
);
4053 bcopy((caddr_t
)bigbuf
+ (off
- bigoff
),
4054 bigbuf_arcbufs
[2 * j
]->b_data
,
4056 bcopy((caddr_t
)bigbuf
+ (off
- bigoff
) +
4058 bigbuf_arcbufs
[2 * j
+ 1]->b_data
,
4063 VERIFY(dmu_buf_hold(os
, bigobj
, off
,
4064 FTAG
, &dbt
, DMU_READ_NO_PREFETCH
) == 0);
4067 dmu_assign_arcbuf(bonus_db
, off
,
4068 bigbuf_arcbufs
[j
], tx
);
4070 dmu_assign_arcbuf(bonus_db
, off
,
4071 bigbuf_arcbufs
[2 * j
], tx
);
4072 dmu_assign_arcbuf(bonus_db
,
4073 off
+ chunksize
/ 2,
4074 bigbuf_arcbufs
[2 * j
+ 1], tx
);
4077 dmu_buf_rele(dbt
, FTAG
);
4083 * Sanity check the stuff we just wrote.
4086 void *packcheck
= umem_alloc(packsize
, UMEM_NOFAIL
);
4087 void *bigcheck
= umem_alloc(bigsize
, UMEM_NOFAIL
);
4089 VERIFY(0 == dmu_read(os
, packobj
, packoff
,
4090 packsize
, packcheck
, DMU_READ_PREFETCH
));
4091 VERIFY(0 == dmu_read(os
, bigobj
, bigoff
,
4092 bigsize
, bigcheck
, DMU_READ_PREFETCH
));
4094 ASSERT(bcmp(packbuf
, packcheck
, packsize
) == 0);
4095 ASSERT(bcmp(bigbuf
, bigcheck
, bigsize
) == 0);
4097 umem_free(packcheck
, packsize
);
4098 umem_free(bigcheck
, bigsize
);
4101 txg_wait_open(dmu_objset_pool(os
), 0);
4102 } else if (i
== 3) {
4103 txg_wait_synced(dmu_objset_pool(os
), 0);
4107 dmu_buf_rele(bonus_db
, FTAG
);
4108 umem_free(packbuf
, packsize
);
4109 umem_free(bigbuf
, bigsize
);
4110 umem_free(bigbuf_arcbufs
, 2 * s
* sizeof (arc_buf_t
*));
4111 umem_free(od
, size
);
4116 ztest_dmu_write_parallel(ztest_ds_t
*zd
, uint64_t id
)
4120 od
= umem_alloc(sizeof(ztest_od_t
), UMEM_NOFAIL
);
4121 uint64_t offset
= (1ULL << (ztest_random(20) + 43)) +
4122 (ztest_random(ZTEST_RANGE_LOCKS
) << SPA_MAXBLOCKSHIFT
);
4125 * Have multiple threads write to large offsets in an object
4126 * to verify that parallel writes to an object -- even to the
4127 * same blocks within the object -- doesn't cause any trouble.
4129 ztest_od_init(od
, ID_PARALLEL
, FTAG
, 0, DMU_OT_UINT64_OTHER
, 0, 0);
4131 if (ztest_object_init(zd
, od
, sizeof (ztest_od_t
), B_FALSE
) != 0)
4134 while (ztest_random(10) != 0)
4135 ztest_io(zd
, od
->od_object
, offset
);
4137 umem_free(od
, sizeof(ztest_od_t
));
4141 ztest_dmu_prealloc(ztest_ds_t
*zd
, uint64_t id
)
4144 uint64_t offset
= (1ULL << (ztest_random(4) + SPA_MAXBLOCKSHIFT
)) +
4145 (ztest_random(ZTEST_RANGE_LOCKS
) << SPA_MAXBLOCKSHIFT
);
4146 uint64_t count
= ztest_random(20) + 1;
4147 uint64_t blocksize
= ztest_random_blocksize();
4150 od
= umem_alloc(sizeof(ztest_od_t
), UMEM_NOFAIL
);
4152 ztest_od_init(od
, id
, FTAG
, 0, DMU_OT_UINT64_OTHER
, blocksize
, 0);
4154 if (ztest_object_init(zd
, od
, sizeof (ztest_od_t
), !ztest_random(2)) != 0) {
4155 umem_free(od
, sizeof(ztest_od_t
));
4159 if (ztest_truncate(zd
, od
->od_object
, offset
, count
* blocksize
) != 0) {
4160 umem_free(od
, sizeof(ztest_od_t
));
4164 ztest_prealloc(zd
, od
->od_object
, offset
, count
* blocksize
);
4166 data
= umem_zalloc(blocksize
, UMEM_NOFAIL
);
4168 while (ztest_random(count
) != 0) {
4169 uint64_t randoff
= offset
+ (ztest_random(count
) * blocksize
);
4170 if (ztest_write(zd
, od
->od_object
, randoff
, blocksize
,
4173 while (ztest_random(4) != 0)
4174 ztest_io(zd
, od
->od_object
, randoff
);
4177 umem_free(data
, blocksize
);
4178 umem_free(od
, sizeof(ztest_od_t
));
4182 * Verify that zap_{create,destroy,add,remove,update} work as expected.
4184 #define ZTEST_ZAP_MIN_INTS 1
4185 #define ZTEST_ZAP_MAX_INTS 4
4186 #define ZTEST_ZAP_MAX_PROPS 1000
4189 ztest_zap(ztest_ds_t
*zd
, uint64_t id
)
4191 objset_t
*os
= zd
->zd_os
;
4194 uint64_t txg
, last_txg
;
4195 uint64_t value
[ZTEST_ZAP_MAX_INTS
];
4196 uint64_t zl_ints
, zl_intsize
, prop
;
4199 char propname
[100], txgname
[100];
4201 char *hc
[2] = { "s.acl.h", ".s.open.h.hyLZlg" };
4203 od
= umem_alloc(sizeof(ztest_od_t
), UMEM_NOFAIL
);
4204 ztest_od_init(od
, id
, FTAG
, 0, DMU_OT_ZAP_OTHER
, 0, 0);
4206 if (ztest_object_init(zd
, od
, sizeof (ztest_od_t
),
4207 !ztest_random(2)) != 0)
4210 object
= od
->od_object
;
4213 * Generate a known hash collision, and verify that
4214 * we can lookup and remove both entries.
4216 tx
= dmu_tx_create(os
);
4217 dmu_tx_hold_zap(tx
, object
, B_TRUE
, NULL
);
4218 txg
= ztest_tx_assign(tx
, TXG_MIGHTWAIT
, FTAG
);
4221 for (i
= 0; i
< 2; i
++) {
4223 VERIFY3U(0, ==, zap_add(os
, object
, hc
[i
], sizeof (uint64_t),
4226 for (i
= 0; i
< 2; i
++) {
4227 VERIFY3U(EEXIST
, ==, zap_add(os
, object
, hc
[i
],
4228 sizeof (uint64_t), 1, &value
[i
], tx
));
4230 zap_length(os
, object
, hc
[i
], &zl_intsize
, &zl_ints
));
4231 ASSERT3U(zl_intsize
, ==, sizeof (uint64_t));
4232 ASSERT3U(zl_ints
, ==, 1);
4234 for (i
= 0; i
< 2; i
++) {
4235 VERIFY3U(0, ==, zap_remove(os
, object
, hc
[i
], tx
));
4240 * Generate a buch of random entries.
4242 ints
= MAX(ZTEST_ZAP_MIN_INTS
, object
% ZTEST_ZAP_MAX_INTS
);
4244 prop
= ztest_random(ZTEST_ZAP_MAX_PROPS
);
4245 (void) sprintf(propname
, "prop_%llu", (u_longlong_t
)prop
);
4246 (void) sprintf(txgname
, "txg_%llu", (u_longlong_t
)prop
);
4247 bzero(value
, sizeof (value
));
4251 * If these zap entries already exist, validate their contents.
4253 error
= zap_length(os
, object
, txgname
, &zl_intsize
, &zl_ints
);
4255 ASSERT3U(zl_intsize
, ==, sizeof (uint64_t));
4256 ASSERT3U(zl_ints
, ==, 1);
4258 VERIFY(zap_lookup(os
, object
, txgname
, zl_intsize
,
4259 zl_ints
, &last_txg
) == 0);
4261 VERIFY(zap_length(os
, object
, propname
, &zl_intsize
,
4264 ASSERT3U(zl_intsize
, ==, sizeof (uint64_t));
4265 ASSERT3U(zl_ints
, ==, ints
);
4267 VERIFY(zap_lookup(os
, object
, propname
, zl_intsize
,
4268 zl_ints
, value
) == 0);
4270 for (i
= 0; i
< ints
; i
++) {
4271 ASSERT3U(value
[i
], ==, last_txg
+ object
+ i
);
4274 ASSERT3U(error
, ==, ENOENT
);
4278 * Atomically update two entries in our zap object.
4279 * The first is named txg_%llu, and contains the txg
4280 * in which the property was last updated. The second
4281 * is named prop_%llu, and the nth element of its value
4282 * should be txg + object + n.
4284 tx
= dmu_tx_create(os
);
4285 dmu_tx_hold_zap(tx
, object
, B_TRUE
, NULL
);
4286 txg
= ztest_tx_assign(tx
, TXG_MIGHTWAIT
, FTAG
);
4291 fatal(0, "zap future leak: old %llu new %llu", last_txg
, txg
);
4293 for (i
= 0; i
< ints
; i
++)
4294 value
[i
] = txg
+ object
+ i
;
4296 VERIFY3U(0, ==, zap_update(os
, object
, txgname
, sizeof (uint64_t),
4298 VERIFY3U(0, ==, zap_update(os
, object
, propname
, sizeof (uint64_t),
4304 * Remove a random pair of entries.
4306 prop
= ztest_random(ZTEST_ZAP_MAX_PROPS
);
4307 (void) sprintf(propname
, "prop_%llu", (u_longlong_t
)prop
);
4308 (void) sprintf(txgname
, "txg_%llu", (u_longlong_t
)prop
);
4310 error
= zap_length(os
, object
, txgname
, &zl_intsize
, &zl_ints
);
4312 if (error
== ENOENT
)
4317 tx
= dmu_tx_create(os
);
4318 dmu_tx_hold_zap(tx
, object
, B_TRUE
, NULL
);
4319 txg
= ztest_tx_assign(tx
, TXG_MIGHTWAIT
, FTAG
);
4322 VERIFY3U(0, ==, zap_remove(os
, object
, txgname
, tx
));
4323 VERIFY3U(0, ==, zap_remove(os
, object
, propname
, tx
));
4326 umem_free(od
, sizeof(ztest_od_t
));
4330 * Testcase to test the upgrading of a microzap to fatzap.
4333 ztest_fzap(ztest_ds_t
*zd
, uint64_t id
)
4335 objset_t
*os
= zd
->zd_os
;
4337 uint64_t object
, txg
;
4340 od
= umem_alloc(sizeof(ztest_od_t
), UMEM_NOFAIL
);
4341 ztest_od_init(od
, id
, FTAG
, 0, DMU_OT_ZAP_OTHER
, 0, 0);
4343 if (ztest_object_init(zd
, od
, sizeof (ztest_od_t
),
4344 !ztest_random(2)) != 0)
4346 object
= od
->od_object
;
4349 * Add entries to this ZAP and make sure it spills over
4350 * and gets upgraded to a fatzap. Also, since we are adding
4351 * 2050 entries we should see ptrtbl growth and leaf-block split.
4353 for (i
= 0; i
< 2050; i
++) {
4354 char name
[MAXNAMELEN
];
4359 (void) snprintf(name
, sizeof (name
), "fzap-%llu-%llu",
4360 (u_longlong_t
)id
, (u_longlong_t
)value
);
4362 tx
= dmu_tx_create(os
);
4363 dmu_tx_hold_zap(tx
, object
, B_TRUE
, name
);
4364 txg
= ztest_tx_assign(tx
, TXG_MIGHTWAIT
, FTAG
);
4367 error
= zap_add(os
, object
, name
, sizeof (uint64_t), 1,
4369 ASSERT(error
== 0 || error
== EEXIST
);
4373 umem_free(od
, sizeof(ztest_od_t
));
4378 ztest_zap_parallel(ztest_ds_t
*zd
, uint64_t id
)
4380 objset_t
*os
= zd
->zd_os
;
4382 uint64_t txg
, object
, count
, wsize
, wc
, zl_wsize
, zl_wc
;
4384 int i
, namelen
, error
;
4385 int micro
= ztest_random(2);
4386 char name
[20], string_value
[20];
4389 od
= umem_alloc(sizeof(ztest_od_t
), UMEM_NOFAIL
);
4390 ztest_od_init(od
, ID_PARALLEL
, FTAG
, micro
, DMU_OT_ZAP_OTHER
, 0, 0);
4392 if (ztest_object_init(zd
, od
, sizeof (ztest_od_t
), B_FALSE
) != 0) {
4393 umem_free(od
, sizeof(ztest_od_t
));
4397 object
= od
->od_object
;
4400 * Generate a random name of the form 'xxx.....' where each
4401 * x is a random printable character and the dots are dots.
4402 * There are 94 such characters, and the name length goes from
4403 * 6 to 20, so there are 94^3 * 15 = 12,458,760 possible names.
4405 namelen
= ztest_random(sizeof (name
) - 5) + 5 + 1;
4407 for (i
= 0; i
< 3; i
++)
4408 name
[i
] = '!' + ztest_random('~' - '!' + 1);
4409 for (; i
< namelen
- 1; i
++)
4413 if ((namelen
& 1) || micro
) {
4414 wsize
= sizeof (txg
);
4420 data
= string_value
;
4424 VERIFY0(zap_count(os
, object
, &count
));
4425 ASSERT(count
!= -1ULL);
4428 * Select an operation: length, lookup, add, update, remove.
4430 i
= ztest_random(5);
4433 tx
= dmu_tx_create(os
);
4434 dmu_tx_hold_zap(tx
, object
, B_TRUE
, NULL
);
4435 txg
= ztest_tx_assign(tx
, TXG_MIGHTWAIT
, FTAG
);
4438 bcopy(name
, string_value
, namelen
);
4442 bzero(string_value
, namelen
);
4448 error
= zap_length(os
, object
, name
, &zl_wsize
, &zl_wc
);
4450 ASSERT3U(wsize
, ==, zl_wsize
);
4451 ASSERT3U(wc
, ==, zl_wc
);
4453 ASSERT3U(error
, ==, ENOENT
);
4458 error
= zap_lookup(os
, object
, name
, wsize
, wc
, data
);
4460 if (data
== string_value
&&
4461 bcmp(name
, data
, namelen
) != 0)
4462 fatal(0, "name '%s' != val '%s' len %d",
4463 name
, data
, namelen
);
4465 ASSERT3U(error
, ==, ENOENT
);
4470 error
= zap_add(os
, object
, name
, wsize
, wc
, data
, tx
);
4471 ASSERT(error
== 0 || error
== EEXIST
);
4475 VERIFY(zap_update(os
, object
, name
, wsize
, wc
, data
, tx
) == 0);
4479 error
= zap_remove(os
, object
, name
, tx
);
4480 ASSERT(error
== 0 || error
== ENOENT
);
4487 umem_free(od
, sizeof(ztest_od_t
));
4491 * Commit callback data.
4493 typedef struct ztest_cb_data
{
4494 list_node_t zcd_node
;
4496 int zcd_expected_err
;
4497 boolean_t zcd_added
;
4498 boolean_t zcd_called
;
4502 /* This is the actual commit callback function */
4504 ztest_commit_callback(void *arg
, int error
)
4506 ztest_cb_data_t
*data
= arg
;
4507 uint64_t synced_txg
;
4509 VERIFY(data
!= NULL
);
4510 VERIFY3S(data
->zcd_expected_err
, ==, error
);
4511 VERIFY(!data
->zcd_called
);
4513 synced_txg
= spa_last_synced_txg(data
->zcd_spa
);
4514 if (data
->zcd_txg
> synced_txg
)
4515 fatal(0, "commit callback of txg %" PRIu64
" called prematurely"
4516 ", last synced txg = %" PRIu64
"\n", data
->zcd_txg
,
4519 data
->zcd_called
= B_TRUE
;
4521 if (error
== ECANCELED
) {
4522 ASSERT0(data
->zcd_txg
);
4523 ASSERT(!data
->zcd_added
);
4526 * The private callback data should be destroyed here, but
4527 * since we are going to check the zcd_called field after
4528 * dmu_tx_abort(), we will destroy it there.
4533 ASSERT(data
->zcd_added
);
4534 ASSERT3U(data
->zcd_txg
, !=, 0);
4536 (void) mutex_enter(&zcl
.zcl_callbacks_lock
);
4538 /* See if this cb was called more quickly */
4539 if ((synced_txg
- data
->zcd_txg
) < zc_min_txg_delay
)
4540 zc_min_txg_delay
= synced_txg
- data
->zcd_txg
;
4542 /* Remove our callback from the list */
4543 list_remove(&zcl
.zcl_callbacks
, data
);
4545 (void) mutex_exit(&zcl
.zcl_callbacks_lock
);
4547 umem_free(data
, sizeof (ztest_cb_data_t
));
4550 /* Allocate and initialize callback data structure */
4551 static ztest_cb_data_t
*
4552 ztest_create_cb_data(objset_t
*os
, uint64_t txg
)
4554 ztest_cb_data_t
*cb_data
;
4556 cb_data
= umem_zalloc(sizeof (ztest_cb_data_t
), UMEM_NOFAIL
);
4558 cb_data
->zcd_txg
= txg
;
4559 cb_data
->zcd_spa
= dmu_objset_spa(os
);
4560 list_link_init(&cb_data
->zcd_node
);
4566 * Commit callback test.
4569 ztest_dmu_commit_callbacks(ztest_ds_t
*zd
, uint64_t id
)
4571 objset_t
*os
= zd
->zd_os
;
4574 ztest_cb_data_t
*cb_data
[3], *tmp_cb
;
4575 uint64_t old_txg
, txg
;
4578 od
= umem_alloc(sizeof(ztest_od_t
), UMEM_NOFAIL
);
4579 ztest_od_init(od
, id
, FTAG
, 0, DMU_OT_UINT64_OTHER
, 0, 0);
4581 if (ztest_object_init(zd
, od
, sizeof (ztest_od_t
), B_FALSE
) != 0) {
4582 umem_free(od
, sizeof(ztest_od_t
));
4586 tx
= dmu_tx_create(os
);
4588 cb_data
[0] = ztest_create_cb_data(os
, 0);
4589 dmu_tx_callback_register(tx
, ztest_commit_callback
, cb_data
[0]);
4591 dmu_tx_hold_write(tx
, od
->od_object
, 0, sizeof (uint64_t));
4593 /* Every once in a while, abort the transaction on purpose */
4594 if (ztest_random(100) == 0)
4598 error
= dmu_tx_assign(tx
, TXG_NOWAIT
);
4600 txg
= error
? 0 : dmu_tx_get_txg(tx
);
4602 cb_data
[0]->zcd_txg
= txg
;
4603 cb_data
[1] = ztest_create_cb_data(os
, txg
);
4604 dmu_tx_callback_register(tx
, ztest_commit_callback
, cb_data
[1]);
4608 * It's not a strict requirement to call the registered
4609 * callbacks from inside dmu_tx_abort(), but that's what
4610 * it's supposed to happen in the current implementation
4611 * so we will check for that.
4613 for (i
= 0; i
< 2; i
++) {
4614 cb_data
[i
]->zcd_expected_err
= ECANCELED
;
4615 VERIFY(!cb_data
[i
]->zcd_called
);
4620 for (i
= 0; i
< 2; i
++) {
4621 VERIFY(cb_data
[i
]->zcd_called
);
4622 umem_free(cb_data
[i
], sizeof (ztest_cb_data_t
));
4625 umem_free(od
, sizeof(ztest_od_t
));
4629 cb_data
[2] = ztest_create_cb_data(os
, txg
);
4630 dmu_tx_callback_register(tx
, ztest_commit_callback
, cb_data
[2]);
4633 * Read existing data to make sure there isn't a future leak.
4635 VERIFY(0 == dmu_read(os
, od
->od_object
, 0, sizeof (uint64_t),
4636 &old_txg
, DMU_READ_PREFETCH
));
4639 fatal(0, "future leak: got %" PRIu64
", open txg is %" PRIu64
,
4642 dmu_write(os
, od
->od_object
, 0, sizeof (uint64_t), &txg
, tx
);
4644 (void) mutex_enter(&zcl
.zcl_callbacks_lock
);
4647 * Since commit callbacks don't have any ordering requirement and since
4648 * it is theoretically possible for a commit callback to be called
4649 * after an arbitrary amount of time has elapsed since its txg has been
4650 * synced, it is difficult to reliably determine whether a commit
4651 * callback hasn't been called due to high load or due to a flawed
4654 * In practice, we will assume that if after a certain number of txgs a
4655 * commit callback hasn't been called, then most likely there's an
4656 * implementation bug..
4658 tmp_cb
= list_head(&zcl
.zcl_callbacks
);
4659 if (tmp_cb
!= NULL
&&
4660 tmp_cb
->zcd_txg
+ ZTEST_COMMIT_CB_THRESH
< txg
) {
4661 fatal(0, "Commit callback threshold exceeded, oldest txg: %"
4662 PRIu64
", open txg: %" PRIu64
"\n", tmp_cb
->zcd_txg
, txg
);
4666 * Let's find the place to insert our callbacks.
4668 * Even though the list is ordered by txg, it is possible for the
4669 * insertion point to not be the end because our txg may already be
4670 * quiescing at this point and other callbacks in the open txg
4671 * (from other objsets) may have sneaked in.
4673 tmp_cb
= list_tail(&zcl
.zcl_callbacks
);
4674 while (tmp_cb
!= NULL
&& tmp_cb
->zcd_txg
> txg
)
4675 tmp_cb
= list_prev(&zcl
.zcl_callbacks
, tmp_cb
);
4677 /* Add the 3 callbacks to the list */
4678 for (i
= 0; i
< 3; i
++) {
4680 list_insert_head(&zcl
.zcl_callbacks
, cb_data
[i
]);
4682 list_insert_after(&zcl
.zcl_callbacks
, tmp_cb
,
4685 cb_data
[i
]->zcd_added
= B_TRUE
;
4686 VERIFY(!cb_data
[i
]->zcd_called
);
4688 tmp_cb
= cb_data
[i
];
4693 (void) mutex_exit(&zcl
.zcl_callbacks_lock
);
4697 umem_free(od
, sizeof(ztest_od_t
));
4702 ztest_dsl_prop_get_set(ztest_ds_t
*zd
, uint64_t id
)
4704 zfs_prop_t proplist
[] = {
4706 ZFS_PROP_COMPRESSION
,
4712 (void) rw_enter(&ztest_name_lock
, RW_READER
);
4714 for (p
= 0; p
< sizeof (proplist
) / sizeof (proplist
[0]); p
++)
4715 (void) ztest_dsl_prop_set_uint64(zd
->zd_name
, proplist
[p
],
4716 ztest_random_dsl_prop(proplist
[p
]), (int)ztest_random(2));
4718 (void) rw_exit(&ztest_name_lock
);
4723 ztest_spa_prop_get_set(ztest_ds_t
*zd
, uint64_t id
)
4725 nvlist_t
*props
= NULL
;
4727 (void) rw_enter(&ztest_name_lock
, RW_READER
);
4729 (void) ztest_spa_prop_set_uint64(ZPOOL_PROP_DEDUPDITTO
,
4730 ZIO_DEDUPDITTO_MIN
+ ztest_random(ZIO_DEDUPDITTO_MIN
));
4732 VERIFY0(spa_prop_get(ztest_spa
, &props
));
4734 if (ztest_opts
.zo_verbose
>= 6)
4735 dump_nvlist(props
, 4);
4739 (void) rw_exit(&ztest_name_lock
);
4743 user_release_one(const char *snapname
, const char *holdname
)
4745 nvlist_t
*snaps
, *holds
;
4748 snaps
= fnvlist_alloc();
4749 holds
= fnvlist_alloc();
4750 fnvlist_add_boolean(holds
, holdname
);
4751 fnvlist_add_nvlist(snaps
, snapname
, holds
);
4752 fnvlist_free(holds
);
4753 error
= dsl_dataset_user_release(snaps
, NULL
);
4754 fnvlist_free(snaps
);
4759 * Test snapshot hold/release and deferred destroy.
4762 ztest_dmu_snapshot_hold(ztest_ds_t
*zd
, uint64_t id
)
4765 objset_t
*os
= zd
->zd_os
;
4769 char clonename
[100];
4771 char osname
[MAXNAMELEN
];
4774 (void) rw_enter(&ztest_name_lock
, RW_READER
);
4776 dmu_objset_name(os
, osname
);
4778 (void) snprintf(snapname
, sizeof (snapname
), "sh1_%llu", (long long unsigned int)id
);
4779 (void) snprintf(fullname
, sizeof (fullname
), "%s@%s", osname
, snapname
);
4780 (void) snprintf(clonename
, sizeof (clonename
),
4781 "%s/ch1_%llu", osname
, (long long unsigned int)id
);
4782 (void) snprintf(tag
, sizeof (tag
), "tag_%llu", (long long unsigned int)id
);
4785 * Clean up from any previous run.
4787 error
= dsl_destroy_head(clonename
);
4788 if (error
!= ENOENT
)
4790 error
= user_release_one(fullname
, tag
);
4791 if (error
!= ESRCH
&& error
!= ENOENT
)
4793 error
= dsl_destroy_snapshot(fullname
, B_FALSE
);
4794 if (error
!= ENOENT
)
4798 * Create snapshot, clone it, mark snap for deferred destroy,
4799 * destroy clone, verify snap was also destroyed.
4801 error
= dmu_objset_snapshot_one(osname
, snapname
);
4803 if (error
== ENOSPC
) {
4804 ztest_record_enospc("dmu_objset_snapshot");
4807 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname
, error
);
4810 error
= dmu_objset_clone(clonename
, fullname
);
4812 if (error
== ENOSPC
) {
4813 ztest_record_enospc("dmu_objset_clone");
4816 fatal(0, "dmu_objset_clone(%s) = %d", clonename
, error
);
4819 error
= dsl_destroy_snapshot(fullname
, B_TRUE
);
4821 fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
4825 error
= dsl_destroy_head(clonename
);
4827 fatal(0, "dsl_destroy_head(%s) = %d", clonename
, error
);
4829 error
= dmu_objset_hold(fullname
, FTAG
, &origin
);
4830 if (error
!= ENOENT
)
4831 fatal(0, "dmu_objset_hold(%s) = %d", fullname
, error
);
4834 * Create snapshot, add temporary hold, verify that we can't
4835 * destroy a held snapshot, mark for deferred destroy,
4836 * release hold, verify snapshot was destroyed.
4838 error
= dmu_objset_snapshot_one(osname
, snapname
);
4840 if (error
== ENOSPC
) {
4841 ztest_record_enospc("dmu_objset_snapshot");
4844 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname
, error
);
4847 holds
= fnvlist_alloc();
4848 fnvlist_add_string(holds
, fullname
, tag
);
4849 error
= dsl_dataset_user_hold(holds
, 0, NULL
);
4850 fnvlist_free(holds
);
4853 fatal(0, "dsl_dataset_user_hold(%s)", fullname
, tag
);
4855 error
= dsl_destroy_snapshot(fullname
, B_FALSE
);
4856 if (error
!= EBUSY
) {
4857 fatal(0, "dsl_destroy_snapshot(%s, B_FALSE) = %d",
4861 error
= dsl_destroy_snapshot(fullname
, B_TRUE
);
4863 fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
4867 error
= user_release_one(fullname
, tag
);
4869 fatal(0, "user_release_one(%s, %s) = %d", fullname
, tag
, error
);
4871 VERIFY3U(dmu_objset_hold(fullname
, FTAG
, &origin
), ==, ENOENT
);
4874 (void) rw_exit(&ztest_name_lock
);
4878 * Inject random faults into the on-disk data.
4882 ztest_fault_inject(ztest_ds_t
*zd
, uint64_t id
)
4884 ztest_shared_t
*zs
= ztest_shared
;
4885 spa_t
*spa
= ztest_spa
;
4889 uint64_t bad
= 0x1990c0ffeedecadeull
;
4894 int bshift
= SPA_MAXBLOCKSHIFT
+ 2; /* don't scrog all labels */
4900 boolean_t islog
= B_FALSE
;
4902 path0
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
4903 pathrand
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
4905 mutex_enter(&ztest_vdev_lock
);
4906 maxfaults
= MAXFAULTS();
4907 leaves
= MAX(zs
->zs_mirrors
, 1) * ztest_opts
.zo_raidz
;
4908 mirror_save
= zs
->zs_mirrors
;
4909 mutex_exit(&ztest_vdev_lock
);
4911 ASSERT(leaves
>= 1);
4914 * We need SCL_STATE here because we're going to look at vd0->vdev_tsd.
4916 spa_config_enter(spa
, SCL_STATE
, FTAG
, RW_READER
);
4918 if (ztest_random(2) == 0) {
4920 * Inject errors on a normal data device or slog device.
4922 top
= ztest_random_vdev_top(spa
, B_TRUE
);
4923 leaf
= ztest_random(leaves
) + zs
->zs_splits
;
4926 * Generate paths to the first leaf in this top-level vdev,
4927 * and to the random leaf we selected. We'll induce transient
4928 * write failures and random online/offline activity on leaf 0,
4929 * and we'll write random garbage to the randomly chosen leaf.
4931 (void) snprintf(path0
, MAXPATHLEN
, ztest_dev_template
,
4932 ztest_opts
.zo_dir
, ztest_opts
.zo_pool
,
4933 top
* leaves
+ zs
->zs_splits
);
4934 (void) snprintf(pathrand
, MAXPATHLEN
, ztest_dev_template
,
4935 ztest_opts
.zo_dir
, ztest_opts
.zo_pool
,
4936 top
* leaves
+ leaf
);
4938 vd0
= vdev_lookup_by_path(spa
->spa_root_vdev
, path0
);
4939 if (vd0
!= NULL
&& vd0
->vdev_top
->vdev_islog
)
4942 if (vd0
!= NULL
&& maxfaults
!= 1) {
4944 * Make vd0 explicitly claim to be unreadable,
4945 * or unwriteable, or reach behind its back
4946 * and close the underlying fd. We can do this if
4947 * maxfaults == 0 because we'll fail and reexecute,
4948 * and we can do it if maxfaults >= 2 because we'll
4949 * have enough redundancy. If maxfaults == 1, the
4950 * combination of this with injection of random data
4951 * corruption below exceeds the pool's fault tolerance.
4953 vdev_file_t
*vf
= vd0
->vdev_tsd
;
4955 if (vf
!= NULL
&& ztest_random(3) == 0) {
4956 (void) close(vf
->vf_vnode
->v_fd
);
4957 vf
->vf_vnode
->v_fd
= -1;
4958 } else if (ztest_random(2) == 0) {
4959 vd0
->vdev_cant_read
= B_TRUE
;
4961 vd0
->vdev_cant_write
= B_TRUE
;
4963 guid0
= vd0
->vdev_guid
;
4967 * Inject errors on an l2cache device.
4969 spa_aux_vdev_t
*sav
= &spa
->spa_l2cache
;
4971 if (sav
->sav_count
== 0) {
4972 spa_config_exit(spa
, SCL_STATE
, FTAG
);
4975 vd0
= sav
->sav_vdevs
[ztest_random(sav
->sav_count
)];
4976 guid0
= vd0
->vdev_guid
;
4977 (void) strcpy(path0
, vd0
->vdev_path
);
4978 (void) strcpy(pathrand
, vd0
->vdev_path
);
4982 maxfaults
= INT_MAX
; /* no limit on cache devices */
4985 spa_config_exit(spa
, SCL_STATE
, FTAG
);
4988 * If we can tolerate two or more faults, or we're dealing
4989 * with a slog, randomly online/offline vd0.
4991 if ((maxfaults
>= 2 || islog
) && guid0
!= 0) {
4992 if (ztest_random(10) < 6) {
4993 int flags
= (ztest_random(2) == 0 ?
4994 ZFS_OFFLINE_TEMPORARY
: 0);
4997 * We have to grab the zs_name_lock as writer to
4998 * prevent a race between offlining a slog and
4999 * destroying a dataset. Offlining the slog will
5000 * grab a reference on the dataset which may cause
5001 * dsl_destroy_head() to fail with EBUSY thus
5002 * leaving the dataset in an inconsistent state.
5005 (void) rw_enter(&ztest_name_lock
,
5008 VERIFY(vdev_offline(spa
, guid0
, flags
) != EBUSY
);
5011 (void) rw_exit(&ztest_name_lock
);
5014 * Ideally we would like to be able to randomly
5015 * call vdev_[on|off]line without holding locks
5016 * to force unpredictable failures but the side
5017 * effects of vdev_[on|off]line prevent us from
5018 * doing so. We grab the ztest_vdev_lock here to
5019 * prevent a race between injection testing and
5022 mutex_enter(&ztest_vdev_lock
);
5023 (void) vdev_online(spa
, guid0
, 0, NULL
);
5024 mutex_exit(&ztest_vdev_lock
);
5032 * We have at least single-fault tolerance, so inject data corruption.
5034 fd
= open(pathrand
, O_RDWR
);
5036 if (fd
== -1) /* we hit a gap in the device namespace */
5039 fsize
= lseek(fd
, 0, SEEK_END
);
5041 while (--iters
!= 0) {
5042 offset
= ztest_random(fsize
/ (leaves
<< bshift
)) *
5043 (leaves
<< bshift
) + (leaf
<< bshift
) +
5044 (ztest_random(1ULL << (bshift
- 1)) & -8ULL);
5046 if (offset
>= fsize
)
5049 mutex_enter(&ztest_vdev_lock
);
5050 if (mirror_save
!= zs
->zs_mirrors
) {
5051 mutex_exit(&ztest_vdev_lock
);
5056 if (pwrite(fd
, &bad
, sizeof (bad
), offset
) != sizeof (bad
))
5057 fatal(1, "can't inject bad word at 0x%llx in %s",
5060 mutex_exit(&ztest_vdev_lock
);
5062 if (ztest_opts
.zo_verbose
>= 7)
5063 (void) printf("injected bad word into %s,"
5064 " offset 0x%llx\n", pathrand
, (u_longlong_t
)offset
);
5069 umem_free(path0
, MAXPATHLEN
);
5070 umem_free(pathrand
, MAXPATHLEN
);
5074 * Verify that DDT repair works as expected.
5077 ztest_ddt_repair(ztest_ds_t
*zd
, uint64_t id
)
5079 ztest_shared_t
*zs
= ztest_shared
;
5080 spa_t
*spa
= ztest_spa
;
5081 objset_t
*os
= zd
->zd_os
;
5083 uint64_t object
, blocksize
, txg
, pattern
, psize
;
5084 enum zio_checksum checksum
= spa_dedup_checksum(spa
);
5089 int copies
= 2 * ZIO_DEDUPDITTO_MIN
;
5092 blocksize
= ztest_random_blocksize();
5093 blocksize
= MIN(blocksize
, 2048); /* because we write so many */
5095 od
= umem_alloc(sizeof(ztest_od_t
), UMEM_NOFAIL
);
5096 ztest_od_init(od
, id
, FTAG
, 0, DMU_OT_UINT64_OTHER
, blocksize
, 0);
5098 if (ztest_object_init(zd
, od
, sizeof (ztest_od_t
), B_FALSE
) != 0) {
5099 umem_free(od
, sizeof(ztest_od_t
));
5104 * Take the name lock as writer to prevent anyone else from changing
5105 * the pool and dataset properies we need to maintain during this test.
5107 (void) rw_enter(&ztest_name_lock
, RW_WRITER
);
5109 if (ztest_dsl_prop_set_uint64(zd
->zd_name
, ZFS_PROP_DEDUP
, checksum
,
5111 ztest_dsl_prop_set_uint64(zd
->zd_name
, ZFS_PROP_COPIES
, 1,
5113 (void) rw_exit(&ztest_name_lock
);
5114 umem_free(od
, sizeof(ztest_od_t
));
5118 object
= od
[0].od_object
;
5119 blocksize
= od
[0].od_blocksize
;
5120 pattern
= zs
->zs_guid
^ dmu_objset_fsid_guid(os
);
5122 ASSERT(object
!= 0);
5124 tx
= dmu_tx_create(os
);
5125 dmu_tx_hold_write(tx
, object
, 0, copies
* blocksize
);
5126 txg
= ztest_tx_assign(tx
, TXG_WAIT
, FTAG
);
5128 (void) rw_exit(&ztest_name_lock
);
5129 umem_free(od
, sizeof(ztest_od_t
));
5134 * Write all the copies of our block.
5136 for (i
= 0; i
< copies
; i
++) {
5137 uint64_t offset
= i
* blocksize
;
5138 int error
= dmu_buf_hold(os
, object
, offset
, FTAG
, &db
,
5139 DMU_READ_NO_PREFETCH
);
5141 fatal(B_FALSE
, "dmu_buf_hold(%p, %llu, %llu) = %u",
5142 os
, (long long)object
, (long long) offset
, error
);
5144 ASSERT(db
->db_offset
== offset
);
5145 ASSERT(db
->db_size
== blocksize
);
5146 ASSERT(ztest_pattern_match(db
->db_data
, db
->db_size
, pattern
) ||
5147 ztest_pattern_match(db
->db_data
, db
->db_size
, 0ULL));
5148 dmu_buf_will_fill(db
, tx
);
5149 ztest_pattern_set(db
->db_data
, db
->db_size
, pattern
);
5150 dmu_buf_rele(db
, FTAG
);
5154 txg_wait_synced(spa_get_dsl(spa
), txg
);
5157 * Find out what block we got.
5159 VERIFY0(dmu_buf_hold(os
, object
, 0, FTAG
, &db
,
5160 DMU_READ_NO_PREFETCH
));
5161 blk
= *((dmu_buf_impl_t
*)db
)->db_blkptr
;
5162 dmu_buf_rele(db
, FTAG
);
5165 * Damage the block. Dedup-ditto will save us when we read it later.
5167 psize
= BP_GET_PSIZE(&blk
);
5168 buf
= zio_buf_alloc(psize
);
5169 ztest_pattern_set(buf
, psize
, ~pattern
);
5171 (void) zio_wait(zio_rewrite(NULL
, spa
, 0, &blk
,
5172 buf
, psize
, NULL
, NULL
, ZIO_PRIORITY_SYNC_WRITE
,
5173 ZIO_FLAG_CANFAIL
| ZIO_FLAG_INDUCE_DAMAGE
, NULL
));
5175 zio_buf_free(buf
, psize
);
5177 (void) rw_exit(&ztest_name_lock
);
5178 umem_free(od
, sizeof(ztest_od_t
));
5186 ztest_scrub(ztest_ds_t
*zd
, uint64_t id
)
5188 spa_t
*spa
= ztest_spa
;
5190 (void) spa_scan(spa
, POOL_SCAN_SCRUB
);
5191 (void) poll(NULL
, 0, 100); /* wait a moment, then force a restart */
5192 (void) spa_scan(spa
, POOL_SCAN_SCRUB
);
5196 * Change the guid for the pool.
5200 ztest_reguid(ztest_ds_t
*zd
, uint64_t id
)
5202 spa_t
*spa
= ztest_spa
;
5203 uint64_t orig
, load
;
5206 orig
= spa_guid(spa
);
5207 load
= spa_load_guid(spa
);
5209 (void) rw_enter(&ztest_name_lock
, RW_WRITER
);
5210 error
= spa_change_guid(spa
);
5211 (void) rw_exit(&ztest_name_lock
);
5216 if (ztest_opts
.zo_verbose
>= 4) {
5217 (void) printf("Changed guid old %llu -> %llu\n",
5218 (u_longlong_t
)orig
, (u_longlong_t
)spa_guid(spa
));
5221 VERIFY3U(orig
, !=, spa_guid(spa
));
5222 VERIFY3U(load
, ==, spa_load_guid(spa
));
5226 * Rename the pool to a different name and then rename it back.
5230 ztest_spa_rename(ztest_ds_t
*zd
, uint64_t id
)
5232 char *oldname
, *newname
;
5235 (void) rw_enter(&ztest_name_lock
, RW_WRITER
);
5237 oldname
= ztest_opts
.zo_pool
;
5238 newname
= umem_alloc(strlen(oldname
) + 5, UMEM_NOFAIL
);
5239 (void) strcpy(newname
, oldname
);
5240 (void) strcat(newname
, "_tmp");
5245 VERIFY3U(0, ==, spa_rename(oldname
, newname
));
5248 * Try to open it under the old name, which shouldn't exist
5250 VERIFY3U(ENOENT
, ==, spa_open(oldname
, &spa
, FTAG
));
5253 * Open it under the new name and make sure it's still the same spa_t.
5255 VERIFY3U(0, ==, spa_open(newname
, &spa
, FTAG
));
5257 ASSERT(spa
== ztest_spa
);
5258 spa_close(spa
, FTAG
);
5261 * Rename it back to the original
5263 VERIFY3U(0, ==, spa_rename(newname
, oldname
));
5266 * Make sure it can still be opened
5268 VERIFY3U(0, ==, spa_open(oldname
, &spa
, FTAG
));
5270 ASSERT(spa
== ztest_spa
);
5271 spa_close(spa
, FTAG
);
5273 umem_free(newname
, strlen(newname
) + 1);
5275 (void) rw_exit(&ztest_name_lock
);
5279 * Verify pool integrity by running zdb.
5282 ztest_run_zdb(char *pool
)
5290 bin
= umem_alloc(MAXPATHLEN
+ MAXNAMELEN
+ 20, UMEM_NOFAIL
);
5291 zdb
= umem_alloc(MAXPATHLEN
+ MAXNAMELEN
+ 20, UMEM_NOFAIL
);
5292 zbuf
= umem_alloc(1024, UMEM_NOFAIL
);
5294 VERIFY(realpath(getexecname(), bin
) != NULL
);
5295 if (strncmp(bin
, "/usr/sbin/ztest", 15) == 0) {
5296 strcpy(bin
, "/usr/sbin/zdb"); /* Installed */
5297 } else if (strncmp(bin
, "/sbin/ztest", 11) == 0) {
5298 strcpy(bin
, "/sbin/zdb"); /* Installed */
5300 strstr(bin
, "/ztest/")[0] = '\0'; /* In-tree */
5301 strcat(bin
, "/zdb/zdb");
5305 "%s -bcc%s%s -U %s %s",
5307 ztest_opts
.zo_verbose
>= 3 ? "s" : "",
5308 ztest_opts
.zo_verbose
>= 4 ? "v" : "",
5312 if (ztest_opts
.zo_verbose
>= 5)
5313 (void) printf("Executing %s\n", strstr(zdb
, "zdb "));
5315 fp
= popen(zdb
, "r");
5317 while (fgets(zbuf
, 1024, fp
) != NULL
)
5318 if (ztest_opts
.zo_verbose
>= 3)
5319 (void) printf("%s", zbuf
);
5321 status
= pclose(fp
);
5326 ztest_dump_core
= 0;
5327 if (WIFEXITED(status
))
5328 fatal(0, "'%s' exit code %d", zdb
, WEXITSTATUS(status
));
5330 fatal(0, "'%s' died with signal %d", zdb
, WTERMSIG(status
));
5332 umem_free(bin
, MAXPATHLEN
+ MAXNAMELEN
+ 20);
5333 umem_free(zdb
, MAXPATHLEN
+ MAXNAMELEN
+ 20);
5334 umem_free(zbuf
, 1024);
5338 ztest_walk_pool_directory(char *header
)
5342 if (ztest_opts
.zo_verbose
>= 6)
5343 (void) printf("%s\n", header
);
5345 mutex_enter(&spa_namespace_lock
);
5346 while ((spa
= spa_next(spa
)) != NULL
)
5347 if (ztest_opts
.zo_verbose
>= 6)
5348 (void) printf("\t%s\n", spa_name(spa
));
5349 mutex_exit(&spa_namespace_lock
);
5353 ztest_spa_import_export(char *oldname
, char *newname
)
5355 nvlist_t
*config
, *newconfig
;
5360 if (ztest_opts
.zo_verbose
>= 4) {
5361 (void) printf("import/export: old = %s, new = %s\n",
5366 * Clean up from previous runs.
5368 (void) spa_destroy(newname
);
5371 * Get the pool's configuration and guid.
5373 VERIFY3U(0, ==, spa_open(oldname
, &spa
, FTAG
));
5376 * Kick off a scrub to tickle scrub/export races.
5378 if (ztest_random(2) == 0)
5379 (void) spa_scan(spa
, POOL_SCAN_SCRUB
);
5381 pool_guid
= spa_guid(spa
);
5382 spa_close(spa
, FTAG
);
5384 ztest_walk_pool_directory("pools before export");
5389 VERIFY3U(0, ==, spa_export(oldname
, &config
, B_FALSE
, B_FALSE
));
5391 ztest_walk_pool_directory("pools after export");
5396 newconfig
= spa_tryimport(config
);
5397 ASSERT(newconfig
!= NULL
);
5398 nvlist_free(newconfig
);
5401 * Import it under the new name.
5403 error
= spa_import(newname
, config
, NULL
, 0);
5405 dump_nvlist(config
, 0);
5406 fatal(B_FALSE
, "couldn't import pool %s as %s: error %u",
5407 oldname
, newname
, error
);
5410 ztest_walk_pool_directory("pools after import");
5413 * Try to import it again -- should fail with EEXIST.
5415 VERIFY3U(EEXIST
, ==, spa_import(newname
, config
, NULL
, 0));
5418 * Try to import it under a different name -- should fail with EEXIST.
5420 VERIFY3U(EEXIST
, ==, spa_import(oldname
, config
, NULL
, 0));
5423 * Verify that the pool is no longer visible under the old name.
5425 VERIFY3U(ENOENT
, ==, spa_open(oldname
, &spa
, FTAG
));
5428 * Verify that we can open and close the pool using the new name.
5430 VERIFY3U(0, ==, spa_open(newname
, &spa
, FTAG
));
5431 ASSERT(pool_guid
== spa_guid(spa
));
5432 spa_close(spa
, FTAG
);
5434 nvlist_free(config
);
5438 ztest_resume(spa_t
*spa
)
5440 if (spa_suspended(spa
) && ztest_opts
.zo_verbose
>= 6)
5441 (void) printf("resuming from suspended state\n");
5442 spa_vdev_state_enter(spa
, SCL_NONE
);
5443 vdev_clear(spa
, NULL
);
5444 (void) spa_vdev_state_exit(spa
, NULL
, 0);
5445 (void) zio_resume(spa
);
5449 ztest_resume_thread(void *arg
)
5453 while (!ztest_exiting
) {
5454 if (spa_suspended(spa
))
5456 (void) poll(NULL
, 0, 100);
5468 ztest_deadman_alarm(int sig
)
5470 fatal(0, "failed to complete within %d seconds of deadline", GRACE
);
5475 ztest_execute(int test
, ztest_info_t
*zi
, uint64_t id
)
5477 ztest_ds_t
*zd
= &ztest_ds
[id
% ztest_opts
.zo_datasets
];
5478 ztest_shared_callstate_t
*zc
= ZTEST_GET_SHARED_CALLSTATE(test
);
5479 hrtime_t functime
= gethrtime();
5482 for (i
= 0; i
< zi
->zi_iters
; i
++)
5483 zi
->zi_func(zd
, id
);
5485 functime
= gethrtime() - functime
;
5487 atomic_add_64(&zc
->zc_count
, 1);
5488 atomic_add_64(&zc
->zc_time
, functime
);
5490 if (ztest_opts
.zo_verbose
>= 4) {
5492 (void) dladdr((void *)zi
->zi_func
, &dli
);
5493 (void) printf("%6.2f sec in %s\n",
5494 (double)functime
/ NANOSEC
, dli
.dli_sname
);
5499 ztest_thread(void *arg
)
5502 uint64_t id
= (uintptr_t)arg
;
5503 ztest_shared_t
*zs
= ztest_shared
;
5507 ztest_shared_callstate_t
*zc
;
5509 while ((now
= gethrtime()) < zs
->zs_thread_stop
) {
5511 * See if it's time to force a crash.
5513 if (now
> zs
->zs_thread_kill
)
5517 * If we're getting ENOSPC with some regularity, stop.
5519 if (zs
->zs_enospc_count
> 10)
5523 * Pick a random function to execute.
5525 rand
= ztest_random(ZTEST_FUNCS
);
5526 zi
= &ztest_info
[rand
];
5527 zc
= ZTEST_GET_SHARED_CALLSTATE(rand
);
5528 call_next
= zc
->zc_next
;
5530 if (now
>= call_next
&&
5531 atomic_cas_64(&zc
->zc_next
, call_next
, call_next
+
5532 ztest_random(2 * zi
->zi_interval
[0] + 1)) == call_next
) {
5533 ztest_execute(rand
, zi
, id
);
5543 ztest_dataset_name(char *dsname
, char *pool
, int d
)
5545 (void) snprintf(dsname
, MAXNAMELEN
, "%s/ds_%d", pool
, d
);
5549 ztest_dataset_destroy(int d
)
5551 char name
[MAXNAMELEN
];
5554 ztest_dataset_name(name
, ztest_opts
.zo_pool
, d
);
5556 if (ztest_opts
.zo_verbose
>= 3)
5557 (void) printf("Destroying %s to free up space\n", name
);
5560 * Cleanup any non-standard clones and snapshots. In general,
5561 * ztest thread t operates on dataset (t % zopt_datasets),
5562 * so there may be more than one thing to clean up.
5564 for (t
= d
; t
< ztest_opts
.zo_threads
;
5565 t
+= ztest_opts
.zo_datasets
)
5566 ztest_dsl_dataset_cleanup(name
, t
);
5568 (void) dmu_objset_find(name
, ztest_objset_destroy_cb
, NULL
,
5569 DS_FIND_SNAPSHOTS
| DS_FIND_CHILDREN
);
5573 ztest_dataset_dirobj_verify(ztest_ds_t
*zd
)
5575 uint64_t usedobjs
, dirobjs
, scratch
;
5578 * ZTEST_DIROBJ is the object directory for the entire dataset.
5579 * Therefore, the number of objects in use should equal the
5580 * number of ZTEST_DIROBJ entries, +1 for ZTEST_DIROBJ itself.
5581 * If not, we have an object leak.
5583 * Note that we can only check this in ztest_dataset_open(),
5584 * when the open-context and syncing-context values agree.
5585 * That's because zap_count() returns the open-context value,
5586 * while dmu_objset_space() returns the rootbp fill count.
5588 VERIFY3U(0, ==, zap_count(zd
->zd_os
, ZTEST_DIROBJ
, &dirobjs
));
5589 dmu_objset_space(zd
->zd_os
, &scratch
, &scratch
, &usedobjs
, &scratch
);
5590 ASSERT3U(dirobjs
+ 1, ==, usedobjs
);
5594 ztest_dataset_open(int d
)
5596 ztest_ds_t
*zd
= &ztest_ds
[d
];
5597 uint64_t committed_seq
= ZTEST_GET_SHARED_DS(d
)->zd_seq
;
5600 char name
[MAXNAMELEN
];
5603 ztest_dataset_name(name
, ztest_opts
.zo_pool
, d
);
5605 (void) rw_enter(&ztest_name_lock
, RW_READER
);
5607 error
= ztest_dataset_create(name
);
5608 if (error
== ENOSPC
) {
5609 (void) rw_exit(&ztest_name_lock
);
5610 ztest_record_enospc(FTAG
);
5613 ASSERT(error
== 0 || error
== EEXIST
);
5615 VERIFY0(dmu_objset_own(name
, DMU_OST_OTHER
, B_FALSE
, zd
, &os
));
5616 (void) rw_exit(&ztest_name_lock
);
5618 ztest_zd_init(zd
, ZTEST_GET_SHARED_DS(d
), os
);
5620 zilog
= zd
->zd_zilog
;
5622 if (zilog
->zl_header
->zh_claim_lr_seq
!= 0 &&
5623 zilog
->zl_header
->zh_claim_lr_seq
< committed_seq
)
5624 fatal(0, "missing log records: claimed %llu < committed %llu",
5625 zilog
->zl_header
->zh_claim_lr_seq
, committed_seq
);
5627 ztest_dataset_dirobj_verify(zd
);
5629 zil_replay(os
, zd
, ztest_replay_vector
);
5631 ztest_dataset_dirobj_verify(zd
);
5633 if (ztest_opts
.zo_verbose
>= 6)
5634 (void) printf("%s replay %llu blocks, %llu records, seq %llu\n",
5636 (u_longlong_t
)zilog
->zl_parse_blk_count
,
5637 (u_longlong_t
)zilog
->zl_parse_lr_count
,
5638 (u_longlong_t
)zilog
->zl_replaying_seq
);
5640 zilog
= zil_open(os
, ztest_get_data
);
5642 if (zilog
->zl_replaying_seq
!= 0 &&
5643 zilog
->zl_replaying_seq
< committed_seq
)
5644 fatal(0, "missing log records: replayed %llu < committed %llu",
5645 zilog
->zl_replaying_seq
, committed_seq
);
5651 ztest_dataset_close(int d
)
5653 ztest_ds_t
*zd
= &ztest_ds
[d
];
5655 zil_close(zd
->zd_zilog
);
5656 dmu_objset_disown(zd
->zd_os
, zd
);
5662 * Kick off threads to run tests on all datasets in parallel.
5665 ztest_run(ztest_shared_t
*zs
)
5670 kthread_t
*resume_thread
;
5675 ztest_exiting
= B_FALSE
;
5678 * Initialize parent/child shared state.
5680 mutex_init(&ztest_vdev_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
5681 rw_init(&ztest_name_lock
, NULL
, RW_DEFAULT
, NULL
);
5683 zs
->zs_thread_start
= gethrtime();
5684 zs
->zs_thread_stop
=
5685 zs
->zs_thread_start
+ ztest_opts
.zo_passtime
* NANOSEC
;
5686 zs
->zs_thread_stop
= MIN(zs
->zs_thread_stop
, zs
->zs_proc_stop
);
5687 zs
->zs_thread_kill
= zs
->zs_thread_stop
;
5688 if (ztest_random(100) < ztest_opts
.zo_killrate
) {
5689 zs
->zs_thread_kill
-=
5690 ztest_random(ztest_opts
.zo_passtime
* NANOSEC
);
5693 mutex_init(&zcl
.zcl_callbacks_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
5695 list_create(&zcl
.zcl_callbacks
, sizeof (ztest_cb_data_t
),
5696 offsetof(ztest_cb_data_t
, zcd_node
));
5701 kernel_init(FREAD
| FWRITE
);
5702 VERIFY0(spa_open(ztest_opts
.zo_pool
, &spa
, FTAG
));
5703 spa
->spa_debug
= B_TRUE
;
5706 VERIFY0(dmu_objset_own(ztest_opts
.zo_pool
,
5707 DMU_OST_ANY
, B_TRUE
, FTAG
, &os
));
5708 zs
->zs_guid
= dmu_objset_fsid_guid(os
);
5709 dmu_objset_disown(os
, FTAG
);
5711 spa
->spa_dedup_ditto
= 2 * ZIO_DEDUPDITTO_MIN
;
5714 * We don't expect the pool to suspend unless maxfaults == 0,
5715 * in which case ztest_fault_inject() temporarily takes away
5716 * the only valid replica.
5718 if (MAXFAULTS() == 0)
5719 spa
->spa_failmode
= ZIO_FAILURE_MODE_WAIT
;
5721 spa
->spa_failmode
= ZIO_FAILURE_MODE_PANIC
;
5724 * Create a thread to periodically resume suspended I/O.
5726 VERIFY3P((resume_thread
= zk_thread_create(NULL
, 0,
5727 (thread_func_t
)ztest_resume_thread
, spa
, TS_RUN
, NULL
, 0, 0,
5728 PTHREAD_CREATE_JOINABLE
)), !=, NULL
);
5732 * Set a deadman alarm to abort() if we hang.
5734 signal(SIGALRM
, ztest_deadman_alarm
);
5735 alarm((zs
->zs_thread_stop
- zs
->zs_thread_start
) / NANOSEC
+ GRACE
);
5739 * Verify that we can safely inquire about about any object,
5740 * whether it's allocated or not. To make it interesting,
5741 * we probe a 5-wide window around each power of two.
5742 * This hits all edge cases, including zero and the max.
5744 for (t
= 0; t
< 64; t
++) {
5745 for (d
= -5; d
<= 5; d
++) {
5746 error
= dmu_object_info(spa
->spa_meta_objset
,
5747 (1ULL << t
) + d
, NULL
);
5748 ASSERT(error
== 0 || error
== ENOENT
||
5754 * If we got any ENOSPC errors on the previous run, destroy something.
5756 if (zs
->zs_enospc_count
!= 0) {
5757 int d
= ztest_random(ztest_opts
.zo_datasets
);
5758 ztest_dataset_destroy(d
);
5760 zs
->zs_enospc_count
= 0;
5762 tid
= umem_zalloc(ztest_opts
.zo_threads
* sizeof (kt_did_t
),
5765 if (ztest_opts
.zo_verbose
>= 4)
5766 (void) printf("starting main threads...\n");
5769 * Kick off all the tests that run in parallel.
5771 for (t
= 0; t
< ztest_opts
.zo_threads
; t
++) {
5774 if (t
< ztest_opts
.zo_datasets
&&
5775 ztest_dataset_open(t
) != 0)
5778 VERIFY3P(thread
= zk_thread_create(NULL
, 0,
5779 (thread_func_t
)ztest_thread
,
5780 (void *)(uintptr_t)t
, TS_RUN
, NULL
, 0, 0,
5781 PTHREAD_CREATE_JOINABLE
), !=, NULL
);
5782 tid
[t
] = thread
->t_tid
;
5786 * Wait for all of the tests to complete. We go in reverse order
5787 * so we don't close datasets while threads are still using them.
5789 for (t
= ztest_opts
.zo_threads
- 1; t
>= 0; t
--) {
5790 thread_join(tid
[t
]);
5791 if (t
< ztest_opts
.zo_datasets
)
5792 ztest_dataset_close(t
);
5795 txg_wait_synced(spa_get_dsl(spa
), 0);
5797 zs
->zs_alloc
= metaslab_class_get_alloc(spa_normal_class(spa
));
5798 zs
->zs_space
= metaslab_class_get_space(spa_normal_class(spa
));
5800 umem_free(tid
, ztest_opts
.zo_threads
* sizeof (kt_did_t
));
5802 /* Kill the resume thread */
5803 ztest_exiting
= B_TRUE
;
5804 thread_join(resume_thread
->t_tid
);
5808 * Right before closing the pool, kick off a bunch of async I/O;
5809 * spa_close() should wait for it to complete.
5811 for (object
= 1; object
< 50; object
++)
5812 dmu_prefetch(spa
->spa_meta_objset
, object
, 0, 1ULL << 20);
5814 /* Verify that at least one commit cb was called in a timely fashion */
5815 if (zc_cb_counter
>= ZTEST_COMMIT_CB_MIN_REG
)
5816 VERIFY0(zc_min_txg_delay
);
5818 spa_close(spa
, FTAG
);
5821 * Verify that we can loop over all pools.
5823 mutex_enter(&spa_namespace_lock
);
5824 for (spa
= spa_next(NULL
); spa
!= NULL
; spa
= spa_next(spa
))
5825 if (ztest_opts
.zo_verbose
> 3)
5826 (void) printf("spa_next: found %s\n", spa_name(spa
));
5827 mutex_exit(&spa_namespace_lock
);
5830 * Verify that we can export the pool and reimport it under a
5833 if (ztest_random(2) == 0) {
5834 char name
[MAXNAMELEN
];
5835 (void) snprintf(name
, MAXNAMELEN
, "%s_import",
5836 ztest_opts
.zo_pool
);
5837 ztest_spa_import_export(ztest_opts
.zo_pool
, name
);
5838 ztest_spa_import_export(name
, ztest_opts
.zo_pool
);
5843 list_destroy(&zcl
.zcl_callbacks
);
5844 mutex_destroy(&zcl
.zcl_callbacks_lock
);
5845 rw_destroy(&ztest_name_lock
);
5846 mutex_destroy(&ztest_vdev_lock
);
5852 ztest_ds_t
*zd
= &ztest_ds
[0];
5856 if (ztest_opts
.zo_verbose
>= 3)
5857 (void) printf("testing spa_freeze()...\n");
5859 kernel_init(FREAD
| FWRITE
);
5860 VERIFY3U(0, ==, spa_open(ztest_opts
.zo_pool
, &spa
, FTAG
));
5861 VERIFY3U(0, ==, ztest_dataset_open(0));
5862 spa
->spa_debug
= B_TRUE
;
5866 * Force the first log block to be transactionally allocated.
5867 * We have to do this before we freeze the pool -- otherwise
5868 * the log chain won't be anchored.
5870 while (BP_IS_HOLE(&zd
->zd_zilog
->zl_header
->zh_log
)) {
5871 ztest_dmu_object_alloc_free(zd
, 0);
5872 zil_commit(zd
->zd_zilog
, 0);
5875 txg_wait_synced(spa_get_dsl(spa
), 0);
5878 * Freeze the pool. This stops spa_sync() from doing anything,
5879 * so that the only way to record changes from now on is the ZIL.
5884 * Run tests that generate log records but don't alter the pool config
5885 * or depend on DSL sync tasks (snapshots, objset create/destroy, etc).
5886 * We do a txg_wait_synced() after each iteration to force the txg
5887 * to increase well beyond the last synced value in the uberblock.
5888 * The ZIL should be OK with that.
5890 while (ztest_random(10) != 0 &&
5891 numloops
++ < ztest_opts
.zo_maxloops
) {
5892 ztest_dmu_write_parallel(zd
, 0);
5893 ztest_dmu_object_alloc_free(zd
, 0);
5894 txg_wait_synced(spa_get_dsl(spa
), 0);
5898 * Commit all of the changes we just generated.
5900 zil_commit(zd
->zd_zilog
, 0);
5901 txg_wait_synced(spa_get_dsl(spa
), 0);
5904 * Close our dataset and close the pool.
5906 ztest_dataset_close(0);
5907 spa_close(spa
, FTAG
);
5911 * Open and close the pool and dataset to induce log replay.
5913 kernel_init(FREAD
| FWRITE
);
5914 VERIFY3U(0, ==, spa_open(ztest_opts
.zo_pool
, &spa
, FTAG
));
5915 ASSERT(spa_freeze_txg(spa
) == UINT64_MAX
);
5916 VERIFY3U(0, ==, ztest_dataset_open(0));
5917 ztest_dataset_close(0);
5919 spa
->spa_debug
= B_TRUE
;
5921 txg_wait_synced(spa_get_dsl(spa
), 0);
5922 ztest_reguid(NULL
, 0);
5924 spa_close(spa
, FTAG
);
5929 print_time(hrtime_t t
, char *timebuf
)
5931 hrtime_t s
= t
/ NANOSEC
;
5932 hrtime_t m
= s
/ 60;
5933 hrtime_t h
= m
/ 60;
5934 hrtime_t d
= h
/ 24;
5943 (void) sprintf(timebuf
,
5944 "%llud%02lluh%02llum%02llus", d
, h
, m
, s
);
5946 (void) sprintf(timebuf
, "%lluh%02llum%02llus", h
, m
, s
);
5948 (void) sprintf(timebuf
, "%llum%02llus", m
, s
);
5950 (void) sprintf(timebuf
, "%llus", s
);
5954 make_random_props(void)
5958 VERIFY(nvlist_alloc(&props
, NV_UNIQUE_NAME
, 0) == 0);
5959 if (ztest_random(2) == 0)
5961 VERIFY(nvlist_add_uint64(props
, "autoreplace", 1) == 0);
5967 * Create a storage pool with the given name and initial vdev size.
5968 * Then test spa_freeze() functionality.
5971 ztest_init(ztest_shared_t
*zs
)
5974 nvlist_t
*nvroot
, *props
;
5977 mutex_init(&ztest_vdev_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
5978 rw_init(&ztest_name_lock
, NULL
, RW_DEFAULT
, NULL
);
5980 kernel_init(FREAD
| FWRITE
);
5983 * Create the storage pool.
5985 (void) spa_destroy(ztest_opts
.zo_pool
);
5986 ztest_shared
->zs_vdev_next_leaf
= 0;
5988 zs
->zs_mirrors
= ztest_opts
.zo_mirrors
;
5989 nvroot
= make_vdev_root(NULL
, NULL
, NULL
, ztest_opts
.zo_vdev_size
, 0,
5990 0, ztest_opts
.zo_raidz
, zs
->zs_mirrors
, 1);
5991 props
= make_random_props();
5992 for (i
= 0; i
< SPA_FEATURES
; i
++) {
5994 VERIFY3S(-1, !=, asprintf(&buf
, "feature@%s",
5995 spa_feature_table
[i
].fi_uname
));
5996 VERIFY3U(0, ==, nvlist_add_uint64(props
, buf
, 0));
5999 VERIFY3U(0, ==, spa_create(ztest_opts
.zo_pool
, nvroot
, props
, NULL
));
6000 nvlist_free(nvroot
);
6002 VERIFY3U(0, ==, spa_open(ztest_opts
.zo_pool
, &spa
, FTAG
));
6003 zs
->zs_metaslab_sz
=
6004 1ULL << spa
->spa_root_vdev
->vdev_child
[0]->vdev_ms_shift
;
6005 spa_close(spa
, FTAG
);
6009 ztest_run_zdb(ztest_opts
.zo_pool
);
6013 ztest_run_zdb(ztest_opts
.zo_pool
);
6015 rw_destroy(&ztest_name_lock
);
6016 mutex_destroy(&ztest_vdev_lock
);
6022 static char ztest_name_data
[] = "/tmp/ztest.data.XXXXXX";
6024 ztest_fd_data
= mkstemp(ztest_name_data
);
6025 ASSERT3S(ztest_fd_data
, >=, 0);
6026 (void) unlink(ztest_name_data
);
6030 shared_data_size(ztest_shared_hdr_t
*hdr
)
6034 size
= hdr
->zh_hdr_size
;
6035 size
+= hdr
->zh_opts_size
;
6036 size
+= hdr
->zh_size
;
6037 size
+= hdr
->zh_stats_size
* hdr
->zh_stats_count
;
6038 size
+= hdr
->zh_ds_size
* hdr
->zh_ds_count
;
6047 ztest_shared_hdr_t
*hdr
;
6049 hdr
= (void *)mmap(0, P2ROUNDUP(sizeof (*hdr
), getpagesize()),
6050 PROT_READ
| PROT_WRITE
, MAP_SHARED
, ztest_fd_data
, 0);
6051 ASSERT(hdr
!= MAP_FAILED
);
6053 VERIFY3U(0, ==, ftruncate(ztest_fd_data
, sizeof (ztest_shared_hdr_t
)));
6055 hdr
->zh_hdr_size
= sizeof (ztest_shared_hdr_t
);
6056 hdr
->zh_opts_size
= sizeof (ztest_shared_opts_t
);
6057 hdr
->zh_size
= sizeof (ztest_shared_t
);
6058 hdr
->zh_stats_size
= sizeof (ztest_shared_callstate_t
);
6059 hdr
->zh_stats_count
= ZTEST_FUNCS
;
6060 hdr
->zh_ds_size
= sizeof (ztest_shared_ds_t
);
6061 hdr
->zh_ds_count
= ztest_opts
.zo_datasets
;
6063 size
= shared_data_size(hdr
);
6064 VERIFY3U(0, ==, ftruncate(ztest_fd_data
, size
));
6066 (void) munmap((caddr_t
)hdr
, P2ROUNDUP(sizeof (*hdr
), getpagesize()));
6073 ztest_shared_hdr_t
*hdr
;
6076 hdr
= (void *)mmap(0, P2ROUNDUP(sizeof (*hdr
), getpagesize()),
6077 PROT_READ
, MAP_SHARED
, ztest_fd_data
, 0);
6078 ASSERT(hdr
!= MAP_FAILED
);
6080 size
= shared_data_size(hdr
);
6082 (void) munmap((caddr_t
)hdr
, P2ROUNDUP(sizeof (*hdr
), getpagesize()));
6083 hdr
= ztest_shared_hdr
= (void *)mmap(0, P2ROUNDUP(size
, getpagesize()),
6084 PROT_READ
| PROT_WRITE
, MAP_SHARED
, ztest_fd_data
, 0);
6085 ASSERT(hdr
!= MAP_FAILED
);
6086 buf
= (uint8_t *)hdr
;
6088 offset
= hdr
->zh_hdr_size
;
6089 ztest_shared_opts
= (void *)&buf
[offset
];
6090 offset
+= hdr
->zh_opts_size
;
6091 ztest_shared
= (void *)&buf
[offset
];
6092 offset
+= hdr
->zh_size
;
6093 ztest_shared_callstate
= (void *)&buf
[offset
];
6094 offset
+= hdr
->zh_stats_size
* hdr
->zh_stats_count
;
6095 ztest_shared_ds
= (void *)&buf
[offset
];
6099 exec_child(char *cmd
, char *libpath
, boolean_t ignorekill
, int *statusp
)
6103 char *cmdbuf
= NULL
;
6108 cmdbuf
= umem_alloc(MAXPATHLEN
, UMEM_NOFAIL
);
6109 (void) strlcpy(cmdbuf
, getexecname(), MAXPATHLEN
);
6114 fatal(1, "fork failed");
6116 if (pid
== 0) { /* child */
6117 char *emptyargv
[2] = { cmd
, NULL
};
6118 char fd_data_str
[12];
6120 struct rlimit rl
= { 1024, 1024 };
6121 (void) setrlimit(RLIMIT_NOFILE
, &rl
);
6123 (void) close(ztest_fd_rand
);
6124 VERIFY(11 >= snprintf(fd_data_str
, 12, "%d", ztest_fd_data
));
6125 VERIFY(0 == setenv("ZTEST_FD_DATA", fd_data_str
, 1));
6127 (void) enable_extended_FILE_stdio(-1, -1);
6128 if (libpath
!= NULL
)
6129 VERIFY(0 == setenv("LD_LIBRARY_PATH", libpath
, 1));
6130 (void) execv(cmd
, emptyargv
);
6131 ztest_dump_core
= B_FALSE
;
6132 fatal(B_TRUE
, "exec failed: %s", cmd
);
6135 if (cmdbuf
!= NULL
) {
6136 umem_free(cmdbuf
, MAXPATHLEN
);
6140 while (waitpid(pid
, &status
, 0) != pid
)
6142 if (statusp
!= NULL
)
6145 if (WIFEXITED(status
)) {
6146 if (WEXITSTATUS(status
) != 0) {
6147 (void) fprintf(stderr
, "child exited with code %d\n",
6148 WEXITSTATUS(status
));
6152 } else if (WIFSIGNALED(status
)) {
6153 if (!ignorekill
|| WTERMSIG(status
) != SIGKILL
) {
6154 (void) fprintf(stderr
, "child died with signal %d\n",
6160 (void) fprintf(stderr
, "something strange happened to child\n");
6167 ztest_run_init(void)
6171 ztest_shared_t
*zs
= ztest_shared
;
6173 ASSERT(ztest_opts
.zo_init
!= 0);
6176 * Blow away any existing copy of zpool.cache
6178 (void) remove(spa_config_path
);
6181 * Create and initialize our storage pool.
6183 for (i
= 1; i
<= ztest_opts
.zo_init
; i
++) {
6184 bzero(zs
, sizeof (ztest_shared_t
));
6185 if (ztest_opts
.zo_verbose
>= 3 &&
6186 ztest_opts
.zo_init
!= 1) {
6187 (void) printf("ztest_init(), pass %d\n", i
);
6194 main(int argc
, char **argv
)
6202 ztest_shared_callstate_t
*zc
;
6209 char *fd_data_str
= getenv("ZTEST_FD_DATA");
6211 (void) setvbuf(stdout
, NULL
, _IOLBF
, 0);
6213 ztest_fd_rand
= open("/dev/urandom", O_RDONLY
);
6214 ASSERT3S(ztest_fd_rand
, >=, 0);
6217 dprintf_setup(&argc
, argv
);
6218 process_options(argc
, argv
);
6223 bcopy(&ztest_opts
, ztest_shared_opts
,
6224 sizeof (*ztest_shared_opts
));
6226 ztest_fd_data
= atoi(fd_data_str
);
6228 bcopy(ztest_shared_opts
, &ztest_opts
, sizeof (ztest_opts
));
6230 ASSERT3U(ztest_opts
.zo_datasets
, ==, ztest_shared_hdr
->zh_ds_count
);
6232 /* Override location of zpool.cache */
6233 VERIFY(asprintf((char **)&spa_config_path
, "%s/zpool.cache",
6234 ztest_opts
.zo_dir
) != -1);
6236 ztest_ds
= umem_alloc(ztest_opts
.zo_datasets
* sizeof (ztest_ds_t
),
6241 metaslab_gang_bang
= ztest_opts
.zo_metaslab_gang_bang
;
6242 metaslab_df_alloc_threshold
=
6243 zs
->zs_metaslab_df_alloc_threshold
;
6252 hasalt
= (strlen(ztest_opts
.zo_alt_ztest
) != 0);
6254 if (ztest_opts
.zo_verbose
>= 1) {
6255 (void) printf("%llu vdevs, %d datasets, %d threads,"
6256 " %llu seconds...\n",
6257 (u_longlong_t
)ztest_opts
.zo_vdevs
,
6258 ztest_opts
.zo_datasets
,
6259 ztest_opts
.zo_threads
,
6260 (u_longlong_t
)ztest_opts
.zo_time
);
6263 cmd
= umem_alloc(MAXNAMELEN
, UMEM_NOFAIL
);
6264 (void) strlcpy(cmd
, getexecname(), MAXNAMELEN
);
6266 zs
->zs_do_init
= B_TRUE
;
6267 if (strlen(ztest_opts
.zo_alt_ztest
) != 0) {
6268 if (ztest_opts
.zo_verbose
>= 1) {
6269 (void) printf("Executing older ztest for "
6270 "initialization: %s\n", ztest_opts
.zo_alt_ztest
);
6272 VERIFY(!exec_child(ztest_opts
.zo_alt_ztest
,
6273 ztest_opts
.zo_alt_libpath
, B_FALSE
, NULL
));
6275 VERIFY(!exec_child(NULL
, NULL
, B_FALSE
, NULL
));
6277 zs
->zs_do_init
= B_FALSE
;
6279 zs
->zs_proc_start
= gethrtime();
6280 zs
->zs_proc_stop
= zs
->zs_proc_start
+ ztest_opts
.zo_time
* NANOSEC
;
6282 for (f
= 0; f
< ZTEST_FUNCS
; f
++) {
6283 zi
= &ztest_info
[f
];
6284 zc
= ZTEST_GET_SHARED_CALLSTATE(f
);
6285 if (zs
->zs_proc_start
+ zi
->zi_interval
[0] > zs
->zs_proc_stop
)
6286 zc
->zc_next
= UINT64_MAX
;
6288 zc
->zc_next
= zs
->zs_proc_start
+
6289 ztest_random(2 * zi
->zi_interval
[0] + 1);
6293 * Run the tests in a loop. These tests include fault injection
6294 * to verify that self-healing data works, and forced crashes
6295 * to verify that we never lose on-disk consistency.
6297 while (gethrtime() < zs
->zs_proc_stop
) {
6302 * Initialize the workload counters for each function.
6304 for (f
= 0; f
< ZTEST_FUNCS
; f
++) {
6305 zc
= ZTEST_GET_SHARED_CALLSTATE(f
);
6310 /* Set the allocation switch size */
6311 zs
->zs_metaslab_df_alloc_threshold
=
6312 ztest_random(zs
->zs_metaslab_sz
/ 4) + 1;
6314 if (!hasalt
|| ztest_random(2) == 0) {
6315 if (hasalt
&& ztest_opts
.zo_verbose
>= 1) {
6316 (void) printf("Executing newer ztest: %s\n",
6320 killed
= exec_child(cmd
, NULL
, B_TRUE
, &status
);
6322 if (hasalt
&& ztest_opts
.zo_verbose
>= 1) {
6323 (void) printf("Executing older ztest: %s\n",
6324 ztest_opts
.zo_alt_ztest
);
6327 killed
= exec_child(ztest_opts
.zo_alt_ztest
,
6328 ztest_opts
.zo_alt_libpath
, B_TRUE
, &status
);
6335 if (ztest_opts
.zo_verbose
>= 1) {
6336 hrtime_t now
= gethrtime();
6338 now
= MIN(now
, zs
->zs_proc_stop
);
6339 print_time(zs
->zs_proc_stop
- now
, timebuf
);
6340 nicenum(zs
->zs_space
, numbuf
);
6342 (void) printf("Pass %3d, %8s, %3llu ENOSPC, "
6343 "%4.1f%% of %5s used, %3.0f%% done, %8s to go\n",
6345 WIFEXITED(status
) ? "Complete" : "SIGKILL",
6346 (u_longlong_t
)zs
->zs_enospc_count
,
6347 100.0 * zs
->zs_alloc
/ zs
->zs_space
,
6349 100.0 * (now
- zs
->zs_proc_start
) /
6350 (ztest_opts
.zo_time
* NANOSEC
), timebuf
);
6353 if (ztest_opts
.zo_verbose
>= 2) {
6354 (void) printf("\nWorkload summary:\n\n");
6355 (void) printf("%7s %9s %s\n",
6356 "Calls", "Time", "Function");
6357 (void) printf("%7s %9s %s\n",
6358 "-----", "----", "--------");
6359 for (f
= 0; f
< ZTEST_FUNCS
; f
++) {
6362 zi
= &ztest_info
[f
];
6363 zc
= ZTEST_GET_SHARED_CALLSTATE(f
);
6364 print_time(zc
->zc_time
, timebuf
);
6365 (void) dladdr((void *)zi
->zi_func
, &dli
);
6366 (void) printf("%7llu %9s %s\n",
6367 (u_longlong_t
)zc
->zc_count
, timebuf
,
6370 (void) printf("\n");
6374 * It's possible that we killed a child during a rename test,
6375 * in which case we'll have a 'ztest_tmp' pool lying around
6376 * instead of 'ztest'. Do a blind rename in case this happened.
6379 if (spa_open(ztest_opts
.zo_pool
, &spa
, FTAG
) == 0) {
6380 spa_close(spa
, FTAG
);
6382 char tmpname
[MAXNAMELEN
];
6384 kernel_init(FREAD
| FWRITE
);
6385 (void) snprintf(tmpname
, sizeof (tmpname
), "%s_tmp",
6386 ztest_opts
.zo_pool
);
6387 (void) spa_rename(tmpname
, ztest_opts
.zo_pool
);
6391 ztest_run_zdb(ztest_opts
.zo_pool
);
6394 if (ztest_opts
.zo_verbose
>= 1) {
6396 (void) printf("%d runs of older ztest: %s\n", older
,
6397 ztest_opts
.zo_alt_ztest
);
6398 (void) printf("%d runs of newer ztest: %s\n", newer
,
6401 (void) printf("%d killed, %d completed, %.0f%% kill rate\n",
6402 kills
, iters
- kills
, (100.0 * kills
) / MAX(1, iters
));
6405 umem_free(cmd
, MAXNAMELEN
);