]> git.proxmox.com Git - mirror_zfs.git/blob - cmd/ztest/ztest.c
projects / mirror_zfs.git / blob
? search:


eed92ec72ebb5546bf75d1f3d2a21c036067b68b
[mirror_zfs.git] / cmd / ztest / ztest.c
1 /*
2 * CDDL HEADER START
3 *
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.
7 *
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.
12 *
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]
18 *
19 * CDDL HEADER END
20 */
21 /*
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 */
24
25 /*
26 * The objective of this program is to provide a DMU/ZAP/SPA stress test
27 * that runs entirely in userland, is easy to use, and easy to extend.
28 *
29 * The overall design of the ztest program is as follows:
30 *
31 * (1) For each major functional area (e.g. adding vdevs to a pool,
32 * creating and destroying datasets, reading and writing objects, etc)
33 * we have a simple routine to test that functionality. These
34 * individual routines do not have to do anything "stressful".
35 *
36 * (2) We turn these simple functionality tests into a stress test by
37 * running them all in parallel, with as many threads as desired,
38 * and spread across as many datasets, objects, and vdevs as desired.
39 *
40 * (3) While all this is happening, we inject faults into the pool to
41 * verify that self-healing data really works.
42 *
43 * (4) Every time we open a dataset, we change its checksum and compression
44 * functions. Thus even individual objects vary from block to block
45 * in which checksum they use and whether they're compressed.
46 *
47 * (5) To verify that we never lose on-disk consistency after a crash,
48 * we run the entire test in a child of the main process.
49 * At random times, the child self-immolates with a SIGKILL.
50 * This is the software equivalent of pulling the power cord.
51 * The parent then runs the test again, using the existing
52 * storage pool, as many times as desired.
53 *
54 * (6) To verify that we don't have future leaks or temporal incursions,
55 * many of the functional tests record the transaction group number
56 * as part of their data. When reading old data, they verify that
57 * the transaction group number is less than the current, open txg.
58 * If you add a new test, please do this if applicable.
59 *
60 * When run with no arguments, ztest runs for about five minutes and
61 * produces no output if successful. To get a little bit of information,
62 * specify -V. To get more information, specify -VV, and so on.
63 *
64 * To turn this into an overnight stress test, use -T to specify run time.
65 *
66 * You can ask more more vdevs [-v], datasets [-d], or threads [-t]
67 * to increase the pool capacity, fanout, and overall stress level.
68 *
69 * The -N(okill) option will suppress kills, so each child runs to completion.
70 * This can be useful when you're trying to distinguish temporal incursions
71 * from plain old race conditions.
72 */
73
74 #include <sys/zfs_context.h>
75 #include <sys/spa.h>
76 #include <sys/dmu.h>
77 #include <sys/txg.h>
78 #include <sys/dbuf.h>
79 #include <sys/zap.h>
80 #include <sys/dmu_objset.h>
81 #include <sys/poll.h>
82 #include <sys/stat.h>
83 #include <sys/time.h>
84 #include <sys/wait.h>
85 #include <sys/mman.h>
86 #include <sys/resource.h>
87 #include <sys/zio.h>
88 #include <sys/zil.h>
89 #include <sys/zil_impl.h>
90 #include <sys/vdev_impl.h>
91 #include <sys/vdev_file.h>
92 #include <sys/spa_impl.h>
93 #include <sys/metaslab_impl.h>
94 #include <sys/dsl_prop.h>
95 #include <sys/dsl_dataset.h>
96 #include <sys/dsl_scan.h>
97 #include <sys/zio_checksum.h>
98 #include <sys/refcount.h>
99 #include <stdio.h>
100 #include <stdio_ext.h>
101 #include <stdlib.h>
102 #include <unistd.h>
103 #include <signal.h>
104 #include <umem.h>
105 #include <dlfcn.h>
106 #include <ctype.h>
107 #include <math.h>
108 #include <sys/fs/zfs.h>
109 #include <libnvpair.h>
110
111 static char cmdname[] = "ztest";
112 static char *zopt_pool = cmdname;
113
114 static uint64_t zopt_vdevs = 5;
115 static uint64_t zopt_vdevtime;
116 static int zopt_ashift = SPA_MINBLOCKSHIFT;
117 static int zopt_mirrors = 2;
118 static int zopt_raidz = 4;
119 static int zopt_raidz_parity = 1;
120 static size_t zopt_vdev_size = SPA_MINDEVSIZE;
121 static int zopt_datasets = 7;
122 static int zopt_threads = 23;
123 static uint64_t zopt_passtime = 60; /* 60 seconds */
124 static uint64_t zopt_killrate = 70; /* 70% kill rate */
125 static int zopt_verbose = 0;
126 static int zopt_init = 1;
127 static char *zopt_dir = "/tmp";
128 static uint64_t zopt_time = 300; /* 5 minutes */
129 static uint64_t zopt_maxloops = 50; /* max loops during spa_freeze() */
130
131 #define BT_MAGIC 0x123456789abcdefULL
132 #define MAXFAULTS() (MAX(zs->zs_mirrors, 1) * (zopt_raidz_parity + 1) - 1)
133
134 enum ztest_io_type {
135 ZTEST_IO_WRITE_TAG,
136 ZTEST_IO_WRITE_PATTERN,
137 ZTEST_IO_WRITE_ZEROES,
138 ZTEST_IO_TRUNCATE,
139 ZTEST_IO_SETATTR,
140 ZTEST_IO_TYPES
141 };
142
143 typedef struct ztest_block_tag {
144 uint64_t bt_magic;
145 uint64_t bt_objset;
146 uint64_t bt_object;
147 uint64_t bt_offset;
148 uint64_t bt_gen;
149 uint64_t bt_txg;
150 uint64_t bt_crtxg;
151 } ztest_block_tag_t;
152
153 typedef struct bufwad {
154 uint64_t bw_index;
155 uint64_t bw_txg;
156 uint64_t bw_data;
157 } bufwad_t;
158
159 /*
160 * XXX -- fix zfs range locks to be generic so we can use them here.
161 */
162 typedef enum {
163 RL_READER,
164 RL_WRITER,
165 RL_APPEND
166 } rl_type_t;
167
168 typedef struct rll {
169 void *rll_writer;
170 int rll_readers;
171 mutex_t rll_lock;
172 cond_t rll_cv;
173 } rll_t;
174
175 typedef struct rl {
176 uint64_t rl_object;
177 uint64_t rl_offset;
178 uint64_t rl_size;
179 rll_t *rl_lock;
180 } rl_t;
181
182 #define ZTEST_RANGE_LOCKS 64
183 #define ZTEST_OBJECT_LOCKS 64
184
185 /*
186 * Object descriptor. Used as a template for object lookup/create/remove.
187 */
188 typedef struct ztest_od {
189 uint64_t od_dir;
190 uint64_t od_object;
191 dmu_object_type_t od_type;
192 dmu_object_type_t od_crtype;
193 uint64_t od_blocksize;
194 uint64_t od_crblocksize;
195 uint64_t od_gen;
196 uint64_t od_crgen;
197 char od_name[MAXNAMELEN];
198 } ztest_od_t;
199
200 /*
201 * Per-dataset state.
202 */
203 typedef struct ztest_ds {
204 objset_t *zd_os;
205 zilog_t *zd_zilog;
206 uint64_t zd_seq;
207 ztest_od_t *zd_od; /* debugging aid */
208 char zd_name[MAXNAMELEN];
209 mutex_t zd_dirobj_lock;
210 rll_t zd_object_lock[ZTEST_OBJECT_LOCKS];
211 rll_t zd_range_lock[ZTEST_RANGE_LOCKS];
212 } ztest_ds_t;
213
214 /*
215 * Per-iteration state.
216 */
217 typedef void ztest_func_t(ztest_ds_t *zd, uint64_t id);
218
219 typedef struct ztest_info {
220 ztest_func_t *zi_func; /* test function */
221 uint64_t zi_iters; /* iterations per execution */
222 uint64_t *zi_interval; /* execute every <interval> seconds */
223 uint64_t zi_call_count; /* per-pass count */
224 uint64_t zi_call_time; /* per-pass time */
225 uint64_t zi_call_next; /* next time to call this function */
226 } ztest_info_t;
227
228 /*
229 * Note: these aren't static because we want dladdr() to work.
230 */
231 ztest_func_t ztest_dmu_read_write;
232 ztest_func_t ztest_dmu_write_parallel;
233 ztest_func_t ztest_dmu_object_alloc_free;
234 ztest_func_t ztest_dmu_commit_callbacks;
235 ztest_func_t ztest_zap;
236 ztest_func_t ztest_zap_parallel;
237 ztest_func_t ztest_zil_commit;
238 ztest_func_t ztest_dmu_read_write_zcopy;
239 ztest_func_t ztest_dmu_objset_create_destroy;
240 ztest_func_t ztest_dmu_prealloc;
241 ztest_func_t ztest_fzap;
242 ztest_func_t ztest_dmu_snapshot_create_destroy;
243 ztest_func_t ztest_dsl_prop_get_set;
244 ztest_func_t ztest_spa_prop_get_set;
245 ztest_func_t ztest_spa_create_destroy;
246 ztest_func_t ztest_fault_inject;
247 ztest_func_t ztest_ddt_repair;
248 ztest_func_t ztest_dmu_snapshot_hold;
249 ztest_func_t ztest_spa_rename;
250 ztest_func_t ztest_scrub;
251 ztest_func_t ztest_dsl_dataset_promote_busy;
252 ztest_func_t ztest_vdev_attach_detach;
253 ztest_func_t ztest_vdev_LUN_growth;
254 ztest_func_t ztest_vdev_add_remove;
255 ztest_func_t ztest_vdev_aux_add_remove;
256 ztest_func_t ztest_split_pool;
257
258 uint64_t zopt_always = 0ULL * NANOSEC; /* all the time */
259 uint64_t zopt_incessant = 1ULL * NANOSEC / 10; /* every 1/10 second */
260 uint64_t zopt_often = 1ULL * NANOSEC; /* every second */
261 uint64_t zopt_sometimes = 10ULL * NANOSEC; /* every 10 seconds */
262 uint64_t zopt_rarely = 60ULL * NANOSEC; /* every 60 seconds */
263
264 ztest_info_t ztest_info[] = {
265 { ztest_dmu_read_write, 1, &zopt_always },
266 { ztest_dmu_write_parallel, 10, &zopt_always },
267 { ztest_dmu_object_alloc_free, 1, &zopt_always },
268 { ztest_dmu_commit_callbacks, 1, &zopt_always },
269 { ztest_zap, 30, &zopt_always },
270 { ztest_zap_parallel, 100, &zopt_always },
271 { ztest_split_pool, 1, &zopt_always },
272 { ztest_zil_commit, 1, &zopt_incessant },
273 { ztest_dmu_read_write_zcopy, 1, &zopt_often },
274 { ztest_dmu_objset_create_destroy, 1, &zopt_often },
275 { ztest_dsl_prop_get_set, 1, &zopt_often },
276 { ztest_spa_prop_get_set, 1, &zopt_sometimes },
277 #if 0
278 { ztest_dmu_prealloc, 1, &zopt_sometimes },
279 #endif
280 { ztest_fzap, 1, &zopt_sometimes },
281 { ztest_dmu_snapshot_create_destroy, 1, &zopt_sometimes },
282 { ztest_spa_create_destroy, 1, &zopt_sometimes },
283 { ztest_fault_inject, 1, &zopt_sometimes },
284 { ztest_ddt_repair, 1, &zopt_sometimes },
285 { ztest_dmu_snapshot_hold, 1, &zopt_sometimes },
286 { ztest_spa_rename, 1, &zopt_rarely },
287 { ztest_scrub, 1, &zopt_rarely },
288 { ztest_dsl_dataset_promote_busy, 1, &zopt_rarely },
289 { ztest_vdev_attach_detach, 1, &zopt_rarely },
290 { ztest_vdev_LUN_growth, 1, &zopt_rarely },
291 { ztest_vdev_add_remove, 1, &zopt_vdevtime },
292 { ztest_vdev_aux_add_remove, 1, &zopt_vdevtime },
293 };
294
295 #define ZTEST_FUNCS (sizeof (ztest_info) / sizeof (ztest_info_t))
296
297 /*
298 * The following struct is used to hold a list of uncalled commit callbacks.
299 * The callbacks are ordered by txg number.
300 */
301 typedef struct ztest_cb_list {
302 mutex_t zcl_callbacks_lock;
303 list_t zcl_callbacks;
304 } ztest_cb_list_t;
305
306 /*
307 * Stuff we need to share writably between parent and child.
308 */
309 typedef struct ztest_shared {
310 char *zs_pool;
311 spa_t *zs_spa;
312 hrtime_t zs_proc_start;
313 hrtime_t zs_proc_stop;
314 hrtime_t zs_thread_start;
315 hrtime_t zs_thread_stop;
316 hrtime_t zs_thread_kill;
317 uint64_t zs_enospc_count;
318 uint64_t zs_vdev_next_leaf;
319 uint64_t zs_vdev_aux;
320 uint64_t zs_alloc;
321 uint64_t zs_space;
322 mutex_t zs_vdev_lock;
323 rwlock_t zs_name_lock;
324 ztest_info_t zs_info[ZTEST_FUNCS];
325 uint64_t zs_splits;
326 uint64_t zs_mirrors;
327 ztest_ds_t zs_zd[];
328 } ztest_shared_t;
329
330 #define ID_PARALLEL -1ULL
331
332 static char ztest_dev_template[] = "%s/%s.%llua";
333 static char ztest_aux_template[] = "%s/%s.%s.%llu";
334 ztest_shared_t *ztest_shared;
335 uint64_t *ztest_seq;
336
337 static int ztest_random_fd;
338 static int ztest_dump_core = 1;
339
340 static boolean_t ztest_exiting;
341
342 /* Global commit callback list */
343 static ztest_cb_list_t zcl;
344
345 extern uint64_t metaslab_gang_bang;
346 extern uint64_t metaslab_df_alloc_threshold;
347 static uint64_t metaslab_sz;
348
349 enum ztest_object {
350 ZTEST_META_DNODE = 0,
351 ZTEST_DIROBJ,
352 ZTEST_OBJECTS
353 };
354
355 static void usage(boolean_t) __NORETURN;
356
357 /*
358 * These libumem hooks provide a reasonable set of defaults for the allocator's
359 * debugging facilities.
360 */
361 const char *
362 _umem_debug_init()
363 {
364 return ("default,verbose"); /* $UMEM_DEBUG setting */
365 }
366
367 const char *
368 _umem_logging_init(void)
369 {
370 return ("fail,contents"); /* $UMEM_LOGGING setting */
371 }
372
373 #define FATAL_MSG_SZ 1024
374
375 char *fatal_msg;
376
377 static void
378 fatal(int do_perror, char *message, ...)
379 {
380 va_list args;
381 int save_errno = errno;
382 char buf[FATAL_MSG_SZ];
383
384 (void) fflush(stdout);
385
386 va_start(args, message);
387 (void) sprintf(buf, "ztest: ");
388 /* LINTED */
389 (void) vsprintf(buf + strlen(buf), message, args);
390 va_end(args);
391 if (do_perror) {
392 (void) snprintf(buf + strlen(buf), FATAL_MSG_SZ - strlen(buf),
393 ": %s", strerror(save_errno));
394 }
395 (void) fprintf(stderr, "%s\n", buf);
396 fatal_msg = buf; /* to ease debugging */
397 if (ztest_dump_core)
398 abort();
399 exit(3);
400 }
401
402 static int
403 str2shift(const char *buf)
404 {
405 const char *ends = "BKMGTPEZ";
406 int i;
407
408 if (buf[0] == '\0')
409 return (0);
410 for (i = 0; i < strlen(ends); i++) {
411 if (toupper(buf[0]) == ends[i])
412 break;
413 }
414 if (i == strlen(ends)) {
415 (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n",
416 buf);
417 usage(B_FALSE);
418 }
419 if (buf[1] == '\0' || (toupper(buf[1]) == 'B' && buf[2] == '\0')) {
420 return (10*i);
421 }
422 (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n", buf);
423 usage(B_FALSE);
424 /* NOTREACHED */
425 }
426
427 static uint64_t
428 nicenumtoull(const char *buf)
429 {
430 char *end;
431 uint64_t val;
432
433 val = strtoull(buf, &end, 0);
434 if (end == buf) {
435 (void) fprintf(stderr, "ztest: bad numeric value: %s\n", buf);
436 usage(B_FALSE);
437 } else if (end[0] == '.') {
438 double fval = strtod(buf, &end);
439 fval *= pow(2, str2shift(end));
440 if (fval > UINT64_MAX) {
441 (void) fprintf(stderr, "ztest: value too large: %s\n",
442 buf);
443 usage(B_FALSE);
444 }
445 val = (uint64_t)fval;
446 } else {
447 int shift = str2shift(end);
448 if (shift >= 64 || (val << shift) >> shift != val) {
449 (void) fprintf(stderr, "ztest: value too large: %s\n",
450 buf);
451 usage(B_FALSE);
452 }
453 val <<= shift;
454 }
455 return (val);
456 }
457
458 static void
459 usage(boolean_t requested)
460 {
461 char nice_vdev_size[10];
462 char nice_gang_bang[10];
463 FILE *fp = requested ? stdout : stderr;
464
465 nicenum(zopt_vdev_size, nice_vdev_size);
466 nicenum(metaslab_gang_bang, nice_gang_bang);
467
468 (void) fprintf(fp, "Usage: %s\n"
469 "\t[-v vdevs (default: %llu)]\n"
470 "\t[-s size_of_each_vdev (default: %s)]\n"
471 "\t[-a alignment_shift (default: %d)] use 0 for random\n"
472 "\t[-m mirror_copies (default: %d)]\n"
473 "\t[-r raidz_disks (default: %d)]\n"
474 "\t[-R raidz_parity (default: %d)]\n"
475 "\t[-d datasets (default: %d)]\n"
476 "\t[-t threads (default: %d)]\n"
477 "\t[-g gang_block_threshold (default: %s)]\n"
478 "\t[-i init_count (default: %d)] initialize pool i times\n"
479 "\t[-k kill_percentage (default: %llu%%)]\n"
480 "\t[-p pool_name (default: %s)]\n"
481 "\t[-f dir (default: %s)] file directory for vdev files\n"
482 "\t[-V] verbose (use multiple times for ever more blather)\n"
483 "\t[-E] use existing pool instead of creating new one\n"
484 "\t[-T time (default: %llu sec)] total run time\n"
485 "\t[-F freezeloops (default: %llu)] max loops in spa_freeze()\n"
486 "\t[-P passtime (default: %llu sec)] time per pass\n"
487 "\t[-h] (print help)\n"
488 "",
489 cmdname,
490 (u_longlong_t)zopt_vdevs, /* -v */
491 nice_vdev_size, /* -s */
492 zopt_ashift, /* -a */
493 zopt_mirrors, /* -m */
494 zopt_raidz, /* -r */
495 zopt_raidz_parity, /* -R */
496 zopt_datasets, /* -d */
497 zopt_threads, /* -t */
498 nice_gang_bang, /* -g */
499 zopt_init, /* -i */
500 (u_longlong_t)zopt_killrate, /* -k */
501 zopt_pool, /* -p */
502 zopt_dir, /* -f */
503 (u_longlong_t)zopt_time, /* -T */
504 (u_longlong_t)zopt_maxloops, /* -F */
505 (u_longlong_t)zopt_passtime); /* -P */
506 exit(requested ? 0 : 1);
507 }
508
509 static void
510 process_options(int argc, char **argv)
511 {
512 int opt;
513 uint64_t value;
514
515 /* By default, test gang blocks for blocks 32K and greater */
516 metaslab_gang_bang = 32 << 10;
517
518 while ((opt = getopt(argc, argv,
519 "v:s:a:m:r:R:d:t:g:i:k:p:f:VET:P:hF:")) != EOF) {
520 value = 0;
521 switch (opt) {
522 case 'v':
523 case 's':
524 case 'a':
525 case 'm':
526 case 'r':
527 case 'R':
528 case 'd':
529 case 't':
530 case 'g':
531 case 'i':
532 case 'k':
533 case 'T':
534 case 'P':
535 case 'F':
536 value = nicenumtoull(optarg);
537 }
538 switch (opt) {
539 case 'v':
540 zopt_vdevs = value;
541 break;
542 case 's':
543 zopt_vdev_size = MAX(SPA_MINDEVSIZE, value);
544 break;
545 case 'a':
546 zopt_ashift = value;
547 break;
548 case 'm':
549 zopt_mirrors = value;
550 break;
551 case 'r':
552 zopt_raidz = MAX(1, value);
553 break;
554 case 'R':
555 zopt_raidz_parity = MIN(MAX(value, 1), 3);
556 break;
557 case 'd':
558 zopt_datasets = MAX(1, value);
559 break;
560 case 't':
561 zopt_threads = MAX(1, value);
562 break;
563 case 'g':
564 metaslab_gang_bang = MAX(SPA_MINBLOCKSIZE << 1, value);
565 break;
566 case 'i':
567 zopt_init = value;
568 break;
569 case 'k':
570 zopt_killrate = value;
571 break;
572 case 'p':
573 zopt_pool = strdup(optarg);
574 break;
575 case 'f':
576 zopt_dir = strdup(optarg);
577 break;
578 case 'V':
579 zopt_verbose++;
580 break;
581 case 'E':
582 zopt_init = 0;
583 break;
584 case 'T':
585 zopt_time = value;
586 break;
587 case 'P':
588 zopt_passtime = MAX(1, value);
589 break;
590 case 'F':
591 zopt_maxloops = MAX(1, value);
592 break;
593 case 'h':
594 usage(B_TRUE);
595 break;
596 case '?':
597 default:
598 usage(B_FALSE);
599 break;
600 }
601 }
602
603 zopt_raidz_parity = MIN(zopt_raidz_parity, zopt_raidz - 1);
604
605 zopt_vdevtime = (zopt_vdevs > 0 ? zopt_time * NANOSEC / zopt_vdevs :
606 UINT64_MAX >> 2);
607 }
608
609 static void
610 ztest_kill(ztest_shared_t *zs)
611 {
612 zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(zs->zs_spa));
613 zs->zs_space = metaslab_class_get_space(spa_normal_class(zs->zs_spa));
614 (void) kill(getpid(), SIGKILL);
615 }
616
617 static uint64_t
618 ztest_random(uint64_t range)
619 {
620 uint64_t r;
621
622 if (range == 0)
623 return (0);
624
625 if (read(ztest_random_fd, &r, sizeof (r)) != sizeof (r))
626 fatal(1, "short read from /dev/urandom");
627
628 return (r % range);
629 }
630
631 /* ARGSUSED */
632 static void
633 ztest_record_enospc(const char *s)
634 {
635 ztest_shared->zs_enospc_count++;
636 }
637
638 static uint64_t
639 ztest_get_ashift(void)
640 {
641 if (zopt_ashift == 0)
642 return (SPA_MINBLOCKSHIFT + ztest_random(3));
643 return (zopt_ashift);
644 }
645
646 static nvlist_t *
647 make_vdev_file(char *path, char *aux, size_t size, uint64_t ashift)
648 {
649 char pathbuf[MAXPATHLEN];
650 uint64_t vdev;
651 nvlist_t *file;
652
653 if (ashift == 0)
654 ashift = ztest_get_ashift();
655
656 if (path == NULL) {
657 path = pathbuf;
658
659 if (aux != NULL) {
660 vdev = ztest_shared->zs_vdev_aux;
661 (void) sprintf(path, ztest_aux_template,
662 zopt_dir, zopt_pool, aux, vdev);
663 } else {
664 vdev = ztest_shared->zs_vdev_next_leaf++;
665 (void) sprintf(path, ztest_dev_template,
666 zopt_dir, zopt_pool, vdev);
667 }
668 }
669
670 if (size != 0) {
671 int fd = open(path, O_RDWR | O_CREAT | O_TRUNC, 0666);
672 if (fd == -1)
673 fatal(1, "can't open %s", path);
674 if (ftruncate(fd, size) != 0)
675 fatal(1, "can't ftruncate %s", path);
676 (void) close(fd);
677 }
678
679 VERIFY(nvlist_alloc(&file, NV_UNIQUE_NAME, 0) == 0);
680 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_TYPE, VDEV_TYPE_FILE) == 0);
681 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_PATH, path) == 0);
682 VERIFY(nvlist_add_uint64(file, ZPOOL_CONFIG_ASHIFT, ashift) == 0);
683
684 return (file);
685 }
686
687 static nvlist_t *
688 make_vdev_raidz(char *path, char *aux, size_t size, uint64_t ashift, int r)
689 {
690 nvlist_t *raidz, **child;
691 int c;
692
693 if (r < 2)
694 return (make_vdev_file(path, aux, size, ashift));
695 child = umem_alloc(r * sizeof (nvlist_t *), UMEM_NOFAIL);
696
697 for (c = 0; c < r; c++)
698 child[c] = make_vdev_file(path, aux, size, ashift);
699
700 VERIFY(nvlist_alloc(&raidz, NV_UNIQUE_NAME, 0) == 0);
701 VERIFY(nvlist_add_string(raidz, ZPOOL_CONFIG_TYPE,
702 VDEV_TYPE_RAIDZ) == 0);
703 VERIFY(nvlist_add_uint64(raidz, ZPOOL_CONFIG_NPARITY,
704 zopt_raidz_parity) == 0);
705 VERIFY(nvlist_add_nvlist_array(raidz, ZPOOL_CONFIG_CHILDREN,
706 child, r) == 0);
707
708 for (c = 0; c < r; c++)
709 nvlist_free(child[c]);
710
711 umem_free(child, r * sizeof (nvlist_t *));
712
713 return (raidz);
714 }
715
716 static nvlist_t *
717 make_vdev_mirror(char *path, char *aux, size_t size, uint64_t ashift,
718 int r, int m)
719 {
720 nvlist_t *mirror, **child;
721 int c;
722
723 if (m < 1)
724 return (make_vdev_raidz(path, aux, size, ashift, r));
725
726 child = umem_alloc(m * sizeof (nvlist_t *), UMEM_NOFAIL);
727
728 for (c = 0; c < m; c++)
729 child[c] = make_vdev_raidz(path, aux, size, ashift, r);
730
731 VERIFY(nvlist_alloc(&mirror, NV_UNIQUE_NAME, 0) == 0);
732 VERIFY(nvlist_add_string(mirror, ZPOOL_CONFIG_TYPE,
733 VDEV_TYPE_MIRROR) == 0);
734 VERIFY(nvlist_add_nvlist_array(mirror, ZPOOL_CONFIG_CHILDREN,
735 child, m) == 0);
736
737 for (c = 0; c < m; c++)
738 nvlist_free(child[c]);
739
740 umem_free(child, m * sizeof (nvlist_t *));
741
742 return (mirror);
743 }
744
745 static nvlist_t *
746 make_vdev_root(char *path, char *aux, size_t size, uint64_t ashift,
747 int log, int r, int m, int t)
748 {
749 nvlist_t *root, **child;
750 int c;
751
752 ASSERT(t > 0);
753
754 child = umem_alloc(t * sizeof (nvlist_t *), UMEM_NOFAIL);
755
756 for (c = 0; c < t; c++) {
757 child[c] = make_vdev_mirror(path, aux, size, ashift, r, m);
758 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_IS_LOG,
759 log) == 0);
760 }
761
762 VERIFY(nvlist_alloc(&root, NV_UNIQUE_NAME, 0) == 0);
763 VERIFY(nvlist_add_string(root, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) == 0);
764 VERIFY(nvlist_add_nvlist_array(root, aux ? aux : ZPOOL_CONFIG_CHILDREN,
765 child, t) == 0);
766
767 for (c = 0; c < t; c++)
768 nvlist_free(child[c]);
769
770 umem_free(child, t * sizeof (nvlist_t *));
771
772 return (root);
773 }
774
775 static int
776 ztest_random_blocksize(void)
777 {
778 return (1 << (SPA_MINBLOCKSHIFT +
779 ztest_random(SPA_MAXBLOCKSHIFT - SPA_MINBLOCKSHIFT + 1)));
780 }
781
782 static int
783 ztest_random_ibshift(void)
784 {
785 return (DN_MIN_INDBLKSHIFT +
786 ztest_random(DN_MAX_INDBLKSHIFT - DN_MIN_INDBLKSHIFT + 1));
787 }
788
789 static uint64_t
790 ztest_random_vdev_top(spa_t *spa, boolean_t log_ok)
791 {
792 uint64_t top;
793 vdev_t *rvd = spa->spa_root_vdev;
794 vdev_t *tvd;
795
796 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
797
798 do {
799 top = ztest_random(rvd->vdev_children);
800 tvd = rvd->vdev_child[top];
801 } while (tvd->vdev_ishole || (tvd->vdev_islog && !log_ok) ||
802 tvd->vdev_mg == NULL || tvd->vdev_mg->mg_class == NULL);
803
804 return (top);
805 }
806
807 static uint64_t
808 ztest_random_dsl_prop(zfs_prop_t prop)
809 {
810 uint64_t value;
811
812 do {
813 value = zfs_prop_random_value(prop, ztest_random(-1ULL));
814 } while (prop == ZFS_PROP_CHECKSUM && value == ZIO_CHECKSUM_OFF);
815
816 return (value);
817 }
818
819 static int
820 ztest_dsl_prop_set_uint64(char *osname, zfs_prop_t prop, uint64_t value,
821 boolean_t inherit)
822 {
823 const char *propname = zfs_prop_to_name(prop);
824 const char *valname;
825 char setpoint[MAXPATHLEN];
826 uint64_t curval;
827 int error;
828
829 error = dsl_prop_set(osname, propname,
830 (inherit ? ZPROP_SRC_NONE : ZPROP_SRC_LOCAL),
831 sizeof (value), 1, &value);
832
833 if (error == ENOSPC) {
834 ztest_record_enospc(FTAG);
835 return (error);
836 }
837 ASSERT3U(error, ==, 0);
838
839 VERIFY3U(dsl_prop_get(osname, propname, sizeof (curval),
840 1, &curval, setpoint), ==, 0);
841
842 if (zopt_verbose >= 6) {
843 VERIFY(zfs_prop_index_to_string(prop, curval, &valname) == 0);
844 (void) printf("%s %s = %s at '%s'\n",
845 osname, propname, valname, setpoint);
846 }
847
848 return (error);
849 }
850
851 static int
852 ztest_spa_prop_set_uint64(ztest_shared_t *zs, zpool_prop_t prop, uint64_t value)
853 {
854 spa_t *spa = zs->zs_spa;
855 nvlist_t *props = NULL;
856 int error;
857
858 VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
859 VERIFY(nvlist_add_uint64(props, zpool_prop_to_name(prop), value) == 0);
860
861 error = spa_prop_set(spa, props);
862
863 nvlist_free(props);
864
865 if (error == ENOSPC) {
866 ztest_record_enospc(FTAG);
867 return (error);
868 }
869 ASSERT3U(error, ==, 0);
870
871 return (error);
872 }
873
874 static void
875 ztest_rll_init(rll_t *rll)
876 {
877 rll->rll_writer = NULL;
878 rll->rll_readers = 0;
879 VERIFY(_mutex_init(&rll->rll_lock, USYNC_THREAD, NULL) == 0);
880 VERIFY(cond_init(&rll->rll_cv, USYNC_THREAD, NULL) == 0);
881 }
882
883 static void
884 ztest_rll_destroy(rll_t *rll)
885 {
886 ASSERT(rll->rll_writer == NULL);
887 ASSERT(rll->rll_readers == 0);
888 VERIFY(_mutex_destroy(&rll->rll_lock) == 0);
889 VERIFY(cond_destroy(&rll->rll_cv) == 0);
890 }
891
892 static void
893 ztest_rll_lock(rll_t *rll, rl_type_t type)
894 {
895 VERIFY(mutex_lock(&rll->rll_lock) == 0);
896
897 if (type == RL_READER) {
898 while (rll->rll_writer != NULL)
899 (void) cond_wait(&rll->rll_cv, &rll->rll_lock);
900 rll->rll_readers++;
901 } else {
902 while (rll->rll_writer != NULL || rll->rll_readers)
903 (void) cond_wait(&rll->rll_cv, &rll->rll_lock);
904 rll->rll_writer = curthread;
905 }
906
907 VERIFY(mutex_unlock(&rll->rll_lock) == 0);
908 }
909
910 static void
911 ztest_rll_unlock(rll_t *rll)
912 {
913 VERIFY(mutex_lock(&rll->rll_lock) == 0);
914
915 if (rll->rll_writer) {
916 ASSERT(rll->rll_readers == 0);
917 rll->rll_writer = NULL;
918 } else {
919 ASSERT(rll->rll_readers != 0);
920 ASSERT(rll->rll_writer == NULL);
921 rll->rll_readers--;
922 }
923
924 if (rll->rll_writer == NULL && rll->rll_readers == 0)
925 VERIFY(cond_broadcast(&rll->rll_cv) == 0);
926
927 VERIFY(mutex_unlock(&rll->rll_lock) == 0);
928 }
929
930 static void
931 ztest_object_lock(ztest_ds_t *zd, uint64_t object, rl_type_t type)
932 {
933 rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
934
935 ztest_rll_lock(rll, type);
936 }
937
938 static void
939 ztest_object_unlock(ztest_ds_t *zd, uint64_t object)
940 {
941 rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
942
943 ztest_rll_unlock(rll);
944 }
945
946 static rl_t *
947 ztest_range_lock(ztest_ds_t *zd, uint64_t object, uint64_t offset,
948 uint64_t size, rl_type_t type)
949 {
950 uint64_t hash = object ^ (offset % (ZTEST_RANGE_LOCKS + 1));
951 rll_t *rll = &zd->zd_range_lock[hash & (ZTEST_RANGE_LOCKS - 1)];
952 rl_t *rl;
953
954 rl = umem_alloc(sizeof (*rl), UMEM_NOFAIL);
955 rl->rl_object = object;
956 rl->rl_offset = offset;
957 rl->rl_size = size;
958 rl->rl_lock = rll;
959
960 ztest_rll_lock(rll, type);
961
962 return (rl);
963 }
964
965 static void
966 ztest_range_unlock(rl_t *rl)
967 {
968 rll_t *rll = rl->rl_lock;
969
970 ztest_rll_unlock(rll);
971
972 umem_free(rl, sizeof (*rl));
973 }
974
975 static void
976 ztest_zd_init(ztest_ds_t *zd, objset_t *os)
977 {
978 zd->zd_os = os;
979 zd->zd_zilog = dmu_objset_zil(os);
980 zd->zd_seq = 0;
981 dmu_objset_name(os, zd->zd_name);
982
983 VERIFY(_mutex_init(&zd->zd_dirobj_lock, USYNC_THREAD, NULL) == 0);
984
985 for (int l = 0; l < ZTEST_OBJECT_LOCKS; l++)
986 ztest_rll_init(&zd->zd_object_lock[l]);
987
988 for (int l = 0; l < ZTEST_RANGE_LOCKS; l++)
989 ztest_rll_init(&zd->zd_range_lock[l]);
990 }
991
992 static void
993 ztest_zd_fini(ztest_ds_t *zd)
994 {
995 VERIFY(_mutex_destroy(&zd->zd_dirobj_lock) == 0);
996
997 for (int l = 0; l < ZTEST_OBJECT_LOCKS; l++)
998 ztest_rll_destroy(&zd->zd_object_lock[l]);
999
1000 for (int l = 0; l < ZTEST_RANGE_LOCKS; l++)
1001 ztest_rll_destroy(&zd->zd_range_lock[l]);
1002 }
1003
1004 #define TXG_MIGHTWAIT (ztest_random(10) == 0 ? TXG_NOWAIT : TXG_WAIT)
1005
1006 static uint64_t
1007 ztest_tx_assign(dmu_tx_t *tx, uint64_t txg_how, const char *tag)
1008 {
1009 uint64_t txg;
1010 int error;
1011
1012 /*
1013 * Attempt to assign tx to some transaction group.
1014 */
1015 error = dmu_tx_assign(tx, txg_how);
1016 if (error) {
1017 if (error == ERESTART) {
1018 ASSERT(txg_how == TXG_NOWAIT);
1019 dmu_tx_wait(tx);
1020 } else {
1021 ASSERT3U(error, ==, ENOSPC);
1022 ztest_record_enospc(tag);
1023 }
1024 dmu_tx_abort(tx);
1025 return (0);
1026 }
1027 txg = dmu_tx_get_txg(tx);
1028 ASSERT(txg != 0);
1029 return (txg);
1030 }
1031
1032 static void
1033 ztest_pattern_set(void *buf, uint64_t size, uint64_t value)
1034 {
1035 uint64_t *ip = buf;
1036 uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1037
1038 while (ip < ip_end)
1039 *ip++ = value;
1040 }
1041
1042 static boolean_t
1043 ztest_pattern_match(void *buf, uint64_t size, uint64_t value)
1044 {
1045 uint64_t *ip = buf;
1046 uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1047 uint64_t diff = 0;
1048
1049 while (ip < ip_end)
1050 diff |= (value - *ip++);
1051
1052 return (diff == 0);
1053 }
1054
1055 static void
1056 ztest_bt_generate(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1057 uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg)
1058 {
1059 bt->bt_magic = BT_MAGIC;
1060 bt->bt_objset = dmu_objset_id(os);
1061 bt->bt_object = object;
1062 bt->bt_offset = offset;
1063 bt->bt_gen = gen;
1064 bt->bt_txg = txg;
1065 bt->bt_crtxg = crtxg;
1066 }
1067
1068 static void
1069 ztest_bt_verify(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1070 uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg)
1071 {
1072 ASSERT(bt->bt_magic == BT_MAGIC);
1073 ASSERT(bt->bt_objset == dmu_objset_id(os));
1074 ASSERT(bt->bt_object == object);
1075 ASSERT(bt->bt_offset == offset);
1076 ASSERT(bt->bt_gen <= gen);
1077 ASSERT(bt->bt_txg <= txg);
1078 ASSERT(bt->bt_crtxg == crtxg);
1079 }
1080
1081 static ztest_block_tag_t *
1082 ztest_bt_bonus(dmu_buf_t *db)
1083 {
1084 dmu_object_info_t doi;
1085 ztest_block_tag_t *bt;
1086
1087 dmu_object_info_from_db(db, &doi);
1088 ASSERT3U(doi.doi_bonus_size, <=, db->db_size);
1089 ASSERT3U(doi.doi_bonus_size, >=, sizeof (*bt));
1090 bt = (void *)((char *)db->db_data + doi.doi_bonus_size - sizeof (*bt));
1091
1092 return (bt);
1093 }
1094
1095 /*
1096 * ZIL logging ops
1097 */
1098
1099 #define lrz_type lr_mode
1100 #define lrz_blocksize lr_uid
1101 #define lrz_ibshift lr_gid
1102 #define lrz_bonustype lr_rdev
1103 #define lrz_bonuslen lr_crtime[1]
1104
1105 static uint64_t
1106 ztest_log_create(ztest_ds_t *zd, dmu_tx_t *tx, lr_create_t *lr)
1107 {
1108 char *name = (void *)(lr + 1); /* name follows lr */
1109 size_t namesize = strlen(name) + 1;
1110 itx_t *itx;
1111
1112 if (zil_replaying(zd->zd_zilog, tx))
1113 return (0);
1114
1115 itx = zil_itx_create(TX_CREATE, sizeof (*lr) + namesize);
1116 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1117 sizeof (*lr) + namesize - sizeof (lr_t));
1118
1119 return (zil_itx_assign(zd->zd_zilog, itx, tx));
1120 }
1121
1122 static uint64_t
1123 ztest_log_remove(ztest_ds_t *zd, dmu_tx_t *tx, lr_remove_t *lr)
1124 {
1125 char *name = (void *)(lr + 1); /* name follows lr */
1126 size_t namesize = strlen(name) + 1;
1127 itx_t *itx;
1128
1129 if (zil_replaying(zd->zd_zilog, tx))
1130 return (0);
1131
1132 itx = zil_itx_create(TX_REMOVE, sizeof (*lr) + namesize);
1133 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1134 sizeof (*lr) + namesize - sizeof (lr_t));
1135
1136 return (zil_itx_assign(zd->zd_zilog, itx, tx));
1137 }
1138
1139 static uint64_t
1140 ztest_log_write(ztest_ds_t *zd, dmu_tx_t *tx, lr_write_t *lr)
1141 {
1142 itx_t *itx;
1143 itx_wr_state_t write_state = ztest_random(WR_NUM_STATES);
1144
1145 if (zil_replaying(zd->zd_zilog, tx))
1146 return (0);
1147
1148 if (lr->lr_length > ZIL_MAX_LOG_DATA)
1149 write_state = WR_INDIRECT;
1150
1151 itx = zil_itx_create(TX_WRITE,
1152 sizeof (*lr) + (write_state == WR_COPIED ? lr->lr_length : 0));
1153
1154 if (write_state == WR_COPIED &&
1155 dmu_read(zd->zd_os, lr->lr_foid, lr->lr_offset, lr->lr_length,
1156 ((lr_write_t *)&itx->itx_lr) + 1, DMU_READ_NO_PREFETCH) != 0) {
1157 zil_itx_destroy(itx);
1158 itx = zil_itx_create(TX_WRITE, sizeof (*lr));
1159 write_state = WR_NEED_COPY;
1160 }
1161 itx->itx_private = zd;
1162 itx->itx_wr_state = write_state;
1163 itx->itx_sync = (ztest_random(8) == 0);
1164 itx->itx_sod += (write_state == WR_NEED_COPY ? lr->lr_length : 0);
1165
1166 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1167 sizeof (*lr) - sizeof (lr_t));
1168
1169 return (zil_itx_assign(zd->zd_zilog, itx, tx));
1170 }
1171
1172 static uint64_t
1173 ztest_log_truncate(ztest_ds_t *zd, dmu_tx_t *tx, lr_truncate_t *lr)
1174 {
1175 itx_t *itx;
1176
1177 if (zil_replaying(zd->zd_zilog, tx))
1178 return (0);
1179
1180 itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr));
1181 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1182 sizeof (*lr) - sizeof (lr_t));
1183
1184 return (zil_itx_assign(zd->zd_zilog, itx, tx));
1185 }
1186
1187 static uint64_t
1188 ztest_log_setattr(ztest_ds_t *zd, dmu_tx_t *tx, lr_setattr_t *lr)
1189 {
1190 itx_t *itx;
1191
1192 if (zil_replaying(zd->zd_zilog, tx))
1193 return (0);
1194
1195 itx = zil_itx_create(TX_SETATTR, sizeof (*lr));
1196 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1197 sizeof (*lr) - sizeof (lr_t));
1198
1199 return (zil_itx_assign(zd->zd_zilog, itx, tx));
1200 }
1201
1202 /*
1203 * ZIL replay ops
1204 */
1205 static int
1206 ztest_replay_create(ztest_ds_t *zd, lr_create_t *lr, boolean_t byteswap)
1207 {
1208 char *name = (void *)(lr + 1); /* name follows lr */
1209 objset_t *os = zd->zd_os;
1210 ztest_block_tag_t *bbt;
1211 dmu_buf_t *db;
1212 dmu_tx_t *tx;
1213 uint64_t txg;
1214 int error = 0;
1215
1216 if (byteswap)
1217 byteswap_uint64_array(lr, sizeof (*lr));
1218
1219 ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1220 ASSERT(name[0] != '\0');
1221
1222 tx = dmu_tx_create(os);
1223
1224 dmu_tx_hold_zap(tx, lr->lr_doid, B_TRUE, name);
1225
1226 if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1227 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, B_TRUE, NULL);
1228 } else {
1229 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1230 }
1231
1232 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1233 if (txg == 0)
1234 return (ENOSPC);
1235
1236 ASSERT(dmu_objset_zil(os)->zl_replay == !!lr->lr_foid);
1237
1238 if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1239 if (lr->lr_foid == 0) {
1240 lr->lr_foid = zap_create(os,
1241 lr->lrz_type, lr->lrz_bonustype,
1242 lr->lrz_bonuslen, tx);
1243 } else {
1244 error = zap_create_claim(os, lr->lr_foid,
1245 lr->lrz_type, lr->lrz_bonustype,
1246 lr->lrz_bonuslen, tx);
1247 }
1248 } else {
1249 if (lr->lr_foid == 0) {
1250 lr->lr_foid = dmu_object_alloc(os,
1251 lr->lrz_type, 0, lr->lrz_bonustype,
1252 lr->lrz_bonuslen, tx);
1253 } else {
1254 error = dmu_object_claim(os, lr->lr_foid,
1255 lr->lrz_type, 0, lr->lrz_bonustype,
1256 lr->lrz_bonuslen, tx);
1257 }
1258 }
1259
1260 if (error) {
1261 ASSERT3U(error, ==, EEXIST);
1262 ASSERT(zd->zd_zilog->zl_replay);
1263 dmu_tx_commit(tx);
1264 return (error);
1265 }
1266
1267 ASSERT(lr->lr_foid != 0);
1268
1269 if (lr->lrz_type != DMU_OT_ZAP_OTHER)
1270 VERIFY3U(0, ==, dmu_object_set_blocksize(os, lr->lr_foid,
1271 lr->lrz_blocksize, lr->lrz_ibshift, tx));
1272
1273 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1274 bbt = ztest_bt_bonus(db);
1275 dmu_buf_will_dirty(db, tx);
1276 ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_gen, txg, txg);
1277 dmu_buf_rele(db, FTAG);
1278
1279 VERIFY3U(0, ==, zap_add(os, lr->lr_doid, name, sizeof (uint64_t), 1,
1280 &lr->lr_foid, tx));
1281
1282 (void) ztest_log_create(zd, tx, lr);
1283
1284 dmu_tx_commit(tx);
1285
1286 return (0);
1287 }
1288
1289 static int
1290 ztest_replay_remove(ztest_ds_t *zd, lr_remove_t *lr, boolean_t byteswap)
1291 {
1292 char *name = (void *)(lr + 1); /* name follows lr */
1293 objset_t *os = zd->zd_os;
1294 dmu_object_info_t doi;
1295 dmu_tx_t *tx;
1296 uint64_t object, txg;
1297
1298 if (byteswap)
1299 byteswap_uint64_array(lr, sizeof (*lr));
1300
1301 ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1302 ASSERT(name[0] != '\0');
1303
1304 VERIFY3U(0, ==,
1305 zap_lookup(os, lr->lr_doid, name, sizeof (object), 1, &object));
1306 ASSERT(object != 0);
1307
1308 ztest_object_lock(zd, object, RL_WRITER);
1309
1310 VERIFY3U(0, ==, dmu_object_info(os, object, &doi));
1311
1312 tx = dmu_tx_create(os);
1313
1314 dmu_tx_hold_zap(tx, lr->lr_doid, B_FALSE, name);
1315 dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END);
1316
1317 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1318 if (txg == 0) {
1319 ztest_object_unlock(zd, object);
1320 return (ENOSPC);
1321 }
1322
1323 if (doi.doi_type == DMU_OT_ZAP_OTHER) {
1324 VERIFY3U(0, ==, zap_destroy(os, object, tx));
1325 } else {
1326 VERIFY3U(0, ==, dmu_object_free(os, object, tx));
1327 }
1328
1329 VERIFY3U(0, ==, zap_remove(os, lr->lr_doid, name, tx));
1330
1331 (void) ztest_log_remove(zd, tx, lr);
1332
1333 dmu_tx_commit(tx);
1334
1335 ztest_object_unlock(zd, object);
1336
1337 return (0);
1338 }
1339
1340 static int
1341 ztest_replay_write(ztest_ds_t *zd, lr_write_t *lr, boolean_t byteswap)
1342 {
1343 objset_t *os = zd->zd_os;
1344 void *data = lr + 1; /* data follows lr */
1345 uint64_t offset, length;
1346 ztest_block_tag_t *bt = data;
1347 ztest_block_tag_t *bbt;
1348 uint64_t gen, txg, lrtxg, crtxg;
1349 dmu_object_info_t doi;
1350 dmu_tx_t *tx;
1351 dmu_buf_t *db;
1352 arc_buf_t *abuf = NULL;
1353 rl_t *rl;
1354
1355 if (byteswap)
1356 byteswap_uint64_array(lr, sizeof (*lr));
1357
1358 offset = lr->lr_offset;
1359 length = lr->lr_length;
1360
1361 /* If it's a dmu_sync() block, write the whole block */
1362 if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
1363 uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
1364 if (length < blocksize) {
1365 offset -= offset % blocksize;
1366 length = blocksize;
1367 }
1368 }
1369
1370 if (bt->bt_magic == BSWAP_64(BT_MAGIC))
1371 byteswap_uint64_array(bt, sizeof (*bt));
1372
1373 if (bt->bt_magic != BT_MAGIC)
1374 bt = NULL;
1375
1376 ztest_object_lock(zd, lr->lr_foid, RL_READER);
1377 rl = ztest_range_lock(zd, lr->lr_foid, offset, length, RL_WRITER);
1378
1379 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1380
1381 dmu_object_info_from_db(db, &doi);
1382
1383 bbt = ztest_bt_bonus(db);
1384 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1385 gen = bbt->bt_gen;
1386 crtxg = bbt->bt_crtxg;
1387 lrtxg = lr->lr_common.lrc_txg;
1388
1389 tx = dmu_tx_create(os);
1390
1391 dmu_tx_hold_write(tx, lr->lr_foid, offset, length);
1392
1393 if (ztest_random(8) == 0 && length == doi.doi_data_block_size &&
1394 P2PHASE(offset, length) == 0)
1395 abuf = dmu_request_arcbuf(db, length);
1396
1397 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1398 if (txg == 0) {
1399 if (abuf != NULL)
1400 dmu_return_arcbuf(abuf);
1401 dmu_buf_rele(db, FTAG);
1402 ztest_range_unlock(rl);
1403 ztest_object_unlock(zd, lr->lr_foid);
1404 return (ENOSPC);
1405 }
1406
1407 if (bt != NULL) {
1408 /*
1409 * Usually, verify the old data before writing new data --
1410 * but not always, because we also want to verify correct
1411 * behavior when the data was not recently read into cache.
1412 */
1413 ASSERT(offset % doi.doi_data_block_size == 0);
1414 if (ztest_random(4) != 0) {
1415 int prefetch = ztest_random(2) ?
1416 DMU_READ_PREFETCH : DMU_READ_NO_PREFETCH;
1417 ztest_block_tag_t rbt;
1418
1419 VERIFY(dmu_read(os, lr->lr_foid, offset,
1420 sizeof (rbt), &rbt, prefetch) == 0);
1421 if (rbt.bt_magic == BT_MAGIC) {
1422 ztest_bt_verify(&rbt, os, lr->lr_foid,
1423 offset, gen, txg, crtxg);
1424 }
1425 }
1426
1427 /*
1428 * Writes can appear to be newer than the bonus buffer because
1429 * the ztest_get_data() callback does a dmu_read() of the
1430 * open-context data, which may be different than the data
1431 * as it was when the write was generated.
1432 */
1433 if (zd->zd_zilog->zl_replay) {
1434 ztest_bt_verify(bt, os, lr->lr_foid, offset,
1435 MAX(gen, bt->bt_gen), MAX(txg, lrtxg),
1436 bt->bt_crtxg);
1437 }
1438
1439 /*
1440 * Set the bt's gen/txg to the bonus buffer's gen/txg
1441 * so that all of the usual ASSERTs will work.
1442 */
1443 ztest_bt_generate(bt, os, lr->lr_foid, offset, gen, txg, crtxg);
1444 }
1445
1446 if (abuf == NULL) {
1447 dmu_write(os, lr->lr_foid, offset, length, data, tx);
1448 } else {
1449 bcopy(data, abuf->b_data, length);
1450 dmu_assign_arcbuf(db, offset, abuf, tx);
1451 }
1452
1453 (void) ztest_log_write(zd, tx, lr);
1454
1455 dmu_buf_rele(db, FTAG);
1456
1457 dmu_tx_commit(tx);
1458
1459 ztest_range_unlock(rl);
1460 ztest_object_unlock(zd, lr->lr_foid);
1461
1462 return (0);
1463 }
1464
1465 static int
1466 ztest_replay_truncate(ztest_ds_t *zd, lr_truncate_t *lr, boolean_t byteswap)
1467 {
1468 objset_t *os = zd->zd_os;
1469 dmu_tx_t *tx;
1470 uint64_t txg;
1471 rl_t *rl;
1472
1473 if (byteswap)
1474 byteswap_uint64_array(lr, sizeof (*lr));
1475
1476 ztest_object_lock(zd, lr->lr_foid, RL_READER);
1477 rl = ztest_range_lock(zd, lr->lr_foid, lr->lr_offset, lr->lr_length,
1478 RL_WRITER);
1479
1480 tx = dmu_tx_create(os);
1481
1482 dmu_tx_hold_free(tx, lr->lr_foid, lr->lr_offset, lr->lr_length);
1483
1484 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1485 if (txg == 0) {
1486 ztest_range_unlock(rl);
1487 ztest_object_unlock(zd, lr->lr_foid);
1488 return (ENOSPC);
1489 }
1490
1491 VERIFY(dmu_free_range(os, lr->lr_foid, lr->lr_offset,
1492 lr->lr_length, tx) == 0);
1493
1494 (void) ztest_log_truncate(zd, tx, lr);
1495
1496 dmu_tx_commit(tx);
1497
1498 ztest_range_unlock(rl);
1499 ztest_object_unlock(zd, lr->lr_foid);
1500
1501 return (0);
1502 }
1503
1504 static int
1505 ztest_replay_setattr(ztest_ds_t *zd, lr_setattr_t *lr, boolean_t byteswap)
1506 {
1507 objset_t *os = zd->zd_os;
1508 dmu_tx_t *tx;
1509 dmu_buf_t *db;
1510 ztest_block_tag_t *bbt;
1511 uint64_t txg, lrtxg, crtxg;
1512
1513 if (byteswap)
1514 byteswap_uint64_array(lr, sizeof (*lr));
1515
1516 ztest_object_lock(zd, lr->lr_foid, RL_WRITER);
1517
1518 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1519
1520 tx = dmu_tx_create(os);
1521 dmu_tx_hold_bonus(tx, lr->lr_foid);
1522
1523 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1524 if (txg == 0) {
1525 dmu_buf_rele(db, FTAG);
1526 ztest_object_unlock(zd, lr->lr_foid);
1527 return (ENOSPC);
1528 }
1529
1530 bbt = ztest_bt_bonus(db);
1531 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1532 crtxg = bbt->bt_crtxg;
1533 lrtxg = lr->lr_common.lrc_txg;
1534
1535 if (zd->zd_zilog->zl_replay) {
1536 ASSERT(lr->lr_size != 0);
1537 ASSERT(lr->lr_mode != 0);
1538 ASSERT(lrtxg != 0);
1539 } else {
1540 /*
1541 * Randomly change the size and increment the generation.
1542 */
1543 lr->lr_size = (ztest_random(db->db_size / sizeof (*bbt)) + 1) *
1544 sizeof (*bbt);
1545 lr->lr_mode = bbt->bt_gen + 1;
1546 ASSERT(lrtxg == 0);
1547 }
1548
1549 /*
1550 * Verify that the current bonus buffer is not newer than our txg.
1551 */
1552 ztest_bt_verify(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode,
1553 MAX(txg, lrtxg), crtxg);
1554
1555 dmu_buf_will_dirty(db, tx);
1556
1557 ASSERT3U(lr->lr_size, >=, sizeof (*bbt));
1558 ASSERT3U(lr->lr_size, <=, db->db_size);
1559 VERIFY3U(dmu_set_bonus(db, lr->lr_size, tx), ==, 0);
1560 bbt = ztest_bt_bonus(db);
1561
1562 ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode, txg, crtxg);
1563
1564 dmu_buf_rele(db, FTAG);
1565
1566 (void) ztest_log_setattr(zd, tx, lr);
1567
1568 dmu_tx_commit(tx);
1569
1570 ztest_object_unlock(zd, lr->lr_foid);
1571
1572 return (0);
1573 }
1574
1575 zil_replay_func_t *ztest_replay_vector[TX_MAX_TYPE] = {
1576 NULL, /* 0 no such transaction type */
1577 ztest_replay_create, /* TX_CREATE */
1578 NULL, /* TX_MKDIR */
1579 NULL, /* TX_MKXATTR */
1580 NULL, /* TX_SYMLINK */
1581 ztest_replay_remove, /* TX_REMOVE */
1582 NULL, /* TX_RMDIR */
1583 NULL, /* TX_LINK */
1584 NULL, /* TX_RENAME */
1585 ztest_replay_write, /* TX_WRITE */
1586 ztest_replay_truncate, /* TX_TRUNCATE */
1587 ztest_replay_setattr, /* TX_SETATTR */
1588 NULL, /* TX_ACL */
1589 NULL, /* TX_CREATE_ACL */
1590 NULL, /* TX_CREATE_ATTR */
1591 NULL, /* TX_CREATE_ACL_ATTR */
1592 NULL, /* TX_MKDIR_ACL */
1593 NULL, /* TX_MKDIR_ATTR */
1594 NULL, /* TX_MKDIR_ACL_ATTR */
1595 NULL, /* TX_WRITE2 */
1596 };
1597
1598 /*
1599 * ZIL get_data callbacks
1600 */
1601
1602 static void
1603 ztest_get_done(zgd_t *zgd, int error)
1604 {
1605 ztest_ds_t *zd = zgd->zgd_private;
1606 uint64_t object = zgd->zgd_rl->rl_object;
1607
1608 if (zgd->zgd_db)
1609 dmu_buf_rele(zgd->zgd_db, zgd);
1610
1611 ztest_range_unlock(zgd->zgd_rl);
1612 ztest_object_unlock(zd, object);
1613
1614 if (error == 0 && zgd->zgd_bp)
1615 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
1616
1617 umem_free(zgd, sizeof (*zgd));
1618 }
1619
1620 static int
1621 ztest_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
1622 {
1623 ztest_ds_t *zd = arg;
1624 objset_t *os = zd->zd_os;
1625 uint64_t object = lr->lr_foid;
1626 uint64_t offset = lr->lr_offset;
1627 uint64_t size = lr->lr_length;
1628 blkptr_t *bp = &lr->lr_blkptr;
1629 uint64_t txg = lr->lr_common.lrc_txg;
1630 uint64_t crtxg;
1631 dmu_object_info_t doi;
1632 dmu_buf_t *db;
1633 zgd_t *zgd;
1634 int error;
1635
1636 ztest_object_lock(zd, object, RL_READER);
1637 error = dmu_bonus_hold(os, object, FTAG, &db);
1638 if (error) {
1639 ztest_object_unlock(zd, object);
1640 return (error);
1641 }
1642
1643 crtxg = ztest_bt_bonus(db)->bt_crtxg;
1644
1645 if (crtxg == 0 || crtxg > txg) {
1646 dmu_buf_rele(db, FTAG);
1647 ztest_object_unlock(zd, object);
1648 return (ENOENT);
1649 }
1650
1651 dmu_object_info_from_db(db, &doi);
1652 dmu_buf_rele(db, FTAG);
1653 db = NULL;
1654
1655 zgd = umem_zalloc(sizeof (*zgd), UMEM_NOFAIL);
1656 zgd->zgd_zilog = zd->zd_zilog;
1657 zgd->zgd_private = zd;
1658
1659 if (buf != NULL) { /* immediate write */
1660 zgd->zgd_rl = ztest_range_lock(zd, object, offset, size,
1661 RL_READER);
1662
1663 error = dmu_read(os, object, offset, size, buf,
1664 DMU_READ_NO_PREFETCH);
1665 ASSERT(error == 0);
1666 } else {
1667 size = doi.doi_data_block_size;
1668 if (ISP2(size)) {
1669 offset = P2ALIGN(offset, size);
1670 } else {
1671 ASSERT(offset < size);
1672 offset = 0;
1673 }
1674
1675 zgd->zgd_rl = ztest_range_lock(zd, object, offset, size,
1676 RL_READER);
1677
1678 error = dmu_buf_hold(os, object, offset, zgd, &db,
1679 DMU_READ_NO_PREFETCH);
1680
1681 if (error == 0) {
1682 zgd->zgd_db = db;
1683 zgd->zgd_bp = bp;
1684
1685 ASSERT(db->db_offset == offset);
1686 ASSERT(db->db_size == size);
1687
1688 error = dmu_sync(zio, lr->lr_common.lrc_txg,
1689 ztest_get_done, zgd);
1690
1691 if (error == 0)
1692 return (0);
1693 }
1694 }
1695
1696 ztest_get_done(zgd, error);
1697
1698 return (error);
1699 }
1700
1701 static void *
1702 ztest_lr_alloc(size_t lrsize, char *name)
1703 {
1704 char *lr;
1705 size_t namesize = name ? strlen(name) + 1 : 0;
1706
1707 lr = umem_zalloc(lrsize + namesize, UMEM_NOFAIL);
1708
1709 if (name)
1710 bcopy(name, lr + lrsize, namesize);
1711
1712 return (lr);
1713 }
1714
1715 void
1716 ztest_lr_free(void *lr, size_t lrsize, char *name)
1717 {
1718 size_t namesize = name ? strlen(name) + 1 : 0;
1719
1720 umem_free(lr, lrsize + namesize);
1721 }
1722
1723 /*
1724 * Lookup a bunch of objects. Returns the number of objects not found.
1725 */
1726 static int
1727 ztest_lookup(ztest_ds_t *zd, ztest_od_t *od, int count)
1728 {
1729 int missing = 0;
1730 int error;
1731
1732 ASSERT(_mutex_held(&zd->zd_dirobj_lock));
1733
1734 for (int i = 0; i < count; i++, od++) {
1735 od->od_object = 0;
1736 error = zap_lookup(zd->zd_os, od->od_dir, od->od_name,
1737 sizeof (uint64_t), 1, &od->od_object);
1738 if (error) {
1739 ASSERT(error == ENOENT);
1740 ASSERT(od->od_object == 0);
1741 missing++;
1742 } else {
1743 dmu_buf_t *db;
1744 ztest_block_tag_t *bbt;
1745 dmu_object_info_t doi;
1746
1747 ASSERT(od->od_object != 0);
1748 ASSERT(missing == 0); /* there should be no gaps */
1749
1750 ztest_object_lock(zd, od->od_object, RL_READER);
1751 VERIFY3U(0, ==, dmu_bonus_hold(zd->zd_os,
1752 od->od_object, FTAG, &db));
1753 dmu_object_info_from_db(db, &doi);
1754 bbt = ztest_bt_bonus(db);
1755 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1756 od->od_type = doi.doi_type;
1757 od->od_blocksize = doi.doi_data_block_size;
1758 od->od_gen = bbt->bt_gen;
1759 dmu_buf_rele(db, FTAG);
1760 ztest_object_unlock(zd, od->od_object);
1761 }
1762 }
1763
1764 return (missing);
1765 }
1766
1767 static int
1768 ztest_create(ztest_ds_t *zd, ztest_od_t *od, int count)
1769 {
1770 int missing = 0;
1771
1772 ASSERT(_mutex_held(&zd->zd_dirobj_lock));
1773
1774 for (int i = 0; i < count; i++, od++) {
1775 if (missing) {
1776 od->od_object = 0;
1777 missing++;
1778 continue;
1779 }
1780
1781 lr_create_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
1782
1783 lr->lr_doid = od->od_dir;
1784 lr->lr_foid = 0; /* 0 to allocate, > 0 to claim */
1785 lr->lrz_type = od->od_crtype;
1786 lr->lrz_blocksize = od->od_crblocksize;
1787 lr->lrz_ibshift = ztest_random_ibshift();
1788 lr->lrz_bonustype = DMU_OT_UINT64_OTHER;
1789 lr->lrz_bonuslen = dmu_bonus_max();
1790 lr->lr_gen = od->od_crgen;
1791 lr->lr_crtime[0] = time(NULL);
1792
1793 if (ztest_replay_create(zd, lr, B_FALSE) != 0) {
1794 ASSERT(missing == 0);
1795 od->od_object = 0;
1796 missing++;
1797 } else {
1798 od->od_object = lr->lr_foid;
1799 od->od_type = od->od_crtype;
1800 od->od_blocksize = od->od_crblocksize;
1801 od->od_gen = od->od_crgen;
1802 ASSERT(od->od_object != 0);
1803 }
1804
1805 ztest_lr_free(lr, sizeof (*lr), od->od_name);
1806 }
1807
1808 return (missing);
1809 }
1810
1811 static int
1812 ztest_remove(ztest_ds_t *zd, ztest_od_t *od, int count)
1813 {
1814 int missing = 0;
1815 int error;
1816
1817 ASSERT(_mutex_held(&zd->zd_dirobj_lock));
1818
1819 od += count - 1;
1820
1821 for (int i = count - 1; i >= 0; i--, od--) {
1822 if (missing) {
1823 missing++;
1824 continue;
1825 }
1826
1827 if (od->od_object == 0)
1828 continue;
1829
1830 lr_remove_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
1831
1832 lr->lr_doid = od->od_dir;
1833
1834 if ((error = ztest_replay_remove(zd, lr, B_FALSE)) != 0) {
1835 ASSERT3U(error, ==, ENOSPC);
1836 missing++;
1837 } else {
1838 od->od_object = 0;
1839 }
1840 ztest_lr_free(lr, sizeof (*lr), od->od_name);
1841 }
1842
1843 return (missing);
1844 }
1845
1846 static int
1847 ztest_write(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size,
1848 void *data)
1849 {
1850 lr_write_t *lr;
1851 int error;
1852
1853 lr = ztest_lr_alloc(sizeof (*lr) + size, NULL);
1854
1855 lr->lr_foid = object;
1856 lr->lr_offset = offset;
1857 lr->lr_length = size;
1858 lr->lr_blkoff = 0;
1859 BP_ZERO(&lr->lr_blkptr);
1860
1861 bcopy(data, lr + 1, size);
1862
1863 error = ztest_replay_write(zd, lr, B_FALSE);
1864
1865 ztest_lr_free(lr, sizeof (*lr) + size, NULL);
1866
1867 return (error);
1868 }
1869
1870 static int
1871 ztest_truncate(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
1872 {
1873 lr_truncate_t *lr;
1874 int error;
1875
1876 lr = ztest_lr_alloc(sizeof (*lr), NULL);
1877
1878 lr->lr_foid = object;
1879 lr->lr_offset = offset;
1880 lr->lr_length = size;
1881
1882 error = ztest_replay_truncate(zd, lr, B_FALSE);
1883
1884 ztest_lr_free(lr, sizeof (*lr), NULL);
1885
1886 return (error);
1887 }
1888
1889 static int
1890 ztest_setattr(ztest_ds_t *zd, uint64_t object)
1891 {
1892 lr_setattr_t *lr;
1893 int error;
1894
1895 lr = ztest_lr_alloc(sizeof (*lr), NULL);
1896
1897 lr->lr_foid = object;
1898 lr->lr_size = 0;
1899 lr->lr_mode = 0;
1900
1901 error = ztest_replay_setattr(zd, lr, B_FALSE);
1902
1903 ztest_lr_free(lr, sizeof (*lr), NULL);
1904
1905 return (error);
1906 }
1907
1908 static void
1909 ztest_prealloc(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
1910 {
1911 objset_t *os = zd->zd_os;
1912 dmu_tx_t *tx;
1913 uint64_t txg;
1914 rl_t *rl;
1915
1916 txg_wait_synced(dmu_objset_pool(os), 0);
1917
1918 ztest_object_lock(zd, object, RL_READER);
1919 rl = ztest_range_lock(zd, object, offset, size, RL_WRITER);
1920
1921 tx = dmu_tx_create(os);
1922
1923 dmu_tx_hold_write(tx, object, offset, size);
1924
1925 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1926
1927 if (txg != 0) {
1928 dmu_prealloc(os, object, offset, size, tx);
1929 dmu_tx_commit(tx);
1930 txg_wait_synced(dmu_objset_pool(os), txg);
1931 } else {
1932 (void) dmu_free_long_range(os, object, offset, size);
1933 }
1934
1935 ztest_range_unlock(rl);
1936 ztest_object_unlock(zd, object);
1937 }
1938
1939 static void
1940 ztest_io(ztest_ds_t *zd, uint64_t object, uint64_t offset)
1941 {
1942 ztest_block_tag_t wbt;
1943 dmu_object_info_t doi;
1944 enum ztest_io_type io_type;
1945 uint64_t blocksize;
1946 void *data;
1947
1948 VERIFY(dmu_object_info(zd->zd_os, object, &doi) == 0);
1949 blocksize = doi.doi_data_block_size;
1950 data = umem_alloc(blocksize, UMEM_NOFAIL);
1951
1952 /*
1953 * Pick an i/o type at random, biased toward writing block tags.
1954 */
1955 io_type = ztest_random(ZTEST_IO_TYPES);
1956 if (ztest_random(2) == 0)
1957 io_type = ZTEST_IO_WRITE_TAG;
1958
1959 switch (io_type) {
1960
1961 case ZTEST_IO_WRITE_TAG:
1962 ztest_bt_generate(&wbt, zd->zd_os, object, offset, 0, 0, 0);
1963 (void) ztest_write(zd, object, offset, sizeof (wbt), &wbt);
1964 break;
1965
1966 case ZTEST_IO_WRITE_PATTERN:
1967 (void) memset(data, 'a' + (object + offset) % 5, blocksize);
1968 if (ztest_random(2) == 0) {
1969 /*
1970 * Induce fletcher2 collisions to ensure that
1971 * zio_ddt_collision() detects and resolves them
1972 * when using fletcher2-verify for deduplication.
1973 */
1974 ((uint64_t *)data)[0] ^= 1ULL << 63;
1975 ((uint64_t *)data)[4] ^= 1ULL << 63;
1976 }
1977 (void) ztest_write(zd, object, offset, blocksize, data);
1978 break;
1979
1980 case ZTEST_IO_WRITE_ZEROES:
1981 bzero(data, blocksize);
1982 (void) ztest_write(zd, object, offset, blocksize, data);
1983 break;
1984
1985 case ZTEST_IO_TRUNCATE:
1986 (void) ztest_truncate(zd, object, offset, blocksize);
1987 break;
1988
1989 case ZTEST_IO_SETATTR:
1990 (void) ztest_setattr(zd, object);
1991 break;
1992 }
1993
1994 umem_free(data, blocksize);
1995 }
1996
1997 /*
1998 * Initialize an object description template.
1999 */
2000 static void
2001 ztest_od_init(ztest_od_t *od, uint64_t id, char *tag, uint64_t index,
2002 dmu_object_type_t type, uint64_t blocksize, uint64_t gen)
2003 {
2004 od->od_dir = ZTEST_DIROBJ;
2005 od->od_object = 0;
2006
2007 od->od_crtype = type;
2008 od->od_crblocksize = blocksize ? blocksize : ztest_random_blocksize();
2009 od->od_crgen = gen;
2010
2011 od->od_type = DMU_OT_NONE;
2012 od->od_blocksize = 0;
2013 od->od_gen = 0;
2014
2015 (void) snprintf(od->od_name, sizeof (od->od_name), "%s(%lld)[%llu]",
2016 tag, (int64_t)id, index);
2017 }
2018
2019 /*
2020 * Lookup or create the objects for a test using the od template.
2021 * If the objects do not all exist, or if 'remove' is specified,
2022 * remove any existing objects and create new ones. Otherwise,
2023 * use the existing objects.
2024 */
2025 static int
2026 ztest_object_init(ztest_ds_t *zd, ztest_od_t *od, size_t size, boolean_t remove)
2027 {
2028 int count = size / sizeof (*od);
2029 int rv = 0;
2030
2031 VERIFY(mutex_lock(&zd->zd_dirobj_lock) == 0);
2032 if ((ztest_lookup(zd, od, count) != 0 || remove) &&
2033 (ztest_remove(zd, od, count) != 0 ||
2034 ztest_create(zd, od, count) != 0))
2035 rv = -1;
2036 zd->zd_od = od;
2037 VERIFY(mutex_unlock(&zd->zd_dirobj_lock) == 0);
2038
2039 return (rv);
2040 }
2041
2042 /* ARGSUSED */
2043 void
2044 ztest_zil_commit(ztest_ds_t *zd, uint64_t id)
2045 {
2046 zilog_t *zilog = zd->zd_zilog;
2047
2048 zil_commit(zilog, UINT64_MAX, ztest_random(ZTEST_OBJECTS));
2049
2050 /*
2051 * Remember the committed values in zd, which is in parent/child
2052 * shared memory. If we die, the next iteration of ztest_run()
2053 * will verify that the log really does contain this record.
2054 */
2055 mutex_enter(&zilog->zl_lock);
2056 ASSERT(zd->zd_seq <= zilog->zl_commit_lr_seq);
2057 zd->zd_seq = zilog->zl_commit_lr_seq;
2058 mutex_exit(&zilog->zl_lock);
2059 }
2060
2061 /*
2062 * Verify that we can't destroy an active pool, create an existing pool,
2063 * or create a pool with a bad vdev spec.
2064 */
2065 /* ARGSUSED */
2066 void
2067 ztest_spa_create_destroy(ztest_ds_t *zd, uint64_t id)
2068 {
2069 ztest_shared_t *zs = ztest_shared;
2070 spa_t *spa;
2071 nvlist_t *nvroot;
2072
2073 /*
2074 * Attempt to create using a bad file.
2075 */
2076 nvroot = make_vdev_root("/dev/bogus", NULL, 0, 0, 0, 0, 0, 1);
2077 VERIFY3U(ENOENT, ==,
2078 spa_create("ztest_bad_file", nvroot, NULL, NULL, NULL));
2079 nvlist_free(nvroot);
2080
2081 /*
2082 * Attempt to create using a bad mirror.
2083 */
2084 nvroot = make_vdev_root("/dev/bogus", NULL, 0, 0, 0, 0, 2, 1);
2085 VERIFY3U(ENOENT, ==,
2086 spa_create("ztest_bad_mirror", nvroot, NULL, NULL, NULL));
2087 nvlist_free(nvroot);
2088
2089 /*
2090 * Attempt to create an existing pool. It shouldn't matter
2091 * what's in the nvroot; we should fail with EEXIST.
2092 */
2093 (void) rw_rdlock(&zs->zs_name_lock);
2094 nvroot = make_vdev_root("/dev/bogus", NULL, 0, 0, 0, 0, 0, 1);
2095 VERIFY3U(EEXIST, ==, spa_create(zs->zs_pool, nvroot, NULL, NULL, NULL));
2096 nvlist_free(nvroot);
2097 VERIFY3U(0, ==, spa_open(zs->zs_pool, &spa, FTAG));
2098 VERIFY3U(EBUSY, ==, spa_destroy(zs->zs_pool));
2099 spa_close(spa, FTAG);
2100
2101 (void) rw_unlock(&zs->zs_name_lock);
2102 }
2103
2104 static vdev_t *
2105 vdev_lookup_by_path(vdev_t *vd, const char *path)
2106 {
2107 vdev_t *mvd;
2108
2109 if (vd->vdev_path != NULL && strcmp(path, vd->vdev_path) == 0)
2110 return (vd);
2111
2112 for (int c = 0; c < vd->vdev_children; c++)
2113 if ((mvd = vdev_lookup_by_path(vd->vdev_child[c], path)) !=
2114 NULL)
2115 return (mvd);
2116
2117 return (NULL);
2118 }
2119
2120 /*
2121 * Find the first available hole which can be used as a top-level.
2122 */
2123 int
2124 find_vdev_hole(spa_t *spa)
2125 {
2126 vdev_t *rvd = spa->spa_root_vdev;
2127 int c;
2128
2129 ASSERT(spa_config_held(spa, SCL_VDEV, RW_READER) == SCL_VDEV);
2130
2131 for (c = 0; c < rvd->vdev_children; c++) {
2132 vdev_t *cvd = rvd->vdev_child[c];
2133
2134 if (cvd->vdev_ishole)
2135 break;
2136 }
2137 return (c);
2138 }
2139
2140 /*
2141 * Verify that vdev_add() works as expected.
2142 */
2143 /* ARGSUSED */
2144 void
2145 ztest_vdev_add_remove(ztest_ds_t *zd, uint64_t id)
2146 {
2147 ztest_shared_t *zs = ztest_shared;
2148 spa_t *spa = zs->zs_spa;
2149 uint64_t leaves;
2150 uint64_t guid;
2151 nvlist_t *nvroot;
2152 int error;
2153
2154 VERIFY(mutex_lock(&zs->zs_vdev_lock) == 0);
2155 leaves = MAX(zs->zs_mirrors + zs->zs_splits, 1) * zopt_raidz;
2156
2157 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2158
2159 ztest_shared->zs_vdev_next_leaf = find_vdev_hole(spa) * leaves;
2160
2161 /*
2162 * If we have slogs then remove them 1/4 of the time.
2163 */
2164 if (spa_has_slogs(spa) && ztest_random(4) == 0) {
2165 /*
2166 * Grab the guid from the head of the log class rotor.
2167 */
2168 guid = spa_log_class(spa)->mc_rotor->mg_vd->vdev_guid;
2169
2170 spa_config_exit(spa, SCL_VDEV, FTAG);
2171
2172 /*
2173 * We have to grab the zs_name_lock as writer to
2174 * prevent a race between removing a slog (dmu_objset_find)
2175 * and destroying a dataset. Removing the slog will
2176 * grab a reference on the dataset which may cause
2177 * dmu_objset_destroy() to fail with EBUSY thus
2178 * leaving the dataset in an inconsistent state.
2179 */
2180 VERIFY(rw_wrlock(&ztest_shared->zs_name_lock) == 0);
2181 error = spa_vdev_remove(spa, guid, B_FALSE);
2182 VERIFY(rw_unlock(&ztest_shared->zs_name_lock) == 0);
2183
2184 if (error && error != EEXIST)
2185 fatal(0, "spa_vdev_remove() = %d", error);
2186 } else {
2187 spa_config_exit(spa, SCL_VDEV, FTAG);
2188
2189 /*
2190 * Make 1/4 of the devices be log devices.
2191 */
2192 nvroot = make_vdev_root(NULL, NULL, zopt_vdev_size, 0,
2193 ztest_random(4) == 0, zopt_raidz, zs->zs_mirrors, 1);
2194
2195 error = spa_vdev_add(spa, nvroot);
2196 nvlist_free(nvroot);
2197
2198 if (error == ENOSPC)
2199 ztest_record_enospc("spa_vdev_add");
2200 else if (error != 0)
2201 fatal(0, "spa_vdev_add() = %d", error);
2202 }
2203
2204 VERIFY(mutex_unlock(&ztest_shared->zs_vdev_lock) == 0);
2205 }
2206
2207 /*
2208 * Verify that adding/removing aux devices (l2arc, hot spare) works as expected.
2209 */
2210 /* ARGSUSED */
2211 void
2212 ztest_vdev_aux_add_remove(ztest_ds_t *zd, uint64_t id)
2213 {
2214 ztest_shared_t *zs = ztest_shared;
2215 spa_t *spa = zs->zs_spa;
2216 vdev_t *rvd = spa->spa_root_vdev;
2217 spa_aux_vdev_t *sav;
2218 char *aux;
2219 uint64_t guid = 0;
2220 int error;
2221
2222 if (ztest_random(2) == 0) {
2223 sav = &spa->spa_spares;
2224 aux = ZPOOL_CONFIG_SPARES;
2225 } else {
2226 sav = &spa->spa_l2cache;
2227 aux = ZPOOL_CONFIG_L2CACHE;
2228 }
2229
2230 VERIFY(mutex_lock(&zs->zs_vdev_lock) == 0);
2231
2232 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2233
2234 if (sav->sav_count != 0 && ztest_random(4) == 0) {
2235 /*
2236 * Pick a random device to remove.
2237 */
2238 guid = sav->sav_vdevs[ztest_random(sav->sav_count)]->vdev_guid;
2239 } else {
2240 /*
2241 * Find an unused device we can add.
2242 */
2243 zs->zs_vdev_aux = 0;
2244 for (;;) {
2245 char path[MAXPATHLEN];
2246 int c;
2247 (void) sprintf(path, ztest_aux_template, zopt_dir,
2248 zopt_pool, aux, zs->zs_vdev_aux);
2249 for (c = 0; c < sav->sav_count; c++)
2250 if (strcmp(sav->sav_vdevs[c]->vdev_path,
2251 path) == 0)
2252 break;
2253 if (c == sav->sav_count &&
2254 vdev_lookup_by_path(rvd, path) == NULL)
2255 break;
2256 zs->zs_vdev_aux++;
2257 }
2258 }
2259
2260 spa_config_exit(spa, SCL_VDEV, FTAG);
2261
2262 if (guid == 0) {
2263 /*
2264 * Add a new device.
2265 */
2266 nvlist_t *nvroot = make_vdev_root(NULL, aux,
2267 (zopt_vdev_size * 5) / 4, 0, 0, 0, 0, 1);
2268 error = spa_vdev_add(spa, nvroot);
2269 if (error != 0)
2270 fatal(0, "spa_vdev_add(%p) = %d", nvroot, error);
2271 nvlist_free(nvroot);
2272 } else {
2273 /*
2274 * Remove an existing device. Sometimes, dirty its
2275 * vdev state first to make sure we handle removal
2276 * of devices that have pending state changes.
2277 */
2278 if (ztest_random(2) == 0)
2279 (void) vdev_online(spa, guid, 0, NULL);
2280
2281 error = spa_vdev_remove(spa, guid, B_FALSE);
2282 if (error != 0 && error != EBUSY)
2283 fatal(0, "spa_vdev_remove(%llu) = %d", guid, error);
2284 }
2285
2286 VERIFY(mutex_unlock(&zs->zs_vdev_lock) == 0);
2287 }
2288
2289 /*
2290 * split a pool if it has mirror tlvdevs
2291 */
2292 /* ARGSUSED */
2293 void
2294 ztest_split_pool(ztest_ds_t *zd, uint64_t id)
2295 {
2296 ztest_shared_t *zs = ztest_shared;
2297 spa_t *spa = zs->zs_spa;
2298 vdev_t *rvd = spa->spa_root_vdev;
2299 nvlist_t *tree, **child, *config, *split, **schild;
2300 uint_t c, children, schildren = 0, lastlogid = 0;
2301 int error = 0;
2302
2303 VERIFY(mutex_lock(&zs->zs_vdev_lock) == 0);
2304
2305 /* ensure we have a useable config; mirrors of raidz aren't supported */
2306 if (zs->zs_mirrors < 3 || zopt_raidz > 1) {
2307 VERIFY(mutex_unlock(&zs->zs_vdev_lock) == 0);
2308 return;
2309 }
2310
2311 /* clean up the old pool, if any */
2312 (void) spa_destroy("splitp");
2313
2314 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2315
2316 /* generate a config from the existing config */
2317 mutex_enter(&spa->spa_props_lock);
2318 VERIFY(nvlist_lookup_nvlist(spa->spa_config, ZPOOL_CONFIG_VDEV_TREE,
2319 &tree) == 0);
2320 mutex_exit(&spa->spa_props_lock);
2321
2322 VERIFY(nvlist_lookup_nvlist_array(tree, ZPOOL_CONFIG_CHILDREN, &child,
2323 &children) == 0);
2324
2325 schild = malloc(rvd->vdev_children * sizeof (nvlist_t *));
2326 for (c = 0; c < children; c++) {
2327 vdev_t *tvd = rvd->vdev_child[c];
2328 nvlist_t **mchild;
2329 uint_t mchildren;
2330
2331 if (tvd->vdev_islog || tvd->vdev_ops == &vdev_hole_ops) {
2332 VERIFY(nvlist_alloc(&schild[schildren], NV_UNIQUE_NAME,
2333 0) == 0);
2334 VERIFY(nvlist_add_string(schild[schildren],
2335 ZPOOL_CONFIG_TYPE, VDEV_TYPE_HOLE) == 0);
2336 VERIFY(nvlist_add_uint64(schild[schildren],
2337 ZPOOL_CONFIG_IS_HOLE, 1) == 0);
2338 if (lastlogid == 0)
2339 lastlogid = schildren;
2340 ++schildren;
2341 continue;
2342 }
2343 lastlogid = 0;
2344 VERIFY(nvlist_lookup_nvlist_array(child[c],
2345 ZPOOL_CONFIG_CHILDREN, &mchild, &mchildren) == 0);
2346 VERIFY(nvlist_dup(mchild[0], &schild[schildren++], 0) == 0);
2347 }
2348
2349 /* OK, create a config that can be used to split */
2350 VERIFY(nvlist_alloc(&split, NV_UNIQUE_NAME, 0) == 0);
2351 VERIFY(nvlist_add_string(split, ZPOOL_CONFIG_TYPE,
2352 VDEV_TYPE_ROOT) == 0);
2353 VERIFY(nvlist_add_nvlist_array(split, ZPOOL_CONFIG_CHILDREN, schild,
2354 lastlogid != 0 ? lastlogid : schildren) == 0);
2355
2356 VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, 0) == 0);
2357 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, split) == 0);
2358
2359 for (c = 0; c < schildren; c++)
2360 nvlist_free(schild[c]);
2361 free(schild);
2362 nvlist_free(split);
2363
2364 spa_config_exit(spa, SCL_VDEV, FTAG);
2365
2366 (void) rw_wrlock(&zs->zs_name_lock);
2367 error = spa_vdev_split_mirror(spa, "splitp", config, NULL, B_FALSE);
2368 (void) rw_unlock(&zs->zs_name_lock);
2369
2370 nvlist_free(config);
2371
2372 if (error == 0) {
2373 (void) printf("successful split - results:\n");
2374 mutex_enter(&spa_namespace_lock);
2375 show_pool_stats(spa);
2376 show_pool_stats(spa_lookup("splitp"));
2377 mutex_exit(&spa_namespace_lock);
2378 ++zs->zs_splits;
2379 --zs->zs_mirrors;
2380 }
2381 VERIFY(mutex_unlock(&zs->zs_vdev_lock) == 0);
2382
2383 }
2384
2385 /*
2386 * Verify that we can attach and detach devices.
2387 */
2388 /* ARGSUSED */
2389 void
2390 ztest_vdev_attach_detach(ztest_ds_t *zd, uint64_t id)
2391 {
2392 ztest_shared_t *zs = ztest_shared;
2393 spa_t *spa = zs->zs_spa;
2394 spa_aux_vdev_t *sav = &spa->spa_spares;
2395 vdev_t *rvd = spa->spa_root_vdev;
2396 vdev_t *oldvd, *newvd, *pvd;
2397 nvlist_t *root;
2398 uint64_t leaves;
2399 uint64_t leaf, top;
2400 uint64_t ashift = ztest_get_ashift();
2401 uint64_t oldguid, pguid;
2402 size_t oldsize, newsize;
2403 char oldpath[MAXPATHLEN], newpath[MAXPATHLEN];
2404 int replacing;
2405 int oldvd_has_siblings = B_FALSE;
2406 int newvd_is_spare = B_FALSE;
2407 int oldvd_is_log;
2408 int error, expected_error;
2409
2410 VERIFY(mutex_lock(&zs->zs_vdev_lock) == 0);
2411 leaves = MAX(zs->zs_mirrors, 1) * zopt_raidz;
2412
2413 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2414
2415 /*
2416 * Decide whether to do an attach or a replace.
2417 */
2418 replacing = ztest_random(2);
2419
2420 /*
2421 * Pick a random top-level vdev.
2422 */
2423 top = ztest_random_vdev_top(spa, B_TRUE);
2424
2425 /*
2426 * Pick a random leaf within it.
2427 */
2428 leaf = ztest_random(leaves);
2429
2430 /*
2431 * Locate this vdev.
2432 */
2433 oldvd = rvd->vdev_child[top];
2434 if (zs->zs_mirrors >= 1) {
2435 ASSERT(oldvd->vdev_ops == &vdev_mirror_ops);
2436 ASSERT(oldvd->vdev_children >= zs->zs_mirrors);
2437 oldvd = oldvd->vdev_child[leaf / zopt_raidz];
2438 }
2439 if (zopt_raidz > 1) {
2440 ASSERT(oldvd->vdev_ops == &vdev_raidz_ops);
2441 ASSERT(oldvd->vdev_children == zopt_raidz);
2442 oldvd = oldvd->vdev_child[leaf % zopt_raidz];
2443 }
2444
2445 /*
2446 * If we're already doing an attach or replace, oldvd may be a
2447 * mirror vdev -- in which case, pick a random child.
2448 */
2449 while (oldvd->vdev_children != 0) {
2450 oldvd_has_siblings = B_TRUE;
2451 ASSERT(oldvd->vdev_children >= 2);
2452 oldvd = oldvd->vdev_child[ztest_random(oldvd->vdev_children)];
2453 }
2454
2455 oldguid = oldvd->vdev_guid;
2456 oldsize = vdev_get_min_asize(oldvd);
2457 oldvd_is_log = oldvd->vdev_top->vdev_islog;
2458 (void) strcpy(oldpath, oldvd->vdev_path);
2459 pvd = oldvd->vdev_parent;
2460 pguid = pvd->vdev_guid;
2461
2462 /*
2463 * If oldvd has siblings, then half of the time, detach it.
2464 */
2465 if (oldvd_has_siblings && ztest_random(2) == 0) {
2466 spa_config_exit(spa, SCL_VDEV, FTAG);
2467 error = spa_vdev_detach(spa, oldguid, pguid, B_FALSE);
2468 if (error != 0 && error != ENODEV && error != EBUSY &&
2469 error != ENOTSUP)
2470 fatal(0, "detach (%s) returned %d", oldpath, error);
2471 VERIFY(mutex_unlock(&zs->zs_vdev_lock) == 0);
2472 return;
2473 }
2474
2475 /*
2476 * For the new vdev, choose with equal probability between the two
2477 * standard paths (ending in either 'a' or 'b') or a random hot spare.
2478 */
2479 if (sav->sav_count != 0 && ztest_random(3) == 0) {
2480 newvd = sav->sav_vdevs[ztest_random(sav->sav_count)];
2481 newvd_is_spare = B_TRUE;
2482 (void) strcpy(newpath, newvd->vdev_path);
2483 } else {
2484 (void) snprintf(newpath, sizeof (newpath), ztest_dev_template,
2485 zopt_dir, zopt_pool, top * leaves + leaf);
2486 if (ztest_random(2) == 0)
2487 newpath[strlen(newpath) - 1] = 'b';
2488 newvd = vdev_lookup_by_path(rvd, newpath);
2489 }
2490
2491 if (newvd) {
2492 newsize = vdev_get_min_asize(newvd);
2493 } else {
2494 /*
2495 * Make newsize a little bigger or smaller than oldsize.
2496 * If it's smaller, the attach should fail.
2497 * If it's larger, and we're doing a replace,
2498 * we should get dynamic LUN growth when we're done.
2499 */
2500 newsize = 10 * oldsize / (9 + ztest_random(3));
2501 }
2502
2503 /*
2504 * If pvd is not a mirror or root, the attach should fail with ENOTSUP,
2505 * unless it's a replace; in that case any non-replacing parent is OK.
2506 *
2507 * If newvd is already part of the pool, it should fail with EBUSY.
2508 *
2509 * If newvd is too small, it should fail with EOVERFLOW.
2510 */
2511 if (pvd->vdev_ops != &vdev_mirror_ops &&
2512 pvd->vdev_ops != &vdev_root_ops && (!replacing ||
2513 pvd->vdev_ops == &vdev_replacing_ops ||
2514 pvd->vdev_ops == &vdev_spare_ops))
2515 expected_error = ENOTSUP;
2516 else if (newvd_is_spare && (!replacing || oldvd_is_log))
2517 expected_error = ENOTSUP;
2518 else if (newvd == oldvd)
2519 expected_error = replacing ? 0 : EBUSY;
2520 else if (vdev_lookup_by_path(rvd, newpath) != NULL)
2521 expected_error = EBUSY;
2522 else if (newsize < oldsize)
2523 expected_error = EOVERFLOW;
2524 else if (ashift > oldvd->vdev_top->vdev_ashift)
2525 expected_error = EDOM;
2526 else
2527 expected_error = 0;
2528
2529 spa_config_exit(spa, SCL_VDEV, FTAG);
2530
2531 /*
2532 * Build the nvlist describing newpath.
2533 */
2534 root = make_vdev_root(newpath, NULL, newvd == NULL ? newsize : 0,
2535 ashift, 0, 0, 0, 1);
2536
2537 error = spa_vdev_attach(spa, oldguid, root, replacing);
2538
2539 nvlist_free(root);
2540
2541 /*
2542 * If our parent was the replacing vdev, but the replace completed,
2543 * then instead of failing with ENOTSUP we may either succeed,
2544 * fail with ENODEV, or fail with EOVERFLOW.
2545 */
2546 if (expected_error == ENOTSUP &&
2547 (error == 0 || error == ENODEV || error == EOVERFLOW))
2548 expected_error = error;
2549
2550 /*
2551 * If someone grew the LUN, the replacement may be too small.
2552 */
2553 if (error == EOVERFLOW || error == EBUSY)
2554 expected_error = error;
2555
2556 /* XXX workaround 6690467 */
2557 if (error != expected_error && expected_error != EBUSY) {
2558 fatal(0, "attach (%s %llu, %s %llu, %d) "
2559 "returned %d, expected %d",
2560 oldpath, (longlong_t)oldsize, newpath,
2561 (longlong_t)newsize, replacing, error, expected_error);
2562 }
2563
2564 VERIFY(mutex_unlock(&zs->zs_vdev_lock) == 0);
2565 }
2566
2567 /*
2568 * Callback function which expands the physical size of the vdev.
2569 */
2570 vdev_t *
2571 grow_vdev(vdev_t *vd, void *arg)
2572 {
2573 spa_t *spa = vd->vdev_spa;
2574 size_t *newsize = arg;
2575 size_t fsize;
2576 int fd;
2577
2578 ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
2579 ASSERT(vd->vdev_ops->vdev_op_leaf);
2580
2581 if ((fd = open(vd->vdev_path, O_RDWR)) == -1)
2582 return (vd);
2583
2584 fsize = lseek(fd, 0, SEEK_END);
2585 (void) ftruncate(fd, *newsize);
2586
2587 if (zopt_verbose >= 6) {
2588 (void) printf("%s grew from %lu to %lu bytes\n",
2589 vd->vdev_path, (ulong_t)fsize, (ulong_t)*newsize);
2590 }
2591 (void) close(fd);
2592 return (NULL);
2593 }
2594
2595 /*
2596 * Callback function which expands a given vdev by calling vdev_online().
2597 */
2598 /* ARGSUSED */
2599 vdev_t *
2600 online_vdev(vdev_t *vd, void *arg)
2601 {
2602 spa_t *spa = vd->vdev_spa;
2603 vdev_t *tvd = vd->vdev_top;
2604 uint64_t guid = vd->vdev_guid;
2605 uint64_t generation = spa->spa_config_generation + 1;
2606 vdev_state_t newstate = VDEV_STATE_UNKNOWN;
2607 int error;
2608
2609 ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
2610 ASSERT(vd->vdev_ops->vdev_op_leaf);
2611
2612 /* Calling vdev_online will initialize the new metaslabs */
2613 spa_config_exit(spa, SCL_STATE, spa);
2614 error = vdev_online(spa, guid, ZFS_ONLINE_EXPAND, &newstate);
2615 spa_config_enter(spa, SCL_STATE, spa, RW_READER);
2616
2617 /*
2618 * If vdev_online returned an error or the underlying vdev_open
2619 * failed then we abort the expand. The only way to know that
2620 * vdev_open fails is by checking the returned newstate.
2621 */
2622 if (error || newstate != VDEV_STATE_HEALTHY) {
2623 if (zopt_verbose >= 5) {
2624 (void) printf("Unable to expand vdev, state %llu, "
2625 "error %d\n", (u_longlong_t)newstate, error);
2626 }
2627 return (vd);
2628 }
2629 ASSERT3U(newstate, ==, VDEV_STATE_HEALTHY);
2630
2631 /*
2632 * Since we dropped the lock we need to ensure that we're
2633 * still talking to the original vdev. It's possible this
2634 * vdev may have been detached/replaced while we were
2635 * trying to online it.
2636 */
2637 if (generation != spa->spa_config_generation) {
2638 if (zopt_verbose >= 5) {
2639 (void) printf("vdev configuration has changed, "
2640 "guid %llu, state %llu, expected gen %llu, "
2641 "got gen %llu\n",
2642 (u_longlong_t)guid,
2643 (u_longlong_t)tvd->vdev_state,
2644 (u_longlong_t)generation,
2645 (u_longlong_t)spa->spa_config_generation);
2646 }
2647 return (vd);
2648 }
2649 return (NULL);
2650 }
2651
2652 /*
2653 * Traverse the vdev tree calling the supplied function.
2654 * We continue to walk the tree until we either have walked all
2655 * children or we receive a non-NULL return from the callback.
2656 * If a NULL callback is passed, then we just return back the first
2657 * leaf vdev we encounter.
2658 */
2659 vdev_t *
2660 vdev_walk_tree(vdev_t *vd, vdev_t *(*func)(vdev_t *, void *), void *arg)
2661 {
2662 if (vd->vdev_ops->vdev_op_leaf) {
2663 if (func == NULL)
2664 return (vd);
2665 else
2666 return (func(vd, arg));
2667 }
2668
2669 for (uint_t c = 0; c < vd->vdev_children; c++) {
2670 vdev_t *cvd = vd->vdev_child[c];
2671 if ((cvd = vdev_walk_tree(cvd, func, arg)) != NULL)
2672 return (cvd);
2673 }
2674 return (NULL);
2675 }
2676
2677 /*
2678 * Verify that dynamic LUN growth works as expected.
2679 */
2680 /* ARGSUSED */
2681 void
2682 ztest_vdev_LUN_growth(ztest_ds_t *zd, uint64_t id)
2683 {
2684 ztest_shared_t *zs = ztest_shared;
2685 spa_t *spa = zs->zs_spa;
2686 vdev_t *vd, *tvd;
2687 metaslab_class_t *mc;
2688 metaslab_group_t *mg;
2689 size_t psize, newsize;
2690 uint64_t top;
2691 uint64_t old_class_space, new_class_space, old_ms_count, new_ms_count;
2692
2693 VERIFY(mutex_lock(&zs->zs_vdev_lock) == 0);
2694 spa_config_enter(spa, SCL_STATE, spa, RW_READER);
2695
2696 top = ztest_random_vdev_top(spa, B_TRUE);
2697
2698 tvd = spa->spa_root_vdev->vdev_child[top];
2699 mg = tvd->vdev_mg;
2700 mc = mg->mg_class;
2701 old_ms_count = tvd->vdev_ms_count;
2702 old_class_space = metaslab_class_get_space(mc);
2703
2704 /*
2705 * Determine the size of the first leaf vdev associated with
2706 * our top-level device.
2707 */
2708 vd = vdev_walk_tree(tvd, NULL, NULL);
2709 ASSERT3P(vd, !=, NULL);
2710 ASSERT(vd->vdev_ops->vdev_op_leaf);
2711
2712 psize = vd->vdev_psize;
2713
2714 /*
2715 * We only try to expand the vdev if it's healthy, less than 4x its
2716 * original size, and it has a valid psize.
2717 */
2718 if (tvd->vdev_state != VDEV_STATE_HEALTHY ||
2719 psize == 0 || psize >= 4 * zopt_vdev_size) {
2720 spa_config_exit(spa, SCL_STATE, spa);
2721 VERIFY(mutex_unlock(&zs->zs_vdev_lock) == 0);
2722 return;
2723 }
2724 ASSERT(psize > 0);
2725 newsize = psize + psize / 8;
2726 ASSERT3U(newsize, >, psize);
2727
2728 if (zopt_verbose >= 6) {
2729 (void) printf("Expanding LUN %s from %lu to %lu\n",
2730 vd->vdev_path, (ulong_t)psize, (ulong_t)newsize);
2731 }
2732
2733 /*
2734 * Growing the vdev is a two step process:
2735 * 1). expand the physical size (i.e. relabel)
2736 * 2). online the vdev to create the new metaslabs
2737 */
2738 if (vdev_walk_tree(tvd, grow_vdev, &newsize) != NULL ||
2739 vdev_walk_tree(tvd, online_vdev, NULL) != NULL ||
2740 tvd->vdev_state != VDEV_STATE_HEALTHY) {
2741 if (zopt_verbose >= 5) {
2742 (void) printf("Could not expand LUN because "
2743 "the vdev configuration changed.\n");
2744 }
2745 spa_config_exit(spa, SCL_STATE, spa);
2746 VERIFY(mutex_unlock(&zs->zs_vdev_lock) == 0);
2747 return;
2748 }
2749
2750 spa_config_exit(spa, SCL_STATE, spa);
2751
2752 /*
2753 * Expanding the LUN will update the config asynchronously,
2754 * thus we must wait for the async thread to complete any
2755 * pending tasks before proceeding.
2756 */
2757 for (;;) {
2758 boolean_t done;
2759 mutex_enter(&spa->spa_async_lock);
2760 done = (spa->spa_async_thread == NULL && !spa->spa_async_tasks);
2761 mutex_exit(&spa->spa_async_lock);
2762 if (done)
2763 break;
2764 txg_wait_synced(spa_get_dsl(spa), 0);
2765 (void) poll(NULL, 0, 100);
2766 }
2767
2768 spa_config_enter(spa, SCL_STATE, spa, RW_READER);
2769
2770 tvd = spa->spa_root_vdev->vdev_child[top];
2771 new_ms_count = tvd->vdev_ms_count;
2772 new_class_space = metaslab_class_get_space(mc);
2773
2774 if (tvd->vdev_mg != mg || mg->mg_class != mc) {
2775 if (zopt_verbose >= 5) {
2776 (void) printf("Could not verify LUN expansion due to "
2777 "intervening vdev offline or remove.\n");
2778 }
2779 spa_config_exit(spa, SCL_STATE, spa);
2780 VERIFY(mutex_unlock(&zs->zs_vdev_lock) == 0);
2781 return;
2782 }
2783
2784 /*
2785 * Make sure we were able to grow the vdev.
2786 */
2787 if (new_ms_count <= old_ms_count)
2788 fatal(0, "LUN expansion failed: ms_count %llu <= %llu\n",
2789 old_ms_count, new_ms_count);
2790
2791 /*
2792 * Make sure we were able to grow the pool.
2793 */
2794 if (new_class_space <= old_class_space)
2795 fatal(0, "LUN expansion failed: class_space %llu <= %llu\n",
2796 old_class_space, new_class_space);
2797
2798 if (zopt_verbose >= 5) {
2799 char oldnumbuf[6], newnumbuf[6];
2800
2801 nicenum(old_class_space, oldnumbuf);
2802 nicenum(new_class_space, newnumbuf);
2803 (void) printf("%s grew from %s to %s\n",
2804 spa->spa_name, oldnumbuf, newnumbuf);
2805 }
2806
2807 spa_config_exit(spa, SCL_STATE, spa);
2808 VERIFY(mutex_unlock(&zs->zs_vdev_lock) == 0);
2809 }
2810
2811 /*
2812 * Verify that dmu_objset_{create,destroy,open,close} work as expected.
2813 */
2814 /* ARGSUSED */
2815 static void
2816 ztest_objset_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
2817 {
2818 /*
2819 * Create the objects common to all ztest datasets.
2820 */
2821 VERIFY(zap_create_claim(os, ZTEST_DIROBJ,
2822 DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx) == 0);
2823 }
2824
2825 static int
2826 ztest_dataset_create(char *dsname)
2827 {
2828 uint64_t zilset = ztest_random(100);
2829 int err = dmu_objset_create(dsname, DMU_OST_OTHER, 0,
2830 ztest_objset_create_cb, NULL);
2831
2832 if (err || zilset < 80)
2833 return (err);
2834
2835 (void) printf("Setting dataset %s to sync always\n", dsname);
2836 return (ztest_dsl_prop_set_uint64(dsname, ZFS_PROP_SYNC,
2837 ZFS_SYNC_ALWAYS, B_FALSE));
2838 }
2839
2840 /* ARGSUSED */
2841 static int
2842 ztest_objset_destroy_cb(const char *name, void *arg)
2843 {
2844 objset_t *os;
2845 dmu_object_info_t doi;
2846 int error;
2847
2848 /*
2849 * Verify that the dataset contains a directory object.
2850 */
2851 VERIFY3U(0, ==, dmu_objset_hold(name, FTAG, &os));
2852 error = dmu_object_info(os, ZTEST_DIROBJ, &doi);
2853 if (error != ENOENT) {
2854 /* We could have crashed in the middle of destroying it */
2855 ASSERT3U(error, ==, 0);
2856 ASSERT3U(doi.doi_type, ==, DMU_OT_ZAP_OTHER);
2857 ASSERT3S(doi.doi_physical_blocks_512, >=, 0);
2858 }
2859 dmu_objset_rele(os, FTAG);
2860
2861 /*
2862 * Destroy the dataset.
2863 */
2864 VERIFY3U(0, ==, dmu_objset_destroy(name, B_FALSE));
2865 return (0);
2866 }
2867
2868 static boolean_t
2869 ztest_snapshot_create(char *osname, uint64_t id)
2870 {
2871 char snapname[MAXNAMELEN];
2872 int error;
2873
2874 (void) snprintf(snapname, MAXNAMELEN, "%s@%llu", osname,
2875 (u_longlong_t)id);
2876
2877 error = dmu_objset_snapshot(osname, strchr(snapname, '@') + 1,
2878 NULL, B_FALSE);
2879 if (error == ENOSPC) {
2880 ztest_record_enospc(FTAG);
2881 return (B_FALSE);
2882 }
2883 if (error != 0 && error != EEXIST)
2884 fatal(0, "ztest_snapshot_create(%s) = %d", snapname, error);
2885 return (B_TRUE);
2886 }
2887
2888 static boolean_t
2889 ztest_snapshot_destroy(char *osname, uint64_t id)
2890 {
2891 char snapname[MAXNAMELEN];
2892 int error;
2893
2894 (void) snprintf(snapname, MAXNAMELEN, "%s@%llu", osname,
2895 (u_longlong_t)id);
2896
2897 error = dmu_objset_destroy(snapname, B_FALSE);
2898 if (error != 0 && error != ENOENT)
2899 fatal(0, "ztest_snapshot_destroy(%s) = %d", snapname, error);
2900 return (B_TRUE);
2901 }
2902
2903 /* ARGSUSED */
2904 void
2905 ztest_dmu_objset_create_destroy(ztest_ds_t *zd, uint64_t id)
2906 {
2907 ztest_shared_t *zs = ztest_shared;
2908 ztest_ds_t zdtmp;
2909 int iters;
2910 int error;
2911 objset_t *os, *os2;
2912 char name[MAXNAMELEN];
2913 zilog_t *zilog;
2914
2915 (void) rw_rdlock(&zs->zs_name_lock);
2916
2917 (void) snprintf(name, MAXNAMELEN, "%s/temp_%llu",
2918 zs->zs_pool, (u_longlong_t)id);
2919
2920 /*
2921 * If this dataset exists from a previous run, process its replay log
2922 * half of the time. If we don't replay it, then dmu_objset_destroy()
2923 * (invoked from ztest_objset_destroy_cb()) should just throw it away.
2924 */
2925 if (ztest_random(2) == 0 &&
2926 dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os) == 0) {
2927 ztest_zd_init(&zdtmp, os);
2928 zil_replay(os, &zdtmp, ztest_replay_vector);
2929 ztest_zd_fini(&zdtmp);
2930 dmu_objset_disown(os, FTAG);
2931 }
2932
2933 /*
2934 * There may be an old instance of the dataset we're about to
2935 * create lying around from a previous run. If so, destroy it
2936 * and all of its snapshots.
2937 */
2938 (void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
2939 DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
2940
2941 /*
2942 * Verify that the destroyed dataset is no longer in the namespace.
2943 */
2944 VERIFY3U(ENOENT, ==, dmu_objset_hold(name, FTAG, &os));
2945
2946 /*
2947 * Verify that we can create a new dataset.
2948 */
2949 error = ztest_dataset_create(name);
2950 if (error) {
2951 if (error == ENOSPC) {
2952 ztest_record_enospc(FTAG);
2953 (void) rw_unlock(&zs->zs_name_lock);
2954 return;
2955 }
2956 fatal(0, "dmu_objset_create(%s) = %d", name, error);
2957 }
2958
2959 VERIFY3U(0, ==,
2960 dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os));
2961
2962 ztest_zd_init(&zdtmp, os);
2963
2964 /*
2965 * Open the intent log for it.
2966 */
2967 zilog = zil_open(os, ztest_get_data);
2968
2969 /*
2970 * Put some objects in there, do a little I/O to them,
2971 * and randomly take a couple of snapshots along the way.
2972 */
2973 iters = ztest_random(5);
2974 for (int i = 0; i < iters; i++) {
2975 ztest_dmu_object_alloc_free(&zdtmp, id);
2976 if (ztest_random(iters) == 0)
2977 (void) ztest_snapshot_create(name, i);
2978 }
2979
2980 /*
2981 * Verify that we cannot create an existing dataset.
2982 */
2983 VERIFY3U(EEXIST, ==,
2984 dmu_objset_create(name, DMU_OST_OTHER, 0, NULL, NULL));
2985
2986 /*
2987 * Verify that we can hold an objset that is also owned.
2988 */
2989 VERIFY3U(0, ==, dmu_objset_hold(name, FTAG, &os2));
2990 dmu_objset_rele(os2, FTAG);
2991
2992 /*
2993 * Verify that we cannot own an objset that is already owned.
2994 */
2995 VERIFY3U(EBUSY, ==,
2996 dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os2));
2997
2998 zil_close(zilog);
2999 dmu_objset_disown(os, FTAG);
3000 ztest_zd_fini(&zdtmp);
3001
3002 (void) rw_unlock(&zs->zs_name_lock);
3003 }
3004
3005 /*
3006 * Verify that dmu_snapshot_{create,destroy,open,close} work as expected.
3007 */
3008 void
3009 ztest_dmu_snapshot_create_destroy(ztest_ds_t *zd, uint64_t id)
3010 {
3011 ztest_shared_t *zs = ztest_shared;
3012
3013 (void) rw_rdlock(&zs->zs_name_lock);
3014 (void) ztest_snapshot_destroy(zd->zd_name, id);
3015 (void) ztest_snapshot_create(zd->zd_name, id);
3016 (void) rw_unlock(&zs->zs_name_lock);
3017 }
3018
3019 /*
3020 * Cleanup non-standard snapshots and clones.
3021 */
3022 void
3023 ztest_dsl_dataset_cleanup(char *osname, uint64_t id)
3024 {
3025 char snap1name[MAXNAMELEN];
3026 char clone1name[MAXNAMELEN];
3027 char snap2name[MAXNAMELEN];
3028 char clone2name[MAXNAMELEN];
3029 char snap3name[MAXNAMELEN];
3030 int error;
3031
3032 (void) snprintf(snap1name, MAXNAMELEN, "%s@s1_%llu", osname, id);
3033 (void) snprintf(clone1name, MAXNAMELEN, "%s/c1_%llu", osname, id);
3034 (void) snprintf(snap2name, MAXNAMELEN, "%s@s2_%llu", clone1name, id);
3035 (void) snprintf(clone2name, MAXNAMELEN, "%s/c2_%llu", osname, id);
3036 (void) snprintf(snap3name, MAXNAMELEN, "%s@s3_%llu", clone1name, id);
3037
3038 error = dmu_objset_destroy(clone2name, B_FALSE);
3039 if (error && error != ENOENT)
3040 fatal(0, "dmu_objset_destroy(%s) = %d", clone2name, error);
3041 error = dmu_objset_destroy(snap3name, B_FALSE);
3042 if (error && error != ENOENT)
3043 fatal(0, "dmu_objset_destroy(%s) = %d", snap3name, error);
3044 error = dmu_objset_destroy(snap2name, B_FALSE);
3045 if (error && error != ENOENT)
3046 fatal(0, "dmu_objset_destroy(%s) = %d", snap2name, error);
3047 error = dmu_objset_destroy(clone1name, B_FALSE);
3048 if (error && error != ENOENT)
3049 fatal(0, "dmu_objset_destroy(%s) = %d", clone1name, error);
3050 error = dmu_objset_destroy(snap1name, B_FALSE);
3051 if (error && error != ENOENT)
3052 fatal(0, "dmu_objset_destroy(%s) = %d", snap1name, error);
3053 }
3054
3055 /*
3056 * Verify dsl_dataset_promote handles EBUSY
3057 */
3058 void
3059 ztest_dsl_dataset_promote_busy(ztest_ds_t *zd, uint64_t id)
3060 {
3061 ztest_shared_t *zs = ztest_shared;
3062 objset_t *clone;
3063 dsl_dataset_t *ds;
3064 char snap1name[MAXNAMELEN];
3065 char clone1name[MAXNAMELEN];
3066 char snap2name[MAXNAMELEN];
3067 char clone2name[MAXNAMELEN];
3068 char snap3name[MAXNAMELEN];
3069 char *osname = zd->zd_name;
3070 int error;
3071
3072 (void) rw_rdlock(&zs->zs_name_lock);
3073
3074 ztest_dsl_dataset_cleanup(osname, id);
3075
3076 (void) snprintf(snap1name, MAXNAMELEN, "%s@s1_%llu", osname, id);
3077 (void) snprintf(clone1name, MAXNAMELEN, "%s/c1_%llu", osname, id);
3078 (void) snprintf(snap2name, MAXNAMELEN, "%s@s2_%llu", clone1name, id);
3079 (void) snprintf(clone2name, MAXNAMELEN, "%s/c2_%llu", osname, id);
3080 (void) snprintf(snap3name, MAXNAMELEN, "%s@s3_%llu", clone1name, id);
3081
3082 error = dmu_objset_snapshot(osname, strchr(snap1name, '@')+1,
3083 NULL, B_FALSE);
3084 if (error && error != EEXIST) {
3085 if (error == ENOSPC) {
3086 ztest_record_enospc(FTAG);
3087 goto out;
3088 }
3089 fatal(0, "dmu_take_snapshot(%s) = %d", snap1name, error);
3090 }
3091
3092 error = dmu_objset_hold(snap1name, FTAG, &clone);
3093 if (error)
3094 fatal(0, "dmu_open_snapshot(%s) = %d", snap1name, error);
3095
3096 error = dmu_objset_clone(clone1name, dmu_objset_ds(clone), 0);
3097 dmu_objset_rele(clone, FTAG);
3098 if (error) {
3099 if (error == ENOSPC) {
3100 ztest_record_enospc(FTAG);
3101 goto out;
3102 }
3103 fatal(0, "dmu_objset_create(%s) = %d", clone1name, error);
3104 }
3105
3106 error = dmu_objset_snapshot(clone1name, strchr(snap2name, '@')+1,
3107 NULL, B_FALSE);
3108 if (error && error != EEXIST) {
3109 if (error == ENOSPC) {
3110 ztest_record_enospc(FTAG);
3111 goto out;
3112 }
3113 fatal(0, "dmu_open_snapshot(%s) = %d", snap2name, error);
3114 }
3115
3116 error = dmu_objset_snapshot(clone1name, strchr(snap3name, '@')+1,
3117 NULL, B_FALSE);
3118 if (error && error != EEXIST) {
3119 if (error == ENOSPC) {
3120 ztest_record_enospc(FTAG);
3121 goto out;
3122 }
3123 fatal(0, "dmu_open_snapshot(%s) = %d", snap3name, error);
3124 }
3125
3126 error = dmu_objset_hold(snap3name, FTAG, &clone);
3127 if (error)
3128 fatal(0, "dmu_open_snapshot(%s) = %d", snap3name, error);
3129
3130 error = dmu_objset_clone(clone2name, dmu_objset_ds(clone), 0);
3131 dmu_objset_rele(clone, FTAG);
3132 if (error) {
3133 if (error == ENOSPC) {
3134 ztest_record_enospc(FTAG);
3135 goto out;
3136 }
3137 fatal(0, "dmu_objset_create(%s) = %d", clone2name, error);
3138 }
3139
3140 error = dsl_dataset_own(snap2name, B_FALSE, FTAG, &ds);
3141 if (error)
3142 fatal(0, "dsl_dataset_own(%s) = %d", snap2name, error);
3143 error = dsl_dataset_promote(clone2name, NULL);
3144 if (error != EBUSY)
3145 fatal(0, "dsl_dataset_promote(%s), %d, not EBUSY", clone2name,
3146 error);
3147 dsl_dataset_disown(ds, FTAG);
3148
3149 out:
3150 ztest_dsl_dataset_cleanup(osname, id);
3151
3152 (void) rw_unlock(&zs->zs_name_lock);
3153 }
3154
3155 /*
3156 * Verify that dmu_object_{alloc,free} work as expected.
3157 */
3158 void
3159 ztest_dmu_object_alloc_free(ztest_ds_t *zd, uint64_t id)
3160 {
3161 ztest_od_t od[4];
3162 int batchsize = sizeof (od) / sizeof (od[0]);
3163
3164 for (int b = 0; b < batchsize; b++)
3165 ztest_od_init(&od[b], id, FTAG, b, DMU_OT_UINT64_OTHER, 0, 0);
3166
3167 /*
3168 * Destroy the previous batch of objects, create a new batch,
3169 * and do some I/O on the new objects.
3170 */
3171 if (ztest_object_init(zd, od, sizeof (od), B_TRUE) != 0)
3172 return;
3173
3174 while (ztest_random(4 * batchsize) != 0)
3175 ztest_io(zd, od[ztest_random(batchsize)].od_object,
3176 ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
3177 }
3178
3179 /*
3180 * Verify that dmu_{read,write} work as expected.
3181 */
3182 void
3183 ztest_dmu_read_write(ztest_ds_t *zd, uint64_t id)
3184 {
3185 objset_t *os = zd->zd_os;
3186 ztest_od_t od[2];
3187 dmu_tx_t *tx;
3188 int i, freeit, error;
3189 uint64_t n, s, txg;
3190 bufwad_t *packbuf, *bigbuf, *pack, *bigH, *bigT;
3191 uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
3192 uint64_t chunksize = (1000 + ztest_random(1000)) * sizeof (uint64_t);
3193 uint64_t regions = 997;
3194 uint64_t stride = 123456789ULL;
3195 uint64_t width = 40;
3196 int free_percent = 5;
3197
3198 /*
3199 * This test uses two objects, packobj and bigobj, that are always
3200 * updated together (i.e. in the same tx) so that their contents are
3201 * in sync and can be compared. Their contents relate to each other
3202 * in a simple way: packobj is a dense array of 'bufwad' structures,
3203 * while bigobj is a sparse array of the same bufwads. Specifically,
3204 * for any index n, there are three bufwads that should be identical:
3205 *
3206 * packobj, at offset n * sizeof (bufwad_t)
3207 * bigobj, at the head of the nth chunk
3208 * bigobj, at the tail of the nth chunk
3209 *
3210 * The chunk size is arbitrary. It doesn't have to be a power of two,
3211 * and it doesn't have any relation to the object blocksize.
3212 * The only requirement is that it can hold at least two bufwads.
3213 *
3214 * Normally, we write the bufwad to each of these locations.
3215 * However, free_percent of the time we instead write zeroes to
3216 * packobj and perform a dmu_free_range() on bigobj. By comparing
3217 * bigobj to packobj, we can verify that the DMU is correctly
3218 * tracking which parts of an object are allocated and free,
3219 * and that the contents of the allocated blocks are correct.
3220 */
3221
3222 /*
3223 * Read the directory info. If it's the first time, set things up.
3224 */
3225 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, chunksize);
3226 ztest_od_init(&od[1], id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize);
3227
3228 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
3229 return;
3230
3231 bigobj = od[0].od_object;
3232 packobj = od[1].od_object;
3233 chunksize = od[0].od_gen;
3234 ASSERT(chunksize == od[1].od_gen);
3235
3236 /*
3237 * Prefetch a random chunk of the big object.
3238 * Our aim here is to get some async reads in flight
3239 * for blocks that we may free below; the DMU should
3240 * handle this race correctly.
3241 */
3242 n = ztest_random(regions) * stride + ztest_random(width);
3243 s = 1 + ztest_random(2 * width - 1);
3244 dmu_prefetch(os, bigobj, n * chunksize, s * chunksize);
3245
3246 /*
3247 * Pick a random index and compute the offsets into packobj and bigobj.
3248 */
3249 n = ztest_random(regions) * stride + ztest_random(width);
3250 s = 1 + ztest_random(width - 1);
3251
3252 packoff = n * sizeof (bufwad_t);
3253 packsize = s * sizeof (bufwad_t);
3254
3255 bigoff = n * chunksize;
3256 bigsize = s * chunksize;
3257
3258 packbuf = umem_alloc(packsize, UMEM_NOFAIL);
3259 bigbuf = umem_alloc(bigsize, UMEM_NOFAIL);
3260
3261 /*
3262 * free_percent of the time, free a range of bigobj rather than
3263 * overwriting it.
3264 */
3265 freeit = (ztest_random(100) < free_percent);
3266
3267 /*
3268 * Read the current contents of our objects.
3269 */
3270 error = dmu_read(os, packobj, packoff, packsize, packbuf,
3271 DMU_READ_PREFETCH);
3272 ASSERT3U(error, ==, 0);
3273 error = dmu_read(os, bigobj, bigoff, bigsize, bigbuf,
3274 DMU_READ_PREFETCH);
3275 ASSERT3U(error, ==, 0);
3276
3277 /*
3278 * Get a tx for the mods to both packobj and bigobj.
3279 */
3280 tx = dmu_tx_create(os);
3281
3282 dmu_tx_hold_write(tx, packobj, packoff, packsize);
3283
3284 if (freeit)
3285 dmu_tx_hold_free(tx, bigobj, bigoff, bigsize);
3286 else
3287 dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
3288
3289 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3290 if (txg == 0) {
3291 umem_free(packbuf, packsize);
3292 umem_free(bigbuf, bigsize);
3293 return;
3294 }
3295
3296 dmu_object_set_checksum(os, bigobj,
3297 (enum zio_checksum)ztest_random_dsl_prop(ZFS_PROP_CHECKSUM), tx);
3298
3299 dmu_object_set_compress(os, bigobj,
3300 (enum zio_compress)ztest_random_dsl_prop(ZFS_PROP_COMPRESSION), tx);
3301
3302 /*
3303 * For each index from n to n + s, verify that the existing bufwad
3304 * in packobj matches the bufwads at the head and tail of the
3305 * corresponding chunk in bigobj. Then update all three bufwads
3306 * with the new values we want to write out.
3307 */
3308 for (i = 0; i < s; i++) {
3309 /* LINTED */
3310 pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
3311 /* LINTED */
3312 bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
3313 /* LINTED */
3314 bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
3315
3316 ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
3317 ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
3318
3319 if (pack->bw_txg > txg)
3320 fatal(0, "future leak: got %llx, open txg is %llx",
3321 pack->bw_txg, txg);
3322
3323 if (pack->bw_data != 0 && pack->bw_index != n + i)
3324 fatal(0, "wrong index: got %llx, wanted %llx+%llx",
3325 pack->bw_index, n, i);
3326
3327 if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
3328 fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
3329
3330 if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
3331 fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
3332
3333 if (freeit) {
3334 bzero(pack, sizeof (bufwad_t));
3335 } else {
3336 pack->bw_index = n + i;
3337 pack->bw_txg = txg;
3338 pack->bw_data = 1 + ztest_random(-2ULL);
3339 }
3340 *bigH = *pack;
3341 *bigT = *pack;
3342 }
3343
3344 /*
3345 * We've verified all the old bufwads, and made new ones.
3346 * Now write them out.
3347 */
3348 dmu_write(os, packobj, packoff, packsize, packbuf, tx);
3349
3350 if (freeit) {
3351 if (zopt_verbose >= 7) {
3352 (void) printf("freeing offset %llx size %llx"
3353 " txg %llx\n",
3354 (u_longlong_t)bigoff,
3355 (u_longlong_t)bigsize,
3356 (u_longlong_t)txg);
3357 }
3358 VERIFY(0 == dmu_free_range(os, bigobj, bigoff, bigsize, tx));
3359 } else {
3360 if (zopt_verbose >= 7) {
3361 (void) printf("writing offset %llx size %llx"
3362 " txg %llx\n",
3363 (u_longlong_t)bigoff,
3364 (u_longlong_t)bigsize,
3365 (u_longlong_t)txg);
3366 }
3367 dmu_write(os, bigobj, bigoff, bigsize, bigbuf, tx);
3368 }
3369
3370 dmu_tx_commit(tx);
3371
3372 /*
3373 * Sanity check the stuff we just wrote.
3374 */
3375 {
3376 void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
3377 void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
3378
3379 VERIFY(0 == dmu_read(os, packobj, packoff,
3380 packsize, packcheck, DMU_READ_PREFETCH));
3381 VERIFY(0 == dmu_read(os, bigobj, bigoff,
3382 bigsize, bigcheck, DMU_READ_PREFETCH));
3383
3384 ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
3385 ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
3386
3387 umem_free(packcheck, packsize);
3388 umem_free(bigcheck, bigsize);
3389 }
3390
3391 umem_free(packbuf, packsize);
3392 umem_free(bigbuf, bigsize);
3393 }
3394
3395 void
3396 compare_and_update_pbbufs(uint64_t s, bufwad_t *packbuf, bufwad_t *bigbuf,
3397 uint64_t bigsize, uint64_t n, uint64_t chunksize, uint64_t txg)
3398 {
3399 uint64_t i;
3400 bufwad_t *pack;
3401 bufwad_t *bigH;
3402 bufwad_t *bigT;
3403
3404 /*
3405 * For each index from n to n + s, verify that the existing bufwad
3406 * in packobj matches the bufwads at the head and tail of the
3407 * corresponding chunk in bigobj. Then update all three bufwads
3408 * with the new values we want to write out.
3409 */
3410 for (i = 0; i < s; i++) {
3411 /* LINTED */
3412 pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
3413 /* LINTED */
3414 bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
3415 /* LINTED */
3416 bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
3417
3418 ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
3419 ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
3420
3421 if (pack->bw_txg > txg)
3422 fatal(0, "future leak: got %llx, open txg is %llx",
3423 pack->bw_txg, txg);
3424
3425 if (pack->bw_data != 0 && pack->bw_index != n + i)
3426 fatal(0, "wrong index: got %llx, wanted %llx+%llx",
3427 pack->bw_index, n, i);
3428
3429 if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
3430 fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
3431
3432 if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
3433 fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
3434
3435 pack->bw_index = n + i;
3436 pack->bw_txg = txg;
3437 pack->bw_data = 1 + ztest_random(-2ULL);
3438
3439 *bigH = *pack;
3440 *bigT = *pack;
3441 }
3442 }
3443
3444 void
3445 ztest_dmu_read_write_zcopy(ztest_ds_t *zd, uint64_t id)
3446 {
3447 objset_t *os = zd->zd_os;
3448 ztest_od_t od[2];
3449 dmu_tx_t *tx;
3450 uint64_t i;
3451 int error;
3452 uint64_t n, s, txg;
3453 bufwad_t *packbuf, *bigbuf;
3454 uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
3455 uint64_t blocksize = ztest_random_blocksize();
3456 uint64_t chunksize = blocksize;
3457 uint64_t regions = 997;
3458 uint64_t stride = 123456789ULL;
3459 uint64_t width = 9;
3460 dmu_buf_t *bonus_db;
3461 arc_buf_t **bigbuf_arcbufs;
3462 dmu_object_info_t doi;
3463
3464 /*
3465 * This test uses two objects, packobj and bigobj, that are always
3466 * updated together (i.e. in the same tx) so that their contents are
3467 * in sync and can be compared. Their contents relate to each other
3468 * in a simple way: packobj is a dense array of 'bufwad' structures,
3469 * while bigobj is a sparse array of the same bufwads. Specifically,
3470 * for any index n, there are three bufwads that should be identical:
3471 *
3472 * packobj, at offset n * sizeof (bufwad_t)
3473 * bigobj, at the head of the nth chunk
3474 * bigobj, at the tail of the nth chunk
3475 *
3476 * The chunk size is set equal to bigobj block size so that
3477 * dmu_assign_arcbuf() can be tested for object updates.
3478 */
3479
3480 /*
3481 * Read the directory info. If it's the first time, set things up.
3482 */
3483 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
3484 ztest_od_init(&od[1], id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize);
3485
3486 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
3487 return;
3488
3489 bigobj = od[0].od_object;
3490 packobj = od[1].od_object;
3491 blocksize = od[0].od_blocksize;
3492 chunksize = blocksize;
3493 ASSERT(chunksize == od[1].od_gen);
3494
3495 VERIFY(dmu_object_info(os, bigobj, &doi) == 0);
3496 VERIFY(ISP2(doi.doi_data_block_size));
3497 VERIFY(chunksize == doi.doi_data_block_size);
3498 VERIFY(chunksize >= 2 * sizeof (bufwad_t));
3499
3500 /*
3501 * Pick a random index and compute the offsets into packobj and bigobj.
3502 */
3503 n = ztest_random(regions) * stride + ztest_random(width);
3504 s = 1 + ztest_random(width - 1);
3505
3506 packoff = n * sizeof (bufwad_t);
3507 packsize = s * sizeof (bufwad_t);
3508
3509 bigoff = n * chunksize;
3510 bigsize = s * chunksize;
3511
3512 packbuf = umem_zalloc(packsize, UMEM_NOFAIL);
3513 bigbuf = umem_zalloc(bigsize, UMEM_NOFAIL);
3514
3515 VERIFY3U(0, ==, dmu_bonus_hold(os, bigobj, FTAG, &bonus_db));
3516
3517 bigbuf_arcbufs = umem_zalloc(2 * s * sizeof (arc_buf_t *), UMEM_NOFAIL);
3518
3519 /*
3520 * Iteration 0 test zcopy for DB_UNCACHED dbufs.
3521 * Iteration 1 test zcopy to already referenced dbufs.
3522 * Iteration 2 test zcopy to dirty dbuf in the same txg.
3523 * Iteration 3 test zcopy to dbuf dirty in previous txg.
3524 * Iteration 4 test zcopy when dbuf is no longer dirty.
3525 * Iteration 5 test zcopy when it can't be done.
3526 * Iteration 6 one more zcopy write.
3527 */
3528 for (i = 0; i < 7; i++) {
3529 uint64_t j;
3530 uint64_t off;
3531
3532 /*
3533 * In iteration 5 (i == 5) use arcbufs
3534 * that don't match bigobj blksz to test
3535 * dmu_assign_arcbuf() when it can't directly
3536 * assign an arcbuf to a dbuf.
3537 */
3538 for (j = 0; j < s; j++) {
3539 if (i != 5) {
3540 bigbuf_arcbufs[j] =
3541 dmu_request_arcbuf(bonus_db, chunksize);
3542 } else {
3543 bigbuf_arcbufs[2 * j] =
3544 dmu_request_arcbuf(bonus_db, chunksize / 2);
3545 bigbuf_arcbufs[2 * j + 1] =
3546 dmu_request_arcbuf(bonus_db, chunksize / 2);
3547 }
3548 }
3549
3550 /*
3551 * Get a tx for the mods to both packobj and bigobj.
3552 */
3553 tx = dmu_tx_create(os);
3554
3555 dmu_tx_hold_write(tx, packobj, packoff, packsize);
3556 dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
3557
3558 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3559 if (txg == 0) {
3560 umem_free(packbuf, packsize);
3561 umem_free(bigbuf, bigsize);
3562 for (j = 0; j < s; j++) {
3563 if (i != 5) {
3564 dmu_return_arcbuf(bigbuf_arcbufs[j]);
3565 } else {
3566 dmu_return_arcbuf(
3567 bigbuf_arcbufs[2 * j]);
3568 dmu_return_arcbuf(
3569 bigbuf_arcbufs[2 * j + 1]);
3570 }
3571 }
3572 umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
3573 dmu_buf_rele(bonus_db, FTAG);
3574 return;
3575 }
3576
3577 /*
3578 * 50% of the time don't read objects in the 1st iteration to
3579 * test dmu_assign_arcbuf() for the case when there're no
3580 * existing dbufs for the specified offsets.
3581 */
3582 if (i != 0 || ztest_random(2) != 0) {
3583 error = dmu_read(os, packobj, packoff,
3584 packsize, packbuf, DMU_READ_PREFETCH);
3585 ASSERT3U(error, ==, 0);
3586 error = dmu_read(os, bigobj, bigoff, bigsize,
3587 bigbuf, DMU_READ_PREFETCH);
3588 ASSERT3U(error, ==, 0);
3589 }
3590 compare_and_update_pbbufs(s, packbuf, bigbuf, bigsize,
3591 n, chunksize, txg);
3592
3593 /*
3594 * We've verified all the old bufwads, and made new ones.
3595 * Now write them out.
3596 */
3597 dmu_write(os, packobj, packoff, packsize, packbuf, tx);
3598 if (zopt_verbose >= 7) {
3599 (void) printf("writing offset %llx size %llx"
3600 " txg %llx\n",
3601 (u_longlong_t)bigoff,
3602 (u_longlong_t)bigsize,
3603 (u_longlong_t)txg);
3604 }
3605 for (off = bigoff, j = 0; j < s; j++, off += chunksize) {
3606 dmu_buf_t *dbt;
3607 if (i != 5) {
3608 bcopy((caddr_t)bigbuf + (off - bigoff),
3609 bigbuf_arcbufs[j]->b_data, chunksize);
3610 } else {
3611 bcopy((caddr_t)bigbuf + (off - bigoff),
3612 bigbuf_arcbufs[2 * j]->b_data,
3613 chunksize / 2);
3614 bcopy((caddr_t)bigbuf + (off - bigoff) +
3615 chunksize / 2,
3616 bigbuf_arcbufs[2 * j + 1]->b_data,
3617 chunksize / 2);
3618 }
3619
3620 if (i == 1) {
3621 VERIFY(dmu_buf_hold(os, bigobj, off,
3622 FTAG, &dbt, DMU_READ_NO_PREFETCH) == 0);
3623 }
3624 if (i != 5) {
3625 dmu_assign_arcbuf(bonus_db, off,
3626 bigbuf_arcbufs[j], tx);
3627 } else {
3628 dmu_assign_arcbuf(bonus_db, off,
3629 bigbuf_arcbufs[2 * j], tx);
3630 dmu_assign_arcbuf(bonus_db,
3631 off + chunksize / 2,
3632 bigbuf_arcbufs[2 * j + 1], tx);
3633 }
3634 if (i == 1) {
3635 dmu_buf_rele(dbt, FTAG);
3636 }
3637 }
3638 dmu_tx_commit(tx);
3639
3640 /*
3641 * Sanity check the stuff we just wrote.
3642 */
3643 {
3644 void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
3645 void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
3646
3647 VERIFY(0 == dmu_read(os, packobj, packoff,
3648 packsize, packcheck, DMU_READ_PREFETCH));
3649 VERIFY(0 == dmu_read(os, bigobj, bigoff,
3650 bigsize, bigcheck, DMU_READ_PREFETCH));
3651
3652 ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
3653 ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
3654
3655 umem_free(packcheck, packsize);
3656 umem_free(bigcheck, bigsize);
3657 }
3658 if (i == 2) {
3659 txg_wait_open(dmu_objset_pool(os), 0);
3660 } else if (i == 3) {
3661 txg_wait_synced(dmu_objset_pool(os), 0);
3662 }
3663 }
3664
3665 dmu_buf_rele(bonus_db, FTAG);
3666 umem_free(packbuf, packsize);
3667 umem_free(bigbuf, bigsize);
3668 umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
3669 }
3670
3671 /* ARGSUSED */
3672 void
3673 ztest_dmu_write_parallel(ztest_ds_t *zd, uint64_t id)
3674 {
3675 ztest_od_t od[1];
3676 uint64_t offset = (1ULL << (ztest_random(20) + 43)) +
3677 (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
3678
3679 /*
3680 * Have multiple threads write to large offsets in an object
3681 * to verify that parallel writes to an object -- even to the
3682 * same blocks within the object -- doesn't cause any trouble.
3683 */
3684 ztest_od_init(&od[0], ID_PARALLEL, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
3685
3686 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
3687 return;
3688
3689 while (ztest_random(10) != 0)
3690 ztest_io(zd, od[0].od_object, offset);
3691 }
3692
3693 void
3694 ztest_dmu_prealloc(ztest_ds_t *zd, uint64_t id)
3695 {
3696 ztest_od_t od[1];
3697 uint64_t offset = (1ULL << (ztest_random(4) + SPA_MAXBLOCKSHIFT)) +
3698 (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
3699 uint64_t count = ztest_random(20) + 1;
3700 uint64_t blocksize = ztest_random_blocksize();
3701 void *data;
3702
3703 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
3704
3705 if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
3706 return;
3707
3708 if (ztest_truncate(zd, od[0].od_object, offset, count * blocksize) != 0)
3709 return;
3710
3711 ztest_prealloc(zd, od[0].od_object, offset, count * blocksize);
3712
3713 data = umem_zalloc(blocksize, UMEM_NOFAIL);
3714
3715 while (ztest_random(count) != 0) {
3716 uint64_t randoff = offset + (ztest_random(count) * blocksize);
3717 if (ztest_write(zd, od[0].od_object, randoff, blocksize,
3718 data) != 0)
3719 break;
3720 while (ztest_random(4) != 0)
3721 ztest_io(zd, od[0].od_object, randoff);
3722 }
3723
3724 umem_free(data, blocksize);
3725 }
3726
3727 /*
3728 * Verify that zap_{create,destroy,add,remove,update} work as expected.
3729 */
3730 #define ZTEST_ZAP_MIN_INTS 1
3731 #define ZTEST_ZAP_MAX_INTS 4
3732 #define ZTEST_ZAP_MAX_PROPS 1000
3733
3734 void
3735 ztest_zap(ztest_ds_t *zd, uint64_t id)
3736 {
3737 objset_t *os = zd->zd_os;
3738 ztest_od_t od[1];
3739 uint64_t object;
3740 uint64_t txg, last_txg;
3741 uint64_t value[ZTEST_ZAP_MAX_INTS];
3742 uint64_t zl_ints, zl_intsize, prop;
3743 int i, ints;
3744 dmu_tx_t *tx;
3745 char propname[100], txgname[100];
3746 int error;
3747 char *hc[2] = { "s.acl.h", ".s.open.h.hyLZlg" };
3748
3749 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0);
3750
3751 if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
3752 return;
3753
3754 object = od[0].od_object;
3755
3756 /*
3757 * Generate a known hash collision, and verify that
3758 * we can lookup and remove both entries.
3759 */
3760 tx = dmu_tx_create(os);
3761 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
3762 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3763 if (txg == 0)
3764 return;
3765 for (i = 0; i < 2; i++) {
3766 value[i] = i;
3767 VERIFY3U(0, ==, zap_add(os, object, hc[i], sizeof (uint64_t),
3768 1, &value[i], tx));
3769 }
3770 for (i = 0; i < 2; i++) {
3771 VERIFY3U(EEXIST, ==, zap_add(os, object, hc[i],
3772 sizeof (uint64_t), 1, &value[i], tx));
3773 VERIFY3U(0, ==,
3774 zap_length(os, object, hc[i], &zl_intsize, &zl_ints));
3775 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
3776 ASSERT3U(zl_ints, ==, 1);
3777 }
3778 for (i = 0; i < 2; i++) {
3779 VERIFY3U(0, ==, zap_remove(os, object, hc[i], tx));
3780 }
3781 dmu_tx_commit(tx);
3782
3783 /*
3784 * Generate a buch of random entries.
3785 */
3786 ints = MAX(ZTEST_ZAP_MIN_INTS, object % ZTEST_ZAP_MAX_INTS);
3787
3788 prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
3789 (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
3790 (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
3791 bzero(value, sizeof (value));
3792 last_txg = 0;
3793
3794 /*
3795 * If these zap entries already exist, validate their contents.
3796 */
3797 error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
3798 if (error == 0) {
3799 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
3800 ASSERT3U(zl_ints, ==, 1);
3801
3802 VERIFY(zap_lookup(os, object, txgname, zl_intsize,
3803 zl_ints, &last_txg) == 0);
3804
3805 VERIFY(zap_length(os, object, propname, &zl_intsize,
3806 &zl_ints) == 0);
3807
3808 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
3809 ASSERT3U(zl_ints, ==, ints);
3810
3811 VERIFY(zap_lookup(os, object, propname, zl_intsize,
3812 zl_ints, value) == 0);
3813
3814 for (i = 0; i < ints; i++) {
3815 ASSERT3U(value[i], ==, last_txg + object + i);
3816 }
3817 } else {
3818 ASSERT3U(error, ==, ENOENT);
3819 }
3820
3821 /*
3822 * Atomically update two entries in our zap object.
3823 * The first is named txg_%llu, and contains the txg
3824 * in which the property was last updated. The second
3825 * is named prop_%llu, and the nth element of its value
3826 * should be txg + object + n.
3827 */
3828 tx = dmu_tx_create(os);
3829 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
3830 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3831 if (txg == 0)
3832 return;
3833
3834 if (last_txg > txg)
3835 fatal(0, "zap future leak: old %llu new %llu", last_txg, txg);
3836
3837 for (i = 0; i < ints; i++)
3838 value[i] = txg + object + i;
3839
3840 VERIFY3U(0, ==, zap_update(os, object, txgname, sizeof (uint64_t),
3841 1, &txg, tx));
3842 VERIFY3U(0, ==, zap_update(os, object, propname, sizeof (uint64_t),
3843 ints, value, tx));
3844
3845 dmu_tx_commit(tx);
3846
3847 /*
3848 * Remove a random pair of entries.
3849 */
3850 prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
3851 (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
3852 (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
3853
3854 error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
3855
3856 if (error == ENOENT)
3857 return;
3858
3859 ASSERT3U(error, ==, 0);
3860
3861 tx = dmu_tx_create(os);
3862 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
3863 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3864 if (txg == 0)
3865 return;
3866 VERIFY3U(0, ==, zap_remove(os, object, txgname, tx));
3867 VERIFY3U(0, ==, zap_remove(os, object, propname, tx));
3868 dmu_tx_commit(tx);
3869 }
3870
3871 /*
3872 * Testcase to test the upgrading of a microzap to fatzap.
3873 */
3874 void
3875 ztest_fzap(ztest_ds_t *zd, uint64_t id)
3876 {
3877 objset_t *os = zd->zd_os;
3878 ztest_od_t od[1];
3879 uint64_t object, txg;
3880
3881 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0);
3882
3883 if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
3884 return;
3885
3886 object = od[0].od_object;
3887
3888 /*
3889 * Add entries to this ZAP and make sure it spills over
3890 * and gets upgraded to a fatzap. Also, since we are adding
3891 * 2050 entries we should see ptrtbl growth and leaf-block split.
3892 */
3893 for (int i = 0; i < 2050; i++) {
3894 char name[MAXNAMELEN];
3895 uint64_t value = i;
3896 dmu_tx_t *tx;
3897 int error;
3898
3899 (void) snprintf(name, sizeof (name), "fzap-%llu-%llu",
3900 id, value);
3901
3902 tx = dmu_tx_create(os);
3903 dmu_tx_hold_zap(tx, object, B_TRUE, name);
3904 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3905 if (txg == 0)
3906 return;
3907 error = zap_add(os, object, name, sizeof (uint64_t), 1,
3908 &value, tx);
3909 ASSERT(error == 0 || error == EEXIST);
3910 dmu_tx_commit(tx);
3911 }
3912 }
3913
3914 /* ARGSUSED */
3915 void
3916 ztest_zap_parallel(ztest_ds_t *zd, uint64_t id)
3917 {
3918 objset_t *os = zd->zd_os;
3919 ztest_od_t od[1];
3920 uint64_t txg, object, count, wsize, wc, zl_wsize, zl_wc;
3921 dmu_tx_t *tx;
3922 int i, namelen, error;
3923 int micro = ztest_random(2);
3924 char name[20], string_value[20];
3925 void *data;
3926
3927 ztest_od_init(&od[0], ID_PARALLEL, FTAG, micro, DMU_OT_ZAP_OTHER, 0, 0);
3928
3929 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
3930 return;
3931
3932 object = od[0].od_object;
3933
3934 /*
3935 * Generate a random name of the form 'xxx.....' where each
3936 * x is a random printable character and the dots are dots.
3937 * There are 94 such characters, and the name length goes from
3938 * 6 to 20, so there are 94^3 * 15 = 12,458,760 possible names.
3939 */
3940 namelen = ztest_random(sizeof (name) - 5) + 5 + 1;
3941
3942 for (i = 0; i < 3; i++)
3943 name[i] = '!' + ztest_random('~' - '!' + 1);
3944 for (; i < namelen - 1; i++)
3945 name[i] = '.';
3946 name[i] = '\0';
3947
3948 if ((namelen & 1) || micro) {
3949 wsize = sizeof (txg);
3950 wc = 1;
3951 data = &txg;
3952 } else {
3953 wsize = 1;
3954 wc = namelen;
3955 data = string_value;
3956 }
3957
3958 count = -1ULL;
3959 VERIFY(zap_count(os, object, &count) == 0);
3960 ASSERT(count != -1ULL);
3961
3962 /*
3963 * Select an operation: length, lookup, add, update, remove.
3964 */
3965 i = ztest_random(5);
3966
3967 if (i >= 2) {
3968 tx = dmu_tx_create(os);
3969 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
3970 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3971 if (txg == 0)
3972 return;
3973 bcopy(name, string_value, namelen);
3974 } else {
3975 tx = NULL;
3976 txg = 0;
3977 bzero(string_value, namelen);
3978 }
3979
3980 switch (i) {
3981
3982 case 0:
3983 error = zap_length(os, object, name, &zl_wsize, &zl_wc);
3984 if (error == 0) {
3985 ASSERT3U(wsize, ==, zl_wsize);
3986 ASSERT3U(wc, ==, zl_wc);
3987 } else {
3988 ASSERT3U(error, ==, ENOENT);
3989 }
3990 break;
3991
3992 case 1:
3993 error = zap_lookup(os, object, name, wsize, wc, data);
3994 if (error == 0) {
3995 if (data == string_value &&
3996 bcmp(name, data, namelen) != 0)
3997 fatal(0, "name '%s' != val '%s' len %d",
3998 name, data, namelen);
3999 } else {
4000 ASSERT3U(error, ==, ENOENT);
4001 }
4002 break;
4003
4004 case 2:
4005 error = zap_add(os, object, name, wsize, wc, data, tx);
4006 ASSERT(error == 0 || error == EEXIST);
4007 break;
4008
4009 case 3:
4010 VERIFY(zap_update(os, object, name, wsize, wc, data, tx) == 0);
4011 break;
4012
4013 case 4:
4014 error = zap_remove(os, object, name, tx);
4015 ASSERT(error == 0 || error == ENOENT);
4016 break;
4017 }
4018
4019 if (tx != NULL)
4020 dmu_tx_commit(tx);
4021 }
4022
4023 /*
4024 * Commit callback data.
4025 */
4026 typedef struct ztest_cb_data {
4027 list_node_t zcd_node;
4028 uint64_t zcd_txg;
4029 int zcd_expected_err;
4030 boolean_t zcd_added;
4031 boolean_t zcd_called;
4032 spa_t *zcd_spa;
4033 } ztest_cb_data_t;
4034
4035 /* This is the actual commit callback function */
4036 static void
4037 ztest_commit_callback(void *arg, int error)
4038 {
4039 ztest_cb_data_t *data = arg;
4040 uint64_t synced_txg;
4041
4042 VERIFY(data != NULL);
4043 VERIFY3S(data->zcd_expected_err, ==, error);
4044 VERIFY(!data->zcd_called);
4045
4046 synced_txg = spa_last_synced_txg(data->zcd_spa);
4047 if (data->zcd_txg > synced_txg)
4048 fatal(0, "commit callback of txg %" PRIu64 " called prematurely"
4049 ", last synced txg = %" PRIu64 "\n", data->zcd_txg,
4050 synced_txg);
4051
4052 data->zcd_called = B_TRUE;
4053
4054 if (error == ECANCELED) {
4055 ASSERT3U(data->zcd_txg, ==, 0);
4056 ASSERT(!data->zcd_added);
4057
4058 /*
4059 * The private callback data should be destroyed here, but
4060 * since we are going to check the zcd_called field after
4061 * dmu_tx_abort(), we will destroy it there.
4062 */
4063 return;
4064 }
4065
4066 /* Was this callback added to the global callback list? */
4067 if (!data->zcd_added)
4068 goto out;
4069
4070 ASSERT3U(data->zcd_txg, !=, 0);
4071
4072 /* Remove our callback from the list */
4073 (void) mutex_lock(&zcl.zcl_callbacks_lock);
4074 list_remove(&zcl.zcl_callbacks, data);
4075 (void) mutex_unlock(&zcl.zcl_callbacks_lock);
4076
4077 out:
4078 umem_free(data, sizeof (ztest_cb_data_t));
4079 }
4080
4081 /* Allocate and initialize callback data structure */
4082 static ztest_cb_data_t *
4083 ztest_create_cb_data(objset_t *os, uint64_t txg)
4084 {
4085 ztest_cb_data_t *cb_data;
4086
4087 cb_data = umem_zalloc(sizeof (ztest_cb_data_t), UMEM_NOFAIL);
4088
4089 cb_data->zcd_txg = txg;
4090 cb_data->zcd_spa = dmu_objset_spa(os);
4091
4092 return (cb_data);
4093 }
4094
4095 /*
4096 * If a number of txgs equal to this threshold have been created after a commit
4097 * callback has been registered but not called, then we assume there is an
4098 * implementation bug.
4099 */
4100 #define ZTEST_COMMIT_CALLBACK_THRESH (TXG_CONCURRENT_STATES + 2)
4101
4102 /*
4103 * Commit callback test.
4104 */
4105 void
4106 ztest_dmu_commit_callbacks(ztest_ds_t *zd, uint64_t id)
4107 {
4108 objset_t *os = zd->zd_os;
4109 ztest_od_t od[1];
4110 dmu_tx_t *tx;
4111 ztest_cb_data_t *cb_data[3], *tmp_cb;
4112 uint64_t old_txg, txg;
4113 int i, error;
4114
4115 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
4116
4117 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4118 return;
4119
4120 tx = dmu_tx_create(os);
4121
4122 cb_data[0] = ztest_create_cb_data(os, 0);
4123 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[0]);
4124
4125 dmu_tx_hold_write(tx, od[0].od_object, 0, sizeof (uint64_t));
4126
4127 /* Every once in a while, abort the transaction on purpose */
4128 if (ztest_random(100) == 0)
4129 error = -1;
4130
4131 if (!error)
4132 error = dmu_tx_assign(tx, TXG_NOWAIT);
4133
4134 txg = error ? 0 : dmu_tx_get_txg(tx);
4135
4136 cb_data[0]->zcd_txg = txg;
4137 cb_data[1] = ztest_create_cb_data(os, txg);
4138 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[1]);
4139
4140 if (error) {
4141 /*
4142 * It's not a strict requirement to call the registered
4143 * callbacks from inside dmu_tx_abort(), but that's what
4144 * it's supposed to happen in the current implementation
4145 * so we will check for that.
4146 */
4147 for (i = 0; i < 2; i++) {
4148 cb_data[i]->zcd_expected_err = ECANCELED;
4149 VERIFY(!cb_data[i]->zcd_called);
4150 }
4151
4152 dmu_tx_abort(tx);
4153
4154 for (i = 0; i < 2; i++) {
4155 VERIFY(cb_data[i]->zcd_called);
4156 umem_free(cb_data[i], sizeof (ztest_cb_data_t));
4157 }
4158
4159 return;
4160 }
4161
4162 cb_data[2] = ztest_create_cb_data(os, txg);
4163 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[2]);
4164
4165 /*
4166 * Read existing data to make sure there isn't a future leak.
4167 */
4168 VERIFY(0 == dmu_read(os, od[0].od_object, 0, sizeof (uint64_t),
4169 &old_txg, DMU_READ_PREFETCH));
4170
4171 if (old_txg > txg)
4172 fatal(0, "future leak: got %" PRIu64 ", open txg is %" PRIu64,
4173 old_txg, txg);
4174
4175 dmu_write(os, od[0].od_object, 0, sizeof (uint64_t), &txg, tx);
4176
4177 (void) mutex_lock(&zcl.zcl_callbacks_lock);
4178
4179 /*
4180 * Since commit callbacks don't have any ordering requirement and since
4181 * it is theoretically possible for a commit callback to be called
4182 * after an arbitrary amount of time has elapsed since its txg has been
4183 * synced, it is difficult to reliably determine whether a commit
4184 * callback hasn't been called due to high load or due to a flawed
4185 * implementation.
4186 *
4187 * In practice, we will assume that if after a certain number of txgs a
4188 * commit callback hasn't been called, then most likely there's an
4189 * implementation bug..
4190 */
4191 tmp_cb = list_head(&zcl.zcl_callbacks);
4192 if (tmp_cb != NULL &&
4193 tmp_cb->zcd_txg > txg - ZTEST_COMMIT_CALLBACK_THRESH) {
4194 fatal(0, "Commit callback threshold exceeded, oldest txg: %"
4195 PRIu64 ", open txg: %" PRIu64 "\n", tmp_cb->zcd_txg, txg);
4196 }
4197
4198 /*
4199 * Let's find the place to insert our callbacks.
4200 *
4201 * Even though the list is ordered by txg, it is possible for the
4202 * insertion point to not be the end because our txg may already be
4203 * quiescing at this point and other callbacks in the open txg
4204 * (from other objsets) may have sneaked in.
4205 */
4206 tmp_cb = list_tail(&zcl.zcl_callbacks);
4207 while (tmp_cb != NULL && tmp_cb->zcd_txg > txg)
4208 tmp_cb = list_prev(&zcl.zcl_callbacks, tmp_cb);
4209
4210 /* Add the 3 callbacks to the list */
4211 for (i = 0; i < 3; i++) {
4212 if (tmp_cb == NULL)
4213 list_insert_head(&zcl.zcl_callbacks, cb_data[i]);
4214 else
4215 list_insert_after(&zcl.zcl_callbacks, tmp_cb,
4216 cb_data[i]);
4217
4218 cb_data[i]->zcd_added = B_TRUE;
4219 VERIFY(!cb_data[i]->zcd_called);
4220
4221 tmp_cb = cb_data[i];
4222 }
4223
4224 (void) mutex_unlock(&zcl.zcl_callbacks_lock);
4225
4226 dmu_tx_commit(tx);
4227 }
4228
4229 /* ARGSUSED */
4230 void
4231 ztest_dsl_prop_get_set(ztest_ds_t *zd, uint64_t id)
4232 {
4233 zfs_prop_t proplist[] = {
4234 ZFS_PROP_CHECKSUM,
4235 ZFS_PROP_COMPRESSION,
4236 ZFS_PROP_COPIES,
4237 ZFS_PROP_DEDUP
4238 };
4239 ztest_shared_t *zs = ztest_shared;
4240
4241 (void) rw_rdlock(&zs->zs_name_lock);
4242
4243 for (int p = 0; p < sizeof (proplist) / sizeof (proplist[0]); p++)
4244 (void) ztest_dsl_prop_set_uint64(zd->zd_name, proplist[p],
4245 ztest_random_dsl_prop(proplist[p]), (int)ztest_random(2));
4246
4247 (void) rw_unlock(&zs->zs_name_lock);
4248 }
4249
4250 /* ARGSUSED */
4251 void
4252 ztest_spa_prop_get_set(ztest_ds_t *zd, uint64_t id)
4253 {
4254 ztest_shared_t *zs = ztest_shared;
4255 nvlist_t *props = NULL;
4256
4257 (void) rw_rdlock(&zs->zs_name_lock);
4258
4259 (void) ztest_spa_prop_set_uint64(zs, ZPOOL_PROP_DEDUPDITTO,
4260 ZIO_DEDUPDITTO_MIN + ztest_random(ZIO_DEDUPDITTO_MIN));
4261
4262 VERIFY3U(spa_prop_get(zs->zs_spa, &props), ==, 0);
4263
4264 if (zopt_verbose >= 6)
4265 dump_nvlist(props, 4);
4266
4267 nvlist_free(props);
4268
4269 (void) rw_unlock(&zs->zs_name_lock);
4270 }
4271
4272 /*
4273 * Test snapshot hold/release and deferred destroy.
4274 */
4275 void
4276 ztest_dmu_snapshot_hold(ztest_ds_t *zd, uint64_t id)
4277 {
4278 int error;
4279 objset_t *os = zd->zd_os;
4280 objset_t *origin;
4281 char snapname[100];
4282 char fullname[100];
4283 char clonename[100];
4284 char tag[100];
4285 char osname[MAXNAMELEN];
4286
4287 (void) rw_rdlock(&ztest_shared->zs_name_lock);
4288
4289 dmu_objset_name(os, osname);
4290
4291 (void) snprintf(snapname, 100, "sh1_%llu", id);
4292 (void) snprintf(fullname, 100, "%s@%s", osname, snapname);
4293 (void) snprintf(clonename, 100, "%s/ch1_%llu", osname, id);
4294 (void) snprintf(tag, 100, "%tag_%llu", id);
4295
4296 /*
4297 * Clean up from any previous run.
4298 */
4299 (void) dmu_objset_destroy(clonename, B_FALSE);
4300 (void) dsl_dataset_user_release(osname, snapname, tag, B_FALSE);
4301 (void) dmu_objset_destroy(fullname, B_FALSE);
4302
4303 /*
4304 * Create snapshot, clone it, mark snap for deferred destroy,
4305 * destroy clone, verify snap was also destroyed.
4306 */
4307 error = dmu_objset_snapshot(osname, snapname, NULL, FALSE);
4308 if (error) {
4309 if (error == ENOSPC) {
4310 ztest_record_enospc("dmu_objset_snapshot");
4311 goto out;
4312 }
4313 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
4314 }
4315
4316 error = dmu_objset_hold(fullname, FTAG, &origin);
4317 if (error)
4318 fatal(0, "dmu_objset_hold(%s) = %d", fullname, error);
4319
4320 error = dmu_objset_clone(clonename, dmu_objset_ds(origin), 0);
4321 dmu_objset_rele(origin, FTAG);
4322 if (error) {
4323 if (error == ENOSPC) {
4324 ztest_record_enospc("dmu_objset_clone");
4325 goto out;
4326 }
4327 fatal(0, "dmu_objset_clone(%s) = %d", clonename, error);
4328 }
4329
4330 error = dmu_objset_destroy(fullname, B_TRUE);
4331 if (error) {
4332 fatal(0, "dmu_objset_destroy(%s, B_TRUE) = %d",
4333 fullname, error);
4334 }
4335
4336 error = dmu_objset_destroy(clonename, B_FALSE);
4337 if (error)
4338 fatal(0, "dmu_objset_destroy(%s) = %d", clonename, error);
4339
4340 error = dmu_objset_hold(fullname, FTAG, &origin);
4341 if (error != ENOENT)
4342 fatal(0, "dmu_objset_hold(%s) = %d", fullname, error);
4343
4344 /*
4345 * Create snapshot, add temporary hold, verify that we can't
4346 * destroy a held snapshot, mark for deferred destroy,
4347 * release hold, verify snapshot was destroyed.
4348 */
4349 error = dmu_objset_snapshot(osname, snapname, NULL, FALSE);
4350 if (error) {
4351 if (error == ENOSPC) {
4352 ztest_record_enospc("dmu_objset_snapshot");
4353 goto out;
4354 }
4355 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
4356 }
4357
4358 error = dsl_dataset_user_hold(osname, snapname, tag, B_FALSE, B_TRUE);
4359 if (error)
4360 fatal(0, "dsl_dataset_user_hold(%s)", fullname, tag);
4361
4362 error = dmu_objset_destroy(fullname, B_FALSE);
4363 if (error != EBUSY) {
4364 fatal(0, "dmu_objset_destroy(%s, B_FALSE) = %d",
4365 fullname, error);
4366 }
4367
4368 error = dmu_objset_destroy(fullname, B_TRUE);
4369 if (error) {
4370 fatal(0, "dmu_objset_destroy(%s, B_TRUE) = %d",
4371 fullname, error);
4372 }
4373
4374 error = dsl_dataset_user_release(osname, snapname, tag, B_FALSE);
4375 if (error)
4376 fatal(0, "dsl_dataset_user_release(%s)", fullname, tag);
4377
4378 VERIFY(dmu_objset_hold(fullname, FTAG, &origin) == ENOENT);
4379
4380 out:
4381 (void) rw_unlock(&ztest_shared->zs_name_lock);
4382 }
4383
4384 /*
4385 * Inject random faults into the on-disk data.
4386 */
4387 /* ARGSUSED */
4388 void
4389 ztest_fault_inject(ztest_ds_t *zd, uint64_t id)
4390 {
4391 ztest_shared_t *zs = ztest_shared;
4392 spa_t *spa = zs->zs_spa;
4393 int fd;
4394 uint64_t offset;
4395 uint64_t leaves;
4396 uint64_t bad = 0x1990c0ffeedecade;
4397 uint64_t top, leaf;
4398 char path0[MAXPATHLEN];
4399 char pathrand[MAXPATHLEN];
4400 size_t fsize;
4401 int bshift = SPA_MAXBLOCKSHIFT + 2; /* don't scrog all labels */
4402 int iters = 1000;
4403 int maxfaults;
4404 int mirror_save;
4405 vdev_t *vd0 = NULL;
4406 uint64_t guid0 = 0;
4407 boolean_t islog = B_FALSE;
4408
4409 VERIFY(mutex_lock(&zs->zs_vdev_lock) == 0);
4410 maxfaults = MAXFAULTS();
4411 leaves = MAX(zs->zs_mirrors, 1) * zopt_raidz;
4412 mirror_save = zs->zs_mirrors;
4413 VERIFY(mutex_unlock(&zs->zs_vdev_lock) == 0);
4414
4415 ASSERT(leaves >= 1);
4416
4417 /*
4418 * We need SCL_STATE here because we're going to look at vd0->vdev_tsd.
4419 */
4420 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
4421
4422 if (ztest_random(2) == 0) {
4423 /*
4424 * Inject errors on a normal data device or slog device.
4425 */
4426 top = ztest_random_vdev_top(spa, B_TRUE);
4427 leaf = ztest_random(leaves) + zs->zs_splits;
4428
4429 /*
4430 * Generate paths to the first leaf in this top-level vdev,
4431 * and to the random leaf we selected. We'll induce transient
4432 * write failures and random online/offline activity on leaf 0,
4433 * and we'll write random garbage to the randomly chosen leaf.
4434 */
4435 (void) snprintf(path0, sizeof (path0), ztest_dev_template,
4436 zopt_dir, zopt_pool, top * leaves + zs->zs_splits);
4437 (void) snprintf(pathrand, sizeof (pathrand), ztest_dev_template,
4438 zopt_dir, zopt_pool, top * leaves + leaf);
4439
4440 vd0 = vdev_lookup_by_path(spa->spa_root_vdev, path0);
4441 if (vd0 != NULL && vd0->vdev_top->vdev_islog)
4442 islog = B_TRUE;
4443
4444 if (vd0 != NULL && maxfaults != 1) {
4445 /*
4446 * Make vd0 explicitly claim to be unreadable,
4447 * or unwriteable, or reach behind its back
4448 * and close the underlying fd. We can do this if
4449 * maxfaults == 0 because we'll fail and reexecute,
4450 * and we can do it if maxfaults >= 2 because we'll
4451 * have enough redundancy. If maxfaults == 1, the
4452 * combination of this with injection of random data
4453 * corruption below exceeds the pool's fault tolerance.
4454 */
4455 vdev_file_t *vf = vd0->vdev_tsd;
4456
4457 if (vf != NULL && ztest_random(3) == 0) {
4458 (void) close(vf->vf_vnode->v_fd);
4459 vf->vf_vnode->v_fd = -1;
4460 } else if (ztest_random(2) == 0) {
4461 vd0->vdev_cant_read = B_TRUE;
4462 } else {
4463 vd0->vdev_cant_write = B_TRUE;
4464 }
4465 guid0 = vd0->vdev_guid;
4466 }
4467 } else {
4468 /*
4469 * Inject errors on an l2cache device.
4470 */
4471 spa_aux_vdev_t *sav = &spa->spa_l2cache;
4472
4473 if (sav->sav_count == 0) {
4474 spa_config_exit(spa, SCL_STATE, FTAG);
4475 return;
4476 }
4477 vd0 = sav->sav_vdevs[ztest_random(sav->sav_count)];
4478 guid0 = vd0->vdev_guid;
4479 (void) strcpy(path0, vd0->vdev_path);
4480 (void) strcpy(pathrand, vd0->vdev_path);
4481
4482 leaf = 0;
4483 leaves = 1;
4484 maxfaults = INT_MAX; /* no limit on cache devices */
4485 }
4486
4487 spa_config_exit(spa, SCL_STATE, FTAG);
4488
4489 /*
4490 * If we can tolerate two or more faults, or we're dealing
4491 * with a slog, randomly online/offline vd0.
4492 */
4493 if ((maxfaults >= 2 || islog) && guid0 != 0) {
4494 if (ztest_random(10) < 6) {
4495 int flags = (ztest_random(2) == 0 ?
4496 ZFS_OFFLINE_TEMPORARY : 0);
4497
4498 /*
4499 * We have to grab the zs_name_lock as writer to
4500 * prevent a race between offlining a slog and
4501 * destroying a dataset. Offlining the slog will
4502 * grab a reference on the dataset which may cause
4503 * dmu_objset_destroy() to fail with EBUSY thus
4504 * leaving the dataset in an inconsistent state.
4505 */
4506 if (islog)
4507 (void) rw_wrlock(&ztest_shared->zs_name_lock);
4508
4509 VERIFY(vdev_offline(spa, guid0, flags) != EBUSY);
4510
4511 if (islog)
4512 (void) rw_unlock(&ztest_shared->zs_name_lock);
4513 } else {
4514 (void) vdev_online(spa, guid0, 0, NULL);
4515 }
4516 }
4517
4518 if (maxfaults == 0)
4519 return;
4520
4521 /*
4522 * We have at least single-fault tolerance, so inject data corruption.
4523 */
4524 fd = open(pathrand, O_RDWR);
4525
4526 if (fd == -1) /* we hit a gap in the device namespace */
4527 return;
4528
4529 fsize = lseek(fd, 0, SEEK_END);
4530
4531 while (--iters != 0) {
4532 offset = ztest_random(fsize / (leaves << bshift)) *
4533 (leaves << bshift) + (leaf << bshift) +
4534 (ztest_random(1ULL << (bshift - 1)) & -8ULL);
4535
4536 if (offset >= fsize)
4537 continue;
4538
4539 VERIFY(mutex_lock(&zs->zs_vdev_lock) == 0);
4540 if (mirror_save != zs->zs_mirrors) {
4541 VERIFY(mutex_unlock(&zs->zs_vdev_lock) == 0);
4542 (void) close(fd);
4543 return;
4544 }
4545
4546 if (pwrite(fd, &bad, sizeof (bad), offset) != sizeof (bad))
4547 fatal(1, "can't inject bad word at 0x%llx in %s",
4548 offset, pathrand);
4549
4550 VERIFY(mutex_unlock(&zs->zs_vdev_lock) == 0);
4551
4552 if (zopt_verbose >= 7)
4553 (void) printf("injected bad word into %s,"
4554 " offset 0x%llx\n", pathrand, (u_longlong_t)offset);
4555 }
4556
4557 (void) close(fd);
4558 }
4559
4560 /*
4561 * Verify that DDT repair works as expected.
4562 */
4563 void
4564 ztest_ddt_repair(ztest_ds_t *zd, uint64_t id)
4565 {
4566 ztest_shared_t *zs = ztest_shared;
4567 spa_t *spa = zs->zs_spa;
4568 objset_t *os = zd->zd_os;
4569 ztest_od_t od[1];
4570 uint64_t object, blocksize, txg, pattern, psize;
4571 enum zio_checksum checksum = spa_dedup_checksum(spa);
4572 dmu_buf_t *db;
4573 dmu_tx_t *tx;
4574 void *buf;
4575 blkptr_t blk;
4576 int copies = 2 * ZIO_DEDUPDITTO_MIN;
4577
4578 blocksize = ztest_random_blocksize();
4579 blocksize = MIN(blocksize, 2048); /* because we write so many */
4580
4581 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
4582
4583 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4584 return;
4585
4586 /*
4587 * Take the name lock as writer to prevent anyone else from changing
4588 * the pool and dataset properies we need to maintain during this test.
4589 */
4590 (void) rw_wrlock(&zs->zs_name_lock);
4591
4592 if (ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_DEDUP, checksum,
4593 B_FALSE) != 0 ||
4594 ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_COPIES, 1,
4595 B_FALSE) != 0) {
4596 (void) rw_unlock(&zs->zs_name_lock);
4597 return;
4598 }
4599
4600 object = od[0].od_object;
4601 blocksize = od[0].od_blocksize;
4602 pattern = spa_guid(spa) ^ dmu_objset_fsid_guid(os);
4603
4604 ASSERT(object != 0);
4605
4606 tx = dmu_tx_create(os);
4607 dmu_tx_hold_write(tx, object, 0, copies * blocksize);
4608 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
4609 if (txg == 0) {
4610 (void) rw_unlock(&zs->zs_name_lock);
4611 return;
4612 }
4613
4614 /*
4615 * Write all the copies of our block.
4616 */
4617 for (int i = 0; i < copies; i++) {
4618 uint64_t offset = i * blocksize;
4619 VERIFY(dmu_buf_hold(os, object, offset, FTAG, &db,
4620 DMU_READ_NO_PREFETCH) == 0);
4621 ASSERT(db->db_offset == offset);
4622 ASSERT(db->db_size == blocksize);
4623 ASSERT(ztest_pattern_match(db->db_data, db->db_size, pattern) ||
4624 ztest_pattern_match(db->db_data, db->db_size, 0ULL));
4625 dmu_buf_will_fill(db, tx);
4626 ztest_pattern_set(db->db_data, db->db_size, pattern);
4627 dmu_buf_rele(db, FTAG);
4628 }
4629
4630 dmu_tx_commit(tx);
4631 txg_wait_synced(spa_get_dsl(spa), txg);
4632
4633 /*
4634 * Find out what block we got.
4635 */
4636 VERIFY(dmu_buf_hold(os, object, 0, FTAG, &db,
4637 DMU_READ_NO_PREFETCH) == 0);
4638 blk = *((dmu_buf_impl_t *)db)->db_blkptr;
4639 dmu_buf_rele(db, FTAG);
4640
4641 /*
4642 * Damage the block. Dedup-ditto will save us when we read it later.
4643 */
4644 psize = BP_GET_PSIZE(&blk);
4645 buf = zio_buf_alloc(psize);
4646 ztest_pattern_set(buf, psize, ~pattern);
4647
4648 (void) zio_wait(zio_rewrite(NULL, spa, 0, &blk,
4649 buf, psize, NULL, NULL, ZIO_PRIORITY_SYNC_WRITE,
4650 ZIO_FLAG_CANFAIL | ZIO_FLAG_INDUCE_DAMAGE, NULL));
4651
4652 zio_buf_free(buf, psize);
4653
4654 (void) rw_unlock(&zs->zs_name_lock);
4655 }
4656
4657 /*
4658 * Scrub the pool.
4659 */
4660 /* ARGSUSED */
4661 void
4662 ztest_scrub(ztest_ds_t *zd, uint64_t id)
4663 {
4664 ztest_shared_t *zs = ztest_shared;
4665 spa_t *spa = zs->zs_spa;
4666
4667 (void) spa_scan(spa, POOL_SCAN_SCRUB);
4668 (void) poll(NULL, 0, 100); /* wait a moment, then force a restart */
4669 (void) spa_scan(spa, POOL_SCAN_SCRUB);
4670 }
4671
4672 /*
4673 * Rename the pool to a different name and then rename it back.
4674 */
4675 /* ARGSUSED */
4676 void
4677 ztest_spa_rename(ztest_ds_t *zd, uint64_t id)
4678 {
4679 ztest_shared_t *zs = ztest_shared;
4680 char *oldname, *newname;
4681 spa_t *spa;
4682
4683 (void) rw_wrlock(&zs->zs_name_lock);
4684
4685 oldname = zs->zs_pool;
4686 newname = umem_alloc(strlen(oldname) + 5, UMEM_NOFAIL);
4687 (void) strcpy(newname, oldname);
4688 (void) strcat(newname, "_tmp");
4689
4690 /*
4691 * Do the rename
4692 */
4693 VERIFY3U(0, ==, spa_rename(oldname, newname));
4694
4695 /*
4696 * Try to open it under the old name, which shouldn't exist
4697 */
4698 VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
4699
4700 /*
4701 * Open it under the new name and make sure it's still the same spa_t.
4702 */
4703 VERIFY3U(0, ==, spa_open(newname, &spa, FTAG));
4704
4705 ASSERT(spa == zs->zs_spa);
4706 spa_close(spa, FTAG);
4707
4708 /*
4709 * Rename it back to the original
4710 */
4711 VERIFY3U(0, ==, spa_rename(newname, oldname));
4712
4713 /*
4714 * Make sure it can still be opened
4715 */
4716 VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG));
4717
4718 ASSERT(spa == zs->zs_spa);
4719 spa_close(spa, FTAG);
4720
4721 umem_free(newname, strlen(newname) + 1);
4722
4723 (void) rw_unlock(&zs->zs_name_lock);
4724 }
4725
4726 /*
4727 * Verify pool integrity by running zdb.
4728 */
4729 static void
4730 ztest_run_zdb(char *pool)
4731 {
4732 int status;
4733 char zdb[MAXPATHLEN + MAXNAMELEN + 20];
4734 char zbuf[1024];
4735 char *bin;
4736 char *ztest;
4737 char *isa;
4738 int isalen;
4739 FILE *fp;
4740
4741 (void) realpath(getexecname(), zdb);
4742
4743 /* zdb lives in /usr/sbin, while ztest lives in /usr/bin */
4744 bin = strstr(zdb, "/usr/bin/");
4745 ztest = strstr(bin, "/ztest");
4746 isa = bin + 8;
4747 isalen = ztest - isa;
4748 isa = strdup(isa);
4749 /* LINTED */
4750 (void) sprintf(bin,
4751 "/usr/sbin%.*s/zdb -bcc%s%s -U %s %s",
4752 isalen,
4753 isa,
4754 zopt_verbose >= 3 ? "s" : "",
4755 zopt_verbose >= 4 ? "v" : "",
4756 spa_config_path,
4757 pool);
4758 free(isa);
4759
4760 if (zopt_verbose >= 5)
4761 (void) printf("Executing %s\n", strstr(zdb, "zdb "));
4762
4763 fp = popen(zdb, "r");
4764
4765 while (fgets(zbuf, sizeof (zbuf), fp) != NULL)
4766 if (zopt_verbose >= 3)
4767 (void) printf("%s", zbuf);
4768
4769 status = pclose(fp);
4770
4771 if (status == 0)
4772 return;
4773
4774 ztest_dump_core = 0;
4775 if (WIFEXITED(status))
4776 fatal(0, "'%s' exit code %d", zdb, WEXITSTATUS(status));
4777 else
4778 fatal(0, "'%s' died with signal %d", zdb, WTERMSIG(status));
4779 }
4780
4781 static void
4782 ztest_walk_pool_directory(char *header)
4783 {
4784 spa_t *spa = NULL;
4785
4786 if (zopt_verbose >= 6)
4787 (void) printf("%s\n", header);
4788
4789 mutex_enter(&spa_namespace_lock);
4790 while ((spa = spa_next(spa)) != NULL)
4791 if (zopt_verbose >= 6)
4792 (void) printf("\t%s\n", spa_name(spa));
4793 mutex_exit(&spa_namespace_lock);
4794 }
4795
4796 static void
4797 ztest_spa_import_export(char *oldname, char *newname)
4798 {
4799 nvlist_t *config, *newconfig;
4800 uint64_t pool_guid;
4801 spa_t *spa;
4802
4803 if (zopt_verbose >= 4) {
4804 (void) printf("import/export: old = %s, new = %s\n",
4805 oldname, newname);
4806 }
4807
4808 /*
4809 * Clean up from previous runs.
4810 */
4811 (void) spa_destroy(newname);
4812
4813 /*
4814 * Get the pool's configuration and guid.
4815 */
4816 VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG));
4817
4818 /*
4819 * Kick off a scrub to tickle scrub/export races.
4820 */
4821 if (ztest_random(2) == 0)
4822 (void) spa_scan(spa, POOL_SCAN_SCRUB);
4823
4824 pool_guid = spa_guid(spa);
4825 spa_close(spa, FTAG);
4826
4827 ztest_walk_pool_directory("pools before export");
4828
4829 /*
4830 * Export it.
4831 */
4832 VERIFY3U(0, ==, spa_export(oldname, &config, B_FALSE, B_FALSE));
4833
4834 ztest_walk_pool_directory("pools after export");
4835
4836 /*
4837 * Try to import it.
4838 */
4839 newconfig = spa_tryimport(config);
4840 ASSERT(newconfig != NULL);
4841 nvlist_free(newconfig);
4842
4843 /*
4844 * Import it under the new name.
4845 */
4846 VERIFY3U(0, ==, spa_import(newname, config, NULL));
4847
4848 ztest_walk_pool_directory("pools after import");
4849
4850 /*
4851 * Try to import it again -- should fail with EEXIST.
4852 */
4853 VERIFY3U(EEXIST, ==, spa_import(newname, config, NULL));
4854
4855 /*
4856 * Try to import it under a different name -- should fail with EEXIST.
4857 */
4858 VERIFY3U(EEXIST, ==, spa_import(oldname, config, NULL));
4859
4860 /*
4861 * Verify that the pool is no longer visible under the old name.
4862 */
4863 VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
4864
4865 /*
4866 * Verify that we can open and close the pool using the new name.
4867 */
4868 VERIFY3U(0, ==, spa_open(newname, &spa, FTAG));
4869 ASSERT(pool_guid == spa_guid(spa));
4870 spa_close(spa, FTAG);
4871
4872 nvlist_free(config);
4873 }
4874
4875 static void
4876 ztest_resume(spa_t *spa)
4877 {
4878 if (spa_suspended(spa) && zopt_verbose >= 6)
4879 (void) printf("resuming from suspended state\n");
4880 spa_vdev_state_enter(spa, SCL_NONE);
4881 vdev_clear(spa, NULL);
4882 (void) spa_vdev_state_exit(spa, NULL, 0);
4883 (void) zio_resume(spa);
4884 }
4885
4886 static void *
4887 ztest_resume_thread(void *arg)
4888 {
4889 spa_t *spa = arg;
4890
4891 while (!ztest_exiting) {
4892 if (spa_suspended(spa))
4893 ztest_resume(spa);
4894 (void) poll(NULL, 0, 100);
4895 }
4896 return (NULL);
4897 }
4898
4899 static void *
4900 ztest_deadman_thread(void *arg)
4901 {
4902 ztest_shared_t *zs = arg;
4903 int grace = 300;
4904 hrtime_t delta;
4905
4906 delta = (zs->zs_thread_stop - zs->zs_thread_start) / NANOSEC + grace;
4907
4908 (void) poll(NULL, 0, (int)(1000 * delta));
4909
4910 fatal(0, "failed to complete within %d seconds of deadline", grace);
4911
4912 return (NULL);
4913 }
4914
4915 static void
4916 ztest_execute(ztest_info_t *zi, uint64_t id)
4917 {
4918 ztest_shared_t *zs = ztest_shared;
4919 ztest_ds_t *zd = &zs->zs_zd[id % zopt_datasets];
4920 hrtime_t functime = gethrtime();
4921
4922 for (int i = 0; i < zi->zi_iters; i++)
4923 zi->zi_func(zd, id);
4924
4925 functime = gethrtime() - functime;
4926
4927 atomic_add_64(&zi->zi_call_count, 1);
4928 atomic_add_64(&zi->zi_call_time, functime);
4929
4930 if (zopt_verbose >= 4) {
4931 Dl_info dli;
4932 (void) dladdr((void *)zi->zi_func, &dli);
4933 (void) printf("%6.2f sec in %s\n",
4934 (double)functime / NANOSEC, dli.dli_sname);
4935 }
4936 }
4937
4938 static void *
4939 ztest_thread(void *arg)
4940 {
4941 uint64_t id = (uintptr_t)arg;
4942 ztest_shared_t *zs = ztest_shared;
4943 uint64_t call_next;
4944 hrtime_t now;
4945 ztest_info_t *zi;
4946
4947 while ((now = gethrtime()) < zs->zs_thread_stop) {
4948 /*
4949 * See if it's time to force a crash.
4950 */
4951 if (now > zs->zs_thread_kill)
4952 ztest_kill(zs);
4953
4954 /*
4955 * If we're getting ENOSPC with some regularity, stop.
4956 */
4957 if (zs->zs_enospc_count > 10)
4958 break;
4959
4960 /*
4961 * Pick a random function to execute.
4962 */
4963 zi = &zs->zs_info[ztest_random(ZTEST_FUNCS)];
4964 call_next = zi->zi_call_next;
4965
4966 if (now >= call_next &&
4967 atomic_cas_64(&zi->zi_call_next, call_next, call_next +
4968 ztest_random(2 * zi->zi_interval[0] + 1)) == call_next)
4969 ztest_execute(zi, id);
4970 }
4971
4972 return (NULL);
4973 }
4974
4975 static void
4976 ztest_dataset_name(char *dsname, char *pool, int d)
4977 {
4978 (void) snprintf(dsname, MAXNAMELEN, "%s/ds_%d", pool, d);
4979 }
4980
4981 static void
4982 ztest_dataset_destroy(ztest_shared_t *zs, int d)
4983 {
4984 char name[MAXNAMELEN];
4985
4986 ztest_dataset_name(name, zs->zs_pool, d);
4987
4988 if (zopt_verbose >= 3)
4989 (void) printf("Destroying %s to free up space\n", name);
4990
4991 /*
4992 * Cleanup any non-standard clones and snapshots. In general,
4993 * ztest thread t operates on dataset (t % zopt_datasets),
4994 * so there may be more than one thing to clean up.
4995 */
4996 for (int t = d; t < zopt_threads; t += zopt_datasets)
4997 ztest_dsl_dataset_cleanup(name, t);
4998
4999 (void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
5000 DS_FIND_SNAPSHOTS | DS_FIND_CHILDREN);
5001 }
5002
5003 static void
5004 ztest_dataset_dirobj_verify(ztest_ds_t *zd)
5005 {
5006 uint64_t usedobjs, dirobjs, scratch;
5007
5008 /*
5009 * ZTEST_DIROBJ is the object directory for the entire dataset.
5010 * Therefore, the number of objects in use should equal the
5011 * number of ZTEST_DIROBJ entries, +1 for ZTEST_DIROBJ itself.
5012 * If not, we have an object leak.
5013 *
5014 * Note that we can only check this in ztest_dataset_open(),
5015 * when the open-context and syncing-context values agree.
5016 * That's because zap_count() returns the open-context value,
5017 * while dmu_objset_space() returns the rootbp fill count.
5018 */
5019 VERIFY3U(0, ==, zap_count(zd->zd_os, ZTEST_DIROBJ, &dirobjs));
5020 dmu_objset_space(zd->zd_os, &scratch, &scratch, &usedobjs, &scratch);
5021 ASSERT3U(dirobjs + 1, ==, usedobjs);
5022 }
5023
5024 static int
5025 ztest_dataset_open(ztest_shared_t *zs, int d)
5026 {
5027 ztest_ds_t *zd = &zs->zs_zd[d];
5028 uint64_t committed_seq = zd->zd_seq;
5029 objset_t *os;
5030 zilog_t *zilog;
5031 char name[MAXNAMELEN];
5032 int error;
5033
5034 ztest_dataset_name(name, zs->zs_pool, d);
5035
5036 (void) rw_rdlock(&zs->zs_name_lock);
5037
5038 error = ztest_dataset_create(name);
5039 if (error == ENOSPC) {
5040 (void) rw_unlock(&zs->zs_name_lock);
5041 ztest_record_enospc(FTAG);
5042 return (error);
5043 }
5044 ASSERT(error == 0 || error == EEXIST);
5045
5046 VERIFY3U(dmu_objset_hold(name, zd, &os), ==, 0);
5047 (void) rw_unlock(&zs->zs_name_lock);
5048
5049 ztest_zd_init(zd, os);
5050
5051 zilog = zd->zd_zilog;
5052
5053 if (zilog->zl_header->zh_claim_lr_seq != 0 &&
5054 zilog->zl_header->zh_claim_lr_seq < committed_seq)
5055 fatal(0, "missing log records: claimed %llu < committed %llu",
5056 zilog->zl_header->zh_claim_lr_seq, committed_seq);
5057
5058 ztest_dataset_dirobj_verify(zd);
5059
5060 zil_replay(os, zd, ztest_replay_vector);
5061
5062 ztest_dataset_dirobj_verify(zd);
5063
5064 if (zopt_verbose >= 6)
5065 (void) printf("%s replay %llu blocks, %llu records, seq %llu\n",
5066 zd->zd_name,
5067 (u_longlong_t)zilog->zl_parse_blk_count,
5068 (u_longlong_t)zilog->zl_parse_lr_count,
5069 (u_longlong_t)zilog->zl_replaying_seq);
5070
5071 zilog = zil_open(os, ztest_get_data);
5072
5073 if (zilog->zl_replaying_seq != 0 &&
5074 zilog->zl_replaying_seq < committed_seq)
5075 fatal(0, "missing log records: replayed %llu < committed %llu",
5076 zilog->zl_replaying_seq, committed_seq);
5077
5078 return (0);
5079 }
5080
5081 static void
5082 ztest_dataset_close(ztest_shared_t *zs, int d)
5083 {
5084 ztest_ds_t *zd = &zs->zs_zd[d];
5085
5086 zil_close(zd->zd_zilog);
5087 dmu_objset_rele(zd->zd_os, zd);
5088
5089 ztest_zd_fini(zd);
5090 }
5091
5092 /*
5093 * Kick off threads to run tests on all datasets in parallel.
5094 */
5095 static void
5096 ztest_run(ztest_shared_t *zs)
5097 {
5098 thread_t *tid;
5099 spa_t *spa;
5100 thread_t resume_tid;
5101 int error;
5102
5103 ztest_exiting = B_FALSE;
5104
5105 /*
5106 * Initialize parent/child shared state.
5107 */
5108 VERIFY(_mutex_init(&zs->zs_vdev_lock, USYNC_THREAD, NULL) == 0);
5109 VERIFY(rwlock_init(&zs->zs_name_lock, USYNC_THREAD, NULL) == 0);
5110
5111 zs->zs_thread_start = gethrtime();
5112 zs->zs_thread_stop = zs->zs_thread_start + zopt_passtime * NANOSEC;
5113 zs->zs_thread_stop = MIN(zs->zs_thread_stop, zs->zs_proc_stop);
5114 zs->zs_thread_kill = zs->zs_thread_stop;
5115 if (ztest_random(100) < zopt_killrate)
5116 zs->zs_thread_kill -= ztest_random(zopt_passtime * NANOSEC);
5117
5118 (void) _mutex_init(&zcl.zcl_callbacks_lock, USYNC_THREAD, NULL);
5119
5120 list_create(&zcl.zcl_callbacks, sizeof (ztest_cb_data_t),
5121 offsetof(ztest_cb_data_t, zcd_node));
5122
5123 /*
5124 * Open our pool.
5125 */
5126 kernel_init(FREAD | FWRITE);
5127 VERIFY(spa_open(zs->zs_pool, &spa, FTAG) == 0);
5128 zs->zs_spa = spa;
5129
5130 spa->spa_dedup_ditto = 2 * ZIO_DEDUPDITTO_MIN;
5131
5132 /*
5133 * We don't expect the pool to suspend unless maxfaults == 0,
5134 * in which case ztest_fault_inject() temporarily takes away
5135 * the only valid replica.
5136 */
5137 if (MAXFAULTS() == 0)
5138 spa->spa_failmode = ZIO_FAILURE_MODE_WAIT;
5139 else
5140 spa->spa_failmode = ZIO_FAILURE_MODE_PANIC;
5141
5142 /*
5143 * Create a thread to periodically resume suspended I/O.
5144 */
5145 VERIFY(thr_create(0, 0, ztest_resume_thread, spa, THR_BOUND,
5146 &resume_tid) == 0);
5147
5148 /*
5149 * Create a deadman thread to abort() if we hang.
5150 */
5151 VERIFY(thr_create(0, 0, ztest_deadman_thread, zs, THR_BOUND,
5152 NULL) == 0);
5153
5154 /*
5155 * Verify that we can safely inquire about about any object,
5156 * whether it's allocated or not. To make it interesting,
5157 * we probe a 5-wide window around each power of two.
5158 * This hits all edge cases, including zero and the max.
5159 */
5160 for (int t = 0; t < 64; t++) {
5161 for (int d = -5; d <= 5; d++) {
5162 error = dmu_object_info(spa->spa_meta_objset,
5163 (1ULL << t) + d, NULL);
5164 ASSERT(error == 0 || error == ENOENT ||
5165 error == EINVAL);
5166 }
5167 }
5168
5169 /*
5170 * If we got any ENOSPC errors on the previous run, destroy something.
5171 */
5172 if (zs->zs_enospc_count != 0) {
5173 int d = ztest_random(zopt_datasets);
5174 ztest_dataset_destroy(zs, d);
5175 }
5176 zs->zs_enospc_count = 0;
5177
5178 tid = umem_zalloc(zopt_threads * sizeof (thread_t), UMEM_NOFAIL);
5179
5180 if (zopt_verbose >= 4)
5181 (void) printf("starting main threads...\n");
5182
5183 /*
5184 * Kick off all the tests that run in parallel.
5185 */
5186 for (int t = 0; t < zopt_threads; t++) {
5187 if (t < zopt_datasets && ztest_dataset_open(zs, t) != 0)
5188 return;
5189 VERIFY(thr_create(0, 0, ztest_thread, (void *)(uintptr_t)t,
5190 THR_BOUND, &tid[t]) == 0);
5191 }
5192
5193 /*
5194 * Wait for all of the tests to complete. We go in reverse order
5195 * so we don't close datasets while threads are still using them.
5196 */
5197 for (int t = zopt_threads - 1; t >= 0; t--) {
5198 VERIFY(thr_join(tid[t], NULL, NULL) == 0);
5199 if (t < zopt_datasets)
5200 ztest_dataset_close(zs, t);
5201 }
5202
5203 txg_wait_synced(spa_get_dsl(spa), 0);
5204
5205 zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(spa));
5206 zs->zs_space = metaslab_class_get_space(spa_normal_class(spa));
5207
5208 umem_free(tid, zopt_threads * sizeof (thread_t));
5209
5210 /* Kill the resume thread */
5211 ztest_exiting = B_TRUE;
5212 VERIFY(thr_join(resume_tid, NULL, NULL) == 0);
5213 ztest_resume(spa);
5214
5215 /*
5216 * Right before closing the pool, kick off a bunch of async I/O;
5217 * spa_close() should wait for it to complete.
5218 */
5219 for (uint64_t object = 1; object < 50; object++)
5220 dmu_prefetch(spa->spa_meta_objset, object, 0, 1ULL << 20);
5221
5222 spa_close(spa, FTAG);
5223
5224 /*
5225 * Verify that we can loop over all pools.
5226 */
5227 mutex_enter(&spa_namespace_lock);
5228 for (spa = spa_next(NULL); spa != NULL; spa = spa_next(spa))
5229 if (zopt_verbose > 3)
5230 (void) printf("spa_next: found %s\n", spa_name(spa));
5231 mutex_exit(&spa_namespace_lock);
5232
5233 /*
5234 * Verify that we can export the pool and reimport it under a
5235 * different name.
5236 */
5237 if (ztest_random(2) == 0) {
5238 char name[MAXNAMELEN];
5239 (void) snprintf(name, MAXNAMELEN, "%s_import", zs->zs_pool);
5240 ztest_spa_import_export(zs->zs_pool, name);
5241 ztest_spa_import_export(name, zs->zs_pool);
5242 }
5243
5244 kernel_fini();
5245 }
5246
5247 static void
5248 ztest_freeze(ztest_shared_t *zs)
5249 {
5250 ztest_ds_t *zd = &zs->zs_zd[0];
5251 spa_t *spa;
5252 int numloops = 0;
5253
5254 if (zopt_verbose >= 3)
5255 (void) printf("testing spa_freeze()...\n");
5256
5257 kernel_init(FREAD | FWRITE);
5258 VERIFY3U(0, ==, spa_open(zs->zs_pool, &spa, FTAG));
5259 VERIFY3U(0, ==, ztest_dataset_open(zs, 0));
5260
5261 /*
5262 * Force the first log block to be transactionally allocated.
5263 * We have to do this before we freeze the pool -- otherwise
5264 * the log chain won't be anchored.
5265 */
5266 while (BP_IS_HOLE(&zd->zd_zilog->zl_header->zh_log)) {
5267 ztest_dmu_object_alloc_free(zd, 0);
5268 zil_commit(zd->zd_zilog, UINT64_MAX, 0);
5269 }
5270
5271 txg_wait_synced(spa_get_dsl(spa), 0);
5272
5273 /*
5274 * Freeze the pool. This stops spa_sync() from doing anything,
5275 * so that the only way to record changes from now on is the ZIL.
5276 */
5277 spa_freeze(spa);
5278
5279 /*
5280 * Run tests that generate log records but don't alter the pool config
5281 * or depend on DSL sync tasks (snapshots, objset create/destroy, etc).
5282 * We do a txg_wait_synced() after each iteration to force the txg
5283 * to increase well beyond the last synced value in the uberblock.
5284 * The ZIL should be OK with that.
5285 */
5286 while (ztest_random(10) != 0 && numloops++ < zopt_maxloops) {
5287 ztest_dmu_write_parallel(zd, 0);
5288 ztest_dmu_object_alloc_free(zd, 0);
5289 txg_wait_synced(spa_get_dsl(spa), 0);
5290 }
5291
5292 /*
5293 * Commit all of the changes we just generated.
5294 */
5295 zil_commit(zd->zd_zilog, UINT64_MAX, 0);
5296 txg_wait_synced(spa_get_dsl(spa), 0);
5297
5298 /*
5299 * Close our dataset and close the pool.
5300 */
5301 ztest_dataset_close(zs, 0);
5302 spa_close(spa, FTAG);
5303 kernel_fini();
5304
5305 /*
5306 * Open and close the pool and dataset to induce log replay.
5307 */
5308 kernel_init(FREAD | FWRITE);
5309 VERIFY3U(0, ==, spa_open(zs->zs_pool, &spa, FTAG));
5310 VERIFY3U(0, ==, ztest_dataset_open(zs, 0));
5311 ztest_dataset_close(zs, 0);
5312 spa_close(spa, FTAG);
5313 kernel_fini();
5314
5315 list_destroy(&zcl.zcl_callbacks);
5316
5317 (void) _mutex_destroy(&zcl.zcl_callbacks_lock);
5318
5319 (void) rwlock_destroy(&zs->zs_name_lock);
5320 (void) _mutex_destroy(&zs->zs_vdev_lock);
5321 }
5322
5323 void
5324 print_time(hrtime_t t, char *timebuf)
5325 {
5326 hrtime_t s = t / NANOSEC;
5327 hrtime_t m = s / 60;
5328 hrtime_t h = m / 60;
5329 hrtime_t d = h / 24;
5330
5331 s -= m * 60;
5332 m -= h * 60;
5333 h -= d * 24;
5334
5335 timebuf[0] = '\0';
5336
5337 if (d)
5338 (void) sprintf(timebuf,
5339 "%llud%02lluh%02llum%02llus", d, h, m, s);
5340 else if (h)
5341 (void) sprintf(timebuf, "%lluh%02llum%02llus", h, m, s);
5342 else if (m)
5343 (void) sprintf(timebuf, "%llum%02llus", m, s);
5344 else
5345 (void) sprintf(timebuf, "%llus", s);
5346 }
5347
5348 static nvlist_t *
5349 make_random_props()
5350 {
5351 nvlist_t *props;
5352
5353 if (ztest_random(2) == 0)
5354 return (NULL);
5355
5356 VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
5357 VERIFY(nvlist_add_uint64(props, "autoreplace", 1) == 0);
5358
5359 (void) printf("props:\n");
5360 dump_nvlist(props, 4);
5361
5362 return (props);
5363 }
5364
5365 /*
5366 * Create a storage pool with the given name and initial vdev size.
5367 * Then test spa_freeze() functionality.
5368 */
5369 static void
5370 ztest_init(ztest_shared_t *zs)
5371 {
5372 spa_t *spa;
5373 nvlist_t *nvroot, *props;
5374
5375 VERIFY(_mutex_init(&zs->zs_vdev_lock, USYNC_THREAD, NULL) == 0);
5376 VERIFY(rwlock_init(&zs->zs_name_lock, USYNC_THREAD, NULL) == 0);
5377
5378 kernel_init(FREAD | FWRITE);
5379
5380 /*
5381 * Create the storage pool.
5382 */
5383 (void) spa_destroy(zs->zs_pool);
5384 ztest_shared->zs_vdev_next_leaf = 0;
5385 zs->zs_splits = 0;
5386 zs->zs_mirrors = zopt_mirrors;
5387 nvroot = make_vdev_root(NULL, NULL, zopt_vdev_size, 0,
5388 0, zopt_raidz, zs->zs_mirrors, 1);
5389 props = make_random_props();
5390 VERIFY3U(0, ==, spa_create(zs->zs_pool, nvroot, props, NULL, NULL));
5391 nvlist_free(nvroot);
5392
5393 VERIFY3U(0, ==, spa_open(zs->zs_pool, &spa, FTAG));
5394 metaslab_sz = 1ULL << spa->spa_root_vdev->vdev_child[0]->vdev_ms_shift;
5395 spa_close(spa, FTAG);
5396
5397 kernel_fini();
5398
5399 ztest_run_zdb(zs->zs_pool);
5400
5401 ztest_freeze(zs);
5402
5403 ztest_run_zdb(zs->zs_pool);
5404 }
5405
5406 int
5407 main(int argc, char **argv)
5408 {
5409 int kills = 0;
5410 int iters = 0;
5411 ztest_shared_t *zs;
5412 size_t shared_size;
5413 ztest_info_t *zi;
5414 char timebuf[100];
5415 char numbuf[6];
5416 spa_t *spa;
5417
5418 (void) setvbuf(stdout, NULL, _IOLBF, 0);
5419
5420 ztest_random_fd = open("/dev/urandom", O_RDONLY);
5421
5422 process_options(argc, argv);
5423
5424 /* Override location of zpool.cache */
5425 (void) asprintf((char **)&spa_config_path, "%s/zpool.cache", zopt_dir);
5426
5427 /*
5428 * Blow away any existing copy of zpool.cache
5429 */
5430 if (zopt_init != 0)
5431 (void) remove(spa_config_path);
5432
5433 shared_size = sizeof (*zs) + zopt_datasets * sizeof (ztest_ds_t);
5434
5435 zs = ztest_shared = (void *)mmap(0,
5436 P2ROUNDUP(shared_size, getpagesize()),
5437 PROT_READ | PROT_WRITE, MAP_SHARED | MAP_ANON, -1, 0);
5438
5439 if (zopt_verbose >= 1) {
5440 (void) printf("%llu vdevs, %d datasets, %d threads,"
5441 " %llu seconds...\n",
5442 (u_longlong_t)zopt_vdevs, zopt_datasets, zopt_threads,
5443 (u_longlong_t)zopt_time);
5444 }
5445
5446 /*
5447 * Create and initialize our storage pool.
5448 */
5449 for (int i = 1; i <= zopt_init; i++) {
5450 bzero(zs, sizeof (ztest_shared_t));
5451 if (zopt_verbose >= 3 && zopt_init != 1)
5452 (void) printf("ztest_init(), pass %d\n", i);
5453 zs->zs_pool = zopt_pool;
5454 ztest_init(zs);
5455 }
5456
5457 zs->zs_pool = zopt_pool;
5458 zs->zs_proc_start = gethrtime();
5459 zs->zs_proc_stop = zs->zs_proc_start + zopt_time * NANOSEC;
5460
5461 for (int f = 0; f < ZTEST_FUNCS; f++) {
5462 zi = &zs->zs_info[f];
5463 *zi = ztest_info[f];
5464 if (zs->zs_proc_start + zi->zi_interval[0] > zs->zs_proc_stop)
5465 zi->zi_call_next = UINT64_MAX;
5466 else
5467 zi->zi_call_next = zs->zs_proc_start +
5468 ztest_random(2 * zi->zi_interval[0] + 1);
5469 }
5470
5471 /*
5472 * Run the tests in a loop. These tests include fault injection
5473 * to verify that self-healing data works, and forced crashes
5474 * to verify that we never lose on-disk consistency.
5475 */
5476 while (gethrtime() < zs->zs_proc_stop) {
5477 int status;
5478 pid_t pid;
5479
5480 /*
5481 * Initialize the workload counters for each function.
5482 */
5483 for (int f = 0; f < ZTEST_FUNCS; f++) {
5484 zi = &zs->zs_info[f];
5485 zi->zi_call_count = 0;
5486 zi->zi_call_time = 0;
5487 }
5488
5489 /* Set the allocation switch size */
5490 metaslab_df_alloc_threshold = ztest_random(metaslab_sz / 4) + 1;
5491
5492 pid = fork();
5493
5494 if (pid == -1)
5495 fatal(1, "fork failed");
5496
5497 if (pid == 0) { /* child */
5498 struct rlimit rl = { 1024, 1024 };
5499 (void) setrlimit(RLIMIT_NOFILE, &rl);
5500 (void) enable_extended_FILE_stdio(-1, -1);
5501 ztest_run(zs);
5502 exit(0);
5503 }
5504
5505 while (waitpid(pid, &status, 0) != pid)
5506 continue;
5507
5508 if (WIFEXITED(status)) {
5509 if (WEXITSTATUS(status) != 0) {
5510 (void) fprintf(stderr,
5511 "child exited with code %d\n",
5512 WEXITSTATUS(status));
5513 exit(2);
5514 }
5515 } else if (WIFSIGNALED(status)) {
5516 if (WTERMSIG(status) != SIGKILL) {
5517 (void) fprintf(stderr,
5518 "child died with signal %d\n",
5519 WTERMSIG(status));
5520 exit(3);
5521 }
5522 kills++;
5523 } else {
5524 (void) fprintf(stderr, "something strange happened "
5525 "to child\n");
5526 exit(4);
5527 }
5528
5529 iters++;
5530
5531 if (zopt_verbose >= 1) {
5532 hrtime_t now = gethrtime();
5533
5534 now = MIN(now, zs->zs_proc_stop);
5535 print_time(zs->zs_proc_stop - now, timebuf);
5536 nicenum(zs->zs_space, numbuf);
5537
5538 (void) printf("Pass %3d, %8s, %3llu ENOSPC, "
5539 "%4.1f%% of %5s used, %3.0f%% done, %8s to go\n",
5540 iters,
5541 WIFEXITED(status) ? "Complete" : "SIGKILL",
5542 (u_longlong_t)zs->zs_enospc_count,
5543 100.0 * zs->zs_alloc / zs->zs_space,
5544 numbuf,
5545 100.0 * (now - zs->zs_proc_start) /
5546 (zopt_time * NANOSEC), timebuf);
5547 }
5548
5549 if (zopt_verbose >= 2) {
5550 (void) printf("\nWorkload summary:\n\n");
5551 (void) printf("%7s %9s %s\n",
5552 "Calls", "Time", "Function");
5553 (void) printf("%7s %9s %s\n",
5554 "-----", "----", "--------");
5555 for (int f = 0; f < ZTEST_FUNCS; f++) {
5556 Dl_info dli;
5557
5558 zi = &zs->zs_info[f];
5559 print_time(zi->zi_call_time, timebuf);
5560 (void) dladdr((void *)zi->zi_func, &dli);
5561 (void) printf("%7llu %9s %s\n",
5562 (u_longlong_t)zi->zi_call_count, timebuf,
5563 dli.dli_sname);
5564 }
5565 (void) printf("\n");
5566 }
5567
5568 /*
5569 * It's possible that we killed a child during a rename test,
5570 * in which case we'll have a 'ztest_tmp' pool lying around
5571 * instead of 'ztest'. Do a blind rename in case this happened.
5572 */
5573 kernel_init(FREAD);
5574 if (spa_open(zopt_pool, &spa, FTAG) == 0) {
5575 spa_close(spa, FTAG);
5576 } else {
5577 char tmpname[MAXNAMELEN];
5578 kernel_fini();
5579 kernel_init(FREAD | FWRITE);
5580 (void) snprintf(tmpname, sizeof (tmpname), "%s_tmp",
5581 zopt_pool);
5582 (void) spa_rename(tmpname, zopt_pool);
5583 }
5584 kernel_fini();
5585
5586 ztest_run_zdb(zopt_pool);
5587 }
5588
5589 if (zopt_verbose >= 1) {
5590 (void) printf("%d killed, %d completed, %.0f%% kill rate\n",
5591 kills, iters - kills, (100.0 * kills) / MAX(1, iters));
5592 }
5593
5594 return (0);
5595 }
ZFS On Linux mirror for PVE