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 * Copyright (c) 2012 by Delphix. All rights reserved.
24 * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
25 */
26
27 /*
28 * The objective of this program is to provide a DMU/ZAP/SPA stress test
29 * that runs entirely in userland, is easy to use, and easy to extend.
30 *
31 * The overall design of the ztest program is as follows:
32 *
33 * (1) For each major functional area (e.g. adding vdevs to a pool,
34 * creating and destroying datasets, reading and writing objects, etc)
35 * we have a simple routine to test that functionality. These
36 * individual routines do not have to do anything "stressful".
37 *
38 * (2) We turn these simple functionality tests into a stress test by
39 * running them all in parallel, with as many threads as desired,
40 * and spread across as many datasets, objects, and vdevs as desired.
41 *
42 * (3) While all this is happening, we inject faults into the pool to
43 * verify that self-healing data really works.
44 *
45 * (4) Every time we open a dataset, we change its checksum and compression
46 * functions. Thus even individual objects vary from block to block
47 * in which checksum they use and whether they're compressed.
48 *
49 * (5) To verify that we never lose on-disk consistency after a crash,
50 * we run the entire test in a child of the main process.
51 * At random times, the child self-immolates with a SIGKILL.
52 * This is the software equivalent of pulling the power cord.
53 * The parent then runs the test again, using the existing
54 * storage pool, as many times as desired. If backwards compatability
55 * testing is enabled ztest will sometimes run the "older" version
56 * of ztest after a SIGKILL.
57 *
58 * (6) To verify that we don't have future leaks or temporal incursions,
59 * many of the functional tests record the transaction group number
60 * as part of their data. When reading old data, they verify that
61 * the transaction group number is less than the current, open txg.
62 * If you add a new test, please do this if applicable.
63 *
64 * When run with no arguments, ztest runs for about five minutes and
65 * produces no output if successful. To get a little bit of information,
66 * specify -V. To get more information, specify -VV, and so on.
67 *
68 * To turn this into an overnight stress test, use -T to specify run time.
69 *
70 * You can ask more more vdevs [-v], datasets [-d], or threads [-t]
71 * to increase the pool capacity, fanout, and overall stress level.
72 *
73 * Use the -k option to set the desired frequency of kills.
74 *
75 * When ztest invokes itself it passes all relevant information through a
76 * temporary file which is mmap-ed in the child process. This allows shared
77 * memory to survive the exec syscall. The ztest_shared_hdr_t struct is always
78 * stored at offset 0 of this file and contains information on the size and
79 * number of shared structures in the file. The information stored in this file
80 * must remain backwards compatible with older versions of ztest so that
81 * ztest can invoke them during backwards compatibility testing (-B).
82 */
83
84 #include <sys/zfs_context.h>
85 #include <sys/spa.h>
86 #include <sys/dmu.h>
87 #include <sys/txg.h>
88 #include <sys/dbuf.h>
89 #include <sys/zap.h>
90 #include <sys/dmu_objset.h>
91 #include <sys/poll.h>
92 #include <sys/stat.h>
93 #include <sys/time.h>
94 #include <sys/wait.h>
95 #include <sys/mman.h>
96 #include <sys/resource.h>
97 #include <sys/zio.h>
98 #include <sys/zil.h>
99 #include <sys/zil_impl.h>
100 #include <sys/vdev_impl.h>
101 #include <sys/vdev_file.h>
102 #include <sys/spa_impl.h>
103 #include <sys/metaslab_impl.h>
104 #include <sys/dsl_prop.h>
105 #include <sys/dsl_dataset.h>
106 #include <sys/dsl_scan.h>
107 #include <sys/zio_checksum.h>
108 #include <sys/refcount.h>
109 #include <sys/zfeature.h>
110 #include <stdio.h>
111 #include <stdio_ext.h>
112 #include <stdlib.h>
113 #include <unistd.h>
114 #include <signal.h>
115 #include <umem.h>
116 #include <dlfcn.h>
117 #include <ctype.h>
118 #include <math.h>
119 #include <sys/fs/zfs.h>
120 #include <libnvpair.h>
121
122 #define ZTEST_FD_DATA 3
123 #define ZTEST_FD_RAND 4
124
125 typedef struct ztest_shared_hdr {
126 uint64_t zh_hdr_size;
127 uint64_t zh_opts_size;
128 uint64_t zh_size;
129 uint64_t zh_stats_size;
130 uint64_t zh_stats_count;
131 uint64_t zh_ds_size;
132 uint64_t zh_ds_count;
133 } ztest_shared_hdr_t;
134
135 static ztest_shared_hdr_t *ztest_shared_hdr;
136
137 typedef struct ztest_shared_opts {
138 char zo_pool[MAXNAMELEN];
139 char zo_dir[MAXNAMELEN];
140 char zo_alt_ztest[MAXNAMELEN];
141 char zo_alt_libpath[MAXNAMELEN];
142 uint64_t zo_vdevs;
143 uint64_t zo_vdevtime;
144 size_t zo_vdev_size;
145 int zo_ashift;
146 int zo_mirrors;
147 int zo_raidz;
148 int zo_raidz_parity;
149 int zo_datasets;
150 int zo_threads;
151 uint64_t zo_passtime;
152 uint64_t zo_killrate;
153 int zo_verbose;
154 int zo_init;
155 uint64_t zo_time;
156 uint64_t zo_maxloops;
157 uint64_t zo_metaslab_gang_bang;
158 } ztest_shared_opts_t;
159
160 static const ztest_shared_opts_t ztest_opts_defaults = {
161 .zo_pool = { 'z', 't', 'e', 's', 't', '\0' },
162 .zo_dir = { '/', 't', 'm', 'p', '\0' },
163 .zo_alt_ztest = { '\0' },
164 .zo_alt_libpath = { '\0' },
165 .zo_vdevs = 5,
166 .zo_ashift = SPA_MINBLOCKSHIFT,
167 .zo_mirrors = 2,
168 .zo_raidz = 4,
169 .zo_raidz_parity = 1,
170 .zo_vdev_size = SPA_MINDEVSIZE,
171 .zo_datasets = 7,
172 .zo_threads = 23,
173 .zo_passtime = 60, /* 60 seconds */
174 .zo_killrate = 70, /* 70% kill rate */
175 .zo_verbose = 0,
176 .zo_init = 1,
177 .zo_time = 300, /* 5 minutes */
178 .zo_maxloops = 50, /* max loops during spa_freeze() */
179 .zo_metaslab_gang_bang = 32 << 10
180 };
181
182 extern uint64_t metaslab_gang_bang;
183 extern uint64_t metaslab_df_alloc_threshold;
184
185 static ztest_shared_opts_t *ztest_shared_opts;
186 static ztest_shared_opts_t ztest_opts;
187
188 typedef struct ztest_shared_ds {
189 uint64_t zd_seq;
190 } ztest_shared_ds_t;
191
192 static ztest_shared_ds_t *ztest_shared_ds;
193 #define ZTEST_GET_SHARED_DS(d) (&ztest_shared_ds[d])
194
195 #define BT_MAGIC 0x123456789abcdefULL
196 #define MAXFAULTS() \
197 (MAX(zs->zs_mirrors, 1) * (ztest_opts.zo_raidz_parity + 1) - 1)
198
199 enum ztest_io_type {
200 ZTEST_IO_WRITE_TAG,
201 ZTEST_IO_WRITE_PATTERN,
202 ZTEST_IO_WRITE_ZEROES,
203 ZTEST_IO_TRUNCATE,
204 ZTEST_IO_SETATTR,
205 ZTEST_IO_TYPES
206 };
207
208 typedef struct ztest_block_tag {
209 uint64_t bt_magic;
210 uint64_t bt_objset;
211 uint64_t bt_object;
212 uint64_t bt_offset;
213 uint64_t bt_gen;
214 uint64_t bt_txg;
215 uint64_t bt_crtxg;
216 } ztest_block_tag_t;
217
218 typedef struct bufwad {
219 uint64_t bw_index;
220 uint64_t bw_txg;
221 uint64_t bw_data;
222 } bufwad_t;
223
224 /*
225 * XXX -- fix zfs range locks to be generic so we can use them here.
226 */
227 typedef enum {
228 RL_READER,
229 RL_WRITER,
230 RL_APPEND
231 } rl_type_t;
232
233 typedef struct rll {
234 void *rll_writer;
235 int rll_readers;
236 mutex_t rll_lock;
237 cond_t rll_cv;
238 } rll_t;
239
240 typedef struct rl {
241 uint64_t rl_object;
242 uint64_t rl_offset;
243 uint64_t rl_size;
244 rll_t *rl_lock;
245 } rl_t;
246
247 #define ZTEST_RANGE_LOCKS 64
248 #define ZTEST_OBJECT_LOCKS 64
249
250 /*
251 * Object descriptor. Used as a template for object lookup/create/remove.
252 */
253 typedef struct ztest_od {
254 uint64_t od_dir;
255 uint64_t od_object;
256 dmu_object_type_t od_type;
257 dmu_object_type_t od_crtype;
258 uint64_t od_blocksize;
259 uint64_t od_crblocksize;
260 uint64_t od_gen;
261 uint64_t od_crgen;
262 char od_name[MAXNAMELEN];
263 } ztest_od_t;
264
265 /*
266 * Per-dataset state.
267 */
268 typedef struct ztest_ds {
269 ztest_shared_ds_t *zd_shared;
270 objset_t *zd_os;
271 rwlock_t zd_zilog_lock;
272 zilog_t *zd_zilog;
273 ztest_od_t *zd_od; /* debugging aid */
274 char zd_name[MAXNAMELEN];
275 mutex_t zd_dirobj_lock;
276 rll_t zd_object_lock[ZTEST_OBJECT_LOCKS];
277 rll_t zd_range_lock[ZTEST_RANGE_LOCKS];
278 } ztest_ds_t;
279
280 /*
281 * Per-iteration state.
282 */
283 typedef void ztest_func_t(ztest_ds_t *zd, uint64_t id);
284
285 typedef struct ztest_info {
286 ztest_func_t *zi_func; /* test function */
287 uint64_t zi_iters; /* iterations per execution */
288 uint64_t *zi_interval; /* execute every <interval> seconds */
289 } ztest_info_t;
290
291 typedef struct ztest_shared_callstate {
292 uint64_t zc_count; /* per-pass count */
293 uint64_t zc_time; /* per-pass time */
294 uint64_t zc_next; /* next time to call this function */
295 } ztest_shared_callstate_t;
296
297 static ztest_shared_callstate_t *ztest_shared_callstate;
298 #define ZTEST_GET_SHARED_CALLSTATE(c) (&ztest_shared_callstate[c])
299
300 /*
301 * Note: these aren't static because we want dladdr() to work.
302 */
303 ztest_func_t ztest_dmu_read_write;
304 ztest_func_t ztest_dmu_write_parallel;
305 ztest_func_t ztest_dmu_object_alloc_free;
306 ztest_func_t ztest_dmu_commit_callbacks;
307 ztest_func_t ztest_zap;
308 ztest_func_t ztest_zap_parallel;
309 ztest_func_t ztest_zil_commit;
310 ztest_func_t ztest_zil_remount;
311 ztest_func_t ztest_dmu_read_write_zcopy;
312 ztest_func_t ztest_dmu_objset_create_destroy;
313 ztest_func_t ztest_dmu_prealloc;
314 ztest_func_t ztest_fzap;
315 ztest_func_t ztest_dmu_snapshot_create_destroy;
316 ztest_func_t ztest_dsl_prop_get_set;
317 ztest_func_t ztest_spa_prop_get_set;
318 ztest_func_t ztest_spa_create_destroy;
319 ztest_func_t ztest_fault_inject;
320 ztest_func_t ztest_ddt_repair;
321 ztest_func_t ztest_dmu_snapshot_hold;
322 ztest_func_t ztest_spa_rename;
323 ztest_func_t ztest_scrub;
324 ztest_func_t ztest_dsl_dataset_promote_busy;
325 ztest_func_t ztest_vdev_attach_detach;
326 ztest_func_t ztest_vdev_LUN_growth;
327 ztest_func_t ztest_vdev_add_remove;
328 ztest_func_t ztest_vdev_aux_add_remove;
329 ztest_func_t ztest_split_pool;
330 ztest_func_t ztest_reguid;
331
332 uint64_t zopt_always = 0ULL * NANOSEC; /* all the time */
333 uint64_t zopt_incessant = 1ULL * NANOSEC / 10; /* every 1/10 second */
334 uint64_t zopt_often = 1ULL * NANOSEC; /* every second */
335 uint64_t zopt_sometimes = 10ULL * NANOSEC; /* every 10 seconds */
336 uint64_t zopt_rarely = 60ULL * NANOSEC; /* every 60 seconds */
337
338 ztest_info_t ztest_info[] = {
339 { ztest_dmu_read_write, 1, &zopt_always },
340 { ztest_dmu_write_parallel, 10, &zopt_always },
341 { ztest_dmu_object_alloc_free, 1, &zopt_always },
342 { ztest_dmu_commit_callbacks, 1, &zopt_always },
343 { ztest_zap, 30, &zopt_always },
344 { ztest_zap_parallel, 100, &zopt_always },
345 { ztest_split_pool, 1, &zopt_always },
346 { ztest_zil_commit, 1, &zopt_incessant },
347 { ztest_zil_remount, 1, &zopt_sometimes },
348 { ztest_dmu_read_write_zcopy, 1, &zopt_often },
349 { ztest_dmu_objset_create_destroy, 1, &zopt_often },
350 { ztest_dsl_prop_get_set, 1, &zopt_often },
351 { ztest_spa_prop_get_set, 1, &zopt_sometimes },
352 #if 0
353 { ztest_dmu_prealloc, 1, &zopt_sometimes },
354 #endif
355 { ztest_fzap, 1, &zopt_sometimes },
356 { ztest_dmu_snapshot_create_destroy, 1, &zopt_sometimes },
357 { ztest_spa_create_destroy, 1, &zopt_sometimes },
358 { ztest_fault_inject, 1, &zopt_sometimes },
359 { ztest_ddt_repair, 1, &zopt_sometimes },
360 { ztest_dmu_snapshot_hold, 1, &zopt_sometimes },
361 { ztest_reguid, 1, &zopt_sometimes },
362 { ztest_spa_rename, 1, &zopt_rarely },
363 { ztest_scrub, 1, &zopt_rarely },
364 { ztest_dsl_dataset_promote_busy, 1, &zopt_rarely },
365 { ztest_vdev_attach_detach, 1, &zopt_rarely },
366 { ztest_vdev_LUN_growth, 1, &zopt_rarely },
367 { ztest_vdev_add_remove, 1,
368 &ztest_opts.zo_vdevtime },
369 { ztest_vdev_aux_add_remove, 1,
370 &ztest_opts.zo_vdevtime },
371 };
372
373 #define ZTEST_FUNCS (sizeof (ztest_info) / sizeof (ztest_info_t))
374
375 /*
376 * The following struct is used to hold a list of uncalled commit callbacks.
377 * The callbacks are ordered by txg number.
378 */
379 typedef struct ztest_cb_list {
380 mutex_t zcl_callbacks_lock;
381 list_t zcl_callbacks;
382 } ztest_cb_list_t;
383
384 /*
385 * Stuff we need to share writably between parent and child.
386 */
387 typedef struct ztest_shared {
388 boolean_t zs_do_init;
389 hrtime_t zs_proc_start;
390 hrtime_t zs_proc_stop;
391 hrtime_t zs_thread_start;
392 hrtime_t zs_thread_stop;
393 hrtime_t zs_thread_kill;
394 uint64_t zs_enospc_count;
395 uint64_t zs_vdev_next_leaf;
396 uint64_t zs_vdev_aux;
397 uint64_t zs_alloc;
398 uint64_t zs_space;
399 uint64_t zs_splits;
400 uint64_t zs_mirrors;
401 uint64_t zs_metaslab_sz;
402 uint64_t zs_metaslab_df_alloc_threshold;
403 uint64_t zs_guid;
404 } ztest_shared_t;
405
406 #define ID_PARALLEL -1ULL
407
408 static char ztest_dev_template[] = "%s/%s.%llua";
409 static char ztest_aux_template[] = "%s/%s.%s.%llu";
410 ztest_shared_t *ztest_shared;
411
412 static spa_t *ztest_spa = NULL;
413 static ztest_ds_t *ztest_ds;
414
415 static mutex_t ztest_vdev_lock;
416 static rwlock_t ztest_name_lock;
417
418 static boolean_t ztest_dump_core = B_TRUE;
419 static boolean_t ztest_exiting;
420
421 /* Global commit callback list */
422 static ztest_cb_list_t zcl;
423
424 enum ztest_object {
425 ZTEST_META_DNODE = 0,
426 ZTEST_DIROBJ,
427 ZTEST_OBJECTS
428 };
429
430 static void usage(boolean_t) __NORETURN;
431
432 /*
433 * These libumem hooks provide a reasonable set of defaults for the allocator's
434 * debugging facilities.
435 */
436 const char *
437 _umem_debug_init()
438 {
439 return ("default,verbose"); /* $UMEM_DEBUG setting */
440 }
441
442 const char *
443 _umem_logging_init(void)
444 {
445 return ("fail,contents"); /* $UMEM_LOGGING setting */
446 }
447
448 #define FATAL_MSG_SZ 1024
449
450 char *fatal_msg;
451
452 static void
453 fatal(int do_perror, char *message, ...)
454 {
455 va_list args;
456 int save_errno = errno;
457 char buf[FATAL_MSG_SZ];
458
459 (void) fflush(stdout);
460
461 va_start(args, message);
462 (void) sprintf(buf, "ztest: ");
463 /* LINTED */
464 (void) vsprintf(buf + strlen(buf), message, args);
465 va_end(args);
466 if (do_perror) {
467 (void) snprintf(buf + strlen(buf), FATAL_MSG_SZ - strlen(buf),
468 ": %s", strerror(save_errno));
469 }
470 (void) fprintf(stderr, "%s\n", buf);
471 fatal_msg = buf; /* to ease debugging */
472 if (ztest_dump_core)
473 abort();
474 exit(3);
475 }
476
477 static int
478 str2shift(const char *buf)
479 {
480 const char *ends = "BKMGTPEZ";
481 int i;
482
483 if (buf[0] == '\0')
484 return (0);
485 for (i = 0; i < strlen(ends); i++) {
486 if (toupper(buf[0]) == ends[i])
487 break;
488 }
489 if (i == strlen(ends)) {
490 (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n",
491 buf);
492 usage(B_FALSE);
493 }
494 if (buf[1] == '\0' || (toupper(buf[1]) == 'B' && buf[2] == '\0')) {
495 return (10*i);
496 }
497 (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n", buf);
498 usage(B_FALSE);
499 /* NOTREACHED */
500 }
501
502 static uint64_t
503 nicenumtoull(const char *buf)
504 {
505 char *end;
506 uint64_t val;
507
508 val = strtoull(buf, &end, 0);
509 if (end == buf) {
510 (void) fprintf(stderr, "ztest: bad numeric value: %s\n", buf);
511 usage(B_FALSE);
512 } else if (end[0] == '.') {
513 double fval = strtod(buf, &end);
514 fval *= pow(2, str2shift(end));
515 if (fval > UINT64_MAX) {
516 (void) fprintf(stderr, "ztest: value too large: %s\n",
517 buf);
518 usage(B_FALSE);
519 }
520 val = (uint64_t)fval;
521 } else {
522 int shift = str2shift(end);
523 if (shift >= 64 || (val << shift) >> shift != val) {
524 (void) fprintf(stderr, "ztest: value too large: %s\n",
525 buf);
526 usage(B_FALSE);
527 }
528 val <<= shift;
529 }
530 return (val);
531 }
532
533 static void
534 usage(boolean_t requested)
535 {
536 const ztest_shared_opts_t *zo = &ztest_opts_defaults;
537
538 char nice_vdev_size[10];
539 char nice_gang_bang[10];
540 FILE *fp = requested ? stdout : stderr;
541
542 nicenum(zo->zo_vdev_size, nice_vdev_size);
543 nicenum(zo->zo_metaslab_gang_bang, nice_gang_bang);
544
545 (void) fprintf(fp, "Usage: %s\n"
546 "\t[-v vdevs (default: %llu)]\n"
547 "\t[-s size_of_each_vdev (default: %s)]\n"
548 "\t[-a alignment_shift (default: %d)] use 0 for random\n"
549 "\t[-m mirror_copies (default: %d)]\n"
550 "\t[-r raidz_disks (default: %d)]\n"
551 "\t[-R raidz_parity (default: %d)]\n"
552 "\t[-d datasets (default: %d)]\n"
553 "\t[-t threads (default: %d)]\n"
554 "\t[-g gang_block_threshold (default: %s)]\n"
555 "\t[-i init_count (default: %d)] initialize pool i times\n"
556 "\t[-k kill_percentage (default: %llu%%)]\n"
557 "\t[-p pool_name (default: %s)]\n"
558 "\t[-f dir (default: %s)] file directory for vdev files\n"
559 "\t[-V] verbose (use multiple times for ever more blather)\n"
560 "\t[-E] use existing pool instead of creating new one\n"
561 "\t[-T time (default: %llu sec)] total run time\n"
562 "\t[-F freezeloops (default: %llu)] max loops in spa_freeze()\n"
563 "\t[-P passtime (default: %llu sec)] time per pass\n"
564 "\t[-B alt_ztest (default: <none>)] alternate ztest path\n"
565 "\t[-h] (print help)\n"
566 "",
567 zo->zo_pool,
568 (u_longlong_t)zo->zo_vdevs, /* -v */
569 nice_vdev_size, /* -s */
570 zo->zo_ashift, /* -a */
571 zo->zo_mirrors, /* -m */
572 zo->zo_raidz, /* -r */
573 zo->zo_raidz_parity, /* -R */
574 zo->zo_datasets, /* -d */
575 zo->zo_threads, /* -t */
576 nice_gang_bang, /* -g */
577 zo->zo_init, /* -i */
578 (u_longlong_t)zo->zo_killrate, /* -k */
579 zo->zo_pool, /* -p */
580 zo->zo_dir, /* -f */
581 (u_longlong_t)zo->zo_time, /* -T */
582 (u_longlong_t)zo->zo_maxloops, /* -F */
583 (u_longlong_t)zo->zo_passtime);
584 exit(requested ? 0 : 1);
585 }
586
587 static void
588 process_options(int argc, char **argv)
589 {
590 char *path;
591 ztest_shared_opts_t *zo = &ztest_opts;
592
593 int opt;
594 uint64_t value;
595 char altdir[MAXNAMELEN] = { 0 };
596
597 bcopy(&ztest_opts_defaults, zo, sizeof (*zo));
598
599 while ((opt = getopt(argc, argv,
600 "v:s:a:m:r:R:d:t:g:i:k:p:f:VET:P:hF:B:")) != EOF) {
601 value = 0;
602 switch (opt) {
603 case 'v':
604 case 's':
605 case 'a':
606 case 'm':
607 case 'r':
608 case 'R':
609 case 'd':
610 case 't':
611 case 'g':
612 case 'i':
613 case 'k':
614 case 'T':
615 case 'P':
616 case 'F':
617 value = nicenumtoull(optarg);
618 }
619 switch (opt) {
620 case 'v':
621 zo->zo_vdevs = value;
622 break;
623 case 's':
624 zo->zo_vdev_size = MAX(SPA_MINDEVSIZE, value);
625 break;
626 case 'a':
627 zo->zo_ashift = value;
628 break;
629 case 'm':
630 zo->zo_mirrors = value;
631 break;
632 case 'r':
633 zo->zo_raidz = MAX(1, value);
634 break;
635 case 'R':
636 zo->zo_raidz_parity = MIN(MAX(value, 1), 3);
637 break;
638 case 'd':
639 zo->zo_datasets = MAX(1, value);
640 break;
641 case 't':
642 zo->zo_threads = MAX(1, value);
643 break;
644 case 'g':
645 zo->zo_metaslab_gang_bang = MAX(SPA_MINBLOCKSIZE << 1,
646 value);
647 break;
648 case 'i':
649 zo->zo_init = value;
650 break;
651 case 'k':
652 zo->zo_killrate = value;
653 break;
654 case 'p':
655 (void) strlcpy(zo->zo_pool, optarg,
656 sizeof (zo->zo_pool));
657 break;
658 case 'f':
659 path = realpath(optarg, NULL);
660 if (path == NULL) {
661 (void) fprintf(stderr, "error: %s: %s\n",
662 optarg, strerror(errno));
663 usage(B_FALSE);
664 } else {
665 (void) strlcpy(zo->zo_dir, path,
666 sizeof (zo->zo_dir));
667 }
668 break;
669 case 'V':
670 zo->zo_verbose++;
671 break;
672 case 'E':
673 zo->zo_init = 0;
674 break;
675 case 'T':
676 zo->zo_time = value;
677 break;
678 case 'P':
679 zo->zo_passtime = MAX(1, value);
680 break;
681 case 'F':
682 zo->zo_maxloops = MAX(1, value);
683 break;
684 case 'B':
685 (void) strlcpy(altdir, optarg, sizeof (altdir));
686 break;
687 case 'h':
688 usage(B_TRUE);
689 break;
690 case '?':
691 default:
692 usage(B_FALSE);
693 break;
694 }
695 }
696
697 zo->zo_raidz_parity = MIN(zo->zo_raidz_parity, zo->zo_raidz - 1);
698
699 zo->zo_vdevtime =
700 (zo->zo_vdevs > 0 ? zo->zo_time * NANOSEC / zo->zo_vdevs :
701 UINT64_MAX >> 2);
702
703 if (strlen(altdir) > 0) {
704 char cmd[MAXNAMELEN];
705 char realaltdir[MAXNAMELEN];
706 char *bin;
707 char *ztest;
708 char *isa;
709 int isalen;
710
711 (void) realpath(getexecname(), cmd);
712 if (0 != access(altdir, F_OK)) {
713 ztest_dump_core = B_FALSE;
714 fatal(B_TRUE, "invalid alternate ztest path: %s",
715 altdir);
716 }
717 VERIFY(NULL != realpath(altdir, realaltdir));
718
719 /*
720 * 'cmd' should be of the form "<anything>/usr/bin/<isa>/ztest".
721 * We want to extract <isa> to determine if we should use
722 * 32 or 64 bit binaries.
723 */
724 bin = strstr(cmd, "/usr/bin/");
725 ztest = strstr(bin, "/ztest");
726 isa = bin + 9;
727 isalen = ztest - isa;
728 (void) snprintf(zo->zo_alt_ztest, sizeof (zo->zo_alt_ztest),
729 "%s/usr/bin/%.*s/ztest", realaltdir, isalen, isa);
730 (void) snprintf(zo->zo_alt_libpath, sizeof (zo->zo_alt_libpath),
731 "%s/usr/lib/%.*s", realaltdir, isalen, isa);
732
733 if (0 != access(zo->zo_alt_ztest, X_OK)) {
734 ztest_dump_core = B_FALSE;
735 fatal(B_TRUE, "invalid alternate ztest: %s",
736 zo->zo_alt_ztest);
737 } else if (0 != access(zo->zo_alt_libpath, X_OK)) {
738 ztest_dump_core = B_FALSE;
739 fatal(B_TRUE, "invalid alternate lib directory %s",
740 zo->zo_alt_libpath);
741 }
742 }
743 }
744
745 static void
746 ztest_kill(ztest_shared_t *zs)
747 {
748 zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(ztest_spa));
749 zs->zs_space = metaslab_class_get_space(spa_normal_class(ztest_spa));
750 (void) kill(getpid(), SIGKILL);
751 }
752
753 static uint64_t
754 ztest_random(uint64_t range)
755 {
756 uint64_t r;
757
758 if (range == 0)
759 return (0);
760
761 if (read(ZTEST_FD_RAND, &r, sizeof (r)) != sizeof (r))
762 fatal(1, "short read from /dev/urandom");
763
764 return (r % range);
765 }
766
767 /* ARGSUSED */
768 static void
769 ztest_record_enospc(const char *s)
770 {
771 ztest_shared->zs_enospc_count++;
772 }
773
774 static uint64_t
775 ztest_get_ashift(void)
776 {
777 if (ztest_opts.zo_ashift == 0)
778 return (SPA_MINBLOCKSHIFT + ztest_random(3));
779 return (ztest_opts.zo_ashift);
780 }
781
782 static nvlist_t *
783 make_vdev_file(char *path, char *aux, size_t size, uint64_t ashift)
784 {
785 char pathbuf[MAXPATHLEN];
786 uint64_t vdev;
787 nvlist_t *file;
788
789 if (ashift == 0)
790 ashift = ztest_get_ashift();
791
792 if (path == NULL) {
793 path = pathbuf;
794
795 if (aux != NULL) {
796 vdev = ztest_shared->zs_vdev_aux;
797 (void) snprintf(path, sizeof (pathbuf),
798 ztest_aux_template, ztest_opts.zo_dir,
799 ztest_opts.zo_pool, aux, vdev);
800 } else {
801 vdev = ztest_shared->zs_vdev_next_leaf++;
802 (void) snprintf(path, sizeof (pathbuf),
803 ztest_dev_template, ztest_opts.zo_dir,
804 ztest_opts.zo_pool, vdev);
805 }
806 }
807
808 if (size != 0) {
809 int fd = open(path, O_RDWR | O_CREAT | O_TRUNC, 0666);
810 if (fd == -1)
811 fatal(1, "can't open %s", path);
812 if (ftruncate(fd, size) != 0)
813 fatal(1, "can't ftruncate %s", path);
814 (void) close(fd);
815 }
816
817 VERIFY(nvlist_alloc(&file, NV_UNIQUE_NAME, 0) == 0);
818 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_TYPE, VDEV_TYPE_FILE) == 0);
819 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_PATH, path) == 0);
820 VERIFY(nvlist_add_uint64(file, ZPOOL_CONFIG_ASHIFT, ashift) == 0);
821
822 return (file);
823 }
824
825 static nvlist_t *
826 make_vdev_raidz(char *path, char *aux, size_t size, uint64_t ashift, int r)
827 {
828 nvlist_t *raidz, **child;
829 int c;
830
831 if (r < 2)
832 return (make_vdev_file(path, aux, size, ashift));
833 child = umem_alloc(r * sizeof (nvlist_t *), UMEM_NOFAIL);
834
835 for (c = 0; c < r; c++)
836 child[c] = make_vdev_file(path, aux, size, ashift);
837
838 VERIFY(nvlist_alloc(&raidz, NV_UNIQUE_NAME, 0) == 0);
839 VERIFY(nvlist_add_string(raidz, ZPOOL_CONFIG_TYPE,
840 VDEV_TYPE_RAIDZ) == 0);
841 VERIFY(nvlist_add_uint64(raidz, ZPOOL_CONFIG_NPARITY,
842 ztest_opts.zo_raidz_parity) == 0);
843 VERIFY(nvlist_add_nvlist_array(raidz, ZPOOL_CONFIG_CHILDREN,
844 child, r) == 0);
845
846 for (c = 0; c < r; c++)
847 nvlist_free(child[c]);
848
849 umem_free(child, r * sizeof (nvlist_t *));
850
851 return (raidz);
852 }
853
854 static nvlist_t *
855 make_vdev_mirror(char *path, char *aux, size_t size, uint64_t ashift,
856 int r, int m)
857 {
858 nvlist_t *mirror, **child;
859 int c;
860
861 if (m < 1)
862 return (make_vdev_raidz(path, aux, size, ashift, r));
863
864 child = umem_alloc(m * sizeof (nvlist_t *), UMEM_NOFAIL);
865
866 for (c = 0; c < m; c++)
867 child[c] = make_vdev_raidz(path, aux, size, ashift, r);
868
869 VERIFY(nvlist_alloc(&mirror, NV_UNIQUE_NAME, 0) == 0);
870 VERIFY(nvlist_add_string(mirror, ZPOOL_CONFIG_TYPE,
871 VDEV_TYPE_MIRROR) == 0);
872 VERIFY(nvlist_add_nvlist_array(mirror, ZPOOL_CONFIG_CHILDREN,
873 child, m) == 0);
874
875 for (c = 0; c < m; c++)
876 nvlist_free(child[c]);
877
878 umem_free(child, m * sizeof (nvlist_t *));
879
880 return (mirror);
881 }
882
883 static nvlist_t *
884 make_vdev_root(char *path, char *aux, size_t size, uint64_t ashift,
885 int log, int r, int m, int t)
886 {
887 nvlist_t *root, **child;
888 int c;
889
890 ASSERT(t > 0);
891
892 child = umem_alloc(t * sizeof (nvlist_t *), UMEM_NOFAIL);
893
894 for (c = 0; c < t; c++) {
895 child[c] = make_vdev_mirror(path, aux, size, ashift, r, m);
896 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_IS_LOG,
897 log) == 0);
898 }
899
900 VERIFY(nvlist_alloc(&root, NV_UNIQUE_NAME, 0) == 0);
901 VERIFY(nvlist_add_string(root, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) == 0);
902 VERIFY(nvlist_add_nvlist_array(root, aux ? aux : ZPOOL_CONFIG_CHILDREN,
903 child, t) == 0);
904
905 for (c = 0; c < t; c++)
906 nvlist_free(child[c]);
907
908 umem_free(child, t * sizeof (nvlist_t *));
909
910 return (root);
911 }
912
913 static int
914 ztest_random_blocksize(void)
915 {
916 return (1 << (SPA_MINBLOCKSHIFT +
917 ztest_random(SPA_MAXBLOCKSHIFT - SPA_MINBLOCKSHIFT + 1)));
918 }
919
920 static int
921 ztest_random_ibshift(void)
922 {
923 return (DN_MIN_INDBLKSHIFT +
924 ztest_random(DN_MAX_INDBLKSHIFT - DN_MIN_INDBLKSHIFT + 1));
925 }
926
927 static uint64_t
928 ztest_random_vdev_top(spa_t *spa, boolean_t log_ok)
929 {
930 uint64_t top;
931 vdev_t *rvd = spa->spa_root_vdev;
932 vdev_t *tvd;
933
934 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
935
936 do {
937 top = ztest_random(rvd->vdev_children);
938 tvd = rvd->vdev_child[top];
939 } while (tvd->vdev_ishole || (tvd->vdev_islog && !log_ok) ||
940 tvd->vdev_mg == NULL || tvd->vdev_mg->mg_class == NULL);
941
942 return (top);
943 }
944
945 static uint64_t
946 ztest_random_dsl_prop(zfs_prop_t prop)
947 {
948 uint64_t value;
949
950 do {
951 value = zfs_prop_random_value(prop, ztest_random(-1ULL));
952 } while (prop == ZFS_PROP_CHECKSUM && value == ZIO_CHECKSUM_OFF);
953
954 return (value);
955 }
956
957 static int
958 ztest_dsl_prop_set_uint64(char *osname, zfs_prop_t prop, uint64_t value,
959 boolean_t inherit)
960 {
961 const char *propname = zfs_prop_to_name(prop);
962 const char *valname;
963 char setpoint[MAXPATHLEN];
964 uint64_t curval;
965 int error;
966
967 error = dsl_prop_set(osname, propname,
968 (inherit ? ZPROP_SRC_NONE : ZPROP_SRC_LOCAL),
969 sizeof (value), 1, &value);
970
971 if (error == ENOSPC) {
972 ztest_record_enospc(FTAG);
973 return (error);
974 }
975 ASSERT3U(error, ==, 0);
976
977 VERIFY3U(dsl_prop_get(osname, propname, sizeof (curval),
978 1, &curval, setpoint), ==, 0);
979
980 if (ztest_opts.zo_verbose >= 6) {
981 VERIFY(zfs_prop_index_to_string(prop, curval, &valname) == 0);
982 (void) printf("%s %s = %s at '%s'\n",
983 osname, propname, valname, setpoint);
984 }
985
986 return (error);
987 }
988
989 static int
990 ztest_spa_prop_set_uint64(zpool_prop_t prop, uint64_t value)
991 {
992 spa_t *spa = ztest_spa;
993 nvlist_t *props = NULL;
994 int error;
995
996 VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
997 VERIFY(nvlist_add_uint64(props, zpool_prop_to_name(prop), value) == 0);
998
999 error = spa_prop_set(spa, props);
1000
1001 nvlist_free(props);
1002
1003 if (error == ENOSPC) {
1004 ztest_record_enospc(FTAG);
1005 return (error);
1006 }
1007 ASSERT3U(error, ==, 0);
1008
1009 return (error);
1010 }
1011
1012 static void
1013 ztest_rll_init(rll_t *rll)
1014 {
1015 rll->rll_writer = NULL;
1016 rll->rll_readers = 0;
1017 VERIFY(_mutex_init(&rll->rll_lock, USYNC_THREAD, NULL) == 0);
1018 VERIFY(cond_init(&rll->rll_cv, USYNC_THREAD, NULL) == 0);
1019 }
1020
1021 static void
1022 ztest_rll_destroy(rll_t *rll)
1023 {
1024 ASSERT(rll->rll_writer == NULL);
1025 ASSERT(rll->rll_readers == 0);
1026 VERIFY(_mutex_destroy(&rll->rll_lock) == 0);
1027 VERIFY(cond_destroy(&rll->rll_cv) == 0);
1028 }
1029
1030 static void
1031 ztest_rll_lock(rll_t *rll, rl_type_t type)
1032 {
1033 VERIFY(mutex_lock(&rll->rll_lock) == 0);
1034
1035 if (type == RL_READER) {
1036 while (rll->rll_writer != NULL)
1037 (void) cond_wait(&rll->rll_cv, &rll->rll_lock);
1038 rll->rll_readers++;
1039 } else {
1040 while (rll->rll_writer != NULL || rll->rll_readers)
1041 (void) cond_wait(&rll->rll_cv, &rll->rll_lock);
1042 rll->rll_writer = curthread;
1043 }
1044
1045 VERIFY(mutex_unlock(&rll->rll_lock) == 0);
1046 }
1047
1048 static void
1049 ztest_rll_unlock(rll_t *rll)
1050 {
1051 VERIFY(mutex_lock(&rll->rll_lock) == 0);
1052
1053 if (rll->rll_writer) {
1054 ASSERT(rll->rll_readers == 0);
1055 rll->rll_writer = NULL;
1056 } else {
1057 ASSERT(rll->rll_readers != 0);
1058 ASSERT(rll->rll_writer == NULL);
1059 rll->rll_readers--;
1060 }
1061
1062 if (rll->rll_writer == NULL && rll->rll_readers == 0)
1063 VERIFY(cond_broadcast(&rll->rll_cv) == 0);
1064
1065 VERIFY(mutex_unlock(&rll->rll_lock) == 0);
1066 }
1067
1068 static void
1069 ztest_object_lock(ztest_ds_t *zd, uint64_t object, rl_type_t type)
1070 {
1071 rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1072
1073 ztest_rll_lock(rll, type);
1074 }
1075
1076 static void
1077 ztest_object_unlock(ztest_ds_t *zd, uint64_t object)
1078 {
1079 rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1080
1081 ztest_rll_unlock(rll);
1082 }
1083
1084 static rl_t *
1085 ztest_range_lock(ztest_ds_t *zd, uint64_t object, uint64_t offset,
1086 uint64_t size, rl_type_t type)
1087 {
1088 uint64_t hash = object ^ (offset % (ZTEST_RANGE_LOCKS + 1));
1089 rll_t *rll = &zd->zd_range_lock[hash & (ZTEST_RANGE_LOCKS - 1)];
1090 rl_t *rl;
1091
1092 rl = umem_alloc(sizeof (*rl), UMEM_NOFAIL);
1093 rl->rl_object = object;
1094 rl->rl_offset = offset;
1095 rl->rl_size = size;
1096 rl->rl_lock = rll;
1097
1098 ztest_rll_lock(rll, type);
1099
1100 return (rl);
1101 }
1102
1103 static void
1104 ztest_range_unlock(rl_t *rl)
1105 {
1106 rll_t *rll = rl->rl_lock;
1107
1108 ztest_rll_unlock(rll);
1109
1110 umem_free(rl, sizeof (*rl));
1111 }
1112
1113 static void
1114 ztest_zd_init(ztest_ds_t *zd, ztest_shared_ds_t *szd, objset_t *os)
1115 {
1116 zd->zd_os = os;
1117 zd->zd_zilog = dmu_objset_zil(os);
1118 zd->zd_shared = szd;
1119 dmu_objset_name(os, zd->zd_name);
1120
1121 if (zd->zd_shared != NULL)
1122 zd->zd_shared->zd_seq = 0;
1123
1124 VERIFY(rwlock_init(&zd->zd_zilog_lock, USYNC_THREAD, NULL) == 0);
1125 VERIFY(_mutex_init(&zd->zd_dirobj_lock, USYNC_THREAD, NULL) == 0);
1126
1127 for (int l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1128 ztest_rll_init(&zd->zd_object_lock[l]);
1129
1130 for (int l = 0; l < ZTEST_RANGE_LOCKS; l++)
1131 ztest_rll_init(&zd->zd_range_lock[l]);
1132 }
1133
1134 static void
1135 ztest_zd_fini(ztest_ds_t *zd)
1136 {
1137 VERIFY(_mutex_destroy(&zd->zd_dirobj_lock) == 0);
1138
1139 for (int l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1140 ztest_rll_destroy(&zd->zd_object_lock[l]);
1141
1142 for (int l = 0; l < ZTEST_RANGE_LOCKS; l++)
1143 ztest_rll_destroy(&zd->zd_range_lock[l]);
1144 }
1145
1146 #define TXG_MIGHTWAIT (ztest_random(10) == 0 ? TXG_NOWAIT : TXG_WAIT)
1147
1148 static uint64_t
1149 ztest_tx_assign(dmu_tx_t *tx, uint64_t txg_how, const char *tag)
1150 {
1151 uint64_t txg;
1152 int error;
1153
1154 /*
1155 * Attempt to assign tx to some transaction group.
1156 */
1157 error = dmu_tx_assign(tx, txg_how);
1158 if (error) {
1159 if (error == ERESTART) {
1160 ASSERT(txg_how == TXG_NOWAIT);
1161 dmu_tx_wait(tx);
1162 } else {
1163 ASSERT3U(error, ==, ENOSPC);
1164 ztest_record_enospc(tag);
1165 }
1166 dmu_tx_abort(tx);
1167 return (0);
1168 }
1169 txg = dmu_tx_get_txg(tx);
1170 ASSERT(txg != 0);
1171 return (txg);
1172 }
1173
1174 static void
1175 ztest_pattern_set(void *buf, uint64_t size, uint64_t value)
1176 {
1177 uint64_t *ip = buf;
1178 uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1179
1180 while (ip < ip_end)
1181 *ip++ = value;
1182 }
1183
1184 static boolean_t
1185 ztest_pattern_match(void *buf, uint64_t size, uint64_t value)
1186 {
1187 uint64_t *ip = buf;
1188 uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1189 uint64_t diff = 0;
1190
1191 while (ip < ip_end)
1192 diff |= (value - *ip++);
1193
1194 return (diff == 0);
1195 }
1196
1197 static void
1198 ztest_bt_generate(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1199 uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg)
1200 {
1201 bt->bt_magic = BT_MAGIC;
1202 bt->bt_objset = dmu_objset_id(os);
1203 bt->bt_object = object;
1204 bt->bt_offset = offset;
1205 bt->bt_gen = gen;
1206 bt->bt_txg = txg;
1207 bt->bt_crtxg = crtxg;
1208 }
1209
1210 static void
1211 ztest_bt_verify(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1212 uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg)
1213 {
1214 ASSERT(bt->bt_magic == BT_MAGIC);
1215 ASSERT(bt->bt_objset == dmu_objset_id(os));
1216 ASSERT(bt->bt_object == object);
1217 ASSERT(bt->bt_offset == offset);
1218 ASSERT(bt->bt_gen <= gen);
1219 ASSERT(bt->bt_txg <= txg);
1220 ASSERT(bt->bt_crtxg == crtxg);
1221 }
1222
1223 static ztest_block_tag_t *
1224 ztest_bt_bonus(dmu_buf_t *db)
1225 {
1226 dmu_object_info_t doi;
1227 ztest_block_tag_t *bt;
1228
1229 dmu_object_info_from_db(db, &doi);
1230 ASSERT3U(doi.doi_bonus_size, <=, db->db_size);
1231 ASSERT3U(doi.doi_bonus_size, >=, sizeof (*bt));
1232 bt = (void *)((char *)db->db_data + doi.doi_bonus_size - sizeof (*bt));
1233
1234 return (bt);
1235 }
1236
1237 /*
1238 * ZIL logging ops
1239 */
1240
1241 #define lrz_type lr_mode
1242 #define lrz_blocksize lr_uid
1243 #define lrz_ibshift lr_gid
1244 #define lrz_bonustype lr_rdev
1245 #define lrz_bonuslen lr_crtime[1]
1246
1247 static void
1248 ztest_log_create(ztest_ds_t *zd, dmu_tx_t *tx, lr_create_t *lr)
1249 {
1250 char *name = (void *)(lr + 1); /* name follows lr */
1251 size_t namesize = strlen(name) + 1;
1252 itx_t *itx;
1253
1254 if (zil_replaying(zd->zd_zilog, tx))
1255 return;
1256
1257 itx = zil_itx_create(TX_CREATE, sizeof (*lr) + namesize);
1258 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1259 sizeof (*lr) + namesize - sizeof (lr_t));
1260
1261 zil_itx_assign(zd->zd_zilog, itx, tx);
1262 }
1263
1264 static void
1265 ztest_log_remove(ztest_ds_t *zd, dmu_tx_t *tx, lr_remove_t *lr, uint64_t object)
1266 {
1267 char *name = (void *)(lr + 1); /* name follows lr */
1268 size_t namesize = strlen(name) + 1;
1269 itx_t *itx;
1270
1271 if (zil_replaying(zd->zd_zilog, tx))
1272 return;
1273
1274 itx = zil_itx_create(TX_REMOVE, sizeof (*lr) + namesize);
1275 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1276 sizeof (*lr) + namesize - sizeof (lr_t));
1277
1278 itx->itx_oid = object;
1279 zil_itx_assign(zd->zd_zilog, itx, tx);
1280 }
1281
1282 static void
1283 ztest_log_write(ztest_ds_t *zd, dmu_tx_t *tx, lr_write_t *lr)
1284 {
1285 itx_t *itx;
1286 itx_wr_state_t write_state = ztest_random(WR_NUM_STATES);
1287
1288 if (zil_replaying(zd->zd_zilog, tx))
1289 return;
1290
1291 if (lr->lr_length > ZIL_MAX_LOG_DATA)
1292 write_state = WR_INDIRECT;
1293
1294 itx = zil_itx_create(TX_WRITE,
1295 sizeof (*lr) + (write_state == WR_COPIED ? lr->lr_length : 0));
1296
1297 if (write_state == WR_COPIED &&
1298 dmu_read(zd->zd_os, lr->lr_foid, lr->lr_offset, lr->lr_length,
1299 ((lr_write_t *)&itx->itx_lr) + 1, DMU_READ_NO_PREFETCH) != 0) {
1300 zil_itx_destroy(itx);
1301 itx = zil_itx_create(TX_WRITE, sizeof (*lr));
1302 write_state = WR_NEED_COPY;
1303 }
1304 itx->itx_private = zd;
1305 itx->itx_wr_state = write_state;
1306 itx->itx_sync = (ztest_random(8) == 0);
1307 itx->itx_sod += (write_state == WR_NEED_COPY ? lr->lr_length : 0);
1308
1309 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1310 sizeof (*lr) - sizeof (lr_t));
1311
1312 zil_itx_assign(zd->zd_zilog, itx, tx);
1313 }
1314
1315 static void
1316 ztest_log_truncate(ztest_ds_t *zd, dmu_tx_t *tx, lr_truncate_t *lr)
1317 {
1318 itx_t *itx;
1319
1320 if (zil_replaying(zd->zd_zilog, tx))
1321 return;
1322
1323 itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr));
1324 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1325 sizeof (*lr) - sizeof (lr_t));
1326
1327 itx->itx_sync = B_FALSE;
1328 zil_itx_assign(zd->zd_zilog, itx, tx);
1329 }
1330
1331 static void
1332 ztest_log_setattr(ztest_ds_t *zd, dmu_tx_t *tx, lr_setattr_t *lr)
1333 {
1334 itx_t *itx;
1335
1336 if (zil_replaying(zd->zd_zilog, tx))
1337 return;
1338
1339 itx = zil_itx_create(TX_SETATTR, sizeof (*lr));
1340 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1341 sizeof (*lr) - sizeof (lr_t));
1342
1343 itx->itx_sync = B_FALSE;
1344 zil_itx_assign(zd->zd_zilog, itx, tx);
1345 }
1346
1347 /*
1348 * ZIL replay ops
1349 */
1350 static int
1351 ztest_replay_create(ztest_ds_t *zd, lr_create_t *lr, boolean_t byteswap)
1352 {
1353 char *name = (void *)(lr + 1); /* name follows lr */
1354 objset_t *os = zd->zd_os;
1355 ztest_block_tag_t *bbt;
1356 dmu_buf_t *db;
1357 dmu_tx_t *tx;
1358 uint64_t txg;
1359 int error = 0;
1360
1361 if (byteswap)
1362 byteswap_uint64_array(lr, sizeof (*lr));
1363
1364 ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1365 ASSERT(name[0] != '\0');
1366
1367 tx = dmu_tx_create(os);
1368
1369 dmu_tx_hold_zap(tx, lr->lr_doid, B_TRUE, name);
1370
1371 if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1372 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, B_TRUE, NULL);
1373 } else {
1374 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1375 }
1376
1377 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1378 if (txg == 0)
1379 return (ENOSPC);
1380
1381 ASSERT(dmu_objset_zil(os)->zl_replay == !!lr->lr_foid);
1382
1383 if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1384 if (lr->lr_foid == 0) {
1385 lr->lr_foid = zap_create(os,
1386 lr->lrz_type, lr->lrz_bonustype,
1387 lr->lrz_bonuslen, tx);
1388 } else {
1389 error = zap_create_claim(os, lr->lr_foid,
1390 lr->lrz_type, lr->lrz_bonustype,
1391 lr->lrz_bonuslen, tx);
1392 }
1393 } else {
1394 if (lr->lr_foid == 0) {
1395 lr->lr_foid = dmu_object_alloc(os,
1396 lr->lrz_type, 0, lr->lrz_bonustype,
1397 lr->lrz_bonuslen, tx);
1398 } else {
1399 error = dmu_object_claim(os, lr->lr_foid,
1400 lr->lrz_type, 0, lr->lrz_bonustype,
1401 lr->lrz_bonuslen, tx);
1402 }
1403 }
1404
1405 if (error) {
1406 ASSERT3U(error, ==, EEXIST);
1407 ASSERT(zd->zd_zilog->zl_replay);
1408 dmu_tx_commit(tx);
1409 return (error);
1410 }
1411
1412 ASSERT(lr->lr_foid != 0);
1413
1414 if (lr->lrz_type != DMU_OT_ZAP_OTHER)
1415 VERIFY3U(0, ==, dmu_object_set_blocksize(os, lr->lr_foid,
1416 lr->lrz_blocksize, lr->lrz_ibshift, tx));
1417
1418 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1419 bbt = ztest_bt_bonus(db);
1420 dmu_buf_will_dirty(db, tx);
1421 ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_gen, txg, txg);
1422 dmu_buf_rele(db, FTAG);
1423
1424 VERIFY3U(0, ==, zap_add(os, lr->lr_doid, name, sizeof (uint64_t), 1,
1425 &lr->lr_foid, tx));
1426
1427 (void) ztest_log_create(zd, tx, lr);
1428
1429 dmu_tx_commit(tx);
1430
1431 return (0);
1432 }
1433
1434 static int
1435 ztest_replay_remove(ztest_ds_t *zd, lr_remove_t *lr, boolean_t byteswap)
1436 {
1437 char *name = (void *)(lr + 1); /* name follows lr */
1438 objset_t *os = zd->zd_os;
1439 dmu_object_info_t doi;
1440 dmu_tx_t *tx;
1441 uint64_t object, txg;
1442
1443 if (byteswap)
1444 byteswap_uint64_array(lr, sizeof (*lr));
1445
1446 ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1447 ASSERT(name[0] != '\0');
1448
1449 VERIFY3U(0, ==,
1450 zap_lookup(os, lr->lr_doid, name, sizeof (object), 1, &object));
1451 ASSERT(object != 0);
1452
1453 ztest_object_lock(zd, object, RL_WRITER);
1454
1455 VERIFY3U(0, ==, dmu_object_info(os, object, &doi));
1456
1457 tx = dmu_tx_create(os);
1458
1459 dmu_tx_hold_zap(tx, lr->lr_doid, B_FALSE, name);
1460 dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END);
1461
1462 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1463 if (txg == 0) {
1464 ztest_object_unlock(zd, object);
1465 return (ENOSPC);
1466 }
1467
1468 if (doi.doi_type == DMU_OT_ZAP_OTHER) {
1469 VERIFY3U(0, ==, zap_destroy(os, object, tx));
1470 } else {
1471 VERIFY3U(0, ==, dmu_object_free(os, object, tx));
1472 }
1473
1474 VERIFY3U(0, ==, zap_remove(os, lr->lr_doid, name, tx));
1475
1476 (void) ztest_log_remove(zd, tx, lr, object);
1477
1478 dmu_tx_commit(tx);
1479
1480 ztest_object_unlock(zd, object);
1481
1482 return (0);
1483 }
1484
1485 static int
1486 ztest_replay_write(ztest_ds_t *zd, lr_write_t *lr, boolean_t byteswap)
1487 {
1488 objset_t *os = zd->zd_os;
1489 void *data = lr + 1; /* data follows lr */
1490 uint64_t offset, length;
1491 ztest_block_tag_t *bt = data;
1492 ztest_block_tag_t *bbt;
1493 uint64_t gen, txg, lrtxg, crtxg;
1494 dmu_object_info_t doi;
1495 dmu_tx_t *tx;
1496 dmu_buf_t *db;
1497 arc_buf_t *abuf = NULL;
1498 rl_t *rl;
1499
1500 if (byteswap)
1501 byteswap_uint64_array(lr, sizeof (*lr));
1502
1503 offset = lr->lr_offset;
1504 length = lr->lr_length;
1505
1506 /* If it's a dmu_sync() block, write the whole block */
1507 if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
1508 uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
1509 if (length < blocksize) {
1510 offset -= offset % blocksize;
1511 length = blocksize;
1512 }
1513 }
1514
1515 if (bt->bt_magic == BSWAP_64(BT_MAGIC))
1516 byteswap_uint64_array(bt, sizeof (*bt));
1517
1518 if (bt->bt_magic != BT_MAGIC)
1519 bt = NULL;
1520
1521 ztest_object_lock(zd, lr->lr_foid, RL_READER);
1522 rl = ztest_range_lock(zd, lr->lr_foid, offset, length, RL_WRITER);
1523
1524 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1525
1526 dmu_object_info_from_db(db, &doi);
1527
1528 bbt = ztest_bt_bonus(db);
1529 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1530 gen = bbt->bt_gen;
1531 crtxg = bbt->bt_crtxg;
1532 lrtxg = lr->lr_common.lrc_txg;
1533
1534 tx = dmu_tx_create(os);
1535
1536 dmu_tx_hold_write(tx, lr->lr_foid, offset, length);
1537
1538 if (ztest_random(8) == 0 && length == doi.doi_data_block_size &&
1539 P2PHASE(offset, length) == 0)
1540 abuf = dmu_request_arcbuf(db, length);
1541
1542 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1543 if (txg == 0) {
1544 if (abuf != NULL)
1545 dmu_return_arcbuf(abuf);
1546 dmu_buf_rele(db, FTAG);
1547 ztest_range_unlock(rl);
1548 ztest_object_unlock(zd, lr->lr_foid);
1549 return (ENOSPC);
1550 }
1551
1552 if (bt != NULL) {
1553 /*
1554 * Usually, verify the old data before writing new data --
1555 * but not always, because we also want to verify correct
1556 * behavior when the data was not recently read into cache.
1557 */
1558 ASSERT(offset % doi.doi_data_block_size == 0);
1559 if (ztest_random(4) != 0) {
1560 int prefetch = ztest_random(2) ?
1561 DMU_READ_PREFETCH : DMU_READ_NO_PREFETCH;
1562 ztest_block_tag_t rbt;
1563
1564 VERIFY(dmu_read(os, lr->lr_foid, offset,
1565 sizeof (rbt), &rbt, prefetch) == 0);
1566 if (rbt.bt_magic == BT_MAGIC) {
1567 ztest_bt_verify(&rbt, os, lr->lr_foid,
1568 offset, gen, txg, crtxg);
1569 }
1570 }
1571
1572 /*
1573 * Writes can appear to be newer than the bonus buffer because
1574 * the ztest_get_data() callback does a dmu_read() of the
1575 * open-context data, which may be different than the data
1576 * as it was when the write was generated.
1577 */
1578 if (zd->zd_zilog->zl_replay) {
1579 ztest_bt_verify(bt, os, lr->lr_foid, offset,
1580 MAX(gen, bt->bt_gen), MAX(txg, lrtxg),
1581 bt->bt_crtxg);
1582 }
1583
1584 /*
1585 * Set the bt's gen/txg to the bonus buffer's gen/txg
1586 * so that all of the usual ASSERTs will work.
1587 */
1588 ztest_bt_generate(bt, os, lr->lr_foid, offset, gen, txg, crtxg);
1589 }
1590
1591 if (abuf == NULL) {
1592 dmu_write(os, lr->lr_foid, offset, length, data, tx);
1593 } else {
1594 bcopy(data, abuf->b_data, length);
1595 dmu_assign_arcbuf(db, offset, abuf, tx);
1596 }
1597
1598 (void) ztest_log_write(zd, tx, lr);
1599
1600 dmu_buf_rele(db, FTAG);
1601
1602 dmu_tx_commit(tx);
1603
1604 ztest_range_unlock(rl);
1605 ztest_object_unlock(zd, lr->lr_foid);
1606
1607 return (0);
1608 }
1609
1610 static int
1611 ztest_replay_truncate(ztest_ds_t *zd, lr_truncate_t *lr, boolean_t byteswap)
1612 {
1613 objset_t *os = zd->zd_os;
1614 dmu_tx_t *tx;
1615 uint64_t txg;
1616 rl_t *rl;
1617
1618 if (byteswap)
1619 byteswap_uint64_array(lr, sizeof (*lr));
1620
1621 ztest_object_lock(zd, lr->lr_foid, RL_READER);
1622 rl = ztest_range_lock(zd, lr->lr_foid, lr->lr_offset, lr->lr_length,
1623 RL_WRITER);
1624
1625 tx = dmu_tx_create(os);
1626
1627 dmu_tx_hold_free(tx, lr->lr_foid, lr->lr_offset, lr->lr_length);
1628
1629 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1630 if (txg == 0) {
1631 ztest_range_unlock(rl);
1632 ztest_object_unlock(zd, lr->lr_foid);
1633 return (ENOSPC);
1634 }
1635
1636 VERIFY(dmu_free_range(os, lr->lr_foid, lr->lr_offset,
1637 lr->lr_length, tx) == 0);
1638
1639 (void) ztest_log_truncate(zd, tx, lr);
1640
1641 dmu_tx_commit(tx);
1642
1643 ztest_range_unlock(rl);
1644 ztest_object_unlock(zd, lr->lr_foid);
1645
1646 return (0);
1647 }
1648
1649 static int
1650 ztest_replay_setattr(ztest_ds_t *zd, lr_setattr_t *lr, boolean_t byteswap)
1651 {
1652 objset_t *os = zd->zd_os;
1653 dmu_tx_t *tx;
1654 dmu_buf_t *db;
1655 ztest_block_tag_t *bbt;
1656 uint64_t txg, lrtxg, crtxg;
1657
1658 if (byteswap)
1659 byteswap_uint64_array(lr, sizeof (*lr));
1660
1661 ztest_object_lock(zd, lr->lr_foid, RL_WRITER);
1662
1663 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1664
1665 tx = dmu_tx_create(os);
1666 dmu_tx_hold_bonus(tx, lr->lr_foid);
1667
1668 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1669 if (txg == 0) {
1670 dmu_buf_rele(db, FTAG);
1671 ztest_object_unlock(zd, lr->lr_foid);
1672 return (ENOSPC);
1673 }
1674
1675 bbt = ztest_bt_bonus(db);
1676 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1677 crtxg = bbt->bt_crtxg;
1678 lrtxg = lr->lr_common.lrc_txg;
1679
1680 if (zd->zd_zilog->zl_replay) {
1681 ASSERT(lr->lr_size != 0);
1682 ASSERT(lr->lr_mode != 0);
1683 ASSERT(lrtxg != 0);
1684 } else {
1685 /*
1686 * Randomly change the size and increment the generation.
1687 */
1688 lr->lr_size = (ztest_random(db->db_size / sizeof (*bbt)) + 1) *
1689 sizeof (*bbt);
1690 lr->lr_mode = bbt->bt_gen + 1;
1691 ASSERT(lrtxg == 0);
1692 }
1693
1694 /*
1695 * Verify that the current bonus buffer is not newer than our txg.
1696 */
1697 ztest_bt_verify(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode,
1698 MAX(txg, lrtxg), crtxg);
1699
1700 dmu_buf_will_dirty(db, tx);
1701
1702 ASSERT3U(lr->lr_size, >=, sizeof (*bbt));
1703 ASSERT3U(lr->lr_size, <=, db->db_size);
1704 VERIFY3U(dmu_set_bonus(db, lr->lr_size, tx), ==, 0);
1705 bbt = ztest_bt_bonus(db);
1706
1707 ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode, txg, crtxg);
1708
1709 dmu_buf_rele(db, FTAG);
1710
1711 (void) ztest_log_setattr(zd, tx, lr);
1712
1713 dmu_tx_commit(tx);
1714
1715 ztest_object_unlock(zd, lr->lr_foid);
1716
1717 return (0);
1718 }
1719
1720 zil_replay_func_t *ztest_replay_vector[TX_MAX_TYPE] = {
1721 NULL, /* 0 no such transaction type */
1722 ztest_replay_create, /* TX_CREATE */
1723 NULL, /* TX_MKDIR */
1724 NULL, /* TX_MKXATTR */
1725 NULL, /* TX_SYMLINK */
1726 ztest_replay_remove, /* TX_REMOVE */
1727 NULL, /* TX_RMDIR */
1728 NULL, /* TX_LINK */
1729 NULL, /* TX_RENAME */
1730 ztest_replay_write, /* TX_WRITE */
1731 ztest_replay_truncate, /* TX_TRUNCATE */
1732 ztest_replay_setattr, /* TX_SETATTR */
1733 NULL, /* TX_ACL */
1734 NULL, /* TX_CREATE_ACL */
1735 NULL, /* TX_CREATE_ATTR */
1736 NULL, /* TX_CREATE_ACL_ATTR */
1737 NULL, /* TX_MKDIR_ACL */
1738 NULL, /* TX_MKDIR_ATTR */
1739 NULL, /* TX_MKDIR_ACL_ATTR */
1740 NULL, /* TX_WRITE2 */
1741 };
1742
1743 /*
1744 * ZIL get_data callbacks
1745 */
1746
1747 static void
1748 ztest_get_done(zgd_t *zgd, int error)
1749 {
1750 ztest_ds_t *zd = zgd->zgd_private;
1751 uint64_t object = zgd->zgd_rl->rl_object;
1752
1753 if (zgd->zgd_db)
1754 dmu_buf_rele(zgd->zgd_db, zgd);
1755
1756 ztest_range_unlock(zgd->zgd_rl);
1757 ztest_object_unlock(zd, object);
1758
1759 if (error == 0 && zgd->zgd_bp)
1760 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
1761
1762 umem_free(zgd, sizeof (*zgd));
1763 }
1764
1765 static int
1766 ztest_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
1767 {
1768 ztest_ds_t *zd = arg;
1769 objset_t *os = zd->zd_os;
1770 uint64_t object = lr->lr_foid;
1771 uint64_t offset = lr->lr_offset;
1772 uint64_t size = lr->lr_length;
1773 blkptr_t *bp = &lr->lr_blkptr;
1774 uint64_t txg = lr->lr_common.lrc_txg;
1775 uint64_t crtxg;
1776 dmu_object_info_t doi;
1777 dmu_buf_t *db;
1778 zgd_t *zgd;
1779 int error;
1780
1781 ztest_object_lock(zd, object, RL_READER);
1782 error = dmu_bonus_hold(os, object, FTAG, &db);
1783 if (error) {
1784 ztest_object_unlock(zd, object);
1785 return (error);
1786 }
1787
1788 crtxg = ztest_bt_bonus(db)->bt_crtxg;
1789
1790 if (crtxg == 0 || crtxg > txg) {
1791 dmu_buf_rele(db, FTAG);
1792 ztest_object_unlock(zd, object);
1793 return (ENOENT);
1794 }
1795
1796 dmu_object_info_from_db(db, &doi);
1797 dmu_buf_rele(db, FTAG);
1798 db = NULL;
1799
1800 zgd = umem_zalloc(sizeof (*zgd), UMEM_NOFAIL);
1801 zgd->zgd_zilog = zd->zd_zilog;
1802 zgd->zgd_private = zd;
1803
1804 if (buf != NULL) { /* immediate write */
1805 zgd->zgd_rl = ztest_range_lock(zd, object, offset, size,
1806 RL_READER);
1807
1808 error = dmu_read(os, object, offset, size, buf,
1809 DMU_READ_NO_PREFETCH);
1810 ASSERT(error == 0);
1811 } else {
1812 size = doi.doi_data_block_size;
1813 if (ISP2(size)) {
1814 offset = P2ALIGN(offset, size);
1815 } else {
1816 ASSERT(offset < size);
1817 offset = 0;
1818 }
1819
1820 zgd->zgd_rl = ztest_range_lock(zd, object, offset, size,
1821 RL_READER);
1822
1823 error = dmu_buf_hold(os, object, offset, zgd, &db,
1824 DMU_READ_NO_PREFETCH);
1825
1826 if (error == 0) {
1827 zgd->zgd_db = db;
1828 zgd->zgd_bp = bp;
1829
1830 ASSERT(db->db_offset == offset);
1831 ASSERT(db->db_size == size);
1832
1833 error = dmu_sync(zio, lr->lr_common.lrc_txg,
1834 ztest_get_done, zgd);
1835
1836 if (error == 0)
1837 return (0);
1838 }
1839 }
1840
1841 ztest_get_done(zgd, error);
1842
1843 return (error);
1844 }
1845
1846 static void *
1847 ztest_lr_alloc(size_t lrsize, char *name)
1848 {
1849 char *lr;
1850 size_t namesize = name ? strlen(name) + 1 : 0;
1851
1852 lr = umem_zalloc(lrsize + namesize, UMEM_NOFAIL);
1853
1854 if (name)
1855 bcopy(name, lr + lrsize, namesize);
1856
1857 return (lr);
1858 }
1859
1860 void
1861 ztest_lr_free(void *lr, size_t lrsize, char *name)
1862 {
1863 size_t namesize = name ? strlen(name) + 1 : 0;
1864
1865 umem_free(lr, lrsize + namesize);
1866 }
1867
1868 /*
1869 * Lookup a bunch of objects. Returns the number of objects not found.
1870 */
1871 static int
1872 ztest_lookup(ztest_ds_t *zd, ztest_od_t *od, int count)
1873 {
1874 int missing = 0;
1875 int error;
1876
1877 ASSERT(_mutex_held(&zd->zd_dirobj_lock));
1878
1879 for (int i = 0; i < count; i++, od++) {
1880 od->od_object = 0;
1881 error = zap_lookup(zd->zd_os, od->od_dir, od->od_name,
1882 sizeof (uint64_t), 1, &od->od_object);
1883 if (error) {
1884 ASSERT(error == ENOENT);
1885 ASSERT(od->od_object == 0);
1886 missing++;
1887 } else {
1888 dmu_buf_t *db;
1889 ztest_block_tag_t *bbt;
1890 dmu_object_info_t doi;
1891
1892 ASSERT(od->od_object != 0);
1893 ASSERT(missing == 0); /* there should be no gaps */
1894
1895 ztest_object_lock(zd, od->od_object, RL_READER);
1896 VERIFY3U(0, ==, dmu_bonus_hold(zd->zd_os,
1897 od->od_object, FTAG, &db));
1898 dmu_object_info_from_db(db, &doi);
1899 bbt = ztest_bt_bonus(db);
1900 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1901 od->od_type = doi.doi_type;
1902 od->od_blocksize = doi.doi_data_block_size;
1903 od->od_gen = bbt->bt_gen;
1904 dmu_buf_rele(db, FTAG);
1905 ztest_object_unlock(zd, od->od_object);
1906 }
1907 }
1908
1909 return (missing);
1910 }
1911
1912 static int
1913 ztest_create(ztest_ds_t *zd, ztest_od_t *od, int count)
1914 {
1915 int missing = 0;
1916
1917 ASSERT(_mutex_held(&zd->zd_dirobj_lock));
1918
1919 for (int i = 0; i < count; i++, od++) {
1920 if (missing) {
1921 od->od_object = 0;
1922 missing++;
1923 continue;
1924 }
1925
1926 lr_create_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
1927
1928 lr->lr_doid = od->od_dir;
1929 lr->lr_foid = 0; /* 0 to allocate, > 0 to claim */
1930 lr->lrz_type = od->od_crtype;
1931 lr->lrz_blocksize = od->od_crblocksize;
1932 lr->lrz_ibshift = ztest_random_ibshift();
1933 lr->lrz_bonustype = DMU_OT_UINT64_OTHER;
1934 lr->lrz_bonuslen = dmu_bonus_max();
1935 lr->lr_gen = od->od_crgen;
1936 lr->lr_crtime[0] = time(NULL);
1937
1938 if (ztest_replay_create(zd, lr, B_FALSE) != 0) {
1939 ASSERT(missing == 0);
1940 od->od_object = 0;
1941 missing++;
1942 } else {
1943 od->od_object = lr->lr_foid;
1944 od->od_type = od->od_crtype;
1945 od->od_blocksize = od->od_crblocksize;
1946 od->od_gen = od->od_crgen;
1947 ASSERT(od->od_object != 0);
1948 }
1949
1950 ztest_lr_free(lr, sizeof (*lr), od->od_name);
1951 }
1952
1953 return (missing);
1954 }
1955
1956 static int
1957 ztest_remove(ztest_ds_t *zd, ztest_od_t *od, int count)
1958 {
1959 int missing = 0;
1960 int error;
1961
1962 ASSERT(_mutex_held(&zd->zd_dirobj_lock));
1963
1964 od += count - 1;
1965
1966 for (int i = count - 1; i >= 0; i--, od--) {
1967 if (missing) {
1968 missing++;
1969 continue;
1970 }
1971
1972 if (od->od_object == 0)
1973 continue;
1974
1975 lr_remove_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
1976
1977 lr->lr_doid = od->od_dir;
1978
1979 if ((error = ztest_replay_remove(zd, lr, B_FALSE)) != 0) {
1980 ASSERT3U(error, ==, ENOSPC);
1981 missing++;
1982 } else {
1983 od->od_object = 0;
1984 }
1985 ztest_lr_free(lr, sizeof (*lr), od->od_name);
1986 }
1987
1988 return (missing);
1989 }
1990
1991 static int
1992 ztest_write(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size,
1993 void *data)
1994 {
1995 lr_write_t *lr;
1996 int error;
1997
1998 lr = ztest_lr_alloc(sizeof (*lr) + size, NULL);
1999
2000 lr->lr_foid = object;
2001 lr->lr_offset = offset;
2002 lr->lr_length = size;
2003 lr->lr_blkoff = 0;
2004 BP_ZERO(&lr->lr_blkptr);
2005
2006 bcopy(data, lr + 1, size);
2007
2008 error = ztest_replay_write(zd, lr, B_FALSE);
2009
2010 ztest_lr_free(lr, sizeof (*lr) + size, NULL);
2011
2012 return (error);
2013 }
2014
2015 static int
2016 ztest_truncate(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2017 {
2018 lr_truncate_t *lr;
2019 int error;
2020
2021 lr = ztest_lr_alloc(sizeof (*lr), NULL);
2022
2023 lr->lr_foid = object;
2024 lr->lr_offset = offset;
2025 lr->lr_length = size;
2026
2027 error = ztest_replay_truncate(zd, lr, B_FALSE);
2028
2029 ztest_lr_free(lr, sizeof (*lr), NULL);
2030
2031 return (error);
2032 }
2033
2034 static int
2035 ztest_setattr(ztest_ds_t *zd, uint64_t object)
2036 {
2037 lr_setattr_t *lr;
2038 int error;
2039
2040 lr = ztest_lr_alloc(sizeof (*lr), NULL);
2041
2042 lr->lr_foid = object;
2043 lr->lr_size = 0;
2044 lr->lr_mode = 0;
2045
2046 error = ztest_replay_setattr(zd, lr, B_FALSE);
2047
2048 ztest_lr_free(lr, sizeof (*lr), NULL);
2049
2050 return (error);
2051 }
2052
2053 static void
2054 ztest_prealloc(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2055 {
2056 objset_t *os = zd->zd_os;
2057 dmu_tx_t *tx;
2058 uint64_t txg;
2059 rl_t *rl;
2060
2061 txg_wait_synced(dmu_objset_pool(os), 0);
2062
2063 ztest_object_lock(zd, object, RL_READER);
2064 rl = ztest_range_lock(zd, object, offset, size, RL_WRITER);
2065
2066 tx = dmu_tx_create(os);
2067
2068 dmu_tx_hold_write(tx, object, offset, size);
2069
2070 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
2071
2072 if (txg != 0) {
2073 dmu_prealloc(os, object, offset, size, tx);
2074 dmu_tx_commit(tx);
2075 txg_wait_synced(dmu_objset_pool(os), txg);
2076 } else {
2077 (void) dmu_free_long_range(os, object, offset, size);
2078 }
2079
2080 ztest_range_unlock(rl);
2081 ztest_object_unlock(zd, object);
2082 }
2083
2084 static void
2085 ztest_io(ztest_ds_t *zd, uint64_t object, uint64_t offset)
2086 {
2087 ztest_block_tag_t wbt;
2088 dmu_object_info_t doi;
2089 enum ztest_io_type io_type;
2090 uint64_t blocksize;
2091 void *data;
2092
2093 VERIFY(dmu_object_info(zd->zd_os, object, &doi) == 0);
2094 blocksize = doi.doi_data_block_size;
2095 data = umem_alloc(blocksize, UMEM_NOFAIL);
2096
2097 /*
2098 * Pick an i/o type at random, biased toward writing block tags.
2099 */
2100 io_type = ztest_random(ZTEST_IO_TYPES);
2101 if (ztest_random(2) == 0)
2102 io_type = ZTEST_IO_WRITE_TAG;
2103
2104 (void) rw_rdlock(&zd->zd_zilog_lock);
2105
2106 switch (io_type) {
2107
2108 case ZTEST_IO_WRITE_TAG:
2109 ztest_bt_generate(&wbt, zd->zd_os, object, offset, 0, 0, 0);
2110 (void) ztest_write(zd, object, offset, sizeof (wbt), &wbt);
2111 break;
2112
2113 case ZTEST_IO_WRITE_PATTERN:
2114 (void) memset(data, 'a' + (object + offset) % 5, blocksize);
2115 if (ztest_random(2) == 0) {
2116 /*
2117 * Induce fletcher2 collisions to ensure that
2118 * zio_ddt_collision() detects and resolves them
2119 * when using fletcher2-verify for deduplication.
2120 */
2121 ((uint64_t *)data)[0] ^= 1ULL << 63;
2122 ((uint64_t *)data)[4] ^= 1ULL << 63;
2123 }
2124 (void) ztest_write(zd, object, offset, blocksize, data);
2125 break;
2126
2127 case ZTEST_IO_WRITE_ZEROES:
2128 bzero(data, blocksize);
2129 (void) ztest_write(zd, object, offset, blocksize, data);
2130 break;
2131
2132 case ZTEST_IO_TRUNCATE:
2133 (void) ztest_truncate(zd, object, offset, blocksize);
2134 break;
2135
2136 case ZTEST_IO_SETATTR:
2137 (void) ztest_setattr(zd, object);
2138 break;
2139 }
2140
2141 (void) rw_unlock(&zd->zd_zilog_lock);
2142
2143 umem_free(data, blocksize);
2144 }
2145
2146 /*
2147 * Initialize an object description template.
2148 */
2149 static void
2150 ztest_od_init(ztest_od_t *od, uint64_t id, char *tag, uint64_t index,
2151 dmu_object_type_t type, uint64_t blocksize, uint64_t gen)
2152 {
2153 od->od_dir = ZTEST_DIROBJ;
2154 od->od_object = 0;
2155
2156 od->od_crtype = type;
2157 od->od_crblocksize = blocksize ? blocksize : ztest_random_blocksize();
2158 od->od_crgen = gen;
2159
2160 od->od_type = DMU_OT_NONE;
2161 od->od_blocksize = 0;
2162 od->od_gen = 0;
2163
2164 (void) snprintf(od->od_name, sizeof (od->od_name), "%s(%lld)[%llu]",
2165 tag, (int64_t)id, index);
2166 }
2167
2168 /*
2169 * Lookup or create the objects for a test using the od template.
2170 * If the objects do not all exist, or if 'remove' is specified,
2171 * remove any existing objects and create new ones. Otherwise,
2172 * use the existing objects.
2173 */
2174 static int
2175 ztest_object_init(ztest_ds_t *zd, ztest_od_t *od, size_t size, boolean_t remove)
2176 {
2177 int count = size / sizeof (*od);
2178 int rv = 0;
2179
2180 VERIFY(mutex_lock(&zd->zd_dirobj_lock) == 0);
2181 if ((ztest_lookup(zd, od, count) != 0 || remove) &&
2182 (ztest_remove(zd, od, count) != 0 ||
2183 ztest_create(zd, od, count) != 0))
2184 rv = -1;
2185 zd->zd_od = od;
2186 VERIFY(mutex_unlock(&zd->zd_dirobj_lock) == 0);
2187
2188 return (rv);
2189 }
2190
2191 /* ARGSUSED */
2192 void
2193 ztest_zil_commit(ztest_ds_t *zd, uint64_t id)
2194 {
2195 zilog_t *zilog = zd->zd_zilog;
2196
2197 (void) rw_rdlock(&zd->zd_zilog_lock);
2198
2199 zil_commit(zilog, ztest_random(ZTEST_OBJECTS));
2200
2201 /*
2202 * Remember the committed values in zd, which is in parent/child
2203 * shared memory. If we die, the next iteration of ztest_run()
2204 * will verify that the log really does contain this record.
2205 */
2206 mutex_enter(&zilog->zl_lock);
2207 ASSERT(zd->zd_shared != NULL);
2208 ASSERT3U(zd->zd_shared->zd_seq, <=, zilog->zl_commit_lr_seq);
2209 zd->zd_shared->zd_seq = zilog->zl_commit_lr_seq;
2210 mutex_exit(&zilog->zl_lock);
2211
2212 (void) rw_unlock(&zd->zd_zilog_lock);
2213 }
2214
2215 /*
2216 * This function is designed to simulate the operations that occur during a
2217 * mount/unmount operation. We hold the dataset across these operations in an
2218 * attempt to expose any implicit assumptions about ZIL management.
2219 */
2220 /* ARGSUSED */
2221 void
2222 ztest_zil_remount(ztest_ds_t *zd, uint64_t id)
2223 {
2224 objset_t *os = zd->zd_os;
2225
2226 (void) rw_wrlock(&zd->zd_zilog_lock);
2227
2228 /* zfsvfs_teardown() */
2229 zil_close(zd->zd_zilog);
2230
2231 /* zfsvfs_setup() */
2232 VERIFY(zil_open(os, ztest_get_data) == zd->zd_zilog);
2233 zil_replay(os, zd, ztest_replay_vector);
2234
2235 (void) rw_unlock(&zd->zd_zilog_lock);
2236 }
2237
2238 /*
2239 * Verify that we can't destroy an active pool, create an existing pool,
2240 * or create a pool with a bad vdev spec.
2241 */
2242 /* ARGSUSED */
2243 void
2244 ztest_spa_create_destroy(ztest_ds_t *zd, uint64_t id)
2245 {
2246 ztest_shared_opts_t *zo = &ztest_opts;
2247 spa_t *spa;
2248 nvlist_t *nvroot;
2249
2250 /*
2251 * Attempt to create using a bad file.
2252 */
2253 nvroot = make_vdev_root("/dev/bogus", NULL, 0, 0, 0, 0, 0, 1);
2254 VERIFY3U(ENOENT, ==,
2255 spa_create("ztest_bad_file", nvroot, NULL, NULL, NULL));
2256 nvlist_free(nvroot);
2257
2258 /*
2259 * Attempt to create using a bad mirror.
2260 */
2261 nvroot = make_vdev_root("/dev/bogus", NULL, 0, 0, 0, 0, 2, 1);
2262 VERIFY3U(ENOENT, ==,
2263 spa_create("ztest_bad_mirror", nvroot, NULL, NULL, NULL));
2264 nvlist_free(nvroot);
2265
2266 /*
2267 * Attempt to create an existing pool. It shouldn't matter
2268 * what's in the nvroot; we should fail with EEXIST.
2269 */
2270 (void) rw_rdlock(&ztest_name_lock);
2271 nvroot = make_vdev_root("/dev/bogus", NULL, 0, 0, 0, 0, 0, 1);
2272 VERIFY3U(EEXIST, ==, spa_create(zo->zo_pool, nvroot, NULL, NULL, NULL));
2273 nvlist_free(nvroot);
2274 VERIFY3U(0, ==, spa_open(zo->zo_pool, &spa, FTAG));
2275 VERIFY3U(EBUSY, ==, spa_destroy(zo->zo_pool));
2276 spa_close(spa, FTAG);
2277
2278 (void) rw_unlock(&ztest_name_lock);
2279 }
2280
2281 static vdev_t *
2282 vdev_lookup_by_path(vdev_t *vd, const char *path)
2283 {
2284 vdev_t *mvd;
2285
2286 if (vd->vdev_path != NULL && strcmp(path, vd->vdev_path) == 0)
2287 return (vd);
2288
2289 for (int c = 0; c < vd->vdev_children; c++)
2290 if ((mvd = vdev_lookup_by_path(vd->vdev_child[c], path)) !=
2291 NULL)
2292 return (mvd);
2293
2294 return (NULL);
2295 }
2296
2297 /*
2298 * Find the first available hole which can be used as a top-level.
2299 */
2300 int
2301 find_vdev_hole(spa_t *spa)
2302 {
2303 vdev_t *rvd = spa->spa_root_vdev;
2304 int c;
2305
2306 ASSERT(spa_config_held(spa, SCL_VDEV, RW_READER) == SCL_VDEV);
2307
2308 for (c = 0; c < rvd->vdev_children; c++) {
2309 vdev_t *cvd = rvd->vdev_child[c];
2310
2311 if (cvd->vdev_ishole)
2312 break;
2313 }
2314 return (c);
2315 }
2316
2317 /*
2318 * Verify that vdev_add() works as expected.
2319 */
2320 /* ARGSUSED */
2321 void
2322 ztest_vdev_add_remove(ztest_ds_t *zd, uint64_t id)
2323 {
2324 ztest_shared_t *zs = ztest_shared;
2325 spa_t *spa = ztest_spa;
2326 uint64_t leaves;
2327 uint64_t guid;
2328 nvlist_t *nvroot;
2329 int error;
2330
2331 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2332 leaves =
2333 MAX(zs->zs_mirrors + zs->zs_splits, 1) * ztest_opts.zo_raidz;
2334
2335 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2336
2337 ztest_shared->zs_vdev_next_leaf = find_vdev_hole(spa) * leaves;
2338
2339 /*
2340 * If we have slogs then remove them 1/4 of the time.
2341 */
2342 if (spa_has_slogs(spa) && ztest_random(4) == 0) {
2343 /*
2344 * Grab the guid from the head of the log class rotor.
2345 */
2346 guid = spa_log_class(spa)->mc_rotor->mg_vd->vdev_guid;
2347
2348 spa_config_exit(spa, SCL_VDEV, FTAG);
2349
2350 /*
2351 * We have to grab the zs_name_lock as writer to
2352 * prevent a race between removing a slog (dmu_objset_find)
2353 * and destroying a dataset. Removing the slog will
2354 * grab a reference on the dataset which may cause
2355 * dmu_objset_destroy() to fail with EBUSY thus
2356 * leaving the dataset in an inconsistent state.
2357 */
2358 VERIFY(rw_wrlock(&ztest_name_lock) == 0);
2359 error = spa_vdev_remove(spa, guid, B_FALSE);
2360 VERIFY(rw_unlock(&ztest_name_lock) == 0);
2361
2362 if (error && error != EEXIST)
2363 fatal(0, "spa_vdev_remove() = %d", error);
2364 } else {
2365 spa_config_exit(spa, SCL_VDEV, FTAG);
2366
2367 /*
2368 * Make 1/4 of the devices be log devices.
2369 */
2370 nvroot = make_vdev_root(NULL, NULL,
2371 ztest_opts.zo_vdev_size, 0,
2372 ztest_random(4) == 0, ztest_opts.zo_raidz,
2373 zs->zs_mirrors, 1);
2374
2375 error = spa_vdev_add(spa, nvroot);
2376 nvlist_free(nvroot);
2377
2378 if (error == ENOSPC)
2379 ztest_record_enospc("spa_vdev_add");
2380 else if (error != 0)
2381 fatal(0, "spa_vdev_add() = %d", error);
2382 }
2383
2384 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2385 }
2386
2387 /*
2388 * Verify that adding/removing aux devices (l2arc, hot spare) works as expected.
2389 */
2390 /* ARGSUSED */
2391 void
2392 ztest_vdev_aux_add_remove(ztest_ds_t *zd, uint64_t id)
2393 {
2394 ztest_shared_t *zs = ztest_shared;
2395 spa_t *spa = ztest_spa;
2396 vdev_t *rvd = spa->spa_root_vdev;
2397 spa_aux_vdev_t *sav;
2398 char *aux;
2399 uint64_t guid = 0;
2400 int error;
2401
2402 if (ztest_random(2) == 0) {
2403 sav = &spa->spa_spares;
2404 aux = ZPOOL_CONFIG_SPARES;
2405 } else {
2406 sav = &spa->spa_l2cache;
2407 aux = ZPOOL_CONFIG_L2CACHE;
2408 }
2409
2410 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2411
2412 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2413
2414 if (sav->sav_count != 0 && ztest_random(4) == 0) {
2415 /*
2416 * Pick a random device to remove.
2417 */
2418 guid = sav->sav_vdevs[ztest_random(sav->sav_count)]->vdev_guid;
2419 } else {
2420 /*
2421 * Find an unused device we can add.
2422 */
2423 zs->zs_vdev_aux = 0;
2424 for (;;) {
2425 char path[MAXPATHLEN];
2426 int c;
2427 (void) snprintf(path, sizeof (path), ztest_aux_template,
2428 ztest_opts.zo_dir, ztest_opts.zo_pool, aux,
2429 zs->zs_vdev_aux);
2430 for (c = 0; c < sav->sav_count; c++)
2431 if (strcmp(sav->sav_vdevs[c]->vdev_path,
2432 path) == 0)
2433 break;
2434 if (c == sav->sav_count &&
2435 vdev_lookup_by_path(rvd, path) == NULL)
2436 break;
2437 zs->zs_vdev_aux++;
2438 }
2439 }
2440
2441 spa_config_exit(spa, SCL_VDEV, FTAG);
2442
2443 if (guid == 0) {
2444 /*
2445 * Add a new device.
2446 */
2447 nvlist_t *nvroot = make_vdev_root(NULL, aux,
2448 (ztest_opts.zo_vdev_size * 5) / 4, 0, 0, 0, 0, 1);
2449 error = spa_vdev_add(spa, nvroot);
2450 if (error != 0)
2451 fatal(0, "spa_vdev_add(%p) = %d", nvroot, error);
2452 nvlist_free(nvroot);
2453 } else {
2454 /*
2455 * Remove an existing device. Sometimes, dirty its
2456 * vdev state first to make sure we handle removal
2457 * of devices that have pending state changes.
2458 */
2459 if (ztest_random(2) == 0)
2460 (void) vdev_online(spa, guid, 0, NULL);
2461
2462 error = spa_vdev_remove(spa, guid, B_FALSE);
2463 if (error != 0 && error != EBUSY)
2464 fatal(0, "spa_vdev_remove(%llu) = %d", guid, error);
2465 }
2466
2467 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2468 }
2469
2470 /*
2471 * split a pool if it has mirror tlvdevs
2472 */
2473 /* ARGSUSED */
2474 void
2475 ztest_split_pool(ztest_ds_t *zd, uint64_t id)
2476 {
2477 ztest_shared_t *zs = ztest_shared;
2478 spa_t *spa = ztest_spa;
2479 vdev_t *rvd = spa->spa_root_vdev;
2480 nvlist_t *tree, **child, *config, *split, **schild;
2481 uint_t c, children, schildren = 0, lastlogid = 0;
2482 int error = 0;
2483
2484 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2485
2486 /* ensure we have a useable config; mirrors of raidz aren't supported */
2487 if (zs->zs_mirrors < 3 || ztest_opts.zo_raidz > 1) {
2488 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2489 return;
2490 }
2491
2492 /* clean up the old pool, if any */
2493 (void) spa_destroy("splitp");
2494
2495 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2496
2497 /* generate a config from the existing config */
2498 mutex_enter(&spa->spa_props_lock);
2499 VERIFY(nvlist_lookup_nvlist(spa->spa_config, ZPOOL_CONFIG_VDEV_TREE,
2500 &tree) == 0);
2501 mutex_exit(&spa->spa_props_lock);
2502
2503 VERIFY(nvlist_lookup_nvlist_array(tree, ZPOOL_CONFIG_CHILDREN, &child,
2504 &children) == 0);
2505
2506 schild = malloc(rvd->vdev_children * sizeof (nvlist_t *));
2507 for (c = 0; c < children; c++) {
2508 vdev_t *tvd = rvd->vdev_child[c];
2509 nvlist_t **mchild;
2510 uint_t mchildren;
2511
2512 if (tvd->vdev_islog || tvd->vdev_ops == &vdev_hole_ops) {
2513 VERIFY(nvlist_alloc(&schild[schildren], NV_UNIQUE_NAME,
2514 0) == 0);
2515 VERIFY(nvlist_add_string(schild[schildren],
2516 ZPOOL_CONFIG_TYPE, VDEV_TYPE_HOLE) == 0);
2517 VERIFY(nvlist_add_uint64(schild[schildren],
2518 ZPOOL_CONFIG_IS_HOLE, 1) == 0);
2519 if (lastlogid == 0)
2520 lastlogid = schildren;
2521 ++schildren;
2522 continue;
2523 }
2524 lastlogid = 0;
2525 VERIFY(nvlist_lookup_nvlist_array(child[c],
2526 ZPOOL_CONFIG_CHILDREN, &mchild, &mchildren) == 0);
2527 VERIFY(nvlist_dup(mchild[0], &schild[schildren++], 0) == 0);
2528 }
2529
2530 /* OK, create a config that can be used to split */
2531 VERIFY(nvlist_alloc(&split, NV_UNIQUE_NAME, 0) == 0);
2532 VERIFY(nvlist_add_string(split, ZPOOL_CONFIG_TYPE,
2533 VDEV_TYPE_ROOT) == 0);
2534 VERIFY(nvlist_add_nvlist_array(split, ZPOOL_CONFIG_CHILDREN, schild,
2535 lastlogid != 0 ? lastlogid : schildren) == 0);
2536
2537 VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, 0) == 0);
2538 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, split) == 0);
2539
2540 for (c = 0; c < schildren; c++)
2541 nvlist_free(schild[c]);
2542 free(schild);
2543 nvlist_free(split);
2544
2545 spa_config_exit(spa, SCL_VDEV, FTAG);
2546
2547 (void) rw_wrlock(&ztest_name_lock);
2548 error = spa_vdev_split_mirror(spa, "splitp", config, NULL, B_FALSE);
2549 (void) rw_unlock(&ztest_name_lock);
2550
2551 nvlist_free(config);
2552
2553 if (error == 0) {
2554 (void) printf("successful split - results:\n");
2555 mutex_enter(&spa_namespace_lock);
2556 show_pool_stats(spa);
2557 show_pool_stats(spa_lookup("splitp"));
2558 mutex_exit(&spa_namespace_lock);
2559 ++zs->zs_splits;
2560 --zs->zs_mirrors;
2561 }
2562 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2563
2564 }
2565
2566 /*
2567 * Verify that we can attach and detach devices.
2568 */
2569 /* ARGSUSED */
2570 void
2571 ztest_vdev_attach_detach(ztest_ds_t *zd, uint64_t id)
2572 {
2573 ztest_shared_t *zs = ztest_shared;
2574 spa_t *spa = ztest_spa;
2575 spa_aux_vdev_t *sav = &spa->spa_spares;
2576 vdev_t *rvd = spa->spa_root_vdev;
2577 vdev_t *oldvd, *newvd, *pvd;
2578 nvlist_t *root;
2579 uint64_t leaves;
2580 uint64_t leaf, top;
2581 uint64_t ashift = ztest_get_ashift();
2582 uint64_t oldguid, pguid;
2583 size_t oldsize, newsize;
2584 char oldpath[MAXPATHLEN], newpath[MAXPATHLEN];
2585 int replacing;
2586 int oldvd_has_siblings = B_FALSE;
2587 int newvd_is_spare = B_FALSE;
2588 int oldvd_is_log;
2589 int error, expected_error;
2590
2591 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2592 leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz;
2593
2594 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2595
2596 /*
2597 * Decide whether to do an attach or a replace.
2598 */
2599 replacing = ztest_random(2);
2600
2601 /*
2602 * Pick a random top-level vdev.
2603 */
2604 top = ztest_random_vdev_top(spa, B_TRUE);
2605
2606 /*
2607 * Pick a random leaf within it.
2608 */
2609 leaf = ztest_random(leaves);
2610
2611 /*
2612 * Locate this vdev.
2613 */
2614 oldvd = rvd->vdev_child[top];
2615 if (zs->zs_mirrors >= 1) {
2616 ASSERT(oldvd->vdev_ops == &vdev_mirror_ops);
2617 ASSERT(oldvd->vdev_children >= zs->zs_mirrors);
2618 oldvd = oldvd->vdev_child[leaf / ztest_opts.zo_raidz];
2619 }
2620 if (ztest_opts.zo_raidz > 1) {
2621 ASSERT(oldvd->vdev_ops == &vdev_raidz_ops);
2622 ASSERT(oldvd->vdev_children == ztest_opts.zo_raidz);
2623 oldvd = oldvd->vdev_child[leaf % ztest_opts.zo_raidz];
2624 }
2625
2626 /*
2627 * If we're already doing an attach or replace, oldvd may be a
2628 * mirror vdev -- in which case, pick a random child.
2629 */
2630 while (oldvd->vdev_children != 0) {
2631 oldvd_has_siblings = B_TRUE;
2632 ASSERT(oldvd->vdev_children >= 2);
2633 oldvd = oldvd->vdev_child[ztest_random(oldvd->vdev_children)];
2634 }
2635
2636 oldguid = oldvd->vdev_guid;
2637 oldsize = vdev_get_min_asize(oldvd);
2638 oldvd_is_log = oldvd->vdev_top->vdev_islog;
2639 (void) strcpy(oldpath, oldvd->vdev_path);
2640 pvd = oldvd->vdev_parent;
2641 pguid = pvd->vdev_guid;
2642
2643 /*
2644 * If oldvd has siblings, then half of the time, detach it.
2645 */
2646 if (oldvd_has_siblings && ztest_random(2) == 0) {
2647 spa_config_exit(spa, SCL_VDEV, FTAG);
2648 error = spa_vdev_detach(spa, oldguid, pguid, B_FALSE);
2649 if (error != 0 && error != ENODEV && error != EBUSY &&
2650 error != ENOTSUP)
2651 fatal(0, "detach (%s) returned %d", oldpath, error);
2652 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2653 return;
2654 }
2655
2656 /*
2657 * For the new vdev, choose with equal probability between the two
2658 * standard paths (ending in either 'a' or 'b') or a random hot spare.
2659 */
2660 if (sav->sav_count != 0 && ztest_random(3) == 0) {
2661 newvd = sav->sav_vdevs[ztest_random(sav->sav_count)];
2662 newvd_is_spare = B_TRUE;
2663 (void) strcpy(newpath, newvd->vdev_path);
2664 } else {
2665 (void) snprintf(newpath, sizeof (newpath), ztest_dev_template,
2666 ztest_opts.zo_dir, ztest_opts.zo_pool,
2667 top * leaves + leaf);
2668 if (ztest_random(2) == 0)
2669 newpath[strlen(newpath) - 1] = 'b';
2670 newvd = vdev_lookup_by_path(rvd, newpath);
2671 }
2672
2673 if (newvd) {
2674 newsize = vdev_get_min_asize(newvd);
2675 } else {
2676 /*
2677 * Make newsize a little bigger or smaller than oldsize.
2678 * If it's smaller, the attach should fail.
2679 * If it's larger, and we're doing a replace,
2680 * we should get dynamic LUN growth when we're done.
2681 */
2682 newsize = 10 * oldsize / (9 + ztest_random(3));
2683 }
2684
2685 /*
2686 * If pvd is not a mirror or root, the attach should fail with ENOTSUP,
2687 * unless it's a replace; in that case any non-replacing parent is OK.
2688 *
2689 * If newvd is already part of the pool, it should fail with EBUSY.
2690 *
2691 * If newvd is too small, it should fail with EOVERFLOW.
2692 */
2693 if (pvd->vdev_ops != &vdev_mirror_ops &&
2694 pvd->vdev_ops != &vdev_root_ops && (!replacing ||
2695 pvd->vdev_ops == &vdev_replacing_ops ||
2696 pvd->vdev_ops == &vdev_spare_ops))
2697 expected_error = ENOTSUP;
2698 else if (newvd_is_spare && (!replacing || oldvd_is_log))
2699 expected_error = ENOTSUP;
2700 else if (newvd == oldvd)
2701 expected_error = replacing ? 0 : EBUSY;
2702 else if (vdev_lookup_by_path(rvd, newpath) != NULL)
2703 expected_error = EBUSY;
2704 else if (newsize < oldsize)
2705 expected_error = EOVERFLOW;
2706 else if (ashift > oldvd->vdev_top->vdev_ashift)
2707 expected_error = EDOM;
2708 else
2709 expected_error = 0;
2710
2711 spa_config_exit(spa, SCL_VDEV, FTAG);
2712
2713 /*
2714 * Build the nvlist describing newpath.
2715 */
2716 root = make_vdev_root(newpath, NULL, newvd == NULL ? newsize : 0,
2717 ashift, 0, 0, 0, 1);
2718
2719 error = spa_vdev_attach(spa, oldguid, root, replacing);
2720
2721 nvlist_free(root);
2722
2723 /*
2724 * If our parent was the replacing vdev, but the replace completed,
2725 * then instead of failing with ENOTSUP we may either succeed,
2726 * fail with ENODEV, or fail with EOVERFLOW.
2727 */
2728 if (expected_error == ENOTSUP &&
2729 (error == 0 || error == ENODEV || error == EOVERFLOW))
2730 expected_error = error;
2731
2732 /*
2733 * If someone grew the LUN, the replacement may be too small.
2734 */
2735 if (error == EOVERFLOW || error == EBUSY)
2736 expected_error = error;
2737
2738 /* XXX workaround 6690467 */
2739 if (error != expected_error && expected_error != EBUSY) {
2740 fatal(0, "attach (%s %llu, %s %llu, %d) "
2741 "returned %d, expected %d",
2742 oldpath, (longlong_t)oldsize, newpath,
2743 (longlong_t)newsize, replacing, error, expected_error);
2744 }
2745
2746 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2747 }
2748
2749 /*
2750 * Callback function which expands the physical size of the vdev.
2751 */
2752 vdev_t *
2753 grow_vdev(vdev_t *vd, void *arg)
2754 {
2755 spa_t *spa = vd->vdev_spa;
2756 size_t *newsize = arg;
2757 size_t fsize;
2758 int fd;
2759
2760 ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
2761 ASSERT(vd->vdev_ops->vdev_op_leaf);
2762
2763 if ((fd = open(vd->vdev_path, O_RDWR)) == -1)
2764 return (vd);
2765
2766 fsize = lseek(fd, 0, SEEK_END);
2767 (void) ftruncate(fd, *newsize);
2768
2769 if (ztest_opts.zo_verbose >= 6) {
2770 (void) printf("%s grew from %lu to %lu bytes\n",
2771 vd->vdev_path, (ulong_t)fsize, (ulong_t)*newsize);
2772 }
2773 (void) close(fd);
2774 return (NULL);
2775 }
2776
2777 /*
2778 * Callback function which expands a given vdev by calling vdev_online().
2779 */
2780 /* ARGSUSED */
2781 vdev_t *
2782 online_vdev(vdev_t *vd, void *arg)
2783 {
2784 spa_t *spa = vd->vdev_spa;
2785 vdev_t *tvd = vd->vdev_top;
2786 uint64_t guid = vd->vdev_guid;
2787 uint64_t generation = spa->spa_config_generation + 1;
2788 vdev_state_t newstate = VDEV_STATE_UNKNOWN;
2789 int error;
2790
2791 ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
2792 ASSERT(vd->vdev_ops->vdev_op_leaf);
2793
2794 /* Calling vdev_online will initialize the new metaslabs */
2795 spa_config_exit(spa, SCL_STATE, spa);
2796 error = vdev_online(spa, guid, ZFS_ONLINE_EXPAND, &newstate);
2797 spa_config_enter(spa, SCL_STATE, spa, RW_READER);
2798
2799 /*
2800 * If vdev_online returned an error or the underlying vdev_open
2801 * failed then we abort the expand. The only way to know that
2802 * vdev_open fails is by checking the returned newstate.
2803 */
2804 if (error || newstate != VDEV_STATE_HEALTHY) {
2805 if (ztest_opts.zo_verbose >= 5) {
2806 (void) printf("Unable to expand vdev, state %llu, "
2807 "error %d\n", (u_longlong_t)newstate, error);
2808 }
2809 return (vd);
2810 }
2811 ASSERT3U(newstate, ==, VDEV_STATE_HEALTHY);
2812
2813 /*
2814 * Since we dropped the lock we need to ensure that we're
2815 * still talking to the original vdev. It's possible this
2816 * vdev may have been detached/replaced while we were
2817 * trying to online it.
2818 */
2819 if (generation != spa->spa_config_generation) {
2820 if (ztest_opts.zo_verbose >= 5) {
2821 (void) printf("vdev configuration has changed, "
2822 "guid %llu, state %llu, expected gen %llu, "
2823 "got gen %llu\n",
2824 (u_longlong_t)guid,
2825 (u_longlong_t)tvd->vdev_state,
2826 (u_longlong_t)generation,
2827 (u_longlong_t)spa->spa_config_generation);
2828 }
2829 return (vd);
2830 }
2831 return (NULL);
2832 }
2833
2834 /*
2835 * Traverse the vdev tree calling the supplied function.
2836 * We continue to walk the tree until we either have walked all
2837 * children or we receive a non-NULL return from the callback.
2838 * If a NULL callback is passed, then we just return back the first
2839 * leaf vdev we encounter.
2840 */
2841 vdev_t *
2842 vdev_walk_tree(vdev_t *vd, vdev_t *(*func)(vdev_t *, void *), void *arg)
2843 {
2844 if (vd->vdev_ops->vdev_op_leaf) {
2845 if (func == NULL)
2846 return (vd);
2847 else
2848 return (func(vd, arg));
2849 }
2850
2851 for (uint_t c = 0; c < vd->vdev_children; c++) {
2852 vdev_t *cvd = vd->vdev_child[c];
2853 if ((cvd = vdev_walk_tree(cvd, func, arg)) != NULL)
2854 return (cvd);
2855 }
2856 return (NULL);
2857 }
2858
2859 /*
2860 * Verify that dynamic LUN growth works as expected.
2861 */
2862 /* ARGSUSED */
2863 void
2864 ztest_vdev_LUN_growth(ztest_ds_t *zd, uint64_t id)
2865 {
2866 spa_t *spa = ztest_spa;
2867 vdev_t *vd, *tvd;
2868 metaslab_class_t *mc;
2869 metaslab_group_t *mg;
2870 size_t psize, newsize;
2871 uint64_t top;
2872 uint64_t old_class_space, new_class_space, old_ms_count, new_ms_count;
2873
2874 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2875 spa_config_enter(spa, SCL_STATE, spa, RW_READER);
2876
2877 top = ztest_random_vdev_top(spa, B_TRUE);
2878
2879 tvd = spa->spa_root_vdev->vdev_child[top];
2880 mg = tvd->vdev_mg;
2881 mc = mg->mg_class;
2882 old_ms_count = tvd->vdev_ms_count;
2883 old_class_space = metaslab_class_get_space(mc);
2884
2885 /*
2886 * Determine the size of the first leaf vdev associated with
2887 * our top-level device.
2888 */
2889 vd = vdev_walk_tree(tvd, NULL, NULL);
2890 ASSERT3P(vd, !=, NULL);
2891 ASSERT(vd->vdev_ops->vdev_op_leaf);
2892
2893 psize = vd->vdev_psize;
2894
2895 /*
2896 * We only try to expand the vdev if it's healthy, less than 4x its
2897 * original size, and it has a valid psize.
2898 */
2899 if (tvd->vdev_state != VDEV_STATE_HEALTHY ||
2900 psize == 0 || psize >= 4 * ztest_opts.zo_vdev_size) {
2901 spa_config_exit(spa, SCL_STATE, spa);
2902 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2903 return;
2904 }
2905 ASSERT(psize > 0);
2906 newsize = psize + psize / 8;
2907 ASSERT3U(newsize, >, psize);
2908
2909 if (ztest_opts.zo_verbose >= 6) {
2910 (void) printf("Expanding LUN %s from %lu to %lu\n",
2911 vd->vdev_path, (ulong_t)psize, (ulong_t)newsize);
2912 }
2913
2914 /*
2915 * Growing the vdev is a two step process:
2916 * 1). expand the physical size (i.e. relabel)
2917 * 2). online the vdev to create the new metaslabs
2918 */
2919 if (vdev_walk_tree(tvd, grow_vdev, &newsize) != NULL ||
2920 vdev_walk_tree(tvd, online_vdev, NULL) != NULL ||
2921 tvd->vdev_state != VDEV_STATE_HEALTHY) {
2922 if (ztest_opts.zo_verbose >= 5) {
2923 (void) printf("Could not expand LUN because "
2924 "the vdev configuration changed.\n");
2925 }
2926 spa_config_exit(spa, SCL_STATE, spa);
2927 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2928 return;
2929 }
2930
2931 spa_config_exit(spa, SCL_STATE, spa);
2932
2933 /*
2934 * Expanding the LUN will update the config asynchronously,
2935 * thus we must wait for the async thread to complete any
2936 * pending tasks before proceeding.
2937 */
2938 for (;;) {
2939 boolean_t done;
2940 mutex_enter(&spa->spa_async_lock);
2941 done = (spa->spa_async_thread == NULL && !spa->spa_async_tasks);
2942 mutex_exit(&spa->spa_async_lock);
2943 if (done)
2944 break;
2945 txg_wait_synced(spa_get_dsl(spa), 0);
2946 (void) poll(NULL, 0, 100);
2947 }
2948
2949 spa_config_enter(spa, SCL_STATE, spa, RW_READER);
2950
2951 tvd = spa->spa_root_vdev->vdev_child[top];
2952 new_ms_count = tvd->vdev_ms_count;
2953 new_class_space = metaslab_class_get_space(mc);
2954
2955 if (tvd->vdev_mg != mg || mg->mg_class != mc) {
2956 if (ztest_opts.zo_verbose >= 5) {
2957 (void) printf("Could not verify LUN expansion due to "
2958 "intervening vdev offline or remove.\n");
2959 }
2960 spa_config_exit(spa, SCL_STATE, spa);
2961 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2962 return;
2963 }
2964
2965 /*
2966 * Make sure we were able to grow the vdev.
2967 */
2968 if (new_ms_count <= old_ms_count)
2969 fatal(0, "LUN expansion failed: ms_count %llu <= %llu\n",
2970 old_ms_count, new_ms_count);
2971
2972 /*
2973 * Make sure we were able to grow the pool.
2974 */
2975 if (new_class_space <= old_class_space)
2976 fatal(0, "LUN expansion failed: class_space %llu <= %llu\n",
2977 old_class_space, new_class_space);
2978
2979 if (ztest_opts.zo_verbose >= 5) {
2980 char oldnumbuf[6], newnumbuf[6];
2981
2982 nicenum(old_class_space, oldnumbuf);
2983 nicenum(new_class_space, newnumbuf);
2984 (void) printf("%s grew from %s to %s\n",
2985 spa->spa_name, oldnumbuf, newnumbuf);
2986 }
2987
2988 spa_config_exit(spa, SCL_STATE, spa);
2989 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2990 }
2991
2992 /*
2993 * Verify that dmu_objset_{create,destroy,open,close} work as expected.
2994 */
2995 /* ARGSUSED */
2996 static void
2997 ztest_objset_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
2998 {
2999 /*
3000 * Create the objects common to all ztest datasets.
3001 */
3002 VERIFY(zap_create_claim(os, ZTEST_DIROBJ,
3003 DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx) == 0);
3004 }
3005
3006 static int
3007 ztest_dataset_create(char *dsname)
3008 {
3009 uint64_t zilset = ztest_random(100);
3010 int err = dmu_objset_create(dsname, DMU_OST_OTHER, 0,
3011 ztest_objset_create_cb, NULL);
3012
3013 if (err || zilset < 80)
3014 return (err);
3015
3016 if (ztest_opts.zo_verbose >= 6)
3017 (void) printf("Setting dataset %s to sync always\n", dsname);
3018 return (ztest_dsl_prop_set_uint64(dsname, ZFS_PROP_SYNC,
3019 ZFS_SYNC_ALWAYS, B_FALSE));
3020 }
3021
3022 /* ARGSUSED */
3023 static int
3024 ztest_objset_destroy_cb(const char *name, void *arg)
3025 {
3026 objset_t *os;
3027 dmu_object_info_t doi;
3028 int error;
3029
3030 /*
3031 * Verify that the dataset contains a directory object.
3032 */
3033 VERIFY3U(0, ==, dmu_objset_hold(name, FTAG, &os));
3034 error = dmu_object_info(os, ZTEST_DIROBJ, &doi);
3035 if (error != ENOENT) {
3036 /* We could have crashed in the middle of destroying it */
3037 ASSERT3U(error, ==, 0);
3038 ASSERT3U(doi.doi_type, ==, DMU_OT_ZAP_OTHER);
3039 ASSERT3S(doi.doi_physical_blocks_512, >=, 0);
3040 }
3041 dmu_objset_rele(os, FTAG);
3042
3043 /*
3044 * Destroy the dataset.
3045 */
3046 VERIFY3U(0, ==, dmu_objset_destroy(name, B_FALSE));
3047 return (0);
3048 }
3049
3050 static boolean_t
3051 ztest_snapshot_create(char *osname, uint64_t id)
3052 {
3053 char snapname[MAXNAMELEN];
3054 int error;
3055
3056 (void) snprintf(snapname, MAXNAMELEN, "%s@%llu", osname,
3057 (u_longlong_t)id);
3058
3059 error = dmu_objset_snapshot(osname, strchr(snapname, '@') + 1,
3060 NULL, NULL, B_FALSE, B_FALSE, -1);
3061 if (error == ENOSPC) {
3062 ztest_record_enospc(FTAG);
3063 return (B_FALSE);
3064 }
3065 if (error != 0 && error != EEXIST)
3066 fatal(0, "ztest_snapshot_create(%s) = %d", snapname, error);
3067 return (B_TRUE);
3068 }
3069
3070 static boolean_t
3071 ztest_snapshot_destroy(char *osname, uint64_t id)
3072 {
3073 char snapname[MAXNAMELEN];
3074 int error;
3075
3076 (void) snprintf(snapname, MAXNAMELEN, "%s@%llu", osname,
3077 (u_longlong_t)id);
3078
3079 error = dmu_objset_destroy(snapname, B_FALSE);
3080 if (error != 0 && error != ENOENT)
3081 fatal(0, "ztest_snapshot_destroy(%s) = %d", snapname, error);
3082 return (B_TRUE);
3083 }
3084
3085 /* ARGSUSED */
3086 void
3087 ztest_dmu_objset_create_destroy(ztest_ds_t *zd, uint64_t id)
3088 {
3089 ztest_ds_t zdtmp;
3090 int iters;
3091 int error;
3092 objset_t *os, *os2;
3093 char name[MAXNAMELEN];
3094 zilog_t *zilog;
3095
3096 (void) rw_rdlock(&ztest_name_lock);
3097
3098 (void) snprintf(name, MAXNAMELEN, "%s/temp_%llu",
3099 ztest_opts.zo_pool, (u_longlong_t)id);
3100
3101 /*
3102 * If this dataset exists from a previous run, process its replay log
3103 * half of the time. If we don't replay it, then dmu_objset_destroy()
3104 * (invoked from ztest_objset_destroy_cb()) should just throw it away.
3105 */
3106 if (ztest_random(2) == 0 &&
3107 dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os) == 0) {
3108 ztest_zd_init(&zdtmp, NULL, os);
3109 zil_replay(os, &zdtmp, ztest_replay_vector);
3110 ztest_zd_fini(&zdtmp);
3111 dmu_objset_disown(os, FTAG);
3112 }
3113
3114 /*
3115 * There may be an old instance of the dataset we're about to
3116 * create lying around from a previous run. If so, destroy it
3117 * and all of its snapshots.
3118 */
3119 (void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
3120 DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
3121
3122 /*
3123 * Verify that the destroyed dataset is no longer in the namespace.
3124 */
3125 VERIFY3U(ENOENT, ==, dmu_objset_hold(name, FTAG, &os));
3126
3127 /*
3128 * Verify that we can create a new dataset.
3129 */
3130 error = ztest_dataset_create(name);
3131 if (error) {
3132 if (error == ENOSPC) {
3133 ztest_record_enospc(FTAG);
3134 (void) rw_unlock(&ztest_name_lock);
3135 return;
3136 }
3137 fatal(0, "dmu_objset_create(%s) = %d", name, error);
3138 }
3139
3140 VERIFY3U(0, ==,
3141 dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os));
3142
3143 ztest_zd_init(&zdtmp, NULL, os);
3144
3145 /*
3146 * Open the intent log for it.
3147 */
3148 zilog = zil_open(os, ztest_get_data);
3149
3150 /*
3151 * Put some objects in there, do a little I/O to them,
3152 * and randomly take a couple of snapshots along the way.
3153 */
3154 iters = ztest_random(5);
3155 for (int i = 0; i < iters; i++) {
3156 ztest_dmu_object_alloc_free(&zdtmp, id);
3157 if (ztest_random(iters) == 0)
3158 (void) ztest_snapshot_create(name, i);
3159 }
3160
3161 /*
3162 * Verify that we cannot create an existing dataset.
3163 */
3164 VERIFY3U(EEXIST, ==,
3165 dmu_objset_create(name, DMU_OST_OTHER, 0, NULL, NULL));
3166
3167 /*
3168 * Verify that we can hold an objset that is also owned.
3169 */
3170 VERIFY3U(0, ==, dmu_objset_hold(name, FTAG, &os2));
3171 dmu_objset_rele(os2, FTAG);
3172
3173 /*
3174 * Verify that we cannot own an objset that is already owned.
3175 */
3176 VERIFY3U(EBUSY, ==,
3177 dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os2));
3178
3179 zil_close(zilog);
3180 dmu_objset_disown(os, FTAG);
3181 ztest_zd_fini(&zdtmp);
3182
3183 (void) rw_unlock(&ztest_name_lock);
3184 }
3185
3186 /*
3187 * Verify that dmu_snapshot_{create,destroy,open,close} work as expected.
3188 */
3189 void
3190 ztest_dmu_snapshot_create_destroy(ztest_ds_t *zd, uint64_t id)
3191 {
3192 (void) rw_rdlock(&ztest_name_lock);
3193 (void) ztest_snapshot_destroy(zd->zd_name, id);
3194 (void) ztest_snapshot_create(zd->zd_name, id);
3195 (void) rw_unlock(&ztest_name_lock);
3196 }
3197
3198 /*
3199 * Cleanup non-standard snapshots and clones.
3200 */
3201 void
3202 ztest_dsl_dataset_cleanup(char *osname, uint64_t id)
3203 {
3204 char snap1name[MAXNAMELEN];
3205 char clone1name[MAXNAMELEN];
3206 char snap2name[MAXNAMELEN];
3207 char clone2name[MAXNAMELEN];
3208 char snap3name[MAXNAMELEN];
3209 int error;
3210
3211 (void) snprintf(snap1name, MAXNAMELEN, "%s@s1_%llu", osname, id);
3212 (void) snprintf(clone1name, MAXNAMELEN, "%s/c1_%llu", osname, id);
3213 (void) snprintf(snap2name, MAXNAMELEN, "%s@s2_%llu", clone1name, id);
3214 (void) snprintf(clone2name, MAXNAMELEN, "%s/c2_%llu", osname, id);
3215 (void) snprintf(snap3name, MAXNAMELEN, "%s@s3_%llu", clone1name, id);
3216
3217 error = dmu_objset_destroy(clone2name, B_FALSE);
3218 if (error && error != ENOENT)
3219 fatal(0, "dmu_objset_destroy(%s) = %d", clone2name, error);
3220 error = dmu_objset_destroy(snap3name, B_FALSE);
3221 if (error && error != ENOENT)
3222 fatal(0, "dmu_objset_destroy(%s) = %d", snap3name, error);
3223 error = dmu_objset_destroy(snap2name, B_FALSE);
3224 if (error && error != ENOENT)
3225 fatal(0, "dmu_objset_destroy(%s) = %d", snap2name, error);
3226 error = dmu_objset_destroy(clone1name, B_FALSE);
3227 if (error && error != ENOENT)
3228 fatal(0, "dmu_objset_destroy(%s) = %d", clone1name, error);
3229 error = dmu_objset_destroy(snap1name, B_FALSE);
3230 if (error && error != ENOENT)
3231 fatal(0, "dmu_objset_destroy(%s) = %d", snap1name, error);
3232 }
3233
3234 /*
3235 * Verify dsl_dataset_promote handles EBUSY
3236 */
3237 void
3238 ztest_dsl_dataset_promote_busy(ztest_ds_t *zd, uint64_t id)
3239 {
3240 objset_t *clone;
3241 dsl_dataset_t *ds;
3242 char snap1name[MAXNAMELEN];
3243 char clone1name[MAXNAMELEN];
3244 char snap2name[MAXNAMELEN];
3245 char clone2name[MAXNAMELEN];
3246 char snap3name[MAXNAMELEN];
3247 char *osname = zd->zd_name;
3248 int error;
3249
3250 (void) rw_rdlock(&ztest_name_lock);
3251
3252 ztest_dsl_dataset_cleanup(osname, id);
3253
3254 (void) snprintf(snap1name, MAXNAMELEN, "%s@s1_%llu", osname, id);
3255 (void) snprintf(clone1name, MAXNAMELEN, "%s/c1_%llu", osname, id);
3256 (void) snprintf(snap2name, MAXNAMELEN, "%s@s2_%llu", clone1name, id);
3257 (void) snprintf(clone2name, MAXNAMELEN, "%s/c2_%llu", osname, id);
3258 (void) snprintf(snap3name, MAXNAMELEN, "%s@s3_%llu", clone1name, id);
3259
3260 error = dmu_objset_snapshot(osname, strchr(snap1name, '@')+1,
3261 NULL, NULL, B_FALSE, B_FALSE, -1);
3262 if (error && error != EEXIST) {
3263 if (error == ENOSPC) {
3264 ztest_record_enospc(FTAG);
3265 goto out;
3266 }
3267 fatal(0, "dmu_take_snapshot(%s) = %d", snap1name, error);
3268 }
3269
3270 error = dmu_objset_hold(snap1name, FTAG, &clone);
3271 if (error)
3272 fatal(0, "dmu_open_snapshot(%s) = %d", snap1name, error);
3273
3274 error = dmu_objset_clone(clone1name, dmu_objset_ds(clone), 0);
3275 dmu_objset_rele(clone, FTAG);
3276 if (error) {
3277 if (error == ENOSPC) {
3278 ztest_record_enospc(FTAG);
3279 goto out;
3280 }
3281 fatal(0, "dmu_objset_create(%s) = %d", clone1name, error);
3282 }
3283
3284 error = dmu_objset_snapshot(clone1name, strchr(snap2name, '@')+1,
3285 NULL, NULL, B_FALSE, B_FALSE, -1);
3286 if (error && error != EEXIST) {
3287 if (error == ENOSPC) {
3288 ztest_record_enospc(FTAG);
3289 goto out;
3290 }
3291 fatal(0, "dmu_open_snapshot(%s) = %d", snap2name, error);
3292 }
3293
3294 error = dmu_objset_snapshot(clone1name, strchr(snap3name, '@')+1,
3295 NULL, NULL, B_FALSE, B_FALSE, -1);
3296 if (error && error != EEXIST) {
3297 if (error == ENOSPC) {
3298 ztest_record_enospc(FTAG);
3299 goto out;
3300 }
3301 fatal(0, "dmu_open_snapshot(%s) = %d", snap3name, error);
3302 }
3303
3304 error = dmu_objset_hold(snap3name, FTAG, &clone);
3305 if (error)
3306 fatal(0, "dmu_open_snapshot(%s) = %d", snap3name, error);
3307
3308 error = dmu_objset_clone(clone2name, dmu_objset_ds(clone), 0);
3309 dmu_objset_rele(clone, FTAG);
3310 if (error) {
3311 if (error == ENOSPC) {
3312 ztest_record_enospc(FTAG);
3313 goto out;
3314 }
3315 fatal(0, "dmu_objset_create(%s) = %d", clone2name, error);
3316 }
3317
3318 error = dsl_dataset_own(snap2name, B_FALSE, FTAG, &ds);
3319 if (error)
3320 fatal(0, "dsl_dataset_own(%s) = %d", snap2name, error);
3321 error = dsl_dataset_promote(clone2name, NULL);
3322 if (error != EBUSY)
3323 fatal(0, "dsl_dataset_promote(%s), %d, not EBUSY", clone2name,
3324 error);
3325 dsl_dataset_disown(ds, FTAG);
3326
3327 out:
3328 ztest_dsl_dataset_cleanup(osname, id);
3329
3330 (void) rw_unlock(&ztest_name_lock);
3331 }
3332
3333 /*
3334 * Verify that dmu_object_{alloc,free} work as expected.
3335 */
3336 void
3337 ztest_dmu_object_alloc_free(ztest_ds_t *zd, uint64_t id)
3338 {
3339 ztest_od_t od[4];
3340 int batchsize = sizeof (od) / sizeof (od[0]);
3341
3342 for (int b = 0; b < batchsize; b++)
3343 ztest_od_init(&od[b], id, FTAG, b, DMU_OT_UINT64_OTHER, 0, 0);
3344
3345 /*
3346 * Destroy the previous batch of objects, create a new batch,
3347 * and do some I/O on the new objects.
3348 */
3349 if (ztest_object_init(zd, od, sizeof (od), B_TRUE) != 0)
3350 return;
3351
3352 while (ztest_random(4 * batchsize) != 0)
3353 ztest_io(zd, od[ztest_random(batchsize)].od_object,
3354 ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
3355 }
3356
3357 /*
3358 * Verify that dmu_{read,write} work as expected.
3359 */
3360 void
3361 ztest_dmu_read_write(ztest_ds_t *zd, uint64_t id)
3362 {
3363 objset_t *os = zd->zd_os;
3364 ztest_od_t od[2];
3365 dmu_tx_t *tx;
3366 int i, freeit, error;
3367 uint64_t n, s, txg;
3368 bufwad_t *packbuf, *bigbuf, *pack, *bigH, *bigT;
3369 uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
3370 uint64_t chunksize = (1000 + ztest_random(1000)) * sizeof (uint64_t);
3371 uint64_t regions = 997;
3372 uint64_t stride = 123456789ULL;
3373 uint64_t width = 40;
3374 int free_percent = 5;
3375
3376 /*
3377 * This test uses two objects, packobj and bigobj, that are always
3378 * updated together (i.e. in the same tx) so that their contents are
3379 * in sync and can be compared. Their contents relate to each other
3380 * in a simple way: packobj is a dense array of 'bufwad' structures,
3381 * while bigobj is a sparse array of the same bufwads. Specifically,
3382 * for any index n, there are three bufwads that should be identical:
3383 *
3384 * packobj, at offset n * sizeof (bufwad_t)
3385 * bigobj, at the head of the nth chunk
3386 * bigobj, at the tail of the nth chunk
3387 *
3388 * The chunk size is arbitrary. It doesn't have to be a power of two,
3389 * and it doesn't have any relation to the object blocksize.
3390 * The only requirement is that it can hold at least two bufwads.
3391 *
3392 * Normally, we write the bufwad to each of these locations.
3393 * However, free_percent of the time we instead write zeroes to
3394 * packobj and perform a dmu_free_range() on bigobj. By comparing
3395 * bigobj to packobj, we can verify that the DMU is correctly
3396 * tracking which parts of an object are allocated and free,
3397 * and that the contents of the allocated blocks are correct.
3398 */
3399
3400 /*
3401 * Read the directory info. If it's the first time, set things up.
3402 */
3403 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, chunksize);
3404 ztest_od_init(&od[1], id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize);
3405
3406 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
3407 return;
3408
3409 bigobj = od[0].od_object;
3410 packobj = od[1].od_object;
3411 chunksize = od[0].od_gen;
3412 ASSERT(chunksize == od[1].od_gen);
3413
3414 /*
3415 * Prefetch a random chunk of the big object.
3416 * Our aim here is to get some async reads in flight
3417 * for blocks that we may free below; the DMU should
3418 * handle this race correctly.
3419 */
3420 n = ztest_random(regions) * stride + ztest_random(width);
3421 s = 1 + ztest_random(2 * width - 1);
3422 dmu_prefetch(os, bigobj, n * chunksize, s * chunksize);
3423
3424 /*
3425 * Pick a random index and compute the offsets into packobj and bigobj.
3426 */
3427 n = ztest_random(regions) * stride + ztest_random(width);
3428 s = 1 + ztest_random(width - 1);
3429
3430 packoff = n * sizeof (bufwad_t);
3431 packsize = s * sizeof (bufwad_t);
3432
3433 bigoff = n * chunksize;
3434 bigsize = s * chunksize;
3435
3436 packbuf = umem_alloc(packsize, UMEM_NOFAIL);
3437 bigbuf = umem_alloc(bigsize, UMEM_NOFAIL);
3438
3439 /*
3440 * free_percent of the time, free a range of bigobj rather than
3441 * overwriting it.
3442 */
3443 freeit = (ztest_random(100) < free_percent);
3444
3445 /*
3446 * Read the current contents of our objects.
3447 */
3448 error = dmu_read(os, packobj, packoff, packsize, packbuf,
3449 DMU_READ_PREFETCH);
3450 ASSERT3U(error, ==, 0);
3451 error = dmu_read(os, bigobj, bigoff, bigsize, bigbuf,
3452 DMU_READ_PREFETCH);
3453 ASSERT3U(error, ==, 0);
3454
3455 /*
3456 * Get a tx for the mods to both packobj and bigobj.
3457 */
3458 tx = dmu_tx_create(os);
3459
3460 dmu_tx_hold_write(tx, packobj, packoff, packsize);
3461
3462 if (freeit)
3463 dmu_tx_hold_free(tx, bigobj, bigoff, bigsize);
3464 else
3465 dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
3466
3467 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3468 if (txg == 0) {
3469 umem_free(packbuf, packsize);
3470 umem_free(bigbuf, bigsize);
3471 return;
3472 }
3473
3474 dmu_object_set_checksum(os, bigobj,
3475 (enum zio_checksum)ztest_random_dsl_prop(ZFS_PROP_CHECKSUM), tx);
3476
3477 dmu_object_set_compress(os, bigobj,
3478 (enum zio_compress)ztest_random_dsl_prop(ZFS_PROP_COMPRESSION), tx);
3479
3480 /*
3481 * For each index from n to n + s, verify that the existing bufwad
3482 * in packobj matches the bufwads at the head and tail of the
3483 * corresponding chunk in bigobj. Then update all three bufwads
3484 * with the new values we want to write out.
3485 */
3486 for (i = 0; i < s; i++) {
3487 /* LINTED */
3488 pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
3489 /* LINTED */
3490 bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
3491 /* LINTED */
3492 bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
3493
3494 ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
3495 ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
3496
3497 if (pack->bw_txg > txg)
3498 fatal(0, "future leak: got %llx, open txg is %llx",
3499 pack->bw_txg, txg);
3500
3501 if (pack->bw_data != 0 && pack->bw_index != n + i)
3502 fatal(0, "wrong index: got %llx, wanted %llx+%llx",
3503 pack->bw_index, n, i);
3504
3505 if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
3506 fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
3507
3508 if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
3509 fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
3510
3511 if (freeit) {
3512 bzero(pack, sizeof (bufwad_t));
3513 } else {
3514 pack->bw_index = n + i;
3515 pack->bw_txg = txg;
3516 pack->bw_data = 1 + ztest_random(-2ULL);
3517 }
3518 *bigH = *pack;
3519 *bigT = *pack;
3520 }
3521
3522 /*
3523 * We've verified all the old bufwads, and made new ones.
3524 * Now write them out.
3525 */
3526 dmu_write(os, packobj, packoff, packsize, packbuf, tx);
3527
3528 if (freeit) {
3529 if (ztest_opts.zo_verbose >= 7) {
3530 (void) printf("freeing offset %llx size %llx"
3531 " txg %llx\n",
3532 (u_longlong_t)bigoff,
3533 (u_longlong_t)bigsize,
3534 (u_longlong_t)txg);
3535 }
3536 VERIFY(0 == dmu_free_range(os, bigobj, bigoff, bigsize, tx));
3537 } else {
3538 if (ztest_opts.zo_verbose >= 7) {
3539 (void) printf("writing offset %llx size %llx"
3540 " txg %llx\n",
3541 (u_longlong_t)bigoff,
3542 (u_longlong_t)bigsize,
3543 (u_longlong_t)txg);
3544 }
3545 dmu_write(os, bigobj, bigoff, bigsize, bigbuf, tx);
3546 }
3547
3548 dmu_tx_commit(tx);
3549
3550 /*
3551 * Sanity check the stuff we just wrote.
3552 */
3553 {
3554 void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
3555 void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
3556
3557 VERIFY(0 == dmu_read(os, packobj, packoff,
3558 packsize, packcheck, DMU_READ_PREFETCH));
3559 VERIFY(0 == dmu_read(os, bigobj, bigoff,
3560 bigsize, bigcheck, DMU_READ_PREFETCH));
3561
3562 ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
3563 ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
3564
3565 umem_free(packcheck, packsize);
3566 umem_free(bigcheck, bigsize);
3567 }
3568
3569 umem_free(packbuf, packsize);
3570 umem_free(bigbuf, bigsize);
3571 }
3572
3573 void
3574 compare_and_update_pbbufs(uint64_t s, bufwad_t *packbuf, bufwad_t *bigbuf,
3575 uint64_t bigsize, uint64_t n, uint64_t chunksize, uint64_t txg)
3576 {
3577 uint64_t i;
3578 bufwad_t *pack;
3579 bufwad_t *bigH;
3580 bufwad_t *bigT;
3581
3582 /*
3583 * For each index from n to n + s, verify that the existing bufwad
3584 * in packobj matches the bufwads at the head and tail of the
3585 * corresponding chunk in bigobj. Then update all three bufwads
3586 * with the new values we want to write out.
3587 */
3588 for (i = 0; i < s; i++) {
3589 /* LINTED */
3590 pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
3591 /* LINTED */
3592 bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
3593 /* LINTED */
3594 bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
3595
3596 ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
3597 ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
3598
3599 if (pack->bw_txg > txg)
3600 fatal(0, "future leak: got %llx, open txg is %llx",
3601 pack->bw_txg, txg);
3602
3603 if (pack->bw_data != 0 && pack->bw_index != n + i)
3604 fatal(0, "wrong index: got %llx, wanted %llx+%llx",
3605 pack->bw_index, n, i);
3606
3607 if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
3608 fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
3609
3610 if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
3611 fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
3612
3613 pack->bw_index = n + i;
3614 pack->bw_txg = txg;
3615 pack->bw_data = 1 + ztest_random(-2ULL);
3616
3617 *bigH = *pack;
3618 *bigT = *pack;
3619 }
3620 }
3621
3622 void
3623 ztest_dmu_read_write_zcopy(ztest_ds_t *zd, uint64_t id)
3624 {
3625 objset_t *os = zd->zd_os;
3626 ztest_od_t od[2];
3627 dmu_tx_t *tx;
3628 uint64_t i;
3629 int error;
3630 uint64_t n, s, txg;
3631 bufwad_t *packbuf, *bigbuf;
3632 uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
3633 uint64_t blocksize = ztest_random_blocksize();
3634 uint64_t chunksize = blocksize;
3635 uint64_t regions = 997;
3636 uint64_t stride = 123456789ULL;
3637 uint64_t width = 9;
3638 dmu_buf_t *bonus_db;
3639 arc_buf_t **bigbuf_arcbufs;
3640 dmu_object_info_t doi;
3641
3642 /*
3643 * This test uses two objects, packobj and bigobj, that are always
3644 * updated together (i.e. in the same tx) so that their contents are
3645 * in sync and can be compared. Their contents relate to each other
3646 * in a simple way: packobj is a dense array of 'bufwad' structures,
3647 * while bigobj is a sparse array of the same bufwads. Specifically,
3648 * for any index n, there are three bufwads that should be identical:
3649 *
3650 * packobj, at offset n * sizeof (bufwad_t)
3651 * bigobj, at the head of the nth chunk
3652 * bigobj, at the tail of the nth chunk
3653 *
3654 * The chunk size is set equal to bigobj block size so that
3655 * dmu_assign_arcbuf() can be tested for object updates.
3656 */
3657
3658 /*
3659 * Read the directory info. If it's the first time, set things up.
3660 */
3661 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
3662 ztest_od_init(&od[1], id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize);
3663
3664 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
3665 return;
3666
3667 bigobj = od[0].od_object;
3668 packobj = od[1].od_object;
3669 blocksize = od[0].od_blocksize;
3670 chunksize = blocksize;
3671 ASSERT(chunksize == od[1].od_gen);
3672
3673 VERIFY(dmu_object_info(os, bigobj, &doi) == 0);
3674 VERIFY(ISP2(doi.doi_data_block_size));
3675 VERIFY(chunksize == doi.doi_data_block_size);
3676 VERIFY(chunksize >= 2 * sizeof (bufwad_t));
3677
3678 /*
3679 * Pick a random index and compute the offsets into packobj and bigobj.
3680 */
3681 n = ztest_random(regions) * stride + ztest_random(width);
3682 s = 1 + ztest_random(width - 1);
3683
3684 packoff = n * sizeof (bufwad_t);
3685 packsize = s * sizeof (bufwad_t);
3686
3687 bigoff = n * chunksize;
3688 bigsize = s * chunksize;
3689
3690 packbuf = umem_zalloc(packsize, UMEM_NOFAIL);
3691 bigbuf = umem_zalloc(bigsize, UMEM_NOFAIL);
3692
3693 VERIFY3U(0, ==, dmu_bonus_hold(os, bigobj, FTAG, &bonus_db));
3694
3695 bigbuf_arcbufs = umem_zalloc(2 * s * sizeof (arc_buf_t *), UMEM_NOFAIL);
3696
3697 /*
3698 * Iteration 0 test zcopy for DB_UNCACHED dbufs.
3699 * Iteration 1 test zcopy to already referenced dbufs.
3700 * Iteration 2 test zcopy to dirty dbuf in the same txg.
3701 * Iteration 3 test zcopy to dbuf dirty in previous txg.
3702 * Iteration 4 test zcopy when dbuf is no longer dirty.
3703 * Iteration 5 test zcopy when it can't be done.
3704 * Iteration 6 one more zcopy write.
3705 */
3706 for (i = 0; i < 7; i++) {
3707 uint64_t j;
3708 uint64_t off;
3709
3710 /*
3711 * In iteration 5 (i == 5) use arcbufs
3712 * that don't match bigobj blksz to test
3713 * dmu_assign_arcbuf() when it can't directly
3714 * assign an arcbuf to a dbuf.
3715 */
3716 for (j = 0; j < s; j++) {
3717 if (i != 5) {
3718 bigbuf_arcbufs[j] =
3719 dmu_request_arcbuf(bonus_db, chunksize);
3720 } else {
3721 bigbuf_arcbufs[2 * j] =
3722 dmu_request_arcbuf(bonus_db, chunksize / 2);
3723 bigbuf_arcbufs[2 * j + 1] =
3724 dmu_request_arcbuf(bonus_db, chunksize / 2);
3725 }
3726 }
3727
3728 /*
3729 * Get a tx for the mods to both packobj and bigobj.
3730 */
3731 tx = dmu_tx_create(os);
3732
3733 dmu_tx_hold_write(tx, packobj, packoff, packsize);
3734 dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
3735
3736 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3737 if (txg == 0) {
3738 umem_free(packbuf, packsize);
3739 umem_free(bigbuf, bigsize);
3740 for (j = 0; j < s; j++) {
3741 if (i != 5) {
3742 dmu_return_arcbuf(bigbuf_arcbufs[j]);
3743 } else {
3744 dmu_return_arcbuf(
3745 bigbuf_arcbufs[2 * j]);
3746 dmu_return_arcbuf(
3747 bigbuf_arcbufs[2 * j + 1]);
3748 }
3749 }
3750 umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
3751 dmu_buf_rele(bonus_db, FTAG);
3752 return;
3753 }
3754
3755 /*
3756 * 50% of the time don't read objects in the 1st iteration to
3757 * test dmu_assign_arcbuf() for the case when there're no
3758 * existing dbufs for the specified offsets.
3759 */
3760 if (i != 0 || ztest_random(2) != 0) {
3761 error = dmu_read(os, packobj, packoff,
3762 packsize, packbuf, DMU_READ_PREFETCH);
3763 ASSERT3U(error, ==, 0);
3764 error = dmu_read(os, bigobj, bigoff, bigsize,
3765 bigbuf, DMU_READ_PREFETCH);
3766 ASSERT3U(error, ==, 0);
3767 }
3768 compare_and_update_pbbufs(s, packbuf, bigbuf, bigsize,
3769 n, chunksize, txg);
3770
3771 /*
3772 * We've verified all the old bufwads, and made new ones.
3773 * Now write them out.
3774 */
3775 dmu_write(os, packobj, packoff, packsize, packbuf, tx);
3776 if (ztest_opts.zo_verbose >= 7) {
3777 (void) printf("writing offset %llx size %llx"
3778 " txg %llx\n",
3779 (u_longlong_t)bigoff,
3780 (u_longlong_t)bigsize,
3781 (u_longlong_t)txg);
3782 }
3783 for (off = bigoff, j = 0; j < s; j++, off += chunksize) {
3784 dmu_buf_t *dbt;
3785 if (i != 5) {
3786 bcopy((caddr_t)bigbuf + (off - bigoff),
3787 bigbuf_arcbufs[j]->b_data, chunksize);
3788 } else {
3789 bcopy((caddr_t)bigbuf + (off - bigoff),
3790 bigbuf_arcbufs[2 * j]->b_data,
3791 chunksize / 2);
3792 bcopy((caddr_t)bigbuf + (off - bigoff) +
3793 chunksize / 2,
3794 bigbuf_arcbufs[2 * j + 1]->b_data,
3795 chunksize / 2);
3796 }
3797
3798 if (i == 1) {
3799 VERIFY(dmu_buf_hold(os, bigobj, off,
3800 FTAG, &dbt, DMU_READ_NO_PREFETCH) == 0);
3801 }
3802 if (i != 5) {
3803 dmu_assign_arcbuf(bonus_db, off,
3804 bigbuf_arcbufs[j], tx);
3805 } else {
3806 dmu_assign_arcbuf(bonus_db, off,
3807 bigbuf_arcbufs[2 * j], tx);
3808 dmu_assign_arcbuf(bonus_db,
3809 off + chunksize / 2,
3810 bigbuf_arcbufs[2 * j + 1], tx);
3811 }
3812 if (i == 1) {
3813 dmu_buf_rele(dbt, FTAG);
3814 }
3815 }
3816 dmu_tx_commit(tx);
3817
3818 /*
3819 * Sanity check the stuff we just wrote.
3820 */
3821 {
3822 void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
3823 void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
3824
3825 VERIFY(0 == dmu_read(os, packobj, packoff,
3826 packsize, packcheck, DMU_READ_PREFETCH));
3827 VERIFY(0 == dmu_read(os, bigobj, bigoff,
3828 bigsize, bigcheck, DMU_READ_PREFETCH));
3829
3830 ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
3831 ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
3832
3833 umem_free(packcheck, packsize);
3834 umem_free(bigcheck, bigsize);
3835 }
3836 if (i == 2) {
3837 txg_wait_open(dmu_objset_pool(os), 0);
3838 } else if (i == 3) {
3839 txg_wait_synced(dmu_objset_pool(os), 0);
3840 }
3841 }
3842
3843 dmu_buf_rele(bonus_db, FTAG);
3844 umem_free(packbuf, packsize);
3845 umem_free(bigbuf, bigsize);
3846 umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
3847 }
3848
3849 /* ARGSUSED */
3850 void
3851 ztest_dmu_write_parallel(ztest_ds_t *zd, uint64_t id)
3852 {
3853 ztest_od_t od[1];
3854 uint64_t offset = (1ULL << (ztest_random(20) + 43)) +
3855 (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
3856
3857 /*
3858 * Have multiple threads write to large offsets in an object
3859 * to verify that parallel writes to an object -- even to the
3860 * same blocks within the object -- doesn't cause any trouble.
3861 */
3862 ztest_od_init(&od[0], ID_PARALLEL, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
3863
3864 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
3865 return;
3866
3867 while (ztest_random(10) != 0)
3868 ztest_io(zd, od[0].od_object, offset);
3869 }
3870
3871 void
3872 ztest_dmu_prealloc(ztest_ds_t *zd, uint64_t id)
3873 {
3874 ztest_od_t od[1];
3875 uint64_t offset = (1ULL << (ztest_random(4) + SPA_MAXBLOCKSHIFT)) +
3876 (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
3877 uint64_t count = ztest_random(20) + 1;
3878 uint64_t blocksize = ztest_random_blocksize();
3879 void *data;
3880
3881 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
3882
3883 if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
3884 return;
3885
3886 if (ztest_truncate(zd, od[0].od_object, offset, count * blocksize) != 0)
3887 return;
3888
3889 ztest_prealloc(zd, od[0].od_object, offset, count * blocksize);
3890
3891 data = umem_zalloc(blocksize, UMEM_NOFAIL);
3892
3893 while (ztest_random(count) != 0) {
3894 uint64_t randoff = offset + (ztest_random(count) * blocksize);
3895 if (ztest_write(zd, od[0].od_object, randoff, blocksize,
3896 data) != 0)
3897 break;
3898 while (ztest_random(4) != 0)
3899 ztest_io(zd, od[0].od_object, randoff);
3900 }
3901
3902 umem_free(data, blocksize);
3903 }
3904
3905 /*
3906 * Verify that zap_{create,destroy,add,remove,update} work as expected.
3907 */
3908 #define ZTEST_ZAP_MIN_INTS 1
3909 #define ZTEST_ZAP_MAX_INTS 4
3910 #define ZTEST_ZAP_MAX_PROPS 1000
3911
3912 void
3913 ztest_zap(ztest_ds_t *zd, uint64_t id)
3914 {
3915 objset_t *os = zd->zd_os;
3916 ztest_od_t od[1];
3917 uint64_t object;
3918 uint64_t txg, last_txg;
3919 uint64_t value[ZTEST_ZAP_MAX_INTS];
3920 uint64_t zl_ints, zl_intsize, prop;
3921 int i, ints;
3922 dmu_tx_t *tx;
3923 char propname[100], txgname[100];
3924 int error;
3925 char *hc[2] = { "s.acl.h", ".s.open.h.hyLZlg" };
3926
3927 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0);
3928
3929 if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
3930 return;
3931
3932 object = od[0].od_object;
3933
3934 /*
3935 * Generate a known hash collision, and verify that
3936 * we can lookup and remove both entries.
3937 */
3938 tx = dmu_tx_create(os);
3939 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
3940 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3941 if (txg == 0)
3942 return;
3943 for (i = 0; i < 2; i++) {
3944 value[i] = i;
3945 VERIFY3U(0, ==, zap_add(os, object, hc[i], sizeof (uint64_t),
3946 1, &value[i], tx));
3947 }
3948 for (i = 0; i < 2; i++) {
3949 VERIFY3U(EEXIST, ==, zap_add(os, object, hc[i],
3950 sizeof (uint64_t), 1, &value[i], tx));
3951 VERIFY3U(0, ==,
3952 zap_length(os, object, hc[i], &zl_intsize, &zl_ints));
3953 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
3954 ASSERT3U(zl_ints, ==, 1);
3955 }
3956 for (i = 0; i < 2; i++) {
3957 VERIFY3U(0, ==, zap_remove(os, object, hc[i], tx));
3958 }
3959 dmu_tx_commit(tx);
3960
3961 /*
3962 * Generate a buch of random entries.
3963 */
3964 ints = MAX(ZTEST_ZAP_MIN_INTS, object % ZTEST_ZAP_MAX_INTS);
3965
3966 prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
3967 (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
3968 (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
3969 bzero(value, sizeof (value));
3970 last_txg = 0;
3971
3972 /*
3973 * If these zap entries already exist, validate their contents.
3974 */
3975 error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
3976 if (error == 0) {
3977 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
3978 ASSERT3U(zl_ints, ==, 1);
3979
3980 VERIFY(zap_lookup(os, object, txgname, zl_intsize,
3981 zl_ints, &last_txg) == 0);
3982
3983 VERIFY(zap_length(os, object, propname, &zl_intsize,
3984 &zl_ints) == 0);
3985
3986 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
3987 ASSERT3U(zl_ints, ==, ints);
3988
3989 VERIFY(zap_lookup(os, object, propname, zl_intsize,
3990 zl_ints, value) == 0);
3991
3992 for (i = 0; i < ints; i++) {
3993 ASSERT3U(value[i], ==, last_txg + object + i);
3994 }
3995 } else {
3996 ASSERT3U(error, ==, ENOENT);
3997 }
3998
3999 /*
4000 * Atomically update two entries in our zap object.
4001 * The first is named txg_%llu, and contains the txg
4002 * in which the property was last updated. The second
4003 * is named prop_%llu, and the nth element of its value
4004 * should be txg + object + n.
4005 */
4006 tx = dmu_tx_create(os);
4007 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4008 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4009 if (txg == 0)
4010 return;
4011
4012 if (last_txg > txg)
4013 fatal(0, "zap future leak: old %llu new %llu", last_txg, txg);
4014
4015 for (i = 0; i < ints; i++)
4016 value[i] = txg + object + i;
4017
4018 VERIFY3U(0, ==, zap_update(os, object, txgname, sizeof (uint64_t),
4019 1, &txg, tx));
4020 VERIFY3U(0, ==, zap_update(os, object, propname, sizeof (uint64_t),
4021 ints, value, tx));
4022
4023 dmu_tx_commit(tx);
4024
4025 /*
4026 * Remove a random pair of entries.
4027 */
4028 prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
4029 (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
4030 (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
4031
4032 error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
4033
4034 if (error == ENOENT)
4035 return;
4036
4037 ASSERT3U(error, ==, 0);
4038
4039 tx = dmu_tx_create(os);
4040 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4041 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4042 if (txg == 0)
4043 return;
4044 VERIFY3U(0, ==, zap_remove(os, object, txgname, tx));
4045 VERIFY3U(0, ==, zap_remove(os, object, propname, tx));
4046 dmu_tx_commit(tx);
4047 }
4048
4049 /*
4050 * Testcase to test the upgrading of a microzap to fatzap.
4051 */
4052 void
4053 ztest_fzap(ztest_ds_t *zd, uint64_t id)
4054 {
4055 objset_t *os = zd->zd_os;
4056 ztest_od_t od[1];
4057 uint64_t object, txg;
4058
4059 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0);
4060
4061 if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4062 return;
4063
4064 object = od[0].od_object;
4065
4066 /*
4067 * Add entries to this ZAP and make sure it spills over
4068 * and gets upgraded to a fatzap. Also, since we are adding
4069 * 2050 entries we should see ptrtbl growth and leaf-block split.
4070 */
4071 for (int i = 0; i < 2050; i++) {
4072 char name[MAXNAMELEN];
4073 uint64_t value = i;
4074 dmu_tx_t *tx;
4075 int error;
4076
4077 (void) snprintf(name, sizeof (name), "fzap-%llu-%llu",
4078 id, value);
4079
4080 tx = dmu_tx_create(os);
4081 dmu_tx_hold_zap(tx, object, B_TRUE, name);
4082 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4083 if (txg == 0)
4084 return;
4085 error = zap_add(os, object, name, sizeof (uint64_t), 1,
4086 &value, tx);
4087 ASSERT(error == 0 || error == EEXIST);
4088 dmu_tx_commit(tx);
4089 }
4090 }
4091
4092 /* ARGSUSED */
4093 void
4094 ztest_zap_parallel(ztest_ds_t *zd, uint64_t id)
4095 {
4096 objset_t *os = zd->zd_os;
4097 ztest_od_t od[1];
4098 uint64_t txg, object, count, wsize, wc, zl_wsize, zl_wc;
4099 dmu_tx_t *tx;
4100 int i, namelen, error;
4101 int micro = ztest_random(2);
4102 char name[20], string_value[20];
4103 void *data;
4104
4105 ztest_od_init(&od[0], ID_PARALLEL, FTAG, micro, DMU_OT_ZAP_OTHER, 0, 0);
4106
4107 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4108 return;
4109
4110 object = od[0].od_object;
4111
4112 /*
4113 * Generate a random name of the form 'xxx.....' where each
4114 * x is a random printable character and the dots are dots.
4115 * There are 94 such characters, and the name length goes from
4116 * 6 to 20, so there are 94^3 * 15 = 12,458,760 possible names.
4117 */
4118 namelen = ztest_random(sizeof (name) - 5) + 5 + 1;
4119
4120 for (i = 0; i < 3; i++)
4121 name[i] = '!' + ztest_random('~' - '!' + 1);
4122 for (; i < namelen - 1; i++)
4123 name[i] = '.';
4124 name[i] = '\0';
4125
4126 if ((namelen & 1) || micro) {
4127 wsize = sizeof (txg);
4128 wc = 1;
4129 data = &txg;
4130 } else {
4131 wsize = 1;
4132 wc = namelen;
4133 data = string_value;
4134 }
4135
4136 count = -1ULL;
4137 VERIFY(zap_count(os, object, &count) == 0);
4138 ASSERT(count != -1ULL);
4139
4140 /*
4141 * Select an operation: length, lookup, add, update, remove.
4142 */
4143 i = ztest_random(5);
4144
4145 if (i >= 2) {
4146 tx = dmu_tx_create(os);
4147 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4148 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4149 if (txg == 0)
4150 return;
4151 bcopy(name, string_value, namelen);
4152 } else {
4153 tx = NULL;
4154 txg = 0;
4155 bzero(string_value, namelen);
4156 }
4157
4158 switch (i) {
4159
4160 case 0:
4161 error = zap_length(os, object, name, &zl_wsize, &zl_wc);
4162 if (error == 0) {
4163 ASSERT3U(wsize, ==, zl_wsize);
4164 ASSERT3U(wc, ==, zl_wc);
4165 } else {
4166 ASSERT3U(error, ==, ENOENT);
4167 }
4168 break;
4169
4170 case 1:
4171 error = zap_lookup(os, object, name, wsize, wc, data);
4172 if (error == 0) {
4173 if (data == string_value &&
4174 bcmp(name, data, namelen) != 0)
4175 fatal(0, "name '%s' != val '%s' len %d",
4176 name, data, namelen);
4177 } else {
4178 ASSERT3U(error, ==, ENOENT);
4179 }
4180 break;
4181
4182 case 2:
4183 error = zap_add(os, object, name, wsize, wc, data, tx);
4184 ASSERT(error == 0 || error == EEXIST);
4185 break;
4186
4187 case 3:
4188 VERIFY(zap_update(os, object, name, wsize, wc, data, tx) == 0);
4189 break;
4190
4191 case 4:
4192 error = zap_remove(os, object, name, tx);
4193 ASSERT(error == 0 || error == ENOENT);
4194 break;
4195 }
4196
4197 if (tx != NULL)
4198 dmu_tx_commit(tx);
4199 }
4200
4201 /*
4202 * Commit callback data.
4203 */
4204 typedef struct ztest_cb_data {
4205 list_node_t zcd_node;
4206 uint64_t zcd_txg;
4207 int zcd_expected_err;
4208 boolean_t zcd_added;
4209 boolean_t zcd_called;
4210 spa_t *zcd_spa;
4211 } ztest_cb_data_t;
4212
4213 /* This is the actual commit callback function */
4214 static void
4215 ztest_commit_callback(void *arg, int error)
4216 {
4217 ztest_cb_data_t *data = arg;
4218 uint64_t synced_txg;
4219
4220 VERIFY(data != NULL);
4221 VERIFY3S(data->zcd_expected_err, ==, error);
4222 VERIFY(!data->zcd_called);
4223
4224 synced_txg = spa_last_synced_txg(data->zcd_spa);
4225 if (data->zcd_txg > synced_txg)
4226 fatal(0, "commit callback of txg %" PRIu64 " called prematurely"
4227 ", last synced txg = %" PRIu64 "\n", data->zcd_txg,
4228 synced_txg);
4229
4230 data->zcd_called = B_TRUE;
4231
4232 if (error == ECANCELED) {
4233 ASSERT3U(data->zcd_txg, ==, 0);
4234 ASSERT(!data->zcd_added);
4235
4236 /*
4237 * The private callback data should be destroyed here, but
4238 * since we are going to check the zcd_called field after
4239 * dmu_tx_abort(), we will destroy it there.
4240 */
4241 return;
4242 }
4243
4244 /* Was this callback added to the global callback list? */
4245 if (!data->zcd_added)
4246 goto out;
4247
4248 ASSERT3U(data->zcd_txg, !=, 0);
4249
4250 /* Remove our callback from the list */
4251 (void) mutex_lock(&zcl.zcl_callbacks_lock);
4252 list_remove(&zcl.zcl_callbacks, data);
4253 (void) mutex_unlock(&zcl.zcl_callbacks_lock);
4254
4255 out:
4256 umem_free(data, sizeof (ztest_cb_data_t));
4257 }
4258
4259 /* Allocate and initialize callback data structure */
4260 static ztest_cb_data_t *
4261 ztest_create_cb_data(objset_t *os, uint64_t txg)
4262 {
4263 ztest_cb_data_t *cb_data;
4264
4265 cb_data = umem_zalloc(sizeof (ztest_cb_data_t), UMEM_NOFAIL);
4266
4267 cb_data->zcd_txg = txg;
4268 cb_data->zcd_spa = dmu_objset_spa(os);
4269
4270 return (cb_data);
4271 }
4272
4273 /*
4274 * If a number of txgs equal to this threshold have been created after a commit
4275 * callback has been registered but not called, then we assume there is an
4276 * implementation bug.
4277 */
4278 #define ZTEST_COMMIT_CALLBACK_THRESH (TXG_CONCURRENT_STATES + 2)
4279
4280 /*
4281 * Commit callback test.
4282 */
4283 void
4284 ztest_dmu_commit_callbacks(ztest_ds_t *zd, uint64_t id)
4285 {
4286 objset_t *os = zd->zd_os;
4287 ztest_od_t od[1];
4288 dmu_tx_t *tx;
4289 ztest_cb_data_t *cb_data[3], *tmp_cb;
4290 uint64_t old_txg, txg;
4291 int i, error;
4292
4293 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
4294
4295 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4296 return;
4297
4298 tx = dmu_tx_create(os);
4299
4300 cb_data[0] = ztest_create_cb_data(os, 0);
4301 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[0]);
4302
4303 dmu_tx_hold_write(tx, od[0].od_object, 0, sizeof (uint64_t));
4304
4305 /* Every once in a while, abort the transaction on purpose */
4306 if (ztest_random(100) == 0)
4307 error = -1;
4308
4309 if (!error)
4310 error = dmu_tx_assign(tx, TXG_NOWAIT);
4311
4312 txg = error ? 0 : dmu_tx_get_txg(tx);
4313
4314 cb_data[0]->zcd_txg = txg;
4315 cb_data[1] = ztest_create_cb_data(os, txg);
4316 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[1]);
4317
4318 if (error) {
4319 /*
4320 * It's not a strict requirement to call the registered
4321 * callbacks from inside dmu_tx_abort(), but that's what
4322 * it's supposed to happen in the current implementation
4323 * so we will check for that.
4324 */
4325 for (i = 0; i < 2; i++) {
4326 cb_data[i]->zcd_expected_err = ECANCELED;
4327 VERIFY(!cb_data[i]->zcd_called);
4328 }
4329
4330 dmu_tx_abort(tx);
4331
4332 for (i = 0; i < 2; i++) {
4333 VERIFY(cb_data[i]->zcd_called);
4334 umem_free(cb_data[i], sizeof (ztest_cb_data_t));
4335 }
4336
4337 return;
4338 }
4339
4340 cb_data[2] = ztest_create_cb_data(os, txg);
4341 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[2]);
4342
4343 /*
4344 * Read existing data to make sure there isn't a future leak.
4345 */
4346 VERIFY(0 == dmu_read(os, od[0].od_object, 0, sizeof (uint64_t),
4347 &old_txg, DMU_READ_PREFETCH));
4348
4349 if (old_txg > txg)
4350 fatal(0, "future leak: got %" PRIu64 ", open txg is %" PRIu64,
4351 old_txg, txg);
4352
4353 dmu_write(os, od[0].od_object, 0, sizeof (uint64_t), &txg, tx);
4354
4355 (void) mutex_lock(&zcl.zcl_callbacks_lock);
4356
4357 /*
4358 * Since commit callbacks don't have any ordering requirement and since
4359 * it is theoretically possible for a commit callback to be called
4360 * after an arbitrary amount of time has elapsed since its txg has been
4361 * synced, it is difficult to reliably determine whether a commit
4362 * callback hasn't been called due to high load or due to a flawed
4363 * implementation.
4364 *
4365 * In practice, we will assume that if after a certain number of txgs a
4366 * commit callback hasn't been called, then most likely there's an
4367 * implementation bug..
4368 */
4369 tmp_cb = list_head(&zcl.zcl_callbacks);
4370 if (tmp_cb != NULL &&
4371 tmp_cb->zcd_txg > txg - ZTEST_COMMIT_CALLBACK_THRESH) {
4372 fatal(0, "Commit callback threshold exceeded, oldest txg: %"
4373 PRIu64 ", open txg: %" PRIu64 "\n", tmp_cb->zcd_txg, txg);
4374 }
4375
4376 /*
4377 * Let's find the place to insert our callbacks.
4378 *
4379 * Even though the list is ordered by txg, it is possible for the
4380 * insertion point to not be the end because our txg may already be
4381 * quiescing at this point and other callbacks in the open txg
4382 * (from other objsets) may have sneaked in.
4383 */
4384 tmp_cb = list_tail(&zcl.zcl_callbacks);
4385 while (tmp_cb != NULL && tmp_cb->zcd_txg > txg)
4386 tmp_cb = list_prev(&zcl.zcl_callbacks, tmp_cb);
4387
4388 /* Add the 3 callbacks to the list */
4389 for (i = 0; i < 3; i++) {
4390 if (tmp_cb == NULL)
4391 list_insert_head(&zcl.zcl_callbacks, cb_data[i]);
4392 else
4393 list_insert_after(&zcl.zcl_callbacks, tmp_cb,
4394 cb_data[i]);
4395
4396 cb_data[i]->zcd_added = B_TRUE;
4397 VERIFY(!cb_data[i]->zcd_called);
4398
4399 tmp_cb = cb_data[i];
4400 }
4401
4402 (void) mutex_unlock(&zcl.zcl_callbacks_lock);
4403
4404 dmu_tx_commit(tx);
4405 }
4406
4407 /* ARGSUSED */
4408 void
4409 ztest_dsl_prop_get_set(ztest_ds_t *zd, uint64_t id)
4410 {
4411 zfs_prop_t proplist[] = {
4412 ZFS_PROP_CHECKSUM,
4413 ZFS_PROP_COMPRESSION,
4414 ZFS_PROP_COPIES,
4415 ZFS_PROP_DEDUP
4416 };
4417
4418 (void) rw_rdlock(&ztest_name_lock);
4419
4420 for (int p = 0; p < sizeof (proplist) / sizeof (proplist[0]); p++)
4421 (void) ztest_dsl_prop_set_uint64(zd->zd_name, proplist[p],
4422 ztest_random_dsl_prop(proplist[p]), (int)ztest_random(2));
4423
4424 (void) rw_unlock(&ztest_name_lock);
4425 }
4426
4427 /* ARGSUSED */
4428 void
4429 ztest_spa_prop_get_set(ztest_ds_t *zd, uint64_t id)
4430 {
4431 nvlist_t *props = NULL;
4432
4433 (void) rw_rdlock(&ztest_name_lock);
4434
4435 (void) ztest_spa_prop_set_uint64(ZPOOL_PROP_DEDUPDITTO,
4436 ZIO_DEDUPDITTO_MIN + ztest_random(ZIO_DEDUPDITTO_MIN));
4437
4438 VERIFY3U(spa_prop_get(ztest_spa, &props), ==, 0);
4439
4440 if (ztest_opts.zo_verbose >= 6)
4441 dump_nvlist(props, 4);
4442
4443 nvlist_free(props);
4444
4445 (void) rw_unlock(&ztest_name_lock);
4446 }
4447
4448 /*
4449 * Test snapshot hold/release and deferred destroy.
4450 */
4451 void
4452 ztest_dmu_snapshot_hold(ztest_ds_t *zd, uint64_t id)
4453 {
4454 int error;
4455 objset_t *os = zd->zd_os;
4456 objset_t *origin;
4457 char snapname[100];
4458 char fullname[100];
4459 char clonename[100];
4460 char tag[100];
4461 char osname[MAXNAMELEN];
4462
4463 (void) rw_rdlock(&ztest_name_lock);
4464
4465 dmu_objset_name(os, osname);
4466
4467 (void) snprintf(snapname, 100, "sh1_%llu", id);
4468 (void) snprintf(fullname, 100, "%s@%s", osname, snapname);
4469 (void) snprintf(clonename, 100, "%s/ch1_%llu", osname, id);
4470 (void) snprintf(tag, 100, "%tag_%llu", id);
4471
4472 /*
4473 * Clean up from any previous run.
4474 */
4475 (void) dmu_objset_destroy(clonename, B_FALSE);
4476 (void) dsl_dataset_user_release(osname, snapname, tag, B_FALSE);
4477 (void) dmu_objset_destroy(fullname, B_FALSE);
4478
4479 /*
4480 * Create snapshot, clone it, mark snap for deferred destroy,
4481 * destroy clone, verify snap was also destroyed.
4482 */
4483 error = dmu_objset_snapshot(osname, snapname, NULL, NULL, FALSE,
4484 FALSE, -1);
4485 if (error) {
4486 if (error == ENOSPC) {
4487 ztest_record_enospc("dmu_objset_snapshot");
4488 goto out;
4489 }
4490 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
4491 }
4492
4493 error = dmu_objset_hold(fullname, FTAG, &origin);
4494 if (error)
4495 fatal(0, "dmu_objset_hold(%s) = %d", fullname, error);
4496
4497 error = dmu_objset_clone(clonename, dmu_objset_ds(origin), 0);
4498 dmu_objset_rele(origin, FTAG);
4499 if (error) {
4500 if (error == ENOSPC) {
4501 ztest_record_enospc("dmu_objset_clone");
4502 goto out;
4503 }
4504 fatal(0, "dmu_objset_clone(%s) = %d", clonename, error);
4505 }
4506
4507 error = dmu_objset_destroy(fullname, B_TRUE);
4508 if (error) {
4509 fatal(0, "dmu_objset_destroy(%s, B_TRUE) = %d",
4510 fullname, error);
4511 }
4512
4513 error = dmu_objset_destroy(clonename, B_FALSE);
4514 if (error)
4515 fatal(0, "dmu_objset_destroy(%s) = %d", clonename, error);
4516
4517 error = dmu_objset_hold(fullname, FTAG, &origin);
4518 if (error != ENOENT)
4519 fatal(0, "dmu_objset_hold(%s) = %d", fullname, error);
4520
4521 /*
4522 * Create snapshot, add temporary hold, verify that we can't
4523 * destroy a held snapshot, mark for deferred destroy,
4524 * release hold, verify snapshot was destroyed.
4525 */
4526 error = dmu_objset_snapshot(osname, snapname, NULL, NULL, FALSE,
4527 FALSE, -1);
4528 if (error) {
4529 if (error == ENOSPC) {
4530 ztest_record_enospc("dmu_objset_snapshot");
4531 goto out;
4532 }
4533 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
4534 }
4535
4536 error = dsl_dataset_user_hold(osname, snapname, tag, B_FALSE,
4537 B_TRUE, -1);
4538 if (error)
4539 fatal(0, "dsl_dataset_user_hold(%s)", fullname, tag);
4540
4541 error = dmu_objset_destroy(fullname, B_FALSE);
4542 if (error != EBUSY) {
4543 fatal(0, "dmu_objset_destroy(%s, B_FALSE) = %d",
4544 fullname, error);
4545 }
4546
4547 error = dmu_objset_destroy(fullname, B_TRUE);
4548 if (error) {
4549 fatal(0, "dmu_objset_destroy(%s, B_TRUE) = %d",
4550 fullname, error);
4551 }
4552
4553 error = dsl_dataset_user_release(osname, snapname, tag, B_FALSE);
4554 if (error)
4555 fatal(0, "dsl_dataset_user_release(%s)", fullname, tag);
4556
4557 VERIFY(dmu_objset_hold(fullname, FTAG, &origin) == ENOENT);
4558
4559 out:
4560 (void) rw_unlock(&ztest_name_lock);
4561 }
4562
4563 /*
4564 * Inject random faults into the on-disk data.
4565 */
4566 /* ARGSUSED */
4567 void
4568 ztest_fault_inject(ztest_ds_t *zd, uint64_t id)
4569 {
4570 ztest_shared_t *zs = ztest_shared;
4571 spa_t *spa = ztest_spa;
4572 int fd;
4573 uint64_t offset;
4574 uint64_t leaves;
4575 uint64_t bad = 0x1990c0ffeedecade;
4576 uint64_t top, leaf;
4577 char path0[MAXPATHLEN];
4578 char pathrand[MAXPATHLEN];
4579 size_t fsize;
4580 int bshift = SPA_MAXBLOCKSHIFT + 2; /* don't scrog all labels */
4581 int iters = 1000;
4582 int maxfaults;
4583 int mirror_save;
4584 vdev_t *vd0 = NULL;
4585 uint64_t guid0 = 0;
4586 boolean_t islog = B_FALSE;
4587
4588 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
4589 maxfaults = MAXFAULTS();
4590 leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz;
4591 mirror_save = zs->zs_mirrors;
4592 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4593
4594 ASSERT(leaves >= 1);
4595
4596 /*
4597 * We need SCL_STATE here because we're going to look at vd0->vdev_tsd.
4598 */
4599 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
4600
4601 if (ztest_random(2) == 0) {
4602 /*
4603 * Inject errors on a normal data device or slog device.
4604 */
4605 top = ztest_random_vdev_top(spa, B_TRUE);
4606 leaf = ztest_random(leaves) + zs->zs_splits;
4607
4608 /*
4609 * Generate paths to the first leaf in this top-level vdev,
4610 * and to the random leaf we selected. We'll induce transient
4611 * write failures and random online/offline activity on leaf 0,
4612 * and we'll write random garbage to the randomly chosen leaf.
4613 */
4614 (void) snprintf(path0, sizeof (path0), ztest_dev_template,
4615 ztest_opts.zo_dir, ztest_opts.zo_pool,
4616 top * leaves + zs->zs_splits);
4617 (void) snprintf(pathrand, sizeof (pathrand), ztest_dev_template,
4618 ztest_opts.zo_dir, ztest_opts.zo_pool,
4619 top * leaves + leaf);
4620
4621 vd0 = vdev_lookup_by_path(spa->spa_root_vdev, path0);
4622 if (vd0 != NULL && vd0->vdev_top->vdev_islog)
4623 islog = B_TRUE;
4624
4625 if (vd0 != NULL && maxfaults != 1) {
4626 /*
4627 * Make vd0 explicitly claim to be unreadable,
4628 * or unwriteable, or reach behind its back
4629 * and close the underlying fd. We can do this if
4630 * maxfaults == 0 because we'll fail and reexecute,
4631 * and we can do it if maxfaults >= 2 because we'll
4632 * have enough redundancy. If maxfaults == 1, the
4633 * combination of this with injection of random data
4634 * corruption below exceeds the pool's fault tolerance.
4635 */
4636 vdev_file_t *vf = vd0->vdev_tsd;
4637
4638 if (vf != NULL && ztest_random(3) == 0) {
4639 (void) close(vf->vf_vnode->v_fd);
4640 vf->vf_vnode->v_fd = -1;
4641 } else if (ztest_random(2) == 0) {
4642 vd0->vdev_cant_read = B_TRUE;
4643 } else {
4644 vd0->vdev_cant_write = B_TRUE;
4645 }
4646 guid0 = vd0->vdev_guid;
4647 }
4648 } else {
4649 /*
4650 * Inject errors on an l2cache device.
4651 */
4652 spa_aux_vdev_t *sav = &spa->spa_l2cache;
4653
4654 if (sav->sav_count == 0) {
4655 spa_config_exit(spa, SCL_STATE, FTAG);
4656 return;
4657 }
4658 vd0 = sav->sav_vdevs[ztest_random(sav->sav_count)];
4659 guid0 = vd0->vdev_guid;
4660 (void) strcpy(path0, vd0->vdev_path);
4661 (void) strcpy(pathrand, vd0->vdev_path);
4662
4663 leaf = 0;
4664 leaves = 1;
4665 maxfaults = INT_MAX; /* no limit on cache devices */
4666 }
4667
4668 spa_config_exit(spa, SCL_STATE, FTAG);
4669
4670 /*
4671 * If we can tolerate two or more faults, or we're dealing
4672 * with a slog, randomly online/offline vd0.
4673 */
4674 if ((maxfaults >= 2 || islog) && guid0 != 0) {
4675 if (ztest_random(10) < 6) {
4676 int flags = (ztest_random(2) == 0 ?
4677 ZFS_OFFLINE_TEMPORARY : 0);
4678
4679 /*
4680 * We have to grab the zs_name_lock as writer to
4681 * prevent a race between offlining a slog and
4682 * destroying a dataset. Offlining the slog will
4683 * grab a reference on the dataset which may cause
4684 * dmu_objset_destroy() to fail with EBUSY thus
4685 * leaving the dataset in an inconsistent state.
4686 */
4687 if (islog)
4688 (void) rw_wrlock(&ztest_name_lock);
4689
4690 VERIFY(vdev_offline(spa, guid0, flags) != EBUSY);
4691
4692 if (islog)
4693 (void) rw_unlock(&ztest_name_lock);
4694 } else {
4695 (void) vdev_online(spa, guid0, 0, NULL);
4696 }
4697 }
4698
4699 if (maxfaults == 0)
4700 return;
4701
4702 /*
4703 * We have at least single-fault tolerance, so inject data corruption.
4704 */
4705 fd = open(pathrand, O_RDWR);
4706
4707 if (fd == -1) /* we hit a gap in the device namespace */
4708 return;
4709
4710 fsize = lseek(fd, 0, SEEK_END);
4711
4712 while (--iters != 0) {
4713 offset = ztest_random(fsize / (leaves << bshift)) *
4714 (leaves << bshift) + (leaf << bshift) +
4715 (ztest_random(1ULL << (bshift - 1)) & -8ULL);
4716
4717 if (offset >= fsize)
4718 continue;
4719
4720 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
4721 if (mirror_save != zs->zs_mirrors) {
4722 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4723 (void) close(fd);
4724 return;
4725 }
4726
4727 if (pwrite(fd, &bad, sizeof (bad), offset) != sizeof (bad))
4728 fatal(1, "can't inject bad word at 0x%llx in %s",
4729 offset, pathrand);
4730
4731 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4732
4733 if (ztest_opts.zo_verbose >= 7)
4734 (void) printf("injected bad word into %s,"
4735 " offset 0x%llx\n", pathrand, (u_longlong_t)offset);
4736 }
4737
4738 (void) close(fd);
4739 }
4740
4741 /*
4742 * Verify that DDT repair works as expected.
4743 */
4744 void
4745 ztest_ddt_repair(ztest_ds_t *zd, uint64_t id)
4746 {
4747 ztest_shared_t *zs = ztest_shared;
4748 spa_t *spa = ztest_spa;
4749 objset_t *os = zd->zd_os;
4750 ztest_od_t od[1];
4751 uint64_t object, blocksize, txg, pattern, psize;
4752 enum zio_checksum checksum = spa_dedup_checksum(spa);
4753 dmu_buf_t *db;
4754 dmu_tx_t *tx;
4755 void *buf;
4756 blkptr_t blk;
4757 int copies = 2 * ZIO_DEDUPDITTO_MIN;
4758
4759 blocksize = ztest_random_blocksize();
4760 blocksize = MIN(blocksize, 2048); /* because we write so many */
4761
4762 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
4763
4764 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4765 return;
4766
4767 /*
4768 * Take the name lock as writer to prevent anyone else from changing
4769 * the pool and dataset properies we need to maintain during this test.
4770 */
4771 (void) rw_wrlock(&ztest_name_lock);
4772
4773 if (ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_DEDUP, checksum,
4774 B_FALSE) != 0 ||
4775 ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_COPIES, 1,
4776 B_FALSE) != 0) {
4777 (void) rw_unlock(&ztest_name_lock);
4778 return;
4779 }
4780
4781 object = od[0].od_object;
4782 blocksize = od[0].od_blocksize;
4783 pattern = zs->zs_guid ^ dmu_objset_fsid_guid(os);
4784
4785 ASSERT(object != 0);
4786
4787 tx = dmu_tx_create(os);
4788 dmu_tx_hold_write(tx, object, 0, copies * blocksize);
4789 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
4790 if (txg == 0) {
4791 (void) rw_unlock(&ztest_name_lock);
4792 return;
4793 }
4794
4795 /*
4796 * Write all the copies of our block.
4797 */
4798 for (int i = 0; i < copies; i++) {
4799 uint64_t offset = i * blocksize;
4800 VERIFY(dmu_buf_hold(os, object, offset, FTAG, &db,
4801 DMU_READ_NO_PREFETCH) == 0);
4802 ASSERT(db->db_offset == offset);
4803 ASSERT(db->db_size == blocksize);
4804 ASSERT(ztest_pattern_match(db->db_data, db->db_size, pattern) ||
4805 ztest_pattern_match(db->db_data, db->db_size, 0ULL));
4806 dmu_buf_will_fill(db, tx);
4807 ztest_pattern_set(db->db_data, db->db_size, pattern);
4808 dmu_buf_rele(db, FTAG);
4809 }
4810
4811 dmu_tx_commit(tx);
4812 txg_wait_synced(spa_get_dsl(spa), txg);
4813
4814 /*
4815 * Find out what block we got.
4816 */
4817 VERIFY(dmu_buf_hold(os, object, 0, FTAG, &db,
4818 DMU_READ_NO_PREFETCH) == 0);
4819 blk = *((dmu_buf_impl_t *)db)->db_blkptr;
4820 dmu_buf_rele(db, FTAG);
4821
4822 /*
4823 * Damage the block. Dedup-ditto will save us when we read it later.
4824 */
4825 psize = BP_GET_PSIZE(&blk);
4826 buf = zio_buf_alloc(psize);
4827 ztest_pattern_set(buf, psize, ~pattern);
4828
4829 (void) zio_wait(zio_rewrite(NULL, spa, 0, &blk,
4830 buf, psize, NULL, NULL, ZIO_PRIORITY_SYNC_WRITE,
4831 ZIO_FLAG_CANFAIL | ZIO_FLAG_INDUCE_DAMAGE, NULL));
4832
4833 zio_buf_free(buf, psize);
4834
4835 (void) rw_unlock(&ztest_name_lock);
4836 }
4837
4838 /*
4839 * Scrub the pool.
4840 */
4841 /* ARGSUSED */
4842 void
4843 ztest_scrub(ztest_ds_t *zd, uint64_t id)
4844 {
4845 spa_t *spa = ztest_spa;
4846
4847 (void) spa_scan(spa, POOL_SCAN_SCRUB);
4848 (void) poll(NULL, 0, 100); /* wait a moment, then force a restart */
4849 (void) spa_scan(spa, POOL_SCAN_SCRUB);
4850 }
4851
4852 /*
4853 * Change the guid for the pool.
4854 */
4855 /* ARGSUSED */
4856 void
4857 ztest_reguid(ztest_ds_t *zd, uint64_t id)
4858 {
4859 spa_t *spa = ztest_spa;
4860 uint64_t orig, load;
4861
4862 orig = spa_guid(spa);
4863 load = spa_load_guid(spa);
4864 if (spa_change_guid(spa) != 0)
4865 return;
4866
4867 if (ztest_opts.zo_verbose >= 3) {
4868 (void) printf("Changed guid old %llu -> %llu\n",
4869 (u_longlong_t)orig, (u_longlong_t)spa_guid(spa));
4870 }
4871
4872 VERIFY3U(orig, !=, spa_guid(spa));
4873 VERIFY3U(load, ==, spa_load_guid(spa));
4874 }
4875
4876 /*
4877 * Rename the pool to a different name and then rename it back.
4878 */
4879 /* ARGSUSED */
4880 void
4881 ztest_spa_rename(ztest_ds_t *zd, uint64_t id)
4882 {
4883 char *oldname, *newname;
4884 spa_t *spa;
4885
4886 (void) rw_wrlock(&ztest_name_lock);
4887
4888 oldname = ztest_opts.zo_pool;
4889 newname = umem_alloc(strlen(oldname) + 5, UMEM_NOFAIL);
4890 (void) strcpy(newname, oldname);
4891 (void) strcat(newname, "_tmp");
4892
4893 /*
4894 * Do the rename
4895 */
4896 VERIFY3U(0, ==, spa_rename(oldname, newname));
4897
4898 /*
4899 * Try to open it under the old name, which shouldn't exist
4900 */
4901 VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
4902
4903 /*
4904 * Open it under the new name and make sure it's still the same spa_t.
4905 */
4906 VERIFY3U(0, ==, spa_open(newname, &spa, FTAG));
4907
4908 ASSERT(spa == ztest_spa);
4909 spa_close(spa, FTAG);
4910
4911 /*
4912 * Rename it back to the original
4913 */
4914 VERIFY3U(0, ==, spa_rename(newname, oldname));
4915
4916 /*
4917 * Make sure it can still be opened
4918 */
4919 VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG));
4920
4921 ASSERT(spa == ztest_spa);
4922 spa_close(spa, FTAG);
4923
4924 umem_free(newname, strlen(newname) + 1);
4925
4926 (void) rw_unlock(&ztest_name_lock);
4927 }
4928
4929 /*
4930 * Verify pool integrity by running zdb.
4931 */
4932 static void
4933 ztest_run_zdb(char *pool)
4934 {
4935 int status;
4936 char zdb[MAXPATHLEN + MAXNAMELEN + 20];
4937 char zbuf[1024];
4938 char *bin;
4939 char *ztest;
4940 char *isa;
4941 int isalen;
4942 FILE *fp;
4943
4944 (void) realpath(getexecname(), zdb);
4945
4946 /* zdb lives in /usr/sbin, while ztest lives in /usr/bin */
4947 bin = strstr(zdb, "/usr/bin/");
4948 ztest = strstr(bin, "/ztest");
4949 isa = bin + 8;
4950 isalen = ztest - isa;
4951 isa = strdup(isa);
4952 /* LINTED */
4953 (void) sprintf(bin,
4954 "/usr/sbin%.*s/zdb -bcc%s%s -U %s %s",
4955 isalen,
4956 isa,
4957 ztest_opts.zo_verbose >= 3 ? "s" : "",
4958 ztest_opts.zo_verbose >= 4 ? "v" : "",
4959 spa_config_path,
4960 pool);
4961 free(isa);
4962
4963 if (ztest_opts.zo_verbose >= 5)
4964 (void) printf("Executing %s\n", strstr(zdb, "zdb "));
4965
4966 fp = popen(zdb, "r");
4967
4968 while (fgets(zbuf, sizeof (zbuf), fp) != NULL)
4969 if (ztest_opts.zo_verbose >= 3)
4970 (void) printf("%s", zbuf);
4971
4972 status = pclose(fp);
4973
4974 if (status == 0)
4975 return;
4976
4977 ztest_dump_core = 0;
4978 if (WIFEXITED(status))
4979 fatal(0, "'%s' exit code %d", zdb, WEXITSTATUS(status));
4980 else
4981 fatal(0, "'%s' died with signal %d", zdb, WTERMSIG(status));
4982 }
4983
4984 static void
4985 ztest_walk_pool_directory(char *header)
4986 {
4987 spa_t *spa = NULL;
4988
4989 if (ztest_opts.zo_verbose >= 6)
4990 (void) printf("%s\n", header);
4991
4992 mutex_enter(&spa_namespace_lock);
4993 while ((spa = spa_next(spa)) != NULL)
4994 if (ztest_opts.zo_verbose >= 6)
4995 (void) printf("\t%s\n", spa_name(spa));
4996 mutex_exit(&spa_namespace_lock);
4997 }
4998
4999 static void
5000 ztest_spa_import_export(char *oldname, char *newname)
5001 {
5002 nvlist_t *config, *newconfig;
5003 uint64_t pool_guid;
5004 spa_t *spa;
5005
5006 if (ztest_opts.zo_verbose >= 4) {
5007 (void) printf("import/export: old = %s, new = %s\n",
5008 oldname, newname);
5009 }
5010
5011 /*
5012 * Clean up from previous runs.
5013 */
5014 (void) spa_destroy(newname);
5015
5016 /*
5017 * Get the pool's configuration and guid.
5018 */
5019 VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG));
5020
5021 /*
5022 * Kick off a scrub to tickle scrub/export races.
5023 */
5024 if (ztest_random(2) == 0)
5025 (void) spa_scan(spa, POOL_SCAN_SCRUB);
5026
5027 pool_guid = spa_guid(spa);
5028 spa_close(spa, FTAG);
5029
5030 ztest_walk_pool_directory("pools before export");
5031
5032 /*
5033 * Export it.
5034 */
5035 VERIFY3U(0, ==, spa_export(oldname, &config, B_FALSE, B_FALSE));
5036
5037 ztest_walk_pool_directory("pools after export");
5038
5039 /*
5040 * Try to import it.
5041 */
5042 newconfig = spa_tryimport(config);
5043 ASSERT(newconfig != NULL);
5044 nvlist_free(newconfig);
5045
5046 /*
5047 * Import it under the new name.
5048 */
5049 VERIFY3U(0, ==, spa_import(newname, config, NULL, 0));
5050
5051 ztest_walk_pool_directory("pools after import");
5052
5053 /*
5054 * Try to import it again -- should fail with EEXIST.
5055 */
5056 VERIFY3U(EEXIST, ==, spa_import(newname, config, NULL, 0));
5057
5058 /*
5059 * Try to import it under a different name -- should fail with EEXIST.
5060 */
5061 VERIFY3U(EEXIST, ==, spa_import(oldname, config, NULL, 0));
5062
5063 /*
5064 * Verify that the pool is no longer visible under the old name.
5065 */
5066 VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
5067
5068 /*
5069 * Verify that we can open and close the pool using the new name.
5070 */
5071 VERIFY3U(0, ==, spa_open(newname, &spa, FTAG));
5072 ASSERT(pool_guid == spa_guid(spa));
5073 spa_close(spa, FTAG);
5074
5075 nvlist_free(config);
5076 }
5077
5078 static void
5079 ztest_resume(spa_t *spa)
5080 {
5081 if (spa_suspended(spa) && ztest_opts.zo_verbose >= 6)
5082 (void) printf("resuming from suspended state\n");
5083 spa_vdev_state_enter(spa, SCL_NONE);
5084 vdev_clear(spa, NULL);
5085 (void) spa_vdev_state_exit(spa, NULL, 0);
5086 (void) zio_resume(spa);
5087 }
5088
5089 static void *
5090 ztest_resume_thread(void *arg)
5091 {
5092 spa_t *spa = arg;
5093
5094 while (!ztest_exiting) {
5095 if (spa_suspended(spa))
5096 ztest_resume(spa);
5097 (void) poll(NULL, 0, 100);
5098 }
5099 return (NULL);
5100 }
5101
5102 static void *
5103 ztest_deadman_thread(void *arg)
5104 {
5105 ztest_shared_t *zs = arg;
5106 int grace = 300;
5107 hrtime_t delta;
5108
5109 delta = (zs->zs_thread_stop - zs->zs_thread_start) / NANOSEC + grace;
5110
5111 (void) poll(NULL, 0, (int)(1000 * delta));
5112
5113 fatal(0, "failed to complete within %d seconds of deadline", grace);
5114
5115 return (NULL);
5116 }
5117
5118 static void
5119 ztest_execute(int test, ztest_info_t *zi, uint64_t id)
5120 {
5121 ztest_ds_t *zd = &ztest_ds[id % ztest_opts.zo_datasets];
5122 ztest_shared_callstate_t *zc = ZTEST_GET_SHARED_CALLSTATE(test);
5123 hrtime_t functime = gethrtime();
5124
5125 for (int i = 0; i < zi->zi_iters; i++)
5126 zi->zi_func(zd, id);
5127
5128 functime = gethrtime() - functime;
5129
5130 atomic_add_64(&zc->zc_count, 1);
5131 atomic_add_64(&zc->zc_time, functime);
5132
5133 if (ztest_opts.zo_verbose >= 4) {
5134 Dl_info dli;
5135 (void) dladdr((void *)zi->zi_func, &dli);
5136 (void) printf("%6.2f sec in %s\n",
5137 (double)functime / NANOSEC, dli.dli_sname);
5138 }
5139 }
5140
5141 static void *
5142 ztest_thread(void *arg)
5143 {
5144 int rand;
5145 uint64_t id = (uintptr_t)arg;
5146 ztest_shared_t *zs = ztest_shared;
5147 uint64_t call_next;
5148 hrtime_t now;
5149 ztest_info_t *zi;
5150 ztest_shared_callstate_t *zc;
5151
5152 while ((now = gethrtime()) < zs->zs_thread_stop) {
5153 /*
5154 * See if it's time to force a crash.
5155 */
5156 if (now > zs->zs_thread_kill)
5157 ztest_kill(zs);
5158
5159 /*
5160 * If we're getting ENOSPC with some regularity, stop.
5161 */
5162 if (zs->zs_enospc_count > 10)
5163 break;
5164
5165 /*
5166 * Pick a random function to execute.
5167 */
5168 rand = ztest_random(ZTEST_FUNCS);
5169 zi = &ztest_info[rand];
5170 zc = ZTEST_GET_SHARED_CALLSTATE(rand);
5171 call_next = zc->zc_next;
5172
5173 if (now >= call_next &&
5174 atomic_cas_64(&zc->zc_next, call_next, call_next +
5175 ztest_random(2 * zi->zi_interval[0] + 1)) == call_next) {
5176 ztest_execute(rand, zi, id);
5177 }
5178 }
5179
5180 return (NULL);
5181 }
5182
5183 static void
5184 ztest_dataset_name(char *dsname, char *pool, int d)
5185 {
5186 (void) snprintf(dsname, MAXNAMELEN, "%s/ds_%d", pool, d);
5187 }
5188
5189 static void
5190 ztest_dataset_destroy(int d)
5191 {
5192 char name[MAXNAMELEN];
5193
5194 ztest_dataset_name(name, ztest_opts.zo_pool, d);
5195
5196 if (ztest_opts.zo_verbose >= 3)
5197 (void) printf("Destroying %s to free up space\n", name);
5198
5199 /*
5200 * Cleanup any non-standard clones and snapshots. In general,
5201 * ztest thread t operates on dataset (t % zopt_datasets),
5202 * so there may be more than one thing to clean up.
5203 */
5204 for (int t = d; t < ztest_opts.zo_threads;
5205 t += ztest_opts.zo_datasets) {
5206 ztest_dsl_dataset_cleanup(name, t);
5207 }
5208
5209 (void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
5210 DS_FIND_SNAPSHOTS | DS_FIND_CHILDREN);
5211 }
5212
5213 static void
5214 ztest_dataset_dirobj_verify(ztest_ds_t *zd)
5215 {
5216 uint64_t usedobjs, dirobjs, scratch;
5217
5218 /*
5219 * ZTEST_DIROBJ is the object directory for the entire dataset.
5220 * Therefore, the number of objects in use should equal the
5221 * number of ZTEST_DIROBJ entries, +1 for ZTEST_DIROBJ itself.
5222 * If not, we have an object leak.
5223 *
5224 * Note that we can only check this in ztest_dataset_open(),
5225 * when the open-context and syncing-context values agree.
5226 * That's because zap_count() returns the open-context value,
5227 * while dmu_objset_space() returns the rootbp fill count.
5228 */
5229 VERIFY3U(0, ==, zap_count(zd->zd_os, ZTEST_DIROBJ, &dirobjs));
5230 dmu_objset_space(zd->zd_os, &scratch, &scratch, &usedobjs, &scratch);
5231 ASSERT3U(dirobjs + 1, ==, usedobjs);
5232 }
5233
5234 static int
5235 ztest_dataset_open(int d)
5236 {
5237 ztest_ds_t *zd = &ztest_ds[d];
5238 uint64_t committed_seq = ZTEST_GET_SHARED_DS(d)->zd_seq;
5239 objset_t *os;
5240 zilog_t *zilog;
5241 char name[MAXNAMELEN];
5242 int error;
5243
5244 ztest_dataset_name(name, ztest_opts.zo_pool, d);
5245
5246 (void) rw_rdlock(&ztest_name_lock);
5247
5248 error = ztest_dataset_create(name);
5249 if (error == ENOSPC) {
5250 (void) rw_unlock(&ztest_name_lock);
5251 ztest_record_enospc(FTAG);
5252 return (error);
5253 }
5254 ASSERT(error == 0 || error == EEXIST);
5255
5256 VERIFY3U(dmu_objset_hold(name, zd, &os), ==, 0);
5257 (void) rw_unlock(&ztest_name_lock);
5258
5259 ztest_zd_init(zd, ZTEST_GET_SHARED_DS(d), os);
5260
5261 zilog = zd->zd_zilog;
5262
5263 if (zilog->zl_header->zh_claim_lr_seq != 0 &&
5264 zilog->zl_header->zh_claim_lr_seq < committed_seq)
5265 fatal(0, "missing log records: claimed %llu < committed %llu",
5266 zilog->zl_header->zh_claim_lr_seq, committed_seq);
5267
5268 ztest_dataset_dirobj_verify(zd);
5269
5270 zil_replay(os, zd, ztest_replay_vector);
5271
5272 ztest_dataset_dirobj_verify(zd);
5273
5274 if (ztest_opts.zo_verbose >= 6)
5275 (void) printf("%s replay %llu blocks, %llu records, seq %llu\n",
5276 zd->zd_name,
5277 (u_longlong_t)zilog->zl_parse_blk_count,
5278 (u_longlong_t)zilog->zl_parse_lr_count,
5279 (u_longlong_t)zilog->zl_replaying_seq);
5280
5281 zilog = zil_open(os, ztest_get_data);
5282
5283 if (zilog->zl_replaying_seq != 0 &&
5284 zilog->zl_replaying_seq < committed_seq)
5285 fatal(0, "missing log records: replayed %llu < committed %llu",
5286 zilog->zl_replaying_seq, committed_seq);
5287
5288 return (0);
5289 }
5290
5291 static void
5292 ztest_dataset_close(int d)
5293 {
5294 ztest_ds_t *zd = &ztest_ds[d];
5295
5296 zil_close(zd->zd_zilog);
5297 dmu_objset_rele(zd->zd_os, zd);
5298
5299 ztest_zd_fini(zd);
5300 }
5301
5302 /*
5303 * Kick off threads to run tests on all datasets in parallel.
5304 */
5305 static void
5306 ztest_run(ztest_shared_t *zs)
5307 {
5308 thread_t *tid;
5309 spa_t *spa;
5310 objset_t *os;
5311 thread_t resume_tid;
5312 int error;
5313
5314 ztest_exiting = B_FALSE;
5315
5316 /*
5317 * Initialize parent/child shared state.
5318 */
5319 VERIFY(_mutex_init(&ztest_vdev_lock, USYNC_THREAD, NULL) == 0);
5320 VERIFY(rwlock_init(&ztest_name_lock, USYNC_THREAD, NULL) == 0);
5321
5322 zs->zs_thread_start = gethrtime();
5323 zs->zs_thread_stop =
5324 zs->zs_thread_start + ztest_opts.zo_passtime * NANOSEC;
5325 zs->zs_thread_stop = MIN(zs->zs_thread_stop, zs->zs_proc_stop);
5326 zs->zs_thread_kill = zs->zs_thread_stop;
5327 if (ztest_random(100) < ztest_opts.zo_killrate) {
5328 zs->zs_thread_kill -=
5329 ztest_random(ztest_opts.zo_passtime * NANOSEC);
5330 }
5331
5332 (void) _mutex_init(&zcl.zcl_callbacks_lock, USYNC_THREAD, NULL);
5333
5334 list_create(&zcl.zcl_callbacks, sizeof (ztest_cb_data_t),
5335 offsetof(ztest_cb_data_t, zcd_node));
5336
5337 /*
5338 * Open our pool.
5339 */
5340 kernel_init(FREAD | FWRITE);
5341 VERIFY(spa_open(ztest_opts.zo_pool, &spa, FTAG) == 0);
5342 spa->spa_debug = B_TRUE;
5343 ztest_spa = spa;
5344
5345 VERIFY3U(0, ==, dmu_objset_hold(ztest_opts.zo_pool, FTAG, &os));
5346 zs->zs_guid = dmu_objset_fsid_guid(os);
5347 dmu_objset_rele(os, FTAG);
5348
5349 spa->spa_dedup_ditto = 2 * ZIO_DEDUPDITTO_MIN;
5350
5351 /*
5352 * We don't expect the pool to suspend unless maxfaults == 0,
5353 * in which case ztest_fault_inject() temporarily takes away
5354 * the only valid replica.
5355 */
5356 if (MAXFAULTS() == 0)
5357 spa->spa_failmode = ZIO_FAILURE_MODE_WAIT;
5358 else
5359 spa->spa_failmode = ZIO_FAILURE_MODE_PANIC;
5360
5361 /*
5362 * Create a thread to periodically resume suspended I/O.
5363 */
5364 VERIFY(thr_create(0, 0, ztest_resume_thread, spa, THR_BOUND,
5365 &resume_tid) == 0);
5366
5367 /*
5368 * Create a deadman thread to abort() if we hang.
5369 */
5370 VERIFY(thr_create(0, 0, ztest_deadman_thread, zs, THR_BOUND,
5371 NULL) == 0);
5372
5373 /*
5374 * Verify that we can safely inquire about about any object,
5375 * whether it's allocated or not. To make it interesting,
5376 * we probe a 5-wide window around each power of two.
5377 * This hits all edge cases, including zero and the max.
5378 */
5379 for (int t = 0; t < 64; t++) {
5380 for (int d = -5; d <= 5; d++) {
5381 error = dmu_object_info(spa->spa_meta_objset,
5382 (1ULL << t) + d, NULL);
5383 ASSERT(error == 0 || error == ENOENT ||
5384 error == EINVAL);
5385 }
5386 }
5387
5388 /*
5389 * If we got any ENOSPC errors on the previous run, destroy something.
5390 */
5391 if (zs->zs_enospc_count != 0) {
5392 int d = ztest_random(ztest_opts.zo_datasets);
5393 ztest_dataset_destroy(d);
5394 }
5395 zs->zs_enospc_count = 0;
5396
5397 tid = umem_zalloc(ztest_opts.zo_threads * sizeof (thread_t),
5398 UMEM_NOFAIL);
5399
5400 if (ztest_opts.zo_verbose >= 4)
5401 (void) printf("starting main threads...\n");
5402
5403 /*
5404 * Kick off all the tests that run in parallel.
5405 */
5406 for (int t = 0; t < ztest_opts.zo_threads; t++) {
5407 if (t < ztest_opts.zo_datasets &&
5408 ztest_dataset_open(t) != 0)
5409 return;
5410 VERIFY(thr_create(0, 0, ztest_thread, (void *)(uintptr_t)t,
5411 THR_BOUND, &tid[t]) == 0);
5412 }
5413
5414 /*
5415 * Wait for all of the tests to complete. We go in reverse order
5416 * so we don't close datasets while threads are still using them.
5417 */
5418 for (int t = ztest_opts.zo_threads - 1; t >= 0; t--) {
5419 VERIFY(thr_join(tid[t], NULL, NULL) == 0);
5420 if (t < ztest_opts.zo_datasets)
5421 ztest_dataset_close(t);
5422 }
5423
5424 txg_wait_synced(spa_get_dsl(spa), 0);
5425
5426 zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(spa));
5427 zs->zs_space = metaslab_class_get_space(spa_normal_class(spa));
5428
5429 umem_free(tid, ztest_opts.zo_threads * sizeof (thread_t));
5430
5431 /* Kill the resume thread */
5432 ztest_exiting = B_TRUE;
5433 VERIFY(thr_join(resume_tid, NULL, NULL) == 0);
5434 ztest_resume(spa);
5435
5436 /*
5437 * Right before closing the pool, kick off a bunch of async I/O;
5438 * spa_close() should wait for it to complete.
5439 */
5440 for (uint64_t object = 1; object < 50; object++)
5441 dmu_prefetch(spa->spa_meta_objset, object, 0, 1ULL << 20);
5442
5443 spa_close(spa, FTAG);
5444
5445 /*
5446 * Verify that we can loop over all pools.
5447 */
5448 mutex_enter(&spa_namespace_lock);
5449 for (spa = spa_next(NULL); spa != NULL; spa = spa_next(spa))
5450 if (ztest_opts.zo_verbose > 3)
5451 (void) printf("spa_next: found %s\n", spa_name(spa));
5452 mutex_exit(&spa_namespace_lock);
5453
5454 /*
5455 * Verify that we can export the pool and reimport it under a
5456 * different name.
5457 */
5458 if (ztest_random(2) == 0) {
5459 char name[MAXNAMELEN];
5460 (void) snprintf(name, MAXNAMELEN, "%s_import",
5461 ztest_opts.zo_pool);
5462 ztest_spa_import_export(ztest_opts.zo_pool, name);
5463 ztest_spa_import_export(name, ztest_opts.zo_pool);
5464 }
5465
5466 kernel_fini();
5467
5468 list_destroy(&zcl.zcl_callbacks);
5469
5470 (void) _mutex_destroy(&zcl.zcl_callbacks_lock);
5471
5472 (void) rwlock_destroy(&ztest_name_lock);
5473 (void) _mutex_destroy(&ztest_vdev_lock);
5474 }
5475
5476 static void
5477 ztest_freeze(void)
5478 {
5479 ztest_ds_t *zd = &ztest_ds[0];
5480 spa_t *spa;
5481 int numloops = 0;
5482
5483 if (ztest_opts.zo_verbose >= 3)
5484 (void) printf("testing spa_freeze()...\n");
5485
5486 kernel_init(FREAD | FWRITE);
5487 VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
5488 VERIFY3U(0, ==, ztest_dataset_open(0));
5489
5490 /*
5491 * Force the first log block to be transactionally allocated.
5492 * We have to do this before we freeze the pool -- otherwise
5493 * the log chain won't be anchored.
5494 */
5495 while (BP_IS_HOLE(&zd->zd_zilog->zl_header->zh_log)) {
5496 ztest_dmu_object_alloc_free(zd, 0);
5497 zil_commit(zd->zd_zilog, 0);
5498 }
5499
5500 txg_wait_synced(spa_get_dsl(spa), 0);
5501
5502 /*
5503 * Freeze the pool. This stops spa_sync() from doing anything,
5504 * so that the only way to record changes from now on is the ZIL.
5505 */
5506 spa_freeze(spa);
5507
5508 /*
5509 * Run tests that generate log records but don't alter the pool config
5510 * or depend on DSL sync tasks (snapshots, objset create/destroy, etc).
5511 * We do a txg_wait_synced() after each iteration to force the txg
5512 * to increase well beyond the last synced value in the uberblock.
5513 * The ZIL should be OK with that.
5514 */
5515 while (ztest_random(10) != 0 &&
5516 numloops++ < ztest_opts.zo_maxloops) {
5517 ztest_dmu_write_parallel(zd, 0);
5518 ztest_dmu_object_alloc_free(zd, 0);
5519 txg_wait_synced(spa_get_dsl(spa), 0);
5520 }
5521
5522 /*
5523 * Commit all of the changes we just generated.
5524 */
5525 zil_commit(zd->zd_zilog, 0);
5526 txg_wait_synced(spa_get_dsl(spa), 0);
5527
5528 /*
5529 * Close our dataset and close the pool.
5530 */
5531 ztest_dataset_close(0);
5532 spa_close(spa, FTAG);
5533 kernel_fini();
5534
5535 /*
5536 * Open and close the pool and dataset to induce log replay.
5537 */
5538 kernel_init(FREAD | FWRITE);
5539 VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
5540 VERIFY3U(0, ==, ztest_dataset_open(0));
5541 ztest_dataset_close(0);
5542 spa_close(spa, FTAG);
5543 kernel_fini();
5544 }
5545
5546 void
5547 print_time(hrtime_t t, char *timebuf)
5548 {
5549 hrtime_t s = t / NANOSEC;
5550 hrtime_t m = s / 60;
5551 hrtime_t h = m / 60;
5552 hrtime_t d = h / 24;
5553
5554 s -= m * 60;
5555 m -= h * 60;
5556 h -= d * 24;
5557
5558 timebuf[0] = '\0';
5559
5560 if (d)
5561 (void) sprintf(timebuf,
5562 "%llud%02lluh%02llum%02llus", d, h, m, s);
5563 else if (h)
5564 (void) sprintf(timebuf, "%lluh%02llum%02llus", h, m, s);
5565 else if (m)
5566 (void) sprintf(timebuf, "%llum%02llus", m, s);
5567 else
5568 (void) sprintf(timebuf, "%llus", s);
5569 }
5570
5571 static nvlist_t *
5572 make_random_props()
5573 {
5574 nvlist_t *props;
5575
5576 VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
5577 if (ztest_random(2) == 0)
5578 return (props);
5579 VERIFY(nvlist_add_uint64(props, "autoreplace", 1) == 0);
5580
5581 return (props);
5582 }
5583
5584 /*
5585 * Create a storage pool with the given name and initial vdev size.
5586 * Then test spa_freeze() functionality.
5587 */
5588 static void
5589 ztest_init(ztest_shared_t *zs)
5590 {
5591 spa_t *spa;
5592 nvlist_t *nvroot, *props;
5593
5594 VERIFY(_mutex_init(&ztest_vdev_lock, USYNC_THREAD, NULL) == 0);
5595 VERIFY(rwlock_init(&ztest_name_lock, USYNC_THREAD, NULL) == 0);
5596
5597 kernel_init(FREAD | FWRITE);
5598
5599 /*
5600 * Create the storage pool.
5601 */
5602 (void) spa_destroy(ztest_opts.zo_pool);
5603 ztest_shared->zs_vdev_next_leaf = 0;
5604 zs->zs_splits = 0;
5605 zs->zs_mirrors = ztest_opts.zo_mirrors;
5606 nvroot = make_vdev_root(NULL, NULL, ztest_opts.zo_vdev_size, 0,
5607 0, ztest_opts.zo_raidz, zs->zs_mirrors, 1);
5608 props = make_random_props();
5609 for (int i = 0; i < SPA_FEATURES; i++) {
5610 char buf[1024];
5611 (void) snprintf(buf, sizeof (buf), "feature@%s",
5612 spa_feature_table[i].fi_uname);
5613 VERIFY3U(0, ==, nvlist_add_uint64(props, buf, 0));
5614 }
5615 VERIFY3U(0, ==, spa_create(ztest_opts.zo_pool, nvroot, props,
5616 NULL, NULL));
5617 nvlist_free(nvroot);
5618
5619 VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
5620 zs->zs_metaslab_sz =
5621 1ULL << spa->spa_root_vdev->vdev_child[0]->vdev_ms_shift;
5622
5623 spa_close(spa, FTAG);
5624
5625 kernel_fini();
5626
5627 ztest_run_zdb(ztest_opts.zo_pool);
5628
5629 ztest_freeze();
5630
5631 ztest_run_zdb(ztest_opts.zo_pool);
5632
5633 (void) rwlock_destroy(&ztest_name_lock);
5634 (void) _mutex_destroy(&ztest_vdev_lock);
5635 }
5636
5637 static void
5638 setup_fds(void)
5639 {
5640 int fd;
5641
5642 char *tmp = tempnam(NULL, NULL);
5643 fd = open(tmp, O_RDWR | O_CREAT, 0700);
5644 ASSERT3U(fd, ==, ZTEST_FD_DATA);
5645 (void) unlink(tmp);
5646 free(tmp);
5647
5648 fd = open("/dev/urandom", O_RDONLY);
5649 ASSERT3U(fd, ==, ZTEST_FD_RAND);
5650 }
5651
5652 static int
5653 shared_data_size(ztest_shared_hdr_t *hdr)
5654 {
5655 int size;
5656
5657 size = hdr->zh_hdr_size;
5658 size += hdr->zh_opts_size;
5659 size += hdr->zh_size;
5660 size += hdr->zh_stats_size * hdr->zh_stats_count;
5661 size += hdr->zh_ds_size * hdr->zh_ds_count;
5662
5663 return (size);
5664 }
5665
5666 static void
5667 setup_hdr(void)
5668 {
5669 int size;
5670 ztest_shared_hdr_t *hdr;
5671
5672 hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
5673 PROT_READ | PROT_WRITE, MAP_SHARED, ZTEST_FD_DATA, 0);
5674 ASSERT(hdr != MAP_FAILED);
5675
5676 VERIFY3U(0, ==, ftruncate(ZTEST_FD_DATA, sizeof (ztest_shared_hdr_t)));
5677
5678 hdr->zh_hdr_size = sizeof (ztest_shared_hdr_t);
5679 hdr->zh_opts_size = sizeof (ztest_shared_opts_t);
5680 hdr->zh_size = sizeof (ztest_shared_t);
5681 hdr->zh_stats_size = sizeof (ztest_shared_callstate_t);
5682 hdr->zh_stats_count = ZTEST_FUNCS;
5683 hdr->zh_ds_size = sizeof (ztest_shared_ds_t);
5684 hdr->zh_ds_count = ztest_opts.zo_datasets;
5685
5686 size = shared_data_size(hdr);
5687 VERIFY3U(0, ==, ftruncate(ZTEST_FD_DATA, size));
5688
5689 (void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
5690 }
5691
5692 static void
5693 setup_data(void)
5694 {
5695 int size, offset;
5696 ztest_shared_hdr_t *hdr;
5697 uint8_t *buf;
5698
5699 hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
5700 PROT_READ, MAP_SHARED, ZTEST_FD_DATA, 0);
5701 ASSERT(hdr != MAP_FAILED);
5702
5703 size = shared_data_size(hdr);
5704
5705 (void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
5706 hdr = ztest_shared_hdr = (void *)mmap(0, P2ROUNDUP(size, getpagesize()),
5707 PROT_READ | PROT_WRITE, MAP_SHARED, ZTEST_FD_DATA, 0);
5708 ASSERT(hdr != MAP_FAILED);
5709 buf = (uint8_t *)hdr;
5710
5711 offset = hdr->zh_hdr_size;
5712 ztest_shared_opts = (void *)&buf[offset];
5713 offset += hdr->zh_opts_size;
5714 ztest_shared = (void *)&buf[offset];
5715 offset += hdr->zh_size;
5716 ztest_shared_callstate = (void *)&buf[offset];
5717 offset += hdr->zh_stats_size * hdr->zh_stats_count;
5718 ztest_shared_ds = (void *)&buf[offset];
5719 }
5720
5721 static boolean_t
5722 exec_child(char *cmd, char *libpath, boolean_t ignorekill, int *statusp)
5723 {
5724 pid_t pid;
5725 int status;
5726 char cmdbuf[MAXPATHLEN];
5727
5728 pid = fork();
5729
5730 if (cmd == NULL) {
5731 (void) strlcpy(cmdbuf, getexecname(), sizeof (cmdbuf));
5732 cmd = cmdbuf;
5733 }
5734
5735 if (pid == -1)
5736 fatal(1, "fork failed");
5737
5738 if (pid == 0) { /* child */
5739 char *emptyargv[2] = { cmd, NULL };
5740
5741 struct rlimit rl = { 1024, 1024 };
5742 (void) setrlimit(RLIMIT_NOFILE, &rl);
5743 (void) enable_extended_FILE_stdio(-1, -1);
5744 if (libpath != NULL)
5745 VERIFY(0 == setenv("LD_LIBRARY_PATH", libpath, 1));
5746 (void) execv(cmd, emptyargv);
5747 ztest_dump_core = B_FALSE;
5748 fatal(B_TRUE, "exec failed: %s", cmd);
5749 }
5750
5751 while (waitpid(pid, &status, 0) != pid)
5752 continue;
5753 if (statusp != NULL)
5754 *statusp = status;
5755
5756 if (WIFEXITED(status)) {
5757 if (WEXITSTATUS(status) != 0) {
5758 (void) fprintf(stderr, "child exited with code %d\n",
5759 WEXITSTATUS(status));
5760 exit(2);
5761 }
5762 return (B_FALSE);
5763 } else if (WIFSIGNALED(status)) {
5764 if (!ignorekill || WTERMSIG(status) != SIGKILL) {
5765 (void) fprintf(stderr, "child died with signal %d\n",
5766 WTERMSIG(status));
5767 exit(3);
5768 }
5769 return (B_TRUE);
5770 } else {
5771 (void) fprintf(stderr, "something strange happened to child\n");
5772 exit(4);
5773 /* NOTREACHED */
5774 }
5775 }
5776
5777 static void
5778 ztest_run_init(void)
5779 {
5780 ztest_shared_t *zs = ztest_shared;
5781
5782 ASSERT(ztest_opts.zo_init != 0);
5783
5784 /*
5785 * Blow away any existing copy of zpool.cache
5786 */
5787 (void) remove(spa_config_path);
5788
5789 /*
5790 * Create and initialize our storage pool.
5791 */
5792 for (int i = 1; i <= ztest_opts.zo_init; i++) {
5793 bzero(zs, sizeof (ztest_shared_t));
5794 if (ztest_opts.zo_verbose >= 3 &&
5795 ztest_opts.zo_init != 1) {
5796 (void) printf("ztest_init(), pass %d\n", i);
5797 }
5798 ztest_init(zs);
5799 }
5800 }
5801
5802 int
5803 main(int argc, char **argv)
5804 {
5805 int kills = 0;
5806 int iters = 0;
5807 int older = 0;
5808 int newer = 0;
5809 ztest_shared_t *zs;
5810 ztest_info_t *zi;
5811 ztest_shared_callstate_t *zc;
5812 char timebuf[100];
5813 char numbuf[6];
5814 spa_t *spa;
5815 char cmd[MAXNAMELEN];
5816 boolean_t hasalt;
5817
5818 boolean_t ischild = (0 == lseek(ZTEST_FD_DATA, 0, SEEK_CUR));
5819 ASSERT(ischild || errno == EBADF);
5820
5821 (void) setvbuf(stdout, NULL, _IOLBF, 0);
5822
5823 if (!ischild) {
5824 process_options(argc, argv);
5825
5826 setup_fds();
5827 setup_hdr();
5828 setup_data();
5829 bcopy(&ztest_opts, ztest_shared_opts,
5830 sizeof (*ztest_shared_opts));
5831 } else {
5832 setup_data();
5833 bcopy(ztest_shared_opts, &ztest_opts, sizeof (ztest_opts));
5834 }
5835 ASSERT3U(ztest_opts.zo_datasets, ==, ztest_shared_hdr->zh_ds_count);
5836
5837 /* Override location of zpool.cache */
5838 (void) asprintf((char **)&spa_config_path, "%s/zpool.cache",
5839 ztest_opts.zo_dir);
5840
5841 ztest_ds = umem_alloc(ztest_opts.zo_datasets * sizeof (ztest_ds_t),
5842 UMEM_NOFAIL);
5843 zs = ztest_shared;
5844
5845 if (ischild) {
5846 metaslab_gang_bang = ztest_opts.zo_metaslab_gang_bang;
5847 metaslab_df_alloc_threshold =
5848 zs->zs_metaslab_df_alloc_threshold;
5849
5850 if (zs->zs_do_init)
5851 ztest_run_init();
5852 else
5853 ztest_run(zs);
5854 exit(0);
5855 }
5856
5857 hasalt = (strlen(ztest_opts.zo_alt_ztest) != 0);
5858
5859 if (ztest_opts.zo_verbose >= 1) {
5860 (void) printf("%llu vdevs, %d datasets, %d threads,"
5861 " %llu seconds...\n",
5862 (u_longlong_t)ztest_opts.zo_vdevs,
5863 ztest_opts.zo_datasets,
5864 ztest_opts.zo_threads,
5865 (u_longlong_t)ztest_opts.zo_time);
5866 }
5867
5868 (void) strlcpy(cmd, getexecname(), sizeof (cmd));
5869
5870 zs->zs_do_init = B_TRUE;
5871 if (strlen(ztest_opts.zo_alt_ztest) != 0) {
5872 if (ztest_opts.zo_verbose >= 1) {
5873 (void) printf("Executing older ztest for "
5874 "initialization: %s\n", ztest_opts.zo_alt_ztest);
5875 }
5876 VERIFY(!exec_child(ztest_opts.zo_alt_ztest,
5877 ztest_opts.zo_alt_libpath, B_FALSE, NULL));
5878 } else {
5879 VERIFY(!exec_child(NULL, NULL, B_FALSE, NULL));
5880 }
5881 zs->zs_do_init = B_FALSE;
5882
5883 zs->zs_proc_start = gethrtime();
5884 zs->zs_proc_stop = zs->zs_proc_start + ztest_opts.zo_time * NANOSEC;
5885
5886 for (int f = 0; f < ZTEST_FUNCS; f++) {
5887 zi = &ztest_info[f];
5888 zc = ZTEST_GET_SHARED_CALLSTATE(f);
5889 if (zs->zs_proc_start + zi->zi_interval[0] > zs->zs_proc_stop)
5890 zc->zc_next = UINT64_MAX;
5891 else
5892 zc->zc_next = zs->zs_proc_start +
5893 ztest_random(2 * zi->zi_interval[0] + 1);
5894 }
5895
5896 /*
5897 * Run the tests in a loop. These tests include fault injection
5898 * to verify that self-healing data works, and forced crashes
5899 * to verify that we never lose on-disk consistency.
5900 */
5901 while (gethrtime() < zs->zs_proc_stop) {
5902 int status;
5903 boolean_t killed;
5904
5905 /*
5906 * Initialize the workload counters for each function.
5907 */
5908 for (int f = 0; f < ZTEST_FUNCS; f++) {
5909 zc = ZTEST_GET_SHARED_CALLSTATE(f);
5910 zc->zc_count = 0;
5911 zc->zc_time = 0;
5912 }
5913
5914 /* Set the allocation switch size */
5915 zs->zs_metaslab_df_alloc_threshold =
5916 ztest_random(zs->zs_metaslab_sz / 4) + 1;
5917
5918 if (!hasalt || ztest_random(2) == 0) {
5919 if (hasalt && ztest_opts.zo_verbose >= 1) {
5920 (void) printf("Executing newer ztest: %s\n",
5921 cmd);
5922 }
5923 newer++;
5924 killed = exec_child(cmd, NULL, B_TRUE, &status);
5925 } else {
5926 if (hasalt && ztest_opts.zo_verbose >= 1) {
5927 (void) printf("Executing older ztest: %s\n",
5928 ztest_opts.zo_alt_ztest);
5929 }
5930 older++;
5931 killed = exec_child(ztest_opts.zo_alt_ztest,
5932 ztest_opts.zo_alt_libpath, B_TRUE, &status);
5933 }
5934
5935 if (killed)
5936 kills++;
5937 iters++;
5938
5939 if (ztest_opts.zo_verbose >= 1) {
5940 hrtime_t now = gethrtime();
5941
5942 now = MIN(now, zs->zs_proc_stop);
5943 print_time(zs->zs_proc_stop - now, timebuf);
5944 nicenum(zs->zs_space, numbuf);
5945
5946 (void) printf("Pass %3d, %8s, %3llu ENOSPC, "
5947 "%4.1f%% of %5s used, %3.0f%% done, %8s to go\n",
5948 iters,
5949 WIFEXITED(status) ? "Complete" : "SIGKILL",
5950 (u_longlong_t)zs->zs_enospc_count,
5951 100.0 * zs->zs_alloc / zs->zs_space,
5952 numbuf,
5953 100.0 * (now - zs->zs_proc_start) /
5954 (ztest_opts.zo_time * NANOSEC), timebuf);
5955 }
5956
5957 if (ztest_opts.zo_verbose >= 2) {
5958 (void) printf("\nWorkload summary:\n\n");
5959 (void) printf("%7s %9s %s\n",
5960 "Calls", "Time", "Function");
5961 (void) printf("%7s %9s %s\n",
5962 "-----", "----", "--------");
5963 for (int f = 0; f < ZTEST_FUNCS; f++) {
5964 Dl_info dli;
5965
5966 zi = &ztest_info[f];
5967 zc = ZTEST_GET_SHARED_CALLSTATE(f);
5968 print_time(zc->zc_time, timebuf);
5969 (void) dladdr((void *)zi->zi_func, &dli);
5970 (void) printf("%7llu %9s %s\n",
5971 (u_longlong_t)zc->zc_count, timebuf,
5972 dli.dli_sname);
5973 }
5974 (void) printf("\n");
5975 }
5976
5977 /*
5978 * It's possible that we killed a child during a rename test,
5979 * in which case we'll have a 'ztest_tmp' pool lying around
5980 * instead of 'ztest'. Do a blind rename in case this happened.
5981 */
5982 kernel_init(FREAD);
5983 if (spa_open(ztest_opts.zo_pool, &spa, FTAG) == 0) {
5984 spa_close(spa, FTAG);
5985 } else {
5986 char tmpname[MAXNAMELEN];
5987 kernel_fini();
5988 kernel_init(FREAD | FWRITE);
5989 (void) snprintf(tmpname, sizeof (tmpname), "%s_tmp",
5990 ztest_opts.zo_pool);
5991 (void) spa_rename(tmpname, ztest_opts.zo_pool);
5992 }
5993 kernel_fini();
5994
5995 ztest_run_zdb(ztest_opts.zo_pool);
5996 }
5997
5998 if (ztest_opts.zo_verbose >= 1) {
5999 if (hasalt) {
6000 (void) printf("%d runs of older ztest: %s\n", older,
6001 ztest_opts.zo_alt_ztest);
6002 (void) printf("%d runs of newer ztest: %s\n", newer,
6003 cmd);
6004 }
6005 (void) printf("%d killed, %d completed, %.0f%% kill rate\n",
6006 kills, iters - kills, (100.0 * kills) / MAX(1, iters));
6007 }
6008
6009 return (0);
6010 }