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) 2011, 2016 by Delphix. All rights reserved.
24 * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
25 * Copyright (c) 2013 Steven Hartland. All rights reserved.
26 * Copyright (c) 2014 Integros [integros.com]
27 * Copyright 2017 Joyent, Inc.
28 */
29
30 /*
31 * The objective of this program is to provide a DMU/ZAP/SPA stress test
32 * that runs entirely in userland, is easy to use, and easy to extend.
33 *
34 * The overall design of the ztest program is as follows:
35 *
36 * (1) For each major functional area (e.g. adding vdevs to a pool,
37 * creating and destroying datasets, reading and writing objects, etc)
38 * we have a simple routine to test that functionality. These
39 * individual routines do not have to do anything "stressful".
40 *
41 * (2) We turn these simple functionality tests into a stress test by
42 * running them all in parallel, with as many threads as desired,
43 * and spread across as many datasets, objects, and vdevs as desired.
44 *
45 * (3) While all this is happening, we inject faults into the pool to
46 * verify that self-healing data really works.
47 *
48 * (4) Every time we open a dataset, we change its checksum and compression
49 * functions. Thus even individual objects vary from block to block
50 * in which checksum they use and whether they're compressed.
51 *
52 * (5) To verify that we never lose on-disk consistency after a crash,
53 * we run the entire test in a child of the main process.
54 * At random times, the child self-immolates with a SIGKILL.
55 * This is the software equivalent of pulling the power cord.
56 * The parent then runs the test again, using the existing
57 * storage pool, as many times as desired. If backwards compatibility
58 * testing is enabled ztest will sometimes run the "older" version
59 * of ztest after a SIGKILL.
60 *
61 * (6) To verify that we don't have future leaks or temporal incursions,
62 * many of the functional tests record the transaction group number
63 * as part of their data. When reading old data, they verify that
64 * the transaction group number is less than the current, open txg.
65 * If you add a new test, please do this if applicable.
66 *
67 * When run with no arguments, ztest runs for about five minutes and
68 * produces no output if successful. To get a little bit of information,
69 * specify -V. To get more information, specify -VV, and so on.
70 *
71 * To turn this into an overnight stress test, use -T to specify run time.
72 *
73 * You can ask more more vdevs [-v], datasets [-d], or threads [-t]
74 * to increase the pool capacity, fanout, and overall stress level.
75 *
76 * Use the -k option to set the desired frequency of kills.
77 *
78 * When ztest invokes itself it passes all relevant information through a
79 * temporary file which is mmap-ed in the child process. This allows shared
80 * memory to survive the exec syscall. The ztest_shared_hdr_t struct is always
81 * stored at offset 0 of this file and contains information on the size and
82 * number of shared structures in the file. The information stored in this file
83 * must remain backwards compatible with older versions of ztest so that
84 * ztest can invoke them during backwards compatibility testing (-B).
85 */
86
87 #include <sys/zfs_context.h>
88 #include <sys/spa.h>
89 #include <sys/dmu.h>
90 #include <sys/txg.h>
91 #include <sys/dbuf.h>
92 #include <sys/zap.h>
93 #include <sys/dmu_objset.h>
94 #include <sys/poll.h>
95 #include <sys/stat.h>
96 #include <sys/time.h>
97 #include <sys/wait.h>
98 #include <sys/mman.h>
99 #include <sys/resource.h>
100 #include <sys/zio.h>
101 #include <sys/zil.h>
102 #include <sys/zil_impl.h>
103 #include <sys/vdev_impl.h>
104 #include <sys/vdev_file.h>
105 #include <sys/spa_impl.h>
106 #include <sys/metaslab_impl.h>
107 #include <sys/dsl_prop.h>
108 #include <sys/dsl_dataset.h>
109 #include <sys/dsl_destroy.h>
110 #include <sys/dsl_scan.h>
111 #include <sys/zio_checksum.h>
112 #include <sys/refcount.h>
113 #include <sys/zfeature.h>
114 #include <sys/dsl_userhold.h>
115 #include <sys/abd.h>
116 #include <stdio.h>
117 #include <stdio_ext.h>
118 #include <stdlib.h>
119 #include <unistd.h>
120 #include <signal.h>
121 #include <umem.h>
122 #include <dlfcn.h>
123 #include <ctype.h>
124 #include <math.h>
125 #include <sys/fs/zfs.h>
126 #include <libnvpair.h>
127 #include <libcmdutils.h>
128
129 static int ztest_fd_data = -1;
130 static int ztest_fd_rand = -1;
131
132 typedef struct ztest_shared_hdr {
133 uint64_t zh_hdr_size;
134 uint64_t zh_opts_size;
135 uint64_t zh_size;
136 uint64_t zh_stats_size;
137 uint64_t zh_stats_count;
138 uint64_t zh_ds_size;
139 uint64_t zh_ds_count;
140 } ztest_shared_hdr_t;
141
142 static ztest_shared_hdr_t *ztest_shared_hdr;
143
144 typedef struct ztest_shared_opts {
145 char zo_pool[ZFS_MAX_DATASET_NAME_LEN];
146 char zo_dir[ZFS_MAX_DATASET_NAME_LEN];
147 char zo_alt_ztest[MAXNAMELEN];
148 char zo_alt_libpath[MAXNAMELEN];
149 uint64_t zo_vdevs;
150 uint64_t zo_vdevtime;
151 size_t zo_vdev_size;
152 int zo_ashift;
153 int zo_mirrors;
154 int zo_raidz;
155 int zo_raidz_parity;
156 int zo_datasets;
157 int zo_threads;
158 uint64_t zo_passtime;
159 uint64_t zo_killrate;
160 int zo_verbose;
161 int zo_init;
162 uint64_t zo_time;
163 uint64_t zo_maxloops;
164 uint64_t zo_metaslab_gang_bang;
165 } ztest_shared_opts_t;
166
167 static const ztest_shared_opts_t ztest_opts_defaults = {
168 .zo_pool = { 'z', 't', 'e', 's', 't', '\0' },
169 .zo_dir = { '/', 't', 'm', 'p', '\0' },
170 .zo_alt_ztest = { '\0' },
171 .zo_alt_libpath = { '\0' },
172 .zo_vdevs = 5,
173 .zo_ashift = SPA_MINBLOCKSHIFT,
174 .zo_mirrors = 2,
175 .zo_raidz = 4,
176 .zo_raidz_parity = 1,
177 .zo_vdev_size = SPA_MINDEVSIZE * 4, /* 256m default size */
178 .zo_datasets = 7,
179 .zo_threads = 23,
180 .zo_passtime = 60, /* 60 seconds */
181 .zo_killrate = 70, /* 70% kill rate */
182 .zo_verbose = 0,
183 .zo_init = 1,
184 .zo_time = 300, /* 5 minutes */
185 .zo_maxloops = 50, /* max loops during spa_freeze() */
186 .zo_metaslab_gang_bang = 32 << 10
187 };
188
189 extern uint64_t metaslab_gang_bang;
190 extern uint64_t metaslab_df_alloc_threshold;
191 extern uint64_t zfs_deadman_synctime_ms;
192 extern int metaslab_preload_limit;
193 extern boolean_t zfs_compressed_arc_enabled;
194 extern boolean_t zfs_abd_scatter_enabled;
195
196 static ztest_shared_opts_t *ztest_shared_opts;
197 static ztest_shared_opts_t ztest_opts;
198
199 typedef struct ztest_shared_ds {
200 uint64_t zd_seq;
201 } ztest_shared_ds_t;
202
203 static ztest_shared_ds_t *ztest_shared_ds;
204 #define ZTEST_GET_SHARED_DS(d) (&ztest_shared_ds[d])
205
206 #define BT_MAGIC 0x123456789abcdefULL
207 #define MAXFAULTS() \
208 (MAX(zs->zs_mirrors, 1) * (ztest_opts.zo_raidz_parity + 1) - 1)
209
210 enum ztest_io_type {
211 ZTEST_IO_WRITE_TAG,
212 ZTEST_IO_WRITE_PATTERN,
213 ZTEST_IO_WRITE_ZEROES,
214 ZTEST_IO_TRUNCATE,
215 ZTEST_IO_SETATTR,
216 ZTEST_IO_REWRITE,
217 ZTEST_IO_TYPES
218 };
219
220 typedef struct ztest_block_tag {
221 uint64_t bt_magic;
222 uint64_t bt_objset;
223 uint64_t bt_object;
224 uint64_t bt_offset;
225 uint64_t bt_gen;
226 uint64_t bt_txg;
227 uint64_t bt_crtxg;
228 } ztest_block_tag_t;
229
230 typedef struct bufwad {
231 uint64_t bw_index;
232 uint64_t bw_txg;
233 uint64_t bw_data;
234 } bufwad_t;
235
236 /*
237 * XXX -- fix zfs range locks to be generic so we can use them here.
238 */
239 typedef enum {
240 RL_READER,
241 RL_WRITER,
242 RL_APPEND
243 } rl_type_t;
244
245 typedef struct rll {
246 void *rll_writer;
247 int rll_readers;
248 mutex_t rll_lock;
249 cond_t rll_cv;
250 } rll_t;
251
252 typedef struct rl {
253 uint64_t rl_object;
254 uint64_t rl_offset;
255 uint64_t rl_size;
256 rll_t *rl_lock;
257 } rl_t;
258
259 #define ZTEST_RANGE_LOCKS 64
260 #define ZTEST_OBJECT_LOCKS 64
261
262 /*
263 * Object descriptor. Used as a template for object lookup/create/remove.
264 */
265 typedef struct ztest_od {
266 uint64_t od_dir;
267 uint64_t od_object;
268 dmu_object_type_t od_type;
269 dmu_object_type_t od_crtype;
270 uint64_t od_blocksize;
271 uint64_t od_crblocksize;
272 uint64_t od_gen;
273 uint64_t od_crgen;
274 char od_name[ZFS_MAX_DATASET_NAME_LEN];
275 } ztest_od_t;
276
277 /*
278 * Per-dataset state.
279 */
280 typedef struct ztest_ds {
281 ztest_shared_ds_t *zd_shared;
282 objset_t *zd_os;
283 rwlock_t zd_zilog_lock;
284 zilog_t *zd_zilog;
285 ztest_od_t *zd_od; /* debugging aid */
286 char zd_name[ZFS_MAX_DATASET_NAME_LEN];
287 mutex_t zd_dirobj_lock;
288 rll_t zd_object_lock[ZTEST_OBJECT_LOCKS];
289 rll_t zd_range_lock[ZTEST_RANGE_LOCKS];
290 } ztest_ds_t;
291
292 /*
293 * Per-iteration state.
294 */
295 typedef void ztest_func_t(ztest_ds_t *zd, uint64_t id);
296
297 typedef struct ztest_info {
298 ztest_func_t *zi_func; /* test function */
299 uint64_t zi_iters; /* iterations per execution */
300 uint64_t *zi_interval; /* execute every <interval> seconds */
301 } ztest_info_t;
302
303 typedef struct ztest_shared_callstate {
304 uint64_t zc_count; /* per-pass count */
305 uint64_t zc_time; /* per-pass time */
306 uint64_t zc_next; /* next time to call this function */
307 } ztest_shared_callstate_t;
308
309 static ztest_shared_callstate_t *ztest_shared_callstate;
310 #define ZTEST_GET_SHARED_CALLSTATE(c) (&ztest_shared_callstate[c])
311
312 /*
313 * Note: these aren't static because we want dladdr() to work.
314 */
315 ztest_func_t ztest_dmu_read_write;
316 ztest_func_t ztest_dmu_write_parallel;
317 ztest_func_t ztest_dmu_object_alloc_free;
318 ztest_func_t ztest_dmu_commit_callbacks;
319 ztest_func_t ztest_zap;
320 ztest_func_t ztest_zap_parallel;
321 ztest_func_t ztest_zil_commit;
322 ztest_func_t ztest_zil_remount;
323 ztest_func_t ztest_dmu_read_write_zcopy;
324 ztest_func_t ztest_dmu_objset_create_destroy;
325 ztest_func_t ztest_dmu_prealloc;
326 ztest_func_t ztest_fzap;
327 ztest_func_t ztest_dmu_snapshot_create_destroy;
328 ztest_func_t ztest_dsl_prop_get_set;
329 ztest_func_t ztest_spa_prop_get_set;
330 ztest_func_t ztest_spa_create_destroy;
331 ztest_func_t ztest_fault_inject;
332 ztest_func_t ztest_ddt_repair;
333 ztest_func_t ztest_dmu_snapshot_hold;
334 ztest_func_t ztest_spa_rename;
335 ztest_func_t ztest_scrub;
336 ztest_func_t ztest_dsl_dataset_promote_busy;
337 ztest_func_t ztest_vdev_attach_detach;
338 ztest_func_t ztest_vdev_LUN_growth;
339 ztest_func_t ztest_vdev_add_remove;
340 ztest_func_t ztest_vdev_aux_add_remove;
341 ztest_func_t ztest_split_pool;
342 ztest_func_t ztest_reguid;
343 ztest_func_t ztest_spa_upgrade;
344
345 uint64_t zopt_always = 0ULL * NANOSEC; /* all the time */
346 uint64_t zopt_incessant = 1ULL * NANOSEC / 10; /* every 1/10 second */
347 uint64_t zopt_often = 1ULL * NANOSEC; /* every second */
348 uint64_t zopt_sometimes = 10ULL * NANOSEC; /* every 10 seconds */
349 uint64_t zopt_rarely = 60ULL * NANOSEC; /* every 60 seconds */
350
351 ztest_info_t ztest_info[] = {
352 { ztest_dmu_read_write, 1, &zopt_always },
353 { ztest_dmu_write_parallel, 10, &zopt_always },
354 { ztest_dmu_object_alloc_free, 1, &zopt_always },
355 { ztest_dmu_commit_callbacks, 1, &zopt_always },
356 { ztest_zap, 30, &zopt_always },
357 { ztest_zap_parallel, 100, &zopt_always },
358 { ztest_split_pool, 1, &zopt_always },
359 { ztest_zil_commit, 1, &zopt_incessant },
360 { ztest_zil_remount, 1, &zopt_sometimes },
361 { ztest_dmu_read_write_zcopy, 1, &zopt_often },
362 { ztest_dmu_objset_create_destroy, 1, &zopt_often },
363 { ztest_dsl_prop_get_set, 1, &zopt_often },
364 { ztest_spa_prop_get_set, 1, &zopt_sometimes },
365 #if 0
366 { ztest_dmu_prealloc, 1, &zopt_sometimes },
367 #endif
368 { ztest_fzap, 1, &zopt_sometimes },
369 { ztest_dmu_snapshot_create_destroy, 1, &zopt_sometimes },
370 { ztest_spa_create_destroy, 1, &zopt_sometimes },
371 { ztest_fault_inject, 1, &zopt_sometimes },
372 { ztest_ddt_repair, 1, &zopt_sometimes },
373 { ztest_dmu_snapshot_hold, 1, &zopt_sometimes },
374 { ztest_reguid, 1, &zopt_rarely },
375 { ztest_spa_rename, 1, &zopt_rarely },
376 { ztest_scrub, 1, &zopt_rarely },
377 { ztest_spa_upgrade, 1, &zopt_rarely },
378 { ztest_dsl_dataset_promote_busy, 1, &zopt_rarely },
379 { ztest_vdev_attach_detach, 1, &zopt_sometimes },
380 { ztest_vdev_LUN_growth, 1, &zopt_rarely },
381 { ztest_vdev_add_remove, 1,
382 &ztest_opts.zo_vdevtime },
383 { ztest_vdev_aux_add_remove, 1,
384 &ztest_opts.zo_vdevtime },
385 };
386
387 #define ZTEST_FUNCS (sizeof (ztest_info) / sizeof (ztest_info_t))
388
389 /*
390 * The following struct is used to hold a list of uncalled commit callbacks.
391 * The callbacks are ordered by txg number.
392 */
393 typedef struct ztest_cb_list {
394 mutex_t zcl_callbacks_lock;
395 list_t zcl_callbacks;
396 } ztest_cb_list_t;
397
398 /*
399 * Stuff we need to share writably between parent and child.
400 */
401 typedef struct ztest_shared {
402 boolean_t zs_do_init;
403 hrtime_t zs_proc_start;
404 hrtime_t zs_proc_stop;
405 hrtime_t zs_thread_start;
406 hrtime_t zs_thread_stop;
407 hrtime_t zs_thread_kill;
408 uint64_t zs_enospc_count;
409 uint64_t zs_vdev_next_leaf;
410 uint64_t zs_vdev_aux;
411 uint64_t zs_alloc;
412 uint64_t zs_space;
413 uint64_t zs_splits;
414 uint64_t zs_mirrors;
415 uint64_t zs_metaslab_sz;
416 uint64_t zs_metaslab_df_alloc_threshold;
417 uint64_t zs_guid;
418 } ztest_shared_t;
419
420 #define ID_PARALLEL -1ULL
421
422 static char ztest_dev_template[] = "%s/%s.%llua";
423 static char ztest_aux_template[] = "%s/%s.%s.%llu";
424 ztest_shared_t *ztest_shared;
425
426 static spa_t *ztest_spa = NULL;
427 static ztest_ds_t *ztest_ds;
428
429 static mutex_t ztest_vdev_lock;
430
431 /*
432 * The ztest_name_lock protects the pool and dataset namespace used by
433 * the individual tests. To modify the namespace, consumers must grab
434 * this lock as writer. Grabbing the lock as reader will ensure that the
435 * namespace does not change while the lock is held.
436 */
437 static rwlock_t ztest_name_lock;
438
439 static boolean_t ztest_dump_core = B_TRUE;
440 static boolean_t ztest_exiting;
441
442 /* Global commit callback list */
443 static ztest_cb_list_t zcl;
444
445 enum ztest_object {
446 ZTEST_META_DNODE = 0,
447 ZTEST_DIROBJ,
448 ZTEST_OBJECTS
449 };
450
451 static void usage(boolean_t) __NORETURN;
452
453 /*
454 * These libumem hooks provide a reasonable set of defaults for the allocator's
455 * debugging facilities.
456 */
457 const char *
458 _umem_debug_init()
459 {
460 return ("default,verbose"); /* $UMEM_DEBUG setting */
461 }
462
463 const char *
464 _umem_logging_init(void)
465 {
466 return ("fail,contents"); /* $UMEM_LOGGING setting */
467 }
468
469 #define FATAL_MSG_SZ 1024
470
471 char *fatal_msg;
472
473 static void
474 fatal(int do_perror, char *message, ...)
475 {
476 va_list args;
477 int save_errno = errno;
478 char buf[FATAL_MSG_SZ];
479
480 (void) fflush(stdout);
481
482 va_start(args, message);
483 (void) sprintf(buf, "ztest: ");
484 /* LINTED */
485 (void) vsprintf(buf + strlen(buf), message, args);
486 va_end(args);
487 if (do_perror) {
488 (void) snprintf(buf + strlen(buf), FATAL_MSG_SZ - strlen(buf),
489 ": %s", strerror(save_errno));
490 }
491 (void) fprintf(stderr, "%s\n", buf);
492 fatal_msg = buf; /* to ease debugging */
493 if (ztest_dump_core)
494 abort();
495 exit(3);
496 }
497
498 static int
499 str2shift(const char *buf)
500 {
501 const char *ends = "BKMGTPEZ";
502 int i;
503
504 if (buf[0] == '\0')
505 return (0);
506 for (i = 0; i < strlen(ends); i++) {
507 if (toupper(buf[0]) == ends[i])
508 break;
509 }
510 if (i == strlen(ends)) {
511 (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n",
512 buf);
513 usage(B_FALSE);
514 }
515 if (buf[1] == '\0' || (toupper(buf[1]) == 'B' && buf[2] == '\0')) {
516 return (10*i);
517 }
518 (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n", buf);
519 usage(B_FALSE);
520 /* NOTREACHED */
521 }
522
523 static uint64_t
524 nicenumtoull(const char *buf)
525 {
526 char *end;
527 uint64_t val;
528
529 val = strtoull(buf, &end, 0);
530 if (end == buf) {
531 (void) fprintf(stderr, "ztest: bad numeric value: %s\n", buf);
532 usage(B_FALSE);
533 } else if (end[0] == '.') {
534 double fval = strtod(buf, &end);
535 fval *= pow(2, str2shift(end));
536 if (fval > UINT64_MAX) {
537 (void) fprintf(stderr, "ztest: value too large: %s\n",
538 buf);
539 usage(B_FALSE);
540 }
541 val = (uint64_t)fval;
542 } else {
543 int shift = str2shift(end);
544 if (shift >= 64 || (val << shift) >> shift != val) {
545 (void) fprintf(stderr, "ztest: value too large: %s\n",
546 buf);
547 usage(B_FALSE);
548 }
549 val <<= shift;
550 }
551 return (val);
552 }
553
554 static void
555 usage(boolean_t requested)
556 {
557 const ztest_shared_opts_t *zo = &ztest_opts_defaults;
558
559 char nice_vdev_size[NN_NUMBUF_SZ];
560 char nice_gang_bang[NN_NUMBUF_SZ];
561 FILE *fp = requested ? stdout : stderr;
562
563 nicenum(zo->zo_vdev_size, nice_vdev_size, sizeof (nice_vdev_size));
564 nicenum(zo->zo_metaslab_gang_bang, nice_gang_bang,
565 sizeof (nice_gang_bang));
566
567 (void) fprintf(fp, "Usage: %s\n"
568 "\t[-v vdevs (default: %llu)]\n"
569 "\t[-s size_of_each_vdev (default: %s)]\n"
570 "\t[-a alignment_shift (default: %d)] use 0 for random\n"
571 "\t[-m mirror_copies (default: %d)]\n"
572 "\t[-r raidz_disks (default: %d)]\n"
573 "\t[-R raidz_parity (default: %d)]\n"
574 "\t[-d datasets (default: %d)]\n"
575 "\t[-t threads (default: %d)]\n"
576 "\t[-g gang_block_threshold (default: %s)]\n"
577 "\t[-i init_count (default: %d)] initialize pool i times\n"
578 "\t[-k kill_percentage (default: %llu%%)]\n"
579 "\t[-p pool_name (default: %s)]\n"
580 "\t[-f dir (default: %s)] file directory for vdev files\n"
581 "\t[-V] verbose (use multiple times for ever more blather)\n"
582 "\t[-E] use existing pool instead of creating new one\n"
583 "\t[-T time (default: %llu sec)] total run time\n"
584 "\t[-F freezeloops (default: %llu)] max loops in spa_freeze()\n"
585 "\t[-P passtime (default: %llu sec)] time per pass\n"
586 "\t[-B alt_ztest (default: <none>)] alternate ztest path\n"
587 "\t[-o variable=value] ... set global variable to an unsigned\n"
588 "\t 32-bit integer value\n"
589 "\t[-h] (print help)\n"
590 "",
591 zo->zo_pool,
592 (u_longlong_t)zo->zo_vdevs, /* -v */
593 nice_vdev_size, /* -s */
594 zo->zo_ashift, /* -a */
595 zo->zo_mirrors, /* -m */
596 zo->zo_raidz, /* -r */
597 zo->zo_raidz_parity, /* -R */
598 zo->zo_datasets, /* -d */
599 zo->zo_threads, /* -t */
600 nice_gang_bang, /* -g */
601 zo->zo_init, /* -i */
602 (u_longlong_t)zo->zo_killrate, /* -k */
603 zo->zo_pool, /* -p */
604 zo->zo_dir, /* -f */
605 (u_longlong_t)zo->zo_time, /* -T */
606 (u_longlong_t)zo->zo_maxloops, /* -F */
607 (u_longlong_t)zo->zo_passtime);
608 exit(requested ? 0 : 1);
609 }
610
611 static void
612 process_options(int argc, char **argv)
613 {
614 char *path;
615 ztest_shared_opts_t *zo = &ztest_opts;
616
617 int opt;
618 uint64_t value;
619 char altdir[MAXNAMELEN] = { 0 };
620
621 bcopy(&ztest_opts_defaults, zo, sizeof (*zo));
622
623 while ((opt = getopt(argc, argv,
624 "v:s:a:m:r:R:d:t:g:i:k:p:f:VET:P:hF:B:o:")) != EOF) {
625 value = 0;
626 switch (opt) {
627 case 'v':
628 case 's':
629 case 'a':
630 case 'm':
631 case 'r':
632 case 'R':
633 case 'd':
634 case 't':
635 case 'g':
636 case 'i':
637 case 'k':
638 case 'T':
639 case 'P':
640 case 'F':
641 value = nicenumtoull(optarg);
642 }
643 switch (opt) {
644 case 'v':
645 zo->zo_vdevs = value;
646 break;
647 case 's':
648 zo->zo_vdev_size = MAX(SPA_MINDEVSIZE, value);
649 break;
650 case 'a':
651 zo->zo_ashift = value;
652 break;
653 case 'm':
654 zo->zo_mirrors = value;
655 break;
656 case 'r':
657 zo->zo_raidz = MAX(1, value);
658 break;
659 case 'R':
660 zo->zo_raidz_parity = MIN(MAX(value, 1), 3);
661 break;
662 case 'd':
663 zo->zo_datasets = MAX(1, value);
664 break;
665 case 't':
666 zo->zo_threads = MAX(1, value);
667 break;
668 case 'g':
669 zo->zo_metaslab_gang_bang = MAX(SPA_MINBLOCKSIZE << 1,
670 value);
671 break;
672 case 'i':
673 zo->zo_init = value;
674 break;
675 case 'k':
676 zo->zo_killrate = value;
677 break;
678 case 'p':
679 (void) strlcpy(zo->zo_pool, optarg,
680 sizeof (zo->zo_pool));
681 break;
682 case 'f':
683 path = realpath(optarg, NULL);
684 if (path == NULL) {
685 (void) fprintf(stderr, "error: %s: %s\n",
686 optarg, strerror(errno));
687 usage(B_FALSE);
688 } else {
689 (void) strlcpy(zo->zo_dir, path,
690 sizeof (zo->zo_dir));
691 }
692 break;
693 case 'V':
694 zo->zo_verbose++;
695 break;
696 case 'E':
697 zo->zo_init = 0;
698 break;
699 case 'T':
700 zo->zo_time = value;
701 break;
702 case 'P':
703 zo->zo_passtime = MAX(1, value);
704 break;
705 case 'F':
706 zo->zo_maxloops = MAX(1, value);
707 break;
708 case 'B':
709 (void) strlcpy(altdir, optarg, sizeof (altdir));
710 break;
711 case 'o':
712 if (set_global_var(optarg) != 0)
713 usage(B_FALSE);
714 break;
715 case 'h':
716 usage(B_TRUE);
717 break;
718 case '?':
719 default:
720 usage(B_FALSE);
721 break;
722 }
723 }
724
725 zo->zo_raidz_parity = MIN(zo->zo_raidz_parity, zo->zo_raidz - 1);
726
727 zo->zo_vdevtime =
728 (zo->zo_vdevs > 0 ? zo->zo_time * NANOSEC / zo->zo_vdevs :
729 UINT64_MAX >> 2);
730
731 if (strlen(altdir) > 0) {
732 char *cmd;
733 char *realaltdir;
734 char *bin;
735 char *ztest;
736 char *isa;
737 int isalen;
738
739 cmd = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
740 realaltdir = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
741
742 VERIFY(NULL != realpath(getexecname(), cmd));
743 if (0 != access(altdir, F_OK)) {
744 ztest_dump_core = B_FALSE;
745 fatal(B_TRUE, "invalid alternate ztest path: %s",
746 altdir);
747 }
748 VERIFY(NULL != realpath(altdir, realaltdir));
749
750 /*
751 * 'cmd' should be of the form "<anything>/usr/bin/<isa>/ztest".
752 * We want to extract <isa> to determine if we should use
753 * 32 or 64 bit binaries.
754 */
755 bin = strstr(cmd, "/usr/bin/");
756 ztest = strstr(bin, "/ztest");
757 isa = bin + 9;
758 isalen = ztest - isa;
759 (void) snprintf(zo->zo_alt_ztest, sizeof (zo->zo_alt_ztest),
760 "%s/usr/bin/%.*s/ztest", realaltdir, isalen, isa);
761 (void) snprintf(zo->zo_alt_libpath, sizeof (zo->zo_alt_libpath),
762 "%s/usr/lib/%.*s", realaltdir, isalen, isa);
763
764 if (0 != access(zo->zo_alt_ztest, X_OK)) {
765 ztest_dump_core = B_FALSE;
766 fatal(B_TRUE, "invalid alternate ztest: %s",
767 zo->zo_alt_ztest);
768 } else if (0 != access(zo->zo_alt_libpath, X_OK)) {
769 ztest_dump_core = B_FALSE;
770 fatal(B_TRUE, "invalid alternate lib directory %s",
771 zo->zo_alt_libpath);
772 }
773
774 umem_free(cmd, MAXPATHLEN);
775 umem_free(realaltdir, MAXPATHLEN);
776 }
777 }
778
779 static void
780 ztest_kill(ztest_shared_t *zs)
781 {
782 zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(ztest_spa));
783 zs->zs_space = metaslab_class_get_space(spa_normal_class(ztest_spa));
784
785 /*
786 * Before we kill off ztest, make sure that the config is updated.
787 * See comment above spa_config_sync().
788 */
789 mutex_enter(&spa_namespace_lock);
790 spa_config_sync(ztest_spa, B_FALSE, B_FALSE);
791 mutex_exit(&spa_namespace_lock);
792
793 zfs_dbgmsg_print(FTAG);
794 (void) kill(getpid(), SIGKILL);
795 }
796
797 static uint64_t
798 ztest_random(uint64_t range)
799 {
800 uint64_t r;
801
802 ASSERT3S(ztest_fd_rand, >=, 0);
803
804 if (range == 0)
805 return (0);
806
807 if (read(ztest_fd_rand, &r, sizeof (r)) != sizeof (r))
808 fatal(1, "short read from /dev/urandom");
809
810 return (r % range);
811 }
812
813 /* ARGSUSED */
814 static void
815 ztest_record_enospc(const char *s)
816 {
817 ztest_shared->zs_enospc_count++;
818 }
819
820 static uint64_t
821 ztest_get_ashift(void)
822 {
823 if (ztest_opts.zo_ashift == 0)
824 return (SPA_MINBLOCKSHIFT + ztest_random(5));
825 return (ztest_opts.zo_ashift);
826 }
827
828 static nvlist_t *
829 make_vdev_file(char *path, char *aux, char *pool, size_t size, uint64_t ashift)
830 {
831 char pathbuf[MAXPATHLEN];
832 uint64_t vdev;
833 nvlist_t *file;
834
835 if (ashift == 0)
836 ashift = ztest_get_ashift();
837
838 if (path == NULL) {
839 path = pathbuf;
840
841 if (aux != NULL) {
842 vdev = ztest_shared->zs_vdev_aux;
843 (void) snprintf(path, sizeof (pathbuf),
844 ztest_aux_template, ztest_opts.zo_dir,
845 pool == NULL ? ztest_opts.zo_pool : pool,
846 aux, vdev);
847 } else {
848 vdev = ztest_shared->zs_vdev_next_leaf++;
849 (void) snprintf(path, sizeof (pathbuf),
850 ztest_dev_template, ztest_opts.zo_dir,
851 pool == NULL ? ztest_opts.zo_pool : pool, vdev);
852 }
853 }
854
855 if (size != 0) {
856 int fd = open(path, O_RDWR | O_CREAT | O_TRUNC, 0666);
857 if (fd == -1)
858 fatal(1, "can't open %s", path);
859 if (ftruncate(fd, size) != 0)
860 fatal(1, "can't ftruncate %s", path);
861 (void) close(fd);
862 }
863
864 VERIFY(nvlist_alloc(&file, NV_UNIQUE_NAME, 0) == 0);
865 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_TYPE, VDEV_TYPE_FILE) == 0);
866 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_PATH, path) == 0);
867 VERIFY(nvlist_add_uint64(file, ZPOOL_CONFIG_ASHIFT, ashift) == 0);
868
869 return (file);
870 }
871
872 static nvlist_t *
873 make_vdev_raidz(char *path, char *aux, char *pool, size_t size,
874 uint64_t ashift, int r)
875 {
876 nvlist_t *raidz, **child;
877 int c;
878
879 if (r < 2)
880 return (make_vdev_file(path, aux, pool, size, ashift));
881 child = umem_alloc(r * sizeof (nvlist_t *), UMEM_NOFAIL);
882
883 for (c = 0; c < r; c++)
884 child[c] = make_vdev_file(path, aux, pool, size, ashift);
885
886 VERIFY(nvlist_alloc(&raidz, NV_UNIQUE_NAME, 0) == 0);
887 VERIFY(nvlist_add_string(raidz, ZPOOL_CONFIG_TYPE,
888 VDEV_TYPE_RAIDZ) == 0);
889 VERIFY(nvlist_add_uint64(raidz, ZPOOL_CONFIG_NPARITY,
890 ztest_opts.zo_raidz_parity) == 0);
891 VERIFY(nvlist_add_nvlist_array(raidz, ZPOOL_CONFIG_CHILDREN,
892 child, r) == 0);
893
894 for (c = 0; c < r; c++)
895 nvlist_free(child[c]);
896
897 umem_free(child, r * sizeof (nvlist_t *));
898
899 return (raidz);
900 }
901
902 static nvlist_t *
903 make_vdev_mirror(char *path, char *aux, char *pool, size_t size,
904 uint64_t ashift, int r, int m)
905 {
906 nvlist_t *mirror, **child;
907 int c;
908
909 if (m < 1)
910 return (make_vdev_raidz(path, aux, pool, size, ashift, r));
911
912 child = umem_alloc(m * sizeof (nvlist_t *), UMEM_NOFAIL);
913
914 for (c = 0; c < m; c++)
915 child[c] = make_vdev_raidz(path, aux, pool, size, ashift, r);
916
917 VERIFY(nvlist_alloc(&mirror, NV_UNIQUE_NAME, 0) == 0);
918 VERIFY(nvlist_add_string(mirror, ZPOOL_CONFIG_TYPE,
919 VDEV_TYPE_MIRROR) == 0);
920 VERIFY(nvlist_add_nvlist_array(mirror, ZPOOL_CONFIG_CHILDREN,
921 child, m) == 0);
922
923 for (c = 0; c < m; c++)
924 nvlist_free(child[c]);
925
926 umem_free(child, m * sizeof (nvlist_t *));
927
928 return (mirror);
929 }
930
931 static nvlist_t *
932 make_vdev_root(char *path, char *aux, char *pool, size_t size, uint64_t ashift,
933 int log, int r, int m, int t)
934 {
935 nvlist_t *root, **child;
936 int c;
937
938 ASSERT(t > 0);
939
940 child = umem_alloc(t * sizeof (nvlist_t *), UMEM_NOFAIL);
941
942 for (c = 0; c < t; c++) {
943 child[c] = make_vdev_mirror(path, aux, pool, size, ashift,
944 r, m);
945 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_IS_LOG,
946 log) == 0);
947 }
948
949 VERIFY(nvlist_alloc(&root, NV_UNIQUE_NAME, 0) == 0);
950 VERIFY(nvlist_add_string(root, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) == 0);
951 VERIFY(nvlist_add_nvlist_array(root, aux ? aux : ZPOOL_CONFIG_CHILDREN,
952 child, t) == 0);
953
954 for (c = 0; c < t; c++)
955 nvlist_free(child[c]);
956
957 umem_free(child, t * sizeof (nvlist_t *));
958
959 return (root);
960 }
961
962 /*
963 * Find a random spa version. Returns back a random spa version in the
964 * range [initial_version, SPA_VERSION_FEATURES].
965 */
966 static uint64_t
967 ztest_random_spa_version(uint64_t initial_version)
968 {
969 uint64_t version = initial_version;
970
971 if (version <= SPA_VERSION_BEFORE_FEATURES) {
972 version = version +
973 ztest_random(SPA_VERSION_BEFORE_FEATURES - version + 1);
974 }
975
976 if (version > SPA_VERSION_BEFORE_FEATURES)
977 version = SPA_VERSION_FEATURES;
978
979 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
980 return (version);
981 }
982
983 static int
984 ztest_random_blocksize(void)
985 {
986 uint64_t block_shift;
987 /*
988 * Choose a block size >= the ashift.
989 * If the SPA supports new MAXBLOCKSIZE, test up to 1MB blocks.
990 */
991 int maxbs = SPA_OLD_MAXBLOCKSHIFT;
992 if (spa_maxblocksize(ztest_spa) == SPA_MAXBLOCKSIZE)
993 maxbs = 20;
994 block_shift = ztest_random(maxbs - ztest_spa->spa_max_ashift + 1);
995 return (1 << (SPA_MINBLOCKSHIFT + block_shift));
996 }
997
998 static int
999 ztest_random_ibshift(void)
1000 {
1001 return (DN_MIN_INDBLKSHIFT +
1002 ztest_random(DN_MAX_INDBLKSHIFT - DN_MIN_INDBLKSHIFT + 1));
1003 }
1004
1005 static uint64_t
1006 ztest_random_vdev_top(spa_t *spa, boolean_t log_ok)
1007 {
1008 uint64_t top;
1009 vdev_t *rvd = spa->spa_root_vdev;
1010 vdev_t *tvd;
1011
1012 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1013
1014 do {
1015 top = ztest_random(rvd->vdev_children);
1016 tvd = rvd->vdev_child[top];
1017 } while (tvd->vdev_ishole || (tvd->vdev_islog && !log_ok) ||
1018 tvd->vdev_mg == NULL || tvd->vdev_mg->mg_class == NULL);
1019
1020 return (top);
1021 }
1022
1023 static uint64_t
1024 ztest_random_dsl_prop(zfs_prop_t prop)
1025 {
1026 uint64_t value;
1027
1028 do {
1029 value = zfs_prop_random_value(prop, ztest_random(-1ULL));
1030 } while (prop == ZFS_PROP_CHECKSUM && value == ZIO_CHECKSUM_OFF);
1031
1032 return (value);
1033 }
1034
1035 static int
1036 ztest_dsl_prop_set_uint64(char *osname, zfs_prop_t prop, uint64_t value,
1037 boolean_t inherit)
1038 {
1039 const char *propname = zfs_prop_to_name(prop);
1040 const char *valname;
1041 char setpoint[MAXPATHLEN];
1042 uint64_t curval;
1043 int error;
1044
1045 error = dsl_prop_set_int(osname, propname,
1046 (inherit ? ZPROP_SRC_NONE : ZPROP_SRC_LOCAL), value);
1047
1048 if (error == ENOSPC) {
1049 ztest_record_enospc(FTAG);
1050 return (error);
1051 }
1052 ASSERT0(error);
1053
1054 VERIFY0(dsl_prop_get_integer(osname, propname, &curval, setpoint));
1055
1056 if (ztest_opts.zo_verbose >= 6) {
1057 VERIFY(zfs_prop_index_to_string(prop, curval, &valname) == 0);
1058 (void) printf("%s %s = %s at '%s'\n",
1059 osname, propname, valname, setpoint);
1060 }
1061
1062 return (error);
1063 }
1064
1065 static int
1066 ztest_spa_prop_set_uint64(zpool_prop_t prop, uint64_t value)
1067 {
1068 spa_t *spa = ztest_spa;
1069 nvlist_t *props = NULL;
1070 int error;
1071
1072 VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
1073 VERIFY(nvlist_add_uint64(props, zpool_prop_to_name(prop), value) == 0);
1074
1075 error = spa_prop_set(spa, props);
1076
1077 nvlist_free(props);
1078
1079 if (error == ENOSPC) {
1080 ztest_record_enospc(FTAG);
1081 return (error);
1082 }
1083 ASSERT0(error);
1084
1085 return (error);
1086 }
1087
1088 static void
1089 ztest_rll_init(rll_t *rll)
1090 {
1091 rll->rll_writer = NULL;
1092 rll->rll_readers = 0;
1093 VERIFY(_mutex_init(&rll->rll_lock, USYNC_THREAD, NULL) == 0);
1094 VERIFY(cond_init(&rll->rll_cv, USYNC_THREAD, NULL) == 0);
1095 }
1096
1097 static void
1098 ztest_rll_destroy(rll_t *rll)
1099 {
1100 ASSERT(rll->rll_writer == NULL);
1101 ASSERT(rll->rll_readers == 0);
1102 VERIFY(_mutex_destroy(&rll->rll_lock) == 0);
1103 VERIFY(cond_destroy(&rll->rll_cv) == 0);
1104 }
1105
1106 static void
1107 ztest_rll_lock(rll_t *rll, rl_type_t type)
1108 {
1109 VERIFY(mutex_lock(&rll->rll_lock) == 0);
1110
1111 if (type == RL_READER) {
1112 while (rll->rll_writer != NULL)
1113 (void) cond_wait(&rll->rll_cv, &rll->rll_lock);
1114 rll->rll_readers++;
1115 } else {
1116 while (rll->rll_writer != NULL || rll->rll_readers)
1117 (void) cond_wait(&rll->rll_cv, &rll->rll_lock);
1118 rll->rll_writer = curthread;
1119 }
1120
1121 VERIFY(mutex_unlock(&rll->rll_lock) == 0);
1122 }
1123
1124 static void
1125 ztest_rll_unlock(rll_t *rll)
1126 {
1127 VERIFY(mutex_lock(&rll->rll_lock) == 0);
1128
1129 if (rll->rll_writer) {
1130 ASSERT(rll->rll_readers == 0);
1131 rll->rll_writer = NULL;
1132 } else {
1133 ASSERT(rll->rll_readers != 0);
1134 ASSERT(rll->rll_writer == NULL);
1135 rll->rll_readers--;
1136 }
1137
1138 if (rll->rll_writer == NULL && rll->rll_readers == 0)
1139 VERIFY(cond_broadcast(&rll->rll_cv) == 0);
1140
1141 VERIFY(mutex_unlock(&rll->rll_lock) == 0);
1142 }
1143
1144 static void
1145 ztest_object_lock(ztest_ds_t *zd, uint64_t object, rl_type_t type)
1146 {
1147 rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1148
1149 ztest_rll_lock(rll, type);
1150 }
1151
1152 static void
1153 ztest_object_unlock(ztest_ds_t *zd, uint64_t object)
1154 {
1155 rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1156
1157 ztest_rll_unlock(rll);
1158 }
1159
1160 static rl_t *
1161 ztest_range_lock(ztest_ds_t *zd, uint64_t object, uint64_t offset,
1162 uint64_t size, rl_type_t type)
1163 {
1164 uint64_t hash = object ^ (offset % (ZTEST_RANGE_LOCKS + 1));
1165 rll_t *rll = &zd->zd_range_lock[hash & (ZTEST_RANGE_LOCKS - 1)];
1166 rl_t *rl;
1167
1168 rl = umem_alloc(sizeof (*rl), UMEM_NOFAIL);
1169 rl->rl_object = object;
1170 rl->rl_offset = offset;
1171 rl->rl_size = size;
1172 rl->rl_lock = rll;
1173
1174 ztest_rll_lock(rll, type);
1175
1176 return (rl);
1177 }
1178
1179 static void
1180 ztest_range_unlock(rl_t *rl)
1181 {
1182 rll_t *rll = rl->rl_lock;
1183
1184 ztest_rll_unlock(rll);
1185
1186 umem_free(rl, sizeof (*rl));
1187 }
1188
1189 static void
1190 ztest_zd_init(ztest_ds_t *zd, ztest_shared_ds_t *szd, objset_t *os)
1191 {
1192 zd->zd_os = os;
1193 zd->zd_zilog = dmu_objset_zil(os);
1194 zd->zd_shared = szd;
1195 dmu_objset_name(os, zd->zd_name);
1196
1197 if (zd->zd_shared != NULL)
1198 zd->zd_shared->zd_seq = 0;
1199
1200 VERIFY(rwlock_init(&zd->zd_zilog_lock, USYNC_THREAD, NULL) == 0);
1201 VERIFY(_mutex_init(&zd->zd_dirobj_lock, USYNC_THREAD, NULL) == 0);
1202
1203 for (int l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1204 ztest_rll_init(&zd->zd_object_lock[l]);
1205
1206 for (int l = 0; l < ZTEST_RANGE_LOCKS; l++)
1207 ztest_rll_init(&zd->zd_range_lock[l]);
1208 }
1209
1210 static void
1211 ztest_zd_fini(ztest_ds_t *zd)
1212 {
1213 VERIFY(_mutex_destroy(&zd->zd_dirobj_lock) == 0);
1214
1215 for (int l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1216 ztest_rll_destroy(&zd->zd_object_lock[l]);
1217
1218 for (int l = 0; l < ZTEST_RANGE_LOCKS; l++)
1219 ztest_rll_destroy(&zd->zd_range_lock[l]);
1220 }
1221
1222 #define TXG_MIGHTWAIT (ztest_random(10) == 0 ? TXG_NOWAIT : TXG_WAIT)
1223
1224 static uint64_t
1225 ztest_tx_assign(dmu_tx_t *tx, uint64_t txg_how, const char *tag)
1226 {
1227 uint64_t txg;
1228 int error;
1229
1230 /*
1231 * Attempt to assign tx to some transaction group.
1232 */
1233 error = dmu_tx_assign(tx, txg_how);
1234 if (error) {
1235 if (error == ERESTART) {
1236 ASSERT(txg_how == TXG_NOWAIT);
1237 dmu_tx_wait(tx);
1238 } else {
1239 ASSERT3U(error, ==, ENOSPC);
1240 ztest_record_enospc(tag);
1241 }
1242 dmu_tx_abort(tx);
1243 return (0);
1244 }
1245 txg = dmu_tx_get_txg(tx);
1246 ASSERT(txg != 0);
1247 return (txg);
1248 }
1249
1250 static void
1251 ztest_pattern_set(void *buf, uint64_t size, uint64_t value)
1252 {
1253 uint64_t *ip = buf;
1254 uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1255
1256 while (ip < ip_end)
1257 *ip++ = value;
1258 }
1259
1260 static boolean_t
1261 ztest_pattern_match(void *buf, uint64_t size, uint64_t value)
1262 {
1263 uint64_t *ip = buf;
1264 uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1265 uint64_t diff = 0;
1266
1267 while (ip < ip_end)
1268 diff |= (value - *ip++);
1269
1270 return (diff == 0);
1271 }
1272
1273 static void
1274 ztest_bt_generate(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1275 uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg)
1276 {
1277 bt->bt_magic = BT_MAGIC;
1278 bt->bt_objset = dmu_objset_id(os);
1279 bt->bt_object = object;
1280 bt->bt_offset = offset;
1281 bt->bt_gen = gen;
1282 bt->bt_txg = txg;
1283 bt->bt_crtxg = crtxg;
1284 }
1285
1286 static void
1287 ztest_bt_verify(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1288 uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg)
1289 {
1290 ASSERT3U(bt->bt_magic, ==, BT_MAGIC);
1291 ASSERT3U(bt->bt_objset, ==, dmu_objset_id(os));
1292 ASSERT3U(bt->bt_object, ==, object);
1293 ASSERT3U(bt->bt_offset, ==, offset);
1294 ASSERT3U(bt->bt_gen, <=, gen);
1295 ASSERT3U(bt->bt_txg, <=, txg);
1296 ASSERT3U(bt->bt_crtxg, ==, crtxg);
1297 }
1298
1299 static ztest_block_tag_t *
1300 ztest_bt_bonus(dmu_buf_t *db)
1301 {
1302 dmu_object_info_t doi;
1303 ztest_block_tag_t *bt;
1304
1305 dmu_object_info_from_db(db, &doi);
1306 ASSERT3U(doi.doi_bonus_size, <=, db->db_size);
1307 ASSERT3U(doi.doi_bonus_size, >=, sizeof (*bt));
1308 bt = (void *)((char *)db->db_data + doi.doi_bonus_size - sizeof (*bt));
1309
1310 return (bt);
1311 }
1312
1313 /*
1314 * ZIL logging ops
1315 */
1316
1317 #define lrz_type lr_mode
1318 #define lrz_blocksize lr_uid
1319 #define lrz_ibshift lr_gid
1320 #define lrz_bonustype lr_rdev
1321 #define lrz_bonuslen lr_crtime[1]
1322
1323 static void
1324 ztest_log_create(ztest_ds_t *zd, dmu_tx_t *tx, lr_create_t *lr)
1325 {
1326 char *name = (void *)(lr + 1); /* name follows lr */
1327 size_t namesize = strlen(name) + 1;
1328 itx_t *itx;
1329
1330 if (zil_replaying(zd->zd_zilog, tx))
1331 return;
1332
1333 itx = zil_itx_create(TX_CREATE, sizeof (*lr) + namesize);
1334 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1335 sizeof (*lr) + namesize - sizeof (lr_t));
1336
1337 zil_itx_assign(zd->zd_zilog, itx, tx);
1338 }
1339
1340 static void
1341 ztest_log_remove(ztest_ds_t *zd, dmu_tx_t *tx, lr_remove_t *lr, uint64_t object)
1342 {
1343 char *name = (void *)(lr + 1); /* name follows lr */
1344 size_t namesize = strlen(name) + 1;
1345 itx_t *itx;
1346
1347 if (zil_replaying(zd->zd_zilog, tx))
1348 return;
1349
1350 itx = zil_itx_create(TX_REMOVE, sizeof (*lr) + namesize);
1351 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1352 sizeof (*lr) + namesize - sizeof (lr_t));
1353
1354 itx->itx_oid = object;
1355 zil_itx_assign(zd->zd_zilog, itx, tx);
1356 }
1357
1358 static void
1359 ztest_log_write(ztest_ds_t *zd, dmu_tx_t *tx, lr_write_t *lr)
1360 {
1361 itx_t *itx;
1362 itx_wr_state_t write_state = ztest_random(WR_NUM_STATES);
1363
1364 if (zil_replaying(zd->zd_zilog, tx))
1365 return;
1366
1367 if (lr->lr_length > ZIL_MAX_LOG_DATA)
1368 write_state = WR_INDIRECT;
1369
1370 itx = zil_itx_create(TX_WRITE,
1371 sizeof (*lr) + (write_state == WR_COPIED ? lr->lr_length : 0));
1372
1373 if (write_state == WR_COPIED &&
1374 dmu_read(zd->zd_os, lr->lr_foid, lr->lr_offset, lr->lr_length,
1375 ((lr_write_t *)&itx->itx_lr) + 1, DMU_READ_NO_PREFETCH) != 0) {
1376 zil_itx_destroy(itx);
1377 itx = zil_itx_create(TX_WRITE, sizeof (*lr));
1378 write_state = WR_NEED_COPY;
1379 }
1380 itx->itx_private = zd;
1381 itx->itx_wr_state = write_state;
1382 itx->itx_sync = (ztest_random(8) == 0);
1383
1384 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1385 sizeof (*lr) - sizeof (lr_t));
1386
1387 zil_itx_assign(zd->zd_zilog, itx, tx);
1388 }
1389
1390 static void
1391 ztest_log_truncate(ztest_ds_t *zd, dmu_tx_t *tx, lr_truncate_t *lr)
1392 {
1393 itx_t *itx;
1394
1395 if (zil_replaying(zd->zd_zilog, tx))
1396 return;
1397
1398 itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr));
1399 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1400 sizeof (*lr) - sizeof (lr_t));
1401
1402 itx->itx_sync = B_FALSE;
1403 zil_itx_assign(zd->zd_zilog, itx, tx);
1404 }
1405
1406 static void
1407 ztest_log_setattr(ztest_ds_t *zd, dmu_tx_t *tx, lr_setattr_t *lr)
1408 {
1409 itx_t *itx;
1410
1411 if (zil_replaying(zd->zd_zilog, tx))
1412 return;
1413
1414 itx = zil_itx_create(TX_SETATTR, sizeof (*lr));
1415 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1416 sizeof (*lr) - sizeof (lr_t));
1417
1418 itx->itx_sync = B_FALSE;
1419 zil_itx_assign(zd->zd_zilog, itx, tx);
1420 }
1421
1422 /*
1423 * ZIL replay ops
1424 */
1425 static int
1426 ztest_replay_create(void *arg1, void *arg2, boolean_t byteswap)
1427 {
1428 ztest_ds_t *zd = arg1;
1429 lr_create_t *lr = arg2;
1430 char *name = (void *)(lr + 1); /* name follows lr */
1431 objset_t *os = zd->zd_os;
1432 ztest_block_tag_t *bbt;
1433 dmu_buf_t *db;
1434 dmu_tx_t *tx;
1435 uint64_t txg;
1436 int error = 0;
1437
1438 if (byteswap)
1439 byteswap_uint64_array(lr, sizeof (*lr));
1440
1441 ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1442 ASSERT(name[0] != '\0');
1443
1444 tx = dmu_tx_create(os);
1445
1446 dmu_tx_hold_zap(tx, lr->lr_doid, B_TRUE, name);
1447
1448 if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1449 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, B_TRUE, NULL);
1450 } else {
1451 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1452 }
1453
1454 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1455 if (txg == 0)
1456 return (ENOSPC);
1457
1458 ASSERT(dmu_objset_zil(os)->zl_replay == !!lr->lr_foid);
1459
1460 if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1461 if (lr->lr_foid == 0) {
1462 lr->lr_foid = zap_create(os,
1463 lr->lrz_type, lr->lrz_bonustype,
1464 lr->lrz_bonuslen, tx);
1465 } else {
1466 error = zap_create_claim(os, lr->lr_foid,
1467 lr->lrz_type, lr->lrz_bonustype,
1468 lr->lrz_bonuslen, tx);
1469 }
1470 } else {
1471 if (lr->lr_foid == 0) {
1472 lr->lr_foid = dmu_object_alloc(os,
1473 lr->lrz_type, 0, lr->lrz_bonustype,
1474 lr->lrz_bonuslen, tx);
1475 } else {
1476 error = dmu_object_claim(os, lr->lr_foid,
1477 lr->lrz_type, 0, lr->lrz_bonustype,
1478 lr->lrz_bonuslen, tx);
1479 }
1480 }
1481
1482 if (error) {
1483 ASSERT3U(error, ==, EEXIST);
1484 ASSERT(zd->zd_zilog->zl_replay);
1485 dmu_tx_commit(tx);
1486 return (error);
1487 }
1488
1489 ASSERT(lr->lr_foid != 0);
1490
1491 if (lr->lrz_type != DMU_OT_ZAP_OTHER)
1492 VERIFY3U(0, ==, dmu_object_set_blocksize(os, lr->lr_foid,
1493 lr->lrz_blocksize, lr->lrz_ibshift, tx));
1494
1495 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1496 bbt = ztest_bt_bonus(db);
1497 dmu_buf_will_dirty(db, tx);
1498 ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_gen, txg, txg);
1499 dmu_buf_rele(db, FTAG);
1500
1501 VERIFY3U(0, ==, zap_add(os, lr->lr_doid, name, sizeof (uint64_t), 1,
1502 &lr->lr_foid, tx));
1503
1504 (void) ztest_log_create(zd, tx, lr);
1505
1506 dmu_tx_commit(tx);
1507
1508 return (0);
1509 }
1510
1511 static int
1512 ztest_replay_remove(void *arg1, void *arg2, boolean_t byteswap)
1513 {
1514 ztest_ds_t *zd = arg1;
1515 lr_remove_t *lr = arg2;
1516 char *name = (void *)(lr + 1); /* name follows lr */
1517 objset_t *os = zd->zd_os;
1518 dmu_object_info_t doi;
1519 dmu_tx_t *tx;
1520 uint64_t object, txg;
1521
1522 if (byteswap)
1523 byteswap_uint64_array(lr, sizeof (*lr));
1524
1525 ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1526 ASSERT(name[0] != '\0');
1527
1528 VERIFY3U(0, ==,
1529 zap_lookup(os, lr->lr_doid, name, sizeof (object), 1, &object));
1530 ASSERT(object != 0);
1531
1532 ztest_object_lock(zd, object, RL_WRITER);
1533
1534 VERIFY3U(0, ==, dmu_object_info(os, object, &doi));
1535
1536 tx = dmu_tx_create(os);
1537
1538 dmu_tx_hold_zap(tx, lr->lr_doid, B_FALSE, name);
1539 dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END);
1540
1541 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1542 if (txg == 0) {
1543 ztest_object_unlock(zd, object);
1544 return (ENOSPC);
1545 }
1546
1547 if (doi.doi_type == DMU_OT_ZAP_OTHER) {
1548 VERIFY3U(0, ==, zap_destroy(os, object, tx));
1549 } else {
1550 VERIFY3U(0, ==, dmu_object_free(os, object, tx));
1551 }
1552
1553 VERIFY3U(0, ==, zap_remove(os, lr->lr_doid, name, tx));
1554
1555 (void) ztest_log_remove(zd, tx, lr, object);
1556
1557 dmu_tx_commit(tx);
1558
1559 ztest_object_unlock(zd, object);
1560
1561 return (0);
1562 }
1563
1564 static int
1565 ztest_replay_write(void *arg1, void *arg2, boolean_t byteswap)
1566 {
1567 ztest_ds_t *zd = arg1;
1568 lr_write_t *lr = arg2;
1569 objset_t *os = zd->zd_os;
1570 void *data = lr + 1; /* data follows lr */
1571 uint64_t offset, length;
1572 ztest_block_tag_t *bt = data;
1573 ztest_block_tag_t *bbt;
1574 uint64_t gen, txg, lrtxg, crtxg;
1575 dmu_object_info_t doi;
1576 dmu_tx_t *tx;
1577 dmu_buf_t *db;
1578 arc_buf_t *abuf = NULL;
1579 rl_t *rl;
1580
1581 if (byteswap)
1582 byteswap_uint64_array(lr, sizeof (*lr));
1583
1584 offset = lr->lr_offset;
1585 length = lr->lr_length;
1586
1587 /* If it's a dmu_sync() block, write the whole block */
1588 if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
1589 uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
1590 if (length < blocksize) {
1591 offset -= offset % blocksize;
1592 length = blocksize;
1593 }
1594 }
1595
1596 if (bt->bt_magic == BSWAP_64(BT_MAGIC))
1597 byteswap_uint64_array(bt, sizeof (*bt));
1598
1599 if (bt->bt_magic != BT_MAGIC)
1600 bt = NULL;
1601
1602 ztest_object_lock(zd, lr->lr_foid, RL_READER);
1603 rl = ztest_range_lock(zd, lr->lr_foid, offset, length, RL_WRITER);
1604
1605 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1606
1607 dmu_object_info_from_db(db, &doi);
1608
1609 bbt = ztest_bt_bonus(db);
1610 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1611 gen = bbt->bt_gen;
1612 crtxg = bbt->bt_crtxg;
1613 lrtxg = lr->lr_common.lrc_txg;
1614
1615 tx = dmu_tx_create(os);
1616
1617 dmu_tx_hold_write(tx, lr->lr_foid, offset, length);
1618
1619 if (ztest_random(8) == 0 && length == doi.doi_data_block_size &&
1620 P2PHASE(offset, length) == 0)
1621 abuf = dmu_request_arcbuf(db, length);
1622
1623 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1624 if (txg == 0) {
1625 if (abuf != NULL)
1626 dmu_return_arcbuf(abuf);
1627 dmu_buf_rele(db, FTAG);
1628 ztest_range_unlock(rl);
1629 ztest_object_unlock(zd, lr->lr_foid);
1630 return (ENOSPC);
1631 }
1632
1633 if (bt != NULL) {
1634 /*
1635 * Usually, verify the old data before writing new data --
1636 * but not always, because we also want to verify correct
1637 * behavior when the data was not recently read into cache.
1638 */
1639 ASSERT(offset % doi.doi_data_block_size == 0);
1640 if (ztest_random(4) != 0) {
1641 int prefetch = ztest_random(2) ?
1642 DMU_READ_PREFETCH : DMU_READ_NO_PREFETCH;
1643 ztest_block_tag_t rbt;
1644
1645 VERIFY(dmu_read(os, lr->lr_foid, offset,
1646 sizeof (rbt), &rbt, prefetch) == 0);
1647 if (rbt.bt_magic == BT_MAGIC) {
1648 ztest_bt_verify(&rbt, os, lr->lr_foid,
1649 offset, gen, txg, crtxg);
1650 }
1651 }
1652
1653 /*
1654 * Writes can appear to be newer than the bonus buffer because
1655 * the ztest_get_data() callback does a dmu_read() of the
1656 * open-context data, which may be different than the data
1657 * as it was when the write was generated.
1658 */
1659 if (zd->zd_zilog->zl_replay) {
1660 ztest_bt_verify(bt, os, lr->lr_foid, offset,
1661 MAX(gen, bt->bt_gen), MAX(txg, lrtxg),
1662 bt->bt_crtxg);
1663 }
1664
1665 /*
1666 * Set the bt's gen/txg to the bonus buffer's gen/txg
1667 * so that all of the usual ASSERTs will work.
1668 */
1669 ztest_bt_generate(bt, os, lr->lr_foid, offset, gen, txg, crtxg);
1670 }
1671
1672 if (abuf == NULL) {
1673 dmu_write(os, lr->lr_foid, offset, length, data, tx);
1674 } else {
1675 bcopy(data, abuf->b_data, length);
1676 dmu_assign_arcbuf(db, offset, abuf, tx);
1677 }
1678
1679 (void) ztest_log_write(zd, tx, lr);
1680
1681 dmu_buf_rele(db, FTAG);
1682
1683 dmu_tx_commit(tx);
1684
1685 ztest_range_unlock(rl);
1686 ztest_object_unlock(zd, lr->lr_foid);
1687
1688 return (0);
1689 }
1690
1691 static int
1692 ztest_replay_truncate(void *arg1, void *arg2, boolean_t byteswap)
1693 {
1694 ztest_ds_t *zd = arg1;
1695 lr_truncate_t *lr = arg2;
1696 objset_t *os = zd->zd_os;
1697 dmu_tx_t *tx;
1698 uint64_t txg;
1699 rl_t *rl;
1700
1701 if (byteswap)
1702 byteswap_uint64_array(lr, sizeof (*lr));
1703
1704 ztest_object_lock(zd, lr->lr_foid, RL_READER);
1705 rl = ztest_range_lock(zd, lr->lr_foid, lr->lr_offset, lr->lr_length,
1706 RL_WRITER);
1707
1708 tx = dmu_tx_create(os);
1709
1710 dmu_tx_hold_free(tx, lr->lr_foid, lr->lr_offset, lr->lr_length);
1711
1712 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1713 if (txg == 0) {
1714 ztest_range_unlock(rl);
1715 ztest_object_unlock(zd, lr->lr_foid);
1716 return (ENOSPC);
1717 }
1718
1719 VERIFY(dmu_free_range(os, lr->lr_foid, lr->lr_offset,
1720 lr->lr_length, tx) == 0);
1721
1722 (void) ztest_log_truncate(zd, tx, lr);
1723
1724 dmu_tx_commit(tx);
1725
1726 ztest_range_unlock(rl);
1727 ztest_object_unlock(zd, lr->lr_foid);
1728
1729 return (0);
1730 }
1731
1732 static int
1733 ztest_replay_setattr(void *arg1, void *arg2, boolean_t byteswap)
1734 {
1735 ztest_ds_t *zd = arg1;
1736 lr_setattr_t *lr = arg2;
1737 objset_t *os = zd->zd_os;
1738 dmu_tx_t *tx;
1739 dmu_buf_t *db;
1740 ztest_block_tag_t *bbt;
1741 uint64_t txg, lrtxg, crtxg;
1742
1743 if (byteswap)
1744 byteswap_uint64_array(lr, sizeof (*lr));
1745
1746 ztest_object_lock(zd, lr->lr_foid, RL_WRITER);
1747
1748 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1749
1750 tx = dmu_tx_create(os);
1751 dmu_tx_hold_bonus(tx, lr->lr_foid);
1752
1753 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1754 if (txg == 0) {
1755 dmu_buf_rele(db, FTAG);
1756 ztest_object_unlock(zd, lr->lr_foid);
1757 return (ENOSPC);
1758 }
1759
1760 bbt = ztest_bt_bonus(db);
1761 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1762 crtxg = bbt->bt_crtxg;
1763 lrtxg = lr->lr_common.lrc_txg;
1764
1765 if (zd->zd_zilog->zl_replay) {
1766 ASSERT(lr->lr_size != 0);
1767 ASSERT(lr->lr_mode != 0);
1768 ASSERT(lrtxg != 0);
1769 } else {
1770 /*
1771 * Randomly change the size and increment the generation.
1772 */
1773 lr->lr_size = (ztest_random(db->db_size / sizeof (*bbt)) + 1) *
1774 sizeof (*bbt);
1775 lr->lr_mode = bbt->bt_gen + 1;
1776 ASSERT(lrtxg == 0);
1777 }
1778
1779 /*
1780 * Verify that the current bonus buffer is not newer than our txg.
1781 */
1782 ztest_bt_verify(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode,
1783 MAX(txg, lrtxg), crtxg);
1784
1785 dmu_buf_will_dirty(db, tx);
1786
1787 ASSERT3U(lr->lr_size, >=, sizeof (*bbt));
1788 ASSERT3U(lr->lr_size, <=, db->db_size);
1789 VERIFY0(dmu_set_bonus(db, lr->lr_size, tx));
1790 bbt = ztest_bt_bonus(db);
1791
1792 ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode, txg, crtxg);
1793
1794 dmu_buf_rele(db, FTAG);
1795
1796 (void) ztest_log_setattr(zd, tx, lr);
1797
1798 dmu_tx_commit(tx);
1799
1800 ztest_object_unlock(zd, lr->lr_foid);
1801
1802 return (0);
1803 }
1804
1805 zil_replay_func_t *ztest_replay_vector[TX_MAX_TYPE] = {
1806 NULL, /* 0 no such transaction type */
1807 ztest_replay_create, /* TX_CREATE */
1808 NULL, /* TX_MKDIR */
1809 NULL, /* TX_MKXATTR */
1810 NULL, /* TX_SYMLINK */
1811 ztest_replay_remove, /* TX_REMOVE */
1812 NULL, /* TX_RMDIR */
1813 NULL, /* TX_LINK */
1814 NULL, /* TX_RENAME */
1815 ztest_replay_write, /* TX_WRITE */
1816 ztest_replay_truncate, /* TX_TRUNCATE */
1817 ztest_replay_setattr, /* TX_SETATTR */
1818 NULL, /* TX_ACL */
1819 NULL, /* TX_CREATE_ACL */
1820 NULL, /* TX_CREATE_ATTR */
1821 NULL, /* TX_CREATE_ACL_ATTR */
1822 NULL, /* TX_MKDIR_ACL */
1823 NULL, /* TX_MKDIR_ATTR */
1824 NULL, /* TX_MKDIR_ACL_ATTR */
1825 NULL, /* TX_WRITE2 */
1826 };
1827
1828 /*
1829 * ZIL get_data callbacks
1830 */
1831
1832 static void
1833 ztest_get_done(zgd_t *zgd, int error)
1834 {
1835 ztest_ds_t *zd = zgd->zgd_private;
1836 uint64_t object = zgd->zgd_rl->rl_object;
1837
1838 if (zgd->zgd_db)
1839 dmu_buf_rele(zgd->zgd_db, zgd);
1840
1841 ztest_range_unlock(zgd->zgd_rl);
1842 ztest_object_unlock(zd, object);
1843
1844 if (error == 0 && zgd->zgd_bp)
1845 zil_lwb_add_block(zgd->zgd_lwb, zgd->zgd_bp);
1846
1847 umem_free(zgd, sizeof (*zgd));
1848 }
1849
1850 static int
1851 ztest_get_data(void *arg, lr_write_t *lr, char *buf, struct lwb *lwb,
1852 zio_t *zio)
1853 {
1854 ztest_ds_t *zd = arg;
1855 objset_t *os = zd->zd_os;
1856 uint64_t object = lr->lr_foid;
1857 uint64_t offset = lr->lr_offset;
1858 uint64_t size = lr->lr_length;
1859 uint64_t txg = lr->lr_common.lrc_txg;
1860 uint64_t crtxg;
1861 dmu_object_info_t doi;
1862 dmu_buf_t *db;
1863 zgd_t *zgd;
1864 int error;
1865
1866 ASSERT3P(lwb, !=, NULL);
1867 ASSERT3P(zio, !=, NULL);
1868 ASSERT3U(size, !=, 0);
1869
1870 ztest_object_lock(zd, object, RL_READER);
1871 error = dmu_bonus_hold(os, object, FTAG, &db);
1872 if (error) {
1873 ztest_object_unlock(zd, object);
1874 return (error);
1875 }
1876
1877 crtxg = ztest_bt_bonus(db)->bt_crtxg;
1878
1879 if (crtxg == 0 || crtxg > txg) {
1880 dmu_buf_rele(db, FTAG);
1881 ztest_object_unlock(zd, object);
1882 return (ENOENT);
1883 }
1884
1885 dmu_object_info_from_db(db, &doi);
1886 dmu_buf_rele(db, FTAG);
1887 db = NULL;
1888
1889 zgd = umem_zalloc(sizeof (*zgd), UMEM_NOFAIL);
1890 zgd->zgd_lwb = lwb;
1891 zgd->zgd_private = zd;
1892
1893 if (buf != NULL) { /* immediate write */
1894 zgd->zgd_rl = ztest_range_lock(zd, object, offset, size,
1895 RL_READER);
1896
1897 error = dmu_read(os, object, offset, size, buf,
1898 DMU_READ_NO_PREFETCH);
1899 ASSERT(error == 0);
1900 } else {
1901 size = doi.doi_data_block_size;
1902 if (ISP2(size)) {
1903 offset = P2ALIGN(offset, size);
1904 } else {
1905 ASSERT(offset < size);
1906 offset = 0;
1907 }
1908
1909 zgd->zgd_rl = ztest_range_lock(zd, object, offset, size,
1910 RL_READER);
1911
1912 error = dmu_buf_hold(os, object, offset, zgd, &db,
1913 DMU_READ_NO_PREFETCH);
1914
1915 if (error == 0) {
1916 blkptr_t *bp = &lr->lr_blkptr;
1917
1918 zgd->zgd_db = db;
1919 zgd->zgd_bp = bp;
1920
1921 ASSERT(db->db_offset == offset);
1922 ASSERT(db->db_size == size);
1923
1924 error = dmu_sync(zio, lr->lr_common.lrc_txg,
1925 ztest_get_done, zgd);
1926
1927 if (error == 0)
1928 return (0);
1929 }
1930 }
1931
1932 ztest_get_done(zgd, error);
1933
1934 return (error);
1935 }
1936
1937 static void *
1938 ztest_lr_alloc(size_t lrsize, char *name)
1939 {
1940 char *lr;
1941 size_t namesize = name ? strlen(name) + 1 : 0;
1942
1943 lr = umem_zalloc(lrsize + namesize, UMEM_NOFAIL);
1944
1945 if (name)
1946 bcopy(name, lr + lrsize, namesize);
1947
1948 return (lr);
1949 }
1950
1951 void
1952 ztest_lr_free(void *lr, size_t lrsize, char *name)
1953 {
1954 size_t namesize = name ? strlen(name) + 1 : 0;
1955
1956 umem_free(lr, lrsize + namesize);
1957 }
1958
1959 /*
1960 * Lookup a bunch of objects. Returns the number of objects not found.
1961 */
1962 static int
1963 ztest_lookup(ztest_ds_t *zd, ztest_od_t *od, int count)
1964 {
1965 int missing = 0;
1966 int error;
1967
1968 ASSERT(_mutex_held(&zd->zd_dirobj_lock));
1969
1970 for (int i = 0; i < count; i++, od++) {
1971 od->od_object = 0;
1972 error = zap_lookup(zd->zd_os, od->od_dir, od->od_name,
1973 sizeof (uint64_t), 1, &od->od_object);
1974 if (error) {
1975 ASSERT(error == ENOENT);
1976 ASSERT(od->od_object == 0);
1977 missing++;
1978 } else {
1979 dmu_buf_t *db;
1980 ztest_block_tag_t *bbt;
1981 dmu_object_info_t doi;
1982
1983 ASSERT(od->od_object != 0);
1984 ASSERT(missing == 0); /* there should be no gaps */
1985
1986 ztest_object_lock(zd, od->od_object, RL_READER);
1987 VERIFY3U(0, ==, dmu_bonus_hold(zd->zd_os,
1988 od->od_object, FTAG, &db));
1989 dmu_object_info_from_db(db, &doi);
1990 bbt = ztest_bt_bonus(db);
1991 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1992 od->od_type = doi.doi_type;
1993 od->od_blocksize = doi.doi_data_block_size;
1994 od->od_gen = bbt->bt_gen;
1995 dmu_buf_rele(db, FTAG);
1996 ztest_object_unlock(zd, od->od_object);
1997 }
1998 }
1999
2000 return (missing);
2001 }
2002
2003 static int
2004 ztest_create(ztest_ds_t *zd, ztest_od_t *od, int count)
2005 {
2006 int missing = 0;
2007
2008 ASSERT(_mutex_held(&zd->zd_dirobj_lock));
2009
2010 for (int i = 0; i < count; i++, od++) {
2011 if (missing) {
2012 od->od_object = 0;
2013 missing++;
2014 continue;
2015 }
2016
2017 lr_create_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
2018
2019 lr->lr_doid = od->od_dir;
2020 lr->lr_foid = 0; /* 0 to allocate, > 0 to claim */
2021 lr->lrz_type = od->od_crtype;
2022 lr->lrz_blocksize = od->od_crblocksize;
2023 lr->lrz_ibshift = ztest_random_ibshift();
2024 lr->lrz_bonustype = DMU_OT_UINT64_OTHER;
2025 lr->lrz_bonuslen = dmu_bonus_max();
2026 lr->lr_gen = od->od_crgen;
2027 lr->lr_crtime[0] = time(NULL);
2028
2029 if (ztest_replay_create(zd, lr, B_FALSE) != 0) {
2030 ASSERT(missing == 0);
2031 od->od_object = 0;
2032 missing++;
2033 } else {
2034 od->od_object = lr->lr_foid;
2035 od->od_type = od->od_crtype;
2036 od->od_blocksize = od->od_crblocksize;
2037 od->od_gen = od->od_crgen;
2038 ASSERT(od->od_object != 0);
2039 }
2040
2041 ztest_lr_free(lr, sizeof (*lr), od->od_name);
2042 }
2043
2044 return (missing);
2045 }
2046
2047 static int
2048 ztest_remove(ztest_ds_t *zd, ztest_od_t *od, int count)
2049 {
2050 int missing = 0;
2051 int error;
2052
2053 ASSERT(_mutex_held(&zd->zd_dirobj_lock));
2054
2055 od += count - 1;
2056
2057 for (int i = count - 1; i >= 0; i--, od--) {
2058 if (missing) {
2059 missing++;
2060 continue;
2061 }
2062
2063 /*
2064 * No object was found.
2065 */
2066 if (od->od_object == 0)
2067 continue;
2068
2069 lr_remove_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
2070
2071 lr->lr_doid = od->od_dir;
2072
2073 if ((error = ztest_replay_remove(zd, lr, B_FALSE)) != 0) {
2074 ASSERT3U(error, ==, ENOSPC);
2075 missing++;
2076 } else {
2077 od->od_object = 0;
2078 }
2079 ztest_lr_free(lr, sizeof (*lr), od->od_name);
2080 }
2081
2082 return (missing);
2083 }
2084
2085 static int
2086 ztest_write(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size,
2087 void *data)
2088 {
2089 lr_write_t *lr;
2090 int error;
2091
2092 lr = ztest_lr_alloc(sizeof (*lr) + size, NULL);
2093
2094 lr->lr_foid = object;
2095 lr->lr_offset = offset;
2096 lr->lr_length = size;
2097 lr->lr_blkoff = 0;
2098 BP_ZERO(&lr->lr_blkptr);
2099
2100 bcopy(data, lr + 1, size);
2101
2102 error = ztest_replay_write(zd, lr, B_FALSE);
2103
2104 ztest_lr_free(lr, sizeof (*lr) + size, NULL);
2105
2106 return (error);
2107 }
2108
2109 static int
2110 ztest_truncate(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2111 {
2112 lr_truncate_t *lr;
2113 int error;
2114
2115 lr = ztest_lr_alloc(sizeof (*lr), NULL);
2116
2117 lr->lr_foid = object;
2118 lr->lr_offset = offset;
2119 lr->lr_length = size;
2120
2121 error = ztest_replay_truncate(zd, lr, B_FALSE);
2122
2123 ztest_lr_free(lr, sizeof (*lr), NULL);
2124
2125 return (error);
2126 }
2127
2128 static int
2129 ztest_setattr(ztest_ds_t *zd, uint64_t object)
2130 {
2131 lr_setattr_t *lr;
2132 int error;
2133
2134 lr = ztest_lr_alloc(sizeof (*lr), NULL);
2135
2136 lr->lr_foid = object;
2137 lr->lr_size = 0;
2138 lr->lr_mode = 0;
2139
2140 error = ztest_replay_setattr(zd, lr, B_FALSE);
2141
2142 ztest_lr_free(lr, sizeof (*lr), NULL);
2143
2144 return (error);
2145 }
2146
2147 static void
2148 ztest_prealloc(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2149 {
2150 objset_t *os = zd->zd_os;
2151 dmu_tx_t *tx;
2152 uint64_t txg;
2153 rl_t *rl;
2154
2155 txg_wait_synced(dmu_objset_pool(os), 0);
2156
2157 ztest_object_lock(zd, object, RL_READER);
2158 rl = ztest_range_lock(zd, object, offset, size, RL_WRITER);
2159
2160 tx = dmu_tx_create(os);
2161
2162 dmu_tx_hold_write(tx, object, offset, size);
2163
2164 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
2165
2166 if (txg != 0) {
2167 dmu_prealloc(os, object, offset, size, tx);
2168 dmu_tx_commit(tx);
2169 txg_wait_synced(dmu_objset_pool(os), txg);
2170 } else {
2171 (void) dmu_free_long_range(os, object, offset, size);
2172 }
2173
2174 ztest_range_unlock(rl);
2175 ztest_object_unlock(zd, object);
2176 }
2177
2178 static void
2179 ztest_io(ztest_ds_t *zd, uint64_t object, uint64_t offset)
2180 {
2181 int err;
2182 ztest_block_tag_t wbt;
2183 dmu_object_info_t doi;
2184 enum ztest_io_type io_type;
2185 uint64_t blocksize;
2186 void *data;
2187
2188 VERIFY(dmu_object_info(zd->zd_os, object, &doi) == 0);
2189 blocksize = doi.doi_data_block_size;
2190 data = umem_alloc(blocksize, UMEM_NOFAIL);
2191
2192 /*
2193 * Pick an i/o type at random, biased toward writing block tags.
2194 */
2195 io_type = ztest_random(ZTEST_IO_TYPES);
2196 if (ztest_random(2) == 0)
2197 io_type = ZTEST_IO_WRITE_TAG;
2198
2199 (void) rw_rdlock(&zd->zd_zilog_lock);
2200
2201 switch (io_type) {
2202
2203 case ZTEST_IO_WRITE_TAG:
2204 ztest_bt_generate(&wbt, zd->zd_os, object, offset, 0, 0, 0);
2205 (void) ztest_write(zd, object, offset, sizeof (wbt), &wbt);
2206 break;
2207
2208 case ZTEST_IO_WRITE_PATTERN:
2209 (void) memset(data, 'a' + (object + offset) % 5, blocksize);
2210 if (ztest_random(2) == 0) {
2211 /*
2212 * Induce fletcher2 collisions to ensure that
2213 * zio_ddt_collision() detects and resolves them
2214 * when using fletcher2-verify for deduplication.
2215 */
2216 ((uint64_t *)data)[0] ^= 1ULL << 63;
2217 ((uint64_t *)data)[4] ^= 1ULL << 63;
2218 }
2219 (void) ztest_write(zd, object, offset, blocksize, data);
2220 break;
2221
2222 case ZTEST_IO_WRITE_ZEROES:
2223 bzero(data, blocksize);
2224 (void) ztest_write(zd, object, offset, blocksize, data);
2225 break;
2226
2227 case ZTEST_IO_TRUNCATE:
2228 (void) ztest_truncate(zd, object, offset, blocksize);
2229 break;
2230
2231 case ZTEST_IO_SETATTR:
2232 (void) ztest_setattr(zd, object);
2233 break;
2234
2235 case ZTEST_IO_REWRITE:
2236 (void) rw_rdlock(&ztest_name_lock);
2237 err = ztest_dsl_prop_set_uint64(zd->zd_name,
2238 ZFS_PROP_CHECKSUM, spa_dedup_checksum(ztest_spa),
2239 B_FALSE);
2240 VERIFY(err == 0 || err == ENOSPC);
2241 err = ztest_dsl_prop_set_uint64(zd->zd_name,
2242 ZFS_PROP_COMPRESSION,
2243 ztest_random_dsl_prop(ZFS_PROP_COMPRESSION),
2244 B_FALSE);
2245 VERIFY(err == 0 || err == ENOSPC);
2246 (void) rw_unlock(&ztest_name_lock);
2247
2248 VERIFY0(dmu_read(zd->zd_os, object, offset, blocksize, data,
2249 DMU_READ_NO_PREFETCH));
2250
2251 (void) ztest_write(zd, object, offset, blocksize, data);
2252 break;
2253 }
2254
2255 (void) rw_unlock(&zd->zd_zilog_lock);
2256
2257 umem_free(data, blocksize);
2258 }
2259
2260 /*
2261 * Initialize an object description template.
2262 */
2263 static void
2264 ztest_od_init(ztest_od_t *od, uint64_t id, char *tag, uint64_t index,
2265 dmu_object_type_t type, uint64_t blocksize, uint64_t gen)
2266 {
2267 od->od_dir = ZTEST_DIROBJ;
2268 od->od_object = 0;
2269
2270 od->od_crtype = type;
2271 od->od_crblocksize = blocksize ? blocksize : ztest_random_blocksize();
2272 od->od_crgen = gen;
2273
2274 od->od_type = DMU_OT_NONE;
2275 od->od_blocksize = 0;
2276 od->od_gen = 0;
2277
2278 (void) snprintf(od->od_name, sizeof (od->od_name), "%s(%lld)[%llu]",
2279 tag, (int64_t)id, index);
2280 }
2281
2282 /*
2283 * Lookup or create the objects for a test using the od template.
2284 * If the objects do not all exist, or if 'remove' is specified,
2285 * remove any existing objects and create new ones. Otherwise,
2286 * use the existing objects.
2287 */
2288 static int
2289 ztest_object_init(ztest_ds_t *zd, ztest_od_t *od, size_t size, boolean_t remove)
2290 {
2291 int count = size / sizeof (*od);
2292 int rv = 0;
2293
2294 VERIFY(mutex_lock(&zd->zd_dirobj_lock) == 0);
2295 if ((ztest_lookup(zd, od, count) != 0 || remove) &&
2296 (ztest_remove(zd, od, count) != 0 ||
2297 ztest_create(zd, od, count) != 0))
2298 rv = -1;
2299 zd->zd_od = od;
2300 VERIFY(mutex_unlock(&zd->zd_dirobj_lock) == 0);
2301
2302 return (rv);
2303 }
2304
2305 /* ARGSUSED */
2306 void
2307 ztest_zil_commit(ztest_ds_t *zd, uint64_t id)
2308 {
2309 zilog_t *zilog = zd->zd_zilog;
2310
2311 (void) rw_rdlock(&zd->zd_zilog_lock);
2312
2313 zil_commit(zilog, ztest_random(ZTEST_OBJECTS));
2314
2315 /*
2316 * Remember the committed values in zd, which is in parent/child
2317 * shared memory. If we die, the next iteration of ztest_run()
2318 * will verify that the log really does contain this record.
2319 */
2320 mutex_enter(&zilog->zl_lock);
2321 ASSERT(zd->zd_shared != NULL);
2322 ASSERT3U(zd->zd_shared->zd_seq, <=, zilog->zl_commit_lr_seq);
2323 zd->zd_shared->zd_seq = zilog->zl_commit_lr_seq;
2324 mutex_exit(&zilog->zl_lock);
2325
2326 (void) rw_unlock(&zd->zd_zilog_lock);
2327 }
2328
2329 /*
2330 * This function is designed to simulate the operations that occur during a
2331 * mount/unmount operation. We hold the dataset across these operations in an
2332 * attempt to expose any implicit assumptions about ZIL management.
2333 */
2334 /* ARGSUSED */
2335 void
2336 ztest_zil_remount(ztest_ds_t *zd, uint64_t id)
2337 {
2338 objset_t *os = zd->zd_os;
2339
2340 /*
2341 * We grab the zd_dirobj_lock to ensure that no other thread is
2342 * updating the zil (i.e. adding in-memory log records) and the
2343 * zd_zilog_lock to block any I/O.
2344 */
2345 VERIFY0(mutex_lock(&zd->zd_dirobj_lock));
2346 (void) rw_wrlock(&zd->zd_zilog_lock);
2347
2348 /* zfsvfs_teardown() */
2349 zil_close(zd->zd_zilog);
2350
2351 /* zfsvfs_setup() */
2352 VERIFY(zil_open(os, ztest_get_data) == zd->zd_zilog);
2353 zil_replay(os, zd, ztest_replay_vector);
2354
2355 (void) rw_unlock(&zd->zd_zilog_lock);
2356 VERIFY(mutex_unlock(&zd->zd_dirobj_lock) == 0);
2357 }
2358
2359 /*
2360 * Verify that we can't destroy an active pool, create an existing pool,
2361 * or create a pool with a bad vdev spec.
2362 */
2363 /* ARGSUSED */
2364 void
2365 ztest_spa_create_destroy(ztest_ds_t *zd, uint64_t id)
2366 {
2367 ztest_shared_opts_t *zo = &ztest_opts;
2368 spa_t *spa;
2369 nvlist_t *nvroot;
2370
2371 /*
2372 * Attempt to create using a bad file.
2373 */
2374 nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1);
2375 VERIFY3U(ENOENT, ==,
2376 spa_create("ztest_bad_file", nvroot, NULL, NULL));
2377 nvlist_free(nvroot);
2378
2379 /*
2380 * Attempt to create using a bad mirror.
2381 */
2382 nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 2, 1);
2383 VERIFY3U(ENOENT, ==,
2384 spa_create("ztest_bad_mirror", nvroot, NULL, NULL));
2385 nvlist_free(nvroot);
2386
2387 /*
2388 * Attempt to create an existing pool. It shouldn't matter
2389 * what's in the nvroot; we should fail with EEXIST.
2390 */
2391 (void) rw_rdlock(&ztest_name_lock);
2392 nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1);
2393 VERIFY3U(EEXIST, ==, spa_create(zo->zo_pool, nvroot, NULL, NULL));
2394 nvlist_free(nvroot);
2395 VERIFY3U(0, ==, spa_open(zo->zo_pool, &spa, FTAG));
2396 VERIFY3U(EBUSY, ==, spa_destroy(zo->zo_pool));
2397 spa_close(spa, FTAG);
2398
2399 (void) rw_unlock(&ztest_name_lock);
2400 }
2401
2402 /* ARGSUSED */
2403 void
2404 ztest_spa_upgrade(ztest_ds_t *zd, uint64_t id)
2405 {
2406 spa_t *spa;
2407 uint64_t initial_version = SPA_VERSION_INITIAL;
2408 uint64_t version, newversion;
2409 nvlist_t *nvroot, *props;
2410 char *name;
2411
2412 VERIFY0(mutex_lock(&ztest_vdev_lock));
2413 name = kmem_asprintf("%s_upgrade", ztest_opts.zo_pool);
2414
2415 /*
2416 * Clean up from previous runs.
2417 */
2418 (void) spa_destroy(name);
2419
2420 nvroot = make_vdev_root(NULL, NULL, name, ztest_opts.zo_vdev_size, 0,
2421 0, ztest_opts.zo_raidz, ztest_opts.zo_mirrors, 1);
2422
2423 /*
2424 * If we're configuring a RAIDZ device then make sure that the
2425 * the initial version is capable of supporting that feature.
2426 */
2427 switch (ztest_opts.zo_raidz_parity) {
2428 case 0:
2429 case 1:
2430 initial_version = SPA_VERSION_INITIAL;
2431 break;
2432 case 2:
2433 initial_version = SPA_VERSION_RAIDZ2;
2434 break;
2435 case 3:
2436 initial_version = SPA_VERSION_RAIDZ3;
2437 break;
2438 }
2439
2440 /*
2441 * Create a pool with a spa version that can be upgraded. Pick
2442 * a value between initial_version and SPA_VERSION_BEFORE_FEATURES.
2443 */
2444 do {
2445 version = ztest_random_spa_version(initial_version);
2446 } while (version > SPA_VERSION_BEFORE_FEATURES);
2447
2448 props = fnvlist_alloc();
2449 fnvlist_add_uint64(props,
2450 zpool_prop_to_name(ZPOOL_PROP_VERSION), version);
2451 VERIFY0(spa_create(name, nvroot, props, NULL));
2452 fnvlist_free(nvroot);
2453 fnvlist_free(props);
2454
2455 VERIFY0(spa_open(name, &spa, FTAG));
2456 VERIFY3U(spa_version(spa), ==, version);
2457 newversion = ztest_random_spa_version(version + 1);
2458
2459 if (ztest_opts.zo_verbose >= 4) {
2460 (void) printf("upgrading spa version from %llu to %llu\n",
2461 (u_longlong_t)version, (u_longlong_t)newversion);
2462 }
2463
2464 spa_upgrade(spa, newversion);
2465 VERIFY3U(spa_version(spa), >, version);
2466 VERIFY3U(spa_version(spa), ==, fnvlist_lookup_uint64(spa->spa_config,
2467 zpool_prop_to_name(ZPOOL_PROP_VERSION)));
2468 spa_close(spa, FTAG);
2469
2470 strfree(name);
2471 VERIFY0(mutex_unlock(&ztest_vdev_lock));
2472 }
2473
2474 static vdev_t *
2475 vdev_lookup_by_path(vdev_t *vd, const char *path)
2476 {
2477 vdev_t *mvd;
2478
2479 if (vd->vdev_path != NULL && strcmp(path, vd->vdev_path) == 0)
2480 return (vd);
2481
2482 for (int c = 0; c < vd->vdev_children; c++)
2483 if ((mvd = vdev_lookup_by_path(vd->vdev_child[c], path)) !=
2484 NULL)
2485 return (mvd);
2486
2487 return (NULL);
2488 }
2489
2490 /*
2491 * Find the first available hole which can be used as a top-level.
2492 */
2493 int
2494 find_vdev_hole(spa_t *spa)
2495 {
2496 vdev_t *rvd = spa->spa_root_vdev;
2497 int c;
2498
2499 ASSERT(spa_config_held(spa, SCL_VDEV, RW_READER) == SCL_VDEV);
2500
2501 for (c = 0; c < rvd->vdev_children; c++) {
2502 vdev_t *cvd = rvd->vdev_child[c];
2503
2504 if (cvd->vdev_ishole)
2505 break;
2506 }
2507 return (c);
2508 }
2509
2510 /*
2511 * Verify that vdev_add() works as expected.
2512 */
2513 /* ARGSUSED */
2514 void
2515 ztest_vdev_add_remove(ztest_ds_t *zd, uint64_t id)
2516 {
2517 ztest_shared_t *zs = ztest_shared;
2518 spa_t *spa = ztest_spa;
2519 uint64_t leaves;
2520 uint64_t guid;
2521 nvlist_t *nvroot;
2522 int error;
2523
2524 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2525 leaves = MAX(zs->zs_mirrors + zs->zs_splits, 1) * ztest_opts.zo_raidz;
2526
2527 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2528
2529 ztest_shared->zs_vdev_next_leaf = find_vdev_hole(spa) * leaves;
2530
2531 /*
2532 * If we have slogs then remove them 1/4 of the time.
2533 */
2534 if (spa_has_slogs(spa) && ztest_random(4) == 0) {
2535 /*
2536 * Grab the guid from the head of the log class rotor.
2537 */
2538 guid = spa_log_class(spa)->mc_rotor->mg_vd->vdev_guid;
2539
2540 spa_config_exit(spa, SCL_VDEV, FTAG);
2541
2542 /*
2543 * We have to grab the zs_name_lock as writer to
2544 * prevent a race between removing a slog (dmu_objset_find)
2545 * and destroying a dataset. Removing the slog will
2546 * grab a reference on the dataset which may cause
2547 * dmu_objset_destroy() to fail with EBUSY thus
2548 * leaving the dataset in an inconsistent state.
2549 */
2550 VERIFY(rw_wrlock(&ztest_name_lock) == 0);
2551 error = spa_vdev_remove(spa, guid, B_FALSE);
2552 VERIFY(rw_unlock(&ztest_name_lock) == 0);
2553
2554 if (error && error != EEXIST)
2555 fatal(0, "spa_vdev_remove() = %d", error);
2556 } else {
2557 spa_config_exit(spa, SCL_VDEV, FTAG);
2558
2559 /*
2560 * Make 1/4 of the devices be log devices.
2561 */
2562 nvroot = make_vdev_root(NULL, NULL, NULL,
2563 ztest_opts.zo_vdev_size, 0,
2564 ztest_random(4) == 0, ztest_opts.zo_raidz,
2565 zs->zs_mirrors, 1);
2566
2567 error = spa_vdev_add(spa, nvroot);
2568 nvlist_free(nvroot);
2569
2570 if (error == ENOSPC)
2571 ztest_record_enospc("spa_vdev_add");
2572 else if (error != 0)
2573 fatal(0, "spa_vdev_add() = %d", error);
2574 }
2575
2576 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2577 }
2578
2579 /*
2580 * Verify that adding/removing aux devices (l2arc, hot spare) works as expected.
2581 */
2582 /* ARGSUSED */
2583 void
2584 ztest_vdev_aux_add_remove(ztest_ds_t *zd, uint64_t id)
2585 {
2586 ztest_shared_t *zs = ztest_shared;
2587 spa_t *spa = ztest_spa;
2588 vdev_t *rvd = spa->spa_root_vdev;
2589 spa_aux_vdev_t *sav;
2590 char *aux;
2591 uint64_t guid = 0;
2592 int error;
2593
2594 if (ztest_random(2) == 0) {
2595 sav = &spa->spa_spares;
2596 aux = ZPOOL_CONFIG_SPARES;
2597 } else {
2598 sav = &spa->spa_l2cache;
2599 aux = ZPOOL_CONFIG_L2CACHE;
2600 }
2601
2602 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2603
2604 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2605
2606 if (sav->sav_count != 0 && ztest_random(4) == 0) {
2607 /*
2608 * Pick a random device to remove.
2609 */
2610 guid = sav->sav_vdevs[ztest_random(sav->sav_count)]->vdev_guid;
2611 } else {
2612 /*
2613 * Find an unused device we can add.
2614 */
2615 zs->zs_vdev_aux = 0;
2616 for (;;) {
2617 char path[MAXPATHLEN];
2618 int c;
2619 (void) snprintf(path, sizeof (path), ztest_aux_template,
2620 ztest_opts.zo_dir, ztest_opts.zo_pool, aux,
2621 zs->zs_vdev_aux);
2622 for (c = 0; c < sav->sav_count; c++)
2623 if (strcmp(sav->sav_vdevs[c]->vdev_path,
2624 path) == 0)
2625 break;
2626 if (c == sav->sav_count &&
2627 vdev_lookup_by_path(rvd, path) == NULL)
2628 break;
2629 zs->zs_vdev_aux++;
2630 }
2631 }
2632
2633 spa_config_exit(spa, SCL_VDEV, FTAG);
2634
2635 if (guid == 0) {
2636 /*
2637 * Add a new device.
2638 */
2639 nvlist_t *nvroot = make_vdev_root(NULL, aux, NULL,
2640 (ztest_opts.zo_vdev_size * 5) / 4, 0, 0, 0, 0, 1);
2641 error = spa_vdev_add(spa, nvroot);
2642 if (error != 0)
2643 fatal(0, "spa_vdev_add(%p) = %d", nvroot, error);
2644 nvlist_free(nvroot);
2645 } else {
2646 /*
2647 * Remove an existing device. Sometimes, dirty its
2648 * vdev state first to make sure we handle removal
2649 * of devices that have pending state changes.
2650 */
2651 if (ztest_random(2) == 0)
2652 (void) vdev_online(spa, guid, 0, NULL);
2653
2654 error = spa_vdev_remove(spa, guid, B_FALSE);
2655 if (error != 0 && error != EBUSY)
2656 fatal(0, "spa_vdev_remove(%llu) = %d", guid, error);
2657 }
2658
2659 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2660 }
2661
2662 /*
2663 * split a pool if it has mirror tlvdevs
2664 */
2665 /* ARGSUSED */
2666 void
2667 ztest_split_pool(ztest_ds_t *zd, uint64_t id)
2668 {
2669 ztest_shared_t *zs = ztest_shared;
2670 spa_t *spa = ztest_spa;
2671 vdev_t *rvd = spa->spa_root_vdev;
2672 nvlist_t *tree, **child, *config, *split, **schild;
2673 uint_t c, children, schildren = 0, lastlogid = 0;
2674 int error = 0;
2675
2676 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2677
2678 /* ensure we have a useable config; mirrors of raidz aren't supported */
2679 if (zs->zs_mirrors < 3 || ztest_opts.zo_raidz > 1) {
2680 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2681 return;
2682 }
2683
2684 /* clean up the old pool, if any */
2685 (void) spa_destroy("splitp");
2686
2687 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2688
2689 /* generate a config from the existing config */
2690 mutex_enter(&spa->spa_props_lock);
2691 VERIFY(nvlist_lookup_nvlist(spa->spa_config, ZPOOL_CONFIG_VDEV_TREE,
2692 &tree) == 0);
2693 mutex_exit(&spa->spa_props_lock);
2694
2695 VERIFY(nvlist_lookup_nvlist_array(tree, ZPOOL_CONFIG_CHILDREN, &child,
2696 &children) == 0);
2697
2698 schild = malloc(rvd->vdev_children * sizeof (nvlist_t *));
2699 for (c = 0; c < children; c++) {
2700 vdev_t *tvd = rvd->vdev_child[c];
2701 nvlist_t **mchild;
2702 uint_t mchildren;
2703
2704 if (tvd->vdev_islog || tvd->vdev_ops == &vdev_hole_ops) {
2705 VERIFY(nvlist_alloc(&schild[schildren], NV_UNIQUE_NAME,
2706 0) == 0);
2707 VERIFY(nvlist_add_string(schild[schildren],
2708 ZPOOL_CONFIG_TYPE, VDEV_TYPE_HOLE) == 0);
2709 VERIFY(nvlist_add_uint64(schild[schildren],
2710 ZPOOL_CONFIG_IS_HOLE, 1) == 0);
2711 if (lastlogid == 0)
2712 lastlogid = schildren;
2713 ++schildren;
2714 continue;
2715 }
2716 lastlogid = 0;
2717 VERIFY(nvlist_lookup_nvlist_array(child[c],
2718 ZPOOL_CONFIG_CHILDREN, &mchild, &mchildren) == 0);
2719 VERIFY(nvlist_dup(mchild[0], &schild[schildren++], 0) == 0);
2720 }
2721
2722 /* OK, create a config that can be used to split */
2723 VERIFY(nvlist_alloc(&split, NV_UNIQUE_NAME, 0) == 0);
2724 VERIFY(nvlist_add_string(split, ZPOOL_CONFIG_TYPE,
2725 VDEV_TYPE_ROOT) == 0);
2726 VERIFY(nvlist_add_nvlist_array(split, ZPOOL_CONFIG_CHILDREN, schild,
2727 lastlogid != 0 ? lastlogid : schildren) == 0);
2728
2729 VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, 0) == 0);
2730 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, split) == 0);
2731
2732 for (c = 0; c < schildren; c++)
2733 nvlist_free(schild[c]);
2734 free(schild);
2735 nvlist_free(split);
2736
2737 spa_config_exit(spa, SCL_VDEV, FTAG);
2738
2739 (void) rw_wrlock(&ztest_name_lock);
2740 error = spa_vdev_split_mirror(spa, "splitp", config, NULL, B_FALSE);
2741 (void) rw_unlock(&ztest_name_lock);
2742
2743 nvlist_free(config);
2744
2745 if (error == 0) {
2746 (void) printf("successful split - results:\n");
2747 mutex_enter(&spa_namespace_lock);
2748 show_pool_stats(spa);
2749 show_pool_stats(spa_lookup("splitp"));
2750 mutex_exit(&spa_namespace_lock);
2751 ++zs->zs_splits;
2752 --zs->zs_mirrors;
2753 }
2754 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2755
2756 }
2757
2758 /*
2759 * Verify that we can attach and detach devices.
2760 */
2761 /* ARGSUSED */
2762 void
2763 ztest_vdev_attach_detach(ztest_ds_t *zd, uint64_t id)
2764 {
2765 ztest_shared_t *zs = ztest_shared;
2766 spa_t *spa = ztest_spa;
2767 spa_aux_vdev_t *sav = &spa->spa_spares;
2768 vdev_t *rvd = spa->spa_root_vdev;
2769 vdev_t *oldvd, *newvd, *pvd;
2770 nvlist_t *root;
2771 uint64_t leaves;
2772 uint64_t leaf, top;
2773 uint64_t ashift = ztest_get_ashift();
2774 uint64_t oldguid, pguid;
2775 uint64_t oldsize, newsize;
2776 char oldpath[MAXPATHLEN], newpath[MAXPATHLEN];
2777 int replacing;
2778 int oldvd_has_siblings = B_FALSE;
2779 int newvd_is_spare = B_FALSE;
2780 int oldvd_is_log;
2781 int error, expected_error;
2782
2783 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2784 leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz;
2785
2786 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2787
2788 /*
2789 * Decide whether to do an attach or a replace.
2790 */
2791 replacing = ztest_random(2);
2792
2793 /*
2794 * Pick a random top-level vdev.
2795 */
2796 top = ztest_random_vdev_top(spa, B_TRUE);
2797
2798 /*
2799 * Pick a random leaf within it.
2800 */
2801 leaf = ztest_random(leaves);
2802
2803 /*
2804 * Locate this vdev.
2805 */
2806 oldvd = rvd->vdev_child[top];
2807 if (zs->zs_mirrors >= 1) {
2808 ASSERT(oldvd->vdev_ops == &vdev_mirror_ops);
2809 ASSERT(oldvd->vdev_children >= zs->zs_mirrors);
2810 oldvd = oldvd->vdev_child[leaf / ztest_opts.zo_raidz];
2811 }
2812 if (ztest_opts.zo_raidz > 1) {
2813 ASSERT(oldvd->vdev_ops == &vdev_raidz_ops);
2814 ASSERT(oldvd->vdev_children == ztest_opts.zo_raidz);
2815 oldvd = oldvd->vdev_child[leaf % ztest_opts.zo_raidz];
2816 }
2817
2818 /*
2819 * If we're already doing an attach or replace, oldvd may be a
2820 * mirror vdev -- in which case, pick a random child.
2821 */
2822 while (oldvd->vdev_children != 0) {
2823 oldvd_has_siblings = B_TRUE;
2824 ASSERT(oldvd->vdev_children >= 2);
2825 oldvd = oldvd->vdev_child[ztest_random(oldvd->vdev_children)];
2826 }
2827
2828 oldguid = oldvd->vdev_guid;
2829 oldsize = vdev_get_min_asize(oldvd);
2830 oldvd_is_log = oldvd->vdev_top->vdev_islog;
2831 (void) strcpy(oldpath, oldvd->vdev_path);
2832 pvd = oldvd->vdev_parent;
2833 pguid = pvd->vdev_guid;
2834
2835 /*
2836 * If oldvd has siblings, then half of the time, detach it.
2837 */
2838 if (oldvd_has_siblings && ztest_random(2) == 0) {
2839 spa_config_exit(spa, SCL_VDEV, FTAG);
2840 error = spa_vdev_detach(spa, oldguid, pguid, B_FALSE);
2841 if (error != 0 && error != ENODEV && error != EBUSY &&
2842 error != ENOTSUP)
2843 fatal(0, "detach (%s) returned %d", oldpath, error);
2844 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2845 return;
2846 }
2847
2848 /*
2849 * For the new vdev, choose with equal probability between the two
2850 * standard paths (ending in either 'a' or 'b') or a random hot spare.
2851 */
2852 if (sav->sav_count != 0 && ztest_random(3) == 0) {
2853 newvd = sav->sav_vdevs[ztest_random(sav->sav_count)];
2854 newvd_is_spare = B_TRUE;
2855 (void) strcpy(newpath, newvd->vdev_path);
2856 } else {
2857 (void) snprintf(newpath, sizeof (newpath), ztest_dev_template,
2858 ztest_opts.zo_dir, ztest_opts.zo_pool,
2859 top * leaves + leaf);
2860 if (ztest_random(2) == 0)
2861 newpath[strlen(newpath) - 1] = 'b';
2862 newvd = vdev_lookup_by_path(rvd, newpath);
2863 }
2864
2865 if (newvd) {
2866 newsize = vdev_get_min_asize(newvd);
2867 } else {
2868 /*
2869 * Make newsize a little bigger or smaller than oldsize.
2870 * If it's smaller, the attach should fail.
2871 * If it's larger, and we're doing a replace,
2872 * we should get dynamic LUN growth when we're done.
2873 */
2874 newsize = 10 * oldsize / (9 + ztest_random(3));
2875 }
2876
2877 /*
2878 * If pvd is not a mirror or root, the attach should fail with ENOTSUP,
2879 * unless it's a replace; in that case any non-replacing parent is OK.
2880 *
2881 * If newvd is already part of the pool, it should fail with EBUSY.
2882 *
2883 * If newvd is too small, it should fail with EOVERFLOW.
2884 */
2885 if (pvd->vdev_ops != &vdev_mirror_ops &&
2886 pvd->vdev_ops != &vdev_root_ops && (!replacing ||
2887 pvd->vdev_ops == &vdev_replacing_ops ||
2888 pvd->vdev_ops == &vdev_spare_ops))
2889 expected_error = ENOTSUP;
2890 else if (newvd_is_spare && (!replacing || oldvd_is_log))
2891 expected_error = ENOTSUP;
2892 else if (newvd == oldvd)
2893 expected_error = replacing ? 0 : EBUSY;
2894 else if (vdev_lookup_by_path(rvd, newpath) != NULL)
2895 expected_error = EBUSY;
2896 else if (newsize < oldsize)
2897 expected_error = EOVERFLOW;
2898 else if (ashift > oldvd->vdev_top->vdev_ashift)
2899 expected_error = EDOM;
2900 else
2901 expected_error = 0;
2902
2903 spa_config_exit(spa, SCL_VDEV, FTAG);
2904
2905 /*
2906 * Build the nvlist describing newpath.
2907 */
2908 root = make_vdev_root(newpath, NULL, NULL, newvd == NULL ? newsize : 0,
2909 ashift, 0, 0, 0, 1);
2910
2911 error = spa_vdev_attach(spa, oldguid, root, replacing);
2912
2913 nvlist_free(root);
2914
2915 /*
2916 * If our parent was the replacing vdev, but the replace completed,
2917 * then instead of failing with ENOTSUP we may either succeed,
2918 * fail with ENODEV, or fail with EOVERFLOW.
2919 */
2920 if (expected_error == ENOTSUP &&
2921 (error == 0 || error == ENODEV || error == EOVERFLOW))
2922 expected_error = error;
2923
2924 /*
2925 * If someone grew the LUN, the replacement may be too small.
2926 */
2927 if (error == EOVERFLOW || error == EBUSY)
2928 expected_error = error;
2929
2930 /* XXX workaround 6690467 */
2931 if (error != expected_error && expected_error != EBUSY) {
2932 fatal(0, "attach (%s %llu, %s %llu, %d) "
2933 "returned %d, expected %d",
2934 oldpath, oldsize, newpath,
2935 newsize, replacing, error, expected_error);
2936 }
2937
2938 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2939 }
2940
2941 /*
2942 * Callback function which expands the physical size of the vdev.
2943 */
2944 vdev_t *
2945 grow_vdev(vdev_t *vd, void *arg)
2946 {
2947 spa_t *spa = vd->vdev_spa;
2948 size_t *newsize = arg;
2949 size_t fsize;
2950 int fd;
2951
2952 ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
2953 ASSERT(vd->vdev_ops->vdev_op_leaf);
2954
2955 if ((fd = open(vd->vdev_path, O_RDWR)) == -1)
2956 return (vd);
2957
2958 fsize = lseek(fd, 0, SEEK_END);
2959 (void) ftruncate(fd, *newsize);
2960
2961 if (ztest_opts.zo_verbose >= 6) {
2962 (void) printf("%s grew from %lu to %lu bytes\n",
2963 vd->vdev_path, (ulong_t)fsize, (ulong_t)*newsize);
2964 }
2965 (void) close(fd);
2966 return (NULL);
2967 }
2968
2969 /*
2970 * Callback function which expands a given vdev by calling vdev_online().
2971 */
2972 /* ARGSUSED */
2973 vdev_t *
2974 online_vdev(vdev_t *vd, void *arg)
2975 {
2976 spa_t *spa = vd->vdev_spa;
2977 vdev_t *tvd = vd->vdev_top;
2978 uint64_t guid = vd->vdev_guid;
2979 uint64_t generation = spa->spa_config_generation + 1;
2980 vdev_state_t newstate = VDEV_STATE_UNKNOWN;
2981 int error;
2982
2983 ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
2984 ASSERT(vd->vdev_ops->vdev_op_leaf);
2985
2986 /* Calling vdev_online will initialize the new metaslabs */
2987 spa_config_exit(spa, SCL_STATE, spa);
2988 error = vdev_online(spa, guid, ZFS_ONLINE_EXPAND, &newstate);
2989 spa_config_enter(spa, SCL_STATE, spa, RW_READER);
2990
2991 /*
2992 * If vdev_online returned an error or the underlying vdev_open
2993 * failed then we abort the expand. The only way to know that
2994 * vdev_open fails is by checking the returned newstate.
2995 */
2996 if (error || newstate != VDEV_STATE_HEALTHY) {
2997 if (ztest_opts.zo_verbose >= 5) {
2998 (void) printf("Unable to expand vdev, state %llu, "
2999 "error %d\n", (u_longlong_t)newstate, error);
3000 }
3001 return (vd);
3002 }
3003 ASSERT3U(newstate, ==, VDEV_STATE_HEALTHY);
3004
3005 /*
3006 * Since we dropped the lock we need to ensure that we're
3007 * still talking to the original vdev. It's possible this
3008 * vdev may have been detached/replaced while we were
3009 * trying to online it.
3010 */
3011 if (generation != spa->spa_config_generation) {
3012 if (ztest_opts.zo_verbose >= 5) {
3013 (void) printf("vdev configuration has changed, "
3014 "guid %llu, state %llu, expected gen %llu, "
3015 "got gen %llu\n",
3016 (u_longlong_t)guid,
3017 (u_longlong_t)tvd->vdev_state,
3018 (u_longlong_t)generation,
3019 (u_longlong_t)spa->spa_config_generation);
3020 }
3021 return (vd);
3022 }
3023 return (NULL);
3024 }
3025
3026 /*
3027 * Traverse the vdev tree calling the supplied function.
3028 * We continue to walk the tree until we either have walked all
3029 * children or we receive a non-NULL return from the callback.
3030 * If a NULL callback is passed, then we just return back the first
3031 * leaf vdev we encounter.
3032 */
3033 vdev_t *
3034 vdev_walk_tree(vdev_t *vd, vdev_t *(*func)(vdev_t *, void *), void *arg)
3035 {
3036 if (vd->vdev_ops->vdev_op_leaf) {
3037 if (func == NULL)
3038 return (vd);
3039 else
3040 return (func(vd, arg));
3041 }
3042
3043 for (uint_t c = 0; c < vd->vdev_children; c++) {
3044 vdev_t *cvd = vd->vdev_child[c];
3045 if ((cvd = vdev_walk_tree(cvd, func, arg)) != NULL)
3046 return (cvd);
3047 }
3048 return (NULL);
3049 }
3050
3051 /*
3052 * Verify that dynamic LUN growth works as expected.
3053 */
3054 /* ARGSUSED */
3055 void
3056 ztest_vdev_LUN_growth(ztest_ds_t *zd, uint64_t id)
3057 {
3058 spa_t *spa = ztest_spa;
3059 vdev_t *vd, *tvd;
3060 metaslab_class_t *mc;
3061 metaslab_group_t *mg;
3062 size_t psize, newsize;
3063 uint64_t top;
3064 uint64_t old_class_space, new_class_space, old_ms_count, new_ms_count;
3065
3066 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
3067 spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3068
3069 top = ztest_random_vdev_top(spa, B_TRUE);
3070
3071 tvd = spa->spa_root_vdev->vdev_child[top];
3072 mg = tvd->vdev_mg;
3073 mc = mg->mg_class;
3074 old_ms_count = tvd->vdev_ms_count;
3075 old_class_space = metaslab_class_get_space(mc);
3076
3077 /*
3078 * Determine the size of the first leaf vdev associated with
3079 * our top-level device.
3080 */
3081 vd = vdev_walk_tree(tvd, NULL, NULL);
3082 ASSERT3P(vd, !=, NULL);
3083 ASSERT(vd->vdev_ops->vdev_op_leaf);
3084
3085 psize = vd->vdev_psize;
3086
3087 /*
3088 * We only try to expand the vdev if it's healthy, less than 4x its
3089 * original size, and it has a valid psize.
3090 */
3091 if (tvd->vdev_state != VDEV_STATE_HEALTHY ||
3092 psize == 0 || psize >= 4 * ztest_opts.zo_vdev_size) {
3093 spa_config_exit(spa, SCL_STATE, spa);
3094 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3095 return;
3096 }
3097 ASSERT(psize > 0);
3098 newsize = psize + psize / 8;
3099 ASSERT3U(newsize, >, psize);
3100
3101 if (ztest_opts.zo_verbose >= 6) {
3102 (void) printf("Expanding LUN %s from %lu to %lu\n",
3103 vd->vdev_path, (ulong_t)psize, (ulong_t)newsize);
3104 }
3105
3106 /*
3107 * Growing the vdev is a two step process:
3108 * 1). expand the physical size (i.e. relabel)
3109 * 2). online the vdev to create the new metaslabs
3110 */
3111 if (vdev_walk_tree(tvd, grow_vdev, &newsize) != NULL ||
3112 vdev_walk_tree(tvd, online_vdev, NULL) != NULL ||
3113 tvd->vdev_state != VDEV_STATE_HEALTHY) {
3114 if (ztest_opts.zo_verbose >= 5) {
3115 (void) printf("Could not expand LUN because "
3116 "the vdev configuration changed.\n");
3117 }
3118 spa_config_exit(spa, SCL_STATE, spa);
3119 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3120 return;
3121 }
3122
3123 spa_config_exit(spa, SCL_STATE, spa);
3124
3125 /*
3126 * Expanding the LUN will update the config asynchronously,
3127 * thus we must wait for the async thread to complete any
3128 * pending tasks before proceeding.
3129 */
3130 for (;;) {
3131 boolean_t done;
3132 mutex_enter(&spa->spa_async_lock);
3133 done = (spa->spa_async_thread == NULL && !spa->spa_async_tasks);
3134 mutex_exit(&spa->spa_async_lock);
3135 if (done)
3136 break;
3137 txg_wait_synced(spa_get_dsl(spa), 0);
3138 (void) poll(NULL, 0, 100);
3139 }
3140
3141 spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3142
3143 tvd = spa->spa_root_vdev->vdev_child[top];
3144 new_ms_count = tvd->vdev_ms_count;
3145 new_class_space = metaslab_class_get_space(mc);
3146
3147 if (tvd->vdev_mg != mg || mg->mg_class != mc) {
3148 if (ztest_opts.zo_verbose >= 5) {
3149 (void) printf("Could not verify LUN expansion due to "
3150 "intervening vdev offline or remove.\n");
3151 }
3152 spa_config_exit(spa, SCL_STATE, spa);
3153 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3154 return;
3155 }
3156
3157 /*
3158 * Make sure we were able to grow the vdev.
3159 */
3160 if (new_ms_count <= old_ms_count)
3161 fatal(0, "LUN expansion failed: ms_count %llu <= %llu\n",
3162 old_ms_count, new_ms_count);
3163
3164 /*
3165 * Make sure we were able to grow the pool.
3166 */
3167 if (new_class_space <= old_class_space)
3168 fatal(0, "LUN expansion failed: class_space %llu <= %llu\n",
3169 old_class_space, new_class_space);
3170
3171 if (ztest_opts.zo_verbose >= 5) {
3172 char oldnumbuf[NN_NUMBUF_SZ], newnumbuf[NN_NUMBUF_SZ];
3173
3174 nicenum(old_class_space, oldnumbuf, sizeof (oldnumbuf));
3175 nicenum(new_class_space, newnumbuf, sizeof (newnumbuf));
3176 (void) printf("%s grew from %s to %s\n",
3177 spa->spa_name, oldnumbuf, newnumbuf);
3178 }
3179
3180 spa_config_exit(spa, SCL_STATE, spa);
3181 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3182 }
3183
3184 /*
3185 * Verify that dmu_objset_{create,destroy,open,close} work as expected.
3186 */
3187 /* ARGSUSED */
3188 static void
3189 ztest_objset_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
3190 {
3191 /*
3192 * Create the objects common to all ztest datasets.
3193 */
3194 VERIFY(zap_create_claim(os, ZTEST_DIROBJ,
3195 DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx) == 0);
3196 }
3197
3198 static int
3199 ztest_dataset_create(char *dsname)
3200 {
3201 uint64_t zilset = ztest_random(100);
3202 int err = dmu_objset_create(dsname, DMU_OST_OTHER, 0,
3203 ztest_objset_create_cb, NULL);
3204
3205 if (err || zilset < 80)
3206 return (err);
3207
3208 if (ztest_opts.zo_verbose >= 6)
3209 (void) printf("Setting dataset %s to sync always\n", dsname);
3210 return (ztest_dsl_prop_set_uint64(dsname, ZFS_PROP_SYNC,
3211 ZFS_SYNC_ALWAYS, B_FALSE));
3212 }
3213
3214 /* ARGSUSED */
3215 static int
3216 ztest_objset_destroy_cb(const char *name, void *arg)
3217 {
3218 objset_t *os;
3219 dmu_object_info_t doi;
3220 int error;
3221
3222 /*
3223 * Verify that the dataset contains a directory object.
3224 */
3225 VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_TRUE, FTAG, &os));
3226 error = dmu_object_info(os, ZTEST_DIROBJ, &doi);
3227 if (error != ENOENT) {
3228 /* We could have crashed in the middle of destroying it */
3229 ASSERT0(error);
3230 ASSERT3U(doi.doi_type, ==, DMU_OT_ZAP_OTHER);
3231 ASSERT3S(doi.doi_physical_blocks_512, >=, 0);
3232 }
3233 dmu_objset_disown(os, FTAG);
3234
3235 /*
3236 * Destroy the dataset.
3237 */
3238 if (strchr(name, '@') != NULL) {
3239 VERIFY0(dsl_destroy_snapshot(name, B_TRUE));
3240 } else {
3241 error = dsl_destroy_head(name);
3242 /* There could be a hold on this dataset */
3243 if (error != EBUSY)
3244 ASSERT0(error);
3245 }
3246 return (0);
3247 }
3248
3249 static boolean_t
3250 ztest_snapshot_create(char *osname, uint64_t id)
3251 {
3252 char snapname[ZFS_MAX_DATASET_NAME_LEN];
3253 int error;
3254
3255 (void) snprintf(snapname, sizeof (snapname), "%llu", (u_longlong_t)id);
3256
3257 error = dmu_objset_snapshot_one(osname, snapname);
3258 if (error == ENOSPC) {
3259 ztest_record_enospc(FTAG);
3260 return (B_FALSE);
3261 }
3262 if (error != 0 && error != EEXIST) {
3263 fatal(0, "ztest_snapshot_create(%s@%s) = %d", osname,
3264 snapname, error);
3265 }
3266 return (B_TRUE);
3267 }
3268
3269 static boolean_t
3270 ztest_snapshot_destroy(char *osname, uint64_t id)
3271 {
3272 char snapname[ZFS_MAX_DATASET_NAME_LEN];
3273 int error;
3274
3275 (void) snprintf(snapname, sizeof (snapname), "%s@%llu", osname,
3276 (u_longlong_t)id);
3277
3278 error = dsl_destroy_snapshot(snapname, B_FALSE);
3279 if (error != 0 && error != ENOENT)
3280 fatal(0, "ztest_snapshot_destroy(%s) = %d", snapname, error);
3281 return (B_TRUE);
3282 }
3283
3284 /* ARGSUSED */
3285 void
3286 ztest_dmu_objset_create_destroy(ztest_ds_t *zd, uint64_t id)
3287 {
3288 ztest_ds_t zdtmp;
3289 int iters;
3290 int error;
3291 objset_t *os, *os2;
3292 char name[ZFS_MAX_DATASET_NAME_LEN];
3293 zilog_t *zilog;
3294
3295 (void) rw_rdlock(&ztest_name_lock);
3296
3297 (void) snprintf(name, sizeof (name), "%s/temp_%llu",
3298 ztest_opts.zo_pool, (u_longlong_t)id);
3299
3300 /*
3301 * If this dataset exists from a previous run, process its replay log
3302 * half of the time. If we don't replay it, then dmu_objset_destroy()
3303 * (invoked from ztest_objset_destroy_cb()) should just throw it away.
3304 */
3305 if (ztest_random(2) == 0 &&
3306 dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os) == 0) {
3307 ztest_zd_init(&zdtmp, NULL, os);
3308 zil_replay(os, &zdtmp, ztest_replay_vector);
3309 ztest_zd_fini(&zdtmp);
3310 dmu_objset_disown(os, FTAG);
3311 }
3312
3313 /*
3314 * There may be an old instance of the dataset we're about to
3315 * create lying around from a previous run. If so, destroy it
3316 * and all of its snapshots.
3317 */
3318 (void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
3319 DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
3320
3321 /*
3322 * Verify that the destroyed dataset is no longer in the namespace.
3323 */
3324 VERIFY3U(ENOENT, ==, dmu_objset_own(name, DMU_OST_OTHER, B_TRUE,
3325 FTAG, &os));
3326
3327 /*
3328 * Verify that we can create a new dataset.
3329 */
3330 error = ztest_dataset_create(name);
3331 if (error) {
3332 if (error == ENOSPC) {
3333 ztest_record_enospc(FTAG);
3334 (void) rw_unlock(&ztest_name_lock);
3335 return;
3336 }
3337 fatal(0, "dmu_objset_create(%s) = %d", name, error);
3338 }
3339
3340 VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os));
3341
3342 ztest_zd_init(&zdtmp, NULL, os);
3343
3344 /*
3345 * Open the intent log for it.
3346 */
3347 zilog = zil_open(os, ztest_get_data);
3348
3349 /*
3350 * Put some objects in there, do a little I/O to them,
3351 * and randomly take a couple of snapshots along the way.
3352 */
3353 iters = ztest_random(5);
3354 for (int i = 0; i < iters; i++) {
3355 ztest_dmu_object_alloc_free(&zdtmp, id);
3356 if (ztest_random(iters) == 0)
3357 (void) ztest_snapshot_create(name, i);
3358 }
3359
3360 /*
3361 * Verify that we cannot create an existing dataset.
3362 */
3363 VERIFY3U(EEXIST, ==,
3364 dmu_objset_create(name, DMU_OST_OTHER, 0, NULL, NULL));
3365
3366 /*
3367 * Verify that we can hold an objset that is also owned.
3368 */
3369 VERIFY3U(0, ==, dmu_objset_hold(name, FTAG, &os2));
3370 dmu_objset_rele(os2, FTAG);
3371
3372 /*
3373 * Verify that we cannot own an objset that is already owned.
3374 */
3375 VERIFY3U(EBUSY, ==,
3376 dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os2));
3377
3378 zil_close(zilog);
3379 dmu_objset_disown(os, FTAG);
3380 ztest_zd_fini(&zdtmp);
3381
3382 (void) rw_unlock(&ztest_name_lock);
3383 }
3384
3385 /*
3386 * Verify that dmu_snapshot_{create,destroy,open,close} work as expected.
3387 */
3388 void
3389 ztest_dmu_snapshot_create_destroy(ztest_ds_t *zd, uint64_t id)
3390 {
3391 (void) rw_rdlock(&ztest_name_lock);
3392 (void) ztest_snapshot_destroy(zd->zd_name, id);
3393 (void) ztest_snapshot_create(zd->zd_name, id);
3394 (void) rw_unlock(&ztest_name_lock);
3395 }
3396
3397 /*
3398 * Cleanup non-standard snapshots and clones.
3399 */
3400 void
3401 ztest_dsl_dataset_cleanup(char *osname, uint64_t id)
3402 {
3403 char snap1name[ZFS_MAX_DATASET_NAME_LEN];
3404 char clone1name[ZFS_MAX_DATASET_NAME_LEN];
3405 char snap2name[ZFS_MAX_DATASET_NAME_LEN];
3406 char clone2name[ZFS_MAX_DATASET_NAME_LEN];
3407 char snap3name[ZFS_MAX_DATASET_NAME_LEN];
3408 int error;
3409
3410 (void) snprintf(snap1name, sizeof (snap1name),
3411 "%s@s1_%llu", osname, id);
3412 (void) snprintf(clone1name, sizeof (clone1name),
3413 "%s/c1_%llu", osname, id);
3414 (void) snprintf(snap2name, sizeof (snap2name),
3415 "%s@s2_%llu", clone1name, id);
3416 (void) snprintf(clone2name, sizeof (clone2name),
3417 "%s/c2_%llu", osname, id);
3418 (void) snprintf(snap3name, sizeof (snap3name),
3419 "%s@s3_%llu", clone1name, id);
3420
3421 error = dsl_destroy_head(clone2name);
3422 if (error && error != ENOENT)
3423 fatal(0, "dsl_destroy_head(%s) = %d", clone2name, error);
3424 error = dsl_destroy_snapshot(snap3name, B_FALSE);
3425 if (error && error != ENOENT)
3426 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap3name, error);
3427 error = dsl_destroy_snapshot(snap2name, B_FALSE);
3428 if (error && error != ENOENT)
3429 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap2name, error);
3430 error = dsl_destroy_head(clone1name);
3431 if (error && error != ENOENT)
3432 fatal(0, "dsl_destroy_head(%s) = %d", clone1name, error);
3433 error = dsl_destroy_snapshot(snap1name, B_FALSE);
3434 if (error && error != ENOENT)
3435 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap1name, error);
3436 }
3437
3438 /*
3439 * Verify dsl_dataset_promote handles EBUSY
3440 */
3441 void
3442 ztest_dsl_dataset_promote_busy(ztest_ds_t *zd, uint64_t id)
3443 {
3444 objset_t *os;
3445 char snap1name[ZFS_MAX_DATASET_NAME_LEN];
3446 char clone1name[ZFS_MAX_DATASET_NAME_LEN];
3447 char snap2name[ZFS_MAX_DATASET_NAME_LEN];
3448 char clone2name[ZFS_MAX_DATASET_NAME_LEN];
3449 char snap3name[ZFS_MAX_DATASET_NAME_LEN];
3450 char *osname = zd->zd_name;
3451 int error;
3452
3453 (void) rw_rdlock(&ztest_name_lock);
3454
3455 ztest_dsl_dataset_cleanup(osname, id);
3456
3457 (void) snprintf(snap1name, sizeof (snap1name),
3458 "%s@s1_%llu", osname, id);
3459 (void) snprintf(clone1name, sizeof (clone1name),
3460 "%s/c1_%llu", osname, id);
3461 (void) snprintf(snap2name, sizeof (snap2name),
3462 "%s@s2_%llu", clone1name, id);
3463 (void) snprintf(clone2name, sizeof (clone2name),
3464 "%s/c2_%llu", osname, id);
3465 (void) snprintf(snap3name, sizeof (snap3name),
3466 "%s@s3_%llu", clone1name, id);
3467
3468 error = dmu_objset_snapshot_one(osname, strchr(snap1name, '@') + 1);
3469 if (error && error != EEXIST) {
3470 if (error == ENOSPC) {
3471 ztest_record_enospc(FTAG);
3472 goto out;
3473 }
3474 fatal(0, "dmu_take_snapshot(%s) = %d", snap1name, error);
3475 }
3476
3477 error = dmu_objset_clone(clone1name, snap1name);
3478 if (error) {
3479 if (error == ENOSPC) {
3480 ztest_record_enospc(FTAG);
3481 goto out;
3482 }
3483 fatal(0, "dmu_objset_create(%s) = %d", clone1name, error);
3484 }
3485
3486 error = dmu_objset_snapshot_one(clone1name, strchr(snap2name, '@') + 1);
3487 if (error && error != EEXIST) {
3488 if (error == ENOSPC) {
3489 ztest_record_enospc(FTAG);
3490 goto out;
3491 }
3492 fatal(0, "dmu_open_snapshot(%s) = %d", snap2name, error);
3493 }
3494
3495 error = dmu_objset_snapshot_one(clone1name, strchr(snap3name, '@') + 1);
3496 if (error && error != EEXIST) {
3497 if (error == ENOSPC) {
3498 ztest_record_enospc(FTAG);
3499 goto out;
3500 }
3501 fatal(0, "dmu_open_snapshot(%s) = %d", snap3name, error);
3502 }
3503
3504 error = dmu_objset_clone(clone2name, snap3name);
3505 if (error) {
3506 if (error == ENOSPC) {
3507 ztest_record_enospc(FTAG);
3508 goto out;
3509 }
3510 fatal(0, "dmu_objset_create(%s) = %d", clone2name, error);
3511 }
3512
3513 error = dmu_objset_own(snap2name, DMU_OST_ANY, B_TRUE, FTAG, &os);
3514 if (error)
3515 fatal(0, "dmu_objset_own(%s) = %d", snap2name, error);
3516 error = dsl_dataset_promote(clone2name, NULL);
3517 if (error == ENOSPC) {
3518 dmu_objset_disown(os, FTAG);
3519 ztest_record_enospc(FTAG);
3520 goto out;
3521 }
3522 if (error != EBUSY)
3523 fatal(0, "dsl_dataset_promote(%s), %d, not EBUSY", clone2name,
3524 error);
3525 dmu_objset_disown(os, FTAG);
3526
3527 out:
3528 ztest_dsl_dataset_cleanup(osname, id);
3529
3530 (void) rw_unlock(&ztest_name_lock);
3531 }
3532
3533 /*
3534 * Verify that dmu_object_{alloc,free} work as expected.
3535 */
3536 void
3537 ztest_dmu_object_alloc_free(ztest_ds_t *zd, uint64_t id)
3538 {
3539 ztest_od_t od[4];
3540 int batchsize = sizeof (od) / sizeof (od[0]);
3541
3542 for (int b = 0; b < batchsize; b++)
3543 ztest_od_init(&od[b], id, FTAG, b, DMU_OT_UINT64_OTHER, 0, 0);
3544
3545 /*
3546 * Destroy the previous batch of objects, create a new batch,
3547 * and do some I/O on the new objects.
3548 */
3549 if (ztest_object_init(zd, od, sizeof (od), B_TRUE) != 0)
3550 return;
3551
3552 while (ztest_random(4 * batchsize) != 0)
3553 ztest_io(zd, od[ztest_random(batchsize)].od_object,
3554 ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
3555 }
3556
3557 /*
3558 * Verify that dmu_{read,write} work as expected.
3559 */
3560 void
3561 ztest_dmu_read_write(ztest_ds_t *zd, uint64_t id)
3562 {
3563 objset_t *os = zd->zd_os;
3564 ztest_od_t od[2];
3565 dmu_tx_t *tx;
3566 int i, freeit, error;
3567 uint64_t n, s, txg;
3568 bufwad_t *packbuf, *bigbuf, *pack, *bigH, *bigT;
3569 uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
3570 uint64_t chunksize = (1000 + ztest_random(1000)) * sizeof (uint64_t);
3571 uint64_t regions = 997;
3572 uint64_t stride = 123456789ULL;
3573 uint64_t width = 40;
3574 int free_percent = 5;
3575
3576 /*
3577 * This test uses two objects, packobj and bigobj, that are always
3578 * updated together (i.e. in the same tx) so that their contents are
3579 * in sync and can be compared. Their contents relate to each other
3580 * in a simple way: packobj is a dense array of 'bufwad' structures,
3581 * while bigobj is a sparse array of the same bufwads. Specifically,
3582 * for any index n, there are three bufwads that should be identical:
3583 *
3584 * packobj, at offset n * sizeof (bufwad_t)
3585 * bigobj, at the head of the nth chunk
3586 * bigobj, at the tail of the nth chunk
3587 *
3588 * The chunk size is arbitrary. It doesn't have to be a power of two,
3589 * and it doesn't have any relation to the object blocksize.
3590 * The only requirement is that it can hold at least two bufwads.
3591 *
3592 * Normally, we write the bufwad to each of these locations.
3593 * However, free_percent of the time we instead write zeroes to
3594 * packobj and perform a dmu_free_range() on bigobj. By comparing
3595 * bigobj to packobj, we can verify that the DMU is correctly
3596 * tracking which parts of an object are allocated and free,
3597 * and that the contents of the allocated blocks are correct.
3598 */
3599
3600 /*
3601 * Read the directory info. If it's the first time, set things up.
3602 */
3603 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, chunksize);
3604 ztest_od_init(&od[1], id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize);
3605
3606 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
3607 return;
3608
3609 bigobj = od[0].od_object;
3610 packobj = od[1].od_object;
3611 chunksize = od[0].od_gen;
3612 ASSERT(chunksize == od[1].od_gen);
3613
3614 /*
3615 * Prefetch a random chunk of the big object.
3616 * Our aim here is to get some async reads in flight
3617 * for blocks that we may free below; the DMU should
3618 * handle this race correctly.
3619 */
3620 n = ztest_random(regions) * stride + ztest_random(width);
3621 s = 1 + ztest_random(2 * width - 1);
3622 dmu_prefetch(os, bigobj, 0, n * chunksize, s * chunksize,
3623 ZIO_PRIORITY_SYNC_READ);
3624
3625 /*
3626 * Pick a random index and compute the offsets into packobj and bigobj.
3627 */
3628 n = ztest_random(regions) * stride + ztest_random(width);
3629 s = 1 + ztest_random(width - 1);
3630
3631 packoff = n * sizeof (bufwad_t);
3632 packsize = s * sizeof (bufwad_t);
3633
3634 bigoff = n * chunksize;
3635 bigsize = s * chunksize;
3636
3637 packbuf = umem_alloc(packsize, UMEM_NOFAIL);
3638 bigbuf = umem_alloc(bigsize, UMEM_NOFAIL);
3639
3640 /*
3641 * free_percent of the time, free a range of bigobj rather than
3642 * overwriting it.
3643 */
3644 freeit = (ztest_random(100) < free_percent);
3645
3646 /*
3647 * Read the current contents of our objects.
3648 */
3649 error = dmu_read(os, packobj, packoff, packsize, packbuf,
3650 DMU_READ_PREFETCH);
3651 ASSERT0(error);
3652 error = dmu_read(os, bigobj, bigoff, bigsize, bigbuf,
3653 DMU_READ_PREFETCH);
3654 ASSERT0(error);
3655
3656 /*
3657 * Get a tx for the mods to both packobj and bigobj.
3658 */
3659 tx = dmu_tx_create(os);
3660
3661 dmu_tx_hold_write(tx, packobj, packoff, packsize);
3662
3663 if (freeit)
3664 dmu_tx_hold_free(tx, bigobj, bigoff, bigsize);
3665 else
3666 dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
3667
3668 /* This accounts for setting the checksum/compression. */
3669 dmu_tx_hold_bonus(tx, bigobj);
3670
3671 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3672 if (txg == 0) {
3673 umem_free(packbuf, packsize);
3674 umem_free(bigbuf, bigsize);
3675 return;
3676 }
3677
3678 enum zio_checksum cksum;
3679 do {
3680 cksum = (enum zio_checksum)
3681 ztest_random_dsl_prop(ZFS_PROP_CHECKSUM);
3682 } while (cksum >= ZIO_CHECKSUM_LEGACY_FUNCTIONS);
3683 dmu_object_set_checksum(os, bigobj, cksum, tx);
3684
3685 enum zio_compress comp;
3686 do {
3687 comp = (enum zio_compress)
3688 ztest_random_dsl_prop(ZFS_PROP_COMPRESSION);
3689 } while (comp >= ZIO_COMPRESS_LEGACY_FUNCTIONS);
3690 dmu_object_set_compress(os, bigobj, comp, tx);
3691
3692 /*
3693 * For each index from n to n + s, verify that the existing bufwad
3694 * in packobj matches the bufwads at the head and tail of the
3695 * corresponding chunk in bigobj. Then update all three bufwads
3696 * with the new values we want to write out.
3697 */
3698 for (i = 0; i < s; i++) {
3699 /* LINTED */
3700 pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
3701 /* LINTED */
3702 bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
3703 /* LINTED */
3704 bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
3705
3706 ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
3707 ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
3708
3709 if (pack->bw_txg > txg)
3710 fatal(0, "future leak: got %llx, open txg is %llx",
3711 pack->bw_txg, txg);
3712
3713 if (pack->bw_data != 0 && pack->bw_index != n + i)
3714 fatal(0, "wrong index: got %llx, wanted %llx+%llx",
3715 pack->bw_index, n, i);
3716
3717 if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
3718 fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
3719
3720 if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
3721 fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
3722
3723 if (freeit) {
3724 bzero(pack, sizeof (bufwad_t));
3725 } else {
3726 pack->bw_index = n + i;
3727 pack->bw_txg = txg;
3728 pack->bw_data = 1 + ztest_random(-2ULL);
3729 }
3730 *bigH = *pack;
3731 *bigT = *pack;
3732 }
3733
3734 /*
3735 * We've verified all the old bufwads, and made new ones.
3736 * Now write them out.
3737 */
3738 dmu_write(os, packobj, packoff, packsize, packbuf, tx);
3739
3740 if (freeit) {
3741 if (ztest_opts.zo_verbose >= 7) {
3742 (void) printf("freeing offset %llx size %llx"
3743 " txg %llx\n",
3744 (u_longlong_t)bigoff,
3745 (u_longlong_t)bigsize,
3746 (u_longlong_t)txg);
3747 }
3748 VERIFY(0 == dmu_free_range(os, bigobj, bigoff, bigsize, tx));
3749 } else {
3750 if (ztest_opts.zo_verbose >= 7) {
3751 (void) printf("writing offset %llx size %llx"
3752 " txg %llx\n",
3753 (u_longlong_t)bigoff,
3754 (u_longlong_t)bigsize,
3755 (u_longlong_t)txg);
3756 }
3757 dmu_write(os, bigobj, bigoff, bigsize, bigbuf, tx);
3758 }
3759
3760 dmu_tx_commit(tx);
3761
3762 /*
3763 * Sanity check the stuff we just wrote.
3764 */
3765 {
3766 void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
3767 void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
3768
3769 VERIFY(0 == dmu_read(os, packobj, packoff,
3770 packsize, packcheck, DMU_READ_PREFETCH));
3771 VERIFY(0 == dmu_read(os, bigobj, bigoff,
3772 bigsize, bigcheck, DMU_READ_PREFETCH));
3773
3774 ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
3775 ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
3776
3777 umem_free(packcheck, packsize);
3778 umem_free(bigcheck, bigsize);
3779 }
3780
3781 umem_free(packbuf, packsize);
3782 umem_free(bigbuf, bigsize);
3783 }
3784
3785 void
3786 compare_and_update_pbbufs(uint64_t s, bufwad_t *packbuf, bufwad_t *bigbuf,
3787 uint64_t bigsize, uint64_t n, uint64_t chunksize, uint64_t txg)
3788 {
3789 uint64_t i;
3790 bufwad_t *pack;
3791 bufwad_t *bigH;
3792 bufwad_t *bigT;
3793
3794 /*
3795 * For each index from n to n + s, verify that the existing bufwad
3796 * in packobj matches the bufwads at the head and tail of the
3797 * corresponding chunk in bigobj. Then update all three bufwads
3798 * with the new values we want to write out.
3799 */
3800 for (i = 0; i < s; i++) {
3801 /* LINTED */
3802 pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
3803 /* LINTED */
3804 bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
3805 /* LINTED */
3806 bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
3807
3808 ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
3809 ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
3810
3811 if (pack->bw_txg > txg)
3812 fatal(0, "future leak: got %llx, open txg is %llx",
3813 pack->bw_txg, txg);
3814
3815 if (pack->bw_data != 0 && pack->bw_index != n + i)
3816 fatal(0, "wrong index: got %llx, wanted %llx+%llx",
3817 pack->bw_index, n, i);
3818
3819 if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
3820 fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
3821
3822 if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
3823 fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
3824
3825 pack->bw_index = n + i;
3826 pack->bw_txg = txg;
3827 pack->bw_data = 1 + ztest_random(-2ULL);
3828
3829 *bigH = *pack;
3830 *bigT = *pack;
3831 }
3832 }
3833
3834 void
3835 ztest_dmu_read_write_zcopy(ztest_ds_t *zd, uint64_t id)
3836 {
3837 objset_t *os = zd->zd_os;
3838 ztest_od_t od[2];
3839 dmu_tx_t *tx;
3840 uint64_t i;
3841 int error;
3842 uint64_t n, s, txg;
3843 bufwad_t *packbuf, *bigbuf;
3844 uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
3845 uint64_t blocksize = ztest_random_blocksize();
3846 uint64_t chunksize = blocksize;
3847 uint64_t regions = 997;
3848 uint64_t stride = 123456789ULL;
3849 uint64_t width = 9;
3850 dmu_buf_t *bonus_db;
3851 arc_buf_t **bigbuf_arcbufs;
3852 dmu_object_info_t doi;
3853
3854 /*
3855 * This test uses two objects, packobj and bigobj, that are always
3856 * updated together (i.e. in the same tx) so that their contents are
3857 * in sync and can be compared. Their contents relate to each other
3858 * in a simple way: packobj is a dense array of 'bufwad' structures,
3859 * while bigobj is a sparse array of the same bufwads. Specifically,
3860 * for any index n, there are three bufwads that should be identical:
3861 *
3862 * packobj, at offset n * sizeof (bufwad_t)
3863 * bigobj, at the head of the nth chunk
3864 * bigobj, at the tail of the nth chunk
3865 *
3866 * The chunk size is set equal to bigobj block size so that
3867 * dmu_assign_arcbuf() can be tested for object updates.
3868 */
3869
3870 /*
3871 * Read the directory info. If it's the first time, set things up.
3872 */
3873 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
3874 ztest_od_init(&od[1], id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize);
3875
3876 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
3877 return;
3878
3879 bigobj = od[0].od_object;
3880 packobj = od[1].od_object;
3881 blocksize = od[0].od_blocksize;
3882 chunksize = blocksize;
3883 ASSERT(chunksize == od[1].od_gen);
3884
3885 VERIFY(dmu_object_info(os, bigobj, &doi) == 0);
3886 VERIFY(ISP2(doi.doi_data_block_size));
3887 VERIFY(chunksize == doi.doi_data_block_size);
3888 VERIFY(chunksize >= 2 * sizeof (bufwad_t));
3889
3890 /*
3891 * Pick a random index and compute the offsets into packobj and bigobj.
3892 */
3893 n = ztest_random(regions) * stride + ztest_random(width);
3894 s = 1 + ztest_random(width - 1);
3895
3896 packoff = n * sizeof (bufwad_t);
3897 packsize = s * sizeof (bufwad_t);
3898
3899 bigoff = n * chunksize;
3900 bigsize = s * chunksize;
3901
3902 packbuf = umem_zalloc(packsize, UMEM_NOFAIL);
3903 bigbuf = umem_zalloc(bigsize, UMEM_NOFAIL);
3904
3905 VERIFY3U(0, ==, dmu_bonus_hold(os, bigobj, FTAG, &bonus_db));
3906
3907 bigbuf_arcbufs = umem_zalloc(2 * s * sizeof (arc_buf_t *), UMEM_NOFAIL);
3908
3909 /*
3910 * Iteration 0 test zcopy for DB_UNCACHED dbufs.
3911 * Iteration 1 test zcopy to already referenced dbufs.
3912 * Iteration 2 test zcopy to dirty dbuf in the same txg.
3913 * Iteration 3 test zcopy to dbuf dirty in previous txg.
3914 * Iteration 4 test zcopy when dbuf is no longer dirty.
3915 * Iteration 5 test zcopy when it can't be done.
3916 * Iteration 6 one more zcopy write.
3917 */
3918 for (i = 0; i < 7; i++) {
3919 uint64_t j;
3920 uint64_t off;
3921
3922 /*
3923 * In iteration 5 (i == 5) use arcbufs
3924 * that don't match bigobj blksz to test
3925 * dmu_assign_arcbuf() when it can't directly
3926 * assign an arcbuf to a dbuf.
3927 */
3928 for (j = 0; j < s; j++) {
3929 if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) {
3930 bigbuf_arcbufs[j] =
3931 dmu_request_arcbuf(bonus_db, chunksize);
3932 } else {
3933 bigbuf_arcbufs[2 * j] =
3934 dmu_request_arcbuf(bonus_db, chunksize / 2);
3935 bigbuf_arcbufs[2 * j + 1] =
3936 dmu_request_arcbuf(bonus_db, chunksize / 2);
3937 }
3938 }
3939
3940 /*
3941 * Get a tx for the mods to both packobj and bigobj.
3942 */
3943 tx = dmu_tx_create(os);
3944
3945 dmu_tx_hold_write(tx, packobj, packoff, packsize);
3946 dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
3947
3948 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3949 if (txg == 0) {
3950 umem_free(packbuf, packsize);
3951 umem_free(bigbuf, bigsize);
3952 for (j = 0; j < s; j++) {
3953 if (i != 5 ||
3954 chunksize < (SPA_MINBLOCKSIZE * 2)) {
3955 dmu_return_arcbuf(bigbuf_arcbufs[j]);
3956 } else {
3957 dmu_return_arcbuf(
3958 bigbuf_arcbufs[2 * j]);
3959 dmu_return_arcbuf(
3960 bigbuf_arcbufs[2 * j + 1]);
3961 }
3962 }
3963 umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
3964 dmu_buf_rele(bonus_db, FTAG);
3965 return;
3966 }
3967
3968 /*
3969 * 50% of the time don't read objects in the 1st iteration to
3970 * test dmu_assign_arcbuf() for the case when there're no
3971 * existing dbufs for the specified offsets.
3972 */
3973 if (i != 0 || ztest_random(2) != 0) {
3974 error = dmu_read(os, packobj, packoff,
3975 packsize, packbuf, DMU_READ_PREFETCH);
3976 ASSERT0(error);
3977 error = dmu_read(os, bigobj, bigoff, bigsize,
3978 bigbuf, DMU_READ_PREFETCH);
3979 ASSERT0(error);
3980 }
3981 compare_and_update_pbbufs(s, packbuf, bigbuf, bigsize,
3982 n, chunksize, txg);
3983
3984 /*
3985 * We've verified all the old bufwads, and made new ones.
3986 * Now write them out.
3987 */
3988 dmu_write(os, packobj, packoff, packsize, packbuf, tx);
3989 if (ztest_opts.zo_verbose >= 7) {
3990 (void) printf("writing offset %llx size %llx"
3991 " txg %llx\n",
3992 (u_longlong_t)bigoff,
3993 (u_longlong_t)bigsize,
3994 (u_longlong_t)txg);
3995 }
3996 for (off = bigoff, j = 0; j < s; j++, off += chunksize) {
3997 dmu_buf_t *dbt;
3998 if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) {
3999 bcopy((caddr_t)bigbuf + (off - bigoff),
4000 bigbuf_arcbufs[j]->b_data, chunksize);
4001 } else {
4002 bcopy((caddr_t)bigbuf + (off - bigoff),
4003 bigbuf_arcbufs[2 * j]->b_data,
4004 chunksize / 2);
4005 bcopy((caddr_t)bigbuf + (off - bigoff) +
4006 chunksize / 2,
4007 bigbuf_arcbufs[2 * j + 1]->b_data,
4008 chunksize / 2);
4009 }
4010
4011 if (i == 1) {
4012 VERIFY(dmu_buf_hold(os, bigobj, off,
4013 FTAG, &dbt, DMU_READ_NO_PREFETCH) == 0);
4014 }
4015 if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) {
4016 dmu_assign_arcbuf(bonus_db, off,
4017 bigbuf_arcbufs[j], tx);
4018 } else {
4019 dmu_assign_arcbuf(bonus_db, off,
4020 bigbuf_arcbufs[2 * j], tx);
4021 dmu_assign_arcbuf(bonus_db,
4022 off + chunksize / 2,
4023 bigbuf_arcbufs[2 * j + 1], tx);
4024 }
4025 if (i == 1) {
4026 dmu_buf_rele(dbt, FTAG);
4027 }
4028 }
4029 dmu_tx_commit(tx);
4030
4031 /*
4032 * Sanity check the stuff we just wrote.
4033 */
4034 {
4035 void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
4036 void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
4037
4038 VERIFY(0 == dmu_read(os, packobj, packoff,
4039 packsize, packcheck, DMU_READ_PREFETCH));
4040 VERIFY(0 == dmu_read(os, bigobj, bigoff,
4041 bigsize, bigcheck, DMU_READ_PREFETCH));
4042
4043 ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
4044 ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
4045
4046 umem_free(packcheck, packsize);
4047 umem_free(bigcheck, bigsize);
4048 }
4049 if (i == 2) {
4050 txg_wait_open(dmu_objset_pool(os), 0);
4051 } else if (i == 3) {
4052 txg_wait_synced(dmu_objset_pool(os), 0);
4053 }
4054 }
4055
4056 dmu_buf_rele(bonus_db, FTAG);
4057 umem_free(packbuf, packsize);
4058 umem_free(bigbuf, bigsize);
4059 umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
4060 }
4061
4062 /* ARGSUSED */
4063 void
4064 ztest_dmu_write_parallel(ztest_ds_t *zd, uint64_t id)
4065 {
4066 ztest_od_t od[1];
4067 uint64_t offset = (1ULL << (ztest_random(20) + 43)) +
4068 (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
4069
4070 /*
4071 * Have multiple threads write to large offsets in an object
4072 * to verify that parallel writes to an object -- even to the
4073 * same blocks within the object -- doesn't cause any trouble.
4074 */
4075 ztest_od_init(&od[0], ID_PARALLEL, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
4076
4077 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4078 return;
4079
4080 while (ztest_random(10) != 0)
4081 ztest_io(zd, od[0].od_object, offset);
4082 }
4083
4084 void
4085 ztest_dmu_prealloc(ztest_ds_t *zd, uint64_t id)
4086 {
4087 ztest_od_t od[1];
4088 uint64_t offset = (1ULL << (ztest_random(4) + SPA_MAXBLOCKSHIFT)) +
4089 (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
4090 uint64_t count = ztest_random(20) + 1;
4091 uint64_t blocksize = ztest_random_blocksize();
4092 void *data;
4093
4094 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
4095
4096 if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4097 return;
4098
4099 if (ztest_truncate(zd, od[0].od_object, offset, count * blocksize) != 0)
4100 return;
4101
4102 ztest_prealloc(zd, od[0].od_object, offset, count * blocksize);
4103
4104 data = umem_zalloc(blocksize, UMEM_NOFAIL);
4105
4106 while (ztest_random(count) != 0) {
4107 uint64_t randoff = offset + (ztest_random(count) * blocksize);
4108 if (ztest_write(zd, od[0].od_object, randoff, blocksize,
4109 data) != 0)
4110 break;
4111 while (ztest_random(4) != 0)
4112 ztest_io(zd, od[0].od_object, randoff);
4113 }
4114
4115 umem_free(data, blocksize);
4116 }
4117
4118 /*
4119 * Verify that zap_{create,destroy,add,remove,update} work as expected.
4120 */
4121 #define ZTEST_ZAP_MIN_INTS 1
4122 #define ZTEST_ZAP_MAX_INTS 4
4123 #define ZTEST_ZAP_MAX_PROPS 1000
4124
4125 void
4126 ztest_zap(ztest_ds_t *zd, uint64_t id)
4127 {
4128 objset_t *os = zd->zd_os;
4129 ztest_od_t od[1];
4130 uint64_t object;
4131 uint64_t txg, last_txg;
4132 uint64_t value[ZTEST_ZAP_MAX_INTS];
4133 uint64_t zl_ints, zl_intsize, prop;
4134 int i, ints;
4135 dmu_tx_t *tx;
4136 char propname[100], txgname[100];
4137 int error;
4138 char *hc[2] = { "s.acl.h", ".s.open.h.hyLZlg" };
4139
4140 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0);
4141
4142 if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4143 return;
4144
4145 object = od[0].od_object;
4146
4147 /*
4148 * Generate a known hash collision, and verify that
4149 * we can lookup and remove both entries.
4150 */
4151 tx = dmu_tx_create(os);
4152 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4153 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4154 if (txg == 0)
4155 return;
4156 for (i = 0; i < 2; i++) {
4157 value[i] = i;
4158 VERIFY3U(0, ==, zap_add(os, object, hc[i], sizeof (uint64_t),
4159 1, &value[i], tx));
4160 }
4161 for (i = 0; i < 2; i++) {
4162 VERIFY3U(EEXIST, ==, zap_add(os, object, hc[i],
4163 sizeof (uint64_t), 1, &value[i], tx));
4164 VERIFY3U(0, ==,
4165 zap_length(os, object, hc[i], &zl_intsize, &zl_ints));
4166 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4167 ASSERT3U(zl_ints, ==, 1);
4168 }
4169 for (i = 0; i < 2; i++) {
4170 VERIFY3U(0, ==, zap_remove(os, object, hc[i], tx));
4171 }
4172 dmu_tx_commit(tx);
4173
4174 /*
4175 * Generate a buch of random entries.
4176 */
4177 ints = MAX(ZTEST_ZAP_MIN_INTS, object % ZTEST_ZAP_MAX_INTS);
4178
4179 prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
4180 (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
4181 (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
4182 bzero(value, sizeof (value));
4183 last_txg = 0;
4184
4185 /*
4186 * If these zap entries already exist, validate their contents.
4187 */
4188 error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
4189 if (error == 0) {
4190 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4191 ASSERT3U(zl_ints, ==, 1);
4192
4193 VERIFY(zap_lookup(os, object, txgname, zl_intsize,
4194 zl_ints, &last_txg) == 0);
4195
4196 VERIFY(zap_length(os, object, propname, &zl_intsize,
4197 &zl_ints) == 0);
4198
4199 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4200 ASSERT3U(zl_ints, ==, ints);
4201
4202 VERIFY(zap_lookup(os, object, propname, zl_intsize,
4203 zl_ints, value) == 0);
4204
4205 for (i = 0; i < ints; i++) {
4206 ASSERT3U(value[i], ==, last_txg + object + i);
4207 }
4208 } else {
4209 ASSERT3U(error, ==, ENOENT);
4210 }
4211
4212 /*
4213 * Atomically update two entries in our zap object.
4214 * The first is named txg_%llu, and contains the txg
4215 * in which the property was last updated. The second
4216 * is named prop_%llu, and the nth element of its value
4217 * should be txg + object + n.
4218 */
4219 tx = dmu_tx_create(os);
4220 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4221 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4222 if (txg == 0)
4223 return;
4224
4225 if (last_txg > txg)
4226 fatal(0, "zap future leak: old %llu new %llu", last_txg, txg);
4227
4228 for (i = 0; i < ints; i++)
4229 value[i] = txg + object + i;
4230
4231 VERIFY3U(0, ==, zap_update(os, object, txgname, sizeof (uint64_t),
4232 1, &txg, tx));
4233 VERIFY3U(0, ==, zap_update(os, object, propname, sizeof (uint64_t),
4234 ints, value, tx));
4235
4236 dmu_tx_commit(tx);
4237
4238 /*
4239 * Remove a random pair of entries.
4240 */
4241 prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
4242 (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
4243 (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
4244
4245 error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
4246
4247 if (error == ENOENT)
4248 return;
4249
4250 ASSERT0(error);
4251
4252 tx = dmu_tx_create(os);
4253 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4254 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4255 if (txg == 0)
4256 return;
4257 VERIFY3U(0, ==, zap_remove(os, object, txgname, tx));
4258 VERIFY3U(0, ==, zap_remove(os, object, propname, tx));
4259 dmu_tx_commit(tx);
4260 }
4261
4262 /*
4263 * Testcase to test the upgrading of a microzap to fatzap.
4264 */
4265 void
4266 ztest_fzap(ztest_ds_t *zd, uint64_t id)
4267 {
4268 objset_t *os = zd->zd_os;
4269 ztest_od_t od[1];
4270 uint64_t object, txg;
4271
4272 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0);
4273
4274 if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4275 return;
4276
4277 object = od[0].od_object;
4278
4279 /*
4280 * Add entries to this ZAP and make sure it spills over
4281 * and gets upgraded to a fatzap. Also, since we are adding
4282 * 2050 entries we should see ptrtbl growth and leaf-block split.
4283 */
4284 for (int i = 0; i < 2050; i++) {
4285 char name[ZFS_MAX_DATASET_NAME_LEN];
4286 uint64_t value = i;
4287 dmu_tx_t *tx;
4288 int error;
4289
4290 (void) snprintf(name, sizeof (name), "fzap-%llu-%llu",
4291 id, value);
4292
4293 tx = dmu_tx_create(os);
4294 dmu_tx_hold_zap(tx, object, B_TRUE, name);
4295 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4296 if (txg == 0)
4297 return;
4298 error = zap_add(os, object, name, sizeof (uint64_t), 1,
4299 &value, tx);
4300 ASSERT(error == 0 || error == EEXIST);
4301 dmu_tx_commit(tx);
4302 }
4303 }
4304
4305 /* ARGSUSED */
4306 void
4307 ztest_zap_parallel(ztest_ds_t *zd, uint64_t id)
4308 {
4309 objset_t *os = zd->zd_os;
4310 ztest_od_t od[1];
4311 uint64_t txg, object, count, wsize, wc, zl_wsize, zl_wc;
4312 dmu_tx_t *tx;
4313 int i, namelen, error;
4314 int micro = ztest_random(2);
4315 char name[20], string_value[20];
4316 void *data;
4317
4318 ztest_od_init(&od[0], ID_PARALLEL, FTAG, micro, DMU_OT_ZAP_OTHER, 0, 0);
4319
4320 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4321 return;
4322
4323 object = od[0].od_object;
4324
4325 /*
4326 * Generate a random name of the form 'xxx.....' where each
4327 * x is a random printable character and the dots are dots.
4328 * There are 94 such characters, and the name length goes from
4329 * 6 to 20, so there are 94^3 * 15 = 12,458,760 possible names.
4330 */
4331 namelen = ztest_random(sizeof (name) - 5) + 5 + 1;
4332
4333 for (i = 0; i < 3; i++)
4334 name[i] = '!' + ztest_random('~' - '!' + 1);
4335 for (; i < namelen - 1; i++)
4336 name[i] = '.';
4337 name[i] = '\0';
4338
4339 if ((namelen & 1) || micro) {
4340 wsize = sizeof (txg);
4341 wc = 1;
4342 data = &txg;
4343 } else {
4344 wsize = 1;
4345 wc = namelen;
4346 data = string_value;
4347 }
4348
4349 count = -1ULL;
4350 VERIFY0(zap_count(os, object, &count));
4351 ASSERT(count != -1ULL);
4352
4353 /*
4354 * Select an operation: length, lookup, add, update, remove.
4355 */
4356 i = ztest_random(5);
4357
4358 if (i >= 2) {
4359 tx = dmu_tx_create(os);
4360 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4361 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4362 if (txg == 0)
4363 return;
4364 bcopy(name, string_value, namelen);
4365 } else {
4366 tx = NULL;
4367 txg = 0;
4368 bzero(string_value, namelen);
4369 }
4370
4371 switch (i) {
4372
4373 case 0:
4374 error = zap_length(os, object, name, &zl_wsize, &zl_wc);
4375 if (error == 0) {
4376 ASSERT3U(wsize, ==, zl_wsize);
4377 ASSERT3U(wc, ==, zl_wc);
4378 } else {
4379 ASSERT3U(error, ==, ENOENT);
4380 }
4381 break;
4382
4383 case 1:
4384 error = zap_lookup(os, object, name, wsize, wc, data);
4385 if (error == 0) {
4386 if (data == string_value &&
4387 bcmp(name, data, namelen) != 0)
4388 fatal(0, "name '%s' != val '%s' len %d",
4389 name, data, namelen);
4390 } else {
4391 ASSERT3U(error, ==, ENOENT);
4392 }
4393 break;
4394
4395 case 2:
4396 error = zap_add(os, object, name, wsize, wc, data, tx);
4397 ASSERT(error == 0 || error == EEXIST);
4398 break;
4399
4400 case 3:
4401 VERIFY(zap_update(os, object, name, wsize, wc, data, tx) == 0);
4402 break;
4403
4404 case 4:
4405 error = zap_remove(os, object, name, tx);
4406 ASSERT(error == 0 || error == ENOENT);
4407 break;
4408 }
4409
4410 if (tx != NULL)
4411 dmu_tx_commit(tx);
4412 }
4413
4414 /*
4415 * Commit callback data.
4416 */
4417 typedef struct ztest_cb_data {
4418 list_node_t zcd_node;
4419 uint64_t zcd_txg;
4420 int zcd_expected_err;
4421 boolean_t zcd_added;
4422 boolean_t zcd_called;
4423 spa_t *zcd_spa;
4424 } ztest_cb_data_t;
4425
4426 /* This is the actual commit callback function */
4427 static void
4428 ztest_commit_callback(void *arg, int error)
4429 {
4430 ztest_cb_data_t *data = arg;
4431 uint64_t synced_txg;
4432
4433 VERIFY(data != NULL);
4434 VERIFY3S(data->zcd_expected_err, ==, error);
4435 VERIFY(!data->zcd_called);
4436
4437 synced_txg = spa_last_synced_txg(data->zcd_spa);
4438 if (data->zcd_txg > synced_txg)
4439 fatal(0, "commit callback of txg %" PRIu64 " called prematurely"
4440 ", last synced txg = %" PRIu64 "\n", data->zcd_txg,
4441 synced_txg);
4442
4443 data->zcd_called = B_TRUE;
4444
4445 if (error == ECANCELED) {
4446 ASSERT0(data->zcd_txg);
4447 ASSERT(!data->zcd_added);
4448
4449 /*
4450 * The private callback data should be destroyed here, but
4451 * since we are going to check the zcd_called field after
4452 * dmu_tx_abort(), we will destroy it there.
4453 */
4454 return;
4455 }
4456
4457 /* Was this callback added to the global callback list? */
4458 if (!data->zcd_added)
4459 goto out;
4460
4461 ASSERT3U(data->zcd_txg, !=, 0);
4462
4463 /* Remove our callback from the list */
4464 (void) mutex_lock(&zcl.zcl_callbacks_lock);
4465 list_remove(&zcl.zcl_callbacks, data);
4466 (void) mutex_unlock(&zcl.zcl_callbacks_lock);
4467
4468 out:
4469 umem_free(data, sizeof (ztest_cb_data_t));
4470 }
4471
4472 /* Allocate and initialize callback data structure */
4473 static ztest_cb_data_t *
4474 ztest_create_cb_data(objset_t *os, uint64_t txg)
4475 {
4476 ztest_cb_data_t *cb_data;
4477
4478 cb_data = umem_zalloc(sizeof (ztest_cb_data_t), UMEM_NOFAIL);
4479
4480 cb_data->zcd_txg = txg;
4481 cb_data->zcd_spa = dmu_objset_spa(os);
4482
4483 return (cb_data);
4484 }
4485
4486 /*
4487 * If a number of txgs equal to this threshold have been created after a commit
4488 * callback has been registered but not called, then we assume there is an
4489 * implementation bug.
4490 */
4491 #define ZTEST_COMMIT_CALLBACK_THRESH (TXG_CONCURRENT_STATES + 2)
4492
4493 /*
4494 * Commit callback test.
4495 */
4496 void
4497 ztest_dmu_commit_callbacks(ztest_ds_t *zd, uint64_t id)
4498 {
4499 objset_t *os = zd->zd_os;
4500 ztest_od_t od[1];
4501 dmu_tx_t *tx;
4502 ztest_cb_data_t *cb_data[3], *tmp_cb;
4503 uint64_t old_txg, txg;
4504 int i, error;
4505
4506 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
4507
4508 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4509 return;
4510
4511 tx = dmu_tx_create(os);
4512
4513 cb_data[0] = ztest_create_cb_data(os, 0);
4514 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[0]);
4515
4516 dmu_tx_hold_write(tx, od[0].od_object, 0, sizeof (uint64_t));
4517
4518 /* Every once in a while, abort the transaction on purpose */
4519 if (ztest_random(100) == 0)
4520 error = -1;
4521
4522 if (!error)
4523 error = dmu_tx_assign(tx, TXG_NOWAIT);
4524
4525 txg = error ? 0 : dmu_tx_get_txg(tx);
4526
4527 cb_data[0]->zcd_txg = txg;
4528 cb_data[1] = ztest_create_cb_data(os, txg);
4529 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[1]);
4530
4531 if (error) {
4532 /*
4533 * It's not a strict requirement to call the registered
4534 * callbacks from inside dmu_tx_abort(), but that's what
4535 * it's supposed to happen in the current implementation
4536 * so we will check for that.
4537 */
4538 for (i = 0; i < 2; i++) {
4539 cb_data[i]->zcd_expected_err = ECANCELED;
4540 VERIFY(!cb_data[i]->zcd_called);
4541 }
4542
4543 dmu_tx_abort(tx);
4544
4545 for (i = 0; i < 2; i++) {
4546 VERIFY(cb_data[i]->zcd_called);
4547 umem_free(cb_data[i], sizeof (ztest_cb_data_t));
4548 }
4549
4550 return;
4551 }
4552
4553 cb_data[2] = ztest_create_cb_data(os, txg);
4554 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[2]);
4555
4556 /*
4557 * Read existing data to make sure there isn't a future leak.
4558 */
4559 VERIFY(0 == dmu_read(os, od[0].od_object, 0, sizeof (uint64_t),
4560 &old_txg, DMU_READ_PREFETCH));
4561
4562 if (old_txg > txg)
4563 fatal(0, "future leak: got %" PRIu64 ", open txg is %" PRIu64,
4564 old_txg, txg);
4565
4566 dmu_write(os, od[0].od_object, 0, sizeof (uint64_t), &txg, tx);
4567
4568 (void) mutex_lock(&zcl.zcl_callbacks_lock);
4569
4570 /*
4571 * Since commit callbacks don't have any ordering requirement and since
4572 * it is theoretically possible for a commit callback to be called
4573 * after an arbitrary amount of time has elapsed since its txg has been
4574 * synced, it is difficult to reliably determine whether a commit
4575 * callback hasn't been called due to high load or due to a flawed
4576 * implementation.
4577 *
4578 * In practice, we will assume that if after a certain number of txgs a
4579 * commit callback hasn't been called, then most likely there's an
4580 * implementation bug..
4581 */
4582 tmp_cb = list_head(&zcl.zcl_callbacks);
4583 if (tmp_cb != NULL &&
4584 (txg - ZTEST_COMMIT_CALLBACK_THRESH) > tmp_cb->zcd_txg) {
4585 fatal(0, "Commit callback threshold exceeded, oldest txg: %"
4586 PRIu64 ", open txg: %" PRIu64 "\n", tmp_cb->zcd_txg, txg);
4587 }
4588
4589 /*
4590 * Let's find the place to insert our callbacks.
4591 *
4592 * Even though the list is ordered by txg, it is possible for the
4593 * insertion point to not be the end because our txg may already be
4594 * quiescing at this point and other callbacks in the open txg
4595 * (from other objsets) may have sneaked in.
4596 */
4597 tmp_cb = list_tail(&zcl.zcl_callbacks);
4598 while (tmp_cb != NULL && tmp_cb->zcd_txg > txg)
4599 tmp_cb = list_prev(&zcl.zcl_callbacks, tmp_cb);
4600
4601 /* Add the 3 callbacks to the list */
4602 for (i = 0; i < 3; i++) {
4603 if (tmp_cb == NULL)
4604 list_insert_head(&zcl.zcl_callbacks, cb_data[i]);
4605 else
4606 list_insert_after(&zcl.zcl_callbacks, tmp_cb,
4607 cb_data[i]);
4608
4609 cb_data[i]->zcd_added = B_TRUE;
4610 VERIFY(!cb_data[i]->zcd_called);
4611
4612 tmp_cb = cb_data[i];
4613 }
4614
4615 (void) mutex_unlock(&zcl.zcl_callbacks_lock);
4616
4617 dmu_tx_commit(tx);
4618 }
4619
4620 /* ARGSUSED */
4621 void
4622 ztest_dsl_prop_get_set(ztest_ds_t *zd, uint64_t id)
4623 {
4624 zfs_prop_t proplist[] = {
4625 ZFS_PROP_CHECKSUM,
4626 ZFS_PROP_COMPRESSION,
4627 ZFS_PROP_COPIES,
4628 ZFS_PROP_DEDUP
4629 };
4630
4631 (void) rw_rdlock(&ztest_name_lock);
4632
4633 for (int p = 0; p < sizeof (proplist) / sizeof (proplist[0]); p++)
4634 (void) ztest_dsl_prop_set_uint64(zd->zd_name, proplist[p],
4635 ztest_random_dsl_prop(proplist[p]), (int)ztest_random(2));
4636
4637 (void) rw_unlock(&ztest_name_lock);
4638 }
4639
4640 /* ARGSUSED */
4641 void
4642 ztest_spa_prop_get_set(ztest_ds_t *zd, uint64_t id)
4643 {
4644 nvlist_t *props = NULL;
4645
4646 (void) rw_rdlock(&ztest_name_lock);
4647
4648 (void) ztest_spa_prop_set_uint64(ZPOOL_PROP_DEDUPDITTO,
4649 ZIO_DEDUPDITTO_MIN + ztest_random(ZIO_DEDUPDITTO_MIN));
4650
4651 VERIFY0(spa_prop_get(ztest_spa, &props));
4652
4653 if (ztest_opts.zo_verbose >= 6)
4654 dump_nvlist(props, 4);
4655
4656 nvlist_free(props);
4657
4658 (void) rw_unlock(&ztest_name_lock);
4659 }
4660
4661 static int
4662 user_release_one(const char *snapname, const char *holdname)
4663 {
4664 nvlist_t *snaps, *holds;
4665 int error;
4666
4667 snaps = fnvlist_alloc();
4668 holds = fnvlist_alloc();
4669 fnvlist_add_boolean(holds, holdname);
4670 fnvlist_add_nvlist(snaps, snapname, holds);
4671 fnvlist_free(holds);
4672 error = dsl_dataset_user_release(snaps, NULL);
4673 fnvlist_free(snaps);
4674 return (error);
4675 }
4676
4677 /*
4678 * Test snapshot hold/release and deferred destroy.
4679 */
4680 void
4681 ztest_dmu_snapshot_hold(ztest_ds_t *zd, uint64_t id)
4682 {
4683 int error;
4684 objset_t *os = zd->zd_os;
4685 objset_t *origin;
4686 char snapname[100];
4687 char fullname[100];
4688 char clonename[100];
4689 char tag[100];
4690 char osname[ZFS_MAX_DATASET_NAME_LEN];
4691 nvlist_t *holds;
4692
4693 (void) rw_rdlock(&ztest_name_lock);
4694
4695 dmu_objset_name(os, osname);
4696
4697 (void) snprintf(snapname, sizeof (snapname), "sh1_%llu", id);
4698 (void) snprintf(fullname, sizeof (fullname), "%s@%s", osname, snapname);
4699 (void) snprintf(clonename, sizeof (clonename),
4700 "%s/ch1_%llu", osname, id);
4701 (void) snprintf(tag, sizeof (tag), "tag_%llu", id);
4702
4703 /*
4704 * Clean up from any previous run.
4705 */
4706 error = dsl_destroy_head(clonename);
4707 if (error != ENOENT)
4708 ASSERT0(error);
4709 error = user_release_one(fullname, tag);
4710 if (error != ESRCH && error != ENOENT)
4711 ASSERT0(error);
4712 error = dsl_destroy_snapshot(fullname, B_FALSE);
4713 if (error != ENOENT)
4714 ASSERT0(error);
4715
4716 /*
4717 * Create snapshot, clone it, mark snap for deferred destroy,
4718 * destroy clone, verify snap was also destroyed.
4719 */
4720 error = dmu_objset_snapshot_one(osname, snapname);
4721 if (error) {
4722 if (error == ENOSPC) {
4723 ztest_record_enospc("dmu_objset_snapshot");
4724 goto out;
4725 }
4726 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
4727 }
4728
4729 error = dmu_objset_clone(clonename, fullname);
4730 if (error) {
4731 if (error == ENOSPC) {
4732 ztest_record_enospc("dmu_objset_clone");
4733 goto out;
4734 }
4735 fatal(0, "dmu_objset_clone(%s) = %d", clonename, error);
4736 }
4737
4738 error = dsl_destroy_snapshot(fullname, B_TRUE);
4739 if (error) {
4740 fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
4741 fullname, error);
4742 }
4743
4744 error = dsl_destroy_head(clonename);
4745 if (error)
4746 fatal(0, "dsl_destroy_head(%s) = %d", clonename, error);
4747
4748 error = dmu_objset_hold(fullname, FTAG, &origin);
4749 if (error != ENOENT)
4750 fatal(0, "dmu_objset_hold(%s) = %d", fullname, error);
4751
4752 /*
4753 * Create snapshot, add temporary hold, verify that we can't
4754 * destroy a held snapshot, mark for deferred destroy,
4755 * release hold, verify snapshot was destroyed.
4756 */
4757 error = dmu_objset_snapshot_one(osname, snapname);
4758 if (error) {
4759 if (error == ENOSPC) {
4760 ztest_record_enospc("dmu_objset_snapshot");
4761 goto out;
4762 }
4763 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
4764 }
4765
4766 holds = fnvlist_alloc();
4767 fnvlist_add_string(holds, fullname, tag);
4768 error = dsl_dataset_user_hold(holds, 0, NULL);
4769 fnvlist_free(holds);
4770
4771 if (error == ENOSPC) {
4772 ztest_record_enospc("dsl_dataset_user_hold");
4773 goto out;
4774 } else if (error) {
4775 fatal(0, "dsl_dataset_user_hold(%s, %s) = %u",
4776 fullname, tag, error);
4777 }
4778
4779 error = dsl_destroy_snapshot(fullname, B_FALSE);
4780 if (error != EBUSY) {
4781 fatal(0, "dsl_destroy_snapshot(%s, B_FALSE) = %d",
4782 fullname, error);
4783 }
4784
4785 error = dsl_destroy_snapshot(fullname, B_TRUE);
4786 if (error) {
4787 fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
4788 fullname, error);
4789 }
4790
4791 error = user_release_one(fullname, tag);
4792 if (error)
4793 fatal(0, "user_release_one(%s, %s) = %d", fullname, tag, error);
4794
4795 VERIFY3U(dmu_objset_hold(fullname, FTAG, &origin), ==, ENOENT);
4796
4797 out:
4798 (void) rw_unlock(&ztest_name_lock);
4799 }
4800
4801 /*
4802 * Inject random faults into the on-disk data.
4803 */
4804 /* ARGSUSED */
4805 void
4806 ztest_fault_inject(ztest_ds_t *zd, uint64_t id)
4807 {
4808 ztest_shared_t *zs = ztest_shared;
4809 spa_t *spa = ztest_spa;
4810 int fd;
4811 uint64_t offset;
4812 uint64_t leaves;
4813 uint64_t bad = 0x1990c0ffeedecade;
4814 uint64_t top, leaf;
4815 char path0[MAXPATHLEN];
4816 char pathrand[MAXPATHLEN];
4817 size_t fsize;
4818 int bshift = SPA_MAXBLOCKSHIFT + 2;
4819 int iters = 1000;
4820 int maxfaults;
4821 int mirror_save;
4822 vdev_t *vd0 = NULL;
4823 uint64_t guid0 = 0;
4824 boolean_t islog = B_FALSE;
4825
4826 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
4827 maxfaults = MAXFAULTS();
4828 leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz;
4829 mirror_save = zs->zs_mirrors;
4830 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4831
4832 ASSERT(leaves >= 1);
4833
4834 /*
4835 * Grab the name lock as reader. There are some operations
4836 * which don't like to have their vdevs changed while
4837 * they are in progress (i.e. spa_change_guid). Those
4838 * operations will have grabbed the name lock as writer.
4839 */
4840 (void) rw_rdlock(&ztest_name_lock);
4841
4842 /*
4843 * We need SCL_STATE here because we're going to look at vd0->vdev_tsd.
4844 */
4845 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
4846
4847 if (ztest_random(2) == 0) {
4848 /*
4849 * Inject errors on a normal data device or slog device.
4850 */
4851 top = ztest_random_vdev_top(spa, B_TRUE);
4852 leaf = ztest_random(leaves) + zs->zs_splits;
4853
4854 /*
4855 * Generate paths to the first leaf in this top-level vdev,
4856 * and to the random leaf we selected. We'll induce transient
4857 * write failures and random online/offline activity on leaf 0,
4858 * and we'll write random garbage to the randomly chosen leaf.
4859 */
4860 (void) snprintf(path0, sizeof (path0), ztest_dev_template,
4861 ztest_opts.zo_dir, ztest_opts.zo_pool,
4862 top * leaves + zs->zs_splits);
4863 (void) snprintf(pathrand, sizeof (pathrand), ztest_dev_template,
4864 ztest_opts.zo_dir, ztest_opts.zo_pool,
4865 top * leaves + leaf);
4866
4867 vd0 = vdev_lookup_by_path(spa->spa_root_vdev, path0);
4868 if (vd0 != NULL && vd0->vdev_top->vdev_islog)
4869 islog = B_TRUE;
4870
4871 /*
4872 * If the top-level vdev needs to be resilvered
4873 * then we only allow faults on the device that is
4874 * resilvering.
4875 */
4876 if (vd0 != NULL && maxfaults != 1 &&
4877 (!vdev_resilver_needed(vd0->vdev_top, NULL, NULL) ||
4878 vd0->vdev_resilver_txg != 0)) {
4879 /*
4880 * Make vd0 explicitly claim to be unreadable,
4881 * or unwriteable, or reach behind its back
4882 * and close the underlying fd. We can do this if
4883 * maxfaults == 0 because we'll fail and reexecute,
4884 * and we can do it if maxfaults >= 2 because we'll
4885 * have enough redundancy. If maxfaults == 1, the
4886 * combination of this with injection of random data
4887 * corruption below exceeds the pool's fault tolerance.
4888 */
4889 vdev_file_t *vf = vd0->vdev_tsd;
4890
4891 if (vf != NULL && ztest_random(3) == 0) {
4892 (void) close(vf->vf_vnode->v_fd);
4893 vf->vf_vnode->v_fd = -1;
4894 } else if (ztest_random(2) == 0) {
4895 vd0->vdev_cant_read = B_TRUE;
4896 } else {
4897 vd0->vdev_cant_write = B_TRUE;
4898 }
4899 guid0 = vd0->vdev_guid;
4900 }
4901 } else {
4902 /*
4903 * Inject errors on an l2cache device.
4904 */
4905 spa_aux_vdev_t *sav = &spa->spa_l2cache;
4906
4907 if (sav->sav_count == 0) {
4908 spa_config_exit(spa, SCL_STATE, FTAG);
4909 (void) rw_unlock(&ztest_name_lock);
4910 return;
4911 }
4912 vd0 = sav->sav_vdevs[ztest_random(sav->sav_count)];
4913 guid0 = vd0->vdev_guid;
4914 (void) strcpy(path0, vd0->vdev_path);
4915 (void) strcpy(pathrand, vd0->vdev_path);
4916
4917 leaf = 0;
4918 leaves = 1;
4919 maxfaults = INT_MAX; /* no limit on cache devices */
4920 }
4921
4922 spa_config_exit(spa, SCL_STATE, FTAG);
4923 (void) rw_unlock(&ztest_name_lock);
4924
4925 /*
4926 * If we can tolerate two or more faults, or we're dealing
4927 * with a slog, randomly online/offline vd0.
4928 */
4929 if ((maxfaults >= 2 || islog) && guid0 != 0) {
4930 if (ztest_random(10) < 6) {
4931 int flags = (ztest_random(2) == 0 ?
4932 ZFS_OFFLINE_TEMPORARY : 0);
4933
4934 /*
4935 * We have to grab the zs_name_lock as writer to
4936 * prevent a race between offlining a slog and
4937 * destroying a dataset. Offlining the slog will
4938 * grab a reference on the dataset which may cause
4939 * dmu_objset_destroy() to fail with EBUSY thus
4940 * leaving the dataset in an inconsistent state.
4941 */
4942 if (islog)
4943 (void) rw_wrlock(&ztest_name_lock);
4944
4945 VERIFY(vdev_offline(spa, guid0, flags) != EBUSY);
4946
4947 if (islog)
4948 (void) rw_unlock(&ztest_name_lock);
4949 } else {
4950 /*
4951 * Ideally we would like to be able to randomly
4952 * call vdev_[on|off]line without holding locks
4953 * to force unpredictable failures but the side
4954 * effects of vdev_[on|off]line prevent us from
4955 * doing so. We grab the ztest_vdev_lock here to
4956 * prevent a race between injection testing and
4957 * aux_vdev removal.
4958 */
4959 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
4960 (void) vdev_online(spa, guid0, 0, NULL);
4961 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4962 }
4963 }
4964
4965 if (maxfaults == 0)
4966 return;
4967
4968 /*
4969 * We have at least single-fault tolerance, so inject data corruption.
4970 */
4971 fd = open(pathrand, O_RDWR);
4972
4973 if (fd == -1) /* we hit a gap in the device namespace */
4974 return;
4975
4976 fsize = lseek(fd, 0, SEEK_END);
4977
4978 while (--iters != 0) {
4979 /*
4980 * The offset must be chosen carefully to ensure that
4981 * we do not inject a given logical block with errors
4982 * on two different leaf devices, because ZFS can not
4983 * tolerate that (if maxfaults==1).
4984 *
4985 * We divide each leaf into chunks of size
4986 * (# leaves * SPA_MAXBLOCKSIZE * 4). Within each chunk
4987 * there is a series of ranges to which we can inject errors.
4988 * Each range can accept errors on only a single leaf vdev.
4989 * The error injection ranges are separated by ranges
4990 * which we will not inject errors on any device (DMZs).
4991 * Each DMZ must be large enough such that a single block
4992 * can not straddle it, so that a single block can not be
4993 * a target in two different injection ranges (on different
4994 * leaf vdevs).
4995 *
4996 * For example, with 3 leaves, each chunk looks like:
4997 * 0 to 32M: injection range for leaf 0
4998 * 32M to 64M: DMZ - no injection allowed
4999 * 64M to 96M: injection range for leaf 1
5000 * 96M to 128M: DMZ - no injection allowed
5001 * 128M to 160M: injection range for leaf 2
5002 * 160M to 192M: DMZ - no injection allowed
5003 */
5004 offset = ztest_random(fsize / (leaves << bshift)) *
5005 (leaves << bshift) + (leaf << bshift) +
5006 (ztest_random(1ULL << (bshift - 1)) & -8ULL);
5007
5008 /*
5009 * Only allow damage to the labels at one end of the vdev.
5010 *
5011 * If all labels are damaged, the device will be totally
5012 * inaccessible, which will result in loss of data,
5013 * because we also damage (parts of) the other side of
5014 * the mirror/raidz.
5015 *
5016 * Additionally, we will always have both an even and an
5017 * odd label, so that we can handle crashes in the
5018 * middle of vdev_config_sync().
5019 */
5020 if ((leaf & 1) == 0 && offset < VDEV_LABEL_START_SIZE)
5021 continue;
5022
5023 /*
5024 * The two end labels are stored at the "end" of the disk, but
5025 * the end of the disk (vdev_psize) is aligned to
5026 * sizeof (vdev_label_t).
5027 */
5028 uint64_t psize = P2ALIGN(fsize, sizeof (vdev_label_t));
5029 if ((leaf & 1) == 1 &&
5030 offset + sizeof (bad) > psize - VDEV_LABEL_END_SIZE)
5031 continue;
5032
5033 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
5034 if (mirror_save != zs->zs_mirrors) {
5035 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
5036 (void) close(fd);
5037 return;
5038 }
5039
5040 if (pwrite(fd, &bad, sizeof (bad), offset) != sizeof (bad))
5041 fatal(1, "can't inject bad word at 0x%llx in %s",
5042 offset, pathrand);
5043
5044 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
5045
5046 if (ztest_opts.zo_verbose >= 7)
5047 (void) printf("injected bad word into %s,"
5048 " offset 0x%llx\n", pathrand, (u_longlong_t)offset);
5049 }
5050
5051 (void) close(fd);
5052 }
5053
5054 /*
5055 * Verify that DDT repair works as expected.
5056 */
5057 void
5058 ztest_ddt_repair(ztest_ds_t *zd, uint64_t id)
5059 {
5060 ztest_shared_t *zs = ztest_shared;
5061 spa_t *spa = ztest_spa;
5062 objset_t *os = zd->zd_os;
5063 ztest_od_t od[1];
5064 uint64_t object, blocksize, txg, pattern, psize;
5065 enum zio_checksum checksum = spa_dedup_checksum(spa);
5066 dmu_buf_t *db;
5067 dmu_tx_t *tx;
5068 abd_t *abd;
5069 blkptr_t blk;
5070 int copies = 2 * ZIO_DEDUPDITTO_MIN;
5071
5072 blocksize = ztest_random_blocksize();
5073 blocksize = MIN(blocksize, 2048); /* because we write so many */
5074
5075 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
5076
5077 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
5078 return;
5079
5080 /*
5081 * Take the name lock as writer to prevent anyone else from changing
5082 * the pool and dataset properies we need to maintain during this test.
5083 */
5084 (void) rw_wrlock(&ztest_name_lock);
5085
5086 if (ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_DEDUP, checksum,
5087 B_FALSE) != 0 ||
5088 ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_COPIES, 1,
5089 B_FALSE) != 0) {
5090 (void) rw_unlock(&ztest_name_lock);
5091 return;
5092 }
5093
5094 dmu_objset_stats_t dds;
5095 dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
5096 dmu_objset_fast_stat(os, &dds);
5097 dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
5098
5099 object = od[0].od_object;
5100 blocksize = od[0].od_blocksize;
5101 pattern = zs->zs_guid ^ dds.dds_guid;
5102
5103 ASSERT(object != 0);
5104
5105 tx = dmu_tx_create(os);
5106 dmu_tx_hold_write(tx, object, 0, copies * blocksize);
5107 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
5108 if (txg == 0) {
5109 (void) rw_unlock(&ztest_name_lock);
5110 return;
5111 }
5112
5113 /*
5114 * Write all the copies of our block.
5115 */
5116 for (int i = 0; i < copies; i++) {
5117 uint64_t offset = i * blocksize;
5118 int error = dmu_buf_hold(os, object, offset, FTAG, &db,
5119 DMU_READ_NO_PREFETCH);
5120 if (error != 0) {
5121 fatal(B_FALSE, "dmu_buf_hold(%p, %llu, %llu) = %u",
5122 os, (long long)object, (long long) offset, error);
5123 }
5124 ASSERT(db->db_offset == offset);
5125 ASSERT(db->db_size == blocksize);
5126 ASSERT(ztest_pattern_match(db->db_data, db->db_size, pattern) ||
5127 ztest_pattern_match(db->db_data, db->db_size, 0ULL));
5128 dmu_buf_will_fill(db, tx);
5129 ztest_pattern_set(db->db_data, db->db_size, pattern);
5130 dmu_buf_rele(db, FTAG);
5131 }
5132
5133 dmu_tx_commit(tx);
5134 txg_wait_synced(spa_get_dsl(spa), txg);
5135
5136 /*
5137 * Find out what block we got.
5138 */
5139 VERIFY0(dmu_buf_hold(os, object, 0, FTAG, &db,
5140 DMU_READ_NO_PREFETCH));
5141 blk = *((dmu_buf_impl_t *)db)->db_blkptr;
5142 dmu_buf_rele(db, FTAG);
5143
5144 /*
5145 * Damage the block. Dedup-ditto will save us when we read it later.
5146 */
5147 psize = BP_GET_PSIZE(&blk);
5148 abd = abd_alloc_linear(psize, B_TRUE);
5149 ztest_pattern_set(abd_to_buf(abd), psize, ~pattern);
5150
5151 (void) zio_wait(zio_rewrite(NULL, spa, 0, &blk,
5152 abd, psize, NULL, NULL, ZIO_PRIORITY_SYNC_WRITE,
5153 ZIO_FLAG_CANFAIL | ZIO_FLAG_INDUCE_DAMAGE, NULL));
5154
5155 abd_free(abd);
5156
5157 (void) rw_unlock(&ztest_name_lock);
5158 }
5159
5160 /*
5161 * Scrub the pool.
5162 */
5163 /* ARGSUSED */
5164 void
5165 ztest_scrub(ztest_ds_t *zd, uint64_t id)
5166 {
5167 spa_t *spa = ztest_spa;
5168
5169 (void) spa_scan(spa, POOL_SCAN_SCRUB);
5170 (void) poll(NULL, 0, 100); /* wait a moment, then force a restart */
5171 (void) spa_scan(spa, POOL_SCAN_SCRUB);
5172 }
5173
5174 /*
5175 * Change the guid for the pool.
5176 */
5177 /* ARGSUSED */
5178 void
5179 ztest_reguid(ztest_ds_t *zd, uint64_t id)
5180 {
5181 spa_t *spa = ztest_spa;
5182 uint64_t orig, load;
5183 int error;
5184
5185 orig = spa_guid(spa);
5186 load = spa_load_guid(spa);
5187
5188 (void) rw_wrlock(&ztest_name_lock);
5189 error = spa_change_guid(spa);
5190 (void) rw_unlock(&ztest_name_lock);
5191
5192 if (error != 0)
5193 return;
5194
5195 if (ztest_opts.zo_verbose >= 4) {
5196 (void) printf("Changed guid old %llu -> %llu\n",
5197 (u_longlong_t)orig, (u_longlong_t)spa_guid(spa));
5198 }
5199
5200 VERIFY3U(orig, !=, spa_guid(spa));
5201 VERIFY3U(load, ==, spa_load_guid(spa));
5202 }
5203
5204 /*
5205 * Rename the pool to a different name and then rename it back.
5206 */
5207 /* ARGSUSED */
5208 void
5209 ztest_spa_rename(ztest_ds_t *zd, uint64_t id)
5210 {
5211 char *oldname, *newname;
5212 spa_t *spa;
5213
5214 (void) rw_wrlock(&ztest_name_lock);
5215
5216 oldname = ztest_opts.zo_pool;
5217 newname = umem_alloc(strlen(oldname) + 5, UMEM_NOFAIL);
5218 (void) strcpy(newname, oldname);
5219 (void) strcat(newname, "_tmp");
5220
5221 /*
5222 * Do the rename
5223 */
5224 VERIFY3U(0, ==, spa_rename(oldname, newname));
5225
5226 /*
5227 * Try to open it under the old name, which shouldn't exist
5228 */
5229 VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
5230
5231 /*
5232 * Open it under the new name and make sure it's still the same spa_t.
5233 */
5234 VERIFY3U(0, ==, spa_open(newname, &spa, FTAG));
5235
5236 ASSERT(spa == ztest_spa);
5237 spa_close(spa, FTAG);
5238
5239 /*
5240 * Rename it back to the original
5241 */
5242 VERIFY3U(0, ==, spa_rename(newname, oldname));
5243
5244 /*
5245 * Make sure it can still be opened
5246 */
5247 VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG));
5248
5249 ASSERT(spa == ztest_spa);
5250 spa_close(spa, FTAG);
5251
5252 umem_free(newname, strlen(newname) + 1);
5253
5254 (void) rw_unlock(&ztest_name_lock);
5255 }
5256
5257 /*
5258 * Verify pool integrity by running zdb.
5259 */
5260 static void
5261 ztest_run_zdb(char *pool)
5262 {
5263 int status;
5264 char zdb[MAXPATHLEN + MAXNAMELEN + 20];
5265 char zbuf[1024];
5266 char *bin;
5267 char *ztest;
5268 char *isa;
5269 int isalen;
5270 FILE *fp;
5271
5272 (void) realpath(getexecname(), zdb);
5273
5274 /* zdb lives in /usr/sbin, while ztest lives in /usr/bin */
5275 bin = strstr(zdb, "/usr/bin/");
5276 ztest = strstr(bin, "/ztest");
5277 isa = bin + 8;
5278 isalen = ztest - isa;
5279 isa = strdup(isa);
5280 /* LINTED */
5281 (void) sprintf(bin,
5282 "/usr/sbin%.*s/zdb -bcc%s%s -G -d -U %s %s",
5283 isalen,
5284 isa,
5285 ztest_opts.zo_verbose >= 3 ? "s" : "",
5286 ztest_opts.zo_verbose >= 4 ? "v" : "",
5287 spa_config_path,
5288 pool);
5289 free(isa);
5290
5291 if (ztest_opts.zo_verbose >= 5)
5292 (void) printf("Executing %s\n", strstr(zdb, "zdb "));
5293
5294 fp = popen(zdb, "r");
5295
5296 while (fgets(zbuf, sizeof (zbuf), fp) != NULL)
5297 if (ztest_opts.zo_verbose >= 3)
5298 (void) printf("%s", zbuf);
5299
5300 status = pclose(fp);
5301
5302 if (status == 0)
5303 return;
5304
5305 ztest_dump_core = 0;
5306 if (WIFEXITED(status))
5307 fatal(0, "'%s' exit code %d", zdb, WEXITSTATUS(status));
5308 else
5309 fatal(0, "'%s' died with signal %d", zdb, WTERMSIG(status));
5310 }
5311
5312 static void
5313 ztest_walk_pool_directory(char *header)
5314 {
5315 spa_t *spa = NULL;
5316
5317 if (ztest_opts.zo_verbose >= 6)
5318 (void) printf("%s\n", header);
5319
5320 mutex_enter(&spa_namespace_lock);
5321 while ((spa = spa_next(spa)) != NULL)
5322 if (ztest_opts.zo_verbose >= 6)
5323 (void) printf("\t%s\n", spa_name(spa));
5324 mutex_exit(&spa_namespace_lock);
5325 }
5326
5327 static void
5328 ztest_spa_import_export(char *oldname, char *newname)
5329 {
5330 nvlist_t *config, *newconfig;
5331 uint64_t pool_guid;
5332 spa_t *spa;
5333 int error;
5334
5335 if (ztest_opts.zo_verbose >= 4) {
5336 (void) printf("import/export: old = %s, new = %s\n",
5337 oldname, newname);
5338 }
5339
5340 /*
5341 * Clean up from previous runs.
5342 */
5343 (void) spa_destroy(newname);
5344
5345 /*
5346 * Get the pool's configuration and guid.
5347 */
5348 VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG));
5349
5350 /*
5351 * Kick off a scrub to tickle scrub/export races.
5352 */
5353 if (ztest_random(2) == 0)
5354 (void) spa_scan(spa, POOL_SCAN_SCRUB);
5355
5356 pool_guid = spa_guid(spa);
5357 spa_close(spa, FTAG);
5358
5359 ztest_walk_pool_directory("pools before export");
5360
5361 /*
5362 * Export it.
5363 */
5364 VERIFY3U(0, ==, spa_export(oldname, &config, B_FALSE, B_FALSE));
5365
5366 ztest_walk_pool_directory("pools after export");
5367
5368 /*
5369 * Try to import it.
5370 */
5371 newconfig = spa_tryimport(config);
5372 ASSERT(newconfig != NULL);
5373 nvlist_free(newconfig);
5374
5375 /*
5376 * Import it under the new name.
5377 */
5378 error = spa_import(newname, config, NULL, 0);
5379 if (error != 0) {
5380 dump_nvlist(config, 0);
5381 fatal(B_FALSE, "couldn't import pool %s as %s: error %u",
5382 oldname, newname, error);
5383 }
5384
5385 ztest_walk_pool_directory("pools after import");
5386
5387 /*
5388 * Try to import it again -- should fail with EEXIST.
5389 */
5390 VERIFY3U(EEXIST, ==, spa_import(newname, config, NULL, 0));
5391
5392 /*
5393 * Try to import it under a different name -- should fail with EEXIST.
5394 */
5395 VERIFY3U(EEXIST, ==, spa_import(oldname, config, NULL, 0));
5396
5397 /*
5398 * Verify that the pool is no longer visible under the old name.
5399 */
5400 VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
5401
5402 /*
5403 * Verify that we can open and close the pool using the new name.
5404 */
5405 VERIFY3U(0, ==, spa_open(newname, &spa, FTAG));
5406 ASSERT(pool_guid == spa_guid(spa));
5407 spa_close(spa, FTAG);
5408
5409 nvlist_free(config);
5410 }
5411
5412 static void
5413 ztest_resume(spa_t *spa)
5414 {
5415 if (spa_suspended(spa) && ztest_opts.zo_verbose >= 6)
5416 (void) printf("resuming from suspended state\n");
5417 spa_vdev_state_enter(spa, SCL_NONE);
5418 vdev_clear(spa, NULL);
5419 (void) spa_vdev_state_exit(spa, NULL, 0);
5420 (void) zio_resume(spa);
5421 }
5422
5423 static void *
5424 ztest_resume_thread(void *arg)
5425 {
5426 spa_t *spa = arg;
5427
5428 while (!ztest_exiting) {
5429 if (spa_suspended(spa))
5430 ztest_resume(spa);
5431 (void) poll(NULL, 0, 100);
5432
5433 /*
5434 * Periodically change the zfs_compressed_arc_enabled setting.
5435 */
5436 if (ztest_random(10) == 0)
5437 zfs_compressed_arc_enabled = ztest_random(2);
5438
5439 /*
5440 * Periodically change the zfs_abd_scatter_enabled setting.
5441 */
5442 if (ztest_random(10) == 0)
5443 zfs_abd_scatter_enabled = ztest_random(2);
5444 }
5445 return (NULL);
5446 }
5447
5448 static void *
5449 ztest_deadman_thread(void *arg)
5450 {
5451 ztest_shared_t *zs = arg;
5452 spa_t *spa = ztest_spa;
5453 hrtime_t delta, total = 0;
5454
5455 for (;;) {
5456 delta = zs->zs_thread_stop - zs->zs_thread_start +
5457 MSEC2NSEC(zfs_deadman_synctime_ms);
5458
5459 (void) poll(NULL, 0, (int)NSEC2MSEC(delta));
5460
5461 /*
5462 * If the pool is suspended then fail immediately. Otherwise,
5463 * check to see if the pool is making any progress. If
5464 * vdev_deadman() discovers that there hasn't been any recent
5465 * I/Os then it will end up aborting the tests.
5466 */
5467 if (spa_suspended(spa) || spa->spa_root_vdev == NULL) {
5468 fatal(0, "aborting test after %llu seconds because "
5469 "pool has transitioned to a suspended state.",
5470 zfs_deadman_synctime_ms / 1000);
5471 return (NULL);
5472 }
5473 vdev_deadman(spa->spa_root_vdev);
5474
5475 total += zfs_deadman_synctime_ms/1000;
5476 (void) printf("ztest has been running for %lld seconds\n",
5477 total);
5478 }
5479 }
5480
5481 static void
5482 ztest_execute(int test, ztest_info_t *zi, uint64_t id)
5483 {
5484 ztest_ds_t *zd = &ztest_ds[id % ztest_opts.zo_datasets];
5485 ztest_shared_callstate_t *zc = ZTEST_GET_SHARED_CALLSTATE(test);
5486 hrtime_t functime = gethrtime();
5487
5488 for (int i = 0; i < zi->zi_iters; i++)
5489 zi->zi_func(zd, id);
5490
5491 functime = gethrtime() - functime;
5492
5493 atomic_add_64(&zc->zc_count, 1);
5494 atomic_add_64(&zc->zc_time, functime);
5495
5496 if (ztest_opts.zo_verbose >= 4) {
5497 Dl_info dli;
5498 (void) dladdr((void *)zi->zi_func, &dli);
5499 (void) printf("%6.2f sec in %s\n",
5500 (double)functime / NANOSEC, dli.dli_sname);
5501 }
5502 }
5503
5504 static void *
5505 ztest_thread(void *arg)
5506 {
5507 int rand;
5508 uint64_t id = (uintptr_t)arg;
5509 ztest_shared_t *zs = ztest_shared;
5510 uint64_t call_next;
5511 hrtime_t now;
5512 ztest_info_t *zi;
5513 ztest_shared_callstate_t *zc;
5514
5515 while ((now = gethrtime()) < zs->zs_thread_stop) {
5516 /*
5517 * See if it's time to force a crash.
5518 */
5519 if (now > zs->zs_thread_kill)
5520 ztest_kill(zs);
5521
5522 /*
5523 * If we're getting ENOSPC with some regularity, stop.
5524 */
5525 if (zs->zs_enospc_count > 10)
5526 break;
5527
5528 /*
5529 * Pick a random function to execute.
5530 */
5531 rand = ztest_random(ZTEST_FUNCS);
5532 zi = &ztest_info[rand];
5533 zc = ZTEST_GET_SHARED_CALLSTATE(rand);
5534 call_next = zc->zc_next;
5535
5536 if (now >= call_next &&
5537 atomic_cas_64(&zc->zc_next, call_next, call_next +
5538 ztest_random(2 * zi->zi_interval[0] + 1)) == call_next) {
5539 ztest_execute(rand, zi, id);
5540 }
5541 }
5542
5543 return (NULL);
5544 }
5545
5546 static void
5547 ztest_dataset_name(char *dsname, char *pool, int d)
5548 {
5549 (void) snprintf(dsname, ZFS_MAX_DATASET_NAME_LEN, "%s/ds_%d", pool, d);
5550 }
5551
5552 static void
5553 ztest_dataset_destroy(int d)
5554 {
5555 char name[ZFS_MAX_DATASET_NAME_LEN];
5556
5557 ztest_dataset_name(name, ztest_opts.zo_pool, d);
5558
5559 if (ztest_opts.zo_verbose >= 3)
5560 (void) printf("Destroying %s to free up space\n", name);
5561
5562 /*
5563 * Cleanup any non-standard clones and snapshots. In general,
5564 * ztest thread t operates on dataset (t % zopt_datasets),
5565 * so there may be more than one thing to clean up.
5566 */
5567 for (int t = d; t < ztest_opts.zo_threads;
5568 t += ztest_opts.zo_datasets) {
5569 ztest_dsl_dataset_cleanup(name, t);
5570 }
5571
5572 (void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
5573 DS_FIND_SNAPSHOTS | DS_FIND_CHILDREN);
5574 }
5575
5576 static void
5577 ztest_dataset_dirobj_verify(ztest_ds_t *zd)
5578 {
5579 uint64_t usedobjs, dirobjs, scratch;
5580
5581 /*
5582 * ZTEST_DIROBJ is the object directory for the entire dataset.
5583 * Therefore, the number of objects in use should equal the
5584 * number of ZTEST_DIROBJ entries, +1 for ZTEST_DIROBJ itself.
5585 * If not, we have an object leak.
5586 *
5587 * Note that we can only check this in ztest_dataset_open(),
5588 * when the open-context and syncing-context values agree.
5589 * That's because zap_count() returns the open-context value,
5590 * while dmu_objset_space() returns the rootbp fill count.
5591 */
5592 VERIFY3U(0, ==, zap_count(zd->zd_os, ZTEST_DIROBJ, &dirobjs));
5593 dmu_objset_space(zd->zd_os, &scratch, &scratch, &usedobjs, &scratch);
5594 ASSERT3U(dirobjs + 1, ==, usedobjs);
5595 }
5596
5597 static int
5598 ztest_dataset_open(int d)
5599 {
5600 ztest_ds_t *zd = &ztest_ds[d];
5601 uint64_t committed_seq = ZTEST_GET_SHARED_DS(d)->zd_seq;
5602 objset_t *os;
5603 zilog_t *zilog;
5604 char name[ZFS_MAX_DATASET_NAME_LEN];
5605 int error;
5606
5607 ztest_dataset_name(name, ztest_opts.zo_pool, d);
5608
5609 (void) rw_rdlock(&ztest_name_lock);
5610
5611 error = ztest_dataset_create(name);
5612 if (error == ENOSPC) {
5613 (void) rw_unlock(&ztest_name_lock);
5614 ztest_record_enospc(FTAG);
5615 return (error);
5616 }
5617 ASSERT(error == 0 || error == EEXIST);
5618
5619 VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, zd, &os));
5620 (void) rw_unlock(&ztest_name_lock);
5621
5622 ztest_zd_init(zd, ZTEST_GET_SHARED_DS(d), os);
5623
5624 zilog = zd->zd_zilog;
5625
5626 if (zilog->zl_header->zh_claim_lr_seq != 0 &&
5627 zilog->zl_header->zh_claim_lr_seq < committed_seq)
5628 fatal(0, "missing log records: claimed %llu < committed %llu",
5629 zilog->zl_header->zh_claim_lr_seq, committed_seq);
5630
5631 ztest_dataset_dirobj_verify(zd);
5632
5633 zil_replay(os, zd, ztest_replay_vector);
5634
5635 ztest_dataset_dirobj_verify(zd);
5636
5637 if (ztest_opts.zo_verbose >= 6)
5638 (void) printf("%s replay %llu blocks, %llu records, seq %llu\n",
5639 zd->zd_name,
5640 (u_longlong_t)zilog->zl_parse_blk_count,
5641 (u_longlong_t)zilog->zl_parse_lr_count,
5642 (u_longlong_t)zilog->zl_replaying_seq);
5643
5644 zilog = zil_open(os, ztest_get_data);
5645
5646 if (zilog->zl_replaying_seq != 0 &&
5647 zilog->zl_replaying_seq < committed_seq)
5648 fatal(0, "missing log records: replayed %llu < committed %llu",
5649 zilog->zl_replaying_seq, committed_seq);
5650
5651 return (0);
5652 }
5653
5654 static void
5655 ztest_dataset_close(int d)
5656 {
5657 ztest_ds_t *zd = &ztest_ds[d];
5658
5659 zil_close(zd->zd_zilog);
5660 dmu_objset_disown(zd->zd_os, zd);
5661
5662 ztest_zd_fini(zd);
5663 }
5664
5665 /*
5666 * Kick off threads to run tests on all datasets in parallel.
5667 */
5668 static void
5669 ztest_run(ztest_shared_t *zs)
5670 {
5671 thread_t *tid;
5672 spa_t *spa;
5673 objset_t *os;
5674 thread_t resume_tid;
5675 int error;
5676
5677 ztest_exiting = B_FALSE;
5678
5679 /*
5680 * Initialize parent/child shared state.
5681 */
5682 VERIFY(_mutex_init(&ztest_vdev_lock, USYNC_THREAD, NULL) == 0);
5683 VERIFY(rwlock_init(&ztest_name_lock, USYNC_THREAD, NULL) == 0);
5684
5685 zs->zs_thread_start = gethrtime();
5686 zs->zs_thread_stop =
5687 zs->zs_thread_start + ztest_opts.zo_passtime * NANOSEC;
5688 zs->zs_thread_stop = MIN(zs->zs_thread_stop, zs->zs_proc_stop);
5689 zs->zs_thread_kill = zs->zs_thread_stop;
5690 if (ztest_random(100) < ztest_opts.zo_killrate) {
5691 zs->zs_thread_kill -=
5692 ztest_random(ztest_opts.zo_passtime * NANOSEC);
5693 }
5694
5695 (void) _mutex_init(&zcl.zcl_callbacks_lock, USYNC_THREAD, NULL);
5696
5697 list_create(&zcl.zcl_callbacks, sizeof (ztest_cb_data_t),
5698 offsetof(ztest_cb_data_t, zcd_node));
5699
5700 /*
5701 * Open our pool.
5702 */
5703 kernel_init(FREAD | FWRITE);
5704 VERIFY0(spa_open(ztest_opts.zo_pool, &spa, FTAG));
5705 spa->spa_debug = B_TRUE;
5706 metaslab_preload_limit = ztest_random(20) + 1;
5707 ztest_spa = spa;
5708
5709 dmu_objset_stats_t dds;
5710 VERIFY0(dmu_objset_own(ztest_opts.zo_pool,
5711 DMU_OST_ANY, B_TRUE, FTAG, &os));
5712 dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
5713 dmu_objset_fast_stat(os, &dds);
5714 dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
5715 zs->zs_guid = dds.dds_guid;
5716 dmu_objset_disown(os, FTAG);
5717
5718 spa->spa_dedup_ditto = 2 * ZIO_DEDUPDITTO_MIN;
5719
5720 /*
5721 * We don't expect the pool to suspend unless maxfaults == 0,
5722 * in which case ztest_fault_inject() temporarily takes away
5723 * the only valid replica.
5724 */
5725 if (MAXFAULTS() == 0)
5726 spa->spa_failmode = ZIO_FAILURE_MODE_WAIT;
5727 else
5728 spa->spa_failmode = ZIO_FAILURE_MODE_PANIC;
5729
5730 /*
5731 * Create a thread to periodically resume suspended I/O.
5732 */
5733 VERIFY(thr_create(0, 0, ztest_resume_thread, spa, THR_BOUND,
5734 &resume_tid) == 0);
5735
5736 /*
5737 * Create a deadman thread to abort() if we hang.
5738 */
5739 VERIFY(thr_create(0, 0, ztest_deadman_thread, zs, THR_BOUND,
5740 NULL) == 0);
5741
5742 /*
5743 * Verify that we can safely inquire about about any object,
5744 * whether it's allocated or not. To make it interesting,
5745 * we probe a 5-wide window around each power of two.
5746 * This hits all edge cases, including zero and the max.
5747 */
5748 for (int t = 0; t < 64; t++) {
5749 for (int d = -5; d <= 5; d++) {
5750 error = dmu_object_info(spa->spa_meta_objset,
5751 (1ULL << t) + d, NULL);
5752 ASSERT(error == 0 || error == ENOENT ||
5753 error == EINVAL);
5754 }
5755 }
5756
5757 /*
5758 * If we got any ENOSPC errors on the previous run, destroy something.
5759 */
5760 if (zs->zs_enospc_count != 0) {
5761 int d = ztest_random(ztest_opts.zo_datasets);
5762 ztest_dataset_destroy(d);
5763 }
5764 zs->zs_enospc_count = 0;
5765
5766 tid = umem_zalloc(ztest_opts.zo_threads * sizeof (thread_t),
5767 UMEM_NOFAIL);
5768
5769 if (ztest_opts.zo_verbose >= 4)
5770 (void) printf("starting main threads...\n");
5771
5772 /*
5773 * Kick off all the tests that run in parallel.
5774 */
5775 for (int t = 0; t < ztest_opts.zo_threads; t++) {
5776 if (t < ztest_opts.zo_datasets &&
5777 ztest_dataset_open(t) != 0)
5778 return;
5779 VERIFY(thr_create(0, 0, ztest_thread, (void *)(uintptr_t)t,
5780 THR_BOUND, &tid[t]) == 0);
5781 }
5782
5783 /*
5784 * Wait for all of the tests to complete. We go in reverse order
5785 * so we don't close datasets while threads are still using them.
5786 */
5787 for (int t = ztest_opts.zo_threads - 1; t >= 0; t--) {
5788 VERIFY(thr_join(tid[t], NULL, NULL) == 0);
5789 if (t < ztest_opts.zo_datasets)
5790 ztest_dataset_close(t);
5791 }
5792
5793 txg_wait_synced(spa_get_dsl(spa), 0);
5794
5795 zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(spa));
5796 zs->zs_space = metaslab_class_get_space(spa_normal_class(spa));
5797 zfs_dbgmsg_print(FTAG);
5798
5799 umem_free(tid, ztest_opts.zo_threads * sizeof (thread_t));
5800
5801 /* Kill the resume thread */
5802 ztest_exiting = B_TRUE;
5803 VERIFY(thr_join(resume_tid, NULL, NULL) == 0);
5804 ztest_resume(spa);
5805
5806 /*
5807 * Right before closing the pool, kick off a bunch of async I/O;
5808 * spa_close() should wait for it to complete.
5809 */
5810 for (uint64_t object = 1; object < 50; object++) {
5811 dmu_prefetch(spa->spa_meta_objset, object, 0, 0, 1ULL << 20,
5812 ZIO_PRIORITY_SYNC_READ);
5813 }
5814
5815 spa_close(spa, FTAG);
5816
5817 /*
5818 * Verify that we can loop over all pools.
5819 */
5820 mutex_enter(&spa_namespace_lock);
5821 for (spa = spa_next(NULL); spa != NULL; spa = spa_next(spa))
5822 if (ztest_opts.zo_verbose > 3)
5823 (void) printf("spa_next: found %s\n", spa_name(spa));
5824 mutex_exit(&spa_namespace_lock);
5825
5826 /*
5827 * Verify that we can export the pool and reimport it under a
5828 * different name.
5829 */
5830 if (ztest_random(2) == 0) {
5831 char name[ZFS_MAX_DATASET_NAME_LEN];
5832 (void) snprintf(name, sizeof (name), "%s_import",
5833 ztest_opts.zo_pool);
5834 ztest_spa_import_export(ztest_opts.zo_pool, name);
5835 ztest_spa_import_export(name, ztest_opts.zo_pool);
5836 }
5837
5838 kernel_fini();
5839
5840 list_destroy(&zcl.zcl_callbacks);
5841
5842 (void) _mutex_destroy(&zcl.zcl_callbacks_lock);
5843
5844 (void) rwlock_destroy(&ztest_name_lock);
5845 (void) _mutex_destroy(&ztest_vdev_lock);
5846 }
5847
5848 static void
5849 ztest_freeze(void)
5850 {
5851 ztest_ds_t *zd = &ztest_ds[0];
5852 spa_t *spa;
5853 int numloops = 0;
5854
5855 if (ztest_opts.zo_verbose >= 3)
5856 (void) printf("testing spa_freeze()...\n");
5857
5858 kernel_init(FREAD | FWRITE);
5859 VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
5860 VERIFY3U(0, ==, ztest_dataset_open(0));
5861 spa->spa_debug = B_TRUE;
5862 ztest_spa = spa;
5863
5864 /*
5865 * Force the first log block to be transactionally allocated.
5866 * We have to do this before we freeze the pool -- otherwise
5867 * the log chain won't be anchored.
5868 */
5869 while (BP_IS_HOLE(&zd->zd_zilog->zl_header->zh_log)) {
5870 ztest_dmu_object_alloc_free(zd, 0);
5871 zil_commit(zd->zd_zilog, 0);
5872 }
5873
5874 txg_wait_synced(spa_get_dsl(spa), 0);
5875
5876 /*
5877 * Freeze the pool. This stops spa_sync() from doing anything,
5878 * so that the only way to record changes from now on is the ZIL.
5879 */
5880 spa_freeze(spa);
5881
5882 /*
5883 * Because it is hard to predict how much space a write will actually
5884 * require beforehand, we leave ourselves some fudge space to write over
5885 * capacity.
5886 */
5887 uint64_t capacity = metaslab_class_get_space(spa_normal_class(spa)) / 2;
5888
5889 /*
5890 * Run tests that generate log records but don't alter the pool config
5891 * or depend on DSL sync tasks (snapshots, objset create/destroy, etc).
5892 * We do a txg_wait_synced() after each iteration to force the txg
5893 * to increase well beyond the last synced value in the uberblock.
5894 * The ZIL should be OK with that.
5895 *
5896 * Run a random number of times less than zo_maxloops and ensure we do
5897 * not run out of space on the pool.
5898 */
5899 while (ztest_random(10) != 0 &&
5900 numloops++ < ztest_opts.zo_maxloops &&
5901 metaslab_class_get_alloc(spa_normal_class(spa)) < capacity) {
5902 ztest_od_t od;
5903 ztest_od_init(&od, 0, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
5904 VERIFY0(ztest_object_init(zd, &od, sizeof (od), B_FALSE));
5905 ztest_io(zd, od.od_object,
5906 ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
5907 txg_wait_synced(spa_get_dsl(spa), 0);
5908 }
5909
5910 /*
5911 * Commit all of the changes we just generated.
5912 */
5913 zil_commit(zd->zd_zilog, 0);
5914 txg_wait_synced(spa_get_dsl(spa), 0);
5915
5916 /*
5917 * Close our dataset and close the pool.
5918 */
5919 ztest_dataset_close(0);
5920 spa_close(spa, FTAG);
5921 kernel_fini();
5922
5923 /*
5924 * Open and close the pool and dataset to induce log replay.
5925 */
5926 kernel_init(FREAD | FWRITE);
5927 VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
5928 ASSERT(spa_freeze_txg(spa) == UINT64_MAX);
5929 VERIFY3U(0, ==, ztest_dataset_open(0));
5930 ztest_dataset_close(0);
5931
5932 spa->spa_debug = B_TRUE;
5933 ztest_spa = spa;
5934 txg_wait_synced(spa_get_dsl(spa), 0);
5935 ztest_reguid(NULL, 0);
5936
5937 spa_close(spa, FTAG);
5938 kernel_fini();
5939 }
5940
5941 void
5942 print_time(hrtime_t t, char *timebuf)
5943 {
5944 hrtime_t s = t / NANOSEC;
5945 hrtime_t m = s / 60;
5946 hrtime_t h = m / 60;
5947 hrtime_t d = h / 24;
5948
5949 s -= m * 60;
5950 m -= h * 60;
5951 h -= d * 24;
5952
5953 timebuf[0] = '\0';
5954
5955 if (d)
5956 (void) sprintf(timebuf,
5957 "%llud%02lluh%02llum%02llus", d, h, m, s);
5958 else if (h)
5959 (void) sprintf(timebuf, "%lluh%02llum%02llus", h, m, s);
5960 else if (m)
5961 (void) sprintf(timebuf, "%llum%02llus", m, s);
5962 else
5963 (void) sprintf(timebuf, "%llus", s);
5964 }
5965
5966 static nvlist_t *
5967 make_random_props()
5968 {
5969 nvlist_t *props;
5970
5971 VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
5972 if (ztest_random(2) == 0)
5973 return (props);
5974 VERIFY(nvlist_add_uint64(props, "autoreplace", 1) == 0);
5975
5976 return (props);
5977 }
5978
5979 /*
5980 * Create a storage pool with the given name and initial vdev size.
5981 * Then test spa_freeze() functionality.
5982 */
5983 static void
5984 ztest_init(ztest_shared_t *zs)
5985 {
5986 spa_t *spa;
5987 nvlist_t *nvroot, *props;
5988
5989 VERIFY(_mutex_init(&ztest_vdev_lock, USYNC_THREAD, NULL) == 0);
5990 VERIFY(rwlock_init(&ztest_name_lock, USYNC_THREAD, NULL) == 0);
5991
5992 kernel_init(FREAD | FWRITE);
5993
5994 /*
5995 * Create the storage pool.
5996 */
5997 (void) spa_destroy(ztest_opts.zo_pool);
5998 ztest_shared->zs_vdev_next_leaf = 0;
5999 zs->zs_splits = 0;
6000 zs->zs_mirrors = ztest_opts.zo_mirrors;
6001 nvroot = make_vdev_root(NULL, NULL, NULL, ztest_opts.zo_vdev_size, 0,
6002 0, ztest_opts.zo_raidz, zs->zs_mirrors, 1);
6003 props = make_random_props();
6004 for (int i = 0; i < SPA_FEATURES; i++) {
6005 char buf[1024];
6006 (void) snprintf(buf, sizeof (buf), "feature@%s",
6007 spa_feature_table[i].fi_uname);
6008 VERIFY3U(0, ==, nvlist_add_uint64(props, buf, 0));
6009 }
6010 VERIFY3U(0, ==, spa_create(ztest_opts.zo_pool, nvroot, props, NULL));
6011 nvlist_free(nvroot);
6012 nvlist_free(props);
6013
6014 VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
6015 zs->zs_metaslab_sz =
6016 1ULL << spa->spa_root_vdev->vdev_child[0]->vdev_ms_shift;
6017
6018 spa_close(spa, FTAG);
6019
6020 kernel_fini();
6021
6022 ztest_run_zdb(ztest_opts.zo_pool);
6023
6024 ztest_freeze();
6025
6026 ztest_run_zdb(ztest_opts.zo_pool);
6027
6028 (void) rwlock_destroy(&ztest_name_lock);
6029 (void) _mutex_destroy(&ztest_vdev_lock);
6030 }
6031
6032 static void
6033 setup_data_fd(void)
6034 {
6035 static char ztest_name_data[] = "/tmp/ztest.data.XXXXXX";
6036
6037 ztest_fd_data = mkstemp(ztest_name_data);
6038 ASSERT3S(ztest_fd_data, >=, 0);
6039 (void) unlink(ztest_name_data);
6040 }
6041
6042
6043 static int
6044 shared_data_size(ztest_shared_hdr_t *hdr)
6045 {
6046 int size;
6047
6048 size = hdr->zh_hdr_size;
6049 size += hdr->zh_opts_size;
6050 size += hdr->zh_size;
6051 size += hdr->zh_stats_size * hdr->zh_stats_count;
6052 size += hdr->zh_ds_size * hdr->zh_ds_count;
6053
6054 return (size);
6055 }
6056
6057 static void
6058 setup_hdr(void)
6059 {
6060 int size;
6061 ztest_shared_hdr_t *hdr;
6062
6063 hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
6064 PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
6065 ASSERT(hdr != MAP_FAILED);
6066
6067 VERIFY3U(0, ==, ftruncate(ztest_fd_data, sizeof (ztest_shared_hdr_t)));
6068
6069 hdr->zh_hdr_size = sizeof (ztest_shared_hdr_t);
6070 hdr->zh_opts_size = sizeof (ztest_shared_opts_t);
6071 hdr->zh_size = sizeof (ztest_shared_t);
6072 hdr->zh_stats_size = sizeof (ztest_shared_callstate_t);
6073 hdr->zh_stats_count = ZTEST_FUNCS;
6074 hdr->zh_ds_size = sizeof (ztest_shared_ds_t);
6075 hdr->zh_ds_count = ztest_opts.zo_datasets;
6076
6077 size = shared_data_size(hdr);
6078 VERIFY3U(0, ==, ftruncate(ztest_fd_data, size));
6079
6080 (void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
6081 }
6082
6083 static void
6084 setup_data(void)
6085 {
6086 int size, offset;
6087 ztest_shared_hdr_t *hdr;
6088 uint8_t *buf;
6089
6090 hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
6091 PROT_READ, MAP_SHARED, ztest_fd_data, 0);
6092 ASSERT(hdr != MAP_FAILED);
6093
6094 size = shared_data_size(hdr);
6095
6096 (void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
6097 hdr = ztest_shared_hdr = (void *)mmap(0, P2ROUNDUP(size, getpagesize()),
6098 PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
6099 ASSERT(hdr != MAP_FAILED);
6100 buf = (uint8_t *)hdr;
6101
6102 offset = hdr->zh_hdr_size;
6103 ztest_shared_opts = (void *)&buf[offset];
6104 offset += hdr->zh_opts_size;
6105 ztest_shared = (void *)&buf[offset];
6106 offset += hdr->zh_size;
6107 ztest_shared_callstate = (void *)&buf[offset];
6108 offset += hdr->zh_stats_size * hdr->zh_stats_count;
6109 ztest_shared_ds = (void *)&buf[offset];
6110 }
6111
6112 static boolean_t
6113 exec_child(char *cmd, char *libpath, boolean_t ignorekill, int *statusp)
6114 {
6115 pid_t pid;
6116 int status;
6117 char *cmdbuf = NULL;
6118
6119 pid = fork();
6120
6121 if (cmd == NULL) {
6122 cmdbuf = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
6123 (void) strlcpy(cmdbuf, getexecname(), MAXPATHLEN);
6124 cmd = cmdbuf;
6125 }
6126
6127 if (pid == -1)
6128 fatal(1, "fork failed");
6129
6130 if (pid == 0) { /* child */
6131 char *emptyargv[2] = { cmd, NULL };
6132 char fd_data_str[12];
6133
6134 struct rlimit rl = { 1024, 1024 };
6135 (void) setrlimit(RLIMIT_NOFILE, &rl);
6136
6137 (void) close(ztest_fd_rand);
6138 VERIFY3U(11, >=,
6139 snprintf(fd_data_str, 12, "%d", ztest_fd_data));
6140 VERIFY0(setenv("ZTEST_FD_DATA", fd_data_str, 1));
6141
6142 (void) enable_extended_FILE_stdio(-1, -1);
6143 if (libpath != NULL)
6144 VERIFY(0 == setenv("LD_LIBRARY_PATH", libpath, 1));
6145 (void) execv(cmd, emptyargv);
6146 ztest_dump_core = B_FALSE;
6147 fatal(B_TRUE, "exec failed: %s", cmd);
6148 }
6149
6150 if (cmdbuf != NULL) {
6151 umem_free(cmdbuf, MAXPATHLEN);
6152 cmd = NULL;
6153 }
6154
6155 while (waitpid(pid, &status, 0) != pid)
6156 continue;
6157 if (statusp != NULL)
6158 *statusp = status;
6159
6160 if (WIFEXITED(status)) {
6161 if (WEXITSTATUS(status) != 0) {
6162 (void) fprintf(stderr, "child exited with code %d\n",
6163 WEXITSTATUS(status));
6164 exit(2);
6165 }
6166 return (B_FALSE);
6167 } else if (WIFSIGNALED(status)) {
6168 if (!ignorekill || WTERMSIG(status) != SIGKILL) {
6169 (void) fprintf(stderr, "child died with signal %d\n",
6170 WTERMSIG(status));
6171 exit(3);
6172 }
6173 return (B_TRUE);
6174 } else {
6175 (void) fprintf(stderr, "something strange happened to child\n");
6176 exit(4);
6177 /* NOTREACHED */
6178 }
6179 }
6180
6181 static void
6182 ztest_run_init(void)
6183 {
6184 ztest_shared_t *zs = ztest_shared;
6185
6186 ASSERT(ztest_opts.zo_init != 0);
6187
6188 /*
6189 * Blow away any existing copy of zpool.cache
6190 */
6191 (void) remove(spa_config_path);
6192
6193 /*
6194 * Create and initialize our storage pool.
6195 */
6196 for (int i = 1; i <= ztest_opts.zo_init; i++) {
6197 bzero(zs, sizeof (ztest_shared_t));
6198 if (ztest_opts.zo_verbose >= 3 &&
6199 ztest_opts.zo_init != 1) {
6200 (void) printf("ztest_init(), pass %d\n", i);
6201 }
6202 ztest_init(zs);
6203 }
6204 }
6205
6206 int
6207 main(int argc, char **argv)
6208 {
6209 int kills = 0;
6210 int iters = 0;
6211 int older = 0;
6212 int newer = 0;
6213 ztest_shared_t *zs;
6214 ztest_info_t *zi;
6215 ztest_shared_callstate_t *zc;
6216 char timebuf[100];
6217 char numbuf[NN_NUMBUF_SZ];
6218 spa_t *spa;
6219 char *cmd;
6220 boolean_t hasalt;
6221 char *fd_data_str = getenv("ZTEST_FD_DATA");
6222
6223 (void) setvbuf(stdout, NULL, _IOLBF, 0);
6224
6225 dprintf_setup(&argc, argv);
6226 zfs_deadman_synctime_ms = 300000;
6227
6228 ztest_fd_rand = open("/dev/urandom", O_RDONLY);
6229 ASSERT3S(ztest_fd_rand, >=, 0);
6230
6231 if (!fd_data_str) {
6232 process_options(argc, argv);
6233
6234 setup_data_fd();
6235 setup_hdr();
6236 setup_data();
6237 bcopy(&ztest_opts, ztest_shared_opts,
6238 sizeof (*ztest_shared_opts));
6239 } else {
6240 ztest_fd_data = atoi(fd_data_str);
6241 setup_data();
6242 bcopy(ztest_shared_opts, &ztest_opts, sizeof (ztest_opts));
6243 }
6244 ASSERT3U(ztest_opts.zo_datasets, ==, ztest_shared_hdr->zh_ds_count);
6245
6246 /* Override location of zpool.cache */
6247 VERIFY3U(asprintf((char **)&spa_config_path, "%s/zpool.cache",
6248 ztest_opts.zo_dir), !=, -1);
6249
6250 ztest_ds = umem_alloc(ztest_opts.zo_datasets * sizeof (ztest_ds_t),
6251 UMEM_NOFAIL);
6252 zs = ztest_shared;
6253
6254 if (fd_data_str) {
6255 metaslab_gang_bang = ztest_opts.zo_metaslab_gang_bang;
6256 metaslab_df_alloc_threshold =
6257 zs->zs_metaslab_df_alloc_threshold;
6258
6259 if (zs->zs_do_init)
6260 ztest_run_init();
6261 else
6262 ztest_run(zs);
6263 exit(0);
6264 }
6265
6266 hasalt = (strlen(ztest_opts.zo_alt_ztest) != 0);
6267
6268 if (ztest_opts.zo_verbose >= 1) {
6269 (void) printf("%llu vdevs, %d datasets, %d threads,"
6270 " %llu seconds...\n",
6271 (u_longlong_t)ztest_opts.zo_vdevs,
6272 ztest_opts.zo_datasets,
6273 ztest_opts.zo_threads,
6274 (u_longlong_t)ztest_opts.zo_time);
6275 }
6276
6277 cmd = umem_alloc(MAXNAMELEN, UMEM_NOFAIL);
6278 (void) strlcpy(cmd, getexecname(), MAXNAMELEN);
6279
6280 zs->zs_do_init = B_TRUE;
6281 if (strlen(ztest_opts.zo_alt_ztest) != 0) {
6282 if (ztest_opts.zo_verbose >= 1) {
6283 (void) printf("Executing older ztest for "
6284 "initialization: %s\n", ztest_opts.zo_alt_ztest);
6285 }
6286 VERIFY(!exec_child(ztest_opts.zo_alt_ztest,
6287 ztest_opts.zo_alt_libpath, B_FALSE, NULL));
6288 } else {
6289 VERIFY(!exec_child(NULL, NULL, B_FALSE, NULL));
6290 }
6291 zs->zs_do_init = B_FALSE;
6292
6293 zs->zs_proc_start = gethrtime();
6294 zs->zs_proc_stop = zs->zs_proc_start + ztest_opts.zo_time * NANOSEC;
6295
6296 for (int f = 0; f < ZTEST_FUNCS; f++) {
6297 zi = &ztest_info[f];
6298 zc = ZTEST_GET_SHARED_CALLSTATE(f);
6299 if (zs->zs_proc_start + zi->zi_interval[0] > zs->zs_proc_stop)
6300 zc->zc_next = UINT64_MAX;
6301 else
6302 zc->zc_next = zs->zs_proc_start +
6303 ztest_random(2 * zi->zi_interval[0] + 1);
6304 }
6305
6306 /*
6307 * Run the tests in a loop. These tests include fault injection
6308 * to verify that self-healing data works, and forced crashes
6309 * to verify that we never lose on-disk consistency.
6310 */
6311 while (gethrtime() < zs->zs_proc_stop) {
6312 int status;
6313 boolean_t killed;
6314
6315 /*
6316 * Initialize the workload counters for each function.
6317 */
6318 for (int f = 0; f < ZTEST_FUNCS; f++) {
6319 zc = ZTEST_GET_SHARED_CALLSTATE(f);
6320 zc->zc_count = 0;
6321 zc->zc_time = 0;
6322 }
6323
6324 /* Set the allocation switch size */
6325 zs->zs_metaslab_df_alloc_threshold =
6326 ztest_random(zs->zs_metaslab_sz / 4) + 1;
6327
6328 if (!hasalt || ztest_random(2) == 0) {
6329 if (hasalt && ztest_opts.zo_verbose >= 1) {
6330 (void) printf("Executing newer ztest: %s\n",
6331 cmd);
6332 }
6333 newer++;
6334 killed = exec_child(cmd, NULL, B_TRUE, &status);
6335 } else {
6336 if (hasalt && ztest_opts.zo_verbose >= 1) {
6337 (void) printf("Executing older ztest: %s\n",
6338 ztest_opts.zo_alt_ztest);
6339 }
6340 older++;
6341 killed = exec_child(ztest_opts.zo_alt_ztest,
6342 ztest_opts.zo_alt_libpath, B_TRUE, &status);
6343 }
6344
6345 if (killed)
6346 kills++;
6347 iters++;
6348
6349 if (ztest_opts.zo_verbose >= 1) {
6350 hrtime_t now = gethrtime();
6351
6352 now = MIN(now, zs->zs_proc_stop);
6353 print_time(zs->zs_proc_stop - now, timebuf);
6354 nicenum(zs->zs_space, numbuf, sizeof (numbuf));
6355
6356 (void) printf("Pass %3d, %8s, %3llu ENOSPC, "
6357 "%4.1f%% of %5s used, %3.0f%% done, %8s to go\n",
6358 iters,
6359 WIFEXITED(status) ? "Complete" : "SIGKILL",
6360 (u_longlong_t)zs->zs_enospc_count,
6361 100.0 * zs->zs_alloc / zs->zs_space,
6362 numbuf,
6363 100.0 * (now - zs->zs_proc_start) /
6364 (ztest_opts.zo_time * NANOSEC), timebuf);
6365 }
6366
6367 if (ztest_opts.zo_verbose >= 2) {
6368 (void) printf("\nWorkload summary:\n\n");
6369 (void) printf("%7s %9s %s\n",
6370 "Calls", "Time", "Function");
6371 (void) printf("%7s %9s %s\n",
6372 "-----", "----", "--------");
6373 for (int f = 0; f < ZTEST_FUNCS; f++) {
6374 Dl_info dli;
6375
6376 zi = &ztest_info[f];
6377 zc = ZTEST_GET_SHARED_CALLSTATE(f);
6378 print_time(zc->zc_time, timebuf);
6379 (void) dladdr((void *)zi->zi_func, &dli);
6380 (void) printf("%7llu %9s %s\n",
6381 (u_longlong_t)zc->zc_count, timebuf,
6382 dli.dli_sname);
6383 }
6384 (void) printf("\n");
6385 }
6386
6387 /*
6388 * It's possible that we killed a child during a rename test,
6389 * in which case we'll have a 'ztest_tmp' pool lying around
6390 * instead of 'ztest'. Do a blind rename in case this happened.
6391 */
6392 kernel_init(FREAD);
6393 if (spa_open(ztest_opts.zo_pool, &spa, FTAG) == 0) {
6394 spa_close(spa, FTAG);
6395 } else {
6396 char tmpname[ZFS_MAX_DATASET_NAME_LEN];
6397 kernel_fini();
6398 kernel_init(FREAD | FWRITE);
6399 (void) snprintf(tmpname, sizeof (tmpname), "%s_tmp",
6400 ztest_opts.zo_pool);
6401 (void) spa_rename(tmpname, ztest_opts.zo_pool);
6402 }
6403 kernel_fini();
6404
6405 ztest_run_zdb(ztest_opts.zo_pool);
6406 }
6407
6408 if (ztest_opts.zo_verbose >= 1) {
6409 if (hasalt) {
6410 (void) printf("%d runs of older ztest: %s\n", older,
6411 ztest_opts.zo_alt_ztest);
6412 (void) printf("%d runs of newer ztest: %s\n", newer,
6413 cmd);
6414 }
6415 (void) printf("%d killed, %d completed, %.0f%% kill rate\n",
6416 kills, iters - kills, (100.0 * kills) / MAX(1, iters));
6417 }
6418
6419 umem_free(cmd, MAXNAMELEN);
6420
6421 return (0);
6422 }