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 * Copyright 2017 RackTop Systems.
29 */
30
31 /*
32 * The objective of this program is to provide a DMU/ZAP/SPA stress test
33 * that runs entirely in userland, is easy to use, and easy to extend.
34 *
35 * The overall design of the ztest program is as follows:
36 *
37 * (1) For each major functional area (e.g. adding vdevs to a pool,
38 * creating and destroying datasets, reading and writing objects, etc)
39 * we have a simple routine to test that functionality. These
40 * individual routines do not have to do anything "stressful".
41 *
42 * (2) We turn these simple functionality tests into a stress test by
43 * running them all in parallel, with as many threads as desired,
44 * and spread across as many datasets, objects, and vdevs as desired.
45 *
46 * (3) While all this is happening, we inject faults into the pool to
47 * verify that self-healing data really works.
48 *
49 * (4) Every time we open a dataset, we change its checksum and compression
50 * functions. Thus even individual objects vary from block to block
51 * in which checksum they use and whether they're compressed.
52 *
53 * (5) To verify that we never lose on-disk consistency after a crash,
54 * we run the entire test in a child of the main process.
55 * At random times, the child self-immolates with a SIGKILL.
56 * This is the software equivalent of pulling the power cord.
57 * The parent then runs the test again, using the existing
58 * storage pool, as many times as desired. If backwards compatibility
59 * testing is enabled ztest will sometimes run the "older" version
60 * of ztest after a SIGKILL.
61 *
62 * (6) To verify that we don't have future leaks or temporal incursions,
63 * many of the functional tests record the transaction group number
64 * as part of their data. When reading old data, they verify that
65 * the transaction group number is less than the current, open txg.
66 * If you add a new test, please do this if applicable.
67 *
68 * When run with no arguments, ztest runs for about five minutes and
69 * produces no output if successful. To get a little bit of information,
70 * specify -V. To get more information, specify -VV, and so on.
71 *
72 * To turn this into an overnight stress test, use -T to specify run time.
73 *
74 * You can ask more more vdevs [-v], datasets [-d], or threads [-t]
75 * to increase the pool capacity, fanout, and overall stress level.
76 *
77 * Use the -k option to set the desired frequency of kills.
78 *
79 * When ztest invokes itself it passes all relevant information through a
80 * temporary file which is mmap-ed in the child process. This allows shared
81 * memory to survive the exec syscall. The ztest_shared_hdr_t struct is always
82 * stored at offset 0 of this file and contains information on the size and
83 * number of shared structures in the file. The information stored in this file
84 * must remain backwards compatible with older versions of ztest so that
85 * ztest can invoke them during backwards compatibility testing (-B).
86 */
87
88 #include <sys/zfs_context.h>
89 #include <sys/spa.h>
90 #include <sys/dmu.h>
91 #include <sys/txg.h>
92 #include <sys/dbuf.h>
93 #include <sys/zap.h>
94 #include <sys/dmu_objset.h>
95 #include <sys/poll.h>
96 #include <sys/stat.h>
97 #include <sys/time.h>
98 #include <sys/wait.h>
99 #include <sys/mman.h>
100 #include <sys/resource.h>
101 #include <sys/zio.h>
102 #include <sys/zil.h>
103 #include <sys/zil_impl.h>
104 #include <sys/vdev_impl.h>
105 #include <sys/vdev_file.h>
106 #include <sys/spa_impl.h>
107 #include <sys/metaslab_impl.h>
108 #include <sys/dsl_prop.h>
109 #include <sys/dsl_dataset.h>
110 #include <sys/dsl_destroy.h>
111 #include <sys/dsl_scan.h>
112 #include <sys/zio_checksum.h>
113 #include <sys/refcount.h>
114 #include <sys/zfeature.h>
115 #include <sys/dsl_userhold.h>
116 #include <sys/abd.h>
117 #include <stdio.h>
118 #include <stdio_ext.h>
119 #include <stdlib.h>
120 #include <unistd.h>
121 #include <signal.h>
122 #include <umem.h>
123 #include <dlfcn.h>
124 #include <ctype.h>
125 #include <math.h>
126 #include <sys/fs/zfs.h>
127 #include <libnvpair.h>
128 #include <libcmdutils.h>
129
130 static int ztest_fd_data = -1;
131 static int ztest_fd_rand = -1;
132
133 typedef struct ztest_shared_hdr {
134 uint64_t zh_hdr_size;
135 uint64_t zh_opts_size;
136 uint64_t zh_size;
137 uint64_t zh_stats_size;
138 uint64_t zh_stats_count;
139 uint64_t zh_ds_size;
140 uint64_t zh_ds_count;
141 } ztest_shared_hdr_t;
142
143 static ztest_shared_hdr_t *ztest_shared_hdr;
144
145 typedef struct ztest_shared_opts {
146 char zo_pool[ZFS_MAX_DATASET_NAME_LEN];
147 char zo_dir[ZFS_MAX_DATASET_NAME_LEN];
148 char zo_alt_ztest[MAXNAMELEN];
149 char zo_alt_libpath[MAXNAMELEN];
150 uint64_t zo_vdevs;
151 uint64_t zo_vdevtime;
152 size_t zo_vdev_size;
153 int zo_ashift;
154 int zo_mirrors;
155 int zo_raidz;
156 int zo_raidz_parity;
157 int zo_datasets;
158 int zo_threads;
159 uint64_t zo_passtime;
160 uint64_t zo_killrate;
161 int zo_verbose;
162 int zo_init;
163 uint64_t zo_time;
164 uint64_t zo_maxloops;
165 uint64_t zo_metaslab_gang_bang;
166 } ztest_shared_opts_t;
167
168 static const ztest_shared_opts_t ztest_opts_defaults = {
169 .zo_pool = { 'z', 't', 'e', 's', 't', '\0' },
170 .zo_dir = { '/', 't', 'm', 'p', '\0' },
171 .zo_alt_ztest = { '\0' },
172 .zo_alt_libpath = { '\0' },
173 .zo_vdevs = 5,
174 .zo_ashift = SPA_MINBLOCKSHIFT,
175 .zo_mirrors = 2,
176 .zo_raidz = 4,
177 .zo_raidz_parity = 1,
178 .zo_vdev_size = SPA_MINDEVSIZE * 4, /* 256m default size */
179 .zo_datasets = 7,
180 .zo_threads = 23,
181 .zo_passtime = 60, /* 60 seconds */
182 .zo_killrate = 70, /* 70% kill rate */
183 .zo_verbose = 0,
184 .zo_init = 1,
185 .zo_time = 300, /* 5 minutes */
186 .zo_maxloops = 50, /* max loops during spa_freeze() */
187 .zo_metaslab_gang_bang = 32 << 10
188 };
189
190 extern uint64_t metaslab_gang_bang;
191 extern uint64_t metaslab_df_alloc_threshold;
192 extern uint64_t zfs_deadman_synctime_ms;
193 extern int metaslab_preload_limit;
194 extern boolean_t zfs_compressed_arc_enabled;
195 extern boolean_t zfs_abd_scatter_enabled;
196
197 static ztest_shared_opts_t *ztest_shared_opts;
198 static ztest_shared_opts_t ztest_opts;
199
200 typedef struct ztest_shared_ds {
201 uint64_t zd_seq;
202 } ztest_shared_ds_t;
203
204 static ztest_shared_ds_t *ztest_shared_ds;
205 #define ZTEST_GET_SHARED_DS(d) (&ztest_shared_ds[d])
206
207 #define BT_MAGIC 0x123456789abcdefULL
208 #define MAXFAULTS() \
209 (MAX(zs->zs_mirrors, 1) * (ztest_opts.zo_raidz_parity + 1) - 1)
210
211 enum ztest_io_type {
212 ZTEST_IO_WRITE_TAG,
213 ZTEST_IO_WRITE_PATTERN,
214 ZTEST_IO_WRITE_ZEROES,
215 ZTEST_IO_TRUNCATE,
216 ZTEST_IO_SETATTR,
217 ZTEST_IO_REWRITE,
218 ZTEST_IO_TYPES
219 };
220
221 typedef struct ztest_block_tag {
222 uint64_t bt_magic;
223 uint64_t bt_objset;
224 uint64_t bt_object;
225 uint64_t bt_offset;
226 uint64_t bt_gen;
227 uint64_t bt_txg;
228 uint64_t bt_crtxg;
229 } ztest_block_tag_t;
230
231 typedef struct bufwad {
232 uint64_t bw_index;
233 uint64_t bw_txg;
234 uint64_t bw_data;
235 } bufwad_t;
236
237 /*
238 * XXX -- fix zfs range locks to be generic so we can use them here.
239 */
240 typedef enum {
241 RL_READER,
242 RL_WRITER,
243 RL_APPEND
244 } rl_type_t;
245
246 typedef struct rll {
247 void *rll_writer;
248 int rll_readers;
249 kmutex_t rll_lock;
250 kcondvar_t rll_cv;
251 } rll_t;
252
253 typedef struct rl {
254 uint64_t rl_object;
255 uint64_t rl_offset;
256 uint64_t rl_size;
257 rll_t *rl_lock;
258 } rl_t;
259
260 #define ZTEST_RANGE_LOCKS 64
261 #define ZTEST_OBJECT_LOCKS 64
262
263 /*
264 * Object descriptor. Used as a template for object lookup/create/remove.
265 */
266 typedef struct ztest_od {
267 uint64_t od_dir;
268 uint64_t od_object;
269 dmu_object_type_t od_type;
270 dmu_object_type_t od_crtype;
271 uint64_t od_blocksize;
272 uint64_t od_crblocksize;
273 uint64_t od_gen;
274 uint64_t od_crgen;
275 char od_name[ZFS_MAX_DATASET_NAME_LEN];
276 } ztest_od_t;
277
278 /*
279 * Per-dataset state.
280 */
281 typedef struct ztest_ds {
282 ztest_shared_ds_t *zd_shared;
283 objset_t *zd_os;
284 krwlock_t zd_zilog_lock;
285 zilog_t *zd_zilog;
286 ztest_od_t *zd_od; /* debugging aid */
287 char zd_name[ZFS_MAX_DATASET_NAME_LEN];
288 kmutex_t zd_dirobj_lock;
289 rll_t zd_object_lock[ZTEST_OBJECT_LOCKS];
290 rll_t zd_range_lock[ZTEST_RANGE_LOCKS];
291 } ztest_ds_t;
292
293 /*
294 * Per-iteration state.
295 */
296 typedef void ztest_func_t(ztest_ds_t *zd, uint64_t id);
297
298 typedef struct ztest_info {
299 ztest_func_t *zi_func; /* test function */
300 uint64_t zi_iters; /* iterations per execution */
301 uint64_t *zi_interval; /* execute every <interval> seconds */
302 } ztest_info_t;
303
304 typedef struct ztest_shared_callstate {
305 uint64_t zc_count; /* per-pass count */
306 uint64_t zc_time; /* per-pass time */
307 uint64_t zc_next; /* next time to call this function */
308 } ztest_shared_callstate_t;
309
310 static ztest_shared_callstate_t *ztest_shared_callstate;
311 #define ZTEST_GET_SHARED_CALLSTATE(c) (&ztest_shared_callstate[c])
312
313 /*
314 * Note: these aren't static because we want dladdr() to work.
315 */
316 ztest_func_t ztest_dmu_read_write;
317 ztest_func_t ztest_dmu_write_parallel;
318 ztest_func_t ztest_dmu_object_alloc_free;
319 ztest_func_t ztest_dmu_commit_callbacks;
320 ztest_func_t ztest_zap;
321 ztest_func_t ztest_zap_parallel;
322 ztest_func_t ztest_zil_commit;
323 ztest_func_t ztest_zil_remount;
324 ztest_func_t ztest_dmu_read_write_zcopy;
325 ztest_func_t ztest_dmu_objset_create_destroy;
326 ztest_func_t ztest_dmu_prealloc;
327 ztest_func_t ztest_fzap;
328 ztest_func_t ztest_dmu_snapshot_create_destroy;
329 ztest_func_t ztest_dsl_prop_get_set;
330 ztest_func_t ztest_spa_prop_get_set;
331 ztest_func_t ztest_spa_create_destroy;
332 ztest_func_t ztest_fault_inject;
333 ztest_func_t ztest_ddt_repair;
334 ztest_func_t ztest_dmu_snapshot_hold;
335 ztest_func_t ztest_spa_rename;
336 ztest_func_t ztest_scrub;
337 ztest_func_t ztest_dsl_dataset_promote_busy;
338 ztest_func_t ztest_vdev_attach_detach;
339 ztest_func_t ztest_vdev_LUN_growth;
340 ztest_func_t ztest_vdev_add_remove;
341 ztest_func_t ztest_vdev_aux_add_remove;
342 ztest_func_t ztest_split_pool;
343 ztest_func_t ztest_reguid;
344 ztest_func_t ztest_spa_upgrade;
345 ztest_func_t ztest_device_removal;
346 ztest_func_t ztest_remap_blocks;
347
348 uint64_t zopt_always = 0ULL * NANOSEC; /* all the time */
349 uint64_t zopt_incessant = 1ULL * NANOSEC / 10; /* every 1/10 second */
350 uint64_t zopt_often = 1ULL * NANOSEC; /* every second */
351 uint64_t zopt_sometimes = 10ULL * NANOSEC; /* every 10 seconds */
352 uint64_t zopt_rarely = 60ULL * NANOSEC; /* every 60 seconds */
353
354 ztest_info_t ztest_info[] = {
355 { ztest_dmu_read_write, 1, &zopt_always },
356 { ztest_dmu_write_parallel, 10, &zopt_always },
357 { ztest_dmu_object_alloc_free, 1, &zopt_always },
358 { ztest_dmu_commit_callbacks, 1, &zopt_always },
359 { ztest_zap, 30, &zopt_always },
360 { ztest_zap_parallel, 100, &zopt_always },
361 { ztest_split_pool, 1, &zopt_always },
362 { ztest_zil_commit, 1, &zopt_incessant },
363 { ztest_zil_remount, 1, &zopt_sometimes },
364 { ztest_dmu_read_write_zcopy, 1, &zopt_often },
365 { ztest_dmu_objset_create_destroy, 1, &zopt_often },
366 { ztest_dsl_prop_get_set, 1, &zopt_often },
367 { ztest_spa_prop_get_set, 1, &zopt_sometimes },
368 #if 0
369 { ztest_dmu_prealloc, 1, &zopt_sometimes },
370 #endif
371 { ztest_fzap, 1, &zopt_sometimes },
372 { ztest_dmu_snapshot_create_destroy, 1, &zopt_sometimes },
373 { ztest_spa_create_destroy, 1, &zopt_sometimes },
374 { ztest_fault_inject, 1, &zopt_sometimes },
375 { ztest_ddt_repair, 1, &zopt_sometimes },
376 { ztest_dmu_snapshot_hold, 1, &zopt_sometimes },
377 { ztest_reguid, 1, &zopt_rarely },
378 { ztest_spa_rename, 1, &zopt_rarely },
379 { ztest_scrub, 1, &zopt_rarely },
380 { ztest_spa_upgrade, 1, &zopt_rarely },
381 { ztest_dsl_dataset_promote_busy, 1, &zopt_rarely },
382 { ztest_vdev_attach_detach, 1, &zopt_sometimes },
383 { ztest_vdev_LUN_growth, 1, &zopt_rarely },
384 { ztest_vdev_add_remove, 1,
385 &ztest_opts.zo_vdevtime },
386 { ztest_vdev_aux_add_remove, 1,
387 &ztest_opts.zo_vdevtime },
388 { ztest_device_removal, 1, &zopt_sometimes },
389 { ztest_remap_blocks, 1, &zopt_sometimes }
390 };
391
392 #define ZTEST_FUNCS (sizeof (ztest_info) / sizeof (ztest_info_t))
393
394 /*
395 * The following struct is used to hold a list of uncalled commit callbacks.
396 * The callbacks are ordered by txg number.
397 */
398 typedef struct ztest_cb_list {
399 kmutex_t zcl_callbacks_lock;
400 list_t zcl_callbacks;
401 } ztest_cb_list_t;
402
403 /*
404 * Stuff we need to share writably between parent and child.
405 */
406 typedef struct ztest_shared {
407 boolean_t zs_do_init;
408 hrtime_t zs_proc_start;
409 hrtime_t zs_proc_stop;
410 hrtime_t zs_thread_start;
411 hrtime_t zs_thread_stop;
412 hrtime_t zs_thread_kill;
413 uint64_t zs_enospc_count;
414 uint64_t zs_vdev_next_leaf;
415 uint64_t zs_vdev_aux;
416 uint64_t zs_alloc;
417 uint64_t zs_space;
418 uint64_t zs_splits;
419 uint64_t zs_mirrors;
420 uint64_t zs_metaslab_sz;
421 uint64_t zs_metaslab_df_alloc_threshold;
422 uint64_t zs_guid;
423 } ztest_shared_t;
424
425 #define ID_PARALLEL -1ULL
426
427 static char ztest_dev_template[] = "%s/%s.%llua";
428 static char ztest_aux_template[] = "%s/%s.%s.%llu";
429 ztest_shared_t *ztest_shared;
430
431 static spa_t *ztest_spa = NULL;
432 static ztest_ds_t *ztest_ds;
433
434 static kmutex_t ztest_vdev_lock;
435
436 /*
437 * The ztest_name_lock protects the pool and dataset namespace used by
438 * the individual tests. To modify the namespace, consumers must grab
439 * this lock as writer. Grabbing the lock as reader will ensure that the
440 * namespace does not change while the lock is held.
441 */
442 static krwlock_t ztest_name_lock;
443
444 static boolean_t ztest_dump_core = B_TRUE;
445 static boolean_t ztest_exiting;
446
447 /* Global commit callback list */
448 static ztest_cb_list_t zcl;
449
450 enum ztest_object {
451 ZTEST_META_DNODE = 0,
452 ZTEST_DIROBJ,
453 ZTEST_OBJECTS
454 };
455
456 static void usage(boolean_t) __NORETURN;
457
458 /*
459 * These libumem hooks provide a reasonable set of defaults for the allocator's
460 * debugging facilities.
461 */
462 const char *
463 _umem_debug_init()
464 {
465 return ("default,verbose"); /* $UMEM_DEBUG setting */
466 }
467
468 const char *
469 _umem_logging_init(void)
470 {
471 return ("fail,contents"); /* $UMEM_LOGGING setting */
472 }
473
474 #define FATAL_MSG_SZ 1024
475
476 char *fatal_msg;
477
478 static void
479 fatal(int do_perror, char *message, ...)
480 {
481 va_list args;
482 int save_errno = errno;
483 char buf[FATAL_MSG_SZ];
484
485 (void) fflush(stdout);
486
487 va_start(args, message);
488 (void) sprintf(buf, "ztest: ");
489 /* LINTED */
490 (void) vsprintf(buf + strlen(buf), message, args);
491 va_end(args);
492 if (do_perror) {
493 (void) snprintf(buf + strlen(buf), FATAL_MSG_SZ - strlen(buf),
494 ": %s", strerror(save_errno));
495 }
496 (void) fprintf(stderr, "%s\n", buf);
497 fatal_msg = buf; /* to ease debugging */
498 if (ztest_dump_core)
499 abort();
500 exit(3);
501 }
502
503 static int
504 str2shift(const char *buf)
505 {
506 const char *ends = "BKMGTPEZ";
507 int i;
508
509 if (buf[0] == '\0')
510 return (0);
511 for (i = 0; i < strlen(ends); i++) {
512 if (toupper(buf[0]) == ends[i])
513 break;
514 }
515 if (i == strlen(ends)) {
516 (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n",
517 buf);
518 usage(B_FALSE);
519 }
520 if (buf[1] == '\0' || (toupper(buf[1]) == 'B' && buf[2] == '\0')) {
521 return (10*i);
522 }
523 (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n", buf);
524 usage(B_FALSE);
525 /* NOTREACHED */
526 }
527
528 static uint64_t
529 nicenumtoull(const char *buf)
530 {
531 char *end;
532 uint64_t val;
533
534 val = strtoull(buf, &end, 0);
535 if (end == buf) {
536 (void) fprintf(stderr, "ztest: bad numeric value: %s\n", buf);
537 usage(B_FALSE);
538 } else if (end[0] == '.') {
539 double fval = strtod(buf, &end);
540 fval *= pow(2, str2shift(end));
541 if (fval > UINT64_MAX) {
542 (void) fprintf(stderr, "ztest: value too large: %s\n",
543 buf);
544 usage(B_FALSE);
545 }
546 val = (uint64_t)fval;
547 } else {
548 int shift = str2shift(end);
549 if (shift >= 64 || (val << shift) >> shift != val) {
550 (void) fprintf(stderr, "ztest: value too large: %s\n",
551 buf);
552 usage(B_FALSE);
553 }
554 val <<= shift;
555 }
556 return (val);
557 }
558
559 static void
560 usage(boolean_t requested)
561 {
562 const ztest_shared_opts_t *zo = &ztest_opts_defaults;
563
564 char nice_vdev_size[NN_NUMBUF_SZ];
565 char nice_gang_bang[NN_NUMBUF_SZ];
566 FILE *fp = requested ? stdout : stderr;
567
568 nicenum(zo->zo_vdev_size, nice_vdev_size, sizeof (nice_vdev_size));
569 nicenum(zo->zo_metaslab_gang_bang, nice_gang_bang,
570 sizeof (nice_gang_bang));
571
572 (void) fprintf(fp, "Usage: %s\n"
573 "\t[-v vdevs (default: %llu)]\n"
574 "\t[-s size_of_each_vdev (default: %s)]\n"
575 "\t[-a alignment_shift (default: %d)] use 0 for random\n"
576 "\t[-m mirror_copies (default: %d)]\n"
577 "\t[-r raidz_disks (default: %d)]\n"
578 "\t[-R raidz_parity (default: %d)]\n"
579 "\t[-d datasets (default: %d)]\n"
580 "\t[-t threads (default: %d)]\n"
581 "\t[-g gang_block_threshold (default: %s)]\n"
582 "\t[-i init_count (default: %d)] initialize pool i times\n"
583 "\t[-k kill_percentage (default: %llu%%)]\n"
584 "\t[-p pool_name (default: %s)]\n"
585 "\t[-f dir (default: %s)] file directory for vdev files\n"
586 "\t[-V] verbose (use multiple times for ever more blather)\n"
587 "\t[-E] use existing pool instead of creating new one\n"
588 "\t[-T time (default: %llu sec)] total run time\n"
589 "\t[-F freezeloops (default: %llu)] max loops in spa_freeze()\n"
590 "\t[-P passtime (default: %llu sec)] time per pass\n"
591 "\t[-B alt_ztest (default: <none>)] alternate ztest path\n"
592 "\t[-o variable=value] ... set global variable to an unsigned\n"
593 "\t 32-bit integer value\n"
594 "\t[-h] (print help)\n"
595 "",
596 zo->zo_pool,
597 (u_longlong_t)zo->zo_vdevs, /* -v */
598 nice_vdev_size, /* -s */
599 zo->zo_ashift, /* -a */
600 zo->zo_mirrors, /* -m */
601 zo->zo_raidz, /* -r */
602 zo->zo_raidz_parity, /* -R */
603 zo->zo_datasets, /* -d */
604 zo->zo_threads, /* -t */
605 nice_gang_bang, /* -g */
606 zo->zo_init, /* -i */
607 (u_longlong_t)zo->zo_killrate, /* -k */
608 zo->zo_pool, /* -p */
609 zo->zo_dir, /* -f */
610 (u_longlong_t)zo->zo_time, /* -T */
611 (u_longlong_t)zo->zo_maxloops, /* -F */
612 (u_longlong_t)zo->zo_passtime);
613 exit(requested ? 0 : 1);
614 }
615
616 static void
617 process_options(int argc, char **argv)
618 {
619 char *path;
620 ztest_shared_opts_t *zo = &ztest_opts;
621
622 int opt;
623 uint64_t value;
624 char altdir[MAXNAMELEN] = { 0 };
625
626 bcopy(&ztest_opts_defaults, zo, sizeof (*zo));
627
628 while ((opt = getopt(argc, argv,
629 "v:s:a:m:r:R:d:t:g:i:k:p:f:VET:P:hF:B:o:")) != EOF) {
630 value = 0;
631 switch (opt) {
632 case 'v':
633 case 's':
634 case 'a':
635 case 'm':
636 case 'r':
637 case 'R':
638 case 'd':
639 case 't':
640 case 'g':
641 case 'i':
642 case 'k':
643 case 'T':
644 case 'P':
645 case 'F':
646 value = nicenumtoull(optarg);
647 }
648 switch (opt) {
649 case 'v':
650 zo->zo_vdevs = value;
651 break;
652 case 's':
653 zo->zo_vdev_size = MAX(SPA_MINDEVSIZE, value);
654 break;
655 case 'a':
656 zo->zo_ashift = value;
657 break;
658 case 'm':
659 zo->zo_mirrors = value;
660 break;
661 case 'r':
662 zo->zo_raidz = MAX(1, value);
663 break;
664 case 'R':
665 zo->zo_raidz_parity = MIN(MAX(value, 1), 3);
666 break;
667 case 'd':
668 zo->zo_datasets = MAX(1, value);
669 break;
670 case 't':
671 zo->zo_threads = MAX(1, value);
672 break;
673 case 'g':
674 zo->zo_metaslab_gang_bang = MAX(SPA_MINBLOCKSIZE << 1,
675 value);
676 break;
677 case 'i':
678 zo->zo_init = value;
679 break;
680 case 'k':
681 zo->zo_killrate = value;
682 break;
683 case 'p':
684 (void) strlcpy(zo->zo_pool, optarg,
685 sizeof (zo->zo_pool));
686 break;
687 case 'f':
688 path = realpath(optarg, NULL);
689 if (path == NULL) {
690 (void) fprintf(stderr, "error: %s: %s\n",
691 optarg, strerror(errno));
692 usage(B_FALSE);
693 } else {
694 (void) strlcpy(zo->zo_dir, path,
695 sizeof (zo->zo_dir));
696 }
697 break;
698 case 'V':
699 zo->zo_verbose++;
700 break;
701 case 'E':
702 zo->zo_init = 0;
703 break;
704 case 'T':
705 zo->zo_time = value;
706 break;
707 case 'P':
708 zo->zo_passtime = MAX(1, value);
709 break;
710 case 'F':
711 zo->zo_maxloops = MAX(1, value);
712 break;
713 case 'B':
714 (void) strlcpy(altdir, optarg, sizeof (altdir));
715 break;
716 case 'o':
717 if (set_global_var(optarg) != 0)
718 usage(B_FALSE);
719 break;
720 case 'h':
721 usage(B_TRUE);
722 break;
723 case '?':
724 default:
725 usage(B_FALSE);
726 break;
727 }
728 }
729
730 zo->zo_raidz_parity = MIN(zo->zo_raidz_parity, zo->zo_raidz - 1);
731
732 zo->zo_vdevtime =
733 (zo->zo_vdevs > 0 ? zo->zo_time * NANOSEC / zo->zo_vdevs :
734 UINT64_MAX >> 2);
735
736 if (strlen(altdir) > 0) {
737 char *cmd;
738 char *realaltdir;
739 char *bin;
740 char *ztest;
741 char *isa;
742 int isalen;
743
744 cmd = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
745 realaltdir = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
746
747 VERIFY(NULL != realpath(getexecname(), cmd));
748 if (0 != access(altdir, F_OK)) {
749 ztest_dump_core = B_FALSE;
750 fatal(B_TRUE, "invalid alternate ztest path: %s",
751 altdir);
752 }
753 VERIFY(NULL != realpath(altdir, realaltdir));
754
755 /*
756 * 'cmd' should be of the form "<anything>/usr/bin/<isa>/ztest".
757 * We want to extract <isa> to determine if we should use
758 * 32 or 64 bit binaries.
759 */
760 bin = strstr(cmd, "/usr/bin/");
761 ztest = strstr(bin, "/ztest");
762 isa = bin + 9;
763 isalen = ztest - isa;
764 (void) snprintf(zo->zo_alt_ztest, sizeof (zo->zo_alt_ztest),
765 "%s/usr/bin/%.*s/ztest", realaltdir, isalen, isa);
766 (void) snprintf(zo->zo_alt_libpath, sizeof (zo->zo_alt_libpath),
767 "%s/usr/lib/%.*s", realaltdir, isalen, isa);
768
769 if (0 != access(zo->zo_alt_ztest, X_OK)) {
770 ztest_dump_core = B_FALSE;
771 fatal(B_TRUE, "invalid alternate ztest: %s",
772 zo->zo_alt_ztest);
773 } else if (0 != access(zo->zo_alt_libpath, X_OK)) {
774 ztest_dump_core = B_FALSE;
775 fatal(B_TRUE, "invalid alternate lib directory %s",
776 zo->zo_alt_libpath);
777 }
778
779 umem_free(cmd, MAXPATHLEN);
780 umem_free(realaltdir, MAXPATHLEN);
781 }
782 }
783
784 static void
785 ztest_kill(ztest_shared_t *zs)
786 {
787 zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(ztest_spa));
788 zs->zs_space = metaslab_class_get_space(spa_normal_class(ztest_spa));
789
790 /*
791 * Before we kill off ztest, make sure that the config is updated.
792 * See comment above spa_write_cachefile().
793 */
794 mutex_enter(&spa_namespace_lock);
795 spa_write_cachefile(ztest_spa, B_FALSE, B_FALSE);
796 mutex_exit(&spa_namespace_lock);
797
798 zfs_dbgmsg_print(FTAG);
799 (void) kill(getpid(), SIGKILL);
800 }
801
802 static uint64_t
803 ztest_random(uint64_t range)
804 {
805 uint64_t r;
806
807 ASSERT3S(ztest_fd_rand, >=, 0);
808
809 if (range == 0)
810 return (0);
811
812 if (read(ztest_fd_rand, &r, sizeof (r)) != sizeof (r))
813 fatal(1, "short read from /dev/urandom");
814
815 return (r % range);
816 }
817
818 /* ARGSUSED */
819 static void
820 ztest_record_enospc(const char *s)
821 {
822 ztest_shared->zs_enospc_count++;
823 }
824
825 static uint64_t
826 ztest_get_ashift(void)
827 {
828 if (ztest_opts.zo_ashift == 0)
829 return (SPA_MINBLOCKSHIFT + ztest_random(5));
830 return (ztest_opts.zo_ashift);
831 }
832
833 static nvlist_t *
834 make_vdev_file(char *path, char *aux, char *pool, size_t size, uint64_t ashift)
835 {
836 char pathbuf[MAXPATHLEN];
837 uint64_t vdev;
838 nvlist_t *file;
839
840 if (ashift == 0)
841 ashift = ztest_get_ashift();
842
843 if (path == NULL) {
844 path = pathbuf;
845
846 if (aux != NULL) {
847 vdev = ztest_shared->zs_vdev_aux;
848 (void) snprintf(path, sizeof (pathbuf),
849 ztest_aux_template, ztest_opts.zo_dir,
850 pool == NULL ? ztest_opts.zo_pool : pool,
851 aux, vdev);
852 } else {
853 vdev = ztest_shared->zs_vdev_next_leaf++;
854 (void) snprintf(path, sizeof (pathbuf),
855 ztest_dev_template, ztest_opts.zo_dir,
856 pool == NULL ? ztest_opts.zo_pool : pool, vdev);
857 }
858 }
859
860 if (size != 0) {
861 int fd = open(path, O_RDWR | O_CREAT | O_TRUNC, 0666);
862 if (fd == -1)
863 fatal(1, "can't open %s", path);
864 if (ftruncate(fd, size) != 0)
865 fatal(1, "can't ftruncate %s", path);
866 (void) close(fd);
867 }
868
869 VERIFY(nvlist_alloc(&file, NV_UNIQUE_NAME, 0) == 0);
870 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_TYPE, VDEV_TYPE_FILE) == 0);
871 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_PATH, path) == 0);
872 VERIFY(nvlist_add_uint64(file, ZPOOL_CONFIG_ASHIFT, ashift) == 0);
873
874 return (file);
875 }
876
877 static nvlist_t *
878 make_vdev_raidz(char *path, char *aux, char *pool, size_t size,
879 uint64_t ashift, int r)
880 {
881 nvlist_t *raidz, **child;
882 int c;
883
884 if (r < 2)
885 return (make_vdev_file(path, aux, pool, size, ashift));
886 child = umem_alloc(r * sizeof (nvlist_t *), UMEM_NOFAIL);
887
888 for (c = 0; c < r; c++)
889 child[c] = make_vdev_file(path, aux, pool, size, ashift);
890
891 VERIFY(nvlist_alloc(&raidz, NV_UNIQUE_NAME, 0) == 0);
892 VERIFY(nvlist_add_string(raidz, ZPOOL_CONFIG_TYPE,
893 VDEV_TYPE_RAIDZ) == 0);
894 VERIFY(nvlist_add_uint64(raidz, ZPOOL_CONFIG_NPARITY,
895 ztest_opts.zo_raidz_parity) == 0);
896 VERIFY(nvlist_add_nvlist_array(raidz, ZPOOL_CONFIG_CHILDREN,
897 child, r) == 0);
898
899 for (c = 0; c < r; c++)
900 nvlist_free(child[c]);
901
902 umem_free(child, r * sizeof (nvlist_t *));
903
904 return (raidz);
905 }
906
907 static nvlist_t *
908 make_vdev_mirror(char *path, char *aux, char *pool, size_t size,
909 uint64_t ashift, int r, int m)
910 {
911 nvlist_t *mirror, **child;
912 int c;
913
914 if (m < 1)
915 return (make_vdev_raidz(path, aux, pool, size, ashift, r));
916
917 child = umem_alloc(m * sizeof (nvlist_t *), UMEM_NOFAIL);
918
919 for (c = 0; c < m; c++)
920 child[c] = make_vdev_raidz(path, aux, pool, size, ashift, r);
921
922 VERIFY(nvlist_alloc(&mirror, NV_UNIQUE_NAME, 0) == 0);
923 VERIFY(nvlist_add_string(mirror, ZPOOL_CONFIG_TYPE,
924 VDEV_TYPE_MIRROR) == 0);
925 VERIFY(nvlist_add_nvlist_array(mirror, ZPOOL_CONFIG_CHILDREN,
926 child, m) == 0);
927
928 for (c = 0; c < m; c++)
929 nvlist_free(child[c]);
930
931 umem_free(child, m * sizeof (nvlist_t *));
932
933 return (mirror);
934 }
935
936 static nvlist_t *
937 make_vdev_root(char *path, char *aux, char *pool, size_t size, uint64_t ashift,
938 int log, int r, int m, int t)
939 {
940 nvlist_t *root, **child;
941 int c;
942
943 ASSERT(t > 0);
944
945 child = umem_alloc(t * sizeof (nvlist_t *), UMEM_NOFAIL);
946
947 for (c = 0; c < t; c++) {
948 child[c] = make_vdev_mirror(path, aux, pool, size, ashift,
949 r, m);
950 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_IS_LOG,
951 log) == 0);
952 }
953
954 VERIFY(nvlist_alloc(&root, NV_UNIQUE_NAME, 0) == 0);
955 VERIFY(nvlist_add_string(root, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) == 0);
956 VERIFY(nvlist_add_nvlist_array(root, aux ? aux : ZPOOL_CONFIG_CHILDREN,
957 child, t) == 0);
958
959 for (c = 0; c < t; c++)
960 nvlist_free(child[c]);
961
962 umem_free(child, t * sizeof (nvlist_t *));
963
964 return (root);
965 }
966
967 /*
968 * Find a random spa version. Returns back a random spa version in the
969 * range [initial_version, SPA_VERSION_FEATURES].
970 */
971 static uint64_t
972 ztest_random_spa_version(uint64_t initial_version)
973 {
974 uint64_t version = initial_version;
975
976 if (version <= SPA_VERSION_BEFORE_FEATURES) {
977 version = version +
978 ztest_random(SPA_VERSION_BEFORE_FEATURES - version + 1);
979 }
980
981 if (version > SPA_VERSION_BEFORE_FEATURES)
982 version = SPA_VERSION_FEATURES;
983
984 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
985 return (version);
986 }
987
988 static int
989 ztest_random_blocksize(void)
990 {
991 uint64_t block_shift;
992 /*
993 * Choose a block size >= the ashift.
994 * If the SPA supports new MAXBLOCKSIZE, test up to 1MB blocks.
995 */
996 int maxbs = SPA_OLD_MAXBLOCKSHIFT;
997 if (spa_maxblocksize(ztest_spa) == SPA_MAXBLOCKSIZE)
998 maxbs = 20;
999 block_shift = ztest_random(maxbs - ztest_spa->spa_max_ashift + 1);
1000 return (1 << (SPA_MINBLOCKSHIFT + block_shift));
1001 }
1002
1003 static int
1004 ztest_random_ibshift(void)
1005 {
1006 return (DN_MIN_INDBLKSHIFT +
1007 ztest_random(DN_MAX_INDBLKSHIFT - DN_MIN_INDBLKSHIFT + 1));
1008 }
1009
1010 static uint64_t
1011 ztest_random_vdev_top(spa_t *spa, boolean_t log_ok)
1012 {
1013 uint64_t top;
1014 vdev_t *rvd = spa->spa_root_vdev;
1015 vdev_t *tvd;
1016
1017 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1018
1019 do {
1020 top = ztest_random(rvd->vdev_children);
1021 tvd = rvd->vdev_child[top];
1022 } while (!vdev_is_concrete(tvd) || (tvd->vdev_islog && !log_ok) ||
1023 tvd->vdev_mg == NULL || tvd->vdev_mg->mg_class == NULL);
1024
1025 return (top);
1026 }
1027
1028 static uint64_t
1029 ztest_random_dsl_prop(zfs_prop_t prop)
1030 {
1031 uint64_t value;
1032
1033 do {
1034 value = zfs_prop_random_value(prop, ztest_random(-1ULL));
1035 } while (prop == ZFS_PROP_CHECKSUM && value == ZIO_CHECKSUM_OFF);
1036
1037 return (value);
1038 }
1039
1040 static int
1041 ztest_dsl_prop_set_uint64(char *osname, zfs_prop_t prop, uint64_t value,
1042 boolean_t inherit)
1043 {
1044 const char *propname = zfs_prop_to_name(prop);
1045 const char *valname;
1046 char setpoint[MAXPATHLEN];
1047 uint64_t curval;
1048 int error;
1049
1050 error = dsl_prop_set_int(osname, propname,
1051 (inherit ? ZPROP_SRC_NONE : ZPROP_SRC_LOCAL), value);
1052
1053 if (error == ENOSPC) {
1054 ztest_record_enospc(FTAG);
1055 return (error);
1056 }
1057 ASSERT0(error);
1058
1059 VERIFY0(dsl_prop_get_integer(osname, propname, &curval, setpoint));
1060
1061 if (ztest_opts.zo_verbose >= 6) {
1062 VERIFY(zfs_prop_index_to_string(prop, curval, &valname) == 0);
1063 (void) printf("%s %s = %s at '%s'\n",
1064 osname, propname, valname, setpoint);
1065 }
1066
1067 return (error);
1068 }
1069
1070 static int
1071 ztest_spa_prop_set_uint64(zpool_prop_t prop, uint64_t value)
1072 {
1073 spa_t *spa = ztest_spa;
1074 nvlist_t *props = NULL;
1075 int error;
1076
1077 VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
1078 VERIFY(nvlist_add_uint64(props, zpool_prop_to_name(prop), value) == 0);
1079
1080 error = spa_prop_set(spa, props);
1081
1082 nvlist_free(props);
1083
1084 if (error == ENOSPC) {
1085 ztest_record_enospc(FTAG);
1086 return (error);
1087 }
1088 ASSERT0(error);
1089
1090 return (error);
1091 }
1092
1093 static void
1094 ztest_rll_init(rll_t *rll)
1095 {
1096 rll->rll_writer = NULL;
1097 rll->rll_readers = 0;
1098 mutex_init(&rll->rll_lock, NULL, USYNC_THREAD, NULL);
1099 cv_init(&rll->rll_cv, NULL, USYNC_THREAD, NULL);
1100 }
1101
1102 static void
1103 ztest_rll_destroy(rll_t *rll)
1104 {
1105 ASSERT(rll->rll_writer == NULL);
1106 ASSERT(rll->rll_readers == 0);
1107 mutex_destroy(&rll->rll_lock);
1108 cv_destroy(&rll->rll_cv);
1109 }
1110
1111 static void
1112 ztest_rll_lock(rll_t *rll, rl_type_t type)
1113 {
1114 mutex_enter(&rll->rll_lock);
1115
1116 if (type == RL_READER) {
1117 while (rll->rll_writer != NULL)
1118 cv_wait(&rll->rll_cv, &rll->rll_lock);
1119 rll->rll_readers++;
1120 } else {
1121 while (rll->rll_writer != NULL || rll->rll_readers)
1122 cv_wait(&rll->rll_cv, &rll->rll_lock);
1123 rll->rll_writer = curthread;
1124 }
1125
1126 mutex_exit(&rll->rll_lock);
1127 }
1128
1129 static void
1130 ztest_rll_unlock(rll_t *rll)
1131 {
1132 mutex_enter(&rll->rll_lock);
1133
1134 if (rll->rll_writer) {
1135 ASSERT(rll->rll_readers == 0);
1136 rll->rll_writer = NULL;
1137 } else {
1138 ASSERT(rll->rll_readers != 0);
1139 ASSERT(rll->rll_writer == NULL);
1140 rll->rll_readers--;
1141 }
1142
1143 if (rll->rll_writer == NULL && rll->rll_readers == 0)
1144 cv_broadcast(&rll->rll_cv);
1145
1146 mutex_exit(&rll->rll_lock);
1147 }
1148
1149 static void
1150 ztest_object_lock(ztest_ds_t *zd, uint64_t object, rl_type_t type)
1151 {
1152 rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1153
1154 ztest_rll_lock(rll, type);
1155 }
1156
1157 static void
1158 ztest_object_unlock(ztest_ds_t *zd, uint64_t object)
1159 {
1160 rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1161
1162 ztest_rll_unlock(rll);
1163 }
1164
1165 static rl_t *
1166 ztest_range_lock(ztest_ds_t *zd, uint64_t object, uint64_t offset,
1167 uint64_t size, rl_type_t type)
1168 {
1169 uint64_t hash = object ^ (offset % (ZTEST_RANGE_LOCKS + 1));
1170 rll_t *rll = &zd->zd_range_lock[hash & (ZTEST_RANGE_LOCKS - 1)];
1171 rl_t *rl;
1172
1173 rl = umem_alloc(sizeof (*rl), UMEM_NOFAIL);
1174 rl->rl_object = object;
1175 rl->rl_offset = offset;
1176 rl->rl_size = size;
1177 rl->rl_lock = rll;
1178
1179 ztest_rll_lock(rll, type);
1180
1181 return (rl);
1182 }
1183
1184 static void
1185 ztest_range_unlock(rl_t *rl)
1186 {
1187 rll_t *rll = rl->rl_lock;
1188
1189 ztest_rll_unlock(rll);
1190
1191 umem_free(rl, sizeof (*rl));
1192 }
1193
1194 static void
1195 ztest_zd_init(ztest_ds_t *zd, ztest_shared_ds_t *szd, objset_t *os)
1196 {
1197 zd->zd_os = os;
1198 zd->zd_zilog = dmu_objset_zil(os);
1199 zd->zd_shared = szd;
1200 dmu_objset_name(os, zd->zd_name);
1201
1202 if (zd->zd_shared != NULL)
1203 zd->zd_shared->zd_seq = 0;
1204
1205 rw_init(&zd->zd_zilog_lock, NULL, USYNC_THREAD, NULL);
1206 mutex_init(&zd->zd_dirobj_lock, NULL, USYNC_THREAD, NULL);
1207
1208 for (int l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1209 ztest_rll_init(&zd->zd_object_lock[l]);
1210
1211 for (int l = 0; l < ZTEST_RANGE_LOCKS; l++)
1212 ztest_rll_init(&zd->zd_range_lock[l]);
1213 }
1214
1215 static void
1216 ztest_zd_fini(ztest_ds_t *zd)
1217 {
1218 mutex_destroy(&zd->zd_dirobj_lock);
1219
1220 for (int l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1221 ztest_rll_destroy(&zd->zd_object_lock[l]);
1222
1223 for (int l = 0; l < ZTEST_RANGE_LOCKS; l++)
1224 ztest_rll_destroy(&zd->zd_range_lock[l]);
1225 }
1226
1227 #define TXG_MIGHTWAIT (ztest_random(10) == 0 ? TXG_NOWAIT : TXG_WAIT)
1228
1229 static uint64_t
1230 ztest_tx_assign(dmu_tx_t *tx, uint64_t txg_how, const char *tag)
1231 {
1232 uint64_t txg;
1233 int error;
1234
1235 /*
1236 * Attempt to assign tx to some transaction group.
1237 */
1238 error = dmu_tx_assign(tx, txg_how);
1239 if (error) {
1240 if (error == ERESTART) {
1241 ASSERT(txg_how == TXG_NOWAIT);
1242 dmu_tx_wait(tx);
1243 } else {
1244 ASSERT3U(error, ==, ENOSPC);
1245 ztest_record_enospc(tag);
1246 }
1247 dmu_tx_abort(tx);
1248 return (0);
1249 }
1250 txg = dmu_tx_get_txg(tx);
1251 ASSERT(txg != 0);
1252 return (txg);
1253 }
1254
1255 static void
1256 ztest_pattern_set(void *buf, uint64_t size, uint64_t value)
1257 {
1258 uint64_t *ip = buf;
1259 uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1260
1261 while (ip < ip_end)
1262 *ip++ = value;
1263 }
1264
1265 static boolean_t
1266 ztest_pattern_match(void *buf, uint64_t size, uint64_t value)
1267 {
1268 uint64_t *ip = buf;
1269 uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1270 uint64_t diff = 0;
1271
1272 while (ip < ip_end)
1273 diff |= (value - *ip++);
1274
1275 return (diff == 0);
1276 }
1277
1278 static void
1279 ztest_bt_generate(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1280 uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg)
1281 {
1282 bt->bt_magic = BT_MAGIC;
1283 bt->bt_objset = dmu_objset_id(os);
1284 bt->bt_object = object;
1285 bt->bt_offset = offset;
1286 bt->bt_gen = gen;
1287 bt->bt_txg = txg;
1288 bt->bt_crtxg = crtxg;
1289 }
1290
1291 static void
1292 ztest_bt_verify(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1293 uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg)
1294 {
1295 ASSERT3U(bt->bt_magic, ==, BT_MAGIC);
1296 ASSERT3U(bt->bt_objset, ==, dmu_objset_id(os));
1297 ASSERT3U(bt->bt_object, ==, object);
1298 ASSERT3U(bt->bt_offset, ==, offset);
1299 ASSERT3U(bt->bt_gen, <=, gen);
1300 ASSERT3U(bt->bt_txg, <=, txg);
1301 ASSERT3U(bt->bt_crtxg, ==, crtxg);
1302 }
1303
1304 static ztest_block_tag_t *
1305 ztest_bt_bonus(dmu_buf_t *db)
1306 {
1307 dmu_object_info_t doi;
1308 ztest_block_tag_t *bt;
1309
1310 dmu_object_info_from_db(db, &doi);
1311 ASSERT3U(doi.doi_bonus_size, <=, db->db_size);
1312 ASSERT3U(doi.doi_bonus_size, >=, sizeof (*bt));
1313 bt = (void *)((char *)db->db_data + doi.doi_bonus_size - sizeof (*bt));
1314
1315 return (bt);
1316 }
1317
1318 /*
1319 * ZIL logging ops
1320 */
1321
1322 #define lrz_type lr_mode
1323 #define lrz_blocksize lr_uid
1324 #define lrz_ibshift lr_gid
1325 #define lrz_bonustype lr_rdev
1326 #define lrz_bonuslen lr_crtime[1]
1327
1328 static void
1329 ztest_log_create(ztest_ds_t *zd, dmu_tx_t *tx, lr_create_t *lr)
1330 {
1331 char *name = (void *)(lr + 1); /* name follows lr */
1332 size_t namesize = strlen(name) + 1;
1333 itx_t *itx;
1334
1335 if (zil_replaying(zd->zd_zilog, tx))
1336 return;
1337
1338 itx = zil_itx_create(TX_CREATE, sizeof (*lr) + namesize);
1339 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1340 sizeof (*lr) + namesize - sizeof (lr_t));
1341
1342 zil_itx_assign(zd->zd_zilog, itx, tx);
1343 }
1344
1345 static void
1346 ztest_log_remove(ztest_ds_t *zd, dmu_tx_t *tx, lr_remove_t *lr, uint64_t object)
1347 {
1348 char *name = (void *)(lr + 1); /* name follows lr */
1349 size_t namesize = strlen(name) + 1;
1350 itx_t *itx;
1351
1352 if (zil_replaying(zd->zd_zilog, tx))
1353 return;
1354
1355 itx = zil_itx_create(TX_REMOVE, sizeof (*lr) + namesize);
1356 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1357 sizeof (*lr) + namesize - sizeof (lr_t));
1358
1359 itx->itx_oid = object;
1360 zil_itx_assign(zd->zd_zilog, itx, tx);
1361 }
1362
1363 static void
1364 ztest_log_write(ztest_ds_t *zd, dmu_tx_t *tx, lr_write_t *lr)
1365 {
1366 itx_t *itx;
1367 itx_wr_state_t write_state = ztest_random(WR_NUM_STATES);
1368
1369 if (zil_replaying(zd->zd_zilog, tx))
1370 return;
1371
1372 if (lr->lr_length > ZIL_MAX_LOG_DATA)
1373 write_state = WR_INDIRECT;
1374
1375 itx = zil_itx_create(TX_WRITE,
1376 sizeof (*lr) + (write_state == WR_COPIED ? lr->lr_length : 0));
1377
1378 if (write_state == WR_COPIED &&
1379 dmu_read(zd->zd_os, lr->lr_foid, lr->lr_offset, lr->lr_length,
1380 ((lr_write_t *)&itx->itx_lr) + 1, DMU_READ_NO_PREFETCH) != 0) {
1381 zil_itx_destroy(itx);
1382 itx = zil_itx_create(TX_WRITE, sizeof (*lr));
1383 write_state = WR_NEED_COPY;
1384 }
1385 itx->itx_private = zd;
1386 itx->itx_wr_state = write_state;
1387 itx->itx_sync = (ztest_random(8) == 0);
1388
1389 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1390 sizeof (*lr) - sizeof (lr_t));
1391
1392 zil_itx_assign(zd->zd_zilog, itx, tx);
1393 }
1394
1395 static void
1396 ztest_log_truncate(ztest_ds_t *zd, dmu_tx_t *tx, lr_truncate_t *lr)
1397 {
1398 itx_t *itx;
1399
1400 if (zil_replaying(zd->zd_zilog, tx))
1401 return;
1402
1403 itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr));
1404 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1405 sizeof (*lr) - sizeof (lr_t));
1406
1407 itx->itx_sync = B_FALSE;
1408 zil_itx_assign(zd->zd_zilog, itx, tx);
1409 }
1410
1411 static void
1412 ztest_log_setattr(ztest_ds_t *zd, dmu_tx_t *tx, lr_setattr_t *lr)
1413 {
1414 itx_t *itx;
1415
1416 if (zil_replaying(zd->zd_zilog, tx))
1417 return;
1418
1419 itx = zil_itx_create(TX_SETATTR, sizeof (*lr));
1420 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1421 sizeof (*lr) - sizeof (lr_t));
1422
1423 itx->itx_sync = B_FALSE;
1424 zil_itx_assign(zd->zd_zilog, itx, tx);
1425 }
1426
1427 /*
1428 * ZIL replay ops
1429 */
1430 static int
1431 ztest_replay_create(void *arg1, void *arg2, boolean_t byteswap)
1432 {
1433 ztest_ds_t *zd = arg1;
1434 lr_create_t *lr = arg2;
1435 char *name = (void *)(lr + 1); /* name follows lr */
1436 objset_t *os = zd->zd_os;
1437 ztest_block_tag_t *bbt;
1438 dmu_buf_t *db;
1439 dmu_tx_t *tx;
1440 uint64_t txg;
1441 int error = 0;
1442
1443 if (byteswap)
1444 byteswap_uint64_array(lr, sizeof (*lr));
1445
1446 ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1447 ASSERT(name[0] != '\0');
1448
1449 tx = dmu_tx_create(os);
1450
1451 dmu_tx_hold_zap(tx, lr->lr_doid, B_TRUE, name);
1452
1453 if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1454 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, B_TRUE, NULL);
1455 } else {
1456 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1457 }
1458
1459 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1460 if (txg == 0)
1461 return (ENOSPC);
1462
1463 ASSERT(dmu_objset_zil(os)->zl_replay == !!lr->lr_foid);
1464
1465 if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1466 if (lr->lr_foid == 0) {
1467 lr->lr_foid = zap_create(os,
1468 lr->lrz_type, lr->lrz_bonustype,
1469 lr->lrz_bonuslen, tx);
1470 } else {
1471 error = zap_create_claim(os, lr->lr_foid,
1472 lr->lrz_type, lr->lrz_bonustype,
1473 lr->lrz_bonuslen, tx);
1474 }
1475 } else {
1476 if (lr->lr_foid == 0) {
1477 lr->lr_foid = dmu_object_alloc(os,
1478 lr->lrz_type, 0, lr->lrz_bonustype,
1479 lr->lrz_bonuslen, tx);
1480 } else {
1481 error = dmu_object_claim(os, lr->lr_foid,
1482 lr->lrz_type, 0, lr->lrz_bonustype,
1483 lr->lrz_bonuslen, tx);
1484 }
1485 }
1486
1487 if (error) {
1488 ASSERT3U(error, ==, EEXIST);
1489 ASSERT(zd->zd_zilog->zl_replay);
1490 dmu_tx_commit(tx);
1491 return (error);
1492 }
1493
1494 ASSERT(lr->lr_foid != 0);
1495
1496 if (lr->lrz_type != DMU_OT_ZAP_OTHER)
1497 VERIFY3U(0, ==, dmu_object_set_blocksize(os, lr->lr_foid,
1498 lr->lrz_blocksize, lr->lrz_ibshift, tx));
1499
1500 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1501 bbt = ztest_bt_bonus(db);
1502 dmu_buf_will_dirty(db, tx);
1503 ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_gen, txg, txg);
1504 dmu_buf_rele(db, FTAG);
1505
1506 VERIFY3U(0, ==, zap_add(os, lr->lr_doid, name, sizeof (uint64_t), 1,
1507 &lr->lr_foid, tx));
1508
1509 (void) ztest_log_create(zd, tx, lr);
1510
1511 dmu_tx_commit(tx);
1512
1513 return (0);
1514 }
1515
1516 static int
1517 ztest_replay_remove(void *arg1, void *arg2, boolean_t byteswap)
1518 {
1519 ztest_ds_t *zd = arg1;
1520 lr_remove_t *lr = arg2;
1521 char *name = (void *)(lr + 1); /* name follows lr */
1522 objset_t *os = zd->zd_os;
1523 dmu_object_info_t doi;
1524 dmu_tx_t *tx;
1525 uint64_t object, txg;
1526
1527 if (byteswap)
1528 byteswap_uint64_array(lr, sizeof (*lr));
1529
1530 ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1531 ASSERT(name[0] != '\0');
1532
1533 VERIFY3U(0, ==,
1534 zap_lookup(os, lr->lr_doid, name, sizeof (object), 1, &object));
1535 ASSERT(object != 0);
1536
1537 ztest_object_lock(zd, object, RL_WRITER);
1538
1539 VERIFY3U(0, ==, dmu_object_info(os, object, &doi));
1540
1541 tx = dmu_tx_create(os);
1542
1543 dmu_tx_hold_zap(tx, lr->lr_doid, B_FALSE, name);
1544 dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END);
1545
1546 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1547 if (txg == 0) {
1548 ztest_object_unlock(zd, object);
1549 return (ENOSPC);
1550 }
1551
1552 if (doi.doi_type == DMU_OT_ZAP_OTHER) {
1553 VERIFY3U(0, ==, zap_destroy(os, object, tx));
1554 } else {
1555 VERIFY3U(0, ==, dmu_object_free(os, object, tx));
1556 }
1557
1558 VERIFY3U(0, ==, zap_remove(os, lr->lr_doid, name, tx));
1559
1560 (void) ztest_log_remove(zd, tx, lr, object);
1561
1562 dmu_tx_commit(tx);
1563
1564 ztest_object_unlock(zd, object);
1565
1566 return (0);
1567 }
1568
1569 static int
1570 ztest_replay_write(void *arg1, void *arg2, boolean_t byteswap)
1571 {
1572 ztest_ds_t *zd = arg1;
1573 lr_write_t *lr = arg2;
1574 objset_t *os = zd->zd_os;
1575 void *data = lr + 1; /* data follows lr */
1576 uint64_t offset, length;
1577 ztest_block_tag_t *bt = data;
1578 ztest_block_tag_t *bbt;
1579 uint64_t gen, txg, lrtxg, crtxg;
1580 dmu_object_info_t doi;
1581 dmu_tx_t *tx;
1582 dmu_buf_t *db;
1583 arc_buf_t *abuf = NULL;
1584 rl_t *rl;
1585
1586 if (byteswap)
1587 byteswap_uint64_array(lr, sizeof (*lr));
1588
1589 offset = lr->lr_offset;
1590 length = lr->lr_length;
1591
1592 /* If it's a dmu_sync() block, write the whole block */
1593 if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
1594 uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
1595 if (length < blocksize) {
1596 offset -= offset % blocksize;
1597 length = blocksize;
1598 }
1599 }
1600
1601 if (bt->bt_magic == BSWAP_64(BT_MAGIC))
1602 byteswap_uint64_array(bt, sizeof (*bt));
1603
1604 if (bt->bt_magic != BT_MAGIC)
1605 bt = NULL;
1606
1607 ztest_object_lock(zd, lr->lr_foid, RL_READER);
1608 rl = ztest_range_lock(zd, lr->lr_foid, offset, length, RL_WRITER);
1609
1610 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1611
1612 dmu_object_info_from_db(db, &doi);
1613
1614 bbt = ztest_bt_bonus(db);
1615 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1616 gen = bbt->bt_gen;
1617 crtxg = bbt->bt_crtxg;
1618 lrtxg = lr->lr_common.lrc_txg;
1619
1620 tx = dmu_tx_create(os);
1621
1622 dmu_tx_hold_write(tx, lr->lr_foid, offset, length);
1623
1624 if (ztest_random(8) == 0 && length == doi.doi_data_block_size &&
1625 P2PHASE(offset, length) == 0)
1626 abuf = dmu_request_arcbuf(db, length);
1627
1628 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1629 if (txg == 0) {
1630 if (abuf != NULL)
1631 dmu_return_arcbuf(abuf);
1632 dmu_buf_rele(db, FTAG);
1633 ztest_range_unlock(rl);
1634 ztest_object_unlock(zd, lr->lr_foid);
1635 return (ENOSPC);
1636 }
1637
1638 if (bt != NULL) {
1639 /*
1640 * Usually, verify the old data before writing new data --
1641 * but not always, because we also want to verify correct
1642 * behavior when the data was not recently read into cache.
1643 */
1644 ASSERT(offset % doi.doi_data_block_size == 0);
1645 if (ztest_random(4) != 0) {
1646 int prefetch = ztest_random(2) ?
1647 DMU_READ_PREFETCH : DMU_READ_NO_PREFETCH;
1648 ztest_block_tag_t rbt;
1649
1650 VERIFY(dmu_read(os, lr->lr_foid, offset,
1651 sizeof (rbt), &rbt, prefetch) == 0);
1652 if (rbt.bt_magic == BT_MAGIC) {
1653 ztest_bt_verify(&rbt, os, lr->lr_foid,
1654 offset, gen, txg, crtxg);
1655 }
1656 }
1657
1658 /*
1659 * Writes can appear to be newer than the bonus buffer because
1660 * the ztest_get_data() callback does a dmu_read() of the
1661 * open-context data, which may be different than the data
1662 * as it was when the write was generated.
1663 */
1664 if (zd->zd_zilog->zl_replay) {
1665 ztest_bt_verify(bt, os, lr->lr_foid, offset,
1666 MAX(gen, bt->bt_gen), MAX(txg, lrtxg),
1667 bt->bt_crtxg);
1668 }
1669
1670 /*
1671 * Set the bt's gen/txg to the bonus buffer's gen/txg
1672 * so that all of the usual ASSERTs will work.
1673 */
1674 ztest_bt_generate(bt, os, lr->lr_foid, offset, gen, txg, crtxg);
1675 }
1676
1677 if (abuf == NULL) {
1678 dmu_write(os, lr->lr_foid, offset, length, data, tx);
1679 } else {
1680 bcopy(data, abuf->b_data, length);
1681 dmu_assign_arcbuf(db, offset, abuf, tx);
1682 }
1683
1684 (void) ztest_log_write(zd, tx, lr);
1685
1686 dmu_buf_rele(db, FTAG);
1687
1688 dmu_tx_commit(tx);
1689
1690 ztest_range_unlock(rl);
1691 ztest_object_unlock(zd, lr->lr_foid);
1692
1693 return (0);
1694 }
1695
1696 static int
1697 ztest_replay_truncate(void *arg1, void *arg2, boolean_t byteswap)
1698 {
1699 ztest_ds_t *zd = arg1;
1700 lr_truncate_t *lr = arg2;
1701 objset_t *os = zd->zd_os;
1702 dmu_tx_t *tx;
1703 uint64_t txg;
1704 rl_t *rl;
1705
1706 if (byteswap)
1707 byteswap_uint64_array(lr, sizeof (*lr));
1708
1709 ztest_object_lock(zd, lr->lr_foid, RL_READER);
1710 rl = ztest_range_lock(zd, lr->lr_foid, lr->lr_offset, lr->lr_length,
1711 RL_WRITER);
1712
1713 tx = dmu_tx_create(os);
1714
1715 dmu_tx_hold_free(tx, lr->lr_foid, lr->lr_offset, lr->lr_length);
1716
1717 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1718 if (txg == 0) {
1719 ztest_range_unlock(rl);
1720 ztest_object_unlock(zd, lr->lr_foid);
1721 return (ENOSPC);
1722 }
1723
1724 VERIFY(dmu_free_range(os, lr->lr_foid, lr->lr_offset,
1725 lr->lr_length, tx) == 0);
1726
1727 (void) ztest_log_truncate(zd, tx, lr);
1728
1729 dmu_tx_commit(tx);
1730
1731 ztest_range_unlock(rl);
1732 ztest_object_unlock(zd, lr->lr_foid);
1733
1734 return (0);
1735 }
1736
1737 static int
1738 ztest_replay_setattr(void *arg1, void *arg2, boolean_t byteswap)
1739 {
1740 ztest_ds_t *zd = arg1;
1741 lr_setattr_t *lr = arg2;
1742 objset_t *os = zd->zd_os;
1743 dmu_tx_t *tx;
1744 dmu_buf_t *db;
1745 ztest_block_tag_t *bbt;
1746 uint64_t txg, lrtxg, crtxg;
1747
1748 if (byteswap)
1749 byteswap_uint64_array(lr, sizeof (*lr));
1750
1751 ztest_object_lock(zd, lr->lr_foid, RL_WRITER);
1752
1753 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1754
1755 tx = dmu_tx_create(os);
1756 dmu_tx_hold_bonus(tx, lr->lr_foid);
1757
1758 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1759 if (txg == 0) {
1760 dmu_buf_rele(db, FTAG);
1761 ztest_object_unlock(zd, lr->lr_foid);
1762 return (ENOSPC);
1763 }
1764
1765 bbt = ztest_bt_bonus(db);
1766 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1767 crtxg = bbt->bt_crtxg;
1768 lrtxg = lr->lr_common.lrc_txg;
1769
1770 if (zd->zd_zilog->zl_replay) {
1771 ASSERT(lr->lr_size != 0);
1772 ASSERT(lr->lr_mode != 0);
1773 ASSERT(lrtxg != 0);
1774 } else {
1775 /*
1776 * Randomly change the size and increment the generation.
1777 */
1778 lr->lr_size = (ztest_random(db->db_size / sizeof (*bbt)) + 1) *
1779 sizeof (*bbt);
1780 lr->lr_mode = bbt->bt_gen + 1;
1781 ASSERT(lrtxg == 0);
1782 }
1783
1784 /*
1785 * Verify that the current bonus buffer is not newer than our txg.
1786 */
1787 ztest_bt_verify(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode,
1788 MAX(txg, lrtxg), crtxg);
1789
1790 dmu_buf_will_dirty(db, tx);
1791
1792 ASSERT3U(lr->lr_size, >=, sizeof (*bbt));
1793 ASSERT3U(lr->lr_size, <=, db->db_size);
1794 VERIFY0(dmu_set_bonus(db, lr->lr_size, tx));
1795 bbt = ztest_bt_bonus(db);
1796
1797 ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode, txg, crtxg);
1798
1799 dmu_buf_rele(db, FTAG);
1800
1801 (void) ztest_log_setattr(zd, tx, lr);
1802
1803 dmu_tx_commit(tx);
1804
1805 ztest_object_unlock(zd, lr->lr_foid);
1806
1807 return (0);
1808 }
1809
1810 zil_replay_func_t *ztest_replay_vector[TX_MAX_TYPE] = {
1811 NULL, /* 0 no such transaction type */
1812 ztest_replay_create, /* TX_CREATE */
1813 NULL, /* TX_MKDIR */
1814 NULL, /* TX_MKXATTR */
1815 NULL, /* TX_SYMLINK */
1816 ztest_replay_remove, /* TX_REMOVE */
1817 NULL, /* TX_RMDIR */
1818 NULL, /* TX_LINK */
1819 NULL, /* TX_RENAME */
1820 ztest_replay_write, /* TX_WRITE */
1821 ztest_replay_truncate, /* TX_TRUNCATE */
1822 ztest_replay_setattr, /* TX_SETATTR */
1823 NULL, /* TX_ACL */
1824 NULL, /* TX_CREATE_ACL */
1825 NULL, /* TX_CREATE_ATTR */
1826 NULL, /* TX_CREATE_ACL_ATTR */
1827 NULL, /* TX_MKDIR_ACL */
1828 NULL, /* TX_MKDIR_ATTR */
1829 NULL, /* TX_MKDIR_ACL_ATTR */
1830 NULL, /* TX_WRITE2 */
1831 };
1832
1833 /*
1834 * ZIL get_data callbacks
1835 */
1836
1837 static void
1838 ztest_get_done(zgd_t *zgd, int error)
1839 {
1840 ztest_ds_t *zd = zgd->zgd_private;
1841 uint64_t object = zgd->zgd_rl->rl_object;
1842
1843 if (zgd->zgd_db)
1844 dmu_buf_rele(zgd->zgd_db, zgd);
1845
1846 ztest_range_unlock(zgd->zgd_rl);
1847 ztest_object_unlock(zd, object);
1848
1849 if (error == 0 && zgd->zgd_bp)
1850 zil_lwb_add_block(zgd->zgd_lwb, zgd->zgd_bp);
1851
1852 umem_free(zgd, sizeof (*zgd));
1853 }
1854
1855 static int
1856 ztest_get_data(void *arg, lr_write_t *lr, char *buf, struct lwb *lwb,
1857 zio_t *zio)
1858 {
1859 ztest_ds_t *zd = arg;
1860 objset_t *os = zd->zd_os;
1861 uint64_t object = lr->lr_foid;
1862 uint64_t offset = lr->lr_offset;
1863 uint64_t size = lr->lr_length;
1864 uint64_t txg = lr->lr_common.lrc_txg;
1865 uint64_t crtxg;
1866 dmu_object_info_t doi;
1867 dmu_buf_t *db;
1868 zgd_t *zgd;
1869 int error;
1870
1871 ASSERT3P(lwb, !=, NULL);
1872 ASSERT3P(zio, !=, NULL);
1873 ASSERT3U(size, !=, 0);
1874
1875 ztest_object_lock(zd, object, RL_READER);
1876 error = dmu_bonus_hold(os, object, FTAG, &db);
1877 if (error) {
1878 ztest_object_unlock(zd, object);
1879 return (error);
1880 }
1881
1882 crtxg = ztest_bt_bonus(db)->bt_crtxg;
1883
1884 if (crtxg == 0 || crtxg > txg) {
1885 dmu_buf_rele(db, FTAG);
1886 ztest_object_unlock(zd, object);
1887 return (ENOENT);
1888 }
1889
1890 dmu_object_info_from_db(db, &doi);
1891 dmu_buf_rele(db, FTAG);
1892 db = NULL;
1893
1894 zgd = umem_zalloc(sizeof (*zgd), UMEM_NOFAIL);
1895 zgd->zgd_lwb = lwb;
1896 zgd->zgd_private = zd;
1897
1898 if (buf != NULL) { /* immediate write */
1899 zgd->zgd_rl = ztest_range_lock(zd, object, offset, size,
1900 RL_READER);
1901
1902 error = dmu_read(os, object, offset, size, buf,
1903 DMU_READ_NO_PREFETCH);
1904 ASSERT(error == 0);
1905 } else {
1906 size = doi.doi_data_block_size;
1907 if (ISP2(size)) {
1908 offset = P2ALIGN(offset, size);
1909 } else {
1910 ASSERT(offset < size);
1911 offset = 0;
1912 }
1913
1914 zgd->zgd_rl = ztest_range_lock(zd, object, offset, size,
1915 RL_READER);
1916
1917 error = dmu_buf_hold(os, object, offset, zgd, &db,
1918 DMU_READ_NO_PREFETCH);
1919
1920 if (error == 0) {
1921 blkptr_t *bp = &lr->lr_blkptr;
1922
1923 zgd->zgd_db = db;
1924 zgd->zgd_bp = bp;
1925
1926 ASSERT(db->db_offset == offset);
1927 ASSERT(db->db_size == size);
1928
1929 error = dmu_sync(zio, lr->lr_common.lrc_txg,
1930 ztest_get_done, zgd);
1931
1932 if (error == 0)
1933 return (0);
1934 }
1935 }
1936
1937 ztest_get_done(zgd, error);
1938
1939 return (error);
1940 }
1941
1942 static void *
1943 ztest_lr_alloc(size_t lrsize, char *name)
1944 {
1945 char *lr;
1946 size_t namesize = name ? strlen(name) + 1 : 0;
1947
1948 lr = umem_zalloc(lrsize + namesize, UMEM_NOFAIL);
1949
1950 if (name)
1951 bcopy(name, lr + lrsize, namesize);
1952
1953 return (lr);
1954 }
1955
1956 void
1957 ztest_lr_free(void *lr, size_t lrsize, char *name)
1958 {
1959 size_t namesize = name ? strlen(name) + 1 : 0;
1960
1961 umem_free(lr, lrsize + namesize);
1962 }
1963
1964 /*
1965 * Lookup a bunch of objects. Returns the number of objects not found.
1966 */
1967 static int
1968 ztest_lookup(ztest_ds_t *zd, ztest_od_t *od, int count)
1969 {
1970 int missing = 0;
1971 int error;
1972
1973 ASSERT(MUTEX_HELD(&zd->zd_dirobj_lock));
1974
1975 for (int i = 0; i < count; i++, od++) {
1976 od->od_object = 0;
1977 error = zap_lookup(zd->zd_os, od->od_dir, od->od_name,
1978 sizeof (uint64_t), 1, &od->od_object);
1979 if (error) {
1980 ASSERT(error == ENOENT);
1981 ASSERT(od->od_object == 0);
1982 missing++;
1983 } else {
1984 dmu_buf_t *db;
1985 ztest_block_tag_t *bbt;
1986 dmu_object_info_t doi;
1987
1988 ASSERT(od->od_object != 0);
1989 ASSERT(missing == 0); /* there should be no gaps */
1990
1991 ztest_object_lock(zd, od->od_object, RL_READER);
1992 VERIFY3U(0, ==, dmu_bonus_hold(zd->zd_os,
1993 od->od_object, FTAG, &db));
1994 dmu_object_info_from_db(db, &doi);
1995 bbt = ztest_bt_bonus(db);
1996 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1997 od->od_type = doi.doi_type;
1998 od->od_blocksize = doi.doi_data_block_size;
1999 od->od_gen = bbt->bt_gen;
2000 dmu_buf_rele(db, FTAG);
2001 ztest_object_unlock(zd, od->od_object);
2002 }
2003 }
2004
2005 return (missing);
2006 }
2007
2008 static int
2009 ztest_create(ztest_ds_t *zd, ztest_od_t *od, int count)
2010 {
2011 int missing = 0;
2012
2013 ASSERT(MUTEX_HELD(&zd->zd_dirobj_lock));
2014
2015 for (int i = 0; i < count; i++, od++) {
2016 if (missing) {
2017 od->od_object = 0;
2018 missing++;
2019 continue;
2020 }
2021
2022 lr_create_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
2023
2024 lr->lr_doid = od->od_dir;
2025 lr->lr_foid = 0; /* 0 to allocate, > 0 to claim */
2026 lr->lrz_type = od->od_crtype;
2027 lr->lrz_blocksize = od->od_crblocksize;
2028 lr->lrz_ibshift = ztest_random_ibshift();
2029 lr->lrz_bonustype = DMU_OT_UINT64_OTHER;
2030 lr->lrz_bonuslen = dmu_bonus_max();
2031 lr->lr_gen = od->od_crgen;
2032 lr->lr_crtime[0] = time(NULL);
2033
2034 if (ztest_replay_create(zd, lr, B_FALSE) != 0) {
2035 ASSERT(missing == 0);
2036 od->od_object = 0;
2037 missing++;
2038 } else {
2039 od->od_object = lr->lr_foid;
2040 od->od_type = od->od_crtype;
2041 od->od_blocksize = od->od_crblocksize;
2042 od->od_gen = od->od_crgen;
2043 ASSERT(od->od_object != 0);
2044 }
2045
2046 ztest_lr_free(lr, sizeof (*lr), od->od_name);
2047 }
2048
2049 return (missing);
2050 }
2051
2052 static int
2053 ztest_remove(ztest_ds_t *zd, ztest_od_t *od, int count)
2054 {
2055 int missing = 0;
2056 int error;
2057
2058 ASSERT(MUTEX_HELD(&zd->zd_dirobj_lock));
2059
2060 od += count - 1;
2061
2062 for (int i = count - 1; i >= 0; i--, od--) {
2063 if (missing) {
2064 missing++;
2065 continue;
2066 }
2067
2068 /*
2069 * No object was found.
2070 */
2071 if (od->od_object == 0)
2072 continue;
2073
2074 lr_remove_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
2075
2076 lr->lr_doid = od->od_dir;
2077
2078 if ((error = ztest_replay_remove(zd, lr, B_FALSE)) != 0) {
2079 ASSERT3U(error, ==, ENOSPC);
2080 missing++;
2081 } else {
2082 od->od_object = 0;
2083 }
2084 ztest_lr_free(lr, sizeof (*lr), od->od_name);
2085 }
2086
2087 return (missing);
2088 }
2089
2090 static int
2091 ztest_write(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size,
2092 void *data)
2093 {
2094 lr_write_t *lr;
2095 int error;
2096
2097 lr = ztest_lr_alloc(sizeof (*lr) + size, NULL);
2098
2099 lr->lr_foid = object;
2100 lr->lr_offset = offset;
2101 lr->lr_length = size;
2102 lr->lr_blkoff = 0;
2103 BP_ZERO(&lr->lr_blkptr);
2104
2105 bcopy(data, lr + 1, size);
2106
2107 error = ztest_replay_write(zd, lr, B_FALSE);
2108
2109 ztest_lr_free(lr, sizeof (*lr) + size, NULL);
2110
2111 return (error);
2112 }
2113
2114 static int
2115 ztest_truncate(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2116 {
2117 lr_truncate_t *lr;
2118 int error;
2119
2120 lr = ztest_lr_alloc(sizeof (*lr), NULL);
2121
2122 lr->lr_foid = object;
2123 lr->lr_offset = offset;
2124 lr->lr_length = size;
2125
2126 error = ztest_replay_truncate(zd, lr, B_FALSE);
2127
2128 ztest_lr_free(lr, sizeof (*lr), NULL);
2129
2130 return (error);
2131 }
2132
2133 static int
2134 ztest_setattr(ztest_ds_t *zd, uint64_t object)
2135 {
2136 lr_setattr_t *lr;
2137 int error;
2138
2139 lr = ztest_lr_alloc(sizeof (*lr), NULL);
2140
2141 lr->lr_foid = object;
2142 lr->lr_size = 0;
2143 lr->lr_mode = 0;
2144
2145 error = ztest_replay_setattr(zd, lr, B_FALSE);
2146
2147 ztest_lr_free(lr, sizeof (*lr), NULL);
2148
2149 return (error);
2150 }
2151
2152 static void
2153 ztest_prealloc(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2154 {
2155 objset_t *os = zd->zd_os;
2156 dmu_tx_t *tx;
2157 uint64_t txg;
2158 rl_t *rl;
2159
2160 txg_wait_synced(dmu_objset_pool(os), 0);
2161
2162 ztest_object_lock(zd, object, RL_READER);
2163 rl = ztest_range_lock(zd, object, offset, size, RL_WRITER);
2164
2165 tx = dmu_tx_create(os);
2166
2167 dmu_tx_hold_write(tx, object, offset, size);
2168
2169 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
2170
2171 if (txg != 0) {
2172 dmu_prealloc(os, object, offset, size, tx);
2173 dmu_tx_commit(tx);
2174 txg_wait_synced(dmu_objset_pool(os), txg);
2175 } else {
2176 (void) dmu_free_long_range(os, object, offset, size);
2177 }
2178
2179 ztest_range_unlock(rl);
2180 ztest_object_unlock(zd, object);
2181 }
2182
2183 static void
2184 ztest_io(ztest_ds_t *zd, uint64_t object, uint64_t offset)
2185 {
2186 int err;
2187 ztest_block_tag_t wbt;
2188 dmu_object_info_t doi;
2189 enum ztest_io_type io_type;
2190 uint64_t blocksize;
2191 void *data;
2192
2193 VERIFY(dmu_object_info(zd->zd_os, object, &doi) == 0);
2194 blocksize = doi.doi_data_block_size;
2195 data = umem_alloc(blocksize, UMEM_NOFAIL);
2196
2197 /*
2198 * Pick an i/o type at random, biased toward writing block tags.
2199 */
2200 io_type = ztest_random(ZTEST_IO_TYPES);
2201 if (ztest_random(2) == 0)
2202 io_type = ZTEST_IO_WRITE_TAG;
2203
2204 rw_enter(&zd->zd_zilog_lock, RW_READER);
2205
2206 switch (io_type) {
2207
2208 case ZTEST_IO_WRITE_TAG:
2209 ztest_bt_generate(&wbt, zd->zd_os, object, offset, 0, 0, 0);
2210 (void) ztest_write(zd, object, offset, sizeof (wbt), &wbt);
2211 break;
2212
2213 case ZTEST_IO_WRITE_PATTERN:
2214 (void) memset(data, 'a' + (object + offset) % 5, blocksize);
2215 if (ztest_random(2) == 0) {
2216 /*
2217 * Induce fletcher2 collisions to ensure that
2218 * zio_ddt_collision() detects and resolves them
2219 * when using fletcher2-verify for deduplication.
2220 */
2221 ((uint64_t *)data)[0] ^= 1ULL << 63;
2222 ((uint64_t *)data)[4] ^= 1ULL << 63;
2223 }
2224 (void) ztest_write(zd, object, offset, blocksize, data);
2225 break;
2226
2227 case ZTEST_IO_WRITE_ZEROES:
2228 bzero(data, blocksize);
2229 (void) ztest_write(zd, object, offset, blocksize, data);
2230 break;
2231
2232 case ZTEST_IO_TRUNCATE:
2233 (void) ztest_truncate(zd, object, offset, blocksize);
2234 break;
2235
2236 case ZTEST_IO_SETATTR:
2237 (void) ztest_setattr(zd, object);
2238 break;
2239
2240 case ZTEST_IO_REWRITE:
2241 rw_enter(&ztest_name_lock, RW_READER);
2242 err = ztest_dsl_prop_set_uint64(zd->zd_name,
2243 ZFS_PROP_CHECKSUM, spa_dedup_checksum(ztest_spa),
2244 B_FALSE);
2245 VERIFY(err == 0 || err == ENOSPC);
2246 err = ztest_dsl_prop_set_uint64(zd->zd_name,
2247 ZFS_PROP_COMPRESSION,
2248 ztest_random_dsl_prop(ZFS_PROP_COMPRESSION),
2249 B_FALSE);
2250 VERIFY(err == 0 || err == ENOSPC);
2251 rw_exit(&ztest_name_lock);
2252
2253 VERIFY0(dmu_read(zd->zd_os, object, offset, blocksize, data,
2254 DMU_READ_NO_PREFETCH));
2255
2256 (void) ztest_write(zd, object, offset, blocksize, data);
2257 break;
2258 }
2259
2260 rw_exit(&zd->zd_zilog_lock);
2261
2262 umem_free(data, blocksize);
2263 }
2264
2265 /*
2266 * Initialize an object description template.
2267 */
2268 static void
2269 ztest_od_init(ztest_od_t *od, uint64_t id, char *tag, uint64_t index,
2270 dmu_object_type_t type, uint64_t blocksize, uint64_t gen)
2271 {
2272 od->od_dir = ZTEST_DIROBJ;
2273 od->od_object = 0;
2274
2275 od->od_crtype = type;
2276 od->od_crblocksize = blocksize ? blocksize : ztest_random_blocksize();
2277 od->od_crgen = gen;
2278
2279 od->od_type = DMU_OT_NONE;
2280 od->od_blocksize = 0;
2281 od->od_gen = 0;
2282
2283 (void) snprintf(od->od_name, sizeof (od->od_name), "%s(%lld)[%llu]",
2284 tag, (int64_t)id, index);
2285 }
2286
2287 /*
2288 * Lookup or create the objects for a test using the od template.
2289 * If the objects do not all exist, or if 'remove' is specified,
2290 * remove any existing objects and create new ones. Otherwise,
2291 * use the existing objects.
2292 */
2293 static int
2294 ztest_object_init(ztest_ds_t *zd, ztest_od_t *od, size_t size, boolean_t remove)
2295 {
2296 int count = size / sizeof (*od);
2297 int rv = 0;
2298
2299 mutex_enter(&zd->zd_dirobj_lock);
2300 if ((ztest_lookup(zd, od, count) != 0 || remove) &&
2301 (ztest_remove(zd, od, count) != 0 ||
2302 ztest_create(zd, od, count) != 0))
2303 rv = -1;
2304 zd->zd_od = od;
2305 mutex_exit(&zd->zd_dirobj_lock);
2306
2307 return (rv);
2308 }
2309
2310 /* ARGSUSED */
2311 void
2312 ztest_zil_commit(ztest_ds_t *zd, uint64_t id)
2313 {
2314 zilog_t *zilog = zd->zd_zilog;
2315
2316 rw_enter(&zd->zd_zilog_lock, RW_READER);
2317
2318 zil_commit(zilog, ztest_random(ZTEST_OBJECTS));
2319
2320 /*
2321 * Remember the committed values in zd, which is in parent/child
2322 * shared memory. If we die, the next iteration of ztest_run()
2323 * will verify that the log really does contain this record.
2324 */
2325 mutex_enter(&zilog->zl_lock);
2326 ASSERT(zd->zd_shared != NULL);
2327 ASSERT3U(zd->zd_shared->zd_seq, <=, zilog->zl_commit_lr_seq);
2328 zd->zd_shared->zd_seq = zilog->zl_commit_lr_seq;
2329 mutex_exit(&zilog->zl_lock);
2330
2331 rw_exit(&zd->zd_zilog_lock);
2332 }
2333
2334 /*
2335 * This function is designed to simulate the operations that occur during a
2336 * mount/unmount operation. We hold the dataset across these operations in an
2337 * attempt to expose any implicit assumptions about ZIL management.
2338 */
2339 /* ARGSUSED */
2340 void
2341 ztest_zil_remount(ztest_ds_t *zd, uint64_t id)
2342 {
2343 objset_t *os = zd->zd_os;
2344
2345 /*
2346 * We grab the zd_dirobj_lock to ensure that no other thread is
2347 * updating the zil (i.e. adding in-memory log records) and the
2348 * zd_zilog_lock to block any I/O.
2349 */
2350 mutex_enter(&zd->zd_dirobj_lock);
2351 rw_enter(&zd->zd_zilog_lock, RW_WRITER);
2352
2353 /* zfsvfs_teardown() */
2354 zil_close(zd->zd_zilog);
2355
2356 /* zfsvfs_setup() */
2357 VERIFY(zil_open(os, ztest_get_data) == zd->zd_zilog);
2358 zil_replay(os, zd, ztest_replay_vector);
2359
2360 rw_exit(&zd->zd_zilog_lock);
2361 mutex_exit(&zd->zd_dirobj_lock);
2362 }
2363
2364 /*
2365 * Verify that we can't destroy an active pool, create an existing pool,
2366 * or create a pool with a bad vdev spec.
2367 */
2368 /* ARGSUSED */
2369 void
2370 ztest_spa_create_destroy(ztest_ds_t *zd, uint64_t id)
2371 {
2372 ztest_shared_opts_t *zo = &ztest_opts;
2373 spa_t *spa;
2374 nvlist_t *nvroot;
2375
2376 /*
2377 * Attempt to create using a bad file.
2378 */
2379 nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1);
2380 VERIFY3U(ENOENT, ==,
2381 spa_create("ztest_bad_file", nvroot, NULL, NULL));
2382 nvlist_free(nvroot);
2383
2384 /*
2385 * Attempt to create using a bad mirror.
2386 */
2387 nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 2, 1);
2388 VERIFY3U(ENOENT, ==,
2389 spa_create("ztest_bad_mirror", nvroot, NULL, NULL));
2390 nvlist_free(nvroot);
2391
2392 /*
2393 * Attempt to create an existing pool. It shouldn't matter
2394 * what's in the nvroot; we should fail with EEXIST.
2395 */
2396 rw_enter(&ztest_name_lock, RW_READER);
2397 nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1);
2398 VERIFY3U(EEXIST, ==, spa_create(zo->zo_pool, nvroot, NULL, NULL));
2399 nvlist_free(nvroot);
2400 VERIFY3U(0, ==, spa_open(zo->zo_pool, &spa, FTAG));
2401 VERIFY3U(EBUSY, ==, spa_destroy(zo->zo_pool));
2402 spa_close(spa, FTAG);
2403
2404 rw_exit(&ztest_name_lock);
2405 }
2406
2407 /* ARGSUSED */
2408 void
2409 ztest_spa_upgrade(ztest_ds_t *zd, uint64_t id)
2410 {
2411 spa_t *spa;
2412 uint64_t initial_version = SPA_VERSION_INITIAL;
2413 uint64_t version, newversion;
2414 nvlist_t *nvroot, *props;
2415 char *name;
2416
2417 mutex_enter(&ztest_vdev_lock);
2418 name = kmem_asprintf("%s_upgrade", ztest_opts.zo_pool);
2419
2420 /*
2421 * Clean up from previous runs.
2422 */
2423 (void) spa_destroy(name);
2424
2425 nvroot = make_vdev_root(NULL, NULL, name, ztest_opts.zo_vdev_size, 0,
2426 0, ztest_opts.zo_raidz, ztest_opts.zo_mirrors, 1);
2427
2428 /*
2429 * If we're configuring a RAIDZ device then make sure that the
2430 * the initial version is capable of supporting that feature.
2431 */
2432 switch (ztest_opts.zo_raidz_parity) {
2433 case 0:
2434 case 1:
2435 initial_version = SPA_VERSION_INITIAL;
2436 break;
2437 case 2:
2438 initial_version = SPA_VERSION_RAIDZ2;
2439 break;
2440 case 3:
2441 initial_version = SPA_VERSION_RAIDZ3;
2442 break;
2443 }
2444
2445 /*
2446 * Create a pool with a spa version that can be upgraded. Pick
2447 * a value between initial_version and SPA_VERSION_BEFORE_FEATURES.
2448 */
2449 do {
2450 version = ztest_random_spa_version(initial_version);
2451 } while (version > SPA_VERSION_BEFORE_FEATURES);
2452
2453 props = fnvlist_alloc();
2454 fnvlist_add_uint64(props,
2455 zpool_prop_to_name(ZPOOL_PROP_VERSION), version);
2456 VERIFY0(spa_create(name, nvroot, props, NULL));
2457 fnvlist_free(nvroot);
2458 fnvlist_free(props);
2459
2460 VERIFY0(spa_open(name, &spa, FTAG));
2461 VERIFY3U(spa_version(spa), ==, version);
2462 newversion = ztest_random_spa_version(version + 1);
2463
2464 if (ztest_opts.zo_verbose >= 4) {
2465 (void) printf("upgrading spa version from %llu to %llu\n",
2466 (u_longlong_t)version, (u_longlong_t)newversion);
2467 }
2468
2469 spa_upgrade(spa, newversion);
2470 VERIFY3U(spa_version(spa), >, version);
2471 VERIFY3U(spa_version(spa), ==, fnvlist_lookup_uint64(spa->spa_config,
2472 zpool_prop_to_name(ZPOOL_PROP_VERSION)));
2473 spa_close(spa, FTAG);
2474
2475 strfree(name);
2476 mutex_exit(&ztest_vdev_lock);
2477 }
2478
2479 static vdev_t *
2480 vdev_lookup_by_path(vdev_t *vd, const char *path)
2481 {
2482 vdev_t *mvd;
2483
2484 if (vd->vdev_path != NULL && strcmp(path, vd->vdev_path) == 0)
2485 return (vd);
2486
2487 for (int c = 0; c < vd->vdev_children; c++)
2488 if ((mvd = vdev_lookup_by_path(vd->vdev_child[c], path)) !=
2489 NULL)
2490 return (mvd);
2491
2492 return (NULL);
2493 }
2494
2495 /*
2496 * Find the first available hole which can be used as a top-level.
2497 */
2498 int
2499 find_vdev_hole(spa_t *spa)
2500 {
2501 vdev_t *rvd = spa->spa_root_vdev;
2502 int c;
2503
2504 ASSERT(spa_config_held(spa, SCL_VDEV, RW_READER) == SCL_VDEV);
2505
2506 for (c = 0; c < rvd->vdev_children; c++) {
2507 vdev_t *cvd = rvd->vdev_child[c];
2508
2509 if (cvd->vdev_ishole)
2510 break;
2511 }
2512 return (c);
2513 }
2514
2515 /*
2516 * Verify that vdev_add() works as expected.
2517 */
2518 /* ARGSUSED */
2519 void
2520 ztest_vdev_add_remove(ztest_ds_t *zd, uint64_t id)
2521 {
2522 ztest_shared_t *zs = ztest_shared;
2523 spa_t *spa = ztest_spa;
2524 uint64_t leaves;
2525 uint64_t guid;
2526 nvlist_t *nvroot;
2527 int error;
2528
2529 mutex_enter(&ztest_vdev_lock);
2530 leaves = MAX(zs->zs_mirrors + zs->zs_splits, 1) * ztest_opts.zo_raidz;
2531
2532 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2533
2534 ztest_shared->zs_vdev_next_leaf = find_vdev_hole(spa) * leaves;
2535
2536 /*
2537 * If we have slogs then remove them 1/4 of the time.
2538 */
2539 if (spa_has_slogs(spa) && ztest_random(4) == 0) {
2540 /*
2541 * Grab the guid from the head of the log class rotor.
2542 */
2543 guid = spa_log_class(spa)->mc_rotor->mg_vd->vdev_guid;
2544
2545 spa_config_exit(spa, SCL_VDEV, FTAG);
2546
2547 /*
2548 * We have to grab the zs_name_lock as writer to
2549 * prevent a race between removing a slog (dmu_objset_find)
2550 * and destroying a dataset. Removing the slog will
2551 * grab a reference on the dataset which may cause
2552 * dmu_objset_destroy() to fail with EBUSY thus
2553 * leaving the dataset in an inconsistent state.
2554 */
2555 rw_enter(&ztest_name_lock, RW_WRITER);
2556 error = spa_vdev_remove(spa, guid, B_FALSE);
2557 rw_exit(&ztest_name_lock);
2558
2559 if (error && error != EEXIST)
2560 fatal(0, "spa_vdev_remove() = %d", error);
2561 } else {
2562 spa_config_exit(spa, SCL_VDEV, FTAG);
2563
2564 /*
2565 * Make 1/4 of the devices be log devices.
2566 */
2567 nvroot = make_vdev_root(NULL, NULL, NULL,
2568 ztest_opts.zo_vdev_size, 0,
2569 ztest_random(4) == 0, ztest_opts.zo_raidz,
2570 zs->zs_mirrors, 1);
2571
2572 error = spa_vdev_add(spa, nvroot);
2573 nvlist_free(nvroot);
2574
2575 if (error == ENOSPC)
2576 ztest_record_enospc("spa_vdev_add");
2577 else if (error != 0)
2578 fatal(0, "spa_vdev_add() = %d", error);
2579 }
2580
2581 mutex_exit(&ztest_vdev_lock);
2582 }
2583
2584 /*
2585 * Verify that adding/removing aux devices (l2arc, hot spare) works as expected.
2586 */
2587 /* ARGSUSED */
2588 void
2589 ztest_vdev_aux_add_remove(ztest_ds_t *zd, uint64_t id)
2590 {
2591 ztest_shared_t *zs = ztest_shared;
2592 spa_t *spa = ztest_spa;
2593 vdev_t *rvd = spa->spa_root_vdev;
2594 spa_aux_vdev_t *sav;
2595 char *aux;
2596 uint64_t guid = 0;
2597 int error;
2598
2599 if (ztest_random(2) == 0) {
2600 sav = &spa->spa_spares;
2601 aux = ZPOOL_CONFIG_SPARES;
2602 } else {
2603 sav = &spa->spa_l2cache;
2604 aux = ZPOOL_CONFIG_L2CACHE;
2605 }
2606
2607 mutex_enter(&ztest_vdev_lock);
2608
2609 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2610
2611 if (sav->sav_count != 0 && ztest_random(4) == 0) {
2612 /*
2613 * Pick a random device to remove.
2614 */
2615 guid = sav->sav_vdevs[ztest_random(sav->sav_count)]->vdev_guid;
2616 } else {
2617 /*
2618 * Find an unused device we can add.
2619 */
2620 zs->zs_vdev_aux = 0;
2621 for (;;) {
2622 char path[MAXPATHLEN];
2623 int c;
2624 (void) snprintf(path, sizeof (path), ztest_aux_template,
2625 ztest_opts.zo_dir, ztest_opts.zo_pool, aux,
2626 zs->zs_vdev_aux);
2627 for (c = 0; c < sav->sav_count; c++)
2628 if (strcmp(sav->sav_vdevs[c]->vdev_path,
2629 path) == 0)
2630 break;
2631 if (c == sav->sav_count &&
2632 vdev_lookup_by_path(rvd, path) == NULL)
2633 break;
2634 zs->zs_vdev_aux++;
2635 }
2636 }
2637
2638 spa_config_exit(spa, SCL_VDEV, FTAG);
2639
2640 if (guid == 0) {
2641 /*
2642 * Add a new device.
2643 */
2644 nvlist_t *nvroot = make_vdev_root(NULL, aux, NULL,
2645 (ztest_opts.zo_vdev_size * 5) / 4, 0, 0, 0, 0, 1);
2646 error = spa_vdev_add(spa, nvroot);
2647 if (error != 0)
2648 fatal(0, "spa_vdev_add(%p) = %d", nvroot, error);
2649 nvlist_free(nvroot);
2650 } else {
2651 /*
2652 * Remove an existing device. Sometimes, dirty its
2653 * vdev state first to make sure we handle removal
2654 * of devices that have pending state changes.
2655 */
2656 if (ztest_random(2) == 0)
2657 (void) vdev_online(spa, guid, 0, NULL);
2658
2659 error = spa_vdev_remove(spa, guid, B_FALSE);
2660 if (error != 0 && error != EBUSY)
2661 fatal(0, "spa_vdev_remove(%llu) = %d", guid, error);
2662 }
2663
2664 mutex_exit(&ztest_vdev_lock);
2665 }
2666
2667 /*
2668 * split a pool if it has mirror tlvdevs
2669 */
2670 /* ARGSUSED */
2671 void
2672 ztest_split_pool(ztest_ds_t *zd, uint64_t id)
2673 {
2674 ztest_shared_t *zs = ztest_shared;
2675 spa_t *spa = ztest_spa;
2676 vdev_t *rvd = spa->spa_root_vdev;
2677 nvlist_t *tree, **child, *config, *split, **schild;
2678 uint_t c, children, schildren = 0, lastlogid = 0;
2679 int error = 0;
2680
2681 mutex_enter(&ztest_vdev_lock);
2682
2683 /* ensure we have a useable config; mirrors of raidz aren't supported */
2684 if (zs->zs_mirrors < 3 || ztest_opts.zo_raidz > 1) {
2685 mutex_exit(&ztest_vdev_lock);
2686 return;
2687 }
2688
2689 /* clean up the old pool, if any */
2690 (void) spa_destroy("splitp");
2691
2692 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2693
2694 /* generate a config from the existing config */
2695 mutex_enter(&spa->spa_props_lock);
2696 VERIFY(nvlist_lookup_nvlist(spa->spa_config, ZPOOL_CONFIG_VDEV_TREE,
2697 &tree) == 0);
2698 mutex_exit(&spa->spa_props_lock);
2699
2700 VERIFY(nvlist_lookup_nvlist_array(tree, ZPOOL_CONFIG_CHILDREN, &child,
2701 &children) == 0);
2702
2703 schild = malloc(rvd->vdev_children * sizeof (nvlist_t *));
2704 for (c = 0; c < children; c++) {
2705 vdev_t *tvd = rvd->vdev_child[c];
2706 nvlist_t **mchild;
2707 uint_t mchildren;
2708
2709 if (tvd->vdev_islog || tvd->vdev_ops == &vdev_hole_ops) {
2710 VERIFY(nvlist_alloc(&schild[schildren], NV_UNIQUE_NAME,
2711 0) == 0);
2712 VERIFY(nvlist_add_string(schild[schildren],
2713 ZPOOL_CONFIG_TYPE, VDEV_TYPE_HOLE) == 0);
2714 VERIFY(nvlist_add_uint64(schild[schildren],
2715 ZPOOL_CONFIG_IS_HOLE, 1) == 0);
2716 if (lastlogid == 0)
2717 lastlogid = schildren;
2718 ++schildren;
2719 continue;
2720 }
2721 lastlogid = 0;
2722 VERIFY(nvlist_lookup_nvlist_array(child[c],
2723 ZPOOL_CONFIG_CHILDREN, &mchild, &mchildren) == 0);
2724 VERIFY(nvlist_dup(mchild[0], &schild[schildren++], 0) == 0);
2725 }
2726
2727 /* OK, create a config that can be used to split */
2728 VERIFY(nvlist_alloc(&split, NV_UNIQUE_NAME, 0) == 0);
2729 VERIFY(nvlist_add_string(split, ZPOOL_CONFIG_TYPE,
2730 VDEV_TYPE_ROOT) == 0);
2731 VERIFY(nvlist_add_nvlist_array(split, ZPOOL_CONFIG_CHILDREN, schild,
2732 lastlogid != 0 ? lastlogid : schildren) == 0);
2733
2734 VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, 0) == 0);
2735 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, split) == 0);
2736
2737 for (c = 0; c < schildren; c++)
2738 nvlist_free(schild[c]);
2739 free(schild);
2740 nvlist_free(split);
2741
2742 spa_config_exit(spa, SCL_VDEV, FTAG);
2743
2744 rw_enter(&ztest_name_lock, RW_WRITER);
2745 error = spa_vdev_split_mirror(spa, "splitp", config, NULL, B_FALSE);
2746 rw_exit(&ztest_name_lock);
2747
2748 nvlist_free(config);
2749
2750 if (error == 0) {
2751 (void) printf("successful split - results:\n");
2752 mutex_enter(&spa_namespace_lock);
2753 show_pool_stats(spa);
2754 show_pool_stats(spa_lookup("splitp"));
2755 mutex_exit(&spa_namespace_lock);
2756 ++zs->zs_splits;
2757 --zs->zs_mirrors;
2758 }
2759 mutex_exit(&ztest_vdev_lock);
2760
2761 }
2762
2763 /*
2764 * Verify that we can attach and detach devices.
2765 */
2766 /* ARGSUSED */
2767 void
2768 ztest_vdev_attach_detach(ztest_ds_t *zd, uint64_t id)
2769 {
2770 ztest_shared_t *zs = ztest_shared;
2771 spa_t *spa = ztest_spa;
2772 spa_aux_vdev_t *sav = &spa->spa_spares;
2773 vdev_t *rvd = spa->spa_root_vdev;
2774 vdev_t *oldvd, *newvd, *pvd;
2775 nvlist_t *root;
2776 uint64_t leaves;
2777 uint64_t leaf, top;
2778 uint64_t ashift = ztest_get_ashift();
2779 uint64_t oldguid, pguid;
2780 uint64_t oldsize, newsize;
2781 char oldpath[MAXPATHLEN], newpath[MAXPATHLEN];
2782 int replacing;
2783 int oldvd_has_siblings = B_FALSE;
2784 int newvd_is_spare = B_FALSE;
2785 int oldvd_is_log;
2786 int error, expected_error;
2787
2788 mutex_enter(&ztest_vdev_lock);
2789 leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz;
2790
2791 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2792
2793 /*
2794 * If a vdev is in the process of being removed, its removal may
2795 * finish while we are in progress, leading to an unexpected error
2796 * value. Don't bother trying to attach while we are in the middle
2797 * of removal.
2798 */
2799 if (spa->spa_vdev_removal != NULL) {
2800 spa_config_exit(spa, SCL_ALL, FTAG);
2801 mutex_exit(&ztest_vdev_lock);
2802 return;
2803 }
2804
2805 /*
2806 * Decide whether to do an attach or a replace.
2807 */
2808 replacing = ztest_random(2);
2809
2810 /*
2811 * Pick a random top-level vdev.
2812 */
2813 top = ztest_random_vdev_top(spa, B_TRUE);
2814
2815 /*
2816 * Pick a random leaf within it.
2817 */
2818 leaf = ztest_random(leaves);
2819
2820 /*
2821 * Locate this vdev.
2822 */
2823 oldvd = rvd->vdev_child[top];
2824 if (zs->zs_mirrors >= 1) {
2825 ASSERT(oldvd->vdev_ops == &vdev_mirror_ops);
2826 ASSERT(oldvd->vdev_children >= zs->zs_mirrors);
2827 oldvd = oldvd->vdev_child[leaf / ztest_opts.zo_raidz];
2828 }
2829 if (ztest_opts.zo_raidz > 1) {
2830 ASSERT(oldvd->vdev_ops == &vdev_raidz_ops);
2831 ASSERT(oldvd->vdev_children == ztest_opts.zo_raidz);
2832 oldvd = oldvd->vdev_child[leaf % ztest_opts.zo_raidz];
2833 }
2834
2835 /*
2836 * If we're already doing an attach or replace, oldvd may be a
2837 * mirror vdev -- in which case, pick a random child.
2838 */
2839 while (oldvd->vdev_children != 0) {
2840 oldvd_has_siblings = B_TRUE;
2841 ASSERT(oldvd->vdev_children >= 2);
2842 oldvd = oldvd->vdev_child[ztest_random(oldvd->vdev_children)];
2843 }
2844
2845 oldguid = oldvd->vdev_guid;
2846 oldsize = vdev_get_min_asize(oldvd);
2847 oldvd_is_log = oldvd->vdev_top->vdev_islog;
2848 (void) strcpy(oldpath, oldvd->vdev_path);
2849 pvd = oldvd->vdev_parent;
2850 pguid = pvd->vdev_guid;
2851
2852 /*
2853 * If oldvd has siblings, then half of the time, detach it.
2854 */
2855 if (oldvd_has_siblings && ztest_random(2) == 0) {
2856 spa_config_exit(spa, SCL_ALL, FTAG);
2857 error = spa_vdev_detach(spa, oldguid, pguid, B_FALSE);
2858 if (error != 0 && error != ENODEV && error != EBUSY &&
2859 error != ENOTSUP)
2860 fatal(0, "detach (%s) returned %d", oldpath, error);
2861 mutex_exit(&ztest_vdev_lock);
2862 return;
2863 }
2864
2865 /*
2866 * For the new vdev, choose with equal probability between the two
2867 * standard paths (ending in either 'a' or 'b') or a random hot spare.
2868 */
2869 if (sav->sav_count != 0 && ztest_random(3) == 0) {
2870 newvd = sav->sav_vdevs[ztest_random(sav->sav_count)];
2871 newvd_is_spare = B_TRUE;
2872 (void) strcpy(newpath, newvd->vdev_path);
2873 } else {
2874 (void) snprintf(newpath, sizeof (newpath), ztest_dev_template,
2875 ztest_opts.zo_dir, ztest_opts.zo_pool,
2876 top * leaves + leaf);
2877 if (ztest_random(2) == 0)
2878 newpath[strlen(newpath) - 1] = 'b';
2879 newvd = vdev_lookup_by_path(rvd, newpath);
2880 }
2881
2882 if (newvd) {
2883 /*
2884 * Reopen to ensure the vdev's asize field isn't stale.
2885 */
2886 vdev_reopen(newvd);
2887 newsize = vdev_get_min_asize(newvd);
2888 } else {
2889 /*
2890 * Make newsize a little bigger or smaller than oldsize.
2891 * If it's smaller, the attach should fail.
2892 * If it's larger, and we're doing a replace,
2893 * we should get dynamic LUN growth when we're done.
2894 */
2895 newsize = 10 * oldsize / (9 + ztest_random(3));
2896 }
2897
2898 /*
2899 * If pvd is not a mirror or root, the attach should fail with ENOTSUP,
2900 * unless it's a replace; in that case any non-replacing parent is OK.
2901 *
2902 * If newvd is already part of the pool, it should fail with EBUSY.
2903 *
2904 * If newvd is too small, it should fail with EOVERFLOW.
2905 */
2906 if (pvd->vdev_ops != &vdev_mirror_ops &&
2907 pvd->vdev_ops != &vdev_root_ops && (!replacing ||
2908 pvd->vdev_ops == &vdev_replacing_ops ||
2909 pvd->vdev_ops == &vdev_spare_ops))
2910 expected_error = ENOTSUP;
2911 else if (newvd_is_spare && (!replacing || oldvd_is_log))
2912 expected_error = ENOTSUP;
2913 else if (newvd == oldvd)
2914 expected_error = replacing ? 0 : EBUSY;
2915 else if (vdev_lookup_by_path(rvd, newpath) != NULL)
2916 expected_error = EBUSY;
2917 else if (newsize < oldsize)
2918 expected_error = EOVERFLOW;
2919 else if (ashift > oldvd->vdev_top->vdev_ashift)
2920 expected_error = EDOM;
2921 else
2922 expected_error = 0;
2923
2924 spa_config_exit(spa, SCL_ALL, FTAG);
2925
2926 /*
2927 * Build the nvlist describing newpath.
2928 */
2929 root = make_vdev_root(newpath, NULL, NULL, newvd == NULL ? newsize : 0,
2930 ashift, 0, 0, 0, 1);
2931
2932 error = spa_vdev_attach(spa, oldguid, root, replacing);
2933
2934 nvlist_free(root);
2935
2936 /*
2937 * If our parent was the replacing vdev, but the replace completed,
2938 * then instead of failing with ENOTSUP we may either succeed,
2939 * fail with ENODEV, or fail with EOVERFLOW.
2940 */
2941 if (expected_error == ENOTSUP &&
2942 (error == 0 || error == ENODEV || error == EOVERFLOW))
2943 expected_error = error;
2944
2945 /*
2946 * If someone grew the LUN, the replacement may be too small.
2947 */
2948 if (error == EOVERFLOW || error == EBUSY)
2949 expected_error = error;
2950
2951 /* XXX workaround 6690467 */
2952 if (error != expected_error && expected_error != EBUSY) {
2953 fatal(0, "attach (%s %llu, %s %llu, %d) "
2954 "returned %d, expected %d",
2955 oldpath, oldsize, newpath,
2956 newsize, replacing, error, expected_error);
2957 }
2958
2959 mutex_exit(&ztest_vdev_lock);
2960 }
2961
2962 /* ARGSUSED */
2963 void
2964 ztest_device_removal(ztest_ds_t *zd, uint64_t id)
2965 {
2966 spa_t *spa = ztest_spa;
2967 vdev_t *vd;
2968 uint64_t guid;
2969
2970 mutex_enter(&ztest_vdev_lock);
2971
2972 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2973 vd = vdev_lookup_top(spa, ztest_random_vdev_top(spa, B_FALSE));
2974 guid = vd->vdev_guid;
2975 spa_config_exit(spa, SCL_VDEV, FTAG);
2976
2977 (void) spa_vdev_remove(spa, guid, B_FALSE);
2978
2979 mutex_exit(&ztest_vdev_lock);
2980 }
2981
2982 /*
2983 * Callback function which expands the physical size of the vdev.
2984 */
2985 vdev_t *
2986 grow_vdev(vdev_t *vd, void *arg)
2987 {
2988 spa_t *spa = vd->vdev_spa;
2989 size_t *newsize = arg;
2990 size_t fsize;
2991 int fd;
2992
2993 ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
2994 ASSERT(vd->vdev_ops->vdev_op_leaf);
2995
2996 if ((fd = open(vd->vdev_path, O_RDWR)) == -1)
2997 return (vd);
2998
2999 fsize = lseek(fd, 0, SEEK_END);
3000 (void) ftruncate(fd, *newsize);
3001
3002 if (ztest_opts.zo_verbose >= 6) {
3003 (void) printf("%s grew from %lu to %lu bytes\n",
3004 vd->vdev_path, (ulong_t)fsize, (ulong_t)*newsize);
3005 }
3006 (void) close(fd);
3007 return (NULL);
3008 }
3009
3010 /*
3011 * Callback function which expands a given vdev by calling vdev_online().
3012 */
3013 /* ARGSUSED */
3014 vdev_t *
3015 online_vdev(vdev_t *vd, void *arg)
3016 {
3017 spa_t *spa = vd->vdev_spa;
3018 vdev_t *tvd = vd->vdev_top;
3019 uint64_t guid = vd->vdev_guid;
3020 uint64_t generation = spa->spa_config_generation + 1;
3021 vdev_state_t newstate = VDEV_STATE_UNKNOWN;
3022 int error;
3023
3024 ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
3025 ASSERT(vd->vdev_ops->vdev_op_leaf);
3026
3027 /* Calling vdev_online will initialize the new metaslabs */
3028 spa_config_exit(spa, SCL_STATE, spa);
3029 error = vdev_online(spa, guid, ZFS_ONLINE_EXPAND, &newstate);
3030 spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3031
3032 /*
3033 * If vdev_online returned an error or the underlying vdev_open
3034 * failed then we abort the expand. The only way to know that
3035 * vdev_open fails is by checking the returned newstate.
3036 */
3037 if (error || newstate != VDEV_STATE_HEALTHY) {
3038 if (ztest_opts.zo_verbose >= 5) {
3039 (void) printf("Unable to expand vdev, state %llu, "
3040 "error %d\n", (u_longlong_t)newstate, error);
3041 }
3042 return (vd);
3043 }
3044 ASSERT3U(newstate, ==, VDEV_STATE_HEALTHY);
3045
3046 /*
3047 * Since we dropped the lock we need to ensure that we're
3048 * still talking to the original vdev. It's possible this
3049 * vdev may have been detached/replaced while we were
3050 * trying to online it.
3051 */
3052 if (generation != spa->spa_config_generation) {
3053 if (ztest_opts.zo_verbose >= 5) {
3054 (void) printf("vdev configuration has changed, "
3055 "guid %llu, state %llu, expected gen %llu, "
3056 "got gen %llu\n",
3057 (u_longlong_t)guid,
3058 (u_longlong_t)tvd->vdev_state,
3059 (u_longlong_t)generation,
3060 (u_longlong_t)spa->spa_config_generation);
3061 }
3062 return (vd);
3063 }
3064 return (NULL);
3065 }
3066
3067 /*
3068 * Traverse the vdev tree calling the supplied function.
3069 * We continue to walk the tree until we either have walked all
3070 * children or we receive a non-NULL return from the callback.
3071 * If a NULL callback is passed, then we just return back the first
3072 * leaf vdev we encounter.
3073 */
3074 vdev_t *
3075 vdev_walk_tree(vdev_t *vd, vdev_t *(*func)(vdev_t *, void *), void *arg)
3076 {
3077 if (vd->vdev_ops->vdev_op_leaf) {
3078 if (func == NULL)
3079 return (vd);
3080 else
3081 return (func(vd, arg));
3082 }
3083
3084 for (uint_t c = 0; c < vd->vdev_children; c++) {
3085 vdev_t *cvd = vd->vdev_child[c];
3086 if ((cvd = vdev_walk_tree(cvd, func, arg)) != NULL)
3087 return (cvd);
3088 }
3089 return (NULL);
3090 }
3091
3092 /*
3093 * Verify that dynamic LUN growth works as expected.
3094 */
3095 /* ARGSUSED */
3096 void
3097 ztest_vdev_LUN_growth(ztest_ds_t *zd, uint64_t id)
3098 {
3099 spa_t *spa = ztest_spa;
3100 vdev_t *vd, *tvd;
3101 metaslab_class_t *mc;
3102 metaslab_group_t *mg;
3103 size_t psize, newsize;
3104 uint64_t top;
3105 uint64_t old_class_space, new_class_space, old_ms_count, new_ms_count;
3106
3107 mutex_enter(&ztest_vdev_lock);
3108 spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3109
3110 /*
3111 * If there is a vdev removal in progress, it could complete while
3112 * we are running, in which case we would not be able to verify
3113 * that the metaslab_class space increased (because it decreases
3114 * when the device removal completes).
3115 */
3116 if (spa->spa_vdev_removal != NULL) {
3117 spa_config_exit(spa, SCL_STATE, FTAG);
3118 mutex_exit(&ztest_vdev_lock);
3119 return;
3120 }
3121
3122 top = ztest_random_vdev_top(spa, B_TRUE);
3123
3124 tvd = spa->spa_root_vdev->vdev_child[top];
3125 mg = tvd->vdev_mg;
3126 mc = mg->mg_class;
3127 old_ms_count = tvd->vdev_ms_count;
3128 old_class_space = metaslab_class_get_space(mc);
3129
3130 /*
3131 * Determine the size of the first leaf vdev associated with
3132 * our top-level device.
3133 */
3134 vd = vdev_walk_tree(tvd, NULL, NULL);
3135 ASSERT3P(vd, !=, NULL);
3136 ASSERT(vd->vdev_ops->vdev_op_leaf);
3137
3138 psize = vd->vdev_psize;
3139
3140 /*
3141 * We only try to expand the vdev if it's healthy, less than 4x its
3142 * original size, and it has a valid psize.
3143 */
3144 if (tvd->vdev_state != VDEV_STATE_HEALTHY ||
3145 psize == 0 || psize >= 4 * ztest_opts.zo_vdev_size) {
3146 spa_config_exit(spa, SCL_STATE, spa);
3147 mutex_exit(&ztest_vdev_lock);
3148 return;
3149 }
3150 ASSERT(psize > 0);
3151 newsize = psize + psize / 8;
3152 ASSERT3U(newsize, >, psize);
3153
3154 if (ztest_opts.zo_verbose >= 6) {
3155 (void) printf("Expanding LUN %s from %lu to %lu\n",
3156 vd->vdev_path, (ulong_t)psize, (ulong_t)newsize);
3157 }
3158
3159 /*
3160 * Growing the vdev is a two step process:
3161 * 1). expand the physical size (i.e. relabel)
3162 * 2). online the vdev to create the new metaslabs
3163 */
3164 if (vdev_walk_tree(tvd, grow_vdev, &newsize) != NULL ||
3165 vdev_walk_tree(tvd, online_vdev, NULL) != NULL ||
3166 tvd->vdev_state != VDEV_STATE_HEALTHY) {
3167 if (ztest_opts.zo_verbose >= 5) {
3168 (void) printf("Could not expand LUN because "
3169 "the vdev configuration changed.\n");
3170 }
3171 spa_config_exit(spa, SCL_STATE, spa);
3172 mutex_exit(&ztest_vdev_lock);
3173 return;
3174 }
3175
3176 spa_config_exit(spa, SCL_STATE, spa);
3177
3178 /*
3179 * Expanding the LUN will update the config asynchronously,
3180 * thus we must wait for the async thread to complete any
3181 * pending tasks before proceeding.
3182 */
3183 for (;;) {
3184 boolean_t done;
3185 mutex_enter(&spa->spa_async_lock);
3186 done = (spa->spa_async_thread == NULL && !spa->spa_async_tasks);
3187 mutex_exit(&spa->spa_async_lock);
3188 if (done)
3189 break;
3190 txg_wait_synced(spa_get_dsl(spa), 0);
3191 (void) poll(NULL, 0, 100);
3192 }
3193
3194 spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3195
3196 tvd = spa->spa_root_vdev->vdev_child[top];
3197 new_ms_count = tvd->vdev_ms_count;
3198 new_class_space = metaslab_class_get_space(mc);
3199
3200 if (tvd->vdev_mg != mg || mg->mg_class != mc) {
3201 if (ztest_opts.zo_verbose >= 5) {
3202 (void) printf("Could not verify LUN expansion due to "
3203 "intervening vdev offline or remove.\n");
3204 }
3205 spa_config_exit(spa, SCL_STATE, spa);
3206 mutex_exit(&ztest_vdev_lock);
3207 return;
3208 }
3209
3210 /*
3211 * Make sure we were able to grow the vdev.
3212 */
3213 if (new_ms_count <= old_ms_count) {
3214 fatal(0, "LUN expansion failed: ms_count %llu < %llu\n",
3215 old_ms_count, new_ms_count);
3216 }
3217
3218 /*
3219 * Make sure we were able to grow the pool.
3220 */
3221 if (new_class_space <= old_class_space) {
3222 fatal(0, "LUN expansion failed: class_space %llu < %llu\n",
3223 old_class_space, new_class_space);
3224 }
3225
3226 if (ztest_opts.zo_verbose >= 5) {
3227 char oldnumbuf[NN_NUMBUF_SZ], newnumbuf[NN_NUMBUF_SZ];
3228
3229 nicenum(old_class_space, oldnumbuf, sizeof (oldnumbuf));
3230 nicenum(new_class_space, newnumbuf, sizeof (newnumbuf));
3231 (void) printf("%s grew from %s to %s\n",
3232 spa->spa_name, oldnumbuf, newnumbuf);
3233 }
3234
3235 spa_config_exit(spa, SCL_STATE, spa);
3236 mutex_exit(&ztest_vdev_lock);
3237 }
3238
3239 /*
3240 * Verify that dmu_objset_{create,destroy,open,close} work as expected.
3241 */
3242 /* ARGSUSED */
3243 static void
3244 ztest_objset_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
3245 {
3246 /*
3247 * Create the objects common to all ztest datasets.
3248 */
3249 VERIFY(zap_create_claim(os, ZTEST_DIROBJ,
3250 DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx) == 0);
3251 }
3252
3253 static int
3254 ztest_dataset_create(char *dsname)
3255 {
3256 uint64_t zilset = ztest_random(100);
3257 int err = dmu_objset_create(dsname, DMU_OST_OTHER, 0,
3258 ztest_objset_create_cb, NULL);
3259
3260 if (err || zilset < 80)
3261 return (err);
3262
3263 if (ztest_opts.zo_verbose >= 6)
3264 (void) printf("Setting dataset %s to sync always\n", dsname);
3265 return (ztest_dsl_prop_set_uint64(dsname, ZFS_PROP_SYNC,
3266 ZFS_SYNC_ALWAYS, B_FALSE));
3267 }
3268
3269 /* ARGSUSED */
3270 static int
3271 ztest_objset_destroy_cb(const char *name, void *arg)
3272 {
3273 objset_t *os;
3274 dmu_object_info_t doi;
3275 int error;
3276
3277 /*
3278 * Verify that the dataset contains a directory object.
3279 */
3280 VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_TRUE, FTAG, &os));
3281 error = dmu_object_info(os, ZTEST_DIROBJ, &doi);
3282 if (error != ENOENT) {
3283 /* We could have crashed in the middle of destroying it */
3284 ASSERT0(error);
3285 ASSERT3U(doi.doi_type, ==, DMU_OT_ZAP_OTHER);
3286 ASSERT3S(doi.doi_physical_blocks_512, >=, 0);
3287 }
3288 dmu_objset_disown(os, FTAG);
3289
3290 /*
3291 * Destroy the dataset.
3292 */
3293 if (strchr(name, '@') != NULL) {
3294 VERIFY0(dsl_destroy_snapshot(name, B_TRUE));
3295 } else {
3296 error = dsl_destroy_head(name);
3297 /* There could be a hold on this dataset */
3298 if (error != EBUSY)
3299 ASSERT0(error);
3300 }
3301 return (0);
3302 }
3303
3304 static boolean_t
3305 ztest_snapshot_create(char *osname, uint64_t id)
3306 {
3307 char snapname[ZFS_MAX_DATASET_NAME_LEN];
3308 int error;
3309
3310 (void) snprintf(snapname, sizeof (snapname), "%llu", (u_longlong_t)id);
3311
3312 error = dmu_objset_snapshot_one(osname, snapname);
3313 if (error == ENOSPC) {
3314 ztest_record_enospc(FTAG);
3315 return (B_FALSE);
3316 }
3317 if (error != 0 && error != EEXIST) {
3318 fatal(0, "ztest_snapshot_create(%s@%s) = %d", osname,
3319 snapname, error);
3320 }
3321 return (B_TRUE);
3322 }
3323
3324 static boolean_t
3325 ztest_snapshot_destroy(char *osname, uint64_t id)
3326 {
3327 char snapname[ZFS_MAX_DATASET_NAME_LEN];
3328 int error;
3329
3330 (void) snprintf(snapname, sizeof (snapname), "%s@%llu", osname,
3331 (u_longlong_t)id);
3332
3333 error = dsl_destroy_snapshot(snapname, B_FALSE);
3334 if (error != 0 && error != ENOENT)
3335 fatal(0, "ztest_snapshot_destroy(%s) = %d", snapname, error);
3336 return (B_TRUE);
3337 }
3338
3339 /* ARGSUSED */
3340 void
3341 ztest_dmu_objset_create_destroy(ztest_ds_t *zd, uint64_t id)
3342 {
3343 ztest_ds_t zdtmp;
3344 int iters;
3345 int error;
3346 objset_t *os, *os2;
3347 char name[ZFS_MAX_DATASET_NAME_LEN];
3348 zilog_t *zilog;
3349
3350 rw_enter(&ztest_name_lock, RW_READER);
3351
3352 (void) snprintf(name, sizeof (name), "%s/temp_%llu",
3353 ztest_opts.zo_pool, (u_longlong_t)id);
3354
3355 /*
3356 * If this dataset exists from a previous run, process its replay log
3357 * half of the time. If we don't replay it, then dmu_objset_destroy()
3358 * (invoked from ztest_objset_destroy_cb()) should just throw it away.
3359 */
3360 if (ztest_random(2) == 0 &&
3361 dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os) == 0) {
3362 ztest_zd_init(&zdtmp, NULL, os);
3363 zil_replay(os, &zdtmp, ztest_replay_vector);
3364 ztest_zd_fini(&zdtmp);
3365 dmu_objset_disown(os, FTAG);
3366 }
3367
3368 /*
3369 * There may be an old instance of the dataset we're about to
3370 * create lying around from a previous run. If so, destroy it
3371 * and all of its snapshots.
3372 */
3373 (void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
3374 DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
3375
3376 /*
3377 * Verify that the destroyed dataset is no longer in the namespace.
3378 */
3379 VERIFY3U(ENOENT, ==, dmu_objset_own(name, DMU_OST_OTHER, B_TRUE,
3380 FTAG, &os));
3381
3382 /*
3383 * Verify that we can create a new dataset.
3384 */
3385 error = ztest_dataset_create(name);
3386 if (error) {
3387 if (error == ENOSPC) {
3388 ztest_record_enospc(FTAG);
3389 rw_exit(&ztest_name_lock);
3390 return;
3391 }
3392 fatal(0, "dmu_objset_create(%s) = %d", name, error);
3393 }
3394
3395 VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os));
3396
3397 ztest_zd_init(&zdtmp, NULL, os);
3398
3399 /*
3400 * Open the intent log for it.
3401 */
3402 zilog = zil_open(os, ztest_get_data);
3403
3404 /*
3405 * Put some objects in there, do a little I/O to them,
3406 * and randomly take a couple of snapshots along the way.
3407 */
3408 iters = ztest_random(5);
3409 for (int i = 0; i < iters; i++) {
3410 ztest_dmu_object_alloc_free(&zdtmp, id);
3411 if (ztest_random(iters) == 0)
3412 (void) ztest_snapshot_create(name, i);
3413 }
3414
3415 /*
3416 * Verify that we cannot create an existing dataset.
3417 */
3418 VERIFY3U(EEXIST, ==,
3419 dmu_objset_create(name, DMU_OST_OTHER, 0, NULL, NULL));
3420
3421 /*
3422 * Verify that we can hold an objset that is also owned.
3423 */
3424 VERIFY3U(0, ==, dmu_objset_hold(name, FTAG, &os2));
3425 dmu_objset_rele(os2, FTAG);
3426
3427 /*
3428 * Verify that we cannot own an objset that is already owned.
3429 */
3430 VERIFY3U(EBUSY, ==,
3431 dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os2));
3432
3433 zil_close(zilog);
3434 dmu_objset_disown(os, FTAG);
3435 ztest_zd_fini(&zdtmp);
3436
3437 rw_exit(&ztest_name_lock);
3438 }
3439
3440 /*
3441 * Verify that dmu_snapshot_{create,destroy,open,close} work as expected.
3442 */
3443 void
3444 ztest_dmu_snapshot_create_destroy(ztest_ds_t *zd, uint64_t id)
3445 {
3446 rw_enter(&ztest_name_lock, RW_READER);
3447 (void) ztest_snapshot_destroy(zd->zd_name, id);
3448 (void) ztest_snapshot_create(zd->zd_name, id);
3449 rw_exit(&ztest_name_lock);
3450 }
3451
3452 /*
3453 * Cleanup non-standard snapshots and clones.
3454 */
3455 void
3456 ztest_dsl_dataset_cleanup(char *osname, uint64_t id)
3457 {
3458 char snap1name[ZFS_MAX_DATASET_NAME_LEN];
3459 char clone1name[ZFS_MAX_DATASET_NAME_LEN];
3460 char snap2name[ZFS_MAX_DATASET_NAME_LEN];
3461 char clone2name[ZFS_MAX_DATASET_NAME_LEN];
3462 char snap3name[ZFS_MAX_DATASET_NAME_LEN];
3463 int error;
3464
3465 (void) snprintf(snap1name, sizeof (snap1name),
3466 "%s@s1_%llu", osname, id);
3467 (void) snprintf(clone1name, sizeof (clone1name),
3468 "%s/c1_%llu", osname, id);
3469 (void) snprintf(snap2name, sizeof (snap2name),
3470 "%s@s2_%llu", clone1name, id);
3471 (void) snprintf(clone2name, sizeof (clone2name),
3472 "%s/c2_%llu", osname, id);
3473 (void) snprintf(snap3name, sizeof (snap3name),
3474 "%s@s3_%llu", clone1name, id);
3475
3476 error = dsl_destroy_head(clone2name);
3477 if (error && error != ENOENT)
3478 fatal(0, "dsl_destroy_head(%s) = %d", clone2name, error);
3479 error = dsl_destroy_snapshot(snap3name, B_FALSE);
3480 if (error && error != ENOENT)
3481 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap3name, error);
3482 error = dsl_destroy_snapshot(snap2name, B_FALSE);
3483 if (error && error != ENOENT)
3484 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap2name, error);
3485 error = dsl_destroy_head(clone1name);
3486 if (error && error != ENOENT)
3487 fatal(0, "dsl_destroy_head(%s) = %d", clone1name, error);
3488 error = dsl_destroy_snapshot(snap1name, B_FALSE);
3489 if (error && error != ENOENT)
3490 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap1name, error);
3491 }
3492
3493 /*
3494 * Verify dsl_dataset_promote handles EBUSY
3495 */
3496 void
3497 ztest_dsl_dataset_promote_busy(ztest_ds_t *zd, uint64_t id)
3498 {
3499 objset_t *os;
3500 char snap1name[ZFS_MAX_DATASET_NAME_LEN];
3501 char clone1name[ZFS_MAX_DATASET_NAME_LEN];
3502 char snap2name[ZFS_MAX_DATASET_NAME_LEN];
3503 char clone2name[ZFS_MAX_DATASET_NAME_LEN];
3504 char snap3name[ZFS_MAX_DATASET_NAME_LEN];
3505 char *osname = zd->zd_name;
3506 int error;
3507
3508 rw_enter(&ztest_name_lock, RW_READER);
3509
3510 ztest_dsl_dataset_cleanup(osname, id);
3511
3512 (void) snprintf(snap1name, sizeof (snap1name),
3513 "%s@s1_%llu", osname, id);
3514 (void) snprintf(clone1name, sizeof (clone1name),
3515 "%s/c1_%llu", osname, id);
3516 (void) snprintf(snap2name, sizeof (snap2name),
3517 "%s@s2_%llu", clone1name, id);
3518 (void) snprintf(clone2name, sizeof (clone2name),
3519 "%s/c2_%llu", osname, id);
3520 (void) snprintf(snap3name, sizeof (snap3name),
3521 "%s@s3_%llu", clone1name, id);
3522
3523 error = dmu_objset_snapshot_one(osname, strchr(snap1name, '@') + 1);
3524 if (error && error != EEXIST) {
3525 if (error == ENOSPC) {
3526 ztest_record_enospc(FTAG);
3527 goto out;
3528 }
3529 fatal(0, "dmu_take_snapshot(%s) = %d", snap1name, error);
3530 }
3531
3532 error = dmu_objset_clone(clone1name, snap1name);
3533 if (error) {
3534 if (error == ENOSPC) {
3535 ztest_record_enospc(FTAG);
3536 goto out;
3537 }
3538 fatal(0, "dmu_objset_create(%s) = %d", clone1name, error);
3539 }
3540
3541 error = dmu_objset_snapshot_one(clone1name, strchr(snap2name, '@') + 1);
3542 if (error && error != EEXIST) {
3543 if (error == ENOSPC) {
3544 ztest_record_enospc(FTAG);
3545 goto out;
3546 }
3547 fatal(0, "dmu_open_snapshot(%s) = %d", snap2name, error);
3548 }
3549
3550 error = dmu_objset_snapshot_one(clone1name, strchr(snap3name, '@') + 1);
3551 if (error && error != EEXIST) {
3552 if (error == ENOSPC) {
3553 ztest_record_enospc(FTAG);
3554 goto out;
3555 }
3556 fatal(0, "dmu_open_snapshot(%s) = %d", snap3name, error);
3557 }
3558
3559 error = dmu_objset_clone(clone2name, snap3name);
3560 if (error) {
3561 if (error == ENOSPC) {
3562 ztest_record_enospc(FTAG);
3563 goto out;
3564 }
3565 fatal(0, "dmu_objset_create(%s) = %d", clone2name, error);
3566 }
3567
3568 error = dmu_objset_own(snap2name, DMU_OST_ANY, B_TRUE, FTAG, &os);
3569 if (error)
3570 fatal(0, "dmu_objset_own(%s) = %d", snap2name, error);
3571 error = dsl_dataset_promote(clone2name, NULL);
3572 if (error == ENOSPC) {
3573 dmu_objset_disown(os, FTAG);
3574 ztest_record_enospc(FTAG);
3575 goto out;
3576 }
3577 if (error != EBUSY)
3578 fatal(0, "dsl_dataset_promote(%s), %d, not EBUSY", clone2name,
3579 error);
3580 dmu_objset_disown(os, FTAG);
3581
3582 out:
3583 ztest_dsl_dataset_cleanup(osname, id);
3584
3585 rw_exit(&ztest_name_lock);
3586 }
3587
3588 /*
3589 * Verify that dmu_object_{alloc,free} work as expected.
3590 */
3591 void
3592 ztest_dmu_object_alloc_free(ztest_ds_t *zd, uint64_t id)
3593 {
3594 ztest_od_t od[4];
3595 int batchsize = sizeof (od) / sizeof (od[0]);
3596
3597 for (int b = 0; b < batchsize; b++)
3598 ztest_od_init(&od[b], id, FTAG, b, DMU_OT_UINT64_OTHER, 0, 0);
3599
3600 /*
3601 * Destroy the previous batch of objects, create a new batch,
3602 * and do some I/O on the new objects.
3603 */
3604 if (ztest_object_init(zd, od, sizeof (od), B_TRUE) != 0)
3605 return;
3606
3607 while (ztest_random(4 * batchsize) != 0)
3608 ztest_io(zd, od[ztest_random(batchsize)].od_object,
3609 ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
3610 }
3611
3612 /*
3613 * Verify that dmu_{read,write} work as expected.
3614 */
3615 void
3616 ztest_dmu_read_write(ztest_ds_t *zd, uint64_t id)
3617 {
3618 objset_t *os = zd->zd_os;
3619 ztest_od_t od[2];
3620 dmu_tx_t *tx;
3621 int i, freeit, error;
3622 uint64_t n, s, txg;
3623 bufwad_t *packbuf, *bigbuf, *pack, *bigH, *bigT;
3624 uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
3625 uint64_t chunksize = (1000 + ztest_random(1000)) * sizeof (uint64_t);
3626 uint64_t regions = 997;
3627 uint64_t stride = 123456789ULL;
3628 uint64_t width = 40;
3629 int free_percent = 5;
3630
3631 /*
3632 * This test uses two objects, packobj and bigobj, that are always
3633 * updated together (i.e. in the same tx) so that their contents are
3634 * in sync and can be compared. Their contents relate to each other
3635 * in a simple way: packobj is a dense array of 'bufwad' structures,
3636 * while bigobj is a sparse array of the same bufwads. Specifically,
3637 * for any index n, there are three bufwads that should be identical:
3638 *
3639 * packobj, at offset n * sizeof (bufwad_t)
3640 * bigobj, at the head of the nth chunk
3641 * bigobj, at the tail of the nth chunk
3642 *
3643 * The chunk size is arbitrary. It doesn't have to be a power of two,
3644 * and it doesn't have any relation to the object blocksize.
3645 * The only requirement is that it can hold at least two bufwads.
3646 *
3647 * Normally, we write the bufwad to each of these locations.
3648 * However, free_percent of the time we instead write zeroes to
3649 * packobj and perform a dmu_free_range() on bigobj. By comparing
3650 * bigobj to packobj, we can verify that the DMU is correctly
3651 * tracking which parts of an object are allocated and free,
3652 * and that the contents of the allocated blocks are correct.
3653 */
3654
3655 /*
3656 * Read the directory info. If it's the first time, set things up.
3657 */
3658 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, chunksize);
3659 ztest_od_init(&od[1], id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize);
3660
3661 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
3662 return;
3663
3664 bigobj = od[0].od_object;
3665 packobj = od[1].od_object;
3666 chunksize = od[0].od_gen;
3667 ASSERT(chunksize == od[1].od_gen);
3668
3669 /*
3670 * Prefetch a random chunk of the big object.
3671 * Our aim here is to get some async reads in flight
3672 * for blocks that we may free below; the DMU should
3673 * handle this race correctly.
3674 */
3675 n = ztest_random(regions) * stride + ztest_random(width);
3676 s = 1 + ztest_random(2 * width - 1);
3677 dmu_prefetch(os, bigobj, 0, n * chunksize, s * chunksize,
3678 ZIO_PRIORITY_SYNC_READ);
3679
3680 /*
3681 * Pick a random index and compute the offsets into packobj and bigobj.
3682 */
3683 n = ztest_random(regions) * stride + ztest_random(width);
3684 s = 1 + ztest_random(width - 1);
3685
3686 packoff = n * sizeof (bufwad_t);
3687 packsize = s * sizeof (bufwad_t);
3688
3689 bigoff = n * chunksize;
3690 bigsize = s * chunksize;
3691
3692 packbuf = umem_alloc(packsize, UMEM_NOFAIL);
3693 bigbuf = umem_alloc(bigsize, UMEM_NOFAIL);
3694
3695 /*
3696 * free_percent of the time, free a range of bigobj rather than
3697 * overwriting it.
3698 */
3699 freeit = (ztest_random(100) < free_percent);
3700
3701 /*
3702 * Read the current contents of our objects.
3703 */
3704 error = dmu_read(os, packobj, packoff, packsize, packbuf,
3705 DMU_READ_PREFETCH);
3706 ASSERT0(error);
3707 error = dmu_read(os, bigobj, bigoff, bigsize, bigbuf,
3708 DMU_READ_PREFETCH);
3709 ASSERT0(error);
3710
3711 /*
3712 * Get a tx for the mods to both packobj and bigobj.
3713 */
3714 tx = dmu_tx_create(os);
3715
3716 dmu_tx_hold_write(tx, packobj, packoff, packsize);
3717
3718 if (freeit)
3719 dmu_tx_hold_free(tx, bigobj, bigoff, bigsize);
3720 else
3721 dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
3722
3723 /* This accounts for setting the checksum/compression. */
3724 dmu_tx_hold_bonus(tx, bigobj);
3725
3726 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3727 if (txg == 0) {
3728 umem_free(packbuf, packsize);
3729 umem_free(bigbuf, bigsize);
3730 return;
3731 }
3732
3733 enum zio_checksum cksum;
3734 do {
3735 cksum = (enum zio_checksum)
3736 ztest_random_dsl_prop(ZFS_PROP_CHECKSUM);
3737 } while (cksum >= ZIO_CHECKSUM_LEGACY_FUNCTIONS);
3738 dmu_object_set_checksum(os, bigobj, cksum, tx);
3739
3740 enum zio_compress comp;
3741 do {
3742 comp = (enum zio_compress)
3743 ztest_random_dsl_prop(ZFS_PROP_COMPRESSION);
3744 } while (comp >= ZIO_COMPRESS_LEGACY_FUNCTIONS);
3745 dmu_object_set_compress(os, bigobj, comp, tx);
3746
3747 /*
3748 * For each index from n to n + s, verify that the existing bufwad
3749 * in packobj matches the bufwads at the head and tail of the
3750 * corresponding chunk in bigobj. Then update all three bufwads
3751 * with the new values we want to write out.
3752 */
3753 for (i = 0; i < s; i++) {
3754 /* LINTED */
3755 pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
3756 /* LINTED */
3757 bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
3758 /* LINTED */
3759 bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
3760
3761 ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
3762 ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
3763
3764 if (pack->bw_txg > txg)
3765 fatal(0, "future leak: got %llx, open txg is %llx",
3766 pack->bw_txg, txg);
3767
3768 if (pack->bw_data != 0 && pack->bw_index != n + i)
3769 fatal(0, "wrong index: got %llx, wanted %llx+%llx",
3770 pack->bw_index, n, i);
3771
3772 if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
3773 fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
3774
3775 if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
3776 fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
3777
3778 if (freeit) {
3779 bzero(pack, sizeof (bufwad_t));
3780 } else {
3781 pack->bw_index = n + i;
3782 pack->bw_txg = txg;
3783 pack->bw_data = 1 + ztest_random(-2ULL);
3784 }
3785 *bigH = *pack;
3786 *bigT = *pack;
3787 }
3788
3789 /*
3790 * We've verified all the old bufwads, and made new ones.
3791 * Now write them out.
3792 */
3793 dmu_write(os, packobj, packoff, packsize, packbuf, tx);
3794
3795 if (freeit) {
3796 if (ztest_opts.zo_verbose >= 7) {
3797 (void) printf("freeing offset %llx size %llx"
3798 " txg %llx\n",
3799 (u_longlong_t)bigoff,
3800 (u_longlong_t)bigsize,
3801 (u_longlong_t)txg);
3802 }
3803 VERIFY(0 == dmu_free_range(os, bigobj, bigoff, bigsize, tx));
3804 } else {
3805 if (ztest_opts.zo_verbose >= 7) {
3806 (void) printf("writing offset %llx size %llx"
3807 " txg %llx\n",
3808 (u_longlong_t)bigoff,
3809 (u_longlong_t)bigsize,
3810 (u_longlong_t)txg);
3811 }
3812 dmu_write(os, bigobj, bigoff, bigsize, bigbuf, tx);
3813 }
3814
3815 dmu_tx_commit(tx);
3816
3817 /*
3818 * Sanity check the stuff we just wrote.
3819 */
3820 {
3821 void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
3822 void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
3823
3824 VERIFY(0 == dmu_read(os, packobj, packoff,
3825 packsize, packcheck, DMU_READ_PREFETCH));
3826 VERIFY(0 == dmu_read(os, bigobj, bigoff,
3827 bigsize, bigcheck, DMU_READ_PREFETCH));
3828
3829 ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
3830 ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
3831
3832 umem_free(packcheck, packsize);
3833 umem_free(bigcheck, bigsize);
3834 }
3835
3836 umem_free(packbuf, packsize);
3837 umem_free(bigbuf, bigsize);
3838 }
3839
3840 void
3841 compare_and_update_pbbufs(uint64_t s, bufwad_t *packbuf, bufwad_t *bigbuf,
3842 uint64_t bigsize, uint64_t n, uint64_t chunksize, uint64_t txg)
3843 {
3844 uint64_t i;
3845 bufwad_t *pack;
3846 bufwad_t *bigH;
3847 bufwad_t *bigT;
3848
3849 /*
3850 * For each index from n to n + s, verify that the existing bufwad
3851 * in packobj matches the bufwads at the head and tail of the
3852 * corresponding chunk in bigobj. Then update all three bufwads
3853 * with the new values we want to write out.
3854 */
3855 for (i = 0; i < s; i++) {
3856 /* LINTED */
3857 pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
3858 /* LINTED */
3859 bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
3860 /* LINTED */
3861 bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
3862
3863 ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
3864 ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
3865
3866 if (pack->bw_txg > txg)
3867 fatal(0, "future leak: got %llx, open txg is %llx",
3868 pack->bw_txg, txg);
3869
3870 if (pack->bw_data != 0 && pack->bw_index != n + i)
3871 fatal(0, "wrong index: got %llx, wanted %llx+%llx",
3872 pack->bw_index, n, i);
3873
3874 if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
3875 fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
3876
3877 if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
3878 fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
3879
3880 pack->bw_index = n + i;
3881 pack->bw_txg = txg;
3882 pack->bw_data = 1 + ztest_random(-2ULL);
3883
3884 *bigH = *pack;
3885 *bigT = *pack;
3886 }
3887 }
3888
3889 void
3890 ztest_dmu_read_write_zcopy(ztest_ds_t *zd, uint64_t id)
3891 {
3892 objset_t *os = zd->zd_os;
3893 ztest_od_t od[2];
3894 dmu_tx_t *tx;
3895 uint64_t i;
3896 int error;
3897 uint64_t n, s, txg;
3898 bufwad_t *packbuf, *bigbuf;
3899 uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
3900 uint64_t blocksize = ztest_random_blocksize();
3901 uint64_t chunksize = blocksize;
3902 uint64_t regions = 997;
3903 uint64_t stride = 123456789ULL;
3904 uint64_t width = 9;
3905 dmu_buf_t *bonus_db;
3906 arc_buf_t **bigbuf_arcbufs;
3907 dmu_object_info_t doi;
3908
3909 /*
3910 * This test uses two objects, packobj and bigobj, that are always
3911 * updated together (i.e. in the same tx) so that their contents are
3912 * in sync and can be compared. Their contents relate to each other
3913 * in a simple way: packobj is a dense array of 'bufwad' structures,
3914 * while bigobj is a sparse array of the same bufwads. Specifically,
3915 * for any index n, there are three bufwads that should be identical:
3916 *
3917 * packobj, at offset n * sizeof (bufwad_t)
3918 * bigobj, at the head of the nth chunk
3919 * bigobj, at the tail of the nth chunk
3920 *
3921 * The chunk size is set equal to bigobj block size so that
3922 * dmu_assign_arcbuf() can be tested for object updates.
3923 */
3924
3925 /*
3926 * Read the directory info. If it's the first time, set things up.
3927 */
3928 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
3929 ztest_od_init(&od[1], id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize);
3930
3931 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
3932 return;
3933
3934 bigobj = od[0].od_object;
3935 packobj = od[1].od_object;
3936 blocksize = od[0].od_blocksize;
3937 chunksize = blocksize;
3938 ASSERT(chunksize == od[1].od_gen);
3939
3940 VERIFY(dmu_object_info(os, bigobj, &doi) == 0);
3941 VERIFY(ISP2(doi.doi_data_block_size));
3942 VERIFY(chunksize == doi.doi_data_block_size);
3943 VERIFY(chunksize >= 2 * sizeof (bufwad_t));
3944
3945 /*
3946 * Pick a random index and compute the offsets into packobj and bigobj.
3947 */
3948 n = ztest_random(regions) * stride + ztest_random(width);
3949 s = 1 + ztest_random(width - 1);
3950
3951 packoff = n * sizeof (bufwad_t);
3952 packsize = s * sizeof (bufwad_t);
3953
3954 bigoff = n * chunksize;
3955 bigsize = s * chunksize;
3956
3957 packbuf = umem_zalloc(packsize, UMEM_NOFAIL);
3958 bigbuf = umem_zalloc(bigsize, UMEM_NOFAIL);
3959
3960 VERIFY3U(0, ==, dmu_bonus_hold(os, bigobj, FTAG, &bonus_db));
3961
3962 bigbuf_arcbufs = umem_zalloc(2 * s * sizeof (arc_buf_t *), UMEM_NOFAIL);
3963
3964 /*
3965 * Iteration 0 test zcopy for DB_UNCACHED dbufs.
3966 * Iteration 1 test zcopy to already referenced dbufs.
3967 * Iteration 2 test zcopy to dirty dbuf in the same txg.
3968 * Iteration 3 test zcopy to dbuf dirty in previous txg.
3969 * Iteration 4 test zcopy when dbuf is no longer dirty.
3970 * Iteration 5 test zcopy when it can't be done.
3971 * Iteration 6 one more zcopy write.
3972 */
3973 for (i = 0; i < 7; i++) {
3974 uint64_t j;
3975 uint64_t off;
3976
3977 /*
3978 * In iteration 5 (i == 5) use arcbufs
3979 * that don't match bigobj blksz to test
3980 * dmu_assign_arcbuf() when it can't directly
3981 * assign an arcbuf to a dbuf.
3982 */
3983 for (j = 0; j < s; j++) {
3984 if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) {
3985 bigbuf_arcbufs[j] =
3986 dmu_request_arcbuf(bonus_db, chunksize);
3987 } else {
3988 bigbuf_arcbufs[2 * j] =
3989 dmu_request_arcbuf(bonus_db, chunksize / 2);
3990 bigbuf_arcbufs[2 * j + 1] =
3991 dmu_request_arcbuf(bonus_db, chunksize / 2);
3992 }
3993 }
3994
3995 /*
3996 * Get a tx for the mods to both packobj and bigobj.
3997 */
3998 tx = dmu_tx_create(os);
3999
4000 dmu_tx_hold_write(tx, packobj, packoff, packsize);
4001 dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
4002
4003 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4004 if (txg == 0) {
4005 umem_free(packbuf, packsize);
4006 umem_free(bigbuf, bigsize);
4007 for (j = 0; j < s; j++) {
4008 if (i != 5 ||
4009 chunksize < (SPA_MINBLOCKSIZE * 2)) {
4010 dmu_return_arcbuf(bigbuf_arcbufs[j]);
4011 } else {
4012 dmu_return_arcbuf(
4013 bigbuf_arcbufs[2 * j]);
4014 dmu_return_arcbuf(
4015 bigbuf_arcbufs[2 * j + 1]);
4016 }
4017 }
4018 umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
4019 dmu_buf_rele(bonus_db, FTAG);
4020 return;
4021 }
4022
4023 /*
4024 * 50% of the time don't read objects in the 1st iteration to
4025 * test dmu_assign_arcbuf() for the case when there're no
4026 * existing dbufs for the specified offsets.
4027 */
4028 if (i != 0 || ztest_random(2) != 0) {
4029 error = dmu_read(os, packobj, packoff,
4030 packsize, packbuf, DMU_READ_PREFETCH);
4031 ASSERT0(error);
4032 error = dmu_read(os, bigobj, bigoff, bigsize,
4033 bigbuf, DMU_READ_PREFETCH);
4034 ASSERT0(error);
4035 }
4036 compare_and_update_pbbufs(s, packbuf, bigbuf, bigsize,
4037 n, chunksize, txg);
4038
4039 /*
4040 * We've verified all the old bufwads, and made new ones.
4041 * Now write them out.
4042 */
4043 dmu_write(os, packobj, packoff, packsize, packbuf, tx);
4044 if (ztest_opts.zo_verbose >= 7) {
4045 (void) printf("writing offset %llx size %llx"
4046 " txg %llx\n",
4047 (u_longlong_t)bigoff,
4048 (u_longlong_t)bigsize,
4049 (u_longlong_t)txg);
4050 }
4051 for (off = bigoff, j = 0; j < s; j++, off += chunksize) {
4052 dmu_buf_t *dbt;
4053 if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) {
4054 bcopy((caddr_t)bigbuf + (off - bigoff),
4055 bigbuf_arcbufs[j]->b_data, chunksize);
4056 } else {
4057 bcopy((caddr_t)bigbuf + (off - bigoff),
4058 bigbuf_arcbufs[2 * j]->b_data,
4059 chunksize / 2);
4060 bcopy((caddr_t)bigbuf + (off - bigoff) +
4061 chunksize / 2,
4062 bigbuf_arcbufs[2 * j + 1]->b_data,
4063 chunksize / 2);
4064 }
4065
4066 if (i == 1) {
4067 VERIFY(dmu_buf_hold(os, bigobj, off,
4068 FTAG, &dbt, DMU_READ_NO_PREFETCH) == 0);
4069 }
4070 if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) {
4071 dmu_assign_arcbuf(bonus_db, off,
4072 bigbuf_arcbufs[j], tx);
4073 } else {
4074 dmu_assign_arcbuf(bonus_db, off,
4075 bigbuf_arcbufs[2 * j], tx);
4076 dmu_assign_arcbuf(bonus_db,
4077 off + chunksize / 2,
4078 bigbuf_arcbufs[2 * j + 1], tx);
4079 }
4080 if (i == 1) {
4081 dmu_buf_rele(dbt, FTAG);
4082 }
4083 }
4084 dmu_tx_commit(tx);
4085
4086 /*
4087 * Sanity check the stuff we just wrote.
4088 */
4089 {
4090 void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
4091 void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
4092
4093 VERIFY(0 == dmu_read(os, packobj, packoff,
4094 packsize, packcheck, DMU_READ_PREFETCH));
4095 VERIFY(0 == dmu_read(os, bigobj, bigoff,
4096 bigsize, bigcheck, DMU_READ_PREFETCH));
4097
4098 ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
4099 ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
4100
4101 umem_free(packcheck, packsize);
4102 umem_free(bigcheck, bigsize);
4103 }
4104 if (i == 2) {
4105 txg_wait_open(dmu_objset_pool(os), 0);
4106 } else if (i == 3) {
4107 txg_wait_synced(dmu_objset_pool(os), 0);
4108 }
4109 }
4110
4111 dmu_buf_rele(bonus_db, FTAG);
4112 umem_free(packbuf, packsize);
4113 umem_free(bigbuf, bigsize);
4114 umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
4115 }
4116
4117 /* ARGSUSED */
4118 void
4119 ztest_dmu_write_parallel(ztest_ds_t *zd, uint64_t id)
4120 {
4121 ztest_od_t od[1];
4122 uint64_t offset = (1ULL << (ztest_random(20) + 43)) +
4123 (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
4124
4125 /*
4126 * Have multiple threads write to large offsets in an object
4127 * to verify that parallel writes to an object -- even to the
4128 * same blocks within the object -- doesn't cause any trouble.
4129 */
4130 ztest_od_init(&od[0], ID_PARALLEL, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
4131
4132 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4133 return;
4134
4135 while (ztest_random(10) != 0)
4136 ztest_io(zd, od[0].od_object, offset);
4137 }
4138
4139 void
4140 ztest_dmu_prealloc(ztest_ds_t *zd, uint64_t id)
4141 {
4142 ztest_od_t od[1];
4143 uint64_t offset = (1ULL << (ztest_random(4) + SPA_MAXBLOCKSHIFT)) +
4144 (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
4145 uint64_t count = ztest_random(20) + 1;
4146 uint64_t blocksize = ztest_random_blocksize();
4147 void *data;
4148
4149 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
4150
4151 if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4152 return;
4153
4154 if (ztest_truncate(zd, od[0].od_object, offset, count * blocksize) != 0)
4155 return;
4156
4157 ztest_prealloc(zd, od[0].od_object, offset, count * blocksize);
4158
4159 data = umem_zalloc(blocksize, UMEM_NOFAIL);
4160
4161 while (ztest_random(count) != 0) {
4162 uint64_t randoff = offset + (ztest_random(count) * blocksize);
4163 if (ztest_write(zd, od[0].od_object, randoff, blocksize,
4164 data) != 0)
4165 break;
4166 while (ztest_random(4) != 0)
4167 ztest_io(zd, od[0].od_object, randoff);
4168 }
4169
4170 umem_free(data, blocksize);
4171 }
4172
4173 /*
4174 * Verify that zap_{create,destroy,add,remove,update} work as expected.
4175 */
4176 #define ZTEST_ZAP_MIN_INTS 1
4177 #define ZTEST_ZAP_MAX_INTS 4
4178 #define ZTEST_ZAP_MAX_PROPS 1000
4179
4180 void
4181 ztest_zap(ztest_ds_t *zd, uint64_t id)
4182 {
4183 objset_t *os = zd->zd_os;
4184 ztest_od_t od[1];
4185 uint64_t object;
4186 uint64_t txg, last_txg;
4187 uint64_t value[ZTEST_ZAP_MAX_INTS];
4188 uint64_t zl_ints, zl_intsize, prop;
4189 int i, ints;
4190 dmu_tx_t *tx;
4191 char propname[100], txgname[100];
4192 int error;
4193 char *hc[2] = { "s.acl.h", ".s.open.h.hyLZlg" };
4194
4195 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0);
4196
4197 if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4198 return;
4199
4200 object = od[0].od_object;
4201
4202 /*
4203 * Generate a known hash collision, and verify that
4204 * we can lookup and remove both entries.
4205 */
4206 tx = dmu_tx_create(os);
4207 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4208 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4209 if (txg == 0)
4210 return;
4211 for (i = 0; i < 2; i++) {
4212 value[i] = i;
4213 VERIFY3U(0, ==, zap_add(os, object, hc[i], sizeof (uint64_t),
4214 1, &value[i], tx));
4215 }
4216 for (i = 0; i < 2; i++) {
4217 VERIFY3U(EEXIST, ==, zap_add(os, object, hc[i],
4218 sizeof (uint64_t), 1, &value[i], tx));
4219 VERIFY3U(0, ==,
4220 zap_length(os, object, hc[i], &zl_intsize, &zl_ints));
4221 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4222 ASSERT3U(zl_ints, ==, 1);
4223 }
4224 for (i = 0; i < 2; i++) {
4225 VERIFY3U(0, ==, zap_remove(os, object, hc[i], tx));
4226 }
4227 dmu_tx_commit(tx);
4228
4229 /*
4230 * Generate a buch of random entries.
4231 */
4232 ints = MAX(ZTEST_ZAP_MIN_INTS, object % ZTEST_ZAP_MAX_INTS);
4233
4234 prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
4235 (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
4236 (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
4237 bzero(value, sizeof (value));
4238 last_txg = 0;
4239
4240 /*
4241 * If these zap entries already exist, validate their contents.
4242 */
4243 error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
4244 if (error == 0) {
4245 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4246 ASSERT3U(zl_ints, ==, 1);
4247
4248 VERIFY(zap_lookup(os, object, txgname, zl_intsize,
4249 zl_ints, &last_txg) == 0);
4250
4251 VERIFY(zap_length(os, object, propname, &zl_intsize,
4252 &zl_ints) == 0);
4253
4254 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4255 ASSERT3U(zl_ints, ==, ints);
4256
4257 VERIFY(zap_lookup(os, object, propname, zl_intsize,
4258 zl_ints, value) == 0);
4259
4260 for (i = 0; i < ints; i++) {
4261 ASSERT3U(value[i], ==, last_txg + object + i);
4262 }
4263 } else {
4264 ASSERT3U(error, ==, ENOENT);
4265 }
4266
4267 /*
4268 * Atomically update two entries in our zap object.
4269 * The first is named txg_%llu, and contains the txg
4270 * in which the property was last updated. The second
4271 * is named prop_%llu, and the nth element of its value
4272 * should be txg + object + n.
4273 */
4274 tx = dmu_tx_create(os);
4275 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4276 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4277 if (txg == 0)
4278 return;
4279
4280 if (last_txg > txg)
4281 fatal(0, "zap future leak: old %llu new %llu", last_txg, txg);
4282
4283 for (i = 0; i < ints; i++)
4284 value[i] = txg + object + i;
4285
4286 VERIFY3U(0, ==, zap_update(os, object, txgname, sizeof (uint64_t),
4287 1, &txg, tx));
4288 VERIFY3U(0, ==, zap_update(os, object, propname, sizeof (uint64_t),
4289 ints, value, tx));
4290
4291 dmu_tx_commit(tx);
4292
4293 /*
4294 * Remove a random pair of entries.
4295 */
4296 prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
4297 (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
4298 (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
4299
4300 error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
4301
4302 if (error == ENOENT)
4303 return;
4304
4305 ASSERT0(error);
4306
4307 tx = dmu_tx_create(os);
4308 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4309 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4310 if (txg == 0)
4311 return;
4312 VERIFY3U(0, ==, zap_remove(os, object, txgname, tx));
4313 VERIFY3U(0, ==, zap_remove(os, object, propname, tx));
4314 dmu_tx_commit(tx);
4315 }
4316
4317 /*
4318 * Testcase to test the upgrading of a microzap to fatzap.
4319 */
4320 void
4321 ztest_fzap(ztest_ds_t *zd, uint64_t id)
4322 {
4323 objset_t *os = zd->zd_os;
4324 ztest_od_t od[1];
4325 uint64_t object, txg;
4326
4327 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0);
4328
4329 if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4330 return;
4331
4332 object = od[0].od_object;
4333
4334 /*
4335 * Add entries to this ZAP and make sure it spills over
4336 * and gets upgraded to a fatzap. Also, since we are adding
4337 * 2050 entries we should see ptrtbl growth and leaf-block split.
4338 */
4339 for (int i = 0; i < 2050; i++) {
4340 char name[ZFS_MAX_DATASET_NAME_LEN];
4341 uint64_t value = i;
4342 dmu_tx_t *tx;
4343 int error;
4344
4345 (void) snprintf(name, sizeof (name), "fzap-%llu-%llu",
4346 id, value);
4347
4348 tx = dmu_tx_create(os);
4349 dmu_tx_hold_zap(tx, object, B_TRUE, name);
4350 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4351 if (txg == 0)
4352 return;
4353 error = zap_add(os, object, name, sizeof (uint64_t), 1,
4354 &value, tx);
4355 ASSERT(error == 0 || error == EEXIST);
4356 dmu_tx_commit(tx);
4357 }
4358 }
4359
4360 /* ARGSUSED */
4361 void
4362 ztest_zap_parallel(ztest_ds_t *zd, uint64_t id)
4363 {
4364 objset_t *os = zd->zd_os;
4365 ztest_od_t od[1];
4366 uint64_t txg, object, count, wsize, wc, zl_wsize, zl_wc;
4367 dmu_tx_t *tx;
4368 int i, namelen, error;
4369 int micro = ztest_random(2);
4370 char name[20], string_value[20];
4371 void *data;
4372
4373 ztest_od_init(&od[0], ID_PARALLEL, FTAG, micro, DMU_OT_ZAP_OTHER, 0, 0);
4374
4375 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4376 return;
4377
4378 object = od[0].od_object;
4379
4380 /*
4381 * Generate a random name of the form 'xxx.....' where each
4382 * x is a random printable character and the dots are dots.
4383 * There are 94 such characters, and the name length goes from
4384 * 6 to 20, so there are 94^3 * 15 = 12,458,760 possible names.
4385 */
4386 namelen = ztest_random(sizeof (name) - 5) + 5 + 1;
4387
4388 for (i = 0; i < 3; i++)
4389 name[i] = '!' + ztest_random('~' - '!' + 1);
4390 for (; i < namelen - 1; i++)
4391 name[i] = '.';
4392 name[i] = '\0';
4393
4394 if ((namelen & 1) || micro) {
4395 wsize = sizeof (txg);
4396 wc = 1;
4397 data = &txg;
4398 } else {
4399 wsize = 1;
4400 wc = namelen;
4401 data = string_value;
4402 }
4403
4404 count = -1ULL;
4405 VERIFY0(zap_count(os, object, &count));
4406 ASSERT(count != -1ULL);
4407
4408 /*
4409 * Select an operation: length, lookup, add, update, remove.
4410 */
4411 i = ztest_random(5);
4412
4413 if (i >= 2) {
4414 tx = dmu_tx_create(os);
4415 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4416 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4417 if (txg == 0)
4418 return;
4419 bcopy(name, string_value, namelen);
4420 } else {
4421 tx = NULL;
4422 txg = 0;
4423 bzero(string_value, namelen);
4424 }
4425
4426 switch (i) {
4427
4428 case 0:
4429 error = zap_length(os, object, name, &zl_wsize, &zl_wc);
4430 if (error == 0) {
4431 ASSERT3U(wsize, ==, zl_wsize);
4432 ASSERT3U(wc, ==, zl_wc);
4433 } else {
4434 ASSERT3U(error, ==, ENOENT);
4435 }
4436 break;
4437
4438 case 1:
4439 error = zap_lookup(os, object, name, wsize, wc, data);
4440 if (error == 0) {
4441 if (data == string_value &&
4442 bcmp(name, data, namelen) != 0)
4443 fatal(0, "name '%s' != val '%s' len %d",
4444 name, data, namelen);
4445 } else {
4446 ASSERT3U(error, ==, ENOENT);
4447 }
4448 break;
4449
4450 case 2:
4451 error = zap_add(os, object, name, wsize, wc, data, tx);
4452 ASSERT(error == 0 || error == EEXIST);
4453 break;
4454
4455 case 3:
4456 VERIFY(zap_update(os, object, name, wsize, wc, data, tx) == 0);
4457 break;
4458
4459 case 4:
4460 error = zap_remove(os, object, name, tx);
4461 ASSERT(error == 0 || error == ENOENT);
4462 break;
4463 }
4464
4465 if (tx != NULL)
4466 dmu_tx_commit(tx);
4467 }
4468
4469 /*
4470 * Commit callback data.
4471 */
4472 typedef struct ztest_cb_data {
4473 list_node_t zcd_node;
4474 uint64_t zcd_txg;
4475 int zcd_expected_err;
4476 boolean_t zcd_added;
4477 boolean_t zcd_called;
4478 spa_t *zcd_spa;
4479 } ztest_cb_data_t;
4480
4481 /* This is the actual commit callback function */
4482 static void
4483 ztest_commit_callback(void *arg, int error)
4484 {
4485 ztest_cb_data_t *data = arg;
4486 uint64_t synced_txg;
4487
4488 VERIFY(data != NULL);
4489 VERIFY3S(data->zcd_expected_err, ==, error);
4490 VERIFY(!data->zcd_called);
4491
4492 synced_txg = spa_last_synced_txg(data->zcd_spa);
4493 if (data->zcd_txg > synced_txg)
4494 fatal(0, "commit callback of txg %" PRIu64 " called prematurely"
4495 ", last synced txg = %" PRIu64 "\n", data->zcd_txg,
4496 synced_txg);
4497
4498 data->zcd_called = B_TRUE;
4499
4500 if (error == ECANCELED) {
4501 ASSERT0(data->zcd_txg);
4502 ASSERT(!data->zcd_added);
4503
4504 /*
4505 * The private callback data should be destroyed here, but
4506 * since we are going to check the zcd_called field after
4507 * dmu_tx_abort(), we will destroy it there.
4508 */
4509 return;
4510 }
4511
4512 /* Was this callback added to the global callback list? */
4513 if (!data->zcd_added)
4514 goto out;
4515
4516 ASSERT3U(data->zcd_txg, !=, 0);
4517
4518 /* Remove our callback from the list */
4519 mutex_enter(&zcl.zcl_callbacks_lock);
4520 list_remove(&zcl.zcl_callbacks, data);
4521 mutex_exit(&zcl.zcl_callbacks_lock);
4522
4523 out:
4524 umem_free(data, sizeof (ztest_cb_data_t));
4525 }
4526
4527 /* Allocate and initialize callback data structure */
4528 static ztest_cb_data_t *
4529 ztest_create_cb_data(objset_t *os, uint64_t txg)
4530 {
4531 ztest_cb_data_t *cb_data;
4532
4533 cb_data = umem_zalloc(sizeof (ztest_cb_data_t), UMEM_NOFAIL);
4534
4535 cb_data->zcd_txg = txg;
4536 cb_data->zcd_spa = dmu_objset_spa(os);
4537
4538 return (cb_data);
4539 }
4540
4541 /*
4542 * If a number of txgs equal to this threshold have been created after a commit
4543 * callback has been registered but not called, then we assume there is an
4544 * implementation bug.
4545 */
4546 #define ZTEST_COMMIT_CALLBACK_THRESH (TXG_CONCURRENT_STATES + 2)
4547
4548 /*
4549 * Commit callback test.
4550 */
4551 void
4552 ztest_dmu_commit_callbacks(ztest_ds_t *zd, uint64_t id)
4553 {
4554 objset_t *os = zd->zd_os;
4555 ztest_od_t od[1];
4556 dmu_tx_t *tx;
4557 ztest_cb_data_t *cb_data[3], *tmp_cb;
4558 uint64_t old_txg, txg;
4559 int i, error;
4560
4561 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
4562
4563 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4564 return;
4565
4566 tx = dmu_tx_create(os);
4567
4568 cb_data[0] = ztest_create_cb_data(os, 0);
4569 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[0]);
4570
4571 dmu_tx_hold_write(tx, od[0].od_object, 0, sizeof (uint64_t));
4572
4573 /* Every once in a while, abort the transaction on purpose */
4574 if (ztest_random(100) == 0)
4575 error = -1;
4576
4577 if (!error)
4578 error = dmu_tx_assign(tx, TXG_NOWAIT);
4579
4580 txg = error ? 0 : dmu_tx_get_txg(tx);
4581
4582 cb_data[0]->zcd_txg = txg;
4583 cb_data[1] = ztest_create_cb_data(os, txg);
4584 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[1]);
4585
4586 if (error) {
4587 /*
4588 * It's not a strict requirement to call the registered
4589 * callbacks from inside dmu_tx_abort(), but that's what
4590 * it's supposed to happen in the current implementation
4591 * so we will check for that.
4592 */
4593 for (i = 0; i < 2; i++) {
4594 cb_data[i]->zcd_expected_err = ECANCELED;
4595 VERIFY(!cb_data[i]->zcd_called);
4596 }
4597
4598 dmu_tx_abort(tx);
4599
4600 for (i = 0; i < 2; i++) {
4601 VERIFY(cb_data[i]->zcd_called);
4602 umem_free(cb_data[i], sizeof (ztest_cb_data_t));
4603 }
4604
4605 return;
4606 }
4607
4608 cb_data[2] = ztest_create_cb_data(os, txg);
4609 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[2]);
4610
4611 /*
4612 * Read existing data to make sure there isn't a future leak.
4613 */
4614 VERIFY(0 == dmu_read(os, od[0].od_object, 0, sizeof (uint64_t),
4615 &old_txg, DMU_READ_PREFETCH));
4616
4617 if (old_txg > txg)
4618 fatal(0, "future leak: got %" PRIu64 ", open txg is %" PRIu64,
4619 old_txg, txg);
4620
4621 dmu_write(os, od[0].od_object, 0, sizeof (uint64_t), &txg, tx);
4622
4623 mutex_enter(&zcl.zcl_callbacks_lock);
4624
4625 /*
4626 * Since commit callbacks don't have any ordering requirement and since
4627 * it is theoretically possible for a commit callback to be called
4628 * after an arbitrary amount of time has elapsed since its txg has been
4629 * synced, it is difficult to reliably determine whether a commit
4630 * callback hasn't been called due to high load or due to a flawed
4631 * implementation.
4632 *
4633 * In practice, we will assume that if after a certain number of txgs a
4634 * commit callback hasn't been called, then most likely there's an
4635 * implementation bug..
4636 */
4637 tmp_cb = list_head(&zcl.zcl_callbacks);
4638 if (tmp_cb != NULL &&
4639 (txg - ZTEST_COMMIT_CALLBACK_THRESH) > tmp_cb->zcd_txg) {
4640 fatal(0, "Commit callback threshold exceeded, oldest txg: %"
4641 PRIu64 ", open txg: %" PRIu64 "\n", tmp_cb->zcd_txg, txg);
4642 }
4643
4644 /*
4645 * Let's find the place to insert our callbacks.
4646 *
4647 * Even though the list is ordered by txg, it is possible for the
4648 * insertion point to not be the end because our txg may already be
4649 * quiescing at this point and other callbacks in the open txg
4650 * (from other objsets) may have sneaked in.
4651 */
4652 tmp_cb = list_tail(&zcl.zcl_callbacks);
4653 while (tmp_cb != NULL && tmp_cb->zcd_txg > txg)
4654 tmp_cb = list_prev(&zcl.zcl_callbacks, tmp_cb);
4655
4656 /* Add the 3 callbacks to the list */
4657 for (i = 0; i < 3; i++) {
4658 if (tmp_cb == NULL)
4659 list_insert_head(&zcl.zcl_callbacks, cb_data[i]);
4660 else
4661 list_insert_after(&zcl.zcl_callbacks, tmp_cb,
4662 cb_data[i]);
4663
4664 cb_data[i]->zcd_added = B_TRUE;
4665 VERIFY(!cb_data[i]->zcd_called);
4666
4667 tmp_cb = cb_data[i];
4668 }
4669
4670 mutex_exit(&zcl.zcl_callbacks_lock);
4671
4672 dmu_tx_commit(tx);
4673 }
4674
4675 /* ARGSUSED */
4676 void
4677 ztest_dsl_prop_get_set(ztest_ds_t *zd, uint64_t id)
4678 {
4679 zfs_prop_t proplist[] = {
4680 ZFS_PROP_CHECKSUM,
4681 ZFS_PROP_COMPRESSION,
4682 ZFS_PROP_COPIES,
4683 ZFS_PROP_DEDUP
4684 };
4685
4686 rw_enter(&ztest_name_lock, RW_READER);
4687
4688 for (int p = 0; p < sizeof (proplist) / sizeof (proplist[0]); p++)
4689 (void) ztest_dsl_prop_set_uint64(zd->zd_name, proplist[p],
4690 ztest_random_dsl_prop(proplist[p]), (int)ztest_random(2));
4691
4692 rw_exit(&ztest_name_lock);
4693 }
4694
4695 /* ARGSUSED */
4696 void
4697 ztest_remap_blocks(ztest_ds_t *zd, uint64_t id)
4698 {
4699 rw_enter(&ztest_name_lock, RW_READER);
4700
4701 int error = dmu_objset_remap_indirects(zd->zd_name);
4702 if (error == ENOSPC)
4703 error = 0;
4704 ASSERT0(error);
4705
4706 rw_exit(&ztest_name_lock);
4707 }
4708
4709 /* ARGSUSED */
4710 void
4711 ztest_spa_prop_get_set(ztest_ds_t *zd, uint64_t id)
4712 {
4713 nvlist_t *props = NULL;
4714
4715 rw_enter(&ztest_name_lock, RW_READER);
4716
4717 (void) ztest_spa_prop_set_uint64(ZPOOL_PROP_DEDUPDITTO,
4718 ZIO_DEDUPDITTO_MIN + ztest_random(ZIO_DEDUPDITTO_MIN));
4719
4720 VERIFY0(spa_prop_get(ztest_spa, &props));
4721
4722 if (ztest_opts.zo_verbose >= 6)
4723 dump_nvlist(props, 4);
4724
4725 nvlist_free(props);
4726
4727 rw_exit(&ztest_name_lock);
4728 }
4729
4730 static int
4731 user_release_one(const char *snapname, const char *holdname)
4732 {
4733 nvlist_t *snaps, *holds;
4734 int error;
4735
4736 snaps = fnvlist_alloc();
4737 holds = fnvlist_alloc();
4738 fnvlist_add_boolean(holds, holdname);
4739 fnvlist_add_nvlist(snaps, snapname, holds);
4740 fnvlist_free(holds);
4741 error = dsl_dataset_user_release(snaps, NULL);
4742 fnvlist_free(snaps);
4743 return (error);
4744 }
4745
4746 /*
4747 * Test snapshot hold/release and deferred destroy.
4748 */
4749 void
4750 ztest_dmu_snapshot_hold(ztest_ds_t *zd, uint64_t id)
4751 {
4752 int error;
4753 objset_t *os = zd->zd_os;
4754 objset_t *origin;
4755 char snapname[100];
4756 char fullname[100];
4757 char clonename[100];
4758 char tag[100];
4759 char osname[ZFS_MAX_DATASET_NAME_LEN];
4760 nvlist_t *holds;
4761
4762 rw_enter(&ztest_name_lock, RW_READER);
4763
4764 dmu_objset_name(os, osname);
4765
4766 (void) snprintf(snapname, sizeof (snapname), "sh1_%llu", id);
4767 (void) snprintf(fullname, sizeof (fullname), "%s@%s", osname, snapname);
4768 (void) snprintf(clonename, sizeof (clonename),
4769 "%s/ch1_%llu", osname, id);
4770 (void) snprintf(tag, sizeof (tag), "tag_%llu", id);
4771
4772 /*
4773 * Clean up from any previous run.
4774 */
4775 error = dsl_destroy_head(clonename);
4776 if (error != ENOENT)
4777 ASSERT0(error);
4778 error = user_release_one(fullname, tag);
4779 if (error != ESRCH && error != ENOENT)
4780 ASSERT0(error);
4781 error = dsl_destroy_snapshot(fullname, B_FALSE);
4782 if (error != ENOENT)
4783 ASSERT0(error);
4784
4785 /*
4786 * Create snapshot, clone it, mark snap for deferred destroy,
4787 * destroy clone, verify snap was also destroyed.
4788 */
4789 error = dmu_objset_snapshot_one(osname, snapname);
4790 if (error) {
4791 if (error == ENOSPC) {
4792 ztest_record_enospc("dmu_objset_snapshot");
4793 goto out;
4794 }
4795 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
4796 }
4797
4798 error = dmu_objset_clone(clonename, fullname);
4799 if (error) {
4800 if (error == ENOSPC) {
4801 ztest_record_enospc("dmu_objset_clone");
4802 goto out;
4803 }
4804 fatal(0, "dmu_objset_clone(%s) = %d", clonename, error);
4805 }
4806
4807 error = dsl_destroy_snapshot(fullname, B_TRUE);
4808 if (error) {
4809 fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
4810 fullname, error);
4811 }
4812
4813 error = dsl_destroy_head(clonename);
4814 if (error)
4815 fatal(0, "dsl_destroy_head(%s) = %d", clonename, error);
4816
4817 error = dmu_objset_hold(fullname, FTAG, &origin);
4818 if (error != ENOENT)
4819 fatal(0, "dmu_objset_hold(%s) = %d", fullname, error);
4820
4821 /*
4822 * Create snapshot, add temporary hold, verify that we can't
4823 * destroy a held snapshot, mark for deferred destroy,
4824 * release hold, verify snapshot was destroyed.
4825 */
4826 error = dmu_objset_snapshot_one(osname, snapname);
4827 if (error) {
4828 if (error == ENOSPC) {
4829 ztest_record_enospc("dmu_objset_snapshot");
4830 goto out;
4831 }
4832 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
4833 }
4834
4835 holds = fnvlist_alloc();
4836 fnvlist_add_string(holds, fullname, tag);
4837 error = dsl_dataset_user_hold(holds, 0, NULL);
4838 fnvlist_free(holds);
4839
4840 if (error == ENOSPC) {
4841 ztest_record_enospc("dsl_dataset_user_hold");
4842 goto out;
4843 } else if (error) {
4844 fatal(0, "dsl_dataset_user_hold(%s, %s) = %u",
4845 fullname, tag, error);
4846 }
4847
4848 error = dsl_destroy_snapshot(fullname, B_FALSE);
4849 if (error != EBUSY) {
4850 fatal(0, "dsl_destroy_snapshot(%s, B_FALSE) = %d",
4851 fullname, error);
4852 }
4853
4854 error = dsl_destroy_snapshot(fullname, B_TRUE);
4855 if (error) {
4856 fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
4857 fullname, error);
4858 }
4859
4860 error = user_release_one(fullname, tag);
4861 if (error)
4862 fatal(0, "user_release_one(%s, %s) = %d", fullname, tag, error);
4863
4864 VERIFY3U(dmu_objset_hold(fullname, FTAG, &origin), ==, ENOENT);
4865
4866 out:
4867 rw_exit(&ztest_name_lock);
4868 }
4869
4870 /*
4871 * Inject random faults into the on-disk data.
4872 */
4873 /* ARGSUSED */
4874 void
4875 ztest_fault_inject(ztest_ds_t *zd, uint64_t id)
4876 {
4877 ztest_shared_t *zs = ztest_shared;
4878 spa_t *spa = ztest_spa;
4879 int fd;
4880 uint64_t offset;
4881 uint64_t leaves;
4882 uint64_t bad = 0x1990c0ffeedecade;
4883 uint64_t top, leaf;
4884 char path0[MAXPATHLEN];
4885 char pathrand[MAXPATHLEN];
4886 size_t fsize;
4887 int bshift = SPA_MAXBLOCKSHIFT + 2;
4888 int iters = 1000;
4889 int maxfaults;
4890 int mirror_save;
4891 vdev_t *vd0 = NULL;
4892 uint64_t guid0 = 0;
4893 boolean_t islog = B_FALSE;
4894
4895 mutex_enter(&ztest_vdev_lock);
4896 maxfaults = MAXFAULTS();
4897 leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz;
4898 mirror_save = zs->zs_mirrors;
4899 mutex_exit(&ztest_vdev_lock);
4900
4901 ASSERT(leaves >= 1);
4902
4903 /*
4904 * Grab the name lock as reader. There are some operations
4905 * which don't like to have their vdevs changed while
4906 * they are in progress (i.e. spa_change_guid). Those
4907 * operations will have grabbed the name lock as writer.
4908 */
4909 rw_enter(&ztest_name_lock, RW_READER);
4910
4911 /*
4912 * We need SCL_STATE here because we're going to look at vd0->vdev_tsd.
4913 */
4914 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
4915
4916 if (ztest_random(2) == 0) {
4917 /*
4918 * Inject errors on a normal data device or slog device.
4919 */
4920 top = ztest_random_vdev_top(spa, B_TRUE);
4921 leaf = ztest_random(leaves) + zs->zs_splits;
4922
4923 /*
4924 * Generate paths to the first leaf in this top-level vdev,
4925 * and to the random leaf we selected. We'll induce transient
4926 * write failures and random online/offline activity on leaf 0,
4927 * and we'll write random garbage to the randomly chosen leaf.
4928 */
4929 (void) snprintf(path0, sizeof (path0), ztest_dev_template,
4930 ztest_opts.zo_dir, ztest_opts.zo_pool,
4931 top * leaves + zs->zs_splits);
4932 (void) snprintf(pathrand, sizeof (pathrand), ztest_dev_template,
4933 ztest_opts.zo_dir, ztest_opts.zo_pool,
4934 top * leaves + leaf);
4935
4936 vd0 = vdev_lookup_by_path(spa->spa_root_vdev, path0);
4937 if (vd0 != NULL && vd0->vdev_top->vdev_islog)
4938 islog = B_TRUE;
4939
4940 /*
4941 * If the top-level vdev needs to be resilvered
4942 * then we only allow faults on the device that is
4943 * resilvering.
4944 */
4945 if (vd0 != NULL && maxfaults != 1 &&
4946 (!vdev_resilver_needed(vd0->vdev_top, NULL, NULL) ||
4947 vd0->vdev_resilver_txg != 0)) {
4948 /*
4949 * Make vd0 explicitly claim to be unreadable,
4950 * or unwriteable, or reach behind its back
4951 * and close the underlying fd. We can do this if
4952 * maxfaults == 0 because we'll fail and reexecute,
4953 * and we can do it if maxfaults >= 2 because we'll
4954 * have enough redundancy. If maxfaults == 1, the
4955 * combination of this with injection of random data
4956 * corruption below exceeds the pool's fault tolerance.
4957 */
4958 vdev_file_t *vf = vd0->vdev_tsd;
4959
4960 zfs_dbgmsg("injecting fault to vdev %llu; maxfaults=%d",
4961 (long long)vd0->vdev_id, (int)maxfaults);
4962
4963 if (vf != NULL && ztest_random(3) == 0) {
4964 (void) close(vf->vf_vnode->v_fd);
4965 vf->vf_vnode->v_fd = -1;
4966 } else if (ztest_random(2) == 0) {
4967 vd0->vdev_cant_read = B_TRUE;
4968 } else {
4969 vd0->vdev_cant_write = B_TRUE;
4970 }
4971 guid0 = vd0->vdev_guid;
4972 }
4973 } else {
4974 /*
4975 * Inject errors on an l2cache device.
4976 */
4977 spa_aux_vdev_t *sav = &spa->spa_l2cache;
4978
4979 if (sav->sav_count == 0) {
4980 spa_config_exit(spa, SCL_STATE, FTAG);
4981 rw_exit(&ztest_name_lock);
4982 return;
4983 }
4984 vd0 = sav->sav_vdevs[ztest_random(sav->sav_count)];
4985 guid0 = vd0->vdev_guid;
4986 (void) strcpy(path0, vd0->vdev_path);
4987 (void) strcpy(pathrand, vd0->vdev_path);
4988
4989 leaf = 0;
4990 leaves = 1;
4991 maxfaults = INT_MAX; /* no limit on cache devices */
4992 }
4993
4994 spa_config_exit(spa, SCL_STATE, FTAG);
4995 rw_exit(&ztest_name_lock);
4996
4997 /*
4998 * If we can tolerate two or more faults, or we're dealing
4999 * with a slog, randomly online/offline vd0.
5000 */
5001 if ((maxfaults >= 2 || islog) && guid0 != 0) {
5002 if (ztest_random(10) < 6) {
5003 int flags = (ztest_random(2) == 0 ?
5004 ZFS_OFFLINE_TEMPORARY : 0);
5005
5006 /*
5007 * We have to grab the zs_name_lock as writer to
5008 * prevent a race between offlining a slog and
5009 * destroying a dataset. Offlining the slog will
5010 * grab a reference on the dataset which may cause
5011 * dmu_objset_destroy() to fail with EBUSY thus
5012 * leaving the dataset in an inconsistent state.
5013 */
5014 if (islog)
5015 rw_enter(&ztest_name_lock, RW_WRITER);
5016
5017 VERIFY(vdev_offline(spa, guid0, flags) != EBUSY);
5018
5019 if (islog)
5020 rw_exit(&ztest_name_lock);
5021 } else {
5022 /*
5023 * Ideally we would like to be able to randomly
5024 * call vdev_[on|off]line without holding locks
5025 * to force unpredictable failures but the side
5026 * effects of vdev_[on|off]line prevent us from
5027 * doing so. We grab the ztest_vdev_lock here to
5028 * prevent a race between injection testing and
5029 * aux_vdev removal.
5030 */
5031 mutex_enter(&ztest_vdev_lock);
5032 (void) vdev_online(spa, guid0, 0, NULL);
5033 mutex_exit(&ztest_vdev_lock);
5034 }
5035 }
5036
5037 if (maxfaults == 0)
5038 return;
5039
5040 /*
5041 * We have at least single-fault tolerance, so inject data corruption.
5042 */
5043 fd = open(pathrand, O_RDWR);
5044
5045 if (fd == -1) /* we hit a gap in the device namespace */
5046 return;
5047
5048 fsize = lseek(fd, 0, SEEK_END);
5049
5050 while (--iters != 0) {
5051 /*
5052 * The offset must be chosen carefully to ensure that
5053 * we do not inject a given logical block with errors
5054 * on two different leaf devices, because ZFS can not
5055 * tolerate that (if maxfaults==1).
5056 *
5057 * We divide each leaf into chunks of size
5058 * (# leaves * SPA_MAXBLOCKSIZE * 4). Within each chunk
5059 * there is a series of ranges to which we can inject errors.
5060 * Each range can accept errors on only a single leaf vdev.
5061 * The error injection ranges are separated by ranges
5062 * which we will not inject errors on any device (DMZs).
5063 * Each DMZ must be large enough such that a single block
5064 * can not straddle it, so that a single block can not be
5065 * a target in two different injection ranges (on different
5066 * leaf vdevs).
5067 *
5068 * For example, with 3 leaves, each chunk looks like:
5069 * 0 to 32M: injection range for leaf 0
5070 * 32M to 64M: DMZ - no injection allowed
5071 * 64M to 96M: injection range for leaf 1
5072 * 96M to 128M: DMZ - no injection allowed
5073 * 128M to 160M: injection range for leaf 2
5074 * 160M to 192M: DMZ - no injection allowed
5075 */
5076 offset = ztest_random(fsize / (leaves << bshift)) *
5077 (leaves << bshift) + (leaf << bshift) +
5078 (ztest_random(1ULL << (bshift - 1)) & -8ULL);
5079
5080 /*
5081 * Only allow damage to the labels at one end of the vdev.
5082 *
5083 * If all labels are damaged, the device will be totally
5084 * inaccessible, which will result in loss of data,
5085 * because we also damage (parts of) the other side of
5086 * the mirror/raidz.
5087 *
5088 * Additionally, we will always have both an even and an
5089 * odd label, so that we can handle crashes in the
5090 * middle of vdev_config_sync().
5091 */
5092 if ((leaf & 1) == 0 && offset < VDEV_LABEL_START_SIZE)
5093 continue;
5094
5095 /*
5096 * The two end labels are stored at the "end" of the disk, but
5097 * the end of the disk (vdev_psize) is aligned to
5098 * sizeof (vdev_label_t).
5099 */
5100 uint64_t psize = P2ALIGN(fsize, sizeof (vdev_label_t));
5101 if ((leaf & 1) == 1 &&
5102 offset + sizeof (bad) > psize - VDEV_LABEL_END_SIZE)
5103 continue;
5104
5105 mutex_enter(&ztest_vdev_lock);
5106 if (mirror_save != zs->zs_mirrors) {
5107 mutex_exit(&ztest_vdev_lock);
5108 (void) close(fd);
5109 return;
5110 }
5111
5112 if (pwrite(fd, &bad, sizeof (bad), offset) != sizeof (bad))
5113 fatal(1, "can't inject bad word at 0x%llx in %s",
5114 offset, pathrand);
5115
5116 mutex_exit(&ztest_vdev_lock);
5117
5118 if (ztest_opts.zo_verbose >= 7)
5119 (void) printf("injected bad word into %s,"
5120 " offset 0x%llx\n", pathrand, (u_longlong_t)offset);
5121 }
5122
5123 (void) close(fd);
5124 }
5125
5126 /*
5127 * Verify that DDT repair works as expected.
5128 */
5129 void
5130 ztest_ddt_repair(ztest_ds_t *zd, uint64_t id)
5131 {
5132 ztest_shared_t *zs = ztest_shared;
5133 spa_t *spa = ztest_spa;
5134 objset_t *os = zd->zd_os;
5135 ztest_od_t od[1];
5136 uint64_t object, blocksize, txg, pattern, psize;
5137 enum zio_checksum checksum = spa_dedup_checksum(spa);
5138 dmu_buf_t *db;
5139 dmu_tx_t *tx;
5140 abd_t *abd;
5141 blkptr_t blk;
5142 int copies = 2 * ZIO_DEDUPDITTO_MIN;
5143
5144 blocksize = ztest_random_blocksize();
5145 blocksize = MIN(blocksize, 2048); /* because we write so many */
5146
5147 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
5148
5149 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
5150 return;
5151
5152 /*
5153 * Take the name lock as writer to prevent anyone else from changing
5154 * the pool and dataset properies we need to maintain during this test.
5155 */
5156 rw_enter(&ztest_name_lock, RW_WRITER);
5157
5158 if (ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_DEDUP, checksum,
5159 B_FALSE) != 0 ||
5160 ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_COPIES, 1,
5161 B_FALSE) != 0) {
5162 rw_exit(&ztest_name_lock);
5163 return;
5164 }
5165
5166 dmu_objset_stats_t dds;
5167 dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
5168 dmu_objset_fast_stat(os, &dds);
5169 dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
5170
5171 object = od[0].od_object;
5172 blocksize = od[0].od_blocksize;
5173 pattern = zs->zs_guid ^ dds.dds_guid;
5174
5175 ASSERT(object != 0);
5176
5177 tx = dmu_tx_create(os);
5178 dmu_tx_hold_write(tx, object, 0, copies * blocksize);
5179 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
5180 if (txg == 0) {
5181 rw_exit(&ztest_name_lock);
5182 return;
5183 }
5184
5185 /*
5186 * Write all the copies of our block.
5187 */
5188 for (int i = 0; i < copies; i++) {
5189 uint64_t offset = i * blocksize;
5190 int error = dmu_buf_hold(os, object, offset, FTAG, &db,
5191 DMU_READ_NO_PREFETCH);
5192 if (error != 0) {
5193 fatal(B_FALSE, "dmu_buf_hold(%p, %llu, %llu) = %u",
5194 os, (long long)object, (long long) offset, error);
5195 }
5196 ASSERT(db->db_offset == offset);
5197 ASSERT(db->db_size == blocksize);
5198 ASSERT(ztest_pattern_match(db->db_data, db->db_size, pattern) ||
5199 ztest_pattern_match(db->db_data, db->db_size, 0ULL));
5200 dmu_buf_will_fill(db, tx);
5201 ztest_pattern_set(db->db_data, db->db_size, pattern);
5202 dmu_buf_rele(db, FTAG);
5203 }
5204
5205 dmu_tx_commit(tx);
5206 txg_wait_synced(spa_get_dsl(spa), txg);
5207
5208 /*
5209 * Find out what block we got.
5210 */
5211 VERIFY0(dmu_buf_hold(os, object, 0, FTAG, &db,
5212 DMU_READ_NO_PREFETCH));
5213 blk = *((dmu_buf_impl_t *)db)->db_blkptr;
5214 dmu_buf_rele(db, FTAG);
5215
5216 /*
5217 * Damage the block. Dedup-ditto will save us when we read it later.
5218 */
5219 psize = BP_GET_PSIZE(&blk);
5220 abd = abd_alloc_linear(psize, B_TRUE);
5221 ztest_pattern_set(abd_to_buf(abd), psize, ~pattern);
5222
5223 (void) zio_wait(zio_rewrite(NULL, spa, 0, &blk,
5224 abd, psize, NULL, NULL, ZIO_PRIORITY_SYNC_WRITE,
5225 ZIO_FLAG_CANFAIL | ZIO_FLAG_INDUCE_DAMAGE, NULL));
5226
5227 abd_free(abd);
5228
5229 rw_exit(&ztest_name_lock);
5230 }
5231
5232 /*
5233 * Scrub the pool.
5234 */
5235 /* ARGSUSED */
5236 void
5237 ztest_scrub(ztest_ds_t *zd, uint64_t id)
5238 {
5239 spa_t *spa = ztest_spa;
5240
5241 (void) spa_scan(spa, POOL_SCAN_SCRUB);
5242 (void) poll(NULL, 0, 100); /* wait a moment, then force a restart */
5243 (void) spa_scan(spa, POOL_SCAN_SCRUB);
5244 }
5245
5246 /*
5247 * Change the guid for the pool.
5248 */
5249 /* ARGSUSED */
5250 void
5251 ztest_reguid(ztest_ds_t *zd, uint64_t id)
5252 {
5253 spa_t *spa = ztest_spa;
5254 uint64_t orig, load;
5255 int error;
5256
5257 orig = spa_guid(spa);
5258 load = spa_load_guid(spa);
5259
5260 rw_enter(&ztest_name_lock, RW_WRITER);
5261 error = spa_change_guid(spa);
5262 rw_exit(&ztest_name_lock);
5263
5264 if (error != 0)
5265 return;
5266
5267 if (ztest_opts.zo_verbose >= 4) {
5268 (void) printf("Changed guid old %llu -> %llu\n",
5269 (u_longlong_t)orig, (u_longlong_t)spa_guid(spa));
5270 }
5271
5272 VERIFY3U(orig, !=, spa_guid(spa));
5273 VERIFY3U(load, ==, spa_load_guid(spa));
5274 }
5275
5276 /*
5277 * Rename the pool to a different name and then rename it back.
5278 */
5279 /* ARGSUSED */
5280 void
5281 ztest_spa_rename(ztest_ds_t *zd, uint64_t id)
5282 {
5283 char *oldname, *newname;
5284 spa_t *spa;
5285
5286 rw_enter(&ztest_name_lock, RW_WRITER);
5287
5288 oldname = ztest_opts.zo_pool;
5289 newname = umem_alloc(strlen(oldname) + 5, UMEM_NOFAIL);
5290 (void) strcpy(newname, oldname);
5291 (void) strcat(newname, "_tmp");
5292
5293 /*
5294 * Do the rename
5295 */
5296 VERIFY3U(0, ==, spa_rename(oldname, newname));
5297
5298 /*
5299 * Try to open it under the old name, which shouldn't exist
5300 */
5301 VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
5302
5303 /*
5304 * Open it under the new name and make sure it's still the same spa_t.
5305 */
5306 VERIFY3U(0, ==, spa_open(newname, &spa, FTAG));
5307
5308 ASSERT(spa == ztest_spa);
5309 spa_close(spa, FTAG);
5310
5311 /*
5312 * Rename it back to the original
5313 */
5314 VERIFY3U(0, ==, spa_rename(newname, oldname));
5315
5316 /*
5317 * Make sure it can still be opened
5318 */
5319 VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG));
5320
5321 ASSERT(spa == ztest_spa);
5322 spa_close(spa, FTAG);
5323
5324 umem_free(newname, strlen(newname) + 1);
5325
5326 rw_exit(&ztest_name_lock);
5327 }
5328
5329 /*
5330 * Verify pool integrity by running zdb.
5331 */
5332 static void
5333 ztest_run_zdb(char *pool)
5334 {
5335 int status;
5336 char zdb[MAXPATHLEN + MAXNAMELEN + 20];
5337 char zbuf[1024];
5338 char *bin;
5339 char *ztest;
5340 char *isa;
5341 int isalen;
5342 FILE *fp;
5343
5344 (void) realpath(getexecname(), zdb);
5345
5346 /* zdb lives in /usr/sbin, while ztest lives in /usr/bin */
5347 bin = strstr(zdb, "/usr/bin/");
5348 ztest = strstr(bin, "/ztest");
5349 isa = bin + 8;
5350 isalen = ztest - isa;
5351 isa = strdup(isa);
5352 /* LINTED */
5353 (void) sprintf(bin,
5354 "/usr/sbin%.*s/zdb -bcc%s%s -G -d -U %s %s",
5355 isalen,
5356 isa,
5357 ztest_opts.zo_verbose >= 3 ? "s" : "",
5358 ztest_opts.zo_verbose >= 4 ? "v" : "",
5359 spa_config_path,
5360 pool);
5361 free(isa);
5362
5363 if (ztest_opts.zo_verbose >= 5)
5364 (void) printf("Executing %s\n", strstr(zdb, "zdb "));
5365
5366 fp = popen(zdb, "r");
5367
5368 while (fgets(zbuf, sizeof (zbuf), fp) != NULL)
5369 if (ztest_opts.zo_verbose >= 3)
5370 (void) printf("%s", zbuf);
5371
5372 status = pclose(fp);
5373
5374 if (status == 0)
5375 return;
5376
5377 ztest_dump_core = 0;
5378 if (WIFEXITED(status))
5379 fatal(0, "'%s' exit code %d", zdb, WEXITSTATUS(status));
5380 else
5381 fatal(0, "'%s' died with signal %d", zdb, WTERMSIG(status));
5382 }
5383
5384 static void
5385 ztest_walk_pool_directory(char *header)
5386 {
5387 spa_t *spa = NULL;
5388
5389 if (ztest_opts.zo_verbose >= 6)
5390 (void) printf("%s\n", header);
5391
5392 mutex_enter(&spa_namespace_lock);
5393 while ((spa = spa_next(spa)) != NULL)
5394 if (ztest_opts.zo_verbose >= 6)
5395 (void) printf("\t%s\n", spa_name(spa));
5396 mutex_exit(&spa_namespace_lock);
5397 }
5398
5399 static void
5400 ztest_spa_import_export(char *oldname, char *newname)
5401 {
5402 nvlist_t *config, *newconfig;
5403 uint64_t pool_guid;
5404 spa_t *spa;
5405 int error;
5406
5407 if (ztest_opts.zo_verbose >= 4) {
5408 (void) printf("import/export: old = %s, new = %s\n",
5409 oldname, newname);
5410 }
5411
5412 /*
5413 * Clean up from previous runs.
5414 */
5415 (void) spa_destroy(newname);
5416
5417 /*
5418 * Get the pool's configuration and guid.
5419 */
5420 VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG));
5421
5422 /*
5423 * Kick off a scrub to tickle scrub/export races.
5424 */
5425 if (ztest_random(2) == 0)
5426 (void) spa_scan(spa, POOL_SCAN_SCRUB);
5427
5428 pool_guid = spa_guid(spa);
5429 spa_close(spa, FTAG);
5430
5431 ztest_walk_pool_directory("pools before export");
5432
5433 /*
5434 * Export it.
5435 */
5436 VERIFY3U(0, ==, spa_export(oldname, &config, B_FALSE, B_FALSE));
5437
5438 ztest_walk_pool_directory("pools after export");
5439
5440 /*
5441 * Try to import it.
5442 */
5443 newconfig = spa_tryimport(config);
5444 ASSERT(newconfig != NULL);
5445 nvlist_free(newconfig);
5446
5447 /*
5448 * Import it under the new name.
5449 */
5450 error = spa_import(newname, config, NULL, 0);
5451 if (error != 0) {
5452 dump_nvlist(config, 0);
5453 fatal(B_FALSE, "couldn't import pool %s as %s: error %u",
5454 oldname, newname, error);
5455 }
5456
5457 ztest_walk_pool_directory("pools after import");
5458
5459 /*
5460 * Try to import it again -- should fail with EEXIST.
5461 */
5462 VERIFY3U(EEXIST, ==, spa_import(newname, config, NULL, 0));
5463
5464 /*
5465 * Try to import it under a different name -- should fail with EEXIST.
5466 */
5467 VERIFY3U(EEXIST, ==, spa_import(oldname, config, NULL, 0));
5468
5469 /*
5470 * Verify that the pool is no longer visible under the old name.
5471 */
5472 VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
5473
5474 /*
5475 * Verify that we can open and close the pool using the new name.
5476 */
5477 VERIFY3U(0, ==, spa_open(newname, &spa, FTAG));
5478 ASSERT(pool_guid == spa_guid(spa));
5479 spa_close(spa, FTAG);
5480
5481 nvlist_free(config);
5482 }
5483
5484 static void
5485 ztest_resume(spa_t *spa)
5486 {
5487 if (spa_suspended(spa) && ztest_opts.zo_verbose >= 6)
5488 (void) printf("resuming from suspended state\n");
5489 spa_vdev_state_enter(spa, SCL_NONE);
5490 vdev_clear(spa, NULL);
5491 (void) spa_vdev_state_exit(spa, NULL, 0);
5492 (void) zio_resume(spa);
5493 }
5494
5495 static void *
5496 ztest_resume_thread(void *arg)
5497 {
5498 spa_t *spa = arg;
5499
5500 while (!ztest_exiting) {
5501 if (spa_suspended(spa))
5502 ztest_resume(spa);
5503 (void) poll(NULL, 0, 100);
5504
5505 /*
5506 * Periodically change the zfs_compressed_arc_enabled setting.
5507 */
5508 if (ztest_random(10) == 0)
5509 zfs_compressed_arc_enabled = ztest_random(2);
5510
5511 /*
5512 * Periodically change the zfs_abd_scatter_enabled setting.
5513 */
5514 if (ztest_random(10) == 0)
5515 zfs_abd_scatter_enabled = ztest_random(2);
5516 }
5517 return (NULL);
5518 }
5519
5520 static void *
5521 ztest_deadman_thread(void *arg)
5522 {
5523 ztest_shared_t *zs = arg;
5524 spa_t *spa = ztest_spa;
5525 hrtime_t delta, total = 0;
5526
5527 for (;;) {
5528 delta = zs->zs_thread_stop - zs->zs_thread_start +
5529 MSEC2NSEC(zfs_deadman_synctime_ms);
5530
5531 (void) poll(NULL, 0, (int)NSEC2MSEC(delta));
5532
5533 /*
5534 * If the pool is suspended then fail immediately. Otherwise,
5535 * check to see if the pool is making any progress. If
5536 * vdev_deadman() discovers that there hasn't been any recent
5537 * I/Os then it will end up aborting the tests.
5538 */
5539 if (spa_suspended(spa) || spa->spa_root_vdev == NULL) {
5540 fatal(0, "aborting test after %llu seconds because "
5541 "pool has transitioned to a suspended state.",
5542 zfs_deadman_synctime_ms / 1000);
5543 return (NULL);
5544 }
5545 vdev_deadman(spa->spa_root_vdev);
5546
5547 total += zfs_deadman_synctime_ms/1000;
5548 (void) printf("ztest has been running for %lld seconds\n",
5549 total);
5550 }
5551 }
5552
5553 static void
5554 ztest_execute(int test, ztest_info_t *zi, uint64_t id)
5555 {
5556 ztest_ds_t *zd = &ztest_ds[id % ztest_opts.zo_datasets];
5557 ztest_shared_callstate_t *zc = ZTEST_GET_SHARED_CALLSTATE(test);
5558 hrtime_t functime = gethrtime();
5559
5560 for (int i = 0; i < zi->zi_iters; i++)
5561 zi->zi_func(zd, id);
5562
5563 functime = gethrtime() - functime;
5564
5565 atomic_add_64(&zc->zc_count, 1);
5566 atomic_add_64(&zc->zc_time, functime);
5567
5568 if (ztest_opts.zo_verbose >= 4) {
5569 Dl_info dli;
5570 (void) dladdr((void *)zi->zi_func, &dli);
5571 (void) printf("%6.2f sec in %s\n",
5572 (double)functime / NANOSEC, dli.dli_sname);
5573 }
5574 }
5575
5576 static void *
5577 ztest_thread(void *arg)
5578 {
5579 int rand;
5580 uint64_t id = (uintptr_t)arg;
5581 ztest_shared_t *zs = ztest_shared;
5582 uint64_t call_next;
5583 hrtime_t now;
5584 ztest_info_t *zi;
5585 ztest_shared_callstate_t *zc;
5586
5587 while ((now = gethrtime()) < zs->zs_thread_stop) {
5588 /*
5589 * See if it's time to force a crash.
5590 */
5591 if (now > zs->zs_thread_kill)
5592 ztest_kill(zs);
5593
5594 /*
5595 * If we're getting ENOSPC with some regularity, stop.
5596 */
5597 if (zs->zs_enospc_count > 10)
5598 break;
5599
5600 /*
5601 * Pick a random function to execute.
5602 */
5603 rand = ztest_random(ZTEST_FUNCS);
5604 zi = &ztest_info[rand];
5605 zc = ZTEST_GET_SHARED_CALLSTATE(rand);
5606 call_next = zc->zc_next;
5607
5608 if (now >= call_next &&
5609 atomic_cas_64(&zc->zc_next, call_next, call_next +
5610 ztest_random(2 * zi->zi_interval[0] + 1)) == call_next) {
5611 ztest_execute(rand, zi, id);
5612 }
5613 }
5614
5615 return (NULL);
5616 }
5617
5618 static void
5619 ztest_dataset_name(char *dsname, char *pool, int d)
5620 {
5621 (void) snprintf(dsname, ZFS_MAX_DATASET_NAME_LEN, "%s/ds_%d", pool, d);
5622 }
5623
5624 static void
5625 ztest_dataset_destroy(int d)
5626 {
5627 char name[ZFS_MAX_DATASET_NAME_LEN];
5628
5629 ztest_dataset_name(name, ztest_opts.zo_pool, d);
5630
5631 if (ztest_opts.zo_verbose >= 3)
5632 (void) printf("Destroying %s to free up space\n", name);
5633
5634 /*
5635 * Cleanup any non-standard clones and snapshots. In general,
5636 * ztest thread t operates on dataset (t % zopt_datasets),
5637 * so there may be more than one thing to clean up.
5638 */
5639 for (int t = d; t < ztest_opts.zo_threads;
5640 t += ztest_opts.zo_datasets) {
5641 ztest_dsl_dataset_cleanup(name, t);
5642 }
5643
5644 (void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
5645 DS_FIND_SNAPSHOTS | DS_FIND_CHILDREN);
5646 }
5647
5648 static void
5649 ztest_dataset_dirobj_verify(ztest_ds_t *zd)
5650 {
5651 uint64_t usedobjs, dirobjs, scratch;
5652
5653 /*
5654 * ZTEST_DIROBJ is the object directory for the entire dataset.
5655 * Therefore, the number of objects in use should equal the
5656 * number of ZTEST_DIROBJ entries, +1 for ZTEST_DIROBJ itself.
5657 * If not, we have an object leak.
5658 *
5659 * Note that we can only check this in ztest_dataset_open(),
5660 * when the open-context and syncing-context values agree.
5661 * That's because zap_count() returns the open-context value,
5662 * while dmu_objset_space() returns the rootbp fill count.
5663 */
5664 VERIFY3U(0, ==, zap_count(zd->zd_os, ZTEST_DIROBJ, &dirobjs));
5665 dmu_objset_space(zd->zd_os, &scratch, &scratch, &usedobjs, &scratch);
5666 ASSERT3U(dirobjs + 1, ==, usedobjs);
5667 }
5668
5669 static int
5670 ztest_dataset_open(int d)
5671 {
5672 ztest_ds_t *zd = &ztest_ds[d];
5673 uint64_t committed_seq = ZTEST_GET_SHARED_DS(d)->zd_seq;
5674 objset_t *os;
5675 zilog_t *zilog;
5676 char name[ZFS_MAX_DATASET_NAME_LEN];
5677 int error;
5678
5679 ztest_dataset_name(name, ztest_opts.zo_pool, d);
5680
5681 rw_enter(&ztest_name_lock, RW_READER);
5682
5683 error = ztest_dataset_create(name);
5684 if (error == ENOSPC) {
5685 rw_exit(&ztest_name_lock);
5686 ztest_record_enospc(FTAG);
5687 return (error);
5688 }
5689 ASSERT(error == 0 || error == EEXIST);
5690
5691 VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, zd, &os));
5692 rw_exit(&ztest_name_lock);
5693
5694 ztest_zd_init(zd, ZTEST_GET_SHARED_DS(d), os);
5695
5696 zilog = zd->zd_zilog;
5697
5698 if (zilog->zl_header->zh_claim_lr_seq != 0 &&
5699 zilog->zl_header->zh_claim_lr_seq < committed_seq)
5700 fatal(0, "missing log records: claimed %llu < committed %llu",
5701 zilog->zl_header->zh_claim_lr_seq, committed_seq);
5702
5703 ztest_dataset_dirobj_verify(zd);
5704
5705 zil_replay(os, zd, ztest_replay_vector);
5706
5707 ztest_dataset_dirobj_verify(zd);
5708
5709 if (ztest_opts.zo_verbose >= 6)
5710 (void) printf("%s replay %llu blocks, %llu records, seq %llu\n",
5711 zd->zd_name,
5712 (u_longlong_t)zilog->zl_parse_blk_count,
5713 (u_longlong_t)zilog->zl_parse_lr_count,
5714 (u_longlong_t)zilog->zl_replaying_seq);
5715
5716 zilog = zil_open(os, ztest_get_data);
5717
5718 if (zilog->zl_replaying_seq != 0 &&
5719 zilog->zl_replaying_seq < committed_seq)
5720 fatal(0, "missing log records: replayed %llu < committed %llu",
5721 zilog->zl_replaying_seq, committed_seq);
5722
5723 return (0);
5724 }
5725
5726 static void
5727 ztest_dataset_close(int d)
5728 {
5729 ztest_ds_t *zd = &ztest_ds[d];
5730
5731 zil_close(zd->zd_zilog);
5732 dmu_objset_disown(zd->zd_os, zd);
5733
5734 ztest_zd_fini(zd);
5735 }
5736
5737 /*
5738 * Kick off threads to run tests on all datasets in parallel.
5739 */
5740 static void
5741 ztest_run(ztest_shared_t *zs)
5742 {
5743 thread_t *tid;
5744 spa_t *spa;
5745 objset_t *os;
5746 thread_t resume_tid;
5747 int error;
5748
5749 ztest_exiting = B_FALSE;
5750
5751 /*
5752 * Initialize parent/child shared state.
5753 */
5754 mutex_init(&ztest_vdev_lock, NULL, USYNC_THREAD, NULL);
5755 rw_init(&ztest_name_lock, NULL, USYNC_THREAD, NULL);
5756
5757 zs->zs_thread_start = gethrtime();
5758 zs->zs_thread_stop =
5759 zs->zs_thread_start + ztest_opts.zo_passtime * NANOSEC;
5760 zs->zs_thread_stop = MIN(zs->zs_thread_stop, zs->zs_proc_stop);
5761 zs->zs_thread_kill = zs->zs_thread_stop;
5762 if (ztest_random(100) < ztest_opts.zo_killrate) {
5763 zs->zs_thread_kill -=
5764 ztest_random(ztest_opts.zo_passtime * NANOSEC);
5765 }
5766
5767 mutex_init(&zcl.zcl_callbacks_lock, NULL, USYNC_THREAD, NULL);
5768
5769 list_create(&zcl.zcl_callbacks, sizeof (ztest_cb_data_t),
5770 offsetof(ztest_cb_data_t, zcd_node));
5771
5772 /*
5773 * Open our pool.
5774 */
5775 kernel_init(FREAD | FWRITE);
5776 VERIFY0(spa_open(ztest_opts.zo_pool, &spa, FTAG));
5777 spa->spa_debug = B_TRUE;
5778 metaslab_preload_limit = ztest_random(20) + 1;
5779 ztest_spa = spa;
5780
5781 dmu_objset_stats_t dds;
5782 VERIFY0(dmu_objset_own(ztest_opts.zo_pool,
5783 DMU_OST_ANY, B_TRUE, FTAG, &os));
5784 dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
5785 dmu_objset_fast_stat(os, &dds);
5786 dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
5787 zs->zs_guid = dds.dds_guid;
5788 dmu_objset_disown(os, FTAG);
5789
5790 spa->spa_dedup_ditto = 2 * ZIO_DEDUPDITTO_MIN;
5791
5792 /*
5793 * We don't expect the pool to suspend unless maxfaults == 0,
5794 * in which case ztest_fault_inject() temporarily takes away
5795 * the only valid replica.
5796 */
5797 if (MAXFAULTS() == 0)
5798 spa->spa_failmode = ZIO_FAILURE_MODE_WAIT;
5799 else
5800 spa->spa_failmode = ZIO_FAILURE_MODE_PANIC;
5801
5802 /*
5803 * Create a thread to periodically resume suspended I/O.
5804 */
5805 VERIFY(thr_create(0, 0, ztest_resume_thread, spa, THR_BOUND,
5806 &resume_tid) == 0);
5807
5808 /*
5809 * Create a deadman thread to abort() if we hang.
5810 */
5811 VERIFY(thr_create(0, 0, ztest_deadman_thread, zs, THR_BOUND,
5812 NULL) == 0);
5813
5814 /*
5815 * Verify that we can safely inquire about about any object,
5816 * whether it's allocated or not. To make it interesting,
5817 * we probe a 5-wide window around each power of two.
5818 * This hits all edge cases, including zero and the max.
5819 */
5820 for (int t = 0; t < 64; t++) {
5821 for (int d = -5; d <= 5; d++) {
5822 error = dmu_object_info(spa->spa_meta_objset,
5823 (1ULL << t) + d, NULL);
5824 ASSERT(error == 0 || error == ENOENT ||
5825 error == EINVAL);
5826 }
5827 }
5828
5829 /*
5830 * If we got any ENOSPC errors on the previous run, destroy something.
5831 */
5832 if (zs->zs_enospc_count != 0) {
5833 int d = ztest_random(ztest_opts.zo_datasets);
5834 ztest_dataset_destroy(d);
5835 }
5836 zs->zs_enospc_count = 0;
5837
5838 tid = umem_zalloc(ztest_opts.zo_threads * sizeof (thread_t),
5839 UMEM_NOFAIL);
5840
5841 if (ztest_opts.zo_verbose >= 4)
5842 (void) printf("starting main threads...\n");
5843
5844 /*
5845 * Kick off all the tests that run in parallel.
5846 */
5847 for (int t = 0; t < ztest_opts.zo_threads; t++) {
5848 if (t < ztest_opts.zo_datasets &&
5849 ztest_dataset_open(t) != 0)
5850 return;
5851 VERIFY(thr_create(0, 0, ztest_thread, (void *)(uintptr_t)t,
5852 THR_BOUND, &tid[t]) == 0);
5853 }
5854
5855 /*
5856 * Wait for all of the tests to complete. We go in reverse order
5857 * so we don't close datasets while threads are still using them.
5858 */
5859 for (int t = ztest_opts.zo_threads - 1; t >= 0; t--) {
5860 VERIFY(thr_join(tid[t], NULL, NULL) == 0);
5861 if (t < ztest_opts.zo_datasets)
5862 ztest_dataset_close(t);
5863 }
5864
5865 txg_wait_synced(spa_get_dsl(spa), 0);
5866
5867 zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(spa));
5868 zs->zs_space = metaslab_class_get_space(spa_normal_class(spa));
5869 zfs_dbgmsg_print(FTAG);
5870
5871 umem_free(tid, ztest_opts.zo_threads * sizeof (thread_t));
5872
5873 /* Kill the resume thread */
5874 ztest_exiting = B_TRUE;
5875 VERIFY(thr_join(resume_tid, NULL, NULL) == 0);
5876 ztest_resume(spa);
5877
5878 /*
5879 * Right before closing the pool, kick off a bunch of async I/O;
5880 * spa_close() should wait for it to complete.
5881 */
5882 for (uint64_t object = 1; object < 50; object++) {
5883 dmu_prefetch(spa->spa_meta_objset, object, 0, 0, 1ULL << 20,
5884 ZIO_PRIORITY_SYNC_READ);
5885 }
5886
5887 spa_close(spa, FTAG);
5888
5889 /*
5890 * Verify that we can loop over all pools.
5891 */
5892 mutex_enter(&spa_namespace_lock);
5893 for (spa = spa_next(NULL); spa != NULL; spa = spa_next(spa))
5894 if (ztest_opts.zo_verbose > 3)
5895 (void) printf("spa_next: found %s\n", spa_name(spa));
5896 mutex_exit(&spa_namespace_lock);
5897
5898 /*
5899 * Verify that we can export the pool and reimport it under a
5900 * different name.
5901 */
5902 if (ztest_random(2) == 0) {
5903 char name[ZFS_MAX_DATASET_NAME_LEN];
5904 (void) snprintf(name, sizeof (name), "%s_import",
5905 ztest_opts.zo_pool);
5906 ztest_spa_import_export(ztest_opts.zo_pool, name);
5907 ztest_spa_import_export(name, ztest_opts.zo_pool);
5908 }
5909
5910 kernel_fini();
5911
5912 list_destroy(&zcl.zcl_callbacks);
5913
5914 mutex_destroy(&zcl.zcl_callbacks_lock);
5915
5916 rw_destroy(&ztest_name_lock);
5917 mutex_destroy(&ztest_vdev_lock);
5918 }
5919
5920 static void
5921 ztest_freeze(void)
5922 {
5923 ztest_ds_t *zd = &ztest_ds[0];
5924 spa_t *spa;
5925 int numloops = 0;
5926
5927 if (ztest_opts.zo_verbose >= 3)
5928 (void) printf("testing spa_freeze()...\n");
5929
5930 kernel_init(FREAD | FWRITE);
5931 VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
5932 VERIFY3U(0, ==, ztest_dataset_open(0));
5933 spa->spa_debug = B_TRUE;
5934 ztest_spa = spa;
5935
5936 /*
5937 * Force the first log block to be transactionally allocated.
5938 * We have to do this before we freeze the pool -- otherwise
5939 * the log chain won't be anchored.
5940 */
5941 while (BP_IS_HOLE(&zd->zd_zilog->zl_header->zh_log)) {
5942 ztest_dmu_object_alloc_free(zd, 0);
5943 zil_commit(zd->zd_zilog, 0);
5944 }
5945
5946 txg_wait_synced(spa_get_dsl(spa), 0);
5947
5948 /*
5949 * Freeze the pool. This stops spa_sync() from doing anything,
5950 * so that the only way to record changes from now on is the ZIL.
5951 */
5952 spa_freeze(spa);
5953
5954 /*
5955 * Because it is hard to predict how much space a write will actually
5956 * require beforehand, we leave ourselves some fudge space to write over
5957 * capacity.
5958 */
5959 uint64_t capacity = metaslab_class_get_space(spa_normal_class(spa)) / 2;
5960
5961 /*
5962 * Run tests that generate log records but don't alter the pool config
5963 * or depend on DSL sync tasks (snapshots, objset create/destroy, etc).
5964 * We do a txg_wait_synced() after each iteration to force the txg
5965 * to increase well beyond the last synced value in the uberblock.
5966 * The ZIL should be OK with that.
5967 *
5968 * Run a random number of times less than zo_maxloops and ensure we do
5969 * not run out of space on the pool.
5970 */
5971 while (ztest_random(10) != 0 &&
5972 numloops++ < ztest_opts.zo_maxloops &&
5973 metaslab_class_get_alloc(spa_normal_class(spa)) < capacity) {
5974 ztest_od_t od;
5975 ztest_od_init(&od, 0, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
5976 VERIFY0(ztest_object_init(zd, &od, sizeof (od), B_FALSE));
5977 ztest_io(zd, od.od_object,
5978 ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
5979 txg_wait_synced(spa_get_dsl(spa), 0);
5980 }
5981
5982 /*
5983 * Commit all of the changes we just generated.
5984 */
5985 zil_commit(zd->zd_zilog, 0);
5986 txg_wait_synced(spa_get_dsl(spa), 0);
5987
5988 /*
5989 * Close our dataset and close the pool.
5990 */
5991 ztest_dataset_close(0);
5992 spa_close(spa, FTAG);
5993 kernel_fini();
5994
5995 /*
5996 * Open and close the pool and dataset to induce log replay.
5997 */
5998 kernel_init(FREAD | FWRITE);
5999 VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
6000 ASSERT(spa_freeze_txg(spa) == UINT64_MAX);
6001 VERIFY3U(0, ==, ztest_dataset_open(0));
6002 ztest_dataset_close(0);
6003
6004 spa->spa_debug = B_TRUE;
6005 ztest_spa = spa;
6006 txg_wait_synced(spa_get_dsl(spa), 0);
6007 ztest_reguid(NULL, 0);
6008
6009 spa_close(spa, FTAG);
6010 kernel_fini();
6011 }
6012
6013 void
6014 print_time(hrtime_t t, char *timebuf)
6015 {
6016 hrtime_t s = t / NANOSEC;
6017 hrtime_t m = s / 60;
6018 hrtime_t h = m / 60;
6019 hrtime_t d = h / 24;
6020
6021 s -= m * 60;
6022 m -= h * 60;
6023 h -= d * 24;
6024
6025 timebuf[0] = '\0';
6026
6027 if (d)
6028 (void) sprintf(timebuf,
6029 "%llud%02lluh%02llum%02llus", d, h, m, s);
6030 else if (h)
6031 (void) sprintf(timebuf, "%lluh%02llum%02llus", h, m, s);
6032 else if (m)
6033 (void) sprintf(timebuf, "%llum%02llus", m, s);
6034 else
6035 (void) sprintf(timebuf, "%llus", s);
6036 }
6037
6038 static nvlist_t *
6039 make_random_props()
6040 {
6041 nvlist_t *props;
6042
6043 VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
6044 if (ztest_random(2) == 0)
6045 return (props);
6046 VERIFY(nvlist_add_uint64(props, "autoreplace", 1) == 0);
6047
6048 return (props);
6049 }
6050
6051 /*
6052 * Create a storage pool with the given name and initial vdev size.
6053 * Then test spa_freeze() functionality.
6054 */
6055 static void
6056 ztest_init(ztest_shared_t *zs)
6057 {
6058 spa_t *spa;
6059 nvlist_t *nvroot, *props;
6060
6061 mutex_init(&ztest_vdev_lock, NULL, USYNC_THREAD, NULL);
6062 rw_init(&ztest_name_lock, NULL, USYNC_THREAD, NULL);
6063
6064 kernel_init(FREAD | FWRITE);
6065
6066 /*
6067 * Create the storage pool.
6068 */
6069 (void) spa_destroy(ztest_opts.zo_pool);
6070 ztest_shared->zs_vdev_next_leaf = 0;
6071 zs->zs_splits = 0;
6072 zs->zs_mirrors = ztest_opts.zo_mirrors;
6073 nvroot = make_vdev_root(NULL, NULL, NULL, ztest_opts.zo_vdev_size, 0,
6074 0, ztest_opts.zo_raidz, zs->zs_mirrors, 1);
6075 props = make_random_props();
6076 for (int i = 0; i < SPA_FEATURES; i++) {
6077 char buf[1024];
6078 (void) snprintf(buf, sizeof (buf), "feature@%s",
6079 spa_feature_table[i].fi_uname);
6080 VERIFY3U(0, ==, nvlist_add_uint64(props, buf, 0));
6081 }
6082 VERIFY3U(0, ==, spa_create(ztest_opts.zo_pool, nvroot, props, NULL));
6083 nvlist_free(nvroot);
6084 nvlist_free(props);
6085
6086 VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
6087 zs->zs_metaslab_sz =
6088 1ULL << spa->spa_root_vdev->vdev_child[0]->vdev_ms_shift;
6089
6090 spa_close(spa, FTAG);
6091
6092 kernel_fini();
6093
6094 ztest_run_zdb(ztest_opts.zo_pool);
6095
6096 ztest_freeze();
6097
6098 ztest_run_zdb(ztest_opts.zo_pool);
6099
6100 rw_destroy(&ztest_name_lock);
6101 mutex_destroy(&ztest_vdev_lock);
6102 }
6103
6104 static void
6105 setup_data_fd(void)
6106 {
6107 static char ztest_name_data[] = "/tmp/ztest.data.XXXXXX";
6108
6109 ztest_fd_data = mkstemp(ztest_name_data);
6110 ASSERT3S(ztest_fd_data, >=, 0);
6111 (void) unlink(ztest_name_data);
6112 }
6113
6114
6115 static int
6116 shared_data_size(ztest_shared_hdr_t *hdr)
6117 {
6118 int size;
6119
6120 size = hdr->zh_hdr_size;
6121 size += hdr->zh_opts_size;
6122 size += hdr->zh_size;
6123 size += hdr->zh_stats_size * hdr->zh_stats_count;
6124 size += hdr->zh_ds_size * hdr->zh_ds_count;
6125
6126 return (size);
6127 }
6128
6129 static void
6130 setup_hdr(void)
6131 {
6132 int size;
6133 ztest_shared_hdr_t *hdr;
6134
6135 hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
6136 PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
6137 ASSERT(hdr != MAP_FAILED);
6138
6139 VERIFY3U(0, ==, ftruncate(ztest_fd_data, sizeof (ztest_shared_hdr_t)));
6140
6141 hdr->zh_hdr_size = sizeof (ztest_shared_hdr_t);
6142 hdr->zh_opts_size = sizeof (ztest_shared_opts_t);
6143 hdr->zh_size = sizeof (ztest_shared_t);
6144 hdr->zh_stats_size = sizeof (ztest_shared_callstate_t);
6145 hdr->zh_stats_count = ZTEST_FUNCS;
6146 hdr->zh_ds_size = sizeof (ztest_shared_ds_t);
6147 hdr->zh_ds_count = ztest_opts.zo_datasets;
6148
6149 size = shared_data_size(hdr);
6150 VERIFY3U(0, ==, ftruncate(ztest_fd_data, size));
6151
6152 (void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
6153 }
6154
6155 static void
6156 setup_data(void)
6157 {
6158 int size, offset;
6159 ztest_shared_hdr_t *hdr;
6160 uint8_t *buf;
6161
6162 hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
6163 PROT_READ, MAP_SHARED, ztest_fd_data, 0);
6164 ASSERT(hdr != MAP_FAILED);
6165
6166 size = shared_data_size(hdr);
6167
6168 (void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
6169 hdr = ztest_shared_hdr = (void *)mmap(0, P2ROUNDUP(size, getpagesize()),
6170 PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
6171 ASSERT(hdr != MAP_FAILED);
6172 buf = (uint8_t *)hdr;
6173
6174 offset = hdr->zh_hdr_size;
6175 ztest_shared_opts = (void *)&buf[offset];
6176 offset += hdr->zh_opts_size;
6177 ztest_shared = (void *)&buf[offset];
6178 offset += hdr->zh_size;
6179 ztest_shared_callstate = (void *)&buf[offset];
6180 offset += hdr->zh_stats_size * hdr->zh_stats_count;
6181 ztest_shared_ds = (void *)&buf[offset];
6182 }
6183
6184 static boolean_t
6185 exec_child(char *cmd, char *libpath, boolean_t ignorekill, int *statusp)
6186 {
6187 pid_t pid;
6188 int status;
6189 char *cmdbuf = NULL;
6190
6191 pid = fork();
6192
6193 if (cmd == NULL) {
6194 cmdbuf = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
6195 (void) strlcpy(cmdbuf, getexecname(), MAXPATHLEN);
6196 cmd = cmdbuf;
6197 }
6198
6199 if (pid == -1)
6200 fatal(1, "fork failed");
6201
6202 if (pid == 0) { /* child */
6203 char *emptyargv[2] = { cmd, NULL };
6204 char fd_data_str[12];
6205
6206 struct rlimit rl = { 1024, 1024 };
6207 (void) setrlimit(RLIMIT_NOFILE, &rl);
6208
6209 (void) close(ztest_fd_rand);
6210 VERIFY3U(11, >=,
6211 snprintf(fd_data_str, 12, "%d", ztest_fd_data));
6212 VERIFY0(setenv("ZTEST_FD_DATA", fd_data_str, 1));
6213
6214 (void) enable_extended_FILE_stdio(-1, -1);
6215 if (libpath != NULL)
6216 VERIFY(0 == setenv("LD_LIBRARY_PATH", libpath, 1));
6217 (void) execv(cmd, emptyargv);
6218 ztest_dump_core = B_FALSE;
6219 fatal(B_TRUE, "exec failed: %s", cmd);
6220 }
6221
6222 if (cmdbuf != NULL) {
6223 umem_free(cmdbuf, MAXPATHLEN);
6224 cmd = NULL;
6225 }
6226
6227 while (waitpid(pid, &status, 0) != pid)
6228 continue;
6229 if (statusp != NULL)
6230 *statusp = status;
6231
6232 if (WIFEXITED(status)) {
6233 if (WEXITSTATUS(status) != 0) {
6234 (void) fprintf(stderr, "child exited with code %d\n",
6235 WEXITSTATUS(status));
6236 exit(2);
6237 }
6238 return (B_FALSE);
6239 } else if (WIFSIGNALED(status)) {
6240 if (!ignorekill || WTERMSIG(status) != SIGKILL) {
6241 (void) fprintf(stderr, "child died with signal %d\n",
6242 WTERMSIG(status));
6243 exit(3);
6244 }
6245 return (B_TRUE);
6246 } else {
6247 (void) fprintf(stderr, "something strange happened to child\n");
6248 exit(4);
6249 /* NOTREACHED */
6250 }
6251 }
6252
6253 static void
6254 ztest_run_init(void)
6255 {
6256 ztest_shared_t *zs = ztest_shared;
6257
6258 ASSERT(ztest_opts.zo_init != 0);
6259
6260 /*
6261 * Blow away any existing copy of zpool.cache
6262 */
6263 (void) remove(spa_config_path);
6264
6265 /*
6266 * Create and initialize our storage pool.
6267 */
6268 for (int i = 1; i <= ztest_opts.zo_init; i++) {
6269 bzero(zs, sizeof (ztest_shared_t));
6270 if (ztest_opts.zo_verbose >= 3 &&
6271 ztest_opts.zo_init != 1) {
6272 (void) printf("ztest_init(), pass %d\n", i);
6273 }
6274 ztest_init(zs);
6275 }
6276 }
6277
6278 int
6279 main(int argc, char **argv)
6280 {
6281 int kills = 0;
6282 int iters = 0;
6283 int older = 0;
6284 int newer = 0;
6285 ztest_shared_t *zs;
6286 ztest_info_t *zi;
6287 ztest_shared_callstate_t *zc;
6288 char timebuf[100];
6289 char numbuf[NN_NUMBUF_SZ];
6290 spa_t *spa;
6291 char *cmd;
6292 boolean_t hasalt;
6293 char *fd_data_str = getenv("ZTEST_FD_DATA");
6294
6295 (void) setvbuf(stdout, NULL, _IOLBF, 0);
6296
6297 dprintf_setup(&argc, argv);
6298 zfs_deadman_synctime_ms = 300000;
6299
6300 ztest_fd_rand = open("/dev/urandom", O_RDONLY);
6301 ASSERT3S(ztest_fd_rand, >=, 0);
6302
6303 if (!fd_data_str) {
6304 process_options(argc, argv);
6305
6306 setup_data_fd();
6307 setup_hdr();
6308 setup_data();
6309 bcopy(&ztest_opts, ztest_shared_opts,
6310 sizeof (*ztest_shared_opts));
6311 } else {
6312 ztest_fd_data = atoi(fd_data_str);
6313 setup_data();
6314 bcopy(ztest_shared_opts, &ztest_opts, sizeof (ztest_opts));
6315 }
6316 ASSERT3U(ztest_opts.zo_datasets, ==, ztest_shared_hdr->zh_ds_count);
6317
6318 /* Override location of zpool.cache */
6319 VERIFY3U(asprintf((char **)&spa_config_path, "%s/zpool.cache",
6320 ztest_opts.zo_dir), !=, -1);
6321
6322 ztest_ds = umem_alloc(ztest_opts.zo_datasets * sizeof (ztest_ds_t),
6323 UMEM_NOFAIL);
6324 zs = ztest_shared;
6325
6326 if (fd_data_str) {
6327 metaslab_gang_bang = ztest_opts.zo_metaslab_gang_bang;
6328 metaslab_df_alloc_threshold =
6329 zs->zs_metaslab_df_alloc_threshold;
6330
6331 if (zs->zs_do_init)
6332 ztest_run_init();
6333 else
6334 ztest_run(zs);
6335 exit(0);
6336 }
6337
6338 hasalt = (strlen(ztest_opts.zo_alt_ztest) != 0);
6339
6340 if (ztest_opts.zo_verbose >= 1) {
6341 (void) printf("%llu vdevs, %d datasets, %d threads,"
6342 " %llu seconds...\n",
6343 (u_longlong_t)ztest_opts.zo_vdevs,
6344 ztest_opts.zo_datasets,
6345 ztest_opts.zo_threads,
6346 (u_longlong_t)ztest_opts.zo_time);
6347 }
6348
6349 cmd = umem_alloc(MAXNAMELEN, UMEM_NOFAIL);
6350 (void) strlcpy(cmd, getexecname(), MAXNAMELEN);
6351
6352 zs->zs_do_init = B_TRUE;
6353 if (strlen(ztest_opts.zo_alt_ztest) != 0) {
6354 if (ztest_opts.zo_verbose >= 1) {
6355 (void) printf("Executing older ztest for "
6356 "initialization: %s\n", ztest_opts.zo_alt_ztest);
6357 }
6358 VERIFY(!exec_child(ztest_opts.zo_alt_ztest,
6359 ztest_opts.zo_alt_libpath, B_FALSE, NULL));
6360 } else {
6361 VERIFY(!exec_child(NULL, NULL, B_FALSE, NULL));
6362 }
6363 zs->zs_do_init = B_FALSE;
6364
6365 zs->zs_proc_start = gethrtime();
6366 zs->zs_proc_stop = zs->zs_proc_start + ztest_opts.zo_time * NANOSEC;
6367
6368 for (int f = 0; f < ZTEST_FUNCS; f++) {
6369 zi = &ztest_info[f];
6370 zc = ZTEST_GET_SHARED_CALLSTATE(f);
6371 if (zs->zs_proc_start + zi->zi_interval[0] > zs->zs_proc_stop)
6372 zc->zc_next = UINT64_MAX;
6373 else
6374 zc->zc_next = zs->zs_proc_start +
6375 ztest_random(2 * zi->zi_interval[0] + 1);
6376 }
6377
6378 /*
6379 * Run the tests in a loop. These tests include fault injection
6380 * to verify that self-healing data works, and forced crashes
6381 * to verify that we never lose on-disk consistency.
6382 */
6383 while (gethrtime() < zs->zs_proc_stop) {
6384 int status;
6385 boolean_t killed;
6386
6387 /*
6388 * Initialize the workload counters for each function.
6389 */
6390 for (int f = 0; f < ZTEST_FUNCS; f++) {
6391 zc = ZTEST_GET_SHARED_CALLSTATE(f);
6392 zc->zc_count = 0;
6393 zc->zc_time = 0;
6394 }
6395
6396 /* Set the allocation switch size */
6397 zs->zs_metaslab_df_alloc_threshold =
6398 ztest_random(zs->zs_metaslab_sz / 4) + 1;
6399
6400 if (!hasalt || ztest_random(2) == 0) {
6401 if (hasalt && ztest_opts.zo_verbose >= 1) {
6402 (void) printf("Executing newer ztest: %s\n",
6403 cmd);
6404 }
6405 newer++;
6406 killed = exec_child(cmd, NULL, B_TRUE, &status);
6407 } else {
6408 if (hasalt && ztest_opts.zo_verbose >= 1) {
6409 (void) printf("Executing older ztest: %s\n",
6410 ztest_opts.zo_alt_ztest);
6411 }
6412 older++;
6413 killed = exec_child(ztest_opts.zo_alt_ztest,
6414 ztest_opts.zo_alt_libpath, B_TRUE, &status);
6415 }
6416
6417 if (killed)
6418 kills++;
6419 iters++;
6420
6421 if (ztest_opts.zo_verbose >= 1) {
6422 hrtime_t now = gethrtime();
6423
6424 now = MIN(now, zs->zs_proc_stop);
6425 print_time(zs->zs_proc_stop - now, timebuf);
6426 nicenum(zs->zs_space, numbuf, sizeof (numbuf));
6427
6428 (void) printf("Pass %3d, %8s, %3llu ENOSPC, "
6429 "%4.1f%% of %5s used, %3.0f%% done, %8s to go\n",
6430 iters,
6431 WIFEXITED(status) ? "Complete" : "SIGKILL",
6432 (u_longlong_t)zs->zs_enospc_count,
6433 100.0 * zs->zs_alloc / zs->zs_space,
6434 numbuf,
6435 100.0 * (now - zs->zs_proc_start) /
6436 (ztest_opts.zo_time * NANOSEC), timebuf);
6437 }
6438
6439 if (ztest_opts.zo_verbose >= 2) {
6440 (void) printf("\nWorkload summary:\n\n");
6441 (void) printf("%7s %9s %s\n",
6442 "Calls", "Time", "Function");
6443 (void) printf("%7s %9s %s\n",
6444 "-----", "----", "--------");
6445 for (int f = 0; f < ZTEST_FUNCS; f++) {
6446 Dl_info dli;
6447
6448 zi = &ztest_info[f];
6449 zc = ZTEST_GET_SHARED_CALLSTATE(f);
6450 print_time(zc->zc_time, timebuf);
6451 (void) dladdr((void *)zi->zi_func, &dli);
6452 (void) printf("%7llu %9s %s\n",
6453 (u_longlong_t)zc->zc_count, timebuf,
6454 dli.dli_sname);
6455 }
6456 (void) printf("\n");
6457 }
6458
6459 /*
6460 * It's possible that we killed a child during a rename test,
6461 * in which case we'll have a 'ztest_tmp' pool lying around
6462 * instead of 'ztest'. Do a blind rename in case this happened.
6463 */
6464 kernel_init(FREAD);
6465 if (spa_open(ztest_opts.zo_pool, &spa, FTAG) == 0) {
6466 spa_close(spa, FTAG);
6467 } else {
6468 char tmpname[ZFS_MAX_DATASET_NAME_LEN];
6469 kernel_fini();
6470 kernel_init(FREAD | FWRITE);
6471 (void) snprintf(tmpname, sizeof (tmpname), "%s_tmp",
6472 ztest_opts.zo_pool);
6473 (void) spa_rename(tmpname, ztest_opts.zo_pool);
6474 }
6475 kernel_fini();
6476
6477 ztest_run_zdb(ztest_opts.zo_pool);
6478 }
6479
6480 if (ztest_opts.zo_verbose >= 1) {
6481 if (hasalt) {
6482 (void) printf("%d runs of older ztest: %s\n", older,
6483 ztest_opts.zo_alt_ztest);
6484 (void) printf("%d runs of newer ztest: %s\n", newer,
6485 cmd);
6486 }
6487 (void) printf("%d killed, %d completed, %.0f%% kill rate\n",
6488 kills, iters - kills, (100.0 * kills) / MAX(1, iters));
6489 }
6490
6491 umem_free(cmd, MAXNAMELEN);
6492
6493 return (0);
6494 }