?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
9166 zfs storage pool checkpoint
[unleashed.git] / usr / src / cmd / ztest / ztest.c
blobb53d4a091af68bc809640b4d265e709e6fda7f85
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 ztest_func_t ztest_spa_checkpoint_create_discard;
348
349 uint64_t zopt_always = 0ULL * NANOSEC; /* all the time */
350 uint64_t zopt_incessant = 1ULL * NANOSEC / 10; /* every 1/10 second */
351 uint64_t zopt_often = 1ULL * NANOSEC; /* every second */
352 uint64_t zopt_sometimes = 10ULL * NANOSEC; /* every 10 seconds */
353 uint64_t zopt_rarely = 60ULL * NANOSEC; /* every 60 seconds */
354
355 ztest_info_t ztest_info[] = {
356 { ztest_dmu_read_write, 1, &zopt_always },
357 { ztest_dmu_write_parallel, 10, &zopt_always },
358 { ztest_dmu_object_alloc_free, 1, &zopt_always },
359 { ztest_dmu_commit_callbacks, 1, &zopt_always },
360 { ztest_zap, 30, &zopt_always },
361 { ztest_zap_parallel, 100, &zopt_always },
362 { ztest_split_pool, 1, &zopt_always },
363 { ztest_zil_commit, 1, &zopt_incessant },
364 { ztest_zil_remount, 1, &zopt_sometimes },
365 { ztest_dmu_read_write_zcopy, 1, &zopt_often },
366 { ztest_dmu_objset_create_destroy, 1, &zopt_often },
367 { ztest_dsl_prop_get_set, 1, &zopt_often },
368 { ztest_spa_prop_get_set, 1, &zopt_sometimes },
369 #if 0
370 { ztest_dmu_prealloc, 1, &zopt_sometimes },
371 #endif
372 { ztest_fzap, 1, &zopt_sometimes },
373 { ztest_dmu_snapshot_create_destroy, 1, &zopt_sometimes },
374 { ztest_spa_create_destroy, 1, &zopt_sometimes },
375 { ztest_fault_inject, 1, &zopt_sometimes },
376 { ztest_ddt_repair, 1, &zopt_sometimes },
377 { ztest_dmu_snapshot_hold, 1, &zopt_sometimes },
378 { ztest_reguid, 1, &zopt_rarely },
379 { ztest_spa_rename, 1, &zopt_rarely },
380 { ztest_scrub, 1, &zopt_rarely },
381 { ztest_spa_upgrade, 1, &zopt_rarely },
382 { ztest_dsl_dataset_promote_busy, 1, &zopt_rarely },
383 { ztest_vdev_attach_detach, 1, &zopt_sometimes },
384 { ztest_vdev_LUN_growth, 1, &zopt_rarely },
385 { ztest_vdev_add_remove, 1,
386 &ztest_opts.zo_vdevtime },
387 { ztest_vdev_aux_add_remove, 1,
388 &ztest_opts.zo_vdevtime },
389 { ztest_device_removal, 1, &zopt_sometimes },
390 { ztest_remap_blocks, 1, &zopt_sometimes },
391 { ztest_spa_checkpoint_create_discard, 1, &zopt_rarely }
392 };
393
394 #define ZTEST_FUNCS (sizeof (ztest_info) / sizeof (ztest_info_t))
395
396 /*
397 * The following struct is used to hold a list of uncalled commit callbacks.
398 * The callbacks are ordered by txg number.
399 */
400 typedef struct ztest_cb_list {
401 kmutex_t zcl_callbacks_lock;
402 list_t zcl_callbacks;
403 } ztest_cb_list_t;
404
405 /*
406 * Stuff we need to share writably between parent and child.
407 */
408 typedef struct ztest_shared {
409 boolean_t zs_do_init;
410 hrtime_t zs_proc_start;
411 hrtime_t zs_proc_stop;
412 hrtime_t zs_thread_start;
413 hrtime_t zs_thread_stop;
414 hrtime_t zs_thread_kill;
415 uint64_t zs_enospc_count;
416 uint64_t zs_vdev_next_leaf;
417 uint64_t zs_vdev_aux;
418 uint64_t zs_alloc;
419 uint64_t zs_space;
420 uint64_t zs_splits;
421 uint64_t zs_mirrors;
422 uint64_t zs_metaslab_sz;
423 uint64_t zs_metaslab_df_alloc_threshold;
424 uint64_t zs_guid;
425 } ztest_shared_t;
426
427 #define ID_PARALLEL -1ULL
428
429 static char ztest_dev_template[] = "%s/%s.%llua";
430 static char ztest_aux_template[] = "%s/%s.%s.%llu";
431 ztest_shared_t *ztest_shared;
432
433 static spa_t *ztest_spa = NULL;
434 static ztest_ds_t *ztest_ds;
435
436 static kmutex_t ztest_vdev_lock;
437 static kmutex_t ztest_checkpoint_lock;
438
439 /*
440 * The ztest_name_lock protects the pool and dataset namespace used by
441 * the individual tests. To modify the namespace, consumers must grab
442 * this lock as writer. Grabbing the lock as reader will ensure that the
443 * namespace does not change while the lock is held.
444 */
445 static krwlock_t ztest_name_lock;
446
447 static boolean_t ztest_dump_core = B_TRUE;
448 static boolean_t ztest_exiting;
449
450 /* Global commit callback list */
451 static ztest_cb_list_t zcl;
452
453 enum ztest_object {
454 ZTEST_META_DNODE = 0,
455 ZTEST_DIROBJ,
456 ZTEST_OBJECTS
457 };
458
459 static void usage(boolean_t) __NORETURN;
460
461 /*
462 * These libumem hooks provide a reasonable set of defaults for the allocator's
463 * debugging facilities.
464 */
465 const char *
466 _umem_debug_init()
467 {
468 return ("default,verbose"); /* $UMEM_DEBUG setting */
469 }
470
471 const char *
472 _umem_logging_init(void)
473 {
474 return ("fail,contents"); /* $UMEM_LOGGING setting */
475 }
476
477 #define FATAL_MSG_SZ 1024
478
479 char *fatal_msg;
480
481 static void
482 fatal(int do_perror, char *message, ...)
483 {
484 va_list args;
485 int save_errno = errno;
486 char buf[FATAL_MSG_SZ];
487
488 (void) fflush(stdout);
489
490 va_start(args, message);
491 (void) sprintf(buf, "ztest: ");
492 /* LINTED */
493 (void) vsprintf(buf + strlen(buf), message, args);
494 va_end(args);
495 if (do_perror) {
496 (void) snprintf(buf + strlen(buf), FATAL_MSG_SZ - strlen(buf),
497 ": %s", strerror(save_errno));
498 }
499 (void) fprintf(stderr, "%s\n", buf);
500 fatal_msg = buf; /* to ease debugging */
501 if (ztest_dump_core)
502 abort();
503 exit(3);
504 }
505
506 static int
507 str2shift(const char *buf)
508 {
509 const char *ends = "BKMGTPEZ";
510 int i;
511
512 if (buf[0] == '\0')
513 return (0);
514 for (i = 0; i < strlen(ends); i++) {
515 if (toupper(buf[0]) == ends[i])
516 break;
517 }
518 if (i == strlen(ends)) {
519 (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n",
520 buf);
521 usage(B_FALSE);
522 }
523 if (buf[1] == '\0' || (toupper(buf[1]) == 'B' && buf[2] == '\0')) {
524 return (10*i);
525 }
526 (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n", buf);
527 usage(B_FALSE);
528 /* NOTREACHED */
529 }
530
531 static uint64_t
532 nicenumtoull(const char *buf)
533 {
534 char *end;
535 uint64_t val;
536
537 val = strtoull(buf, &end, 0);
538 if (end == buf) {
539 (void) fprintf(stderr, "ztest: bad numeric value: %s\n", buf);
540 usage(B_FALSE);
541 } else if (end[0] == '.') {
542 double fval = strtod(buf, &end);
543 fval *= pow(2, str2shift(end));
544 if (fval > UINT64_MAX) {
545 (void) fprintf(stderr, "ztest: value too large: %s\n",
546 buf);
547 usage(B_FALSE);
548 }
549 val = (uint64_t)fval;
550 } else {
551 int shift = str2shift(end);
552 if (shift >= 64 || (val << shift) >> shift != val) {
553 (void) fprintf(stderr, "ztest: value too large: %s\n",
554 buf);
555 usage(B_FALSE);
556 }
557 val <<= shift;
558 }
559 return (val);
560 }
561
562 static void
563 usage(boolean_t requested)
564 {
565 const ztest_shared_opts_t *zo = &ztest_opts_defaults;
566
567 char nice_vdev_size[NN_NUMBUF_SZ];
568 char nice_gang_bang[NN_NUMBUF_SZ];
569 FILE *fp = requested ? stdout : stderr;
570
571 nicenum(zo->zo_vdev_size, nice_vdev_size, sizeof (nice_vdev_size));
572 nicenum(zo->zo_metaslab_gang_bang, nice_gang_bang,
573 sizeof (nice_gang_bang));
574
575 (void) fprintf(fp, "Usage: %s\n"
576 "\t[-v vdevs (default: %llu)]\n"
577 "\t[-s size_of_each_vdev (default: %s)]\n"
578 "\t[-a alignment_shift (default: %d)] use 0 for random\n"
579 "\t[-m mirror_copies (default: %d)]\n"
580 "\t[-r raidz_disks (default: %d)]\n"
581 "\t[-R raidz_parity (default: %d)]\n"
582 "\t[-d datasets (default: %d)]\n"
583 "\t[-t threads (default: %d)]\n"
584 "\t[-g gang_block_threshold (default: %s)]\n"
585 "\t[-i init_count (default: %d)] initialize pool i times\n"
586 "\t[-k kill_percentage (default: %llu%%)]\n"
587 "\t[-p pool_name (default: %s)]\n"
588 "\t[-f dir (default: %s)] file directory for vdev files\n"
589 "\t[-V] verbose (use multiple times for ever more blather)\n"
590 "\t[-E] use existing pool instead of creating new one\n"
591 "\t[-T time (default: %llu sec)] total run time\n"
592 "\t[-F freezeloops (default: %llu)] max loops in spa_freeze()\n"
593 "\t[-P passtime (default: %llu sec)] time per pass\n"
594 "\t[-B alt_ztest (default: <none>)] alternate ztest path\n"
595 "\t[-o variable=value] ... set global variable to an unsigned\n"
596 "\t 32-bit integer value\n"
597 "\t[-h] (print help)\n"
598 "",
599 zo->zo_pool,
600 (u_longlong_t)zo->zo_vdevs, /* -v */
601 nice_vdev_size, /* -s */
602 zo->zo_ashift, /* -a */
603 zo->zo_mirrors, /* -m */
604 zo->zo_raidz, /* -r */
605 zo->zo_raidz_parity, /* -R */
606 zo->zo_datasets, /* -d */
607 zo->zo_threads, /* -t */
608 nice_gang_bang, /* -g */
609 zo->zo_init, /* -i */
610 (u_longlong_t)zo->zo_killrate, /* -k */
611 zo->zo_pool, /* -p */
612 zo->zo_dir, /* -f */
613 (u_longlong_t)zo->zo_time, /* -T */
614 (u_longlong_t)zo->zo_maxloops, /* -F */
615 (u_longlong_t)zo->zo_passtime);
616 exit(requested ? 0 : 1);
617 }
618
619 static void
620 process_options(int argc, char **argv)
621 {
622 char *path;
623 ztest_shared_opts_t *zo = &ztest_opts;
624
625 int opt;
626 uint64_t value;
627 char altdir[MAXNAMELEN] = { 0 };
628
629 bcopy(&ztest_opts_defaults, zo, sizeof (*zo));
630
631 while ((opt = getopt(argc, argv,
632 "v:s:a:m:r:R:d:t:g:i:k:p:f:VET:P:hF:B:o:")) != EOF) {
633 value = 0;
634 switch (opt) {
635 case 'v':
636 case 's':
637 case 'a':
638 case 'm':
639 case 'r':
640 case 'R':
641 case 'd':
642 case 't':
643 case 'g':
644 case 'i':
645 case 'k':
646 case 'T':
647 case 'P':
648 case 'F':
649 value = nicenumtoull(optarg);
650 }
651 switch (opt) {
652 case 'v':
653 zo->zo_vdevs = value;
654 break;
655 case 's':
656 zo->zo_vdev_size = MAX(SPA_MINDEVSIZE, value);
657 break;
658 case 'a':
659 zo->zo_ashift = value;
660 break;
661 case 'm':
662 zo->zo_mirrors = value;
663 break;
664 case 'r':
665 zo->zo_raidz = MAX(1, value);
666 break;
667 case 'R':
668 zo->zo_raidz_parity = MIN(MAX(value, 1), 3);
669 break;
670 case 'd':
671 zo->zo_datasets = MAX(1, value);
672 break;
673 case 't':
674 zo->zo_threads = MAX(1, value);
675 break;
676 case 'g':
677 zo->zo_metaslab_gang_bang = MAX(SPA_MINBLOCKSIZE << 1,
678 value);
679 break;
680 case 'i':
681 zo->zo_init = value;
682 break;
683 case 'k':
684 zo->zo_killrate = value;
685 break;
686 case 'p':
687 (void) strlcpy(zo->zo_pool, optarg,
688 sizeof (zo->zo_pool));
689 break;
690 case 'f':
691 path = realpath(optarg, NULL);
692 if (path == NULL) {
693 (void) fprintf(stderr, "error: %s: %s\n",
694 optarg, strerror(errno));
695 usage(B_FALSE);
696 } else {
697 (void) strlcpy(zo->zo_dir, path,
698 sizeof (zo->zo_dir));
699 }
700 break;
701 case 'V':
702 zo->zo_verbose++;
703 break;
704 case 'E':
705 zo->zo_init = 0;
706 break;
707 case 'T':
708 zo->zo_time = value;
709 break;
710 case 'P':
711 zo->zo_passtime = MAX(1, value);
712 break;
713 case 'F':
714 zo->zo_maxloops = MAX(1, value);
715 break;
716 case 'B':
717 (void) strlcpy(altdir, optarg, sizeof (altdir));
718 break;
719 case 'o':
720 if (set_global_var(optarg) != 0)
721 usage(B_FALSE);
722 break;
723 case 'h':
724 usage(B_TRUE);
725 break;
726 case '?':
727 default:
728 usage(B_FALSE);
729 break;
730 }
731 }
732
733 zo->zo_raidz_parity = MIN(zo->zo_raidz_parity, zo->zo_raidz - 1);
734
735 zo->zo_vdevtime =
736 (zo->zo_vdevs > 0 ? zo->zo_time * NANOSEC / zo->zo_vdevs :
737 UINT64_MAX >> 2);
738
739 if (strlen(altdir) > 0) {
740 char *cmd;
741 char *realaltdir;
742 char *bin;
743 char *ztest;
744 char *isa;
745 int isalen;
746
747 cmd = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
748 realaltdir = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
749
750 VERIFY(NULL != realpath(getexecname(), cmd));
751 if (0 != access(altdir, F_OK)) {
752 ztest_dump_core = B_FALSE;
753 fatal(B_TRUE, "invalid alternate ztest path: %s",
754 altdir);
755 }
756 VERIFY(NULL != realpath(altdir, realaltdir));
757
758 /*
759 * 'cmd' should be of the form "<anything>/usr/bin/<isa>/ztest".
760 * We want to extract <isa> to determine if we should use
761 * 32 or 64 bit binaries.
762 */
763 bin = strstr(cmd, "/usr/bin/");
764 ztest = strstr(bin, "/ztest");
765 isa = bin + 9;
766 isalen = ztest - isa;
767 (void) snprintf(zo->zo_alt_ztest, sizeof (zo->zo_alt_ztest),
768 "%s/usr/bin/%.*s/ztest", realaltdir, isalen, isa);
769 (void) snprintf(zo->zo_alt_libpath, sizeof (zo->zo_alt_libpath),
770 "%s/usr/lib/%.*s", realaltdir, isalen, isa);
771
772 if (0 != access(zo->zo_alt_ztest, X_OK)) {
773 ztest_dump_core = B_FALSE;
774 fatal(B_TRUE, "invalid alternate ztest: %s",
775 zo->zo_alt_ztest);
776 } else if (0 != access(zo->zo_alt_libpath, X_OK)) {
777 ztest_dump_core = B_FALSE;
778 fatal(B_TRUE, "invalid alternate lib directory %s",
779 zo->zo_alt_libpath);
780 }
781
782 umem_free(cmd, MAXPATHLEN);
783 umem_free(realaltdir, MAXPATHLEN);
784 }
785 }
786
787 static void
788 ztest_kill(ztest_shared_t *zs)
789 {
790 zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(ztest_spa));
791 zs->zs_space = metaslab_class_get_space(spa_normal_class(ztest_spa));
792
793 /*
794 * Before we kill off ztest, make sure that the config is updated.
795 * See comment above spa_write_cachefile().
796 */
797 mutex_enter(&spa_namespace_lock);
798 spa_write_cachefile(ztest_spa, B_FALSE, B_FALSE);
799 mutex_exit(&spa_namespace_lock);
800
801 zfs_dbgmsg_print(FTAG);
802 (void) kill(getpid(), SIGKILL);
803 }
804
805 static uint64_t
806 ztest_random(uint64_t range)
807 {
808 uint64_t r;
809
810 ASSERT3S(ztest_fd_rand, >=, 0);
811
812 if (range == 0)
813 return (0);
814
815 if (read(ztest_fd_rand, &r, sizeof (r)) != sizeof (r))
816 fatal(1, "short read from /dev/urandom");
817
818 return (r % range);
819 }
820
821 /* ARGSUSED */
822 static void
823 ztest_record_enospc(const char *s)
824 {
825 ztest_shared->zs_enospc_count++;
826 }
827
828 static uint64_t
829 ztest_get_ashift(void)
830 {
831 if (ztest_opts.zo_ashift == 0)
832 return (SPA_MINBLOCKSHIFT + ztest_random(5));
833 return (ztest_opts.zo_ashift);
834 }
835
836 static nvlist_t *
837 make_vdev_file(char *path, char *aux, char *pool, size_t size, uint64_t ashift)
838 {
839 char pathbuf[MAXPATHLEN];
840 uint64_t vdev;
841 nvlist_t *file;
842
843 if (ashift == 0)
844 ashift = ztest_get_ashift();
845
846 if (path == NULL) {
847 path = pathbuf;
848
849 if (aux != NULL) {
850 vdev = ztest_shared->zs_vdev_aux;
851 (void) snprintf(path, sizeof (pathbuf),
852 ztest_aux_template, ztest_opts.zo_dir,
853 pool == NULL ? ztest_opts.zo_pool : pool,
854 aux, vdev);
855 } else {
856 vdev = ztest_shared->zs_vdev_next_leaf++;
857 (void) snprintf(path, sizeof (pathbuf),
858 ztest_dev_template, ztest_opts.zo_dir,
859 pool == NULL ? ztest_opts.zo_pool : pool, vdev);
860 }
861 }
862
863 if (size != 0) {
864 int fd = open(path, O_RDWR | O_CREAT | O_TRUNC, 0666);
865 if (fd == -1)
866 fatal(1, "can't open %s", path);
867 if (ftruncate(fd, size) != 0)
868 fatal(1, "can't ftruncate %s", path);
869 (void) close(fd);
870 }
871
872 VERIFY(nvlist_alloc(&file, NV_UNIQUE_NAME, 0) == 0);
873 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_TYPE, VDEV_TYPE_FILE) == 0);
874 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_PATH, path) == 0);
875 VERIFY(nvlist_add_uint64(file, ZPOOL_CONFIG_ASHIFT, ashift) == 0);
876
877 return (file);
878 }
879
880 static nvlist_t *
881 make_vdev_raidz(char *path, char *aux, char *pool, size_t size,
882 uint64_t ashift, int r)
883 {
884 nvlist_t *raidz, **child;
885 int c;
886
887 if (r < 2)
888 return (make_vdev_file(path, aux, pool, size, ashift));
889 child = umem_alloc(r * sizeof (nvlist_t *), UMEM_NOFAIL);
890
891 for (c = 0; c < r; c++)
892 child[c] = make_vdev_file(path, aux, pool, size, ashift);
893
894 VERIFY(nvlist_alloc(&raidz, NV_UNIQUE_NAME, 0) == 0);
895 VERIFY(nvlist_add_string(raidz, ZPOOL_CONFIG_TYPE,
896 VDEV_TYPE_RAIDZ) == 0);
897 VERIFY(nvlist_add_uint64(raidz, ZPOOL_CONFIG_NPARITY,
898 ztest_opts.zo_raidz_parity) == 0);
899 VERIFY(nvlist_add_nvlist_array(raidz, ZPOOL_CONFIG_CHILDREN,
900 child, r) == 0);
901
902 for (c = 0; c < r; c++)
903 nvlist_free(child[c]);
904
905 umem_free(child, r * sizeof (nvlist_t *));
906
907 return (raidz);
908 }
909
910 static nvlist_t *
911 make_vdev_mirror(char *path, char *aux, char *pool, size_t size,
912 uint64_t ashift, int r, int m)
913 {
914 nvlist_t *mirror, **child;
915 int c;
916
917 if (m < 1)
918 return (make_vdev_raidz(path, aux, pool, size, ashift, r));
919
920 child = umem_alloc(m * sizeof (nvlist_t *), UMEM_NOFAIL);
921
922 for (c = 0; c < m; c++)
923 child[c] = make_vdev_raidz(path, aux, pool, size, ashift, r);
924
925 VERIFY(nvlist_alloc(&mirror, NV_UNIQUE_NAME, 0) == 0);
926 VERIFY(nvlist_add_string(mirror, ZPOOL_CONFIG_TYPE,
927 VDEV_TYPE_MIRROR) == 0);
928 VERIFY(nvlist_add_nvlist_array(mirror, ZPOOL_CONFIG_CHILDREN,
929 child, m) == 0);
930
931 for (c = 0; c < m; c++)
932 nvlist_free(child[c]);
933
934 umem_free(child, m * sizeof (nvlist_t *));
935
936 return (mirror);
937 }
938
939 static nvlist_t *
940 make_vdev_root(char *path, char *aux, char *pool, size_t size, uint64_t ashift,
941 int log, int r, int m, int t)
942 {
943 nvlist_t *root, **child;
944 int c;
945
946 ASSERT(t > 0);
947
948 child = umem_alloc(t * sizeof (nvlist_t *), UMEM_NOFAIL);
949
950 for (c = 0; c < t; c++) {
951 child[c] = make_vdev_mirror(path, aux, pool, size, ashift,
952 r, m);
953 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_IS_LOG,
954 log) == 0);
955 }
956
957 VERIFY(nvlist_alloc(&root, NV_UNIQUE_NAME, 0) == 0);
958 VERIFY(nvlist_add_string(root, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) == 0);
959 VERIFY(nvlist_add_nvlist_array(root, aux ? aux : ZPOOL_CONFIG_CHILDREN,
960 child, t) == 0);
961
962 for (c = 0; c < t; c++)
963 nvlist_free(child[c]);
964
965 umem_free(child, t * sizeof (nvlist_t *));
966
967 return (root);
968 }
969
970 /*
971 * Find a random spa version. Returns back a random spa version in the
972 * range [initial_version, SPA_VERSION_FEATURES].
973 */
974 static uint64_t
975 ztest_random_spa_version(uint64_t initial_version)
976 {
977 uint64_t version = initial_version;
978
979 if (version <= SPA_VERSION_BEFORE_FEATURES) {
980 version = version +
981 ztest_random(SPA_VERSION_BEFORE_FEATURES - version + 1);
982 }
983
984 if (version > SPA_VERSION_BEFORE_FEATURES)
985 version = SPA_VERSION_FEATURES;
986
987 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
988 return (version);
989 }
990
991 static int
992 ztest_random_blocksize(void)
993 {
994 uint64_t block_shift;
995 /*
996 * Choose a block size >= the ashift.
997 * If the SPA supports new MAXBLOCKSIZE, test up to 1MB blocks.
998 */
999 int maxbs = SPA_OLD_MAXBLOCKSHIFT;
1000 if (spa_maxblocksize(ztest_spa) == SPA_MAXBLOCKSIZE)
1001 maxbs = 20;
1002 block_shift = ztest_random(maxbs - ztest_spa->spa_max_ashift + 1);
1003 return (1 << (SPA_MINBLOCKSHIFT + block_shift));
1004 }
1005
1006 static int
1007 ztest_random_ibshift(void)
1008 {
1009 return (DN_MIN_INDBLKSHIFT +
1010 ztest_random(DN_MAX_INDBLKSHIFT - DN_MIN_INDBLKSHIFT + 1));
1011 }
1012
1013 static uint64_t
1014 ztest_random_vdev_top(spa_t *spa, boolean_t log_ok)
1015 {
1016 uint64_t top;
1017 vdev_t *rvd = spa->spa_root_vdev;
1018 vdev_t *tvd;
1019
1020 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1021
1022 do {
1023 top = ztest_random(rvd->vdev_children);
1024 tvd = rvd->vdev_child[top];
1025 } while (!vdev_is_concrete(tvd) || (tvd->vdev_islog && !log_ok) ||
1026 tvd->vdev_mg == NULL || tvd->vdev_mg->mg_class == NULL);
1027
1028 return (top);
1029 }
1030
1031 static uint64_t
1032 ztest_random_dsl_prop(zfs_prop_t prop)
1033 {
1034 uint64_t value;
1035
1036 do {
1037 value = zfs_prop_random_value(prop, ztest_random(-1ULL));
1038 } while (prop == ZFS_PROP_CHECKSUM && value == ZIO_CHECKSUM_OFF);
1039
1040 return (value);
1041 }
1042
1043 static int
1044 ztest_dsl_prop_set_uint64(char *osname, zfs_prop_t prop, uint64_t value,
1045 boolean_t inherit)
1046 {
1047 const char *propname = zfs_prop_to_name(prop);
1048 const char *valname;
1049 char setpoint[MAXPATHLEN];
1050 uint64_t curval;
1051 int error;
1052
1053 error = dsl_prop_set_int(osname, propname,
1054 (inherit ? ZPROP_SRC_NONE : ZPROP_SRC_LOCAL), value);
1055
1056 if (error == ENOSPC) {
1057 ztest_record_enospc(FTAG);
1058 return (error);
1059 }
1060 ASSERT0(error);
1061
1062 VERIFY0(dsl_prop_get_integer(osname, propname, &curval, setpoint));
1063
1064 if (ztest_opts.zo_verbose >= 6) {
1065 VERIFY(zfs_prop_index_to_string(prop, curval, &valname) == 0);
1066 (void) printf("%s %s = %s at '%s'\n",
1067 osname, propname, valname, setpoint);
1068 }
1069
1070 return (error);
1071 }
1072
1073 static int
1074 ztest_spa_prop_set_uint64(zpool_prop_t prop, uint64_t value)
1075 {
1076 spa_t *spa = ztest_spa;
1077 nvlist_t *props = NULL;
1078 int error;
1079
1080 VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
1081 VERIFY(nvlist_add_uint64(props, zpool_prop_to_name(prop), value) == 0);
1082
1083 error = spa_prop_set(spa, props);
1084
1085 nvlist_free(props);
1086
1087 if (error == ENOSPC) {
1088 ztest_record_enospc(FTAG);
1089 return (error);
1090 }
1091 ASSERT0(error);
1092
1093 return (error);
1094 }
1095
1096 static void
1097 ztest_rll_init(rll_t *rll)
1098 {
1099 rll->rll_writer = NULL;
1100 rll->rll_readers = 0;
1101 mutex_init(&rll->rll_lock, NULL, USYNC_THREAD, NULL);
1102 cv_init(&rll->rll_cv, NULL, USYNC_THREAD, NULL);
1103 }
1104
1105 static void
1106 ztest_rll_destroy(rll_t *rll)
1107 {
1108 ASSERT(rll->rll_writer == NULL);
1109 ASSERT(rll->rll_readers == 0);
1110 mutex_destroy(&rll->rll_lock);
1111 cv_destroy(&rll->rll_cv);
1112 }
1113
1114 static void
1115 ztest_rll_lock(rll_t *rll, rl_type_t type)
1116 {
1117 mutex_enter(&rll->rll_lock);
1118
1119 if (type == RL_READER) {
1120 while (rll->rll_writer != NULL)
1121 cv_wait(&rll->rll_cv, &rll->rll_lock);
1122 rll->rll_readers++;
1123 } else {
1124 while (rll->rll_writer != NULL || rll->rll_readers)
1125 cv_wait(&rll->rll_cv, &rll->rll_lock);
1126 rll->rll_writer = curthread;
1127 }
1128
1129 mutex_exit(&rll->rll_lock);
1130 }
1131
1132 static void
1133 ztest_rll_unlock(rll_t *rll)
1134 {
1135 mutex_enter(&rll->rll_lock);
1136
1137 if (rll->rll_writer) {
1138 ASSERT(rll->rll_readers == 0);
1139 rll->rll_writer = NULL;
1140 } else {
1141 ASSERT(rll->rll_readers != 0);
1142 ASSERT(rll->rll_writer == NULL);
1143 rll->rll_readers--;
1144 }
1145
1146 if (rll->rll_writer == NULL && rll->rll_readers == 0)
1147 cv_broadcast(&rll->rll_cv);
1148
1149 mutex_exit(&rll->rll_lock);
1150 }
1151
1152 static void
1153 ztest_object_lock(ztest_ds_t *zd, uint64_t object, rl_type_t type)
1154 {
1155 rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1156
1157 ztest_rll_lock(rll, type);
1158 }
1159
1160 static void
1161 ztest_object_unlock(ztest_ds_t *zd, uint64_t object)
1162 {
1163 rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1164
1165 ztest_rll_unlock(rll);
1166 }
1167
1168 static rl_t *
1169 ztest_range_lock(ztest_ds_t *zd, uint64_t object, uint64_t offset,
1170 uint64_t size, rl_type_t type)
1171 {
1172 uint64_t hash = object ^ (offset % (ZTEST_RANGE_LOCKS + 1));
1173 rll_t *rll = &zd->zd_range_lock[hash & (ZTEST_RANGE_LOCKS - 1)];
1174 rl_t *rl;
1175
1176 rl = umem_alloc(sizeof (*rl), UMEM_NOFAIL);
1177 rl->rl_object = object;
1178 rl->rl_offset = offset;
1179 rl->rl_size = size;
1180 rl->rl_lock = rll;
1181
1182 ztest_rll_lock(rll, type);
1183
1184 return (rl);
1185 }
1186
1187 static void
1188 ztest_range_unlock(rl_t *rl)
1189 {
1190 rll_t *rll = rl->rl_lock;
1191
1192 ztest_rll_unlock(rll);
1193
1194 umem_free(rl, sizeof (*rl));
1195 }
1196
1197 static void
1198 ztest_zd_init(ztest_ds_t *zd, ztest_shared_ds_t *szd, objset_t *os)
1199 {
1200 zd->zd_os = os;
1201 zd->zd_zilog = dmu_objset_zil(os);
1202 zd->zd_shared = szd;
1203 dmu_objset_name(os, zd->zd_name);
1204
1205 if (zd->zd_shared != NULL)
1206 zd->zd_shared->zd_seq = 0;
1207
1208 rw_init(&zd->zd_zilog_lock, NULL, USYNC_THREAD, NULL);
1209 mutex_init(&zd->zd_dirobj_lock, NULL, USYNC_THREAD, NULL);
1210
1211 for (int l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1212 ztest_rll_init(&zd->zd_object_lock[l]);
1213
1214 for (int l = 0; l < ZTEST_RANGE_LOCKS; l++)
1215 ztest_rll_init(&zd->zd_range_lock[l]);
1216 }
1217
1218 static void
1219 ztest_zd_fini(ztest_ds_t *zd)
1220 {
1221 mutex_destroy(&zd->zd_dirobj_lock);
1222
1223 for (int l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1224 ztest_rll_destroy(&zd->zd_object_lock[l]);
1225
1226 for (int l = 0; l < ZTEST_RANGE_LOCKS; l++)
1227 ztest_rll_destroy(&zd->zd_range_lock[l]);
1228 }
1229
1230 #define TXG_MIGHTWAIT (ztest_random(10) == 0 ? TXG_NOWAIT : TXG_WAIT)
1231
1232 static uint64_t
1233 ztest_tx_assign(dmu_tx_t *tx, uint64_t txg_how, const char *tag)
1234 {
1235 uint64_t txg;
1236 int error;
1237
1238 /*
1239 * Attempt to assign tx to some transaction group.
1240 */
1241 error = dmu_tx_assign(tx, txg_how);
1242 if (error) {
1243 if (error == ERESTART) {
1244 ASSERT(txg_how == TXG_NOWAIT);
1245 dmu_tx_wait(tx);
1246 } else {
1247 ASSERT3U(error, ==, ENOSPC);
1248 ztest_record_enospc(tag);
1249 }
1250 dmu_tx_abort(tx);
1251 return (0);
1252 }
1253 txg = dmu_tx_get_txg(tx);
1254 ASSERT(txg != 0);
1255 return (txg);
1256 }
1257
1258 static void
1259 ztest_pattern_set(void *buf, uint64_t size, uint64_t value)
1260 {
1261 uint64_t *ip = buf;
1262 uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1263
1264 while (ip < ip_end)
1265 *ip++ = value;
1266 }
1267
1268 static boolean_t
1269 ztest_pattern_match(void *buf, uint64_t size, uint64_t value)
1270 {
1271 uint64_t *ip = buf;
1272 uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1273 uint64_t diff = 0;
1274
1275 while (ip < ip_end)
1276 diff |= (value - *ip++);
1277
1278 return (diff == 0);
1279 }
1280
1281 static void
1282 ztest_bt_generate(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1283 uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg)
1284 {
1285 bt->bt_magic = BT_MAGIC;
1286 bt->bt_objset = dmu_objset_id(os);
1287 bt->bt_object = object;
1288 bt->bt_offset = offset;
1289 bt->bt_gen = gen;
1290 bt->bt_txg = txg;
1291 bt->bt_crtxg = crtxg;
1292 }
1293
1294 static void
1295 ztest_bt_verify(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1296 uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg)
1297 {
1298 ASSERT3U(bt->bt_magic, ==, BT_MAGIC);
1299 ASSERT3U(bt->bt_objset, ==, dmu_objset_id(os));
1300 ASSERT3U(bt->bt_object, ==, object);
1301 ASSERT3U(bt->bt_offset, ==, offset);
1302 ASSERT3U(bt->bt_gen, <=, gen);
1303 ASSERT3U(bt->bt_txg, <=, txg);
1304 ASSERT3U(bt->bt_crtxg, ==, crtxg);
1305 }
1306
1307 static ztest_block_tag_t *
1308 ztest_bt_bonus(dmu_buf_t *db)
1309 {
1310 dmu_object_info_t doi;
1311 ztest_block_tag_t *bt;
1312
1313 dmu_object_info_from_db(db, &doi);
1314 ASSERT3U(doi.doi_bonus_size, <=, db->db_size);
1315 ASSERT3U(doi.doi_bonus_size, >=, sizeof (*bt));
1316 bt = (void *)((char *)db->db_data + doi.doi_bonus_size - sizeof (*bt));
1317
1318 return (bt);
1319 }
1320
1321 /*
1322 * ZIL logging ops
1323 */
1324
1325 #define lrz_type lr_mode
1326 #define lrz_blocksize lr_uid
1327 #define lrz_ibshift lr_gid
1328 #define lrz_bonustype lr_rdev
1329 #define lrz_bonuslen lr_crtime[1]
1330
1331 static void
1332 ztest_log_create(ztest_ds_t *zd, dmu_tx_t *tx, lr_create_t *lr)
1333 {
1334 char *name = (void *)(lr + 1); /* name follows lr */
1335 size_t namesize = strlen(name) + 1;
1336 itx_t *itx;
1337
1338 if (zil_replaying(zd->zd_zilog, tx))
1339 return;
1340
1341 itx = zil_itx_create(TX_CREATE, sizeof (*lr) + namesize);
1342 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1343 sizeof (*lr) + namesize - sizeof (lr_t));
1344
1345 zil_itx_assign(zd->zd_zilog, itx, tx);
1346 }
1347
1348 static void
1349 ztest_log_remove(ztest_ds_t *zd, dmu_tx_t *tx, lr_remove_t *lr, uint64_t object)
1350 {
1351 char *name = (void *)(lr + 1); /* name follows lr */
1352 size_t namesize = strlen(name) + 1;
1353 itx_t *itx;
1354
1355 if (zil_replaying(zd->zd_zilog, tx))
1356 return;
1357
1358 itx = zil_itx_create(TX_REMOVE, sizeof (*lr) + namesize);
1359 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1360 sizeof (*lr) + namesize - sizeof (lr_t));
1361
1362 itx->itx_oid = object;
1363 zil_itx_assign(zd->zd_zilog, itx, tx);
1364 }
1365
1366 static void
1367 ztest_log_write(ztest_ds_t *zd, dmu_tx_t *tx, lr_write_t *lr)
1368 {
1369 itx_t *itx;
1370 itx_wr_state_t write_state = ztest_random(WR_NUM_STATES);
1371
1372 if (zil_replaying(zd->zd_zilog, tx))
1373 return;
1374
1375 if (lr->lr_length > ZIL_MAX_LOG_DATA)
1376 write_state = WR_INDIRECT;
1377
1378 itx = zil_itx_create(TX_WRITE,
1379 sizeof (*lr) + (write_state == WR_COPIED ? lr->lr_length : 0));
1380
1381 if (write_state == WR_COPIED &&
1382 dmu_read(zd->zd_os, lr->lr_foid, lr->lr_offset, lr->lr_length,
1383 ((lr_write_t *)&itx->itx_lr) + 1, DMU_READ_NO_PREFETCH) != 0) {
1384 zil_itx_destroy(itx);
1385 itx = zil_itx_create(TX_WRITE, sizeof (*lr));
1386 write_state = WR_NEED_COPY;
1387 }
1388 itx->itx_private = zd;
1389 itx->itx_wr_state = write_state;
1390 itx->itx_sync = (ztest_random(8) == 0);
1391
1392 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1393 sizeof (*lr) - sizeof (lr_t));
1394
1395 zil_itx_assign(zd->zd_zilog, itx, tx);
1396 }
1397
1398 static void
1399 ztest_log_truncate(ztest_ds_t *zd, dmu_tx_t *tx, lr_truncate_t *lr)
1400 {
1401 itx_t *itx;
1402
1403 if (zil_replaying(zd->zd_zilog, tx))
1404 return;
1405
1406 itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr));
1407 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1408 sizeof (*lr) - sizeof (lr_t));
1409
1410 itx->itx_sync = B_FALSE;
1411 zil_itx_assign(zd->zd_zilog, itx, tx);
1412 }
1413
1414 static void
1415 ztest_log_setattr(ztest_ds_t *zd, dmu_tx_t *tx, lr_setattr_t *lr)
1416 {
1417 itx_t *itx;
1418
1419 if (zil_replaying(zd->zd_zilog, tx))
1420 return;
1421
1422 itx = zil_itx_create(TX_SETATTR, sizeof (*lr));
1423 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1424 sizeof (*lr) - sizeof (lr_t));
1425
1426 itx->itx_sync = B_FALSE;
1427 zil_itx_assign(zd->zd_zilog, itx, tx);
1428 }
1429
1430 /*
1431 * ZIL replay ops
1432 */
1433 static int
1434 ztest_replay_create(void *arg1, void *arg2, boolean_t byteswap)
1435 {
1436 ztest_ds_t *zd = arg1;
1437 lr_create_t *lr = arg2;
1438 char *name = (void *)(lr + 1); /* name follows lr */
1439 objset_t *os = zd->zd_os;
1440 ztest_block_tag_t *bbt;
1441 dmu_buf_t *db;
1442 dmu_tx_t *tx;
1443 uint64_t txg;
1444 int error = 0;
1445
1446 if (byteswap)
1447 byteswap_uint64_array(lr, sizeof (*lr));
1448
1449 ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1450 ASSERT(name[0] != '\0');
1451
1452 tx = dmu_tx_create(os);
1453
1454 dmu_tx_hold_zap(tx, lr->lr_doid, B_TRUE, name);
1455
1456 if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1457 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, B_TRUE, NULL);
1458 } else {
1459 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1460 }
1461
1462 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1463 if (txg == 0)
1464 return (ENOSPC);
1465
1466 ASSERT(dmu_objset_zil(os)->zl_replay == !!lr->lr_foid);
1467
1468 if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1469 if (lr->lr_foid == 0) {
1470 lr->lr_foid = zap_create(os,
1471 lr->lrz_type, lr->lrz_bonustype,
1472 lr->lrz_bonuslen, tx);
1473 } else {
1474 error = zap_create_claim(os, lr->lr_foid,
1475 lr->lrz_type, lr->lrz_bonustype,
1476 lr->lrz_bonuslen, tx);
1477 }
1478 } else {
1479 if (lr->lr_foid == 0) {
1480 lr->lr_foid = dmu_object_alloc(os,
1481 lr->lrz_type, 0, lr->lrz_bonustype,
1482 lr->lrz_bonuslen, tx);
1483 } else {
1484 error = dmu_object_claim(os, lr->lr_foid,
1485 lr->lrz_type, 0, lr->lrz_bonustype,
1486 lr->lrz_bonuslen, tx);
1487 }
1488 }
1489
1490 if (error) {
1491 ASSERT3U(error, ==, EEXIST);
1492 ASSERT(zd->zd_zilog->zl_replay);
1493 dmu_tx_commit(tx);
1494 return (error);
1495 }
1496
1497 ASSERT(lr->lr_foid != 0);
1498
1499 if (lr->lrz_type != DMU_OT_ZAP_OTHER)
1500 VERIFY3U(0, ==, dmu_object_set_blocksize(os, lr->lr_foid,
1501 lr->lrz_blocksize, lr->lrz_ibshift, tx));
1502
1503 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1504 bbt = ztest_bt_bonus(db);
1505 dmu_buf_will_dirty(db, tx);
1506 ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_gen, txg, txg);
1507 dmu_buf_rele(db, FTAG);
1508
1509 VERIFY3U(0, ==, zap_add(os, lr->lr_doid, name, sizeof (uint64_t), 1,
1510 &lr->lr_foid, tx));
1511
1512 (void) ztest_log_create(zd, tx, lr);
1513
1514 dmu_tx_commit(tx);
1515
1516 return (0);
1517 }
1518
1519 static int
1520 ztest_replay_remove(void *arg1, void *arg2, boolean_t byteswap)
1521 {
1522 ztest_ds_t *zd = arg1;
1523 lr_remove_t *lr = arg2;
1524 char *name = (void *)(lr + 1); /* name follows lr */
1525 objset_t *os = zd->zd_os;
1526 dmu_object_info_t doi;
1527 dmu_tx_t *tx;
1528 uint64_t object, txg;
1529
1530 if (byteswap)
1531 byteswap_uint64_array(lr, sizeof (*lr));
1532
1533 ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1534 ASSERT(name[0] != '\0');
1535
1536 VERIFY3U(0, ==,
1537 zap_lookup(os, lr->lr_doid, name, sizeof (object), 1, &object));
1538 ASSERT(object != 0);
1539
1540 ztest_object_lock(zd, object, RL_WRITER);
1541
1542 VERIFY3U(0, ==, dmu_object_info(os, object, &doi));
1543
1544 tx = dmu_tx_create(os);
1545
1546 dmu_tx_hold_zap(tx, lr->lr_doid, B_FALSE, name);
1547 dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END);
1548
1549 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1550 if (txg == 0) {
1551 ztest_object_unlock(zd, object);
1552 return (ENOSPC);
1553 }
1554
1555 if (doi.doi_type == DMU_OT_ZAP_OTHER) {
1556 VERIFY3U(0, ==, zap_destroy(os, object, tx));
1557 } else {
1558 VERIFY3U(0, ==, dmu_object_free(os, object, tx));
1559 }
1560
1561 VERIFY3U(0, ==, zap_remove(os, lr->lr_doid, name, tx));
1562
1563 (void) ztest_log_remove(zd, tx, lr, object);
1564
1565 dmu_tx_commit(tx);
1566
1567 ztest_object_unlock(zd, object);
1568
1569 return (0);
1570 }
1571
1572 static int
1573 ztest_replay_write(void *arg1, void *arg2, boolean_t byteswap)
1574 {
1575 ztest_ds_t *zd = arg1;
1576 lr_write_t *lr = arg2;
1577 objset_t *os = zd->zd_os;
1578 void *data = lr + 1; /* data follows lr */
1579 uint64_t offset, length;
1580 ztest_block_tag_t *bt = data;
1581 ztest_block_tag_t *bbt;
1582 uint64_t gen, txg, lrtxg, crtxg;
1583 dmu_object_info_t doi;
1584 dmu_tx_t *tx;
1585 dmu_buf_t *db;
1586 arc_buf_t *abuf = NULL;
1587 rl_t *rl;
1588
1589 if (byteswap)
1590 byteswap_uint64_array(lr, sizeof (*lr));
1591
1592 offset = lr->lr_offset;
1593 length = lr->lr_length;
1594
1595 /* If it's a dmu_sync() block, write the whole block */
1596 if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
1597 uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
1598 if (length < blocksize) {
1599 offset -= offset % blocksize;
1600 length = blocksize;
1601 }
1602 }
1603
1604 if (bt->bt_magic == BSWAP_64(BT_MAGIC))
1605 byteswap_uint64_array(bt, sizeof (*bt));
1606
1607 if (bt->bt_magic != BT_MAGIC)
1608 bt = NULL;
1609
1610 ztest_object_lock(zd, lr->lr_foid, RL_READER);
1611 rl = ztest_range_lock(zd, lr->lr_foid, offset, length, RL_WRITER);
1612
1613 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1614
1615 dmu_object_info_from_db(db, &doi);
1616
1617 bbt = ztest_bt_bonus(db);
1618 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1619 gen = bbt->bt_gen;
1620 crtxg = bbt->bt_crtxg;
1621 lrtxg = lr->lr_common.lrc_txg;
1622
1623 tx = dmu_tx_create(os);
1624
1625 dmu_tx_hold_write(tx, lr->lr_foid, offset, length);
1626
1627 if (ztest_random(8) == 0 && length == doi.doi_data_block_size &&
1628 P2PHASE(offset, length) == 0)
1629 abuf = dmu_request_arcbuf(db, length);
1630
1631 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1632 if (txg == 0) {
1633 if (abuf != NULL)
1634 dmu_return_arcbuf(abuf);
1635 dmu_buf_rele(db, FTAG);
1636 ztest_range_unlock(rl);
1637 ztest_object_unlock(zd, lr->lr_foid);
1638 return (ENOSPC);
1639 }
1640
1641 if (bt != NULL) {
1642 /*
1643 * Usually, verify the old data before writing new data --
1644 * but not always, because we also want to verify correct
1645 * behavior when the data was not recently read into cache.
1646 */
1647 ASSERT(offset % doi.doi_data_block_size == 0);
1648 if (ztest_random(4) != 0) {
1649 int prefetch = ztest_random(2) ?
1650 DMU_READ_PREFETCH : DMU_READ_NO_PREFETCH;
1651 ztest_block_tag_t rbt;
1652
1653 VERIFY(dmu_read(os, lr->lr_foid, offset,
1654 sizeof (rbt), &rbt, prefetch) == 0);
1655 if (rbt.bt_magic == BT_MAGIC) {
1656 ztest_bt_verify(&rbt, os, lr->lr_foid,
1657 offset, gen, txg, crtxg);
1658 }
1659 }
1660
1661 /*
1662 * Writes can appear to be newer than the bonus buffer because
1663 * the ztest_get_data() callback does a dmu_read() of the
1664 * open-context data, which may be different than the data
1665 * as it was when the write was generated.
1666 */
1667 if (zd->zd_zilog->zl_replay) {
1668 ztest_bt_verify(bt, os, lr->lr_foid, offset,
1669 MAX(gen, bt->bt_gen), MAX(txg, lrtxg),
1670 bt->bt_crtxg);
1671 }
1672
1673 /*
1674 * Set the bt's gen/txg to the bonus buffer's gen/txg
1675 * so that all of the usual ASSERTs will work.
1676 */
1677 ztest_bt_generate(bt, os, lr->lr_foid, offset, gen, txg, crtxg);
1678 }
1679
1680 if (abuf == NULL) {
1681 dmu_write(os, lr->lr_foid, offset, length, data, tx);
1682 } else {
1683 bcopy(data, abuf->b_data, length);
1684 dmu_assign_arcbuf(db, offset, abuf, tx);
1685 }
1686
1687 (void) ztest_log_write(zd, tx, lr);
1688
1689 dmu_buf_rele(db, FTAG);
1690
1691 dmu_tx_commit(tx);
1692
1693 ztest_range_unlock(rl);
1694 ztest_object_unlock(zd, lr->lr_foid);
1695
1696 return (0);
1697 }
1698
1699 static int
1700 ztest_replay_truncate(void *arg1, void *arg2, boolean_t byteswap)
1701 {
1702 ztest_ds_t *zd = arg1;
1703 lr_truncate_t *lr = arg2;
1704 objset_t *os = zd->zd_os;
1705 dmu_tx_t *tx;
1706 uint64_t txg;
1707 rl_t *rl;
1708
1709 if (byteswap)
1710 byteswap_uint64_array(lr, sizeof (*lr));
1711
1712 ztest_object_lock(zd, lr->lr_foid, RL_READER);
1713 rl = ztest_range_lock(zd, lr->lr_foid, lr->lr_offset, lr->lr_length,
1714 RL_WRITER);
1715
1716 tx = dmu_tx_create(os);
1717
1718 dmu_tx_hold_free(tx, lr->lr_foid, lr->lr_offset, lr->lr_length);
1719
1720 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1721 if (txg == 0) {
1722 ztest_range_unlock(rl);
1723 ztest_object_unlock(zd, lr->lr_foid);
1724 return (ENOSPC);
1725 }
1726
1727 VERIFY(dmu_free_range(os, lr->lr_foid, lr->lr_offset,
1728 lr->lr_length, tx) == 0);
1729
1730 (void) ztest_log_truncate(zd, tx, lr);
1731
1732 dmu_tx_commit(tx);
1733
1734 ztest_range_unlock(rl);
1735 ztest_object_unlock(zd, lr->lr_foid);
1736
1737 return (0);
1738 }
1739
1740 static int
1741 ztest_replay_setattr(void *arg1, void *arg2, boolean_t byteswap)
1742 {
1743 ztest_ds_t *zd = arg1;
1744 lr_setattr_t *lr = arg2;
1745 objset_t *os = zd->zd_os;
1746 dmu_tx_t *tx;
1747 dmu_buf_t *db;
1748 ztest_block_tag_t *bbt;
1749 uint64_t txg, lrtxg, crtxg;
1750
1751 if (byteswap)
1752 byteswap_uint64_array(lr, sizeof (*lr));
1753
1754 ztest_object_lock(zd, lr->lr_foid, RL_WRITER);
1755
1756 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1757
1758 tx = dmu_tx_create(os);
1759 dmu_tx_hold_bonus(tx, lr->lr_foid);
1760
1761 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1762 if (txg == 0) {
1763 dmu_buf_rele(db, FTAG);
1764 ztest_object_unlock(zd, lr->lr_foid);
1765 return (ENOSPC);
1766 }
1767
1768 bbt = ztest_bt_bonus(db);
1769 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1770 crtxg = bbt->bt_crtxg;
1771 lrtxg = lr->lr_common.lrc_txg;
1772
1773 if (zd->zd_zilog->zl_replay) {
1774 ASSERT(lr->lr_size != 0);
1775 ASSERT(lr->lr_mode != 0);
1776 ASSERT(lrtxg != 0);
1777 } else {
1778 /*
1779 * Randomly change the size and increment the generation.
1780 */
1781 lr->lr_size = (ztest_random(db->db_size / sizeof (*bbt)) + 1) *
1782 sizeof (*bbt);
1783 lr->lr_mode = bbt->bt_gen + 1;
1784 ASSERT(lrtxg == 0);
1785 }
1786
1787 /*
1788 * Verify that the current bonus buffer is not newer than our txg.
1789 */
1790 ztest_bt_verify(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode,
1791 MAX(txg, lrtxg), crtxg);
1792
1793 dmu_buf_will_dirty(db, tx);
1794
1795 ASSERT3U(lr->lr_size, >=, sizeof (*bbt));
1796 ASSERT3U(lr->lr_size, <=, db->db_size);
1797 VERIFY0(dmu_set_bonus(db, lr->lr_size, tx));
1798 bbt = ztest_bt_bonus(db);
1799
1800 ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode, txg, crtxg);
1801
1802 dmu_buf_rele(db, FTAG);
1803
1804 (void) ztest_log_setattr(zd, tx, lr);
1805
1806 dmu_tx_commit(tx);
1807
1808 ztest_object_unlock(zd, lr->lr_foid);
1809
1810 return (0);
1811 }
1812
1813 zil_replay_func_t *ztest_replay_vector[TX_MAX_TYPE] = {
1814 NULL, /* 0 no such transaction type */
1815 ztest_replay_create, /* TX_CREATE */
1816 NULL, /* TX_MKDIR */
1817 NULL, /* TX_MKXATTR */
1818 NULL, /* TX_SYMLINK */
1819 ztest_replay_remove, /* TX_REMOVE */
1820 NULL, /* TX_RMDIR */
1821 NULL, /* TX_LINK */
1822 NULL, /* TX_RENAME */
1823 ztest_replay_write, /* TX_WRITE */
1824 ztest_replay_truncate, /* TX_TRUNCATE */
1825 ztest_replay_setattr, /* TX_SETATTR */
1826 NULL, /* TX_ACL */
1827 NULL, /* TX_CREATE_ACL */
1828 NULL, /* TX_CREATE_ATTR */
1829 NULL, /* TX_CREATE_ACL_ATTR */
1830 NULL, /* TX_MKDIR_ACL */
1831 NULL, /* TX_MKDIR_ATTR */
1832 NULL, /* TX_MKDIR_ACL_ATTR */
1833 NULL, /* TX_WRITE2 */
1834 };
1835
1836 /*
1837 * ZIL get_data callbacks
1838 */
1839
1840 static void
1841 ztest_get_done(zgd_t *zgd, int error)
1842 {
1843 ztest_ds_t *zd = zgd->zgd_private;
1844 uint64_t object = zgd->zgd_rl->rl_object;
1845
1846 if (zgd->zgd_db)
1847 dmu_buf_rele(zgd->zgd_db, zgd);
1848
1849 ztest_range_unlock(zgd->zgd_rl);
1850 ztest_object_unlock(zd, object);
1851
1852 if (error == 0 && zgd->zgd_bp)
1853 zil_lwb_add_block(zgd->zgd_lwb, zgd->zgd_bp);
1854
1855 umem_free(zgd, sizeof (*zgd));
1856 }
1857
1858 static int
1859 ztest_get_data(void *arg, lr_write_t *lr, char *buf, struct lwb *lwb,
1860 zio_t *zio)
1861 {
1862 ztest_ds_t *zd = arg;
1863 objset_t *os = zd->zd_os;
1864 uint64_t object = lr->lr_foid;
1865 uint64_t offset = lr->lr_offset;
1866 uint64_t size = lr->lr_length;
1867 uint64_t txg = lr->lr_common.lrc_txg;
1868 uint64_t crtxg;
1869 dmu_object_info_t doi;
1870 dmu_buf_t *db;
1871 zgd_t *zgd;
1872 int error;
1873
1874 ASSERT3P(lwb, !=, NULL);
1875 ASSERT3P(zio, !=, NULL);
1876 ASSERT3U(size, !=, 0);
1877
1878 ztest_object_lock(zd, object, RL_READER);
1879 error = dmu_bonus_hold(os, object, FTAG, &db);
1880 if (error) {
1881 ztest_object_unlock(zd, object);
1882 return (error);
1883 }
1884
1885 crtxg = ztest_bt_bonus(db)->bt_crtxg;
1886
1887 if (crtxg == 0 || crtxg > txg) {
1888 dmu_buf_rele(db, FTAG);
1889 ztest_object_unlock(zd, object);
1890 return (ENOENT);
1891 }
1892
1893 dmu_object_info_from_db(db, &doi);
1894 dmu_buf_rele(db, FTAG);
1895 db = NULL;
1896
1897 zgd = umem_zalloc(sizeof (*zgd), UMEM_NOFAIL);
1898 zgd->zgd_lwb = lwb;
1899 zgd->zgd_private = zd;
1900
1901 if (buf != NULL) { /* immediate write */
1902 zgd->zgd_rl = ztest_range_lock(zd, object, offset, size,
1903 RL_READER);
1904
1905 error = dmu_read(os, object, offset, size, buf,
1906 DMU_READ_NO_PREFETCH);
1907 ASSERT(error == 0);
1908 } else {
1909 size = doi.doi_data_block_size;
1910 if (ISP2(size)) {
1911 offset = P2ALIGN(offset, size);
1912 } else {
1913 ASSERT(offset < size);
1914 offset = 0;
1915 }
1916
1917 zgd->zgd_rl = ztest_range_lock(zd, object, offset, size,
1918 RL_READER);
1919
1920 error = dmu_buf_hold(os, object, offset, zgd, &db,
1921 DMU_READ_NO_PREFETCH);
1922
1923 if (error == 0) {
1924 blkptr_t *bp = &lr->lr_blkptr;
1925
1926 zgd->zgd_db = db;
1927 zgd->zgd_bp = bp;
1928
1929 ASSERT(db->db_offset == offset);
1930 ASSERT(db->db_size == size);
1931
1932 error = dmu_sync(zio, lr->lr_common.lrc_txg,
1933 ztest_get_done, zgd);
1934
1935 if (error == 0)
1936 return (0);
1937 }
1938 }
1939
1940 ztest_get_done(zgd, error);
1941
1942 return (error);
1943 }
1944
1945 static void *
1946 ztest_lr_alloc(size_t lrsize, char *name)
1947 {
1948 char *lr;
1949 size_t namesize = name ? strlen(name) + 1 : 0;
1950
1951 lr = umem_zalloc(lrsize + namesize, UMEM_NOFAIL);
1952
1953 if (name)
1954 bcopy(name, lr + lrsize, namesize);
1955
1956 return (lr);
1957 }
1958
1959 void
1960 ztest_lr_free(void *lr, size_t lrsize, char *name)
1961 {
1962 size_t namesize = name ? strlen(name) + 1 : 0;
1963
1964 umem_free(lr, lrsize + namesize);
1965 }
1966
1967 /*
1968 * Lookup a bunch of objects. Returns the number of objects not found.
1969 */
1970 static int
1971 ztest_lookup(ztest_ds_t *zd, ztest_od_t *od, int count)
1972 {
1973 int missing = 0;
1974 int error;
1975
1976 ASSERT(MUTEX_HELD(&zd->zd_dirobj_lock));
1977
1978 for (int i = 0; i < count; i++, od++) {
1979 od->od_object = 0;
1980 error = zap_lookup(zd->zd_os, od->od_dir, od->od_name,
1981 sizeof (uint64_t), 1, &od->od_object);
1982 if (error) {
1983 ASSERT(error == ENOENT);
1984 ASSERT(od->od_object == 0);
1985 missing++;
1986 } else {
1987 dmu_buf_t *db;
1988 ztest_block_tag_t *bbt;
1989 dmu_object_info_t doi;
1990
1991 ASSERT(od->od_object != 0);
1992 ASSERT(missing == 0); /* there should be no gaps */
1993
1994 ztest_object_lock(zd, od->od_object, RL_READER);
1995 VERIFY3U(0, ==, dmu_bonus_hold(zd->zd_os,
1996 od->od_object, FTAG, &db));
1997 dmu_object_info_from_db(db, &doi);
1998 bbt = ztest_bt_bonus(db);
1999 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
2000 od->od_type = doi.doi_type;
2001 od->od_blocksize = doi.doi_data_block_size;
2002 od->od_gen = bbt->bt_gen;
2003 dmu_buf_rele(db, FTAG);
2004 ztest_object_unlock(zd, od->od_object);
2005 }
2006 }
2007
2008 return (missing);
2009 }
2010
2011 static int
2012 ztest_create(ztest_ds_t *zd, ztest_od_t *od, int count)
2013 {
2014 int missing = 0;
2015
2016 ASSERT(MUTEX_HELD(&zd->zd_dirobj_lock));
2017
2018 for (int i = 0; i < count; i++, od++) {
2019 if (missing) {
2020 od->od_object = 0;
2021 missing++;
2022 continue;
2023 }
2024
2025 lr_create_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
2026
2027 lr->lr_doid = od->od_dir;
2028 lr->lr_foid = 0; /* 0 to allocate, > 0 to claim */
2029 lr->lrz_type = od->od_crtype;
2030 lr->lrz_blocksize = od->od_crblocksize;
2031 lr->lrz_ibshift = ztest_random_ibshift();
2032 lr->lrz_bonustype = DMU_OT_UINT64_OTHER;
2033 lr->lrz_bonuslen = dmu_bonus_max();
2034 lr->lr_gen = od->od_crgen;
2035 lr->lr_crtime[0] = time(NULL);
2036
2037 if (ztest_replay_create(zd, lr, B_FALSE) != 0) {
2038 ASSERT(missing == 0);
2039 od->od_object = 0;
2040 missing++;
2041 } else {
2042 od->od_object = lr->lr_foid;
2043 od->od_type = od->od_crtype;
2044 od->od_blocksize = od->od_crblocksize;
2045 od->od_gen = od->od_crgen;
2046 ASSERT(od->od_object != 0);
2047 }
2048
2049 ztest_lr_free(lr, sizeof (*lr), od->od_name);
2050 }
2051
2052 return (missing);
2053 }
2054
2055 static int
2056 ztest_remove(ztest_ds_t *zd, ztest_od_t *od, int count)
2057 {
2058 int missing = 0;
2059 int error;
2060
2061 ASSERT(MUTEX_HELD(&zd->zd_dirobj_lock));
2062
2063 od += count - 1;
2064
2065 for (int i = count - 1; i >= 0; i--, od--) {
2066 if (missing) {
2067 missing++;
2068 continue;
2069 }
2070
2071 /*
2072 * No object was found.
2073 */
2074 if (od->od_object == 0)
2075 continue;
2076
2077 lr_remove_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
2078
2079 lr->lr_doid = od->od_dir;
2080
2081 if ((error = ztest_replay_remove(zd, lr, B_FALSE)) != 0) {
2082 ASSERT3U(error, ==, ENOSPC);
2083 missing++;
2084 } else {
2085 od->od_object = 0;
2086 }
2087 ztest_lr_free(lr, sizeof (*lr), od->od_name);
2088 }
2089
2090 return (missing);
2091 }
2092
2093 static int
2094 ztest_write(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size,
2095 void *data)
2096 {
2097 lr_write_t *lr;
2098 int error;
2099
2100 lr = ztest_lr_alloc(sizeof (*lr) + size, NULL);
2101
2102 lr->lr_foid = object;
2103 lr->lr_offset = offset;
2104 lr->lr_length = size;
2105 lr->lr_blkoff = 0;
2106 BP_ZERO(&lr->lr_blkptr);
2107
2108 bcopy(data, lr + 1, size);
2109
2110 error = ztest_replay_write(zd, lr, B_FALSE);
2111
2112 ztest_lr_free(lr, sizeof (*lr) + size, NULL);
2113
2114 return (error);
2115 }
2116
2117 static int
2118 ztest_truncate(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2119 {
2120 lr_truncate_t *lr;
2121 int error;
2122
2123 lr = ztest_lr_alloc(sizeof (*lr), NULL);
2124
2125 lr->lr_foid = object;
2126 lr->lr_offset = offset;
2127 lr->lr_length = size;
2128
2129 error = ztest_replay_truncate(zd, lr, B_FALSE);
2130
2131 ztest_lr_free(lr, sizeof (*lr), NULL);
2132
2133 return (error);
2134 }
2135
2136 static int
2137 ztest_setattr(ztest_ds_t *zd, uint64_t object)
2138 {
2139 lr_setattr_t *lr;
2140 int error;
2141
2142 lr = ztest_lr_alloc(sizeof (*lr), NULL);
2143
2144 lr->lr_foid = object;
2145 lr->lr_size = 0;
2146 lr->lr_mode = 0;
2147
2148 error = ztest_replay_setattr(zd, lr, B_FALSE);
2149
2150 ztest_lr_free(lr, sizeof (*lr), NULL);
2151
2152 return (error);
2153 }
2154
2155 static void
2156 ztest_prealloc(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2157 {
2158 objset_t *os = zd->zd_os;
2159 dmu_tx_t *tx;
2160 uint64_t txg;
2161 rl_t *rl;
2162
2163 txg_wait_synced(dmu_objset_pool(os), 0);
2164
2165 ztest_object_lock(zd, object, RL_READER);
2166 rl = ztest_range_lock(zd, object, offset, size, RL_WRITER);
2167
2168 tx = dmu_tx_create(os);
2169
2170 dmu_tx_hold_write(tx, object, offset, size);
2171
2172 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
2173
2174 if (txg != 0) {
2175 dmu_prealloc(os, object, offset, size, tx);
2176 dmu_tx_commit(tx);
2177 txg_wait_synced(dmu_objset_pool(os), txg);
2178 } else {
2179 (void) dmu_free_long_range(os, object, offset, size);
2180 }
2181
2182 ztest_range_unlock(rl);
2183 ztest_object_unlock(zd, object);
2184 }
2185
2186 static void
2187 ztest_io(ztest_ds_t *zd, uint64_t object, uint64_t offset)
2188 {
2189 int err;
2190 ztest_block_tag_t wbt;
2191 dmu_object_info_t doi;
2192 enum ztest_io_type io_type;
2193 uint64_t blocksize;
2194 void *data;
2195
2196 VERIFY(dmu_object_info(zd->zd_os, object, &doi) == 0);
2197 blocksize = doi.doi_data_block_size;
2198 data = umem_alloc(blocksize, UMEM_NOFAIL);
2199
2200 /*
2201 * Pick an i/o type at random, biased toward writing block tags.
2202 */
2203 io_type = ztest_random(ZTEST_IO_TYPES);
2204 if (ztest_random(2) == 0)
2205 io_type = ZTEST_IO_WRITE_TAG;
2206
2207 rw_enter(&zd->zd_zilog_lock, RW_READER);
2208
2209 switch (io_type) {
2210
2211 case ZTEST_IO_WRITE_TAG:
2212 ztest_bt_generate(&wbt, zd->zd_os, object, offset, 0, 0, 0);
2213 (void) ztest_write(zd, object, offset, sizeof (wbt), &wbt);
2214 break;
2215
2216 case ZTEST_IO_WRITE_PATTERN:
2217 (void) memset(data, 'a' + (object + offset) % 5, blocksize);
2218 if (ztest_random(2) == 0) {
2219 /*
2220 * Induce fletcher2 collisions to ensure that
2221 * zio_ddt_collision() detects and resolves them
2222 * when using fletcher2-verify for deduplication.
2223 */
2224 ((uint64_t *)data)[0] ^= 1ULL << 63;
2225 ((uint64_t *)data)[4] ^= 1ULL << 63;
2226 }
2227 (void) ztest_write(zd, object, offset, blocksize, data);
2228 break;
2229
2230 case ZTEST_IO_WRITE_ZEROES:
2231 bzero(data, blocksize);
2232 (void) ztest_write(zd, object, offset, blocksize, data);
2233 break;
2234
2235 case ZTEST_IO_TRUNCATE:
2236 (void) ztest_truncate(zd, object, offset, blocksize);
2237 break;
2238
2239 case ZTEST_IO_SETATTR:
2240 (void) ztest_setattr(zd, object);
2241 break;
2242
2243 case ZTEST_IO_REWRITE:
2244 rw_enter(&ztest_name_lock, RW_READER);
2245 err = ztest_dsl_prop_set_uint64(zd->zd_name,
2246 ZFS_PROP_CHECKSUM, spa_dedup_checksum(ztest_spa),
2247 B_FALSE);
2248 VERIFY(err == 0 || err == ENOSPC);
2249 err = ztest_dsl_prop_set_uint64(zd->zd_name,
2250 ZFS_PROP_COMPRESSION,
2251 ztest_random_dsl_prop(ZFS_PROP_COMPRESSION),
2252 B_FALSE);
2253 VERIFY(err == 0 || err == ENOSPC);
2254 rw_exit(&ztest_name_lock);
2255
2256 VERIFY0(dmu_read(zd->zd_os, object, offset, blocksize, data,
2257 DMU_READ_NO_PREFETCH));
2258
2259 (void) ztest_write(zd, object, offset, blocksize, data);
2260 break;
2261 }
2262
2263 rw_exit(&zd->zd_zilog_lock);
2264
2265 umem_free(data, blocksize);
2266 }
2267
2268 /*
2269 * Initialize an object description template.
2270 */
2271 static void
2272 ztest_od_init(ztest_od_t *od, uint64_t id, char *tag, uint64_t index,
2273 dmu_object_type_t type, uint64_t blocksize, uint64_t gen)
2274 {
2275 od->od_dir = ZTEST_DIROBJ;
2276 od->od_object = 0;
2277
2278 od->od_crtype = type;
2279 od->od_crblocksize = blocksize ? blocksize : ztest_random_blocksize();
2280 od->od_crgen = gen;
2281
2282 od->od_type = DMU_OT_NONE;
2283 od->od_blocksize = 0;
2284 od->od_gen = 0;
2285
2286 (void) snprintf(od->od_name, sizeof (od->od_name), "%s(%lld)[%llu]",
2287 tag, (int64_t)id, index);
2288 }
2289
2290 /*
2291 * Lookup or create the objects for a test using the od template.
2292 * If the objects do not all exist, or if 'remove' is specified,
2293 * remove any existing objects and create new ones. Otherwise,
2294 * use the existing objects.
2295 */
2296 static int
2297 ztest_object_init(ztest_ds_t *zd, ztest_od_t *od, size_t size, boolean_t remove)
2298 {
2299 int count = size / sizeof (*od);
2300 int rv = 0;
2301
2302 mutex_enter(&zd->zd_dirobj_lock);
2303 if ((ztest_lookup(zd, od, count) != 0 || remove) &&
2304 (ztest_remove(zd, od, count) != 0 ||
2305 ztest_create(zd, od, count) != 0))
2306 rv = -1;
2307 zd->zd_od = od;
2308 mutex_exit(&zd->zd_dirobj_lock);
2309
2310 return (rv);
2311 }
2312
2313 /* ARGSUSED */
2314 void
2315 ztest_zil_commit(ztest_ds_t *zd, uint64_t id)
2316 {
2317 zilog_t *zilog = zd->zd_zilog;
2318
2319 rw_enter(&zd->zd_zilog_lock, RW_READER);
2320
2321 zil_commit(zilog, ztest_random(ZTEST_OBJECTS));
2322
2323 /*
2324 * Remember the committed values in zd, which is in parent/child
2325 * shared memory. If we die, the next iteration of ztest_run()
2326 * will verify that the log really does contain this record.
2327 */
2328 mutex_enter(&zilog->zl_lock);
2329 ASSERT(zd->zd_shared != NULL);
2330 ASSERT3U(zd->zd_shared->zd_seq, <=, zilog->zl_commit_lr_seq);
2331 zd->zd_shared->zd_seq = zilog->zl_commit_lr_seq;
2332 mutex_exit(&zilog->zl_lock);
2333
2334 rw_exit(&zd->zd_zilog_lock);
2335 }
2336
2337 /*
2338 * This function is designed to simulate the operations that occur during a
2339 * mount/unmount operation. We hold the dataset across these operations in an
2340 * attempt to expose any implicit assumptions about ZIL management.
2341 */
2342 /* ARGSUSED */
2343 void
2344 ztest_zil_remount(ztest_ds_t *zd, uint64_t id)
2345 {
2346 objset_t *os = zd->zd_os;
2347
2348 /*
2349 * We grab the zd_dirobj_lock to ensure that no other thread is
2350 * updating the zil (i.e. adding in-memory log records) and the
2351 * zd_zilog_lock to block any I/O.
2352 */
2353 mutex_enter(&zd->zd_dirobj_lock);
2354 rw_enter(&zd->zd_zilog_lock, RW_WRITER);
2355
2356 /* zfsvfs_teardown() */
2357 zil_close(zd->zd_zilog);
2358
2359 /* zfsvfs_setup() */
2360 VERIFY(zil_open(os, ztest_get_data) == zd->zd_zilog);
2361 zil_replay(os, zd, ztest_replay_vector);
2362
2363 rw_exit(&zd->zd_zilog_lock);
2364 mutex_exit(&zd->zd_dirobj_lock);
2365 }
2366
2367 /*
2368 * Verify that we can't destroy an active pool, create an existing pool,
2369 * or create a pool with a bad vdev spec.
2370 */
2371 /* ARGSUSED */
2372 void
2373 ztest_spa_create_destroy(ztest_ds_t *zd, uint64_t id)
2374 {
2375 ztest_shared_opts_t *zo = &ztest_opts;
2376 spa_t *spa;
2377 nvlist_t *nvroot;
2378
2379 /*
2380 * Attempt to create using a bad file.
2381 */
2382 nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1);
2383 VERIFY3U(ENOENT, ==,
2384 spa_create("ztest_bad_file", nvroot, NULL, NULL));
2385 nvlist_free(nvroot);
2386
2387 /*
2388 * Attempt to create using a bad mirror.
2389 */
2390 nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 2, 1);
2391 VERIFY3U(ENOENT, ==,
2392 spa_create("ztest_bad_mirror", nvroot, NULL, NULL));
2393 nvlist_free(nvroot);
2394
2395 /*
2396 * Attempt to create an existing pool. It shouldn't matter
2397 * what's in the nvroot; we should fail with EEXIST.
2398 */
2399 rw_enter(&ztest_name_lock, RW_READER);
2400 nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1);
2401 VERIFY3U(EEXIST, ==, spa_create(zo->zo_pool, nvroot, NULL, NULL));
2402 nvlist_free(nvroot);
2403 VERIFY3U(0, ==, spa_open(zo->zo_pool, &spa, FTAG));
2404 VERIFY3U(EBUSY, ==, spa_destroy(zo->zo_pool));
2405 spa_close(spa, FTAG);
2406
2407 rw_exit(&ztest_name_lock);
2408 }
2409
2410 /* ARGSUSED */
2411 void
2412 ztest_spa_upgrade(ztest_ds_t *zd, uint64_t id)
2413 {
2414 spa_t *spa;
2415 uint64_t initial_version = SPA_VERSION_INITIAL;
2416 uint64_t version, newversion;
2417 nvlist_t *nvroot, *props;
2418 char *name;
2419
2420 mutex_enter(&ztest_vdev_lock);
2421 name = kmem_asprintf("%s_upgrade", ztest_opts.zo_pool);
2422
2423 /*
2424 * Clean up from previous runs.
2425 */
2426 (void) spa_destroy(name);
2427
2428 nvroot = make_vdev_root(NULL, NULL, name, ztest_opts.zo_vdev_size, 0,
2429 0, ztest_opts.zo_raidz, ztest_opts.zo_mirrors, 1);
2430
2431 /*
2432 * If we're configuring a RAIDZ device then make sure that the
2433 * the initial version is capable of supporting that feature.
2434 */
2435 switch (ztest_opts.zo_raidz_parity) {
2436 case 0:
2437 case 1:
2438 initial_version = SPA_VERSION_INITIAL;
2439 break;
2440 case 2:
2441 initial_version = SPA_VERSION_RAIDZ2;
2442 break;
2443 case 3:
2444 initial_version = SPA_VERSION_RAIDZ3;
2445 break;
2446 }
2447
2448 /*
2449 * Create a pool with a spa version that can be upgraded. Pick
2450 * a value between initial_version and SPA_VERSION_BEFORE_FEATURES.
2451 */
2452 do {
2453 version = ztest_random_spa_version(initial_version);
2454 } while (version > SPA_VERSION_BEFORE_FEATURES);
2455
2456 props = fnvlist_alloc();
2457 fnvlist_add_uint64(props,
2458 zpool_prop_to_name(ZPOOL_PROP_VERSION), version);
2459 VERIFY0(spa_create(name, nvroot, props, NULL));
2460 fnvlist_free(nvroot);
2461 fnvlist_free(props);
2462
2463 VERIFY0(spa_open(name, &spa, FTAG));
2464 VERIFY3U(spa_version(spa), ==, version);
2465 newversion = ztest_random_spa_version(version + 1);
2466
2467 if (ztest_opts.zo_verbose >= 4) {
2468 (void) printf("upgrading spa version from %llu to %llu\n",
2469 (u_longlong_t)version, (u_longlong_t)newversion);
2470 }
2471
2472 spa_upgrade(spa, newversion);
2473 VERIFY3U(spa_version(spa), >, version);
2474 VERIFY3U(spa_version(spa), ==, fnvlist_lookup_uint64(spa->spa_config,
2475 zpool_prop_to_name(ZPOOL_PROP_VERSION)));
2476 spa_close(spa, FTAG);
2477
2478 strfree(name);
2479 mutex_exit(&ztest_vdev_lock);
2480 }
2481
2482 static void
2483 ztest_spa_checkpoint(spa_t *spa)
2484 {
2485 ASSERT(MUTEX_HELD(&ztest_checkpoint_lock));
2486
2487 int error = spa_checkpoint(spa->spa_name);
2488
2489 switch (error) {
2490 case 0:
2491 case ZFS_ERR_DEVRM_IN_PROGRESS:
2492 case ZFS_ERR_DISCARDING_CHECKPOINT:
2493 case ZFS_ERR_CHECKPOINT_EXISTS:
2494 break;
2495 case ENOSPC:
2496 ztest_record_enospc(FTAG);
2497 break;
2498 default:
2499 fatal(0, "spa_checkpoint(%s) = %d", spa->spa_name, error);
2500 }
2501 }
2502
2503 static void
2504 ztest_spa_discard_checkpoint(spa_t *spa)
2505 {
2506 ASSERT(MUTEX_HELD(&ztest_checkpoint_lock));
2507
2508 int error = spa_checkpoint_discard(spa->spa_name);
2509
2510 switch (error) {
2511 case 0:
2512 case ZFS_ERR_DISCARDING_CHECKPOINT:
2513 case ZFS_ERR_NO_CHECKPOINT:
2514 break;
2515 default:
2516 fatal(0, "spa_discard_checkpoint(%s) = %d",
2517 spa->spa_name, error);
2518 }
2519
2520 }
2521
2522 /* ARGSUSED */
2523 void
2524 ztest_spa_checkpoint_create_discard(ztest_ds_t *zd, uint64_t id)
2525 {
2526 spa_t *spa = ztest_spa;
2527
2528 mutex_enter(&ztest_checkpoint_lock);
2529 if (ztest_random(2) == 0) {
2530 ztest_spa_checkpoint(spa);
2531 } else {
2532 ztest_spa_discard_checkpoint(spa);
2533 }
2534 mutex_exit(&ztest_checkpoint_lock);
2535 }
2536
2537
2538 static vdev_t *
2539 vdev_lookup_by_path(vdev_t *vd, const char *path)
2540 {
2541 vdev_t *mvd;
2542
2543 if (vd->vdev_path != NULL && strcmp(path, vd->vdev_path) == 0)
2544 return (vd);
2545
2546 for (int c = 0; c < vd->vdev_children; c++)
2547 if ((mvd = vdev_lookup_by_path(vd->vdev_child[c], path)) !=
2548 NULL)
2549 return (mvd);
2550
2551 return (NULL);
2552 }
2553
2554 /*
2555 * Find the first available hole which can be used as a top-level.
2556 */
2557 int
2558 find_vdev_hole(spa_t *spa)
2559 {
2560 vdev_t *rvd = spa->spa_root_vdev;
2561 int c;
2562
2563 ASSERT(spa_config_held(spa, SCL_VDEV, RW_READER) == SCL_VDEV);
2564
2565 for (c = 0; c < rvd->vdev_children; c++) {
2566 vdev_t *cvd = rvd->vdev_child[c];
2567
2568 if (cvd->vdev_ishole)
2569 break;
2570 }
2571 return (c);
2572 }
2573
2574 /*
2575 * Verify that vdev_add() works as expected.
2576 */
2577 /* ARGSUSED */
2578 void
2579 ztest_vdev_add_remove(ztest_ds_t *zd, uint64_t id)
2580 {
2581 ztest_shared_t *zs = ztest_shared;
2582 spa_t *spa = ztest_spa;
2583 uint64_t leaves;
2584 uint64_t guid;
2585 nvlist_t *nvroot;
2586 int error;
2587
2588 mutex_enter(&ztest_vdev_lock);
2589 leaves = MAX(zs->zs_mirrors + zs->zs_splits, 1) * ztest_opts.zo_raidz;
2590
2591 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2592
2593 ztest_shared->zs_vdev_next_leaf = find_vdev_hole(spa) * leaves;
2594
2595 /*
2596 * If we have slogs then remove them 1/4 of the time.
2597 */
2598 if (spa_has_slogs(spa) && ztest_random(4) == 0) {
2599 /*
2600 * Grab the guid from the head of the log class rotor.
2601 */
2602 guid = spa_log_class(spa)->mc_rotor->mg_vd->vdev_guid;
2603
2604 spa_config_exit(spa, SCL_VDEV, FTAG);
2605
2606 /*
2607 * We have to grab the zs_name_lock as writer to
2608 * prevent a race between removing a slog (dmu_objset_find)
2609 * and destroying a dataset. Removing the slog will
2610 * grab a reference on the dataset which may cause
2611 * dmu_objset_destroy() to fail with EBUSY thus
2612 * leaving the dataset in an inconsistent state.
2613 */
2614 rw_enter(&ztest_name_lock, RW_WRITER);
2615 error = spa_vdev_remove(spa, guid, B_FALSE);
2616 rw_exit(&ztest_name_lock);
2617
2618 switch (error) {
2619 case 0:
2620 case EEXIST:
2621 case ZFS_ERR_CHECKPOINT_EXISTS:
2622 case ZFS_ERR_DISCARDING_CHECKPOINT:
2623 break;
2624 default:
2625 fatal(0, "spa_vdev_remove() = %d", error);
2626 }
2627 } else {
2628 spa_config_exit(spa, SCL_VDEV, FTAG);
2629
2630 /*
2631 * Make 1/4 of the devices be log devices.
2632 */
2633 nvroot = make_vdev_root(NULL, NULL, NULL,
2634 ztest_opts.zo_vdev_size, 0,
2635 ztest_random(4) == 0, ztest_opts.zo_raidz,
2636 zs->zs_mirrors, 1);
2637
2638 error = spa_vdev_add(spa, nvroot);
2639 nvlist_free(nvroot);
2640
2641 switch (error) {
2642 case 0:
2643 break;
2644 case ENOSPC:
2645 ztest_record_enospc("spa_vdev_add");
2646 break;
2647 default:
2648 fatal(0, "spa_vdev_add() = %d", error);
2649 }
2650 }
2651
2652 mutex_exit(&ztest_vdev_lock);
2653 }
2654
2655 /*
2656 * Verify that adding/removing aux devices (l2arc, hot spare) works as expected.
2657 */
2658 /* ARGSUSED */
2659 void
2660 ztest_vdev_aux_add_remove(ztest_ds_t *zd, uint64_t id)
2661 {
2662 ztest_shared_t *zs = ztest_shared;
2663 spa_t *spa = ztest_spa;
2664 vdev_t *rvd = spa->spa_root_vdev;
2665 spa_aux_vdev_t *sav;
2666 char *aux;
2667 uint64_t guid = 0;
2668 int error;
2669
2670 if (ztest_random(2) == 0) {
2671 sav = &spa->spa_spares;
2672 aux = ZPOOL_CONFIG_SPARES;
2673 } else {
2674 sav = &spa->spa_l2cache;
2675 aux = ZPOOL_CONFIG_L2CACHE;
2676 }
2677
2678 mutex_enter(&ztest_vdev_lock);
2679
2680 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2681
2682 if (sav->sav_count != 0 && ztest_random(4) == 0) {
2683 /*
2684 * Pick a random device to remove.
2685 */
2686 guid = sav->sav_vdevs[ztest_random(sav->sav_count)]->vdev_guid;
2687 } else {
2688 /*
2689 * Find an unused device we can add.
2690 */
2691 zs->zs_vdev_aux = 0;
2692 for (;;) {
2693 char path[MAXPATHLEN];
2694 int c;
2695 (void) snprintf(path, sizeof (path), ztest_aux_template,
2696 ztest_opts.zo_dir, ztest_opts.zo_pool, aux,
2697 zs->zs_vdev_aux);
2698 for (c = 0; c < sav->sav_count; c++)
2699 if (strcmp(sav->sav_vdevs[c]->vdev_path,
2700 path) == 0)
2701 break;
2702 if (c == sav->sav_count &&
2703 vdev_lookup_by_path(rvd, path) == NULL)
2704 break;
2705 zs->zs_vdev_aux++;
2706 }
2707 }
2708
2709 spa_config_exit(spa, SCL_VDEV, FTAG);
2710
2711 if (guid == 0) {
2712 /*
2713 * Add a new device.
2714 */
2715 nvlist_t *nvroot = make_vdev_root(NULL, aux, NULL,
2716 (ztest_opts.zo_vdev_size * 5) / 4, 0, 0, 0, 0, 1);
2717 error = spa_vdev_add(spa, nvroot);
2718
2719 switch (error) {
2720 case 0:
2721 break;
2722 default:
2723 fatal(0, "spa_vdev_add(%p) = %d", nvroot, error);
2724 }
2725 nvlist_free(nvroot);
2726 } else {
2727 /*
2728 * Remove an existing device. Sometimes, dirty its
2729 * vdev state first to make sure we handle removal
2730 * of devices that have pending state changes.
2731 */
2732 if (ztest_random(2) == 0)
2733 (void) vdev_online(spa, guid, 0, NULL);
2734
2735 error = spa_vdev_remove(spa, guid, B_FALSE);
2736
2737 switch (error) {
2738 case 0:
2739 case EBUSY:
2740 case ZFS_ERR_CHECKPOINT_EXISTS:
2741 case ZFS_ERR_DISCARDING_CHECKPOINT:
2742 break;
2743 default:
2744 fatal(0, "spa_vdev_remove(%llu) = %d", guid, error);
2745 }
2746 }
2747
2748 mutex_exit(&ztest_vdev_lock);
2749 }
2750
2751 /*
2752 * split a pool if it has mirror tlvdevs
2753 */
2754 /* ARGSUSED */
2755 void
2756 ztest_split_pool(ztest_ds_t *zd, uint64_t id)
2757 {
2758 ztest_shared_t *zs = ztest_shared;
2759 spa_t *spa = ztest_spa;
2760 vdev_t *rvd = spa->spa_root_vdev;
2761 nvlist_t *tree, **child, *config, *split, **schild;
2762 uint_t c, children, schildren = 0, lastlogid = 0;
2763 int error = 0;
2764
2765 mutex_enter(&ztest_vdev_lock);
2766
2767 /* ensure we have a useable config; mirrors of raidz aren't supported */
2768 if (zs->zs_mirrors < 3 || ztest_opts.zo_raidz > 1) {
2769 mutex_exit(&ztest_vdev_lock);
2770 return;
2771 }
2772
2773 /* clean up the old pool, if any */
2774 (void) spa_destroy("splitp");
2775
2776 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2777
2778 /* generate a config from the existing config */
2779 mutex_enter(&spa->spa_props_lock);
2780 VERIFY(nvlist_lookup_nvlist(spa->spa_config, ZPOOL_CONFIG_VDEV_TREE,
2781 &tree) == 0);
2782 mutex_exit(&spa->spa_props_lock);
2783
2784 VERIFY(nvlist_lookup_nvlist_array(tree, ZPOOL_CONFIG_CHILDREN, &child,
2785 &children) == 0);
2786
2787 schild = malloc(rvd->vdev_children * sizeof (nvlist_t *));
2788 for (c = 0; c < children; c++) {
2789 vdev_t *tvd = rvd->vdev_child[c];
2790 nvlist_t **mchild;
2791 uint_t mchildren;
2792
2793 if (tvd->vdev_islog || tvd->vdev_ops == &vdev_hole_ops) {
2794 VERIFY(nvlist_alloc(&schild[schildren], NV_UNIQUE_NAME,
2795 0) == 0);
2796 VERIFY(nvlist_add_string(schild[schildren],
2797 ZPOOL_CONFIG_TYPE, VDEV_TYPE_HOLE) == 0);
2798 VERIFY(nvlist_add_uint64(schild[schildren],
2799 ZPOOL_CONFIG_IS_HOLE, 1) == 0);
2800 if (lastlogid == 0)
2801 lastlogid = schildren;
2802 ++schildren;
2803 continue;
2804 }
2805 lastlogid = 0;
2806 VERIFY(nvlist_lookup_nvlist_array(child[c],
2807 ZPOOL_CONFIG_CHILDREN, &mchild, &mchildren) == 0);
2808 VERIFY(nvlist_dup(mchild[0], &schild[schildren++], 0) == 0);
2809 }
2810
2811 /* OK, create a config that can be used to split */
2812 VERIFY(nvlist_alloc(&split, NV_UNIQUE_NAME, 0) == 0);
2813 VERIFY(nvlist_add_string(split, ZPOOL_CONFIG_TYPE,
2814 VDEV_TYPE_ROOT) == 0);
2815 VERIFY(nvlist_add_nvlist_array(split, ZPOOL_CONFIG_CHILDREN, schild,
2816 lastlogid != 0 ? lastlogid : schildren) == 0);
2817
2818 VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, 0) == 0);
2819 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, split) == 0);
2820
2821 for (c = 0; c < schildren; c++)
2822 nvlist_free(schild[c]);
2823 free(schild);
2824 nvlist_free(split);
2825
2826 spa_config_exit(spa, SCL_VDEV, FTAG);
2827
2828 rw_enter(&ztest_name_lock, RW_WRITER);
2829 error = spa_vdev_split_mirror(spa, "splitp", config, NULL, B_FALSE);
2830 rw_exit(&ztest_name_lock);
2831
2832 nvlist_free(config);
2833
2834 if (error == 0) {
2835 (void) printf("successful split - results:\n");
2836 mutex_enter(&spa_namespace_lock);
2837 show_pool_stats(spa);
2838 show_pool_stats(spa_lookup("splitp"));
2839 mutex_exit(&spa_namespace_lock);
2840 ++zs->zs_splits;
2841 --zs->zs_mirrors;
2842 }
2843 mutex_exit(&ztest_vdev_lock);
2844 }
2845
2846 /*
2847 * Verify that we can attach and detach devices.
2848 */
2849 /* ARGSUSED */
2850 void
2851 ztest_vdev_attach_detach(ztest_ds_t *zd, uint64_t id)
2852 {
2853 ztest_shared_t *zs = ztest_shared;
2854 spa_t *spa = ztest_spa;
2855 spa_aux_vdev_t *sav = &spa->spa_spares;
2856 vdev_t *rvd = spa->spa_root_vdev;
2857 vdev_t *oldvd, *newvd, *pvd;
2858 nvlist_t *root;
2859 uint64_t leaves;
2860 uint64_t leaf, top;
2861 uint64_t ashift = ztest_get_ashift();
2862 uint64_t oldguid, pguid;
2863 uint64_t oldsize, newsize;
2864 char oldpath[MAXPATHLEN], newpath[MAXPATHLEN];
2865 int replacing;
2866 int oldvd_has_siblings = B_FALSE;
2867 int newvd_is_spare = B_FALSE;
2868 int oldvd_is_log;
2869 int error, expected_error;
2870
2871 mutex_enter(&ztest_vdev_lock);
2872 leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz;
2873
2874 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2875
2876 /*
2877 * If a vdev is in the process of being removed, its removal may
2878 * finish while we are in progress, leading to an unexpected error
2879 * value. Don't bother trying to attach while we are in the middle
2880 * of removal.
2881 */
2882 if (spa->spa_vdev_removal != NULL) {
2883 spa_config_exit(spa, SCL_ALL, FTAG);
2884 mutex_exit(&ztest_vdev_lock);
2885 return;
2886 }
2887
2888 /*
2889 * Decide whether to do an attach or a replace.
2890 */
2891 replacing = ztest_random(2);
2892
2893 /*
2894 * Pick a random top-level vdev.
2895 */
2896 top = ztest_random_vdev_top(spa, B_TRUE);
2897
2898 /*
2899 * Pick a random leaf within it.
2900 */
2901 leaf = ztest_random(leaves);
2902
2903 /*
2904 * Locate this vdev.
2905 */
2906 oldvd = rvd->vdev_child[top];
2907 if (zs->zs_mirrors >= 1) {
2908 ASSERT(oldvd->vdev_ops == &vdev_mirror_ops);
2909 ASSERT(oldvd->vdev_children >= zs->zs_mirrors);
2910 oldvd = oldvd->vdev_child[leaf / ztest_opts.zo_raidz];
2911 }
2912 if (ztest_opts.zo_raidz > 1) {
2913 ASSERT(oldvd->vdev_ops == &vdev_raidz_ops);
2914 ASSERT(oldvd->vdev_children == ztest_opts.zo_raidz);
2915 oldvd = oldvd->vdev_child[leaf % ztest_opts.zo_raidz];
2916 }
2917
2918 /*
2919 * If we're already doing an attach or replace, oldvd may be a
2920 * mirror vdev -- in which case, pick a random child.
2921 */
2922 while (oldvd->vdev_children != 0) {
2923 oldvd_has_siblings = B_TRUE;
2924 ASSERT(oldvd->vdev_children >= 2);
2925 oldvd = oldvd->vdev_child[ztest_random(oldvd->vdev_children)];
2926 }
2927
2928 oldguid = oldvd->vdev_guid;
2929 oldsize = vdev_get_min_asize(oldvd);
2930 oldvd_is_log = oldvd->vdev_top->vdev_islog;
2931 (void) strcpy(oldpath, oldvd->vdev_path);
2932 pvd = oldvd->vdev_parent;
2933 pguid = pvd->vdev_guid;
2934
2935 /*
2936 * If oldvd has siblings, then half of the time, detach it.
2937 */
2938 if (oldvd_has_siblings && ztest_random(2) == 0) {
2939 spa_config_exit(spa, SCL_ALL, FTAG);
2940 error = spa_vdev_detach(spa, oldguid, pguid, B_FALSE);
2941 if (error != 0 && error != ENODEV && error != EBUSY &&
2942 error != ENOTSUP && error != ZFS_ERR_CHECKPOINT_EXISTS &&
2943 error != ZFS_ERR_DISCARDING_CHECKPOINT)
2944 fatal(0, "detach (%s) returned %d", oldpath, error);
2945 mutex_exit(&ztest_vdev_lock);
2946 return;
2947 }
2948
2949 /*
2950 * For the new vdev, choose with equal probability between the two
2951 * standard paths (ending in either 'a' or 'b') or a random hot spare.
2952 */
2953 if (sav->sav_count != 0 && ztest_random(3) == 0) {
2954 newvd = sav->sav_vdevs[ztest_random(sav->sav_count)];
2955 newvd_is_spare = B_TRUE;
2956 (void) strcpy(newpath, newvd->vdev_path);
2957 } else {
2958 (void) snprintf(newpath, sizeof (newpath), ztest_dev_template,
2959 ztest_opts.zo_dir, ztest_opts.zo_pool,
2960 top * leaves + leaf);
2961 if (ztest_random(2) == 0)
2962 newpath[strlen(newpath) - 1] = 'b';
2963 newvd = vdev_lookup_by_path(rvd, newpath);
2964 }
2965
2966 if (newvd) {
2967 /*
2968 * Reopen to ensure the vdev's asize field isn't stale.
2969 */
2970 vdev_reopen(newvd);
2971 newsize = vdev_get_min_asize(newvd);
2972 } else {
2973 /*
2974 * Make newsize a little bigger or smaller than oldsize.
2975 * If it's smaller, the attach should fail.
2976 * If it's larger, and we're doing a replace,
2977 * we should get dynamic LUN growth when we're done.
2978 */
2979 newsize = 10 * oldsize / (9 + ztest_random(3));
2980 }
2981
2982 /*
2983 * If pvd is not a mirror or root, the attach should fail with ENOTSUP,
2984 * unless it's a replace; in that case any non-replacing parent is OK.
2985 *
2986 * If newvd is already part of the pool, it should fail with EBUSY.
2987 *
2988 * If newvd is too small, it should fail with EOVERFLOW.
2989 */
2990 if (pvd->vdev_ops != &vdev_mirror_ops &&
2991 pvd->vdev_ops != &vdev_root_ops && (!replacing ||
2992 pvd->vdev_ops == &vdev_replacing_ops ||
2993 pvd->vdev_ops == &vdev_spare_ops))
2994 expected_error = ENOTSUP;
2995 else if (newvd_is_spare && (!replacing || oldvd_is_log))
2996 expected_error = ENOTSUP;
2997 else if (newvd == oldvd)
2998 expected_error = replacing ? 0 : EBUSY;
2999 else if (vdev_lookup_by_path(rvd, newpath) != NULL)
3000 expected_error = EBUSY;
3001 else if (newsize < oldsize)
3002 expected_error = EOVERFLOW;
3003 else if (ashift > oldvd->vdev_top->vdev_ashift)
3004 expected_error = EDOM;
3005 else
3006 expected_error = 0;
3007
3008 spa_config_exit(spa, SCL_ALL, FTAG);
3009
3010 /*
3011 * Build the nvlist describing newpath.
3012 */
3013 root = make_vdev_root(newpath, NULL, NULL, newvd == NULL ? newsize : 0,
3014 ashift, 0, 0, 0, 1);
3015
3016 error = spa_vdev_attach(spa, oldguid, root, replacing);
3017
3018 nvlist_free(root);
3019
3020 /*
3021 * If our parent was the replacing vdev, but the replace completed,
3022 * then instead of failing with ENOTSUP we may either succeed,
3023 * fail with ENODEV, or fail with EOVERFLOW.
3024 */
3025 if (expected_error == ENOTSUP &&
3026 (error == 0 || error == ENODEV || error == EOVERFLOW))
3027 expected_error = error;
3028
3029 /*
3030 * If someone grew the LUN, the replacement may be too small.
3031 */
3032 if (error == EOVERFLOW || error == EBUSY)
3033 expected_error = error;
3034
3035 if (error == ZFS_ERR_CHECKPOINT_EXISTS ||
3036 error == ZFS_ERR_DISCARDING_CHECKPOINT)
3037 expected_error = error;
3038
3039 /* XXX workaround 6690467 */
3040 if (error != expected_error && expected_error != EBUSY) {
3041 fatal(0, "attach (%s %llu, %s %llu, %d) "
3042 "returned %d, expected %d",
3043 oldpath, oldsize, newpath,
3044 newsize, replacing, error, expected_error);
3045 }
3046
3047 mutex_exit(&ztest_vdev_lock);
3048 }
3049
3050 /* ARGSUSED */
3051 void
3052 ztest_device_removal(ztest_ds_t *zd, uint64_t id)
3053 {
3054 spa_t *spa = ztest_spa;
3055 vdev_t *vd;
3056 uint64_t guid;
3057
3058 mutex_enter(&ztest_vdev_lock);
3059
3060 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
3061 vd = vdev_lookup_top(spa, ztest_random_vdev_top(spa, B_FALSE));
3062 guid = vd->vdev_guid;
3063 spa_config_exit(spa, SCL_VDEV, FTAG);
3064
3065 (void) spa_vdev_remove(spa, guid, B_FALSE);
3066
3067 mutex_exit(&ztest_vdev_lock);
3068 }
3069
3070 /*
3071 * Callback function which expands the physical size of the vdev.
3072 */
3073 vdev_t *
3074 grow_vdev(vdev_t *vd, void *arg)
3075 {
3076 spa_t *spa = vd->vdev_spa;
3077 size_t *newsize = arg;
3078 size_t fsize;
3079 int fd;
3080
3081 ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
3082 ASSERT(vd->vdev_ops->vdev_op_leaf);
3083
3084 if ((fd = open(vd->vdev_path, O_RDWR)) == -1)
3085 return (vd);
3086
3087 fsize = lseek(fd, 0, SEEK_END);
3088 (void) ftruncate(fd, *newsize);
3089
3090 if (ztest_opts.zo_verbose >= 6) {
3091 (void) printf("%s grew from %lu to %lu bytes\n",
3092 vd->vdev_path, (ulong_t)fsize, (ulong_t)*newsize);
3093 }
3094 (void) close(fd);
3095 return (NULL);
3096 }
3097
3098 /*
3099 * Callback function which expands a given vdev by calling vdev_online().
3100 */
3101 /* ARGSUSED */
3102 vdev_t *
3103 online_vdev(vdev_t *vd, void *arg)
3104 {
3105 spa_t *spa = vd->vdev_spa;
3106 vdev_t *tvd = vd->vdev_top;
3107 uint64_t guid = vd->vdev_guid;
3108 uint64_t generation = spa->spa_config_generation + 1;
3109 vdev_state_t newstate = VDEV_STATE_UNKNOWN;
3110 int error;
3111
3112 ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
3113 ASSERT(vd->vdev_ops->vdev_op_leaf);
3114
3115 /* Calling vdev_online will initialize the new metaslabs */
3116 spa_config_exit(spa, SCL_STATE, spa);
3117 error = vdev_online(spa, guid, ZFS_ONLINE_EXPAND, &newstate);
3118 spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3119
3120 /*
3121 * If vdev_online returned an error or the underlying vdev_open
3122 * failed then we abort the expand. The only way to know that
3123 * vdev_open fails is by checking the returned newstate.
3124 */
3125 if (error || newstate != VDEV_STATE_HEALTHY) {
3126 if (ztest_opts.zo_verbose >= 5) {
3127 (void) printf("Unable to expand vdev, state %llu, "
3128 "error %d\n", (u_longlong_t)newstate, error);
3129 }
3130 return (vd);
3131 }
3132 ASSERT3U(newstate, ==, VDEV_STATE_HEALTHY);
3133
3134 /*
3135 * Since we dropped the lock we need to ensure that we're
3136 * still talking to the original vdev. It's possible this
3137 * vdev may have been detached/replaced while we were
3138 * trying to online it.
3139 */
3140 if (generation != spa->spa_config_generation) {
3141 if (ztest_opts.zo_verbose >= 5) {
3142 (void) printf("vdev configuration has changed, "
3143 "guid %llu, state %llu, expected gen %llu, "
3144 "got gen %llu\n",
3145 (u_longlong_t)guid,
3146 (u_longlong_t)tvd->vdev_state,
3147 (u_longlong_t)generation,
3148 (u_longlong_t)spa->spa_config_generation);
3149 }
3150 return (vd);
3151 }
3152 return (NULL);
3153 }
3154
3155 /*
3156 * Traverse the vdev tree calling the supplied function.
3157 * We continue to walk the tree until we either have walked all
3158 * children or we receive a non-NULL return from the callback.
3159 * If a NULL callback is passed, then we just return back the first
3160 * leaf vdev we encounter.
3161 */
3162 vdev_t *
3163 vdev_walk_tree(vdev_t *vd, vdev_t *(*func)(vdev_t *, void *), void *arg)
3164 {
3165 if (vd->vdev_ops->vdev_op_leaf) {
3166 if (func == NULL)
3167 return (vd);
3168 else
3169 return (func(vd, arg));
3170 }
3171
3172 for (uint_t c = 0; c < vd->vdev_children; c++) {
3173 vdev_t *cvd = vd->vdev_child[c];
3174 if ((cvd = vdev_walk_tree(cvd, func, arg)) != NULL)
3175 return (cvd);
3176 }
3177 return (NULL);
3178 }
3179
3180 /*
3181 * Verify that dynamic LUN growth works as expected.
3182 */
3183 /* ARGSUSED */
3184 void
3185 ztest_vdev_LUN_growth(ztest_ds_t *zd, uint64_t id)
3186 {
3187 spa_t *spa = ztest_spa;
3188 vdev_t *vd, *tvd;
3189 metaslab_class_t *mc;
3190 metaslab_group_t *mg;
3191 size_t psize, newsize;
3192 uint64_t top;
3193 uint64_t old_class_space, new_class_space, old_ms_count, new_ms_count;
3194
3195 mutex_enter(&ztest_checkpoint_lock);
3196 mutex_enter(&ztest_vdev_lock);
3197 spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3198
3199 /*
3200 * If there is a vdev removal in progress, it could complete while
3201 * we are running, in which case we would not be able to verify
3202 * that the metaslab_class space increased (because it decreases
3203 * when the device removal completes).
3204 */
3205 if (spa->spa_vdev_removal != NULL) {
3206 spa_config_exit(spa, SCL_STATE, spa);
3207 mutex_exit(&ztest_vdev_lock);
3208 mutex_exit(&ztest_checkpoint_lock);
3209 return;
3210 }
3211
3212 top = ztest_random_vdev_top(spa, B_TRUE);
3213
3214 tvd = spa->spa_root_vdev->vdev_child[top];
3215 mg = tvd->vdev_mg;
3216 mc = mg->mg_class;
3217 old_ms_count = tvd->vdev_ms_count;
3218 old_class_space = metaslab_class_get_space(mc);
3219
3220 /*
3221 * Determine the size of the first leaf vdev associated with
3222 * our top-level device.
3223 */
3224 vd = vdev_walk_tree(tvd, NULL, NULL);
3225 ASSERT3P(vd, !=, NULL);
3226 ASSERT(vd->vdev_ops->vdev_op_leaf);
3227
3228 psize = vd->vdev_psize;
3229
3230 /*
3231 * We only try to expand the vdev if it's healthy, less than 4x its
3232 * original size, and it has a valid psize.
3233 */
3234 if (tvd->vdev_state != VDEV_STATE_HEALTHY ||
3235 psize == 0 || psize >= 4 * ztest_opts.zo_vdev_size) {
3236 spa_config_exit(spa, SCL_STATE, spa);
3237 mutex_exit(&ztest_vdev_lock);
3238 mutex_exit(&ztest_checkpoint_lock);
3239 return;
3240 }
3241 ASSERT(psize > 0);
3242 newsize = psize + psize / 8;
3243 ASSERT3U(newsize, >, psize);
3244
3245 if (ztest_opts.zo_verbose >= 6) {
3246 (void) printf("Expanding LUN %s from %lu to %lu\n",
3247 vd->vdev_path, (ulong_t)psize, (ulong_t)newsize);
3248 }
3249
3250 /*
3251 * Growing the vdev is a two step process:
3252 * 1). expand the physical size (i.e. relabel)
3253 * 2). online the vdev to create the new metaslabs
3254 */
3255 if (vdev_walk_tree(tvd, grow_vdev, &newsize) != NULL ||
3256 vdev_walk_tree(tvd, online_vdev, NULL) != NULL ||
3257 tvd->vdev_state != VDEV_STATE_HEALTHY) {
3258 if (ztest_opts.zo_verbose >= 5) {
3259 (void) printf("Could not expand LUN because "
3260 "the vdev configuration changed.\n");
3261 }
3262 spa_config_exit(spa, SCL_STATE, spa);
3263 mutex_exit(&ztest_vdev_lock);
3264 mutex_exit(&ztest_checkpoint_lock);
3265 return;
3266 }
3267
3268 spa_config_exit(spa, SCL_STATE, spa);
3269
3270 /*
3271 * Expanding the LUN will update the config asynchronously,
3272 * thus we must wait for the async thread to complete any
3273 * pending tasks before proceeding.
3274 */
3275 for (;;) {
3276 boolean_t done;
3277 mutex_enter(&spa->spa_async_lock);
3278 done = (spa->spa_async_thread == NULL && !spa->spa_async_tasks);
3279 mutex_exit(&spa->spa_async_lock);
3280 if (done)
3281 break;
3282 txg_wait_synced(spa_get_dsl(spa), 0);
3283 (void) poll(NULL, 0, 100);
3284 }
3285
3286 spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3287
3288 tvd = spa->spa_root_vdev->vdev_child[top];
3289 new_ms_count = tvd->vdev_ms_count;
3290 new_class_space = metaslab_class_get_space(mc);
3291
3292 if (tvd->vdev_mg != mg || mg->mg_class != mc) {
3293 if (ztest_opts.zo_verbose >= 5) {
3294 (void) printf("Could not verify LUN expansion due to "
3295 "intervening vdev offline or remove.\n");
3296 }
3297 spa_config_exit(spa, SCL_STATE, spa);
3298 mutex_exit(&ztest_vdev_lock);
3299 mutex_exit(&ztest_checkpoint_lock);
3300 return;
3301 }
3302
3303 /*
3304 * Make sure we were able to grow the vdev.
3305 */
3306 if (new_ms_count <= old_ms_count) {
3307 fatal(0, "LUN expansion failed: ms_count %llu < %llu\n",
3308 old_ms_count, new_ms_count);
3309 }
3310
3311 /*
3312 * Make sure we were able to grow the pool.
3313 */
3314 if (new_class_space <= old_class_space) {
3315 fatal(0, "LUN expansion failed: class_space %llu < %llu\n",
3316 old_class_space, new_class_space);
3317 }
3318
3319 if (ztest_opts.zo_verbose >= 5) {
3320 char oldnumbuf[NN_NUMBUF_SZ], newnumbuf[NN_NUMBUF_SZ];
3321
3322 nicenum(old_class_space, oldnumbuf, sizeof (oldnumbuf));
3323 nicenum(new_class_space, newnumbuf, sizeof (newnumbuf));
3324 (void) printf("%s grew from %s to %s\n",
3325 spa->spa_name, oldnumbuf, newnumbuf);
3326 }
3327
3328 spa_config_exit(spa, SCL_STATE, spa);
3329 mutex_exit(&ztest_vdev_lock);
3330 mutex_exit(&ztest_checkpoint_lock);
3331 }
3332
3333 /*
3334 * Verify that dmu_objset_{create,destroy,open,close} work as expected.
3335 */
3336 /* ARGSUSED */
3337 static void
3338 ztest_objset_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
3339 {
3340 /*
3341 * Create the objects common to all ztest datasets.
3342 */
3343 VERIFY(zap_create_claim(os, ZTEST_DIROBJ,
3344 DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx) == 0);
3345 }
3346
3347 static int
3348 ztest_dataset_create(char *dsname)
3349 {
3350 uint64_t zilset = ztest_random(100);
3351 int err = dmu_objset_create(dsname, DMU_OST_OTHER, 0,
3352 ztest_objset_create_cb, NULL);
3353
3354 if (err || zilset < 80)
3355 return (err);
3356
3357 if (ztest_opts.zo_verbose >= 6)
3358 (void) printf("Setting dataset %s to sync always\n", dsname);
3359 return (ztest_dsl_prop_set_uint64(dsname, ZFS_PROP_SYNC,
3360 ZFS_SYNC_ALWAYS, B_FALSE));
3361 }
3362
3363 /* ARGSUSED */
3364 static int
3365 ztest_objset_destroy_cb(const char *name, void *arg)
3366 {
3367 objset_t *os;
3368 dmu_object_info_t doi;
3369 int error;
3370
3371 /*
3372 * Verify that the dataset contains a directory object.
3373 */
3374 VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_TRUE, FTAG, &os));
3375 error = dmu_object_info(os, ZTEST_DIROBJ, &doi);
3376 if (error != ENOENT) {
3377 /* We could have crashed in the middle of destroying it */
3378 ASSERT0(error);
3379 ASSERT3U(doi.doi_type, ==, DMU_OT_ZAP_OTHER);
3380 ASSERT3S(doi.doi_physical_blocks_512, >=, 0);
3381 }
3382 dmu_objset_disown(os, FTAG);
3383
3384 /*
3385 * Destroy the dataset.
3386 */
3387 if (strchr(name, '@') != NULL) {
3388 VERIFY0(dsl_destroy_snapshot(name, B_TRUE));
3389 } else {
3390 error = dsl_destroy_head(name);
3391 /* There could be a hold on this dataset */
3392 if (error != EBUSY)
3393 ASSERT0(error);
3394 }
3395 return (0);
3396 }
3397
3398 static boolean_t
3399 ztest_snapshot_create(char *osname, uint64_t id)
3400 {
3401 char snapname[ZFS_MAX_DATASET_NAME_LEN];
3402 int error;
3403
3404 (void) snprintf(snapname, sizeof (snapname), "%llu", (u_longlong_t)id);
3405
3406 error = dmu_objset_snapshot_one(osname, snapname);
3407 if (error == ENOSPC) {
3408 ztest_record_enospc(FTAG);
3409 return (B_FALSE);
3410 }
3411 if (error != 0 && error != EEXIST) {
3412 fatal(0, "ztest_snapshot_create(%s@%s) = %d", osname,
3413 snapname, error);
3414 }
3415 return (B_TRUE);
3416 }
3417
3418 static boolean_t
3419 ztest_snapshot_destroy(char *osname, uint64_t id)
3420 {
3421 char snapname[ZFS_MAX_DATASET_NAME_LEN];
3422 int error;
3423
3424 (void) snprintf(snapname, sizeof (snapname), "%s@%llu", osname,
3425 (u_longlong_t)id);
3426
3427 error = dsl_destroy_snapshot(snapname, B_FALSE);
3428 if (error != 0 && error != ENOENT)
3429 fatal(0, "ztest_snapshot_destroy(%s) = %d", snapname, error);
3430 return (B_TRUE);
3431 }
3432
3433 /* ARGSUSED */
3434 void
3435 ztest_dmu_objset_create_destroy(ztest_ds_t *zd, uint64_t id)
3436 {
3437 ztest_ds_t zdtmp;
3438 int iters;
3439 int error;
3440 objset_t *os, *os2;
3441 char name[ZFS_MAX_DATASET_NAME_LEN];
3442 zilog_t *zilog;
3443
3444 rw_enter(&ztest_name_lock, RW_READER);
3445
3446 (void) snprintf(name, sizeof (name), "%s/temp_%llu",
3447 ztest_opts.zo_pool, (u_longlong_t)id);
3448
3449 /*
3450 * If this dataset exists from a previous run, process its replay log
3451 * half of the time. If we don't replay it, then dmu_objset_destroy()
3452 * (invoked from ztest_objset_destroy_cb()) should just throw it away.
3453 */
3454 if (ztest_random(2) == 0 &&
3455 dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os) == 0) {
3456 ztest_zd_init(&zdtmp, NULL, os);
3457 zil_replay(os, &zdtmp, ztest_replay_vector);
3458 ztest_zd_fini(&zdtmp);
3459 dmu_objset_disown(os, FTAG);
3460 }
3461
3462 /*
3463 * There may be an old instance of the dataset we're about to
3464 * create lying around from a previous run. If so, destroy it
3465 * and all of its snapshots.
3466 */
3467 (void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
3468 DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
3469
3470 /*
3471 * Verify that the destroyed dataset is no longer in the namespace.
3472 */
3473 VERIFY3U(ENOENT, ==, dmu_objset_own(name, DMU_OST_OTHER, B_TRUE,
3474 FTAG, &os));
3475
3476 /*
3477 * Verify that we can create a new dataset.
3478 */
3479 error = ztest_dataset_create(name);
3480 if (error) {
3481 if (error == ENOSPC) {
3482 ztest_record_enospc(FTAG);
3483 rw_exit(&ztest_name_lock);
3484 return;
3485 }
3486 fatal(0, "dmu_objset_create(%s) = %d", name, error);
3487 }
3488
3489 VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os));
3490
3491 ztest_zd_init(&zdtmp, NULL, os);
3492
3493 /*
3494 * Open the intent log for it.
3495 */
3496 zilog = zil_open(os, ztest_get_data);
3497
3498 /*
3499 * Put some objects in there, do a little I/O to them,
3500 * and randomly take a couple of snapshots along the way.
3501 */
3502 iters = ztest_random(5);
3503 for (int i = 0; i < iters; i++) {
3504 ztest_dmu_object_alloc_free(&zdtmp, id);
3505 if (ztest_random(iters) == 0)
3506 (void) ztest_snapshot_create(name, i);
3507 }
3508
3509 /*
3510 * Verify that we cannot create an existing dataset.
3511 */
3512 VERIFY3U(EEXIST, ==,
3513 dmu_objset_create(name, DMU_OST_OTHER, 0, NULL, NULL));
3514
3515 /*
3516 * Verify that we can hold an objset that is also owned.
3517 */
3518 VERIFY3U(0, ==, dmu_objset_hold(name, FTAG, &os2));
3519 dmu_objset_rele(os2, FTAG);
3520
3521 /*
3522 * Verify that we cannot own an objset that is already owned.
3523 */
3524 VERIFY3U(EBUSY, ==,
3525 dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os2));
3526
3527 zil_close(zilog);
3528 dmu_objset_disown(os, FTAG);
3529 ztest_zd_fini(&zdtmp);
3530
3531 rw_exit(&ztest_name_lock);
3532 }
3533
3534 /*
3535 * Verify that dmu_snapshot_{create,destroy,open,close} work as expected.
3536 */
3537 void
3538 ztest_dmu_snapshot_create_destroy(ztest_ds_t *zd, uint64_t id)
3539 {
3540 rw_enter(&ztest_name_lock, RW_READER);
3541 (void) ztest_snapshot_destroy(zd->zd_name, id);
3542 (void) ztest_snapshot_create(zd->zd_name, id);
3543 rw_exit(&ztest_name_lock);
3544 }
3545
3546 /*
3547 * Cleanup non-standard snapshots and clones.
3548 */
3549 void
3550 ztest_dsl_dataset_cleanup(char *osname, uint64_t id)
3551 {
3552 char snap1name[ZFS_MAX_DATASET_NAME_LEN];
3553 char clone1name[ZFS_MAX_DATASET_NAME_LEN];
3554 char snap2name[ZFS_MAX_DATASET_NAME_LEN];
3555 char clone2name[ZFS_MAX_DATASET_NAME_LEN];
3556 char snap3name[ZFS_MAX_DATASET_NAME_LEN];
3557 int error;
3558
3559 (void) snprintf(snap1name, sizeof (snap1name),
3560 "%s@s1_%llu", osname, id);
3561 (void) snprintf(clone1name, sizeof (clone1name),
3562 "%s/c1_%llu", osname, id);
3563 (void) snprintf(snap2name, sizeof (snap2name),
3564 "%s@s2_%llu", clone1name, id);
3565 (void) snprintf(clone2name, sizeof (clone2name),
3566 "%s/c2_%llu", osname, id);
3567 (void) snprintf(snap3name, sizeof (snap3name),
3568 "%s@s3_%llu", clone1name, id);
3569
3570 error = dsl_destroy_head(clone2name);
3571 if (error && error != ENOENT)
3572 fatal(0, "dsl_destroy_head(%s) = %d", clone2name, error);
3573 error = dsl_destroy_snapshot(snap3name, B_FALSE);
3574 if (error && error != ENOENT)
3575 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap3name, error);
3576 error = dsl_destroy_snapshot(snap2name, B_FALSE);
3577 if (error && error != ENOENT)
3578 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap2name, error);
3579 error = dsl_destroy_head(clone1name);
3580 if (error && error != ENOENT)
3581 fatal(0, "dsl_destroy_head(%s) = %d", clone1name, error);
3582 error = dsl_destroy_snapshot(snap1name, B_FALSE);
3583 if (error && error != ENOENT)
3584 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap1name, error);
3585 }
3586
3587 /*
3588 * Verify dsl_dataset_promote handles EBUSY
3589 */
3590 void
3591 ztest_dsl_dataset_promote_busy(ztest_ds_t *zd, uint64_t id)
3592 {
3593 objset_t *os;
3594 char snap1name[ZFS_MAX_DATASET_NAME_LEN];
3595 char clone1name[ZFS_MAX_DATASET_NAME_LEN];
3596 char snap2name[ZFS_MAX_DATASET_NAME_LEN];
3597 char clone2name[ZFS_MAX_DATASET_NAME_LEN];
3598 char snap3name[ZFS_MAX_DATASET_NAME_LEN];
3599 char *osname = zd->zd_name;
3600 int error;
3601
3602 rw_enter(&ztest_name_lock, RW_READER);
3603
3604 ztest_dsl_dataset_cleanup(osname, id);
3605
3606 (void) snprintf(snap1name, sizeof (snap1name),
3607 "%s@s1_%llu", osname, id);
3608 (void) snprintf(clone1name, sizeof (clone1name),
3609 "%s/c1_%llu", osname, id);
3610 (void) snprintf(snap2name, sizeof (snap2name),
3611 "%s@s2_%llu", clone1name, id);
3612 (void) snprintf(clone2name, sizeof (clone2name),
3613 "%s/c2_%llu", osname, id);
3614 (void) snprintf(snap3name, sizeof (snap3name),
3615 "%s@s3_%llu", clone1name, id);
3616
3617 error = dmu_objset_snapshot_one(osname, strchr(snap1name, '@') + 1);
3618 if (error && error != EEXIST) {
3619 if (error == ENOSPC) {
3620 ztest_record_enospc(FTAG);
3621 goto out;
3622 }
3623 fatal(0, "dmu_take_snapshot(%s) = %d", snap1name, error);
3624 }
3625
3626 error = dmu_objset_clone(clone1name, snap1name);
3627 if (error) {
3628 if (error == ENOSPC) {
3629 ztest_record_enospc(FTAG);
3630 goto out;
3631 }
3632 fatal(0, "dmu_objset_create(%s) = %d", clone1name, error);
3633 }
3634
3635 error = dmu_objset_snapshot_one(clone1name, strchr(snap2name, '@') + 1);
3636 if (error && error != EEXIST) {
3637 if (error == ENOSPC) {
3638 ztest_record_enospc(FTAG);
3639 goto out;
3640 }
3641 fatal(0, "dmu_open_snapshot(%s) = %d", snap2name, error);
3642 }
3643
3644 error = dmu_objset_snapshot_one(clone1name, strchr(snap3name, '@') + 1);
3645 if (error && error != EEXIST) {
3646 if (error == ENOSPC) {
3647 ztest_record_enospc(FTAG);
3648 goto out;
3649 }
3650 fatal(0, "dmu_open_snapshot(%s) = %d", snap3name, error);
3651 }
3652
3653 error = dmu_objset_clone(clone2name, snap3name);
3654 if (error) {
3655 if (error == ENOSPC) {
3656 ztest_record_enospc(FTAG);
3657 goto out;
3658 }
3659 fatal(0, "dmu_objset_create(%s) = %d", clone2name, error);
3660 }
3661
3662 error = dmu_objset_own(snap2name, DMU_OST_ANY, B_TRUE, FTAG, &os);
3663 if (error)
3664 fatal(0, "dmu_objset_own(%s) = %d", snap2name, error);
3665 error = dsl_dataset_promote(clone2name, NULL);
3666 if (error == ENOSPC) {
3667 dmu_objset_disown(os, FTAG);
3668 ztest_record_enospc(FTAG);
3669 goto out;
3670 }
3671 if (error != EBUSY)
3672 fatal(0, "dsl_dataset_promote(%s), %d, not EBUSY", clone2name,
3673 error);
3674 dmu_objset_disown(os, FTAG);
3675
3676 out:
3677 ztest_dsl_dataset_cleanup(osname, id);
3678
3679 rw_exit(&ztest_name_lock);
3680 }
3681
3682 /*
3683 * Verify that dmu_object_{alloc,free} work as expected.
3684 */
3685 void
3686 ztest_dmu_object_alloc_free(ztest_ds_t *zd, uint64_t id)
3687 {
3688 ztest_od_t od[4];
3689 int batchsize = sizeof (od) / sizeof (od[0]);
3690
3691 for (int b = 0; b < batchsize; b++)
3692 ztest_od_init(&od[b], id, FTAG, b, DMU_OT_UINT64_OTHER, 0, 0);
3693
3694 /*
3695 * Destroy the previous batch of objects, create a new batch,
3696 * and do some I/O on the new objects.
3697 */
3698 if (ztest_object_init(zd, od, sizeof (od), B_TRUE) != 0)
3699 return;
3700
3701 while (ztest_random(4 * batchsize) != 0)
3702 ztest_io(zd, od[ztest_random(batchsize)].od_object,
3703 ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
3704 }
3705
3706 /*
3707 * Verify that dmu_{read,write} work as expected.
3708 */
3709 void
3710 ztest_dmu_read_write(ztest_ds_t *zd, uint64_t id)
3711 {
3712 objset_t *os = zd->zd_os;
3713 ztest_od_t od[2];
3714 dmu_tx_t *tx;
3715 int i, freeit, error;
3716 uint64_t n, s, txg;
3717 bufwad_t *packbuf, *bigbuf, *pack, *bigH, *bigT;
3718 uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
3719 uint64_t chunksize = (1000 + ztest_random(1000)) * sizeof (uint64_t);
3720 uint64_t regions = 997;
3721 uint64_t stride = 123456789ULL;
3722 uint64_t width = 40;
3723 int free_percent = 5;
3724
3725 /*
3726 * This test uses two objects, packobj and bigobj, that are always
3727 * updated together (i.e. in the same tx) so that their contents are
3728 * in sync and can be compared. Their contents relate to each other
3729 * in a simple way: packobj is a dense array of 'bufwad' structures,
3730 * while bigobj is a sparse array of the same bufwads. Specifically,
3731 * for any index n, there are three bufwads that should be identical:
3732 *
3733 * packobj, at offset n * sizeof (bufwad_t)
3734 * bigobj, at the head of the nth chunk
3735 * bigobj, at the tail of the nth chunk
3736 *
3737 * The chunk size is arbitrary. It doesn't have to be a power of two,
3738 * and it doesn't have any relation to the object blocksize.
3739 * The only requirement is that it can hold at least two bufwads.
3740 *
3741 * Normally, we write the bufwad to each of these locations.
3742 * However, free_percent of the time we instead write zeroes to
3743 * packobj and perform a dmu_free_range() on bigobj. By comparing
3744 * bigobj to packobj, we can verify that the DMU is correctly
3745 * tracking which parts of an object are allocated and free,
3746 * and that the contents of the allocated blocks are correct.
3747 */
3748
3749 /*
3750 * Read the directory info. If it's the first time, set things up.
3751 */
3752 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, chunksize);
3753 ztest_od_init(&od[1], id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize);
3754
3755 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
3756 return;
3757
3758 bigobj = od[0].od_object;
3759 packobj = od[1].od_object;
3760 chunksize = od[0].od_gen;
3761 ASSERT(chunksize == od[1].od_gen);
3762
3763 /*
3764 * Prefetch a random chunk of the big object.
3765 * Our aim here is to get some async reads in flight
3766 * for blocks that we may free below; the DMU should
3767 * handle this race correctly.
3768 */
3769 n = ztest_random(regions) * stride + ztest_random(width);
3770 s = 1 + ztest_random(2 * width - 1);
3771 dmu_prefetch(os, bigobj, 0, n * chunksize, s * chunksize,
3772 ZIO_PRIORITY_SYNC_READ);
3773
3774 /*
3775 * Pick a random index and compute the offsets into packobj and bigobj.
3776 */
3777 n = ztest_random(regions) * stride + ztest_random(width);
3778 s = 1 + ztest_random(width - 1);
3779
3780 packoff = n * sizeof (bufwad_t);
3781 packsize = s * sizeof (bufwad_t);
3782
3783 bigoff = n * chunksize;
3784 bigsize = s * chunksize;
3785
3786 packbuf = umem_alloc(packsize, UMEM_NOFAIL);
3787 bigbuf = umem_alloc(bigsize, UMEM_NOFAIL);
3788
3789 /*
3790 * free_percent of the time, free a range of bigobj rather than
3791 * overwriting it.
3792 */
3793 freeit = (ztest_random(100) < free_percent);
3794
3795 /*
3796 * Read the current contents of our objects.
3797 */
3798 error = dmu_read(os, packobj, packoff, packsize, packbuf,
3799 DMU_READ_PREFETCH);
3800 ASSERT0(error);
3801 error = dmu_read(os, bigobj, bigoff, bigsize, bigbuf,
3802 DMU_READ_PREFETCH);
3803 ASSERT0(error);
3804
3805 /*
3806 * Get a tx for the mods to both packobj and bigobj.
3807 */
3808 tx = dmu_tx_create(os);
3809
3810 dmu_tx_hold_write(tx, packobj, packoff, packsize);
3811
3812 if (freeit)
3813 dmu_tx_hold_free(tx, bigobj, bigoff, bigsize);
3814 else
3815 dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
3816
3817 /* This accounts for setting the checksum/compression. */
3818 dmu_tx_hold_bonus(tx, bigobj);
3819
3820 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3821 if (txg == 0) {
3822 umem_free(packbuf, packsize);
3823 umem_free(bigbuf, bigsize);
3824 return;
3825 }
3826
3827 enum zio_checksum cksum;
3828 do {
3829 cksum = (enum zio_checksum)
3830 ztest_random_dsl_prop(ZFS_PROP_CHECKSUM);
3831 } while (cksum >= ZIO_CHECKSUM_LEGACY_FUNCTIONS);
3832 dmu_object_set_checksum(os, bigobj, cksum, tx);
3833
3834 enum zio_compress comp;
3835 do {
3836 comp = (enum zio_compress)
3837 ztest_random_dsl_prop(ZFS_PROP_COMPRESSION);
3838 } while (comp >= ZIO_COMPRESS_LEGACY_FUNCTIONS);
3839 dmu_object_set_compress(os, bigobj, comp, tx);
3840
3841 /*
3842 * For each index from n to n + s, verify that the existing bufwad
3843 * in packobj matches the bufwads at the head and tail of the
3844 * corresponding chunk in bigobj. Then update all three bufwads
3845 * with the new values we want to write out.
3846 */
3847 for (i = 0; i < s; i++) {
3848 /* LINTED */
3849 pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
3850 /* LINTED */
3851 bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
3852 /* LINTED */
3853 bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
3854
3855 ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
3856 ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
3857
3858 if (pack->bw_txg > txg)
3859 fatal(0, "future leak: got %llx, open txg is %llx",
3860 pack->bw_txg, txg);
3861
3862 if (pack->bw_data != 0 && pack->bw_index != n + i)
3863 fatal(0, "wrong index: got %llx, wanted %llx+%llx",
3864 pack->bw_index, n, i);
3865
3866 if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
3867 fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
3868
3869 if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
3870 fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
3871
3872 if (freeit) {
3873 bzero(pack, sizeof (bufwad_t));
3874 } else {
3875 pack->bw_index = n + i;
3876 pack->bw_txg = txg;
3877 pack->bw_data = 1 + ztest_random(-2ULL);
3878 }
3879 *bigH = *pack;
3880 *bigT = *pack;
3881 }
3882
3883 /*
3884 * We've verified all the old bufwads, and made new ones.
3885 * Now write them out.
3886 */
3887 dmu_write(os, packobj, packoff, packsize, packbuf, tx);
3888
3889 if (freeit) {
3890 if (ztest_opts.zo_verbose >= 7) {
3891 (void) printf("freeing offset %llx size %llx"
3892 " txg %llx\n",
3893 (u_longlong_t)bigoff,
3894 (u_longlong_t)bigsize,
3895 (u_longlong_t)txg);
3896 }
3897 VERIFY(0 == dmu_free_range(os, bigobj, bigoff, bigsize, tx));
3898 } else {
3899 if (ztest_opts.zo_verbose >= 7) {
3900 (void) printf("writing offset %llx size %llx"
3901 " txg %llx\n",
3902 (u_longlong_t)bigoff,
3903 (u_longlong_t)bigsize,
3904 (u_longlong_t)txg);
3905 }
3906 dmu_write(os, bigobj, bigoff, bigsize, bigbuf, tx);
3907 }
3908
3909 dmu_tx_commit(tx);
3910
3911 /*
3912 * Sanity check the stuff we just wrote.
3913 */
3914 {
3915 void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
3916 void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
3917
3918 VERIFY(0 == dmu_read(os, packobj, packoff,
3919 packsize, packcheck, DMU_READ_PREFETCH));
3920 VERIFY(0 == dmu_read(os, bigobj, bigoff,
3921 bigsize, bigcheck, DMU_READ_PREFETCH));
3922
3923 ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
3924 ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
3925
3926 umem_free(packcheck, packsize);
3927 umem_free(bigcheck, bigsize);
3928 }
3929
3930 umem_free(packbuf, packsize);
3931 umem_free(bigbuf, bigsize);
3932 }
3933
3934 void
3935 compare_and_update_pbbufs(uint64_t s, bufwad_t *packbuf, bufwad_t *bigbuf,
3936 uint64_t bigsize, uint64_t n, uint64_t chunksize, uint64_t txg)
3937 {
3938 uint64_t i;
3939 bufwad_t *pack;
3940 bufwad_t *bigH;
3941 bufwad_t *bigT;
3942
3943 /*
3944 * For each index from n to n + s, verify that the existing bufwad
3945 * in packobj matches the bufwads at the head and tail of the
3946 * corresponding chunk in bigobj. Then update all three bufwads
3947 * with the new values we want to write out.
3948 */
3949 for (i = 0; i < s; i++) {
3950 /* LINTED */
3951 pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
3952 /* LINTED */
3953 bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
3954 /* LINTED */
3955 bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
3956
3957 ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
3958 ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
3959
3960 if (pack->bw_txg > txg)
3961 fatal(0, "future leak: got %llx, open txg is %llx",
3962 pack->bw_txg, txg);
3963
3964 if (pack->bw_data != 0 && pack->bw_index != n + i)
3965 fatal(0, "wrong index: got %llx, wanted %llx+%llx",
3966 pack->bw_index, n, i);
3967
3968 if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
3969 fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
3970
3971 if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
3972 fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
3973
3974 pack->bw_index = n + i;
3975 pack->bw_txg = txg;
3976 pack->bw_data = 1 + ztest_random(-2ULL);
3977
3978 *bigH = *pack;
3979 *bigT = *pack;
3980 }
3981 }
3982
3983 void
3984 ztest_dmu_read_write_zcopy(ztest_ds_t *zd, uint64_t id)
3985 {
3986 objset_t *os = zd->zd_os;
3987 ztest_od_t od[2];
3988 dmu_tx_t *tx;
3989 uint64_t i;
3990 int error;
3991 uint64_t n, s, txg;
3992 bufwad_t *packbuf, *bigbuf;
3993 uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
3994 uint64_t blocksize = ztest_random_blocksize();
3995 uint64_t chunksize = blocksize;
3996 uint64_t regions = 997;
3997 uint64_t stride = 123456789ULL;
3998 uint64_t width = 9;
3999 dmu_buf_t *bonus_db;
4000 arc_buf_t **bigbuf_arcbufs;
4001 dmu_object_info_t doi;
4002
4003 /*
4004 * This test uses two objects, packobj and bigobj, that are always
4005 * updated together (i.e. in the same tx) so that their contents are
4006 * in sync and can be compared. Their contents relate to each other
4007 * in a simple way: packobj is a dense array of 'bufwad' structures,
4008 * while bigobj is a sparse array of the same bufwads. Specifically,
4009 * for any index n, there are three bufwads that should be identical:
4010 *
4011 * packobj, at offset n * sizeof (bufwad_t)
4012 * bigobj, at the head of the nth chunk
4013 * bigobj, at the tail of the nth chunk
4014 *
4015 * The chunk size is set equal to bigobj block size so that
4016 * dmu_assign_arcbuf() can be tested for object updates.
4017 */
4018
4019 /*
4020 * Read the directory info. If it's the first time, set things up.
4021 */
4022 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
4023 ztest_od_init(&od[1], id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize);
4024
4025 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4026 return;
4027
4028 bigobj = od[0].od_object;
4029 packobj = od[1].od_object;
4030 blocksize = od[0].od_blocksize;
4031 chunksize = blocksize;
4032 ASSERT(chunksize == od[1].od_gen);
4033
4034 VERIFY(dmu_object_info(os, bigobj, &doi) == 0);
4035 VERIFY(ISP2(doi.doi_data_block_size));
4036 VERIFY(chunksize == doi.doi_data_block_size);
4037 VERIFY(chunksize >= 2 * sizeof (bufwad_t));
4038
4039 /*
4040 * Pick a random index and compute the offsets into packobj and bigobj.
4041 */
4042 n = ztest_random(regions) * stride + ztest_random(width);
4043 s = 1 + ztest_random(width - 1);
4044
4045 packoff = n * sizeof (bufwad_t);
4046 packsize = s * sizeof (bufwad_t);
4047
4048 bigoff = n * chunksize;
4049 bigsize = s * chunksize;
4050
4051 packbuf = umem_zalloc(packsize, UMEM_NOFAIL);
4052 bigbuf = umem_zalloc(bigsize, UMEM_NOFAIL);
4053
4054 VERIFY3U(0, ==, dmu_bonus_hold(os, bigobj, FTAG, &bonus_db));
4055
4056 bigbuf_arcbufs = umem_zalloc(2 * s * sizeof (arc_buf_t *), UMEM_NOFAIL);
4057
4058 /*
4059 * Iteration 0 test zcopy for DB_UNCACHED dbufs.
4060 * Iteration 1 test zcopy to already referenced dbufs.
4061 * Iteration 2 test zcopy to dirty dbuf in the same txg.
4062 * Iteration 3 test zcopy to dbuf dirty in previous txg.
4063 * Iteration 4 test zcopy when dbuf is no longer dirty.
4064 * Iteration 5 test zcopy when it can't be done.
4065 * Iteration 6 one more zcopy write.
4066 */
4067 for (i = 0; i < 7; i++) {
4068 uint64_t j;
4069 uint64_t off;
4070
4071 /*
4072 * In iteration 5 (i == 5) use arcbufs
4073 * that don't match bigobj blksz to test
4074 * dmu_assign_arcbuf() when it can't directly
4075 * assign an arcbuf to a dbuf.
4076 */
4077 for (j = 0; j < s; j++) {
4078 if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) {
4079 bigbuf_arcbufs[j] =
4080 dmu_request_arcbuf(bonus_db, chunksize);
4081 } else {
4082 bigbuf_arcbufs[2 * j] =
4083 dmu_request_arcbuf(bonus_db, chunksize / 2);
4084 bigbuf_arcbufs[2 * j + 1] =
4085 dmu_request_arcbuf(bonus_db, chunksize / 2);
4086 }
4087 }
4088
4089 /*
4090 * Get a tx for the mods to both packobj and bigobj.
4091 */
4092 tx = dmu_tx_create(os);
4093
4094 dmu_tx_hold_write(tx, packobj, packoff, packsize);
4095 dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
4096
4097 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4098 if (txg == 0) {
4099 umem_free(packbuf, packsize);
4100 umem_free(bigbuf, bigsize);
4101 for (j = 0; j < s; j++) {
4102 if (i != 5 ||
4103 chunksize < (SPA_MINBLOCKSIZE * 2)) {
4104 dmu_return_arcbuf(bigbuf_arcbufs[j]);
4105 } else {
4106 dmu_return_arcbuf(
4107 bigbuf_arcbufs[2 * j]);
4108 dmu_return_arcbuf(
4109 bigbuf_arcbufs[2 * j + 1]);
4110 }
4111 }
4112 umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
4113 dmu_buf_rele(bonus_db, FTAG);
4114 return;
4115 }
4116
4117 /*
4118 * 50% of the time don't read objects in the 1st iteration to
4119 * test dmu_assign_arcbuf() for the case when there're no
4120 * existing dbufs for the specified offsets.
4121 */
4122 if (i != 0 || ztest_random(2) != 0) {
4123 error = dmu_read(os, packobj, packoff,
4124 packsize, packbuf, DMU_READ_PREFETCH);
4125 ASSERT0(error);
4126 error = dmu_read(os, bigobj, bigoff, bigsize,
4127 bigbuf, DMU_READ_PREFETCH);
4128 ASSERT0(error);
4129 }
4130 compare_and_update_pbbufs(s, packbuf, bigbuf, bigsize,
4131 n, chunksize, txg);
4132
4133 /*
4134 * We've verified all the old bufwads, and made new ones.
4135 * Now write them out.
4136 */
4137 dmu_write(os, packobj, packoff, packsize, packbuf, tx);
4138 if (ztest_opts.zo_verbose >= 7) {
4139 (void) printf("writing offset %llx size %llx"
4140 " txg %llx\n",
4141 (u_longlong_t)bigoff,
4142 (u_longlong_t)bigsize,
4143 (u_longlong_t)txg);
4144 }
4145 for (off = bigoff, j = 0; j < s; j++, off += chunksize) {
4146 dmu_buf_t *dbt;
4147 if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) {
4148 bcopy((caddr_t)bigbuf + (off - bigoff),
4149 bigbuf_arcbufs[j]->b_data, chunksize);
4150 } else {
4151 bcopy((caddr_t)bigbuf + (off - bigoff),
4152 bigbuf_arcbufs[2 * j]->b_data,
4153 chunksize / 2);
4154 bcopy((caddr_t)bigbuf + (off - bigoff) +
4155 chunksize / 2,
4156 bigbuf_arcbufs[2 * j + 1]->b_data,
4157 chunksize / 2);
4158 }
4159
4160 if (i == 1) {
4161 VERIFY(dmu_buf_hold(os, bigobj, off,
4162 FTAG, &dbt, DMU_READ_NO_PREFETCH) == 0);
4163 }
4164 if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) {
4165 dmu_assign_arcbuf(bonus_db, off,
4166 bigbuf_arcbufs[j], tx);
4167 } else {
4168 dmu_assign_arcbuf(bonus_db, off,
4169 bigbuf_arcbufs[2 * j], tx);
4170 dmu_assign_arcbuf(bonus_db,
4171 off + chunksize / 2,
4172 bigbuf_arcbufs[2 * j + 1], tx);
4173 }
4174 if (i == 1) {
4175 dmu_buf_rele(dbt, FTAG);
4176 }
4177 }
4178 dmu_tx_commit(tx);
4179
4180 /*
4181 * Sanity check the stuff we just wrote.
4182 */
4183 {
4184 void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
4185 void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
4186
4187 VERIFY(0 == dmu_read(os, packobj, packoff,
4188 packsize, packcheck, DMU_READ_PREFETCH));
4189 VERIFY(0 == dmu_read(os, bigobj, bigoff,
4190 bigsize, bigcheck, DMU_READ_PREFETCH));
4191
4192 ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
4193 ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
4194
4195 umem_free(packcheck, packsize);
4196 umem_free(bigcheck, bigsize);
4197 }
4198 if (i == 2) {
4199 txg_wait_open(dmu_objset_pool(os), 0);
4200 } else if (i == 3) {
4201 txg_wait_synced(dmu_objset_pool(os), 0);
4202 }
4203 }
4204
4205 dmu_buf_rele(bonus_db, FTAG);
4206 umem_free(packbuf, packsize);
4207 umem_free(bigbuf, bigsize);
4208 umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
4209 }
4210
4211 /* ARGSUSED */
4212 void
4213 ztest_dmu_write_parallel(ztest_ds_t *zd, uint64_t id)
4214 {
4215 ztest_od_t od[1];
4216 uint64_t offset = (1ULL << (ztest_random(20) + 43)) +
4217 (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
4218
4219 /*
4220 * Have multiple threads write to large offsets in an object
4221 * to verify that parallel writes to an object -- even to the
4222 * same blocks within the object -- doesn't cause any trouble.
4223 */
4224 ztest_od_init(&od[0], ID_PARALLEL, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
4225
4226 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4227 return;
4228
4229 while (ztest_random(10) != 0)
4230 ztest_io(zd, od[0].od_object, offset);
4231 }
4232
4233 void
4234 ztest_dmu_prealloc(ztest_ds_t *zd, uint64_t id)
4235 {
4236 ztest_od_t od[1];
4237 uint64_t offset = (1ULL << (ztest_random(4) + SPA_MAXBLOCKSHIFT)) +
4238 (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
4239 uint64_t count = ztest_random(20) + 1;
4240 uint64_t blocksize = ztest_random_blocksize();
4241 void *data;
4242
4243 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
4244
4245 if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4246 return;
4247
4248 if (ztest_truncate(zd, od[0].od_object, offset, count * blocksize) != 0)
4249 return;
4250
4251 ztest_prealloc(zd, od[0].od_object, offset, count * blocksize);
4252
4253 data = umem_zalloc(blocksize, UMEM_NOFAIL);
4254
4255 while (ztest_random(count) != 0) {
4256 uint64_t randoff = offset + (ztest_random(count) * blocksize);
4257 if (ztest_write(zd, od[0].od_object, randoff, blocksize,
4258 data) != 0)
4259 break;
4260 while (ztest_random(4) != 0)
4261 ztest_io(zd, od[0].od_object, randoff);
4262 }
4263
4264 umem_free(data, blocksize);
4265 }
4266
4267 /*
4268 * Verify that zap_{create,destroy,add,remove,update} work as expected.
4269 */
4270 #define ZTEST_ZAP_MIN_INTS 1
4271 #define ZTEST_ZAP_MAX_INTS 4
4272 #define ZTEST_ZAP_MAX_PROPS 1000
4273
4274 void
4275 ztest_zap(ztest_ds_t *zd, uint64_t id)
4276 {
4277 objset_t *os = zd->zd_os;
4278 ztest_od_t od[1];
4279 uint64_t object;
4280 uint64_t txg, last_txg;
4281 uint64_t value[ZTEST_ZAP_MAX_INTS];
4282 uint64_t zl_ints, zl_intsize, prop;
4283 int i, ints;
4284 dmu_tx_t *tx;
4285 char propname[100], txgname[100];
4286 int error;
4287 char *hc[2] = { "s.acl.h", ".s.open.h.hyLZlg" };
4288
4289 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0);
4290
4291 if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4292 return;
4293
4294 object = od[0].od_object;
4295
4296 /*
4297 * Generate a known hash collision, and verify that
4298 * we can lookup and remove both entries.
4299 */
4300 tx = dmu_tx_create(os);
4301 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4302 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4303 if (txg == 0)
4304 return;
4305 for (i = 0; i < 2; i++) {
4306 value[i] = i;
4307 VERIFY3U(0, ==, zap_add(os, object, hc[i], sizeof (uint64_t),
4308 1, &value[i], tx));
4309 }
4310 for (i = 0; i < 2; i++) {
4311 VERIFY3U(EEXIST, ==, zap_add(os, object, hc[i],
4312 sizeof (uint64_t), 1, &value[i], tx));
4313 VERIFY3U(0, ==,
4314 zap_length(os, object, hc[i], &zl_intsize, &zl_ints));
4315 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4316 ASSERT3U(zl_ints, ==, 1);
4317 }
4318 for (i = 0; i < 2; i++) {
4319 VERIFY3U(0, ==, zap_remove(os, object, hc[i], tx));
4320 }
4321 dmu_tx_commit(tx);
4322
4323 /*
4324 * Generate a buch of random entries.
4325 */
4326 ints = MAX(ZTEST_ZAP_MIN_INTS, object % ZTEST_ZAP_MAX_INTS);
4327
4328 prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
4329 (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
4330 (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
4331 bzero(value, sizeof (value));
4332 last_txg = 0;
4333
4334 /*
4335 * If these zap entries already exist, validate their contents.
4336 */
4337 error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
4338 if (error == 0) {
4339 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4340 ASSERT3U(zl_ints, ==, 1);
4341
4342 VERIFY(zap_lookup(os, object, txgname, zl_intsize,
4343 zl_ints, &last_txg) == 0);
4344
4345 VERIFY(zap_length(os, object, propname, &zl_intsize,
4346 &zl_ints) == 0);
4347
4348 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4349 ASSERT3U(zl_ints, ==, ints);
4350
4351 VERIFY(zap_lookup(os, object, propname, zl_intsize,
4352 zl_ints, value) == 0);
4353
4354 for (i = 0; i < ints; i++) {
4355 ASSERT3U(value[i], ==, last_txg + object + i);
4356 }
4357 } else {
4358 ASSERT3U(error, ==, ENOENT);
4359 }
4360
4361 /*
4362 * Atomically update two entries in our zap object.
4363 * The first is named txg_%llu, and contains the txg
4364 * in which the property was last updated. The second
4365 * is named prop_%llu, and the nth element of its value
4366 * should be txg + object + n.
4367 */
4368 tx = dmu_tx_create(os);
4369 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4370 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4371 if (txg == 0)
4372 return;
4373
4374 if (last_txg > txg)
4375 fatal(0, "zap future leak: old %llu new %llu", last_txg, txg);
4376
4377 for (i = 0; i < ints; i++)
4378 value[i] = txg + object + i;
4379
4380 VERIFY3U(0, ==, zap_update(os, object, txgname, sizeof (uint64_t),
4381 1, &txg, tx));
4382 VERIFY3U(0, ==, zap_update(os, object, propname, sizeof (uint64_t),
4383 ints, value, tx));
4384
4385 dmu_tx_commit(tx);
4386
4387 /*
4388 * Remove a random pair of entries.
4389 */
4390 prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
4391 (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
4392 (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
4393
4394 error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
4395
4396 if (error == ENOENT)
4397 return;
4398
4399 ASSERT0(error);
4400
4401 tx = dmu_tx_create(os);
4402 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4403 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4404 if (txg == 0)
4405 return;
4406 VERIFY3U(0, ==, zap_remove(os, object, txgname, tx));
4407 VERIFY3U(0, ==, zap_remove(os, object, propname, tx));
4408 dmu_tx_commit(tx);
4409 }
4410
4411 /*
4412 * Testcase to test the upgrading of a microzap to fatzap.
4413 */
4414 void
4415 ztest_fzap(ztest_ds_t *zd, uint64_t id)
4416 {
4417 objset_t *os = zd->zd_os;
4418 ztest_od_t od[1];
4419 uint64_t object, txg;
4420
4421 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0);
4422
4423 if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4424 return;
4425
4426 object = od[0].od_object;
4427
4428 /*
4429 * Add entries to this ZAP and make sure it spills over
4430 * and gets upgraded to a fatzap. Also, since we are adding
4431 * 2050 entries we should see ptrtbl growth and leaf-block split.
4432 */
4433 for (int i = 0; i < 2050; i++) {
4434 char name[ZFS_MAX_DATASET_NAME_LEN];
4435 uint64_t value = i;
4436 dmu_tx_t *tx;
4437 int error;
4438
4439 (void) snprintf(name, sizeof (name), "fzap-%llu-%llu",
4440 id, value);
4441
4442 tx = dmu_tx_create(os);
4443 dmu_tx_hold_zap(tx, object, B_TRUE, name);
4444 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4445 if (txg == 0)
4446 return;
4447 error = zap_add(os, object, name, sizeof (uint64_t), 1,
4448 &value, tx);
4449 ASSERT(error == 0 || error == EEXIST);
4450 dmu_tx_commit(tx);
4451 }
4452 }
4453
4454 /* ARGSUSED */
4455 void
4456 ztest_zap_parallel(ztest_ds_t *zd, uint64_t id)
4457 {
4458 objset_t *os = zd->zd_os;
4459 ztest_od_t od[1];
4460 uint64_t txg, object, count, wsize, wc, zl_wsize, zl_wc;
4461 dmu_tx_t *tx;
4462 int i, namelen, error;
4463 int micro = ztest_random(2);
4464 char name[20], string_value[20];
4465 void *data;
4466
4467 ztest_od_init(&od[0], ID_PARALLEL, FTAG, micro, DMU_OT_ZAP_OTHER, 0, 0);
4468
4469 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4470 return;
4471
4472 object = od[0].od_object;
4473
4474 /*
4475 * Generate a random name of the form 'xxx.....' where each
4476 * x is a random printable character and the dots are dots.
4477 * There are 94 such characters, and the name length goes from
4478 * 6 to 20, so there are 94^3 * 15 = 12,458,760 possible names.
4479 */
4480 namelen = ztest_random(sizeof (name) - 5) + 5 + 1;
4481
4482 for (i = 0; i < 3; i++)
4483 name[i] = '!' + ztest_random('~' - '!' + 1);
4484 for (; i < namelen - 1; i++)
4485 name[i] = '.';
4486 name[i] = '\0';
4487
4488 if ((namelen & 1) || micro) {
4489 wsize = sizeof (txg);
4490 wc = 1;
4491 data = &txg;
4492 } else {
4493 wsize = 1;
4494 wc = namelen;
4495 data = string_value;
4496 }
4497
4498 count = -1ULL;
4499 VERIFY0(zap_count(os, object, &count));
4500 ASSERT(count != -1ULL);
4501
4502 /*
4503 * Select an operation: length, lookup, add, update, remove.
4504 */
4505 i = ztest_random(5);
4506
4507 if (i >= 2) {
4508 tx = dmu_tx_create(os);
4509 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4510 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4511 if (txg == 0)
4512 return;
4513 bcopy(name, string_value, namelen);
4514 } else {
4515 tx = NULL;
4516 txg = 0;
4517 bzero(string_value, namelen);
4518 }
4519
4520 switch (i) {
4521
4522 case 0:
4523 error = zap_length(os, object, name, &zl_wsize, &zl_wc);
4524 if (error == 0) {
4525 ASSERT3U(wsize, ==, zl_wsize);
4526 ASSERT3U(wc, ==, zl_wc);
4527 } else {
4528 ASSERT3U(error, ==, ENOENT);
4529 }
4530 break;
4531
4532 case 1:
4533 error = zap_lookup(os, object, name, wsize, wc, data);
4534 if (error == 0) {
4535 if (data == string_value &&
4536 bcmp(name, data, namelen) != 0)
4537 fatal(0, "name '%s' != val '%s' len %d",
4538 name, data, namelen);
4539 } else {
4540 ASSERT3U(error, ==, ENOENT);
4541 }
4542 break;
4543
4544 case 2:
4545 error = zap_add(os, object, name, wsize, wc, data, tx);
4546 ASSERT(error == 0 || error == EEXIST);
4547 break;
4548
4549 case 3:
4550 VERIFY(zap_update(os, object, name, wsize, wc, data, tx) == 0);
4551 break;
4552
4553 case 4:
4554 error = zap_remove(os, object, name, tx);
4555 ASSERT(error == 0 || error == ENOENT);
4556 break;
4557 }
4558
4559 if (tx != NULL)
4560 dmu_tx_commit(tx);
4561 }
4562
4563 /*
4564 * Commit callback data.
4565 */
4566 typedef struct ztest_cb_data {
4567 list_node_t zcd_node;
4568 uint64_t zcd_txg;
4569 int zcd_expected_err;
4570 boolean_t zcd_added;
4571 boolean_t zcd_called;
4572 spa_t *zcd_spa;
4573 } ztest_cb_data_t;
4574
4575 /* This is the actual commit callback function */
4576 static void
4577 ztest_commit_callback(void *arg, int error)
4578 {
4579 ztest_cb_data_t *data = arg;
4580 uint64_t synced_txg;
4581
4582 VERIFY(data != NULL);
4583 VERIFY3S(data->zcd_expected_err, ==, error);
4584 VERIFY(!data->zcd_called);
4585
4586 synced_txg = spa_last_synced_txg(data->zcd_spa);
4587 if (data->zcd_txg > synced_txg)
4588 fatal(0, "commit callback of txg %" PRIu64 " called prematurely"
4589 ", last synced txg = %" PRIu64 "\n", data->zcd_txg,
4590 synced_txg);
4591
4592 data->zcd_called = B_TRUE;
4593
4594 if (error == ECANCELED) {
4595 ASSERT0(data->zcd_txg);
4596 ASSERT(!data->zcd_added);
4597
4598 /*
4599 * The private callback data should be destroyed here, but
4600 * since we are going to check the zcd_called field after
4601 * dmu_tx_abort(), we will destroy it there.
4602 */
4603 return;
4604 }
4605
4606 /* Was this callback added to the global callback list? */
4607 if (!data->zcd_added)
4608 goto out;
4609
4610 ASSERT3U(data->zcd_txg, !=, 0);
4611
4612 /* Remove our callback from the list */
4613 mutex_enter(&zcl.zcl_callbacks_lock);
4614 list_remove(&zcl.zcl_callbacks, data);
4615 mutex_exit(&zcl.zcl_callbacks_lock);
4616
4617 out:
4618 umem_free(data, sizeof (ztest_cb_data_t));
4619 }
4620
4621 /* Allocate and initialize callback data structure */
4622 static ztest_cb_data_t *
4623 ztest_create_cb_data(objset_t *os, uint64_t txg)
4624 {
4625 ztest_cb_data_t *cb_data;
4626
4627 cb_data = umem_zalloc(sizeof (ztest_cb_data_t), UMEM_NOFAIL);
4628
4629 cb_data->zcd_txg = txg;
4630 cb_data->zcd_spa = dmu_objset_spa(os);
4631
4632 return (cb_data);
4633 }
4634
4635 /*
4636 * If a number of txgs equal to this threshold have been created after a commit
4637 * callback has been registered but not called, then we assume there is an
4638 * implementation bug.
4639 */
4640 #define ZTEST_COMMIT_CALLBACK_THRESH (TXG_CONCURRENT_STATES + 2)
4641
4642 /*
4643 * Commit callback test.
4644 */
4645 void
4646 ztest_dmu_commit_callbacks(ztest_ds_t *zd, uint64_t id)
4647 {
4648 objset_t *os = zd->zd_os;
4649 ztest_od_t od[1];
4650 dmu_tx_t *tx;
4651 ztest_cb_data_t *cb_data[3], *tmp_cb;
4652 uint64_t old_txg, txg;
4653 int i, error;
4654
4655 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
4656
4657 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4658 return;
4659
4660 tx = dmu_tx_create(os);
4661
4662 cb_data[0] = ztest_create_cb_data(os, 0);
4663 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[0]);
4664
4665 dmu_tx_hold_write(tx, od[0].od_object, 0, sizeof (uint64_t));
4666
4667 /* Every once in a while, abort the transaction on purpose */
4668 if (ztest_random(100) == 0)
4669 error = -1;
4670
4671 if (!error)
4672 error = dmu_tx_assign(tx, TXG_NOWAIT);
4673
4674 txg = error ? 0 : dmu_tx_get_txg(tx);
4675
4676 cb_data[0]->zcd_txg = txg;
4677 cb_data[1] = ztest_create_cb_data(os, txg);
4678 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[1]);
4679
4680 if (error) {
4681 /*
4682 * It's not a strict requirement to call the registered
4683 * callbacks from inside dmu_tx_abort(), but that's what
4684 * it's supposed to happen in the current implementation
4685 * so we will check for that.
4686 */
4687 for (i = 0; i < 2; i++) {
4688 cb_data[i]->zcd_expected_err = ECANCELED;
4689 VERIFY(!cb_data[i]->zcd_called);
4690 }
4691
4692 dmu_tx_abort(tx);
4693
4694 for (i = 0; i < 2; i++) {
4695 VERIFY(cb_data[i]->zcd_called);
4696 umem_free(cb_data[i], sizeof (ztest_cb_data_t));
4697 }
4698
4699 return;
4700 }
4701
4702 cb_data[2] = ztest_create_cb_data(os, txg);
4703 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[2]);
4704
4705 /*
4706 * Read existing data to make sure there isn't a future leak.
4707 */
4708 VERIFY(0 == dmu_read(os, od[0].od_object, 0, sizeof (uint64_t),
4709 &old_txg, DMU_READ_PREFETCH));
4710
4711 if (old_txg > txg)
4712 fatal(0, "future leak: got %" PRIu64 ", open txg is %" PRIu64,
4713 old_txg, txg);
4714
4715 dmu_write(os, od[0].od_object, 0, sizeof (uint64_t), &txg, tx);
4716
4717 mutex_enter(&zcl.zcl_callbacks_lock);
4718
4719 /*
4720 * Since commit callbacks don't have any ordering requirement and since
4721 * it is theoretically possible for a commit callback to be called
4722 * after an arbitrary amount of time has elapsed since its txg has been
4723 * synced, it is difficult to reliably determine whether a commit
4724 * callback hasn't been called due to high load or due to a flawed
4725 * implementation.
4726 *
4727 * In practice, we will assume that if after a certain number of txgs a
4728 * commit callback hasn't been called, then most likely there's an
4729 * implementation bug..
4730 */
4731 tmp_cb = list_head(&zcl.zcl_callbacks);
4732 if (tmp_cb != NULL &&
4733 (txg - ZTEST_COMMIT_CALLBACK_THRESH) > tmp_cb->zcd_txg) {
4734 fatal(0, "Commit callback threshold exceeded, oldest txg: %"
4735 PRIu64 ", open txg: %" PRIu64 "\n", tmp_cb->zcd_txg, txg);
4736 }
4737
4738 /*
4739 * Let's find the place to insert our callbacks.
4740 *
4741 * Even though the list is ordered by txg, it is possible for the
4742 * insertion point to not be the end because our txg may already be
4743 * quiescing at this point and other callbacks in the open txg
4744 * (from other objsets) may have sneaked in.
4745 */
4746 tmp_cb = list_tail(&zcl.zcl_callbacks);
4747 while (tmp_cb != NULL && tmp_cb->zcd_txg > txg)
4748 tmp_cb = list_prev(&zcl.zcl_callbacks, tmp_cb);
4749
4750 /* Add the 3 callbacks to the list */
4751 for (i = 0; i < 3; i++) {
4752 if (tmp_cb == NULL)
4753 list_insert_head(&zcl.zcl_callbacks, cb_data[i]);
4754 else
4755 list_insert_after(&zcl.zcl_callbacks, tmp_cb,
4756 cb_data[i]);
4757
4758 cb_data[i]->zcd_added = B_TRUE;
4759 VERIFY(!cb_data[i]->zcd_called);
4760
4761 tmp_cb = cb_data[i];
4762 }
4763
4764 mutex_exit(&zcl.zcl_callbacks_lock);
4765
4766 dmu_tx_commit(tx);
4767 }
4768
4769 /* ARGSUSED */
4770 void
4771 ztest_dsl_prop_get_set(ztest_ds_t *zd, uint64_t id)
4772 {
4773 zfs_prop_t proplist[] = {
4774 ZFS_PROP_CHECKSUM,
4775 ZFS_PROP_COMPRESSION,
4776 ZFS_PROP_COPIES,
4777 ZFS_PROP_DEDUP
4778 };
4779
4780 rw_enter(&ztest_name_lock, RW_READER);
4781
4782 for (int p = 0; p < sizeof (proplist) / sizeof (proplist[0]); p++)
4783 (void) ztest_dsl_prop_set_uint64(zd->zd_name, proplist[p],
4784 ztest_random_dsl_prop(proplist[p]), (int)ztest_random(2));
4785
4786 rw_exit(&ztest_name_lock);
4787 }
4788
4789 /* ARGSUSED */
4790 void
4791 ztest_remap_blocks(ztest_ds_t *zd, uint64_t id)
4792 {
4793 rw_enter(&ztest_name_lock, RW_READER);
4794
4795 int error = dmu_objset_remap_indirects(zd->zd_name);
4796 if (error == ENOSPC)
4797 error = 0;
4798 ASSERT0(error);
4799
4800 rw_exit(&ztest_name_lock);
4801 }
4802
4803 /* ARGSUSED */
4804 void
4805 ztest_spa_prop_get_set(ztest_ds_t *zd, uint64_t id)
4806 {
4807 nvlist_t *props = NULL;
4808
4809 rw_enter(&ztest_name_lock, RW_READER);
4810
4811 (void) ztest_spa_prop_set_uint64(ZPOOL_PROP_DEDUPDITTO,
4812 ZIO_DEDUPDITTO_MIN + ztest_random(ZIO_DEDUPDITTO_MIN));
4813
4814 VERIFY0(spa_prop_get(ztest_spa, &props));
4815
4816 if (ztest_opts.zo_verbose >= 6)
4817 dump_nvlist(props, 4);
4818
4819 nvlist_free(props);
4820
4821 rw_exit(&ztest_name_lock);
4822 }
4823
4824 static int
4825 user_release_one(const char *snapname, const char *holdname)
4826 {
4827 nvlist_t *snaps, *holds;
4828 int error;
4829
4830 snaps = fnvlist_alloc();
4831 holds = fnvlist_alloc();
4832 fnvlist_add_boolean(holds, holdname);
4833 fnvlist_add_nvlist(snaps, snapname, holds);
4834 fnvlist_free(holds);
4835 error = dsl_dataset_user_release(snaps, NULL);
4836 fnvlist_free(snaps);
4837 return (error);
4838 }
4839
4840 /*
4841 * Test snapshot hold/release and deferred destroy.
4842 */
4843 void
4844 ztest_dmu_snapshot_hold(ztest_ds_t *zd, uint64_t id)
4845 {
4846 int error;
4847 objset_t *os = zd->zd_os;
4848 objset_t *origin;
4849 char snapname[100];
4850 char fullname[100];
4851 char clonename[100];
4852 char tag[100];
4853 char osname[ZFS_MAX_DATASET_NAME_LEN];
4854 nvlist_t *holds;
4855
4856 rw_enter(&ztest_name_lock, RW_READER);
4857
4858 dmu_objset_name(os, osname);
4859
4860 (void) snprintf(snapname, sizeof (snapname), "sh1_%llu", id);
4861 (void) snprintf(fullname, sizeof (fullname), "%s@%s", osname, snapname);
4862 (void) snprintf(clonename, sizeof (clonename),
4863 "%s/ch1_%llu", osname, id);
4864 (void) snprintf(tag, sizeof (tag), "tag_%llu", id);
4865
4866 /*
4867 * Clean up from any previous run.
4868 */
4869 error = dsl_destroy_head(clonename);
4870 if (error != ENOENT)
4871 ASSERT0(error);
4872 error = user_release_one(fullname, tag);
4873 if (error != ESRCH && error != ENOENT)
4874 ASSERT0(error);
4875 error = dsl_destroy_snapshot(fullname, B_FALSE);
4876 if (error != ENOENT)
4877 ASSERT0(error);
4878
4879 /*
4880 * Create snapshot, clone it, mark snap for deferred destroy,
4881 * destroy clone, verify snap was also destroyed.
4882 */
4883 error = dmu_objset_snapshot_one(osname, snapname);
4884 if (error) {
4885 if (error == ENOSPC) {
4886 ztest_record_enospc("dmu_objset_snapshot");
4887 goto out;
4888 }
4889 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
4890 }
4891
4892 error = dmu_objset_clone(clonename, fullname);
4893 if (error) {
4894 if (error == ENOSPC) {
4895 ztest_record_enospc("dmu_objset_clone");
4896 goto out;
4897 }
4898 fatal(0, "dmu_objset_clone(%s) = %d", clonename, error);
4899 }
4900
4901 error = dsl_destroy_snapshot(fullname, B_TRUE);
4902 if (error) {
4903 fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
4904 fullname, error);
4905 }
4906
4907 error = dsl_destroy_head(clonename);
4908 if (error)
4909 fatal(0, "dsl_destroy_head(%s) = %d", clonename, error);
4910
4911 error = dmu_objset_hold(fullname, FTAG, &origin);
4912 if (error != ENOENT)
4913 fatal(0, "dmu_objset_hold(%s) = %d", fullname, error);
4914
4915 /*
4916 * Create snapshot, add temporary hold, verify that we can't
4917 * destroy a held snapshot, mark for deferred destroy,
4918 * release hold, verify snapshot was destroyed.
4919 */
4920 error = dmu_objset_snapshot_one(osname, snapname);
4921 if (error) {
4922 if (error == ENOSPC) {
4923 ztest_record_enospc("dmu_objset_snapshot");
4924 goto out;
4925 }
4926 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
4927 }
4928
4929 holds = fnvlist_alloc();
4930 fnvlist_add_string(holds, fullname, tag);
4931 error = dsl_dataset_user_hold(holds, 0, NULL);
4932 fnvlist_free(holds);
4933
4934 if (error == ENOSPC) {
4935 ztest_record_enospc("dsl_dataset_user_hold");
4936 goto out;
4937 } else if (error) {
4938 fatal(0, "dsl_dataset_user_hold(%s, %s) = %u",
4939 fullname, tag, error);
4940 }
4941
4942 error = dsl_destroy_snapshot(fullname, B_FALSE);
4943 if (error != EBUSY) {
4944 fatal(0, "dsl_destroy_snapshot(%s, B_FALSE) = %d",
4945 fullname, error);
4946 }
4947
4948 error = dsl_destroy_snapshot(fullname, B_TRUE);
4949 if (error) {
4950 fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
4951 fullname, error);
4952 }
4953
4954 error = user_release_one(fullname, tag);
4955 if (error)
4956 fatal(0, "user_release_one(%s, %s) = %d", fullname, tag, error);
4957
4958 VERIFY3U(dmu_objset_hold(fullname, FTAG, &origin), ==, ENOENT);
4959
4960 out:
4961 rw_exit(&ztest_name_lock);
4962 }
4963
4964 /*
4965 * Inject random faults into the on-disk data.
4966 */
4967 /* ARGSUSED */
4968 void
4969 ztest_fault_inject(ztest_ds_t *zd, uint64_t id)
4970 {
4971 ztest_shared_t *zs = ztest_shared;
4972 spa_t *spa = ztest_spa;
4973 int fd;
4974 uint64_t offset;
4975 uint64_t leaves;
4976 uint64_t bad = 0x1990c0ffeedecade;
4977 uint64_t top, leaf;
4978 char path0[MAXPATHLEN];
4979 char pathrand[MAXPATHLEN];
4980 size_t fsize;
4981 int bshift = SPA_MAXBLOCKSHIFT + 2;
4982 int iters = 1000;
4983 int maxfaults;
4984 int mirror_save;
4985 vdev_t *vd0 = NULL;
4986 uint64_t guid0 = 0;
4987 boolean_t islog = B_FALSE;
4988
4989 mutex_enter(&ztest_vdev_lock);
4990 maxfaults = MAXFAULTS();
4991 leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz;
4992 mirror_save = zs->zs_mirrors;
4993 mutex_exit(&ztest_vdev_lock);
4994
4995 ASSERT(leaves >= 1);
4996
4997 /*
4998 * Grab the name lock as reader. There are some operations
4999 * which don't like to have their vdevs changed while
5000 * they are in progress (i.e. spa_change_guid). Those
5001 * operations will have grabbed the name lock as writer.
5002 */
5003 rw_enter(&ztest_name_lock, RW_READER);
5004
5005 /*
5006 * We need SCL_STATE here because we're going to look at vd0->vdev_tsd.
5007 */
5008 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5009
5010 if (ztest_random(2) == 0) {
5011 /*
5012 * Inject errors on a normal data device or slog device.
5013 */
5014 top = ztest_random_vdev_top(spa, B_TRUE);
5015 leaf = ztest_random(leaves) + zs->zs_splits;
5016
5017 /*
5018 * Generate paths to the first leaf in this top-level vdev,
5019 * and to the random leaf we selected. We'll induce transient
5020 * write failures and random online/offline activity on leaf 0,
5021 * and we'll write random garbage to the randomly chosen leaf.
5022 */
5023 (void) snprintf(path0, sizeof (path0), ztest_dev_template,
5024 ztest_opts.zo_dir, ztest_opts.zo_pool,
5025 top * leaves + zs->zs_splits);
5026 (void) snprintf(pathrand, sizeof (pathrand), ztest_dev_template,
5027 ztest_opts.zo_dir, ztest_opts.zo_pool,
5028 top * leaves + leaf);
5029
5030 vd0 = vdev_lookup_by_path(spa->spa_root_vdev, path0);
5031 if (vd0 != NULL && vd0->vdev_top->vdev_islog)
5032 islog = B_TRUE;
5033
5034 /*
5035 * If the top-level vdev needs to be resilvered
5036 * then we only allow faults on the device that is
5037 * resilvering.
5038 */
5039 if (vd0 != NULL && maxfaults != 1 &&
5040 (!vdev_resilver_needed(vd0->vdev_top, NULL, NULL) ||
5041 vd0->vdev_resilver_txg != 0)) {
5042 /*
5043 * Make vd0 explicitly claim to be unreadable,
5044 * or unwriteable, or reach behind its back
5045 * and close the underlying fd. We can do this if
5046 * maxfaults == 0 because we'll fail and reexecute,
5047 * and we can do it if maxfaults >= 2 because we'll
5048 * have enough redundancy. If maxfaults == 1, the
5049 * combination of this with injection of random data
5050 * corruption below exceeds the pool's fault tolerance.
5051 */
5052 vdev_file_t *vf = vd0->vdev_tsd;
5053
5054 zfs_dbgmsg("injecting fault to vdev %llu; maxfaults=%d",
5055 (long long)vd0->vdev_id, (int)maxfaults);
5056
5057 if (vf != NULL && ztest_random(3) == 0) {
5058 (void) close(vf->vf_vnode->v_fd);
5059 vf->vf_vnode->v_fd = -1;
5060 } else if (ztest_random(2) == 0) {
5061 vd0->vdev_cant_read = B_TRUE;
5062 } else {
5063 vd0->vdev_cant_write = B_TRUE;
5064 }
5065 guid0 = vd0->vdev_guid;
5066 }
5067 } else {
5068 /*
5069 * Inject errors on an l2cache device.
5070 */
5071 spa_aux_vdev_t *sav = &spa->spa_l2cache;
5072
5073 if (sav->sav_count == 0) {
5074 spa_config_exit(spa, SCL_STATE, FTAG);
5075 rw_exit(&ztest_name_lock);
5076 return;
5077 }
5078 vd0 = sav->sav_vdevs[ztest_random(sav->sav_count)];
5079 guid0 = vd0->vdev_guid;
5080 (void) strcpy(path0, vd0->vdev_path);
5081 (void) strcpy(pathrand, vd0->vdev_path);
5082
5083 leaf = 0;
5084 leaves = 1;
5085 maxfaults = INT_MAX; /* no limit on cache devices */
5086 }
5087
5088 spa_config_exit(spa, SCL_STATE, FTAG);
5089 rw_exit(&ztest_name_lock);
5090
5091 /*
5092 * If we can tolerate two or more faults, or we're dealing
5093 * with a slog, randomly online/offline vd0.
5094 */
5095 if ((maxfaults >= 2 || islog) && guid0 != 0) {
5096 if (ztest_random(10) < 6) {
5097 int flags = (ztest_random(2) == 0 ?
5098 ZFS_OFFLINE_TEMPORARY : 0);
5099
5100 /*
5101 * We have to grab the zs_name_lock as writer to
5102 * prevent a race between offlining a slog and
5103 * destroying a dataset. Offlining the slog will
5104 * grab a reference on the dataset which may cause
5105 * dmu_objset_destroy() to fail with EBUSY thus
5106 * leaving the dataset in an inconsistent state.
5107 */
5108 if (islog)
5109 rw_enter(&ztest_name_lock, RW_WRITER);
5110
5111 VERIFY(vdev_offline(spa, guid0, flags) != EBUSY);
5112
5113 if (islog)
5114 rw_exit(&ztest_name_lock);
5115 } else {
5116 /*
5117 * Ideally we would like to be able to randomly
5118 * call vdev_[on|off]line without holding locks
5119 * to force unpredictable failures but the side
5120 * effects of vdev_[on|off]line prevent us from
5121 * doing so. We grab the ztest_vdev_lock here to
5122 * prevent a race between injection testing and
5123 * aux_vdev removal.
5124 */
5125 mutex_enter(&ztest_vdev_lock);
5126 (void) vdev_online(spa, guid0, 0, NULL);
5127 mutex_exit(&ztest_vdev_lock);
5128 }
5129 }
5130
5131 if (maxfaults == 0)
5132 return;
5133
5134 /*
5135 * We have at least single-fault tolerance, so inject data corruption.
5136 */
5137 fd = open(pathrand, O_RDWR);
5138
5139 if (fd == -1) /* we hit a gap in the device namespace */
5140 return;
5141
5142 fsize = lseek(fd, 0, SEEK_END);
5143
5144 while (--iters != 0) {
5145 /*
5146 * The offset must be chosen carefully to ensure that
5147 * we do not inject a given logical block with errors
5148 * on two different leaf devices, because ZFS can not
5149 * tolerate that (if maxfaults==1).
5150 *
5151 * We divide each leaf into chunks of size
5152 * (# leaves * SPA_MAXBLOCKSIZE * 4). Within each chunk
5153 * there is a series of ranges to which we can inject errors.
5154 * Each range can accept errors on only a single leaf vdev.
5155 * The error injection ranges are separated by ranges
5156 * which we will not inject errors on any device (DMZs).
5157 * Each DMZ must be large enough such that a single block
5158 * can not straddle it, so that a single block can not be
5159 * a target in two different injection ranges (on different
5160 * leaf vdevs).
5161 *
5162 * For example, with 3 leaves, each chunk looks like:
5163 * 0 to 32M: injection range for leaf 0
5164 * 32M to 64M: DMZ - no injection allowed
5165 * 64M to 96M: injection range for leaf 1
5166 * 96M to 128M: DMZ - no injection allowed
5167 * 128M to 160M: injection range for leaf 2
5168 * 160M to 192M: DMZ - no injection allowed
5169 */
5170 offset = ztest_random(fsize / (leaves << bshift)) *
5171 (leaves << bshift) + (leaf << bshift) +
5172 (ztest_random(1ULL << (bshift - 1)) & -8ULL);
5173
5174 /*
5175 * Only allow damage to the labels at one end of the vdev.
5176 *
5177 * If all labels are damaged, the device will be totally
5178 * inaccessible, which will result in loss of data,
5179 * because we also damage (parts of) the other side of
5180 * the mirror/raidz.
5181 *
5182 * Additionally, we will always have both an even and an
5183 * odd label, so that we can handle crashes in the
5184 * middle of vdev_config_sync().
5185 */
5186 if ((leaf & 1) == 0 && offset < VDEV_LABEL_START_SIZE)
5187 continue;
5188
5189 /*
5190 * The two end labels are stored at the "end" of the disk, but
5191 * the end of the disk (vdev_psize) is aligned to
5192 * sizeof (vdev_label_t).
5193 */
5194 uint64_t psize = P2ALIGN(fsize, sizeof (vdev_label_t));
5195 if ((leaf & 1) == 1 &&
5196 offset + sizeof (bad) > psize - VDEV_LABEL_END_SIZE)
5197 continue;
5198
5199 mutex_enter(&ztest_vdev_lock);
5200 if (mirror_save != zs->zs_mirrors) {
5201 mutex_exit(&ztest_vdev_lock);
5202 (void) close(fd);
5203 return;
5204 }
5205
5206 if (pwrite(fd, &bad, sizeof (bad), offset) != sizeof (bad))
5207 fatal(1, "can't inject bad word at 0x%llx in %s",
5208 offset, pathrand);
5209
5210 mutex_exit(&ztest_vdev_lock);
5211
5212 if (ztest_opts.zo_verbose >= 7)
5213 (void) printf("injected bad word into %s,"
5214 " offset 0x%llx\n", pathrand, (u_longlong_t)offset);
5215 }
5216
5217 (void) close(fd);
5218 }
5219
5220 /*
5221 * Verify that DDT repair works as expected.
5222 */
5223 void
5224 ztest_ddt_repair(ztest_ds_t *zd, uint64_t id)
5225 {
5226 ztest_shared_t *zs = ztest_shared;
5227 spa_t *spa = ztest_spa;
5228 objset_t *os = zd->zd_os;
5229 ztest_od_t od[1];
5230 uint64_t object, blocksize, txg, pattern, psize;
5231 enum zio_checksum checksum = spa_dedup_checksum(spa);
5232 dmu_buf_t *db;
5233 dmu_tx_t *tx;
5234 abd_t *abd;
5235 blkptr_t blk;
5236 int copies = 2 * ZIO_DEDUPDITTO_MIN;
5237
5238 blocksize = ztest_random_blocksize();
5239 blocksize = MIN(blocksize, 2048); /* because we write so many */
5240
5241 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
5242
5243 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
5244 return;
5245
5246 /*
5247 * Take the name lock as writer to prevent anyone else from changing
5248 * the pool and dataset properies we need to maintain during this test.
5249 */
5250 rw_enter(&ztest_name_lock, RW_WRITER);
5251
5252 if (ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_DEDUP, checksum,
5253 B_FALSE) != 0 ||
5254 ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_COPIES, 1,
5255 B_FALSE) != 0) {
5256 rw_exit(&ztest_name_lock);
5257 return;
5258 }
5259
5260 dmu_objset_stats_t dds;
5261 dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
5262 dmu_objset_fast_stat(os, &dds);
5263 dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
5264
5265 object = od[0].od_object;
5266 blocksize = od[0].od_blocksize;
5267 pattern = zs->zs_guid ^ dds.dds_guid;
5268
5269 ASSERT(object != 0);
5270
5271 tx = dmu_tx_create(os);
5272 dmu_tx_hold_write(tx, object, 0, copies * blocksize);
5273 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
5274 if (txg == 0) {
5275 rw_exit(&ztest_name_lock);
5276 return;
5277 }
5278
5279 /*
5280 * Write all the copies of our block.
5281 */
5282 for (int i = 0; i < copies; i++) {
5283 uint64_t offset = i * blocksize;
5284 int error = dmu_buf_hold(os, object, offset, FTAG, &db,
5285 DMU_READ_NO_PREFETCH);
5286 if (error != 0) {
5287 fatal(B_FALSE, "dmu_buf_hold(%p, %llu, %llu) = %u",
5288 os, (long long)object, (long long) offset, error);
5289 }
5290 ASSERT(db->db_offset == offset);
5291 ASSERT(db->db_size == blocksize);
5292 ASSERT(ztest_pattern_match(db->db_data, db->db_size, pattern) ||
5293 ztest_pattern_match(db->db_data, db->db_size, 0ULL));
5294 dmu_buf_will_fill(db, tx);
5295 ztest_pattern_set(db->db_data, db->db_size, pattern);
5296 dmu_buf_rele(db, FTAG);
5297 }
5298
5299 dmu_tx_commit(tx);
5300 txg_wait_synced(spa_get_dsl(spa), txg);
5301
5302 /*
5303 * Find out what block we got.
5304 */
5305 VERIFY0(dmu_buf_hold(os, object, 0, FTAG, &db,
5306 DMU_READ_NO_PREFETCH));
5307 blk = *((dmu_buf_impl_t *)db)->db_blkptr;
5308 dmu_buf_rele(db, FTAG);
5309
5310 /*
5311 * Damage the block. Dedup-ditto will save us when we read it later.
5312 */
5313 psize = BP_GET_PSIZE(&blk);
5314 abd = abd_alloc_linear(psize, B_TRUE);
5315 ztest_pattern_set(abd_to_buf(abd), psize, ~pattern);
5316
5317 (void) zio_wait(zio_rewrite(NULL, spa, 0, &blk,
5318 abd, psize, NULL, NULL, ZIO_PRIORITY_SYNC_WRITE,
5319 ZIO_FLAG_CANFAIL | ZIO_FLAG_INDUCE_DAMAGE, NULL));
5320
5321 abd_free(abd);
5322
5323 rw_exit(&ztest_name_lock);
5324 }
5325
5326 /*
5327 * Scrub the pool.
5328 */
5329 /* ARGSUSED */
5330 void
5331 ztest_scrub(ztest_ds_t *zd, uint64_t id)
5332 {
5333 spa_t *spa = ztest_spa;
5334
5335 (void) spa_scan(spa, POOL_SCAN_SCRUB);
5336 (void) poll(NULL, 0, 100); /* wait a moment, then force a restart */
5337 (void) spa_scan(spa, POOL_SCAN_SCRUB);
5338 }
5339
5340 /*
5341 * Change the guid for the pool.
5342 */
5343 /* ARGSUSED */
5344 void
5345 ztest_reguid(ztest_ds_t *zd, uint64_t id)
5346 {
5347 spa_t *spa = ztest_spa;
5348 uint64_t orig, load;
5349 int error;
5350
5351 orig = spa_guid(spa);
5352 load = spa_load_guid(spa);
5353
5354 rw_enter(&ztest_name_lock, RW_WRITER);
5355 error = spa_change_guid(spa);
5356 rw_exit(&ztest_name_lock);
5357
5358 if (error != 0)
5359 return;
5360
5361 if (ztest_opts.zo_verbose >= 4) {
5362 (void) printf("Changed guid old %llu -> %llu\n",
5363 (u_longlong_t)orig, (u_longlong_t)spa_guid(spa));
5364 }
5365
5366 VERIFY3U(orig, !=, spa_guid(spa));
5367 VERIFY3U(load, ==, spa_load_guid(spa));
5368 }
5369
5370 /*
5371 * Rename the pool to a different name and then rename it back.
5372 */
5373 /* ARGSUSED */
5374 void
5375 ztest_spa_rename(ztest_ds_t *zd, uint64_t id)
5376 {
5377 char *oldname, *newname;
5378 spa_t *spa;
5379
5380 rw_enter(&ztest_name_lock, RW_WRITER);
5381
5382 oldname = ztest_opts.zo_pool;
5383 newname = umem_alloc(strlen(oldname) + 5, UMEM_NOFAIL);
5384 (void) strcpy(newname, oldname);
5385 (void) strcat(newname, "_tmp");
5386
5387 /*
5388 * Do the rename
5389 */
5390 VERIFY3U(0, ==, spa_rename(oldname, newname));
5391
5392 /*
5393 * Try to open it under the old name, which shouldn't exist
5394 */
5395 VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
5396
5397 /*
5398 * Open it under the new name and make sure it's still the same spa_t.
5399 */
5400 VERIFY3U(0, ==, spa_open(newname, &spa, FTAG));
5401
5402 ASSERT(spa == ztest_spa);
5403 spa_close(spa, FTAG);
5404
5405 /*
5406 * Rename it back to the original
5407 */
5408 VERIFY3U(0, ==, spa_rename(newname, oldname));
5409
5410 /*
5411 * Make sure it can still be opened
5412 */
5413 VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG));
5414
5415 ASSERT(spa == ztest_spa);
5416 spa_close(spa, FTAG);
5417
5418 umem_free(newname, strlen(newname) + 1);
5419
5420 rw_exit(&ztest_name_lock);
5421 }
5422
5423 /*
5424 * Verify pool integrity by running zdb.
5425 */
5426 static void
5427 ztest_run_zdb(char *pool)
5428 {
5429 int status;
5430 char zdb[MAXPATHLEN + MAXNAMELEN + 20];
5431 char zbuf[1024];
5432 char *bin;
5433 char *ztest;
5434 char *isa;
5435 int isalen;
5436 FILE *fp;
5437
5438 (void) realpath(getexecname(), zdb);
5439
5440 /* zdb lives in /usr/sbin, while ztest lives in /usr/bin */
5441 bin = strstr(zdb, "/usr/bin/");
5442 ztest = strstr(bin, "/ztest");
5443 isa = bin + 8;
5444 isalen = ztest - isa;
5445 isa = strdup(isa);
5446 /* LINTED */
5447 (void) sprintf(bin,
5448 "/usr/sbin%.*s/zdb -bcc%s%s -G -d -U %s %s",
5449 isalen,
5450 isa,
5451 ztest_opts.zo_verbose >= 3 ? "s" : "",
5452 ztest_opts.zo_verbose >= 4 ? "v" : "",
5453 spa_config_path,
5454 pool);
5455 free(isa);
5456
5457 if (ztest_opts.zo_verbose >= 5)
5458 (void) printf("Executing %s\n", strstr(zdb, "zdb "));
5459
5460 fp = popen(zdb, "r");
5461
5462 while (fgets(zbuf, sizeof (zbuf), fp) != NULL)
5463 if (ztest_opts.zo_verbose >= 3)
5464 (void) printf("%s", zbuf);
5465
5466 status = pclose(fp);
5467
5468 if (status == 0)
5469 return;
5470
5471 ztest_dump_core = 0;
5472 if (WIFEXITED(status))
5473 fatal(0, "'%s' exit code %d", zdb, WEXITSTATUS(status));
5474 else
5475 fatal(0, "'%s' died with signal %d", zdb, WTERMSIG(status));
5476 }
5477
5478 static void
5479 ztest_walk_pool_directory(char *header)
5480 {
5481 spa_t *spa = NULL;
5482
5483 if (ztest_opts.zo_verbose >= 6)
5484 (void) printf("%s\n", header);
5485
5486 mutex_enter(&spa_namespace_lock);
5487 while ((spa = spa_next(spa)) != NULL)
5488 if (ztest_opts.zo_verbose >= 6)
5489 (void) printf("\t%s\n", spa_name(spa));
5490 mutex_exit(&spa_namespace_lock);
5491 }
5492
5493 static void
5494 ztest_spa_import_export(char *oldname, char *newname)
5495 {
5496 nvlist_t *config, *newconfig;
5497 uint64_t pool_guid;
5498 spa_t *spa;
5499 int error;
5500
5501 if (ztest_opts.zo_verbose >= 4) {
5502 (void) printf("import/export: old = %s, new = %s\n",
5503 oldname, newname);
5504 }
5505
5506 /*
5507 * Clean up from previous runs.
5508 */
5509 (void) spa_destroy(newname);
5510
5511 /*
5512 * Get the pool's configuration and guid.
5513 */
5514 VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG));
5515
5516 /*
5517 * Kick off a scrub to tickle scrub/export races.
5518 */
5519 if (ztest_random(2) == 0)
5520 (void) spa_scan(spa, POOL_SCAN_SCRUB);
5521
5522 pool_guid = spa_guid(spa);
5523 spa_close(spa, FTAG);
5524
5525 ztest_walk_pool_directory("pools before export");
5526
5527 /*
5528 * Export it.
5529 */
5530 VERIFY3U(0, ==, spa_export(oldname, &config, B_FALSE, B_FALSE));
5531
5532 ztest_walk_pool_directory("pools after export");
5533
5534 /*
5535 * Try to import it.
5536 */
5537 newconfig = spa_tryimport(config);
5538 ASSERT(newconfig != NULL);
5539 nvlist_free(newconfig);
5540
5541 /*
5542 * Import it under the new name.
5543 */
5544 error = spa_import(newname, config, NULL, 0);
5545 if (error != 0) {
5546 dump_nvlist(config, 0);
5547 fatal(B_FALSE, "couldn't import pool %s as %s: error %u",
5548 oldname, newname, error);
5549 }
5550
5551 ztest_walk_pool_directory("pools after import");
5552
5553 /*
5554 * Try to import it again -- should fail with EEXIST.
5555 */
5556 VERIFY3U(EEXIST, ==, spa_import(newname, config, NULL, 0));
5557
5558 /*
5559 * Try to import it under a different name -- should fail with EEXIST.
5560 */
5561 VERIFY3U(EEXIST, ==, spa_import(oldname, config, NULL, 0));
5562
5563 /*
5564 * Verify that the pool is no longer visible under the old name.
5565 */
5566 VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
5567
5568 /*
5569 * Verify that we can open and close the pool using the new name.
5570 */
5571 VERIFY3U(0, ==, spa_open(newname, &spa, FTAG));
5572 ASSERT(pool_guid == spa_guid(spa));
5573 spa_close(spa, FTAG);
5574
5575 nvlist_free(config);
5576 }
5577
5578 static void
5579 ztest_resume(spa_t *spa)
5580 {
5581 if (spa_suspended(spa) && ztest_opts.zo_verbose >= 6)
5582 (void) printf("resuming from suspended state\n");
5583 spa_vdev_state_enter(spa, SCL_NONE);
5584 vdev_clear(spa, NULL);
5585 (void) spa_vdev_state_exit(spa, NULL, 0);
5586 (void) zio_resume(spa);
5587 }
5588
5589 static void *
5590 ztest_resume_thread(void *arg)
5591 {
5592 spa_t *spa = arg;
5593
5594 while (!ztest_exiting) {
5595 if (spa_suspended(spa))
5596 ztest_resume(spa);
5597 (void) poll(NULL, 0, 100);
5598
5599 /*
5600 * Periodically change the zfs_compressed_arc_enabled setting.
5601 */
5602 if (ztest_random(10) == 0)
5603 zfs_compressed_arc_enabled = ztest_random(2);
5604
5605 /*
5606 * Periodically change the zfs_abd_scatter_enabled setting.
5607 */
5608 if (ztest_random(10) == 0)
5609 zfs_abd_scatter_enabled = ztest_random(2);
5610 }
5611 return (NULL);
5612 }
5613
5614 static void *
5615 ztest_deadman_thread(void *arg)
5616 {
5617 ztest_shared_t *zs = arg;
5618 spa_t *spa = ztest_spa;
5619 hrtime_t delta, total = 0;
5620
5621 for (;;) {
5622 delta = zs->zs_thread_stop - zs->zs_thread_start +
5623 MSEC2NSEC(zfs_deadman_synctime_ms);
5624
5625 (void) poll(NULL, 0, (int)NSEC2MSEC(delta));
5626
5627 /*
5628 * If the pool is suspended then fail immediately. Otherwise,
5629 * check to see if the pool is making any progress. If
5630 * vdev_deadman() discovers that there hasn't been any recent
5631 * I/Os then it will end up aborting the tests.
5632 */
5633 if (spa_suspended(spa) || spa->spa_root_vdev == NULL) {
5634 fatal(0, "aborting test after %llu seconds because "
5635 "pool has transitioned to a suspended state.",
5636 zfs_deadman_synctime_ms / 1000);
5637 return (NULL);
5638 }
5639 vdev_deadman(spa->spa_root_vdev);
5640
5641 total += zfs_deadman_synctime_ms/1000;
5642 (void) printf("ztest has been running for %lld seconds\n",
5643 total);
5644 }
5645 }
5646
5647 static void
5648 ztest_execute(int test, ztest_info_t *zi, uint64_t id)
5649 {
5650 ztest_ds_t *zd = &ztest_ds[id % ztest_opts.zo_datasets];
5651 ztest_shared_callstate_t *zc = ZTEST_GET_SHARED_CALLSTATE(test);
5652 hrtime_t functime = gethrtime();
5653
5654 for (int i = 0; i < zi->zi_iters; i++)
5655 zi->zi_func(zd, id);
5656
5657 functime = gethrtime() - functime;
5658
5659 atomic_add_64(&zc->zc_count, 1);
5660 atomic_add_64(&zc->zc_time, functime);
5661
5662 if (ztest_opts.zo_verbose >= 4) {
5663 Dl_info dli;
5664 (void) dladdr((void *)zi->zi_func, &dli);
5665 (void) printf("%6.2f sec in %s\n",
5666 (double)functime / NANOSEC, dli.dli_sname);
5667 }
5668 }
5669
5670 static void *
5671 ztest_thread(void *arg)
5672 {
5673 int rand;
5674 uint64_t id = (uintptr_t)arg;
5675 ztest_shared_t *zs = ztest_shared;
5676 uint64_t call_next;
5677 hrtime_t now;
5678 ztest_info_t *zi;
5679 ztest_shared_callstate_t *zc;
5680
5681 while ((now = gethrtime()) < zs->zs_thread_stop) {
5682 /*
5683 * See if it's time to force a crash.
5684 */
5685 if (now > zs->zs_thread_kill)
5686 ztest_kill(zs);
5687
5688 /*
5689 * If we're getting ENOSPC with some regularity, stop.
5690 */
5691 if (zs->zs_enospc_count > 10)
5692 break;
5693
5694 /*
5695 * Pick a random function to execute.
5696 */
5697 rand = ztest_random(ZTEST_FUNCS);
5698 zi = &ztest_info[rand];
5699 zc = ZTEST_GET_SHARED_CALLSTATE(rand);
5700 call_next = zc->zc_next;
5701
5702 if (now >= call_next &&
5703 atomic_cas_64(&zc->zc_next, call_next, call_next +
5704 ztest_random(2 * zi->zi_interval[0] + 1)) == call_next) {
5705 ztest_execute(rand, zi, id);
5706 }
5707 }
5708
5709 return (NULL);
5710 }
5711
5712 static void
5713 ztest_dataset_name(char *dsname, char *pool, int d)
5714 {
5715 (void) snprintf(dsname, ZFS_MAX_DATASET_NAME_LEN, "%s/ds_%d", pool, d);
5716 }
5717
5718 static void
5719 ztest_dataset_destroy(int d)
5720 {
5721 char name[ZFS_MAX_DATASET_NAME_LEN];
5722
5723 ztest_dataset_name(name, ztest_opts.zo_pool, d);
5724
5725 if (ztest_opts.zo_verbose >= 3)
5726 (void) printf("Destroying %s to free up space\n", name);
5727
5728 /*
5729 * Cleanup any non-standard clones and snapshots. In general,
5730 * ztest thread t operates on dataset (t % zopt_datasets),
5731 * so there may be more than one thing to clean up.
5732 */
5733 for (int t = d; t < ztest_opts.zo_threads;
5734 t += ztest_opts.zo_datasets) {
5735 ztest_dsl_dataset_cleanup(name, t);
5736 }
5737
5738 (void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
5739 DS_FIND_SNAPSHOTS | DS_FIND_CHILDREN);
5740 }
5741
5742 static void
5743 ztest_dataset_dirobj_verify(ztest_ds_t *zd)
5744 {
5745 uint64_t usedobjs, dirobjs, scratch;
5746
5747 /*
5748 * ZTEST_DIROBJ is the object directory for the entire dataset.
5749 * Therefore, the number of objects in use should equal the
5750 * number of ZTEST_DIROBJ entries, +1 for ZTEST_DIROBJ itself.
5751 * If not, we have an object leak.
5752 *
5753 * Note that we can only check this in ztest_dataset_open(),
5754 * when the open-context and syncing-context values agree.
5755 * That's because zap_count() returns the open-context value,
5756 * while dmu_objset_space() returns the rootbp fill count.
5757 */
5758 VERIFY3U(0, ==, zap_count(zd->zd_os, ZTEST_DIROBJ, &dirobjs));
5759 dmu_objset_space(zd->zd_os, &scratch, &scratch, &usedobjs, &scratch);
5760 ASSERT3U(dirobjs + 1, ==, usedobjs);
5761 }
5762
5763 static int
5764 ztest_dataset_open(int d)
5765 {
5766 ztest_ds_t *zd = &ztest_ds[d];
5767 uint64_t committed_seq = ZTEST_GET_SHARED_DS(d)->zd_seq;
5768 objset_t *os;
5769 zilog_t *zilog;
5770 char name[ZFS_MAX_DATASET_NAME_LEN];
5771 int error;
5772
5773 ztest_dataset_name(name, ztest_opts.zo_pool, d);
5774
5775 rw_enter(&ztest_name_lock, RW_READER);
5776
5777 error = ztest_dataset_create(name);
5778 if (error == ENOSPC) {
5779 rw_exit(&ztest_name_lock);
5780 ztest_record_enospc(FTAG);
5781 return (error);
5782 }
5783 ASSERT(error == 0 || error == EEXIST);
5784
5785 VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, zd, &os));
5786 rw_exit(&ztest_name_lock);
5787
5788 ztest_zd_init(zd, ZTEST_GET_SHARED_DS(d), os);
5789
5790 zilog = zd->zd_zilog;
5791
5792 if (zilog->zl_header->zh_claim_lr_seq != 0 &&
5793 zilog->zl_header->zh_claim_lr_seq < committed_seq)
5794 fatal(0, "missing log records: claimed %llu < committed %llu",
5795 zilog->zl_header->zh_claim_lr_seq, committed_seq);
5796
5797 ztest_dataset_dirobj_verify(zd);
5798
5799 zil_replay(os, zd, ztest_replay_vector);
5800
5801 ztest_dataset_dirobj_verify(zd);
5802
5803 if (ztest_opts.zo_verbose >= 6)
5804 (void) printf("%s replay %llu blocks, %llu records, seq %llu\n",
5805 zd->zd_name,
5806 (u_longlong_t)zilog->zl_parse_blk_count,
5807 (u_longlong_t)zilog->zl_parse_lr_count,
5808 (u_longlong_t)zilog->zl_replaying_seq);
5809
5810 zilog = zil_open(os, ztest_get_data);
5811
5812 if (zilog->zl_replaying_seq != 0 &&
5813 zilog->zl_replaying_seq < committed_seq)
5814 fatal(0, "missing log records: replayed %llu < committed %llu",
5815 zilog->zl_replaying_seq, committed_seq);
5816
5817 return (0);
5818 }
5819
5820 static void
5821 ztest_dataset_close(int d)
5822 {
5823 ztest_ds_t *zd = &ztest_ds[d];
5824
5825 zil_close(zd->zd_zilog);
5826 dmu_objset_disown(zd->zd_os, zd);
5827
5828 ztest_zd_fini(zd);
5829 }
5830
5831 /*
5832 * Kick off threads to run tests on all datasets in parallel.
5833 */
5834 static void
5835 ztest_run(ztest_shared_t *zs)
5836 {
5837 thread_t *tid;
5838 spa_t *spa;
5839 objset_t *os;
5840 thread_t resume_tid;
5841 int error;
5842
5843 ztest_exiting = B_FALSE;
5844
5845 /*
5846 * Initialize parent/child shared state.
5847 */
5848 mutex_init(&ztest_checkpoint_lock, NULL, USYNC_THREAD, NULL);
5849 mutex_init(&ztest_vdev_lock, NULL, USYNC_THREAD, NULL);
5850 rw_init(&ztest_name_lock, NULL, USYNC_THREAD, NULL);
5851
5852 zs->zs_thread_start = gethrtime();
5853 zs->zs_thread_stop =
5854 zs->zs_thread_start + ztest_opts.zo_passtime * NANOSEC;
5855 zs->zs_thread_stop = MIN(zs->zs_thread_stop, zs->zs_proc_stop);
5856 zs->zs_thread_kill = zs->zs_thread_stop;
5857 if (ztest_random(100) < ztest_opts.zo_killrate) {
5858 zs->zs_thread_kill -=
5859 ztest_random(ztest_opts.zo_passtime * NANOSEC);
5860 }
5861
5862 mutex_init(&zcl.zcl_callbacks_lock, NULL, USYNC_THREAD, NULL);
5863
5864 list_create(&zcl.zcl_callbacks, sizeof (ztest_cb_data_t),
5865 offsetof(ztest_cb_data_t, zcd_node));
5866
5867 /*
5868 * Open our pool.
5869 */
5870 kernel_init(FREAD | FWRITE);
5871 VERIFY0(spa_open(ztest_opts.zo_pool, &spa, FTAG));
5872 spa->spa_debug = B_TRUE;
5873 metaslab_preload_limit = ztest_random(20) + 1;
5874 ztest_spa = spa;
5875
5876 dmu_objset_stats_t dds;
5877 VERIFY0(dmu_objset_own(ztest_opts.zo_pool,
5878 DMU_OST_ANY, B_TRUE, FTAG, &os));
5879 dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
5880 dmu_objset_fast_stat(os, &dds);
5881 dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
5882 zs->zs_guid = dds.dds_guid;
5883 dmu_objset_disown(os, FTAG);
5884
5885 spa->spa_dedup_ditto = 2 * ZIO_DEDUPDITTO_MIN;
5886
5887 /*
5888 * We don't expect the pool to suspend unless maxfaults == 0,
5889 * in which case ztest_fault_inject() temporarily takes away
5890 * the only valid replica.
5891 */
5892 if (MAXFAULTS() == 0)
5893 spa->spa_failmode = ZIO_FAILURE_MODE_WAIT;
5894 else
5895 spa->spa_failmode = ZIO_FAILURE_MODE_PANIC;
5896
5897 /*
5898 * Create a thread to periodically resume suspended I/O.
5899 */
5900 VERIFY(thr_create(0, 0, ztest_resume_thread, spa, THR_BOUND,
5901 &resume_tid) == 0);
5902
5903 /*
5904 * Create a deadman thread to abort() if we hang.
5905 */
5906 VERIFY(thr_create(0, 0, ztest_deadman_thread, zs, THR_BOUND,
5907 NULL) == 0);
5908
5909 /*
5910 * Verify that we can safely inquire about any object,
5911 * whether it's allocated or not. To make it interesting,
5912 * we probe a 5-wide window around each power of two.
5913 * This hits all edge cases, including zero and the max.
5914 */
5915 for (int t = 0; t < 64; t++) {
5916 for (int d = -5; d <= 5; d++) {
5917 error = dmu_object_info(spa->spa_meta_objset,
5918 (1ULL << t) + d, NULL);
5919 ASSERT(error == 0 || error == ENOENT ||
5920 error == EINVAL);
5921 }
5922 }
5923
5924 /*
5925 * If we got any ENOSPC errors on the previous run, destroy something.
5926 */
5927 if (zs->zs_enospc_count != 0) {
5928 int d = ztest_random(ztest_opts.zo_datasets);
5929 ztest_dataset_destroy(d);
5930 }
5931 zs->zs_enospc_count = 0;
5932
5933 tid = umem_zalloc(ztest_opts.zo_threads * sizeof (thread_t),
5934 UMEM_NOFAIL);
5935
5936 if (ztest_opts.zo_verbose >= 4)
5937 (void) printf("starting main threads...\n");
5938
5939 /*
5940 * Kick off all the tests that run in parallel.
5941 */
5942 for (int t = 0; t < ztest_opts.zo_threads; t++) {
5943 if (t < ztest_opts.zo_datasets &&
5944 ztest_dataset_open(t) != 0)
5945 return;
5946 VERIFY(thr_create(0, 0, ztest_thread, (void *)(uintptr_t)t,
5947 THR_BOUND, &tid[t]) == 0);
5948 }
5949
5950 /*
5951 * Wait for all of the tests to complete. We go in reverse order
5952 * so we don't close datasets while threads are still using them.
5953 */
5954 for (int t = ztest_opts.zo_threads - 1; t >= 0; t--) {
5955 VERIFY(thr_join(tid[t], NULL, NULL) == 0);
5956 if (t < ztest_opts.zo_datasets)
5957 ztest_dataset_close(t);
5958 }
5959
5960 txg_wait_synced(spa_get_dsl(spa), 0);
5961
5962 zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(spa));
5963 zs->zs_space = metaslab_class_get_space(spa_normal_class(spa));
5964 zfs_dbgmsg_print(FTAG);
5965
5966 umem_free(tid, ztest_opts.zo_threads * sizeof (thread_t));
5967
5968 /* Kill the resume thread */
5969 ztest_exiting = B_TRUE;
5970 VERIFY(thr_join(resume_tid, NULL, NULL) == 0);
5971 ztest_resume(spa);
5972
5973 /*
5974 * Right before closing the pool, kick off a bunch of async I/O;
5975 * spa_close() should wait for it to complete.
5976 */
5977 for (uint64_t object = 1; object < 50; object++) {
5978 dmu_prefetch(spa->spa_meta_objset, object, 0, 0, 1ULL << 20,
5979 ZIO_PRIORITY_SYNC_READ);
5980 }
5981
5982 spa_close(spa, FTAG);
5983
5984 /*
5985 * Verify that we can loop over all pools.
5986 */
5987 mutex_enter(&spa_namespace_lock);
5988 for (spa = spa_next(NULL); spa != NULL; spa = spa_next(spa))
5989 if (ztest_opts.zo_verbose > 3)
5990 (void) printf("spa_next: found %s\n", spa_name(spa));
5991 mutex_exit(&spa_namespace_lock);
5992
5993 /*
5994 * Verify that we can export the pool and reimport it under a
5995 * different name.
5996 */
5997 if (ztest_random(2) == 0) {
5998 char name[ZFS_MAX_DATASET_NAME_LEN];
5999 (void) snprintf(name, sizeof (name), "%s_import",
6000 ztest_opts.zo_pool);
6001 ztest_spa_import_export(ztest_opts.zo_pool, name);
6002 ztest_spa_import_export(name, ztest_opts.zo_pool);
6003 }
6004
6005 kernel_fini();
6006
6007 list_destroy(&zcl.zcl_callbacks);
6008
6009 mutex_destroy(&zcl.zcl_callbacks_lock);
6010
6011 rw_destroy(&ztest_name_lock);
6012 mutex_destroy(&ztest_vdev_lock);
6013 mutex_destroy(&ztest_checkpoint_lock);
6014 }
6015
6016 static void
6017 ztest_freeze(void)
6018 {
6019 ztest_ds_t *zd = &ztest_ds[0];
6020 spa_t *spa;
6021 int numloops = 0;
6022
6023 if (ztest_opts.zo_verbose >= 3)
6024 (void) printf("testing spa_freeze()...\n");
6025
6026 kernel_init(FREAD | FWRITE);
6027 VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
6028 VERIFY3U(0, ==, ztest_dataset_open(0));
6029 spa->spa_debug = B_TRUE;
6030 ztest_spa = spa;
6031
6032 /*
6033 * Force the first log block to be transactionally allocated.
6034 * We have to do this before we freeze the pool -- otherwise
6035 * the log chain won't be anchored.
6036 */
6037 while (BP_IS_HOLE(&zd->zd_zilog->zl_header->zh_log)) {
6038 ztest_dmu_object_alloc_free(zd, 0);
6039 zil_commit(zd->zd_zilog, 0);
6040 }
6041
6042 txg_wait_synced(spa_get_dsl(spa), 0);
6043
6044 /*
6045 * Freeze the pool. This stops spa_sync() from doing anything,
6046 * so that the only way to record changes from now on is the ZIL.
6047 */
6048 spa_freeze(spa);
6049
6050 /*
6051 * Because it is hard to predict how much space a write will actually
6052 * require beforehand, we leave ourselves some fudge space to write over
6053 * capacity.
6054 */
6055 uint64_t capacity = metaslab_class_get_space(spa_normal_class(spa)) / 2;
6056
6057 /*
6058 * Run tests that generate log records but don't alter the pool config
6059 * or depend on DSL sync tasks (snapshots, objset create/destroy, etc).
6060 * We do a txg_wait_synced() after each iteration to force the txg
6061 * to increase well beyond the last synced value in the uberblock.
6062 * The ZIL should be OK with that.
6063 *
6064 * Run a random number of times less than zo_maxloops and ensure we do
6065 * not run out of space on the pool.
6066 */
6067 while (ztest_random(10) != 0 &&
6068 numloops++ < ztest_opts.zo_maxloops &&
6069 metaslab_class_get_alloc(spa_normal_class(spa)) < capacity) {
6070 ztest_od_t od;
6071 ztest_od_init(&od, 0, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
6072 VERIFY0(ztest_object_init(zd, &od, sizeof (od), B_FALSE));
6073 ztest_io(zd, od.od_object,
6074 ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
6075 txg_wait_synced(spa_get_dsl(spa), 0);
6076 }
6077
6078 /*
6079 * Commit all of the changes we just generated.
6080 */
6081 zil_commit(zd->zd_zilog, 0);
6082 txg_wait_synced(spa_get_dsl(spa), 0);
6083
6084 /*
6085 * Close our dataset and close the pool.
6086 */
6087 ztest_dataset_close(0);
6088 spa_close(spa, FTAG);
6089 kernel_fini();
6090
6091 /*
6092 * Open and close the pool and dataset to induce log replay.
6093 */
6094 kernel_init(FREAD | FWRITE);
6095 VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
6096 ASSERT(spa_freeze_txg(spa) == UINT64_MAX);
6097 VERIFY3U(0, ==, ztest_dataset_open(0));
6098 ztest_dataset_close(0);
6099
6100 spa->spa_debug = B_TRUE;
6101 ztest_spa = spa;
6102 txg_wait_synced(spa_get_dsl(spa), 0);
6103 ztest_reguid(NULL, 0);
6104
6105 spa_close(spa, FTAG);
6106 kernel_fini();
6107 }
6108
6109 void
6110 print_time(hrtime_t t, char *timebuf)
6111 {
6112 hrtime_t s = t / NANOSEC;
6113 hrtime_t m = s / 60;
6114 hrtime_t h = m / 60;
6115 hrtime_t d = h / 24;
6116
6117 s -= m * 60;
6118 m -= h * 60;
6119 h -= d * 24;
6120
6121 timebuf[0] = '\0';
6122
6123 if (d)
6124 (void) sprintf(timebuf,
6125 "%llud%02lluh%02llum%02llus", d, h, m, s);
6126 else if (h)
6127 (void) sprintf(timebuf, "%lluh%02llum%02llus", h, m, s);
6128 else if (m)
6129 (void) sprintf(timebuf, "%llum%02llus", m, s);
6130 else
6131 (void) sprintf(timebuf, "%llus", s);
6132 }
6133
6134 static nvlist_t *
6135 make_random_props()
6136 {
6137 nvlist_t *props;
6138
6139 VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
6140 if (ztest_random(2) == 0)
6141 return (props);
6142 VERIFY(nvlist_add_uint64(props, "autoreplace", 1) == 0);
6143
6144 return (props);
6145 }
6146
6147 /*
6148 * Create a storage pool with the given name and initial vdev size.
6149 * Then test spa_freeze() functionality.
6150 */
6151 static void
6152 ztest_init(ztest_shared_t *zs)
6153 {
6154 spa_t *spa;
6155 nvlist_t *nvroot, *props;
6156
6157 mutex_init(&ztest_vdev_lock, NULL, USYNC_THREAD, NULL);
6158 mutex_init(&ztest_checkpoint_lock, NULL, USYNC_THREAD, NULL);
6159 rw_init(&ztest_name_lock, NULL, USYNC_THREAD, NULL);
6160
6161 kernel_init(FREAD | FWRITE);
6162
6163 /*
6164 * Create the storage pool.
6165 */
6166 (void) spa_destroy(ztest_opts.zo_pool);
6167 ztest_shared->zs_vdev_next_leaf = 0;
6168 zs->zs_splits = 0;
6169 zs->zs_mirrors = ztest_opts.zo_mirrors;
6170 nvroot = make_vdev_root(NULL, NULL, NULL, ztest_opts.zo_vdev_size, 0,
6171 0, ztest_opts.zo_raidz, zs->zs_mirrors, 1);
6172 props = make_random_props();
6173 for (int i = 0; i < SPA_FEATURES; i++) {
6174 char buf[1024];
6175 (void) snprintf(buf, sizeof (buf), "feature@%s",
6176 spa_feature_table[i].fi_uname);
6177 VERIFY3U(0, ==, nvlist_add_uint64(props, buf, 0));
6178 }
6179 VERIFY3U(0, ==, spa_create(ztest_opts.zo_pool, nvroot, props, NULL));
6180 nvlist_free(nvroot);
6181 nvlist_free(props);
6182
6183 VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
6184 zs->zs_metaslab_sz =
6185 1ULL << spa->spa_root_vdev->vdev_child[0]->vdev_ms_shift;
6186
6187 spa_close(spa, FTAG);
6188
6189 kernel_fini();
6190
6191 ztest_run_zdb(ztest_opts.zo_pool);
6192
6193 ztest_freeze();
6194
6195 ztest_run_zdb(ztest_opts.zo_pool);
6196
6197 rw_destroy(&ztest_name_lock);
6198 mutex_destroy(&ztest_vdev_lock);
6199 mutex_destroy(&ztest_checkpoint_lock);
6200 }
6201
6202 static void
6203 setup_data_fd(void)
6204 {
6205 static char ztest_name_data[] = "/tmp/ztest.data.XXXXXX";
6206
6207 ztest_fd_data = mkstemp(ztest_name_data);
6208 ASSERT3S(ztest_fd_data, >=, 0);
6209 (void) unlink(ztest_name_data);
6210 }
6211
6212
6213 static int
6214 shared_data_size(ztest_shared_hdr_t *hdr)
6215 {
6216 int size;
6217
6218 size = hdr->zh_hdr_size;
6219 size += hdr->zh_opts_size;
6220 size += hdr->zh_size;
6221 size += hdr->zh_stats_size * hdr->zh_stats_count;
6222 size += hdr->zh_ds_size * hdr->zh_ds_count;
6223
6224 return (size);
6225 }
6226
6227 static void
6228 setup_hdr(void)
6229 {
6230 int size;
6231 ztest_shared_hdr_t *hdr;
6232
6233 hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
6234 PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
6235 ASSERT(hdr != MAP_FAILED);
6236
6237 VERIFY3U(0, ==, ftruncate(ztest_fd_data, sizeof (ztest_shared_hdr_t)));
6238
6239 hdr->zh_hdr_size = sizeof (ztest_shared_hdr_t);
6240 hdr->zh_opts_size = sizeof (ztest_shared_opts_t);
6241 hdr->zh_size = sizeof (ztest_shared_t);
6242 hdr->zh_stats_size = sizeof (ztest_shared_callstate_t);
6243 hdr->zh_stats_count = ZTEST_FUNCS;
6244 hdr->zh_ds_size = sizeof (ztest_shared_ds_t);
6245 hdr->zh_ds_count = ztest_opts.zo_datasets;
6246
6247 size = shared_data_size(hdr);
6248 VERIFY3U(0, ==, ftruncate(ztest_fd_data, size));
6249
6250 (void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
6251 }
6252
6253 static void
6254 setup_data(void)
6255 {
6256 int size, offset;
6257 ztest_shared_hdr_t *hdr;
6258 uint8_t *buf;
6259
6260 hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
6261 PROT_READ, MAP_SHARED, ztest_fd_data, 0);
6262 ASSERT(hdr != MAP_FAILED);
6263
6264 size = shared_data_size(hdr);
6265
6266 (void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
6267 hdr = ztest_shared_hdr = (void *)mmap(0, P2ROUNDUP(size, getpagesize()),
6268 PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
6269 ASSERT(hdr != MAP_FAILED);
6270 buf = (uint8_t *)hdr;
6271
6272 offset = hdr->zh_hdr_size;
6273 ztest_shared_opts = (void *)&buf[offset];
6274 offset += hdr->zh_opts_size;
6275 ztest_shared = (void *)&buf[offset];
6276 offset += hdr->zh_size;
6277 ztest_shared_callstate = (void *)&buf[offset];
6278 offset += hdr->zh_stats_size * hdr->zh_stats_count;
6279 ztest_shared_ds = (void *)&buf[offset];
6280 }
6281
6282 static boolean_t
6283 exec_child(char *cmd, char *libpath, boolean_t ignorekill, int *statusp)
6284 {
6285 pid_t pid;
6286 int status;
6287 char *cmdbuf = NULL;
6288
6289 pid = fork();
6290
6291 if (cmd == NULL) {
6292 cmdbuf = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
6293 (void) strlcpy(cmdbuf, getexecname(), MAXPATHLEN);
6294 cmd = cmdbuf;
6295 }
6296
6297 if (pid == -1)
6298 fatal(1, "fork failed");
6299
6300 if (pid == 0) { /* child */
6301 char *emptyargv[2] = { cmd, NULL };
6302 char fd_data_str[12];
6303
6304 struct rlimit rl = { 1024, 1024 };
6305 (void) setrlimit(RLIMIT_NOFILE, &rl);
6306
6307 (void) close(ztest_fd_rand);
6308 VERIFY3U(11, >=,
6309 snprintf(fd_data_str, 12, "%d", ztest_fd_data));
6310 VERIFY0(setenv("ZTEST_FD_DATA", fd_data_str, 1));
6311
6312 (void) enable_extended_FILE_stdio(-1, -1);
6313 if (libpath != NULL)
6314 VERIFY(0 == setenv("LD_LIBRARY_PATH", libpath, 1));
6315 (void) execv(cmd, emptyargv);
6316 ztest_dump_core = B_FALSE;
6317 fatal(B_TRUE, "exec failed: %s", cmd);
6318 }
6319
6320 if (cmdbuf != NULL) {
6321 umem_free(cmdbuf, MAXPATHLEN);
6322 cmd = NULL;
6323 }
6324
6325 while (waitpid(pid, &status, 0) != pid)
6326 continue;
6327 if (statusp != NULL)
6328 *statusp = status;
6329
6330 if (WIFEXITED(status)) {
6331 if (WEXITSTATUS(status) != 0) {
6332 (void) fprintf(stderr, "child exited with code %d\n",
6333 WEXITSTATUS(status));
6334 exit(2);
6335 }
6336 return (B_FALSE);
6337 } else if (WIFSIGNALED(status)) {
6338 if (!ignorekill || WTERMSIG(status) != SIGKILL) {
6339 (void) fprintf(stderr, "child died with signal %d\n",
6340 WTERMSIG(status));
6341 exit(3);
6342 }
6343 return (B_TRUE);
6344 } else {
6345 (void) fprintf(stderr, "something strange happened to child\n");
6346 exit(4);
6347 /* NOTREACHED */
6348 }
6349 }
6350
6351 static void
6352 ztest_run_init(void)
6353 {
6354 ztest_shared_t *zs = ztest_shared;
6355
6356 ASSERT(ztest_opts.zo_init != 0);
6357
6358 /*
6359 * Blow away any existing copy of zpool.cache
6360 */
6361 (void) remove(spa_config_path);
6362
6363 /*
6364 * Create and initialize our storage pool.
6365 */
6366 for (int i = 1; i <= ztest_opts.zo_init; i++) {
6367 bzero(zs, sizeof (ztest_shared_t));
6368 if (ztest_opts.zo_verbose >= 3 &&
6369 ztest_opts.zo_init != 1) {
6370 (void) printf("ztest_init(), pass %d\n", i);
6371 }
6372 ztest_init(zs);
6373 }
6374 }
6375
6376 int
6377 main(int argc, char **argv)
6378 {
6379 int kills = 0;
6380 int iters = 0;
6381 int older = 0;
6382 int newer = 0;
6383 ztest_shared_t *zs;
6384 ztest_info_t *zi;
6385 ztest_shared_callstate_t *zc;
6386 char timebuf[100];
6387 char numbuf[NN_NUMBUF_SZ];
6388 spa_t *spa;
6389 char *cmd;
6390 boolean_t hasalt;
6391 char *fd_data_str = getenv("ZTEST_FD_DATA");
6392
6393 (void) setvbuf(stdout, NULL, _IOLBF, 0);
6394
6395 dprintf_setup(&argc, argv);
6396 zfs_deadman_synctime_ms = 300000;
6397
6398 ztest_fd_rand = open("/dev/urandom", O_RDONLY);
6399 ASSERT3S(ztest_fd_rand, >=, 0);
6400
6401 if (!fd_data_str) {
6402 process_options(argc, argv);
6403
6404 setup_data_fd();
6405 setup_hdr();
6406 setup_data();
6407 bcopy(&ztest_opts, ztest_shared_opts,
6408 sizeof (*ztest_shared_opts));
6409 } else {
6410 ztest_fd_data = atoi(fd_data_str);
6411 setup_data();
6412 bcopy(ztest_shared_opts, &ztest_opts, sizeof (ztest_opts));
6413 }
6414 ASSERT3U(ztest_opts.zo_datasets, ==, ztest_shared_hdr->zh_ds_count);
6415
6416 /* Override location of zpool.cache */
6417 VERIFY3U(asprintf((char **)&spa_config_path, "%s/zpool.cache",
6418 ztest_opts.zo_dir), !=, -1);
6419
6420 ztest_ds = umem_alloc(ztest_opts.zo_datasets * sizeof (ztest_ds_t),
6421 UMEM_NOFAIL);
6422 zs = ztest_shared;
6423
6424 if (fd_data_str) {
6425 metaslab_gang_bang = ztest_opts.zo_metaslab_gang_bang;
6426 metaslab_df_alloc_threshold =
6427 zs->zs_metaslab_df_alloc_threshold;
6428
6429 if (zs->zs_do_init)
6430 ztest_run_init();
6431 else
6432 ztest_run(zs);
6433 exit(0);
6434 }
6435
6436 hasalt = (strlen(ztest_opts.zo_alt_ztest) != 0);
6437
6438 if (ztest_opts.zo_verbose >= 1) {
6439 (void) printf("%llu vdevs, %d datasets, %d threads,"
6440 " %llu seconds...\n",
6441 (u_longlong_t)ztest_opts.zo_vdevs,
6442 ztest_opts.zo_datasets,
6443 ztest_opts.zo_threads,
6444 (u_longlong_t)ztest_opts.zo_time);
6445 }
6446
6447 cmd = umem_alloc(MAXNAMELEN, UMEM_NOFAIL);
6448 (void) strlcpy(cmd, getexecname(), MAXNAMELEN);
6449
6450 zs->zs_do_init = B_TRUE;
6451 if (strlen(ztest_opts.zo_alt_ztest) != 0) {
6452 if (ztest_opts.zo_verbose >= 1) {
6453 (void) printf("Executing older ztest for "
6454 "initialization: %s\n", ztest_opts.zo_alt_ztest);
6455 }
6456 VERIFY(!exec_child(ztest_opts.zo_alt_ztest,
6457 ztest_opts.zo_alt_libpath, B_FALSE, NULL));
6458 } else {
6459 VERIFY(!exec_child(NULL, NULL, B_FALSE, NULL));
6460 }
6461 zs->zs_do_init = B_FALSE;
6462
6463 zs->zs_proc_start = gethrtime();
6464 zs->zs_proc_stop = zs->zs_proc_start + ztest_opts.zo_time * NANOSEC;
6465
6466 for (int f = 0; f < ZTEST_FUNCS; f++) {
6467 zi = &ztest_info[f];
6468 zc = ZTEST_GET_SHARED_CALLSTATE(f);
6469 if (zs->zs_proc_start + zi->zi_interval[0] > zs->zs_proc_stop)
6470 zc->zc_next = UINT64_MAX;
6471 else
6472 zc->zc_next = zs->zs_proc_start +
6473 ztest_random(2 * zi->zi_interval[0] + 1);
6474 }
6475
6476 /*
6477 * Run the tests in a loop. These tests include fault injection
6478 * to verify that self-healing data works, and forced crashes
6479 * to verify that we never lose on-disk consistency.
6480 */
6481 while (gethrtime() < zs->zs_proc_stop) {
6482 int status;
6483 boolean_t killed;
6484
6485 /*
6486 * Initialize the workload counters for each function.
6487 */
6488 for (int f = 0; f < ZTEST_FUNCS; f++) {
6489 zc = ZTEST_GET_SHARED_CALLSTATE(f);
6490 zc->zc_count = 0;
6491 zc->zc_time = 0;
6492 }
6493
6494 /* Set the allocation switch size */
6495 zs->zs_metaslab_df_alloc_threshold =
6496 ztest_random(zs->zs_metaslab_sz / 4) + 1;
6497
6498 if (!hasalt || ztest_random(2) == 0) {
6499 if (hasalt && ztest_opts.zo_verbose >= 1) {
6500 (void) printf("Executing newer ztest: %s\n",
6501 cmd);
6502 }
6503 newer++;
6504 killed = exec_child(cmd, NULL, B_TRUE, &status);
6505 } else {
6506 if (hasalt && ztest_opts.zo_verbose >= 1) {
6507 (void) printf("Executing older ztest: %s\n",
6508 ztest_opts.zo_alt_ztest);
6509 }
6510 older++;
6511 killed = exec_child(ztest_opts.zo_alt_ztest,
6512 ztest_opts.zo_alt_libpath, B_TRUE, &status);
6513 }
6514
6515 if (killed)
6516 kills++;
6517 iters++;
6518
6519 if (ztest_opts.zo_verbose >= 1) {
6520 hrtime_t now = gethrtime();
6521
6522 now = MIN(now, zs->zs_proc_stop);
6523 print_time(zs->zs_proc_stop - now, timebuf);
6524 nicenum(zs->zs_space, numbuf, sizeof (numbuf));
6525
6526 (void) printf("Pass %3d, %8s, %3llu ENOSPC, "
6527 "%4.1f%% of %5s used, %3.0f%% done, %8s to go\n",
6528 iters,
6529 WIFEXITED(status) ? "Complete" : "SIGKILL",
6530 (u_longlong_t)zs->zs_enospc_count,
6531 100.0 * zs->zs_alloc / zs->zs_space,
6532 numbuf,
6533 100.0 * (now - zs->zs_proc_start) /
6534 (ztest_opts.zo_time * NANOSEC), timebuf);
6535 }
6536
6537 if (ztest_opts.zo_verbose >= 2) {
6538 (void) printf("\nWorkload summary:\n\n");
6539 (void) printf("%7s %9s %s\n",
6540 "Calls", "Time", "Function");
6541 (void) printf("%7s %9s %s\n",
6542 "-----", "----", "--------");
6543 for (int f = 0; f < ZTEST_FUNCS; f++) {
6544 Dl_info dli;
6545
6546 zi = &ztest_info[f];
6547 zc = ZTEST_GET_SHARED_CALLSTATE(f);
6548 print_time(zc->zc_time, timebuf);
6549 (void) dladdr((void *)zi->zi_func, &dli);
6550 (void) printf("%7llu %9s %s\n",
6551 (u_longlong_t)zc->zc_count, timebuf,
6552 dli.dli_sname);
6553 }
6554 (void) printf("\n");
6555 }
6556
6557 /*
6558 * It's possible that we killed a child during a rename test,
6559 * in which case we'll have a 'ztest_tmp' pool lying around
6560 * instead of 'ztest'. Do a blind rename in case this happened.
6561 */
6562 kernel_init(FREAD);
6563 if (spa_open(ztest_opts.zo_pool, &spa, FTAG) == 0) {
6564 spa_close(spa, FTAG);
6565 } else {
6566 char tmpname[ZFS_MAX_DATASET_NAME_LEN];
6567 kernel_fini();
6568 kernel_init(FREAD | FWRITE);
6569 (void) snprintf(tmpname, sizeof (tmpname), "%s_tmp",
6570 ztest_opts.zo_pool);
6571 (void) spa_rename(tmpname, ztest_opts.zo_pool);
6572 }
6573 kernel_fini();
6574
6575 ztest_run_zdb(ztest_opts.zo_pool);
6576 }
6577
6578 if (ztest_opts.zo_verbose >= 1) {
6579 if (hasalt) {
6580 (void) printf("%d runs of older ztest: %s\n", older,
6581 ztest_opts.zo_alt_ztest);
6582 (void) printf("%d runs of newer ztest: %s\n", newer,
6583 cmd);
6584 }
6585 (void) printf("%d killed, %d completed, %.0f%% kill rate\n",
6586 kills, iters - kills, (100.0 * kills) / MAX(1, iters));
6587 }
6588
6589 umem_free(cmd, MAXNAMELEN);
6590
6591 return (0);
6592 }
Unleashed OS