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