?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
640 number_to_scaled_string is duplicated in several commands
[unleashed.git] / usr / src / cmd / ztest / ztest.c
blob32ecef6b18a1a66aa5e3bd3be79b7ca20c10317d
1 /*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21 /*
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2011, 2016 by Delphix. All rights reserved.
24 * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
25 * Copyright (c) 2013 Steven Hartland. All rights reserved.
26 * Copyright (c) 2014 Integros [integros.com]
27 * Copyright 2017 Joyent, Inc.
28 */
29
30 /*
31 * The objective of this program is to provide a DMU/ZAP/SPA stress test
32 * that runs entirely in userland, is easy to use, and easy to extend.
33 *
34 * The overall design of the ztest program is as follows:
35 *
36 * (1) For each major functional area (e.g. adding vdevs to a pool,
37 * creating and destroying datasets, reading and writing objects, etc)
38 * we have a simple routine to test that functionality. These
39 * individual routines do not have to do anything "stressful".
40 *
41 * (2) We turn these simple functionality tests into a stress test by
42 * running them all in parallel, with as many threads as desired,
43 * and spread across as many datasets, objects, and vdevs as desired.
44 *
45 * (3) While all this is happening, we inject faults into the pool to
46 * verify that self-healing data really works.
47 *
48 * (4) Every time we open a dataset, we change its checksum and compression
49 * functions. Thus even individual objects vary from block to block
50 * in which checksum they use and whether they're compressed.
51 *
52 * (5) To verify that we never lose on-disk consistency after a crash,
53 * we run the entire test in a child of the main process.
54 * At random times, the child self-immolates with a SIGKILL.
55 * This is the software equivalent of pulling the power cord.
56 * The parent then runs the test again, using the existing
57 * storage pool, as many times as desired. If backwards compatibility
58 * testing is enabled ztest will sometimes run the "older" version
59 * of ztest after a SIGKILL.
60 *
61 * (6) To verify that we don't have future leaks or temporal incursions,
62 * many of the functional tests record the transaction group number
63 * as part of their data. When reading old data, they verify that
64 * the transaction group number is less than the current, open txg.
65 * If you add a new test, please do this if applicable.
66 *
67 * When run with no arguments, ztest runs for about five minutes and
68 * produces no output if successful. To get a little bit of information,
69 * specify -V. To get more information, specify -VV, and so on.
70 *
71 * To turn this into an overnight stress test, use -T to specify run time.
72 *
73 * You can ask more more vdevs [-v], datasets [-d], or threads [-t]
74 * to increase the pool capacity, fanout, and overall stress level.
75 *
76 * Use the -k option to set the desired frequency of kills.
77 *
78 * When ztest invokes itself it passes all relevant information through a
79 * temporary file which is mmap-ed in the child process. This allows shared
80 * memory to survive the exec syscall. The ztest_shared_hdr_t struct is always
81 * stored at offset 0 of this file and contains information on the size and
82 * number of shared structures in the file. The information stored in this file
83 * must remain backwards compatible with older versions of ztest so that
84 * ztest can invoke them during backwards compatibility testing (-B).
85 */
86
87 #include <sys/zfs_context.h>
88 #include <sys/spa.h>
89 #include <sys/dmu.h>
90 #include <sys/txg.h>
91 #include <sys/dbuf.h>
92 #include <sys/zap.h>
93 #include <sys/dmu_objset.h>
94 #include <sys/poll.h>
95 #include <sys/stat.h>
96 #include <sys/time.h>
97 #include <sys/wait.h>
98 #include <sys/mman.h>
99 #include <sys/resource.h>
100 #include <sys/zio.h>
101 #include <sys/zil.h>
102 #include <sys/zil_impl.h>
103 #include <sys/vdev_impl.h>
104 #include <sys/vdev_file.h>
105 #include <sys/spa_impl.h>
106 #include <sys/metaslab_impl.h>
107 #include <sys/dsl_prop.h>
108 #include <sys/dsl_dataset.h>
109 #include <sys/dsl_destroy.h>
110 #include <sys/dsl_scan.h>
111 #include <sys/zio_checksum.h>
112 #include <sys/refcount.h>
113 #include <sys/zfeature.h>
114 #include <sys/dsl_userhold.h>
115 #include <sys/abd.h>
116 #include <stdio.h>
117 #include <stdio_ext.h>
118 #include <stdlib.h>
119 #include <unistd.h>
120 #include <signal.h>
121 #include <umem.h>
122 #include <dlfcn.h>
123 #include <ctype.h>
124 #include <math.h>
125 #include <sys/fs/zfs.h>
126 #include <libnvpair.h>
127 #include <libcmdutils.h>
128
129 static int ztest_fd_data = -1;
130 static int ztest_fd_rand = -1;
131
132 typedef struct ztest_shared_hdr {
133 uint64_t zh_hdr_size;
134 uint64_t zh_opts_size;
135 uint64_t zh_size;
136 uint64_t zh_stats_size;
137 uint64_t zh_stats_count;
138 uint64_t zh_ds_size;
139 uint64_t zh_ds_count;
140 } ztest_shared_hdr_t;
141
142 static ztest_shared_hdr_t *ztest_shared_hdr;
143
144 typedef struct ztest_shared_opts {
145 char zo_pool[ZFS_MAX_DATASET_NAME_LEN];
146 char zo_dir[ZFS_MAX_DATASET_NAME_LEN];
147 char zo_alt_ztest[MAXNAMELEN];
148 char zo_alt_libpath[MAXNAMELEN];
149 uint64_t zo_vdevs;
150 uint64_t zo_vdevtime;
151 size_t zo_vdev_size;
152 int zo_ashift;
153 int zo_mirrors;
154 int zo_raidz;
155 int zo_raidz_parity;
156 int zo_datasets;
157 int zo_threads;
158 uint64_t zo_passtime;
159 uint64_t zo_killrate;
160 int zo_verbose;
161 int zo_init;
162 uint64_t zo_time;
163 uint64_t zo_maxloops;
164 uint64_t zo_metaslab_gang_bang;
165 } ztest_shared_opts_t;
166
167 static const ztest_shared_opts_t ztest_opts_defaults = {
168 .zo_pool = { 'z', 't', 'e', 's', 't', '\0' },
169 .zo_dir = { '/', 't', 'm', 'p', '\0' },
170 .zo_alt_ztest = { '\0' },
171 .zo_alt_libpath = { '\0' },
172 .zo_vdevs = 5,
173 .zo_ashift = SPA_MINBLOCKSHIFT,
174 .zo_mirrors = 2,
175 .zo_raidz = 4,
176 .zo_raidz_parity = 1,
177 .zo_vdev_size = SPA_MINDEVSIZE * 4, /* 256m default size */
178 .zo_datasets = 7,
179 .zo_threads = 23,
180 .zo_passtime = 60, /* 60 seconds */
181 .zo_killrate = 70, /* 70% kill rate */
182 .zo_verbose = 0,
183 .zo_init = 1,
184 .zo_time = 300, /* 5 minutes */
185 .zo_maxloops = 50, /* max loops during spa_freeze() */
186 .zo_metaslab_gang_bang = 32 << 10
187 };
188
189 extern uint64_t metaslab_gang_bang;
190 extern uint64_t metaslab_df_alloc_threshold;
191 extern uint64_t zfs_deadman_synctime_ms;
192 extern int metaslab_preload_limit;
193 extern boolean_t zfs_compressed_arc_enabled;
194 extern boolean_t zfs_abd_scatter_enabled;
195
196 static ztest_shared_opts_t *ztest_shared_opts;
197 static ztest_shared_opts_t ztest_opts;
198
199 typedef struct ztest_shared_ds {
200 uint64_t zd_seq;
201 } ztest_shared_ds_t;
202
203 static ztest_shared_ds_t *ztest_shared_ds;
204 #define ZTEST_GET_SHARED_DS(d) (&ztest_shared_ds[d])
205
206 #define BT_MAGIC 0x123456789abcdefULL
207 #define MAXFAULTS() \
208 (MAX(zs->zs_mirrors, 1) * (ztest_opts.zo_raidz_parity + 1) - 1)
209
210 enum ztest_io_type {
211 ZTEST_IO_WRITE_TAG,
212 ZTEST_IO_WRITE_PATTERN,
213 ZTEST_IO_WRITE_ZEROES,
214 ZTEST_IO_TRUNCATE,
215 ZTEST_IO_SETATTR,
216 ZTEST_IO_REWRITE,
217 ZTEST_IO_TYPES
218 };
219
220 typedef struct ztest_block_tag {
221 uint64_t bt_magic;
222 uint64_t bt_objset;
223 uint64_t bt_object;
224 uint64_t bt_offset;
225 uint64_t bt_gen;
226 uint64_t bt_txg;
227 uint64_t bt_crtxg;
228 } ztest_block_tag_t;
229
230 typedef struct bufwad {
231 uint64_t bw_index;
232 uint64_t bw_txg;
233 uint64_t bw_data;
234 } bufwad_t;
235
236 /*
237 * XXX -- fix zfs range locks to be generic so we can use them here.
238 */
239 typedef enum {
240 RL_READER,
241 RL_WRITER,
242 RL_APPEND
243 } rl_type_t;
244
245 typedef struct rll {
246 void *rll_writer;
247 int rll_readers;
248 mutex_t rll_lock;
249 cond_t rll_cv;
250 } rll_t;
251
252 typedef struct rl {
253 uint64_t rl_object;
254 uint64_t rl_offset;
255 uint64_t rl_size;
256 rll_t *rl_lock;
257 } rl_t;
258
259 #define ZTEST_RANGE_LOCKS 64
260 #define ZTEST_OBJECT_LOCKS 64
261
262 /*
263 * Object descriptor. Used as a template for object lookup/create/remove.
264 */
265 typedef struct ztest_od {
266 uint64_t od_dir;
267 uint64_t od_object;
268 dmu_object_type_t od_type;
269 dmu_object_type_t od_crtype;
270 uint64_t od_blocksize;
271 uint64_t od_crblocksize;
272 uint64_t od_gen;
273 uint64_t od_crgen;
274 char od_name[ZFS_MAX_DATASET_NAME_LEN];
275 } ztest_od_t;
276
277 /*
278 * Per-dataset state.
279 */
280 typedef struct ztest_ds {
281 ztest_shared_ds_t *zd_shared;
282 objset_t *zd_os;
283 rwlock_t zd_zilog_lock;
284 zilog_t *zd_zilog;
285 ztest_od_t *zd_od; /* debugging aid */
286 char zd_name[ZFS_MAX_DATASET_NAME_LEN];
287 mutex_t zd_dirobj_lock;
288 rll_t zd_object_lock[ZTEST_OBJECT_LOCKS];
289 rll_t zd_range_lock[ZTEST_RANGE_LOCKS];
290 } ztest_ds_t;
291
292 /*
293 * Per-iteration state.
294 */
295 typedef void ztest_func_t(ztest_ds_t *zd, uint64_t id);
296
297 typedef struct ztest_info {
298 ztest_func_t *zi_func; /* test function */
299 uint64_t zi_iters; /* iterations per execution */
300 uint64_t *zi_interval; /* execute every <interval> seconds */
301 } ztest_info_t;
302
303 typedef struct ztest_shared_callstate {
304 uint64_t zc_count; /* per-pass count */
305 uint64_t zc_time; /* per-pass time */
306 uint64_t zc_next; /* next time to call this function */
307 } ztest_shared_callstate_t;
308
309 static ztest_shared_callstate_t *ztest_shared_callstate;
310 #define ZTEST_GET_SHARED_CALLSTATE(c) (&ztest_shared_callstate[c])
311
312 /*
313 * Note: these aren't static because we want dladdr() to work.
314 */
315 ztest_func_t ztest_dmu_read_write;
316 ztest_func_t ztest_dmu_write_parallel;
317 ztest_func_t ztest_dmu_object_alloc_free;
318 ztest_func_t ztest_dmu_commit_callbacks;
319 ztest_func_t ztest_zap;
320 ztest_func_t ztest_zap_parallel;
321 ztest_func_t ztest_zil_commit;
322 ztest_func_t ztest_zil_remount;
323 ztest_func_t ztest_dmu_read_write_zcopy;
324 ztest_func_t ztest_dmu_objset_create_destroy;
325 ztest_func_t ztest_dmu_prealloc;
326 ztest_func_t ztest_fzap;
327 ztest_func_t ztest_dmu_snapshot_create_destroy;
328 ztest_func_t ztest_dsl_prop_get_set;
329 ztest_func_t ztest_spa_prop_get_set;
330 ztest_func_t ztest_spa_create_destroy;
331 ztest_func_t ztest_fault_inject;
332 ztest_func_t ztest_ddt_repair;
333 ztest_func_t ztest_dmu_snapshot_hold;
334 ztest_func_t ztest_spa_rename;
335 ztest_func_t ztest_scrub;
336 ztest_func_t ztest_dsl_dataset_promote_busy;
337 ztest_func_t ztest_vdev_attach_detach;
338 ztest_func_t ztest_vdev_LUN_growth;
339 ztest_func_t ztest_vdev_add_remove;
340 ztest_func_t ztest_vdev_aux_add_remove;
341 ztest_func_t ztest_split_pool;
342 ztest_func_t ztest_reguid;
343 ztest_func_t ztest_spa_upgrade;
344
345 uint64_t zopt_always = 0ULL * NANOSEC; /* all the time */
346 uint64_t zopt_incessant = 1ULL * NANOSEC / 10; /* every 1/10 second */
347 uint64_t zopt_often = 1ULL * NANOSEC; /* every second */
348 uint64_t zopt_sometimes = 10ULL * NANOSEC; /* every 10 seconds */
349 uint64_t zopt_rarely = 60ULL * NANOSEC; /* every 60 seconds */
350
351 ztest_info_t ztest_info[] = {
352 { ztest_dmu_read_write, 1, &zopt_always },
353 { ztest_dmu_write_parallel, 10, &zopt_always },
354 { ztest_dmu_object_alloc_free, 1, &zopt_always },
355 { ztest_dmu_commit_callbacks, 1, &zopt_always },
356 { ztest_zap, 30, &zopt_always },
357 { ztest_zap_parallel, 100, &zopt_always },
358 { ztest_split_pool, 1, &zopt_always },
359 { ztest_zil_commit, 1, &zopt_incessant },
360 { ztest_zil_remount, 1, &zopt_sometimes },
361 { ztest_dmu_read_write_zcopy, 1, &zopt_often },
362 { ztest_dmu_objset_create_destroy, 1, &zopt_often },
363 { ztest_dsl_prop_get_set, 1, &zopt_often },
364 { ztest_spa_prop_get_set, 1, &zopt_sometimes },
365 #if 0
366 { ztest_dmu_prealloc, 1, &zopt_sometimes },
367 #endif
368 { ztest_fzap, 1, &zopt_sometimes },
369 { ztest_dmu_snapshot_create_destroy, 1, &zopt_sometimes },
370 { ztest_spa_create_destroy, 1, &zopt_sometimes },
371 { ztest_fault_inject, 1, &zopt_sometimes },
372 { ztest_ddt_repair, 1, &zopt_sometimes },
373 { ztest_dmu_snapshot_hold, 1, &zopt_sometimes },
374 { ztest_reguid, 1, &zopt_rarely },
375 { ztest_spa_rename, 1, &zopt_rarely },
376 { ztest_scrub, 1, &zopt_rarely },
377 { ztest_spa_upgrade, 1, &zopt_rarely },
378 { ztest_dsl_dataset_promote_busy, 1, &zopt_rarely },
379 { ztest_vdev_attach_detach, 1, &zopt_sometimes },
380 { ztest_vdev_LUN_growth, 1, &zopt_rarely },
381 { ztest_vdev_add_remove, 1,
382 &ztest_opts.zo_vdevtime },
383 { ztest_vdev_aux_add_remove, 1,
384 &ztest_opts.zo_vdevtime },
385 };
386
387 #define ZTEST_FUNCS (sizeof (ztest_info) / sizeof (ztest_info_t))
388
389 /*
390 * The following struct is used to hold a list of uncalled commit callbacks.
391 * The callbacks are ordered by txg number.
392 */
393 typedef struct ztest_cb_list {
394 mutex_t zcl_callbacks_lock;
395 list_t zcl_callbacks;
396 } ztest_cb_list_t;
397
398 /*
399 * Stuff we need to share writably between parent and child.
400 */
401 typedef struct ztest_shared {
402 boolean_t zs_do_init;
403 hrtime_t zs_proc_start;
404 hrtime_t zs_proc_stop;
405 hrtime_t zs_thread_start;
406 hrtime_t zs_thread_stop;
407 hrtime_t zs_thread_kill;
408 uint64_t zs_enospc_count;
409 uint64_t zs_vdev_next_leaf;
410 uint64_t zs_vdev_aux;
411 uint64_t zs_alloc;
412 uint64_t zs_space;
413 uint64_t zs_splits;
414 uint64_t zs_mirrors;
415 uint64_t zs_metaslab_sz;
416 uint64_t zs_metaslab_df_alloc_threshold;
417 uint64_t zs_guid;
418 } ztest_shared_t;
419
420 #define ID_PARALLEL -1ULL
421
422 static char ztest_dev_template[] = "%s/%s.%llua";
423 static char ztest_aux_template[] = "%s/%s.%s.%llu";
424 ztest_shared_t *ztest_shared;
425
426 static spa_t *ztest_spa = NULL;
427 static ztest_ds_t *ztest_ds;
428
429 static mutex_t ztest_vdev_lock;
430
431 /*
432 * The ztest_name_lock protects the pool and dataset namespace used by
433 * the individual tests. To modify the namespace, consumers must grab
434 * this lock as writer. Grabbing the lock as reader will ensure that the
435 * namespace does not change while the lock is held.
436 */
437 static rwlock_t ztest_name_lock;
438
439 static boolean_t ztest_dump_core = B_TRUE;
440 static boolean_t ztest_exiting;
441
442 /* Global commit callback list */
443 static ztest_cb_list_t zcl;
444
445 enum ztest_object {
446 ZTEST_META_DNODE = 0,
447 ZTEST_DIROBJ,
448 ZTEST_OBJECTS
449 };
450
451 static void usage(boolean_t) __NORETURN;
452
453 /*
454 * These libumem hooks provide a reasonable set of defaults for the allocator's
455 * debugging facilities.
456 */
457 const char *
458 _umem_debug_init()
459 {
460 return ("default,verbose"); /* $UMEM_DEBUG setting */
461 }
462
463 const char *
464 _umem_logging_init(void)
465 {
466 return ("fail,contents"); /* $UMEM_LOGGING setting */
467 }
468
469 #define FATAL_MSG_SZ 1024
470
471 char *fatal_msg;
472
473 static void
474 fatal(int do_perror, char *message, ...)
475 {
476 va_list args;
477 int save_errno = errno;
478 char buf[FATAL_MSG_SZ];
479
480 (void) fflush(stdout);
481
482 va_start(args, message);
483 (void) sprintf(buf, "ztest: ");
484 /* LINTED */
485 (void) vsprintf(buf + strlen(buf), message, args);
486 va_end(args);
487 if (do_perror) {
488 (void) snprintf(buf + strlen(buf), FATAL_MSG_SZ - strlen(buf),
489 ": %s", strerror(save_errno));
490 }
491 (void) fprintf(stderr, "%s\n", buf);
492 fatal_msg = buf; /* to ease debugging */
493 if (ztest_dump_core)
494 abort();
495 exit(3);
496 }
497
498 static int
499 str2shift(const char *buf)
500 {
501 const char *ends = "BKMGTPEZ";
502 int i;
503
504 if (buf[0] == '\0')
505 return (0);
506 for (i = 0; i < strlen(ends); i++) {
507 if (toupper(buf[0]) == ends[i])
508 break;
509 }
510 if (i == strlen(ends)) {
511 (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n",
512 buf);
513 usage(B_FALSE);
514 }
515 if (buf[1] == '\0' || (toupper(buf[1]) == 'B' && buf[2] == '\0')) {
516 return (10*i);
517 }
518 (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n", buf);
519 usage(B_FALSE);
520 /* NOTREACHED */
521 }
522
523 static uint64_t
524 nicenumtoull(const char *buf)
525 {
526 char *end;
527 uint64_t val;
528
529 val = strtoull(buf, &end, 0);
530 if (end == buf) {
531 (void) fprintf(stderr, "ztest: bad numeric value: %s\n", buf);
532 usage(B_FALSE);
533 } else if (end[0] == '.') {
534 double fval = strtod(buf, &end);
535 fval *= pow(2, str2shift(end));
536 if (fval > UINT64_MAX) {
537 (void) fprintf(stderr, "ztest: value too large: %s\n",
538 buf);
539 usage(B_FALSE);
540 }
541 val = (uint64_t)fval;
542 } else {
543 int shift = str2shift(end);
544 if (shift >= 64 || (val << shift) >> shift != val) {
545 (void) fprintf(stderr, "ztest: value too large: %s\n",
546 buf);
547 usage(B_FALSE);
548 }
549 val <<= shift;
550 }
551 return (val);
552 }
553
554 static void
555 usage(boolean_t requested)
556 {
557 const ztest_shared_opts_t *zo = &ztest_opts_defaults;
558
559 char nice_vdev_size[NN_NUMBUF_SZ];
560 char nice_gang_bang[NN_NUMBUF_SZ];
561 FILE *fp = requested ? stdout : stderr;
562
563 nicenum(zo->zo_vdev_size, nice_vdev_size, sizeof (nice_vdev_size));
564 nicenum(zo->zo_metaslab_gang_bang, nice_gang_bang,
565 sizeof (nice_gang_bang));
566
567 (void) fprintf(fp, "Usage: %s\n"
568 "\t[-v vdevs (default: %llu)]\n"
569 "\t[-s size_of_each_vdev (default: %s)]\n"
570 "\t[-a alignment_shift (default: %d)] use 0 for random\n"
571 "\t[-m mirror_copies (default: %d)]\n"
572 "\t[-r raidz_disks (default: %d)]\n"
573 "\t[-R raidz_parity (default: %d)]\n"
574 "\t[-d datasets (default: %d)]\n"
575 "\t[-t threads (default: %d)]\n"
576 "\t[-g gang_block_threshold (default: %s)]\n"
577 "\t[-i init_count (default: %d)] initialize pool i times\n"
578 "\t[-k kill_percentage (default: %llu%%)]\n"
579 "\t[-p pool_name (default: %s)]\n"
580 "\t[-f dir (default: %s)] file directory for vdev files\n"
581 "\t[-V] verbose (use multiple times for ever more blather)\n"
582 "\t[-E] use existing pool instead of creating new one\n"
583 "\t[-T time (default: %llu sec)] total run time\n"
584 "\t[-F freezeloops (default: %llu)] max loops in spa_freeze()\n"
585 "\t[-P passtime (default: %llu sec)] time per pass\n"
586 "\t[-B alt_ztest (default: <none>)] alternate ztest path\n"
587 "\t[-o variable=value] ... set global variable to an unsigned\n"
588 "\t 32-bit integer value\n"
589 "\t[-h] (print help)\n"
590 "",
591 zo->zo_pool,
592 (u_longlong_t)zo->zo_vdevs, /* -v */
593 nice_vdev_size, /* -s */
594 zo->zo_ashift, /* -a */
595 zo->zo_mirrors, /* -m */
596 zo->zo_raidz, /* -r */
597 zo->zo_raidz_parity, /* -R */
598 zo->zo_datasets, /* -d */
599 zo->zo_threads, /* -t */
600 nice_gang_bang, /* -g */
601 zo->zo_init, /* -i */
602 (u_longlong_t)zo->zo_killrate, /* -k */
603 zo->zo_pool, /* -p */
604 zo->zo_dir, /* -f */
605 (u_longlong_t)zo->zo_time, /* -T */
606 (u_longlong_t)zo->zo_maxloops, /* -F */
607 (u_longlong_t)zo->zo_passtime);
608 exit(requested ? 0 : 1);
609 }
610
611 static void
612 process_options(int argc, char **argv)
613 {
614 char *path;
615 ztest_shared_opts_t *zo = &ztest_opts;
616
617 int opt;
618 uint64_t value;
619 char altdir[MAXNAMELEN] = { 0 };
620
621 bcopy(&ztest_opts_defaults, zo, sizeof (*zo));
622
623 while ((opt = getopt(argc, argv,
624 "v:s:a:m:r:R:d:t:g:i:k:p:f:VET:P:hF:B:o:")) != EOF) {
625 value = 0;
626 switch (opt) {
627 case 'v':
628 case 's':
629 case 'a':
630 case 'm':
631 case 'r':
632 case 'R':
633 case 'd':
634 case 't':
635 case 'g':
636 case 'i':
637 case 'k':
638 case 'T':
639 case 'P':
640 case 'F':
641 value = nicenumtoull(optarg);
642 }
643 switch (opt) {
644 case 'v':
645 zo->zo_vdevs = value;
646 break;
647 case 's':
648 zo->zo_vdev_size = MAX(SPA_MINDEVSIZE, value);
649 break;
650 case 'a':
651 zo->zo_ashift = value;
652 break;
653 case 'm':
654 zo->zo_mirrors = value;
655 break;
656 case 'r':
657 zo->zo_raidz = MAX(1, value);
658 break;
659 case 'R':
660 zo->zo_raidz_parity = MIN(MAX(value, 1), 3);
661 break;
662 case 'd':
663 zo->zo_datasets = MAX(1, value);
664 break;
665 case 't':
666 zo->zo_threads = MAX(1, value);
667 break;
668 case 'g':
669 zo->zo_metaslab_gang_bang = MAX(SPA_MINBLOCKSIZE << 1,
670 value);
671 break;
672 case 'i':
673 zo->zo_init = value;
674 break;
675 case 'k':
676 zo->zo_killrate = value;
677 break;
678 case 'p':
679 (void) strlcpy(zo->zo_pool, optarg,
680 sizeof (zo->zo_pool));
681 break;
682 case 'f':
683 path = realpath(optarg, NULL);
684 if (path == NULL) {
685 (void) fprintf(stderr, "error: %s: %s\n",
686 optarg, strerror(errno));
687 usage(B_FALSE);
688 } else {
689 (void) strlcpy(zo->zo_dir, path,
690 sizeof (zo->zo_dir));
691 }
692 break;
693 case 'V':
694 zo->zo_verbose++;
695 break;
696 case 'E':
697 zo->zo_init = 0;
698 break;
699 case 'T':
700 zo->zo_time = value;
701 break;
702 case 'P':
703 zo->zo_passtime = MAX(1, value);
704 break;
705 case 'F':
706 zo->zo_maxloops = MAX(1, value);
707 break;
708 case 'B':
709 (void) strlcpy(altdir, optarg, sizeof (altdir));
710 break;
711 case 'o':
712 if (set_global_var(optarg) != 0)
713 usage(B_FALSE);
714 break;
715 case 'h':
716 usage(B_TRUE);
717 break;
718 case '?':
719 default:
720 usage(B_FALSE);
721 break;
722 }
723 }
724
725 zo->zo_raidz_parity = MIN(zo->zo_raidz_parity, zo->zo_raidz - 1);
726
727 zo->zo_vdevtime =
728 (zo->zo_vdevs > 0 ? zo->zo_time * NANOSEC / zo->zo_vdevs :
729 UINT64_MAX >> 2);
730
731 if (strlen(altdir) > 0) {
732 char *cmd;
733 char *realaltdir;
734 char *bin;
735 char *ztest;
736 char *isa;
737 int isalen;
738
739 cmd = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
740 realaltdir = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
741
742 VERIFY(NULL != realpath(getexecname(), cmd));
743 if (0 != access(altdir, F_OK)) {
744 ztest_dump_core = B_FALSE;
745 fatal(B_TRUE, "invalid alternate ztest path: %s",
746 altdir);
747 }
748 VERIFY(NULL != realpath(altdir, realaltdir));
749
750 /*
751 * 'cmd' should be of the form "<anything>/usr/bin/<isa>/ztest".
752 * We want to extract <isa> to determine if we should use
753 * 32 or 64 bit binaries.
754 */
755 bin = strstr(cmd, "/usr/bin/");
756 ztest = strstr(bin, "/ztest");
757 isa = bin + 9;
758 isalen = ztest - isa;
759 (void) snprintf(zo->zo_alt_ztest, sizeof (zo->zo_alt_ztest),
760 "%s/usr/bin/%.*s/ztest", realaltdir, isalen, isa);
761 (void) snprintf(zo->zo_alt_libpath, sizeof (zo->zo_alt_libpath),
762 "%s/usr/lib/%.*s", realaltdir, isalen, isa);
763
764 if (0 != access(zo->zo_alt_ztest, X_OK)) {
765 ztest_dump_core = B_FALSE;
766 fatal(B_TRUE, "invalid alternate ztest: %s",
767 zo->zo_alt_ztest);
768 } else if (0 != access(zo->zo_alt_libpath, X_OK)) {
769 ztest_dump_core = B_FALSE;
770 fatal(B_TRUE, "invalid alternate lib directory %s",
771 zo->zo_alt_libpath);
772 }
773
774 umem_free(cmd, MAXPATHLEN);
775 umem_free(realaltdir, MAXPATHLEN);
776 }
777 }
778
779 static void
780 ztest_kill(ztest_shared_t *zs)
781 {
782 zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(ztest_spa));
783 zs->zs_space = metaslab_class_get_space(spa_normal_class(ztest_spa));
784
785 /*
786 * Before we kill off ztest, make sure that the config is updated.
787 * See comment above spa_config_sync().
788 */
789 mutex_enter(&spa_namespace_lock);
790 spa_config_sync(ztest_spa, B_FALSE, B_FALSE);
791 mutex_exit(&spa_namespace_lock);
792
793 zfs_dbgmsg_print(FTAG);
794 (void) kill(getpid(), SIGKILL);
795 }
796
797 static uint64_t
798 ztest_random(uint64_t range)
799 {
800 uint64_t r;
801
802 ASSERT3S(ztest_fd_rand, >=, 0);
803
804 if (range == 0)
805 return (0);
806
807 if (read(ztest_fd_rand, &r, sizeof (r)) != sizeof (r))
808 fatal(1, "short read from /dev/urandom");
809
810 return (r % range);
811 }
812
813 /* ARGSUSED */
814 static void
815 ztest_record_enospc(const char *s)
816 {
817 ztest_shared->zs_enospc_count++;
818 }
819
820 static uint64_t
821 ztest_get_ashift(void)
822 {
823 if (ztest_opts.zo_ashift == 0)
824 return (SPA_MINBLOCKSHIFT + ztest_random(5));
825 return (ztest_opts.zo_ashift);
826 }
827
828 static nvlist_t *
829 make_vdev_file(char *path, char *aux, char *pool, size_t size, uint64_t ashift)
830 {
831 char pathbuf[MAXPATHLEN];
832 uint64_t vdev;
833 nvlist_t *file;
834
835 if (ashift == 0)
836 ashift = ztest_get_ashift();
837
838 if (path == NULL) {
839 path = pathbuf;
840
841 if (aux != NULL) {
842 vdev = ztest_shared->zs_vdev_aux;
843 (void) snprintf(path, sizeof (pathbuf),
844 ztest_aux_template, ztest_opts.zo_dir,
845 pool == NULL ? ztest_opts.zo_pool : pool,
846 aux, vdev);
847 } else {
848 vdev = ztest_shared->zs_vdev_next_leaf++;
849 (void) snprintf(path, sizeof (pathbuf),
850 ztest_dev_template, ztest_opts.zo_dir,
851 pool == NULL ? ztest_opts.zo_pool : pool, vdev);
852 }
853 }
854
855 if (size != 0) {
856 int fd = open(path, O_RDWR | O_CREAT | O_TRUNC, 0666);
857 if (fd == -1)
858 fatal(1, "can't open %s", path);
859 if (ftruncate(fd, size) != 0)
860 fatal(1, "can't ftruncate %s", path);
861 (void) close(fd);
862 }
863
864 VERIFY(nvlist_alloc(&file, NV_UNIQUE_NAME, 0) == 0);
865 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_TYPE, VDEV_TYPE_FILE) == 0);
866 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_PATH, path) == 0);
867 VERIFY(nvlist_add_uint64(file, ZPOOL_CONFIG_ASHIFT, ashift) == 0);
868
869 return (file);
870 }
871
872 static nvlist_t *
873 make_vdev_raidz(char *path, char *aux, char *pool, size_t size,
874 uint64_t ashift, int r)
875 {
876 nvlist_t *raidz, **child;
877 int c;
878
879 if (r < 2)
880 return (make_vdev_file(path, aux, pool, size, ashift));
881 child = umem_alloc(r * sizeof (nvlist_t *), UMEM_NOFAIL);
882
883 for (c = 0; c < r; c++)
884 child[c] = make_vdev_file(path, aux, pool, size, ashift);
885
886 VERIFY(nvlist_alloc(&raidz, NV_UNIQUE_NAME, 0) == 0);
887 VERIFY(nvlist_add_string(raidz, ZPOOL_CONFIG_TYPE,
888 VDEV_TYPE_RAIDZ) == 0);
889 VERIFY(nvlist_add_uint64(raidz, ZPOOL_CONFIG_NPARITY,
890 ztest_opts.zo_raidz_parity) == 0);
891 VERIFY(nvlist_add_nvlist_array(raidz, ZPOOL_CONFIG_CHILDREN,
892 child, r) == 0);
893
894 for (c = 0; c < r; c++)
895 nvlist_free(child[c]);
896
897 umem_free(child, r * sizeof (nvlist_t *));
898
899 return (raidz);
900 }
901
902 static nvlist_t *
903 make_vdev_mirror(char *path, char *aux, char *pool, size_t size,
904 uint64_t ashift, int r, int m)
905 {
906 nvlist_t *mirror, **child;
907 int c;
908
909 if (m < 1)
910 return (make_vdev_raidz(path, aux, pool, size, ashift, r));
911
912 child = umem_alloc(m * sizeof (nvlist_t *), UMEM_NOFAIL);
913
914 for (c = 0; c < m; c++)
915 child[c] = make_vdev_raidz(path, aux, pool, size, ashift, r);
916
917 VERIFY(nvlist_alloc(&mirror, NV_UNIQUE_NAME, 0) == 0);
918 VERIFY(nvlist_add_string(mirror, ZPOOL_CONFIG_TYPE,
919 VDEV_TYPE_MIRROR) == 0);
920 VERIFY(nvlist_add_nvlist_array(mirror, ZPOOL_CONFIG_CHILDREN,
921 child, m) == 0);
922
923 for (c = 0; c < m; c++)
924 nvlist_free(child[c]);
925
926 umem_free(child, m * sizeof (nvlist_t *));
927
928 return (mirror);
929 }
930
931 static nvlist_t *
932 make_vdev_root(char *path, char *aux, char *pool, size_t size, uint64_t ashift,
933 int log, int r, int m, int t)
934 {
935 nvlist_t *root, **child;
936 int c;
937
938 ASSERT(t > 0);
939
940 child = umem_alloc(t * sizeof (nvlist_t *), UMEM_NOFAIL);
941
942 for (c = 0; c < t; c++) {
943 child[c] = make_vdev_mirror(path, aux, pool, size, ashift,
944 r, m);
945 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_IS_LOG,
946 log) == 0);
947 }
948
949 VERIFY(nvlist_alloc(&root, NV_UNIQUE_NAME, 0) == 0);
950 VERIFY(nvlist_add_string(root, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) == 0);
951 VERIFY(nvlist_add_nvlist_array(root, aux ? aux : ZPOOL_CONFIG_CHILDREN,
952 child, t) == 0);
953
954 for (c = 0; c < t; c++)
955 nvlist_free(child[c]);
956
957 umem_free(child, t * sizeof (nvlist_t *));
958
959 return (root);
960 }
961
962 /*
963 * Find a random spa version. Returns back a random spa version in the
964 * range [initial_version, SPA_VERSION_FEATURES].
965 */
966 static uint64_t
967 ztest_random_spa_version(uint64_t initial_version)
968 {
969 uint64_t version = initial_version;
970
971 if (version <= SPA_VERSION_BEFORE_FEATURES) {
972 version = version +
973 ztest_random(SPA_VERSION_BEFORE_FEATURES - version + 1);
974 }
975
976 if (version > SPA_VERSION_BEFORE_FEATURES)
977 version = SPA_VERSION_FEATURES;
978
979 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
980 return (version);
981 }
982
983 static int
984 ztest_random_blocksize(void)
985 {
986 uint64_t block_shift;
987 /*
988 * Choose a block size >= the ashift.
989 * If the SPA supports new MAXBLOCKSIZE, test up to 1MB blocks.
990 */
991 int maxbs = SPA_OLD_MAXBLOCKSHIFT;
992 if (spa_maxblocksize(ztest_spa) == SPA_MAXBLOCKSIZE)
993 maxbs = 20;
994 block_shift = ztest_random(maxbs - ztest_spa->spa_max_ashift + 1);
995 return (1 << (SPA_MINBLOCKSHIFT + block_shift));
996 }
997
998 static int
999 ztest_random_ibshift(void)
1000 {
1001 return (DN_MIN_INDBLKSHIFT +
1002 ztest_random(DN_MAX_INDBLKSHIFT - DN_MIN_INDBLKSHIFT + 1));
1003 }
1004
1005 static uint64_t
1006 ztest_random_vdev_top(spa_t *spa, boolean_t log_ok)
1007 {
1008 uint64_t top;
1009 vdev_t *rvd = spa->spa_root_vdev;
1010 vdev_t *tvd;
1011
1012 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1013
1014 do {
1015 top = ztest_random(rvd->vdev_children);
1016 tvd = rvd->vdev_child[top];
1017 } while (tvd->vdev_ishole || (tvd->vdev_islog && !log_ok) ||
1018 tvd->vdev_mg == NULL || tvd->vdev_mg->mg_class == NULL);
1019
1020 return (top);
1021 }
1022
1023 static uint64_t
1024 ztest_random_dsl_prop(zfs_prop_t prop)
1025 {
1026 uint64_t value;
1027
1028 do {
1029 value = zfs_prop_random_value(prop, ztest_random(-1ULL));
1030 } while (prop == ZFS_PROP_CHECKSUM && value == ZIO_CHECKSUM_OFF);
1031
1032 return (value);
1033 }
1034
1035 static int
1036 ztest_dsl_prop_set_uint64(char *osname, zfs_prop_t prop, uint64_t value,
1037 boolean_t inherit)
1038 {
1039 const char *propname = zfs_prop_to_name(prop);
1040 const char *valname;
1041 char setpoint[MAXPATHLEN];
1042 uint64_t curval;
1043 int error;
1044
1045 error = dsl_prop_set_int(osname, propname,
1046 (inherit ? ZPROP_SRC_NONE : ZPROP_SRC_LOCAL), value);
1047
1048 if (error == ENOSPC) {
1049 ztest_record_enospc(FTAG);
1050 return (error);
1051 }
1052 ASSERT0(error);
1053
1054 VERIFY0(dsl_prop_get_integer(osname, propname, &curval, setpoint));
1055
1056 if (ztest_opts.zo_verbose >= 6) {
1057 VERIFY(zfs_prop_index_to_string(prop, curval, &valname) == 0);
1058 (void) printf("%s %s = %s at '%s'\n",
1059 osname, propname, valname, setpoint);
1060 }
1061
1062 return (error);
1063 }
1064
1065 static int
1066 ztest_spa_prop_set_uint64(zpool_prop_t prop, uint64_t value)
1067 {
1068 spa_t *spa = ztest_spa;
1069 nvlist_t *props = NULL;
1070 int error;
1071
1072 VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
1073 VERIFY(nvlist_add_uint64(props, zpool_prop_to_name(prop), value) == 0);
1074
1075 error = spa_prop_set(spa, props);
1076
1077 nvlist_free(props);
1078
1079 if (error == ENOSPC) {
1080 ztest_record_enospc(FTAG);
1081 return (error);
1082 }
1083 ASSERT0(error);
1084
1085 return (error);
1086 }
1087
1088 static void
1089 ztest_rll_init(rll_t *rll)
1090 {
1091 rll->rll_writer = NULL;
1092 rll->rll_readers = 0;
1093 VERIFY(_mutex_init(&rll->rll_lock, USYNC_THREAD, NULL) == 0);
1094 VERIFY(cond_init(&rll->rll_cv, USYNC_THREAD, NULL) == 0);
1095 }
1096
1097 static void
1098 ztest_rll_destroy(rll_t *rll)
1099 {
1100 ASSERT(rll->rll_writer == NULL);
1101 ASSERT(rll->rll_readers == 0);
1102 VERIFY(_mutex_destroy(&rll->rll_lock) == 0);
1103 VERIFY(cond_destroy(&rll->rll_cv) == 0);
1104 }
1105
1106 static void
1107 ztest_rll_lock(rll_t *rll, rl_type_t type)
1108 {
1109 VERIFY(mutex_lock(&rll->rll_lock) == 0);
1110
1111 if (type == RL_READER) {
1112 while (rll->rll_writer != NULL)
1113 (void) cond_wait(&rll->rll_cv, &rll->rll_lock);
1114 rll->rll_readers++;
1115 } else {
1116 while (rll->rll_writer != NULL || rll->rll_readers)
1117 (void) cond_wait(&rll->rll_cv, &rll->rll_lock);
1118 rll->rll_writer = curthread;
1119 }
1120
1121 VERIFY(mutex_unlock(&rll->rll_lock) == 0);
1122 }
1123
1124 static void
1125 ztest_rll_unlock(rll_t *rll)
1126 {
1127 VERIFY(mutex_lock(&rll->rll_lock) == 0);
1128
1129 if (rll->rll_writer) {
1130 ASSERT(rll->rll_readers == 0);
1131 rll->rll_writer = NULL;
1132 } else {
1133 ASSERT(rll->rll_readers != 0);
1134 ASSERT(rll->rll_writer == NULL);
1135 rll->rll_readers--;
1136 }
1137
1138 if (rll->rll_writer == NULL && rll->rll_readers == 0)
1139 VERIFY(cond_broadcast(&rll->rll_cv) == 0);
1140
1141 VERIFY(mutex_unlock(&rll->rll_lock) == 0);
1142 }
1143
1144 static void
1145 ztest_object_lock(ztest_ds_t *zd, uint64_t object, rl_type_t type)
1146 {
1147 rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1148
1149 ztest_rll_lock(rll, type);
1150 }
1151
1152 static void
1153 ztest_object_unlock(ztest_ds_t *zd, uint64_t object)
1154 {
1155 rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1156
1157 ztest_rll_unlock(rll);
1158 }
1159
1160 static rl_t *
1161 ztest_range_lock(ztest_ds_t *zd, uint64_t object, uint64_t offset,
1162 uint64_t size, rl_type_t type)
1163 {
1164 uint64_t hash = object ^ (offset % (ZTEST_RANGE_LOCKS + 1));
1165 rll_t *rll = &zd->zd_range_lock[hash & (ZTEST_RANGE_LOCKS - 1)];
1166 rl_t *rl;
1167
1168 rl = umem_alloc(sizeof (*rl), UMEM_NOFAIL);
1169 rl->rl_object = object;
1170 rl->rl_offset = offset;
1171 rl->rl_size = size;
1172 rl->rl_lock = rll;
1173
1174 ztest_rll_lock(rll, type);
1175
1176 return (rl);
1177 }
1178
1179 static void
1180 ztest_range_unlock(rl_t *rl)
1181 {
1182 rll_t *rll = rl->rl_lock;
1183
1184 ztest_rll_unlock(rll);
1185
1186 umem_free(rl, sizeof (*rl));
1187 }
1188
1189 static void
1190 ztest_zd_init(ztest_ds_t *zd, ztest_shared_ds_t *szd, objset_t *os)
1191 {
1192 zd->zd_os = os;
1193 zd->zd_zilog = dmu_objset_zil(os);
1194 zd->zd_shared = szd;
1195 dmu_objset_name(os, zd->zd_name);
1196
1197 if (zd->zd_shared != NULL)
1198 zd->zd_shared->zd_seq = 0;
1199
1200 VERIFY(rwlock_init(&zd->zd_zilog_lock, USYNC_THREAD, NULL) == 0);
1201 VERIFY(_mutex_init(&zd->zd_dirobj_lock, USYNC_THREAD, NULL) == 0);
1202
1203 for (int l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1204 ztest_rll_init(&zd->zd_object_lock[l]);
1205
1206 for (int l = 0; l < ZTEST_RANGE_LOCKS; l++)
1207 ztest_rll_init(&zd->zd_range_lock[l]);
1208 }
1209
1210 static void
1211 ztest_zd_fini(ztest_ds_t *zd)
1212 {
1213 VERIFY(_mutex_destroy(&zd->zd_dirobj_lock) == 0);
1214
1215 for (int l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1216 ztest_rll_destroy(&zd->zd_object_lock[l]);
1217
1218 for (int l = 0; l < ZTEST_RANGE_LOCKS; l++)
1219 ztest_rll_destroy(&zd->zd_range_lock[l]);
1220 }
1221
1222 #define TXG_MIGHTWAIT (ztest_random(10) == 0 ? TXG_NOWAIT : TXG_WAIT)
1223
1224 static uint64_t
1225 ztest_tx_assign(dmu_tx_t *tx, uint64_t txg_how, const char *tag)
1226 {
1227 uint64_t txg;
1228 int error;
1229
1230 /*
1231 * Attempt to assign tx to some transaction group.
1232 */
1233 error = dmu_tx_assign(tx, txg_how);
1234 if (error) {
1235 if (error == ERESTART) {
1236 ASSERT(txg_how == TXG_NOWAIT);
1237 dmu_tx_wait(tx);
1238 } else {
1239 ASSERT3U(error, ==, ENOSPC);
1240 ztest_record_enospc(tag);
1241 }
1242 dmu_tx_abort(tx);
1243 return (0);
1244 }
1245 txg = dmu_tx_get_txg(tx);
1246 ASSERT(txg != 0);
1247 return (txg);
1248 }
1249
1250 static void
1251 ztest_pattern_set(void *buf, uint64_t size, uint64_t value)
1252 {
1253 uint64_t *ip = buf;
1254 uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1255
1256 while (ip < ip_end)
1257 *ip++ = value;
1258 }
1259
1260 static boolean_t
1261 ztest_pattern_match(void *buf, uint64_t size, uint64_t value)
1262 {
1263 uint64_t *ip = buf;
1264 uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1265 uint64_t diff = 0;
1266
1267 while (ip < ip_end)
1268 diff |= (value - *ip++);
1269
1270 return (diff == 0);
1271 }
1272
1273 static void
1274 ztest_bt_generate(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1275 uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg)
1276 {
1277 bt->bt_magic = BT_MAGIC;
1278 bt->bt_objset = dmu_objset_id(os);
1279 bt->bt_object = object;
1280 bt->bt_offset = offset;
1281 bt->bt_gen = gen;
1282 bt->bt_txg = txg;
1283 bt->bt_crtxg = crtxg;
1284 }
1285
1286 static void
1287 ztest_bt_verify(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1288 uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg)
1289 {
1290 ASSERT3U(bt->bt_magic, ==, BT_MAGIC);
1291 ASSERT3U(bt->bt_objset, ==, dmu_objset_id(os));
1292 ASSERT3U(bt->bt_object, ==, object);
1293 ASSERT3U(bt->bt_offset, ==, offset);
1294 ASSERT3U(bt->bt_gen, <=, gen);
1295 ASSERT3U(bt->bt_txg, <=, txg);
1296 ASSERT3U(bt->bt_crtxg, ==, crtxg);
1297 }
1298
1299 static ztest_block_tag_t *
1300 ztest_bt_bonus(dmu_buf_t *db)
1301 {
1302 dmu_object_info_t doi;
1303 ztest_block_tag_t *bt;
1304
1305 dmu_object_info_from_db(db, &doi);
1306 ASSERT3U(doi.doi_bonus_size, <=, db->db_size);
1307 ASSERT3U(doi.doi_bonus_size, >=, sizeof (*bt));
1308 bt = (void *)((char *)db->db_data + doi.doi_bonus_size - sizeof (*bt));
1309
1310 return (bt);
1311 }
1312
1313 /*
1314 * ZIL logging ops
1315 */
1316
1317 #define lrz_type lr_mode
1318 #define lrz_blocksize lr_uid
1319 #define lrz_ibshift lr_gid
1320 #define lrz_bonustype lr_rdev
1321 #define lrz_bonuslen lr_crtime[1]
1322
1323 static void
1324 ztest_log_create(ztest_ds_t *zd, dmu_tx_t *tx, lr_create_t *lr)
1325 {
1326 char *name = (void *)(lr + 1); /* name follows lr */
1327 size_t namesize = strlen(name) + 1;
1328 itx_t *itx;
1329
1330 if (zil_replaying(zd->zd_zilog, tx))
1331 return;
1332
1333 itx = zil_itx_create(TX_CREATE, sizeof (*lr) + namesize);
1334 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1335 sizeof (*lr) + namesize - sizeof (lr_t));
1336
1337 zil_itx_assign(zd->zd_zilog, itx, tx);
1338 }
1339
1340 static void
1341 ztest_log_remove(ztest_ds_t *zd, dmu_tx_t *tx, lr_remove_t *lr, uint64_t object)
1342 {
1343 char *name = (void *)(lr + 1); /* name follows lr */
1344 size_t namesize = strlen(name) + 1;
1345 itx_t *itx;
1346
1347 if (zil_replaying(zd->zd_zilog, tx))
1348 return;
1349
1350 itx = zil_itx_create(TX_REMOVE, sizeof (*lr) + namesize);
1351 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1352 sizeof (*lr) + namesize - sizeof (lr_t));
1353
1354 itx->itx_oid = object;
1355 zil_itx_assign(zd->zd_zilog, itx, tx);
1356 }
1357
1358 static void
1359 ztest_log_write(ztest_ds_t *zd, dmu_tx_t *tx, lr_write_t *lr)
1360 {
1361 itx_t *itx;
1362 itx_wr_state_t write_state = ztest_random(WR_NUM_STATES);
1363
1364 if (zil_replaying(zd->zd_zilog, tx))
1365 return;
1366
1367 if (lr->lr_length > ZIL_MAX_LOG_DATA)
1368 write_state = WR_INDIRECT;
1369
1370 itx = zil_itx_create(TX_WRITE,
1371 sizeof (*lr) + (write_state == WR_COPIED ? lr->lr_length : 0));
1372
1373 if (write_state == WR_COPIED &&
1374 dmu_read(zd->zd_os, lr->lr_foid, lr->lr_offset, lr->lr_length,
1375 ((lr_write_t *)&itx->itx_lr) + 1, DMU_READ_NO_PREFETCH) != 0) {
1376 zil_itx_destroy(itx);
1377 itx = zil_itx_create(TX_WRITE, sizeof (*lr));
1378 write_state = WR_NEED_COPY;
1379 }
1380 itx->itx_private = zd;
1381 itx->itx_wr_state = write_state;
1382 itx->itx_sync = (ztest_random(8) == 0);
1383
1384 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1385 sizeof (*lr) - sizeof (lr_t));
1386
1387 zil_itx_assign(zd->zd_zilog, itx, tx);
1388 }
1389
1390 static void
1391 ztest_log_truncate(ztest_ds_t *zd, dmu_tx_t *tx, lr_truncate_t *lr)
1392 {
1393 itx_t *itx;
1394
1395 if (zil_replaying(zd->zd_zilog, tx))
1396 return;
1397
1398 itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr));
1399 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1400 sizeof (*lr) - sizeof (lr_t));
1401
1402 itx->itx_sync = B_FALSE;
1403 zil_itx_assign(zd->zd_zilog, itx, tx);
1404 }
1405
1406 static void
1407 ztest_log_setattr(ztest_ds_t *zd, dmu_tx_t *tx, lr_setattr_t *lr)
1408 {
1409 itx_t *itx;
1410
1411 if (zil_replaying(zd->zd_zilog, tx))
1412 return;
1413
1414 itx = zil_itx_create(TX_SETATTR, sizeof (*lr));
1415 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1416 sizeof (*lr) - sizeof (lr_t));
1417
1418 itx->itx_sync = B_FALSE;
1419 zil_itx_assign(zd->zd_zilog, itx, tx);
1420 }
1421
1422 /*
1423 * ZIL replay ops
1424 */
1425 static int
1426 ztest_replay_create(ztest_ds_t *zd, lr_create_t *lr, boolean_t byteswap)
1427 {
1428 char *name = (void *)(lr + 1); /* name follows lr */
1429 objset_t *os = zd->zd_os;
1430 ztest_block_tag_t *bbt;
1431 dmu_buf_t *db;
1432 dmu_tx_t *tx;
1433 uint64_t txg;
1434 int error = 0;
1435
1436 if (byteswap)
1437 byteswap_uint64_array(lr, sizeof (*lr));
1438
1439 ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1440 ASSERT(name[0] != '\0');
1441
1442 tx = dmu_tx_create(os);
1443
1444 dmu_tx_hold_zap(tx, lr->lr_doid, B_TRUE, name);
1445
1446 if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1447 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, B_TRUE, NULL);
1448 } else {
1449 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1450 }
1451
1452 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1453 if (txg == 0)
1454 return (ENOSPC);
1455
1456 ASSERT(dmu_objset_zil(os)->zl_replay == !!lr->lr_foid);
1457
1458 if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1459 if (lr->lr_foid == 0) {
1460 lr->lr_foid = zap_create(os,
1461 lr->lrz_type, lr->lrz_bonustype,
1462 lr->lrz_bonuslen, tx);
1463 } else {
1464 error = zap_create_claim(os, lr->lr_foid,
1465 lr->lrz_type, lr->lrz_bonustype,
1466 lr->lrz_bonuslen, tx);
1467 }
1468 } else {
1469 if (lr->lr_foid == 0) {
1470 lr->lr_foid = dmu_object_alloc(os,
1471 lr->lrz_type, 0, lr->lrz_bonustype,
1472 lr->lrz_bonuslen, tx);
1473 } else {
1474 error = dmu_object_claim(os, lr->lr_foid,
1475 lr->lrz_type, 0, lr->lrz_bonustype,
1476 lr->lrz_bonuslen, tx);
1477 }
1478 }
1479
1480 if (error) {
1481 ASSERT3U(error, ==, EEXIST);
1482 ASSERT(zd->zd_zilog->zl_replay);
1483 dmu_tx_commit(tx);
1484 return (error);
1485 }
1486
1487 ASSERT(lr->lr_foid != 0);
1488
1489 if (lr->lrz_type != DMU_OT_ZAP_OTHER)
1490 VERIFY3U(0, ==, dmu_object_set_blocksize(os, lr->lr_foid,
1491 lr->lrz_blocksize, lr->lrz_ibshift, tx));
1492
1493 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1494 bbt = ztest_bt_bonus(db);
1495 dmu_buf_will_dirty(db, tx);
1496 ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_gen, txg, txg);
1497 dmu_buf_rele(db, FTAG);
1498
1499 VERIFY3U(0, ==, zap_add(os, lr->lr_doid, name, sizeof (uint64_t), 1,
1500 &lr->lr_foid, tx));
1501
1502 (void) ztest_log_create(zd, tx, lr);
1503
1504 dmu_tx_commit(tx);
1505
1506 return (0);
1507 }
1508
1509 static int
1510 ztest_replay_remove(ztest_ds_t *zd, lr_remove_t *lr, boolean_t byteswap)
1511 {
1512 char *name = (void *)(lr + 1); /* name follows lr */
1513 objset_t *os = zd->zd_os;
1514 dmu_object_info_t doi;
1515 dmu_tx_t *tx;
1516 uint64_t object, txg;
1517
1518 if (byteswap)
1519 byteswap_uint64_array(lr, sizeof (*lr));
1520
1521 ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1522 ASSERT(name[0] != '\0');
1523
1524 VERIFY3U(0, ==,
1525 zap_lookup(os, lr->lr_doid, name, sizeof (object), 1, &object));
1526 ASSERT(object != 0);
1527
1528 ztest_object_lock(zd, object, RL_WRITER);
1529
1530 VERIFY3U(0, ==, dmu_object_info(os, object, &doi));
1531
1532 tx = dmu_tx_create(os);
1533
1534 dmu_tx_hold_zap(tx, lr->lr_doid, B_FALSE, name);
1535 dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END);
1536
1537 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1538 if (txg == 0) {
1539 ztest_object_unlock(zd, object);
1540 return (ENOSPC);
1541 }
1542
1543 if (doi.doi_type == DMU_OT_ZAP_OTHER) {
1544 VERIFY3U(0, ==, zap_destroy(os, object, tx));
1545 } else {
1546 VERIFY3U(0, ==, dmu_object_free(os, object, tx));
1547 }
1548
1549 VERIFY3U(0, ==, zap_remove(os, lr->lr_doid, name, tx));
1550
1551 (void) ztest_log_remove(zd, tx, lr, object);
1552
1553 dmu_tx_commit(tx);
1554
1555 ztest_object_unlock(zd, object);
1556
1557 return (0);
1558 }
1559
1560 static int
1561 ztest_replay_write(ztest_ds_t *zd, lr_write_t *lr, boolean_t byteswap)
1562 {
1563 objset_t *os = zd->zd_os;
1564 void *data = lr + 1; /* data follows lr */
1565 uint64_t offset, length;
1566 ztest_block_tag_t *bt = data;
1567 ztest_block_tag_t *bbt;
1568 uint64_t gen, txg, lrtxg, crtxg;
1569 dmu_object_info_t doi;
1570 dmu_tx_t *tx;
1571 dmu_buf_t *db;
1572 arc_buf_t *abuf = NULL;
1573 rl_t *rl;
1574
1575 if (byteswap)
1576 byteswap_uint64_array(lr, sizeof (*lr));
1577
1578 offset = lr->lr_offset;
1579 length = lr->lr_length;
1580
1581 /* If it's a dmu_sync() block, write the whole block */
1582 if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
1583 uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
1584 if (length < blocksize) {
1585 offset -= offset % blocksize;
1586 length = blocksize;
1587 }
1588 }
1589
1590 if (bt->bt_magic == BSWAP_64(BT_MAGIC))
1591 byteswap_uint64_array(bt, sizeof (*bt));
1592
1593 if (bt->bt_magic != BT_MAGIC)
1594 bt = NULL;
1595
1596 ztest_object_lock(zd, lr->lr_foid, RL_READER);
1597 rl = ztest_range_lock(zd, lr->lr_foid, offset, length, RL_WRITER);
1598
1599 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1600
1601 dmu_object_info_from_db(db, &doi);
1602
1603 bbt = ztest_bt_bonus(db);
1604 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1605 gen = bbt->bt_gen;
1606 crtxg = bbt->bt_crtxg;
1607 lrtxg = lr->lr_common.lrc_txg;
1608
1609 tx = dmu_tx_create(os);
1610
1611 dmu_tx_hold_write(tx, lr->lr_foid, offset, length);
1612
1613 if (ztest_random(8) == 0 && length == doi.doi_data_block_size &&
1614 P2PHASE(offset, length) == 0)
1615 abuf = dmu_request_arcbuf(db, length);
1616
1617 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1618 if (txg == 0) {
1619 if (abuf != NULL)
1620 dmu_return_arcbuf(abuf);
1621 dmu_buf_rele(db, FTAG);
1622 ztest_range_unlock(rl);
1623 ztest_object_unlock(zd, lr->lr_foid);
1624 return (ENOSPC);
1625 }
1626
1627 if (bt != NULL) {
1628 /*
1629 * Usually, verify the old data before writing new data --
1630 * but not always, because we also want to verify correct
1631 * behavior when the data was not recently read into cache.
1632 */
1633 ASSERT(offset % doi.doi_data_block_size == 0);
1634 if (ztest_random(4) != 0) {
1635 int prefetch = ztest_random(2) ?
1636 DMU_READ_PREFETCH : DMU_READ_NO_PREFETCH;
1637 ztest_block_tag_t rbt;
1638
1639 VERIFY(dmu_read(os, lr->lr_foid, offset,
1640 sizeof (rbt), &rbt, prefetch) == 0);
1641 if (rbt.bt_magic == BT_MAGIC) {
1642 ztest_bt_verify(&rbt, os, lr->lr_foid,
1643 offset, gen, txg, crtxg);
1644 }
1645 }
1646
1647 /*
1648 * Writes can appear to be newer than the bonus buffer because
1649 * the ztest_get_data() callback does a dmu_read() of the
1650 * open-context data, which may be different than the data
1651 * as it was when the write was generated.
1652 */
1653 if (zd->zd_zilog->zl_replay) {
1654 ztest_bt_verify(bt, os, lr->lr_foid, offset,
1655 MAX(gen, bt->bt_gen), MAX(txg, lrtxg),
1656 bt->bt_crtxg);
1657 }
1658
1659 /*
1660 * Set the bt's gen/txg to the bonus buffer's gen/txg
1661 * so that all of the usual ASSERTs will work.
1662 */
1663 ztest_bt_generate(bt, os, lr->lr_foid, offset, gen, txg, crtxg);
1664 }
1665
1666 if (abuf == NULL) {
1667 dmu_write(os, lr->lr_foid, offset, length, data, tx);
1668 } else {
1669 bcopy(data, abuf->b_data, length);
1670 dmu_assign_arcbuf(db, offset, abuf, tx);
1671 }
1672
1673 (void) ztest_log_write(zd, tx, lr);
1674
1675 dmu_buf_rele(db, FTAG);
1676
1677 dmu_tx_commit(tx);
1678
1679 ztest_range_unlock(rl);
1680 ztest_object_unlock(zd, lr->lr_foid);
1681
1682 return (0);
1683 }
1684
1685 static int
1686 ztest_replay_truncate(ztest_ds_t *zd, lr_truncate_t *lr, boolean_t byteswap)
1687 {
1688 objset_t *os = zd->zd_os;
1689 dmu_tx_t *tx;
1690 uint64_t txg;
1691 rl_t *rl;
1692
1693 if (byteswap)
1694 byteswap_uint64_array(lr, sizeof (*lr));
1695
1696 ztest_object_lock(zd, lr->lr_foid, RL_READER);
1697 rl = ztest_range_lock(zd, lr->lr_foid, lr->lr_offset, lr->lr_length,
1698 RL_WRITER);
1699
1700 tx = dmu_tx_create(os);
1701
1702 dmu_tx_hold_free(tx, lr->lr_foid, lr->lr_offset, lr->lr_length);
1703
1704 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1705 if (txg == 0) {
1706 ztest_range_unlock(rl);
1707 ztest_object_unlock(zd, lr->lr_foid);
1708 return (ENOSPC);
1709 }
1710
1711 VERIFY(dmu_free_range(os, lr->lr_foid, lr->lr_offset,
1712 lr->lr_length, tx) == 0);
1713
1714 (void) ztest_log_truncate(zd, tx, lr);
1715
1716 dmu_tx_commit(tx);
1717
1718 ztest_range_unlock(rl);
1719 ztest_object_unlock(zd, lr->lr_foid);
1720
1721 return (0);
1722 }
1723
1724 static int
1725 ztest_replay_setattr(ztest_ds_t *zd, lr_setattr_t *lr, boolean_t byteswap)
1726 {
1727 objset_t *os = zd->zd_os;
1728 dmu_tx_t *tx;
1729 dmu_buf_t *db;
1730 ztest_block_tag_t *bbt;
1731 uint64_t txg, lrtxg, crtxg;
1732
1733 if (byteswap)
1734 byteswap_uint64_array(lr, sizeof (*lr));
1735
1736 ztest_object_lock(zd, lr->lr_foid, RL_WRITER);
1737
1738 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1739
1740 tx = dmu_tx_create(os);
1741 dmu_tx_hold_bonus(tx, lr->lr_foid);
1742
1743 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1744 if (txg == 0) {
1745 dmu_buf_rele(db, FTAG);
1746 ztest_object_unlock(zd, lr->lr_foid);
1747 return (ENOSPC);
1748 }
1749
1750 bbt = ztest_bt_bonus(db);
1751 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1752 crtxg = bbt->bt_crtxg;
1753 lrtxg = lr->lr_common.lrc_txg;
1754
1755 if (zd->zd_zilog->zl_replay) {
1756 ASSERT(lr->lr_size != 0);
1757 ASSERT(lr->lr_mode != 0);
1758 ASSERT(lrtxg != 0);
1759 } else {
1760 /*
1761 * Randomly change the size and increment the generation.
1762 */
1763 lr->lr_size = (ztest_random(db->db_size / sizeof (*bbt)) + 1) *
1764 sizeof (*bbt);
1765 lr->lr_mode = bbt->bt_gen + 1;
1766 ASSERT(lrtxg == 0);
1767 }
1768
1769 /*
1770 * Verify that the current bonus buffer is not newer than our txg.
1771 */
1772 ztest_bt_verify(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode,
1773 MAX(txg, lrtxg), crtxg);
1774
1775 dmu_buf_will_dirty(db, tx);
1776
1777 ASSERT3U(lr->lr_size, >=, sizeof (*bbt));
1778 ASSERT3U(lr->lr_size, <=, db->db_size);
1779 VERIFY0(dmu_set_bonus(db, lr->lr_size, tx));
1780 bbt = ztest_bt_bonus(db);
1781
1782 ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode, txg, crtxg);
1783
1784 dmu_buf_rele(db, FTAG);
1785
1786 (void) ztest_log_setattr(zd, tx, lr);
1787
1788 dmu_tx_commit(tx);
1789
1790 ztest_object_unlock(zd, lr->lr_foid);
1791
1792 return (0);
1793 }
1794
1795 zil_replay_func_t *ztest_replay_vector[TX_MAX_TYPE] = {
1796 NULL, /* 0 no such transaction type */
1797 ztest_replay_create, /* TX_CREATE */
1798 NULL, /* TX_MKDIR */
1799 NULL, /* TX_MKXATTR */
1800 NULL, /* TX_SYMLINK */
1801 ztest_replay_remove, /* TX_REMOVE */
1802 NULL, /* TX_RMDIR */
1803 NULL, /* TX_LINK */
1804 NULL, /* TX_RENAME */
1805 ztest_replay_write, /* TX_WRITE */
1806 ztest_replay_truncate, /* TX_TRUNCATE */
1807 ztest_replay_setattr, /* TX_SETATTR */
1808 NULL, /* TX_ACL */
1809 NULL, /* TX_CREATE_ACL */
1810 NULL, /* TX_CREATE_ATTR */
1811 NULL, /* TX_CREATE_ACL_ATTR */
1812 NULL, /* TX_MKDIR_ACL */
1813 NULL, /* TX_MKDIR_ATTR */
1814 NULL, /* TX_MKDIR_ACL_ATTR */
1815 NULL, /* TX_WRITE2 */
1816 };
1817
1818 /*
1819 * ZIL get_data callbacks
1820 */
1821
1822 static void
1823 ztest_get_done(zgd_t *zgd, int error)
1824 {
1825 ztest_ds_t *zd = zgd->zgd_private;
1826 uint64_t object = zgd->zgd_rl->rl_object;
1827
1828 if (zgd->zgd_db)
1829 dmu_buf_rele(zgd->zgd_db, zgd);
1830
1831 ztest_range_unlock(zgd->zgd_rl);
1832 ztest_object_unlock(zd, object);
1833
1834 if (error == 0 && zgd->zgd_bp)
1835 zil_lwb_add_block(zgd->zgd_lwb, zgd->zgd_bp);
1836
1837 umem_free(zgd, sizeof (*zgd));
1838 }
1839
1840 static int
1841 ztest_get_data(void *arg, lr_write_t *lr, char *buf, struct lwb *lwb,
1842 zio_t *zio)
1843 {
1844 ztest_ds_t *zd = arg;
1845 objset_t *os = zd->zd_os;
1846 uint64_t object = lr->lr_foid;
1847 uint64_t offset = lr->lr_offset;
1848 uint64_t size = lr->lr_length;
1849 uint64_t txg = lr->lr_common.lrc_txg;
1850 uint64_t crtxg;
1851 dmu_object_info_t doi;
1852 dmu_buf_t *db;
1853 zgd_t *zgd;
1854 int error;
1855
1856 ASSERT3P(lwb, !=, NULL);
1857 ASSERT3P(zio, !=, NULL);
1858 ASSERT3U(size, !=, 0);
1859
1860 ztest_object_lock(zd, object, RL_READER);
1861 error = dmu_bonus_hold(os, object, FTAG, &db);
1862 if (error) {
1863 ztest_object_unlock(zd, object);
1864 return (error);
1865 }
1866
1867 crtxg = ztest_bt_bonus(db)->bt_crtxg;
1868
1869 if (crtxg == 0 || crtxg > txg) {
1870 dmu_buf_rele(db, FTAG);
1871 ztest_object_unlock(zd, object);
1872 return (ENOENT);
1873 }
1874
1875 dmu_object_info_from_db(db, &doi);
1876 dmu_buf_rele(db, FTAG);
1877 db = NULL;
1878
1879 zgd = umem_zalloc(sizeof (*zgd), UMEM_NOFAIL);
1880 zgd->zgd_lwb = lwb;
1881 zgd->zgd_private = zd;
1882
1883 if (buf != NULL) { /* immediate write */
1884 zgd->zgd_rl = ztest_range_lock(zd, object, offset, size,
1885 RL_READER);
1886
1887 error = dmu_read(os, object, offset, size, buf,
1888 DMU_READ_NO_PREFETCH);
1889 ASSERT(error == 0);
1890 } else {
1891 size = doi.doi_data_block_size;
1892 if (ISP2(size)) {
1893 offset = P2ALIGN(offset, size);
1894 } else {
1895 ASSERT(offset < size);
1896 offset = 0;
1897 }
1898
1899 zgd->zgd_rl = ztest_range_lock(zd, object, offset, size,
1900 RL_READER);
1901
1902 error = dmu_buf_hold(os, object, offset, zgd, &db,
1903 DMU_READ_NO_PREFETCH);
1904
1905 if (error == 0) {
1906 blkptr_t *bp = &lr->lr_blkptr;
1907
1908 zgd->zgd_db = db;
1909 zgd->zgd_bp = bp;
1910
1911 ASSERT(db->db_offset == offset);
1912 ASSERT(db->db_size == size);
1913
1914 error = dmu_sync(zio, lr->lr_common.lrc_txg,
1915 ztest_get_done, zgd);
1916
1917 if (error == 0)
1918 return (0);
1919 }
1920 }
1921
1922 ztest_get_done(zgd, error);
1923
1924 return (error);
1925 }
1926
1927 static void *
1928 ztest_lr_alloc(size_t lrsize, char *name)
1929 {
1930 char *lr;
1931 size_t namesize = name ? strlen(name) + 1 : 0;
1932
1933 lr = umem_zalloc(lrsize + namesize, UMEM_NOFAIL);
1934
1935 if (name)
1936 bcopy(name, lr + lrsize, namesize);
1937
1938 return (lr);
1939 }
1940
1941 void
1942 ztest_lr_free(void *lr, size_t lrsize, char *name)
1943 {
1944 size_t namesize = name ? strlen(name) + 1 : 0;
1945
1946 umem_free(lr, lrsize + namesize);
1947 }
1948
1949 /*
1950 * Lookup a bunch of objects. Returns the number of objects not found.
1951 */
1952 static int
1953 ztest_lookup(ztest_ds_t *zd, ztest_od_t *od, int count)
1954 {
1955 int missing = 0;
1956 int error;
1957
1958 ASSERT(_mutex_held(&zd->zd_dirobj_lock));
1959
1960 for (int i = 0; i < count; i++, od++) {
1961 od->od_object = 0;
1962 error = zap_lookup(zd->zd_os, od->od_dir, od->od_name,
1963 sizeof (uint64_t), 1, &od->od_object);
1964 if (error) {
1965 ASSERT(error == ENOENT);
1966 ASSERT(od->od_object == 0);
1967 missing++;
1968 } else {
1969 dmu_buf_t *db;
1970 ztest_block_tag_t *bbt;
1971 dmu_object_info_t doi;
1972
1973 ASSERT(od->od_object != 0);
1974 ASSERT(missing == 0); /* there should be no gaps */
1975
1976 ztest_object_lock(zd, od->od_object, RL_READER);
1977 VERIFY3U(0, ==, dmu_bonus_hold(zd->zd_os,
1978 od->od_object, FTAG, &db));
1979 dmu_object_info_from_db(db, &doi);
1980 bbt = ztest_bt_bonus(db);
1981 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1982 od->od_type = doi.doi_type;
1983 od->od_blocksize = doi.doi_data_block_size;
1984 od->od_gen = bbt->bt_gen;
1985 dmu_buf_rele(db, FTAG);
1986 ztest_object_unlock(zd, od->od_object);
1987 }
1988 }
1989
1990 return (missing);
1991 }
1992
1993 static int
1994 ztest_create(ztest_ds_t *zd, ztest_od_t *od, int count)
1995 {
1996 int missing = 0;
1997
1998 ASSERT(_mutex_held(&zd->zd_dirobj_lock));
1999
2000 for (int i = 0; i < count; i++, od++) {
2001 if (missing) {
2002 od->od_object = 0;
2003 missing++;
2004 continue;
2005 }
2006
2007 lr_create_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
2008
2009 lr->lr_doid = od->od_dir;
2010 lr->lr_foid = 0; /* 0 to allocate, > 0 to claim */
2011 lr->lrz_type = od->od_crtype;
2012 lr->lrz_blocksize = od->od_crblocksize;
2013 lr->lrz_ibshift = ztest_random_ibshift();
2014 lr->lrz_bonustype = DMU_OT_UINT64_OTHER;
2015 lr->lrz_bonuslen = dmu_bonus_max();
2016 lr->lr_gen = od->od_crgen;
2017 lr->lr_crtime[0] = time(NULL);
2018
2019 if (ztest_replay_create(zd, lr, B_FALSE) != 0) {
2020 ASSERT(missing == 0);
2021 od->od_object = 0;
2022 missing++;
2023 } else {
2024 od->od_object = lr->lr_foid;
2025 od->od_type = od->od_crtype;
2026 od->od_blocksize = od->od_crblocksize;
2027 od->od_gen = od->od_crgen;
2028 ASSERT(od->od_object != 0);
2029 }
2030
2031 ztest_lr_free(lr, sizeof (*lr), od->od_name);
2032 }
2033
2034 return (missing);
2035 }
2036
2037 static int
2038 ztest_remove(ztest_ds_t *zd, ztest_od_t *od, int count)
2039 {
2040 int missing = 0;
2041 int error;
2042
2043 ASSERT(_mutex_held(&zd->zd_dirobj_lock));
2044
2045 od += count - 1;
2046
2047 for (int i = count - 1; i >= 0; i--, od--) {
2048 if (missing) {
2049 missing++;
2050 continue;
2051 }
2052
2053 /*
2054 * No object was found.
2055 */
2056 if (od->od_object == 0)
2057 continue;
2058
2059 lr_remove_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
2060
2061 lr->lr_doid = od->od_dir;
2062
2063 if ((error = ztest_replay_remove(zd, lr, B_FALSE)) != 0) {
2064 ASSERT3U(error, ==, ENOSPC);
2065 missing++;
2066 } else {
2067 od->od_object = 0;
2068 }
2069 ztest_lr_free(lr, sizeof (*lr), od->od_name);
2070 }
2071
2072 return (missing);
2073 }
2074
2075 static int
2076 ztest_write(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size,
2077 void *data)
2078 {
2079 lr_write_t *lr;
2080 int error;
2081
2082 lr = ztest_lr_alloc(sizeof (*lr) + size, NULL);
2083
2084 lr->lr_foid = object;
2085 lr->lr_offset = offset;
2086 lr->lr_length = size;
2087 lr->lr_blkoff = 0;
2088 BP_ZERO(&lr->lr_blkptr);
2089
2090 bcopy(data, lr + 1, size);
2091
2092 error = ztest_replay_write(zd, lr, B_FALSE);
2093
2094 ztest_lr_free(lr, sizeof (*lr) + size, NULL);
2095
2096 return (error);
2097 }
2098
2099 static int
2100 ztest_truncate(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2101 {
2102 lr_truncate_t *lr;
2103 int error;
2104
2105 lr = ztest_lr_alloc(sizeof (*lr), NULL);
2106
2107 lr->lr_foid = object;
2108 lr->lr_offset = offset;
2109 lr->lr_length = size;
2110
2111 error = ztest_replay_truncate(zd, lr, B_FALSE);
2112
2113 ztest_lr_free(lr, sizeof (*lr), NULL);
2114
2115 return (error);
2116 }
2117
2118 static int
2119 ztest_setattr(ztest_ds_t *zd, uint64_t object)
2120 {
2121 lr_setattr_t *lr;
2122 int error;
2123
2124 lr = ztest_lr_alloc(sizeof (*lr), NULL);
2125
2126 lr->lr_foid = object;
2127 lr->lr_size = 0;
2128 lr->lr_mode = 0;
2129
2130 error = ztest_replay_setattr(zd, lr, B_FALSE);
2131
2132 ztest_lr_free(lr, sizeof (*lr), NULL);
2133
2134 return (error);
2135 }
2136
2137 static void
2138 ztest_prealloc(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2139 {
2140 objset_t *os = zd->zd_os;
2141 dmu_tx_t *tx;
2142 uint64_t txg;
2143 rl_t *rl;
2144
2145 txg_wait_synced(dmu_objset_pool(os), 0);
2146
2147 ztest_object_lock(zd, object, RL_READER);
2148 rl = ztest_range_lock(zd, object, offset, size, RL_WRITER);
2149
2150 tx = dmu_tx_create(os);
2151
2152 dmu_tx_hold_write(tx, object, offset, size);
2153
2154 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
2155
2156 if (txg != 0) {
2157 dmu_prealloc(os, object, offset, size, tx);
2158 dmu_tx_commit(tx);
2159 txg_wait_synced(dmu_objset_pool(os), txg);
2160 } else {
2161 (void) dmu_free_long_range(os, object, offset, size);
2162 }
2163
2164 ztest_range_unlock(rl);
2165 ztest_object_unlock(zd, object);
2166 }
2167
2168 static void
2169 ztest_io(ztest_ds_t *zd, uint64_t object, uint64_t offset)
2170 {
2171 int err;
2172 ztest_block_tag_t wbt;
2173 dmu_object_info_t doi;
2174 enum ztest_io_type io_type;
2175 uint64_t blocksize;
2176 void *data;
2177
2178 VERIFY(dmu_object_info(zd->zd_os, object, &doi) == 0);
2179 blocksize = doi.doi_data_block_size;
2180 data = umem_alloc(blocksize, UMEM_NOFAIL);
2181
2182 /*
2183 * Pick an i/o type at random, biased toward writing block tags.
2184 */
2185 io_type = ztest_random(ZTEST_IO_TYPES);
2186 if (ztest_random(2) == 0)
2187 io_type = ZTEST_IO_WRITE_TAG;
2188
2189 (void) rw_rdlock(&zd->zd_zilog_lock);
2190
2191 switch (io_type) {
2192
2193 case ZTEST_IO_WRITE_TAG:
2194 ztest_bt_generate(&wbt, zd->zd_os, object, offset, 0, 0, 0);
2195 (void) ztest_write(zd, object, offset, sizeof (wbt), &wbt);
2196 break;
2197
2198 case ZTEST_IO_WRITE_PATTERN:
2199 (void) memset(data, 'a' + (object + offset) % 5, blocksize);
2200 if (ztest_random(2) == 0) {
2201 /*
2202 * Induce fletcher2 collisions to ensure that
2203 * zio_ddt_collision() detects and resolves them
2204 * when using fletcher2-verify for deduplication.
2205 */
2206 ((uint64_t *)data)[0] ^= 1ULL << 63;
2207 ((uint64_t *)data)[4] ^= 1ULL << 63;
2208 }
2209 (void) ztest_write(zd, object, offset, blocksize, data);
2210 break;
2211
2212 case ZTEST_IO_WRITE_ZEROES:
2213 bzero(data, blocksize);
2214 (void) ztest_write(zd, object, offset, blocksize, data);
2215 break;
2216
2217 case ZTEST_IO_TRUNCATE:
2218 (void) ztest_truncate(zd, object, offset, blocksize);
2219 break;
2220
2221 case ZTEST_IO_SETATTR:
2222 (void) ztest_setattr(zd, object);
2223 break;
2224
2225 case ZTEST_IO_REWRITE:
2226 (void) rw_rdlock(&ztest_name_lock);
2227 err = ztest_dsl_prop_set_uint64(zd->zd_name,
2228 ZFS_PROP_CHECKSUM, spa_dedup_checksum(ztest_spa),
2229 B_FALSE);
2230 VERIFY(err == 0 || err == ENOSPC);
2231 err = ztest_dsl_prop_set_uint64(zd->zd_name,
2232 ZFS_PROP_COMPRESSION,
2233 ztest_random_dsl_prop(ZFS_PROP_COMPRESSION),
2234 B_FALSE);
2235 VERIFY(err == 0 || err == ENOSPC);
2236 (void) rw_unlock(&ztest_name_lock);
2237
2238 VERIFY0(dmu_read(zd->zd_os, object, offset, blocksize, data,
2239 DMU_READ_NO_PREFETCH));
2240
2241 (void) ztest_write(zd, object, offset, blocksize, data);
2242 break;
2243 }
2244
2245 (void) rw_unlock(&zd->zd_zilog_lock);
2246
2247 umem_free(data, blocksize);
2248 }
2249
2250 /*
2251 * Initialize an object description template.
2252 */
2253 static void
2254 ztest_od_init(ztest_od_t *od, uint64_t id, char *tag, uint64_t index,
2255 dmu_object_type_t type, uint64_t blocksize, uint64_t gen)
2256 {
2257 od->od_dir = ZTEST_DIROBJ;
2258 od->od_object = 0;
2259
2260 od->od_crtype = type;
2261 od->od_crblocksize = blocksize ? blocksize : ztest_random_blocksize();
2262 od->od_crgen = gen;
2263
2264 od->od_type = DMU_OT_NONE;
2265 od->od_blocksize = 0;
2266 od->od_gen = 0;
2267
2268 (void) snprintf(od->od_name, sizeof (od->od_name), "%s(%lld)[%llu]",
2269 tag, (int64_t)id, index);
2270 }
2271
2272 /*
2273 * Lookup or create the objects for a test using the od template.
2274 * If the objects do not all exist, or if 'remove' is specified,
2275 * remove any existing objects and create new ones. Otherwise,
2276 * use the existing objects.
2277 */
2278 static int
2279 ztest_object_init(ztest_ds_t *zd, ztest_od_t *od, size_t size, boolean_t remove)
2280 {
2281 int count = size / sizeof (*od);
2282 int rv = 0;
2283
2284 VERIFY(mutex_lock(&zd->zd_dirobj_lock) == 0);
2285 if ((ztest_lookup(zd, od, count) != 0 || remove) &&
2286 (ztest_remove(zd, od, count) != 0 ||
2287 ztest_create(zd, od, count) != 0))
2288 rv = -1;
2289 zd->zd_od = od;
2290 VERIFY(mutex_unlock(&zd->zd_dirobj_lock) == 0);
2291
2292 return (rv);
2293 }
2294
2295 /* ARGSUSED */
2296 void
2297 ztest_zil_commit(ztest_ds_t *zd, uint64_t id)
2298 {
2299 zilog_t *zilog = zd->zd_zilog;
2300
2301 (void) rw_rdlock(&zd->zd_zilog_lock);
2302
2303 zil_commit(zilog, ztest_random(ZTEST_OBJECTS));
2304
2305 /*
2306 * Remember the committed values in zd, which is in parent/child
2307 * shared memory. If we die, the next iteration of ztest_run()
2308 * will verify that the log really does contain this record.
2309 */
2310 mutex_enter(&zilog->zl_lock);
2311 ASSERT(zd->zd_shared != NULL);
2312 ASSERT3U(zd->zd_shared->zd_seq, <=, zilog->zl_commit_lr_seq);
2313 zd->zd_shared->zd_seq = zilog->zl_commit_lr_seq;
2314 mutex_exit(&zilog->zl_lock);
2315
2316 (void) rw_unlock(&zd->zd_zilog_lock);
2317 }
2318
2319 /*
2320 * This function is designed to simulate the operations that occur during a
2321 * mount/unmount operation. We hold the dataset across these operations in an
2322 * attempt to expose any implicit assumptions about ZIL management.
2323 */
2324 /* ARGSUSED */
2325 void
2326 ztest_zil_remount(ztest_ds_t *zd, uint64_t id)
2327 {
2328 objset_t *os = zd->zd_os;
2329
2330 /*
2331 * We grab the zd_dirobj_lock to ensure that no other thread is
2332 * updating the zil (i.e. adding in-memory log records) and the
2333 * zd_zilog_lock to block any I/O.
2334 */
2335 VERIFY0(mutex_lock(&zd->zd_dirobj_lock));
2336 (void) rw_wrlock(&zd->zd_zilog_lock);
2337
2338 /* zfsvfs_teardown() */
2339 zil_close(zd->zd_zilog);
2340
2341 /* zfsvfs_setup() */
2342 VERIFY(zil_open(os, ztest_get_data) == zd->zd_zilog);
2343 zil_replay(os, zd, ztest_replay_vector);
2344
2345 (void) rw_unlock(&zd->zd_zilog_lock);
2346 VERIFY(mutex_unlock(&zd->zd_dirobj_lock) == 0);
2347 }
2348
2349 /*
2350 * Verify that we can't destroy an active pool, create an existing pool,
2351 * or create a pool with a bad vdev spec.
2352 */
2353 /* ARGSUSED */
2354 void
2355 ztest_spa_create_destroy(ztest_ds_t *zd, uint64_t id)
2356 {
2357 ztest_shared_opts_t *zo = &ztest_opts;
2358 spa_t *spa;
2359 nvlist_t *nvroot;
2360
2361 /*
2362 * Attempt to create using a bad file.
2363 */
2364 nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1);
2365 VERIFY3U(ENOENT, ==,
2366 spa_create("ztest_bad_file", nvroot, NULL, NULL));
2367 nvlist_free(nvroot);
2368
2369 /*
2370 * Attempt to create using a bad mirror.
2371 */
2372 nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 2, 1);
2373 VERIFY3U(ENOENT, ==,
2374 spa_create("ztest_bad_mirror", nvroot, NULL, NULL));
2375 nvlist_free(nvroot);
2376
2377 /*
2378 * Attempt to create an existing pool. It shouldn't matter
2379 * what's in the nvroot; we should fail with EEXIST.
2380 */
2381 (void) rw_rdlock(&ztest_name_lock);
2382 nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1);
2383 VERIFY3U(EEXIST, ==, spa_create(zo->zo_pool, nvroot, NULL, NULL));
2384 nvlist_free(nvroot);
2385 VERIFY3U(0, ==, spa_open(zo->zo_pool, &spa, FTAG));
2386 VERIFY3U(EBUSY, ==, spa_destroy(zo->zo_pool));
2387 spa_close(spa, FTAG);
2388
2389 (void) rw_unlock(&ztest_name_lock);
2390 }
2391
2392 /* ARGSUSED */
2393 void
2394 ztest_spa_upgrade(ztest_ds_t *zd, uint64_t id)
2395 {
2396 spa_t *spa;
2397 uint64_t initial_version = SPA_VERSION_INITIAL;
2398 uint64_t version, newversion;
2399 nvlist_t *nvroot, *props;
2400 char *name;
2401
2402 VERIFY0(mutex_lock(&ztest_vdev_lock));
2403 name = kmem_asprintf("%s_upgrade", ztest_opts.zo_pool);
2404
2405 /*
2406 * Clean up from previous runs.
2407 */
2408 (void) spa_destroy(name);
2409
2410 nvroot = make_vdev_root(NULL, NULL, name, ztest_opts.zo_vdev_size, 0,
2411 0, ztest_opts.zo_raidz, ztest_opts.zo_mirrors, 1);
2412
2413 /*
2414 * If we're configuring a RAIDZ device then make sure that the
2415 * the initial version is capable of supporting that feature.
2416 */
2417 switch (ztest_opts.zo_raidz_parity) {
2418 case 0:
2419 case 1:
2420 initial_version = SPA_VERSION_INITIAL;
2421 break;
2422 case 2:
2423 initial_version = SPA_VERSION_RAIDZ2;
2424 break;
2425 case 3:
2426 initial_version = SPA_VERSION_RAIDZ3;
2427 break;
2428 }
2429
2430 /*
2431 * Create a pool with a spa version that can be upgraded. Pick
2432 * a value between initial_version and SPA_VERSION_BEFORE_FEATURES.
2433 */
2434 do {
2435 version = ztest_random_spa_version(initial_version);
2436 } while (version > SPA_VERSION_BEFORE_FEATURES);
2437
2438 props = fnvlist_alloc();
2439 fnvlist_add_uint64(props,
2440 zpool_prop_to_name(ZPOOL_PROP_VERSION), version);
2441 VERIFY0(spa_create(name, nvroot, props, NULL));
2442 fnvlist_free(nvroot);
2443 fnvlist_free(props);
2444
2445 VERIFY0(spa_open(name, &spa, FTAG));
2446 VERIFY3U(spa_version(spa), ==, version);
2447 newversion = ztest_random_spa_version(version + 1);
2448
2449 if (ztest_opts.zo_verbose >= 4) {
2450 (void) printf("upgrading spa version from %llu to %llu\n",
2451 (u_longlong_t)version, (u_longlong_t)newversion);
2452 }
2453
2454 spa_upgrade(spa, newversion);
2455 VERIFY3U(spa_version(spa), >, version);
2456 VERIFY3U(spa_version(spa), ==, fnvlist_lookup_uint64(spa->spa_config,
2457 zpool_prop_to_name(ZPOOL_PROP_VERSION)));
2458 spa_close(spa, FTAG);
2459
2460 strfree(name);
2461 VERIFY0(mutex_unlock(&ztest_vdev_lock));
2462 }
2463
2464 static vdev_t *
2465 vdev_lookup_by_path(vdev_t *vd, const char *path)
2466 {
2467 vdev_t *mvd;
2468
2469 if (vd->vdev_path != NULL && strcmp(path, vd->vdev_path) == 0)
2470 return (vd);
2471
2472 for (int c = 0; c < vd->vdev_children; c++)
2473 if ((mvd = vdev_lookup_by_path(vd->vdev_child[c], path)) !=
2474 NULL)
2475 return (mvd);
2476
2477 return (NULL);
2478 }
2479
2480 /*
2481 * Find the first available hole which can be used as a top-level.
2482 */
2483 int
2484 find_vdev_hole(spa_t *spa)
2485 {
2486 vdev_t *rvd = spa->spa_root_vdev;
2487 int c;
2488
2489 ASSERT(spa_config_held(spa, SCL_VDEV, RW_READER) == SCL_VDEV);
2490
2491 for (c = 0; c < rvd->vdev_children; c++) {
2492 vdev_t *cvd = rvd->vdev_child[c];
2493
2494 if (cvd->vdev_ishole)
2495 break;
2496 }
2497 return (c);
2498 }
2499
2500 /*
2501 * Verify that vdev_add() works as expected.
2502 */
2503 /* ARGSUSED */
2504 void
2505 ztest_vdev_add_remove(ztest_ds_t *zd, uint64_t id)
2506 {
2507 ztest_shared_t *zs = ztest_shared;
2508 spa_t *spa = ztest_spa;
2509 uint64_t leaves;
2510 uint64_t guid;
2511 nvlist_t *nvroot;
2512 int error;
2513
2514 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2515 leaves = MAX(zs->zs_mirrors + zs->zs_splits, 1) * ztest_opts.zo_raidz;
2516
2517 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2518
2519 ztest_shared->zs_vdev_next_leaf = find_vdev_hole(spa) * leaves;
2520
2521 /*
2522 * If we have slogs then remove them 1/4 of the time.
2523 */
2524 if (spa_has_slogs(spa) && ztest_random(4) == 0) {
2525 /*
2526 * Grab the guid from the head of the log class rotor.
2527 */
2528 guid = spa_log_class(spa)->mc_rotor->mg_vd->vdev_guid;
2529
2530 spa_config_exit(spa, SCL_VDEV, FTAG);
2531
2532 /*
2533 * We have to grab the zs_name_lock as writer to
2534 * prevent a race between removing a slog (dmu_objset_find)
2535 * and destroying a dataset. Removing the slog will
2536 * grab a reference on the dataset which may cause
2537 * dmu_objset_destroy() to fail with EBUSY thus
2538 * leaving the dataset in an inconsistent state.
2539 */
2540 VERIFY(rw_wrlock(&ztest_name_lock) == 0);
2541 error = spa_vdev_remove(spa, guid, B_FALSE);
2542 VERIFY(rw_unlock(&ztest_name_lock) == 0);
2543
2544 if (error && error != EEXIST)
2545 fatal(0, "spa_vdev_remove() = %d", error);
2546 } else {
2547 spa_config_exit(spa, SCL_VDEV, FTAG);
2548
2549 /*
2550 * Make 1/4 of the devices be log devices.
2551 */
2552 nvroot = make_vdev_root(NULL, NULL, NULL,
2553 ztest_opts.zo_vdev_size, 0,
2554 ztest_random(4) == 0, ztest_opts.zo_raidz,
2555 zs->zs_mirrors, 1);
2556
2557 error = spa_vdev_add(spa, nvroot);
2558 nvlist_free(nvroot);
2559
2560 if (error == ENOSPC)
2561 ztest_record_enospc("spa_vdev_add");
2562 else if (error != 0)
2563 fatal(0, "spa_vdev_add() = %d", error);
2564 }
2565
2566 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2567 }
2568
2569 /*
2570 * Verify that adding/removing aux devices (l2arc, hot spare) works as expected.
2571 */
2572 /* ARGSUSED */
2573 void
2574 ztest_vdev_aux_add_remove(ztest_ds_t *zd, uint64_t id)
2575 {
2576 ztest_shared_t *zs = ztest_shared;
2577 spa_t *spa = ztest_spa;
2578 vdev_t *rvd = spa->spa_root_vdev;
2579 spa_aux_vdev_t *sav;
2580 char *aux;
2581 uint64_t guid = 0;
2582 int error;
2583
2584 if (ztest_random(2) == 0) {
2585 sav = &spa->spa_spares;
2586 aux = ZPOOL_CONFIG_SPARES;
2587 } else {
2588 sav = &spa->spa_l2cache;
2589 aux = ZPOOL_CONFIG_L2CACHE;
2590 }
2591
2592 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2593
2594 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2595
2596 if (sav->sav_count != 0 && ztest_random(4) == 0) {
2597 /*
2598 * Pick a random device to remove.
2599 */
2600 guid = sav->sav_vdevs[ztest_random(sav->sav_count)]->vdev_guid;
2601 } else {
2602 /*
2603 * Find an unused device we can add.
2604 */
2605 zs->zs_vdev_aux = 0;
2606 for (;;) {
2607 char path[MAXPATHLEN];
2608 int c;
2609 (void) snprintf(path, sizeof (path), ztest_aux_template,
2610 ztest_opts.zo_dir, ztest_opts.zo_pool, aux,
2611 zs->zs_vdev_aux);
2612 for (c = 0; c < sav->sav_count; c++)
2613 if (strcmp(sav->sav_vdevs[c]->vdev_path,
2614 path) == 0)
2615 break;
2616 if (c == sav->sav_count &&
2617 vdev_lookup_by_path(rvd, path) == NULL)
2618 break;
2619 zs->zs_vdev_aux++;
2620 }
2621 }
2622
2623 spa_config_exit(spa, SCL_VDEV, FTAG);
2624
2625 if (guid == 0) {
2626 /*
2627 * Add a new device.
2628 */
2629 nvlist_t *nvroot = make_vdev_root(NULL, aux, NULL,
2630 (ztest_opts.zo_vdev_size * 5) / 4, 0, 0, 0, 0, 1);
2631 error = spa_vdev_add(spa, nvroot);
2632 if (error != 0)
2633 fatal(0, "spa_vdev_add(%p) = %d", nvroot, error);
2634 nvlist_free(nvroot);
2635 } else {
2636 /*
2637 * Remove an existing device. Sometimes, dirty its
2638 * vdev state first to make sure we handle removal
2639 * of devices that have pending state changes.
2640 */
2641 if (ztest_random(2) == 0)
2642 (void) vdev_online(spa, guid, 0, NULL);
2643
2644 error = spa_vdev_remove(spa, guid, B_FALSE);
2645 if (error != 0 && error != EBUSY)
2646 fatal(0, "spa_vdev_remove(%llu) = %d", guid, error);
2647 }
2648
2649 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2650 }
2651
2652 /*
2653 * split a pool if it has mirror tlvdevs
2654 */
2655 /* ARGSUSED */
2656 void
2657 ztest_split_pool(ztest_ds_t *zd, uint64_t id)
2658 {
2659 ztest_shared_t *zs = ztest_shared;
2660 spa_t *spa = ztest_spa;
2661 vdev_t *rvd = spa->spa_root_vdev;
2662 nvlist_t *tree, **child, *config, *split, **schild;
2663 uint_t c, children, schildren = 0, lastlogid = 0;
2664 int error = 0;
2665
2666 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2667
2668 /* ensure we have a useable config; mirrors of raidz aren't supported */
2669 if (zs->zs_mirrors < 3 || ztest_opts.zo_raidz > 1) {
2670 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2671 return;
2672 }
2673
2674 /* clean up the old pool, if any */
2675 (void) spa_destroy("splitp");
2676
2677 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2678
2679 /* generate a config from the existing config */
2680 mutex_enter(&spa->spa_props_lock);
2681 VERIFY(nvlist_lookup_nvlist(spa->spa_config, ZPOOL_CONFIG_VDEV_TREE,
2682 &tree) == 0);
2683 mutex_exit(&spa->spa_props_lock);
2684
2685 VERIFY(nvlist_lookup_nvlist_array(tree, ZPOOL_CONFIG_CHILDREN, &child,
2686 &children) == 0);
2687
2688 schild = malloc(rvd->vdev_children * sizeof (nvlist_t *));
2689 for (c = 0; c < children; c++) {
2690 vdev_t *tvd = rvd->vdev_child[c];
2691 nvlist_t **mchild;
2692 uint_t mchildren;
2693
2694 if (tvd->vdev_islog || tvd->vdev_ops == &vdev_hole_ops) {
2695 VERIFY(nvlist_alloc(&schild[schildren], NV_UNIQUE_NAME,
2696 0) == 0);
2697 VERIFY(nvlist_add_string(schild[schildren],
2698 ZPOOL_CONFIG_TYPE, VDEV_TYPE_HOLE) == 0);
2699 VERIFY(nvlist_add_uint64(schild[schildren],
2700 ZPOOL_CONFIG_IS_HOLE, 1) == 0);
2701 if (lastlogid == 0)
2702 lastlogid = schildren;
2703 ++schildren;
2704 continue;
2705 }
2706 lastlogid = 0;
2707 VERIFY(nvlist_lookup_nvlist_array(child[c],
2708 ZPOOL_CONFIG_CHILDREN, &mchild, &mchildren) == 0);
2709 VERIFY(nvlist_dup(mchild[0], &schild[schildren++], 0) == 0);
2710 }
2711
2712 /* OK, create a config that can be used to split */
2713 VERIFY(nvlist_alloc(&split, NV_UNIQUE_NAME, 0) == 0);
2714 VERIFY(nvlist_add_string(split, ZPOOL_CONFIG_TYPE,
2715 VDEV_TYPE_ROOT) == 0);
2716 VERIFY(nvlist_add_nvlist_array(split, ZPOOL_CONFIG_CHILDREN, schild,
2717 lastlogid != 0 ? lastlogid : schildren) == 0);
2718
2719 VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, 0) == 0);
2720 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, split) == 0);
2721
2722 for (c = 0; c < schildren; c++)
2723 nvlist_free(schild[c]);
2724 free(schild);
2725 nvlist_free(split);
2726
2727 spa_config_exit(spa, SCL_VDEV, FTAG);
2728
2729 (void) rw_wrlock(&ztest_name_lock);
2730 error = spa_vdev_split_mirror(spa, "splitp", config, NULL, B_FALSE);
2731 (void) rw_unlock(&ztest_name_lock);
2732
2733 nvlist_free(config);
2734
2735 if (error == 0) {
2736 (void) printf("successful split - results:\n");
2737 mutex_enter(&spa_namespace_lock);
2738 show_pool_stats(spa);
2739 show_pool_stats(spa_lookup("splitp"));
2740 mutex_exit(&spa_namespace_lock);
2741 ++zs->zs_splits;
2742 --zs->zs_mirrors;
2743 }
2744 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2745
2746 }
2747
2748 /*
2749 * Verify that we can attach and detach devices.
2750 */
2751 /* ARGSUSED */
2752 void
2753 ztest_vdev_attach_detach(ztest_ds_t *zd, uint64_t id)
2754 {
2755 ztest_shared_t *zs = ztest_shared;
2756 spa_t *spa = ztest_spa;
2757 spa_aux_vdev_t *sav = &spa->spa_spares;
2758 vdev_t *rvd = spa->spa_root_vdev;
2759 vdev_t *oldvd, *newvd, *pvd;
2760 nvlist_t *root;
2761 uint64_t leaves;
2762 uint64_t leaf, top;
2763 uint64_t ashift = ztest_get_ashift();
2764 uint64_t oldguid, pguid;
2765 uint64_t oldsize, newsize;
2766 char oldpath[MAXPATHLEN], newpath[MAXPATHLEN];
2767 int replacing;
2768 int oldvd_has_siblings = B_FALSE;
2769 int newvd_is_spare = B_FALSE;
2770 int oldvd_is_log;
2771 int error, expected_error;
2772
2773 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2774 leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz;
2775
2776 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2777
2778 /*
2779 * Decide whether to do an attach or a replace.
2780 */
2781 replacing = ztest_random(2);
2782
2783 /*
2784 * Pick a random top-level vdev.
2785 */
2786 top = ztest_random_vdev_top(spa, B_TRUE);
2787
2788 /*
2789 * Pick a random leaf within it.
2790 */
2791 leaf = ztest_random(leaves);
2792
2793 /*
2794 * Locate this vdev.
2795 */
2796 oldvd = rvd->vdev_child[top];
2797 if (zs->zs_mirrors >= 1) {
2798 ASSERT(oldvd->vdev_ops == &vdev_mirror_ops);
2799 ASSERT(oldvd->vdev_children >= zs->zs_mirrors);
2800 oldvd = oldvd->vdev_child[leaf / ztest_opts.zo_raidz];
2801 }
2802 if (ztest_opts.zo_raidz > 1) {
2803 ASSERT(oldvd->vdev_ops == &vdev_raidz_ops);
2804 ASSERT(oldvd->vdev_children == ztest_opts.zo_raidz);
2805 oldvd = oldvd->vdev_child[leaf % ztest_opts.zo_raidz];
2806 }
2807
2808 /*
2809 * If we're already doing an attach or replace, oldvd may be a
2810 * mirror vdev -- in which case, pick a random child.
2811 */
2812 while (oldvd->vdev_children != 0) {
2813 oldvd_has_siblings = B_TRUE;
2814 ASSERT(oldvd->vdev_children >= 2);
2815 oldvd = oldvd->vdev_child[ztest_random(oldvd->vdev_children)];
2816 }
2817
2818 oldguid = oldvd->vdev_guid;
2819 oldsize = vdev_get_min_asize(oldvd);
2820 oldvd_is_log = oldvd->vdev_top->vdev_islog;
2821 (void) strcpy(oldpath, oldvd->vdev_path);
2822 pvd = oldvd->vdev_parent;
2823 pguid = pvd->vdev_guid;
2824
2825 /*
2826 * If oldvd has siblings, then half of the time, detach it.
2827 */
2828 if (oldvd_has_siblings && ztest_random(2) == 0) {
2829 spa_config_exit(spa, SCL_VDEV, FTAG);
2830 error = spa_vdev_detach(spa, oldguid, pguid, B_FALSE);
2831 if (error != 0 && error != ENODEV && error != EBUSY &&
2832 error != ENOTSUP)
2833 fatal(0, "detach (%s) returned %d", oldpath, error);
2834 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2835 return;
2836 }
2837
2838 /*
2839 * For the new vdev, choose with equal probability between the two
2840 * standard paths (ending in either 'a' or 'b') or a random hot spare.
2841 */
2842 if (sav->sav_count != 0 && ztest_random(3) == 0) {
2843 newvd = sav->sav_vdevs[ztest_random(sav->sav_count)];
2844 newvd_is_spare = B_TRUE;
2845 (void) strcpy(newpath, newvd->vdev_path);
2846 } else {
2847 (void) snprintf(newpath, sizeof (newpath), ztest_dev_template,
2848 ztest_opts.zo_dir, ztest_opts.zo_pool,
2849 top * leaves + leaf);
2850 if (ztest_random(2) == 0)
2851 newpath[strlen(newpath) - 1] = 'b';
2852 newvd = vdev_lookup_by_path(rvd, newpath);
2853 }
2854
2855 if (newvd) {
2856 newsize = vdev_get_min_asize(newvd);
2857 } else {
2858 /*
2859 * Make newsize a little bigger or smaller than oldsize.
2860 * If it's smaller, the attach should fail.
2861 * If it's larger, and we're doing a replace,
2862 * we should get dynamic LUN growth when we're done.
2863 */
2864 newsize = 10 * oldsize / (9 + ztest_random(3));
2865 }
2866
2867 /*
2868 * If pvd is not a mirror or root, the attach should fail with ENOTSUP,
2869 * unless it's a replace; in that case any non-replacing parent is OK.
2870 *
2871 * If newvd is already part of the pool, it should fail with EBUSY.
2872 *
2873 * If newvd is too small, it should fail with EOVERFLOW.
2874 */
2875 if (pvd->vdev_ops != &vdev_mirror_ops &&
2876 pvd->vdev_ops != &vdev_root_ops && (!replacing ||
2877 pvd->vdev_ops == &vdev_replacing_ops ||
2878 pvd->vdev_ops == &vdev_spare_ops))
2879 expected_error = ENOTSUP;
2880 else if (newvd_is_spare && (!replacing || oldvd_is_log))
2881 expected_error = ENOTSUP;
2882 else if (newvd == oldvd)
2883 expected_error = replacing ? 0 : EBUSY;
2884 else if (vdev_lookup_by_path(rvd, newpath) != NULL)
2885 expected_error = EBUSY;
2886 else if (newsize < oldsize)
2887 expected_error = EOVERFLOW;
2888 else if (ashift > oldvd->vdev_top->vdev_ashift)
2889 expected_error = EDOM;
2890 else
2891 expected_error = 0;
2892
2893 spa_config_exit(spa, SCL_VDEV, FTAG);
2894
2895 /*
2896 * Build the nvlist describing newpath.
2897 */
2898 root = make_vdev_root(newpath, NULL, NULL, newvd == NULL ? newsize : 0,
2899 ashift, 0, 0, 0, 1);
2900
2901 error = spa_vdev_attach(spa, oldguid, root, replacing);
2902
2903 nvlist_free(root);
2904
2905 /*
2906 * If our parent was the replacing vdev, but the replace completed,
2907 * then instead of failing with ENOTSUP we may either succeed,
2908 * fail with ENODEV, or fail with EOVERFLOW.
2909 */
2910 if (expected_error == ENOTSUP &&
2911 (error == 0 || error == ENODEV || error == EOVERFLOW))
2912 expected_error = error;
2913
2914 /*
2915 * If someone grew the LUN, the replacement may be too small.
2916 */
2917 if (error == EOVERFLOW || error == EBUSY)
2918 expected_error = error;
2919
2920 /* XXX workaround 6690467 */
2921 if (error != expected_error && expected_error != EBUSY) {
2922 fatal(0, "attach (%s %llu, %s %llu, %d) "
2923 "returned %d, expected %d",
2924 oldpath, oldsize, newpath,
2925 newsize, replacing, error, expected_error);
2926 }
2927
2928 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2929 }
2930
2931 /*
2932 * Callback function which expands the physical size of the vdev.
2933 */
2934 vdev_t *
2935 grow_vdev(vdev_t *vd, void *arg)
2936 {
2937 spa_t *spa = vd->vdev_spa;
2938 size_t *newsize = arg;
2939 size_t fsize;
2940 int fd;
2941
2942 ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
2943 ASSERT(vd->vdev_ops->vdev_op_leaf);
2944
2945 if ((fd = open(vd->vdev_path, O_RDWR)) == -1)
2946 return (vd);
2947
2948 fsize = lseek(fd, 0, SEEK_END);
2949 (void) ftruncate(fd, *newsize);
2950
2951 if (ztest_opts.zo_verbose >= 6) {
2952 (void) printf("%s grew from %lu to %lu bytes\n",
2953 vd->vdev_path, (ulong_t)fsize, (ulong_t)*newsize);
2954 }
2955 (void) close(fd);
2956 return (NULL);
2957 }
2958
2959 /*
2960 * Callback function which expands a given vdev by calling vdev_online().
2961 */
2962 /* ARGSUSED */
2963 vdev_t *
2964 online_vdev(vdev_t *vd, void *arg)
2965 {
2966 spa_t *spa = vd->vdev_spa;
2967 vdev_t *tvd = vd->vdev_top;
2968 uint64_t guid = vd->vdev_guid;
2969 uint64_t generation = spa->spa_config_generation + 1;
2970 vdev_state_t newstate = VDEV_STATE_UNKNOWN;
2971 int error;
2972
2973 ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
2974 ASSERT(vd->vdev_ops->vdev_op_leaf);
2975
2976 /* Calling vdev_online will initialize the new metaslabs */
2977 spa_config_exit(spa, SCL_STATE, spa);
2978 error = vdev_online(spa, guid, ZFS_ONLINE_EXPAND, &newstate);
2979 spa_config_enter(spa, SCL_STATE, spa, RW_READER);
2980
2981 /*
2982 * If vdev_online returned an error or the underlying vdev_open
2983 * failed then we abort the expand. The only way to know that
2984 * vdev_open fails is by checking the returned newstate.
2985 */
2986 if (error || newstate != VDEV_STATE_HEALTHY) {
2987 if (ztest_opts.zo_verbose >= 5) {
2988 (void) printf("Unable to expand vdev, state %llu, "
2989 "error %d\n", (u_longlong_t)newstate, error);
2990 }
2991 return (vd);
2992 }
2993 ASSERT3U(newstate, ==, VDEV_STATE_HEALTHY);
2994
2995 /*
2996 * Since we dropped the lock we need to ensure that we're
2997 * still talking to the original vdev. It's possible this
2998 * vdev may have been detached/replaced while we were
2999 * trying to online it.
3000 */
3001 if (generation != spa->spa_config_generation) {
3002 if (ztest_opts.zo_verbose >= 5) {
3003 (void) printf("vdev configuration has changed, "
3004 "guid %llu, state %llu, expected gen %llu, "
3005 "got gen %llu\n",
3006 (u_longlong_t)guid,
3007 (u_longlong_t)tvd->vdev_state,
3008 (u_longlong_t)generation,
3009 (u_longlong_t)spa->spa_config_generation);
3010 }
3011 return (vd);
3012 }
3013 return (NULL);
3014 }
3015
3016 /*
3017 * Traverse the vdev tree calling the supplied function.
3018 * We continue to walk the tree until we either have walked all
3019 * children or we receive a non-NULL return from the callback.
3020 * If a NULL callback is passed, then we just return back the first
3021 * leaf vdev we encounter.
3022 */
3023 vdev_t *
3024 vdev_walk_tree(vdev_t *vd, vdev_t *(*func)(vdev_t *, void *), void *arg)
3025 {
3026 if (vd->vdev_ops->vdev_op_leaf) {
3027 if (func == NULL)
3028 return (vd);
3029 else
3030 return (func(vd, arg));
3031 }
3032
3033 for (uint_t c = 0; c < vd->vdev_children; c++) {
3034 vdev_t *cvd = vd->vdev_child[c];
3035 if ((cvd = vdev_walk_tree(cvd, func, arg)) != NULL)
3036 return (cvd);
3037 }
3038 return (NULL);
3039 }
3040
3041 /*
3042 * Verify that dynamic LUN growth works as expected.
3043 */
3044 /* ARGSUSED */
3045 void
3046 ztest_vdev_LUN_growth(ztest_ds_t *zd, uint64_t id)
3047 {
3048 spa_t *spa = ztest_spa;
3049 vdev_t *vd, *tvd;
3050 metaslab_class_t *mc;
3051 metaslab_group_t *mg;
3052 size_t psize, newsize;
3053 uint64_t top;
3054 uint64_t old_class_space, new_class_space, old_ms_count, new_ms_count;
3055
3056 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
3057 spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3058
3059 top = ztest_random_vdev_top(spa, B_TRUE);
3060
3061 tvd = spa->spa_root_vdev->vdev_child[top];
3062 mg = tvd->vdev_mg;
3063 mc = mg->mg_class;
3064 old_ms_count = tvd->vdev_ms_count;
3065 old_class_space = metaslab_class_get_space(mc);
3066
3067 /*
3068 * Determine the size of the first leaf vdev associated with
3069 * our top-level device.
3070 */
3071 vd = vdev_walk_tree(tvd, NULL, NULL);
3072 ASSERT3P(vd, !=, NULL);
3073 ASSERT(vd->vdev_ops->vdev_op_leaf);
3074
3075 psize = vd->vdev_psize;
3076
3077 /*
3078 * We only try to expand the vdev if it's healthy, less than 4x its
3079 * original size, and it has a valid psize.
3080 */
3081 if (tvd->vdev_state != VDEV_STATE_HEALTHY ||
3082 psize == 0 || psize >= 4 * ztest_opts.zo_vdev_size) {
3083 spa_config_exit(spa, SCL_STATE, spa);
3084 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3085 return;
3086 }
3087 ASSERT(psize > 0);
3088 newsize = psize + psize / 8;
3089 ASSERT3U(newsize, >, psize);
3090
3091 if (ztest_opts.zo_verbose >= 6) {
3092 (void) printf("Expanding LUN %s from %lu to %lu\n",
3093 vd->vdev_path, (ulong_t)psize, (ulong_t)newsize);
3094 }
3095
3096 /*
3097 * Growing the vdev is a two step process:
3098 * 1). expand the physical size (i.e. relabel)
3099 * 2). online the vdev to create the new metaslabs
3100 */
3101 if (vdev_walk_tree(tvd, grow_vdev, &newsize) != NULL ||
3102 vdev_walk_tree(tvd, online_vdev, NULL) != NULL ||
3103 tvd->vdev_state != VDEV_STATE_HEALTHY) {
3104 if (ztest_opts.zo_verbose >= 5) {
3105 (void) printf("Could not expand LUN because "
3106 "the vdev configuration changed.\n");
3107 }
3108 spa_config_exit(spa, SCL_STATE, spa);
3109 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3110 return;
3111 }
3112
3113 spa_config_exit(spa, SCL_STATE, spa);
3114
3115 /*
3116 * Expanding the LUN will update the config asynchronously,
3117 * thus we must wait for the async thread to complete any
3118 * pending tasks before proceeding.
3119 */
3120 for (;;) {
3121 boolean_t done;
3122 mutex_enter(&spa->spa_async_lock);
3123 done = (spa->spa_async_thread == NULL && !spa->spa_async_tasks);
3124 mutex_exit(&spa->spa_async_lock);
3125 if (done)
3126 break;
3127 txg_wait_synced(spa_get_dsl(spa), 0);
3128 (void) poll(NULL, 0, 100);
3129 }
3130
3131 spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3132
3133 tvd = spa->spa_root_vdev->vdev_child[top];
3134 new_ms_count = tvd->vdev_ms_count;
3135 new_class_space = metaslab_class_get_space(mc);
3136
3137 if (tvd->vdev_mg != mg || mg->mg_class != mc) {
3138 if (ztest_opts.zo_verbose >= 5) {
3139 (void) printf("Could not verify LUN expansion due to "
3140 "intervening vdev offline or remove.\n");
3141 }
3142 spa_config_exit(spa, SCL_STATE, spa);
3143 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3144 return;
3145 }
3146
3147 /*
3148 * Make sure we were able to grow the vdev.
3149 */
3150 if (new_ms_count <= old_ms_count)
3151 fatal(0, "LUN expansion failed: ms_count %llu <= %llu\n",
3152 old_ms_count, new_ms_count);
3153
3154 /*
3155 * Make sure we were able to grow the pool.
3156 */
3157 if (new_class_space <= old_class_space)
3158 fatal(0, "LUN expansion failed: class_space %llu <= %llu\n",
3159 old_class_space, new_class_space);
3160
3161 if (ztest_opts.zo_verbose >= 5) {
3162 char oldnumbuf[NN_NUMBUF_SZ], newnumbuf[NN_NUMBUF_SZ];
3163
3164 nicenum(old_class_space, oldnumbuf, sizeof (oldnumbuf));
3165 nicenum(new_class_space, newnumbuf, sizeof (newnumbuf));
3166 (void) printf("%s grew from %s to %s\n",
3167 spa->spa_name, oldnumbuf, newnumbuf);
3168 }
3169
3170 spa_config_exit(spa, SCL_STATE, spa);
3171 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3172 }
3173
3174 /*
3175 * Verify that dmu_objset_{create,destroy,open,close} work as expected.
3176 */
3177 /* ARGSUSED */
3178 static void
3179 ztest_objset_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
3180 {
3181 /*
3182 * Create the objects common to all ztest datasets.
3183 */
3184 VERIFY(zap_create_claim(os, ZTEST_DIROBJ,
3185 DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx) == 0);
3186 }
3187
3188 static int
3189 ztest_dataset_create(char *dsname)
3190 {
3191 uint64_t zilset = ztest_random(100);
3192 int err = dmu_objset_create(dsname, DMU_OST_OTHER, 0,
3193 ztest_objset_create_cb, NULL);
3194
3195 if (err || zilset < 80)
3196 return (err);
3197
3198 if (ztest_opts.zo_verbose >= 6)
3199 (void) printf("Setting dataset %s to sync always\n", dsname);
3200 return (ztest_dsl_prop_set_uint64(dsname, ZFS_PROP_SYNC,
3201 ZFS_SYNC_ALWAYS, B_FALSE));
3202 }
3203
3204 /* ARGSUSED */
3205 static int
3206 ztest_objset_destroy_cb(const char *name, void *arg)
3207 {
3208 objset_t *os;
3209 dmu_object_info_t doi;
3210 int error;
3211
3212 /*
3213 * Verify that the dataset contains a directory object.
3214 */
3215 VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_TRUE, FTAG, &os));
3216 error = dmu_object_info(os, ZTEST_DIROBJ, &doi);
3217 if (error != ENOENT) {
3218 /* We could have crashed in the middle of destroying it */
3219 ASSERT0(error);
3220 ASSERT3U(doi.doi_type, ==, DMU_OT_ZAP_OTHER);
3221 ASSERT3S(doi.doi_physical_blocks_512, >=, 0);
3222 }
3223 dmu_objset_disown(os, FTAG);
3224
3225 /*
3226 * Destroy the dataset.
3227 */
3228 if (strchr(name, '@') != NULL) {
3229 VERIFY0(dsl_destroy_snapshot(name, B_TRUE));
3230 } else {
3231 error = dsl_destroy_head(name);
3232 /* There could be a hold on this dataset */
3233 if (error != EBUSY)
3234 ASSERT0(error);
3235 }
3236 return (0);
3237 }
3238
3239 static boolean_t
3240 ztest_snapshot_create(char *osname, uint64_t id)
3241 {
3242 char snapname[ZFS_MAX_DATASET_NAME_LEN];
3243 int error;
3244
3245 (void) snprintf(snapname, sizeof (snapname), "%llu", (u_longlong_t)id);
3246
3247 error = dmu_objset_snapshot_one(osname, snapname);
3248 if (error == ENOSPC) {
3249 ztest_record_enospc(FTAG);
3250 return (B_FALSE);
3251 }
3252 if (error != 0 && error != EEXIST) {
3253 fatal(0, "ztest_snapshot_create(%s@%s) = %d", osname,
3254 snapname, error);
3255 }
3256 return (B_TRUE);
3257 }
3258
3259 static boolean_t
3260 ztest_snapshot_destroy(char *osname, uint64_t id)
3261 {
3262 char snapname[ZFS_MAX_DATASET_NAME_LEN];
3263 int error;
3264
3265 (void) snprintf(snapname, sizeof (snapname), "%s@%llu", osname,
3266 (u_longlong_t)id);
3267
3268 error = dsl_destroy_snapshot(snapname, B_FALSE);
3269 if (error != 0 && error != ENOENT)
3270 fatal(0, "ztest_snapshot_destroy(%s) = %d", snapname, error);
3271 return (B_TRUE);
3272 }
3273
3274 /* ARGSUSED */
3275 void
3276 ztest_dmu_objset_create_destroy(ztest_ds_t *zd, uint64_t id)
3277 {
3278 ztest_ds_t zdtmp;
3279 int iters;
3280 int error;
3281 objset_t *os, *os2;
3282 char name[ZFS_MAX_DATASET_NAME_LEN];
3283 zilog_t *zilog;
3284
3285 (void) rw_rdlock(&ztest_name_lock);
3286
3287 (void) snprintf(name, sizeof (name), "%s/temp_%llu",
3288 ztest_opts.zo_pool, (u_longlong_t)id);
3289
3290 /*
3291 * If this dataset exists from a previous run, process its replay log
3292 * half of the time. If we don't replay it, then dmu_objset_destroy()
3293 * (invoked from ztest_objset_destroy_cb()) should just throw it away.
3294 */
3295 if (ztest_random(2) == 0 &&
3296 dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os) == 0) {
3297 ztest_zd_init(&zdtmp, NULL, os);
3298 zil_replay(os, &zdtmp, ztest_replay_vector);
3299 ztest_zd_fini(&zdtmp);
3300 dmu_objset_disown(os, FTAG);
3301 }
3302
3303 /*
3304 * There may be an old instance of the dataset we're about to
3305 * create lying around from a previous run. If so, destroy it
3306 * and all of its snapshots.
3307 */
3308 (void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
3309 DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
3310
3311 /*
3312 * Verify that the destroyed dataset is no longer in the namespace.
3313 */
3314 VERIFY3U(ENOENT, ==, dmu_objset_own(name, DMU_OST_OTHER, B_TRUE,
3315 FTAG, &os));
3316
3317 /*
3318 * Verify that we can create a new dataset.
3319 */
3320 error = ztest_dataset_create(name);
3321 if (error) {
3322 if (error == ENOSPC) {
3323 ztest_record_enospc(FTAG);
3324 (void) rw_unlock(&ztest_name_lock);
3325 return;
3326 }
3327 fatal(0, "dmu_objset_create(%s) = %d", name, error);
3328 }
3329
3330 VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os));
3331
3332 ztest_zd_init(&zdtmp, NULL, os);
3333
3334 /*
3335 * Open the intent log for it.
3336 */
3337 zilog = zil_open(os, ztest_get_data);
3338
3339 /*
3340 * Put some objects in there, do a little I/O to them,
3341 * and randomly take a couple of snapshots along the way.
3342 */
3343 iters = ztest_random(5);
3344 for (int i = 0; i < iters; i++) {
3345 ztest_dmu_object_alloc_free(&zdtmp, id);
3346 if (ztest_random(iters) == 0)
3347 (void) ztest_snapshot_create(name, i);
3348 }
3349
3350 /*
3351 * Verify that we cannot create an existing dataset.
3352 */
3353 VERIFY3U(EEXIST, ==,
3354 dmu_objset_create(name, DMU_OST_OTHER, 0, NULL, NULL));
3355
3356 /*
3357 * Verify that we can hold an objset that is also owned.
3358 */
3359 VERIFY3U(0, ==, dmu_objset_hold(name, FTAG, &os2));
3360 dmu_objset_rele(os2, FTAG);
3361
3362 /*
3363 * Verify that we cannot own an objset that is already owned.
3364 */
3365 VERIFY3U(EBUSY, ==,
3366 dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os2));
3367
3368 zil_close(zilog);
3369 dmu_objset_disown(os, FTAG);
3370 ztest_zd_fini(&zdtmp);
3371
3372 (void) rw_unlock(&ztest_name_lock);
3373 }
3374
3375 /*
3376 * Verify that dmu_snapshot_{create,destroy,open,close} work as expected.
3377 */
3378 void
3379 ztest_dmu_snapshot_create_destroy(ztest_ds_t *zd, uint64_t id)
3380 {
3381 (void) rw_rdlock(&ztest_name_lock);
3382 (void) ztest_snapshot_destroy(zd->zd_name, id);
3383 (void) ztest_snapshot_create(zd->zd_name, id);
3384 (void) rw_unlock(&ztest_name_lock);
3385 }
3386
3387 /*
3388 * Cleanup non-standard snapshots and clones.
3389 */
3390 void
3391 ztest_dsl_dataset_cleanup(char *osname, uint64_t id)
3392 {
3393 char snap1name[ZFS_MAX_DATASET_NAME_LEN];
3394 char clone1name[ZFS_MAX_DATASET_NAME_LEN];
3395 char snap2name[ZFS_MAX_DATASET_NAME_LEN];
3396 char clone2name[ZFS_MAX_DATASET_NAME_LEN];
3397 char snap3name[ZFS_MAX_DATASET_NAME_LEN];
3398 int error;
3399
3400 (void) snprintf(snap1name, sizeof (snap1name),
3401 "%s@s1_%llu", osname, id);
3402 (void) snprintf(clone1name, sizeof (clone1name),
3403 "%s/c1_%llu", osname, id);
3404 (void) snprintf(snap2name, sizeof (snap2name),
3405 "%s@s2_%llu", clone1name, id);
3406 (void) snprintf(clone2name, sizeof (clone2name),
3407 "%s/c2_%llu", osname, id);
3408 (void) snprintf(snap3name, sizeof (snap3name),
3409 "%s@s3_%llu", clone1name, id);
3410
3411 error = dsl_destroy_head(clone2name);
3412 if (error && error != ENOENT)
3413 fatal(0, "dsl_destroy_head(%s) = %d", clone2name, error);
3414 error = dsl_destroy_snapshot(snap3name, B_FALSE);
3415 if (error && error != ENOENT)
3416 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap3name, error);
3417 error = dsl_destroy_snapshot(snap2name, B_FALSE);
3418 if (error && error != ENOENT)
3419 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap2name, error);
3420 error = dsl_destroy_head(clone1name);
3421 if (error && error != ENOENT)
3422 fatal(0, "dsl_destroy_head(%s) = %d", clone1name, error);
3423 error = dsl_destroy_snapshot(snap1name, B_FALSE);
3424 if (error && error != ENOENT)
3425 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap1name, error);
3426 }
3427
3428 /*
3429 * Verify dsl_dataset_promote handles EBUSY
3430 */
3431 void
3432 ztest_dsl_dataset_promote_busy(ztest_ds_t *zd, uint64_t id)
3433 {
3434 objset_t *os;
3435 char snap1name[ZFS_MAX_DATASET_NAME_LEN];
3436 char clone1name[ZFS_MAX_DATASET_NAME_LEN];
3437 char snap2name[ZFS_MAX_DATASET_NAME_LEN];
3438 char clone2name[ZFS_MAX_DATASET_NAME_LEN];
3439 char snap3name[ZFS_MAX_DATASET_NAME_LEN];
3440 char *osname = zd->zd_name;
3441 int error;
3442
3443 (void) rw_rdlock(&ztest_name_lock);
3444
3445 ztest_dsl_dataset_cleanup(osname, id);
3446
3447 (void) snprintf(snap1name, sizeof (snap1name),
3448 "%s@s1_%llu", osname, id);
3449 (void) snprintf(clone1name, sizeof (clone1name),
3450 "%s/c1_%llu", osname, id);
3451 (void) snprintf(snap2name, sizeof (snap2name),
3452 "%s@s2_%llu", clone1name, id);
3453 (void) snprintf(clone2name, sizeof (clone2name),
3454 "%s/c2_%llu", osname, id);
3455 (void) snprintf(snap3name, sizeof (snap3name),
3456 "%s@s3_%llu", clone1name, id);
3457
3458 error = dmu_objset_snapshot_one(osname, strchr(snap1name, '@') + 1);
3459 if (error && error != EEXIST) {
3460 if (error == ENOSPC) {
3461 ztest_record_enospc(FTAG);
3462 goto out;
3463 }
3464 fatal(0, "dmu_take_snapshot(%s) = %d", snap1name, error);
3465 }
3466
3467 error = dmu_objset_clone(clone1name, snap1name);
3468 if (error) {
3469 if (error == ENOSPC) {
3470 ztest_record_enospc(FTAG);
3471 goto out;
3472 }
3473 fatal(0, "dmu_objset_create(%s) = %d", clone1name, error);
3474 }
3475
3476 error = dmu_objset_snapshot_one(clone1name, strchr(snap2name, '@') + 1);
3477 if (error && error != EEXIST) {
3478 if (error == ENOSPC) {
3479 ztest_record_enospc(FTAG);
3480 goto out;
3481 }
3482 fatal(0, "dmu_open_snapshot(%s) = %d", snap2name, error);
3483 }
3484
3485 error = dmu_objset_snapshot_one(clone1name, strchr(snap3name, '@') + 1);
3486 if (error && error != EEXIST) {
3487 if (error == ENOSPC) {
3488 ztest_record_enospc(FTAG);
3489 goto out;
3490 }
3491 fatal(0, "dmu_open_snapshot(%s) = %d", snap3name, error);
3492 }
3493
3494 error = dmu_objset_clone(clone2name, snap3name);
3495 if (error) {
3496 if (error == ENOSPC) {
3497 ztest_record_enospc(FTAG);
3498 goto out;
3499 }
3500 fatal(0, "dmu_objset_create(%s) = %d", clone2name, error);
3501 }
3502
3503 error = dmu_objset_own(snap2name, DMU_OST_ANY, B_TRUE, FTAG, &os);
3504 if (error)
3505 fatal(0, "dmu_objset_own(%s) = %d", snap2name, error);
3506 error = dsl_dataset_promote(clone2name, NULL);
3507 if (error == ENOSPC) {
3508 dmu_objset_disown(os, FTAG);
3509 ztest_record_enospc(FTAG);
3510 goto out;
3511 }
3512 if (error != EBUSY)
3513 fatal(0, "dsl_dataset_promote(%s), %d, not EBUSY", clone2name,
3514 error);
3515 dmu_objset_disown(os, FTAG);
3516
3517 out:
3518 ztest_dsl_dataset_cleanup(osname, id);
3519
3520 (void) rw_unlock(&ztest_name_lock);
3521 }
3522
3523 /*
3524 * Verify that dmu_object_{alloc,free} work as expected.
3525 */
3526 void
3527 ztest_dmu_object_alloc_free(ztest_ds_t *zd, uint64_t id)
3528 {
3529 ztest_od_t od[4];
3530 int batchsize = sizeof (od) / sizeof (od[0]);
3531
3532 for (int b = 0; b < batchsize; b++)
3533 ztest_od_init(&od[b], id, FTAG, b, DMU_OT_UINT64_OTHER, 0, 0);
3534
3535 /*
3536 * Destroy the previous batch of objects, create a new batch,
3537 * and do some I/O on the new objects.
3538 */
3539 if (ztest_object_init(zd, od, sizeof (od), B_TRUE) != 0)
3540 return;
3541
3542 while (ztest_random(4 * batchsize) != 0)
3543 ztest_io(zd, od[ztest_random(batchsize)].od_object,
3544 ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
3545 }
3546
3547 /*
3548 * Verify that dmu_{read,write} work as expected.
3549 */
3550 void
3551 ztest_dmu_read_write(ztest_ds_t *zd, uint64_t id)
3552 {
3553 objset_t *os = zd->zd_os;
3554 ztest_od_t od[2];
3555 dmu_tx_t *tx;
3556 int i, freeit, error;
3557 uint64_t n, s, txg;
3558 bufwad_t *packbuf, *bigbuf, *pack, *bigH, *bigT;
3559 uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
3560 uint64_t chunksize = (1000 + ztest_random(1000)) * sizeof (uint64_t);
3561 uint64_t regions = 997;
3562 uint64_t stride = 123456789ULL;
3563 uint64_t width = 40;
3564 int free_percent = 5;
3565
3566 /*
3567 * This test uses two objects, packobj and bigobj, that are always
3568 * updated together (i.e. in the same tx) so that their contents are
3569 * in sync and can be compared. Their contents relate to each other
3570 * in a simple way: packobj is a dense array of 'bufwad' structures,
3571 * while bigobj is a sparse array of the same bufwads. Specifically,
3572 * for any index n, there are three bufwads that should be identical:
3573 *
3574 * packobj, at offset n * sizeof (bufwad_t)
3575 * bigobj, at the head of the nth chunk
3576 * bigobj, at the tail of the nth chunk
3577 *
3578 * The chunk size is arbitrary. It doesn't have to be a power of two,
3579 * and it doesn't have any relation to the object blocksize.
3580 * The only requirement is that it can hold at least two bufwads.
3581 *
3582 * Normally, we write the bufwad to each of these locations.
3583 * However, free_percent of the time we instead write zeroes to
3584 * packobj and perform a dmu_free_range() on bigobj. By comparing
3585 * bigobj to packobj, we can verify that the DMU is correctly
3586 * tracking which parts of an object are allocated and free,
3587 * and that the contents of the allocated blocks are correct.
3588 */
3589
3590 /*
3591 * Read the directory info. If it's the first time, set things up.
3592 */
3593 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, chunksize);
3594 ztest_od_init(&od[1], id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize);
3595
3596 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
3597 return;
3598
3599 bigobj = od[0].od_object;
3600 packobj = od[1].od_object;
3601 chunksize = od[0].od_gen;
3602 ASSERT(chunksize == od[1].od_gen);
3603
3604 /*
3605 * Prefetch a random chunk of the big object.
3606 * Our aim here is to get some async reads in flight
3607 * for blocks that we may free below; the DMU should
3608 * handle this race correctly.
3609 */
3610 n = ztest_random(regions) * stride + ztest_random(width);
3611 s = 1 + ztest_random(2 * width - 1);
3612 dmu_prefetch(os, bigobj, 0, n * chunksize, s * chunksize,
3613 ZIO_PRIORITY_SYNC_READ);
3614
3615 /*
3616 * Pick a random index and compute the offsets into packobj and bigobj.
3617 */
3618 n = ztest_random(regions) * stride + ztest_random(width);
3619 s = 1 + ztest_random(width - 1);
3620
3621 packoff = n * sizeof (bufwad_t);
3622 packsize = s * sizeof (bufwad_t);
3623
3624 bigoff = n * chunksize;
3625 bigsize = s * chunksize;
3626
3627 packbuf = umem_alloc(packsize, UMEM_NOFAIL);
3628 bigbuf = umem_alloc(bigsize, UMEM_NOFAIL);
3629
3630 /*
3631 * free_percent of the time, free a range of bigobj rather than
3632 * overwriting it.
3633 */
3634 freeit = (ztest_random(100) < free_percent);
3635
3636 /*
3637 * Read the current contents of our objects.
3638 */
3639 error = dmu_read(os, packobj, packoff, packsize, packbuf,
3640 DMU_READ_PREFETCH);
3641 ASSERT0(error);
3642 error = dmu_read(os, bigobj, bigoff, bigsize, bigbuf,
3643 DMU_READ_PREFETCH);
3644 ASSERT0(error);
3645
3646 /*
3647 * Get a tx for the mods to both packobj and bigobj.
3648 */
3649 tx = dmu_tx_create(os);
3650
3651 dmu_tx_hold_write(tx, packobj, packoff, packsize);
3652
3653 if (freeit)
3654 dmu_tx_hold_free(tx, bigobj, bigoff, bigsize);
3655 else
3656 dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
3657
3658 /* This accounts for setting the checksum/compression. */
3659 dmu_tx_hold_bonus(tx, bigobj);
3660
3661 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3662 if (txg == 0) {
3663 umem_free(packbuf, packsize);
3664 umem_free(bigbuf, bigsize);
3665 return;
3666 }
3667
3668 enum zio_checksum cksum;
3669 do {
3670 cksum = (enum zio_checksum)
3671 ztest_random_dsl_prop(ZFS_PROP_CHECKSUM);
3672 } while (cksum >= ZIO_CHECKSUM_LEGACY_FUNCTIONS);
3673 dmu_object_set_checksum(os, bigobj, cksum, tx);
3674
3675 enum zio_compress comp;
3676 do {
3677 comp = (enum zio_compress)
3678 ztest_random_dsl_prop(ZFS_PROP_COMPRESSION);
3679 } while (comp >= ZIO_COMPRESS_LEGACY_FUNCTIONS);
3680 dmu_object_set_compress(os, bigobj, comp, tx);
3681
3682 /*
3683 * For each index from n to n + s, verify that the existing bufwad
3684 * in packobj matches the bufwads at the head and tail of the
3685 * corresponding chunk in bigobj. Then update all three bufwads
3686 * with the new values we want to write out.
3687 */
3688 for (i = 0; i < s; i++) {
3689 /* LINTED */
3690 pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
3691 /* LINTED */
3692 bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
3693 /* LINTED */
3694 bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
3695
3696 ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
3697 ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
3698
3699 if (pack->bw_txg > txg)
3700 fatal(0, "future leak: got %llx, open txg is %llx",
3701 pack->bw_txg, txg);
3702
3703 if (pack->bw_data != 0 && pack->bw_index != n + i)
3704 fatal(0, "wrong index: got %llx, wanted %llx+%llx",
3705 pack->bw_index, n, i);
3706
3707 if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
3708 fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
3709
3710 if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
3711 fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
3712
3713 if (freeit) {
3714 bzero(pack, sizeof (bufwad_t));
3715 } else {
3716 pack->bw_index = n + i;
3717 pack->bw_txg = txg;
3718 pack->bw_data = 1 + ztest_random(-2ULL);
3719 }
3720 *bigH = *pack;
3721 *bigT = *pack;
3722 }
3723
3724 /*
3725 * We've verified all the old bufwads, and made new ones.
3726 * Now write them out.
3727 */
3728 dmu_write(os, packobj, packoff, packsize, packbuf, tx);
3729
3730 if (freeit) {
3731 if (ztest_opts.zo_verbose >= 7) {
3732 (void) printf("freeing offset %llx size %llx"
3733 " txg %llx\n",
3734 (u_longlong_t)bigoff,
3735 (u_longlong_t)bigsize,
3736 (u_longlong_t)txg);
3737 }
3738 VERIFY(0 == dmu_free_range(os, bigobj, bigoff, bigsize, tx));
3739 } else {
3740 if (ztest_opts.zo_verbose >= 7) {
3741 (void) printf("writing offset %llx size %llx"
3742 " txg %llx\n",
3743 (u_longlong_t)bigoff,
3744 (u_longlong_t)bigsize,
3745 (u_longlong_t)txg);
3746 }
3747 dmu_write(os, bigobj, bigoff, bigsize, bigbuf, tx);
3748 }
3749
3750 dmu_tx_commit(tx);
3751
3752 /*
3753 * Sanity check the stuff we just wrote.
3754 */
3755 {
3756 void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
3757 void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
3758
3759 VERIFY(0 == dmu_read(os, packobj, packoff,
3760 packsize, packcheck, DMU_READ_PREFETCH));
3761 VERIFY(0 == dmu_read(os, bigobj, bigoff,
3762 bigsize, bigcheck, DMU_READ_PREFETCH));
3763
3764 ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
3765 ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
3766
3767 umem_free(packcheck, packsize);
3768 umem_free(bigcheck, bigsize);
3769 }
3770
3771 umem_free(packbuf, packsize);
3772 umem_free(bigbuf, bigsize);
3773 }
3774
3775 void
3776 compare_and_update_pbbufs(uint64_t s, bufwad_t *packbuf, bufwad_t *bigbuf,
3777 uint64_t bigsize, uint64_t n, uint64_t chunksize, uint64_t txg)
3778 {
3779 uint64_t i;
3780 bufwad_t *pack;
3781 bufwad_t *bigH;
3782 bufwad_t *bigT;
3783
3784 /*
3785 * For each index from n to n + s, verify that the existing bufwad
3786 * in packobj matches the bufwads at the head and tail of the
3787 * corresponding chunk in bigobj. Then update all three bufwads
3788 * with the new values we want to write out.
3789 */
3790 for (i = 0; i < s; i++) {
3791 /* LINTED */
3792 pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
3793 /* LINTED */
3794 bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
3795 /* LINTED */
3796 bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
3797
3798 ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
3799 ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
3800
3801 if (pack->bw_txg > txg)
3802 fatal(0, "future leak: got %llx, open txg is %llx",
3803 pack->bw_txg, txg);
3804
3805 if (pack->bw_data != 0 && pack->bw_index != n + i)
3806 fatal(0, "wrong index: got %llx, wanted %llx+%llx",
3807 pack->bw_index, n, i);
3808
3809 if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
3810 fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
3811
3812 if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
3813 fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
3814
3815 pack->bw_index = n + i;
3816 pack->bw_txg = txg;
3817 pack->bw_data = 1 + ztest_random(-2ULL);
3818
3819 *bigH = *pack;
3820 *bigT = *pack;
3821 }
3822 }
3823
3824 void
3825 ztest_dmu_read_write_zcopy(ztest_ds_t *zd, uint64_t id)
3826 {
3827 objset_t *os = zd->zd_os;
3828 ztest_od_t od[2];
3829 dmu_tx_t *tx;
3830 uint64_t i;
3831 int error;
3832 uint64_t n, s, txg;
3833 bufwad_t *packbuf, *bigbuf;
3834 uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
3835 uint64_t blocksize = ztest_random_blocksize();
3836 uint64_t chunksize = blocksize;
3837 uint64_t regions = 997;
3838 uint64_t stride = 123456789ULL;
3839 uint64_t width = 9;
3840 dmu_buf_t *bonus_db;
3841 arc_buf_t **bigbuf_arcbufs;
3842 dmu_object_info_t doi;
3843
3844 /*
3845 * This test uses two objects, packobj and bigobj, that are always
3846 * updated together (i.e. in the same tx) so that their contents are
3847 * in sync and can be compared. Their contents relate to each other
3848 * in a simple way: packobj is a dense array of 'bufwad' structures,
3849 * while bigobj is a sparse array of the same bufwads. Specifically,
3850 * for any index n, there are three bufwads that should be identical:
3851 *
3852 * packobj, at offset n * sizeof (bufwad_t)
3853 * bigobj, at the head of the nth chunk
3854 * bigobj, at the tail of the nth chunk
3855 *
3856 * The chunk size is set equal to bigobj block size so that
3857 * dmu_assign_arcbuf() can be tested for object updates.
3858 */
3859
3860 /*
3861 * Read the directory info. If it's the first time, set things up.
3862 */
3863 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
3864 ztest_od_init(&od[1], id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize);
3865
3866 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
3867 return;
3868
3869 bigobj = od[0].od_object;
3870 packobj = od[1].od_object;
3871 blocksize = od[0].od_blocksize;
3872 chunksize = blocksize;
3873 ASSERT(chunksize == od[1].od_gen);
3874
3875 VERIFY(dmu_object_info(os, bigobj, &doi) == 0);
3876 VERIFY(ISP2(doi.doi_data_block_size));
3877 VERIFY(chunksize == doi.doi_data_block_size);
3878 VERIFY(chunksize >= 2 * sizeof (bufwad_t));
3879
3880 /*
3881 * Pick a random index and compute the offsets into packobj and bigobj.
3882 */
3883 n = ztest_random(regions) * stride + ztest_random(width);
3884 s = 1 + ztest_random(width - 1);
3885
3886 packoff = n * sizeof (bufwad_t);
3887 packsize = s * sizeof (bufwad_t);
3888
3889 bigoff = n * chunksize;
3890 bigsize = s * chunksize;
3891
3892 packbuf = umem_zalloc(packsize, UMEM_NOFAIL);
3893 bigbuf = umem_zalloc(bigsize, UMEM_NOFAIL);
3894
3895 VERIFY3U(0, ==, dmu_bonus_hold(os, bigobj, FTAG, &bonus_db));
3896
3897 bigbuf_arcbufs = umem_zalloc(2 * s * sizeof (arc_buf_t *), UMEM_NOFAIL);
3898
3899 /*
3900 * Iteration 0 test zcopy for DB_UNCACHED dbufs.
3901 * Iteration 1 test zcopy to already referenced dbufs.
3902 * Iteration 2 test zcopy to dirty dbuf in the same txg.
3903 * Iteration 3 test zcopy to dbuf dirty in previous txg.
3904 * Iteration 4 test zcopy when dbuf is no longer dirty.
3905 * Iteration 5 test zcopy when it can't be done.
3906 * Iteration 6 one more zcopy write.
3907 */
3908 for (i = 0; i < 7; i++) {
3909 uint64_t j;
3910 uint64_t off;
3911
3912 /*
3913 * In iteration 5 (i == 5) use arcbufs
3914 * that don't match bigobj blksz to test
3915 * dmu_assign_arcbuf() when it can't directly
3916 * assign an arcbuf to a dbuf.
3917 */
3918 for (j = 0; j < s; j++) {
3919 if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) {
3920 bigbuf_arcbufs[j] =
3921 dmu_request_arcbuf(bonus_db, chunksize);
3922 } else {
3923 bigbuf_arcbufs[2 * j] =
3924 dmu_request_arcbuf(bonus_db, chunksize / 2);
3925 bigbuf_arcbufs[2 * j + 1] =
3926 dmu_request_arcbuf(bonus_db, chunksize / 2);
3927 }
3928 }
3929
3930 /*
3931 * Get a tx for the mods to both packobj and bigobj.
3932 */
3933 tx = dmu_tx_create(os);
3934
3935 dmu_tx_hold_write(tx, packobj, packoff, packsize);
3936 dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
3937
3938 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3939 if (txg == 0) {
3940 umem_free(packbuf, packsize);
3941 umem_free(bigbuf, bigsize);
3942 for (j = 0; j < s; j++) {
3943 if (i != 5 ||
3944 chunksize < (SPA_MINBLOCKSIZE * 2)) {
3945 dmu_return_arcbuf(bigbuf_arcbufs[j]);
3946 } else {
3947 dmu_return_arcbuf(
3948 bigbuf_arcbufs[2 * j]);
3949 dmu_return_arcbuf(
3950 bigbuf_arcbufs[2 * j + 1]);
3951 }
3952 }
3953 umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
3954 dmu_buf_rele(bonus_db, FTAG);
3955 return;
3956 }
3957
3958 /*
3959 * 50% of the time don't read objects in the 1st iteration to
3960 * test dmu_assign_arcbuf() for the case when there're no
3961 * existing dbufs for the specified offsets.
3962 */
3963 if (i != 0 || ztest_random(2) != 0) {
3964 error = dmu_read(os, packobj, packoff,
3965 packsize, packbuf, DMU_READ_PREFETCH);
3966 ASSERT0(error);
3967 error = dmu_read(os, bigobj, bigoff, bigsize,
3968 bigbuf, DMU_READ_PREFETCH);
3969 ASSERT0(error);
3970 }
3971 compare_and_update_pbbufs(s, packbuf, bigbuf, bigsize,
3972 n, chunksize, txg);
3973
3974 /*
3975 * We've verified all the old bufwads, and made new ones.
3976 * Now write them out.
3977 */
3978 dmu_write(os, packobj, packoff, packsize, packbuf, tx);
3979 if (ztest_opts.zo_verbose >= 7) {
3980 (void) printf("writing offset %llx size %llx"
3981 " txg %llx\n",
3982 (u_longlong_t)bigoff,
3983 (u_longlong_t)bigsize,
3984 (u_longlong_t)txg);
3985 }
3986 for (off = bigoff, j = 0; j < s; j++, off += chunksize) {
3987 dmu_buf_t *dbt;
3988 if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) {
3989 bcopy((caddr_t)bigbuf + (off - bigoff),
3990 bigbuf_arcbufs[j]->b_data, chunksize);
3991 } else {
3992 bcopy((caddr_t)bigbuf + (off - bigoff),
3993 bigbuf_arcbufs[2 * j]->b_data,
3994 chunksize / 2);
3995 bcopy((caddr_t)bigbuf + (off - bigoff) +
3996 chunksize / 2,
3997 bigbuf_arcbufs[2 * j + 1]->b_data,
3998 chunksize / 2);
3999 }
4000
4001 if (i == 1) {
4002 VERIFY(dmu_buf_hold(os, bigobj, off,
4003 FTAG, &dbt, DMU_READ_NO_PREFETCH) == 0);
4004 }
4005 if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) {
4006 dmu_assign_arcbuf(bonus_db, off,
4007 bigbuf_arcbufs[j], tx);
4008 } else {
4009 dmu_assign_arcbuf(bonus_db, off,
4010 bigbuf_arcbufs[2 * j], tx);
4011 dmu_assign_arcbuf(bonus_db,
4012 off + chunksize / 2,
4013 bigbuf_arcbufs[2 * j + 1], tx);
4014 }
4015 if (i == 1) {
4016 dmu_buf_rele(dbt, FTAG);
4017 }
4018 }
4019 dmu_tx_commit(tx);
4020
4021 /*
4022 * Sanity check the stuff we just wrote.
4023 */
4024 {
4025 void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
4026 void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
4027
4028 VERIFY(0 == dmu_read(os, packobj, packoff,
4029 packsize, packcheck, DMU_READ_PREFETCH));
4030 VERIFY(0 == dmu_read(os, bigobj, bigoff,
4031 bigsize, bigcheck, DMU_READ_PREFETCH));
4032
4033 ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
4034 ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
4035
4036 umem_free(packcheck, packsize);
4037 umem_free(bigcheck, bigsize);
4038 }
4039 if (i == 2) {
4040 txg_wait_open(dmu_objset_pool(os), 0);
4041 } else if (i == 3) {
4042 txg_wait_synced(dmu_objset_pool(os), 0);
4043 }
4044 }
4045
4046 dmu_buf_rele(bonus_db, FTAG);
4047 umem_free(packbuf, packsize);
4048 umem_free(bigbuf, bigsize);
4049 umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
4050 }
4051
4052 /* ARGSUSED */
4053 void
4054 ztest_dmu_write_parallel(ztest_ds_t *zd, uint64_t id)
4055 {
4056 ztest_od_t od[1];
4057 uint64_t offset = (1ULL << (ztest_random(20) + 43)) +
4058 (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
4059
4060 /*
4061 * Have multiple threads write to large offsets in an object
4062 * to verify that parallel writes to an object -- even to the
4063 * same blocks within the object -- doesn't cause any trouble.
4064 */
4065 ztest_od_init(&od[0], ID_PARALLEL, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
4066
4067 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4068 return;
4069
4070 while (ztest_random(10) != 0)
4071 ztest_io(zd, od[0].od_object, offset);
4072 }
4073
4074 void
4075 ztest_dmu_prealloc(ztest_ds_t *zd, uint64_t id)
4076 {
4077 ztest_od_t od[1];
4078 uint64_t offset = (1ULL << (ztest_random(4) + SPA_MAXBLOCKSHIFT)) +
4079 (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
4080 uint64_t count = ztest_random(20) + 1;
4081 uint64_t blocksize = ztest_random_blocksize();
4082 void *data;
4083
4084 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
4085
4086 if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4087 return;
4088
4089 if (ztest_truncate(zd, od[0].od_object, offset, count * blocksize) != 0)
4090 return;
4091
4092 ztest_prealloc(zd, od[0].od_object, offset, count * blocksize);
4093
4094 data = umem_zalloc(blocksize, UMEM_NOFAIL);
4095
4096 while (ztest_random(count) != 0) {
4097 uint64_t randoff = offset + (ztest_random(count) * blocksize);
4098 if (ztest_write(zd, od[0].od_object, randoff, blocksize,
4099 data) != 0)
4100 break;
4101 while (ztest_random(4) != 0)
4102 ztest_io(zd, od[0].od_object, randoff);
4103 }
4104
4105 umem_free(data, blocksize);
4106 }
4107
4108 /*
4109 * Verify that zap_{create,destroy,add,remove,update} work as expected.
4110 */
4111 #define ZTEST_ZAP_MIN_INTS 1
4112 #define ZTEST_ZAP_MAX_INTS 4
4113 #define ZTEST_ZAP_MAX_PROPS 1000
4114
4115 void
4116 ztest_zap(ztest_ds_t *zd, uint64_t id)
4117 {
4118 objset_t *os = zd->zd_os;
4119 ztest_od_t od[1];
4120 uint64_t object;
4121 uint64_t txg, last_txg;
4122 uint64_t value[ZTEST_ZAP_MAX_INTS];
4123 uint64_t zl_ints, zl_intsize, prop;
4124 int i, ints;
4125 dmu_tx_t *tx;
4126 char propname[100], txgname[100];
4127 int error;
4128 char *hc[2] = { "s.acl.h", ".s.open.h.hyLZlg" };
4129
4130 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0);
4131
4132 if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4133 return;
4134
4135 object = od[0].od_object;
4136
4137 /*
4138 * Generate a known hash collision, and verify that
4139 * we can lookup and remove both entries.
4140 */
4141 tx = dmu_tx_create(os);
4142 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4143 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4144 if (txg == 0)
4145 return;
4146 for (i = 0; i < 2; i++) {
4147 value[i] = i;
4148 VERIFY3U(0, ==, zap_add(os, object, hc[i], sizeof (uint64_t),
4149 1, &value[i], tx));
4150 }
4151 for (i = 0; i < 2; i++) {
4152 VERIFY3U(EEXIST, ==, zap_add(os, object, hc[i],
4153 sizeof (uint64_t), 1, &value[i], tx));
4154 VERIFY3U(0, ==,
4155 zap_length(os, object, hc[i], &zl_intsize, &zl_ints));
4156 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4157 ASSERT3U(zl_ints, ==, 1);
4158 }
4159 for (i = 0; i < 2; i++) {
4160 VERIFY3U(0, ==, zap_remove(os, object, hc[i], tx));
4161 }
4162 dmu_tx_commit(tx);
4163
4164 /*
4165 * Generate a buch of random entries.
4166 */
4167 ints = MAX(ZTEST_ZAP_MIN_INTS, object % ZTEST_ZAP_MAX_INTS);
4168
4169 prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
4170 (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
4171 (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
4172 bzero(value, sizeof (value));
4173 last_txg = 0;
4174
4175 /*
4176 * If these zap entries already exist, validate their contents.
4177 */
4178 error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
4179 if (error == 0) {
4180 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4181 ASSERT3U(zl_ints, ==, 1);
4182
4183 VERIFY(zap_lookup(os, object, txgname, zl_intsize,
4184 zl_ints, &last_txg) == 0);
4185
4186 VERIFY(zap_length(os, object, propname, &zl_intsize,
4187 &zl_ints) == 0);
4188
4189 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4190 ASSERT3U(zl_ints, ==, ints);
4191
4192 VERIFY(zap_lookup(os, object, propname, zl_intsize,
4193 zl_ints, value) == 0);
4194
4195 for (i = 0; i < ints; i++) {
4196 ASSERT3U(value[i], ==, last_txg + object + i);
4197 }
4198 } else {
4199 ASSERT3U(error, ==, ENOENT);
4200 }
4201
4202 /*
4203 * Atomically update two entries in our zap object.
4204 * The first is named txg_%llu, and contains the txg
4205 * in which the property was last updated. The second
4206 * is named prop_%llu, and the nth element of its value
4207 * should be txg + object + n.
4208 */
4209 tx = dmu_tx_create(os);
4210 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4211 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4212 if (txg == 0)
4213 return;
4214
4215 if (last_txg > txg)
4216 fatal(0, "zap future leak: old %llu new %llu", last_txg, txg);
4217
4218 for (i = 0; i < ints; i++)
4219 value[i] = txg + object + i;
4220
4221 VERIFY3U(0, ==, zap_update(os, object, txgname, sizeof (uint64_t),
4222 1, &txg, tx));
4223 VERIFY3U(0, ==, zap_update(os, object, propname, sizeof (uint64_t),
4224 ints, value, tx));
4225
4226 dmu_tx_commit(tx);
4227
4228 /*
4229 * Remove a random pair of entries.
4230 */
4231 prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
4232 (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
4233 (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
4234
4235 error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
4236
4237 if (error == ENOENT)
4238 return;
4239
4240 ASSERT0(error);
4241
4242 tx = dmu_tx_create(os);
4243 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4244 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4245 if (txg == 0)
4246 return;
4247 VERIFY3U(0, ==, zap_remove(os, object, txgname, tx));
4248 VERIFY3U(0, ==, zap_remove(os, object, propname, tx));
4249 dmu_tx_commit(tx);
4250 }
4251
4252 /*
4253 * Testcase to test the upgrading of a microzap to fatzap.
4254 */
4255 void
4256 ztest_fzap(ztest_ds_t *zd, uint64_t id)
4257 {
4258 objset_t *os = zd->zd_os;
4259 ztest_od_t od[1];
4260 uint64_t object, txg;
4261
4262 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0);
4263
4264 if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4265 return;
4266
4267 object = od[0].od_object;
4268
4269 /*
4270 * Add entries to this ZAP and make sure it spills over
4271 * and gets upgraded to a fatzap. Also, since we are adding
4272 * 2050 entries we should see ptrtbl growth and leaf-block split.
4273 */
4274 for (int i = 0; i < 2050; i++) {
4275 char name[ZFS_MAX_DATASET_NAME_LEN];
4276 uint64_t value = i;
4277 dmu_tx_t *tx;
4278 int error;
4279
4280 (void) snprintf(name, sizeof (name), "fzap-%llu-%llu",
4281 id, value);
4282
4283 tx = dmu_tx_create(os);
4284 dmu_tx_hold_zap(tx, object, B_TRUE, name);
4285 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4286 if (txg == 0)
4287 return;
4288 error = zap_add(os, object, name, sizeof (uint64_t), 1,
4289 &value, tx);
4290 ASSERT(error == 0 || error == EEXIST);
4291 dmu_tx_commit(tx);
4292 }
4293 }
4294
4295 /* ARGSUSED */
4296 void
4297 ztest_zap_parallel(ztest_ds_t *zd, uint64_t id)
4298 {
4299 objset_t *os = zd->zd_os;
4300 ztest_od_t od[1];
4301 uint64_t txg, object, count, wsize, wc, zl_wsize, zl_wc;
4302 dmu_tx_t *tx;
4303 int i, namelen, error;
4304 int micro = ztest_random(2);
4305 char name[20], string_value[20];
4306 void *data;
4307
4308 ztest_od_init(&od[0], ID_PARALLEL, FTAG, micro, DMU_OT_ZAP_OTHER, 0, 0);
4309
4310 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4311 return;
4312
4313 object = od[0].od_object;
4314
4315 /*
4316 * Generate a random name of the form 'xxx.....' where each
4317 * x is a random printable character and the dots are dots.
4318 * There are 94 such characters, and the name length goes from
4319 * 6 to 20, so there are 94^3 * 15 = 12,458,760 possible names.
4320 */
4321 namelen = ztest_random(sizeof (name) - 5) + 5 + 1;
4322
4323 for (i = 0; i < 3; i++)
4324 name[i] = '!' + ztest_random('~' - '!' + 1);
4325 for (; i < namelen - 1; i++)
4326 name[i] = '.';
4327 name[i] = '\0';
4328
4329 if ((namelen & 1) || micro) {
4330 wsize = sizeof (txg);
4331 wc = 1;
4332 data = &txg;
4333 } else {
4334 wsize = 1;
4335 wc = namelen;
4336 data = string_value;
4337 }
4338
4339 count = -1ULL;
4340 VERIFY0(zap_count(os, object, &count));
4341 ASSERT(count != -1ULL);
4342
4343 /*
4344 * Select an operation: length, lookup, add, update, remove.
4345 */
4346 i = ztest_random(5);
4347
4348 if (i >= 2) {
4349 tx = dmu_tx_create(os);
4350 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4351 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4352 if (txg == 0)
4353 return;
4354 bcopy(name, string_value, namelen);
4355 } else {
4356 tx = NULL;
4357 txg = 0;
4358 bzero(string_value, namelen);
4359 }
4360
4361 switch (i) {
4362
4363 case 0:
4364 error = zap_length(os, object, name, &zl_wsize, &zl_wc);
4365 if (error == 0) {
4366 ASSERT3U(wsize, ==, zl_wsize);
4367 ASSERT3U(wc, ==, zl_wc);
4368 } else {
4369 ASSERT3U(error, ==, ENOENT);
4370 }
4371 break;
4372
4373 case 1:
4374 error = zap_lookup(os, object, name, wsize, wc, data);
4375 if (error == 0) {
4376 if (data == string_value &&
4377 bcmp(name, data, namelen) != 0)
4378 fatal(0, "name '%s' != val '%s' len %d",
4379 name, data, namelen);
4380 } else {
4381 ASSERT3U(error, ==, ENOENT);
4382 }
4383 break;
4384
4385 case 2:
4386 error = zap_add(os, object, name, wsize, wc, data, tx);
4387 ASSERT(error == 0 || error == EEXIST);
4388 break;
4389
4390 case 3:
4391 VERIFY(zap_update(os, object, name, wsize, wc, data, tx) == 0);
4392 break;
4393
4394 case 4:
4395 error = zap_remove(os, object, name, tx);
4396 ASSERT(error == 0 || error == ENOENT);
4397 break;
4398 }
4399
4400 if (tx != NULL)
4401 dmu_tx_commit(tx);
4402 }
4403
4404 /*
4405 * Commit callback data.
4406 */
4407 typedef struct ztest_cb_data {
4408 list_node_t zcd_node;
4409 uint64_t zcd_txg;
4410 int zcd_expected_err;
4411 boolean_t zcd_added;
4412 boolean_t zcd_called;
4413 spa_t *zcd_spa;
4414 } ztest_cb_data_t;
4415
4416 /* This is the actual commit callback function */
4417 static void
4418 ztest_commit_callback(void *arg, int error)
4419 {
4420 ztest_cb_data_t *data = arg;
4421 uint64_t synced_txg;
4422
4423 VERIFY(data != NULL);
4424 VERIFY3S(data->zcd_expected_err, ==, error);
4425 VERIFY(!data->zcd_called);
4426
4427 synced_txg = spa_last_synced_txg(data->zcd_spa);
4428 if (data->zcd_txg > synced_txg)
4429 fatal(0, "commit callback of txg %" PRIu64 " called prematurely"
4430 ", last synced txg = %" PRIu64 "\n", data->zcd_txg,
4431 synced_txg);
4432
4433 data->zcd_called = B_TRUE;
4434
4435 if (error == ECANCELED) {
4436 ASSERT0(data->zcd_txg);
4437 ASSERT(!data->zcd_added);
4438
4439 /*
4440 * The private callback data should be destroyed here, but
4441 * since we are going to check the zcd_called field after
4442 * dmu_tx_abort(), we will destroy it there.
4443 */
4444 return;
4445 }
4446
4447 /* Was this callback added to the global callback list? */
4448 if (!data->zcd_added)
4449 goto out;
4450
4451 ASSERT3U(data->zcd_txg, !=, 0);
4452
4453 /* Remove our callback from the list */
4454 (void) mutex_lock(&zcl.zcl_callbacks_lock);
4455 list_remove(&zcl.zcl_callbacks, data);
4456 (void) mutex_unlock(&zcl.zcl_callbacks_lock);
4457
4458 out:
4459 umem_free(data, sizeof (ztest_cb_data_t));
4460 }
4461
4462 /* Allocate and initialize callback data structure */
4463 static ztest_cb_data_t *
4464 ztest_create_cb_data(objset_t *os, uint64_t txg)
4465 {
4466 ztest_cb_data_t *cb_data;
4467
4468 cb_data = umem_zalloc(sizeof (ztest_cb_data_t), UMEM_NOFAIL);
4469
4470 cb_data->zcd_txg = txg;
4471 cb_data->zcd_spa = dmu_objset_spa(os);
4472
4473 return (cb_data);
4474 }
4475
4476 /*
4477 * If a number of txgs equal to this threshold have been created after a commit
4478 * callback has been registered but not called, then we assume there is an
4479 * implementation bug.
4480 */
4481 #define ZTEST_COMMIT_CALLBACK_THRESH (TXG_CONCURRENT_STATES + 2)
4482
4483 /*
4484 * Commit callback test.
4485 */
4486 void
4487 ztest_dmu_commit_callbacks(ztest_ds_t *zd, uint64_t id)
4488 {
4489 objset_t *os = zd->zd_os;
4490 ztest_od_t od[1];
4491 dmu_tx_t *tx;
4492 ztest_cb_data_t *cb_data[3], *tmp_cb;
4493 uint64_t old_txg, txg;
4494 int i, error;
4495
4496 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
4497
4498 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4499 return;
4500
4501 tx = dmu_tx_create(os);
4502
4503 cb_data[0] = ztest_create_cb_data(os, 0);
4504 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[0]);
4505
4506 dmu_tx_hold_write(tx, od[0].od_object, 0, sizeof (uint64_t));
4507
4508 /* Every once in a while, abort the transaction on purpose */
4509 if (ztest_random(100) == 0)
4510 error = -1;
4511
4512 if (!error)
4513 error = dmu_tx_assign(tx, TXG_NOWAIT);
4514
4515 txg = error ? 0 : dmu_tx_get_txg(tx);
4516
4517 cb_data[0]->zcd_txg = txg;
4518 cb_data[1] = ztest_create_cb_data(os, txg);
4519 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[1]);
4520
4521 if (error) {
4522 /*
4523 * It's not a strict requirement to call the registered
4524 * callbacks from inside dmu_tx_abort(), but that's what
4525 * it's supposed to happen in the current implementation
4526 * so we will check for that.
4527 */
4528 for (i = 0; i < 2; i++) {
4529 cb_data[i]->zcd_expected_err = ECANCELED;
4530 VERIFY(!cb_data[i]->zcd_called);
4531 }
4532
4533 dmu_tx_abort(tx);
4534
4535 for (i = 0; i < 2; i++) {
4536 VERIFY(cb_data[i]->zcd_called);
4537 umem_free(cb_data[i], sizeof (ztest_cb_data_t));
4538 }
4539
4540 return;
4541 }
4542
4543 cb_data[2] = ztest_create_cb_data(os, txg);
4544 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[2]);
4545
4546 /*
4547 * Read existing data to make sure there isn't a future leak.
4548 */
4549 VERIFY(0 == dmu_read(os, od[0].od_object, 0, sizeof (uint64_t),
4550 &old_txg, DMU_READ_PREFETCH));
4551
4552 if (old_txg > txg)
4553 fatal(0, "future leak: got %" PRIu64 ", open txg is %" PRIu64,
4554 old_txg, txg);
4555
4556 dmu_write(os, od[0].od_object, 0, sizeof (uint64_t), &txg, tx);
4557
4558 (void) mutex_lock(&zcl.zcl_callbacks_lock);
4559
4560 /*
4561 * Since commit callbacks don't have any ordering requirement and since
4562 * it is theoretically possible for a commit callback to be called
4563 * after an arbitrary amount of time has elapsed since its txg has been
4564 * synced, it is difficult to reliably determine whether a commit
4565 * callback hasn't been called due to high load or due to a flawed
4566 * implementation.
4567 *
4568 * In practice, we will assume that if after a certain number of txgs a
4569 * commit callback hasn't been called, then most likely there's an
4570 * implementation bug..
4571 */
4572 tmp_cb = list_head(&zcl.zcl_callbacks);
4573 if (tmp_cb != NULL &&
4574 (txg - ZTEST_COMMIT_CALLBACK_THRESH) > tmp_cb->zcd_txg) {
4575 fatal(0, "Commit callback threshold exceeded, oldest txg: %"
4576 PRIu64 ", open txg: %" PRIu64 "\n", tmp_cb->zcd_txg, txg);
4577 }
4578
4579 /*
4580 * Let's find the place to insert our callbacks.
4581 *
4582 * Even though the list is ordered by txg, it is possible for the
4583 * insertion point to not be the end because our txg may already be
4584 * quiescing at this point and other callbacks in the open txg
4585 * (from other objsets) may have sneaked in.
4586 */
4587 tmp_cb = list_tail(&zcl.zcl_callbacks);
4588 while (tmp_cb != NULL && tmp_cb->zcd_txg > txg)
4589 tmp_cb = list_prev(&zcl.zcl_callbacks, tmp_cb);
4590
4591 /* Add the 3 callbacks to the list */
4592 for (i = 0; i < 3; i++) {
4593 if (tmp_cb == NULL)
4594 list_insert_head(&zcl.zcl_callbacks, cb_data[i]);
4595 else
4596 list_insert_after(&zcl.zcl_callbacks, tmp_cb,
4597 cb_data[i]);
4598
4599 cb_data[i]->zcd_added = B_TRUE;
4600 VERIFY(!cb_data[i]->zcd_called);
4601
4602 tmp_cb = cb_data[i];
4603 }
4604
4605 (void) mutex_unlock(&zcl.zcl_callbacks_lock);
4606
4607 dmu_tx_commit(tx);
4608 }
4609
4610 /* ARGSUSED */
4611 void
4612 ztest_dsl_prop_get_set(ztest_ds_t *zd, uint64_t id)
4613 {
4614 zfs_prop_t proplist[] = {
4615 ZFS_PROP_CHECKSUM,
4616 ZFS_PROP_COMPRESSION,
4617 ZFS_PROP_COPIES,
4618 ZFS_PROP_DEDUP
4619 };
4620
4621 (void) rw_rdlock(&ztest_name_lock);
4622
4623 for (int p = 0; p < sizeof (proplist) / sizeof (proplist[0]); p++)
4624 (void) ztest_dsl_prop_set_uint64(zd->zd_name, proplist[p],
4625 ztest_random_dsl_prop(proplist[p]), (int)ztest_random(2));
4626
4627 (void) rw_unlock(&ztest_name_lock);
4628 }
4629
4630 /* ARGSUSED */
4631 void
4632 ztest_spa_prop_get_set(ztest_ds_t *zd, uint64_t id)
4633 {
4634 nvlist_t *props = NULL;
4635
4636 (void) rw_rdlock(&ztest_name_lock);
4637
4638 (void) ztest_spa_prop_set_uint64(ZPOOL_PROP_DEDUPDITTO,
4639 ZIO_DEDUPDITTO_MIN + ztest_random(ZIO_DEDUPDITTO_MIN));
4640
4641 VERIFY0(spa_prop_get(ztest_spa, &props));
4642
4643 if (ztest_opts.zo_verbose >= 6)
4644 dump_nvlist(props, 4);
4645
4646 nvlist_free(props);
4647
4648 (void) rw_unlock(&ztest_name_lock);
4649 }
4650
4651 static int
4652 user_release_one(const char *snapname, const char *holdname)
4653 {
4654 nvlist_t *snaps, *holds;
4655 int error;
4656
4657 snaps = fnvlist_alloc();
4658 holds = fnvlist_alloc();
4659 fnvlist_add_boolean(holds, holdname);
4660 fnvlist_add_nvlist(snaps, snapname, holds);
4661 fnvlist_free(holds);
4662 error = dsl_dataset_user_release(snaps, NULL);
4663 fnvlist_free(snaps);
4664 return (error);
4665 }
4666
4667 /*
4668 * Test snapshot hold/release and deferred destroy.
4669 */
4670 void
4671 ztest_dmu_snapshot_hold(ztest_ds_t *zd, uint64_t id)
4672 {
4673 int error;
4674 objset_t *os = zd->zd_os;
4675 objset_t *origin;
4676 char snapname[100];
4677 char fullname[100];
4678 char clonename[100];
4679 char tag[100];
4680 char osname[ZFS_MAX_DATASET_NAME_LEN];
4681 nvlist_t *holds;
4682
4683 (void) rw_rdlock(&ztest_name_lock);
4684
4685 dmu_objset_name(os, osname);
4686
4687 (void) snprintf(snapname, sizeof (snapname), "sh1_%llu", id);
4688 (void) snprintf(fullname, sizeof (fullname), "%s@%s", osname, snapname);
4689 (void) snprintf(clonename, sizeof (clonename),
4690 "%s/ch1_%llu", osname, id);
4691 (void) snprintf(tag, sizeof (tag), "tag_%llu", id);
4692
4693 /*
4694 * Clean up from any previous run.
4695 */
4696 error = dsl_destroy_head(clonename);
4697 if (error != ENOENT)
4698 ASSERT0(error);
4699 error = user_release_one(fullname, tag);
4700 if (error != ESRCH && error != ENOENT)
4701 ASSERT0(error);
4702 error = dsl_destroy_snapshot(fullname, B_FALSE);
4703 if (error != ENOENT)
4704 ASSERT0(error);
4705
4706 /*
4707 * Create snapshot, clone it, mark snap for deferred destroy,
4708 * destroy clone, verify snap was also destroyed.
4709 */
4710 error = dmu_objset_snapshot_one(osname, snapname);
4711 if (error) {
4712 if (error == ENOSPC) {
4713 ztest_record_enospc("dmu_objset_snapshot");
4714 goto out;
4715 }
4716 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
4717 }
4718
4719 error = dmu_objset_clone(clonename, fullname);
4720 if (error) {
4721 if (error == ENOSPC) {
4722 ztest_record_enospc("dmu_objset_clone");
4723 goto out;
4724 }
4725 fatal(0, "dmu_objset_clone(%s) = %d", clonename, error);
4726 }
4727
4728 error = dsl_destroy_snapshot(fullname, B_TRUE);
4729 if (error) {
4730 fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
4731 fullname, error);
4732 }
4733
4734 error = dsl_destroy_head(clonename);
4735 if (error)
4736 fatal(0, "dsl_destroy_head(%s) = %d", clonename, error);
4737
4738 error = dmu_objset_hold(fullname, FTAG, &origin);
4739 if (error != ENOENT)
4740 fatal(0, "dmu_objset_hold(%s) = %d", fullname, error);
4741
4742 /*
4743 * Create snapshot, add temporary hold, verify that we can't
4744 * destroy a held snapshot, mark for deferred destroy,
4745 * release hold, verify snapshot was destroyed.
4746 */
4747 error = dmu_objset_snapshot_one(osname, snapname);
4748 if (error) {
4749 if (error == ENOSPC) {
4750 ztest_record_enospc("dmu_objset_snapshot");
4751 goto out;
4752 }
4753 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
4754 }
4755
4756 holds = fnvlist_alloc();
4757 fnvlist_add_string(holds, fullname, tag);
4758 error = dsl_dataset_user_hold(holds, 0, NULL);
4759 fnvlist_free(holds);
4760
4761 if (error == ENOSPC) {
4762 ztest_record_enospc("dsl_dataset_user_hold");
4763 goto out;
4764 } else if (error) {
4765 fatal(0, "dsl_dataset_user_hold(%s, %s) = %u",
4766 fullname, tag, error);
4767 }
4768
4769 error = dsl_destroy_snapshot(fullname, B_FALSE);
4770 if (error != EBUSY) {
4771 fatal(0, "dsl_destroy_snapshot(%s, B_FALSE) = %d",
4772 fullname, error);
4773 }
4774
4775 error = dsl_destroy_snapshot(fullname, B_TRUE);
4776 if (error) {
4777 fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
4778 fullname, error);
4779 }
4780
4781 error = user_release_one(fullname, tag);
4782 if (error)
4783 fatal(0, "user_release_one(%s, %s) = %d", fullname, tag, error);
4784
4785 VERIFY3U(dmu_objset_hold(fullname, FTAG, &origin), ==, ENOENT);
4786
4787 out:
4788 (void) rw_unlock(&ztest_name_lock);
4789 }
4790
4791 /*
4792 * Inject random faults into the on-disk data.
4793 */
4794 /* ARGSUSED */
4795 void
4796 ztest_fault_inject(ztest_ds_t *zd, uint64_t id)
4797 {
4798 ztest_shared_t *zs = ztest_shared;
4799 spa_t *spa = ztest_spa;
4800 int fd;
4801 uint64_t offset;
4802 uint64_t leaves;
4803 uint64_t bad = 0x1990c0ffeedecade;
4804 uint64_t top, leaf;
4805 char path0[MAXPATHLEN];
4806 char pathrand[MAXPATHLEN];
4807 size_t fsize;
4808 int bshift = SPA_MAXBLOCKSHIFT + 2;
4809 int iters = 1000;
4810 int maxfaults;
4811 int mirror_save;
4812 vdev_t *vd0 = NULL;
4813 uint64_t guid0 = 0;
4814 boolean_t islog = B_FALSE;
4815
4816 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
4817 maxfaults = MAXFAULTS();
4818 leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz;
4819 mirror_save = zs->zs_mirrors;
4820 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4821
4822 ASSERT(leaves >= 1);
4823
4824 /*
4825 * Grab the name lock as reader. There are some operations
4826 * which don't like to have their vdevs changed while
4827 * they are in progress (i.e. spa_change_guid). Those
4828 * operations will have grabbed the name lock as writer.
4829 */
4830 (void) rw_rdlock(&ztest_name_lock);
4831
4832 /*
4833 * We need SCL_STATE here because we're going to look at vd0->vdev_tsd.
4834 */
4835 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
4836
4837 if (ztest_random(2) == 0) {
4838 /*
4839 * Inject errors on a normal data device or slog device.
4840 */
4841 top = ztest_random_vdev_top(spa, B_TRUE);
4842 leaf = ztest_random(leaves) + zs->zs_splits;
4843
4844 /*
4845 * Generate paths to the first leaf in this top-level vdev,
4846 * and to the random leaf we selected. We'll induce transient
4847 * write failures and random online/offline activity on leaf 0,
4848 * and we'll write random garbage to the randomly chosen leaf.
4849 */
4850 (void) snprintf(path0, sizeof (path0), ztest_dev_template,
4851 ztest_opts.zo_dir, ztest_opts.zo_pool,
4852 top * leaves + zs->zs_splits);
4853 (void) snprintf(pathrand, sizeof (pathrand), ztest_dev_template,
4854 ztest_opts.zo_dir, ztest_opts.zo_pool,
4855 top * leaves + leaf);
4856
4857 vd0 = vdev_lookup_by_path(spa->spa_root_vdev, path0);
4858 if (vd0 != NULL && vd0->vdev_top->vdev_islog)
4859 islog = B_TRUE;
4860
4861 /*
4862 * If the top-level vdev needs to be resilvered
4863 * then we only allow faults on the device that is
4864 * resilvering.
4865 */
4866 if (vd0 != NULL && maxfaults != 1 &&
4867 (!vdev_resilver_needed(vd0->vdev_top, NULL, NULL) ||
4868 vd0->vdev_resilver_txg != 0)) {
4869 /*
4870 * Make vd0 explicitly claim to be unreadable,
4871 * or unwriteable, or reach behind its back
4872 * and close the underlying fd. We can do this if
4873 * maxfaults == 0 because we'll fail and reexecute,
4874 * and we can do it if maxfaults >= 2 because we'll
4875 * have enough redundancy. If maxfaults == 1, the
4876 * combination of this with injection of random data
4877 * corruption below exceeds the pool's fault tolerance.
4878 */
4879 vdev_file_t *vf = vd0->vdev_tsd;
4880
4881 if (vf != NULL && ztest_random(3) == 0) {
4882 (void) close(vf->vf_vnode->v_fd);
4883 vf->vf_vnode->v_fd = -1;
4884 } else if (ztest_random(2) == 0) {
4885 vd0->vdev_cant_read = B_TRUE;
4886 } else {
4887 vd0->vdev_cant_write = B_TRUE;
4888 }
4889 guid0 = vd0->vdev_guid;
4890 }
4891 } else {
4892 /*
4893 * Inject errors on an l2cache device.
4894 */
4895 spa_aux_vdev_t *sav = &spa->spa_l2cache;
4896
4897 if (sav->sav_count == 0) {
4898 spa_config_exit(spa, SCL_STATE, FTAG);
4899 (void) rw_unlock(&ztest_name_lock);
4900 return;
4901 }
4902 vd0 = sav->sav_vdevs[ztest_random(sav->sav_count)];
4903 guid0 = vd0->vdev_guid;
4904 (void) strcpy(path0, vd0->vdev_path);
4905 (void) strcpy(pathrand, vd0->vdev_path);
4906
4907 leaf = 0;
4908 leaves = 1;
4909 maxfaults = INT_MAX; /* no limit on cache devices */
4910 }
4911
4912 spa_config_exit(spa, SCL_STATE, FTAG);
4913 (void) rw_unlock(&ztest_name_lock);
4914
4915 /*
4916 * If we can tolerate two or more faults, or we're dealing
4917 * with a slog, randomly online/offline vd0.
4918 */
4919 if ((maxfaults >= 2 || islog) && guid0 != 0) {
4920 if (ztest_random(10) < 6) {
4921 int flags = (ztest_random(2) == 0 ?
4922 ZFS_OFFLINE_TEMPORARY : 0);
4923
4924 /*
4925 * We have to grab the zs_name_lock as writer to
4926 * prevent a race between offlining a slog and
4927 * destroying a dataset. Offlining the slog will
4928 * grab a reference on the dataset which may cause
4929 * dmu_objset_destroy() to fail with EBUSY thus
4930 * leaving the dataset in an inconsistent state.
4931 */
4932 if (islog)
4933 (void) rw_wrlock(&ztest_name_lock);
4934
4935 VERIFY(vdev_offline(spa, guid0, flags) != EBUSY);
4936
4937 if (islog)
4938 (void) rw_unlock(&ztest_name_lock);
4939 } else {
4940 /*
4941 * Ideally we would like to be able to randomly
4942 * call vdev_[on|off]line without holding locks
4943 * to force unpredictable failures but the side
4944 * effects of vdev_[on|off]line prevent us from
4945 * doing so. We grab the ztest_vdev_lock here to
4946 * prevent a race between injection testing and
4947 * aux_vdev removal.
4948 */
4949 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
4950 (void) vdev_online(spa, guid0, 0, NULL);
4951 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4952 }
4953 }
4954
4955 if (maxfaults == 0)
4956 return;
4957
4958 /*
4959 * We have at least single-fault tolerance, so inject data corruption.
4960 */
4961 fd = open(pathrand, O_RDWR);
4962
4963 if (fd == -1) /* we hit a gap in the device namespace */
4964 return;
4965
4966 fsize = lseek(fd, 0, SEEK_END);
4967
4968 while (--iters != 0) {
4969 /*
4970 * The offset must be chosen carefully to ensure that
4971 * we do not inject a given logical block with errors
4972 * on two different leaf devices, because ZFS can not
4973 * tolerate that (if maxfaults==1).
4974 *
4975 * We divide each leaf into chunks of size
4976 * (# leaves * SPA_MAXBLOCKSIZE * 4). Within each chunk
4977 * there is a series of ranges to which we can inject errors.
4978 * Each range can accept errors on only a single leaf vdev.
4979 * The error injection ranges are separated by ranges
4980 * which we will not inject errors on any device (DMZs).
4981 * Each DMZ must be large enough such that a single block
4982 * can not straddle it, so that a single block can not be
4983 * a target in two different injection ranges (on different
4984 * leaf vdevs).
4985 *
4986 * For example, with 3 leaves, each chunk looks like:
4987 * 0 to 32M: injection range for leaf 0
4988 * 32M to 64M: DMZ - no injection allowed
4989 * 64M to 96M: injection range for leaf 1
4990 * 96M to 128M: DMZ - no injection allowed
4991 * 128M to 160M: injection range for leaf 2
4992 * 160M to 192M: DMZ - no injection allowed
4993 */
4994 offset = ztest_random(fsize / (leaves << bshift)) *
4995 (leaves << bshift) + (leaf << bshift) +
4996 (ztest_random(1ULL << (bshift - 1)) & -8ULL);
4997
4998 /*
4999 * Only allow damage to the labels at one end of the vdev.
5000 *
5001 * If all labels are damaged, the device will be totally
5002 * inaccessible, which will result in loss of data,
5003 * because we also damage (parts of) the other side of
5004 * the mirror/raidz.
5005 *
5006 * Additionally, we will always have both an even and an
5007 * odd label, so that we can handle crashes in the
5008 * middle of vdev_config_sync().
5009 */
5010 if ((leaf & 1) == 0 && offset < VDEV_LABEL_START_SIZE)
5011 continue;
5012
5013 /*
5014 * The two end labels are stored at the "end" of the disk, but
5015 * the end of the disk (vdev_psize) is aligned to
5016 * sizeof (vdev_label_t).
5017 */
5018 uint64_t psize = P2ALIGN(fsize, sizeof (vdev_label_t));
5019 if ((leaf & 1) == 1 &&
5020 offset + sizeof (bad) > psize - VDEV_LABEL_END_SIZE)
5021 continue;
5022
5023 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
5024 if (mirror_save != zs->zs_mirrors) {
5025 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
5026 (void) close(fd);
5027 return;
5028 }
5029
5030 if (pwrite(fd, &bad, sizeof (bad), offset) != sizeof (bad))
5031 fatal(1, "can't inject bad word at 0x%llx in %s",
5032 offset, pathrand);
5033
5034 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
5035
5036 if (ztest_opts.zo_verbose >= 7)
5037 (void) printf("injected bad word into %s,"
5038 " offset 0x%llx\n", pathrand, (u_longlong_t)offset);
5039 }
5040
5041 (void) close(fd);
5042 }
5043
5044 /*
5045 * Verify that DDT repair works as expected.
5046 */
5047 void
5048 ztest_ddt_repair(ztest_ds_t *zd, uint64_t id)
5049 {
5050 ztest_shared_t *zs = ztest_shared;
5051 spa_t *spa = ztest_spa;
5052 objset_t *os = zd->zd_os;
5053 ztest_od_t od[1];
5054 uint64_t object, blocksize, txg, pattern, psize;
5055 enum zio_checksum checksum = spa_dedup_checksum(spa);
5056 dmu_buf_t *db;
5057 dmu_tx_t *tx;
5058 abd_t *abd;
5059 blkptr_t blk;
5060 int copies = 2 * ZIO_DEDUPDITTO_MIN;
5061
5062 blocksize = ztest_random_blocksize();
5063 blocksize = MIN(blocksize, 2048); /* because we write so many */
5064
5065 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
5066
5067 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
5068 return;
5069
5070 /*
5071 * Take the name lock as writer to prevent anyone else from changing
5072 * the pool and dataset properies we need to maintain during this test.
5073 */
5074 (void) rw_wrlock(&ztest_name_lock);
5075
5076 if (ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_DEDUP, checksum,
5077 B_FALSE) != 0 ||
5078 ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_COPIES, 1,
5079 B_FALSE) != 0) {
5080 (void) rw_unlock(&ztest_name_lock);
5081 return;
5082 }
5083
5084 dmu_objset_stats_t dds;
5085 dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
5086 dmu_objset_fast_stat(os, &dds);
5087 dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
5088
5089 object = od[0].od_object;
5090 blocksize = od[0].od_blocksize;
5091 pattern = zs->zs_guid ^ dds.dds_guid;
5092
5093 ASSERT(object != 0);
5094
5095 tx = dmu_tx_create(os);
5096 dmu_tx_hold_write(tx, object, 0, copies * blocksize);
5097 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
5098 if (txg == 0) {
5099 (void) rw_unlock(&ztest_name_lock);
5100 return;
5101 }
5102
5103 /*
5104 * Write all the copies of our block.
5105 */
5106 for (int i = 0; i < copies; i++) {
5107 uint64_t offset = i * blocksize;
5108 int error = dmu_buf_hold(os, object, offset, FTAG, &db,
5109 DMU_READ_NO_PREFETCH);
5110 if (error != 0) {
5111 fatal(B_FALSE, "dmu_buf_hold(%p, %llu, %llu) = %u",
5112 os, (long long)object, (long long) offset, error);
5113 }
5114 ASSERT(db->db_offset == offset);
5115 ASSERT(db->db_size == blocksize);
5116 ASSERT(ztest_pattern_match(db->db_data, db->db_size, pattern) ||
5117 ztest_pattern_match(db->db_data, db->db_size, 0ULL));
5118 dmu_buf_will_fill(db, tx);
5119 ztest_pattern_set(db->db_data, db->db_size, pattern);
5120 dmu_buf_rele(db, FTAG);
5121 }
5122
5123 dmu_tx_commit(tx);
5124 txg_wait_synced(spa_get_dsl(spa), txg);
5125
5126 /*
5127 * Find out what block we got.
5128 */
5129 VERIFY0(dmu_buf_hold(os, object, 0, FTAG, &db,
5130 DMU_READ_NO_PREFETCH));
5131 blk = *((dmu_buf_impl_t *)db)->db_blkptr;
5132 dmu_buf_rele(db, FTAG);
5133
5134 /*
5135 * Damage the block. Dedup-ditto will save us when we read it later.
5136 */
5137 psize = BP_GET_PSIZE(&blk);
5138 abd = abd_alloc_linear(psize, B_TRUE);
5139 ztest_pattern_set(abd_to_buf(abd), psize, ~pattern);
5140
5141 (void) zio_wait(zio_rewrite(NULL, spa, 0, &blk,
5142 abd, psize, NULL, NULL, ZIO_PRIORITY_SYNC_WRITE,
5143 ZIO_FLAG_CANFAIL | ZIO_FLAG_INDUCE_DAMAGE, NULL));
5144
5145 abd_free(abd);
5146
5147 (void) rw_unlock(&ztest_name_lock);
5148 }
5149
5150 /*
5151 * Scrub the pool.
5152 */
5153 /* ARGSUSED */
5154 void
5155 ztest_scrub(ztest_ds_t *zd, uint64_t id)
5156 {
5157 spa_t *spa = ztest_spa;
5158
5159 (void) spa_scan(spa, POOL_SCAN_SCRUB);
5160 (void) poll(NULL, 0, 100); /* wait a moment, then force a restart */
5161 (void) spa_scan(spa, POOL_SCAN_SCRUB);
5162 }
5163
5164 /*
5165 * Change the guid for the pool.
5166 */
5167 /* ARGSUSED */
5168 void
5169 ztest_reguid(ztest_ds_t *zd, uint64_t id)
5170 {
5171 spa_t *spa = ztest_spa;
5172 uint64_t orig, load;
5173 int error;
5174
5175 orig = spa_guid(spa);
5176 load = spa_load_guid(spa);
5177
5178 (void) rw_wrlock(&ztest_name_lock);
5179 error = spa_change_guid(spa);
5180 (void) rw_unlock(&ztest_name_lock);
5181
5182 if (error != 0)
5183 return;
5184
5185 if (ztest_opts.zo_verbose >= 4) {
5186 (void) printf("Changed guid old %llu -> %llu\n",
5187 (u_longlong_t)orig, (u_longlong_t)spa_guid(spa));
5188 }
5189
5190 VERIFY3U(orig, !=, spa_guid(spa));
5191 VERIFY3U(load, ==, spa_load_guid(spa));
5192 }
5193
5194 /*
5195 * Rename the pool to a different name and then rename it back.
5196 */
5197 /* ARGSUSED */
5198 void
5199 ztest_spa_rename(ztest_ds_t *zd, uint64_t id)
5200 {
5201 char *oldname, *newname;
5202 spa_t *spa;
5203
5204 (void) rw_wrlock(&ztest_name_lock);
5205
5206 oldname = ztest_opts.zo_pool;
5207 newname = umem_alloc(strlen(oldname) + 5, UMEM_NOFAIL);
5208 (void) strcpy(newname, oldname);
5209 (void) strcat(newname, "_tmp");
5210
5211 /*
5212 * Do the rename
5213 */
5214 VERIFY3U(0, ==, spa_rename(oldname, newname));
5215
5216 /*
5217 * Try to open it under the old name, which shouldn't exist
5218 */
5219 VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
5220
5221 /*
5222 * Open it under the new name and make sure it's still the same spa_t.
5223 */
5224 VERIFY3U(0, ==, spa_open(newname, &spa, FTAG));
5225
5226 ASSERT(spa == ztest_spa);
5227 spa_close(spa, FTAG);
5228
5229 /*
5230 * Rename it back to the original
5231 */
5232 VERIFY3U(0, ==, spa_rename(newname, oldname));
5233
5234 /*
5235 * Make sure it can still be opened
5236 */
5237 VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG));
5238
5239 ASSERT(spa == ztest_spa);
5240 spa_close(spa, FTAG);
5241
5242 umem_free(newname, strlen(newname) + 1);
5243
5244 (void) rw_unlock(&ztest_name_lock);
5245 }
5246
5247 /*
5248 * Verify pool integrity by running zdb.
5249 */
5250 static void
5251 ztest_run_zdb(char *pool)
5252 {
5253 int status;
5254 char zdb[MAXPATHLEN + MAXNAMELEN + 20];
5255 char zbuf[1024];
5256 char *bin;
5257 char *ztest;
5258 char *isa;
5259 int isalen;
5260 FILE *fp;
5261
5262 (void) realpath(getexecname(), zdb);
5263
5264 /* zdb lives in /usr/sbin, while ztest lives in /usr/bin */
5265 bin = strstr(zdb, "/usr/bin/");
5266 ztest = strstr(bin, "/ztest");
5267 isa = bin + 8;
5268 isalen = ztest - isa;
5269 isa = strdup(isa);
5270 /* LINTED */
5271 (void) sprintf(bin,
5272 "/usr/sbin%.*s/zdb -bcc%s%s -G -d -U %s %s",
5273 isalen,
5274 isa,
5275 ztest_opts.zo_verbose >= 3 ? "s" : "",
5276 ztest_opts.zo_verbose >= 4 ? "v" : "",
5277 spa_config_path,
5278 pool);
5279 free(isa);
5280
5281 if (ztest_opts.zo_verbose >= 5)
5282 (void) printf("Executing %s\n", strstr(zdb, "zdb "));
5283
5284 fp = popen(zdb, "r");
5285
5286 while (fgets(zbuf, sizeof (zbuf), fp) != NULL)
5287 if (ztest_opts.zo_verbose >= 3)
5288 (void) printf("%s", zbuf);
5289
5290 status = pclose(fp);
5291
5292 if (status == 0)
5293 return;
5294
5295 ztest_dump_core = 0;
5296 if (WIFEXITED(status))
5297 fatal(0, "'%s' exit code %d", zdb, WEXITSTATUS(status));
5298 else
5299 fatal(0, "'%s' died with signal %d", zdb, WTERMSIG(status));
5300 }
5301
5302 static void
5303 ztest_walk_pool_directory(char *header)
5304 {
5305 spa_t *spa = NULL;
5306
5307 if (ztest_opts.zo_verbose >= 6)
5308 (void) printf("%s\n", header);
5309
5310 mutex_enter(&spa_namespace_lock);
5311 while ((spa = spa_next(spa)) != NULL)
5312 if (ztest_opts.zo_verbose >= 6)
5313 (void) printf("\t%s\n", spa_name(spa));
5314 mutex_exit(&spa_namespace_lock);
5315 }
5316
5317 static void
5318 ztest_spa_import_export(char *oldname, char *newname)
5319 {
5320 nvlist_t *config, *newconfig;
5321 uint64_t pool_guid;
5322 spa_t *spa;
5323 int error;
5324
5325 if (ztest_opts.zo_verbose >= 4) {
5326 (void) printf("import/export: old = %s, new = %s\n",
5327 oldname, newname);
5328 }
5329
5330 /*
5331 * Clean up from previous runs.
5332 */
5333 (void) spa_destroy(newname);
5334
5335 /*
5336 * Get the pool's configuration and guid.
5337 */
5338 VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG));
5339
5340 /*
5341 * Kick off a scrub to tickle scrub/export races.
5342 */
5343 if (ztest_random(2) == 0)
5344 (void) spa_scan(spa, POOL_SCAN_SCRUB);
5345
5346 pool_guid = spa_guid(spa);
5347 spa_close(spa, FTAG);
5348
5349 ztest_walk_pool_directory("pools before export");
5350
5351 /*
5352 * Export it.
5353 */
5354 VERIFY3U(0, ==, spa_export(oldname, &config, B_FALSE, B_FALSE));
5355
5356 ztest_walk_pool_directory("pools after export");
5357
5358 /*
5359 * Try to import it.
5360 */
5361 newconfig = spa_tryimport(config);
5362 ASSERT(newconfig != NULL);
5363 nvlist_free(newconfig);
5364
5365 /*
5366 * Import it under the new name.
5367 */
5368 error = spa_import(newname, config, NULL, 0);
5369 if (error != 0) {
5370 dump_nvlist(config, 0);
5371 fatal(B_FALSE, "couldn't import pool %s as %s: error %u",
5372 oldname, newname, error);
5373 }
5374
5375 ztest_walk_pool_directory("pools after import");
5376
5377 /*
5378 * Try to import it again -- should fail with EEXIST.
5379 */
5380 VERIFY3U(EEXIST, ==, spa_import(newname, config, NULL, 0));
5381
5382 /*
5383 * Try to import it under a different name -- should fail with EEXIST.
5384 */
5385 VERIFY3U(EEXIST, ==, spa_import(oldname, config, NULL, 0));
5386
5387 /*
5388 * Verify that the pool is no longer visible under the old name.
5389 */
5390 VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
5391
5392 /*
5393 * Verify that we can open and close the pool using the new name.
5394 */
5395 VERIFY3U(0, ==, spa_open(newname, &spa, FTAG));
5396 ASSERT(pool_guid == spa_guid(spa));
5397 spa_close(spa, FTAG);
5398
5399 nvlist_free(config);
5400 }
5401
5402 static void
5403 ztest_resume(spa_t *spa)
5404 {
5405 if (spa_suspended(spa) && ztest_opts.zo_verbose >= 6)
5406 (void) printf("resuming from suspended state\n");
5407 spa_vdev_state_enter(spa, SCL_NONE);
5408 vdev_clear(spa, NULL);
5409 (void) spa_vdev_state_exit(spa, NULL, 0);
5410 (void) zio_resume(spa);
5411 }
5412
5413 static void *
5414 ztest_resume_thread(void *arg)
5415 {
5416 spa_t *spa = arg;
5417
5418 while (!ztest_exiting) {
5419 if (spa_suspended(spa))
5420 ztest_resume(spa);
5421 (void) poll(NULL, 0, 100);
5422
5423 /*
5424 * Periodically change the zfs_compressed_arc_enabled setting.
5425 */
5426 if (ztest_random(10) == 0)
5427 zfs_compressed_arc_enabled = ztest_random(2);
5428
5429 /*
5430 * Periodically change the zfs_abd_scatter_enabled setting.
5431 */
5432 if (ztest_random(10) == 0)
5433 zfs_abd_scatter_enabled = ztest_random(2);
5434 }
5435 return (NULL);
5436 }
5437
5438 static void *
5439 ztest_deadman_thread(void *arg)
5440 {
5441 ztest_shared_t *zs = arg;
5442 spa_t *spa = ztest_spa;
5443 hrtime_t delta, total = 0;
5444
5445 for (;;) {
5446 delta = zs->zs_thread_stop - zs->zs_thread_start +
5447 MSEC2NSEC(zfs_deadman_synctime_ms);
5448
5449 (void) poll(NULL, 0, (int)NSEC2MSEC(delta));
5450
5451 /*
5452 * If the pool is suspended then fail immediately. Otherwise,
5453 * check to see if the pool is making any progress. If
5454 * vdev_deadman() discovers that there hasn't been any recent
5455 * I/Os then it will end up aborting the tests.
5456 */
5457 if (spa_suspended(spa) || spa->spa_root_vdev == NULL) {
5458 fatal(0, "aborting test after %llu seconds because "
5459 "pool has transitioned to a suspended state.",
5460 zfs_deadman_synctime_ms / 1000);
5461 return (NULL);
5462 }
5463 vdev_deadman(spa->spa_root_vdev);
5464
5465 total += zfs_deadman_synctime_ms/1000;
5466 (void) printf("ztest has been running for %lld seconds\n",
5467 total);
5468 }
5469 }
5470
5471 static void
5472 ztest_execute(int test, ztest_info_t *zi, uint64_t id)
5473 {
5474 ztest_ds_t *zd = &ztest_ds[id % ztest_opts.zo_datasets];
5475 ztest_shared_callstate_t *zc = ZTEST_GET_SHARED_CALLSTATE(test);
5476 hrtime_t functime = gethrtime();
5477
5478 for (int i = 0; i < zi->zi_iters; i++)
5479 zi->zi_func(zd, id);
5480
5481 functime = gethrtime() - functime;
5482
5483 atomic_add_64(&zc->zc_count, 1);
5484 atomic_add_64(&zc->zc_time, functime);
5485
5486 if (ztest_opts.zo_verbose >= 4) {
5487 Dl_info dli;
5488 (void) dladdr((void *)zi->zi_func, &dli);
5489 (void) printf("%6.2f sec in %s\n",
5490 (double)functime / NANOSEC, dli.dli_sname);
5491 }
5492 }
5493
5494 static void *
5495 ztest_thread(void *arg)
5496 {
5497 int rand;
5498 uint64_t id = (uintptr_t)arg;
5499 ztest_shared_t *zs = ztest_shared;
5500 uint64_t call_next;
5501 hrtime_t now;
5502 ztest_info_t *zi;
5503 ztest_shared_callstate_t *zc;
5504
5505 while ((now = gethrtime()) < zs->zs_thread_stop) {
5506 /*
5507 * See if it's time to force a crash.
5508 */
5509 if (now > zs->zs_thread_kill)
5510 ztest_kill(zs);
5511
5512 /*
5513 * If we're getting ENOSPC with some regularity, stop.
5514 */
5515 if (zs->zs_enospc_count > 10)
5516 break;
5517
5518 /*
5519 * Pick a random function to execute.
5520 */
5521 rand = ztest_random(ZTEST_FUNCS);
5522 zi = &ztest_info[rand];
5523 zc = ZTEST_GET_SHARED_CALLSTATE(rand);
5524 call_next = zc->zc_next;
5525
5526 if (now >= call_next &&
5527 atomic_cas_64(&zc->zc_next, call_next, call_next +
5528 ztest_random(2 * zi->zi_interval[0] + 1)) == call_next) {
5529 ztest_execute(rand, zi, id);
5530 }
5531 }
5532
5533 return (NULL);
5534 }
5535
5536 static void
5537 ztest_dataset_name(char *dsname, char *pool, int d)
5538 {
5539 (void) snprintf(dsname, ZFS_MAX_DATASET_NAME_LEN, "%s/ds_%d", pool, d);
5540 }
5541
5542 static void
5543 ztest_dataset_destroy(int d)
5544 {
5545 char name[ZFS_MAX_DATASET_NAME_LEN];
5546
5547 ztest_dataset_name(name, ztest_opts.zo_pool, d);
5548
5549 if (ztest_opts.zo_verbose >= 3)
5550 (void) printf("Destroying %s to free up space\n", name);
5551
5552 /*
5553 * Cleanup any non-standard clones and snapshots. In general,
5554 * ztest thread t operates on dataset (t % zopt_datasets),
5555 * so there may be more than one thing to clean up.
5556 */
5557 for (int t = d; t < ztest_opts.zo_threads;
5558 t += ztest_opts.zo_datasets) {
5559 ztest_dsl_dataset_cleanup(name, t);
5560 }
5561
5562 (void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
5563 DS_FIND_SNAPSHOTS | DS_FIND_CHILDREN);
5564 }
5565
5566 static void
5567 ztest_dataset_dirobj_verify(ztest_ds_t *zd)
5568 {
5569 uint64_t usedobjs, dirobjs, scratch;
5570
5571 /*
5572 * ZTEST_DIROBJ is the object directory for the entire dataset.
5573 * Therefore, the number of objects in use should equal the
5574 * number of ZTEST_DIROBJ entries, +1 for ZTEST_DIROBJ itself.
5575 * If not, we have an object leak.
5576 *
5577 * Note that we can only check this in ztest_dataset_open(),
5578 * when the open-context and syncing-context values agree.
5579 * That's because zap_count() returns the open-context value,
5580 * while dmu_objset_space() returns the rootbp fill count.
5581 */
5582 VERIFY3U(0, ==, zap_count(zd->zd_os, ZTEST_DIROBJ, &dirobjs));
5583 dmu_objset_space(zd->zd_os, &scratch, &scratch, &usedobjs, &scratch);
5584 ASSERT3U(dirobjs + 1, ==, usedobjs);
5585 }
5586
5587 static int
5588 ztest_dataset_open(int d)
5589 {
5590 ztest_ds_t *zd = &ztest_ds[d];
5591 uint64_t committed_seq = ZTEST_GET_SHARED_DS(d)->zd_seq;
5592 objset_t *os;
5593 zilog_t *zilog;
5594 char name[ZFS_MAX_DATASET_NAME_LEN];
5595 int error;
5596
5597 ztest_dataset_name(name, ztest_opts.zo_pool, d);
5598
5599 (void) rw_rdlock(&ztest_name_lock);
5600
5601 error = ztest_dataset_create(name);
5602 if (error == ENOSPC) {
5603 (void) rw_unlock(&ztest_name_lock);
5604 ztest_record_enospc(FTAG);
5605 return (error);
5606 }
5607 ASSERT(error == 0 || error == EEXIST);
5608
5609 VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, zd, &os));
5610 (void) rw_unlock(&ztest_name_lock);
5611
5612 ztest_zd_init(zd, ZTEST_GET_SHARED_DS(d), os);
5613
5614 zilog = zd->zd_zilog;
5615
5616 if (zilog->zl_header->zh_claim_lr_seq != 0 &&
5617 zilog->zl_header->zh_claim_lr_seq < committed_seq)
5618 fatal(0, "missing log records: claimed %llu < committed %llu",
5619 zilog->zl_header->zh_claim_lr_seq, committed_seq);
5620
5621 ztest_dataset_dirobj_verify(zd);
5622
5623 zil_replay(os, zd, ztest_replay_vector);
5624
5625 ztest_dataset_dirobj_verify(zd);
5626
5627 if (ztest_opts.zo_verbose >= 6)
5628 (void) printf("%s replay %llu blocks, %llu records, seq %llu\n",
5629 zd->zd_name,
5630 (u_longlong_t)zilog->zl_parse_blk_count,
5631 (u_longlong_t)zilog->zl_parse_lr_count,
5632 (u_longlong_t)zilog->zl_replaying_seq);
5633
5634 zilog = zil_open(os, ztest_get_data);
5635
5636 if (zilog->zl_replaying_seq != 0 &&
5637 zilog->zl_replaying_seq < committed_seq)
5638 fatal(0, "missing log records: replayed %llu < committed %llu",
5639 zilog->zl_replaying_seq, committed_seq);
5640
5641 return (0);
5642 }
5643
5644 static void
5645 ztest_dataset_close(int d)
5646 {
5647 ztest_ds_t *zd = &ztest_ds[d];
5648
5649 zil_close(zd->zd_zilog);
5650 dmu_objset_disown(zd->zd_os, zd);
5651
5652 ztest_zd_fini(zd);
5653 }
5654
5655 /*
5656 * Kick off threads to run tests on all datasets in parallel.
5657 */
5658 static void
5659 ztest_run(ztest_shared_t *zs)
5660 {
5661 thread_t *tid;
5662 spa_t *spa;
5663 objset_t *os;
5664 thread_t resume_tid;
5665 int error;
5666
5667 ztest_exiting = B_FALSE;
5668
5669 /*
5670 * Initialize parent/child shared state.
5671 */
5672 VERIFY(_mutex_init(&ztest_vdev_lock, USYNC_THREAD, NULL) == 0);
5673 VERIFY(rwlock_init(&ztest_name_lock, USYNC_THREAD, NULL) == 0);
5674
5675 zs->zs_thread_start = gethrtime();
5676 zs->zs_thread_stop =
5677 zs->zs_thread_start + ztest_opts.zo_passtime * NANOSEC;
5678 zs->zs_thread_stop = MIN(zs->zs_thread_stop, zs->zs_proc_stop);
5679 zs->zs_thread_kill = zs->zs_thread_stop;
5680 if (ztest_random(100) < ztest_opts.zo_killrate) {
5681 zs->zs_thread_kill -=
5682 ztest_random(ztest_opts.zo_passtime * NANOSEC);
5683 }
5684
5685 (void) _mutex_init(&zcl.zcl_callbacks_lock, USYNC_THREAD, NULL);
5686
5687 list_create(&zcl.zcl_callbacks, sizeof (ztest_cb_data_t),
5688 offsetof(ztest_cb_data_t, zcd_node));
5689
5690 /*
5691 * Open our pool.
5692 */
5693 kernel_init(FREAD | FWRITE);
5694 VERIFY0(spa_open(ztest_opts.zo_pool, &spa, FTAG));
5695 spa->spa_debug = B_TRUE;
5696 metaslab_preload_limit = ztest_random(20) + 1;
5697 ztest_spa = spa;
5698
5699 dmu_objset_stats_t dds;
5700 VERIFY0(dmu_objset_own(ztest_opts.zo_pool,
5701 DMU_OST_ANY, B_TRUE, FTAG, &os));
5702 dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
5703 dmu_objset_fast_stat(os, &dds);
5704 dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
5705 zs->zs_guid = dds.dds_guid;
5706 dmu_objset_disown(os, FTAG);
5707
5708 spa->spa_dedup_ditto = 2 * ZIO_DEDUPDITTO_MIN;
5709
5710 /*
5711 * We don't expect the pool to suspend unless maxfaults == 0,
5712 * in which case ztest_fault_inject() temporarily takes away
5713 * the only valid replica.
5714 */
5715 if (MAXFAULTS() == 0)
5716 spa->spa_failmode = ZIO_FAILURE_MODE_WAIT;
5717 else
5718 spa->spa_failmode = ZIO_FAILURE_MODE_PANIC;
5719
5720 /*
5721 * Create a thread to periodically resume suspended I/O.
5722 */
5723 VERIFY(thr_create(0, 0, ztest_resume_thread, spa, THR_BOUND,
5724 &resume_tid) == 0);
5725
5726 /*
5727 * Create a deadman thread to abort() if we hang.
5728 */
5729 VERIFY(thr_create(0, 0, ztest_deadman_thread, zs, THR_BOUND,
5730 NULL) == 0);
5731
5732 /*
5733 * Verify that we can safely inquire about about any object,
5734 * whether it's allocated or not. To make it interesting,
5735 * we probe a 5-wide window around each power of two.
5736 * This hits all edge cases, including zero and the max.
5737 */
5738 for (int t = 0; t < 64; t++) {
5739 for (int d = -5; d <= 5; d++) {
5740 error = dmu_object_info(spa->spa_meta_objset,
5741 (1ULL << t) + d, NULL);
5742 ASSERT(error == 0 || error == ENOENT ||
5743 error == EINVAL);
5744 }
5745 }
5746
5747 /*
5748 * If we got any ENOSPC errors on the previous run, destroy something.
5749 */
5750 if (zs->zs_enospc_count != 0) {
5751 int d = ztest_random(ztest_opts.zo_datasets);
5752 ztest_dataset_destroy(d);
5753 }
5754 zs->zs_enospc_count = 0;
5755
5756 tid = umem_zalloc(ztest_opts.zo_threads * sizeof (thread_t),
5757 UMEM_NOFAIL);
5758
5759 if (ztest_opts.zo_verbose >= 4)
5760 (void) printf("starting main threads...\n");
5761
5762 /*
5763 * Kick off all the tests that run in parallel.
5764 */
5765 for (int t = 0; t < ztest_opts.zo_threads; t++) {
5766 if (t < ztest_opts.zo_datasets &&
5767 ztest_dataset_open(t) != 0)
5768 return;
5769 VERIFY(thr_create(0, 0, ztest_thread, (void *)(uintptr_t)t,
5770 THR_BOUND, &tid[t]) == 0);
5771 }
5772
5773 /*
5774 * Wait for all of the tests to complete. We go in reverse order
5775 * so we don't close datasets while threads are still using them.
5776 */
5777 for (int t = ztest_opts.zo_threads - 1; t >= 0; t--) {
5778 VERIFY(thr_join(tid[t], NULL, NULL) == 0);
5779 if (t < ztest_opts.zo_datasets)
5780 ztest_dataset_close(t);
5781 }
5782
5783 txg_wait_synced(spa_get_dsl(spa), 0);
5784
5785 zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(spa));
5786 zs->zs_space = metaslab_class_get_space(spa_normal_class(spa));
5787 zfs_dbgmsg_print(FTAG);
5788
5789 umem_free(tid, ztest_opts.zo_threads * sizeof (thread_t));
5790
5791 /* Kill the resume thread */
5792 ztest_exiting = B_TRUE;
5793 VERIFY(thr_join(resume_tid, NULL, NULL) == 0);
5794 ztest_resume(spa);
5795
5796 /*
5797 * Right before closing the pool, kick off a bunch of async I/O;
5798 * spa_close() should wait for it to complete.
5799 */
5800 for (uint64_t object = 1; object < 50; object++) {
5801 dmu_prefetch(spa->spa_meta_objset, object, 0, 0, 1ULL << 20,
5802 ZIO_PRIORITY_SYNC_READ);
5803 }
5804
5805 spa_close(spa, FTAG);
5806
5807 /*
5808 * Verify that we can loop over all pools.
5809 */
5810 mutex_enter(&spa_namespace_lock);
5811 for (spa = spa_next(NULL); spa != NULL; spa = spa_next(spa))
5812 if (ztest_opts.zo_verbose > 3)
5813 (void) printf("spa_next: found %s\n", spa_name(spa));
5814 mutex_exit(&spa_namespace_lock);
5815
5816 /*
5817 * Verify that we can export the pool and reimport it under a
5818 * different name.
5819 */
5820 if (ztest_random(2) == 0) {
5821 char name[ZFS_MAX_DATASET_NAME_LEN];
5822 (void) snprintf(name, sizeof (name), "%s_import",
5823 ztest_opts.zo_pool);
5824 ztest_spa_import_export(ztest_opts.zo_pool, name);
5825 ztest_spa_import_export(name, ztest_opts.zo_pool);
5826 }
5827
5828 kernel_fini();
5829
5830 list_destroy(&zcl.zcl_callbacks);
5831
5832 (void) _mutex_destroy(&zcl.zcl_callbacks_lock);
5833
5834 (void) rwlock_destroy(&ztest_name_lock);
5835 (void) _mutex_destroy(&ztest_vdev_lock);
5836 }
5837
5838 static void
5839 ztest_freeze(void)
5840 {
5841 ztest_ds_t *zd = &ztest_ds[0];
5842 spa_t *spa;
5843 int numloops = 0;
5844
5845 if (ztest_opts.zo_verbose >= 3)
5846 (void) printf("testing spa_freeze()...\n");
5847
5848 kernel_init(FREAD | FWRITE);
5849 VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
5850 VERIFY3U(0, ==, ztest_dataset_open(0));
5851 spa->spa_debug = B_TRUE;
5852 ztest_spa = spa;
5853
5854 /*
5855 * Force the first log block to be transactionally allocated.
5856 * We have to do this before we freeze the pool -- otherwise
5857 * the log chain won't be anchored.
5858 */
5859 while (BP_IS_HOLE(&zd->zd_zilog->zl_header->zh_log)) {
5860 ztest_dmu_object_alloc_free(zd, 0);
5861 zil_commit(zd->zd_zilog, 0);
5862 }
5863
5864 txg_wait_synced(spa_get_dsl(spa), 0);
5865
5866 /*
5867 * Freeze the pool. This stops spa_sync() from doing anything,
5868 * so that the only way to record changes from now on is the ZIL.
5869 */
5870 spa_freeze(spa);
5871
5872 /*
5873 * Because it is hard to predict how much space a write will actually
5874 * require beforehand, we leave ourselves some fudge space to write over
5875 * capacity.
5876 */
5877 uint64_t capacity = metaslab_class_get_space(spa_normal_class(spa)) / 2;
5878
5879 /*
5880 * Run tests that generate log records but don't alter the pool config
5881 * or depend on DSL sync tasks (snapshots, objset create/destroy, etc).
5882 * We do a txg_wait_synced() after each iteration to force the txg
5883 * to increase well beyond the last synced value in the uberblock.
5884 * The ZIL should be OK with that.
5885 *
5886 * Run a random number of times less than zo_maxloops and ensure we do
5887 * not run out of space on the pool.
5888 */
5889 while (ztest_random(10) != 0 &&
5890 numloops++ < ztest_opts.zo_maxloops &&
5891 metaslab_class_get_alloc(spa_normal_class(spa)) < capacity) {
5892 ztest_od_t od;
5893 ztest_od_init(&od, 0, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
5894 VERIFY0(ztest_object_init(zd, &od, sizeof (od), B_FALSE));
5895 ztest_io(zd, od.od_object,
5896 ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
5897 txg_wait_synced(spa_get_dsl(spa), 0);
5898 }
5899
5900 /*
5901 * Commit all of the changes we just generated.
5902 */
5903 zil_commit(zd->zd_zilog, 0);
5904 txg_wait_synced(spa_get_dsl(spa), 0);
5905
5906 /*
5907 * Close our dataset and close the pool.
5908 */
5909 ztest_dataset_close(0);
5910 spa_close(spa, FTAG);
5911 kernel_fini();
5912
5913 /*
5914 * Open and close the pool and dataset to induce log replay.
5915 */
5916 kernel_init(FREAD | FWRITE);
5917 VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
5918 ASSERT(spa_freeze_txg(spa) == UINT64_MAX);
5919 VERIFY3U(0, ==, ztest_dataset_open(0));
5920 ztest_dataset_close(0);
5921
5922 spa->spa_debug = B_TRUE;
5923 ztest_spa = spa;
5924 txg_wait_synced(spa_get_dsl(spa), 0);
5925 ztest_reguid(NULL, 0);
5926
5927 spa_close(spa, FTAG);
5928 kernel_fini();
5929 }
5930
5931 void
5932 print_time(hrtime_t t, char *timebuf)
5933 {
5934 hrtime_t s = t / NANOSEC;
5935 hrtime_t m = s / 60;
5936 hrtime_t h = m / 60;
5937 hrtime_t d = h / 24;
5938
5939 s -= m * 60;
5940 m -= h * 60;
5941 h -= d * 24;
5942
5943 timebuf[0] = '\0';
5944
5945 if (d)
5946 (void) sprintf(timebuf,
5947 "%llud%02lluh%02llum%02llus", d, h, m, s);
5948 else if (h)
5949 (void) sprintf(timebuf, "%lluh%02llum%02llus", h, m, s);
5950 else if (m)
5951 (void) sprintf(timebuf, "%llum%02llus", m, s);
5952 else
5953 (void) sprintf(timebuf, "%llus", s);
5954 }
5955
5956 static nvlist_t *
5957 make_random_props()
5958 {
5959 nvlist_t *props;
5960
5961 VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
5962 if (ztest_random(2) == 0)
5963 return (props);
5964 VERIFY(nvlist_add_uint64(props, "autoreplace", 1) == 0);
5965
5966 return (props);
5967 }
5968
5969 /*
5970 * Create a storage pool with the given name and initial vdev size.
5971 * Then test spa_freeze() functionality.
5972 */
5973 static void
5974 ztest_init(ztest_shared_t *zs)
5975 {
5976 spa_t *spa;
5977 nvlist_t *nvroot, *props;
5978
5979 VERIFY(_mutex_init(&ztest_vdev_lock, USYNC_THREAD, NULL) == 0);
5980 VERIFY(rwlock_init(&ztest_name_lock, USYNC_THREAD, NULL) == 0);
5981
5982 kernel_init(FREAD | FWRITE);
5983
5984 /*
5985 * Create the storage pool.
5986 */
5987 (void) spa_destroy(ztest_opts.zo_pool);
5988 ztest_shared->zs_vdev_next_leaf = 0;
5989 zs->zs_splits = 0;
5990 zs->zs_mirrors = ztest_opts.zo_mirrors;
5991 nvroot = make_vdev_root(NULL, NULL, NULL, ztest_opts.zo_vdev_size, 0,
5992 0, ztest_opts.zo_raidz, zs->zs_mirrors, 1);
5993 props = make_random_props();
5994 for (int i = 0; i < SPA_FEATURES; i++) {
5995 char buf[1024];
5996 (void) snprintf(buf, sizeof (buf), "feature@%s",
5997 spa_feature_table[i].fi_uname);
5998 VERIFY3U(0, ==, nvlist_add_uint64(props, buf, 0));
5999 }
6000 VERIFY3U(0, ==, spa_create(ztest_opts.zo_pool, nvroot, props, NULL));
6001 nvlist_free(nvroot);
6002 nvlist_free(props);
6003
6004 VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
6005 zs->zs_metaslab_sz =
6006 1ULL << spa->spa_root_vdev->vdev_child[0]->vdev_ms_shift;
6007
6008 spa_close(spa, FTAG);
6009
6010 kernel_fini();
6011
6012 ztest_run_zdb(ztest_opts.zo_pool);
6013
6014 ztest_freeze();
6015
6016 ztest_run_zdb(ztest_opts.zo_pool);
6017
6018 (void) rwlock_destroy(&ztest_name_lock);
6019 (void) _mutex_destroy(&ztest_vdev_lock);
6020 }
6021
6022 static void
6023 setup_data_fd(void)
6024 {
6025 static char ztest_name_data[] = "/tmp/ztest.data.XXXXXX";
6026
6027 ztest_fd_data = mkstemp(ztest_name_data);
6028 ASSERT3S(ztest_fd_data, >=, 0);
6029 (void) unlink(ztest_name_data);
6030 }
6031
6032
6033 static int
6034 shared_data_size(ztest_shared_hdr_t *hdr)
6035 {
6036 int size;
6037
6038 size = hdr->zh_hdr_size;
6039 size += hdr->zh_opts_size;
6040 size += hdr->zh_size;
6041 size += hdr->zh_stats_size * hdr->zh_stats_count;
6042 size += hdr->zh_ds_size * hdr->zh_ds_count;
6043
6044 return (size);
6045 }
6046
6047 static void
6048 setup_hdr(void)
6049 {
6050 int size;
6051 ztest_shared_hdr_t *hdr;
6052
6053 hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
6054 PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
6055 ASSERT(hdr != MAP_FAILED);
6056
6057 VERIFY3U(0, ==, ftruncate(ztest_fd_data, sizeof (ztest_shared_hdr_t)));
6058
6059 hdr->zh_hdr_size = sizeof (ztest_shared_hdr_t);
6060 hdr->zh_opts_size = sizeof (ztest_shared_opts_t);
6061 hdr->zh_size = sizeof (ztest_shared_t);
6062 hdr->zh_stats_size = sizeof (ztest_shared_callstate_t);
6063 hdr->zh_stats_count = ZTEST_FUNCS;
6064 hdr->zh_ds_size = sizeof (ztest_shared_ds_t);
6065 hdr->zh_ds_count = ztest_opts.zo_datasets;
6066
6067 size = shared_data_size(hdr);
6068 VERIFY3U(0, ==, ftruncate(ztest_fd_data, size));
6069
6070 (void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
6071 }
6072
6073 static void
6074 setup_data(void)
6075 {
6076 int size, offset;
6077 ztest_shared_hdr_t *hdr;
6078 uint8_t *buf;
6079
6080 hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
6081 PROT_READ, MAP_SHARED, ztest_fd_data, 0);
6082 ASSERT(hdr != MAP_FAILED);
6083
6084 size = shared_data_size(hdr);
6085
6086 (void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
6087 hdr = ztest_shared_hdr = (void *)mmap(0, P2ROUNDUP(size, getpagesize()),
6088 PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
6089 ASSERT(hdr != MAP_FAILED);
6090 buf = (uint8_t *)hdr;
6091
6092 offset = hdr->zh_hdr_size;
6093 ztest_shared_opts = (void *)&buf[offset];
6094 offset += hdr->zh_opts_size;
6095 ztest_shared = (void *)&buf[offset];
6096 offset += hdr->zh_size;
6097 ztest_shared_callstate = (void *)&buf[offset];
6098 offset += hdr->zh_stats_size * hdr->zh_stats_count;
6099 ztest_shared_ds = (void *)&buf[offset];
6100 }
6101
6102 static boolean_t
6103 exec_child(char *cmd, char *libpath, boolean_t ignorekill, int *statusp)
6104 {
6105 pid_t pid;
6106 int status;
6107 char *cmdbuf = NULL;
6108
6109 pid = fork();
6110
6111 if (cmd == NULL) {
6112 cmdbuf = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
6113 (void) strlcpy(cmdbuf, getexecname(), MAXPATHLEN);
6114 cmd = cmdbuf;
6115 }
6116
6117 if (pid == -1)
6118 fatal(1, "fork failed");
6119
6120 if (pid == 0) { /* child */
6121 char *emptyargv[2] = { cmd, NULL };
6122 char fd_data_str[12];
6123
6124 struct rlimit rl = { 1024, 1024 };
6125 (void) setrlimit(RLIMIT_NOFILE, &rl);
6126
6127 (void) close(ztest_fd_rand);
6128 VERIFY3U(11, >=,
6129 snprintf(fd_data_str, 12, "%d", ztest_fd_data));
6130 VERIFY0(setenv("ZTEST_FD_DATA", fd_data_str, 1));
6131
6132 (void) enable_extended_FILE_stdio(-1, -1);
6133 if (libpath != NULL)
6134 VERIFY(0 == setenv("LD_LIBRARY_PATH", libpath, 1));
6135 (void) execv(cmd, emptyargv);
6136 ztest_dump_core = B_FALSE;
6137 fatal(B_TRUE, "exec failed: %s", cmd);
6138 }
6139
6140 if (cmdbuf != NULL) {
6141 umem_free(cmdbuf, MAXPATHLEN);
6142 cmd = NULL;
6143 }
6144
6145 while (waitpid(pid, &status, 0) != pid)
6146 continue;
6147 if (statusp != NULL)
6148 *statusp = status;
6149
6150 if (WIFEXITED(status)) {
6151 if (WEXITSTATUS(status) != 0) {
6152 (void) fprintf(stderr, "child exited with code %d\n",
6153 WEXITSTATUS(status));
6154 exit(2);
6155 }
6156 return (B_FALSE);
6157 } else if (WIFSIGNALED(status)) {
6158 if (!ignorekill || WTERMSIG(status) != SIGKILL) {
6159 (void) fprintf(stderr, "child died with signal %d\n",
6160 WTERMSIG(status));
6161 exit(3);
6162 }
6163 return (B_TRUE);
6164 } else {
6165 (void) fprintf(stderr, "something strange happened to child\n");
6166 exit(4);
6167 /* NOTREACHED */
6168 }
6169 }
6170
6171 static void
6172 ztest_run_init(void)
6173 {
6174 ztest_shared_t *zs = ztest_shared;
6175
6176 ASSERT(ztest_opts.zo_init != 0);
6177
6178 /*
6179 * Blow away any existing copy of zpool.cache
6180 */
6181 (void) remove(spa_config_path);
6182
6183 /*
6184 * Create and initialize our storage pool.
6185 */
6186 for (int i = 1; i <= ztest_opts.zo_init; i++) {
6187 bzero(zs, sizeof (ztest_shared_t));
6188 if (ztest_opts.zo_verbose >= 3 &&
6189 ztest_opts.zo_init != 1) {
6190 (void) printf("ztest_init(), pass %d\n", i);
6191 }
6192 ztest_init(zs);
6193 }
6194 }
6195
6196 int
6197 main(int argc, char **argv)
6198 {
6199 int kills = 0;
6200 int iters = 0;
6201 int older = 0;
6202 int newer = 0;
6203 ztest_shared_t *zs;
6204 ztest_info_t *zi;
6205 ztest_shared_callstate_t *zc;
6206 char timebuf[100];
6207 char numbuf[NN_NUMBUF_SZ];
6208 spa_t *spa;
6209 char *cmd;
6210 boolean_t hasalt;
6211 char *fd_data_str = getenv("ZTEST_FD_DATA");
6212
6213 (void) setvbuf(stdout, NULL, _IOLBF, 0);
6214
6215 dprintf_setup(&argc, argv);
6216 zfs_deadman_synctime_ms = 300000;
6217
6218 ztest_fd_rand = open("/dev/urandom", O_RDONLY);
6219 ASSERT3S(ztest_fd_rand, >=, 0);
6220
6221 if (!fd_data_str) {
6222 process_options(argc, argv);
6223
6224 setup_data_fd();
6225 setup_hdr();
6226 setup_data();
6227 bcopy(&ztest_opts, ztest_shared_opts,
6228 sizeof (*ztest_shared_opts));
6229 } else {
6230 ztest_fd_data = atoi(fd_data_str);
6231 setup_data();
6232 bcopy(ztest_shared_opts, &ztest_opts, sizeof (ztest_opts));
6233 }
6234 ASSERT3U(ztest_opts.zo_datasets, ==, ztest_shared_hdr->zh_ds_count);
6235
6236 /* Override location of zpool.cache */
6237 VERIFY3U(asprintf((char **)&spa_config_path, "%s/zpool.cache",
6238 ztest_opts.zo_dir), !=, -1);
6239
6240 ztest_ds = umem_alloc(ztest_opts.zo_datasets * sizeof (ztest_ds_t),
6241 UMEM_NOFAIL);
6242 zs = ztest_shared;
6243
6244 if (fd_data_str) {
6245 metaslab_gang_bang = ztest_opts.zo_metaslab_gang_bang;
6246 metaslab_df_alloc_threshold =
6247 zs->zs_metaslab_df_alloc_threshold;
6248
6249 if (zs->zs_do_init)
6250 ztest_run_init();
6251 else
6252 ztest_run(zs);
6253 exit(0);
6254 }
6255
6256 hasalt = (strlen(ztest_opts.zo_alt_ztest) != 0);
6257
6258 if (ztest_opts.zo_verbose >= 1) {
6259 (void) printf("%llu vdevs, %d datasets, %d threads,"
6260 " %llu seconds...\n",
6261 (u_longlong_t)ztest_opts.zo_vdevs,
6262 ztest_opts.zo_datasets,
6263 ztest_opts.zo_threads,
6264 (u_longlong_t)ztest_opts.zo_time);
6265 }
6266
6267 cmd = umem_alloc(MAXNAMELEN, UMEM_NOFAIL);
6268 (void) strlcpy(cmd, getexecname(), MAXNAMELEN);
6269
6270 zs->zs_do_init = B_TRUE;
6271 if (strlen(ztest_opts.zo_alt_ztest) != 0) {
6272 if (ztest_opts.zo_verbose >= 1) {
6273 (void) printf("Executing older ztest for "
6274 "initialization: %s\n", ztest_opts.zo_alt_ztest);
6275 }
6276 VERIFY(!exec_child(ztest_opts.zo_alt_ztest,
6277 ztest_opts.zo_alt_libpath, B_FALSE, NULL));
6278 } else {
6279 VERIFY(!exec_child(NULL, NULL, B_FALSE, NULL));
6280 }
6281 zs->zs_do_init = B_FALSE;
6282
6283 zs->zs_proc_start = gethrtime();
6284 zs->zs_proc_stop = zs->zs_proc_start + ztest_opts.zo_time * NANOSEC;
6285
6286 for (int f = 0; f < ZTEST_FUNCS; f++) {
6287 zi = &ztest_info[f];
6288 zc = ZTEST_GET_SHARED_CALLSTATE(f);
6289 if (zs->zs_proc_start + zi->zi_interval[0] > zs->zs_proc_stop)
6290 zc->zc_next = UINT64_MAX;
6291 else
6292 zc->zc_next = zs->zs_proc_start +
6293 ztest_random(2 * zi->zi_interval[0] + 1);
6294 }
6295
6296 /*
6297 * Run the tests in a loop. These tests include fault injection
6298 * to verify that self-healing data works, and forced crashes
6299 * to verify that we never lose on-disk consistency.
6300 */
6301 while (gethrtime() < zs->zs_proc_stop) {
6302 int status;
6303 boolean_t killed;
6304
6305 /*
6306 * Initialize the workload counters for each function.
6307 */
6308 for (int f = 0; f < ZTEST_FUNCS; f++) {
6309 zc = ZTEST_GET_SHARED_CALLSTATE(f);
6310 zc->zc_count = 0;
6311 zc->zc_time = 0;
6312 }
6313
6314 /* Set the allocation switch size */
6315 zs->zs_metaslab_df_alloc_threshold =
6316 ztest_random(zs->zs_metaslab_sz / 4) + 1;
6317
6318 if (!hasalt || ztest_random(2) == 0) {
6319 if (hasalt && ztest_opts.zo_verbose >= 1) {
6320 (void) printf("Executing newer ztest: %s\n",
6321 cmd);
6322 }
6323 newer++;
6324 killed = exec_child(cmd, NULL, B_TRUE, &status);
6325 } else {
6326 if (hasalt && ztest_opts.zo_verbose >= 1) {
6327 (void) printf("Executing older ztest: %s\n",
6328 ztest_opts.zo_alt_ztest);
6329 }
6330 older++;
6331 killed = exec_child(ztest_opts.zo_alt_ztest,
6332 ztest_opts.zo_alt_libpath, B_TRUE, &status);
6333 }
6334
6335 if (killed)
6336 kills++;
6337 iters++;
6338
6339 if (ztest_opts.zo_verbose >= 1) {
6340 hrtime_t now = gethrtime();
6341
6342 now = MIN(now, zs->zs_proc_stop);
6343 print_time(zs->zs_proc_stop - now, timebuf);
6344 nicenum(zs->zs_space, numbuf, sizeof (numbuf));
6345
6346 (void) printf("Pass %3d, %8s, %3llu ENOSPC, "
6347 "%4.1f%% of %5s used, %3.0f%% done, %8s to go\n",
6348 iters,
6349 WIFEXITED(status) ? "Complete" : "SIGKILL",
6350 (u_longlong_t)zs->zs_enospc_count,
6351 100.0 * zs->zs_alloc / zs->zs_space,
6352 numbuf,
6353 100.0 * (now - zs->zs_proc_start) /
6354 (ztest_opts.zo_time * NANOSEC), timebuf);
6355 }
6356
6357 if (ztest_opts.zo_verbose >= 2) {
6358 (void) printf("\nWorkload summary:\n\n");
6359 (void) printf("%7s %9s %s\n",
6360 "Calls", "Time", "Function");
6361 (void) printf("%7s %9s %s\n",
6362 "-----", "----", "--------");
6363 for (int f = 0; f < ZTEST_FUNCS; f++) {
6364 Dl_info dli;
6365
6366 zi = &ztest_info[f];
6367 zc = ZTEST_GET_SHARED_CALLSTATE(f);
6368 print_time(zc->zc_time, timebuf);
6369 (void) dladdr((void *)zi->zi_func, &dli);
6370 (void) printf("%7llu %9s %s\n",
6371 (u_longlong_t)zc->zc_count, timebuf,
6372 dli.dli_sname);
6373 }
6374 (void) printf("\n");
6375 }
6376
6377 /*
6378 * It's possible that we killed a child during a rename test,
6379 * in which case we'll have a 'ztest_tmp' pool lying around
6380 * instead of 'ztest'. Do a blind rename in case this happened.
6381 */
6382 kernel_init(FREAD);
6383 if (spa_open(ztest_opts.zo_pool, &spa, FTAG) == 0) {
6384 spa_close(spa, FTAG);
6385 } else {
6386 char tmpname[ZFS_MAX_DATASET_NAME_LEN];
6387 kernel_fini();
6388 kernel_init(FREAD | FWRITE);
6389 (void) snprintf(tmpname, sizeof (tmpname), "%s_tmp",
6390 ztest_opts.zo_pool);
6391 (void) spa_rename(tmpname, ztest_opts.zo_pool);
6392 }
6393 kernel_fini();
6394
6395 ztest_run_zdb(ztest_opts.zo_pool);
6396 }
6397
6398 if (ztest_opts.zo_verbose >= 1) {
6399 if (hasalt) {
6400 (void) printf("%d runs of older ztest: %s\n", older,
6401 ztest_opts.zo_alt_ztest);
6402 (void) printf("%d runs of newer ztest: %s\n", newer,
6403 cmd);
6404 }
6405 (void) printf("%d killed, %d completed, %.0f%% kill rate\n",
6406 kills, iters - kills, (100.0 * kills) / MAX(1, iters));
6407 }
6408
6409 umem_free(cmd, MAXNAMELEN);
6410
6411 return (0);
6412 }
Unleashed OS