?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
8322 nl: misleading-indentation
[unleashed/tickless.git] / usr / src / cmd / ztest / ztest.c
blob91abd30e9c57271d278aab4bdc2fb92b7bd17417
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 */
28
29 /*
30 * The objective of this program is to provide a DMU/ZAP/SPA stress test
31 * that runs entirely in userland, is easy to use, and easy to extend.
32 *
33 * The overall design of the ztest program is as follows:
34 *
35 * (1) For each major functional area (e.g. adding vdevs to a pool,
36 * creating and destroying datasets, reading and writing objects, etc)
37 * we have a simple routine to test that functionality. These
38 * individual routines do not have to do anything "stressful".
39 *
40 * (2) We turn these simple functionality tests into a stress test by
41 * running them all in parallel, with as many threads as desired,
42 * and spread across as many datasets, objects, and vdevs as desired.
43 *
44 * (3) While all this is happening, we inject faults into the pool to
45 * verify that self-healing data really works.
46 *
47 * (4) Every time we open a dataset, we change its checksum and compression
48 * functions. Thus even individual objects vary from block to block
49 * in which checksum they use and whether they're compressed.
50 *
51 * (5) To verify that we never lose on-disk consistency after a crash,
52 * we run the entire test in a child of the main process.
53 * At random times, the child self-immolates with a SIGKILL.
54 * This is the software equivalent of pulling the power cord.
55 * The parent then runs the test again, using the existing
56 * storage pool, as many times as desired. If backwards compatibility
57 * testing is enabled ztest will sometimes run the "older" version
58 * of ztest after a SIGKILL.
59 *
60 * (6) To verify that we don't have future leaks or temporal incursions,
61 * many of the functional tests record the transaction group number
62 * as part of their data. When reading old data, they verify that
63 * the transaction group number is less than the current, open txg.
64 * If you add a new test, please do this if applicable.
65 *
66 * When run with no arguments, ztest runs for about five minutes and
67 * produces no output if successful. To get a little bit of information,
68 * specify -V. To get more information, specify -VV, and so on.
69 *
70 * To turn this into an overnight stress test, use -T to specify run time.
71 *
72 * You can ask more more vdevs [-v], datasets [-d], or threads [-t]
73 * to increase the pool capacity, fanout, and overall stress level.
74 *
75 * Use the -k option to set the desired frequency of kills.
76 *
77 * When ztest invokes itself it passes all relevant information through a
78 * temporary file which is mmap-ed in the child process. This allows shared
79 * memory to survive the exec syscall. The ztest_shared_hdr_t struct is always
80 * stored at offset 0 of this file and contains information on the size and
81 * number of shared structures in the file. The information stored in this file
82 * must remain backwards compatible with older versions of ztest so that
83 * ztest can invoke them during backwards compatibility testing (-B).
84 */
85
86 #include <sys/zfs_context.h>
87 #include <sys/spa.h>
88 #include <sys/dmu.h>
89 #include <sys/txg.h>
90 #include <sys/dbuf.h>
91 #include <sys/zap.h>
92 #include <sys/dmu_objset.h>
93 #include <sys/poll.h>
94 #include <sys/stat.h>
95 #include <sys/time.h>
96 #include <sys/wait.h>
97 #include <sys/mman.h>
98 #include <sys/resource.h>
99 #include <sys/zio.h>
100 #include <sys/zil.h>
101 #include <sys/zil_impl.h>
102 #include <sys/vdev_impl.h>
103 #include <sys/vdev_file.h>
104 #include <sys/spa_impl.h>
105 #include <sys/metaslab_impl.h>
106 #include <sys/dsl_prop.h>
107 #include <sys/dsl_dataset.h>
108 #include <sys/dsl_destroy.h>
109 #include <sys/dsl_scan.h>
110 #include <sys/zio_checksum.h>
111 #include <sys/refcount.h>
112 #include <sys/zfeature.h>
113 #include <sys/dsl_userhold.h>
114 #include <sys/abd.h>
115 #include <stdio.h>
116 #include <stdio_ext.h>
117 #include <stdlib.h>
118 #include <unistd.h>
119 #include <signal.h>
120 #include <umem.h>
121 #include <dlfcn.h>
122 #include <ctype.h>
123 #include <math.h>
124 #include <sys/fs/zfs.h>
125 #include <libnvpair.h>
126
127 static int ztest_fd_data = -1;
128 static int ztest_fd_rand = -1;
129
130 typedef struct ztest_shared_hdr {
131 uint64_t zh_hdr_size;
132 uint64_t zh_opts_size;
133 uint64_t zh_size;
134 uint64_t zh_stats_size;
135 uint64_t zh_stats_count;
136 uint64_t zh_ds_size;
137 uint64_t zh_ds_count;
138 } ztest_shared_hdr_t;
139
140 static ztest_shared_hdr_t *ztest_shared_hdr;
141
142 typedef struct ztest_shared_opts {
143 char zo_pool[ZFS_MAX_DATASET_NAME_LEN];
144 char zo_dir[ZFS_MAX_DATASET_NAME_LEN];
145 char zo_alt_ztest[MAXNAMELEN];
146 char zo_alt_libpath[MAXNAMELEN];
147 uint64_t zo_vdevs;
148 uint64_t zo_vdevtime;
149 size_t zo_vdev_size;
150 int zo_ashift;
151 int zo_mirrors;
152 int zo_raidz;
153 int zo_raidz_parity;
154 int zo_datasets;
155 int zo_threads;
156 uint64_t zo_passtime;
157 uint64_t zo_killrate;
158 int zo_verbose;
159 int zo_init;
160 uint64_t zo_time;
161 uint64_t zo_maxloops;
162 uint64_t zo_metaslab_gang_bang;
163 } ztest_shared_opts_t;
164
165 static const ztest_shared_opts_t ztest_opts_defaults = {
166 .zo_pool = { 'z', 't', 'e', 's', 't', '\0' },
167 .zo_dir = { '/', 't', 'm', 'p', '\0' },
168 .zo_alt_ztest = { '\0' },
169 .zo_alt_libpath = { '\0' },
170 .zo_vdevs = 5,
171 .zo_ashift = SPA_MINBLOCKSHIFT,
172 .zo_mirrors = 2,
173 .zo_raidz = 4,
174 .zo_raidz_parity = 1,
175 .zo_vdev_size = SPA_MINDEVSIZE * 4, /* 256m default size */
176 .zo_datasets = 7,
177 .zo_threads = 23,
178 .zo_passtime = 60, /* 60 seconds */
179 .zo_killrate = 70, /* 70% kill rate */
180 .zo_verbose = 0,
181 .zo_init = 1,
182 .zo_time = 300, /* 5 minutes */
183 .zo_maxloops = 50, /* max loops during spa_freeze() */
184 .zo_metaslab_gang_bang = 32 << 10
185 };
186
187 extern uint64_t metaslab_gang_bang;
188 extern uint64_t metaslab_df_alloc_threshold;
189 extern uint64_t zfs_deadman_synctime_ms;
190 extern int metaslab_preload_limit;
191 extern boolean_t zfs_compressed_arc_enabled;
192 extern boolean_t zfs_abd_scatter_enabled;
193
194 static ztest_shared_opts_t *ztest_shared_opts;
195 static ztest_shared_opts_t ztest_opts;
196
197 typedef struct ztest_shared_ds {
198 uint64_t zd_seq;
199 } ztest_shared_ds_t;
200
201 static ztest_shared_ds_t *ztest_shared_ds;
202 #define ZTEST_GET_SHARED_DS(d) (&ztest_shared_ds[d])
203
204 #define BT_MAGIC 0x123456789abcdefULL
205 #define MAXFAULTS() \
206 (MAX(zs->zs_mirrors, 1) * (ztest_opts.zo_raidz_parity + 1) - 1)
207
208 enum ztest_io_type {
209 ZTEST_IO_WRITE_TAG,
210 ZTEST_IO_WRITE_PATTERN,
211 ZTEST_IO_WRITE_ZEROES,
212 ZTEST_IO_TRUNCATE,
213 ZTEST_IO_SETATTR,
214 ZTEST_IO_REWRITE,
215 ZTEST_IO_TYPES
216 };
217
218 typedef struct ztest_block_tag {
219 uint64_t bt_magic;
220 uint64_t bt_objset;
221 uint64_t bt_object;
222 uint64_t bt_offset;
223 uint64_t bt_gen;
224 uint64_t bt_txg;
225 uint64_t bt_crtxg;
226 } ztest_block_tag_t;
227
228 typedef struct bufwad {
229 uint64_t bw_index;
230 uint64_t bw_txg;
231 uint64_t bw_data;
232 } bufwad_t;
233
234 /*
235 * XXX -- fix zfs range locks to be generic so we can use them here.
236 */
237 typedef enum {
238 RL_READER,
239 RL_WRITER,
240 RL_APPEND
241 } rl_type_t;
242
243 typedef struct rll {
244 void *rll_writer;
245 int rll_readers;
246 mutex_t rll_lock;
247 cond_t rll_cv;
248 } rll_t;
249
250 typedef struct rl {
251 uint64_t rl_object;
252 uint64_t rl_offset;
253 uint64_t rl_size;
254 rll_t *rl_lock;
255 } rl_t;
256
257 #define ZTEST_RANGE_LOCKS 64
258 #define ZTEST_OBJECT_LOCKS 64
259
260 /*
261 * Object descriptor. Used as a template for object lookup/create/remove.
262 */
263 typedef struct ztest_od {
264 uint64_t od_dir;
265 uint64_t od_object;
266 dmu_object_type_t od_type;
267 dmu_object_type_t od_crtype;
268 uint64_t od_blocksize;
269 uint64_t od_crblocksize;
270 uint64_t od_gen;
271 uint64_t od_crgen;
272 char od_name[ZFS_MAX_DATASET_NAME_LEN];
273 } ztest_od_t;
274
275 /*
276 * Per-dataset state.
277 */
278 typedef struct ztest_ds {
279 ztest_shared_ds_t *zd_shared;
280 objset_t *zd_os;
281 rwlock_t zd_zilog_lock;
282 zilog_t *zd_zilog;
283 ztest_od_t *zd_od; /* debugging aid */
284 char zd_name[ZFS_MAX_DATASET_NAME_LEN];
285 mutex_t zd_dirobj_lock;
286 rll_t zd_object_lock[ZTEST_OBJECT_LOCKS];
287 rll_t zd_range_lock[ZTEST_RANGE_LOCKS];
288 } ztest_ds_t;
289
290 /*
291 * Per-iteration state.
292 */
293 typedef void ztest_func_t(ztest_ds_t *zd, uint64_t id);
294
295 typedef struct ztest_info {
296 ztest_func_t *zi_func; /* test function */
297 uint64_t zi_iters; /* iterations per execution */
298 uint64_t *zi_interval; /* execute every <interval> seconds */
299 } ztest_info_t;
300
301 typedef struct ztest_shared_callstate {
302 uint64_t zc_count; /* per-pass count */
303 uint64_t zc_time; /* per-pass time */
304 uint64_t zc_next; /* next time to call this function */
305 } ztest_shared_callstate_t;
306
307 static ztest_shared_callstate_t *ztest_shared_callstate;
308 #define ZTEST_GET_SHARED_CALLSTATE(c) (&ztest_shared_callstate[c])
309
310 /*
311 * Note: these aren't static because we want dladdr() to work.
312 */
313 ztest_func_t ztest_dmu_read_write;
314 ztest_func_t ztest_dmu_write_parallel;
315 ztest_func_t ztest_dmu_object_alloc_free;
316 ztest_func_t ztest_dmu_commit_callbacks;
317 ztest_func_t ztest_zap;
318 ztest_func_t ztest_zap_parallel;
319 ztest_func_t ztest_zil_commit;
320 ztest_func_t ztest_zil_remount;
321 ztest_func_t ztest_dmu_read_write_zcopy;
322 ztest_func_t ztest_dmu_objset_create_destroy;
323 ztest_func_t ztest_dmu_prealloc;
324 ztest_func_t ztest_fzap;
325 ztest_func_t ztest_dmu_snapshot_create_destroy;
326 ztest_func_t ztest_dsl_prop_get_set;
327 ztest_func_t ztest_spa_prop_get_set;
328 ztest_func_t ztest_spa_create_destroy;
329 ztest_func_t ztest_fault_inject;
330 ztest_func_t ztest_ddt_repair;
331 ztest_func_t ztest_dmu_snapshot_hold;
332 ztest_func_t ztest_spa_rename;
333 ztest_func_t ztest_scrub;
334 ztest_func_t ztest_dsl_dataset_promote_busy;
335 ztest_func_t ztest_vdev_attach_detach;
336 ztest_func_t ztest_vdev_LUN_growth;
337 ztest_func_t ztest_vdev_add_remove;
338 ztest_func_t ztest_vdev_aux_add_remove;
339 ztest_func_t ztest_split_pool;
340 ztest_func_t ztest_reguid;
341 ztest_func_t ztest_spa_upgrade;
342
343 uint64_t zopt_always = 0ULL * NANOSEC; /* all the time */
344 uint64_t zopt_incessant = 1ULL * NANOSEC / 10; /* every 1/10 second */
345 uint64_t zopt_often = 1ULL * NANOSEC; /* every second */
346 uint64_t zopt_sometimes = 10ULL * NANOSEC; /* every 10 seconds */
347 uint64_t zopt_rarely = 60ULL * NANOSEC; /* every 60 seconds */
348
349 ztest_info_t ztest_info[] = {
350 { ztest_dmu_read_write, 1, &zopt_always },
351 { ztest_dmu_write_parallel, 10, &zopt_always },
352 { ztest_dmu_object_alloc_free, 1, &zopt_always },
353 { ztest_dmu_commit_callbacks, 1, &zopt_always },
354 { ztest_zap, 30, &zopt_always },
355 { ztest_zap_parallel, 100, &zopt_always },
356 { ztest_split_pool, 1, &zopt_always },
357 { ztest_zil_commit, 1, &zopt_incessant },
358 { ztest_zil_remount, 1, &zopt_sometimes },
359 { ztest_dmu_read_write_zcopy, 1, &zopt_often },
360 { ztest_dmu_objset_create_destroy, 1, &zopt_often },
361 { ztest_dsl_prop_get_set, 1, &zopt_often },
362 { ztest_spa_prop_get_set, 1, &zopt_sometimes },
363 #if 0
364 { ztest_dmu_prealloc, 1, &zopt_sometimes },
365 #endif
366 { ztest_fzap, 1, &zopt_sometimes },
367 { ztest_dmu_snapshot_create_destroy, 1, &zopt_sometimes },
368 { ztest_spa_create_destroy, 1, &zopt_sometimes },
369 { ztest_fault_inject, 1, &zopt_sometimes },
370 { ztest_ddt_repair, 1, &zopt_sometimes },
371 { ztest_dmu_snapshot_hold, 1, &zopt_sometimes },
372 { ztest_reguid, 1, &zopt_rarely },
373 { ztest_spa_rename, 1, &zopt_rarely },
374 { ztest_scrub, 1, &zopt_rarely },
375 { ztest_spa_upgrade, 1, &zopt_rarely },
376 { ztest_dsl_dataset_promote_busy, 1, &zopt_rarely },
377 { ztest_vdev_attach_detach, 1, &zopt_sometimes },
378 { ztest_vdev_LUN_growth, 1, &zopt_rarely },
379 { ztest_vdev_add_remove, 1,
380 &ztest_opts.zo_vdevtime },
381 { ztest_vdev_aux_add_remove, 1,
382 &ztest_opts.zo_vdevtime },
383 };
384
385 #define ZTEST_FUNCS (sizeof (ztest_info) / sizeof (ztest_info_t))
386
387 /*
388 * The following struct is used to hold a list of uncalled commit callbacks.
389 * The callbacks are ordered by txg number.
390 */
391 typedef struct ztest_cb_list {
392 mutex_t zcl_callbacks_lock;
393 list_t zcl_callbacks;
394 } ztest_cb_list_t;
395
396 /*
397 * Stuff we need to share writably between parent and child.
398 */
399 typedef struct ztest_shared {
400 boolean_t zs_do_init;
401 hrtime_t zs_proc_start;
402 hrtime_t zs_proc_stop;
403 hrtime_t zs_thread_start;
404 hrtime_t zs_thread_stop;
405 hrtime_t zs_thread_kill;
406 uint64_t zs_enospc_count;
407 uint64_t zs_vdev_next_leaf;
408 uint64_t zs_vdev_aux;
409 uint64_t zs_alloc;
410 uint64_t zs_space;
411 uint64_t zs_splits;
412 uint64_t zs_mirrors;
413 uint64_t zs_metaslab_sz;
414 uint64_t zs_metaslab_df_alloc_threshold;
415 uint64_t zs_guid;
416 } ztest_shared_t;
417
418 #define ID_PARALLEL -1ULL
419
420 static char ztest_dev_template[] = "%s/%s.%llua";
421 static char ztest_aux_template[] = "%s/%s.%s.%llu";
422 ztest_shared_t *ztest_shared;
423
424 static spa_t *ztest_spa = NULL;
425 static ztest_ds_t *ztest_ds;
426
427 static mutex_t ztest_vdev_lock;
428
429 /*
430 * The ztest_name_lock protects the pool and dataset namespace used by
431 * the individual tests. To modify the namespace, consumers must grab
432 * this lock as writer. Grabbing the lock as reader will ensure that the
433 * namespace does not change while the lock is held.
434 */
435 static rwlock_t ztest_name_lock;
436
437 static boolean_t ztest_dump_core = B_TRUE;
438 static boolean_t ztest_exiting;
439
440 /* Global commit callback list */
441 static ztest_cb_list_t zcl;
442
443 enum ztest_object {
444 ZTEST_META_DNODE = 0,
445 ZTEST_DIROBJ,
446 ZTEST_OBJECTS
447 };
448
449 static void usage(boolean_t) __NORETURN;
450
451 /*
452 * These libumem hooks provide a reasonable set of defaults for the allocator's
453 * debugging facilities.
454 */
455 const char *
456 _umem_debug_init()
457 {
458 return ("default,verbose"); /* $UMEM_DEBUG setting */
459 }
460
461 const char *
462 _umem_logging_init(void)
463 {
464 return ("fail,contents"); /* $UMEM_LOGGING setting */
465 }
466
467 #define FATAL_MSG_SZ 1024
468
469 char *fatal_msg;
470
471 static void
472 fatal(int do_perror, char *message, ...)
473 {
474 va_list args;
475 int save_errno = errno;
476 char buf[FATAL_MSG_SZ];
477
478 (void) fflush(stdout);
479
480 va_start(args, message);
481 (void) sprintf(buf, "ztest: ");
482 /* LINTED */
483 (void) vsprintf(buf + strlen(buf), message, args);
484 va_end(args);
485 if (do_perror) {
486 (void) snprintf(buf + strlen(buf), FATAL_MSG_SZ - strlen(buf),
487 ": %s", strerror(save_errno));
488 }
489 (void) fprintf(stderr, "%s\n", buf);
490 fatal_msg = buf; /* to ease debugging */
491 if (ztest_dump_core)
492 abort();
493 exit(3);
494 }
495
496 static int
497 str2shift(const char *buf)
498 {
499 const char *ends = "BKMGTPEZ";
500 int i;
501
502 if (buf[0] == '\0')
503 return (0);
504 for (i = 0; i < strlen(ends); i++) {
505 if (toupper(buf[0]) == ends[i])
506 break;
507 }
508 if (i == strlen(ends)) {
509 (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n",
510 buf);
511 usage(B_FALSE);
512 }
513 if (buf[1] == '\0' || (toupper(buf[1]) == 'B' && buf[2] == '\0')) {
514 return (10*i);
515 }
516 (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n", buf);
517 usage(B_FALSE);
518 /* NOTREACHED */
519 }
520
521 static uint64_t
522 nicenumtoull(const char *buf)
523 {
524 char *end;
525 uint64_t val;
526
527 val = strtoull(buf, &end, 0);
528 if (end == buf) {
529 (void) fprintf(stderr, "ztest: bad numeric value: %s\n", buf);
530 usage(B_FALSE);
531 } else if (end[0] == '.') {
532 double fval = strtod(buf, &end);
533 fval *= pow(2, str2shift(end));
534 if (fval > UINT64_MAX) {
535 (void) fprintf(stderr, "ztest: value too large: %s\n",
536 buf);
537 usage(B_FALSE);
538 }
539 val = (uint64_t)fval;
540 } else {
541 int shift = str2shift(end);
542 if (shift >= 64 || (val << shift) >> shift != val) {
543 (void) fprintf(stderr, "ztest: value too large: %s\n",
544 buf);
545 usage(B_FALSE);
546 }
547 val <<= shift;
548 }
549 return (val);
550 }
551
552 static void
553 usage(boolean_t requested)
554 {
555 const ztest_shared_opts_t *zo = &ztest_opts_defaults;
556
557 char nice_vdev_size[10];
558 char nice_gang_bang[10];
559 FILE *fp = requested ? stdout : stderr;
560
561 nicenum(zo->zo_vdev_size, nice_vdev_size);
562 nicenum(zo->zo_metaslab_gang_bang, nice_gang_bang);
563
564 (void) fprintf(fp, "Usage: %s\n"
565 "\t[-v vdevs (default: %llu)]\n"
566 "\t[-s size_of_each_vdev (default: %s)]\n"
567 "\t[-a alignment_shift (default: %d)] use 0 for random\n"
568 "\t[-m mirror_copies (default: %d)]\n"
569 "\t[-r raidz_disks (default: %d)]\n"
570 "\t[-R raidz_parity (default: %d)]\n"
571 "\t[-d datasets (default: %d)]\n"
572 "\t[-t threads (default: %d)]\n"
573 "\t[-g gang_block_threshold (default: %s)]\n"
574 "\t[-i init_count (default: %d)] initialize pool i times\n"
575 "\t[-k kill_percentage (default: %llu%%)]\n"
576 "\t[-p pool_name (default: %s)]\n"
577 "\t[-f dir (default: %s)] file directory for vdev files\n"
578 "\t[-V] verbose (use multiple times for ever more blather)\n"
579 "\t[-E] use existing pool instead of creating new one\n"
580 "\t[-T time (default: %llu sec)] total run time\n"
581 "\t[-F freezeloops (default: %llu)] max loops in spa_freeze()\n"
582 "\t[-P passtime (default: %llu sec)] time per pass\n"
583 "\t[-B alt_ztest (default: <none>)] alternate ztest path\n"
584 "\t[-o variable=value] ... set global variable to an unsigned\n"
585 "\t 32-bit integer value\n"
586 "\t[-h] (print help)\n"
587 "",
588 zo->zo_pool,
589 (u_longlong_t)zo->zo_vdevs, /* -v */
590 nice_vdev_size, /* -s */
591 zo->zo_ashift, /* -a */
592 zo->zo_mirrors, /* -m */
593 zo->zo_raidz, /* -r */
594 zo->zo_raidz_parity, /* -R */
595 zo->zo_datasets, /* -d */
596 zo->zo_threads, /* -t */
597 nice_gang_bang, /* -g */
598 zo->zo_init, /* -i */
599 (u_longlong_t)zo->zo_killrate, /* -k */
600 zo->zo_pool, /* -p */
601 zo->zo_dir, /* -f */
602 (u_longlong_t)zo->zo_time, /* -T */
603 (u_longlong_t)zo->zo_maxloops, /* -F */
604 (u_longlong_t)zo->zo_passtime);
605 exit(requested ? 0 : 1);
606 }
607
608 static void
609 process_options(int argc, char **argv)
610 {
611 char *path;
612 ztest_shared_opts_t *zo = &ztest_opts;
613
614 int opt;
615 uint64_t value;
616 char altdir[MAXNAMELEN] = { 0 };
617
618 bcopy(&ztest_opts_defaults, zo, sizeof (*zo));
619
620 while ((opt = getopt(argc, argv,
621 "v:s:a:m:r:R:d:t:g:i:k:p:f:VET:P:hF:B:o:")) != EOF) {
622 value = 0;
623 switch (opt) {
624 case 'v':
625 case 's':
626 case 'a':
627 case 'm':
628 case 'r':
629 case 'R':
630 case 'd':
631 case 't':
632 case 'g':
633 case 'i':
634 case 'k':
635 case 'T':
636 case 'P':
637 case 'F':
638 value = nicenumtoull(optarg);
639 }
640 switch (opt) {
641 case 'v':
642 zo->zo_vdevs = value;
643 break;
644 case 's':
645 zo->zo_vdev_size = MAX(SPA_MINDEVSIZE, value);
646 break;
647 case 'a':
648 zo->zo_ashift = value;
649 break;
650 case 'm':
651 zo->zo_mirrors = value;
652 break;
653 case 'r':
654 zo->zo_raidz = MAX(1, value);
655 break;
656 case 'R':
657 zo->zo_raidz_parity = MIN(MAX(value, 1), 3);
658 break;
659 case 'd':
660 zo->zo_datasets = MAX(1, value);
661 break;
662 case 't':
663 zo->zo_threads = MAX(1, value);
664 break;
665 case 'g':
666 zo->zo_metaslab_gang_bang = MAX(SPA_MINBLOCKSIZE << 1,
667 value);
668 break;
669 case 'i':
670 zo->zo_init = value;
671 break;
672 case 'k':
673 zo->zo_killrate = value;
674 break;
675 case 'p':
676 (void) strlcpy(zo->zo_pool, optarg,
677 sizeof (zo->zo_pool));
678 break;
679 case 'f':
680 path = realpath(optarg, NULL);
681 if (path == NULL) {
682 (void) fprintf(stderr, "error: %s: %s\n",
683 optarg, strerror(errno));
684 usage(B_FALSE);
685 } else {
686 (void) strlcpy(zo->zo_dir, path,
687 sizeof (zo->zo_dir));
688 }
689 break;
690 case 'V':
691 zo->zo_verbose++;
692 break;
693 case 'E':
694 zo->zo_init = 0;
695 break;
696 case 'T':
697 zo->zo_time = value;
698 break;
699 case 'P':
700 zo->zo_passtime = MAX(1, value);
701 break;
702 case 'F':
703 zo->zo_maxloops = MAX(1, value);
704 break;
705 case 'B':
706 (void) strlcpy(altdir, optarg, sizeof (altdir));
707 break;
708 case 'o':
709 if (set_global_var(optarg) != 0)
710 usage(B_FALSE);
711 break;
712 case 'h':
713 usage(B_TRUE);
714 break;
715 case '?':
716 default:
717 usage(B_FALSE);
718 break;
719 }
720 }
721
722 zo->zo_raidz_parity = MIN(zo->zo_raidz_parity, zo->zo_raidz - 1);
723
724 zo->zo_vdevtime =
725 (zo->zo_vdevs > 0 ? zo->zo_time * NANOSEC / zo->zo_vdevs :
726 UINT64_MAX >> 2);
727
728 if (strlen(altdir) > 0) {
729 char *cmd;
730 char *realaltdir;
731 char *bin;
732 char *ztest;
733 char *isa;
734 int isalen;
735
736 cmd = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
737 realaltdir = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
738
739 VERIFY(NULL != realpath(getexecname(), cmd));
740 if (0 != access(altdir, F_OK)) {
741 ztest_dump_core = B_FALSE;
742 fatal(B_TRUE, "invalid alternate ztest path: %s",
743 altdir);
744 }
745 VERIFY(NULL != realpath(altdir, realaltdir));
746
747 /*
748 * 'cmd' should be of the form "<anything>/usr/bin/<isa>/ztest".
749 * We want to extract <isa> to determine if we should use
750 * 32 or 64 bit binaries.
751 */
752 bin = strstr(cmd, "/usr/bin/");
753 ztest = strstr(bin, "/ztest");
754 isa = bin + 9;
755 isalen = ztest - isa;
756 (void) snprintf(zo->zo_alt_ztest, sizeof (zo->zo_alt_ztest),
757 "%s/usr/bin/%.*s/ztest", realaltdir, isalen, isa);
758 (void) snprintf(zo->zo_alt_libpath, sizeof (zo->zo_alt_libpath),
759 "%s/usr/lib/%.*s", realaltdir, isalen, isa);
760
761 if (0 != access(zo->zo_alt_ztest, X_OK)) {
762 ztest_dump_core = B_FALSE;
763 fatal(B_TRUE, "invalid alternate ztest: %s",
764 zo->zo_alt_ztest);
765 } else if (0 != access(zo->zo_alt_libpath, X_OK)) {
766 ztest_dump_core = B_FALSE;
767 fatal(B_TRUE, "invalid alternate lib directory %s",
768 zo->zo_alt_libpath);
769 }
770
771 umem_free(cmd, MAXPATHLEN);
772 umem_free(realaltdir, MAXPATHLEN);
773 }
774 }
775
776 static void
777 ztest_kill(ztest_shared_t *zs)
778 {
779 zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(ztest_spa));
780 zs->zs_space = metaslab_class_get_space(spa_normal_class(ztest_spa));
781
782 /*
783 * Before we kill off ztest, make sure that the config is updated.
784 * See comment above spa_config_sync().
785 */
786 mutex_enter(&spa_namespace_lock);
787 spa_config_sync(ztest_spa, B_FALSE, B_FALSE);
788 mutex_exit(&spa_namespace_lock);
789
790 zfs_dbgmsg_print(FTAG);
791 (void) kill(getpid(), SIGKILL);
792 }
793
794 static uint64_t
795 ztest_random(uint64_t range)
796 {
797 uint64_t r;
798
799 ASSERT3S(ztest_fd_rand, >=, 0);
800
801 if (range == 0)
802 return (0);
803
804 if (read(ztest_fd_rand, &r, sizeof (r)) != sizeof (r))
805 fatal(1, "short read from /dev/urandom");
806
807 return (r % range);
808 }
809
810 /* ARGSUSED */
811 static void
812 ztest_record_enospc(const char *s)
813 {
814 ztest_shared->zs_enospc_count++;
815 }
816
817 static uint64_t
818 ztest_get_ashift(void)
819 {
820 if (ztest_opts.zo_ashift == 0)
821 return (SPA_MINBLOCKSHIFT + ztest_random(5));
822 return (ztest_opts.zo_ashift);
823 }
824
825 static nvlist_t *
826 make_vdev_file(char *path, char *aux, char *pool, size_t size, uint64_t ashift)
827 {
828 char pathbuf[MAXPATHLEN];
829 uint64_t vdev;
830 nvlist_t *file;
831
832 if (ashift == 0)
833 ashift = ztest_get_ashift();
834
835 if (path == NULL) {
836 path = pathbuf;
837
838 if (aux != NULL) {
839 vdev = ztest_shared->zs_vdev_aux;
840 (void) snprintf(path, sizeof (pathbuf),
841 ztest_aux_template, ztest_opts.zo_dir,
842 pool == NULL ? ztest_opts.zo_pool : pool,
843 aux, vdev);
844 } else {
845 vdev = ztest_shared->zs_vdev_next_leaf++;
846 (void) snprintf(path, sizeof (pathbuf),
847 ztest_dev_template, ztest_opts.zo_dir,
848 pool == NULL ? ztest_opts.zo_pool : pool, vdev);
849 }
850 }
851
852 if (size != 0) {
853 int fd = open(path, O_RDWR | O_CREAT | O_TRUNC, 0666);
854 if (fd == -1)
855 fatal(1, "can't open %s", path);
856 if (ftruncate(fd, size) != 0)
857 fatal(1, "can't ftruncate %s", path);
858 (void) close(fd);
859 }
860
861 VERIFY(nvlist_alloc(&file, NV_UNIQUE_NAME, 0) == 0);
862 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_TYPE, VDEV_TYPE_FILE) == 0);
863 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_PATH, path) == 0);
864 VERIFY(nvlist_add_uint64(file, ZPOOL_CONFIG_ASHIFT, ashift) == 0);
865
866 return (file);
867 }
868
869 static nvlist_t *
870 make_vdev_raidz(char *path, char *aux, char *pool, size_t size,
871 uint64_t ashift, int r)
872 {
873 nvlist_t *raidz, **child;
874 int c;
875
876 if (r < 2)
877 return (make_vdev_file(path, aux, pool, size, ashift));
878 child = umem_alloc(r * sizeof (nvlist_t *), UMEM_NOFAIL);
879
880 for (c = 0; c < r; c++)
881 child[c] = make_vdev_file(path, aux, pool, size, ashift);
882
883 VERIFY(nvlist_alloc(&raidz, NV_UNIQUE_NAME, 0) == 0);
884 VERIFY(nvlist_add_string(raidz, ZPOOL_CONFIG_TYPE,
885 VDEV_TYPE_RAIDZ) == 0);
886 VERIFY(nvlist_add_uint64(raidz, ZPOOL_CONFIG_NPARITY,
887 ztest_opts.zo_raidz_parity) == 0);
888 VERIFY(nvlist_add_nvlist_array(raidz, ZPOOL_CONFIG_CHILDREN,
889 child, r) == 0);
890
891 for (c = 0; c < r; c++)
892 nvlist_free(child[c]);
893
894 umem_free(child, r * sizeof (nvlist_t *));
895
896 return (raidz);
897 }
898
899 static nvlist_t *
900 make_vdev_mirror(char *path, char *aux, char *pool, size_t size,
901 uint64_t ashift, int r, int m)
902 {
903 nvlist_t *mirror, **child;
904 int c;
905
906 if (m < 1)
907 return (make_vdev_raidz(path, aux, pool, size, ashift, r));
908
909 child = umem_alloc(m * sizeof (nvlist_t *), UMEM_NOFAIL);
910
911 for (c = 0; c < m; c++)
912 child[c] = make_vdev_raidz(path, aux, pool, size, ashift, r);
913
914 VERIFY(nvlist_alloc(&mirror, NV_UNIQUE_NAME, 0) == 0);
915 VERIFY(nvlist_add_string(mirror, ZPOOL_CONFIG_TYPE,
916 VDEV_TYPE_MIRROR) == 0);
917 VERIFY(nvlist_add_nvlist_array(mirror, ZPOOL_CONFIG_CHILDREN,
918 child, m) == 0);
919
920 for (c = 0; c < m; c++)
921 nvlist_free(child[c]);
922
923 umem_free(child, m * sizeof (nvlist_t *));
924
925 return (mirror);
926 }
927
928 static nvlist_t *
929 make_vdev_root(char *path, char *aux, char *pool, size_t size, uint64_t ashift,
930 int log, int r, int m, int t)
931 {
932 nvlist_t *root, **child;
933 int c;
934
935 ASSERT(t > 0);
936
937 child = umem_alloc(t * sizeof (nvlist_t *), UMEM_NOFAIL);
938
939 for (c = 0; c < t; c++) {
940 child[c] = make_vdev_mirror(path, aux, pool, size, ashift,
941 r, m);
942 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_IS_LOG,
943 log) == 0);
944 }
945
946 VERIFY(nvlist_alloc(&root, NV_UNIQUE_NAME, 0) == 0);
947 VERIFY(nvlist_add_string(root, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) == 0);
948 VERIFY(nvlist_add_nvlist_array(root, aux ? aux : ZPOOL_CONFIG_CHILDREN,
949 child, t) == 0);
950
951 for (c = 0; c < t; c++)
952 nvlist_free(child[c]);
953
954 umem_free(child, t * sizeof (nvlist_t *));
955
956 return (root);
957 }
958
959 /*
960 * Find a random spa version. Returns back a random spa version in the
961 * range [initial_version, SPA_VERSION_FEATURES].
962 */
963 static uint64_t
964 ztest_random_spa_version(uint64_t initial_version)
965 {
966 uint64_t version = initial_version;
967
968 if (version <= SPA_VERSION_BEFORE_FEATURES) {
969 version = version +
970 ztest_random(SPA_VERSION_BEFORE_FEATURES - version + 1);
971 }
972
973 if (version > SPA_VERSION_BEFORE_FEATURES)
974 version = SPA_VERSION_FEATURES;
975
976 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
977 return (version);
978 }
979
980 static int
981 ztest_random_blocksize(void)
982 {
983 uint64_t block_shift;
984 /*
985 * Choose a block size >= the ashift.
986 * If the SPA supports new MAXBLOCKSIZE, test up to 1MB blocks.
987 */
988 int maxbs = SPA_OLD_MAXBLOCKSHIFT;
989 if (spa_maxblocksize(ztest_spa) == SPA_MAXBLOCKSIZE)
990 maxbs = 20;
991 block_shift = ztest_random(maxbs - ztest_spa->spa_max_ashift + 1);
992 return (1 << (SPA_MINBLOCKSHIFT + block_shift));
993 }
994
995 static int
996 ztest_random_ibshift(void)
997 {
998 return (DN_MIN_INDBLKSHIFT +
999 ztest_random(DN_MAX_INDBLKSHIFT - DN_MIN_INDBLKSHIFT + 1));
1000 }
1001
1002 static uint64_t
1003 ztest_random_vdev_top(spa_t *spa, boolean_t log_ok)
1004 {
1005 uint64_t top;
1006 vdev_t *rvd = spa->spa_root_vdev;
1007 vdev_t *tvd;
1008
1009 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1010
1011 do {
1012 top = ztest_random(rvd->vdev_children);
1013 tvd = rvd->vdev_child[top];
1014 } while (tvd->vdev_ishole || (tvd->vdev_islog && !log_ok) ||
1015 tvd->vdev_mg == NULL || tvd->vdev_mg->mg_class == NULL);
1016
1017 return (top);
1018 }
1019
1020 static uint64_t
1021 ztest_random_dsl_prop(zfs_prop_t prop)
1022 {
1023 uint64_t value;
1024
1025 do {
1026 value = zfs_prop_random_value(prop, ztest_random(-1ULL));
1027 } while (prop == ZFS_PROP_CHECKSUM && value == ZIO_CHECKSUM_OFF);
1028
1029 return (value);
1030 }
1031
1032 static int
1033 ztest_dsl_prop_set_uint64(char *osname, zfs_prop_t prop, uint64_t value,
1034 boolean_t inherit)
1035 {
1036 const char *propname = zfs_prop_to_name(prop);
1037 const char *valname;
1038 char setpoint[MAXPATHLEN];
1039 uint64_t curval;
1040 int error;
1041
1042 error = dsl_prop_set_int(osname, propname,
1043 (inherit ? ZPROP_SRC_NONE : ZPROP_SRC_LOCAL), value);
1044
1045 if (error == ENOSPC) {
1046 ztest_record_enospc(FTAG);
1047 return (error);
1048 }
1049 ASSERT0(error);
1050
1051 VERIFY0(dsl_prop_get_integer(osname, propname, &curval, setpoint));
1052
1053 if (ztest_opts.zo_verbose >= 6) {
1054 VERIFY(zfs_prop_index_to_string(prop, curval, &valname) == 0);
1055 (void) printf("%s %s = %s at '%s'\n",
1056 osname, propname, valname, setpoint);
1057 }
1058
1059 return (error);
1060 }
1061
1062 static int
1063 ztest_spa_prop_set_uint64(zpool_prop_t prop, uint64_t value)
1064 {
1065 spa_t *spa = ztest_spa;
1066 nvlist_t *props = NULL;
1067 int error;
1068
1069 VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
1070 VERIFY(nvlist_add_uint64(props, zpool_prop_to_name(prop), value) == 0);
1071
1072 error = spa_prop_set(spa, props);
1073
1074 nvlist_free(props);
1075
1076 if (error == ENOSPC) {
1077 ztest_record_enospc(FTAG);
1078 return (error);
1079 }
1080 ASSERT0(error);
1081
1082 return (error);
1083 }
1084
1085 static void
1086 ztest_rll_init(rll_t *rll)
1087 {
1088 rll->rll_writer = NULL;
1089 rll->rll_readers = 0;
1090 VERIFY(_mutex_init(&rll->rll_lock, USYNC_THREAD, NULL) == 0);
1091 VERIFY(cond_init(&rll->rll_cv, USYNC_THREAD, NULL) == 0);
1092 }
1093
1094 static void
1095 ztest_rll_destroy(rll_t *rll)
1096 {
1097 ASSERT(rll->rll_writer == NULL);
1098 ASSERT(rll->rll_readers == 0);
1099 VERIFY(_mutex_destroy(&rll->rll_lock) == 0);
1100 VERIFY(cond_destroy(&rll->rll_cv) == 0);
1101 }
1102
1103 static void
1104 ztest_rll_lock(rll_t *rll, rl_type_t type)
1105 {
1106 VERIFY(mutex_lock(&rll->rll_lock) == 0);
1107
1108 if (type == RL_READER) {
1109 while (rll->rll_writer != NULL)
1110 (void) cond_wait(&rll->rll_cv, &rll->rll_lock);
1111 rll->rll_readers++;
1112 } else {
1113 while (rll->rll_writer != NULL || rll->rll_readers)
1114 (void) cond_wait(&rll->rll_cv, &rll->rll_lock);
1115 rll->rll_writer = curthread;
1116 }
1117
1118 VERIFY(mutex_unlock(&rll->rll_lock) == 0);
1119 }
1120
1121 static void
1122 ztest_rll_unlock(rll_t *rll)
1123 {
1124 VERIFY(mutex_lock(&rll->rll_lock) == 0);
1125
1126 if (rll->rll_writer) {
1127 ASSERT(rll->rll_readers == 0);
1128 rll->rll_writer = NULL;
1129 } else {
1130 ASSERT(rll->rll_readers != 0);
1131 ASSERT(rll->rll_writer == NULL);
1132 rll->rll_readers--;
1133 }
1134
1135 if (rll->rll_writer == NULL && rll->rll_readers == 0)
1136 VERIFY(cond_broadcast(&rll->rll_cv) == 0);
1137
1138 VERIFY(mutex_unlock(&rll->rll_lock) == 0);
1139 }
1140
1141 static void
1142 ztest_object_lock(ztest_ds_t *zd, uint64_t object, rl_type_t type)
1143 {
1144 rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1145
1146 ztest_rll_lock(rll, type);
1147 }
1148
1149 static void
1150 ztest_object_unlock(ztest_ds_t *zd, uint64_t object)
1151 {
1152 rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1153
1154 ztest_rll_unlock(rll);
1155 }
1156
1157 static rl_t *
1158 ztest_range_lock(ztest_ds_t *zd, uint64_t object, uint64_t offset,
1159 uint64_t size, rl_type_t type)
1160 {
1161 uint64_t hash = object ^ (offset % (ZTEST_RANGE_LOCKS + 1));
1162 rll_t *rll = &zd->zd_range_lock[hash & (ZTEST_RANGE_LOCKS - 1)];
1163 rl_t *rl;
1164
1165 rl = umem_alloc(sizeof (*rl), UMEM_NOFAIL);
1166 rl->rl_object = object;
1167 rl->rl_offset = offset;
1168 rl->rl_size = size;
1169 rl->rl_lock = rll;
1170
1171 ztest_rll_lock(rll, type);
1172
1173 return (rl);
1174 }
1175
1176 static void
1177 ztest_range_unlock(rl_t *rl)
1178 {
1179 rll_t *rll = rl->rl_lock;
1180
1181 ztest_rll_unlock(rll);
1182
1183 umem_free(rl, sizeof (*rl));
1184 }
1185
1186 static void
1187 ztest_zd_init(ztest_ds_t *zd, ztest_shared_ds_t *szd, objset_t *os)
1188 {
1189 zd->zd_os = os;
1190 zd->zd_zilog = dmu_objset_zil(os);
1191 zd->zd_shared = szd;
1192 dmu_objset_name(os, zd->zd_name);
1193
1194 if (zd->zd_shared != NULL)
1195 zd->zd_shared->zd_seq = 0;
1196
1197 VERIFY(rwlock_init(&zd->zd_zilog_lock, USYNC_THREAD, NULL) == 0);
1198 VERIFY(_mutex_init(&zd->zd_dirobj_lock, USYNC_THREAD, NULL) == 0);
1199
1200 for (int l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1201 ztest_rll_init(&zd->zd_object_lock[l]);
1202
1203 for (int l = 0; l < ZTEST_RANGE_LOCKS; l++)
1204 ztest_rll_init(&zd->zd_range_lock[l]);
1205 }
1206
1207 static void
1208 ztest_zd_fini(ztest_ds_t *zd)
1209 {
1210 VERIFY(_mutex_destroy(&zd->zd_dirobj_lock) == 0);
1211
1212 for (int l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1213 ztest_rll_destroy(&zd->zd_object_lock[l]);
1214
1215 for (int l = 0; l < ZTEST_RANGE_LOCKS; l++)
1216 ztest_rll_destroy(&zd->zd_range_lock[l]);
1217 }
1218
1219 #define TXG_MIGHTWAIT (ztest_random(10) == 0 ? TXG_NOWAIT : TXG_WAIT)
1220
1221 static uint64_t
1222 ztest_tx_assign(dmu_tx_t *tx, uint64_t txg_how, const char *tag)
1223 {
1224 uint64_t txg;
1225 int error;
1226
1227 /*
1228 * Attempt to assign tx to some transaction group.
1229 */
1230 error = dmu_tx_assign(tx, txg_how);
1231 if (error) {
1232 if (error == ERESTART) {
1233 ASSERT(txg_how == TXG_NOWAIT);
1234 dmu_tx_wait(tx);
1235 } else {
1236 ASSERT3U(error, ==, ENOSPC);
1237 ztest_record_enospc(tag);
1238 }
1239 dmu_tx_abort(tx);
1240 return (0);
1241 }
1242 txg = dmu_tx_get_txg(tx);
1243 ASSERT(txg != 0);
1244 return (txg);
1245 }
1246
1247 static void
1248 ztest_pattern_set(void *buf, uint64_t size, uint64_t value)
1249 {
1250 uint64_t *ip = buf;
1251 uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1252
1253 while (ip < ip_end)
1254 *ip++ = value;
1255 }
1256
1257 static boolean_t
1258 ztest_pattern_match(void *buf, uint64_t size, uint64_t value)
1259 {
1260 uint64_t *ip = buf;
1261 uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1262 uint64_t diff = 0;
1263
1264 while (ip < ip_end)
1265 diff |= (value - *ip++);
1266
1267 return (diff == 0);
1268 }
1269
1270 static void
1271 ztest_bt_generate(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1272 uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg)
1273 {
1274 bt->bt_magic = BT_MAGIC;
1275 bt->bt_objset = dmu_objset_id(os);
1276 bt->bt_object = object;
1277 bt->bt_offset = offset;
1278 bt->bt_gen = gen;
1279 bt->bt_txg = txg;
1280 bt->bt_crtxg = crtxg;
1281 }
1282
1283 static void
1284 ztest_bt_verify(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1285 uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg)
1286 {
1287 ASSERT3U(bt->bt_magic, ==, BT_MAGIC);
1288 ASSERT3U(bt->bt_objset, ==, dmu_objset_id(os));
1289 ASSERT3U(bt->bt_object, ==, object);
1290 ASSERT3U(bt->bt_offset, ==, offset);
1291 ASSERT3U(bt->bt_gen, <=, gen);
1292 ASSERT3U(bt->bt_txg, <=, txg);
1293 ASSERT3U(bt->bt_crtxg, ==, crtxg);
1294 }
1295
1296 static ztest_block_tag_t *
1297 ztest_bt_bonus(dmu_buf_t *db)
1298 {
1299 dmu_object_info_t doi;
1300 ztest_block_tag_t *bt;
1301
1302 dmu_object_info_from_db(db, &doi);
1303 ASSERT3U(doi.doi_bonus_size, <=, db->db_size);
1304 ASSERT3U(doi.doi_bonus_size, >=, sizeof (*bt));
1305 bt = (void *)((char *)db->db_data + doi.doi_bonus_size - sizeof (*bt));
1306
1307 return (bt);
1308 }
1309
1310 /*
1311 * ZIL logging ops
1312 */
1313
1314 #define lrz_type lr_mode
1315 #define lrz_blocksize lr_uid
1316 #define lrz_ibshift lr_gid
1317 #define lrz_bonustype lr_rdev
1318 #define lrz_bonuslen lr_crtime[1]
1319
1320 static void
1321 ztest_log_create(ztest_ds_t *zd, dmu_tx_t *tx, lr_create_t *lr)
1322 {
1323 char *name = (void *)(lr + 1); /* name follows lr */
1324 size_t namesize = strlen(name) + 1;
1325 itx_t *itx;
1326
1327 if (zil_replaying(zd->zd_zilog, tx))
1328 return;
1329
1330 itx = zil_itx_create(TX_CREATE, sizeof (*lr) + namesize);
1331 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1332 sizeof (*lr) + namesize - sizeof (lr_t));
1333
1334 zil_itx_assign(zd->zd_zilog, itx, tx);
1335 }
1336
1337 static void
1338 ztest_log_remove(ztest_ds_t *zd, dmu_tx_t *tx, lr_remove_t *lr, uint64_t object)
1339 {
1340 char *name = (void *)(lr + 1); /* name follows lr */
1341 size_t namesize = strlen(name) + 1;
1342 itx_t *itx;
1343
1344 if (zil_replaying(zd->zd_zilog, tx))
1345 return;
1346
1347 itx = zil_itx_create(TX_REMOVE, sizeof (*lr) + namesize);
1348 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1349 sizeof (*lr) + namesize - sizeof (lr_t));
1350
1351 itx->itx_oid = object;
1352 zil_itx_assign(zd->zd_zilog, itx, tx);
1353 }
1354
1355 static void
1356 ztest_log_write(ztest_ds_t *zd, dmu_tx_t *tx, lr_write_t *lr)
1357 {
1358 itx_t *itx;
1359 itx_wr_state_t write_state = ztest_random(WR_NUM_STATES);
1360
1361 if (zil_replaying(zd->zd_zilog, tx))
1362 return;
1363
1364 if (lr->lr_length > ZIL_MAX_LOG_DATA)
1365 write_state = WR_INDIRECT;
1366
1367 itx = zil_itx_create(TX_WRITE,
1368 sizeof (*lr) + (write_state == WR_COPIED ? lr->lr_length : 0));
1369
1370 if (write_state == WR_COPIED &&
1371 dmu_read(zd->zd_os, lr->lr_foid, lr->lr_offset, lr->lr_length,
1372 ((lr_write_t *)&itx->itx_lr) + 1, DMU_READ_NO_PREFETCH) != 0) {
1373 zil_itx_destroy(itx);
1374 itx = zil_itx_create(TX_WRITE, sizeof (*lr));
1375 write_state = WR_NEED_COPY;
1376 }
1377 itx->itx_private = zd;
1378 itx->itx_wr_state = write_state;
1379 itx->itx_sync = (ztest_random(8) == 0);
1380
1381 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1382 sizeof (*lr) - sizeof (lr_t));
1383
1384 zil_itx_assign(zd->zd_zilog, itx, tx);
1385 }
1386
1387 static void
1388 ztest_log_truncate(ztest_ds_t *zd, dmu_tx_t *tx, lr_truncate_t *lr)
1389 {
1390 itx_t *itx;
1391
1392 if (zil_replaying(zd->zd_zilog, tx))
1393 return;
1394
1395 itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr));
1396 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1397 sizeof (*lr) - sizeof (lr_t));
1398
1399 itx->itx_sync = B_FALSE;
1400 zil_itx_assign(zd->zd_zilog, itx, tx);
1401 }
1402
1403 static void
1404 ztest_log_setattr(ztest_ds_t *zd, dmu_tx_t *tx, lr_setattr_t *lr)
1405 {
1406 itx_t *itx;
1407
1408 if (zil_replaying(zd->zd_zilog, tx))
1409 return;
1410
1411 itx = zil_itx_create(TX_SETATTR, sizeof (*lr));
1412 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1413 sizeof (*lr) - sizeof (lr_t));
1414
1415 itx->itx_sync = B_FALSE;
1416 zil_itx_assign(zd->zd_zilog, itx, tx);
1417 }
1418
1419 /*
1420 * ZIL replay ops
1421 */
1422 static int
1423 ztest_replay_create(ztest_ds_t *zd, lr_create_t *lr, boolean_t byteswap)
1424 {
1425 char *name = (void *)(lr + 1); /* name follows lr */
1426 objset_t *os = zd->zd_os;
1427 ztest_block_tag_t *bbt;
1428 dmu_buf_t *db;
1429 dmu_tx_t *tx;
1430 uint64_t txg;
1431 int error = 0;
1432
1433 if (byteswap)
1434 byteswap_uint64_array(lr, sizeof (*lr));
1435
1436 ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1437 ASSERT(name[0] != '\0');
1438
1439 tx = dmu_tx_create(os);
1440
1441 dmu_tx_hold_zap(tx, lr->lr_doid, B_TRUE, name);
1442
1443 if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1444 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, B_TRUE, NULL);
1445 } else {
1446 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1447 }
1448
1449 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1450 if (txg == 0)
1451 return (ENOSPC);
1452
1453 ASSERT(dmu_objset_zil(os)->zl_replay == !!lr->lr_foid);
1454
1455 if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1456 if (lr->lr_foid == 0) {
1457 lr->lr_foid = zap_create(os,
1458 lr->lrz_type, lr->lrz_bonustype,
1459 lr->lrz_bonuslen, tx);
1460 } else {
1461 error = zap_create_claim(os, lr->lr_foid,
1462 lr->lrz_type, lr->lrz_bonustype,
1463 lr->lrz_bonuslen, tx);
1464 }
1465 } else {
1466 if (lr->lr_foid == 0) {
1467 lr->lr_foid = dmu_object_alloc(os,
1468 lr->lrz_type, 0, lr->lrz_bonustype,
1469 lr->lrz_bonuslen, tx);
1470 } else {
1471 error = dmu_object_claim(os, lr->lr_foid,
1472 lr->lrz_type, 0, lr->lrz_bonustype,
1473 lr->lrz_bonuslen, tx);
1474 }
1475 }
1476
1477 if (error) {
1478 ASSERT3U(error, ==, EEXIST);
1479 ASSERT(zd->zd_zilog->zl_replay);
1480 dmu_tx_commit(tx);
1481 return (error);
1482 }
1483
1484 ASSERT(lr->lr_foid != 0);
1485
1486 if (lr->lrz_type != DMU_OT_ZAP_OTHER)
1487 VERIFY3U(0, ==, dmu_object_set_blocksize(os, lr->lr_foid,
1488 lr->lrz_blocksize, lr->lrz_ibshift, tx));
1489
1490 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1491 bbt = ztest_bt_bonus(db);
1492 dmu_buf_will_dirty(db, tx);
1493 ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_gen, txg, txg);
1494 dmu_buf_rele(db, FTAG);
1495
1496 VERIFY3U(0, ==, zap_add(os, lr->lr_doid, name, sizeof (uint64_t), 1,
1497 &lr->lr_foid, tx));
1498
1499 (void) ztest_log_create(zd, tx, lr);
1500
1501 dmu_tx_commit(tx);
1502
1503 return (0);
1504 }
1505
1506 static int
1507 ztest_replay_remove(ztest_ds_t *zd, lr_remove_t *lr, boolean_t byteswap)
1508 {
1509 char *name = (void *)(lr + 1); /* name follows lr */
1510 objset_t *os = zd->zd_os;
1511 dmu_object_info_t doi;
1512 dmu_tx_t *tx;
1513 uint64_t object, txg;
1514
1515 if (byteswap)
1516 byteswap_uint64_array(lr, sizeof (*lr));
1517
1518 ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1519 ASSERT(name[0] != '\0');
1520
1521 VERIFY3U(0, ==,
1522 zap_lookup(os, lr->lr_doid, name, sizeof (object), 1, &object));
1523 ASSERT(object != 0);
1524
1525 ztest_object_lock(zd, object, RL_WRITER);
1526
1527 VERIFY3U(0, ==, dmu_object_info(os, object, &doi));
1528
1529 tx = dmu_tx_create(os);
1530
1531 dmu_tx_hold_zap(tx, lr->lr_doid, B_FALSE, name);
1532 dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END);
1533
1534 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1535 if (txg == 0) {
1536 ztest_object_unlock(zd, object);
1537 return (ENOSPC);
1538 }
1539
1540 if (doi.doi_type == DMU_OT_ZAP_OTHER) {
1541 VERIFY3U(0, ==, zap_destroy(os, object, tx));
1542 } else {
1543 VERIFY3U(0, ==, dmu_object_free(os, object, tx));
1544 }
1545
1546 VERIFY3U(0, ==, zap_remove(os, lr->lr_doid, name, tx));
1547
1548 (void) ztest_log_remove(zd, tx, lr, object);
1549
1550 dmu_tx_commit(tx);
1551
1552 ztest_object_unlock(zd, object);
1553
1554 return (0);
1555 }
1556
1557 static int
1558 ztest_replay_write(ztest_ds_t *zd, lr_write_t *lr, boolean_t byteswap)
1559 {
1560 objset_t *os = zd->zd_os;
1561 void *data = lr + 1; /* data follows lr */
1562 uint64_t offset, length;
1563 ztest_block_tag_t *bt = data;
1564 ztest_block_tag_t *bbt;
1565 uint64_t gen, txg, lrtxg, crtxg;
1566 dmu_object_info_t doi;
1567 dmu_tx_t *tx;
1568 dmu_buf_t *db;
1569 arc_buf_t *abuf = NULL;
1570 rl_t *rl;
1571
1572 if (byteswap)
1573 byteswap_uint64_array(lr, sizeof (*lr));
1574
1575 offset = lr->lr_offset;
1576 length = lr->lr_length;
1577
1578 /* If it's a dmu_sync() block, write the whole block */
1579 if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
1580 uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
1581 if (length < blocksize) {
1582 offset -= offset % blocksize;
1583 length = blocksize;
1584 }
1585 }
1586
1587 if (bt->bt_magic == BSWAP_64(BT_MAGIC))
1588 byteswap_uint64_array(bt, sizeof (*bt));
1589
1590 if (bt->bt_magic != BT_MAGIC)
1591 bt = NULL;
1592
1593 ztest_object_lock(zd, lr->lr_foid, RL_READER);
1594 rl = ztest_range_lock(zd, lr->lr_foid, offset, length, RL_WRITER);
1595
1596 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1597
1598 dmu_object_info_from_db(db, &doi);
1599
1600 bbt = ztest_bt_bonus(db);
1601 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1602 gen = bbt->bt_gen;
1603 crtxg = bbt->bt_crtxg;
1604 lrtxg = lr->lr_common.lrc_txg;
1605
1606 tx = dmu_tx_create(os);
1607
1608 dmu_tx_hold_write(tx, lr->lr_foid, offset, length);
1609
1610 if (ztest_random(8) == 0 && length == doi.doi_data_block_size &&
1611 P2PHASE(offset, length) == 0)
1612 abuf = dmu_request_arcbuf(db, length);
1613
1614 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1615 if (txg == 0) {
1616 if (abuf != NULL)
1617 dmu_return_arcbuf(abuf);
1618 dmu_buf_rele(db, FTAG);
1619 ztest_range_unlock(rl);
1620 ztest_object_unlock(zd, lr->lr_foid);
1621 return (ENOSPC);
1622 }
1623
1624 if (bt != NULL) {
1625 /*
1626 * Usually, verify the old data before writing new data --
1627 * but not always, because we also want to verify correct
1628 * behavior when the data was not recently read into cache.
1629 */
1630 ASSERT(offset % doi.doi_data_block_size == 0);
1631 if (ztest_random(4) != 0) {
1632 int prefetch = ztest_random(2) ?
1633 DMU_READ_PREFETCH : DMU_READ_NO_PREFETCH;
1634 ztest_block_tag_t rbt;
1635
1636 VERIFY(dmu_read(os, lr->lr_foid, offset,
1637 sizeof (rbt), &rbt, prefetch) == 0);
1638 if (rbt.bt_magic == BT_MAGIC) {
1639 ztest_bt_verify(&rbt, os, lr->lr_foid,
1640 offset, gen, txg, crtxg);
1641 }
1642 }
1643
1644 /*
1645 * Writes can appear to be newer than the bonus buffer because
1646 * the ztest_get_data() callback does a dmu_read() of the
1647 * open-context data, which may be different than the data
1648 * as it was when the write was generated.
1649 */
1650 if (zd->zd_zilog->zl_replay) {
1651 ztest_bt_verify(bt, os, lr->lr_foid, offset,
1652 MAX(gen, bt->bt_gen), MAX(txg, lrtxg),
1653 bt->bt_crtxg);
1654 }
1655
1656 /*
1657 * Set the bt's gen/txg to the bonus buffer's gen/txg
1658 * so that all of the usual ASSERTs will work.
1659 */
1660 ztest_bt_generate(bt, os, lr->lr_foid, offset, gen, txg, crtxg);
1661 }
1662
1663 if (abuf == NULL) {
1664 dmu_write(os, lr->lr_foid, offset, length, data, tx);
1665 } else {
1666 bcopy(data, abuf->b_data, length);
1667 dmu_assign_arcbuf(db, offset, abuf, tx);
1668 }
1669
1670 (void) ztest_log_write(zd, tx, lr);
1671
1672 dmu_buf_rele(db, FTAG);
1673
1674 dmu_tx_commit(tx);
1675
1676 ztest_range_unlock(rl);
1677 ztest_object_unlock(zd, lr->lr_foid);
1678
1679 return (0);
1680 }
1681
1682 static int
1683 ztest_replay_truncate(ztest_ds_t *zd, lr_truncate_t *lr, boolean_t byteswap)
1684 {
1685 objset_t *os = zd->zd_os;
1686 dmu_tx_t *tx;
1687 uint64_t txg;
1688 rl_t *rl;
1689
1690 if (byteswap)
1691 byteswap_uint64_array(lr, sizeof (*lr));
1692
1693 ztest_object_lock(zd, lr->lr_foid, RL_READER);
1694 rl = ztest_range_lock(zd, lr->lr_foid, lr->lr_offset, lr->lr_length,
1695 RL_WRITER);
1696
1697 tx = dmu_tx_create(os);
1698
1699 dmu_tx_hold_free(tx, lr->lr_foid, lr->lr_offset, lr->lr_length);
1700
1701 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1702 if (txg == 0) {
1703 ztest_range_unlock(rl);
1704 ztest_object_unlock(zd, lr->lr_foid);
1705 return (ENOSPC);
1706 }
1707
1708 VERIFY(dmu_free_range(os, lr->lr_foid, lr->lr_offset,
1709 lr->lr_length, tx) == 0);
1710
1711 (void) ztest_log_truncate(zd, tx, lr);
1712
1713 dmu_tx_commit(tx);
1714
1715 ztest_range_unlock(rl);
1716 ztest_object_unlock(zd, lr->lr_foid);
1717
1718 return (0);
1719 }
1720
1721 static int
1722 ztest_replay_setattr(ztest_ds_t *zd, lr_setattr_t *lr, boolean_t byteswap)
1723 {
1724 objset_t *os = zd->zd_os;
1725 dmu_tx_t *tx;
1726 dmu_buf_t *db;
1727 ztest_block_tag_t *bbt;
1728 uint64_t txg, lrtxg, crtxg;
1729
1730 if (byteswap)
1731 byteswap_uint64_array(lr, sizeof (*lr));
1732
1733 ztest_object_lock(zd, lr->lr_foid, RL_WRITER);
1734
1735 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1736
1737 tx = dmu_tx_create(os);
1738 dmu_tx_hold_bonus(tx, lr->lr_foid);
1739
1740 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1741 if (txg == 0) {
1742 dmu_buf_rele(db, FTAG);
1743 ztest_object_unlock(zd, lr->lr_foid);
1744 return (ENOSPC);
1745 }
1746
1747 bbt = ztest_bt_bonus(db);
1748 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1749 crtxg = bbt->bt_crtxg;
1750 lrtxg = lr->lr_common.lrc_txg;
1751
1752 if (zd->zd_zilog->zl_replay) {
1753 ASSERT(lr->lr_size != 0);
1754 ASSERT(lr->lr_mode != 0);
1755 ASSERT(lrtxg != 0);
1756 } else {
1757 /*
1758 * Randomly change the size and increment the generation.
1759 */
1760 lr->lr_size = (ztest_random(db->db_size / sizeof (*bbt)) + 1) *
1761 sizeof (*bbt);
1762 lr->lr_mode = bbt->bt_gen + 1;
1763 ASSERT(lrtxg == 0);
1764 }
1765
1766 /*
1767 * Verify that the current bonus buffer is not newer than our txg.
1768 */
1769 ztest_bt_verify(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode,
1770 MAX(txg, lrtxg), crtxg);
1771
1772 dmu_buf_will_dirty(db, tx);
1773
1774 ASSERT3U(lr->lr_size, >=, sizeof (*bbt));
1775 ASSERT3U(lr->lr_size, <=, db->db_size);
1776 VERIFY0(dmu_set_bonus(db, lr->lr_size, tx));
1777 bbt = ztest_bt_bonus(db);
1778
1779 ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode, txg, crtxg);
1780
1781 dmu_buf_rele(db, FTAG);
1782
1783 (void) ztest_log_setattr(zd, tx, lr);
1784
1785 dmu_tx_commit(tx);
1786
1787 ztest_object_unlock(zd, lr->lr_foid);
1788
1789 return (0);
1790 }
1791
1792 zil_replay_func_t *ztest_replay_vector[TX_MAX_TYPE] = {
1793 NULL, /* 0 no such transaction type */
1794 ztest_replay_create, /* TX_CREATE */
1795 NULL, /* TX_MKDIR */
1796 NULL, /* TX_MKXATTR */
1797 NULL, /* TX_SYMLINK */
1798 ztest_replay_remove, /* TX_REMOVE */
1799 NULL, /* TX_RMDIR */
1800 NULL, /* TX_LINK */
1801 NULL, /* TX_RENAME */
1802 ztest_replay_write, /* TX_WRITE */
1803 ztest_replay_truncate, /* TX_TRUNCATE */
1804 ztest_replay_setattr, /* TX_SETATTR */
1805 NULL, /* TX_ACL */
1806 NULL, /* TX_CREATE_ACL */
1807 NULL, /* TX_CREATE_ATTR */
1808 NULL, /* TX_CREATE_ACL_ATTR */
1809 NULL, /* TX_MKDIR_ACL */
1810 NULL, /* TX_MKDIR_ATTR */
1811 NULL, /* TX_MKDIR_ACL_ATTR */
1812 NULL, /* TX_WRITE2 */
1813 };
1814
1815 /*
1816 * ZIL get_data callbacks
1817 */
1818
1819 static void
1820 ztest_get_done(zgd_t *zgd, int error)
1821 {
1822 ztest_ds_t *zd = zgd->zgd_private;
1823 uint64_t object = zgd->zgd_rl->rl_object;
1824
1825 if (zgd->zgd_db)
1826 dmu_buf_rele(zgd->zgd_db, zgd);
1827
1828 ztest_range_unlock(zgd->zgd_rl);
1829 ztest_object_unlock(zd, object);
1830
1831 if (error == 0 && zgd->zgd_bp)
1832 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
1833
1834 umem_free(zgd, sizeof (*zgd));
1835 }
1836
1837 static int
1838 ztest_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
1839 {
1840 ztest_ds_t *zd = arg;
1841 objset_t *os = zd->zd_os;
1842 uint64_t object = lr->lr_foid;
1843 uint64_t offset = lr->lr_offset;
1844 uint64_t size = lr->lr_length;
1845 blkptr_t *bp = &lr->lr_blkptr;
1846 uint64_t txg = lr->lr_common.lrc_txg;
1847 uint64_t crtxg;
1848 dmu_object_info_t doi;
1849 dmu_buf_t *db;
1850 zgd_t *zgd;
1851 int error;
1852
1853 ztest_object_lock(zd, object, RL_READER);
1854 error = dmu_bonus_hold(os, object, FTAG, &db);
1855 if (error) {
1856 ztest_object_unlock(zd, object);
1857 return (error);
1858 }
1859
1860 crtxg = ztest_bt_bonus(db)->bt_crtxg;
1861
1862 if (crtxg == 0 || crtxg > txg) {
1863 dmu_buf_rele(db, FTAG);
1864 ztest_object_unlock(zd, object);
1865 return (ENOENT);
1866 }
1867
1868 dmu_object_info_from_db(db, &doi);
1869 dmu_buf_rele(db, FTAG);
1870 db = NULL;
1871
1872 zgd = umem_zalloc(sizeof (*zgd), UMEM_NOFAIL);
1873 zgd->zgd_zilog = zd->zd_zilog;
1874 zgd->zgd_private = zd;
1875
1876 if (buf != NULL) { /* immediate write */
1877 zgd->zgd_rl = ztest_range_lock(zd, object, offset, size,
1878 RL_READER);
1879
1880 error = dmu_read(os, object, offset, size, buf,
1881 DMU_READ_NO_PREFETCH);
1882 ASSERT(error == 0);
1883 } else {
1884 size = doi.doi_data_block_size;
1885 if (ISP2(size)) {
1886 offset = P2ALIGN(offset, size);
1887 } else {
1888 ASSERT(offset < size);
1889 offset = 0;
1890 }
1891
1892 zgd->zgd_rl = ztest_range_lock(zd, object, offset, size,
1893 RL_READER);
1894
1895 error = dmu_buf_hold(os, object, offset, zgd, &db,
1896 DMU_READ_NO_PREFETCH);
1897
1898 if (error == 0) {
1899 blkptr_t *obp = dmu_buf_get_blkptr(db);
1900 if (obp) {
1901 ASSERT(BP_IS_HOLE(bp));
1902 *bp = *obp;
1903 }
1904
1905 zgd->zgd_db = db;
1906 zgd->zgd_bp = bp;
1907
1908 ASSERT(db->db_offset == offset);
1909 ASSERT(db->db_size == size);
1910
1911 error = dmu_sync(zio, lr->lr_common.lrc_txg,
1912 ztest_get_done, zgd);
1913
1914 if (error == 0)
1915 return (0);
1916 }
1917 }
1918
1919 ztest_get_done(zgd, error);
1920
1921 return (error);
1922 }
1923
1924 static void *
1925 ztest_lr_alloc(size_t lrsize, char *name)
1926 {
1927 char *lr;
1928 size_t namesize = name ? strlen(name) + 1 : 0;
1929
1930 lr = umem_zalloc(lrsize + namesize, UMEM_NOFAIL);
1931
1932 if (name)
1933 bcopy(name, lr + lrsize, namesize);
1934
1935 return (lr);
1936 }
1937
1938 void
1939 ztest_lr_free(void *lr, size_t lrsize, char *name)
1940 {
1941 size_t namesize = name ? strlen(name) + 1 : 0;
1942
1943 umem_free(lr, lrsize + namesize);
1944 }
1945
1946 /*
1947 * Lookup a bunch of objects. Returns the number of objects not found.
1948 */
1949 static int
1950 ztest_lookup(ztest_ds_t *zd, ztest_od_t *od, int count)
1951 {
1952 int missing = 0;
1953 int error;
1954
1955 ASSERT(_mutex_held(&zd->zd_dirobj_lock));
1956
1957 for (int i = 0; i < count; i++, od++) {
1958 od->od_object = 0;
1959 error = zap_lookup(zd->zd_os, od->od_dir, od->od_name,
1960 sizeof (uint64_t), 1, &od->od_object);
1961 if (error) {
1962 ASSERT(error == ENOENT);
1963 ASSERT(od->od_object == 0);
1964 missing++;
1965 } else {
1966 dmu_buf_t *db;
1967 ztest_block_tag_t *bbt;
1968 dmu_object_info_t doi;
1969
1970 ASSERT(od->od_object != 0);
1971 ASSERT(missing == 0); /* there should be no gaps */
1972
1973 ztest_object_lock(zd, od->od_object, RL_READER);
1974 VERIFY3U(0, ==, dmu_bonus_hold(zd->zd_os,
1975 od->od_object, FTAG, &db));
1976 dmu_object_info_from_db(db, &doi);
1977 bbt = ztest_bt_bonus(db);
1978 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1979 od->od_type = doi.doi_type;
1980 od->od_blocksize = doi.doi_data_block_size;
1981 od->od_gen = bbt->bt_gen;
1982 dmu_buf_rele(db, FTAG);
1983 ztest_object_unlock(zd, od->od_object);
1984 }
1985 }
1986
1987 return (missing);
1988 }
1989
1990 static int
1991 ztest_create(ztest_ds_t *zd, ztest_od_t *od, int count)
1992 {
1993 int missing = 0;
1994
1995 ASSERT(_mutex_held(&zd->zd_dirobj_lock));
1996
1997 for (int i = 0; i < count; i++, od++) {
1998 if (missing) {
1999 od->od_object = 0;
2000 missing++;
2001 continue;
2002 }
2003
2004 lr_create_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
2005
2006 lr->lr_doid = od->od_dir;
2007 lr->lr_foid = 0; /* 0 to allocate, > 0 to claim */
2008 lr->lrz_type = od->od_crtype;
2009 lr->lrz_blocksize = od->od_crblocksize;
2010 lr->lrz_ibshift = ztest_random_ibshift();
2011 lr->lrz_bonustype = DMU_OT_UINT64_OTHER;
2012 lr->lrz_bonuslen = dmu_bonus_max();
2013 lr->lr_gen = od->od_crgen;
2014 lr->lr_crtime[0] = time(NULL);
2015
2016 if (ztest_replay_create(zd, lr, B_FALSE) != 0) {
2017 ASSERT(missing == 0);
2018 od->od_object = 0;
2019 missing++;
2020 } else {
2021 od->od_object = lr->lr_foid;
2022 od->od_type = od->od_crtype;
2023 od->od_blocksize = od->od_crblocksize;
2024 od->od_gen = od->od_crgen;
2025 ASSERT(od->od_object != 0);
2026 }
2027
2028 ztest_lr_free(lr, sizeof (*lr), od->od_name);
2029 }
2030
2031 return (missing);
2032 }
2033
2034 static int
2035 ztest_remove(ztest_ds_t *zd, ztest_od_t *od, int count)
2036 {
2037 int missing = 0;
2038 int error;
2039
2040 ASSERT(_mutex_held(&zd->zd_dirobj_lock));
2041
2042 od += count - 1;
2043
2044 for (int i = count - 1; i >= 0; i--, od--) {
2045 if (missing) {
2046 missing++;
2047 continue;
2048 }
2049
2050 /*
2051 * No object was found.
2052 */
2053 if (od->od_object == 0)
2054 continue;
2055
2056 lr_remove_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
2057
2058 lr->lr_doid = od->od_dir;
2059
2060 if ((error = ztest_replay_remove(zd, lr, B_FALSE)) != 0) {
2061 ASSERT3U(error, ==, ENOSPC);
2062 missing++;
2063 } else {
2064 od->od_object = 0;
2065 }
2066 ztest_lr_free(lr, sizeof (*lr), od->od_name);
2067 }
2068
2069 return (missing);
2070 }
2071
2072 static int
2073 ztest_write(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size,
2074 void *data)
2075 {
2076 lr_write_t *lr;
2077 int error;
2078
2079 lr = ztest_lr_alloc(sizeof (*lr) + size, NULL);
2080
2081 lr->lr_foid = object;
2082 lr->lr_offset = offset;
2083 lr->lr_length = size;
2084 lr->lr_blkoff = 0;
2085 BP_ZERO(&lr->lr_blkptr);
2086
2087 bcopy(data, lr + 1, size);
2088
2089 error = ztest_replay_write(zd, lr, B_FALSE);
2090
2091 ztest_lr_free(lr, sizeof (*lr) + size, NULL);
2092
2093 return (error);
2094 }
2095
2096 static int
2097 ztest_truncate(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2098 {
2099 lr_truncate_t *lr;
2100 int error;
2101
2102 lr = ztest_lr_alloc(sizeof (*lr), NULL);
2103
2104 lr->lr_foid = object;
2105 lr->lr_offset = offset;
2106 lr->lr_length = size;
2107
2108 error = ztest_replay_truncate(zd, lr, B_FALSE);
2109
2110 ztest_lr_free(lr, sizeof (*lr), NULL);
2111
2112 return (error);
2113 }
2114
2115 static int
2116 ztest_setattr(ztest_ds_t *zd, uint64_t object)
2117 {
2118 lr_setattr_t *lr;
2119 int error;
2120
2121 lr = ztest_lr_alloc(sizeof (*lr), NULL);
2122
2123 lr->lr_foid = object;
2124 lr->lr_size = 0;
2125 lr->lr_mode = 0;
2126
2127 error = ztest_replay_setattr(zd, lr, B_FALSE);
2128
2129 ztest_lr_free(lr, sizeof (*lr), NULL);
2130
2131 return (error);
2132 }
2133
2134 static void
2135 ztest_prealloc(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2136 {
2137 objset_t *os = zd->zd_os;
2138 dmu_tx_t *tx;
2139 uint64_t txg;
2140 rl_t *rl;
2141
2142 txg_wait_synced(dmu_objset_pool(os), 0);
2143
2144 ztest_object_lock(zd, object, RL_READER);
2145 rl = ztest_range_lock(zd, object, offset, size, RL_WRITER);
2146
2147 tx = dmu_tx_create(os);
2148
2149 dmu_tx_hold_write(tx, object, offset, size);
2150
2151 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
2152
2153 if (txg != 0) {
2154 dmu_prealloc(os, object, offset, size, tx);
2155 dmu_tx_commit(tx);
2156 txg_wait_synced(dmu_objset_pool(os), txg);
2157 } else {
2158 (void) dmu_free_long_range(os, object, offset, size);
2159 }
2160
2161 ztest_range_unlock(rl);
2162 ztest_object_unlock(zd, object);
2163 }
2164
2165 static void
2166 ztest_io(ztest_ds_t *zd, uint64_t object, uint64_t offset)
2167 {
2168 int err;
2169 ztest_block_tag_t wbt;
2170 dmu_object_info_t doi;
2171 enum ztest_io_type io_type;
2172 uint64_t blocksize;
2173 void *data;
2174
2175 VERIFY(dmu_object_info(zd->zd_os, object, &doi) == 0);
2176 blocksize = doi.doi_data_block_size;
2177 data = umem_alloc(blocksize, UMEM_NOFAIL);
2178
2179 /*
2180 * Pick an i/o type at random, biased toward writing block tags.
2181 */
2182 io_type = ztest_random(ZTEST_IO_TYPES);
2183 if (ztest_random(2) == 0)
2184 io_type = ZTEST_IO_WRITE_TAG;
2185
2186 (void) rw_rdlock(&zd->zd_zilog_lock);
2187
2188 switch (io_type) {
2189
2190 case ZTEST_IO_WRITE_TAG:
2191 ztest_bt_generate(&wbt, zd->zd_os, object, offset, 0, 0, 0);
2192 (void) ztest_write(zd, object, offset, sizeof (wbt), &wbt);
2193 break;
2194
2195 case ZTEST_IO_WRITE_PATTERN:
2196 (void) memset(data, 'a' + (object + offset) % 5, blocksize);
2197 if (ztest_random(2) == 0) {
2198 /*
2199 * Induce fletcher2 collisions to ensure that
2200 * zio_ddt_collision() detects and resolves them
2201 * when using fletcher2-verify for deduplication.
2202 */
2203 ((uint64_t *)data)[0] ^= 1ULL << 63;
2204 ((uint64_t *)data)[4] ^= 1ULL << 63;
2205 }
2206 (void) ztest_write(zd, object, offset, blocksize, data);
2207 break;
2208
2209 case ZTEST_IO_WRITE_ZEROES:
2210 bzero(data, blocksize);
2211 (void) ztest_write(zd, object, offset, blocksize, data);
2212 break;
2213
2214 case ZTEST_IO_TRUNCATE:
2215 (void) ztest_truncate(zd, object, offset, blocksize);
2216 break;
2217
2218 case ZTEST_IO_SETATTR:
2219 (void) ztest_setattr(zd, object);
2220 break;
2221
2222 case ZTEST_IO_REWRITE:
2223 (void) rw_rdlock(&ztest_name_lock);
2224 err = ztest_dsl_prop_set_uint64(zd->zd_name,
2225 ZFS_PROP_CHECKSUM, spa_dedup_checksum(ztest_spa),
2226 B_FALSE);
2227 VERIFY(err == 0 || err == ENOSPC);
2228 err = ztest_dsl_prop_set_uint64(zd->zd_name,
2229 ZFS_PROP_COMPRESSION,
2230 ztest_random_dsl_prop(ZFS_PROP_COMPRESSION),
2231 B_FALSE);
2232 VERIFY(err == 0 || err == ENOSPC);
2233 (void) rw_unlock(&ztest_name_lock);
2234
2235 VERIFY0(dmu_read(zd->zd_os, object, offset, blocksize, data,
2236 DMU_READ_NO_PREFETCH));
2237
2238 (void) ztest_write(zd, object, offset, blocksize, data);
2239 break;
2240 }
2241
2242 (void) rw_unlock(&zd->zd_zilog_lock);
2243
2244 umem_free(data, blocksize);
2245 }
2246
2247 /*
2248 * Initialize an object description template.
2249 */
2250 static void
2251 ztest_od_init(ztest_od_t *od, uint64_t id, char *tag, uint64_t index,
2252 dmu_object_type_t type, uint64_t blocksize, uint64_t gen)
2253 {
2254 od->od_dir = ZTEST_DIROBJ;
2255 od->od_object = 0;
2256
2257 od->od_crtype = type;
2258 od->od_crblocksize = blocksize ? blocksize : ztest_random_blocksize();
2259 od->od_crgen = gen;
2260
2261 od->od_type = DMU_OT_NONE;
2262 od->od_blocksize = 0;
2263 od->od_gen = 0;
2264
2265 (void) snprintf(od->od_name, sizeof (od->od_name), "%s(%lld)[%llu]",
2266 tag, (int64_t)id, index);
2267 }
2268
2269 /*
2270 * Lookup or create the objects for a test using the od template.
2271 * If the objects do not all exist, or if 'remove' is specified,
2272 * remove any existing objects and create new ones. Otherwise,
2273 * use the existing objects.
2274 */
2275 static int
2276 ztest_object_init(ztest_ds_t *zd, ztest_od_t *od, size_t size, boolean_t remove)
2277 {
2278 int count = size / sizeof (*od);
2279 int rv = 0;
2280
2281 VERIFY(mutex_lock(&zd->zd_dirobj_lock) == 0);
2282 if ((ztest_lookup(zd, od, count) != 0 || remove) &&
2283 (ztest_remove(zd, od, count) != 0 ||
2284 ztest_create(zd, od, count) != 0))
2285 rv = -1;
2286 zd->zd_od = od;
2287 VERIFY(mutex_unlock(&zd->zd_dirobj_lock) == 0);
2288
2289 return (rv);
2290 }
2291
2292 /* ARGSUSED */
2293 void
2294 ztest_zil_commit(ztest_ds_t *zd, uint64_t id)
2295 {
2296 zilog_t *zilog = zd->zd_zilog;
2297
2298 (void) rw_rdlock(&zd->zd_zilog_lock);
2299
2300 zil_commit(zilog, ztest_random(ZTEST_OBJECTS));
2301
2302 /*
2303 * Remember the committed values in zd, which is in parent/child
2304 * shared memory. If we die, the next iteration of ztest_run()
2305 * will verify that the log really does contain this record.
2306 */
2307 mutex_enter(&zilog->zl_lock);
2308 ASSERT(zd->zd_shared != NULL);
2309 ASSERT3U(zd->zd_shared->zd_seq, <=, zilog->zl_commit_lr_seq);
2310 zd->zd_shared->zd_seq = zilog->zl_commit_lr_seq;
2311 mutex_exit(&zilog->zl_lock);
2312
2313 (void) rw_unlock(&zd->zd_zilog_lock);
2314 }
2315
2316 /*
2317 * This function is designed to simulate the operations that occur during a
2318 * mount/unmount operation. We hold the dataset across these operations in an
2319 * attempt to expose any implicit assumptions about ZIL management.
2320 */
2321 /* ARGSUSED */
2322 void
2323 ztest_zil_remount(ztest_ds_t *zd, uint64_t id)
2324 {
2325 objset_t *os = zd->zd_os;
2326
2327 /*
2328 * We grab the zd_dirobj_lock to ensure that no other thread is
2329 * updating the zil (i.e. adding in-memory log records) and the
2330 * zd_zilog_lock to block any I/O.
2331 */
2332 VERIFY0(mutex_lock(&zd->zd_dirobj_lock));
2333 (void) rw_wrlock(&zd->zd_zilog_lock);
2334
2335 /* zfsvfs_teardown() */
2336 zil_close(zd->zd_zilog);
2337
2338 /* zfsvfs_setup() */
2339 VERIFY(zil_open(os, ztest_get_data) == zd->zd_zilog);
2340 zil_replay(os, zd, ztest_replay_vector);
2341
2342 (void) rw_unlock(&zd->zd_zilog_lock);
2343 VERIFY(mutex_unlock(&zd->zd_dirobj_lock) == 0);
2344 }
2345
2346 /*
2347 * Verify that we can't destroy an active pool, create an existing pool,
2348 * or create a pool with a bad vdev spec.
2349 */
2350 /* ARGSUSED */
2351 void
2352 ztest_spa_create_destroy(ztest_ds_t *zd, uint64_t id)
2353 {
2354 ztest_shared_opts_t *zo = &ztest_opts;
2355 spa_t *spa;
2356 nvlist_t *nvroot;
2357
2358 /*
2359 * Attempt to create using a bad file.
2360 */
2361 nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1);
2362 VERIFY3U(ENOENT, ==,
2363 spa_create("ztest_bad_file", nvroot, NULL, NULL));
2364 nvlist_free(nvroot);
2365
2366 /*
2367 * Attempt to create using a bad mirror.
2368 */
2369 nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 2, 1);
2370 VERIFY3U(ENOENT, ==,
2371 spa_create("ztest_bad_mirror", nvroot, NULL, NULL));
2372 nvlist_free(nvroot);
2373
2374 /*
2375 * Attempt to create an existing pool. It shouldn't matter
2376 * what's in the nvroot; we should fail with EEXIST.
2377 */
2378 (void) rw_rdlock(&ztest_name_lock);
2379 nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1);
2380 VERIFY3U(EEXIST, ==, spa_create(zo->zo_pool, nvroot, NULL, NULL));
2381 nvlist_free(nvroot);
2382 VERIFY3U(0, ==, spa_open(zo->zo_pool, &spa, FTAG));
2383 VERIFY3U(EBUSY, ==, spa_destroy(zo->zo_pool));
2384 spa_close(spa, FTAG);
2385
2386 (void) rw_unlock(&ztest_name_lock);
2387 }
2388
2389 /* ARGSUSED */
2390 void
2391 ztest_spa_upgrade(ztest_ds_t *zd, uint64_t id)
2392 {
2393 spa_t *spa;
2394 uint64_t initial_version = SPA_VERSION_INITIAL;
2395 uint64_t version, newversion;
2396 nvlist_t *nvroot, *props;
2397 char *name;
2398
2399 VERIFY0(mutex_lock(&ztest_vdev_lock));
2400 name = kmem_asprintf("%s_upgrade", ztest_opts.zo_pool);
2401
2402 /*
2403 * Clean up from previous runs.
2404 */
2405 (void) spa_destroy(name);
2406
2407 nvroot = make_vdev_root(NULL, NULL, name, ztest_opts.zo_vdev_size, 0,
2408 0, ztest_opts.zo_raidz, ztest_opts.zo_mirrors, 1);
2409
2410 /*
2411 * If we're configuring a RAIDZ device then make sure that the
2412 * the initial version is capable of supporting that feature.
2413 */
2414 switch (ztest_opts.zo_raidz_parity) {
2415 case 0:
2416 case 1:
2417 initial_version = SPA_VERSION_INITIAL;
2418 break;
2419 case 2:
2420 initial_version = SPA_VERSION_RAIDZ2;
2421 break;
2422 case 3:
2423 initial_version = SPA_VERSION_RAIDZ3;
2424 break;
2425 }
2426
2427 /*
2428 * Create a pool with a spa version that can be upgraded. Pick
2429 * a value between initial_version and SPA_VERSION_BEFORE_FEATURES.
2430 */
2431 do {
2432 version = ztest_random_spa_version(initial_version);
2433 } while (version > SPA_VERSION_BEFORE_FEATURES);
2434
2435 props = fnvlist_alloc();
2436 fnvlist_add_uint64(props,
2437 zpool_prop_to_name(ZPOOL_PROP_VERSION), version);
2438 VERIFY0(spa_create(name, nvroot, props, NULL));
2439 fnvlist_free(nvroot);
2440 fnvlist_free(props);
2441
2442 VERIFY0(spa_open(name, &spa, FTAG));
2443 VERIFY3U(spa_version(spa), ==, version);
2444 newversion = ztest_random_spa_version(version + 1);
2445
2446 if (ztest_opts.zo_verbose >= 4) {
2447 (void) printf("upgrading spa version from %llu to %llu\n",
2448 (u_longlong_t)version, (u_longlong_t)newversion);
2449 }
2450
2451 spa_upgrade(spa, newversion);
2452 VERIFY3U(spa_version(spa), >, version);
2453 VERIFY3U(spa_version(spa), ==, fnvlist_lookup_uint64(spa->spa_config,
2454 zpool_prop_to_name(ZPOOL_PROP_VERSION)));
2455 spa_close(spa, FTAG);
2456
2457 strfree(name);
2458 VERIFY0(mutex_unlock(&ztest_vdev_lock));
2459 }
2460
2461 static vdev_t *
2462 vdev_lookup_by_path(vdev_t *vd, const char *path)
2463 {
2464 vdev_t *mvd;
2465
2466 if (vd->vdev_path != NULL && strcmp(path, vd->vdev_path) == 0)
2467 return (vd);
2468
2469 for (int c = 0; c < vd->vdev_children; c++)
2470 if ((mvd = vdev_lookup_by_path(vd->vdev_child[c], path)) !=
2471 NULL)
2472 return (mvd);
2473
2474 return (NULL);
2475 }
2476
2477 /*
2478 * Find the first available hole which can be used as a top-level.
2479 */
2480 int
2481 find_vdev_hole(spa_t *spa)
2482 {
2483 vdev_t *rvd = spa->spa_root_vdev;
2484 int c;
2485
2486 ASSERT(spa_config_held(spa, SCL_VDEV, RW_READER) == SCL_VDEV);
2487
2488 for (c = 0; c < rvd->vdev_children; c++) {
2489 vdev_t *cvd = rvd->vdev_child[c];
2490
2491 if (cvd->vdev_ishole)
2492 break;
2493 }
2494 return (c);
2495 }
2496
2497 /*
2498 * Verify that vdev_add() works as expected.
2499 */
2500 /* ARGSUSED */
2501 void
2502 ztest_vdev_add_remove(ztest_ds_t *zd, uint64_t id)
2503 {
2504 ztest_shared_t *zs = ztest_shared;
2505 spa_t *spa = ztest_spa;
2506 uint64_t leaves;
2507 uint64_t guid;
2508 nvlist_t *nvroot;
2509 int error;
2510
2511 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2512 leaves = MAX(zs->zs_mirrors + zs->zs_splits, 1) * ztest_opts.zo_raidz;
2513
2514 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2515
2516 ztest_shared->zs_vdev_next_leaf = find_vdev_hole(spa) * leaves;
2517
2518 /*
2519 * If we have slogs then remove them 1/4 of the time.
2520 */
2521 if (spa_has_slogs(spa) && ztest_random(4) == 0) {
2522 /*
2523 * Grab the guid from the head of the log class rotor.
2524 */
2525 guid = spa_log_class(spa)->mc_rotor->mg_vd->vdev_guid;
2526
2527 spa_config_exit(spa, SCL_VDEV, FTAG);
2528
2529 /*
2530 * We have to grab the zs_name_lock as writer to
2531 * prevent a race between removing a slog (dmu_objset_find)
2532 * and destroying a dataset. Removing the slog will
2533 * grab a reference on the dataset which may cause
2534 * dmu_objset_destroy() to fail with EBUSY thus
2535 * leaving the dataset in an inconsistent state.
2536 */
2537 VERIFY(rw_wrlock(&ztest_name_lock) == 0);
2538 error = spa_vdev_remove(spa, guid, B_FALSE);
2539 VERIFY(rw_unlock(&ztest_name_lock) == 0);
2540
2541 if (error && error != EEXIST)
2542 fatal(0, "spa_vdev_remove() = %d", error);
2543 } else {
2544 spa_config_exit(spa, SCL_VDEV, FTAG);
2545
2546 /*
2547 * Make 1/4 of the devices be log devices.
2548 */
2549 nvroot = make_vdev_root(NULL, NULL, NULL,
2550 ztest_opts.zo_vdev_size, 0,
2551 ztest_random(4) == 0, ztest_opts.zo_raidz,
2552 zs->zs_mirrors, 1);
2553
2554 error = spa_vdev_add(spa, nvroot);
2555 nvlist_free(nvroot);
2556
2557 if (error == ENOSPC)
2558 ztest_record_enospc("spa_vdev_add");
2559 else if (error != 0)
2560 fatal(0, "spa_vdev_add() = %d", error);
2561 }
2562
2563 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2564 }
2565
2566 /*
2567 * Verify that adding/removing aux devices (l2arc, hot spare) works as expected.
2568 */
2569 /* ARGSUSED */
2570 void
2571 ztest_vdev_aux_add_remove(ztest_ds_t *zd, uint64_t id)
2572 {
2573 ztest_shared_t *zs = ztest_shared;
2574 spa_t *spa = ztest_spa;
2575 vdev_t *rvd = spa->spa_root_vdev;
2576 spa_aux_vdev_t *sav;
2577 char *aux;
2578 uint64_t guid = 0;
2579 int error;
2580
2581 if (ztest_random(2) == 0) {
2582 sav = &spa->spa_spares;
2583 aux = ZPOOL_CONFIG_SPARES;
2584 } else {
2585 sav = &spa->spa_l2cache;
2586 aux = ZPOOL_CONFIG_L2CACHE;
2587 }
2588
2589 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2590
2591 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2592
2593 if (sav->sav_count != 0 && ztest_random(4) == 0) {
2594 /*
2595 * Pick a random device to remove.
2596 */
2597 guid = sav->sav_vdevs[ztest_random(sav->sav_count)]->vdev_guid;
2598 } else {
2599 /*
2600 * Find an unused device we can add.
2601 */
2602 zs->zs_vdev_aux = 0;
2603 for (;;) {
2604 char path[MAXPATHLEN];
2605 int c;
2606 (void) snprintf(path, sizeof (path), ztest_aux_template,
2607 ztest_opts.zo_dir, ztest_opts.zo_pool, aux,
2608 zs->zs_vdev_aux);
2609 for (c = 0; c < sav->sav_count; c++)
2610 if (strcmp(sav->sav_vdevs[c]->vdev_path,
2611 path) == 0)
2612 break;
2613 if (c == sav->sav_count &&
2614 vdev_lookup_by_path(rvd, path) == NULL)
2615 break;
2616 zs->zs_vdev_aux++;
2617 }
2618 }
2619
2620 spa_config_exit(spa, SCL_VDEV, FTAG);
2621
2622 if (guid == 0) {
2623 /*
2624 * Add a new device.
2625 */
2626 nvlist_t *nvroot = make_vdev_root(NULL, aux, NULL,
2627 (ztest_opts.zo_vdev_size * 5) / 4, 0, 0, 0, 0, 1);
2628 error = spa_vdev_add(spa, nvroot);
2629 if (error != 0)
2630 fatal(0, "spa_vdev_add(%p) = %d", nvroot, error);
2631 nvlist_free(nvroot);
2632 } else {
2633 /*
2634 * Remove an existing device. Sometimes, dirty its
2635 * vdev state first to make sure we handle removal
2636 * of devices that have pending state changes.
2637 */
2638 if (ztest_random(2) == 0)
2639 (void) vdev_online(spa, guid, 0, NULL);
2640
2641 error = spa_vdev_remove(spa, guid, B_FALSE);
2642 if (error != 0 && error != EBUSY)
2643 fatal(0, "spa_vdev_remove(%llu) = %d", guid, error);
2644 }
2645
2646 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2647 }
2648
2649 /*
2650 * split a pool if it has mirror tlvdevs
2651 */
2652 /* ARGSUSED */
2653 void
2654 ztest_split_pool(ztest_ds_t *zd, uint64_t id)
2655 {
2656 ztest_shared_t *zs = ztest_shared;
2657 spa_t *spa = ztest_spa;
2658 vdev_t *rvd = spa->spa_root_vdev;
2659 nvlist_t *tree, **child, *config, *split, **schild;
2660 uint_t c, children, schildren = 0, lastlogid = 0;
2661 int error = 0;
2662
2663 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2664
2665 /* ensure we have a useable config; mirrors of raidz aren't supported */
2666 if (zs->zs_mirrors < 3 || ztest_opts.zo_raidz > 1) {
2667 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2668 return;
2669 }
2670
2671 /* clean up the old pool, if any */
2672 (void) spa_destroy("splitp");
2673
2674 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2675
2676 /* generate a config from the existing config */
2677 mutex_enter(&spa->spa_props_lock);
2678 VERIFY(nvlist_lookup_nvlist(spa->spa_config, ZPOOL_CONFIG_VDEV_TREE,
2679 &tree) == 0);
2680 mutex_exit(&spa->spa_props_lock);
2681
2682 VERIFY(nvlist_lookup_nvlist_array(tree, ZPOOL_CONFIG_CHILDREN, &child,
2683 &children) == 0);
2684
2685 schild = malloc(rvd->vdev_children * sizeof (nvlist_t *));
2686 for (c = 0; c < children; c++) {
2687 vdev_t *tvd = rvd->vdev_child[c];
2688 nvlist_t **mchild;
2689 uint_t mchildren;
2690
2691 if (tvd->vdev_islog || tvd->vdev_ops == &vdev_hole_ops) {
2692 VERIFY(nvlist_alloc(&schild[schildren], NV_UNIQUE_NAME,
2693 0) == 0);
2694 VERIFY(nvlist_add_string(schild[schildren],
2695 ZPOOL_CONFIG_TYPE, VDEV_TYPE_HOLE) == 0);
2696 VERIFY(nvlist_add_uint64(schild[schildren],
2697 ZPOOL_CONFIG_IS_HOLE, 1) == 0);
2698 if (lastlogid == 0)
2699 lastlogid = schildren;
2700 ++schildren;
2701 continue;
2702 }
2703 lastlogid = 0;
2704 VERIFY(nvlist_lookup_nvlist_array(child[c],
2705 ZPOOL_CONFIG_CHILDREN, &mchild, &mchildren) == 0);
2706 VERIFY(nvlist_dup(mchild[0], &schild[schildren++], 0) == 0);
2707 }
2708
2709 /* OK, create a config that can be used to split */
2710 VERIFY(nvlist_alloc(&split, NV_UNIQUE_NAME, 0) == 0);
2711 VERIFY(nvlist_add_string(split, ZPOOL_CONFIG_TYPE,
2712 VDEV_TYPE_ROOT) == 0);
2713 VERIFY(nvlist_add_nvlist_array(split, ZPOOL_CONFIG_CHILDREN, schild,
2714 lastlogid != 0 ? lastlogid : schildren) == 0);
2715
2716 VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, 0) == 0);
2717 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, split) == 0);
2718
2719 for (c = 0; c < schildren; c++)
2720 nvlist_free(schild[c]);
2721 free(schild);
2722 nvlist_free(split);
2723
2724 spa_config_exit(spa, SCL_VDEV, FTAG);
2725
2726 (void) rw_wrlock(&ztest_name_lock);
2727 error = spa_vdev_split_mirror(spa, "splitp", config, NULL, B_FALSE);
2728 (void) rw_unlock(&ztest_name_lock);
2729
2730 nvlist_free(config);
2731
2732 if (error == 0) {
2733 (void) printf("successful split - results:\n");
2734 mutex_enter(&spa_namespace_lock);
2735 show_pool_stats(spa);
2736 show_pool_stats(spa_lookup("splitp"));
2737 mutex_exit(&spa_namespace_lock);
2738 ++zs->zs_splits;
2739 --zs->zs_mirrors;
2740 }
2741 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2742
2743 }
2744
2745 /*
2746 * Verify that we can attach and detach devices.
2747 */
2748 /* ARGSUSED */
2749 void
2750 ztest_vdev_attach_detach(ztest_ds_t *zd, uint64_t id)
2751 {
2752 ztest_shared_t *zs = ztest_shared;
2753 spa_t *spa = ztest_spa;
2754 spa_aux_vdev_t *sav = &spa->spa_spares;
2755 vdev_t *rvd = spa->spa_root_vdev;
2756 vdev_t *oldvd, *newvd, *pvd;
2757 nvlist_t *root;
2758 uint64_t leaves;
2759 uint64_t leaf, top;
2760 uint64_t ashift = ztest_get_ashift();
2761 uint64_t oldguid, pguid;
2762 uint64_t oldsize, newsize;
2763 char oldpath[MAXPATHLEN], newpath[MAXPATHLEN];
2764 int replacing;
2765 int oldvd_has_siblings = B_FALSE;
2766 int newvd_is_spare = B_FALSE;
2767 int oldvd_is_log;
2768 int error, expected_error;
2769
2770 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2771 leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz;
2772
2773 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2774
2775 /*
2776 * Decide whether to do an attach or a replace.
2777 */
2778 replacing = ztest_random(2);
2779
2780 /*
2781 * Pick a random top-level vdev.
2782 */
2783 top = ztest_random_vdev_top(spa, B_TRUE);
2784
2785 /*
2786 * Pick a random leaf within it.
2787 */
2788 leaf = ztest_random(leaves);
2789
2790 /*
2791 * Locate this vdev.
2792 */
2793 oldvd = rvd->vdev_child[top];
2794 if (zs->zs_mirrors >= 1) {
2795 ASSERT(oldvd->vdev_ops == &vdev_mirror_ops);
2796 ASSERT(oldvd->vdev_children >= zs->zs_mirrors);
2797 oldvd = oldvd->vdev_child[leaf / ztest_opts.zo_raidz];
2798 }
2799 if (ztest_opts.zo_raidz > 1) {
2800 ASSERT(oldvd->vdev_ops == &vdev_raidz_ops);
2801 ASSERT(oldvd->vdev_children == ztest_opts.zo_raidz);
2802 oldvd = oldvd->vdev_child[leaf % ztest_opts.zo_raidz];
2803 }
2804
2805 /*
2806 * If we're already doing an attach or replace, oldvd may be a
2807 * mirror vdev -- in which case, pick a random child.
2808 */
2809 while (oldvd->vdev_children != 0) {
2810 oldvd_has_siblings = B_TRUE;
2811 ASSERT(oldvd->vdev_children >= 2);
2812 oldvd = oldvd->vdev_child[ztest_random(oldvd->vdev_children)];
2813 }
2814
2815 oldguid = oldvd->vdev_guid;
2816 oldsize = vdev_get_min_asize(oldvd);
2817 oldvd_is_log = oldvd->vdev_top->vdev_islog;
2818 (void) strcpy(oldpath, oldvd->vdev_path);
2819 pvd = oldvd->vdev_parent;
2820 pguid = pvd->vdev_guid;
2821
2822 /*
2823 * If oldvd has siblings, then half of the time, detach it.
2824 */
2825 if (oldvd_has_siblings && ztest_random(2) == 0) {
2826 spa_config_exit(spa, SCL_VDEV, FTAG);
2827 error = spa_vdev_detach(spa, oldguid, pguid, B_FALSE);
2828 if (error != 0 && error != ENODEV && error != EBUSY &&
2829 error != ENOTSUP)
2830 fatal(0, "detach (%s) returned %d", oldpath, error);
2831 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2832 return;
2833 }
2834
2835 /*
2836 * For the new vdev, choose with equal probability between the two
2837 * standard paths (ending in either 'a' or 'b') or a random hot spare.
2838 */
2839 if (sav->sav_count != 0 && ztest_random(3) == 0) {
2840 newvd = sav->sav_vdevs[ztest_random(sav->sav_count)];
2841 newvd_is_spare = B_TRUE;
2842 (void) strcpy(newpath, newvd->vdev_path);
2843 } else {
2844 (void) snprintf(newpath, sizeof (newpath), ztest_dev_template,
2845 ztest_opts.zo_dir, ztest_opts.zo_pool,
2846 top * leaves + leaf);
2847 if (ztest_random(2) == 0)
2848 newpath[strlen(newpath) - 1] = 'b';
2849 newvd = vdev_lookup_by_path(rvd, newpath);
2850 }
2851
2852 if (newvd) {
2853 newsize = vdev_get_min_asize(newvd);
2854 } else {
2855 /*
2856 * Make newsize a little bigger or smaller than oldsize.
2857 * If it's smaller, the attach should fail.
2858 * If it's larger, and we're doing a replace,
2859 * we should get dynamic LUN growth when we're done.
2860 */
2861 newsize = 10 * oldsize / (9 + ztest_random(3));
2862 }
2863
2864 /*
2865 * If pvd is not a mirror or root, the attach should fail with ENOTSUP,
2866 * unless it's a replace; in that case any non-replacing parent is OK.
2867 *
2868 * If newvd is already part of the pool, it should fail with EBUSY.
2869 *
2870 * If newvd is too small, it should fail with EOVERFLOW.
2871 */
2872 if (pvd->vdev_ops != &vdev_mirror_ops &&
2873 pvd->vdev_ops != &vdev_root_ops && (!replacing ||
2874 pvd->vdev_ops == &vdev_replacing_ops ||
2875 pvd->vdev_ops == &vdev_spare_ops))
2876 expected_error = ENOTSUP;
2877 else if (newvd_is_spare && (!replacing || oldvd_is_log))
2878 expected_error = ENOTSUP;
2879 else if (newvd == oldvd)
2880 expected_error = replacing ? 0 : EBUSY;
2881 else if (vdev_lookup_by_path(rvd, newpath) != NULL)
2882 expected_error = EBUSY;
2883 else if (newsize < oldsize)
2884 expected_error = EOVERFLOW;
2885 else if (ashift > oldvd->vdev_top->vdev_ashift)
2886 expected_error = EDOM;
2887 else
2888 expected_error = 0;
2889
2890 spa_config_exit(spa, SCL_VDEV, FTAG);
2891
2892 /*
2893 * Build the nvlist describing newpath.
2894 */
2895 root = make_vdev_root(newpath, NULL, NULL, newvd == NULL ? newsize : 0,
2896 ashift, 0, 0, 0, 1);
2897
2898 error = spa_vdev_attach(spa, oldguid, root, replacing);
2899
2900 nvlist_free(root);
2901
2902 /*
2903 * If our parent was the replacing vdev, but the replace completed,
2904 * then instead of failing with ENOTSUP we may either succeed,
2905 * fail with ENODEV, or fail with EOVERFLOW.
2906 */
2907 if (expected_error == ENOTSUP &&
2908 (error == 0 || error == ENODEV || error == EOVERFLOW))
2909 expected_error = error;
2910
2911 /*
2912 * If someone grew the LUN, the replacement may be too small.
2913 */
2914 if (error == EOVERFLOW || error == EBUSY)
2915 expected_error = error;
2916
2917 /* XXX workaround 6690467 */
2918 if (error != expected_error && expected_error != EBUSY) {
2919 fatal(0, "attach (%s %llu, %s %llu, %d) "
2920 "returned %d, expected %d",
2921 oldpath, oldsize, newpath,
2922 newsize, replacing, error, expected_error);
2923 }
2924
2925 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2926 }
2927
2928 /*
2929 * Callback function which expands the physical size of the vdev.
2930 */
2931 vdev_t *
2932 grow_vdev(vdev_t *vd, void *arg)
2933 {
2934 spa_t *spa = vd->vdev_spa;
2935 size_t *newsize = arg;
2936 size_t fsize;
2937 int fd;
2938
2939 ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
2940 ASSERT(vd->vdev_ops->vdev_op_leaf);
2941
2942 if ((fd = open(vd->vdev_path, O_RDWR)) == -1)
2943 return (vd);
2944
2945 fsize = lseek(fd, 0, SEEK_END);
2946 (void) ftruncate(fd, *newsize);
2947
2948 if (ztest_opts.zo_verbose >= 6) {
2949 (void) printf("%s grew from %lu to %lu bytes\n",
2950 vd->vdev_path, (ulong_t)fsize, (ulong_t)*newsize);
2951 }
2952 (void) close(fd);
2953 return (NULL);
2954 }
2955
2956 /*
2957 * Callback function which expands a given vdev by calling vdev_online().
2958 */
2959 /* ARGSUSED */
2960 vdev_t *
2961 online_vdev(vdev_t *vd, void *arg)
2962 {
2963 spa_t *spa = vd->vdev_spa;
2964 vdev_t *tvd = vd->vdev_top;
2965 uint64_t guid = vd->vdev_guid;
2966 uint64_t generation = spa->spa_config_generation + 1;
2967 vdev_state_t newstate = VDEV_STATE_UNKNOWN;
2968 int error;
2969
2970 ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
2971 ASSERT(vd->vdev_ops->vdev_op_leaf);
2972
2973 /* Calling vdev_online will initialize the new metaslabs */
2974 spa_config_exit(spa, SCL_STATE, spa);
2975 error = vdev_online(spa, guid, ZFS_ONLINE_EXPAND, &newstate);
2976 spa_config_enter(spa, SCL_STATE, spa, RW_READER);
2977
2978 /*
2979 * If vdev_online returned an error or the underlying vdev_open
2980 * failed then we abort the expand. The only way to know that
2981 * vdev_open fails is by checking the returned newstate.
2982 */
2983 if (error || newstate != VDEV_STATE_HEALTHY) {
2984 if (ztest_opts.zo_verbose >= 5) {
2985 (void) printf("Unable to expand vdev, state %llu, "
2986 "error %d\n", (u_longlong_t)newstate, error);
2987 }
2988 return (vd);
2989 }
2990 ASSERT3U(newstate, ==, VDEV_STATE_HEALTHY);
2991
2992 /*
2993 * Since we dropped the lock we need to ensure that we're
2994 * still talking to the original vdev. It's possible this
2995 * vdev may have been detached/replaced while we were
2996 * trying to online it.
2997 */
2998 if (generation != spa->spa_config_generation) {
2999 if (ztest_opts.zo_verbose >= 5) {
3000 (void) printf("vdev configuration has changed, "
3001 "guid %llu, state %llu, expected gen %llu, "
3002 "got gen %llu\n",
3003 (u_longlong_t)guid,
3004 (u_longlong_t)tvd->vdev_state,
3005 (u_longlong_t)generation,
3006 (u_longlong_t)spa->spa_config_generation);
3007 }
3008 return (vd);
3009 }
3010 return (NULL);
3011 }
3012
3013 /*
3014 * Traverse the vdev tree calling the supplied function.
3015 * We continue to walk the tree until we either have walked all
3016 * children or we receive a non-NULL return from the callback.
3017 * If a NULL callback is passed, then we just return back the first
3018 * leaf vdev we encounter.
3019 */
3020 vdev_t *
3021 vdev_walk_tree(vdev_t *vd, vdev_t *(*func)(vdev_t *, void *), void *arg)
3022 {
3023 if (vd->vdev_ops->vdev_op_leaf) {
3024 if (func == NULL)
3025 return (vd);
3026 else
3027 return (func(vd, arg));
3028 }
3029
3030 for (uint_t c = 0; c < vd->vdev_children; c++) {
3031 vdev_t *cvd = vd->vdev_child[c];
3032 if ((cvd = vdev_walk_tree(cvd, func, arg)) != NULL)
3033 return (cvd);
3034 }
3035 return (NULL);
3036 }
3037
3038 /*
3039 * Verify that dynamic LUN growth works as expected.
3040 */
3041 /* ARGSUSED */
3042 void
3043 ztest_vdev_LUN_growth(ztest_ds_t *zd, uint64_t id)
3044 {
3045 spa_t *spa = ztest_spa;
3046 vdev_t *vd, *tvd;
3047 metaslab_class_t *mc;
3048 metaslab_group_t *mg;
3049 size_t psize, newsize;
3050 uint64_t top;
3051 uint64_t old_class_space, new_class_space, old_ms_count, new_ms_count;
3052
3053 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
3054 spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3055
3056 top = ztest_random_vdev_top(spa, B_TRUE);
3057
3058 tvd = spa->spa_root_vdev->vdev_child[top];
3059 mg = tvd->vdev_mg;
3060 mc = mg->mg_class;
3061 old_ms_count = tvd->vdev_ms_count;
3062 old_class_space = metaslab_class_get_space(mc);
3063
3064 /*
3065 * Determine the size of the first leaf vdev associated with
3066 * our top-level device.
3067 */
3068 vd = vdev_walk_tree(tvd, NULL, NULL);
3069 ASSERT3P(vd, !=, NULL);
3070 ASSERT(vd->vdev_ops->vdev_op_leaf);
3071
3072 psize = vd->vdev_psize;
3073
3074 /*
3075 * We only try to expand the vdev if it's healthy, less than 4x its
3076 * original size, and it has a valid psize.
3077 */
3078 if (tvd->vdev_state != VDEV_STATE_HEALTHY ||
3079 psize == 0 || psize >= 4 * ztest_opts.zo_vdev_size) {
3080 spa_config_exit(spa, SCL_STATE, spa);
3081 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3082 return;
3083 }
3084 ASSERT(psize > 0);
3085 newsize = psize + psize / 8;
3086 ASSERT3U(newsize, >, psize);
3087
3088 if (ztest_opts.zo_verbose >= 6) {
3089 (void) printf("Expanding LUN %s from %lu to %lu\n",
3090 vd->vdev_path, (ulong_t)psize, (ulong_t)newsize);
3091 }
3092
3093 /*
3094 * Growing the vdev is a two step process:
3095 * 1). expand the physical size (i.e. relabel)
3096 * 2). online the vdev to create the new metaslabs
3097 */
3098 if (vdev_walk_tree(tvd, grow_vdev, &newsize) != NULL ||
3099 vdev_walk_tree(tvd, online_vdev, NULL) != NULL ||
3100 tvd->vdev_state != VDEV_STATE_HEALTHY) {
3101 if (ztest_opts.zo_verbose >= 5) {
3102 (void) printf("Could not expand LUN because "
3103 "the vdev configuration changed.\n");
3104 }
3105 spa_config_exit(spa, SCL_STATE, spa);
3106 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3107 return;
3108 }
3109
3110 spa_config_exit(spa, SCL_STATE, spa);
3111
3112 /*
3113 * Expanding the LUN will update the config asynchronously,
3114 * thus we must wait for the async thread to complete any
3115 * pending tasks before proceeding.
3116 */
3117 for (;;) {
3118 boolean_t done;
3119 mutex_enter(&spa->spa_async_lock);
3120 done = (spa->spa_async_thread == NULL && !spa->spa_async_tasks);
3121 mutex_exit(&spa->spa_async_lock);
3122 if (done)
3123 break;
3124 txg_wait_synced(spa_get_dsl(spa), 0);
3125 (void) poll(NULL, 0, 100);
3126 }
3127
3128 spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3129
3130 tvd = spa->spa_root_vdev->vdev_child[top];
3131 new_ms_count = tvd->vdev_ms_count;
3132 new_class_space = metaslab_class_get_space(mc);
3133
3134 if (tvd->vdev_mg != mg || mg->mg_class != mc) {
3135 if (ztest_opts.zo_verbose >= 5) {
3136 (void) printf("Could not verify LUN expansion due to "
3137 "intervening vdev offline or remove.\n");
3138 }
3139 spa_config_exit(spa, SCL_STATE, spa);
3140 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3141 return;
3142 }
3143
3144 /*
3145 * Make sure we were able to grow the vdev.
3146 */
3147 if (new_ms_count <= old_ms_count)
3148 fatal(0, "LUN expansion failed: ms_count %llu <= %llu\n",
3149 old_ms_count, new_ms_count);
3150
3151 /*
3152 * Make sure we were able to grow the pool.
3153 */
3154 if (new_class_space <= old_class_space)
3155 fatal(0, "LUN expansion failed: class_space %llu <= %llu\n",
3156 old_class_space, new_class_space);
3157
3158 if (ztest_opts.zo_verbose >= 5) {
3159 char oldnumbuf[6], newnumbuf[6];
3160
3161 nicenum(old_class_space, oldnumbuf);
3162 nicenum(new_class_space, newnumbuf);
3163 (void) printf("%s grew from %s to %s\n",
3164 spa->spa_name, oldnumbuf, newnumbuf);
3165 }
3166
3167 spa_config_exit(spa, SCL_STATE, spa);
3168 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3169 }
3170
3171 /*
3172 * Verify that dmu_objset_{create,destroy,open,close} work as expected.
3173 */
3174 /* ARGSUSED */
3175 static void
3176 ztest_objset_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
3177 {
3178 /*
3179 * Create the objects common to all ztest datasets.
3180 */
3181 VERIFY(zap_create_claim(os, ZTEST_DIROBJ,
3182 DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx) == 0);
3183 }
3184
3185 static int
3186 ztest_dataset_create(char *dsname)
3187 {
3188 uint64_t zilset = ztest_random(100);
3189 int err = dmu_objset_create(dsname, DMU_OST_OTHER, 0,
3190 ztest_objset_create_cb, NULL);
3191
3192 if (err || zilset < 80)
3193 return (err);
3194
3195 if (ztest_opts.zo_verbose >= 6)
3196 (void) printf("Setting dataset %s to sync always\n", dsname);
3197 return (ztest_dsl_prop_set_uint64(dsname, ZFS_PROP_SYNC,
3198 ZFS_SYNC_ALWAYS, B_FALSE));
3199 }
3200
3201 /* ARGSUSED */
3202 static int
3203 ztest_objset_destroy_cb(const char *name, void *arg)
3204 {
3205 objset_t *os;
3206 dmu_object_info_t doi;
3207 int error;
3208
3209 /*
3210 * Verify that the dataset contains a directory object.
3211 */
3212 VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_TRUE, FTAG, &os));
3213 error = dmu_object_info(os, ZTEST_DIROBJ, &doi);
3214 if (error != ENOENT) {
3215 /* We could have crashed in the middle of destroying it */
3216 ASSERT0(error);
3217 ASSERT3U(doi.doi_type, ==, DMU_OT_ZAP_OTHER);
3218 ASSERT3S(doi.doi_physical_blocks_512, >=, 0);
3219 }
3220 dmu_objset_disown(os, FTAG);
3221
3222 /*
3223 * Destroy the dataset.
3224 */
3225 if (strchr(name, '@') != NULL) {
3226 VERIFY0(dsl_destroy_snapshot(name, B_TRUE));
3227 } else {
3228 error = dsl_destroy_head(name);
3229 /* There could be a hold on this dataset */
3230 if (error != EBUSY)
3231 ASSERT0(error);
3232 }
3233 return (0);
3234 }
3235
3236 static boolean_t
3237 ztest_snapshot_create(char *osname, uint64_t id)
3238 {
3239 char snapname[ZFS_MAX_DATASET_NAME_LEN];
3240 int error;
3241
3242 (void) snprintf(snapname, sizeof (snapname), "%llu", (u_longlong_t)id);
3243
3244 error = dmu_objset_snapshot_one(osname, snapname);
3245 if (error == ENOSPC) {
3246 ztest_record_enospc(FTAG);
3247 return (B_FALSE);
3248 }
3249 if (error != 0 && error != EEXIST) {
3250 fatal(0, "ztest_snapshot_create(%s@%s) = %d", osname,
3251 snapname, error);
3252 }
3253 return (B_TRUE);
3254 }
3255
3256 static boolean_t
3257 ztest_snapshot_destroy(char *osname, uint64_t id)
3258 {
3259 char snapname[ZFS_MAX_DATASET_NAME_LEN];
3260 int error;
3261
3262 (void) snprintf(snapname, sizeof (snapname), "%s@%llu", osname,
3263 (u_longlong_t)id);
3264
3265 error = dsl_destroy_snapshot(snapname, B_FALSE);
3266 if (error != 0 && error != ENOENT)
3267 fatal(0, "ztest_snapshot_destroy(%s) = %d", snapname, error);
3268 return (B_TRUE);
3269 }
3270
3271 /* ARGSUSED */
3272 void
3273 ztest_dmu_objset_create_destroy(ztest_ds_t *zd, uint64_t id)
3274 {
3275 ztest_ds_t zdtmp;
3276 int iters;
3277 int error;
3278 objset_t *os, *os2;
3279 char name[ZFS_MAX_DATASET_NAME_LEN];
3280 zilog_t *zilog;
3281
3282 (void) rw_rdlock(&ztest_name_lock);
3283
3284 (void) snprintf(name, sizeof (name), "%s/temp_%llu",
3285 ztest_opts.zo_pool, (u_longlong_t)id);
3286
3287 /*
3288 * If this dataset exists from a previous run, process its replay log
3289 * half of the time. If we don't replay it, then dmu_objset_destroy()
3290 * (invoked from ztest_objset_destroy_cb()) should just throw it away.
3291 */
3292 if (ztest_random(2) == 0 &&
3293 dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os) == 0) {
3294 ztest_zd_init(&zdtmp, NULL, os);
3295 zil_replay(os, &zdtmp, ztest_replay_vector);
3296 ztest_zd_fini(&zdtmp);
3297 dmu_objset_disown(os, FTAG);
3298 }
3299
3300 /*
3301 * There may be an old instance of the dataset we're about to
3302 * create lying around from a previous run. If so, destroy it
3303 * and all of its snapshots.
3304 */
3305 (void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
3306 DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
3307
3308 /*
3309 * Verify that the destroyed dataset is no longer in the namespace.
3310 */
3311 VERIFY3U(ENOENT, ==, dmu_objset_own(name, DMU_OST_OTHER, B_TRUE,
3312 FTAG, &os));
3313
3314 /*
3315 * Verify that we can create a new dataset.
3316 */
3317 error = ztest_dataset_create(name);
3318 if (error) {
3319 if (error == ENOSPC) {
3320 ztest_record_enospc(FTAG);
3321 (void) rw_unlock(&ztest_name_lock);
3322 return;
3323 }
3324 fatal(0, "dmu_objset_create(%s) = %d", name, error);
3325 }
3326
3327 VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os));
3328
3329 ztest_zd_init(&zdtmp, NULL, os);
3330
3331 /*
3332 * Open the intent log for it.
3333 */
3334 zilog = zil_open(os, ztest_get_data);
3335
3336 /*
3337 * Put some objects in there, do a little I/O to them,
3338 * and randomly take a couple of snapshots along the way.
3339 */
3340 iters = ztest_random(5);
3341 for (int i = 0; i < iters; i++) {
3342 ztest_dmu_object_alloc_free(&zdtmp, id);
3343 if (ztest_random(iters) == 0)
3344 (void) ztest_snapshot_create(name, i);
3345 }
3346
3347 /*
3348 * Verify that we cannot create an existing dataset.
3349 */
3350 VERIFY3U(EEXIST, ==,
3351 dmu_objset_create(name, DMU_OST_OTHER, 0, NULL, NULL));
3352
3353 /*
3354 * Verify that we can hold an objset that is also owned.
3355 */
3356 VERIFY3U(0, ==, dmu_objset_hold(name, FTAG, &os2));
3357 dmu_objset_rele(os2, FTAG);
3358
3359 /*
3360 * Verify that we cannot own an objset that is already owned.
3361 */
3362 VERIFY3U(EBUSY, ==,
3363 dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os2));
3364
3365 zil_close(zilog);
3366 dmu_objset_disown(os, FTAG);
3367 ztest_zd_fini(&zdtmp);
3368
3369 (void) rw_unlock(&ztest_name_lock);
3370 }
3371
3372 /*
3373 * Verify that dmu_snapshot_{create,destroy,open,close} work as expected.
3374 */
3375 void
3376 ztest_dmu_snapshot_create_destroy(ztest_ds_t *zd, uint64_t id)
3377 {
3378 (void) rw_rdlock(&ztest_name_lock);
3379 (void) ztest_snapshot_destroy(zd->zd_name, id);
3380 (void) ztest_snapshot_create(zd->zd_name, id);
3381 (void) rw_unlock(&ztest_name_lock);
3382 }
3383
3384 /*
3385 * Cleanup non-standard snapshots and clones.
3386 */
3387 void
3388 ztest_dsl_dataset_cleanup(char *osname, uint64_t id)
3389 {
3390 char snap1name[ZFS_MAX_DATASET_NAME_LEN];
3391 char clone1name[ZFS_MAX_DATASET_NAME_LEN];
3392 char snap2name[ZFS_MAX_DATASET_NAME_LEN];
3393 char clone2name[ZFS_MAX_DATASET_NAME_LEN];
3394 char snap3name[ZFS_MAX_DATASET_NAME_LEN];
3395 int error;
3396
3397 (void) snprintf(snap1name, sizeof (snap1name),
3398 "%s@s1_%llu", osname, id);
3399 (void) snprintf(clone1name, sizeof (clone1name),
3400 "%s/c1_%llu", osname, id);
3401 (void) snprintf(snap2name, sizeof (snap2name),
3402 "%s@s2_%llu", clone1name, id);
3403 (void) snprintf(clone2name, sizeof (clone2name),
3404 "%s/c2_%llu", osname, id);
3405 (void) snprintf(snap3name, sizeof (snap3name),
3406 "%s@s3_%llu", clone1name, id);
3407
3408 error = dsl_destroy_head(clone2name);
3409 if (error && error != ENOENT)
3410 fatal(0, "dsl_destroy_head(%s) = %d", clone2name, error);
3411 error = dsl_destroy_snapshot(snap3name, B_FALSE);
3412 if (error && error != ENOENT)
3413 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap3name, error);
3414 error = dsl_destroy_snapshot(snap2name, B_FALSE);
3415 if (error && error != ENOENT)
3416 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap2name, error);
3417 error = dsl_destroy_head(clone1name);
3418 if (error && error != ENOENT)
3419 fatal(0, "dsl_destroy_head(%s) = %d", clone1name, error);
3420 error = dsl_destroy_snapshot(snap1name, B_FALSE);
3421 if (error && error != ENOENT)
3422 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap1name, error);
3423 }
3424
3425 /*
3426 * Verify dsl_dataset_promote handles EBUSY
3427 */
3428 void
3429 ztest_dsl_dataset_promote_busy(ztest_ds_t *zd, uint64_t id)
3430 {
3431 objset_t *os;
3432 char snap1name[ZFS_MAX_DATASET_NAME_LEN];
3433 char clone1name[ZFS_MAX_DATASET_NAME_LEN];
3434 char snap2name[ZFS_MAX_DATASET_NAME_LEN];
3435 char clone2name[ZFS_MAX_DATASET_NAME_LEN];
3436 char snap3name[ZFS_MAX_DATASET_NAME_LEN];
3437 char *osname = zd->zd_name;
3438 int error;
3439
3440 (void) rw_rdlock(&ztest_name_lock);
3441
3442 ztest_dsl_dataset_cleanup(osname, id);
3443
3444 (void) snprintf(snap1name, sizeof (snap1name),
3445 "%s@s1_%llu", osname, id);
3446 (void) snprintf(clone1name, sizeof (clone1name),
3447 "%s/c1_%llu", osname, id);
3448 (void) snprintf(snap2name, sizeof (snap2name),
3449 "%s@s2_%llu", clone1name, id);
3450 (void) snprintf(clone2name, sizeof (clone2name),
3451 "%s/c2_%llu", osname, id);
3452 (void) snprintf(snap3name, sizeof (snap3name),
3453 "%s@s3_%llu", clone1name, id);
3454
3455 error = dmu_objset_snapshot_one(osname, strchr(snap1name, '@') + 1);
3456 if (error && error != EEXIST) {
3457 if (error == ENOSPC) {
3458 ztest_record_enospc(FTAG);
3459 goto out;
3460 }
3461 fatal(0, "dmu_take_snapshot(%s) = %d", snap1name, error);
3462 }
3463
3464 error = dmu_objset_clone(clone1name, snap1name);
3465 if (error) {
3466 if (error == ENOSPC) {
3467 ztest_record_enospc(FTAG);
3468 goto out;
3469 }
3470 fatal(0, "dmu_objset_create(%s) = %d", clone1name, error);
3471 }
3472
3473 error = dmu_objset_snapshot_one(clone1name, strchr(snap2name, '@') + 1);
3474 if (error && error != EEXIST) {
3475 if (error == ENOSPC) {
3476 ztest_record_enospc(FTAG);
3477 goto out;
3478 }
3479 fatal(0, "dmu_open_snapshot(%s) = %d", snap2name, error);
3480 }
3481
3482 error = dmu_objset_snapshot_one(clone1name, strchr(snap3name, '@') + 1);
3483 if (error && error != EEXIST) {
3484 if (error == ENOSPC) {
3485 ztest_record_enospc(FTAG);
3486 goto out;
3487 }
3488 fatal(0, "dmu_open_snapshot(%s) = %d", snap3name, error);
3489 }
3490
3491 error = dmu_objset_clone(clone2name, snap3name);
3492 if (error) {
3493 if (error == ENOSPC) {
3494 ztest_record_enospc(FTAG);
3495 goto out;
3496 }
3497 fatal(0, "dmu_objset_create(%s) = %d", clone2name, error);
3498 }
3499
3500 error = dmu_objset_own(snap2name, DMU_OST_ANY, B_TRUE, FTAG, &os);
3501 if (error)
3502 fatal(0, "dmu_objset_own(%s) = %d", snap2name, error);
3503 error = dsl_dataset_promote(clone2name, NULL);
3504 if (error == ENOSPC) {
3505 dmu_objset_disown(os, FTAG);
3506 ztest_record_enospc(FTAG);
3507 goto out;
3508 }
3509 if (error != EBUSY)
3510 fatal(0, "dsl_dataset_promote(%s), %d, not EBUSY", clone2name,
3511 error);
3512 dmu_objset_disown(os, FTAG);
3513
3514 out:
3515 ztest_dsl_dataset_cleanup(osname, id);
3516
3517 (void) rw_unlock(&ztest_name_lock);
3518 }
3519
3520 /*
3521 * Verify that dmu_object_{alloc,free} work as expected.
3522 */
3523 void
3524 ztest_dmu_object_alloc_free(ztest_ds_t *zd, uint64_t id)
3525 {
3526 ztest_od_t od[4];
3527 int batchsize = sizeof (od) / sizeof (od[0]);
3528
3529 for (int b = 0; b < batchsize; b++)
3530 ztest_od_init(&od[b], id, FTAG, b, DMU_OT_UINT64_OTHER, 0, 0);
3531
3532 /*
3533 * Destroy the previous batch of objects, create a new batch,
3534 * and do some I/O on the new objects.
3535 */
3536 if (ztest_object_init(zd, od, sizeof (od), B_TRUE) != 0)
3537 return;
3538
3539 while (ztest_random(4 * batchsize) != 0)
3540 ztest_io(zd, od[ztest_random(batchsize)].od_object,
3541 ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
3542 }
3543
3544 /*
3545 * Verify that dmu_{read,write} work as expected.
3546 */
3547 void
3548 ztest_dmu_read_write(ztest_ds_t *zd, uint64_t id)
3549 {
3550 objset_t *os = zd->zd_os;
3551 ztest_od_t od[2];
3552 dmu_tx_t *tx;
3553 int i, freeit, error;
3554 uint64_t n, s, txg;
3555 bufwad_t *packbuf, *bigbuf, *pack, *bigH, *bigT;
3556 uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
3557 uint64_t chunksize = (1000 + ztest_random(1000)) * sizeof (uint64_t);
3558 uint64_t regions = 997;
3559 uint64_t stride = 123456789ULL;
3560 uint64_t width = 40;
3561 int free_percent = 5;
3562
3563 /*
3564 * This test uses two objects, packobj and bigobj, that are always
3565 * updated together (i.e. in the same tx) so that their contents are
3566 * in sync and can be compared. Their contents relate to each other
3567 * in a simple way: packobj is a dense array of 'bufwad' structures,
3568 * while bigobj is a sparse array of the same bufwads. Specifically,
3569 * for any index n, there are three bufwads that should be identical:
3570 *
3571 * packobj, at offset n * sizeof (bufwad_t)
3572 * bigobj, at the head of the nth chunk
3573 * bigobj, at the tail of the nth chunk
3574 *
3575 * The chunk size is arbitrary. It doesn't have to be a power of two,
3576 * and it doesn't have any relation to the object blocksize.
3577 * The only requirement is that it can hold at least two bufwads.
3578 *
3579 * Normally, we write the bufwad to each of these locations.
3580 * However, free_percent of the time we instead write zeroes to
3581 * packobj and perform a dmu_free_range() on bigobj. By comparing
3582 * bigobj to packobj, we can verify that the DMU is correctly
3583 * tracking which parts of an object are allocated and free,
3584 * and that the contents of the allocated blocks are correct.
3585 */
3586
3587 /*
3588 * Read the directory info. If it's the first time, set things up.
3589 */
3590 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, chunksize);
3591 ztest_od_init(&od[1], id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize);
3592
3593 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
3594 return;
3595
3596 bigobj = od[0].od_object;
3597 packobj = od[1].od_object;
3598 chunksize = od[0].od_gen;
3599 ASSERT(chunksize == od[1].od_gen);
3600
3601 /*
3602 * Prefetch a random chunk of the big object.
3603 * Our aim here is to get some async reads in flight
3604 * for blocks that we may free below; the DMU should
3605 * handle this race correctly.
3606 */
3607 n = ztest_random(regions) * stride + ztest_random(width);
3608 s = 1 + ztest_random(2 * width - 1);
3609 dmu_prefetch(os, bigobj, 0, n * chunksize, s * chunksize,
3610 ZIO_PRIORITY_SYNC_READ);
3611
3612 /*
3613 * Pick a random index and compute the offsets into packobj and bigobj.
3614 */
3615 n = ztest_random(regions) * stride + ztest_random(width);
3616 s = 1 + ztest_random(width - 1);
3617
3618 packoff = n * sizeof (bufwad_t);
3619 packsize = s * sizeof (bufwad_t);
3620
3621 bigoff = n * chunksize;
3622 bigsize = s * chunksize;
3623
3624 packbuf = umem_alloc(packsize, UMEM_NOFAIL);
3625 bigbuf = umem_alloc(bigsize, UMEM_NOFAIL);
3626
3627 /*
3628 * free_percent of the time, free a range of bigobj rather than
3629 * overwriting it.
3630 */
3631 freeit = (ztest_random(100) < free_percent);
3632
3633 /*
3634 * Read the current contents of our objects.
3635 */
3636 error = dmu_read(os, packobj, packoff, packsize, packbuf,
3637 DMU_READ_PREFETCH);
3638 ASSERT0(error);
3639 error = dmu_read(os, bigobj, bigoff, bigsize, bigbuf,
3640 DMU_READ_PREFETCH);
3641 ASSERT0(error);
3642
3643 /*
3644 * Get a tx for the mods to both packobj and bigobj.
3645 */
3646 tx = dmu_tx_create(os);
3647
3648 dmu_tx_hold_write(tx, packobj, packoff, packsize);
3649
3650 if (freeit)
3651 dmu_tx_hold_free(tx, bigobj, bigoff, bigsize);
3652 else
3653 dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
3654
3655 /* This accounts for setting the checksum/compression. */
3656 dmu_tx_hold_bonus(tx, bigobj);
3657
3658 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3659 if (txg == 0) {
3660 umem_free(packbuf, packsize);
3661 umem_free(bigbuf, bigsize);
3662 return;
3663 }
3664
3665 enum zio_checksum cksum;
3666 do {
3667 cksum = (enum zio_checksum)
3668 ztest_random_dsl_prop(ZFS_PROP_CHECKSUM);
3669 } while (cksum >= ZIO_CHECKSUM_LEGACY_FUNCTIONS);
3670 dmu_object_set_checksum(os, bigobj, cksum, tx);
3671
3672 enum zio_compress comp;
3673 do {
3674 comp = (enum zio_compress)
3675 ztest_random_dsl_prop(ZFS_PROP_COMPRESSION);
3676 } while (comp >= ZIO_COMPRESS_LEGACY_FUNCTIONS);
3677 dmu_object_set_compress(os, bigobj, comp, tx);
3678
3679 /*
3680 * For each index from n to n + s, verify that the existing bufwad
3681 * in packobj matches the bufwads at the head and tail of the
3682 * corresponding chunk in bigobj. Then update all three bufwads
3683 * with the new values we want to write out.
3684 */
3685 for (i = 0; i < s; i++) {
3686 /* LINTED */
3687 pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
3688 /* LINTED */
3689 bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
3690 /* LINTED */
3691 bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
3692
3693 ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
3694 ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
3695
3696 if (pack->bw_txg > txg)
3697 fatal(0, "future leak: got %llx, open txg is %llx",
3698 pack->bw_txg, txg);
3699
3700 if (pack->bw_data != 0 && pack->bw_index != n + i)
3701 fatal(0, "wrong index: got %llx, wanted %llx+%llx",
3702 pack->bw_index, n, i);
3703
3704 if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
3705 fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
3706
3707 if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
3708 fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
3709
3710 if (freeit) {
3711 bzero(pack, sizeof (bufwad_t));
3712 } else {
3713 pack->bw_index = n + i;
3714 pack->bw_txg = txg;
3715 pack->bw_data = 1 + ztest_random(-2ULL);
3716 }
3717 *bigH = *pack;
3718 *bigT = *pack;
3719 }
3720
3721 /*
3722 * We've verified all the old bufwads, and made new ones.
3723 * Now write them out.
3724 */
3725 dmu_write(os, packobj, packoff, packsize, packbuf, tx);
3726
3727 if (freeit) {
3728 if (ztest_opts.zo_verbose >= 7) {
3729 (void) printf("freeing offset %llx size %llx"
3730 " txg %llx\n",
3731 (u_longlong_t)bigoff,
3732 (u_longlong_t)bigsize,
3733 (u_longlong_t)txg);
3734 }
3735 VERIFY(0 == dmu_free_range(os, bigobj, bigoff, bigsize, tx));
3736 } else {
3737 if (ztest_opts.zo_verbose >= 7) {
3738 (void) printf("writing offset %llx size %llx"
3739 " txg %llx\n",
3740 (u_longlong_t)bigoff,
3741 (u_longlong_t)bigsize,
3742 (u_longlong_t)txg);
3743 }
3744 dmu_write(os, bigobj, bigoff, bigsize, bigbuf, tx);
3745 }
3746
3747 dmu_tx_commit(tx);
3748
3749 /*
3750 * Sanity check the stuff we just wrote.
3751 */
3752 {
3753 void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
3754 void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
3755
3756 VERIFY(0 == dmu_read(os, packobj, packoff,
3757 packsize, packcheck, DMU_READ_PREFETCH));
3758 VERIFY(0 == dmu_read(os, bigobj, bigoff,
3759 bigsize, bigcheck, DMU_READ_PREFETCH));
3760
3761 ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
3762 ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
3763
3764 umem_free(packcheck, packsize);
3765 umem_free(bigcheck, bigsize);
3766 }
3767
3768 umem_free(packbuf, packsize);
3769 umem_free(bigbuf, bigsize);
3770 }
3771
3772 void
3773 compare_and_update_pbbufs(uint64_t s, bufwad_t *packbuf, bufwad_t *bigbuf,
3774 uint64_t bigsize, uint64_t n, uint64_t chunksize, uint64_t txg)
3775 {
3776 uint64_t i;
3777 bufwad_t *pack;
3778 bufwad_t *bigH;
3779 bufwad_t *bigT;
3780
3781 /*
3782 * For each index from n to n + s, verify that the existing bufwad
3783 * in packobj matches the bufwads at the head and tail of the
3784 * corresponding chunk in bigobj. Then update all three bufwads
3785 * with the new values we want to write out.
3786 */
3787 for (i = 0; i < s; i++) {
3788 /* LINTED */
3789 pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
3790 /* LINTED */
3791 bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
3792 /* LINTED */
3793 bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
3794
3795 ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
3796 ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
3797
3798 if (pack->bw_txg > txg)
3799 fatal(0, "future leak: got %llx, open txg is %llx",
3800 pack->bw_txg, txg);
3801
3802 if (pack->bw_data != 0 && pack->bw_index != n + i)
3803 fatal(0, "wrong index: got %llx, wanted %llx+%llx",
3804 pack->bw_index, n, i);
3805
3806 if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
3807 fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
3808
3809 if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
3810 fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
3811
3812 pack->bw_index = n + i;
3813 pack->bw_txg = txg;
3814 pack->bw_data = 1 + ztest_random(-2ULL);
3815
3816 *bigH = *pack;
3817 *bigT = *pack;
3818 }
3819 }
3820
3821 void
3822 ztest_dmu_read_write_zcopy(ztest_ds_t *zd, uint64_t id)
3823 {
3824 objset_t *os = zd->zd_os;
3825 ztest_od_t od[2];
3826 dmu_tx_t *tx;
3827 uint64_t i;
3828 int error;
3829 uint64_t n, s, txg;
3830 bufwad_t *packbuf, *bigbuf;
3831 uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
3832 uint64_t blocksize = ztest_random_blocksize();
3833 uint64_t chunksize = blocksize;
3834 uint64_t regions = 997;
3835 uint64_t stride = 123456789ULL;
3836 uint64_t width = 9;
3837 dmu_buf_t *bonus_db;
3838 arc_buf_t **bigbuf_arcbufs;
3839 dmu_object_info_t doi;
3840
3841 /*
3842 * This test uses two objects, packobj and bigobj, that are always
3843 * updated together (i.e. in the same tx) so that their contents are
3844 * in sync and can be compared. Their contents relate to each other
3845 * in a simple way: packobj is a dense array of 'bufwad' structures,
3846 * while bigobj is a sparse array of the same bufwads. Specifically,
3847 * for any index n, there are three bufwads that should be identical:
3848 *
3849 * packobj, at offset n * sizeof (bufwad_t)
3850 * bigobj, at the head of the nth chunk
3851 * bigobj, at the tail of the nth chunk
3852 *
3853 * The chunk size is set equal to bigobj block size so that
3854 * dmu_assign_arcbuf() can be tested for object updates.
3855 */
3856
3857 /*
3858 * Read the directory info. If it's the first time, set things up.
3859 */
3860 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
3861 ztest_od_init(&od[1], id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize);
3862
3863 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
3864 return;
3865
3866 bigobj = od[0].od_object;
3867 packobj = od[1].od_object;
3868 blocksize = od[0].od_blocksize;
3869 chunksize = blocksize;
3870 ASSERT(chunksize == od[1].od_gen);
3871
3872 VERIFY(dmu_object_info(os, bigobj, &doi) == 0);
3873 VERIFY(ISP2(doi.doi_data_block_size));
3874 VERIFY(chunksize == doi.doi_data_block_size);
3875 VERIFY(chunksize >= 2 * sizeof (bufwad_t));
3876
3877 /*
3878 * Pick a random index and compute the offsets into packobj and bigobj.
3879 */
3880 n = ztest_random(regions) * stride + ztest_random(width);
3881 s = 1 + ztest_random(width - 1);
3882
3883 packoff = n * sizeof (bufwad_t);
3884 packsize = s * sizeof (bufwad_t);
3885
3886 bigoff = n * chunksize;
3887 bigsize = s * chunksize;
3888
3889 packbuf = umem_zalloc(packsize, UMEM_NOFAIL);
3890 bigbuf = umem_zalloc(bigsize, UMEM_NOFAIL);
3891
3892 VERIFY3U(0, ==, dmu_bonus_hold(os, bigobj, FTAG, &bonus_db));
3893
3894 bigbuf_arcbufs = umem_zalloc(2 * s * sizeof (arc_buf_t *), UMEM_NOFAIL);
3895
3896 /*
3897 * Iteration 0 test zcopy for DB_UNCACHED dbufs.
3898 * Iteration 1 test zcopy to already referenced dbufs.
3899 * Iteration 2 test zcopy to dirty dbuf in the same txg.
3900 * Iteration 3 test zcopy to dbuf dirty in previous txg.
3901 * Iteration 4 test zcopy when dbuf is no longer dirty.
3902 * Iteration 5 test zcopy when it can't be done.
3903 * Iteration 6 one more zcopy write.
3904 */
3905 for (i = 0; i < 7; i++) {
3906 uint64_t j;
3907 uint64_t off;
3908
3909 /*
3910 * In iteration 5 (i == 5) use arcbufs
3911 * that don't match bigobj blksz to test
3912 * dmu_assign_arcbuf() when it can't directly
3913 * assign an arcbuf to a dbuf.
3914 */
3915 for (j = 0; j < s; j++) {
3916 if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) {
3917 bigbuf_arcbufs[j] =
3918 dmu_request_arcbuf(bonus_db, chunksize);
3919 } else {
3920 bigbuf_arcbufs[2 * j] =
3921 dmu_request_arcbuf(bonus_db, chunksize / 2);
3922 bigbuf_arcbufs[2 * j + 1] =
3923 dmu_request_arcbuf(bonus_db, chunksize / 2);
3924 }
3925 }
3926
3927 /*
3928 * Get a tx for the mods to both packobj and bigobj.
3929 */
3930 tx = dmu_tx_create(os);
3931
3932 dmu_tx_hold_write(tx, packobj, packoff, packsize);
3933 dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
3934
3935 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3936 if (txg == 0) {
3937 umem_free(packbuf, packsize);
3938 umem_free(bigbuf, bigsize);
3939 for (j = 0; j < s; j++) {
3940 if (i != 5 ||
3941 chunksize < (SPA_MINBLOCKSIZE * 2)) {
3942 dmu_return_arcbuf(bigbuf_arcbufs[j]);
3943 } else {
3944 dmu_return_arcbuf(
3945 bigbuf_arcbufs[2 * j]);
3946 dmu_return_arcbuf(
3947 bigbuf_arcbufs[2 * j + 1]);
3948 }
3949 }
3950 umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
3951 dmu_buf_rele(bonus_db, FTAG);
3952 return;
3953 }
3954
3955 /*
3956 * 50% of the time don't read objects in the 1st iteration to
3957 * test dmu_assign_arcbuf() for the case when there're no
3958 * existing dbufs for the specified offsets.
3959 */
3960 if (i != 0 || ztest_random(2) != 0) {
3961 error = dmu_read(os, packobj, packoff,
3962 packsize, packbuf, DMU_READ_PREFETCH);
3963 ASSERT0(error);
3964 error = dmu_read(os, bigobj, bigoff, bigsize,
3965 bigbuf, DMU_READ_PREFETCH);
3966 ASSERT0(error);
3967 }
3968 compare_and_update_pbbufs(s, packbuf, bigbuf, bigsize,
3969 n, chunksize, txg);
3970
3971 /*
3972 * We've verified all the old bufwads, and made new ones.
3973 * Now write them out.
3974 */
3975 dmu_write(os, packobj, packoff, packsize, packbuf, tx);
3976 if (ztest_opts.zo_verbose >= 7) {
3977 (void) printf("writing offset %llx size %llx"
3978 " txg %llx\n",
3979 (u_longlong_t)bigoff,
3980 (u_longlong_t)bigsize,
3981 (u_longlong_t)txg);
3982 }
3983 for (off = bigoff, j = 0; j < s; j++, off += chunksize) {
3984 dmu_buf_t *dbt;
3985 if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) {
3986 bcopy((caddr_t)bigbuf + (off - bigoff),
3987 bigbuf_arcbufs[j]->b_data, chunksize);
3988 } else {
3989 bcopy((caddr_t)bigbuf + (off - bigoff),
3990 bigbuf_arcbufs[2 * j]->b_data,
3991 chunksize / 2);
3992 bcopy((caddr_t)bigbuf + (off - bigoff) +
3993 chunksize / 2,
3994 bigbuf_arcbufs[2 * j + 1]->b_data,
3995 chunksize / 2);
3996 }
3997
3998 if (i == 1) {
3999 VERIFY(dmu_buf_hold(os, bigobj, off,
4000 FTAG, &dbt, DMU_READ_NO_PREFETCH) == 0);
4001 }
4002 if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) {
4003 dmu_assign_arcbuf(bonus_db, off,
4004 bigbuf_arcbufs[j], tx);
4005 } else {
4006 dmu_assign_arcbuf(bonus_db, off,
4007 bigbuf_arcbufs[2 * j], tx);
4008 dmu_assign_arcbuf(bonus_db,
4009 off + chunksize / 2,
4010 bigbuf_arcbufs[2 * j + 1], tx);
4011 }
4012 if (i == 1) {
4013 dmu_buf_rele(dbt, FTAG);
4014 }
4015 }
4016 dmu_tx_commit(tx);
4017
4018 /*
4019 * Sanity check the stuff we just wrote.
4020 */
4021 {
4022 void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
4023 void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
4024
4025 VERIFY(0 == dmu_read(os, packobj, packoff,
4026 packsize, packcheck, DMU_READ_PREFETCH));
4027 VERIFY(0 == dmu_read(os, bigobj, bigoff,
4028 bigsize, bigcheck, DMU_READ_PREFETCH));
4029
4030 ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
4031 ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
4032
4033 umem_free(packcheck, packsize);
4034 umem_free(bigcheck, bigsize);
4035 }
4036 if (i == 2) {
4037 txg_wait_open(dmu_objset_pool(os), 0);
4038 } else if (i == 3) {
4039 txg_wait_synced(dmu_objset_pool(os), 0);
4040 }
4041 }
4042
4043 dmu_buf_rele(bonus_db, FTAG);
4044 umem_free(packbuf, packsize);
4045 umem_free(bigbuf, bigsize);
4046 umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
4047 }
4048
4049 /* ARGSUSED */
4050 void
4051 ztest_dmu_write_parallel(ztest_ds_t *zd, uint64_t id)
4052 {
4053 ztest_od_t od[1];
4054 uint64_t offset = (1ULL << (ztest_random(20) + 43)) +
4055 (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
4056
4057 /*
4058 * Have multiple threads write to large offsets in an object
4059 * to verify that parallel writes to an object -- even to the
4060 * same blocks within the object -- doesn't cause any trouble.
4061 */
4062 ztest_od_init(&od[0], ID_PARALLEL, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
4063
4064 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4065 return;
4066
4067 while (ztest_random(10) != 0)
4068 ztest_io(zd, od[0].od_object, offset);
4069 }
4070
4071 void
4072 ztest_dmu_prealloc(ztest_ds_t *zd, uint64_t id)
4073 {
4074 ztest_od_t od[1];
4075 uint64_t offset = (1ULL << (ztest_random(4) + SPA_MAXBLOCKSHIFT)) +
4076 (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
4077 uint64_t count = ztest_random(20) + 1;
4078 uint64_t blocksize = ztest_random_blocksize();
4079 void *data;
4080
4081 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
4082
4083 if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4084 return;
4085
4086 if (ztest_truncate(zd, od[0].od_object, offset, count * blocksize) != 0)
4087 return;
4088
4089 ztest_prealloc(zd, od[0].od_object, offset, count * blocksize);
4090
4091 data = umem_zalloc(blocksize, UMEM_NOFAIL);
4092
4093 while (ztest_random(count) != 0) {
4094 uint64_t randoff = offset + (ztest_random(count) * blocksize);
4095 if (ztest_write(zd, od[0].od_object, randoff, blocksize,
4096 data) != 0)
4097 break;
4098 while (ztest_random(4) != 0)
4099 ztest_io(zd, od[0].od_object, randoff);
4100 }
4101
4102 umem_free(data, blocksize);
4103 }
4104
4105 /*
4106 * Verify that zap_{create,destroy,add,remove,update} work as expected.
4107 */
4108 #define ZTEST_ZAP_MIN_INTS 1
4109 #define ZTEST_ZAP_MAX_INTS 4
4110 #define ZTEST_ZAP_MAX_PROPS 1000
4111
4112 void
4113 ztest_zap(ztest_ds_t *zd, uint64_t id)
4114 {
4115 objset_t *os = zd->zd_os;
4116 ztest_od_t od[1];
4117 uint64_t object;
4118 uint64_t txg, last_txg;
4119 uint64_t value[ZTEST_ZAP_MAX_INTS];
4120 uint64_t zl_ints, zl_intsize, prop;
4121 int i, ints;
4122 dmu_tx_t *tx;
4123 char propname[100], txgname[100];
4124 int error;
4125 char *hc[2] = { "s.acl.h", ".s.open.h.hyLZlg" };
4126
4127 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0);
4128
4129 if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4130 return;
4131
4132 object = od[0].od_object;
4133
4134 /*
4135 * Generate a known hash collision, and verify that
4136 * we can lookup and remove both entries.
4137 */
4138 tx = dmu_tx_create(os);
4139 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4140 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4141 if (txg == 0)
4142 return;
4143 for (i = 0; i < 2; i++) {
4144 value[i] = i;
4145 VERIFY3U(0, ==, zap_add(os, object, hc[i], sizeof (uint64_t),
4146 1, &value[i], tx));
4147 }
4148 for (i = 0; i < 2; i++) {
4149 VERIFY3U(EEXIST, ==, zap_add(os, object, hc[i],
4150 sizeof (uint64_t), 1, &value[i], tx));
4151 VERIFY3U(0, ==,
4152 zap_length(os, object, hc[i], &zl_intsize, &zl_ints));
4153 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4154 ASSERT3U(zl_ints, ==, 1);
4155 }
4156 for (i = 0; i < 2; i++) {
4157 VERIFY3U(0, ==, zap_remove(os, object, hc[i], tx));
4158 }
4159 dmu_tx_commit(tx);
4160
4161 /*
4162 * Generate a buch of random entries.
4163 */
4164 ints = MAX(ZTEST_ZAP_MIN_INTS, object % ZTEST_ZAP_MAX_INTS);
4165
4166 prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
4167 (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
4168 (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
4169 bzero(value, sizeof (value));
4170 last_txg = 0;
4171
4172 /*
4173 * If these zap entries already exist, validate their contents.
4174 */
4175 error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
4176 if (error == 0) {
4177 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4178 ASSERT3U(zl_ints, ==, 1);
4179
4180 VERIFY(zap_lookup(os, object, txgname, zl_intsize,
4181 zl_ints, &last_txg) == 0);
4182
4183 VERIFY(zap_length(os, object, propname, &zl_intsize,
4184 &zl_ints) == 0);
4185
4186 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4187 ASSERT3U(zl_ints, ==, ints);
4188
4189 VERIFY(zap_lookup(os, object, propname, zl_intsize,
4190 zl_ints, value) == 0);
4191
4192 for (i = 0; i < ints; i++) {
4193 ASSERT3U(value[i], ==, last_txg + object + i);
4194 }
4195 } else {
4196 ASSERT3U(error, ==, ENOENT);
4197 }
4198
4199 /*
4200 * Atomically update two entries in our zap object.
4201 * The first is named txg_%llu, and contains the txg
4202 * in which the property was last updated. The second
4203 * is named prop_%llu, and the nth element of its value
4204 * should be txg + object + n.
4205 */
4206 tx = dmu_tx_create(os);
4207 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4208 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4209 if (txg == 0)
4210 return;
4211
4212 if (last_txg > txg)
4213 fatal(0, "zap future leak: old %llu new %llu", last_txg, txg);
4214
4215 for (i = 0; i < ints; i++)
4216 value[i] = txg + object + i;
4217
4218 VERIFY3U(0, ==, zap_update(os, object, txgname, sizeof (uint64_t),
4219 1, &txg, tx));
4220 VERIFY3U(0, ==, zap_update(os, object, propname, sizeof (uint64_t),
4221 ints, value, tx));
4222
4223 dmu_tx_commit(tx);
4224
4225 /*
4226 * Remove a random pair of entries.
4227 */
4228 prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
4229 (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
4230 (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
4231
4232 error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
4233
4234 if (error == ENOENT)
4235 return;
4236
4237 ASSERT0(error);
4238
4239 tx = dmu_tx_create(os);
4240 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4241 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4242 if (txg == 0)
4243 return;
4244 VERIFY3U(0, ==, zap_remove(os, object, txgname, tx));
4245 VERIFY3U(0, ==, zap_remove(os, object, propname, tx));
4246 dmu_tx_commit(tx);
4247 }
4248
4249 /*
4250 * Testcase to test the upgrading of a microzap to fatzap.
4251 */
4252 void
4253 ztest_fzap(ztest_ds_t *zd, uint64_t id)
4254 {
4255 objset_t *os = zd->zd_os;
4256 ztest_od_t od[1];
4257 uint64_t object, txg;
4258
4259 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0);
4260
4261 if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4262 return;
4263
4264 object = od[0].od_object;
4265
4266 /*
4267 * Add entries to this ZAP and make sure it spills over
4268 * and gets upgraded to a fatzap. Also, since we are adding
4269 * 2050 entries we should see ptrtbl growth and leaf-block split.
4270 */
4271 for (int i = 0; i < 2050; i++) {
4272 char name[ZFS_MAX_DATASET_NAME_LEN];
4273 uint64_t value = i;
4274 dmu_tx_t *tx;
4275 int error;
4276
4277 (void) snprintf(name, sizeof (name), "fzap-%llu-%llu",
4278 id, value);
4279
4280 tx = dmu_tx_create(os);
4281 dmu_tx_hold_zap(tx, object, B_TRUE, name);
4282 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4283 if (txg == 0)
4284 return;
4285 error = zap_add(os, object, name, sizeof (uint64_t), 1,
4286 &value, tx);
4287 ASSERT(error == 0 || error == EEXIST);
4288 dmu_tx_commit(tx);
4289 }
4290 }
4291
4292 /* ARGSUSED */
4293 void
4294 ztest_zap_parallel(ztest_ds_t *zd, uint64_t id)
4295 {
4296 objset_t *os = zd->zd_os;
4297 ztest_od_t od[1];
4298 uint64_t txg, object, count, wsize, wc, zl_wsize, zl_wc;
4299 dmu_tx_t *tx;
4300 int i, namelen, error;
4301 int micro = ztest_random(2);
4302 char name[20], string_value[20];
4303 void *data;
4304
4305 ztest_od_init(&od[0], ID_PARALLEL, FTAG, micro, DMU_OT_ZAP_OTHER, 0, 0);
4306
4307 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4308 return;
4309
4310 object = od[0].od_object;
4311
4312 /*
4313 * Generate a random name of the form 'xxx.....' where each
4314 * x is a random printable character and the dots are dots.
4315 * There are 94 such characters, and the name length goes from
4316 * 6 to 20, so there are 94^3 * 15 = 12,458,760 possible names.
4317 */
4318 namelen = ztest_random(sizeof (name) - 5) + 5 + 1;
4319
4320 for (i = 0; i < 3; i++)
4321 name[i] = '!' + ztest_random('~' - '!' + 1);
4322 for (; i < namelen - 1; i++)
4323 name[i] = '.';
4324 name[i] = '\0';
4325
4326 if ((namelen & 1) || micro) {
4327 wsize = sizeof (txg);
4328 wc = 1;
4329 data = &txg;
4330 } else {
4331 wsize = 1;
4332 wc = namelen;
4333 data = string_value;
4334 }
4335
4336 count = -1ULL;
4337 VERIFY0(zap_count(os, object, &count));
4338 ASSERT(count != -1ULL);
4339
4340 /*
4341 * Select an operation: length, lookup, add, update, remove.
4342 */
4343 i = ztest_random(5);
4344
4345 if (i >= 2) {
4346 tx = dmu_tx_create(os);
4347 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4348 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4349 if (txg == 0)
4350 return;
4351 bcopy(name, string_value, namelen);
4352 } else {
4353 tx = NULL;
4354 txg = 0;
4355 bzero(string_value, namelen);
4356 }
4357
4358 switch (i) {
4359
4360 case 0:
4361 error = zap_length(os, object, name, &zl_wsize, &zl_wc);
4362 if (error == 0) {
4363 ASSERT3U(wsize, ==, zl_wsize);
4364 ASSERT3U(wc, ==, zl_wc);
4365 } else {
4366 ASSERT3U(error, ==, ENOENT);
4367 }
4368 break;
4369
4370 case 1:
4371 error = zap_lookup(os, object, name, wsize, wc, data);
4372 if (error == 0) {
4373 if (data == string_value &&
4374 bcmp(name, data, namelen) != 0)
4375 fatal(0, "name '%s' != val '%s' len %d",
4376 name, data, namelen);
4377 } else {
4378 ASSERT3U(error, ==, ENOENT);
4379 }
4380 break;
4381
4382 case 2:
4383 error = zap_add(os, object, name, wsize, wc, data, tx);
4384 ASSERT(error == 0 || error == EEXIST);
4385 break;
4386
4387 case 3:
4388 VERIFY(zap_update(os, object, name, wsize, wc, data, tx) == 0);
4389 break;
4390
4391 case 4:
4392 error = zap_remove(os, object, name, tx);
4393 ASSERT(error == 0 || error == ENOENT);
4394 break;
4395 }
4396
4397 if (tx != NULL)
4398 dmu_tx_commit(tx);
4399 }
4400
4401 /*
4402 * Commit callback data.
4403 */
4404 typedef struct ztest_cb_data {
4405 list_node_t zcd_node;
4406 uint64_t zcd_txg;
4407 int zcd_expected_err;
4408 boolean_t zcd_added;
4409 boolean_t zcd_called;
4410 spa_t *zcd_spa;
4411 } ztest_cb_data_t;
4412
4413 /* This is the actual commit callback function */
4414 static void
4415 ztest_commit_callback(void *arg, int error)
4416 {
4417 ztest_cb_data_t *data = arg;
4418 uint64_t synced_txg;
4419
4420 VERIFY(data != NULL);
4421 VERIFY3S(data->zcd_expected_err, ==, error);
4422 VERIFY(!data->zcd_called);
4423
4424 synced_txg = spa_last_synced_txg(data->zcd_spa);
4425 if (data->zcd_txg > synced_txg)
4426 fatal(0, "commit callback of txg %" PRIu64 " called prematurely"
4427 ", last synced txg = %" PRIu64 "\n", data->zcd_txg,
4428 synced_txg);
4429
4430 data->zcd_called = B_TRUE;
4431
4432 if (error == ECANCELED) {
4433 ASSERT0(data->zcd_txg);
4434 ASSERT(!data->zcd_added);
4435
4436 /*
4437 * The private callback data should be destroyed here, but
4438 * since we are going to check the zcd_called field after
4439 * dmu_tx_abort(), we will destroy it there.
4440 */
4441 return;
4442 }
4443
4444 /* Was this callback added to the global callback list? */
4445 if (!data->zcd_added)
4446 goto out;
4447
4448 ASSERT3U(data->zcd_txg, !=, 0);
4449
4450 /* Remove our callback from the list */
4451 (void) mutex_lock(&zcl.zcl_callbacks_lock);
4452 list_remove(&zcl.zcl_callbacks, data);
4453 (void) mutex_unlock(&zcl.zcl_callbacks_lock);
4454
4455 out:
4456 umem_free(data, sizeof (ztest_cb_data_t));
4457 }
4458
4459 /* Allocate and initialize callback data structure */
4460 static ztest_cb_data_t *
4461 ztest_create_cb_data(objset_t *os, uint64_t txg)
4462 {
4463 ztest_cb_data_t *cb_data;
4464
4465 cb_data = umem_zalloc(sizeof (ztest_cb_data_t), UMEM_NOFAIL);
4466
4467 cb_data->zcd_txg = txg;
4468 cb_data->zcd_spa = dmu_objset_spa(os);
4469
4470 return (cb_data);
4471 }
4472
4473 /*
4474 * If a number of txgs equal to this threshold have been created after a commit
4475 * callback has been registered but not called, then we assume there is an
4476 * implementation bug.
4477 */
4478 #define ZTEST_COMMIT_CALLBACK_THRESH (TXG_CONCURRENT_STATES + 2)
4479
4480 /*
4481 * Commit callback test.
4482 */
4483 void
4484 ztest_dmu_commit_callbacks(ztest_ds_t *zd, uint64_t id)
4485 {
4486 objset_t *os = zd->zd_os;
4487 ztest_od_t od[1];
4488 dmu_tx_t *tx;
4489 ztest_cb_data_t *cb_data[3], *tmp_cb;
4490 uint64_t old_txg, txg;
4491 int i, error;
4492
4493 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
4494
4495 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4496 return;
4497
4498 tx = dmu_tx_create(os);
4499
4500 cb_data[0] = ztest_create_cb_data(os, 0);
4501 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[0]);
4502
4503 dmu_tx_hold_write(tx, od[0].od_object, 0, sizeof (uint64_t));
4504
4505 /* Every once in a while, abort the transaction on purpose */
4506 if (ztest_random(100) == 0)
4507 error = -1;
4508
4509 if (!error)
4510 error = dmu_tx_assign(tx, TXG_NOWAIT);
4511
4512 txg = error ? 0 : dmu_tx_get_txg(tx);
4513
4514 cb_data[0]->zcd_txg = txg;
4515 cb_data[1] = ztest_create_cb_data(os, txg);
4516 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[1]);
4517
4518 if (error) {
4519 /*
4520 * It's not a strict requirement to call the registered
4521 * callbacks from inside dmu_tx_abort(), but that's what
4522 * it's supposed to happen in the current implementation
4523 * so we will check for that.
4524 */
4525 for (i = 0; i < 2; i++) {
4526 cb_data[i]->zcd_expected_err = ECANCELED;
4527 VERIFY(!cb_data[i]->zcd_called);
4528 }
4529
4530 dmu_tx_abort(tx);
4531
4532 for (i = 0; i < 2; i++) {
4533 VERIFY(cb_data[i]->zcd_called);
4534 umem_free(cb_data[i], sizeof (ztest_cb_data_t));
4535 }
4536
4537 return;
4538 }
4539
4540 cb_data[2] = ztest_create_cb_data(os, txg);
4541 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[2]);
4542
4543 /*
4544 * Read existing data to make sure there isn't a future leak.
4545 */
4546 VERIFY(0 == dmu_read(os, od[0].od_object, 0, sizeof (uint64_t),
4547 &old_txg, DMU_READ_PREFETCH));
4548
4549 if (old_txg > txg)
4550 fatal(0, "future leak: got %" PRIu64 ", open txg is %" PRIu64,
4551 old_txg, txg);
4552
4553 dmu_write(os, od[0].od_object, 0, sizeof (uint64_t), &txg, tx);
4554
4555 (void) mutex_lock(&zcl.zcl_callbacks_lock);
4556
4557 /*
4558 * Since commit callbacks don't have any ordering requirement and since
4559 * it is theoretically possible for a commit callback to be called
4560 * after an arbitrary amount of time has elapsed since its txg has been
4561 * synced, it is difficult to reliably determine whether a commit
4562 * callback hasn't been called due to high load or due to a flawed
4563 * implementation.
4564 *
4565 * In practice, we will assume that if after a certain number of txgs a
4566 * commit callback hasn't been called, then most likely there's an
4567 * implementation bug..
4568 */
4569 tmp_cb = list_head(&zcl.zcl_callbacks);
4570 if (tmp_cb != NULL &&
4571 (txg - ZTEST_COMMIT_CALLBACK_THRESH) > tmp_cb->zcd_txg) {
4572 fatal(0, "Commit callback threshold exceeded, oldest txg: %"
4573 PRIu64 ", open txg: %" PRIu64 "\n", tmp_cb->zcd_txg, txg);
4574 }
4575
4576 /*
4577 * Let's find the place to insert our callbacks.
4578 *
4579 * Even though the list is ordered by txg, it is possible for the
4580 * insertion point to not be the end because our txg may already be
4581 * quiescing at this point and other callbacks in the open txg
4582 * (from other objsets) may have sneaked in.
4583 */
4584 tmp_cb = list_tail(&zcl.zcl_callbacks);
4585 while (tmp_cb != NULL && tmp_cb->zcd_txg > txg)
4586 tmp_cb = list_prev(&zcl.zcl_callbacks, tmp_cb);
4587
4588 /* Add the 3 callbacks to the list */
4589 for (i = 0; i < 3; i++) {
4590 if (tmp_cb == NULL)
4591 list_insert_head(&zcl.zcl_callbacks, cb_data[i]);
4592 else
4593 list_insert_after(&zcl.zcl_callbacks, tmp_cb,
4594 cb_data[i]);
4595
4596 cb_data[i]->zcd_added = B_TRUE;
4597 VERIFY(!cb_data[i]->zcd_called);
4598
4599 tmp_cb = cb_data[i];
4600 }
4601
4602 (void) mutex_unlock(&zcl.zcl_callbacks_lock);
4603
4604 dmu_tx_commit(tx);
4605 }
4606
4607 /* ARGSUSED */
4608 void
4609 ztest_dsl_prop_get_set(ztest_ds_t *zd, uint64_t id)
4610 {
4611 zfs_prop_t proplist[] = {
4612 ZFS_PROP_CHECKSUM,
4613 ZFS_PROP_COMPRESSION,
4614 ZFS_PROP_COPIES,
4615 ZFS_PROP_DEDUP
4616 };
4617
4618 (void) rw_rdlock(&ztest_name_lock);
4619
4620 for (int p = 0; p < sizeof (proplist) / sizeof (proplist[0]); p++)
4621 (void) ztest_dsl_prop_set_uint64(zd->zd_name, proplist[p],
4622 ztest_random_dsl_prop(proplist[p]), (int)ztest_random(2));
4623
4624 (void) rw_unlock(&ztest_name_lock);
4625 }
4626
4627 /* ARGSUSED */
4628 void
4629 ztest_spa_prop_get_set(ztest_ds_t *zd, uint64_t id)
4630 {
4631 nvlist_t *props = NULL;
4632
4633 (void) rw_rdlock(&ztest_name_lock);
4634
4635 (void) ztest_spa_prop_set_uint64(ZPOOL_PROP_DEDUPDITTO,
4636 ZIO_DEDUPDITTO_MIN + ztest_random(ZIO_DEDUPDITTO_MIN));
4637
4638 VERIFY0(spa_prop_get(ztest_spa, &props));
4639
4640 if (ztest_opts.zo_verbose >= 6)
4641 dump_nvlist(props, 4);
4642
4643 nvlist_free(props);
4644
4645 (void) rw_unlock(&ztest_name_lock);
4646 }
4647
4648 static int
4649 user_release_one(const char *snapname, const char *holdname)
4650 {
4651 nvlist_t *snaps, *holds;
4652 int error;
4653
4654 snaps = fnvlist_alloc();
4655 holds = fnvlist_alloc();
4656 fnvlist_add_boolean(holds, holdname);
4657 fnvlist_add_nvlist(snaps, snapname, holds);
4658 fnvlist_free(holds);
4659 error = dsl_dataset_user_release(snaps, NULL);
4660 fnvlist_free(snaps);
4661 return (error);
4662 }
4663
4664 /*
4665 * Test snapshot hold/release and deferred destroy.
4666 */
4667 void
4668 ztest_dmu_snapshot_hold(ztest_ds_t *zd, uint64_t id)
4669 {
4670 int error;
4671 objset_t *os = zd->zd_os;
4672 objset_t *origin;
4673 char snapname[100];
4674 char fullname[100];
4675 char clonename[100];
4676 char tag[100];
4677 char osname[ZFS_MAX_DATASET_NAME_LEN];
4678 nvlist_t *holds;
4679
4680 (void) rw_rdlock(&ztest_name_lock);
4681
4682 dmu_objset_name(os, osname);
4683
4684 (void) snprintf(snapname, sizeof (snapname), "sh1_%llu", id);
4685 (void) snprintf(fullname, sizeof (fullname), "%s@%s", osname, snapname);
4686 (void) snprintf(clonename, sizeof (clonename),
4687 "%s/ch1_%llu", osname, id);
4688 (void) snprintf(tag, sizeof (tag), "tag_%llu", id);
4689
4690 /*
4691 * Clean up from any previous run.
4692 */
4693 error = dsl_destroy_head(clonename);
4694 if (error != ENOENT)
4695 ASSERT0(error);
4696 error = user_release_one(fullname, tag);
4697 if (error != ESRCH && error != ENOENT)
4698 ASSERT0(error);
4699 error = dsl_destroy_snapshot(fullname, B_FALSE);
4700 if (error != ENOENT)
4701 ASSERT0(error);
4702
4703 /*
4704 * Create snapshot, clone it, mark snap for deferred destroy,
4705 * destroy clone, verify snap was also destroyed.
4706 */
4707 error = dmu_objset_snapshot_one(osname, snapname);
4708 if (error) {
4709 if (error == ENOSPC) {
4710 ztest_record_enospc("dmu_objset_snapshot");
4711 goto out;
4712 }
4713 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
4714 }
4715
4716 error = dmu_objset_clone(clonename, fullname);
4717 if (error) {
4718 if (error == ENOSPC) {
4719 ztest_record_enospc("dmu_objset_clone");
4720 goto out;
4721 }
4722 fatal(0, "dmu_objset_clone(%s) = %d", clonename, error);
4723 }
4724
4725 error = dsl_destroy_snapshot(fullname, B_TRUE);
4726 if (error) {
4727 fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
4728 fullname, error);
4729 }
4730
4731 error = dsl_destroy_head(clonename);
4732 if (error)
4733 fatal(0, "dsl_destroy_head(%s) = %d", clonename, error);
4734
4735 error = dmu_objset_hold(fullname, FTAG, &origin);
4736 if (error != ENOENT)
4737 fatal(0, "dmu_objset_hold(%s) = %d", fullname, error);
4738
4739 /*
4740 * Create snapshot, add temporary hold, verify that we can't
4741 * destroy a held snapshot, mark for deferred destroy,
4742 * release hold, verify snapshot was destroyed.
4743 */
4744 error = dmu_objset_snapshot_one(osname, snapname);
4745 if (error) {
4746 if (error == ENOSPC) {
4747 ztest_record_enospc("dmu_objset_snapshot");
4748 goto out;
4749 }
4750 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
4751 }
4752
4753 holds = fnvlist_alloc();
4754 fnvlist_add_string(holds, fullname, tag);
4755 error = dsl_dataset_user_hold(holds, 0, NULL);
4756 fnvlist_free(holds);
4757
4758 if (error == ENOSPC) {
4759 ztest_record_enospc("dsl_dataset_user_hold");
4760 goto out;
4761 } else if (error) {
4762 fatal(0, "dsl_dataset_user_hold(%s, %s) = %u",
4763 fullname, tag, error);
4764 }
4765
4766 error = dsl_destroy_snapshot(fullname, B_FALSE);
4767 if (error != EBUSY) {
4768 fatal(0, "dsl_destroy_snapshot(%s, B_FALSE) = %d",
4769 fullname, error);
4770 }
4771
4772 error = dsl_destroy_snapshot(fullname, B_TRUE);
4773 if (error) {
4774 fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
4775 fullname, error);
4776 }
4777
4778 error = user_release_one(fullname, tag);
4779 if (error)
4780 fatal(0, "user_release_one(%s, %s) = %d", fullname, tag, error);
4781
4782 VERIFY3U(dmu_objset_hold(fullname, FTAG, &origin), ==, ENOENT);
4783
4784 out:
4785 (void) rw_unlock(&ztest_name_lock);
4786 }
4787
4788 /*
4789 * Inject random faults into the on-disk data.
4790 */
4791 /* ARGSUSED */
4792 void
4793 ztest_fault_inject(ztest_ds_t *zd, uint64_t id)
4794 {
4795 ztest_shared_t *zs = ztest_shared;
4796 spa_t *spa = ztest_spa;
4797 int fd;
4798 uint64_t offset;
4799 uint64_t leaves;
4800 uint64_t bad = 0x1990c0ffeedecade;
4801 uint64_t top, leaf;
4802 char path0[MAXPATHLEN];
4803 char pathrand[MAXPATHLEN];
4804 size_t fsize;
4805 int bshift = SPA_MAXBLOCKSHIFT + 2;
4806 int iters = 1000;
4807 int maxfaults;
4808 int mirror_save;
4809 vdev_t *vd0 = NULL;
4810 uint64_t guid0 = 0;
4811 boolean_t islog = B_FALSE;
4812
4813 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
4814 maxfaults = MAXFAULTS();
4815 leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz;
4816 mirror_save = zs->zs_mirrors;
4817 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4818
4819 ASSERT(leaves >= 1);
4820
4821 /*
4822 * Grab the name lock as reader. There are some operations
4823 * which don't like to have their vdevs changed while
4824 * they are in progress (i.e. spa_change_guid). Those
4825 * operations will have grabbed the name lock as writer.
4826 */
4827 (void) rw_rdlock(&ztest_name_lock);
4828
4829 /*
4830 * We need SCL_STATE here because we're going to look at vd0->vdev_tsd.
4831 */
4832 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
4833
4834 if (ztest_random(2) == 0) {
4835 /*
4836 * Inject errors on a normal data device or slog device.
4837 */
4838 top = ztest_random_vdev_top(spa, B_TRUE);
4839 leaf = ztest_random(leaves) + zs->zs_splits;
4840
4841 /*
4842 * Generate paths to the first leaf in this top-level vdev,
4843 * and to the random leaf we selected. We'll induce transient
4844 * write failures and random online/offline activity on leaf 0,
4845 * and we'll write random garbage to the randomly chosen leaf.
4846 */
4847 (void) snprintf(path0, sizeof (path0), ztest_dev_template,
4848 ztest_opts.zo_dir, ztest_opts.zo_pool,
4849 top * leaves + zs->zs_splits);
4850 (void) snprintf(pathrand, sizeof (pathrand), ztest_dev_template,
4851 ztest_opts.zo_dir, ztest_opts.zo_pool,
4852 top * leaves + leaf);
4853
4854 vd0 = vdev_lookup_by_path(spa->spa_root_vdev, path0);
4855 if (vd0 != NULL && vd0->vdev_top->vdev_islog)
4856 islog = B_TRUE;
4857
4858 /*
4859 * If the top-level vdev needs to be resilvered
4860 * then we only allow faults on the device that is
4861 * resilvering.
4862 */
4863 if (vd0 != NULL && maxfaults != 1 &&
4864 (!vdev_resilver_needed(vd0->vdev_top, NULL, NULL) ||
4865 vd0->vdev_resilver_txg != 0)) {
4866 /*
4867 * Make vd0 explicitly claim to be unreadable,
4868 * or unwriteable, or reach behind its back
4869 * and close the underlying fd. We can do this if
4870 * maxfaults == 0 because we'll fail and reexecute,
4871 * and we can do it if maxfaults >= 2 because we'll
4872 * have enough redundancy. If maxfaults == 1, the
4873 * combination of this with injection of random data
4874 * corruption below exceeds the pool's fault tolerance.
4875 */
4876 vdev_file_t *vf = vd0->vdev_tsd;
4877
4878 if (vf != NULL && ztest_random(3) == 0) {
4879 (void) close(vf->vf_vnode->v_fd);
4880 vf->vf_vnode->v_fd = -1;
4881 } else if (ztest_random(2) == 0) {
4882 vd0->vdev_cant_read = B_TRUE;
4883 } else {
4884 vd0->vdev_cant_write = B_TRUE;
4885 }
4886 guid0 = vd0->vdev_guid;
4887 }
4888 } else {
4889 /*
4890 * Inject errors on an l2cache device.
4891 */
4892 spa_aux_vdev_t *sav = &spa->spa_l2cache;
4893
4894 if (sav->sav_count == 0) {
4895 spa_config_exit(spa, SCL_STATE, FTAG);
4896 (void) rw_unlock(&ztest_name_lock);
4897 return;
4898 }
4899 vd0 = sav->sav_vdevs[ztest_random(sav->sav_count)];
4900 guid0 = vd0->vdev_guid;
4901 (void) strcpy(path0, vd0->vdev_path);
4902 (void) strcpy(pathrand, vd0->vdev_path);
4903
4904 leaf = 0;
4905 leaves = 1;
4906 maxfaults = INT_MAX; /* no limit on cache devices */
4907 }
4908
4909 spa_config_exit(spa, SCL_STATE, FTAG);
4910 (void) rw_unlock(&ztest_name_lock);
4911
4912 /*
4913 * If we can tolerate two or more faults, or we're dealing
4914 * with a slog, randomly online/offline vd0.
4915 */
4916 if ((maxfaults >= 2 || islog) && guid0 != 0) {
4917 if (ztest_random(10) < 6) {
4918 int flags = (ztest_random(2) == 0 ?
4919 ZFS_OFFLINE_TEMPORARY : 0);
4920
4921 /*
4922 * We have to grab the zs_name_lock as writer to
4923 * prevent a race between offlining a slog and
4924 * destroying a dataset. Offlining the slog will
4925 * grab a reference on the dataset which may cause
4926 * dmu_objset_destroy() to fail with EBUSY thus
4927 * leaving the dataset in an inconsistent state.
4928 */
4929 if (islog)
4930 (void) rw_wrlock(&ztest_name_lock);
4931
4932 VERIFY(vdev_offline(spa, guid0, flags) != EBUSY);
4933
4934 if (islog)
4935 (void) rw_unlock(&ztest_name_lock);
4936 } else {
4937 /*
4938 * Ideally we would like to be able to randomly
4939 * call vdev_[on|off]line without holding locks
4940 * to force unpredictable failures but the side
4941 * effects of vdev_[on|off]line prevent us from
4942 * doing so. We grab the ztest_vdev_lock here to
4943 * prevent a race between injection testing and
4944 * aux_vdev removal.
4945 */
4946 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
4947 (void) vdev_online(spa, guid0, 0, NULL);
4948 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4949 }
4950 }
4951
4952 if (maxfaults == 0)
4953 return;
4954
4955 /*
4956 * We have at least single-fault tolerance, so inject data corruption.
4957 */
4958 fd = open(pathrand, O_RDWR);
4959
4960 if (fd == -1) /* we hit a gap in the device namespace */
4961 return;
4962
4963 fsize = lseek(fd, 0, SEEK_END);
4964
4965 while (--iters != 0) {
4966 /*
4967 * The offset must be chosen carefully to ensure that
4968 * we do not inject a given logical block with errors
4969 * on two different leaf devices, because ZFS can not
4970 * tolerate that (if maxfaults==1).
4971 *
4972 * We divide each leaf into chunks of size
4973 * (# leaves * SPA_MAXBLOCKSIZE * 4). Within each chunk
4974 * there is a series of ranges to which we can inject errors.
4975 * Each range can accept errors on only a single leaf vdev.
4976 * The error injection ranges are separated by ranges
4977 * which we will not inject errors on any device (DMZs).
4978 * Each DMZ must be large enough such that a single block
4979 * can not straddle it, so that a single block can not be
4980 * a target in two different injection ranges (on different
4981 * leaf vdevs).
4982 *
4983 * For example, with 3 leaves, each chunk looks like:
4984 * 0 to 32M: injection range for leaf 0
4985 * 32M to 64M: DMZ - no injection allowed
4986 * 64M to 96M: injection range for leaf 1
4987 * 96M to 128M: DMZ - no injection allowed
4988 * 128M to 160M: injection range for leaf 2
4989 * 160M to 192M: DMZ - no injection allowed
4990 */
4991 offset = ztest_random(fsize / (leaves << bshift)) *
4992 (leaves << bshift) + (leaf << bshift) +
4993 (ztest_random(1ULL << (bshift - 1)) & -8ULL);
4994
4995 /*
4996 * Only allow damage to the labels at one end of the vdev.
4997 *
4998 * If all labels are damaged, the device will be totally
4999 * inaccessible, which will result in loss of data,
5000 * because we also damage (parts of) the other side of
5001 * the mirror/raidz.
5002 *
5003 * Additionally, we will always have both an even and an
5004 * odd label, so that we can handle crashes in the
5005 * middle of vdev_config_sync().
5006 */
5007 if ((leaf & 1) == 0 && offset < VDEV_LABEL_START_SIZE)
5008 continue;
5009
5010 /*
5011 * The two end labels are stored at the "end" of the disk, but
5012 * the end of the disk (vdev_psize) is aligned to
5013 * sizeof (vdev_label_t).
5014 */
5015 uint64_t psize = P2ALIGN(fsize, sizeof (vdev_label_t));
5016 if ((leaf & 1) == 1 &&
5017 offset + sizeof (bad) > psize - VDEV_LABEL_END_SIZE)
5018 continue;
5019
5020 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
5021 if (mirror_save != zs->zs_mirrors) {
5022 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
5023 (void) close(fd);
5024 return;
5025 }
5026
5027 if (pwrite(fd, &bad, sizeof (bad), offset) != sizeof (bad))
5028 fatal(1, "can't inject bad word at 0x%llx in %s",
5029 offset, pathrand);
5030
5031 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
5032
5033 if (ztest_opts.zo_verbose >= 7)
5034 (void) printf("injected bad word into %s,"
5035 " offset 0x%llx\n", pathrand, (u_longlong_t)offset);
5036 }
5037
5038 (void) close(fd);
5039 }
5040
5041 /*
5042 * Verify that DDT repair works as expected.
5043 */
5044 void
5045 ztest_ddt_repair(ztest_ds_t *zd, uint64_t id)
5046 {
5047 ztest_shared_t *zs = ztest_shared;
5048 spa_t *spa = ztest_spa;
5049 objset_t *os = zd->zd_os;
5050 ztest_od_t od[1];
5051 uint64_t object, blocksize, txg, pattern, psize;
5052 enum zio_checksum checksum = spa_dedup_checksum(spa);
5053 dmu_buf_t *db;
5054 dmu_tx_t *tx;
5055 abd_t *abd;
5056 blkptr_t blk;
5057 int copies = 2 * ZIO_DEDUPDITTO_MIN;
5058
5059 blocksize = ztest_random_blocksize();
5060 blocksize = MIN(blocksize, 2048); /* because we write so many */
5061
5062 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
5063
5064 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
5065 return;
5066
5067 /*
5068 * Take the name lock as writer to prevent anyone else from changing
5069 * the pool and dataset properies we need to maintain during this test.
5070 */
5071 (void) rw_wrlock(&ztest_name_lock);
5072
5073 if (ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_DEDUP, checksum,
5074 B_FALSE) != 0 ||
5075 ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_COPIES, 1,
5076 B_FALSE) != 0) {
5077 (void) rw_unlock(&ztest_name_lock);
5078 return;
5079 }
5080
5081 dmu_objset_stats_t dds;
5082 dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
5083 dmu_objset_fast_stat(os, &dds);
5084 dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
5085
5086 object = od[0].od_object;
5087 blocksize = od[0].od_blocksize;
5088 pattern = zs->zs_guid ^ dds.dds_guid;
5089
5090 ASSERT(object != 0);
5091
5092 tx = dmu_tx_create(os);
5093 dmu_tx_hold_write(tx, object, 0, copies * blocksize);
5094 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
5095 if (txg == 0) {
5096 (void) rw_unlock(&ztest_name_lock);
5097 return;
5098 }
5099
5100 /*
5101 * Write all the copies of our block.
5102 */
5103 for (int i = 0; i < copies; i++) {
5104 uint64_t offset = i * blocksize;
5105 int error = dmu_buf_hold(os, object, offset, FTAG, &db,
5106 DMU_READ_NO_PREFETCH);
5107 if (error != 0) {
5108 fatal(B_FALSE, "dmu_buf_hold(%p, %llu, %llu) = %u",
5109 os, (long long)object, (long long) offset, error);
5110 }
5111 ASSERT(db->db_offset == offset);
5112 ASSERT(db->db_size == blocksize);
5113 ASSERT(ztest_pattern_match(db->db_data, db->db_size, pattern) ||
5114 ztest_pattern_match(db->db_data, db->db_size, 0ULL));
5115 dmu_buf_will_fill(db, tx);
5116 ztest_pattern_set(db->db_data, db->db_size, pattern);
5117 dmu_buf_rele(db, FTAG);
5118 }
5119
5120 dmu_tx_commit(tx);
5121 txg_wait_synced(spa_get_dsl(spa), txg);
5122
5123 /*
5124 * Find out what block we got.
5125 */
5126 VERIFY0(dmu_buf_hold(os, object, 0, FTAG, &db,
5127 DMU_READ_NO_PREFETCH));
5128 blk = *((dmu_buf_impl_t *)db)->db_blkptr;
5129 dmu_buf_rele(db, FTAG);
5130
5131 /*
5132 * Damage the block. Dedup-ditto will save us when we read it later.
5133 */
5134 psize = BP_GET_PSIZE(&blk);
5135 abd = abd_alloc_linear(psize, B_TRUE);
5136 ztest_pattern_set(abd_to_buf(abd), psize, ~pattern);
5137
5138 (void) zio_wait(zio_rewrite(NULL, spa, 0, &blk,
5139 abd, psize, NULL, NULL, ZIO_PRIORITY_SYNC_WRITE,
5140 ZIO_FLAG_CANFAIL | ZIO_FLAG_INDUCE_DAMAGE, NULL));
5141
5142 abd_free(abd);
5143
5144 (void) rw_unlock(&ztest_name_lock);
5145 }
5146
5147 /*
5148 * Scrub the pool.
5149 */
5150 /* ARGSUSED */
5151 void
5152 ztest_scrub(ztest_ds_t *zd, uint64_t id)
5153 {
5154 spa_t *spa = ztest_spa;
5155
5156 (void) spa_scan(spa, POOL_SCAN_SCRUB);
5157 (void) poll(NULL, 0, 100); /* wait a moment, then force a restart */
5158 (void) spa_scan(spa, POOL_SCAN_SCRUB);
5159 }
5160
5161 /*
5162 * Change the guid for the pool.
5163 */
5164 /* ARGSUSED */
5165 void
5166 ztest_reguid(ztest_ds_t *zd, uint64_t id)
5167 {
5168 spa_t *spa = ztest_spa;
5169 uint64_t orig, load;
5170 int error;
5171
5172 orig = spa_guid(spa);
5173 load = spa_load_guid(spa);
5174
5175 (void) rw_wrlock(&ztest_name_lock);
5176 error = spa_change_guid(spa);
5177 (void) rw_unlock(&ztest_name_lock);
5178
5179 if (error != 0)
5180 return;
5181
5182 if (ztest_opts.zo_verbose >= 4) {
5183 (void) printf("Changed guid old %llu -> %llu\n",
5184 (u_longlong_t)orig, (u_longlong_t)spa_guid(spa));
5185 }
5186
5187 VERIFY3U(orig, !=, spa_guid(spa));
5188 VERIFY3U(load, ==, spa_load_guid(spa));
5189 }
5190
5191 /*
5192 * Rename the pool to a different name and then rename it back.
5193 */
5194 /* ARGSUSED */
5195 void
5196 ztest_spa_rename(ztest_ds_t *zd, uint64_t id)
5197 {
5198 char *oldname, *newname;
5199 spa_t *spa;
5200
5201 (void) rw_wrlock(&ztest_name_lock);
5202
5203 oldname = ztest_opts.zo_pool;
5204 newname = umem_alloc(strlen(oldname) + 5, UMEM_NOFAIL);
5205 (void) strcpy(newname, oldname);
5206 (void) strcat(newname, "_tmp");
5207
5208 /*
5209 * Do the rename
5210 */
5211 VERIFY3U(0, ==, spa_rename(oldname, newname));
5212
5213 /*
5214 * Try to open it under the old name, which shouldn't exist
5215 */
5216 VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
5217
5218 /*
5219 * Open it under the new name and make sure it's still the same spa_t.
5220 */
5221 VERIFY3U(0, ==, spa_open(newname, &spa, FTAG));
5222
5223 ASSERT(spa == ztest_spa);
5224 spa_close(spa, FTAG);
5225
5226 /*
5227 * Rename it back to the original
5228 */
5229 VERIFY3U(0, ==, spa_rename(newname, oldname));
5230
5231 /*
5232 * Make sure it can still be opened
5233 */
5234 VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG));
5235
5236 ASSERT(spa == ztest_spa);
5237 spa_close(spa, FTAG);
5238
5239 umem_free(newname, strlen(newname) + 1);
5240
5241 (void) rw_unlock(&ztest_name_lock);
5242 }
5243
5244 /*
5245 * Verify pool integrity by running zdb.
5246 */
5247 static void
5248 ztest_run_zdb(char *pool)
5249 {
5250 int status;
5251 char zdb[MAXPATHLEN + MAXNAMELEN + 20];
5252 char zbuf[1024];
5253 char *bin;
5254 char *ztest;
5255 char *isa;
5256 int isalen;
5257 FILE *fp;
5258
5259 (void) realpath(getexecname(), zdb);
5260
5261 /* zdb lives in /usr/sbin, while ztest lives in /usr/bin */
5262 bin = strstr(zdb, "/usr/bin/");
5263 ztest = strstr(bin, "/ztest");
5264 isa = bin + 8;
5265 isalen = ztest - isa;
5266 isa = strdup(isa);
5267 /* LINTED */
5268 (void) sprintf(bin,
5269 "/usr/sbin%.*s/zdb -bcc%s%s -G -d -U %s %s",
5270 isalen,
5271 isa,
5272 ztest_opts.zo_verbose >= 3 ? "s" : "",
5273 ztest_opts.zo_verbose >= 4 ? "v" : "",
5274 spa_config_path,
5275 pool);
5276 free(isa);
5277
5278 if (ztest_opts.zo_verbose >= 5)
5279 (void) printf("Executing %s\n", strstr(zdb, "zdb "));
5280
5281 fp = popen(zdb, "r");
5282
5283 while (fgets(zbuf, sizeof (zbuf), fp) != NULL)
5284 if (ztest_opts.zo_verbose >= 3)
5285 (void) printf("%s", zbuf);
5286
5287 status = pclose(fp);
5288
5289 if (status == 0)
5290 return;
5291
5292 ztest_dump_core = 0;
5293 if (WIFEXITED(status))
5294 fatal(0, "'%s' exit code %d", zdb, WEXITSTATUS(status));
5295 else
5296 fatal(0, "'%s' died with signal %d", zdb, WTERMSIG(status));
5297 }
5298
5299 static void
5300 ztest_walk_pool_directory(char *header)
5301 {
5302 spa_t *spa = NULL;
5303
5304 if (ztest_opts.zo_verbose >= 6)
5305 (void) printf("%s\n", header);
5306
5307 mutex_enter(&spa_namespace_lock);
5308 while ((spa = spa_next(spa)) != NULL)
5309 if (ztest_opts.zo_verbose >= 6)
5310 (void) printf("\t%s\n", spa_name(spa));
5311 mutex_exit(&spa_namespace_lock);
5312 }
5313
5314 static void
5315 ztest_spa_import_export(char *oldname, char *newname)
5316 {
5317 nvlist_t *config, *newconfig;
5318 uint64_t pool_guid;
5319 spa_t *spa;
5320 int error;
5321
5322 if (ztest_opts.zo_verbose >= 4) {
5323 (void) printf("import/export: old = %s, new = %s\n",
5324 oldname, newname);
5325 }
5326
5327 /*
5328 * Clean up from previous runs.
5329 */
5330 (void) spa_destroy(newname);
5331
5332 /*
5333 * Get the pool's configuration and guid.
5334 */
5335 VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG));
5336
5337 /*
5338 * Kick off a scrub to tickle scrub/export races.
5339 */
5340 if (ztest_random(2) == 0)
5341 (void) spa_scan(spa, POOL_SCAN_SCRUB);
5342
5343 pool_guid = spa_guid(spa);
5344 spa_close(spa, FTAG);
5345
5346 ztest_walk_pool_directory("pools before export");
5347
5348 /*
5349 * Export it.
5350 */
5351 VERIFY3U(0, ==, spa_export(oldname, &config, B_FALSE, B_FALSE));
5352
5353 ztest_walk_pool_directory("pools after export");
5354
5355 /*
5356 * Try to import it.
5357 */
5358 newconfig = spa_tryimport(config);
5359 ASSERT(newconfig != NULL);
5360 nvlist_free(newconfig);
5361
5362 /*
5363 * Import it under the new name.
5364 */
5365 error = spa_import(newname, config, NULL, 0);
5366 if (error != 0) {
5367 dump_nvlist(config, 0);
5368 fatal(B_FALSE, "couldn't import pool %s as %s: error %u",
5369 oldname, newname, error);
5370 }
5371
5372 ztest_walk_pool_directory("pools after import");
5373
5374 /*
5375 * Try to import it again -- should fail with EEXIST.
5376 */
5377 VERIFY3U(EEXIST, ==, spa_import(newname, config, NULL, 0));
5378
5379 /*
5380 * Try to import it under a different name -- should fail with EEXIST.
5381 */
5382 VERIFY3U(EEXIST, ==, spa_import(oldname, config, NULL, 0));
5383
5384 /*
5385 * Verify that the pool is no longer visible under the old name.
5386 */
5387 VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
5388
5389 /*
5390 * Verify that we can open and close the pool using the new name.
5391 */
5392 VERIFY3U(0, ==, spa_open(newname, &spa, FTAG));
5393 ASSERT(pool_guid == spa_guid(spa));
5394 spa_close(spa, FTAG);
5395
5396 nvlist_free(config);
5397 }
5398
5399 static void
5400 ztest_resume(spa_t *spa)
5401 {
5402 if (spa_suspended(spa) && ztest_opts.zo_verbose >= 6)
5403 (void) printf("resuming from suspended state\n");
5404 spa_vdev_state_enter(spa, SCL_NONE);
5405 vdev_clear(spa, NULL);
5406 (void) spa_vdev_state_exit(spa, NULL, 0);
5407 (void) zio_resume(spa);
5408 }
5409
5410 static void *
5411 ztest_resume_thread(void *arg)
5412 {
5413 spa_t *spa = arg;
5414
5415 while (!ztest_exiting) {
5416 if (spa_suspended(spa))
5417 ztest_resume(spa);
5418 (void) poll(NULL, 0, 100);
5419
5420 /*
5421 * Periodically change the zfs_compressed_arc_enabled setting.
5422 */
5423 if (ztest_random(10) == 0)
5424 zfs_compressed_arc_enabled = ztest_random(2);
5425
5426 /*
5427 * Periodically change the zfs_abd_scatter_enabled setting.
5428 */
5429 if (ztest_random(10) == 0)
5430 zfs_abd_scatter_enabled = ztest_random(2);
5431 }
5432 return (NULL);
5433 }
5434
5435 static void *
5436 ztest_deadman_thread(void *arg)
5437 {
5438 ztest_shared_t *zs = arg;
5439 spa_t *spa = ztest_spa;
5440 hrtime_t delta, total = 0;
5441
5442 for (;;) {
5443 delta = zs->zs_thread_stop - zs->zs_thread_start +
5444 MSEC2NSEC(zfs_deadman_synctime_ms);
5445
5446 (void) poll(NULL, 0, (int)NSEC2MSEC(delta));
5447
5448 /*
5449 * If the pool is suspended then fail immediately. Otherwise,
5450 * check to see if the pool is making any progress. If
5451 * vdev_deadman() discovers that there hasn't been any recent
5452 * I/Os then it will end up aborting the tests.
5453 */
5454 if (spa_suspended(spa) || spa->spa_root_vdev == NULL) {
5455 fatal(0, "aborting test after %llu seconds because "
5456 "pool has transitioned to a suspended state.",
5457 zfs_deadman_synctime_ms / 1000);
5458 return (NULL);
5459 }
5460 vdev_deadman(spa->spa_root_vdev);
5461
5462 total += zfs_deadman_synctime_ms/1000;
5463 (void) printf("ztest has been running for %lld seconds\n",
5464 total);
5465 }
5466 }
5467
5468 static void
5469 ztest_execute(int test, ztest_info_t *zi, uint64_t id)
5470 {
5471 ztest_ds_t *zd = &ztest_ds[id % ztest_opts.zo_datasets];
5472 ztest_shared_callstate_t *zc = ZTEST_GET_SHARED_CALLSTATE(test);
5473 hrtime_t functime = gethrtime();
5474
5475 for (int i = 0; i < zi->zi_iters; i++)
5476 zi->zi_func(zd, id);
5477
5478 functime = gethrtime() - functime;
5479
5480 atomic_add_64(&zc->zc_count, 1);
5481 atomic_add_64(&zc->zc_time, functime);
5482
5483 if (ztest_opts.zo_verbose >= 4) {
5484 Dl_info dli;
5485 (void) dladdr((void *)zi->zi_func, &dli);
5486 (void) printf("%6.2f sec in %s\n",
5487 (double)functime / NANOSEC, dli.dli_sname);
5488 }
5489 }
5490
5491 static void *
5492 ztest_thread(void *arg)
5493 {
5494 int rand;
5495 uint64_t id = (uintptr_t)arg;
5496 ztest_shared_t *zs = ztest_shared;
5497 uint64_t call_next;
5498 hrtime_t now;
5499 ztest_info_t *zi;
5500 ztest_shared_callstate_t *zc;
5501
5502 while ((now = gethrtime()) < zs->zs_thread_stop) {
5503 /*
5504 * See if it's time to force a crash.
5505 */
5506 if (now > zs->zs_thread_kill)
5507 ztest_kill(zs);
5508
5509 /*
5510 * If we're getting ENOSPC with some regularity, stop.
5511 */
5512 if (zs->zs_enospc_count > 10)
5513 break;
5514
5515 /*
5516 * Pick a random function to execute.
5517 */
5518 rand = ztest_random(ZTEST_FUNCS);
5519 zi = &ztest_info[rand];
5520 zc = ZTEST_GET_SHARED_CALLSTATE(rand);
5521 call_next = zc->zc_next;
5522
5523 if (now >= call_next &&
5524 atomic_cas_64(&zc->zc_next, call_next, call_next +
5525 ztest_random(2 * zi->zi_interval[0] + 1)) == call_next) {
5526 ztest_execute(rand, zi, id);
5527 }
5528 }
5529
5530 return (NULL);
5531 }
5532
5533 static void
5534 ztest_dataset_name(char *dsname, char *pool, int d)
5535 {
5536 (void) snprintf(dsname, ZFS_MAX_DATASET_NAME_LEN, "%s/ds_%d", pool, d);
5537 }
5538
5539 static void
5540 ztest_dataset_destroy(int d)
5541 {
5542 char name[ZFS_MAX_DATASET_NAME_LEN];
5543
5544 ztest_dataset_name(name, ztest_opts.zo_pool, d);
5545
5546 if (ztest_opts.zo_verbose >= 3)
5547 (void) printf("Destroying %s to free up space\n", name);
5548
5549 /*
5550 * Cleanup any non-standard clones and snapshots. In general,
5551 * ztest thread t operates on dataset (t % zopt_datasets),
5552 * so there may be more than one thing to clean up.
5553 */
5554 for (int t = d; t < ztest_opts.zo_threads;
5555 t += ztest_opts.zo_datasets) {
5556 ztest_dsl_dataset_cleanup(name, t);
5557 }
5558
5559 (void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
5560 DS_FIND_SNAPSHOTS | DS_FIND_CHILDREN);
5561 }
5562
5563 static void
5564 ztest_dataset_dirobj_verify(ztest_ds_t *zd)
5565 {
5566 uint64_t usedobjs, dirobjs, scratch;
5567
5568 /*
5569 * ZTEST_DIROBJ is the object directory for the entire dataset.
5570 * Therefore, the number of objects in use should equal the
5571 * number of ZTEST_DIROBJ entries, +1 for ZTEST_DIROBJ itself.
5572 * If not, we have an object leak.
5573 *
5574 * Note that we can only check this in ztest_dataset_open(),
5575 * when the open-context and syncing-context values agree.
5576 * That's because zap_count() returns the open-context value,
5577 * while dmu_objset_space() returns the rootbp fill count.
5578 */
5579 VERIFY3U(0, ==, zap_count(zd->zd_os, ZTEST_DIROBJ, &dirobjs));
5580 dmu_objset_space(zd->zd_os, &scratch, &scratch, &usedobjs, &scratch);
5581 ASSERT3U(dirobjs + 1, ==, usedobjs);
5582 }
5583
5584 static int
5585 ztest_dataset_open(int d)
5586 {
5587 ztest_ds_t *zd = &ztest_ds[d];
5588 uint64_t committed_seq = ZTEST_GET_SHARED_DS(d)->zd_seq;
5589 objset_t *os;
5590 zilog_t *zilog;
5591 char name[ZFS_MAX_DATASET_NAME_LEN];
5592 int error;
5593
5594 ztest_dataset_name(name, ztest_opts.zo_pool, d);
5595
5596 (void) rw_rdlock(&ztest_name_lock);
5597
5598 error = ztest_dataset_create(name);
5599 if (error == ENOSPC) {
5600 (void) rw_unlock(&ztest_name_lock);
5601 ztest_record_enospc(FTAG);
5602 return (error);
5603 }
5604 ASSERT(error == 0 || error == EEXIST);
5605
5606 VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, zd, &os));
5607 (void) rw_unlock(&ztest_name_lock);
5608
5609 ztest_zd_init(zd, ZTEST_GET_SHARED_DS(d), os);
5610
5611 zilog = zd->zd_zilog;
5612
5613 if (zilog->zl_header->zh_claim_lr_seq != 0 &&
5614 zilog->zl_header->zh_claim_lr_seq < committed_seq)
5615 fatal(0, "missing log records: claimed %llu < committed %llu",
5616 zilog->zl_header->zh_claim_lr_seq, committed_seq);
5617
5618 ztest_dataset_dirobj_verify(zd);
5619
5620 zil_replay(os, zd, ztest_replay_vector);
5621
5622 ztest_dataset_dirobj_verify(zd);
5623
5624 if (ztest_opts.zo_verbose >= 6)
5625 (void) printf("%s replay %llu blocks, %llu records, seq %llu\n",
5626 zd->zd_name,
5627 (u_longlong_t)zilog->zl_parse_blk_count,
5628 (u_longlong_t)zilog->zl_parse_lr_count,
5629 (u_longlong_t)zilog->zl_replaying_seq);
5630
5631 zilog = zil_open(os, ztest_get_data);
5632
5633 if (zilog->zl_replaying_seq != 0 &&
5634 zilog->zl_replaying_seq < committed_seq)
5635 fatal(0, "missing log records: replayed %llu < committed %llu",
5636 zilog->zl_replaying_seq, committed_seq);
5637
5638 return (0);
5639 }
5640
5641 static void
5642 ztest_dataset_close(int d)
5643 {
5644 ztest_ds_t *zd = &ztest_ds[d];
5645
5646 zil_close(zd->zd_zilog);
5647 dmu_objset_disown(zd->zd_os, zd);
5648
5649 ztest_zd_fini(zd);
5650 }
5651
5652 /*
5653 * Kick off threads to run tests on all datasets in parallel.
5654 */
5655 static void
5656 ztest_run(ztest_shared_t *zs)
5657 {
5658 thread_t *tid;
5659 spa_t *spa;
5660 objset_t *os;
5661 thread_t resume_tid;
5662 int error;
5663
5664 ztest_exiting = B_FALSE;
5665
5666 /*
5667 * Initialize parent/child shared state.
5668 */
5669 VERIFY(_mutex_init(&ztest_vdev_lock, USYNC_THREAD, NULL) == 0);
5670 VERIFY(rwlock_init(&ztest_name_lock, USYNC_THREAD, NULL) == 0);
5671
5672 zs->zs_thread_start = gethrtime();
5673 zs->zs_thread_stop =
5674 zs->zs_thread_start + ztest_opts.zo_passtime * NANOSEC;
5675 zs->zs_thread_stop = MIN(zs->zs_thread_stop, zs->zs_proc_stop);
5676 zs->zs_thread_kill = zs->zs_thread_stop;
5677 if (ztest_random(100) < ztest_opts.zo_killrate) {
5678 zs->zs_thread_kill -=
5679 ztest_random(ztest_opts.zo_passtime * NANOSEC);
5680 }
5681
5682 (void) _mutex_init(&zcl.zcl_callbacks_lock, USYNC_THREAD, NULL);
5683
5684 list_create(&zcl.zcl_callbacks, sizeof (ztest_cb_data_t),
5685 offsetof(ztest_cb_data_t, zcd_node));
5686
5687 /*
5688 * Open our pool.
5689 */
5690 kernel_init(FREAD | FWRITE);
5691 VERIFY0(spa_open(ztest_opts.zo_pool, &spa, FTAG));
5692 spa->spa_debug = B_TRUE;
5693 metaslab_preload_limit = ztest_random(20) + 1;
5694 ztest_spa = spa;
5695
5696 dmu_objset_stats_t dds;
5697 VERIFY0(dmu_objset_own(ztest_opts.zo_pool,
5698 DMU_OST_ANY, B_TRUE, FTAG, &os));
5699 dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
5700 dmu_objset_fast_stat(os, &dds);
5701 dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
5702 zs->zs_guid = dds.dds_guid;
5703 dmu_objset_disown(os, FTAG);
5704
5705 spa->spa_dedup_ditto = 2 * ZIO_DEDUPDITTO_MIN;
5706
5707 /*
5708 * We don't expect the pool to suspend unless maxfaults == 0,
5709 * in which case ztest_fault_inject() temporarily takes away
5710 * the only valid replica.
5711 */
5712 if (MAXFAULTS() == 0)
5713 spa->spa_failmode = ZIO_FAILURE_MODE_WAIT;
5714 else
5715 spa->spa_failmode = ZIO_FAILURE_MODE_PANIC;
5716
5717 /*
5718 * Create a thread to periodically resume suspended I/O.
5719 */
5720 VERIFY(thr_create(0, 0, ztest_resume_thread, spa, THR_BOUND,
5721 &resume_tid) == 0);
5722
5723 /*
5724 * Create a deadman thread to abort() if we hang.
5725 */
5726 VERIFY(thr_create(0, 0, ztest_deadman_thread, zs, THR_BOUND,
5727 NULL) == 0);
5728
5729 /*
5730 * Verify that we can safely inquire about about any object,
5731 * whether it's allocated or not. To make it interesting,
5732 * we probe a 5-wide window around each power of two.
5733 * This hits all edge cases, including zero and the max.
5734 */
5735 for (int t = 0; t < 64; t++) {
5736 for (int d = -5; d <= 5; d++) {
5737 error = dmu_object_info(spa->spa_meta_objset,
5738 (1ULL << t) + d, NULL);
5739 ASSERT(error == 0 || error == ENOENT ||
5740 error == EINVAL);
5741 }
5742 }
5743
5744 /*
5745 * If we got any ENOSPC errors on the previous run, destroy something.
5746 */
5747 if (zs->zs_enospc_count != 0) {
5748 int d = ztest_random(ztest_opts.zo_datasets);
5749 ztest_dataset_destroy(d);
5750 }
5751 zs->zs_enospc_count = 0;
5752
5753 tid = umem_zalloc(ztest_opts.zo_threads * sizeof (thread_t),
5754 UMEM_NOFAIL);
5755
5756 if (ztest_opts.zo_verbose >= 4)
5757 (void) printf("starting main threads...\n");
5758
5759 /*
5760 * Kick off all the tests that run in parallel.
5761 */
5762 for (int t = 0; t < ztest_opts.zo_threads; t++) {
5763 if (t < ztest_opts.zo_datasets &&
5764 ztest_dataset_open(t) != 0)
5765 return;
5766 VERIFY(thr_create(0, 0, ztest_thread, (void *)(uintptr_t)t,
5767 THR_BOUND, &tid[t]) == 0);
5768 }
5769
5770 /*
5771 * Wait for all of the tests to complete. We go in reverse order
5772 * so we don't close datasets while threads are still using them.
5773 */
5774 for (int t = ztest_opts.zo_threads - 1; t >= 0; t--) {
5775 VERIFY(thr_join(tid[t], NULL, NULL) == 0);
5776 if (t < ztest_opts.zo_datasets)
5777 ztest_dataset_close(t);
5778 }
5779
5780 txg_wait_synced(spa_get_dsl(spa), 0);
5781
5782 zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(spa));
5783 zs->zs_space = metaslab_class_get_space(spa_normal_class(spa));
5784 zfs_dbgmsg_print(FTAG);
5785
5786 umem_free(tid, ztest_opts.zo_threads * sizeof (thread_t));
5787
5788 /* Kill the resume thread */
5789 ztest_exiting = B_TRUE;
5790 VERIFY(thr_join(resume_tid, NULL, NULL) == 0);
5791 ztest_resume(spa);
5792
5793 /*
5794 * Right before closing the pool, kick off a bunch of async I/O;
5795 * spa_close() should wait for it to complete.
5796 */
5797 for (uint64_t object = 1; object < 50; object++) {
5798 dmu_prefetch(spa->spa_meta_objset, object, 0, 0, 1ULL << 20,
5799 ZIO_PRIORITY_SYNC_READ);
5800 }
5801
5802 spa_close(spa, FTAG);
5803
5804 /*
5805 * Verify that we can loop over all pools.
5806 */
5807 mutex_enter(&spa_namespace_lock);
5808 for (spa = spa_next(NULL); spa != NULL; spa = spa_next(spa))
5809 if (ztest_opts.zo_verbose > 3)
5810 (void) printf("spa_next: found %s\n", spa_name(spa));
5811 mutex_exit(&spa_namespace_lock);
5812
5813 /*
5814 * Verify that we can export the pool and reimport it under a
5815 * different name.
5816 */
5817 if (ztest_random(2) == 0) {
5818 char name[ZFS_MAX_DATASET_NAME_LEN];
5819 (void) snprintf(name, sizeof (name), "%s_import",
5820 ztest_opts.zo_pool);
5821 ztest_spa_import_export(ztest_opts.zo_pool, name);
5822 ztest_spa_import_export(name, ztest_opts.zo_pool);
5823 }
5824
5825 kernel_fini();
5826
5827 list_destroy(&zcl.zcl_callbacks);
5828
5829 (void) _mutex_destroy(&zcl.zcl_callbacks_lock);
5830
5831 (void) rwlock_destroy(&ztest_name_lock);
5832 (void) _mutex_destroy(&ztest_vdev_lock);
5833 }
5834
5835 static void
5836 ztest_freeze(void)
5837 {
5838 ztest_ds_t *zd = &ztest_ds[0];
5839 spa_t *spa;
5840 int numloops = 0;
5841
5842 if (ztest_opts.zo_verbose >= 3)
5843 (void) printf("testing spa_freeze()...\n");
5844
5845 kernel_init(FREAD | FWRITE);
5846 VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
5847 VERIFY3U(0, ==, ztest_dataset_open(0));
5848 spa->spa_debug = B_TRUE;
5849 ztest_spa = spa;
5850
5851 /*
5852 * Force the first log block to be transactionally allocated.
5853 * We have to do this before we freeze the pool -- otherwise
5854 * the log chain won't be anchored.
5855 */
5856 while (BP_IS_HOLE(&zd->zd_zilog->zl_header->zh_log)) {
5857 ztest_dmu_object_alloc_free(zd, 0);
5858 zil_commit(zd->zd_zilog, 0);
5859 }
5860
5861 txg_wait_synced(spa_get_dsl(spa), 0);
5862
5863 /*
5864 * Freeze the pool. This stops spa_sync() from doing anything,
5865 * so that the only way to record changes from now on is the ZIL.
5866 */
5867 spa_freeze(spa);
5868
5869 /*
5870 * Because it is hard to predict how much space a write will actually
5871 * require beforehand, we leave ourselves some fudge space to write over
5872 * capacity.
5873 */
5874 uint64_t capacity = metaslab_class_get_space(spa_normal_class(spa)) / 2;
5875
5876 /*
5877 * Run tests that generate log records but don't alter the pool config
5878 * or depend on DSL sync tasks (snapshots, objset create/destroy, etc).
5879 * We do a txg_wait_synced() after each iteration to force the txg
5880 * to increase well beyond the last synced value in the uberblock.
5881 * The ZIL should be OK with that.
5882 *
5883 * Run a random number of times less than zo_maxloops and ensure we do
5884 * not run out of space on the pool.
5885 */
5886 while (ztest_random(10) != 0 &&
5887 numloops++ < ztest_opts.zo_maxloops &&
5888 metaslab_class_get_alloc(spa_normal_class(spa)) < capacity) {
5889 ztest_od_t od;
5890 ztest_od_init(&od, 0, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
5891 VERIFY0(ztest_object_init(zd, &od, sizeof (od), B_FALSE));
5892 ztest_io(zd, od.od_object,
5893 ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
5894 txg_wait_synced(spa_get_dsl(spa), 0);
5895 }
5896
5897 /*
5898 * Commit all of the changes we just generated.
5899 */
5900 zil_commit(zd->zd_zilog, 0);
5901 txg_wait_synced(spa_get_dsl(spa), 0);
5902
5903 /*
5904 * Close our dataset and close the pool.
5905 */
5906 ztest_dataset_close(0);
5907 spa_close(spa, FTAG);
5908 kernel_fini();
5909
5910 /*
5911 * Open and close the pool and dataset to induce log replay.
5912 */
5913 kernel_init(FREAD | FWRITE);
5914 VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
5915 ASSERT(spa_freeze_txg(spa) == UINT64_MAX);
5916 VERIFY3U(0, ==, ztest_dataset_open(0));
5917 ztest_dataset_close(0);
5918
5919 spa->spa_debug = B_TRUE;
5920 ztest_spa = spa;
5921 txg_wait_synced(spa_get_dsl(spa), 0);
5922 ztest_reguid(NULL, 0);
5923
5924 spa_close(spa, FTAG);
5925 kernel_fini();
5926 }
5927
5928 void
5929 print_time(hrtime_t t, char *timebuf)
5930 {
5931 hrtime_t s = t / NANOSEC;
5932 hrtime_t m = s / 60;
5933 hrtime_t h = m / 60;
5934 hrtime_t d = h / 24;
5935
5936 s -= m * 60;
5937 m -= h * 60;
5938 h -= d * 24;
5939
5940 timebuf[0] = '\0';
5941
5942 if (d)
5943 (void) sprintf(timebuf,
5944 "%llud%02lluh%02llum%02llus", d, h, m, s);
5945 else if (h)
5946 (void) sprintf(timebuf, "%lluh%02llum%02llus", h, m, s);
5947 else if (m)
5948 (void) sprintf(timebuf, "%llum%02llus", m, s);
5949 else
5950 (void) sprintf(timebuf, "%llus", s);
5951 }
5952
5953 static nvlist_t *
5954 make_random_props()
5955 {
5956 nvlist_t *props;
5957
5958 VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
5959 if (ztest_random(2) == 0)
5960 return (props);
5961 VERIFY(nvlist_add_uint64(props, "autoreplace", 1) == 0);
5962
5963 return (props);
5964 }
5965
5966 /*
5967 * Create a storage pool with the given name and initial vdev size.
5968 * Then test spa_freeze() functionality.
5969 */
5970 static void
5971 ztest_init(ztest_shared_t *zs)
5972 {
5973 spa_t *spa;
5974 nvlist_t *nvroot, *props;
5975
5976 VERIFY(_mutex_init(&ztest_vdev_lock, USYNC_THREAD, NULL) == 0);
5977 VERIFY(rwlock_init(&ztest_name_lock, USYNC_THREAD, NULL) == 0);
5978
5979 kernel_init(FREAD | FWRITE);
5980
5981 /*
5982 * Create the storage pool.
5983 */
5984 (void) spa_destroy(ztest_opts.zo_pool);
5985 ztest_shared->zs_vdev_next_leaf = 0;
5986 zs->zs_splits = 0;
5987 zs->zs_mirrors = ztest_opts.zo_mirrors;
5988 nvroot = make_vdev_root(NULL, NULL, NULL, ztest_opts.zo_vdev_size, 0,
5989 0, ztest_opts.zo_raidz, zs->zs_mirrors, 1);
5990 props = make_random_props();
5991 for (int i = 0; i < SPA_FEATURES; i++) {
5992 char buf[1024];
5993 (void) snprintf(buf, sizeof (buf), "feature@%s",
5994 spa_feature_table[i].fi_uname);
5995 VERIFY3U(0, ==, nvlist_add_uint64(props, buf, 0));
5996 }
5997 VERIFY3U(0, ==, spa_create(ztest_opts.zo_pool, nvroot, props, NULL));
5998 nvlist_free(nvroot);
5999 nvlist_free(props);
6000
6001 VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
6002 zs->zs_metaslab_sz =
6003 1ULL << spa->spa_root_vdev->vdev_child[0]->vdev_ms_shift;
6004
6005 spa_close(spa, FTAG);
6006
6007 kernel_fini();
6008
6009 ztest_run_zdb(ztest_opts.zo_pool);
6010
6011 ztest_freeze();
6012
6013 ztest_run_zdb(ztest_opts.zo_pool);
6014
6015 (void) rwlock_destroy(&ztest_name_lock);
6016 (void) _mutex_destroy(&ztest_vdev_lock);
6017 }
6018
6019 static void
6020 setup_data_fd(void)
6021 {
6022 static char ztest_name_data[] = "/tmp/ztest.data.XXXXXX";
6023
6024 ztest_fd_data = mkstemp(ztest_name_data);
6025 ASSERT3S(ztest_fd_data, >=, 0);
6026 (void) unlink(ztest_name_data);
6027 }
6028
6029
6030 static int
6031 shared_data_size(ztest_shared_hdr_t *hdr)
6032 {
6033 int size;
6034
6035 size = hdr->zh_hdr_size;
6036 size += hdr->zh_opts_size;
6037 size += hdr->zh_size;
6038 size += hdr->zh_stats_size * hdr->zh_stats_count;
6039 size += hdr->zh_ds_size * hdr->zh_ds_count;
6040
6041 return (size);
6042 }
6043
6044 static void
6045 setup_hdr(void)
6046 {
6047 int size;
6048 ztest_shared_hdr_t *hdr;
6049
6050 hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
6051 PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
6052 ASSERT(hdr != MAP_FAILED);
6053
6054 VERIFY3U(0, ==, ftruncate(ztest_fd_data, sizeof (ztest_shared_hdr_t)));
6055
6056 hdr->zh_hdr_size = sizeof (ztest_shared_hdr_t);
6057 hdr->zh_opts_size = sizeof (ztest_shared_opts_t);
6058 hdr->zh_size = sizeof (ztest_shared_t);
6059 hdr->zh_stats_size = sizeof (ztest_shared_callstate_t);
6060 hdr->zh_stats_count = ZTEST_FUNCS;
6061 hdr->zh_ds_size = sizeof (ztest_shared_ds_t);
6062 hdr->zh_ds_count = ztest_opts.zo_datasets;
6063
6064 size = shared_data_size(hdr);
6065 VERIFY3U(0, ==, ftruncate(ztest_fd_data, size));
6066
6067 (void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
6068 }
6069
6070 static void
6071 setup_data(void)
6072 {
6073 int size, offset;
6074 ztest_shared_hdr_t *hdr;
6075 uint8_t *buf;
6076
6077 hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
6078 PROT_READ, MAP_SHARED, ztest_fd_data, 0);
6079 ASSERT(hdr != MAP_FAILED);
6080
6081 size = shared_data_size(hdr);
6082
6083 (void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
6084 hdr = ztest_shared_hdr = (void *)mmap(0, P2ROUNDUP(size, getpagesize()),
6085 PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
6086 ASSERT(hdr != MAP_FAILED);
6087 buf = (uint8_t *)hdr;
6088
6089 offset = hdr->zh_hdr_size;
6090 ztest_shared_opts = (void *)&buf[offset];
6091 offset += hdr->zh_opts_size;
6092 ztest_shared = (void *)&buf[offset];
6093 offset += hdr->zh_size;
6094 ztest_shared_callstate = (void *)&buf[offset];
6095 offset += hdr->zh_stats_size * hdr->zh_stats_count;
6096 ztest_shared_ds = (void *)&buf[offset];
6097 }
6098
6099 static boolean_t
6100 exec_child(char *cmd, char *libpath, boolean_t ignorekill, int *statusp)
6101 {
6102 pid_t pid;
6103 int status;
6104 char *cmdbuf = NULL;
6105
6106 pid = fork();
6107
6108 if (cmd == NULL) {
6109 cmdbuf = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
6110 (void) strlcpy(cmdbuf, getexecname(), MAXPATHLEN);
6111 cmd = cmdbuf;
6112 }
6113
6114 if (pid == -1)
6115 fatal(1, "fork failed");
6116
6117 if (pid == 0) { /* child */
6118 char *emptyargv[2] = { cmd, NULL };
6119 char fd_data_str[12];
6120
6121 struct rlimit rl = { 1024, 1024 };
6122 (void) setrlimit(RLIMIT_NOFILE, &rl);
6123
6124 (void) close(ztest_fd_rand);
6125 VERIFY3U(11, >=,
6126 snprintf(fd_data_str, 12, "%d", ztest_fd_data));
6127 VERIFY0(setenv("ZTEST_FD_DATA", fd_data_str, 1));
6128
6129 (void) enable_extended_FILE_stdio(-1, -1);
6130 if (libpath != NULL)
6131 VERIFY(0 == setenv("LD_LIBRARY_PATH", libpath, 1));
6132 (void) execv(cmd, emptyargv);
6133 ztest_dump_core = B_FALSE;
6134 fatal(B_TRUE, "exec failed: %s", cmd);
6135 }
6136
6137 if (cmdbuf != NULL) {
6138 umem_free(cmdbuf, MAXPATHLEN);
6139 cmd = NULL;
6140 }
6141
6142 while (waitpid(pid, &status, 0) != pid)
6143 continue;
6144 if (statusp != NULL)
6145 *statusp = status;
6146
6147 if (WIFEXITED(status)) {
6148 if (WEXITSTATUS(status) != 0) {
6149 (void) fprintf(stderr, "child exited with code %d\n",
6150 WEXITSTATUS(status));
6151 exit(2);
6152 }
6153 return (B_FALSE);
6154 } else if (WIFSIGNALED(status)) {
6155 if (!ignorekill || WTERMSIG(status) != SIGKILL) {
6156 (void) fprintf(stderr, "child died with signal %d\n",
6157 WTERMSIG(status));
6158 exit(3);
6159 }
6160 return (B_TRUE);
6161 } else {
6162 (void) fprintf(stderr, "something strange happened to child\n");
6163 exit(4);
6164 /* NOTREACHED */
6165 }
6166 }
6167
6168 static void
6169 ztest_run_init(void)
6170 {
6171 ztest_shared_t *zs = ztest_shared;
6172
6173 ASSERT(ztest_opts.zo_init != 0);
6174
6175 /*
6176 * Blow away any existing copy of zpool.cache
6177 */
6178 (void) remove(spa_config_path);
6179
6180 /*
6181 * Create and initialize our storage pool.
6182 */
6183 for (int i = 1; i <= ztest_opts.zo_init; i++) {
6184 bzero(zs, sizeof (ztest_shared_t));
6185 if (ztest_opts.zo_verbose >= 3 &&
6186 ztest_opts.zo_init != 1) {
6187 (void) printf("ztest_init(), pass %d\n", i);
6188 }
6189 ztest_init(zs);
6190 }
6191 }
6192
6193 int
6194 main(int argc, char **argv)
6195 {
6196 int kills = 0;
6197 int iters = 0;
6198 int older = 0;
6199 int newer = 0;
6200 ztest_shared_t *zs;
6201 ztest_info_t *zi;
6202 ztest_shared_callstate_t *zc;
6203 char timebuf[100];
6204 char numbuf[6];
6205 spa_t *spa;
6206 char *cmd;
6207 boolean_t hasalt;
6208 char *fd_data_str = getenv("ZTEST_FD_DATA");
6209
6210 (void) setvbuf(stdout, NULL, _IOLBF, 0);
6211
6212 dprintf_setup(&argc, argv);
6213 zfs_deadman_synctime_ms = 300000;
6214
6215 ztest_fd_rand = open("/dev/urandom", O_RDONLY);
6216 ASSERT3S(ztest_fd_rand, >=, 0);
6217
6218 if (!fd_data_str) {
6219 process_options(argc, argv);
6220
6221 setup_data_fd();
6222 setup_hdr();
6223 setup_data();
6224 bcopy(&ztest_opts, ztest_shared_opts,
6225 sizeof (*ztest_shared_opts));
6226 } else {
6227 ztest_fd_data = atoi(fd_data_str);
6228 setup_data();
6229 bcopy(ztest_shared_opts, &ztest_opts, sizeof (ztest_opts));
6230 }
6231 ASSERT3U(ztest_opts.zo_datasets, ==, ztest_shared_hdr->zh_ds_count);
6232
6233 /* Override location of zpool.cache */
6234 VERIFY3U(asprintf((char **)&spa_config_path, "%s/zpool.cache",
6235 ztest_opts.zo_dir), !=, -1);
6236
6237 ztest_ds = umem_alloc(ztest_opts.zo_datasets * sizeof (ztest_ds_t),
6238 UMEM_NOFAIL);
6239 zs = ztest_shared;
6240
6241 if (fd_data_str) {
6242 metaslab_gang_bang = ztest_opts.zo_metaslab_gang_bang;
6243 metaslab_df_alloc_threshold =
6244 zs->zs_metaslab_df_alloc_threshold;
6245
6246 if (zs->zs_do_init)
6247 ztest_run_init();
6248 else
6249 ztest_run(zs);
6250 exit(0);
6251 }
6252
6253 hasalt = (strlen(ztest_opts.zo_alt_ztest) != 0);
6254
6255 if (ztest_opts.zo_verbose >= 1) {
6256 (void) printf("%llu vdevs, %d datasets, %d threads,"
6257 " %llu seconds...\n",
6258 (u_longlong_t)ztest_opts.zo_vdevs,
6259 ztest_opts.zo_datasets,
6260 ztest_opts.zo_threads,
6261 (u_longlong_t)ztest_opts.zo_time);
6262 }
6263
6264 cmd = umem_alloc(MAXNAMELEN, UMEM_NOFAIL);
6265 (void) strlcpy(cmd, getexecname(), MAXNAMELEN);
6266
6267 zs->zs_do_init = B_TRUE;
6268 if (strlen(ztest_opts.zo_alt_ztest) != 0) {
6269 if (ztest_opts.zo_verbose >= 1) {
6270 (void) printf("Executing older ztest for "
6271 "initialization: %s\n", ztest_opts.zo_alt_ztest);
6272 }
6273 VERIFY(!exec_child(ztest_opts.zo_alt_ztest,
6274 ztest_opts.zo_alt_libpath, B_FALSE, NULL));
6275 } else {
6276 VERIFY(!exec_child(NULL, NULL, B_FALSE, NULL));
6277 }
6278 zs->zs_do_init = B_FALSE;
6279
6280 zs->zs_proc_start = gethrtime();
6281 zs->zs_proc_stop = zs->zs_proc_start + ztest_opts.zo_time * NANOSEC;
6282
6283 for (int f = 0; f < ZTEST_FUNCS; f++) {
6284 zi = &ztest_info[f];
6285 zc = ZTEST_GET_SHARED_CALLSTATE(f);
6286 if (zs->zs_proc_start + zi->zi_interval[0] > zs->zs_proc_stop)
6287 zc->zc_next = UINT64_MAX;
6288 else
6289 zc->zc_next = zs->zs_proc_start +
6290 ztest_random(2 * zi->zi_interval[0] + 1);
6291 }
6292
6293 /*
6294 * Run the tests in a loop. These tests include fault injection
6295 * to verify that self-healing data works, and forced crashes
6296 * to verify that we never lose on-disk consistency.
6297 */
6298 while (gethrtime() < zs->zs_proc_stop) {
6299 int status;
6300 boolean_t killed;
6301
6302 /*
6303 * Initialize the workload counters for each function.
6304 */
6305 for (int f = 0; f < ZTEST_FUNCS; f++) {
6306 zc = ZTEST_GET_SHARED_CALLSTATE(f);
6307 zc->zc_count = 0;
6308 zc->zc_time = 0;
6309 }
6310
6311 /* Set the allocation switch size */
6312 zs->zs_metaslab_df_alloc_threshold =
6313 ztest_random(zs->zs_metaslab_sz / 4) + 1;
6314
6315 if (!hasalt || ztest_random(2) == 0) {
6316 if (hasalt && ztest_opts.zo_verbose >= 1) {
6317 (void) printf("Executing newer ztest: %s\n",
6318 cmd);
6319 }
6320 newer++;
6321 killed = exec_child(cmd, NULL, B_TRUE, &status);
6322 } else {
6323 if (hasalt && ztest_opts.zo_verbose >= 1) {
6324 (void) printf("Executing older ztest: %s\n",
6325 ztest_opts.zo_alt_ztest);
6326 }
6327 older++;
6328 killed = exec_child(ztest_opts.zo_alt_ztest,
6329 ztest_opts.zo_alt_libpath, B_TRUE, &status);
6330 }
6331
6332 if (killed)
6333 kills++;
6334 iters++;
6335
6336 if (ztest_opts.zo_verbose >= 1) {
6337 hrtime_t now = gethrtime();
6338
6339 now = MIN(now, zs->zs_proc_stop);
6340 print_time(zs->zs_proc_stop - now, timebuf);
6341 nicenum(zs->zs_space, numbuf);
6342
6343 (void) printf("Pass %3d, %8s, %3llu ENOSPC, "
6344 "%4.1f%% of %5s used, %3.0f%% done, %8s to go\n",
6345 iters,
6346 WIFEXITED(status) ? "Complete" : "SIGKILL",
6347 (u_longlong_t)zs->zs_enospc_count,
6348 100.0 * zs->zs_alloc / zs->zs_space,
6349 numbuf,
6350 100.0 * (now - zs->zs_proc_start) /
6351 (ztest_opts.zo_time * NANOSEC), timebuf);
6352 }
6353
6354 if (ztest_opts.zo_verbose >= 2) {
6355 (void) printf("\nWorkload summary:\n\n");
6356 (void) printf("%7s %9s %s\n",
6357 "Calls", "Time", "Function");
6358 (void) printf("%7s %9s %s\n",
6359 "-----", "----", "--------");
6360 for (int f = 0; f < ZTEST_FUNCS; f++) {
6361 Dl_info dli;
6362
6363 zi = &ztest_info[f];
6364 zc = ZTEST_GET_SHARED_CALLSTATE(f);
6365 print_time(zc->zc_time, timebuf);
6366 (void) dladdr((void *)zi->zi_func, &dli);
6367 (void) printf("%7llu %9s %s\n",
6368 (u_longlong_t)zc->zc_count, timebuf,
6369 dli.dli_sname);
6370 }
6371 (void) printf("\n");
6372 }
6373
6374 /*
6375 * It's possible that we killed a child during a rename test,
6376 * in which case we'll have a 'ztest_tmp' pool lying around
6377 * instead of 'ztest'. Do a blind rename in case this happened.
6378 */
6379 kernel_init(FREAD);
6380 if (spa_open(ztest_opts.zo_pool, &spa, FTAG) == 0) {
6381 spa_close(spa, FTAG);
6382 } else {
6383 char tmpname[ZFS_MAX_DATASET_NAME_LEN];
6384 kernel_fini();
6385 kernel_init(FREAD | FWRITE);
6386 (void) snprintf(tmpname, sizeof (tmpname), "%s_tmp",
6387 ztest_opts.zo_pool);
6388 (void) spa_rename(tmpname, ztest_opts.zo_pool);
6389 }
6390 kernel_fini();
6391
6392 ztest_run_zdb(ztest_opts.zo_pool);
6393 }
6394
6395 if (ztest_opts.zo_verbose >= 1) {
6396 if (hasalt) {
6397 (void) printf("%d runs of older ztest: %s\n", older,
6398 ztest_opts.zo_alt_ztest);
6399 (void) printf("%d runs of newer ztest: %s\n", newer,
6400 cmd);
6401 }
6402 (void) printf("%d killed, %d completed, %.0f%% kill rate\n",
6403 kills, iters - kills, (100.0 * kills) / MAX(1, iters));
6404 }
6405
6406 umem_free(cmd, MAXNAMELEN);
6407
6408 return (0);
6409 }
unleashed repo fork