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) 2012 Martin Matuska <mm@FreeBSD.org>. All rights reserved.
26 * Copyright (c) 2013 Steven Hartland. All rights reserved.
27 * Copyright (c) 2014 Integros [integros.com]
28 */
29
30 /*
31 * The objective of this program is to provide a DMU/ZAP/SPA stress test
32 * that runs entirely in userland, is easy to use, and easy to extend.
33 *
34 * The overall design of the ztest program is as follows:
35 *
36 * (1) For each major functional area (e.g. adding vdevs to a pool,
37 * creating and destroying datasets, reading and writing objects, etc)
38 * we have a simple routine to test that functionality. These
39 * individual routines do not have to do anything "stressful".
40 *
41 * (2) We turn these simple functionality tests into a stress test by
42 * running them all in parallel, with as many threads as desired,
43 * and spread across as many datasets, objects, and vdevs as desired.
44 *
45 * (3) While all this is happening, we inject faults into the pool to
46 * verify that self-healing data really works.
47 *
48 * (4) Every time we open a dataset, we change its checksum and compression
49 * functions. Thus even individual objects vary from block to block
50 * in which checksum they use and whether they're compressed.
51 *
52 * (5) To verify that we never lose on-disk consistency after a crash,
53 * we run the entire test in a child of the main process.
54 * At random times, the child self-immolates with a SIGKILL.
55 * This is the software equivalent of pulling the power cord.
56 * The parent then runs the test again, using the existing
57 * storage pool, as many times as desired. If backwards compatibility
58 * testing is enabled ztest will sometimes run the "older" version
59 * of ztest after a SIGKILL.
60 *
61 * (6) To verify that we don't have future leaks or temporal incursions,
62 * many of the functional tests record the transaction group number
63 * as part of their data. When reading old data, they verify that
64 * the transaction group number is less than the current, open txg.
65 * If you add a new test, please do this if applicable.
66 *
67 * When run with no arguments, ztest runs for about five minutes and
68 * produces no output if successful. To get a little bit of information,
69 * specify -V. To get more information, specify -VV, and so on.
70 *
71 * To turn this into an overnight stress test, use -T to specify run time.
72 *
73 * You can ask more more vdevs [-v], datasets [-d], or threads [-t]
74 * to increase the pool capacity, fanout, and overall stress level.
75 *
76 * Use the -k option to set the desired frequency of kills.
77 *
78 * When ztest invokes itself it passes all relevant information through a
79 * temporary file which is mmap-ed in the child process. This allows shared
80 * memory to survive the exec syscall. The ztest_shared_hdr_t struct is always
81 * stored at offset 0 of this file and contains information on the size and
82 * number of shared structures in the file. The information stored in this file
83 * must remain backwards compatible with older versions of ztest so that
84 * ztest can invoke them during backwards compatibility testing (-B).
85 */
86
87 #include <sys/zfs_context.h>
88 #include <sys/spa.h>
89 #include <sys/dmu.h>
90 #include <sys/txg.h>
91 #include <sys/dbuf.h>
92 #include <sys/zap.h>
93 #include <sys/dmu_objset.h>
94 #include <sys/poll.h>
95 #include <sys/stat.h>
96 #include <sys/time.h>
97 #include <sys/wait.h>
98 #include <sys/mman.h>
99 #include <sys/resource.h>
100 #include <sys/zio.h>
101 #include <sys/zil.h>
102 #include <sys/zil_impl.h>
103 #include <sys/vdev_impl.h>
104 #include <sys/vdev_file.h>
105 #include <sys/spa_impl.h>
106 #include <sys/metaslab_impl.h>
107 #include <sys/dsl_prop.h>
108 #include <sys/dsl_dataset.h>
109 #include <sys/dsl_destroy.h>
110 #include <sys/dsl_scan.h>
111 #include <sys/zio_checksum.h>
112 #include <sys/refcount.h>
113 #include <sys/zfeature.h>
114 #include <sys/dsl_userhold.h>
115 #include <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 <errno.h>
125 #include <sys/fs/zfs.h>
126 #include <libnvpair.h>
127
128 static int ztest_fd_data = -1;
129 static int ztest_fd_rand = -1;
130
131 typedef struct ztest_shared_hdr {
132 uint64_t zh_hdr_size;
133 uint64_t zh_opts_size;
134 uint64_t zh_size;
135 uint64_t zh_stats_size;
136 uint64_t zh_stats_count;
137 uint64_t zh_ds_size;
138 uint64_t zh_ds_count;
139 } ztest_shared_hdr_t;
140
141 static ztest_shared_hdr_t *ztest_shared_hdr;
142
143 typedef struct ztest_shared_opts {
144 char zo_pool[ZFS_MAX_DATASET_NAME_LEN];
145 char zo_dir[ZFS_MAX_DATASET_NAME_LEN];
146 char zo_alt_ztest[MAXNAMELEN];
147 char zo_alt_libpath[MAXNAMELEN];
148 uint64_t zo_vdevs;
149 uint64_t zo_vdevtime;
150 size_t zo_vdev_size;
151 int zo_ashift;
152 int zo_mirrors;
153 int zo_raidz;
154 int zo_raidz_parity;
155 int zo_datasets;
156 int zo_threads;
157 uint64_t zo_passtime;
158 uint64_t zo_killrate;
159 int zo_verbose;
160 int zo_init;
161 uint64_t zo_time;
162 uint64_t zo_maxloops;
163 uint64_t zo_metaslab_gang_bang;
164 } ztest_shared_opts_t;
165
166 static const ztest_shared_opts_t ztest_opts_defaults = {
167 .zo_pool = { 'z', 't', 'e', 's', 't', '\0' },
168 .zo_dir = { '/', 't', 'm', 'p', '\0' },
169 .zo_alt_ztest = { '\0' },
170 .zo_alt_libpath = { '\0' },
171 .zo_vdevs = 5,
172 .zo_ashift = SPA_MINBLOCKSHIFT,
173 .zo_mirrors = 2,
174 .zo_raidz = 4,
175 .zo_raidz_parity = 1,
176 .zo_vdev_size = SPA_MINDEVSIZE * 4, /* 256m default size */
177 .zo_datasets = 7,
178 .zo_threads = 23,
179 .zo_passtime = 60, /* 60 seconds */
180 .zo_killrate = 70, /* 70% kill rate */
181 .zo_verbose = 0,
182 .zo_init = 1,
183 .zo_time = 300, /* 5 minutes */
184 .zo_maxloops = 50, /* max loops during spa_freeze() */
185 .zo_metaslab_gang_bang = 32 << 10
186 };
187
188 extern uint64_t metaslab_gang_bang;
189 extern uint64_t metaslab_df_alloc_threshold;
190 extern uint64_t zfs_deadman_synctime_ms;
191 extern int metaslab_preload_limit;
192 extern boolean_t zfs_compressed_arc_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 *
_umem_debug_init()456 _umem_debug_init()
457 {
458 return ("default,verbose"); /* $UMEM_DEBUG setting */
459 }
460
461 const char *
_umem_logging_init(void)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
fatal(int do_perror,char * message,...)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
str2shift(const char * buf)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
nicenumtoull(const char * buf)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
usage(boolean_t requested)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[-h] (print help)\n"
585 "",
586 zo->zo_pool,
587 (u_longlong_t)zo->zo_vdevs, /* -v */
588 nice_vdev_size, /* -s */
589 zo->zo_ashift, /* -a */
590 zo->zo_mirrors, /* -m */
591 zo->zo_raidz, /* -r */
592 zo->zo_raidz_parity, /* -R */
593 zo->zo_datasets, /* -d */
594 zo->zo_threads, /* -t */
595 nice_gang_bang, /* -g */
596 zo->zo_init, /* -i */
597 (u_longlong_t)zo->zo_killrate, /* -k */
598 zo->zo_pool, /* -p */
599 zo->zo_dir, /* -f */
600 (u_longlong_t)zo->zo_time, /* -T */
601 (u_longlong_t)zo->zo_maxloops, /* -F */
602 (u_longlong_t)zo->zo_passtime);
603 exit(requested ? 0 : 1);
604 }
605
606 static void
process_options(int argc,char ** argv)607 process_options(int argc, char **argv)
608 {
609 char *path;
610 ztest_shared_opts_t *zo = &ztest_opts;
611
612 int opt;
613 uint64_t value;
614 char altdir[MAXNAMELEN] = { 0 };
615
616 bcopy(&ztest_opts_defaults, zo, sizeof (*zo));
617
618 while ((opt = getopt(argc, argv,
619 "v:s:a:m:r:R:d:t:g:i:k:p:f:VET:P:hF:B:")) != EOF) {
620 value = 0;
621 switch (opt) {
622 case 'v':
623 case 's':
624 case 'a':
625 case 'm':
626 case 'r':
627 case 'R':
628 case 'd':
629 case 't':
630 case 'g':
631 case 'i':
632 case 'k':
633 case 'T':
634 case 'P':
635 case 'F':
636 value = nicenumtoull(optarg);
637 }
638 switch (opt) {
639 case 'v':
640 zo->zo_vdevs = value;
641 break;
642 case 's':
643 zo->zo_vdev_size = MAX(SPA_MINDEVSIZE, value);
644 break;
645 case 'a':
646 zo->zo_ashift = value;
647 break;
648 case 'm':
649 zo->zo_mirrors = value;
650 break;
651 case 'r':
652 zo->zo_raidz = MAX(1, value);
653 break;
654 case 'R':
655 zo->zo_raidz_parity = MIN(MAX(value, 1), 3);
656 break;
657 case 'd':
658 zo->zo_datasets = MAX(1, value);
659 break;
660 case 't':
661 zo->zo_threads = MAX(1, value);
662 break;
663 case 'g':
664 zo->zo_metaslab_gang_bang = MAX(SPA_MINBLOCKSIZE << 1,
665 value);
666 break;
667 case 'i':
668 zo->zo_init = value;
669 break;
670 case 'k':
671 zo->zo_killrate = value;
672 break;
673 case 'p':
674 (void) strlcpy(zo->zo_pool, optarg,
675 sizeof (zo->zo_pool));
676 break;
677 case 'f':
678 path = realpath(optarg, NULL);
679 if (path == NULL) {
680 (void) fprintf(stderr, "error: %s: %s\n",
681 optarg, strerror(errno));
682 usage(B_FALSE);
683 } else {
684 (void) strlcpy(zo->zo_dir, path,
685 sizeof (zo->zo_dir));
686 }
687 break;
688 case 'V':
689 zo->zo_verbose++;
690 break;
691 case 'E':
692 zo->zo_init = 0;
693 break;
694 case 'T':
695 zo->zo_time = value;
696 break;
697 case 'P':
698 zo->zo_passtime = MAX(1, value);
699 break;
700 case 'F':
701 zo->zo_maxloops = MAX(1, value);
702 break;
703 case 'B':
704 (void) strlcpy(altdir, optarg, sizeof (altdir));
705 break;
706 case 'h':
707 usage(B_TRUE);
708 break;
709 case '?':
710 default:
711 usage(B_FALSE);
712 break;
713 }
714 }
715
716 zo->zo_raidz_parity = MIN(zo->zo_raidz_parity, zo->zo_raidz - 1);
717
718 zo->zo_vdevtime =
719 (zo->zo_vdevs > 0 ? zo->zo_time * NANOSEC / zo->zo_vdevs :
720 UINT64_MAX >> 2);
721
722 if (strlen(altdir) > 0) {
723 char *cmd;
724 char *realaltdir;
725 char *bin;
726 char *ztest;
727 char *isa;
728 int isalen;
729
730 cmd = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
731 realaltdir = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
732
733 VERIFY(NULL != realpath(getexecname(), cmd));
734 if (0 != access(altdir, F_OK)) {
735 ztest_dump_core = B_FALSE;
736 fatal(B_TRUE, "invalid alternate ztest path: %s",
737 altdir);
738 }
739 VERIFY(NULL != realpath(altdir, realaltdir));
740
741 /*
742 * 'cmd' should be of the form "<anything>/usr/bin/<isa>/ztest".
743 * We want to extract <isa> to determine if we should use
744 * 32 or 64 bit binaries.
745 */
746 bin = strstr(cmd, "/usr/bin/");
747 ztest = strstr(bin, "/ztest");
748 isa = bin + 9;
749 isalen = ztest - isa;
750 (void) snprintf(zo->zo_alt_ztest, sizeof (zo->zo_alt_ztest),
751 "%s/usr/bin/%.*s/ztest", realaltdir, isalen, isa);
752 (void) snprintf(zo->zo_alt_libpath, sizeof (zo->zo_alt_libpath),
753 "%s/usr/lib/%.*s", realaltdir, isalen, isa);
754
755 if (0 != access(zo->zo_alt_ztest, X_OK)) {
756 ztest_dump_core = B_FALSE;
757 fatal(B_TRUE, "invalid alternate ztest: %s",
758 zo->zo_alt_ztest);
759 } else if (0 != access(zo->zo_alt_libpath, X_OK)) {
760 ztest_dump_core = B_FALSE;
761 fatal(B_TRUE, "invalid alternate lib directory %s",
762 zo->zo_alt_libpath);
763 }
764
765 umem_free(cmd, MAXPATHLEN);
766 umem_free(realaltdir, MAXPATHLEN);
767 }
768 }
769
770 static void
ztest_kill(ztest_shared_t * zs)771 ztest_kill(ztest_shared_t *zs)
772 {
773 zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(ztest_spa));
774 zs->zs_space = metaslab_class_get_space(spa_normal_class(ztest_spa));
775
776 /*
777 * Before we kill off ztest, make sure that the config is updated.
778 * See comment above spa_config_sync().
779 */
780 mutex_enter(&spa_namespace_lock);
781 spa_config_sync(ztest_spa, B_FALSE, B_FALSE);
782 mutex_exit(&spa_namespace_lock);
783
784 zfs_dbgmsg_print(FTAG);
785 (void) kill(getpid(), SIGKILL);
786 }
787
788 static uint64_t
ztest_random(uint64_t range)789 ztest_random(uint64_t range)
790 {
791 uint64_t r;
792
793 ASSERT3S(ztest_fd_rand, >=, 0);
794
795 if (range == 0)
796 return (0);
797
798 if (read(ztest_fd_rand, &r, sizeof (r)) != sizeof (r))
799 fatal(1, "short read from /dev/urandom");
800
801 return (r % range);
802 }
803
804 /* ARGSUSED */
805 static void
ztest_record_enospc(const char * s)806 ztest_record_enospc(const char *s)
807 {
808 ztest_shared->zs_enospc_count++;
809 }
810
811 static uint64_t
ztest_get_ashift(void)812 ztest_get_ashift(void)
813 {
814 if (ztest_opts.zo_ashift == 0)
815 return (SPA_MINBLOCKSHIFT + ztest_random(5));
816 return (ztest_opts.zo_ashift);
817 }
818
819 static nvlist_t *
make_vdev_file(char * path,char * aux,char * pool,size_t size,uint64_t ashift)820 make_vdev_file(char *path, char *aux, char *pool, size_t size, uint64_t ashift)
821 {
822 char pathbuf[MAXPATHLEN];
823 uint64_t vdev;
824 nvlist_t *file;
825
826 if (ashift == 0)
827 ashift = ztest_get_ashift();
828
829 if (path == NULL) {
830 path = pathbuf;
831
832 if (aux != NULL) {
833 vdev = ztest_shared->zs_vdev_aux;
834 (void) snprintf(path, sizeof (pathbuf),
835 ztest_aux_template, ztest_opts.zo_dir,
836 pool == NULL ? ztest_opts.zo_pool : pool,
837 aux, vdev);
838 } else {
839 vdev = ztest_shared->zs_vdev_next_leaf++;
840 (void) snprintf(path, sizeof (pathbuf),
841 ztest_dev_template, ztest_opts.zo_dir,
842 pool == NULL ? ztest_opts.zo_pool : pool, vdev);
843 }
844 }
845
846 if (size != 0) {
847 int fd = open(path, O_RDWR | O_CREAT | O_TRUNC, 0666);
848 if (fd == -1)
849 fatal(1, "can't open %s", path);
850 if (ftruncate(fd, size) != 0)
851 fatal(1, "can't ftruncate %s", path);
852 (void) close(fd);
853 }
854
855 VERIFY(nvlist_alloc(&file, NV_UNIQUE_NAME, 0) == 0);
856 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_TYPE, VDEV_TYPE_FILE) == 0);
857 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_PATH, path) == 0);
858 VERIFY(nvlist_add_uint64(file, ZPOOL_CONFIG_ASHIFT, ashift) == 0);
859
860 return (file);
861 }
862
863 static nvlist_t *
make_vdev_raidz(char * path,char * aux,char * pool,size_t size,uint64_t ashift,int r)864 make_vdev_raidz(char *path, char *aux, char *pool, size_t size,
865 uint64_t ashift, int r)
866 {
867 nvlist_t *raidz, **child;
868 int c;
869
870 if (r < 2)
871 return (make_vdev_file(path, aux, pool, size, ashift));
872 child = umem_alloc(r * sizeof (nvlist_t *), UMEM_NOFAIL);
873
874 for (c = 0; c < r; c++)
875 child[c] = make_vdev_file(path, aux, pool, size, ashift);
876
877 VERIFY(nvlist_alloc(&raidz, NV_UNIQUE_NAME, 0) == 0);
878 VERIFY(nvlist_add_string(raidz, ZPOOL_CONFIG_TYPE,
879 VDEV_TYPE_RAIDZ) == 0);
880 VERIFY(nvlist_add_uint64(raidz, ZPOOL_CONFIG_NPARITY,
881 ztest_opts.zo_raidz_parity) == 0);
882 VERIFY(nvlist_add_nvlist_array(raidz, ZPOOL_CONFIG_CHILDREN,
883 child, r) == 0);
884
885 for (c = 0; c < r; c++)
886 nvlist_free(child[c]);
887
888 umem_free(child, r * sizeof (nvlist_t *));
889
890 return (raidz);
891 }
892
893 static nvlist_t *
make_vdev_mirror(char * path,char * aux,char * pool,size_t size,uint64_t ashift,int r,int m)894 make_vdev_mirror(char *path, char *aux, char *pool, size_t size,
895 uint64_t ashift, int r, int m)
896 {
897 nvlist_t *mirror, **child;
898 int c;
899
900 if (m < 1)
901 return (make_vdev_raidz(path, aux, pool, size, ashift, r));
902
903 child = umem_alloc(m * sizeof (nvlist_t *), UMEM_NOFAIL);
904
905 for (c = 0; c < m; c++)
906 child[c] = make_vdev_raidz(path, aux, pool, size, ashift, r);
907
908 VERIFY(nvlist_alloc(&mirror, NV_UNIQUE_NAME, 0) == 0);
909 VERIFY(nvlist_add_string(mirror, ZPOOL_CONFIG_TYPE,
910 VDEV_TYPE_MIRROR) == 0);
911 VERIFY(nvlist_add_nvlist_array(mirror, ZPOOL_CONFIG_CHILDREN,
912 child, m) == 0);
913
914 for (c = 0; c < m; c++)
915 nvlist_free(child[c]);
916
917 umem_free(child, m * sizeof (nvlist_t *));
918
919 return (mirror);
920 }
921
922 static nvlist_t *
make_vdev_root(char * path,char * aux,char * pool,size_t size,uint64_t ashift,int log,int r,int m,int t)923 make_vdev_root(char *path, char *aux, char *pool, size_t size, uint64_t ashift,
924 int log, int r, int m, int t)
925 {
926 nvlist_t *root, **child;
927 int c;
928
929 ASSERT(t > 0);
930
931 child = umem_alloc(t * sizeof (nvlist_t *), UMEM_NOFAIL);
932
933 for (c = 0; c < t; c++) {
934 child[c] = make_vdev_mirror(path, aux, pool, size, ashift,
935 r, m);
936 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_IS_LOG,
937 log) == 0);
938 }
939
940 VERIFY(nvlist_alloc(&root, NV_UNIQUE_NAME, 0) == 0);
941 VERIFY(nvlist_add_string(root, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) == 0);
942 VERIFY(nvlist_add_nvlist_array(root, aux ? aux : ZPOOL_CONFIG_CHILDREN,
943 child, t) == 0);
944
945 for (c = 0; c < t; c++)
946 nvlist_free(child[c]);
947
948 umem_free(child, t * sizeof (nvlist_t *));
949
950 return (root);
951 }
952
953 /*
954 * Find a random spa version. Returns back a random spa version in the
955 * range [initial_version, SPA_VERSION_FEATURES].
956 */
957 static uint64_t
ztest_random_spa_version(uint64_t initial_version)958 ztest_random_spa_version(uint64_t initial_version)
959 {
960 uint64_t version = initial_version;
961
962 if (version <= SPA_VERSION_BEFORE_FEATURES) {
963 version = version +
964 ztest_random(SPA_VERSION_BEFORE_FEATURES - version + 1);
965 }
966
967 if (version > SPA_VERSION_BEFORE_FEATURES)
968 version = SPA_VERSION_FEATURES;
969
970 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
971 return (version);
972 }
973
974 static int
ztest_random_blocksize(void)975 ztest_random_blocksize(void)
976 {
977 uint64_t block_shift;
978 /*
979 * Choose a block size >= the ashift.
980 * If the SPA supports new MAXBLOCKSIZE, test up to 1MB blocks.
981 */
982 int maxbs = SPA_OLD_MAXBLOCKSHIFT;
983 if (spa_maxblocksize(ztest_spa) == SPA_MAXBLOCKSIZE)
984 maxbs = 20;
985 block_shift = ztest_random(maxbs - ztest_spa->spa_max_ashift + 1);
986 return (1 << (SPA_MINBLOCKSHIFT + block_shift));
987 }
988
989 static int
ztest_random_ibshift(void)990 ztest_random_ibshift(void)
991 {
992 return (DN_MIN_INDBLKSHIFT +
993 ztest_random(DN_MAX_INDBLKSHIFT - DN_MIN_INDBLKSHIFT + 1));
994 }
995
996 static uint64_t
ztest_random_vdev_top(spa_t * spa,boolean_t log_ok)997 ztest_random_vdev_top(spa_t *spa, boolean_t log_ok)
998 {
999 uint64_t top;
1000 vdev_t *rvd = spa->spa_root_vdev;
1001 vdev_t *tvd;
1002
1003 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1004
1005 do {
1006 top = ztest_random(rvd->vdev_children);
1007 tvd = rvd->vdev_child[top];
1008 } while (tvd->vdev_ishole || (tvd->vdev_islog && !log_ok) ||
1009 tvd->vdev_mg == NULL || tvd->vdev_mg->mg_class == NULL);
1010
1011 return (top);
1012 }
1013
1014 static uint64_t
ztest_random_dsl_prop(zfs_prop_t prop)1015 ztest_random_dsl_prop(zfs_prop_t prop)
1016 {
1017 uint64_t value;
1018
1019 do {
1020 value = zfs_prop_random_value(prop, ztest_random(-1ULL));
1021 } while (prop == ZFS_PROP_CHECKSUM && value == ZIO_CHECKSUM_OFF);
1022
1023 return (value);
1024 }
1025
1026 static int
ztest_dsl_prop_set_uint64(char * osname,zfs_prop_t prop,uint64_t value,boolean_t inherit)1027 ztest_dsl_prop_set_uint64(char *osname, zfs_prop_t prop, uint64_t value,
1028 boolean_t inherit)
1029 {
1030 const char *propname = zfs_prop_to_name(prop);
1031 const char *valname;
1032 char setpoint[MAXPATHLEN];
1033 uint64_t curval;
1034 int error;
1035
1036 error = dsl_prop_set_int(osname, propname,
1037 (inherit ? ZPROP_SRC_NONE : ZPROP_SRC_LOCAL), value);
1038
1039 if (error == ENOSPC) {
1040 ztest_record_enospc(FTAG);
1041 return (error);
1042 }
1043 ASSERT0(error);
1044
1045 VERIFY0(dsl_prop_get_integer(osname, propname, &curval, setpoint));
1046
1047 if (ztest_opts.zo_verbose >= 6) {
1048 VERIFY(zfs_prop_index_to_string(prop, curval, &valname) == 0);
1049 (void) printf("%s %s = %s at '%s'\n",
1050 osname, propname, valname, setpoint);
1051 }
1052
1053 return (error);
1054 }
1055
1056 static int
ztest_spa_prop_set_uint64(zpool_prop_t prop,uint64_t value)1057 ztest_spa_prop_set_uint64(zpool_prop_t prop, uint64_t value)
1058 {
1059 spa_t *spa = ztest_spa;
1060 nvlist_t *props = NULL;
1061 int error;
1062
1063 VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
1064 VERIFY(nvlist_add_uint64(props, zpool_prop_to_name(prop), value) == 0);
1065
1066 error = spa_prop_set(spa, props);
1067
1068 nvlist_free(props);
1069
1070 if (error == ENOSPC) {
1071 ztest_record_enospc(FTAG);
1072 return (error);
1073 }
1074 ASSERT0(error);
1075
1076 return (error);
1077 }
1078
1079 static void
ztest_rll_init(rll_t * rll)1080 ztest_rll_init(rll_t *rll)
1081 {
1082 rll->rll_writer = NULL;
1083 rll->rll_readers = 0;
1084 VERIFY(_mutex_init(&rll->rll_lock, USYNC_THREAD, NULL) == 0);
1085 VERIFY(cond_init(&rll->rll_cv, USYNC_THREAD, NULL) == 0);
1086 }
1087
1088 static void
ztest_rll_destroy(rll_t * rll)1089 ztest_rll_destroy(rll_t *rll)
1090 {
1091 ASSERT(rll->rll_writer == NULL);
1092 ASSERT(rll->rll_readers == 0);
1093 VERIFY(_mutex_destroy(&rll->rll_lock) == 0);
1094 VERIFY(cond_destroy(&rll->rll_cv) == 0);
1095 }
1096
1097 static void
ztest_rll_lock(rll_t * rll,rl_type_t type)1098 ztest_rll_lock(rll_t *rll, rl_type_t type)
1099 {
1100 VERIFY(mutex_lock(&rll->rll_lock) == 0);
1101
1102 if (type == RL_READER) {
1103 while (rll->rll_writer != NULL)
1104 (void) cond_wait(&rll->rll_cv, &rll->rll_lock);
1105 rll->rll_readers++;
1106 } else {
1107 while (rll->rll_writer != NULL || rll->rll_readers)
1108 (void) cond_wait(&rll->rll_cv, &rll->rll_lock);
1109 rll->rll_writer = curthread;
1110 }
1111
1112 VERIFY(mutex_unlock(&rll->rll_lock) == 0);
1113 }
1114
1115 static void
ztest_rll_unlock(rll_t * rll)1116 ztest_rll_unlock(rll_t *rll)
1117 {
1118 VERIFY(mutex_lock(&rll->rll_lock) == 0);
1119
1120 if (rll->rll_writer) {
1121 ASSERT(rll->rll_readers == 0);
1122 rll->rll_writer = NULL;
1123 } else {
1124 ASSERT(rll->rll_readers != 0);
1125 ASSERT(rll->rll_writer == NULL);
1126 rll->rll_readers--;
1127 }
1128
1129 if (rll->rll_writer == NULL && rll->rll_readers == 0)
1130 VERIFY(cond_broadcast(&rll->rll_cv) == 0);
1131
1132 VERIFY(mutex_unlock(&rll->rll_lock) == 0);
1133 }
1134
1135 static void
ztest_object_lock(ztest_ds_t * zd,uint64_t object,rl_type_t type)1136 ztest_object_lock(ztest_ds_t *zd, uint64_t object, rl_type_t type)
1137 {
1138 rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1139
1140 ztest_rll_lock(rll, type);
1141 }
1142
1143 static void
ztest_object_unlock(ztest_ds_t * zd,uint64_t object)1144 ztest_object_unlock(ztest_ds_t *zd, uint64_t object)
1145 {
1146 rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1147
1148 ztest_rll_unlock(rll);
1149 }
1150
1151 static rl_t *
ztest_range_lock(ztest_ds_t * zd,uint64_t object,uint64_t offset,uint64_t size,rl_type_t type)1152 ztest_range_lock(ztest_ds_t *zd, uint64_t object, uint64_t offset,
1153 uint64_t size, rl_type_t type)
1154 {
1155 uint64_t hash = object ^ (offset % (ZTEST_RANGE_LOCKS + 1));
1156 rll_t *rll = &zd->zd_range_lock[hash & (ZTEST_RANGE_LOCKS - 1)];
1157 rl_t *rl;
1158
1159 rl = umem_alloc(sizeof (*rl), UMEM_NOFAIL);
1160 rl->rl_object = object;
1161 rl->rl_offset = offset;
1162 rl->rl_size = size;
1163 rl->rl_lock = rll;
1164
1165 ztest_rll_lock(rll, type);
1166
1167 return (rl);
1168 }
1169
1170 static void
ztest_range_unlock(rl_t * rl)1171 ztest_range_unlock(rl_t *rl)
1172 {
1173 rll_t *rll = rl->rl_lock;
1174
1175 ztest_rll_unlock(rll);
1176
1177 umem_free(rl, sizeof (*rl));
1178 }
1179
1180 static void
ztest_zd_init(ztest_ds_t * zd,ztest_shared_ds_t * szd,objset_t * os)1181 ztest_zd_init(ztest_ds_t *zd, ztest_shared_ds_t *szd, objset_t *os)
1182 {
1183 zd->zd_os = os;
1184 zd->zd_zilog = dmu_objset_zil(os);
1185 zd->zd_shared = szd;
1186 dmu_objset_name(os, zd->zd_name);
1187
1188 if (zd->zd_shared != NULL)
1189 zd->zd_shared->zd_seq = 0;
1190
1191 VERIFY(rwlock_init(&zd->zd_zilog_lock, USYNC_THREAD, NULL) == 0);
1192 VERIFY(_mutex_init(&zd->zd_dirobj_lock, USYNC_THREAD, NULL) == 0);
1193
1194 for (int l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1195 ztest_rll_init(&zd->zd_object_lock[l]);
1196
1197 for (int l = 0; l < ZTEST_RANGE_LOCKS; l++)
1198 ztest_rll_init(&zd->zd_range_lock[l]);
1199 }
1200
1201 static void
ztest_zd_fini(ztest_ds_t * zd)1202 ztest_zd_fini(ztest_ds_t *zd)
1203 {
1204 VERIFY(_mutex_destroy(&zd->zd_dirobj_lock) == 0);
1205
1206 for (int l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1207 ztest_rll_destroy(&zd->zd_object_lock[l]);
1208
1209 for (int l = 0; l < ZTEST_RANGE_LOCKS; l++)
1210 ztest_rll_destroy(&zd->zd_range_lock[l]);
1211 }
1212
1213 #define TXG_MIGHTWAIT (ztest_random(10) == 0 ? TXG_NOWAIT : TXG_WAIT)
1214
1215 static uint64_t
ztest_tx_assign(dmu_tx_t * tx,uint64_t txg_how,const char * tag)1216 ztest_tx_assign(dmu_tx_t *tx, uint64_t txg_how, const char *tag)
1217 {
1218 uint64_t txg;
1219 int error;
1220
1221 /*
1222 * Attempt to assign tx to some transaction group.
1223 */
1224 error = dmu_tx_assign(tx, txg_how);
1225 if (error) {
1226 if (error == ERESTART) {
1227 ASSERT(txg_how == TXG_NOWAIT);
1228 dmu_tx_wait(tx);
1229 } else {
1230 ASSERT3U(error, ==, ENOSPC);
1231 ztest_record_enospc(tag);
1232 }
1233 dmu_tx_abort(tx);
1234 return (0);
1235 }
1236 txg = dmu_tx_get_txg(tx);
1237 ASSERT(txg != 0);
1238 return (txg);
1239 }
1240
1241 static void
ztest_pattern_set(void * buf,uint64_t size,uint64_t value)1242 ztest_pattern_set(void *buf, uint64_t size, uint64_t value)
1243 {
1244 uint64_t *ip = buf;
1245 uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1246
1247 while (ip < ip_end)
1248 *ip++ = value;
1249 }
1250
1251 static boolean_t
ztest_pattern_match(void * buf,uint64_t size,uint64_t value)1252 ztest_pattern_match(void *buf, uint64_t size, uint64_t value)
1253 {
1254 uint64_t *ip = buf;
1255 uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1256 uint64_t diff = 0;
1257
1258 while (ip < ip_end)
1259 diff |= (value - *ip++);
1260
1261 return (diff == 0);
1262 }
1263
1264 static void
ztest_bt_generate(ztest_block_tag_t * bt,objset_t * os,uint64_t object,uint64_t offset,uint64_t gen,uint64_t txg,uint64_t crtxg)1265 ztest_bt_generate(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1266 uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg)
1267 {
1268 bt->bt_magic = BT_MAGIC;
1269 bt->bt_objset = dmu_objset_id(os);
1270 bt->bt_object = object;
1271 bt->bt_offset = offset;
1272 bt->bt_gen = gen;
1273 bt->bt_txg = txg;
1274 bt->bt_crtxg = crtxg;
1275 }
1276
1277 static void
ztest_bt_verify(ztest_block_tag_t * bt,objset_t * os,uint64_t object,uint64_t offset,uint64_t gen,uint64_t txg,uint64_t crtxg)1278 ztest_bt_verify(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1279 uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg)
1280 {
1281 ASSERT3U(bt->bt_magic, ==, BT_MAGIC);
1282 ASSERT3U(bt->bt_objset, ==, dmu_objset_id(os));
1283 ASSERT3U(bt->bt_object, ==, object);
1284 ASSERT3U(bt->bt_offset, ==, offset);
1285 ASSERT3U(bt->bt_gen, <=, gen);
1286 ASSERT3U(bt->bt_txg, <=, txg);
1287 ASSERT3U(bt->bt_crtxg, ==, crtxg);
1288 }
1289
1290 static ztest_block_tag_t *
ztest_bt_bonus(dmu_buf_t * db)1291 ztest_bt_bonus(dmu_buf_t *db)
1292 {
1293 dmu_object_info_t doi;
1294 ztest_block_tag_t *bt;
1295
1296 dmu_object_info_from_db(db, &doi);
1297 ASSERT3U(doi.doi_bonus_size, <=, db->db_size);
1298 ASSERT3U(doi.doi_bonus_size, >=, sizeof (*bt));
1299 bt = (void *)((char *)db->db_data + doi.doi_bonus_size - sizeof (*bt));
1300
1301 return (bt);
1302 }
1303
1304 /*
1305 * ZIL logging ops
1306 */
1307
1308 #define lrz_type lr_mode
1309 #define lrz_blocksize lr_uid
1310 #define lrz_ibshift lr_gid
1311 #define lrz_bonustype lr_rdev
1312 #define lrz_bonuslen lr_crtime[1]
1313
1314 static void
ztest_log_create(ztest_ds_t * zd,dmu_tx_t * tx,lr_create_t * lr)1315 ztest_log_create(ztest_ds_t *zd, dmu_tx_t *tx, lr_create_t *lr)
1316 {
1317 char *name = (void *)(lr + 1); /* name follows lr */
1318 size_t namesize = strlen(name) + 1;
1319 itx_t *itx;
1320
1321 if (zil_replaying(zd->zd_zilog, tx))
1322 return;
1323
1324 itx = zil_itx_create(TX_CREATE, sizeof (*lr) + namesize);
1325 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1326 sizeof (*lr) + namesize - sizeof (lr_t));
1327
1328 zil_itx_assign(zd->zd_zilog, itx, tx);
1329 }
1330
1331 static void
ztest_log_remove(ztest_ds_t * zd,dmu_tx_t * tx,lr_remove_t * lr,uint64_t object)1332 ztest_log_remove(ztest_ds_t *zd, dmu_tx_t *tx, lr_remove_t *lr, uint64_t object)
1333 {
1334 char *name = (void *)(lr + 1); /* name follows lr */
1335 size_t namesize = strlen(name) + 1;
1336 itx_t *itx;
1337
1338 if (zil_replaying(zd->zd_zilog, tx))
1339 return;
1340
1341 itx = zil_itx_create(TX_REMOVE, sizeof (*lr) + namesize);
1342 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1343 sizeof (*lr) + namesize - sizeof (lr_t));
1344
1345 itx->itx_oid = object;
1346 zil_itx_assign(zd->zd_zilog, itx, tx);
1347 }
1348
1349 static void
ztest_log_write(ztest_ds_t * zd,dmu_tx_t * tx,lr_write_t * lr)1350 ztest_log_write(ztest_ds_t *zd, dmu_tx_t *tx, lr_write_t *lr)
1351 {
1352 itx_t *itx;
1353 itx_wr_state_t write_state = ztest_random(WR_NUM_STATES);
1354
1355 if (zil_replaying(zd->zd_zilog, tx))
1356 return;
1357
1358 if (lr->lr_length > ZIL_MAX_LOG_DATA)
1359 write_state = WR_INDIRECT;
1360
1361 itx = zil_itx_create(TX_WRITE,
1362 sizeof (*lr) + (write_state == WR_COPIED ? lr->lr_length : 0));
1363
1364 if (write_state == WR_COPIED &&
1365 dmu_read(zd->zd_os, lr->lr_foid, lr->lr_offset, lr->lr_length,
1366 ((lr_write_t *)&itx->itx_lr) + 1, DMU_READ_NO_PREFETCH) != 0) {
1367 zil_itx_destroy(itx);
1368 itx = zil_itx_create(TX_WRITE, sizeof (*lr));
1369 write_state = WR_NEED_COPY;
1370 }
1371 itx->itx_private = zd;
1372 itx->itx_wr_state = write_state;
1373 itx->itx_sync = (ztest_random(8) == 0);
1374
1375 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1376 sizeof (*lr) - sizeof (lr_t));
1377
1378 zil_itx_assign(zd->zd_zilog, itx, tx);
1379 }
1380
1381 static void
ztest_log_truncate(ztest_ds_t * zd,dmu_tx_t * tx,lr_truncate_t * lr)1382 ztest_log_truncate(ztest_ds_t *zd, dmu_tx_t *tx, lr_truncate_t *lr)
1383 {
1384 itx_t *itx;
1385
1386 if (zil_replaying(zd->zd_zilog, tx))
1387 return;
1388
1389 itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr));
1390 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1391 sizeof (*lr) - sizeof (lr_t));
1392
1393 itx->itx_sync = B_FALSE;
1394 zil_itx_assign(zd->zd_zilog, itx, tx);
1395 }
1396
1397 static void
ztest_log_setattr(ztest_ds_t * zd,dmu_tx_t * tx,lr_setattr_t * lr)1398 ztest_log_setattr(ztest_ds_t *zd, dmu_tx_t *tx, lr_setattr_t *lr)
1399 {
1400 itx_t *itx;
1401
1402 if (zil_replaying(zd->zd_zilog, tx))
1403 return;
1404
1405 itx = zil_itx_create(TX_SETATTR, sizeof (*lr));
1406 bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1407 sizeof (*lr) - sizeof (lr_t));
1408
1409 itx->itx_sync = B_FALSE;
1410 zil_itx_assign(zd->zd_zilog, itx, tx);
1411 }
1412
1413 /*
1414 * ZIL replay ops
1415 */
1416 static int
ztest_replay_create(ztest_ds_t * zd,lr_create_t * lr,boolean_t byteswap)1417 ztest_replay_create(ztest_ds_t *zd, lr_create_t *lr, boolean_t byteswap)
1418 {
1419 char *name = (void *)(lr + 1); /* name follows lr */
1420 objset_t *os = zd->zd_os;
1421 ztest_block_tag_t *bbt;
1422 dmu_buf_t *db;
1423 dmu_tx_t *tx;
1424 uint64_t txg;
1425 int error = 0;
1426
1427 if (byteswap)
1428 byteswap_uint64_array(lr, sizeof (*lr));
1429
1430 ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1431 ASSERT(name[0] != '\0');
1432
1433 tx = dmu_tx_create(os);
1434
1435 dmu_tx_hold_zap(tx, lr->lr_doid, B_TRUE, name);
1436
1437 if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1438 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, B_TRUE, NULL);
1439 } else {
1440 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1441 }
1442
1443 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1444 if (txg == 0)
1445 return (ENOSPC);
1446
1447 ASSERT(dmu_objset_zil(os)->zl_replay == !!lr->lr_foid);
1448
1449 if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1450 if (lr->lr_foid == 0) {
1451 lr->lr_foid = zap_create(os,
1452 lr->lrz_type, lr->lrz_bonustype,
1453 lr->lrz_bonuslen, tx);
1454 } else {
1455 error = zap_create_claim(os, lr->lr_foid,
1456 lr->lrz_type, lr->lrz_bonustype,
1457 lr->lrz_bonuslen, tx);
1458 }
1459 } else {
1460 if (lr->lr_foid == 0) {
1461 lr->lr_foid = dmu_object_alloc(os,
1462 lr->lrz_type, 0, lr->lrz_bonustype,
1463 lr->lrz_bonuslen, tx);
1464 } else {
1465 error = dmu_object_claim(os, lr->lr_foid,
1466 lr->lrz_type, 0, lr->lrz_bonustype,
1467 lr->lrz_bonuslen, tx);
1468 }
1469 }
1470
1471 if (error) {
1472 ASSERT3U(error, ==, EEXIST);
1473 ASSERT(zd->zd_zilog->zl_replay);
1474 dmu_tx_commit(tx);
1475 return (error);
1476 }
1477
1478 ASSERT(lr->lr_foid != 0);
1479
1480 if (lr->lrz_type != DMU_OT_ZAP_OTHER)
1481 VERIFY3U(0, ==, dmu_object_set_blocksize(os, lr->lr_foid,
1482 lr->lrz_blocksize, lr->lrz_ibshift, tx));
1483
1484 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1485 bbt = ztest_bt_bonus(db);
1486 dmu_buf_will_dirty(db, tx);
1487 ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_gen, txg, txg);
1488 dmu_buf_rele(db, FTAG);
1489
1490 VERIFY3U(0, ==, zap_add(os, lr->lr_doid, name, sizeof (uint64_t), 1,
1491 &lr->lr_foid, tx));
1492
1493 (void) ztest_log_create(zd, tx, lr);
1494
1495 dmu_tx_commit(tx);
1496
1497 return (0);
1498 }
1499
1500 static int
ztest_replay_remove(ztest_ds_t * zd,lr_remove_t * lr,boolean_t byteswap)1501 ztest_replay_remove(ztest_ds_t *zd, lr_remove_t *lr, boolean_t byteswap)
1502 {
1503 char *name = (void *)(lr + 1); /* name follows lr */
1504 objset_t *os = zd->zd_os;
1505 dmu_object_info_t doi;
1506 dmu_tx_t *tx;
1507 uint64_t object, txg;
1508
1509 if (byteswap)
1510 byteswap_uint64_array(lr, sizeof (*lr));
1511
1512 ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1513 ASSERT(name[0] != '\0');
1514
1515 VERIFY3U(0, ==,
1516 zap_lookup(os, lr->lr_doid, name, sizeof (object), 1, &object));
1517 ASSERT(object != 0);
1518
1519 ztest_object_lock(zd, object, RL_WRITER);
1520
1521 VERIFY3U(0, ==, dmu_object_info(os, object, &doi));
1522
1523 tx = dmu_tx_create(os);
1524
1525 dmu_tx_hold_zap(tx, lr->lr_doid, B_FALSE, name);
1526 dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END);
1527
1528 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1529 if (txg == 0) {
1530 ztest_object_unlock(zd, object);
1531 return (ENOSPC);
1532 }
1533
1534 if (doi.doi_type == DMU_OT_ZAP_OTHER) {
1535 VERIFY3U(0, ==, zap_destroy(os, object, tx));
1536 } else {
1537 VERIFY3U(0, ==, dmu_object_free(os, object, tx));
1538 }
1539
1540 VERIFY3U(0, ==, zap_remove(os, lr->lr_doid, name, tx));
1541
1542 (void) ztest_log_remove(zd, tx, lr, object);
1543
1544 dmu_tx_commit(tx);
1545
1546 ztest_object_unlock(zd, object);
1547
1548 return (0);
1549 }
1550
1551 static int
ztest_replay_write(ztest_ds_t * zd,lr_write_t * lr,boolean_t byteswap)1552 ztest_replay_write(ztest_ds_t *zd, lr_write_t *lr, boolean_t byteswap)
1553 {
1554 objset_t *os = zd->zd_os;
1555 void *data = lr + 1; /* data follows lr */
1556 uint64_t offset, length;
1557 ztest_block_tag_t *bt = data;
1558 ztest_block_tag_t *bbt;
1559 uint64_t gen, txg, lrtxg, crtxg;
1560 dmu_object_info_t doi;
1561 dmu_tx_t *tx;
1562 dmu_buf_t *db;
1563 arc_buf_t *abuf = NULL;
1564 rl_t *rl;
1565
1566 if (byteswap)
1567 byteswap_uint64_array(lr, sizeof (*lr));
1568
1569 offset = lr->lr_offset;
1570 length = lr->lr_length;
1571
1572 /* If it's a dmu_sync() block, write the whole block */
1573 if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
1574 uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
1575 if (length < blocksize) {
1576 offset -= offset % blocksize;
1577 length = blocksize;
1578 }
1579 }
1580
1581 if (bt->bt_magic == BSWAP_64(BT_MAGIC))
1582 byteswap_uint64_array(bt, sizeof (*bt));
1583
1584 if (bt->bt_magic != BT_MAGIC)
1585 bt = NULL;
1586
1587 ztest_object_lock(zd, lr->lr_foid, RL_READER);
1588 rl = ztest_range_lock(zd, lr->lr_foid, offset, length, RL_WRITER);
1589
1590 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1591
1592 dmu_object_info_from_db(db, &doi);
1593
1594 bbt = ztest_bt_bonus(db);
1595 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1596 gen = bbt->bt_gen;
1597 crtxg = bbt->bt_crtxg;
1598 lrtxg = lr->lr_common.lrc_txg;
1599
1600 tx = dmu_tx_create(os);
1601
1602 dmu_tx_hold_write(tx, lr->lr_foid, offset, length);
1603
1604 if (ztest_random(8) == 0 && length == doi.doi_data_block_size &&
1605 P2PHASE(offset, length) == 0)
1606 abuf = dmu_request_arcbuf(db, length);
1607
1608 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1609 if (txg == 0) {
1610 if (abuf != NULL)
1611 dmu_return_arcbuf(abuf);
1612 dmu_buf_rele(db, FTAG);
1613 ztest_range_unlock(rl);
1614 ztest_object_unlock(zd, lr->lr_foid);
1615 return (ENOSPC);
1616 }
1617
1618 if (bt != NULL) {
1619 /*
1620 * Usually, verify the old data before writing new data --
1621 * but not always, because we also want to verify correct
1622 * behavior when the data was not recently read into cache.
1623 */
1624 ASSERT(offset % doi.doi_data_block_size == 0);
1625 if (ztest_random(4) != 0) {
1626 int prefetch = ztest_random(2) ?
1627 DMU_READ_PREFETCH : DMU_READ_NO_PREFETCH;
1628 ztest_block_tag_t rbt;
1629
1630 VERIFY(dmu_read(os, lr->lr_foid, offset,
1631 sizeof (rbt), &rbt, prefetch) == 0);
1632 if (rbt.bt_magic == BT_MAGIC) {
1633 ztest_bt_verify(&rbt, os, lr->lr_foid,
1634 offset, gen, txg, crtxg);
1635 }
1636 }
1637
1638 /*
1639 * Writes can appear to be newer than the bonus buffer because
1640 * the ztest_get_data() callback does a dmu_read() of the
1641 * open-context data, which may be different than the data
1642 * as it was when the write was generated.
1643 */
1644 if (zd->zd_zilog->zl_replay) {
1645 ztest_bt_verify(bt, os, lr->lr_foid, offset,
1646 MAX(gen, bt->bt_gen), MAX(txg, lrtxg),
1647 bt->bt_crtxg);
1648 }
1649
1650 /*
1651 * Set the bt's gen/txg to the bonus buffer's gen/txg
1652 * so that all of the usual ASSERTs will work.
1653 */
1654 ztest_bt_generate(bt, os, lr->lr_foid, offset, gen, txg, crtxg);
1655 }
1656
1657 if (abuf == NULL) {
1658 dmu_write(os, lr->lr_foid, offset, length, data, tx);
1659 } else {
1660 bcopy(data, abuf->b_data, length);
1661 dmu_assign_arcbuf(db, offset, abuf, tx);
1662 }
1663
1664 (void) ztest_log_write(zd, tx, lr);
1665
1666 dmu_buf_rele(db, FTAG);
1667
1668 dmu_tx_commit(tx);
1669
1670 ztest_range_unlock(rl);
1671 ztest_object_unlock(zd, lr->lr_foid);
1672
1673 return (0);
1674 }
1675
1676 static int
ztest_replay_truncate(ztest_ds_t * zd,lr_truncate_t * lr,boolean_t byteswap)1677 ztest_replay_truncate(ztest_ds_t *zd, lr_truncate_t *lr, boolean_t byteswap)
1678 {
1679 objset_t *os = zd->zd_os;
1680 dmu_tx_t *tx;
1681 uint64_t txg;
1682 rl_t *rl;
1683
1684 if (byteswap)
1685 byteswap_uint64_array(lr, sizeof (*lr));
1686
1687 ztest_object_lock(zd, lr->lr_foid, RL_READER);
1688 rl = ztest_range_lock(zd, lr->lr_foid, lr->lr_offset, lr->lr_length,
1689 RL_WRITER);
1690
1691 tx = dmu_tx_create(os);
1692
1693 dmu_tx_hold_free(tx, lr->lr_foid, lr->lr_offset, lr->lr_length);
1694
1695 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1696 if (txg == 0) {
1697 ztest_range_unlock(rl);
1698 ztest_object_unlock(zd, lr->lr_foid);
1699 return (ENOSPC);
1700 }
1701
1702 VERIFY(dmu_free_range(os, lr->lr_foid, lr->lr_offset,
1703 lr->lr_length, tx) == 0);
1704
1705 (void) ztest_log_truncate(zd, tx, lr);
1706
1707 dmu_tx_commit(tx);
1708
1709 ztest_range_unlock(rl);
1710 ztest_object_unlock(zd, lr->lr_foid);
1711
1712 return (0);
1713 }
1714
1715 static int
ztest_replay_setattr(ztest_ds_t * zd,lr_setattr_t * lr,boolean_t byteswap)1716 ztest_replay_setattr(ztest_ds_t *zd, lr_setattr_t *lr, boolean_t byteswap)
1717 {
1718 objset_t *os = zd->zd_os;
1719 dmu_tx_t *tx;
1720 dmu_buf_t *db;
1721 ztest_block_tag_t *bbt;
1722 uint64_t txg, lrtxg, crtxg;
1723
1724 if (byteswap)
1725 byteswap_uint64_array(lr, sizeof (*lr));
1726
1727 ztest_object_lock(zd, lr->lr_foid, RL_WRITER);
1728
1729 VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1730
1731 tx = dmu_tx_create(os);
1732 dmu_tx_hold_bonus(tx, lr->lr_foid);
1733
1734 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1735 if (txg == 0) {
1736 dmu_buf_rele(db, FTAG);
1737 ztest_object_unlock(zd, lr->lr_foid);
1738 return (ENOSPC);
1739 }
1740
1741 bbt = ztest_bt_bonus(db);
1742 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1743 crtxg = bbt->bt_crtxg;
1744 lrtxg = lr->lr_common.lrc_txg;
1745
1746 if (zd->zd_zilog->zl_replay) {
1747 ASSERT(lr->lr_size != 0);
1748 ASSERT(lr->lr_mode != 0);
1749 ASSERT(lrtxg != 0);
1750 } else {
1751 /*
1752 * Randomly change the size and increment the generation.
1753 */
1754 lr->lr_size = (ztest_random(db->db_size / sizeof (*bbt)) + 1) *
1755 sizeof (*bbt);
1756 lr->lr_mode = bbt->bt_gen + 1;
1757 ASSERT(lrtxg == 0);
1758 }
1759
1760 /*
1761 * Verify that the current bonus buffer is not newer than our txg.
1762 */
1763 ztest_bt_verify(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode,
1764 MAX(txg, lrtxg), crtxg);
1765
1766 dmu_buf_will_dirty(db, tx);
1767
1768 ASSERT3U(lr->lr_size, >=, sizeof (*bbt));
1769 ASSERT3U(lr->lr_size, <=, db->db_size);
1770 VERIFY0(dmu_set_bonus(db, lr->lr_size, tx));
1771 bbt = ztest_bt_bonus(db);
1772
1773 ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode, txg, crtxg);
1774
1775 dmu_buf_rele(db, FTAG);
1776
1777 (void) ztest_log_setattr(zd, tx, lr);
1778
1779 dmu_tx_commit(tx);
1780
1781 ztest_object_unlock(zd, lr->lr_foid);
1782
1783 return (0);
1784 }
1785
1786 zil_replay_func_t *ztest_replay_vector[TX_MAX_TYPE] = {
1787 NULL, /* 0 no such transaction type */
1788 ztest_replay_create, /* TX_CREATE */
1789 NULL, /* TX_MKDIR */
1790 NULL, /* TX_MKXATTR */
1791 NULL, /* TX_SYMLINK */
1792 ztest_replay_remove, /* TX_REMOVE */
1793 NULL, /* TX_RMDIR */
1794 NULL, /* TX_LINK */
1795 NULL, /* TX_RENAME */
1796 ztest_replay_write, /* TX_WRITE */
1797 ztest_replay_truncate, /* TX_TRUNCATE */
1798 ztest_replay_setattr, /* TX_SETATTR */
1799 NULL, /* TX_ACL */
1800 NULL, /* TX_CREATE_ACL */
1801 NULL, /* TX_CREATE_ATTR */
1802 NULL, /* TX_CREATE_ACL_ATTR */
1803 NULL, /* TX_MKDIR_ACL */
1804 NULL, /* TX_MKDIR_ATTR */
1805 NULL, /* TX_MKDIR_ACL_ATTR */
1806 NULL, /* TX_WRITE2 */
1807 };
1808
1809 /*
1810 * ZIL get_data callbacks
1811 */
1812
1813 static void
ztest_get_done(zgd_t * zgd,int error)1814 ztest_get_done(zgd_t *zgd, int error)
1815 {
1816 ztest_ds_t *zd = zgd->zgd_private;
1817 uint64_t object = zgd->zgd_rl->rl_object;
1818
1819 if (zgd->zgd_db)
1820 dmu_buf_rele(zgd->zgd_db, zgd);
1821
1822 ztest_range_unlock(zgd->zgd_rl);
1823 ztest_object_unlock(zd, object);
1824
1825 if (error == 0 && zgd->zgd_bp)
1826 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
1827
1828 umem_free(zgd, sizeof (*zgd));
1829 }
1830
1831 static int
ztest_get_data(void * arg,lr_write_t * lr,char * buf,zio_t * zio)1832 ztest_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
1833 {
1834 ztest_ds_t *zd = arg;
1835 objset_t *os = zd->zd_os;
1836 uint64_t object = lr->lr_foid;
1837 uint64_t offset = lr->lr_offset;
1838 uint64_t size = lr->lr_length;
1839 blkptr_t *bp = &lr->lr_blkptr;
1840 uint64_t txg = lr->lr_common.lrc_txg;
1841 uint64_t crtxg;
1842 dmu_object_info_t doi;
1843 dmu_buf_t *db;
1844 zgd_t *zgd;
1845 int error;
1846
1847 ztest_object_lock(zd, object, RL_READER);
1848 error = dmu_bonus_hold(os, object, FTAG, &db);
1849 if (error) {
1850 ztest_object_unlock(zd, object);
1851 return (error);
1852 }
1853
1854 crtxg = ztest_bt_bonus(db)->bt_crtxg;
1855
1856 if (crtxg == 0 || crtxg > txg) {
1857 dmu_buf_rele(db, FTAG);
1858 ztest_object_unlock(zd, object);
1859 return (ENOENT);
1860 }
1861
1862 dmu_object_info_from_db(db, &doi);
1863 dmu_buf_rele(db, FTAG);
1864 db = NULL;
1865
1866 zgd = umem_zalloc(sizeof (*zgd), UMEM_NOFAIL);
1867 zgd->zgd_zilog = zd->zd_zilog;
1868 zgd->zgd_private = zd;
1869
1870 if (buf != NULL) { /* immediate write */
1871 zgd->zgd_rl = ztest_range_lock(zd, object, offset, size,
1872 RL_READER);
1873
1874 error = dmu_read(os, object, offset, size, buf,
1875 DMU_READ_NO_PREFETCH);
1876 ASSERT(error == 0);
1877 } else {
1878 size = doi.doi_data_block_size;
1879 if (ISP2(size)) {
1880 offset = P2ALIGN(offset, size);
1881 } else {
1882 ASSERT(offset < size);
1883 offset = 0;
1884 }
1885
1886 zgd->zgd_rl = ztest_range_lock(zd, object, offset, size,
1887 RL_READER);
1888
1889 error = dmu_buf_hold(os, object, offset, zgd, &db,
1890 DMU_READ_NO_PREFETCH);
1891
1892 if (error == 0) {
1893 blkptr_t *obp = dmu_buf_get_blkptr(db);
1894 if (obp) {
1895 ASSERT(BP_IS_HOLE(bp));
1896 *bp = *obp;
1897 }
1898
1899 zgd->zgd_db = db;
1900 zgd->zgd_bp = bp;
1901
1902 ASSERT(db->db_offset == offset);
1903 ASSERT(db->db_size == size);
1904
1905 error = dmu_sync(zio, lr->lr_common.lrc_txg,
1906 ztest_get_done, zgd);
1907
1908 if (error == 0)
1909 return (0);
1910 }
1911 }
1912
1913 ztest_get_done(zgd, error);
1914
1915 return (error);
1916 }
1917
1918 static void *
ztest_lr_alloc(size_t lrsize,char * name)1919 ztest_lr_alloc(size_t lrsize, char *name)
1920 {
1921 char *lr;
1922 size_t namesize = name ? strlen(name) + 1 : 0;
1923
1924 lr = umem_zalloc(lrsize + namesize, UMEM_NOFAIL);
1925
1926 if (name)
1927 bcopy(name, lr + lrsize, namesize);
1928
1929 return (lr);
1930 }
1931
1932 void
ztest_lr_free(void * lr,size_t lrsize,char * name)1933 ztest_lr_free(void *lr, size_t lrsize, char *name)
1934 {
1935 size_t namesize = name ? strlen(name) + 1 : 0;
1936
1937 umem_free(lr, lrsize + namesize);
1938 }
1939
1940 /*
1941 * Lookup a bunch of objects. Returns the number of objects not found.
1942 */
1943 static int
ztest_lookup(ztest_ds_t * zd,ztest_od_t * od,int count)1944 ztest_lookup(ztest_ds_t *zd, ztest_od_t *od, int count)
1945 {
1946 int missing = 0;
1947 int error;
1948
1949 ASSERT(_mutex_held(&zd->zd_dirobj_lock));
1950
1951 for (int i = 0; i < count; i++, od++) {
1952 od->od_object = 0;
1953 error = zap_lookup(zd->zd_os, od->od_dir, od->od_name,
1954 sizeof (uint64_t), 1, &od->od_object);
1955 if (error) {
1956 ASSERT(error == ENOENT);
1957 ASSERT(od->od_object == 0);
1958 missing++;
1959 } else {
1960 dmu_buf_t *db;
1961 ztest_block_tag_t *bbt;
1962 dmu_object_info_t doi;
1963
1964 ASSERT(od->od_object != 0);
1965 ASSERT(missing == 0); /* there should be no gaps */
1966
1967 ztest_object_lock(zd, od->od_object, RL_READER);
1968 VERIFY3U(0, ==, dmu_bonus_hold(zd->zd_os,
1969 od->od_object, FTAG, &db));
1970 dmu_object_info_from_db(db, &doi);
1971 bbt = ztest_bt_bonus(db);
1972 ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1973 od->od_type = doi.doi_type;
1974 od->od_blocksize = doi.doi_data_block_size;
1975 od->od_gen = bbt->bt_gen;
1976 dmu_buf_rele(db, FTAG);
1977 ztest_object_unlock(zd, od->od_object);
1978 }
1979 }
1980
1981 return (missing);
1982 }
1983
1984 static int
ztest_create(ztest_ds_t * zd,ztest_od_t * od,int count)1985 ztest_create(ztest_ds_t *zd, ztest_od_t *od, int count)
1986 {
1987 int missing = 0;
1988
1989 ASSERT(_mutex_held(&zd->zd_dirobj_lock));
1990
1991 for (int i = 0; i < count; i++, od++) {
1992 if (missing) {
1993 od->od_object = 0;
1994 missing++;
1995 continue;
1996 }
1997
1998 lr_create_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
1999
2000 lr->lr_doid = od->od_dir;
2001 lr->lr_foid = 0; /* 0 to allocate, > 0 to claim */
2002 lr->lrz_type = od->od_crtype;
2003 lr->lrz_blocksize = od->od_crblocksize;
2004 lr->lrz_ibshift = ztest_random_ibshift();
2005 lr->lrz_bonustype = DMU_OT_UINT64_OTHER;
2006 lr->lrz_bonuslen = dmu_bonus_max();
2007 lr->lr_gen = od->od_crgen;
2008 lr->lr_crtime[0] = time(NULL);
2009
2010 if (ztest_replay_create(zd, lr, B_FALSE) != 0) {
2011 ASSERT(missing == 0);
2012 od->od_object = 0;
2013 missing++;
2014 } else {
2015 od->od_object = lr->lr_foid;
2016 od->od_type = od->od_crtype;
2017 od->od_blocksize = od->od_crblocksize;
2018 od->od_gen = od->od_crgen;
2019 ASSERT(od->od_object != 0);
2020 }
2021
2022 ztest_lr_free(lr, sizeof (*lr), od->od_name);
2023 }
2024
2025 return (missing);
2026 }
2027
2028 static int
ztest_remove(ztest_ds_t * zd,ztest_od_t * od,int count)2029 ztest_remove(ztest_ds_t *zd, ztest_od_t *od, int count)
2030 {
2031 int missing = 0;
2032 int error;
2033
2034 ASSERT(_mutex_held(&zd->zd_dirobj_lock));
2035
2036 od += count - 1;
2037
2038 for (int i = count - 1; i >= 0; i--, od--) {
2039 if (missing) {
2040 missing++;
2041 continue;
2042 }
2043
2044 /*
2045 * No object was found.
2046 */
2047 if (od->od_object == 0)
2048 continue;
2049
2050 lr_remove_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
2051
2052 lr->lr_doid = od->od_dir;
2053
2054 if ((error = ztest_replay_remove(zd, lr, B_FALSE)) != 0) {
2055 ASSERT3U(error, ==, ENOSPC);
2056 missing++;
2057 } else {
2058 od->od_object = 0;
2059 }
2060 ztest_lr_free(lr, sizeof (*lr), od->od_name);
2061 }
2062
2063 return (missing);
2064 }
2065
2066 static int
ztest_write(ztest_ds_t * zd,uint64_t object,uint64_t offset,uint64_t size,void * data)2067 ztest_write(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size,
2068 void *data)
2069 {
2070 lr_write_t *lr;
2071 int error;
2072
2073 lr = ztest_lr_alloc(sizeof (*lr) + size, NULL);
2074
2075 lr->lr_foid = object;
2076 lr->lr_offset = offset;
2077 lr->lr_length = size;
2078 lr->lr_blkoff = 0;
2079 BP_ZERO(&lr->lr_blkptr);
2080
2081 bcopy(data, lr + 1, size);
2082
2083 error = ztest_replay_write(zd, lr, B_FALSE);
2084
2085 ztest_lr_free(lr, sizeof (*lr) + size, NULL);
2086
2087 return (error);
2088 }
2089
2090 static int
ztest_truncate(ztest_ds_t * zd,uint64_t object,uint64_t offset,uint64_t size)2091 ztest_truncate(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2092 {
2093 lr_truncate_t *lr;
2094 int error;
2095
2096 lr = ztest_lr_alloc(sizeof (*lr), NULL);
2097
2098 lr->lr_foid = object;
2099 lr->lr_offset = offset;
2100 lr->lr_length = size;
2101
2102 error = ztest_replay_truncate(zd, lr, B_FALSE);
2103
2104 ztest_lr_free(lr, sizeof (*lr), NULL);
2105
2106 return (error);
2107 }
2108
2109 static int
ztest_setattr(ztest_ds_t * zd,uint64_t object)2110 ztest_setattr(ztest_ds_t *zd, uint64_t object)
2111 {
2112 lr_setattr_t *lr;
2113 int error;
2114
2115 lr = ztest_lr_alloc(sizeof (*lr), NULL);
2116
2117 lr->lr_foid = object;
2118 lr->lr_size = 0;
2119 lr->lr_mode = 0;
2120
2121 error = ztest_replay_setattr(zd, lr, B_FALSE);
2122
2123 ztest_lr_free(lr, sizeof (*lr), NULL);
2124
2125 return (error);
2126 }
2127
2128 static void
ztest_prealloc(ztest_ds_t * zd,uint64_t object,uint64_t offset,uint64_t size)2129 ztest_prealloc(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2130 {
2131 objset_t *os = zd->zd_os;
2132 dmu_tx_t *tx;
2133 uint64_t txg;
2134 rl_t *rl;
2135
2136 txg_wait_synced(dmu_objset_pool(os), 0);
2137
2138 ztest_object_lock(zd, object, RL_READER);
2139 rl = ztest_range_lock(zd, object, offset, size, RL_WRITER);
2140
2141 tx = dmu_tx_create(os);
2142
2143 dmu_tx_hold_write(tx, object, offset, size);
2144
2145 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
2146
2147 if (txg != 0) {
2148 dmu_prealloc(os, object, offset, size, tx);
2149 dmu_tx_commit(tx);
2150 txg_wait_synced(dmu_objset_pool(os), txg);
2151 } else {
2152 (void) dmu_free_long_range(os, object, offset, size);
2153 }
2154
2155 ztest_range_unlock(rl);
2156 ztest_object_unlock(zd, object);
2157 }
2158
2159 static void
ztest_io(ztest_ds_t * zd,uint64_t object,uint64_t offset)2160 ztest_io(ztest_ds_t *zd, uint64_t object, uint64_t offset)
2161 {
2162 int err;
2163 ztest_block_tag_t wbt;
2164 dmu_object_info_t doi;
2165 enum ztest_io_type io_type;
2166 uint64_t blocksize;
2167 void *data;
2168
2169 VERIFY(dmu_object_info(zd->zd_os, object, &doi) == 0);
2170 blocksize = doi.doi_data_block_size;
2171 data = umem_alloc(blocksize, UMEM_NOFAIL);
2172
2173 /*
2174 * Pick an i/o type at random, biased toward writing block tags.
2175 */
2176 io_type = ztest_random(ZTEST_IO_TYPES);
2177 if (ztest_random(2) == 0)
2178 io_type = ZTEST_IO_WRITE_TAG;
2179
2180 (void) rw_rdlock(&zd->zd_zilog_lock);
2181
2182 switch (io_type) {
2183
2184 case ZTEST_IO_WRITE_TAG:
2185 ztest_bt_generate(&wbt, zd->zd_os, object, offset, 0, 0, 0);
2186 (void) ztest_write(zd, object, offset, sizeof (wbt), &wbt);
2187 break;
2188
2189 case ZTEST_IO_WRITE_PATTERN:
2190 (void) memset(data, 'a' + (object + offset) % 5, blocksize);
2191 if (ztest_random(2) == 0) {
2192 /*
2193 * Induce fletcher2 collisions to ensure that
2194 * zio_ddt_collision() detects and resolves them
2195 * when using fletcher2-verify for deduplication.
2196 */
2197 ((uint64_t *)data)[0] ^= 1ULL << 63;
2198 ((uint64_t *)data)[4] ^= 1ULL << 63;
2199 }
2200 (void) ztest_write(zd, object, offset, blocksize, data);
2201 break;
2202
2203 case ZTEST_IO_WRITE_ZEROES:
2204 bzero(data, blocksize);
2205 (void) ztest_write(zd, object, offset, blocksize, data);
2206 break;
2207
2208 case ZTEST_IO_TRUNCATE:
2209 (void) ztest_truncate(zd, object, offset, blocksize);
2210 break;
2211
2212 case ZTEST_IO_SETATTR:
2213 (void) ztest_setattr(zd, object);
2214 break;
2215
2216 case ZTEST_IO_REWRITE:
2217 (void) rw_rdlock(&ztest_name_lock);
2218 err = ztest_dsl_prop_set_uint64(zd->zd_name,
2219 ZFS_PROP_CHECKSUM, spa_dedup_checksum(ztest_spa),
2220 B_FALSE);
2221 VERIFY(err == 0 || err == ENOSPC);
2222 err = ztest_dsl_prop_set_uint64(zd->zd_name,
2223 ZFS_PROP_COMPRESSION,
2224 ztest_random_dsl_prop(ZFS_PROP_COMPRESSION),
2225 B_FALSE);
2226 VERIFY(err == 0 || err == ENOSPC);
2227 (void) rw_unlock(&ztest_name_lock);
2228
2229 VERIFY0(dmu_read(zd->zd_os, object, offset, blocksize, data,
2230 DMU_READ_NO_PREFETCH));
2231
2232 (void) ztest_write(zd, object, offset, blocksize, data);
2233 break;
2234 }
2235
2236 (void) rw_unlock(&zd->zd_zilog_lock);
2237
2238 umem_free(data, blocksize);
2239 }
2240
2241 /*
2242 * Initialize an object description template.
2243 */
2244 static void
ztest_od_init(ztest_od_t * od,uint64_t id,char * tag,uint64_t index,dmu_object_type_t type,uint64_t blocksize,uint64_t gen)2245 ztest_od_init(ztest_od_t *od, uint64_t id, char *tag, uint64_t index,
2246 dmu_object_type_t type, uint64_t blocksize, uint64_t gen)
2247 {
2248 od->od_dir = ZTEST_DIROBJ;
2249 od->od_object = 0;
2250
2251 od->od_crtype = type;
2252 od->od_crblocksize = blocksize ? blocksize : ztest_random_blocksize();
2253 od->od_crgen = gen;
2254
2255 od->od_type = DMU_OT_NONE;
2256 od->od_blocksize = 0;
2257 od->od_gen = 0;
2258
2259 (void) snprintf(od->od_name, sizeof (od->od_name), "%s(%lld)[%llu]",
2260 tag, (int64_t)id, index);
2261 }
2262
2263 /*
2264 * Lookup or create the objects for a test using the od template.
2265 * If the objects do not all exist, or if 'remove' is specified,
2266 * remove any existing objects and create new ones. Otherwise,
2267 * use the existing objects.
2268 */
2269 static int
ztest_object_init(ztest_ds_t * zd,ztest_od_t * od,size_t size,boolean_t remove)2270 ztest_object_init(ztest_ds_t *zd, ztest_od_t *od, size_t size, boolean_t remove)
2271 {
2272 int count = size / sizeof (*od);
2273 int rv = 0;
2274
2275 VERIFY(mutex_lock(&zd->zd_dirobj_lock) == 0);
2276 if ((ztest_lookup(zd, od, count) != 0 || remove) &&
2277 (ztest_remove(zd, od, count) != 0 ||
2278 ztest_create(zd, od, count) != 0))
2279 rv = -1;
2280 zd->zd_od = od;
2281 VERIFY(mutex_unlock(&zd->zd_dirobj_lock) == 0);
2282
2283 return (rv);
2284 }
2285
2286 /* ARGSUSED */
2287 void
ztest_zil_commit(ztest_ds_t * zd,uint64_t id)2288 ztest_zil_commit(ztest_ds_t *zd, uint64_t id)
2289 {
2290 zilog_t *zilog = zd->zd_zilog;
2291
2292 (void) rw_rdlock(&zd->zd_zilog_lock);
2293
2294 zil_commit(zilog, ztest_random(ZTEST_OBJECTS));
2295
2296 /*
2297 * Remember the committed values in zd, which is in parent/child
2298 * shared memory. If we die, the next iteration of ztest_run()
2299 * will verify that the log really does contain this record.
2300 */
2301 mutex_enter(&zilog->zl_lock);
2302 ASSERT(zd->zd_shared != NULL);
2303 ASSERT3U(zd->zd_shared->zd_seq, <=, zilog->zl_commit_lr_seq);
2304 zd->zd_shared->zd_seq = zilog->zl_commit_lr_seq;
2305 mutex_exit(&zilog->zl_lock);
2306
2307 (void) rw_unlock(&zd->zd_zilog_lock);
2308 }
2309
2310 /*
2311 * This function is designed to simulate the operations that occur during a
2312 * mount/unmount operation. We hold the dataset across these operations in an
2313 * attempt to expose any implicit assumptions about ZIL management.
2314 */
2315 /* ARGSUSED */
2316 void
ztest_zil_remount(ztest_ds_t * zd,uint64_t id)2317 ztest_zil_remount(ztest_ds_t *zd, uint64_t id)
2318 {
2319 objset_t *os = zd->zd_os;
2320
2321 /*
2322 * We grab the zd_dirobj_lock to ensure that no other thread is
2323 * updating the zil (i.e. adding in-memory log records) and the
2324 * zd_zilog_lock to block any I/O.
2325 */
2326 VERIFY0(mutex_lock(&zd->zd_dirobj_lock));
2327 (void) rw_wrlock(&zd->zd_zilog_lock);
2328
2329 /* zfsvfs_teardown() */
2330 zil_close(zd->zd_zilog);
2331
2332 /* zfsvfs_setup() */
2333 VERIFY(zil_open(os, ztest_get_data) == zd->zd_zilog);
2334 zil_replay(os, zd, ztest_replay_vector);
2335
2336 (void) rw_unlock(&zd->zd_zilog_lock);
2337 VERIFY(mutex_unlock(&zd->zd_dirobj_lock) == 0);
2338 }
2339
2340 /*
2341 * Verify that we can't destroy an active pool, create an existing pool,
2342 * or create a pool with a bad vdev spec.
2343 */
2344 /* ARGSUSED */
2345 void
ztest_spa_create_destroy(ztest_ds_t * zd,uint64_t id)2346 ztest_spa_create_destroy(ztest_ds_t *zd, uint64_t id)
2347 {
2348 ztest_shared_opts_t *zo = &ztest_opts;
2349 spa_t *spa;
2350 nvlist_t *nvroot;
2351
2352 /*
2353 * Attempt to create using a bad file.
2354 */
2355 nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1);
2356 VERIFY3U(ENOENT, ==,
2357 spa_create("ztest_bad_file", nvroot, NULL, NULL));
2358 nvlist_free(nvroot);
2359
2360 /*
2361 * Attempt to create using a bad mirror.
2362 */
2363 nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 2, 1);
2364 VERIFY3U(ENOENT, ==,
2365 spa_create("ztest_bad_mirror", nvroot, NULL, NULL));
2366 nvlist_free(nvroot);
2367
2368 /*
2369 * Attempt to create an existing pool. It shouldn't matter
2370 * what's in the nvroot; we should fail with EEXIST.
2371 */
2372 (void) rw_rdlock(&ztest_name_lock);
2373 nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1);
2374 VERIFY3U(EEXIST, ==, spa_create(zo->zo_pool, nvroot, NULL, NULL));
2375 nvlist_free(nvroot);
2376 VERIFY3U(0, ==, spa_open(zo->zo_pool, &spa, FTAG));
2377 VERIFY3U(EBUSY, ==, spa_destroy(zo->zo_pool));
2378 spa_close(spa, FTAG);
2379
2380 (void) rw_unlock(&ztest_name_lock);
2381 }
2382
2383 /* ARGSUSED */
2384 void
ztest_spa_upgrade(ztest_ds_t * zd,uint64_t id)2385 ztest_spa_upgrade(ztest_ds_t *zd, uint64_t id)
2386 {
2387 spa_t *spa;
2388 uint64_t initial_version = SPA_VERSION_INITIAL;
2389 uint64_t version, newversion;
2390 nvlist_t *nvroot, *props;
2391 char *name;
2392
2393 VERIFY0(mutex_lock(&ztest_vdev_lock));
2394 name = kmem_asprintf("%s_upgrade", ztest_opts.zo_pool);
2395
2396 /*
2397 * Clean up from previous runs.
2398 */
2399 (void) spa_destroy(name);
2400
2401 nvroot = make_vdev_root(NULL, NULL, name, ztest_opts.zo_vdev_size, 0,
2402 0, ztest_opts.zo_raidz, ztest_opts.zo_mirrors, 1);
2403
2404 /*
2405 * If we're configuring a RAIDZ device then make sure that the
2406 * the initial version is capable of supporting that feature.
2407 */
2408 switch (ztest_opts.zo_raidz_parity) {
2409 case 0:
2410 case 1:
2411 initial_version = SPA_VERSION_INITIAL;
2412 break;
2413 case 2:
2414 initial_version = SPA_VERSION_RAIDZ2;
2415 break;
2416 case 3:
2417 initial_version = SPA_VERSION_RAIDZ3;
2418 break;
2419 }
2420
2421 /*
2422 * Create a pool with a spa version that can be upgraded. Pick
2423 * a value between initial_version and SPA_VERSION_BEFORE_FEATURES.
2424 */
2425 do {
2426 version = ztest_random_spa_version(initial_version);
2427 } while (version > SPA_VERSION_BEFORE_FEATURES);
2428
2429 props = fnvlist_alloc();
2430 fnvlist_add_uint64(props,
2431 zpool_prop_to_name(ZPOOL_PROP_VERSION), version);
2432 VERIFY0(spa_create(name, nvroot, props, NULL));
2433 fnvlist_free(nvroot);
2434 fnvlist_free(props);
2435
2436 VERIFY0(spa_open(name, &spa, FTAG));
2437 VERIFY3U(spa_version(spa), ==, version);
2438 newversion = ztest_random_spa_version(version + 1);
2439
2440 if (ztest_opts.zo_verbose >= 4) {
2441 (void) printf("upgrading spa version from %llu to %llu\n",
2442 (u_longlong_t)version, (u_longlong_t)newversion);
2443 }
2444
2445 spa_upgrade(spa, newversion);
2446 VERIFY3U(spa_version(spa), >, version);
2447 VERIFY3U(spa_version(spa), ==, fnvlist_lookup_uint64(spa->spa_config,
2448 zpool_prop_to_name(ZPOOL_PROP_VERSION)));
2449 spa_close(spa, FTAG);
2450
2451 strfree(name);
2452 VERIFY0(mutex_unlock(&ztest_vdev_lock));
2453 }
2454
2455 static vdev_t *
vdev_lookup_by_path(vdev_t * vd,const char * path)2456 vdev_lookup_by_path(vdev_t *vd, const char *path)
2457 {
2458 vdev_t *mvd;
2459
2460 if (vd->vdev_path != NULL && strcmp(path, vd->vdev_path) == 0)
2461 return (vd);
2462
2463 for (int c = 0; c < vd->vdev_children; c++)
2464 if ((mvd = vdev_lookup_by_path(vd->vdev_child[c], path)) !=
2465 NULL)
2466 return (mvd);
2467
2468 return (NULL);
2469 }
2470
2471 /*
2472 * Find the first available hole which can be used as a top-level.
2473 */
2474 int
find_vdev_hole(spa_t * spa)2475 find_vdev_hole(spa_t *spa)
2476 {
2477 vdev_t *rvd = spa->spa_root_vdev;
2478 int c;
2479
2480 ASSERT(spa_config_held(spa, SCL_VDEV, RW_READER) == SCL_VDEV);
2481
2482 for (c = 0; c < rvd->vdev_children; c++) {
2483 vdev_t *cvd = rvd->vdev_child[c];
2484
2485 if (cvd->vdev_ishole)
2486 break;
2487 }
2488 return (c);
2489 }
2490
2491 /*
2492 * Verify that vdev_add() works as expected.
2493 */
2494 /* ARGSUSED */
2495 void
ztest_vdev_add_remove(ztest_ds_t * zd,uint64_t id)2496 ztest_vdev_add_remove(ztest_ds_t *zd, uint64_t id)
2497 {
2498 ztest_shared_t *zs = ztest_shared;
2499 spa_t *spa = ztest_spa;
2500 uint64_t leaves;
2501 uint64_t guid;
2502 nvlist_t *nvroot;
2503 int error;
2504
2505 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2506 leaves = MAX(zs->zs_mirrors + zs->zs_splits, 1) * ztest_opts.zo_raidz;
2507
2508 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2509
2510 ztest_shared->zs_vdev_next_leaf = find_vdev_hole(spa) * leaves;
2511
2512 /*
2513 * If we have slogs then remove them 1/4 of the time.
2514 */
2515 if (spa_has_slogs(spa) && ztest_random(4) == 0) {
2516 /*
2517 * Grab the guid from the head of the log class rotor.
2518 */
2519 guid = spa_log_class(spa)->mc_rotor->mg_vd->vdev_guid;
2520
2521 spa_config_exit(spa, SCL_VDEV, FTAG);
2522
2523 /*
2524 * We have to grab the zs_name_lock as writer to
2525 * prevent a race between removing a slog (dmu_objset_find)
2526 * and destroying a dataset. Removing the slog will
2527 * grab a reference on the dataset which may cause
2528 * dmu_objset_destroy() to fail with EBUSY thus
2529 * leaving the dataset in an inconsistent state.
2530 */
2531 VERIFY(rw_wrlock(&ztest_name_lock) == 0);
2532 error = spa_vdev_remove(spa, guid, B_FALSE);
2533 VERIFY(rw_unlock(&ztest_name_lock) == 0);
2534
2535 if (error && error != EEXIST)
2536 fatal(0, "spa_vdev_remove() = %d", error);
2537 } else {
2538 spa_config_exit(spa, SCL_VDEV, FTAG);
2539
2540 /*
2541 * Make 1/4 of the devices be log devices.
2542 */
2543 nvroot = make_vdev_root(NULL, NULL, NULL,
2544 ztest_opts.zo_vdev_size, 0,
2545 ztest_random(4) == 0, ztest_opts.zo_raidz,
2546 zs->zs_mirrors, 1);
2547
2548 error = spa_vdev_add(spa, nvroot);
2549 nvlist_free(nvroot);
2550
2551 if (error == ENOSPC)
2552 ztest_record_enospc("spa_vdev_add");
2553 else if (error != 0)
2554 fatal(0, "spa_vdev_add() = %d", error);
2555 }
2556
2557 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2558 }
2559
2560 /*
2561 * Verify that adding/removing aux devices (l2arc, hot spare) works as expected.
2562 */
2563 /* ARGSUSED */
2564 void
ztest_vdev_aux_add_remove(ztest_ds_t * zd,uint64_t id)2565 ztest_vdev_aux_add_remove(ztest_ds_t *zd, uint64_t id)
2566 {
2567 ztest_shared_t *zs = ztest_shared;
2568 spa_t *spa = ztest_spa;
2569 vdev_t *rvd = spa->spa_root_vdev;
2570 spa_aux_vdev_t *sav;
2571 char *aux;
2572 uint64_t guid = 0;
2573 int error;
2574
2575 if (ztest_random(2) == 0) {
2576 sav = &spa->spa_spares;
2577 aux = ZPOOL_CONFIG_SPARES;
2578 } else {
2579 sav = &spa->spa_l2cache;
2580 aux = ZPOOL_CONFIG_L2CACHE;
2581 }
2582
2583 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2584
2585 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2586
2587 if (sav->sav_count != 0 && ztest_random(4) == 0) {
2588 /*
2589 * Pick a random device to remove.
2590 */
2591 guid = sav->sav_vdevs[ztest_random(sav->sav_count)]->vdev_guid;
2592 } else {
2593 /*
2594 * Find an unused device we can add.
2595 */
2596 zs->zs_vdev_aux = 0;
2597 for (;;) {
2598 char path[MAXPATHLEN];
2599 int c;
2600 (void) snprintf(path, sizeof (path), ztest_aux_template,
2601 ztest_opts.zo_dir, ztest_opts.zo_pool, aux,
2602 zs->zs_vdev_aux);
2603 for (c = 0; c < sav->sav_count; c++)
2604 if (strcmp(sav->sav_vdevs[c]->vdev_path,
2605 path) == 0)
2606 break;
2607 if (c == sav->sav_count &&
2608 vdev_lookup_by_path(rvd, path) == NULL)
2609 break;
2610 zs->zs_vdev_aux++;
2611 }
2612 }
2613
2614 spa_config_exit(spa, SCL_VDEV, FTAG);
2615
2616 if (guid == 0) {
2617 /*
2618 * Add a new device.
2619 */
2620 nvlist_t *nvroot = make_vdev_root(NULL, aux, NULL,
2621 (ztest_opts.zo_vdev_size * 5) / 4, 0, 0, 0, 0, 1);
2622 error = spa_vdev_add(spa, nvroot);
2623 if (error != 0)
2624 fatal(0, "spa_vdev_add(%p) = %d", nvroot, error);
2625 nvlist_free(nvroot);
2626 } else {
2627 /*
2628 * Remove an existing device. Sometimes, dirty its
2629 * vdev state first to make sure we handle removal
2630 * of devices that have pending state changes.
2631 */
2632 if (ztest_random(2) == 0)
2633 (void) vdev_online(spa, guid, 0, NULL);
2634
2635 error = spa_vdev_remove(spa, guid, B_FALSE);
2636 if (error != 0 && error != EBUSY)
2637 fatal(0, "spa_vdev_remove(%llu) = %d", guid, error);
2638 }
2639
2640 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2641 }
2642
2643 /*
2644 * split a pool if it has mirror tlvdevs
2645 */
2646 /* ARGSUSED */
2647 void
ztest_split_pool(ztest_ds_t * zd,uint64_t id)2648 ztest_split_pool(ztest_ds_t *zd, uint64_t id)
2649 {
2650 ztest_shared_t *zs = ztest_shared;
2651 spa_t *spa = ztest_spa;
2652 vdev_t *rvd = spa->spa_root_vdev;
2653 nvlist_t *tree, **child, *config, *split, **schild;
2654 uint_t c, children, schildren = 0, lastlogid = 0;
2655 int error = 0;
2656
2657 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2658
2659 /* ensure we have a useable config; mirrors of raidz aren't supported */
2660 if (zs->zs_mirrors < 3 || ztest_opts.zo_raidz > 1) {
2661 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2662 return;
2663 }
2664
2665 /* clean up the old pool, if any */
2666 (void) spa_destroy("splitp");
2667
2668 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2669
2670 /* generate a config from the existing config */
2671 mutex_enter(&spa->spa_props_lock);
2672 VERIFY(nvlist_lookup_nvlist(spa->spa_config, ZPOOL_CONFIG_VDEV_TREE,
2673 &tree) == 0);
2674 mutex_exit(&spa->spa_props_lock);
2675
2676 VERIFY(nvlist_lookup_nvlist_array(tree, ZPOOL_CONFIG_CHILDREN, &child,
2677 &children) == 0);
2678
2679 schild = malloc(rvd->vdev_children * sizeof (nvlist_t *));
2680 for (c = 0; c < children; c++) {
2681 vdev_t *tvd = rvd->vdev_child[c];
2682 nvlist_t **mchild;
2683 uint_t mchildren;
2684
2685 if (tvd->vdev_islog || tvd->vdev_ops == &vdev_hole_ops) {
2686 VERIFY(nvlist_alloc(&schild[schildren], NV_UNIQUE_NAME,
2687 0) == 0);
2688 VERIFY(nvlist_add_string(schild[schildren],
2689 ZPOOL_CONFIG_TYPE, VDEV_TYPE_HOLE) == 0);
2690 VERIFY(nvlist_add_uint64(schild[schildren],
2691 ZPOOL_CONFIG_IS_HOLE, 1) == 0);
2692 if (lastlogid == 0)
2693 lastlogid = schildren;
2694 ++schildren;
2695 continue;
2696 }
2697 lastlogid = 0;
2698 VERIFY(nvlist_lookup_nvlist_array(child[c],
2699 ZPOOL_CONFIG_CHILDREN, &mchild, &mchildren) == 0);
2700 VERIFY(nvlist_dup(mchild[0], &schild[schildren++], 0) == 0);
2701 }
2702
2703 /* OK, create a config that can be used to split */
2704 VERIFY(nvlist_alloc(&split, NV_UNIQUE_NAME, 0) == 0);
2705 VERIFY(nvlist_add_string(split, ZPOOL_CONFIG_TYPE,
2706 VDEV_TYPE_ROOT) == 0);
2707 VERIFY(nvlist_add_nvlist_array(split, ZPOOL_CONFIG_CHILDREN, schild,
2708 lastlogid != 0 ? lastlogid : schildren) == 0);
2709
2710 VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, 0) == 0);
2711 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, split) == 0);
2712
2713 for (c = 0; c < schildren; c++)
2714 nvlist_free(schild[c]);
2715 free(schild);
2716 nvlist_free(split);
2717
2718 spa_config_exit(spa, SCL_VDEV, FTAG);
2719
2720 (void) rw_wrlock(&ztest_name_lock);
2721 error = spa_vdev_split_mirror(spa, "splitp", config, NULL, B_FALSE);
2722 (void) rw_unlock(&ztest_name_lock);
2723
2724 nvlist_free(config);
2725
2726 if (error == 0) {
2727 (void) printf("successful split - results:\n");
2728 mutex_enter(&spa_namespace_lock);
2729 show_pool_stats(spa);
2730 show_pool_stats(spa_lookup("splitp"));
2731 mutex_exit(&spa_namespace_lock);
2732 ++zs->zs_splits;
2733 --zs->zs_mirrors;
2734 }
2735 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2736
2737 }
2738
2739 /*
2740 * Verify that we can attach and detach devices.
2741 */
2742 /* ARGSUSED */
2743 void
ztest_vdev_attach_detach(ztest_ds_t * zd,uint64_t id)2744 ztest_vdev_attach_detach(ztest_ds_t *zd, uint64_t id)
2745 {
2746 ztest_shared_t *zs = ztest_shared;
2747 spa_t *spa = ztest_spa;
2748 spa_aux_vdev_t *sav = &spa->spa_spares;
2749 vdev_t *rvd = spa->spa_root_vdev;
2750 vdev_t *oldvd, *newvd, *pvd;
2751 nvlist_t *root;
2752 uint64_t leaves;
2753 uint64_t leaf, top;
2754 uint64_t ashift = ztest_get_ashift();
2755 uint64_t oldguid, pguid;
2756 uint64_t oldsize, newsize;
2757 char oldpath[MAXPATHLEN], newpath[MAXPATHLEN];
2758 int replacing;
2759 int oldvd_has_siblings = B_FALSE;
2760 int newvd_is_spare = B_FALSE;
2761 int oldvd_is_log;
2762 int error, expected_error;
2763
2764 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2765 leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz;
2766
2767 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2768
2769 /*
2770 * Decide whether to do an attach or a replace.
2771 */
2772 replacing = ztest_random(2);
2773
2774 /*
2775 * Pick a random top-level vdev.
2776 */
2777 top = ztest_random_vdev_top(spa, B_TRUE);
2778
2779 /*
2780 * Pick a random leaf within it.
2781 */
2782 leaf = ztest_random(leaves);
2783
2784 /*
2785 * Locate this vdev.
2786 */
2787 oldvd = rvd->vdev_child[top];
2788 if (zs->zs_mirrors >= 1) {
2789 ASSERT(oldvd->vdev_ops == &vdev_mirror_ops);
2790 ASSERT(oldvd->vdev_children >= zs->zs_mirrors);
2791 oldvd = oldvd->vdev_child[leaf / ztest_opts.zo_raidz];
2792 }
2793 if (ztest_opts.zo_raidz > 1) {
2794 ASSERT(oldvd->vdev_ops == &vdev_raidz_ops);
2795 ASSERT(oldvd->vdev_children == ztest_opts.zo_raidz);
2796 oldvd = oldvd->vdev_child[leaf % ztest_opts.zo_raidz];
2797 }
2798
2799 /*
2800 * If we're already doing an attach or replace, oldvd may be a
2801 * mirror vdev -- in which case, pick a random child.
2802 */
2803 while (oldvd->vdev_children != 0) {
2804 oldvd_has_siblings = B_TRUE;
2805 ASSERT(oldvd->vdev_children >= 2);
2806 oldvd = oldvd->vdev_child[ztest_random(oldvd->vdev_children)];
2807 }
2808
2809 oldguid = oldvd->vdev_guid;
2810 oldsize = vdev_get_min_asize(oldvd);
2811 oldvd_is_log = oldvd->vdev_top->vdev_islog;
2812 (void) strcpy(oldpath, oldvd->vdev_path);
2813 pvd = oldvd->vdev_parent;
2814 pguid = pvd->vdev_guid;
2815
2816 /*
2817 * If oldvd has siblings, then half of the time, detach it.
2818 */
2819 if (oldvd_has_siblings && ztest_random(2) == 0) {
2820 spa_config_exit(spa, SCL_VDEV, FTAG);
2821 error = spa_vdev_detach(spa, oldguid, pguid, B_FALSE);
2822 if (error != 0 && error != ENODEV && error != EBUSY &&
2823 error != ENOTSUP)
2824 fatal(0, "detach (%s) returned %d", oldpath, error);
2825 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2826 return;
2827 }
2828
2829 /*
2830 * For the new vdev, choose with equal probability between the two
2831 * standard paths (ending in either 'a' or 'b') or a random hot spare.
2832 */
2833 if (sav->sav_count != 0 && ztest_random(3) == 0) {
2834 newvd = sav->sav_vdevs[ztest_random(sav->sav_count)];
2835 newvd_is_spare = B_TRUE;
2836 (void) strcpy(newpath, newvd->vdev_path);
2837 } else {
2838 (void) snprintf(newpath, sizeof (newpath), ztest_dev_template,
2839 ztest_opts.zo_dir, ztest_opts.zo_pool,
2840 top * leaves + leaf);
2841 if (ztest_random(2) == 0)
2842 newpath[strlen(newpath) - 1] = 'b';
2843 newvd = vdev_lookup_by_path(rvd, newpath);
2844 }
2845
2846 if (newvd) {
2847 newsize = vdev_get_min_asize(newvd);
2848 } else {
2849 /*
2850 * Make newsize a little bigger or smaller than oldsize.
2851 * If it's smaller, the attach should fail.
2852 * If it's larger, and we're doing a replace,
2853 * we should get dynamic LUN growth when we're done.
2854 */
2855 newsize = 10 * oldsize / (9 + ztest_random(3));
2856 }
2857
2858 /*
2859 * If pvd is not a mirror or root, the attach should fail with ENOTSUP,
2860 * unless it's a replace; in that case any non-replacing parent is OK.
2861 *
2862 * If newvd is already part of the pool, it should fail with EBUSY.
2863 *
2864 * If newvd is too small, it should fail with EOVERFLOW.
2865 */
2866 if (pvd->vdev_ops != &vdev_mirror_ops &&
2867 pvd->vdev_ops != &vdev_root_ops && (!replacing ||
2868 pvd->vdev_ops == &vdev_replacing_ops ||
2869 pvd->vdev_ops == &vdev_spare_ops))
2870 expected_error = ENOTSUP;
2871 else if (newvd_is_spare && (!replacing || oldvd_is_log))
2872 expected_error = ENOTSUP;
2873 else if (newvd == oldvd)
2874 expected_error = replacing ? 0 : EBUSY;
2875 else if (vdev_lookup_by_path(rvd, newpath) != NULL)
2876 expected_error = EBUSY;
2877 else if (newsize < oldsize)
2878 expected_error = EOVERFLOW;
2879 else if (ashift > oldvd->vdev_top->vdev_ashift)
2880 expected_error = EDOM;
2881 else
2882 expected_error = 0;
2883
2884 spa_config_exit(spa, SCL_VDEV, FTAG);
2885
2886 /*
2887 * Build the nvlist describing newpath.
2888 */
2889 root = make_vdev_root(newpath, NULL, NULL, newvd == NULL ? newsize : 0,
2890 ashift, 0, 0, 0, 1);
2891
2892 error = spa_vdev_attach(spa, oldguid, root, replacing);
2893
2894 nvlist_free(root);
2895
2896 /*
2897 * If our parent was the replacing vdev, but the replace completed,
2898 * then instead of failing with ENOTSUP we may either succeed,
2899 * fail with ENODEV, or fail with EOVERFLOW.
2900 */
2901 if (expected_error == ENOTSUP &&
2902 (error == 0 || error == ENODEV || error == EOVERFLOW))
2903 expected_error = error;
2904
2905 /*
2906 * If someone grew the LUN, the replacement may be too small.
2907 */
2908 if (error == EOVERFLOW || error == EBUSY)
2909 expected_error = error;
2910
2911 /* XXX workaround 6690467 */
2912 if (error != expected_error && expected_error != EBUSY) {
2913 fatal(0, "attach (%s %llu, %s %llu, %d) "
2914 "returned %d, expected %d",
2915 oldpath, oldsize, newpath,
2916 newsize, replacing, error, expected_error);
2917 }
2918
2919 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2920 }
2921
2922 /*
2923 * Callback function which expands the physical size of the vdev.
2924 */
2925 vdev_t *
grow_vdev(vdev_t * vd,void * arg)2926 grow_vdev(vdev_t *vd, void *arg)
2927 {
2928 spa_t *spa = vd->vdev_spa;
2929 size_t *newsize = arg;
2930 size_t fsize;
2931 int fd;
2932
2933 ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
2934 ASSERT(vd->vdev_ops->vdev_op_leaf);
2935
2936 if ((fd = open(vd->vdev_path, O_RDWR)) == -1)
2937 return (vd);
2938
2939 fsize = lseek(fd, 0, SEEK_END);
2940 (void) ftruncate(fd, *newsize);
2941
2942 if (ztest_opts.zo_verbose >= 6) {
2943 (void) printf("%s grew from %lu to %lu bytes\n",
2944 vd->vdev_path, (ulong_t)fsize, (ulong_t)*newsize);
2945 }
2946 (void) close(fd);
2947 return (NULL);
2948 }
2949
2950 /*
2951 * Callback function which expands a given vdev by calling vdev_online().
2952 */
2953 /* ARGSUSED */
2954 vdev_t *
online_vdev(vdev_t * vd,void * arg)2955 online_vdev(vdev_t *vd, void *arg)
2956 {
2957 spa_t *spa = vd->vdev_spa;
2958 vdev_t *tvd = vd->vdev_top;
2959 uint64_t guid = vd->vdev_guid;
2960 uint64_t generation = spa->spa_config_generation + 1;
2961 vdev_state_t newstate = VDEV_STATE_UNKNOWN;
2962 int error;
2963
2964 ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
2965 ASSERT(vd->vdev_ops->vdev_op_leaf);
2966
2967 /* Calling vdev_online will initialize the new metaslabs */
2968 spa_config_exit(spa, SCL_STATE, spa);
2969 error = vdev_online(spa, guid, ZFS_ONLINE_EXPAND, &newstate);
2970 spa_config_enter(spa, SCL_STATE, spa, RW_READER);
2971
2972 /*
2973 * If vdev_online returned an error or the underlying vdev_open
2974 * failed then we abort the expand. The only way to know that
2975 * vdev_open fails is by checking the returned newstate.
2976 */
2977 if (error || newstate != VDEV_STATE_HEALTHY) {
2978 if (ztest_opts.zo_verbose >= 5) {
2979 (void) printf("Unable to expand vdev, state %llu, "
2980 "error %d\n", (u_longlong_t)newstate, error);
2981 }
2982 return (vd);
2983 }
2984 ASSERT3U(newstate, ==, VDEV_STATE_HEALTHY);
2985
2986 /*
2987 * Since we dropped the lock we need to ensure that we're
2988 * still talking to the original vdev. It's possible this
2989 * vdev may have been detached/replaced while we were
2990 * trying to online it.
2991 */
2992 if (generation != spa->spa_config_generation) {
2993 if (ztest_opts.zo_verbose >= 5) {
2994 (void) printf("vdev configuration has changed, "
2995 "guid %llu, state %llu, expected gen %llu, "
2996 "got gen %llu\n",
2997 (u_longlong_t)guid,
2998 (u_longlong_t)tvd->vdev_state,
2999 (u_longlong_t)generation,
3000 (u_longlong_t)spa->spa_config_generation);
3001 }
3002 return (vd);
3003 }
3004 return (NULL);
3005 }
3006
3007 /*
3008 * Traverse the vdev tree calling the supplied function.
3009 * We continue to walk the tree until we either have walked all
3010 * children or we receive a non-NULL return from the callback.
3011 * If a NULL callback is passed, then we just return back the first
3012 * leaf vdev we encounter.
3013 */
3014 vdev_t *
vdev_walk_tree(vdev_t * vd,vdev_t * (* func)(vdev_t *,void *),void * arg)3015 vdev_walk_tree(vdev_t *vd, vdev_t *(*func)(vdev_t *, void *), void *arg)
3016 {
3017 if (vd->vdev_ops->vdev_op_leaf) {
3018 if (func == NULL)
3019 return (vd);
3020 else
3021 return (func(vd, arg));
3022 }
3023
3024 for (uint_t c = 0; c < vd->vdev_children; c++) {
3025 vdev_t *cvd = vd->vdev_child[c];
3026 if ((cvd = vdev_walk_tree(cvd, func, arg)) != NULL)
3027 return (cvd);
3028 }
3029 return (NULL);
3030 }
3031
3032 /*
3033 * Verify that dynamic LUN growth works as expected.
3034 */
3035 /* ARGSUSED */
3036 void
ztest_vdev_LUN_growth(ztest_ds_t * zd,uint64_t id)3037 ztest_vdev_LUN_growth(ztest_ds_t *zd, uint64_t id)
3038 {
3039 spa_t *spa = ztest_spa;
3040 vdev_t *vd, *tvd;
3041 metaslab_class_t *mc;
3042 metaslab_group_t *mg;
3043 size_t psize, newsize;
3044 uint64_t top;
3045 uint64_t old_class_space, new_class_space, old_ms_count, new_ms_count;
3046
3047 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
3048 spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3049
3050 top = ztest_random_vdev_top(spa, B_TRUE);
3051
3052 tvd = spa->spa_root_vdev->vdev_child[top];
3053 mg = tvd->vdev_mg;
3054 mc = mg->mg_class;
3055 old_ms_count = tvd->vdev_ms_count;
3056 old_class_space = metaslab_class_get_space(mc);
3057
3058 /*
3059 * Determine the size of the first leaf vdev associated with
3060 * our top-level device.
3061 */
3062 vd = vdev_walk_tree(tvd, NULL, NULL);
3063 ASSERT3P(vd, !=, NULL);
3064 ASSERT(vd->vdev_ops->vdev_op_leaf);
3065
3066 psize = vd->vdev_psize;
3067
3068 /*
3069 * We only try to expand the vdev if it's healthy, less than 4x its
3070 * original size, and it has a valid psize.
3071 */
3072 if (tvd->vdev_state != VDEV_STATE_HEALTHY ||
3073 psize == 0 || psize >= 4 * ztest_opts.zo_vdev_size) {
3074 spa_config_exit(spa, SCL_STATE, spa);
3075 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3076 return;
3077 }
3078 ASSERT(psize > 0);
3079 newsize = psize + psize / 8;
3080 ASSERT3U(newsize, >, psize);
3081
3082 if (ztest_opts.zo_verbose >= 6) {
3083 (void) printf("Expanding LUN %s from %lu to %lu\n",
3084 vd->vdev_path, (ulong_t)psize, (ulong_t)newsize);
3085 }
3086
3087 /*
3088 * Growing the vdev is a two step process:
3089 * 1). expand the physical size (i.e. relabel)
3090 * 2). online the vdev to create the new metaslabs
3091 */
3092 if (vdev_walk_tree(tvd, grow_vdev, &newsize) != NULL ||
3093 vdev_walk_tree(tvd, online_vdev, NULL) != NULL ||
3094 tvd->vdev_state != VDEV_STATE_HEALTHY) {
3095 if (ztest_opts.zo_verbose >= 5) {
3096 (void) printf("Could not expand LUN because "
3097 "the vdev configuration changed.\n");
3098 }
3099 spa_config_exit(spa, SCL_STATE, spa);
3100 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3101 return;
3102 }
3103
3104 spa_config_exit(spa, SCL_STATE, spa);
3105
3106 /*
3107 * Expanding the LUN will update the config asynchronously,
3108 * thus we must wait for the async thread to complete any
3109 * pending tasks before proceeding.
3110 */
3111 for (;;) {
3112 boolean_t done;
3113 mutex_enter(&spa->spa_async_lock);
3114 done = (spa->spa_async_thread == NULL && !spa->spa_async_tasks);
3115 mutex_exit(&spa->spa_async_lock);
3116 if (done)
3117 break;
3118 txg_wait_synced(spa_get_dsl(spa), 0);
3119 (void) poll(NULL, 0, 100);
3120 }
3121
3122 spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3123
3124 tvd = spa->spa_root_vdev->vdev_child[top];
3125 new_ms_count = tvd->vdev_ms_count;
3126 new_class_space = metaslab_class_get_space(mc);
3127
3128 if (tvd->vdev_mg != mg || mg->mg_class != mc) {
3129 if (ztest_opts.zo_verbose >= 5) {
3130 (void) printf("Could not verify LUN expansion due to "
3131 "intervening vdev offline or remove.\n");
3132 }
3133 spa_config_exit(spa, SCL_STATE, spa);
3134 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3135 return;
3136 }
3137
3138 /*
3139 * Make sure we were able to grow the vdev.
3140 */
3141 if (new_ms_count <= old_ms_count)
3142 fatal(0, "LUN expansion failed: ms_count %llu <= %llu\n",
3143 old_ms_count, new_ms_count);
3144
3145 /*
3146 * Make sure we were able to grow the pool.
3147 */
3148 if (new_class_space <= old_class_space)
3149 fatal(0, "LUN expansion failed: class_space %llu <= %llu\n",
3150 old_class_space, new_class_space);
3151
3152 if (ztest_opts.zo_verbose >= 5) {
3153 char oldnumbuf[6], newnumbuf[6];
3154
3155 nicenum(old_class_space, oldnumbuf);
3156 nicenum(new_class_space, newnumbuf);
3157 (void) printf("%s grew from %s to %s\n",
3158 spa->spa_name, oldnumbuf, newnumbuf);
3159 }
3160
3161 spa_config_exit(spa, SCL_STATE, spa);
3162 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3163 }
3164
3165 /*
3166 * Verify that dmu_objset_{create,destroy,open,close} work as expected.
3167 */
3168 /* ARGSUSED */
3169 static void
ztest_objset_create_cb(objset_t * os,void * arg,cred_t * cr,dmu_tx_t * tx)3170 ztest_objset_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
3171 {
3172 /*
3173 * Create the objects common to all ztest datasets.
3174 */
3175 VERIFY(zap_create_claim(os, ZTEST_DIROBJ,
3176 DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx) == 0);
3177 }
3178
3179 static int
ztest_dataset_create(char * dsname)3180 ztest_dataset_create(char *dsname)
3181 {
3182 uint64_t zilset = ztest_random(100);
3183 int err = dmu_objset_create(dsname, DMU_OST_OTHER, 0,
3184 ztest_objset_create_cb, NULL);
3185
3186 if (err || zilset < 80)
3187 return (err);
3188
3189 if (ztest_opts.zo_verbose >= 6)
3190 (void) printf("Setting dataset %s to sync always\n", dsname);
3191 return (ztest_dsl_prop_set_uint64(dsname, ZFS_PROP_SYNC,
3192 ZFS_SYNC_ALWAYS, B_FALSE));
3193 }
3194
3195 /* ARGSUSED */
3196 static int
ztest_objset_destroy_cb(const char * name,void * arg)3197 ztest_objset_destroy_cb(const char *name, void *arg)
3198 {
3199 objset_t *os;
3200 dmu_object_info_t doi;
3201 int error;
3202
3203 /*
3204 * Verify that the dataset contains a directory object.
3205 */
3206 VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_TRUE, FTAG, &os));
3207 error = dmu_object_info(os, ZTEST_DIROBJ, &doi);
3208 if (error != ENOENT) {
3209 /* We could have crashed in the middle of destroying it */
3210 ASSERT0(error);
3211 ASSERT3U(doi.doi_type, ==, DMU_OT_ZAP_OTHER);
3212 ASSERT3S(doi.doi_physical_blocks_512, >=, 0);
3213 }
3214 dmu_objset_disown(os, FTAG);
3215
3216 /*
3217 * Destroy the dataset.
3218 */
3219 if (strchr(name, '@') != NULL) {
3220 VERIFY0(dsl_destroy_snapshot(name, B_FALSE));
3221 } else {
3222 VERIFY0(dsl_destroy_head(name));
3223 }
3224 return (0);
3225 }
3226
3227 static boolean_t
ztest_snapshot_create(char * osname,uint64_t id)3228 ztest_snapshot_create(char *osname, uint64_t id)
3229 {
3230 char snapname[ZFS_MAX_DATASET_NAME_LEN];
3231 int error;
3232
3233 (void) snprintf(snapname, sizeof (snapname), "%llu", (u_longlong_t)id);
3234
3235 error = dmu_objset_snapshot_one(osname, snapname);
3236 if (error == ENOSPC) {
3237 ztest_record_enospc(FTAG);
3238 return (B_FALSE);
3239 }
3240 if (error != 0 && error != EEXIST) {
3241 fatal(0, "ztest_snapshot_create(%s@%s) = %d", osname,
3242 snapname, error);
3243 }
3244 return (B_TRUE);
3245 }
3246
3247 static boolean_t
ztest_snapshot_destroy(char * osname,uint64_t id)3248 ztest_snapshot_destroy(char *osname, uint64_t id)
3249 {
3250 char snapname[ZFS_MAX_DATASET_NAME_LEN];
3251 int error;
3252
3253 (void) snprintf(snapname, sizeof (snapname), "%s@%llu", osname,
3254 (u_longlong_t)id);
3255
3256 error = dsl_destroy_snapshot(snapname, B_FALSE);
3257 if (error != 0 && error != ENOENT)
3258 fatal(0, "ztest_snapshot_destroy(%s) = %d", snapname, error);
3259 return (B_TRUE);
3260 }
3261
3262 /* ARGSUSED */
3263 void
ztest_dmu_objset_create_destroy(ztest_ds_t * zd,uint64_t id)3264 ztest_dmu_objset_create_destroy(ztest_ds_t *zd, uint64_t id)
3265 {
3266 ztest_ds_t zdtmp;
3267 int iters;
3268 int error;
3269 objset_t *os, *os2;
3270 char name[ZFS_MAX_DATASET_NAME_LEN];
3271 zilog_t *zilog;
3272
3273 (void) rw_rdlock(&ztest_name_lock);
3274
3275 (void) snprintf(name, sizeof (name), "%s/temp_%llu",
3276 ztest_opts.zo_pool, (u_longlong_t)id);
3277
3278 /*
3279 * If this dataset exists from a previous run, process its replay log
3280 * half of the time. If we don't replay it, then dmu_objset_destroy()
3281 * (invoked from ztest_objset_destroy_cb()) should just throw it away.
3282 */
3283 if (ztest_random(2) == 0 &&
3284 dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os) == 0) {
3285 ztest_zd_init(&zdtmp, NULL, os);
3286 zil_replay(os, &zdtmp, ztest_replay_vector);
3287 ztest_zd_fini(&zdtmp);
3288 dmu_objset_disown(os, FTAG);
3289 }
3290
3291 /*
3292 * There may be an old instance of the dataset we're about to
3293 * create lying around from a previous run. If so, destroy it
3294 * and all of its snapshots.
3295 */
3296 (void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
3297 DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
3298
3299 /*
3300 * Verify that the destroyed dataset is no longer in the namespace.
3301 */
3302 VERIFY3U(ENOENT, ==, dmu_objset_own(name, DMU_OST_OTHER, B_TRUE,
3303 FTAG, &os));
3304
3305 /*
3306 * Verify that we can create a new dataset.
3307 */
3308 error = ztest_dataset_create(name);
3309 if (error) {
3310 if (error == ENOSPC) {
3311 ztest_record_enospc(FTAG);
3312 (void) rw_unlock(&ztest_name_lock);
3313 return;
3314 }
3315 fatal(0, "dmu_objset_create(%s) = %d", name, error);
3316 }
3317
3318 VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os));
3319
3320 ztest_zd_init(&zdtmp, NULL, os);
3321
3322 /*
3323 * Open the intent log for it.
3324 */
3325 zilog = zil_open(os, ztest_get_data);
3326
3327 /*
3328 * Put some objects in there, do a little I/O to them,
3329 * and randomly take a couple of snapshots along the way.
3330 */
3331 iters = ztest_random(5);
3332 for (int i = 0; i < iters; i++) {
3333 ztest_dmu_object_alloc_free(&zdtmp, id);
3334 if (ztest_random(iters) == 0)
3335 (void) ztest_snapshot_create(name, i);
3336 }
3337
3338 /*
3339 * Verify that we cannot create an existing dataset.
3340 */
3341 VERIFY3U(EEXIST, ==,
3342 dmu_objset_create(name, DMU_OST_OTHER, 0, NULL, NULL));
3343
3344 /*
3345 * Verify that we can hold an objset that is also owned.
3346 */
3347 VERIFY3U(0, ==, dmu_objset_hold(name, FTAG, &os2));
3348 dmu_objset_rele(os2, FTAG);
3349
3350 /*
3351 * Verify that we cannot own an objset that is already owned.
3352 */
3353 VERIFY3U(EBUSY, ==,
3354 dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os2));
3355
3356 zil_close(zilog);
3357 dmu_objset_disown(os, FTAG);
3358 ztest_zd_fini(&zdtmp);
3359
3360 (void) rw_unlock(&ztest_name_lock);
3361 }
3362
3363 /*
3364 * Verify that dmu_snapshot_{create,destroy,open,close} work as expected.
3365 */
3366 void
ztest_dmu_snapshot_create_destroy(ztest_ds_t * zd,uint64_t id)3367 ztest_dmu_snapshot_create_destroy(ztest_ds_t *zd, uint64_t id)
3368 {
3369 (void) rw_rdlock(&ztest_name_lock);
3370 (void) ztest_snapshot_destroy(zd->zd_name, id);
3371 (void) ztest_snapshot_create(zd->zd_name, id);
3372 (void) rw_unlock(&ztest_name_lock);
3373 }
3374
3375 /*
3376 * Cleanup non-standard snapshots and clones.
3377 */
3378 void
ztest_dsl_dataset_cleanup(char * osname,uint64_t id)3379 ztest_dsl_dataset_cleanup(char *osname, uint64_t id)
3380 {
3381 char snap1name[ZFS_MAX_DATASET_NAME_LEN];
3382 char clone1name[ZFS_MAX_DATASET_NAME_LEN];
3383 char snap2name[ZFS_MAX_DATASET_NAME_LEN];
3384 char clone2name[ZFS_MAX_DATASET_NAME_LEN];
3385 char snap3name[ZFS_MAX_DATASET_NAME_LEN];
3386 int error;
3387
3388 (void) snprintf(snap1name, sizeof (snap1name),
3389 "%s@s1_%llu", osname, id);
3390 (void) snprintf(clone1name, sizeof (clone1name),
3391 "%s/c1_%llu", osname, id);
3392 (void) snprintf(snap2name, sizeof (snap2name),
3393 "%s@s2_%llu", clone1name, id);
3394 (void) snprintf(clone2name, sizeof (clone2name),
3395 "%s/c2_%llu", osname, id);
3396 (void) snprintf(snap3name, sizeof (snap3name),
3397 "%s@s3_%llu", clone1name, id);
3398
3399 error = dsl_destroy_head(clone2name);
3400 if (error && error != ENOENT)
3401 fatal(0, "dsl_destroy_head(%s) = %d", clone2name, error);
3402 error = dsl_destroy_snapshot(snap3name, B_FALSE);
3403 if (error && error != ENOENT)
3404 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap3name, error);
3405 error = dsl_destroy_snapshot(snap2name, B_FALSE);
3406 if (error && error != ENOENT)
3407 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap2name, error);
3408 error = dsl_destroy_head(clone1name);
3409 if (error && error != ENOENT)
3410 fatal(0, "dsl_destroy_head(%s) = %d", clone1name, error);
3411 error = dsl_destroy_snapshot(snap1name, B_FALSE);
3412 if (error && error != ENOENT)
3413 fatal(0, "dsl_destroy_snapshot(%s) = %d", snap1name, error);
3414 }
3415
3416 /*
3417 * Verify dsl_dataset_promote handles EBUSY
3418 */
3419 void
ztest_dsl_dataset_promote_busy(ztest_ds_t * zd,uint64_t id)3420 ztest_dsl_dataset_promote_busy(ztest_ds_t *zd, uint64_t id)
3421 {
3422 objset_t *os;
3423 char snap1name[ZFS_MAX_DATASET_NAME_LEN];
3424 char clone1name[ZFS_MAX_DATASET_NAME_LEN];
3425 char snap2name[ZFS_MAX_DATASET_NAME_LEN];
3426 char clone2name[ZFS_MAX_DATASET_NAME_LEN];
3427 char snap3name[ZFS_MAX_DATASET_NAME_LEN];
3428 char *osname = zd->zd_name;
3429 int error;
3430
3431 (void) rw_rdlock(&ztest_name_lock);
3432
3433 ztest_dsl_dataset_cleanup(osname, id);
3434
3435 (void) snprintf(snap1name, sizeof (snap1name),
3436 "%s@s1_%llu", osname, id);
3437 (void) snprintf(clone1name, sizeof (clone1name),
3438 "%s/c1_%llu", osname, id);
3439 (void) snprintf(snap2name, sizeof (snap2name),
3440 "%s@s2_%llu", clone1name, id);
3441 (void) snprintf(clone2name, sizeof (clone2name),
3442 "%s/c2_%llu", osname, id);
3443 (void) snprintf(snap3name, sizeof (snap3name),
3444 "%s@s3_%llu", clone1name, id);
3445
3446 error = dmu_objset_snapshot_one(osname, strchr(snap1name, '@') + 1);
3447 if (error && error != EEXIST) {
3448 if (error == ENOSPC) {
3449 ztest_record_enospc(FTAG);
3450 goto out;
3451 }
3452 fatal(0, "dmu_take_snapshot(%s) = %d", snap1name, error);
3453 }
3454
3455 error = dmu_objset_clone(clone1name, snap1name);
3456 if (error) {
3457 if (error == ENOSPC) {
3458 ztest_record_enospc(FTAG);
3459 goto out;
3460 }
3461 fatal(0, "dmu_objset_create(%s) = %d", clone1name, error);
3462 }
3463
3464 error = dmu_objset_snapshot_one(clone1name, strchr(snap2name, '@') + 1);
3465 if (error && error != EEXIST) {
3466 if (error == ENOSPC) {
3467 ztest_record_enospc(FTAG);
3468 goto out;
3469 }
3470 fatal(0, "dmu_open_snapshot(%s) = %d", snap2name, error);
3471 }
3472
3473 error = dmu_objset_snapshot_one(clone1name, strchr(snap3name, '@') + 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", snap3name, error);
3480 }
3481
3482 error = dmu_objset_clone(clone2name, snap3name);
3483 if (error) {
3484 if (error == ENOSPC) {
3485 ztest_record_enospc(FTAG);
3486 goto out;
3487 }
3488 fatal(0, "dmu_objset_create(%s) = %d", clone2name, error);
3489 }
3490
3491 error = dmu_objset_own(snap2name, DMU_OST_ANY, B_TRUE, FTAG, &os);
3492 if (error)
3493 fatal(0, "dmu_objset_own(%s) = %d", snap2name, error);
3494 error = dsl_dataset_promote(clone2name, NULL);
3495 if (error == ENOSPC) {
3496 dmu_objset_disown(os, FTAG);
3497 ztest_record_enospc(FTAG);
3498 goto out;
3499 }
3500 if (error != EBUSY)
3501 fatal(0, "dsl_dataset_promote(%s), %d, not EBUSY", clone2name,
3502 error);
3503 dmu_objset_disown(os, FTAG);
3504
3505 out:
3506 ztest_dsl_dataset_cleanup(osname, id);
3507
3508 (void) rw_unlock(&ztest_name_lock);
3509 }
3510
3511 /*
3512 * Verify that dmu_object_{alloc,free} work as expected.
3513 */
3514 void
ztest_dmu_object_alloc_free(ztest_ds_t * zd,uint64_t id)3515 ztest_dmu_object_alloc_free(ztest_ds_t *zd, uint64_t id)
3516 {
3517 ztest_od_t od[4];
3518 int batchsize = sizeof (od) / sizeof (od[0]);
3519
3520 for (int b = 0; b < batchsize; b++)
3521 ztest_od_init(&od[b], id, FTAG, b, DMU_OT_UINT64_OTHER, 0, 0);
3522
3523 /*
3524 * Destroy the previous batch of objects, create a new batch,
3525 * and do some I/O on the new objects.
3526 */
3527 if (ztest_object_init(zd, od, sizeof (od), B_TRUE) != 0)
3528 return;
3529
3530 while (ztest_random(4 * batchsize) != 0)
3531 ztest_io(zd, od[ztest_random(batchsize)].od_object,
3532 ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
3533 }
3534
3535 /*
3536 * Verify that dmu_{read,write} work as expected.
3537 */
3538 void
ztest_dmu_read_write(ztest_ds_t * zd,uint64_t id)3539 ztest_dmu_read_write(ztest_ds_t *zd, uint64_t id)
3540 {
3541 objset_t *os = zd->zd_os;
3542 ztest_od_t od[2];
3543 dmu_tx_t *tx;
3544 int i, freeit, error;
3545 uint64_t n, s, txg;
3546 bufwad_t *packbuf, *bigbuf, *pack, *bigH, *bigT;
3547 uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
3548 uint64_t chunksize = (1000 + ztest_random(1000)) * sizeof (uint64_t);
3549 uint64_t regions = 997;
3550 uint64_t stride = 123456789ULL;
3551 uint64_t width = 40;
3552 int free_percent = 5;
3553
3554 /*
3555 * This test uses two objects, packobj and bigobj, that are always
3556 * updated together (i.e. in the same tx) so that their contents are
3557 * in sync and can be compared. Their contents relate to each other
3558 * in a simple way: packobj is a dense array of 'bufwad' structures,
3559 * while bigobj is a sparse array of the same bufwads. Specifically,
3560 * for any index n, there are three bufwads that should be identical:
3561 *
3562 * packobj, at offset n * sizeof (bufwad_t)
3563 * bigobj, at the head of the nth chunk
3564 * bigobj, at the tail of the nth chunk
3565 *
3566 * The chunk size is arbitrary. It doesn't have to be a power of two,
3567 * and it doesn't have any relation to the object blocksize.
3568 * The only requirement is that it can hold at least two bufwads.
3569 *
3570 * Normally, we write the bufwad to each of these locations.
3571 * However, free_percent of the time we instead write zeroes to
3572 * packobj and perform a dmu_free_range() on bigobj. By comparing
3573 * bigobj to packobj, we can verify that the DMU is correctly
3574 * tracking which parts of an object are allocated and free,
3575 * and that the contents of the allocated blocks are correct.
3576 */
3577
3578 /*
3579 * Read the directory info. If it's the first time, set things up.
3580 */
3581 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, chunksize);
3582 ztest_od_init(&od[1], id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize);
3583
3584 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
3585 return;
3586
3587 bigobj = od[0].od_object;
3588 packobj = od[1].od_object;
3589 chunksize = od[0].od_gen;
3590 ASSERT(chunksize == od[1].od_gen);
3591
3592 /*
3593 * Prefetch a random chunk of the big object.
3594 * Our aim here is to get some async reads in flight
3595 * for blocks that we may free below; the DMU should
3596 * handle this race correctly.
3597 */
3598 n = ztest_random(regions) * stride + ztest_random(width);
3599 s = 1 + ztest_random(2 * width - 1);
3600 dmu_prefetch(os, bigobj, 0, n * chunksize, s * chunksize,
3601 ZIO_PRIORITY_SYNC_READ);
3602
3603 /*
3604 * Pick a random index and compute the offsets into packobj and bigobj.
3605 */
3606 n = ztest_random(regions) * stride + ztest_random(width);
3607 s = 1 + ztest_random(width - 1);
3608
3609 packoff = n * sizeof (bufwad_t);
3610 packsize = s * sizeof (bufwad_t);
3611
3612 bigoff = n * chunksize;
3613 bigsize = s * chunksize;
3614
3615 packbuf = umem_alloc(packsize, UMEM_NOFAIL);
3616 bigbuf = umem_alloc(bigsize, UMEM_NOFAIL);
3617
3618 /*
3619 * free_percent of the time, free a range of bigobj rather than
3620 * overwriting it.
3621 */
3622 freeit = (ztest_random(100) < free_percent);
3623
3624 /*
3625 * Read the current contents of our objects.
3626 */
3627 error = dmu_read(os, packobj, packoff, packsize, packbuf,
3628 DMU_READ_PREFETCH);
3629 ASSERT0(error);
3630 error = dmu_read(os, bigobj, bigoff, bigsize, bigbuf,
3631 DMU_READ_PREFETCH);
3632 ASSERT0(error);
3633
3634 /*
3635 * Get a tx for the mods to both packobj and bigobj.
3636 */
3637 tx = dmu_tx_create(os);
3638
3639 dmu_tx_hold_write(tx, packobj, packoff, packsize);
3640
3641 if (freeit)
3642 dmu_tx_hold_free(tx, bigobj, bigoff, bigsize);
3643 else
3644 dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
3645
3646 /* This accounts for setting the checksum/compression. */
3647 dmu_tx_hold_bonus(tx, bigobj);
3648
3649 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3650 if (txg == 0) {
3651 umem_free(packbuf, packsize);
3652 umem_free(bigbuf, bigsize);
3653 return;
3654 }
3655
3656 enum zio_checksum cksum;
3657 do {
3658 cksum = (enum zio_checksum)
3659 ztest_random_dsl_prop(ZFS_PROP_CHECKSUM);
3660 } while (cksum >= ZIO_CHECKSUM_LEGACY_FUNCTIONS);
3661 dmu_object_set_checksum(os, bigobj, cksum, tx);
3662
3663 enum zio_compress comp;
3664 do {
3665 comp = (enum zio_compress)
3666 ztest_random_dsl_prop(ZFS_PROP_COMPRESSION);
3667 } while (comp >= ZIO_COMPRESS_LEGACY_FUNCTIONS);
3668 dmu_object_set_compress(os, bigobj, comp, tx);
3669
3670 /*
3671 * For each index from n to n + s, verify that the existing bufwad
3672 * in packobj matches the bufwads at the head and tail of the
3673 * corresponding chunk in bigobj. Then update all three bufwads
3674 * with the new values we want to write out.
3675 */
3676 for (i = 0; i < s; i++) {
3677 /* LINTED */
3678 pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
3679 /* LINTED */
3680 bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
3681 /* LINTED */
3682 bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
3683
3684 ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
3685 ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
3686
3687 if (pack->bw_txg > txg)
3688 fatal(0, "future leak: got %llx, open txg is %llx",
3689 pack->bw_txg, txg);
3690
3691 if (pack->bw_data != 0 && pack->bw_index != n + i)
3692 fatal(0, "wrong index: got %llx, wanted %llx+%llx",
3693 pack->bw_index, n, i);
3694
3695 if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
3696 fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
3697
3698 if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
3699 fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
3700
3701 if (freeit) {
3702 bzero(pack, sizeof (bufwad_t));
3703 } else {
3704 pack->bw_index = n + i;
3705 pack->bw_txg = txg;
3706 pack->bw_data = 1 + ztest_random(-2ULL);
3707 }
3708 *bigH = *pack;
3709 *bigT = *pack;
3710 }
3711
3712 /*
3713 * We've verified all the old bufwads, and made new ones.
3714 * Now write them out.
3715 */
3716 dmu_write(os, packobj, packoff, packsize, packbuf, tx);
3717
3718 if (freeit) {
3719 if (ztest_opts.zo_verbose >= 7) {
3720 (void) printf("freeing offset %llx size %llx"
3721 " txg %llx\n",
3722 (u_longlong_t)bigoff,
3723 (u_longlong_t)bigsize,
3724 (u_longlong_t)txg);
3725 }
3726 VERIFY(0 == dmu_free_range(os, bigobj, bigoff, bigsize, tx));
3727 } else {
3728 if (ztest_opts.zo_verbose >= 7) {
3729 (void) printf("writing 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 dmu_write(os, bigobj, bigoff, bigsize, bigbuf, tx);
3736 }
3737
3738 dmu_tx_commit(tx);
3739
3740 /*
3741 * Sanity check the stuff we just wrote.
3742 */
3743 {
3744 void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
3745 void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
3746
3747 VERIFY(0 == dmu_read(os, packobj, packoff,
3748 packsize, packcheck, DMU_READ_PREFETCH));
3749 VERIFY(0 == dmu_read(os, bigobj, bigoff,
3750 bigsize, bigcheck, DMU_READ_PREFETCH));
3751
3752 ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
3753 ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
3754
3755 umem_free(packcheck, packsize);
3756 umem_free(bigcheck, bigsize);
3757 }
3758
3759 umem_free(packbuf, packsize);
3760 umem_free(bigbuf, bigsize);
3761 }
3762
3763 void
compare_and_update_pbbufs(uint64_t s,bufwad_t * packbuf,bufwad_t * bigbuf,uint64_t bigsize,uint64_t n,uint64_t chunksize,uint64_t txg)3764 compare_and_update_pbbufs(uint64_t s, bufwad_t *packbuf, bufwad_t *bigbuf,
3765 uint64_t bigsize, uint64_t n, uint64_t chunksize, uint64_t txg)
3766 {
3767 uint64_t i;
3768 bufwad_t *pack;
3769 bufwad_t *bigH;
3770 bufwad_t *bigT;
3771
3772 /*
3773 * For each index from n to n + s, verify that the existing bufwad
3774 * in packobj matches the bufwads at the head and tail of the
3775 * corresponding chunk in bigobj. Then update all three bufwads
3776 * with the new values we want to write out.
3777 */
3778 for (i = 0; i < s; i++) {
3779 /* LINTED */
3780 pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
3781 /* LINTED */
3782 bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
3783 /* LINTED */
3784 bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
3785
3786 ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
3787 ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
3788
3789 if (pack->bw_txg > txg)
3790 fatal(0, "future leak: got %llx, open txg is %llx",
3791 pack->bw_txg, txg);
3792
3793 if (pack->bw_data != 0 && pack->bw_index != n + i)
3794 fatal(0, "wrong index: got %llx, wanted %llx+%llx",
3795 pack->bw_index, n, i);
3796
3797 if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
3798 fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
3799
3800 if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
3801 fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
3802
3803 pack->bw_index = n + i;
3804 pack->bw_txg = txg;
3805 pack->bw_data = 1 + ztest_random(-2ULL);
3806
3807 *bigH = *pack;
3808 *bigT = *pack;
3809 }
3810 }
3811
3812 void
ztest_dmu_read_write_zcopy(ztest_ds_t * zd,uint64_t id)3813 ztest_dmu_read_write_zcopy(ztest_ds_t *zd, uint64_t id)
3814 {
3815 objset_t *os = zd->zd_os;
3816 ztest_od_t od[2];
3817 dmu_tx_t *tx;
3818 uint64_t i;
3819 int error;
3820 uint64_t n, s, txg;
3821 bufwad_t *packbuf, *bigbuf;
3822 uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
3823 uint64_t blocksize = ztest_random_blocksize();
3824 uint64_t chunksize = blocksize;
3825 uint64_t regions = 997;
3826 uint64_t stride = 123456789ULL;
3827 uint64_t width = 9;
3828 dmu_buf_t *bonus_db;
3829 arc_buf_t **bigbuf_arcbufs;
3830 dmu_object_info_t doi;
3831
3832 /*
3833 * This test uses two objects, packobj and bigobj, that are always
3834 * updated together (i.e. in the same tx) so that their contents are
3835 * in sync and can be compared. Their contents relate to each other
3836 * in a simple way: packobj is a dense array of 'bufwad' structures,
3837 * while bigobj is a sparse array of the same bufwads. Specifically,
3838 * for any index n, there are three bufwads that should be identical:
3839 *
3840 * packobj, at offset n * sizeof (bufwad_t)
3841 * bigobj, at the head of the nth chunk
3842 * bigobj, at the tail of the nth chunk
3843 *
3844 * The chunk size is set equal to bigobj block size so that
3845 * dmu_assign_arcbuf() can be tested for object updates.
3846 */
3847
3848 /*
3849 * Read the directory info. If it's the first time, set things up.
3850 */
3851 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
3852 ztest_od_init(&od[1], id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize);
3853
3854 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
3855 return;
3856
3857 bigobj = od[0].od_object;
3858 packobj = od[1].od_object;
3859 blocksize = od[0].od_blocksize;
3860 chunksize = blocksize;
3861 ASSERT(chunksize == od[1].od_gen);
3862
3863 VERIFY(dmu_object_info(os, bigobj, &doi) == 0);
3864 VERIFY(ISP2(doi.doi_data_block_size));
3865 VERIFY(chunksize == doi.doi_data_block_size);
3866 VERIFY(chunksize >= 2 * sizeof (bufwad_t));
3867
3868 /*
3869 * Pick a random index and compute the offsets into packobj and bigobj.
3870 */
3871 n = ztest_random(regions) * stride + ztest_random(width);
3872 s = 1 + ztest_random(width - 1);
3873
3874 packoff = n * sizeof (bufwad_t);
3875 packsize = s * sizeof (bufwad_t);
3876
3877 bigoff = n * chunksize;
3878 bigsize = s * chunksize;
3879
3880 packbuf = umem_zalloc(packsize, UMEM_NOFAIL);
3881 bigbuf = umem_zalloc(bigsize, UMEM_NOFAIL);
3882
3883 VERIFY3U(0, ==, dmu_bonus_hold(os, bigobj, FTAG, &bonus_db));
3884
3885 bigbuf_arcbufs = umem_zalloc(2 * s * sizeof (arc_buf_t *), UMEM_NOFAIL);
3886
3887 /*
3888 * Iteration 0 test zcopy for DB_UNCACHED dbufs.
3889 * Iteration 1 test zcopy to already referenced dbufs.
3890 * Iteration 2 test zcopy to dirty dbuf in the same txg.
3891 * Iteration 3 test zcopy to dbuf dirty in previous txg.
3892 * Iteration 4 test zcopy when dbuf is no longer dirty.
3893 * Iteration 5 test zcopy when it can't be done.
3894 * Iteration 6 one more zcopy write.
3895 */
3896 for (i = 0; i < 7; i++) {
3897 uint64_t j;
3898 uint64_t off;
3899
3900 /*
3901 * In iteration 5 (i == 5) use arcbufs
3902 * that don't match bigobj blksz to test
3903 * dmu_assign_arcbuf() when it can't directly
3904 * assign an arcbuf to a dbuf.
3905 */
3906 for (j = 0; j < s; j++) {
3907 if (i != 5) {
3908 bigbuf_arcbufs[j] =
3909 dmu_request_arcbuf(bonus_db, chunksize);
3910 } else {
3911 bigbuf_arcbufs[2 * j] =
3912 dmu_request_arcbuf(bonus_db, chunksize / 2);
3913 bigbuf_arcbufs[2 * j + 1] =
3914 dmu_request_arcbuf(bonus_db, chunksize / 2);
3915 }
3916 }
3917
3918 /*
3919 * Get a tx for the mods to both packobj and bigobj.
3920 */
3921 tx = dmu_tx_create(os);
3922
3923 dmu_tx_hold_write(tx, packobj, packoff, packsize);
3924 dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
3925
3926 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3927 if (txg == 0) {
3928 umem_free(packbuf, packsize);
3929 umem_free(bigbuf, bigsize);
3930 for (j = 0; j < s; j++) {
3931 if (i != 5) {
3932 dmu_return_arcbuf(bigbuf_arcbufs[j]);
3933 } else {
3934 dmu_return_arcbuf(
3935 bigbuf_arcbufs[2 * j]);
3936 dmu_return_arcbuf(
3937 bigbuf_arcbufs[2 * j + 1]);
3938 }
3939 }
3940 umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
3941 dmu_buf_rele(bonus_db, FTAG);
3942 return;
3943 }
3944
3945 /*
3946 * 50% of the time don't read objects in the 1st iteration to
3947 * test dmu_assign_arcbuf() for the case when there're no
3948 * existing dbufs for the specified offsets.
3949 */
3950 if (i != 0 || ztest_random(2) != 0) {
3951 error = dmu_read(os, packobj, packoff,
3952 packsize, packbuf, DMU_READ_PREFETCH);
3953 ASSERT0(error);
3954 error = dmu_read(os, bigobj, bigoff, bigsize,
3955 bigbuf, DMU_READ_PREFETCH);
3956 ASSERT0(error);
3957 }
3958 compare_and_update_pbbufs(s, packbuf, bigbuf, bigsize,
3959 n, chunksize, txg);
3960
3961 /*
3962 * We've verified all the old bufwads, and made new ones.
3963 * Now write them out.
3964 */
3965 dmu_write(os, packobj, packoff, packsize, packbuf, tx);
3966 if (ztest_opts.zo_verbose >= 7) {
3967 (void) printf("writing offset %llx size %llx"
3968 " txg %llx\n",
3969 (u_longlong_t)bigoff,
3970 (u_longlong_t)bigsize,
3971 (u_longlong_t)txg);
3972 }
3973 for (off = bigoff, j = 0; j < s; j++, off += chunksize) {
3974 dmu_buf_t *dbt;
3975 if (i != 5) {
3976 bcopy((caddr_t)bigbuf + (off - bigoff),
3977 bigbuf_arcbufs[j]->b_data, chunksize);
3978 } else {
3979 bcopy((caddr_t)bigbuf + (off - bigoff),
3980 bigbuf_arcbufs[2 * j]->b_data,
3981 chunksize / 2);
3982 bcopy((caddr_t)bigbuf + (off - bigoff) +
3983 chunksize / 2,
3984 bigbuf_arcbufs[2 * j + 1]->b_data,
3985 chunksize / 2);
3986 }
3987
3988 if (i == 1) {
3989 VERIFY(dmu_buf_hold(os, bigobj, off,
3990 FTAG, &dbt, DMU_READ_NO_PREFETCH) == 0);
3991 }
3992 if (i != 5) {
3993 dmu_assign_arcbuf(bonus_db, off,
3994 bigbuf_arcbufs[j], tx);
3995 } else {
3996 dmu_assign_arcbuf(bonus_db, off,
3997 bigbuf_arcbufs[2 * j], tx);
3998 dmu_assign_arcbuf(bonus_db,
3999 off + chunksize / 2,
4000 bigbuf_arcbufs[2 * j + 1], tx);
4001 }
4002 if (i == 1) {
4003 dmu_buf_rele(dbt, FTAG);
4004 }
4005 }
4006 dmu_tx_commit(tx);
4007
4008 /*
4009 * Sanity check the stuff we just wrote.
4010 */
4011 {
4012 void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
4013 void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
4014
4015 VERIFY(0 == dmu_read(os, packobj, packoff,
4016 packsize, packcheck, DMU_READ_PREFETCH));
4017 VERIFY(0 == dmu_read(os, bigobj, bigoff,
4018 bigsize, bigcheck, DMU_READ_PREFETCH));
4019
4020 ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
4021 ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
4022
4023 umem_free(packcheck, packsize);
4024 umem_free(bigcheck, bigsize);
4025 }
4026 if (i == 2) {
4027 txg_wait_open(dmu_objset_pool(os), 0);
4028 } else if (i == 3) {
4029 txg_wait_synced(dmu_objset_pool(os), 0);
4030 }
4031 }
4032
4033 dmu_buf_rele(bonus_db, FTAG);
4034 umem_free(packbuf, packsize);
4035 umem_free(bigbuf, bigsize);
4036 umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
4037 }
4038
4039 /* ARGSUSED */
4040 void
ztest_dmu_write_parallel(ztest_ds_t * zd,uint64_t id)4041 ztest_dmu_write_parallel(ztest_ds_t *zd, uint64_t id)
4042 {
4043 ztest_od_t od[1];
4044 uint64_t offset = (1ULL << (ztest_random(20) + 43)) +
4045 (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
4046
4047 /*
4048 * Have multiple threads write to large offsets in an object
4049 * to verify that parallel writes to an object -- even to the
4050 * same blocks within the object -- doesn't cause any trouble.
4051 */
4052 ztest_od_init(&od[0], ID_PARALLEL, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
4053
4054 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4055 return;
4056
4057 while (ztest_random(10) != 0)
4058 ztest_io(zd, od[0].od_object, offset);
4059 }
4060
4061 void
ztest_dmu_prealloc(ztest_ds_t * zd,uint64_t id)4062 ztest_dmu_prealloc(ztest_ds_t *zd, uint64_t id)
4063 {
4064 ztest_od_t od[1];
4065 uint64_t offset = (1ULL << (ztest_random(4) + SPA_MAXBLOCKSHIFT)) +
4066 (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
4067 uint64_t count = ztest_random(20) + 1;
4068 uint64_t blocksize = ztest_random_blocksize();
4069 void *data;
4070
4071 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
4072
4073 if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4074 return;
4075
4076 if (ztest_truncate(zd, od[0].od_object, offset, count * blocksize) != 0)
4077 return;
4078
4079 ztest_prealloc(zd, od[0].od_object, offset, count * blocksize);
4080
4081 data = umem_zalloc(blocksize, UMEM_NOFAIL);
4082
4083 while (ztest_random(count) != 0) {
4084 uint64_t randoff = offset + (ztest_random(count) * blocksize);
4085 if (ztest_write(zd, od[0].od_object, randoff, blocksize,
4086 data) != 0)
4087 break;
4088 while (ztest_random(4) != 0)
4089 ztest_io(zd, od[0].od_object, randoff);
4090 }
4091
4092 umem_free(data, blocksize);
4093 }
4094
4095 /*
4096 * Verify that zap_{create,destroy,add,remove,update} work as expected.
4097 */
4098 #define ZTEST_ZAP_MIN_INTS 1
4099 #define ZTEST_ZAP_MAX_INTS 4
4100 #define ZTEST_ZAP_MAX_PROPS 1000
4101
4102 void
ztest_zap(ztest_ds_t * zd,uint64_t id)4103 ztest_zap(ztest_ds_t *zd, uint64_t id)
4104 {
4105 objset_t *os = zd->zd_os;
4106 ztest_od_t od[1];
4107 uint64_t object;
4108 uint64_t txg, last_txg;
4109 uint64_t value[ZTEST_ZAP_MAX_INTS];
4110 uint64_t zl_ints, zl_intsize, prop;
4111 int i, ints;
4112 dmu_tx_t *tx;
4113 char propname[100], txgname[100];
4114 int error;
4115 char *hc[2] = { "s.acl.h", ".s.open.h.hyLZlg" };
4116
4117 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0);
4118
4119 if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4120 return;
4121
4122 object = od[0].od_object;
4123
4124 /*
4125 * Generate a known hash collision, and verify that
4126 * we can lookup and remove both entries.
4127 */
4128 tx = dmu_tx_create(os);
4129 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4130 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4131 if (txg == 0)
4132 return;
4133 for (i = 0; i < 2; i++) {
4134 value[i] = i;
4135 VERIFY3U(0, ==, zap_add(os, object, hc[i], sizeof (uint64_t),
4136 1, &value[i], tx));
4137 }
4138 for (i = 0; i < 2; i++) {
4139 VERIFY3U(EEXIST, ==, zap_add(os, object, hc[i],
4140 sizeof (uint64_t), 1, &value[i], tx));
4141 VERIFY3U(0, ==,
4142 zap_length(os, object, hc[i], &zl_intsize, &zl_ints));
4143 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4144 ASSERT3U(zl_ints, ==, 1);
4145 }
4146 for (i = 0; i < 2; i++) {
4147 VERIFY3U(0, ==, zap_remove(os, object, hc[i], tx));
4148 }
4149 dmu_tx_commit(tx);
4150
4151 /*
4152 * Generate a buch of random entries.
4153 */
4154 ints = MAX(ZTEST_ZAP_MIN_INTS, object % ZTEST_ZAP_MAX_INTS);
4155
4156 prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
4157 (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
4158 (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
4159 bzero(value, sizeof (value));
4160 last_txg = 0;
4161
4162 /*
4163 * If these zap entries already exist, validate their contents.
4164 */
4165 error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
4166 if (error == 0) {
4167 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4168 ASSERT3U(zl_ints, ==, 1);
4169
4170 VERIFY(zap_lookup(os, object, txgname, zl_intsize,
4171 zl_ints, &last_txg) == 0);
4172
4173 VERIFY(zap_length(os, object, propname, &zl_intsize,
4174 &zl_ints) == 0);
4175
4176 ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4177 ASSERT3U(zl_ints, ==, ints);
4178
4179 VERIFY(zap_lookup(os, object, propname, zl_intsize,
4180 zl_ints, value) == 0);
4181
4182 for (i = 0; i < ints; i++) {
4183 ASSERT3U(value[i], ==, last_txg + object + i);
4184 }
4185 } else {
4186 ASSERT3U(error, ==, ENOENT);
4187 }
4188
4189 /*
4190 * Atomically update two entries in our zap object.
4191 * The first is named txg_%llu, and contains the txg
4192 * in which the property was last updated. The second
4193 * is named prop_%llu, and the nth element of its value
4194 * should be txg + object + n.
4195 */
4196 tx = dmu_tx_create(os);
4197 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4198 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4199 if (txg == 0)
4200 return;
4201
4202 if (last_txg > txg)
4203 fatal(0, "zap future leak: old %llu new %llu", last_txg, txg);
4204
4205 for (i = 0; i < ints; i++)
4206 value[i] = txg + object + i;
4207
4208 VERIFY3U(0, ==, zap_update(os, object, txgname, sizeof (uint64_t),
4209 1, &txg, tx));
4210 VERIFY3U(0, ==, zap_update(os, object, propname, sizeof (uint64_t),
4211 ints, value, tx));
4212
4213 dmu_tx_commit(tx);
4214
4215 /*
4216 * Remove a random pair of entries.
4217 */
4218 prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
4219 (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
4220 (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
4221
4222 error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
4223
4224 if (error == ENOENT)
4225 return;
4226
4227 ASSERT0(error);
4228
4229 tx = dmu_tx_create(os);
4230 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4231 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4232 if (txg == 0)
4233 return;
4234 VERIFY3U(0, ==, zap_remove(os, object, txgname, tx));
4235 VERIFY3U(0, ==, zap_remove(os, object, propname, tx));
4236 dmu_tx_commit(tx);
4237 }
4238
4239 /*
4240 * Testcase to test the upgrading of a microzap to fatzap.
4241 */
4242 void
ztest_fzap(ztest_ds_t * zd,uint64_t id)4243 ztest_fzap(ztest_ds_t *zd, uint64_t id)
4244 {
4245 objset_t *os = zd->zd_os;
4246 ztest_od_t od[1];
4247 uint64_t object, txg;
4248
4249 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0);
4250
4251 if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4252 return;
4253
4254 object = od[0].od_object;
4255
4256 /*
4257 * Add entries to this ZAP and make sure it spills over
4258 * and gets upgraded to a fatzap. Also, since we are adding
4259 * 2050 entries we should see ptrtbl growth and leaf-block split.
4260 */
4261 for (int i = 0; i < 2050; i++) {
4262 char name[ZFS_MAX_DATASET_NAME_LEN];
4263 uint64_t value = i;
4264 dmu_tx_t *tx;
4265 int error;
4266
4267 (void) snprintf(name, sizeof (name), "fzap-%llu-%llu",
4268 id, value);
4269
4270 tx = dmu_tx_create(os);
4271 dmu_tx_hold_zap(tx, object, B_TRUE, name);
4272 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4273 if (txg == 0)
4274 return;
4275 error = zap_add(os, object, name, sizeof (uint64_t), 1,
4276 &value, tx);
4277 ASSERT(error == 0 || error == EEXIST);
4278 dmu_tx_commit(tx);
4279 }
4280 }
4281
4282 /* ARGSUSED */
4283 void
ztest_zap_parallel(ztest_ds_t * zd,uint64_t id)4284 ztest_zap_parallel(ztest_ds_t *zd, uint64_t id)
4285 {
4286 objset_t *os = zd->zd_os;
4287 ztest_od_t od[1];
4288 uint64_t txg, object, count, wsize, wc, zl_wsize, zl_wc;
4289 dmu_tx_t *tx;
4290 int i, namelen, error;
4291 int micro = ztest_random(2);
4292 char name[20], string_value[20];
4293 void *data;
4294
4295 ztest_od_init(&od[0], ID_PARALLEL, FTAG, micro, DMU_OT_ZAP_OTHER, 0, 0);
4296
4297 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4298 return;
4299
4300 object = od[0].od_object;
4301
4302 /*
4303 * Generate a random name of the form 'xxx.....' where each
4304 * x is a random printable character and the dots are dots.
4305 * There are 94 such characters, and the name length goes from
4306 * 6 to 20, so there are 94^3 * 15 = 12,458,760 possible names.
4307 */
4308 namelen = ztest_random(sizeof (name) - 5) + 5 + 1;
4309
4310 for (i = 0; i < 3; i++)
4311 name[i] = '!' + ztest_random('~' - '!' + 1);
4312 for (; i < namelen - 1; i++)
4313 name[i] = '.';
4314 name[i] = '\0';
4315
4316 if ((namelen & 1) || micro) {
4317 wsize = sizeof (txg);
4318 wc = 1;
4319 data = &txg;
4320 } else {
4321 wsize = 1;
4322 wc = namelen;
4323 data = string_value;
4324 }
4325
4326 count = -1ULL;
4327 VERIFY0(zap_count(os, object, &count));
4328 ASSERT(count != -1ULL);
4329
4330 /*
4331 * Select an operation: length, lookup, add, update, remove.
4332 */
4333 i = ztest_random(5);
4334
4335 if (i >= 2) {
4336 tx = dmu_tx_create(os);
4337 dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4338 txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4339 if (txg == 0)
4340 return;
4341 bcopy(name, string_value, namelen);
4342 } else {
4343 tx = NULL;
4344 txg = 0;
4345 bzero(string_value, namelen);
4346 }
4347
4348 switch (i) {
4349
4350 case 0:
4351 error = zap_length(os, object, name, &zl_wsize, &zl_wc);
4352 if (error == 0) {
4353 ASSERT3U(wsize, ==, zl_wsize);
4354 ASSERT3U(wc, ==, zl_wc);
4355 } else {
4356 ASSERT3U(error, ==, ENOENT);
4357 }
4358 break;
4359
4360 case 1:
4361 error = zap_lookup(os, object, name, wsize, wc, data);
4362 if (error == 0) {
4363 if (data == string_value &&
4364 bcmp(name, data, namelen) != 0)
4365 fatal(0, "name '%s' != val '%s' len %d",
4366 name, data, namelen);
4367 } else {
4368 ASSERT3U(error, ==, ENOENT);
4369 }
4370 break;
4371
4372 case 2:
4373 error = zap_add(os, object, name, wsize, wc, data, tx);
4374 ASSERT(error == 0 || error == EEXIST);
4375 break;
4376
4377 case 3:
4378 VERIFY(zap_update(os, object, name, wsize, wc, data, tx) == 0);
4379 break;
4380
4381 case 4:
4382 error = zap_remove(os, object, name, tx);
4383 ASSERT(error == 0 || error == ENOENT);
4384 break;
4385 }
4386
4387 if (tx != NULL)
4388 dmu_tx_commit(tx);
4389 }
4390
4391 /*
4392 * Commit callback data.
4393 */
4394 typedef struct ztest_cb_data {
4395 list_node_t zcd_node;
4396 uint64_t zcd_txg;
4397 int zcd_expected_err;
4398 boolean_t zcd_added;
4399 boolean_t zcd_called;
4400 spa_t *zcd_spa;
4401 } ztest_cb_data_t;
4402
4403 /* This is the actual commit callback function */
4404 static void
ztest_commit_callback(void * arg,int error)4405 ztest_commit_callback(void *arg, int error)
4406 {
4407 ztest_cb_data_t *data = arg;
4408 uint64_t synced_txg;
4409
4410 VERIFY(data != NULL);
4411 VERIFY3S(data->zcd_expected_err, ==, error);
4412 VERIFY(!data->zcd_called);
4413
4414 synced_txg = spa_last_synced_txg(data->zcd_spa);
4415 if (data->zcd_txg > synced_txg)
4416 fatal(0, "commit callback of txg %" PRIu64 " called prematurely"
4417 ", last synced txg = %" PRIu64 "\n", data->zcd_txg,
4418 synced_txg);
4419
4420 data->zcd_called = B_TRUE;
4421
4422 if (error == ECANCELED) {
4423 ASSERT0(data->zcd_txg);
4424 ASSERT(!data->zcd_added);
4425
4426 /*
4427 * The private callback data should be destroyed here, but
4428 * since we are going to check the zcd_called field after
4429 * dmu_tx_abort(), we will destroy it there.
4430 */
4431 return;
4432 }
4433
4434 /* Was this callback added to the global callback list? */
4435 if (!data->zcd_added)
4436 goto out;
4437
4438 ASSERT3U(data->zcd_txg, !=, 0);
4439
4440 /* Remove our callback from the list */
4441 (void) mutex_lock(&zcl.zcl_callbacks_lock);
4442 list_remove(&zcl.zcl_callbacks, data);
4443 (void) mutex_unlock(&zcl.zcl_callbacks_lock);
4444
4445 out:
4446 umem_free(data, sizeof (ztest_cb_data_t));
4447 }
4448
4449 /* Allocate and initialize callback data structure */
4450 static ztest_cb_data_t *
ztest_create_cb_data(objset_t * os,uint64_t txg)4451 ztest_create_cb_data(objset_t *os, uint64_t txg)
4452 {
4453 ztest_cb_data_t *cb_data;
4454
4455 cb_data = umem_zalloc(sizeof (ztest_cb_data_t), UMEM_NOFAIL);
4456
4457 cb_data->zcd_txg = txg;
4458 cb_data->zcd_spa = dmu_objset_spa(os);
4459
4460 return (cb_data);
4461 }
4462
4463 /*
4464 * If a number of txgs equal to this threshold have been created after a commit
4465 * callback has been registered but not called, then we assume there is an
4466 * implementation bug.
4467 */
4468 #define ZTEST_COMMIT_CALLBACK_THRESH (TXG_CONCURRENT_STATES + 2)
4469
4470 /*
4471 * Commit callback test.
4472 */
4473 void
ztest_dmu_commit_callbacks(ztest_ds_t * zd,uint64_t id)4474 ztest_dmu_commit_callbacks(ztest_ds_t *zd, uint64_t id)
4475 {
4476 objset_t *os = zd->zd_os;
4477 ztest_od_t od[1];
4478 dmu_tx_t *tx;
4479 ztest_cb_data_t *cb_data[3], *tmp_cb;
4480 uint64_t old_txg, txg;
4481 int i, error;
4482
4483 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
4484
4485 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4486 return;
4487
4488 tx = dmu_tx_create(os);
4489
4490 cb_data[0] = ztest_create_cb_data(os, 0);
4491 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[0]);
4492
4493 dmu_tx_hold_write(tx, od[0].od_object, 0, sizeof (uint64_t));
4494
4495 /* Every once in a while, abort the transaction on purpose */
4496 if (ztest_random(100) == 0)
4497 error = -1;
4498
4499 if (!error)
4500 error = dmu_tx_assign(tx, TXG_NOWAIT);
4501
4502 txg = error ? 0 : dmu_tx_get_txg(tx);
4503
4504 cb_data[0]->zcd_txg = txg;
4505 cb_data[1] = ztest_create_cb_data(os, txg);
4506 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[1]);
4507
4508 if (error) {
4509 /*
4510 * It's not a strict requirement to call the registered
4511 * callbacks from inside dmu_tx_abort(), but that's what
4512 * it's supposed to happen in the current implementation
4513 * so we will check for that.
4514 */
4515 for (i = 0; i < 2; i++) {
4516 cb_data[i]->zcd_expected_err = ECANCELED;
4517 VERIFY(!cb_data[i]->zcd_called);
4518 }
4519
4520 dmu_tx_abort(tx);
4521
4522 for (i = 0; i < 2; i++) {
4523 VERIFY(cb_data[i]->zcd_called);
4524 umem_free(cb_data[i], sizeof (ztest_cb_data_t));
4525 }
4526
4527 return;
4528 }
4529
4530 cb_data[2] = ztest_create_cb_data(os, txg);
4531 dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[2]);
4532
4533 /*
4534 * Read existing data to make sure there isn't a future leak.
4535 */
4536 VERIFY(0 == dmu_read(os, od[0].od_object, 0, sizeof (uint64_t),
4537 &old_txg, DMU_READ_PREFETCH));
4538
4539 if (old_txg > txg)
4540 fatal(0, "future leak: got %" PRIu64 ", open txg is %" PRIu64,
4541 old_txg, txg);
4542
4543 dmu_write(os, od[0].od_object, 0, sizeof (uint64_t), &txg, tx);
4544
4545 (void) mutex_lock(&zcl.zcl_callbacks_lock);
4546
4547 /*
4548 * Since commit callbacks don't have any ordering requirement and since
4549 * it is theoretically possible for a commit callback to be called
4550 * after an arbitrary amount of time has elapsed since its txg has been
4551 * synced, it is difficult to reliably determine whether a commit
4552 * callback hasn't been called due to high load or due to a flawed
4553 * implementation.
4554 *
4555 * In practice, we will assume that if after a certain number of txgs a
4556 * commit callback hasn't been called, then most likely there's an
4557 * implementation bug..
4558 */
4559 tmp_cb = list_head(&zcl.zcl_callbacks);
4560 if (tmp_cb != NULL &&
4561 (txg - ZTEST_COMMIT_CALLBACK_THRESH) > tmp_cb->zcd_txg) {
4562 fatal(0, "Commit callback threshold exceeded, oldest txg: %"
4563 PRIu64 ", open txg: %" PRIu64 "\n", tmp_cb->zcd_txg, txg);
4564 }
4565
4566 /*
4567 * Let's find the place to insert our callbacks.
4568 *
4569 * Even though the list is ordered by txg, it is possible for the
4570 * insertion point to not be the end because our txg may already be
4571 * quiescing at this point and other callbacks in the open txg
4572 * (from other objsets) may have sneaked in.
4573 */
4574 tmp_cb = list_tail(&zcl.zcl_callbacks);
4575 while (tmp_cb != NULL && tmp_cb->zcd_txg > txg)
4576 tmp_cb = list_prev(&zcl.zcl_callbacks, tmp_cb);
4577
4578 /* Add the 3 callbacks to the list */
4579 for (i = 0; i < 3; i++) {
4580 if (tmp_cb == NULL)
4581 list_insert_head(&zcl.zcl_callbacks, cb_data[i]);
4582 else
4583 list_insert_after(&zcl.zcl_callbacks, tmp_cb,
4584 cb_data[i]);
4585
4586 cb_data[i]->zcd_added = B_TRUE;
4587 VERIFY(!cb_data[i]->zcd_called);
4588
4589 tmp_cb = cb_data[i];
4590 }
4591
4592 (void) mutex_unlock(&zcl.zcl_callbacks_lock);
4593
4594 dmu_tx_commit(tx);
4595 }
4596
4597 /* ARGSUSED */
4598 void
ztest_dsl_prop_get_set(ztest_ds_t * zd,uint64_t id)4599 ztest_dsl_prop_get_set(ztest_ds_t *zd, uint64_t id)
4600 {
4601 zfs_prop_t proplist[] = {
4602 ZFS_PROP_CHECKSUM,
4603 ZFS_PROP_COMPRESSION,
4604 ZFS_PROP_COPIES,
4605 ZFS_PROP_DEDUP
4606 };
4607
4608 (void) rw_rdlock(&ztest_name_lock);
4609
4610 for (int p = 0; p < sizeof (proplist) / sizeof (proplist[0]); p++)
4611 (void) ztest_dsl_prop_set_uint64(zd->zd_name, proplist[p],
4612 ztest_random_dsl_prop(proplist[p]), (int)ztest_random(2));
4613
4614 (void) rw_unlock(&ztest_name_lock);
4615 }
4616
4617 /* ARGSUSED */
4618 void
ztest_spa_prop_get_set(ztest_ds_t * zd,uint64_t id)4619 ztest_spa_prop_get_set(ztest_ds_t *zd, uint64_t id)
4620 {
4621 nvlist_t *props = NULL;
4622
4623 (void) rw_rdlock(&ztest_name_lock);
4624
4625 (void) ztest_spa_prop_set_uint64(ZPOOL_PROP_DEDUPDITTO,
4626 ZIO_DEDUPDITTO_MIN + ztest_random(ZIO_DEDUPDITTO_MIN));
4627
4628 VERIFY0(spa_prop_get(ztest_spa, &props));
4629
4630 if (ztest_opts.zo_verbose >= 6)
4631 dump_nvlist(props, 4);
4632
4633 nvlist_free(props);
4634
4635 (void) rw_unlock(&ztest_name_lock);
4636 }
4637
4638 static int
user_release_one(const char * snapname,const char * holdname)4639 user_release_one(const char *snapname, const char *holdname)
4640 {
4641 nvlist_t *snaps, *holds;
4642 int error;
4643
4644 snaps = fnvlist_alloc();
4645 holds = fnvlist_alloc();
4646 fnvlist_add_boolean(holds, holdname);
4647 fnvlist_add_nvlist(snaps, snapname, holds);
4648 fnvlist_free(holds);
4649 error = dsl_dataset_user_release(snaps, NULL);
4650 fnvlist_free(snaps);
4651 return (error);
4652 }
4653
4654 /*
4655 * Test snapshot hold/release and deferred destroy.
4656 */
4657 void
ztest_dmu_snapshot_hold(ztest_ds_t * zd,uint64_t id)4658 ztest_dmu_snapshot_hold(ztest_ds_t *zd, uint64_t id)
4659 {
4660 int error;
4661 objset_t *os = zd->zd_os;
4662 objset_t *origin;
4663 char snapname[100];
4664 char fullname[100];
4665 char clonename[100];
4666 char tag[100];
4667 char osname[ZFS_MAX_DATASET_NAME_LEN];
4668 nvlist_t *holds;
4669
4670 (void) rw_rdlock(&ztest_name_lock);
4671
4672 dmu_objset_name(os, osname);
4673
4674 (void) snprintf(snapname, sizeof (snapname), "sh1_%llu", id);
4675 (void) snprintf(fullname, sizeof (fullname), "%s@%s", osname, snapname);
4676 (void) snprintf(clonename, sizeof (clonename),
4677 "%s/ch1_%llu", osname, id);
4678 (void) snprintf(tag, sizeof (tag), "tag_%llu", id);
4679
4680 /*
4681 * Clean up from any previous run.
4682 */
4683 error = dsl_destroy_head(clonename);
4684 if (error != ENOENT)
4685 ASSERT0(error);
4686 error = user_release_one(fullname, tag);
4687 if (error != ESRCH && error != ENOENT)
4688 ASSERT0(error);
4689 error = dsl_destroy_snapshot(fullname, B_FALSE);
4690 if (error != ENOENT)
4691 ASSERT0(error);
4692
4693 /*
4694 * Create snapshot, clone it, mark snap for deferred destroy,
4695 * destroy clone, verify snap was also destroyed.
4696 */
4697 error = dmu_objset_snapshot_one(osname, snapname);
4698 if (error) {
4699 if (error == ENOSPC) {
4700 ztest_record_enospc("dmu_objset_snapshot");
4701 goto out;
4702 }
4703 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
4704 }
4705
4706 error = dmu_objset_clone(clonename, fullname);
4707 if (error) {
4708 if (error == ENOSPC) {
4709 ztest_record_enospc("dmu_objset_clone");
4710 goto out;
4711 }
4712 fatal(0, "dmu_objset_clone(%s) = %d", clonename, error);
4713 }
4714
4715 error = dsl_destroy_snapshot(fullname, B_TRUE);
4716 if (error) {
4717 fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
4718 fullname, error);
4719 }
4720
4721 error = dsl_destroy_head(clonename);
4722 if (error)
4723 fatal(0, "dsl_destroy_head(%s) = %d", clonename, error);
4724
4725 error = dmu_objset_hold(fullname, FTAG, &origin);
4726 if (error != ENOENT)
4727 fatal(0, "dmu_objset_hold(%s) = %d", fullname, error);
4728
4729 /*
4730 * Create snapshot, add temporary hold, verify that we can't
4731 * destroy a held snapshot, mark for deferred destroy,
4732 * release hold, verify snapshot was destroyed.
4733 */
4734 error = dmu_objset_snapshot_one(osname, snapname);
4735 if (error) {
4736 if (error == ENOSPC) {
4737 ztest_record_enospc("dmu_objset_snapshot");
4738 goto out;
4739 }
4740 fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
4741 }
4742
4743 holds = fnvlist_alloc();
4744 fnvlist_add_string(holds, fullname, tag);
4745 error = dsl_dataset_user_hold(holds, 0, NULL);
4746 fnvlist_free(holds);
4747
4748 if (error == ENOSPC) {
4749 ztest_record_enospc("dsl_dataset_user_hold");
4750 goto out;
4751 } else if (error) {
4752 fatal(0, "dsl_dataset_user_hold(%s, %s) = %u",
4753 fullname, tag, error);
4754 }
4755
4756 error = dsl_destroy_snapshot(fullname, B_FALSE);
4757 if (error != EBUSY) {
4758 fatal(0, "dsl_destroy_snapshot(%s, B_FALSE) = %d",
4759 fullname, error);
4760 }
4761
4762 error = dsl_destroy_snapshot(fullname, B_TRUE);
4763 if (error) {
4764 fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
4765 fullname, error);
4766 }
4767
4768 error = user_release_one(fullname, tag);
4769 if (error)
4770 fatal(0, "user_release_one(%s, %s) = %d", fullname, tag, error);
4771
4772 VERIFY3U(dmu_objset_hold(fullname, FTAG, &origin), ==, ENOENT);
4773
4774 out:
4775 (void) rw_unlock(&ztest_name_lock);
4776 }
4777
4778 /*
4779 * Inject random faults into the on-disk data.
4780 */
4781 /* ARGSUSED */
4782 void
ztest_fault_inject(ztest_ds_t * zd,uint64_t id)4783 ztest_fault_inject(ztest_ds_t *zd, uint64_t id)
4784 {
4785 ztest_shared_t *zs = ztest_shared;
4786 spa_t *spa = ztest_spa;
4787 int fd;
4788 uint64_t offset;
4789 uint64_t leaves;
4790 uint64_t bad = 0x1990c0ffeedecadeULL;
4791 uint64_t top, leaf;
4792 char path0[MAXPATHLEN];
4793 char pathrand[MAXPATHLEN];
4794 size_t fsize;
4795 int bshift = SPA_MAXBLOCKSHIFT + 2;
4796 int iters = 1000;
4797 int maxfaults;
4798 int mirror_save;
4799 vdev_t *vd0 = NULL;
4800 uint64_t guid0 = 0;
4801 boolean_t islog = B_FALSE;
4802
4803 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
4804 maxfaults = MAXFAULTS();
4805 leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz;
4806 mirror_save = zs->zs_mirrors;
4807 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4808
4809 ASSERT(leaves >= 1);
4810
4811 /*
4812 * Grab the name lock as reader. There are some operations
4813 * which don't like to have their vdevs changed while
4814 * they are in progress (i.e. spa_change_guid). Those
4815 * operations will have grabbed the name lock as writer.
4816 */
4817 (void) rw_rdlock(&ztest_name_lock);
4818
4819 /*
4820 * We need SCL_STATE here because we're going to look at vd0->vdev_tsd.
4821 */
4822 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
4823
4824 if (ztest_random(2) == 0) {
4825 /*
4826 * Inject errors on a normal data device or slog device.
4827 */
4828 top = ztest_random_vdev_top(spa, B_TRUE);
4829 leaf = ztest_random(leaves) + zs->zs_splits;
4830
4831 /*
4832 * Generate paths to the first leaf in this top-level vdev,
4833 * and to the random leaf we selected. We'll induce transient
4834 * write failures and random online/offline activity on leaf 0,
4835 * and we'll write random garbage to the randomly chosen leaf.
4836 */
4837 (void) snprintf(path0, sizeof (path0), ztest_dev_template,
4838 ztest_opts.zo_dir, ztest_opts.zo_pool,
4839 top * leaves + zs->zs_splits);
4840 (void) snprintf(pathrand, sizeof (pathrand), ztest_dev_template,
4841 ztest_opts.zo_dir, ztest_opts.zo_pool,
4842 top * leaves + leaf);
4843
4844 vd0 = vdev_lookup_by_path(spa->spa_root_vdev, path0);
4845 if (vd0 != NULL && vd0->vdev_top->vdev_islog)
4846 islog = B_TRUE;
4847
4848 /*
4849 * If the top-level vdev needs to be resilvered
4850 * then we only allow faults on the device that is
4851 * resilvering.
4852 */
4853 if (vd0 != NULL && maxfaults != 1 &&
4854 (!vdev_resilver_needed(vd0->vdev_top, NULL, NULL) ||
4855 vd0->vdev_resilver_txg != 0)) {
4856 /*
4857 * Make vd0 explicitly claim to be unreadable,
4858 * or unwriteable, or reach behind its back
4859 * and close the underlying fd. We can do this if
4860 * maxfaults == 0 because we'll fail and reexecute,
4861 * and we can do it if maxfaults >= 2 because we'll
4862 * have enough redundancy. If maxfaults == 1, the
4863 * combination of this with injection of random data
4864 * corruption below exceeds the pool's fault tolerance.
4865 */
4866 vdev_file_t *vf = vd0->vdev_tsd;
4867
4868 if (vf != NULL && ztest_random(3) == 0) {
4869 (void) close(vf->vf_vnode->v_fd);
4870 vf->vf_vnode->v_fd = -1;
4871 } else if (ztest_random(2) == 0) {
4872 vd0->vdev_cant_read = B_TRUE;
4873 } else {
4874 vd0->vdev_cant_write = B_TRUE;
4875 }
4876 guid0 = vd0->vdev_guid;
4877 }
4878 } else {
4879 /*
4880 * Inject errors on an l2cache device.
4881 */
4882 spa_aux_vdev_t *sav = &spa->spa_l2cache;
4883
4884 if (sav->sav_count == 0) {
4885 spa_config_exit(spa, SCL_STATE, FTAG);
4886 (void) rw_unlock(&ztest_name_lock);
4887 return;
4888 }
4889 vd0 = sav->sav_vdevs[ztest_random(sav->sav_count)];
4890 guid0 = vd0->vdev_guid;
4891 (void) strcpy(path0, vd0->vdev_path);
4892 (void) strcpy(pathrand, vd0->vdev_path);
4893
4894 leaf = 0;
4895 leaves = 1;
4896 maxfaults = INT_MAX; /* no limit on cache devices */
4897 }
4898
4899 spa_config_exit(spa, SCL_STATE, FTAG);
4900 (void) rw_unlock(&ztest_name_lock);
4901
4902 /*
4903 * If we can tolerate two or more faults, or we're dealing
4904 * with a slog, randomly online/offline vd0.
4905 */
4906 if ((maxfaults >= 2 || islog) && guid0 != 0) {
4907 if (ztest_random(10) < 6) {
4908 int flags = (ztest_random(2) == 0 ?
4909 ZFS_OFFLINE_TEMPORARY : 0);
4910
4911 /*
4912 * We have to grab the zs_name_lock as writer to
4913 * prevent a race between offlining a slog and
4914 * destroying a dataset. Offlining the slog will
4915 * grab a reference on the dataset which may cause
4916 * dmu_objset_destroy() to fail with EBUSY thus
4917 * leaving the dataset in an inconsistent state.
4918 */
4919 if (islog)
4920 (void) rw_wrlock(&ztest_name_lock);
4921
4922 VERIFY(vdev_offline(spa, guid0, flags) != EBUSY);
4923
4924 if (islog)
4925 (void) rw_unlock(&ztest_name_lock);
4926 } else {
4927 /*
4928 * Ideally we would like to be able to randomly
4929 * call vdev_[on|off]line without holding locks
4930 * to force unpredictable failures but the side
4931 * effects of vdev_[on|off]line prevent us from
4932 * doing so. We grab the ztest_vdev_lock here to
4933 * prevent a race between injection testing and
4934 * aux_vdev removal.
4935 */
4936 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
4937 (void) vdev_online(spa, guid0, 0, NULL);
4938 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4939 }
4940 }
4941
4942 if (maxfaults == 0)
4943 return;
4944
4945 /*
4946 * We have at least single-fault tolerance, so inject data corruption.
4947 */
4948 fd = open(pathrand, O_RDWR);
4949
4950 if (fd == -1) /* we hit a gap in the device namespace */
4951 return;
4952
4953 fsize = lseek(fd, 0, SEEK_END);
4954
4955 while (--iters != 0) {
4956 /*
4957 * The offset must be chosen carefully to ensure that
4958 * we do not inject a given logical block with errors
4959 * on two different leaf devices, because ZFS can not
4960 * tolerate that (if maxfaults==1).
4961 *
4962 * We divide each leaf into chunks of size
4963 * (# leaves * SPA_MAXBLOCKSIZE * 4). Within each chunk
4964 * there is a series of ranges to which we can inject errors.
4965 * Each range can accept errors on only a single leaf vdev.
4966 * The error injection ranges are separated by ranges
4967 * which we will not inject errors on any device (DMZs).
4968 * Each DMZ must be large enough such that a single block
4969 * can not straddle it, so that a single block can not be
4970 * a target in two different injection ranges (on different
4971 * leaf vdevs).
4972 *
4973 * For example, with 3 leaves, each chunk looks like:
4974 * 0 to 32M: injection range for leaf 0
4975 * 32M to 64M: DMZ - no injection allowed
4976 * 64M to 96M: injection range for leaf 1
4977 * 96M to 128M: DMZ - no injection allowed
4978 * 128M to 160M: injection range for leaf 2
4979 * 160M to 192M: DMZ - no injection allowed
4980 */
4981 offset = ztest_random(fsize / (leaves << bshift)) *
4982 (leaves << bshift) + (leaf << bshift) +
4983 (ztest_random(1ULL << (bshift - 1)) & -8ULL);
4984
4985 /*
4986 * Only allow damage to the labels at one end of the vdev.
4987 *
4988 * If all labels are damaged, the device will be totally
4989 * inaccessible, which will result in loss of data,
4990 * because we also damage (parts of) the other side of
4991 * the mirror/raidz.
4992 *
4993 * Additionally, we will always have both an even and an
4994 * odd label, so that we can handle crashes in the
4995 * middle of vdev_config_sync().
4996 */
4997 if ((leaf & 1) == 0 && offset < VDEV_LABEL_START_SIZE)
4998 continue;
4999
5000 /*
5001 * The two end labels are stored at the "end" of the disk, but
5002 * the end of the disk (vdev_psize) is aligned to
5003 * sizeof (vdev_label_t).
5004 */
5005 uint64_t psize = P2ALIGN(fsize, sizeof (vdev_label_t));
5006 if ((leaf & 1) == 1 &&
5007 offset + sizeof (bad) > psize - VDEV_LABEL_END_SIZE)
5008 continue;
5009
5010 VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
5011 if (mirror_save != zs->zs_mirrors) {
5012 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
5013 (void) close(fd);
5014 return;
5015 }
5016
5017 if (pwrite(fd, &bad, sizeof (bad), offset) != sizeof (bad))
5018 fatal(1, "can't inject bad word at 0x%llx in %s",
5019 offset, pathrand);
5020
5021 VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
5022
5023 if (ztest_opts.zo_verbose >= 7)
5024 (void) printf("injected bad word into %s,"
5025 " offset 0x%llx\n", pathrand, (u_longlong_t)offset);
5026 }
5027
5028 (void) close(fd);
5029 }
5030
5031 /*
5032 * Verify that DDT repair works as expected.
5033 */
5034 void
ztest_ddt_repair(ztest_ds_t * zd,uint64_t id)5035 ztest_ddt_repair(ztest_ds_t *zd, uint64_t id)
5036 {
5037 ztest_shared_t *zs = ztest_shared;
5038 spa_t *spa = ztest_spa;
5039 objset_t *os = zd->zd_os;
5040 ztest_od_t od[1];
5041 uint64_t object, blocksize, txg, pattern, psize;
5042 enum zio_checksum checksum = spa_dedup_checksum(spa);
5043 dmu_buf_t *db;
5044 dmu_tx_t *tx;
5045 void *buf;
5046 blkptr_t blk;
5047 int copies = 2 * ZIO_DEDUPDITTO_MIN;
5048
5049 blocksize = ztest_random_blocksize();
5050 blocksize = MIN(blocksize, 2048); /* because we write so many */
5051
5052 ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
5053
5054 if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
5055 return;
5056
5057 /*
5058 * Take the name lock as writer to prevent anyone else from changing
5059 * the pool and dataset properies we need to maintain during this test.
5060 */
5061 (void) rw_wrlock(&ztest_name_lock);
5062
5063 if (ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_DEDUP, checksum,
5064 B_FALSE) != 0 ||
5065 ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_COPIES, 1,
5066 B_FALSE) != 0) {
5067 (void) rw_unlock(&ztest_name_lock);
5068 return;
5069 }
5070
5071 dmu_objset_stats_t dds;
5072 dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
5073 dmu_objset_fast_stat(os, &dds);
5074 dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
5075
5076 object = od[0].od_object;
5077 blocksize = od[0].od_blocksize;
5078 pattern = zs->zs_guid ^ dds.dds_guid;
5079
5080 ASSERT(object != 0);
5081
5082 tx = dmu_tx_create(os);
5083 dmu_tx_hold_write(tx, object, 0, copies * blocksize);
5084 txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
5085 if (txg == 0) {
5086 (void) rw_unlock(&ztest_name_lock);
5087 return;
5088 }
5089
5090 /*
5091 * Write all the copies of our block.
5092 */
5093 for (int i = 0; i < copies; i++) {
5094 uint64_t offset = i * blocksize;
5095 int error = dmu_buf_hold(os, object, offset, FTAG, &db,
5096 DMU_READ_NO_PREFETCH);
5097 if (error != 0) {
5098 fatal(B_FALSE, "dmu_buf_hold(%p, %llu, %llu) = %u",
5099 os, (long long)object, (long long) offset, error);
5100 }
5101 ASSERT(db->db_offset == offset);
5102 ASSERT(db->db_size == blocksize);
5103 ASSERT(ztest_pattern_match(db->db_data, db->db_size, pattern) ||
5104 ztest_pattern_match(db->db_data, db->db_size, 0ULL));
5105 dmu_buf_will_fill(db, tx);
5106 ztest_pattern_set(db->db_data, db->db_size, pattern);
5107 dmu_buf_rele(db, FTAG);
5108 }
5109
5110 dmu_tx_commit(tx);
5111 txg_wait_synced(spa_get_dsl(spa), txg);
5112
5113 /*
5114 * Find out what block we got.
5115 */
5116 VERIFY0(dmu_buf_hold(os, object, 0, FTAG, &db,
5117 DMU_READ_NO_PREFETCH));
5118 blk = *((dmu_buf_impl_t *)db)->db_blkptr;
5119 dmu_buf_rele(db, FTAG);
5120
5121 /*
5122 * Damage the block. Dedup-ditto will save us when we read it later.
5123 */
5124 psize = BP_GET_PSIZE(&blk);
5125 buf = zio_buf_alloc(psize);
5126 ztest_pattern_set(buf, psize, ~pattern);
5127
5128 (void) zio_wait(zio_rewrite(NULL, spa, 0, &blk,
5129 buf, psize, NULL, NULL, ZIO_PRIORITY_SYNC_WRITE,
5130 ZIO_FLAG_CANFAIL | ZIO_FLAG_INDUCE_DAMAGE, NULL));
5131
5132 zio_buf_free(buf, psize);
5133
5134 (void) rw_unlock(&ztest_name_lock);
5135 }
5136
5137 /*
5138 * Scrub the pool.
5139 */
5140 /* ARGSUSED */
5141 void
ztest_scrub(ztest_ds_t * zd,uint64_t id)5142 ztest_scrub(ztest_ds_t *zd, uint64_t id)
5143 {
5144 spa_t *spa = ztest_spa;
5145
5146 (void) spa_scan(spa, POOL_SCAN_SCRUB);
5147 (void) poll(NULL, 0, 100); /* wait a moment, then force a restart */
5148 (void) spa_scan(spa, POOL_SCAN_SCRUB);
5149 }
5150
5151 /*
5152 * Change the guid for the pool.
5153 */
5154 /* ARGSUSED */
5155 void
ztest_reguid(ztest_ds_t * zd,uint64_t id)5156 ztest_reguid(ztest_ds_t *zd, uint64_t id)
5157 {
5158 spa_t *spa = ztest_spa;
5159 uint64_t orig, load;
5160 int error;
5161
5162 orig = spa_guid(spa);
5163 load = spa_load_guid(spa);
5164
5165 (void) rw_wrlock(&ztest_name_lock);
5166 error = spa_change_guid(spa);
5167 (void) rw_unlock(&ztest_name_lock);
5168
5169 if (error != 0)
5170 return;
5171
5172 if (ztest_opts.zo_verbose >= 4) {
5173 (void) printf("Changed guid old %llu -> %llu\n",
5174 (u_longlong_t)orig, (u_longlong_t)spa_guid(spa));
5175 }
5176
5177 VERIFY3U(orig, !=, spa_guid(spa));
5178 VERIFY3U(load, ==, spa_load_guid(spa));
5179 }
5180
5181 /*
5182 * Rename the pool to a different name and then rename it back.
5183 */
5184 /* ARGSUSED */
5185 void
ztest_spa_rename(ztest_ds_t * zd,uint64_t id)5186 ztest_spa_rename(ztest_ds_t *zd, uint64_t id)
5187 {
5188 char *oldname, *newname;
5189 spa_t *spa;
5190
5191 (void) rw_wrlock(&ztest_name_lock);
5192
5193 oldname = ztest_opts.zo_pool;
5194 newname = umem_alloc(strlen(oldname) + 5, UMEM_NOFAIL);
5195 (void) strcpy(newname, oldname);
5196 (void) strcat(newname, "_tmp");
5197
5198 /*
5199 * Do the rename
5200 */
5201 VERIFY3U(0, ==, spa_rename(oldname, newname));
5202
5203 /*
5204 * Try to open it under the old name, which shouldn't exist
5205 */
5206 VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
5207
5208 /*
5209 * Open it under the new name and make sure it's still the same spa_t.
5210 */
5211 VERIFY3U(0, ==, spa_open(newname, &spa, FTAG));
5212
5213 ASSERT(spa == ztest_spa);
5214 spa_close(spa, FTAG);
5215
5216 /*
5217 * Rename it back to the original
5218 */
5219 VERIFY3U(0, ==, spa_rename(newname, oldname));
5220
5221 /*
5222 * Make sure it can still be opened
5223 */
5224 VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG));
5225
5226 ASSERT(spa == ztest_spa);
5227 spa_close(spa, FTAG);
5228
5229 umem_free(newname, strlen(newname) + 1);
5230
5231 (void) rw_unlock(&ztest_name_lock);
5232 }
5233
5234 /*
5235 * Verify pool integrity by running zdb.
5236 */
5237 static void
ztest_run_zdb(char * pool)5238 ztest_run_zdb(char *pool)
5239 {
5240 int status;
5241 char zdb[MAXPATHLEN + MAXNAMELEN + 20];
5242 char zbuf[1024];
5243 char *bin;
5244 char *ztest;
5245 char *isa;
5246 int isalen;
5247 FILE *fp;
5248
5249 strlcpy(zdb, "/usr/bin/ztest", sizeof(zdb));
5250
5251 /* zdb lives in /usr/sbin, while ztest lives in /usr/bin */
5252 bin = strstr(zdb, "/usr/bin/");
5253 ztest = strstr(bin, "/ztest");
5254 isa = bin + 8;
5255 isalen = ztest - isa;
5256 isa = strdup(isa);
5257 /* LINTED */
5258 (void) sprintf(bin,
5259 "/usr/sbin%.*s/zdb -bcc%s%s -d -U %s %s",
5260 isalen,
5261 isa,
5262 ztest_opts.zo_verbose >= 3 ? "s" : "",
5263 ztest_opts.zo_verbose >= 4 ? "v" : "",
5264 spa_config_path,
5265 pool);
5266 free(isa);
5267
5268 if (ztest_opts.zo_verbose >= 5)
5269 (void) printf("Executing %s\n", strstr(zdb, "zdb "));
5270
5271 fp = popen(zdb, "r");
5272 assert(fp != NULL);
5273
5274 while (fgets(zbuf, sizeof (zbuf), fp) != NULL)
5275 if (ztest_opts.zo_verbose >= 3)
5276 (void) printf("%s", zbuf);
5277
5278 status = pclose(fp);
5279
5280 if (status == 0)
5281 return;
5282
5283 ztest_dump_core = 0;
5284 if (WIFEXITED(status))
5285 fatal(0, "'%s' exit code %d", zdb, WEXITSTATUS(status));
5286 else
5287 fatal(0, "'%s' died with signal %d", zdb, WTERMSIG(status));
5288 }
5289
5290 static void
ztest_walk_pool_directory(char * header)5291 ztest_walk_pool_directory(char *header)
5292 {
5293 spa_t *spa = NULL;
5294
5295 if (ztest_opts.zo_verbose >= 6)
5296 (void) printf("%s\n", header);
5297
5298 mutex_enter(&spa_namespace_lock);
5299 while ((spa = spa_next(spa)) != NULL)
5300 if (ztest_opts.zo_verbose >= 6)
5301 (void) printf("\t%s\n", spa_name(spa));
5302 mutex_exit(&spa_namespace_lock);
5303 }
5304
5305 static void
ztest_spa_import_export(char * oldname,char * newname)5306 ztest_spa_import_export(char *oldname, char *newname)
5307 {
5308 nvlist_t *config, *newconfig;
5309 uint64_t pool_guid;
5310 spa_t *spa;
5311 int error;
5312
5313 if (ztest_opts.zo_verbose >= 4) {
5314 (void) printf("import/export: old = %s, new = %s\n",
5315 oldname, newname);
5316 }
5317
5318 /*
5319 * Clean up from previous runs.
5320 */
5321 (void) spa_destroy(newname);
5322
5323 /*
5324 * Get the pool's configuration and guid.
5325 */
5326 VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG));
5327
5328 /*
5329 * Kick off a scrub to tickle scrub/export races.
5330 */
5331 if (ztest_random(2) == 0)
5332 (void) spa_scan(spa, POOL_SCAN_SCRUB);
5333
5334 pool_guid = spa_guid(spa);
5335 spa_close(spa, FTAG);
5336
5337 ztest_walk_pool_directory("pools before export");
5338
5339 /*
5340 * Export it.
5341 */
5342 VERIFY3U(0, ==, spa_export(oldname, &config, B_FALSE, B_FALSE));
5343
5344 ztest_walk_pool_directory("pools after export");
5345
5346 /*
5347 * Try to import it.
5348 */
5349 newconfig = spa_tryimport(config);
5350 ASSERT(newconfig != NULL);
5351 nvlist_free(newconfig);
5352
5353 /*
5354 * Import it under the new name.
5355 */
5356 error = spa_import(newname, config, NULL, 0);
5357 if (error != 0) {
5358 dump_nvlist(config, 0);
5359 fatal(B_FALSE, "couldn't import pool %s as %s: error %u",
5360 oldname, newname, error);
5361 }
5362
5363 ztest_walk_pool_directory("pools after import");
5364
5365 /*
5366 * Try to import it again -- should fail with EEXIST.
5367 */
5368 VERIFY3U(EEXIST, ==, spa_import(newname, config, NULL, 0));
5369
5370 /*
5371 * Try to import it under a different name -- should fail with EEXIST.
5372 */
5373 VERIFY3U(EEXIST, ==, spa_import(oldname, config, NULL, 0));
5374
5375 /*
5376 * Verify that the pool is no longer visible under the old name.
5377 */
5378 VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
5379
5380 /*
5381 * Verify that we can open and close the pool using the new name.
5382 */
5383 VERIFY3U(0, ==, spa_open(newname, &spa, FTAG));
5384 ASSERT(pool_guid == spa_guid(spa));
5385 spa_close(spa, FTAG);
5386
5387 nvlist_free(config);
5388 }
5389
5390 static void
ztest_resume(spa_t * spa)5391 ztest_resume(spa_t *spa)
5392 {
5393 if (spa_suspended(spa) && ztest_opts.zo_verbose >= 6)
5394 (void) printf("resuming from suspended state\n");
5395 spa_vdev_state_enter(spa, SCL_NONE);
5396 vdev_clear(spa, NULL);
5397 (void) spa_vdev_state_exit(spa, NULL, 0);
5398 (void) zio_resume(spa);
5399 }
5400
5401 static void *
ztest_resume_thread(void * arg)5402 ztest_resume_thread(void *arg)
5403 {
5404 spa_t *spa = arg;
5405
5406 while (!ztest_exiting) {
5407 if (spa_suspended(spa))
5408 ztest_resume(spa);
5409 (void) poll(NULL, 0, 100);
5410
5411 /*
5412 * Periodically change the zfs_compressed_arc_enabled setting.
5413 */
5414 if (ztest_random(10) == 0)
5415 zfs_compressed_arc_enabled = ztest_random(2);
5416 }
5417 return (NULL);
5418 }
5419
5420 static void *
ztest_deadman_thread(void * arg)5421 ztest_deadman_thread(void *arg)
5422 {
5423 ztest_shared_t *zs = arg;
5424 spa_t *spa = ztest_spa;
5425 hrtime_t delta, total = 0;
5426
5427 for (;;) {
5428 delta = zs->zs_thread_stop - zs->zs_thread_start +
5429 MSEC2NSEC(zfs_deadman_synctime_ms);
5430
5431 (void) poll(NULL, 0, (int)NSEC2MSEC(delta));
5432
5433 /*
5434 * If the pool is suspended then fail immediately. Otherwise,
5435 * check to see if the pool is making any progress. If
5436 * vdev_deadman() discovers that there hasn't been any recent
5437 * I/Os then it will end up aborting the tests.
5438 */
5439 if (spa_suspended(spa) || spa->spa_root_vdev == NULL) {
5440 fatal(0, "aborting test after %llu seconds because "
5441 "pool has transitioned to a suspended state.",
5442 zfs_deadman_synctime_ms / 1000);
5443 return (NULL);
5444 }
5445 vdev_deadman(spa->spa_root_vdev);
5446
5447 total += zfs_deadman_synctime_ms/1000;
5448 (void) printf("ztest has been running for %lld seconds\n",
5449 total);
5450 }
5451 }
5452
5453 static void
ztest_execute(int test,ztest_info_t * zi,uint64_t id)5454 ztest_execute(int test, ztest_info_t *zi, uint64_t id)
5455 {
5456 ztest_ds_t *zd = &ztest_ds[id % ztest_opts.zo_datasets];
5457 ztest_shared_callstate_t *zc = ZTEST_GET_SHARED_CALLSTATE(test);
5458 hrtime_t functime = gethrtime();
5459
5460 for (int i = 0; i < zi->zi_iters; i++)
5461 zi->zi_func(zd, id);
5462
5463 functime = gethrtime() - functime;
5464
5465 atomic_add_64(&zc->zc_count, 1);
5466 atomic_add_64(&zc->zc_time, functime);
5467
5468 if (ztest_opts.zo_verbose >= 4) {
5469 Dl_info dli;
5470 (void) dladdr((void *)zi->zi_func, &dli);
5471 (void) printf("%6.2f sec in %s\n",
5472 (double)functime / NANOSEC, dli.dli_sname);
5473 }
5474 }
5475
5476 static void *
ztest_thread(void * arg)5477 ztest_thread(void *arg)
5478 {
5479 int rand;
5480 uint64_t id = (uintptr_t)arg;
5481 ztest_shared_t *zs = ztest_shared;
5482 uint64_t call_next;
5483 hrtime_t now;
5484 ztest_info_t *zi;
5485 ztest_shared_callstate_t *zc;
5486
5487 while ((now = gethrtime()) < zs->zs_thread_stop) {
5488 /*
5489 * See if it's time to force a crash.
5490 */
5491 if (now > zs->zs_thread_kill)
5492 ztest_kill(zs);
5493
5494 /*
5495 * If we're getting ENOSPC with some regularity, stop.
5496 */
5497 if (zs->zs_enospc_count > 10)
5498 break;
5499
5500 /*
5501 * Pick a random function to execute.
5502 */
5503 rand = ztest_random(ZTEST_FUNCS);
5504 zi = &ztest_info[rand];
5505 zc = ZTEST_GET_SHARED_CALLSTATE(rand);
5506 call_next = zc->zc_next;
5507
5508 if (now >= call_next &&
5509 atomic_cas_64(&zc->zc_next, call_next, call_next +
5510 ztest_random(2 * zi->zi_interval[0] + 1)) == call_next) {
5511 ztest_execute(rand, zi, id);
5512 }
5513 }
5514
5515 return (NULL);
5516 }
5517
5518 static void
ztest_dataset_name(char * dsname,char * pool,int d)5519 ztest_dataset_name(char *dsname, char *pool, int d)
5520 {
5521 (void) snprintf(dsname, ZFS_MAX_DATASET_NAME_LEN, "%s/ds_%d", pool, d);
5522 }
5523
5524 static void
ztest_dataset_destroy(int d)5525 ztest_dataset_destroy(int d)
5526 {
5527 char name[ZFS_MAX_DATASET_NAME_LEN];
5528
5529 ztest_dataset_name(name, ztest_opts.zo_pool, d);
5530
5531 if (ztest_opts.zo_verbose >= 3)
5532 (void) printf("Destroying %s to free up space\n", name);
5533
5534 /*
5535 * Cleanup any non-standard clones and snapshots. In general,
5536 * ztest thread t operates on dataset (t % zopt_datasets),
5537 * so there may be more than one thing to clean up.
5538 */
5539 for (int t = d; t < ztest_opts.zo_threads;
5540 t += ztest_opts.zo_datasets) {
5541 ztest_dsl_dataset_cleanup(name, t);
5542 }
5543
5544 (void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
5545 DS_FIND_SNAPSHOTS | DS_FIND_CHILDREN);
5546 }
5547
5548 static void
ztest_dataset_dirobj_verify(ztest_ds_t * zd)5549 ztest_dataset_dirobj_verify(ztest_ds_t *zd)
5550 {
5551 uint64_t usedobjs, dirobjs, scratch;
5552
5553 /*
5554 * ZTEST_DIROBJ is the object directory for the entire dataset.
5555 * Therefore, the number of objects in use should equal the
5556 * number of ZTEST_DIROBJ entries, +1 for ZTEST_DIROBJ itself.
5557 * If not, we have an object leak.
5558 *
5559 * Note that we can only check this in ztest_dataset_open(),
5560 * when the open-context and syncing-context values agree.
5561 * That's because zap_count() returns the open-context value,
5562 * while dmu_objset_space() returns the rootbp fill count.
5563 */
5564 VERIFY3U(0, ==, zap_count(zd->zd_os, ZTEST_DIROBJ, &dirobjs));
5565 dmu_objset_space(zd->zd_os, &scratch, &scratch, &usedobjs, &scratch);
5566 ASSERT3U(dirobjs + 1, ==, usedobjs);
5567 }
5568
5569 static int
ztest_dataset_open(int d)5570 ztest_dataset_open(int d)
5571 {
5572 ztest_ds_t *zd = &ztest_ds[d];
5573 uint64_t committed_seq = ZTEST_GET_SHARED_DS(d)->zd_seq;
5574 objset_t *os;
5575 zilog_t *zilog;
5576 char name[ZFS_MAX_DATASET_NAME_LEN];
5577 int error;
5578
5579 ztest_dataset_name(name, ztest_opts.zo_pool, d);
5580
5581 (void) rw_rdlock(&ztest_name_lock);
5582
5583 error = ztest_dataset_create(name);
5584 if (error == ENOSPC) {
5585 (void) rw_unlock(&ztest_name_lock);
5586 ztest_record_enospc(FTAG);
5587 return (error);
5588 }
5589 ASSERT(error == 0 || error == EEXIST);
5590
5591 VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, zd, &os));
5592 (void) rw_unlock(&ztest_name_lock);
5593
5594 ztest_zd_init(zd, ZTEST_GET_SHARED_DS(d), os);
5595
5596 zilog = zd->zd_zilog;
5597
5598 if (zilog->zl_header->zh_claim_lr_seq != 0 &&
5599 zilog->zl_header->zh_claim_lr_seq < committed_seq)
5600 fatal(0, "missing log records: claimed %llu < committed %llu",
5601 zilog->zl_header->zh_claim_lr_seq, committed_seq);
5602
5603 ztest_dataset_dirobj_verify(zd);
5604
5605 zil_replay(os, zd, ztest_replay_vector);
5606
5607 ztest_dataset_dirobj_verify(zd);
5608
5609 if (ztest_opts.zo_verbose >= 6)
5610 (void) printf("%s replay %llu blocks, %llu records, seq %llu\n",
5611 zd->zd_name,
5612 (u_longlong_t)zilog->zl_parse_blk_count,
5613 (u_longlong_t)zilog->zl_parse_lr_count,
5614 (u_longlong_t)zilog->zl_replaying_seq);
5615
5616 zilog = zil_open(os, ztest_get_data);
5617
5618 if (zilog->zl_replaying_seq != 0 &&
5619 zilog->zl_replaying_seq < committed_seq)
5620 fatal(0, "missing log records: replayed %llu < committed %llu",
5621 zilog->zl_replaying_seq, committed_seq);
5622
5623 return (0);
5624 }
5625
5626 static void
ztest_dataset_close(int d)5627 ztest_dataset_close(int d)
5628 {
5629 ztest_ds_t *zd = &ztest_ds[d];
5630
5631 zil_close(zd->zd_zilog);
5632 dmu_objset_disown(zd->zd_os, zd);
5633
5634 ztest_zd_fini(zd);
5635 }
5636
5637 /*
5638 * Kick off threads to run tests on all datasets in parallel.
5639 */
5640 static void
ztest_run(ztest_shared_t * zs)5641 ztest_run(ztest_shared_t *zs)
5642 {
5643 thread_t *tid;
5644 spa_t *spa;
5645 objset_t *os;
5646 thread_t resume_tid;
5647 int error;
5648
5649 ztest_exiting = B_FALSE;
5650
5651 /*
5652 * Initialize parent/child shared state.
5653 */
5654 VERIFY(_mutex_init(&ztest_vdev_lock, USYNC_THREAD, NULL) == 0);
5655 VERIFY(rwlock_init(&ztest_name_lock, USYNC_THREAD, NULL) == 0);
5656
5657 zs->zs_thread_start = gethrtime();
5658 zs->zs_thread_stop =
5659 zs->zs_thread_start + ztest_opts.zo_passtime * NANOSEC;
5660 zs->zs_thread_stop = MIN(zs->zs_thread_stop, zs->zs_proc_stop);
5661 zs->zs_thread_kill = zs->zs_thread_stop;
5662 if (ztest_random(100) < ztest_opts.zo_killrate) {
5663 zs->zs_thread_kill -=
5664 ztest_random(ztest_opts.zo_passtime * NANOSEC);
5665 }
5666
5667 (void) _mutex_init(&zcl.zcl_callbacks_lock, USYNC_THREAD, NULL);
5668
5669 list_create(&zcl.zcl_callbacks, sizeof (ztest_cb_data_t),
5670 offsetof(ztest_cb_data_t, zcd_node));
5671
5672 /*
5673 * Open our pool.
5674 */
5675 kernel_init(FREAD | FWRITE);
5676 VERIFY0(spa_open(ztest_opts.zo_pool, &spa, FTAG));
5677 spa->spa_debug = B_TRUE;
5678 metaslab_preload_limit = ztest_random(20) + 1;
5679 ztest_spa = spa;
5680
5681 dmu_objset_stats_t dds;
5682 VERIFY0(dmu_objset_own(ztest_opts.zo_pool,
5683 DMU_OST_ANY, B_TRUE, FTAG, &os));
5684 dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
5685 dmu_objset_fast_stat(os, &dds);
5686 dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
5687 zs->zs_guid = dds.dds_guid;
5688 dmu_objset_disown(os, FTAG);
5689
5690 spa->spa_dedup_ditto = 2 * ZIO_DEDUPDITTO_MIN;
5691
5692 /*
5693 * We don't expect the pool to suspend unless maxfaults == 0,
5694 * in which case ztest_fault_inject() temporarily takes away
5695 * the only valid replica.
5696 */
5697 if (MAXFAULTS() == 0)
5698 spa->spa_failmode = ZIO_FAILURE_MODE_WAIT;
5699 else
5700 spa->spa_failmode = ZIO_FAILURE_MODE_PANIC;
5701
5702 /*
5703 * Create a thread to periodically resume suspended I/O.
5704 */
5705 VERIFY(thr_create(0, 0, ztest_resume_thread, spa, THR_BOUND,
5706 &resume_tid) == 0);
5707
5708 /*
5709 * Create a deadman thread to abort() if we hang.
5710 */
5711 VERIFY(thr_create(0, 0, ztest_deadman_thread, zs, THR_BOUND,
5712 NULL) == 0);
5713
5714 /*
5715 * Verify that we can safely inquire about about any object,
5716 * whether it's allocated or not. To make it interesting,
5717 * we probe a 5-wide window around each power of two.
5718 * This hits all edge cases, including zero and the max.
5719 */
5720 for (int t = 0; t < 64; t++) {
5721 for (int d = -5; d <= 5; d++) {
5722 error = dmu_object_info(spa->spa_meta_objset,
5723 (1ULL << t) + d, NULL);
5724 ASSERT(error == 0 || error == ENOENT ||
5725 error == EINVAL);
5726 }
5727 }
5728
5729 /*
5730 * If we got any ENOSPC errors on the previous run, destroy something.
5731 */
5732 if (zs->zs_enospc_count != 0) {
5733 int d = ztest_random(ztest_opts.zo_datasets);
5734 ztest_dataset_destroy(d);
5735 }
5736 zs->zs_enospc_count = 0;
5737
5738 tid = umem_zalloc(ztest_opts.zo_threads * sizeof (thread_t),
5739 UMEM_NOFAIL);
5740
5741 if (ztest_opts.zo_verbose >= 4)
5742 (void) printf("starting main threads...\n");
5743
5744 /*
5745 * Kick off all the tests that run in parallel.
5746 */
5747 for (int t = 0; t < ztest_opts.zo_threads; t++) {
5748 if (t < ztest_opts.zo_datasets &&
5749 ztest_dataset_open(t) != 0)
5750 return;
5751 VERIFY(thr_create(0, 0, ztest_thread, (void *)(uintptr_t)t,
5752 THR_BOUND, &tid[t]) == 0);
5753 }
5754
5755 /*
5756 * Wait for all of the tests to complete. We go in reverse order
5757 * so we don't close datasets while threads are still using them.
5758 */
5759 for (int t = ztest_opts.zo_threads - 1; t >= 0; t--) {
5760 VERIFY(thr_join(tid[t], NULL, NULL) == 0);
5761 if (t < ztest_opts.zo_datasets)
5762 ztest_dataset_close(t);
5763 }
5764
5765 txg_wait_synced(spa_get_dsl(spa), 0);
5766
5767 zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(spa));
5768 zs->zs_space = metaslab_class_get_space(spa_normal_class(spa));
5769 zfs_dbgmsg_print(FTAG);
5770
5771 umem_free(tid, ztest_opts.zo_threads * sizeof (thread_t));
5772
5773 /* Kill the resume thread */
5774 ztest_exiting = B_TRUE;
5775 VERIFY(thr_join(resume_tid, NULL, NULL) == 0);
5776 ztest_resume(spa);
5777
5778 /*
5779 * Right before closing the pool, kick off a bunch of async I/O;
5780 * spa_close() should wait for it to complete.
5781 */
5782 for (uint64_t object = 1; object < 50; object++) {
5783 dmu_prefetch(spa->spa_meta_objset, object, 0, 0, 1ULL << 20,
5784 ZIO_PRIORITY_SYNC_READ);
5785 }
5786
5787 spa_close(spa, FTAG);
5788
5789 /*
5790 * Verify that we can loop over all pools.
5791 */
5792 mutex_enter(&spa_namespace_lock);
5793 for (spa = spa_next(NULL); spa != NULL; spa = spa_next(spa))
5794 if (ztest_opts.zo_verbose > 3)
5795 (void) printf("spa_next: found %s\n", spa_name(spa));
5796 mutex_exit(&spa_namespace_lock);
5797
5798 /*
5799 * Verify that we can export the pool and reimport it under a
5800 * different name.
5801 */
5802 if (ztest_random(2) == 0) {
5803 char name[ZFS_MAX_DATASET_NAME_LEN];
5804 (void) snprintf(name, sizeof (name), "%s_import",
5805 ztest_opts.zo_pool);
5806 ztest_spa_import_export(ztest_opts.zo_pool, name);
5807 ztest_spa_import_export(name, ztest_opts.zo_pool);
5808 }
5809
5810 kernel_fini();
5811
5812 list_destroy(&zcl.zcl_callbacks);
5813
5814 (void) _mutex_destroy(&zcl.zcl_callbacks_lock);
5815
5816 (void) rwlock_destroy(&ztest_name_lock);
5817 (void) _mutex_destroy(&ztest_vdev_lock);
5818 }
5819
5820 static void
ztest_freeze(void)5821 ztest_freeze(void)
5822 {
5823 ztest_ds_t *zd = &ztest_ds[0];
5824 spa_t *spa;
5825 int numloops = 0;
5826
5827 if (ztest_opts.zo_verbose >= 3)
5828 (void) printf("testing spa_freeze()...\n");
5829
5830 kernel_init(FREAD | FWRITE);
5831 VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
5832 VERIFY3U(0, ==, ztest_dataset_open(0));
5833 spa->spa_debug = B_TRUE;
5834 ztest_spa = spa;
5835
5836 /*
5837 * Force the first log block to be transactionally allocated.
5838 * We have to do this before we freeze the pool -- otherwise
5839 * the log chain won't be anchored.
5840 */
5841 while (BP_IS_HOLE(&zd->zd_zilog->zl_header->zh_log)) {
5842 ztest_dmu_object_alloc_free(zd, 0);
5843 zil_commit(zd->zd_zilog, 0);
5844 }
5845
5846 txg_wait_synced(spa_get_dsl(spa), 0);
5847
5848 /*
5849 * Freeze the pool. This stops spa_sync() from doing anything,
5850 * so that the only way to record changes from now on is the ZIL.
5851 */
5852 spa_freeze(spa);
5853
5854 /*
5855 * Because it is hard to predict how much space a write will actually
5856 * require beforehand, we leave ourselves some fudge space to write over
5857 * capacity.
5858 */
5859 uint64_t capacity = metaslab_class_get_space(spa_normal_class(spa)) / 2;
5860
5861 /*
5862 * Run tests that generate log records but don't alter the pool config
5863 * or depend on DSL sync tasks (snapshots, objset create/destroy, etc).
5864 * We do a txg_wait_synced() after each iteration to force the txg
5865 * to increase well beyond the last synced value in the uberblock.
5866 * The ZIL should be OK with that.
5867 *
5868 * Run a random number of times less than zo_maxloops and ensure we do
5869 * not run out of space on the pool.
5870 */
5871 while (ztest_random(10) != 0 &&
5872 numloops++ < ztest_opts.zo_maxloops &&
5873 metaslab_class_get_alloc(spa_normal_class(spa)) < capacity) {
5874 ztest_od_t od;
5875 ztest_od_init(&od, 0, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
5876 VERIFY0(ztest_object_init(zd, &od, sizeof (od), B_FALSE));
5877 ztest_io(zd, od.od_object,
5878 ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
5879 txg_wait_synced(spa_get_dsl(spa), 0);
5880 }
5881
5882 /*
5883 * Commit all of the changes we just generated.
5884 */
5885 zil_commit(zd->zd_zilog, 0);
5886 txg_wait_synced(spa_get_dsl(spa), 0);
5887
5888 /*
5889 * Close our dataset and close the pool.
5890 */
5891 ztest_dataset_close(0);
5892 spa_close(spa, FTAG);
5893 kernel_fini();
5894
5895 /*
5896 * Open and close the pool and dataset to induce log replay.
5897 */
5898 kernel_init(FREAD | FWRITE);
5899 VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
5900 ASSERT(spa_freeze_txg(spa) == UINT64_MAX);
5901 VERIFY3U(0, ==, ztest_dataset_open(0));
5902 ztest_dataset_close(0);
5903
5904 spa->spa_debug = B_TRUE;
5905 ztest_spa = spa;
5906 txg_wait_synced(spa_get_dsl(spa), 0);
5907 ztest_reguid(NULL, 0);
5908
5909 spa_close(spa, FTAG);
5910 kernel_fini();
5911 }
5912
5913 void
print_time(hrtime_t t,char * timebuf)5914 print_time(hrtime_t t, char *timebuf)
5915 {
5916 hrtime_t s = t / NANOSEC;
5917 hrtime_t m = s / 60;
5918 hrtime_t h = m / 60;
5919 hrtime_t d = h / 24;
5920
5921 s -= m * 60;
5922 m -= h * 60;
5923 h -= d * 24;
5924
5925 timebuf[0] = '\0';
5926
5927 if (d)
5928 (void) sprintf(timebuf,
5929 "%llud%02lluh%02llum%02llus", d, h, m, s);
5930 else if (h)
5931 (void) sprintf(timebuf, "%lluh%02llum%02llus", h, m, s);
5932 else if (m)
5933 (void) sprintf(timebuf, "%llum%02llus", m, s);
5934 else
5935 (void) sprintf(timebuf, "%llus", s);
5936 }
5937
5938 static nvlist_t *
make_random_props()5939 make_random_props()
5940 {
5941 nvlist_t *props;
5942
5943 VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
5944 if (ztest_random(2) == 0)
5945 return (props);
5946 VERIFY(nvlist_add_uint64(props, "autoreplace", 1) == 0);
5947
5948 return (props);
5949 }
5950
5951 /*
5952 * Create a storage pool with the given name and initial vdev size.
5953 * Then test spa_freeze() functionality.
5954 */
5955 static void
ztest_init(ztest_shared_t * zs)5956 ztest_init(ztest_shared_t *zs)
5957 {
5958 spa_t *spa;
5959 nvlist_t *nvroot, *props;
5960
5961 VERIFY(_mutex_init(&ztest_vdev_lock, USYNC_THREAD, NULL) == 0);
5962 VERIFY(rwlock_init(&ztest_name_lock, USYNC_THREAD, NULL) == 0);
5963
5964 kernel_init(FREAD | FWRITE);
5965
5966 /*
5967 * Create the storage pool.
5968 */
5969 (void) spa_destroy(ztest_opts.zo_pool);
5970 ztest_shared->zs_vdev_next_leaf = 0;
5971 zs->zs_splits = 0;
5972 zs->zs_mirrors = ztest_opts.zo_mirrors;
5973 nvroot = make_vdev_root(NULL, NULL, NULL, ztest_opts.zo_vdev_size, 0,
5974 0, ztest_opts.zo_raidz, zs->zs_mirrors, 1);
5975 props = make_random_props();
5976 for (int i = 0; i < SPA_FEATURES; i++) {
5977 char buf[1024];
5978 (void) snprintf(buf, sizeof (buf), "feature@%s",
5979 spa_feature_table[i].fi_uname);
5980 VERIFY3U(0, ==, nvlist_add_uint64(props, buf, 0));
5981 }
5982 VERIFY3U(0, ==, spa_create(ztest_opts.zo_pool, nvroot, props, NULL));
5983 nvlist_free(nvroot);
5984 nvlist_free(props);
5985
5986 VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
5987 zs->zs_metaslab_sz =
5988 1ULL << spa->spa_root_vdev->vdev_child[0]->vdev_ms_shift;
5989
5990 spa_close(spa, FTAG);
5991
5992 kernel_fini();
5993
5994 ztest_run_zdb(ztest_opts.zo_pool);
5995
5996 ztest_freeze();
5997
5998 ztest_run_zdb(ztest_opts.zo_pool);
5999
6000 (void) rwlock_destroy(&ztest_name_lock);
6001 (void) _mutex_destroy(&ztest_vdev_lock);
6002 }
6003
6004 static void
setup_data_fd(void)6005 setup_data_fd(void)
6006 {
6007 static char ztest_name_data[] = "/tmp/ztest.data.XXXXXX";
6008
6009 ztest_fd_data = mkstemp(ztest_name_data);
6010 ASSERT3S(ztest_fd_data, >=, 0);
6011 (void) unlink(ztest_name_data);
6012 }
6013
6014
6015 static int
shared_data_size(ztest_shared_hdr_t * hdr)6016 shared_data_size(ztest_shared_hdr_t *hdr)
6017 {
6018 int size;
6019
6020 size = hdr->zh_hdr_size;
6021 size += hdr->zh_opts_size;
6022 size += hdr->zh_size;
6023 size += hdr->zh_stats_size * hdr->zh_stats_count;
6024 size += hdr->zh_ds_size * hdr->zh_ds_count;
6025
6026 return (size);
6027 }
6028
6029 static void
setup_hdr(void)6030 setup_hdr(void)
6031 {
6032 int size;
6033 ztest_shared_hdr_t *hdr;
6034
6035 hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
6036 PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
6037 VERIFY(hdr != MAP_FAILED);
6038
6039 VERIFY3U(0, ==, ftruncate(ztest_fd_data, sizeof (ztest_shared_hdr_t)));
6040
6041 hdr->zh_hdr_size = sizeof (ztest_shared_hdr_t);
6042 hdr->zh_opts_size = sizeof (ztest_shared_opts_t);
6043 hdr->zh_size = sizeof (ztest_shared_t);
6044 hdr->zh_stats_size = sizeof (ztest_shared_callstate_t);
6045 hdr->zh_stats_count = ZTEST_FUNCS;
6046 hdr->zh_ds_size = sizeof (ztest_shared_ds_t);
6047 hdr->zh_ds_count = ztest_opts.zo_datasets;
6048
6049 size = shared_data_size(hdr);
6050 VERIFY3U(0, ==, ftruncate(ztest_fd_data, size));
6051
6052 (void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
6053 }
6054
6055 static void
setup_data(void)6056 setup_data(void)
6057 {
6058 int size, offset;
6059 ztest_shared_hdr_t *hdr;
6060 uint8_t *buf;
6061
6062 hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
6063 PROT_READ, MAP_SHARED, ztest_fd_data, 0);
6064 VERIFY(hdr != MAP_FAILED);
6065
6066 size = shared_data_size(hdr);
6067
6068 (void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
6069 hdr = ztest_shared_hdr = (void *)mmap(0, P2ROUNDUP(size, getpagesize()),
6070 PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
6071 VERIFY(hdr != MAP_FAILED);
6072 buf = (uint8_t *)hdr;
6073
6074 offset = hdr->zh_hdr_size;
6075 ztest_shared_opts = (void *)&buf[offset];
6076 offset += hdr->zh_opts_size;
6077 ztest_shared = (void *)&buf[offset];
6078 offset += hdr->zh_size;
6079 ztest_shared_callstate = (void *)&buf[offset];
6080 offset += hdr->zh_stats_size * hdr->zh_stats_count;
6081 ztest_shared_ds = (void *)&buf[offset];
6082 }
6083
6084 static boolean_t
exec_child(char * cmd,char * libpath,boolean_t ignorekill,int * statusp)6085 exec_child(char *cmd, char *libpath, boolean_t ignorekill, int *statusp)
6086 {
6087 pid_t pid;
6088 int status;
6089 char *cmdbuf = NULL;
6090
6091 pid = fork();
6092
6093 if (cmd == NULL) {
6094 cmdbuf = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
6095 (void) strlcpy(cmdbuf, getexecname(), MAXPATHLEN);
6096 cmd = cmdbuf;
6097 }
6098
6099 if (pid == -1)
6100 fatal(1, "fork failed");
6101
6102 if (pid == 0) { /* child */
6103 char *emptyargv[2] = { cmd, NULL };
6104 char fd_data_str[12];
6105
6106 struct rlimit rl = { 1024, 1024 };
6107 (void) setrlimit(RLIMIT_NOFILE, &rl);
6108
6109 (void) close(ztest_fd_rand);
6110 VERIFY3U(11, >=,
6111 snprintf(fd_data_str, 12, "%d", ztest_fd_data));
6112 VERIFY0(setenv("ZTEST_FD_DATA", fd_data_str, 1));
6113
6114 (void) enable_extended_FILE_stdio(-1, -1);
6115 if (libpath != NULL)
6116 VERIFY(0 == setenv("LD_LIBRARY_PATH", libpath, 1));
6117 #ifdef illumos
6118 (void) execv(cmd, emptyargv);
6119 #else
6120 (void) execvp(cmd, emptyargv);
6121 #endif
6122 ztest_dump_core = B_FALSE;
6123 fatal(B_TRUE, "exec failed: %s", cmd);
6124 }
6125
6126 if (cmdbuf != NULL) {
6127 umem_free(cmdbuf, MAXPATHLEN);
6128 cmd = NULL;
6129 }
6130
6131 while (waitpid(pid, &status, 0) != pid)
6132 continue;
6133 if (statusp != NULL)
6134 *statusp = status;
6135
6136 if (WIFEXITED(status)) {
6137 if (WEXITSTATUS(status) != 0) {
6138 (void) fprintf(stderr, "child exited with code %d\n",
6139 WEXITSTATUS(status));
6140 exit(2);
6141 }
6142 return (B_FALSE);
6143 } else if (WIFSIGNALED(status)) {
6144 if (!ignorekill || WTERMSIG(status) != SIGKILL) {
6145 (void) fprintf(stderr, "child died with signal %d\n",
6146 WTERMSIG(status));
6147 exit(3);
6148 }
6149 return (B_TRUE);
6150 } else {
6151 (void) fprintf(stderr, "something strange happened to child\n");
6152 exit(4);
6153 /* NOTREACHED */
6154 }
6155 }
6156
6157 static void
ztest_run_init(void)6158 ztest_run_init(void)
6159 {
6160 ztest_shared_t *zs = ztest_shared;
6161
6162 ASSERT(ztest_opts.zo_init != 0);
6163
6164 /*
6165 * Blow away any existing copy of zpool.cache
6166 */
6167 (void) remove(spa_config_path);
6168
6169 /*
6170 * Create and initialize our storage pool.
6171 */
6172 for (int i = 1; i <= ztest_opts.zo_init; i++) {
6173 bzero(zs, sizeof (ztest_shared_t));
6174 if (ztest_opts.zo_verbose >= 3 &&
6175 ztest_opts.zo_init != 1) {
6176 (void) printf("ztest_init(), pass %d\n", i);
6177 }
6178 ztest_init(zs);
6179 }
6180 }
6181
6182 int
main(int argc,char ** argv)6183 main(int argc, char **argv)
6184 {
6185 int kills = 0;
6186 int iters = 0;
6187 int older = 0;
6188 int newer = 0;
6189 ztest_shared_t *zs;
6190 ztest_info_t *zi;
6191 ztest_shared_callstate_t *zc;
6192 char timebuf[100];
6193 char numbuf[6];
6194 spa_t *spa;
6195 char *cmd;
6196 boolean_t hasalt;
6197 char *fd_data_str = getenv("ZTEST_FD_DATA");
6198
6199 (void) setvbuf(stdout, NULL, _IOLBF, 0);
6200
6201 dprintf_setup(&argc, argv);
6202 zfs_deadman_synctime_ms = 300000;
6203
6204 ztest_fd_rand = open("/dev/urandom", O_RDONLY);
6205 ASSERT3S(ztest_fd_rand, >=, 0);
6206
6207 if (!fd_data_str) {
6208 process_options(argc, argv);
6209
6210 setup_data_fd();
6211 setup_hdr();
6212 setup_data();
6213 bcopy(&ztest_opts, ztest_shared_opts,
6214 sizeof (*ztest_shared_opts));
6215 } else {
6216 ztest_fd_data = atoi(fd_data_str);
6217 setup_data();
6218 bcopy(ztest_shared_opts, &ztest_opts, sizeof (ztest_opts));
6219 }
6220 ASSERT3U(ztest_opts.zo_datasets, ==, ztest_shared_hdr->zh_ds_count);
6221
6222 /* Override location of zpool.cache */
6223 VERIFY3U(asprintf((char **)&spa_config_path, "%s/zpool.cache",
6224 ztest_opts.zo_dir), !=, -1);
6225
6226 ztest_ds = umem_alloc(ztest_opts.zo_datasets * sizeof (ztest_ds_t),
6227 UMEM_NOFAIL);
6228 zs = ztest_shared;
6229
6230 if (fd_data_str) {
6231 metaslab_gang_bang = ztest_opts.zo_metaslab_gang_bang;
6232 metaslab_df_alloc_threshold =
6233 zs->zs_metaslab_df_alloc_threshold;
6234
6235 if (zs->zs_do_init)
6236 ztest_run_init();
6237 else
6238 ztest_run(zs);
6239 exit(0);
6240 }
6241
6242 hasalt = (strlen(ztest_opts.zo_alt_ztest) != 0);
6243
6244 if (ztest_opts.zo_verbose >= 1) {
6245 (void) printf("%llu vdevs, %d datasets, %d threads,"
6246 " %llu seconds...\n",
6247 (u_longlong_t)ztest_opts.zo_vdevs,
6248 ztest_opts.zo_datasets,
6249 ztest_opts.zo_threads,
6250 (u_longlong_t)ztest_opts.zo_time);
6251 }
6252
6253 cmd = umem_alloc(MAXNAMELEN, UMEM_NOFAIL);
6254 (void) strlcpy(cmd, getexecname(), MAXNAMELEN);
6255
6256 zs->zs_do_init = B_TRUE;
6257 if (strlen(ztest_opts.zo_alt_ztest) != 0) {
6258 if (ztest_opts.zo_verbose >= 1) {
6259 (void) printf("Executing older ztest for "
6260 "initialization: %s\n", ztest_opts.zo_alt_ztest);
6261 }
6262 VERIFY(!exec_child(ztest_opts.zo_alt_ztest,
6263 ztest_opts.zo_alt_libpath, B_FALSE, NULL));
6264 } else {
6265 VERIFY(!exec_child(NULL, NULL, B_FALSE, NULL));
6266 }
6267 zs->zs_do_init = B_FALSE;
6268
6269 zs->zs_proc_start = gethrtime();
6270 zs->zs_proc_stop = zs->zs_proc_start + ztest_opts.zo_time * NANOSEC;
6271
6272 for (int f = 0; f < ZTEST_FUNCS; f++) {
6273 zi = &ztest_info[f];
6274 zc = ZTEST_GET_SHARED_CALLSTATE(f);
6275 if (zs->zs_proc_start + zi->zi_interval[0] > zs->zs_proc_stop)
6276 zc->zc_next = UINT64_MAX;
6277 else
6278 zc->zc_next = zs->zs_proc_start +
6279 ztest_random(2 * zi->zi_interval[0] + 1);
6280 }
6281
6282 /*
6283 * Run the tests in a loop. These tests include fault injection
6284 * to verify that self-healing data works, and forced crashes
6285 * to verify that we never lose on-disk consistency.
6286 */
6287 while (gethrtime() < zs->zs_proc_stop) {
6288 int status;
6289 boolean_t killed;
6290
6291 /*
6292 * Initialize the workload counters for each function.
6293 */
6294 for (int f = 0; f < ZTEST_FUNCS; f++) {
6295 zc = ZTEST_GET_SHARED_CALLSTATE(f);
6296 zc->zc_count = 0;
6297 zc->zc_time = 0;
6298 }
6299
6300 /* Set the allocation switch size */
6301 zs->zs_metaslab_df_alloc_threshold =
6302 ztest_random(zs->zs_metaslab_sz / 4) + 1;
6303
6304 if (!hasalt || ztest_random(2) == 0) {
6305 if (hasalt && ztest_opts.zo_verbose >= 1) {
6306 (void) printf("Executing newer ztest: %s\n",
6307 cmd);
6308 }
6309 newer++;
6310 killed = exec_child(cmd, NULL, B_TRUE, &status);
6311 } else {
6312 if (hasalt && ztest_opts.zo_verbose >= 1) {
6313 (void) printf("Executing older ztest: %s\n",
6314 ztest_opts.zo_alt_ztest);
6315 }
6316 older++;
6317 killed = exec_child(ztest_opts.zo_alt_ztest,
6318 ztest_opts.zo_alt_libpath, B_TRUE, &status);
6319 }
6320
6321 if (killed)
6322 kills++;
6323 iters++;
6324
6325 if (ztest_opts.zo_verbose >= 1) {
6326 hrtime_t now = gethrtime();
6327
6328 now = MIN(now, zs->zs_proc_stop);
6329 print_time(zs->zs_proc_stop - now, timebuf);
6330 nicenum(zs->zs_space, numbuf);
6331
6332 (void) printf("Pass %3d, %8s, %3llu ENOSPC, "
6333 "%4.1f%% of %5s used, %3.0f%% done, %8s to go\n",
6334 iters,
6335 WIFEXITED(status) ? "Complete" : "SIGKILL",
6336 (u_longlong_t)zs->zs_enospc_count,
6337 100.0 * zs->zs_alloc / zs->zs_space,
6338 numbuf,
6339 100.0 * (now - zs->zs_proc_start) /
6340 (ztest_opts.zo_time * NANOSEC), timebuf);
6341 }
6342
6343 if (ztest_opts.zo_verbose >= 2) {
6344 (void) printf("\nWorkload summary:\n\n");
6345 (void) printf("%7s %9s %s\n",
6346 "Calls", "Time", "Function");
6347 (void) printf("%7s %9s %s\n",
6348 "-----", "----", "--------");
6349 for (int f = 0; f < ZTEST_FUNCS; f++) {
6350 Dl_info dli;
6351
6352 zi = &ztest_info[f];
6353 zc = ZTEST_GET_SHARED_CALLSTATE(f);
6354 print_time(zc->zc_time, timebuf);
6355 (void) dladdr((void *)zi->zi_func, &dli);
6356 (void) printf("%7llu %9s %s\n",
6357 (u_longlong_t)zc->zc_count, timebuf,
6358 dli.dli_sname);
6359 }
6360 (void) printf("\n");
6361 }
6362
6363 /*
6364 * It's possible that we killed a child during a rename test,
6365 * in which case we'll have a 'ztest_tmp' pool lying around
6366 * instead of 'ztest'. Do a blind rename in case this happened.
6367 */
6368 kernel_init(FREAD);
6369 if (spa_open(ztest_opts.zo_pool, &spa, FTAG) == 0) {
6370 spa_close(spa, FTAG);
6371 } else {
6372 char tmpname[ZFS_MAX_DATASET_NAME_LEN];
6373 kernel_fini();
6374 kernel_init(FREAD | FWRITE);
6375 (void) snprintf(tmpname, sizeof (tmpname), "%s_tmp",
6376 ztest_opts.zo_pool);
6377 (void) spa_rename(tmpname, ztest_opts.zo_pool);
6378 }
6379 kernel_fini();
6380
6381 ztest_run_zdb(ztest_opts.zo_pool);
6382 }
6383
6384 if (ztest_opts.zo_verbose >= 1) {
6385 if (hasalt) {
6386 (void) printf("%d runs of older ztest: %s\n", older,
6387 ztest_opts.zo_alt_ztest);
6388 (void) printf("%d runs of newer ztest: %s\n", newer,
6389 cmd);
6390 }
6391 (void) printf("%d killed, %d completed, %.0f%% kill rate\n",
6392 kills, iters - kills, (100.0 * kills) / MAX(1, iters));
6393 }
6394
6395 umem_free(cmd, MAXNAMELEN);
6396
6397 return (0);
6398 }
6399