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