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