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 2008 Sun Microsystems, Inc. All rights reserved. 23 * Use is subject to license terms. 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/zap.h> 80 #include <sys/dmu_traverse.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/zio_checksum.h> 90 #include <sys/zio_compress.h> 91 #include <sys/zil.h> 92 #include <sys/vdev_impl.h> 93 #include <sys/vdev_file.h> 94 #include <sys/spa_impl.h> 95 #include <sys/dsl_prop.h> 96 #include <sys/refcount.h> 97 #include <stdio.h> 98 #include <stdio_ext.h> 99 #include <stdlib.h> 100 #include <unistd.h> 101 #include <signal.h> 102 #include <umem.h> 103 #include <dlfcn.h> 104 #include <ctype.h> 105 #include <math.h> 106 #include <sys/fs/zfs.h> 107 108 static char cmdname[] = "ztest"; 109 static char *zopt_pool = cmdname; 110 111 static uint64_t zopt_vdevs = 5; 112 static uint64_t zopt_vdevtime; 113 static int zopt_ashift = SPA_MINBLOCKSHIFT; 114 static int zopt_mirrors = 2; 115 static int zopt_raidz = 4; 116 static int zopt_raidz_parity = 1; 117 static size_t zopt_vdev_size = SPA_MINDEVSIZE; 118 static int zopt_datasets = 7; 119 static int zopt_threads = 23; 120 static uint64_t zopt_passtime = 60; /* 60 seconds */ 121 static uint64_t zopt_killrate = 70; /* 70% kill rate */ 122 static int zopt_verbose = 0; 123 static int zopt_init = 1; 124 static char *zopt_dir = "/tmp"; 125 static uint64_t zopt_time = 300; /* 5 minutes */ 126 static int zopt_maxfaults; 127 128 typedef struct ztest_block_tag { 129 uint64_t bt_objset; 130 uint64_t bt_object; 131 uint64_t bt_offset; 132 uint64_t bt_txg; 133 uint64_t bt_thread; 134 uint64_t bt_seq; 135 } ztest_block_tag_t; 136 137 typedef struct ztest_args { 138 char za_pool[MAXNAMELEN]; 139 spa_t *za_spa; 140 objset_t *za_os; 141 zilog_t *za_zilog; 142 thread_t za_thread; 143 uint64_t za_instance; 144 uint64_t za_random; 145 uint64_t za_diroff; 146 uint64_t za_diroff_shared; 147 uint64_t za_zil_seq; 148 hrtime_t za_start; 149 hrtime_t za_stop; 150 hrtime_t za_kill; 151 traverse_handle_t *za_th; 152 /* 153 * Thread-local variables can go here to aid debugging. 154 */ 155 ztest_block_tag_t za_rbt; 156 ztest_block_tag_t za_wbt; 157 dmu_object_info_t za_doi; 158 dmu_buf_t *za_dbuf; 159 } ztest_args_t; 160 161 typedef void ztest_func_t(ztest_args_t *); 162 163 /* 164 * Note: these aren't static because we want dladdr() to work. 165 */ 166 ztest_func_t ztest_dmu_read_write; 167 ztest_func_t ztest_dmu_write_parallel; 168 ztest_func_t ztest_dmu_object_alloc_free; 169 ztest_func_t ztest_zap; 170 ztest_func_t ztest_zap_parallel; 171 ztest_func_t ztest_traverse; 172 ztest_func_t ztest_dsl_prop_get_set; 173 ztest_func_t ztest_dmu_objset_create_destroy; 174 ztest_func_t ztest_dmu_snapshot_create_destroy; 175 ztest_func_t ztest_spa_create_destroy; 176 ztest_func_t ztest_fault_inject; 177 ztest_func_t ztest_spa_rename; 178 ztest_func_t ztest_vdev_attach_detach; 179 ztest_func_t ztest_vdev_LUN_growth; 180 ztest_func_t ztest_vdev_add_remove; 181 ztest_func_t ztest_vdev_aux_add_remove; 182 ztest_func_t ztest_scrub; 183 184 typedef struct ztest_info { 185 ztest_func_t *zi_func; /* test function */ 186 uint64_t zi_iters; /* iterations per execution */ 187 uint64_t *zi_interval; /* execute every <interval> seconds */ 188 uint64_t zi_calls; /* per-pass count */ 189 uint64_t zi_call_time; /* per-pass time */ 190 uint64_t zi_call_total; /* cumulative total */ 191 uint64_t zi_call_target; /* target cumulative total */ 192 } ztest_info_t; 193 194 uint64_t zopt_always = 0; /* all the time */ 195 uint64_t zopt_often = 1; /* every second */ 196 uint64_t zopt_sometimes = 10; /* every 10 seconds */ 197 uint64_t zopt_rarely = 60; /* every 60 seconds */ 198 199 ztest_info_t ztest_info[] = { 200 { ztest_dmu_read_write, 1, &zopt_always }, 201 { ztest_dmu_write_parallel, 30, &zopt_always }, 202 { ztest_dmu_object_alloc_free, 1, &zopt_always }, 203 { ztest_zap, 30, &zopt_always }, 204 { ztest_zap_parallel, 100, &zopt_always }, 205 { ztest_traverse, 1, &zopt_often }, 206 { ztest_dsl_prop_get_set, 1, &zopt_sometimes }, 207 { ztest_dmu_objset_create_destroy, 1, &zopt_sometimes }, 208 { ztest_dmu_snapshot_create_destroy, 1, &zopt_sometimes }, 209 { ztest_spa_create_destroy, 1, &zopt_sometimes }, 210 { ztest_fault_inject, 1, &zopt_sometimes }, 211 { ztest_spa_rename, 1, &zopt_rarely }, 212 { ztest_vdev_attach_detach, 1, &zopt_rarely }, 213 { ztest_vdev_LUN_growth, 1, &zopt_rarely }, 214 { ztest_vdev_add_remove, 1, &zopt_vdevtime }, 215 { ztest_vdev_aux_add_remove, 1, &zopt_vdevtime }, 216 { ztest_scrub, 1, &zopt_vdevtime }, 217 }; 218 219 #define ZTEST_FUNCS (sizeof (ztest_info) / sizeof (ztest_info_t)) 220 221 #define ZTEST_SYNC_LOCKS 16 222 223 /* 224 * Stuff we need to share writably between parent and child. 225 */ 226 typedef struct ztest_shared { 227 mutex_t zs_vdev_lock; 228 rwlock_t zs_name_lock; 229 uint64_t zs_vdev_primaries; 230 uint64_t zs_vdev_aux; 231 uint64_t zs_enospc_count; 232 hrtime_t zs_start_time; 233 hrtime_t zs_stop_time; 234 uint64_t zs_alloc; 235 uint64_t zs_space; 236 ztest_info_t zs_info[ZTEST_FUNCS]; 237 mutex_t zs_sync_lock[ZTEST_SYNC_LOCKS]; 238 uint64_t zs_seq[ZTEST_SYNC_LOCKS]; 239 } ztest_shared_t; 240 241 static char ztest_dev_template[] = "%s/%s.%llua"; 242 static char ztest_aux_template[] = "%s/%s.%s.%llu"; 243 static ztest_shared_t *ztest_shared; 244 245 static int ztest_random_fd; 246 static int ztest_dump_core = 1; 247 248 static boolean_t ztest_exiting; 249 250 extern uint64_t metaslab_gang_bang; 251 252 #define ZTEST_DIROBJ 1 253 #define ZTEST_MICROZAP_OBJ 2 254 #define ZTEST_FATZAP_OBJ 3 255 256 #define ZTEST_DIROBJ_BLOCKSIZE (1 << 10) 257 #define ZTEST_DIRSIZE 256 258 259 static void usage(boolean_t) __NORETURN; 260 261 /* 262 * These libumem hooks provide a reasonable set of defaults for the allocator's 263 * debugging facilities. 264 */ 265 const char * 266 _umem_debug_init() 267 { 268 return ("default,verbose"); /* $UMEM_DEBUG setting */ 269 } 270 271 const char * 272 _umem_logging_init(void) 273 { 274 return ("fail,contents"); /* $UMEM_LOGGING setting */ 275 } 276 277 #define FATAL_MSG_SZ 1024 278 279 char *fatal_msg; 280 281 static void 282 fatal(int do_perror, char *message, ...) 283 { 284 va_list args; 285 int save_errno = errno; 286 char buf[FATAL_MSG_SZ]; 287 288 (void) fflush(stdout); 289 290 va_start(args, message); 291 (void) sprintf(buf, "ztest: "); 292 /* LINTED */ 293 (void) vsprintf(buf + strlen(buf), message, args); 294 va_end(args); 295 if (do_perror) { 296 (void) snprintf(buf + strlen(buf), FATAL_MSG_SZ - strlen(buf), 297 ": %s", strerror(save_errno)); 298 } 299 (void) fprintf(stderr, "%s\n", buf); 300 fatal_msg = buf; /* to ease debugging */ 301 if (ztest_dump_core) 302 abort(); 303 exit(3); 304 } 305 306 static int 307 str2shift(const char *buf) 308 { 309 const char *ends = "BKMGTPEZ"; 310 int i; 311 312 if (buf[0] == '\0') 313 return (0); 314 for (i = 0; i < strlen(ends); i++) { 315 if (toupper(buf[0]) == ends[i]) 316 break; 317 } 318 if (i == strlen(ends)) { 319 (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n", 320 buf); 321 usage(B_FALSE); 322 } 323 if (buf[1] == '\0' || (toupper(buf[1]) == 'B' && buf[2] == '\0')) { 324 return (10*i); 325 } 326 (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n", buf); 327 usage(B_FALSE); 328 /* NOTREACHED */ 329 } 330 331 static uint64_t 332 nicenumtoull(const char *buf) 333 { 334 char *end; 335 uint64_t val; 336 337 val = strtoull(buf, &end, 0); 338 if (end == buf) { 339 (void) fprintf(stderr, "ztest: bad numeric value: %s\n", buf); 340 usage(B_FALSE); 341 } else if (end[0] == '.') { 342 double fval = strtod(buf, &end); 343 fval *= pow(2, str2shift(end)); 344 if (fval > UINT64_MAX) { 345 (void) fprintf(stderr, "ztest: value too large: %s\n", 346 buf); 347 usage(B_FALSE); 348 } 349 val = (uint64_t)fval; 350 } else { 351 int shift = str2shift(end); 352 if (shift >= 64 || (val << shift) >> shift != val) { 353 (void) fprintf(stderr, "ztest: value too large: %s\n", 354 buf); 355 usage(B_FALSE); 356 } 357 val <<= shift; 358 } 359 return (val); 360 } 361 362 static void 363 usage(boolean_t requested) 364 { 365 char nice_vdev_size[10]; 366 char nice_gang_bang[10]; 367 FILE *fp = requested ? stdout : stderr; 368 369 nicenum(zopt_vdev_size, nice_vdev_size); 370 nicenum(metaslab_gang_bang, nice_gang_bang); 371 372 (void) fprintf(fp, "Usage: %s\n" 373 "\t[-v vdevs (default: %llu)]\n" 374 "\t[-s size_of_each_vdev (default: %s)]\n" 375 "\t[-a alignment_shift (default: %d) (use 0 for random)]\n" 376 "\t[-m mirror_copies (default: %d)]\n" 377 "\t[-r raidz_disks (default: %d)]\n" 378 "\t[-R raidz_parity (default: %d)]\n" 379 "\t[-d datasets (default: %d)]\n" 380 "\t[-t threads (default: %d)]\n" 381 "\t[-g gang_block_threshold (default: %s)]\n" 382 "\t[-i initialize pool i times (default: %d)]\n" 383 "\t[-k kill percentage (default: %llu%%)]\n" 384 "\t[-p pool_name (default: %s)]\n" 385 "\t[-f file directory for vdev files (default: %s)]\n" 386 "\t[-V(erbose)] (use multiple times for ever more blather)\n" 387 "\t[-E(xisting)] (use existing pool instead of creating new one)\n" 388 "\t[-T time] total run time (default: %llu sec)\n" 389 "\t[-P passtime] time per pass (default: %llu sec)\n" 390 "\t[-h] (print help)\n" 391 "", 392 cmdname, 393 (u_longlong_t)zopt_vdevs, /* -v */ 394 nice_vdev_size, /* -s */ 395 zopt_ashift, /* -a */ 396 zopt_mirrors, /* -m */ 397 zopt_raidz, /* -r */ 398 zopt_raidz_parity, /* -R */ 399 zopt_datasets, /* -d */ 400 zopt_threads, /* -t */ 401 nice_gang_bang, /* -g */ 402 zopt_init, /* -i */ 403 (u_longlong_t)zopt_killrate, /* -k */ 404 zopt_pool, /* -p */ 405 zopt_dir, /* -f */ 406 (u_longlong_t)zopt_time, /* -T */ 407 (u_longlong_t)zopt_passtime); /* -P */ 408 exit(requested ? 0 : 1); 409 } 410 411 static uint64_t 412 ztest_random(uint64_t range) 413 { 414 uint64_t r; 415 416 if (range == 0) 417 return (0); 418 419 if (read(ztest_random_fd, &r, sizeof (r)) != sizeof (r)) 420 fatal(1, "short read from /dev/urandom"); 421 422 return (r % range); 423 } 424 425 static void 426 ztest_record_enospc(char *s) 427 { 428 dprintf("ENOSPC doing: %s\n", s ? s : "<unknown>"); 429 ztest_shared->zs_enospc_count++; 430 } 431 432 static void 433 process_options(int argc, char **argv) 434 { 435 int opt; 436 uint64_t value; 437 438 /* By default, test gang blocks for blocks 32K and greater */ 439 metaslab_gang_bang = 32 << 10; 440 441 while ((opt = getopt(argc, argv, 442 "v:s:a:m:r:R:d:t:g:i:k:p:f:VET:P:h")) != EOF) { 443 value = 0; 444 switch (opt) { 445 case 'v': 446 case 's': 447 case 'a': 448 case 'm': 449 case 'r': 450 case 'R': 451 case 'd': 452 case 't': 453 case 'g': 454 case 'i': 455 case 'k': 456 case 'T': 457 case 'P': 458 value = nicenumtoull(optarg); 459 } 460 switch (opt) { 461 case 'v': 462 zopt_vdevs = value; 463 break; 464 case 's': 465 zopt_vdev_size = MAX(SPA_MINDEVSIZE, value); 466 break; 467 case 'a': 468 zopt_ashift = value; 469 break; 470 case 'm': 471 zopt_mirrors = value; 472 break; 473 case 'r': 474 zopt_raidz = MAX(1, value); 475 break; 476 case 'R': 477 zopt_raidz_parity = MIN(MAX(value, 1), 2); 478 break; 479 case 'd': 480 zopt_datasets = MAX(1, value); 481 break; 482 case 't': 483 zopt_threads = MAX(1, value); 484 break; 485 case 'g': 486 metaslab_gang_bang = MAX(SPA_MINBLOCKSIZE << 1, value); 487 break; 488 case 'i': 489 zopt_init = value; 490 break; 491 case 'k': 492 zopt_killrate = value; 493 break; 494 case 'p': 495 zopt_pool = strdup(optarg); 496 break; 497 case 'f': 498 zopt_dir = strdup(optarg); 499 break; 500 case 'V': 501 zopt_verbose++; 502 break; 503 case 'E': 504 zopt_init = 0; 505 break; 506 case 'T': 507 zopt_time = value; 508 break; 509 case 'P': 510 zopt_passtime = MAX(1, value); 511 break; 512 case 'h': 513 usage(B_TRUE); 514 break; 515 case '?': 516 default: 517 usage(B_FALSE); 518 break; 519 } 520 } 521 522 zopt_raidz_parity = MIN(zopt_raidz_parity, zopt_raidz - 1); 523 524 zopt_vdevtime = (zopt_vdevs > 0 ? zopt_time / zopt_vdevs : UINT64_MAX); 525 zopt_maxfaults = MAX(zopt_mirrors, 1) * (zopt_raidz_parity + 1) - 1; 526 } 527 528 static uint64_t 529 ztest_get_ashift(void) 530 { 531 if (zopt_ashift == 0) 532 return (SPA_MINBLOCKSHIFT + ztest_random(3)); 533 return (zopt_ashift); 534 } 535 536 static nvlist_t * 537 make_vdev_file(char *path, char *aux, size_t size, uint64_t ashift) 538 { 539 char pathbuf[MAXPATHLEN]; 540 uint64_t vdev; 541 nvlist_t *file; 542 543 if (ashift == 0) 544 ashift = ztest_get_ashift(); 545 546 if (path == NULL) { 547 path = pathbuf; 548 549 if (aux != NULL) { 550 vdev = ztest_shared->zs_vdev_aux; 551 (void) sprintf(path, ztest_aux_template, 552 zopt_dir, zopt_pool, aux, vdev); 553 } else { 554 vdev = ztest_shared->zs_vdev_primaries++; 555 (void) sprintf(path, ztest_dev_template, 556 zopt_dir, zopt_pool, vdev); 557 } 558 } 559 560 if (size != 0) { 561 int fd = open(path, O_RDWR | O_CREAT | O_TRUNC, 0666); 562 if (fd == -1) 563 fatal(1, "can't open %s", path); 564 if (ftruncate(fd, size) != 0) 565 fatal(1, "can't ftruncate %s", path); 566 (void) close(fd); 567 } 568 569 VERIFY(nvlist_alloc(&file, NV_UNIQUE_NAME, 0) == 0); 570 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_TYPE, VDEV_TYPE_FILE) == 0); 571 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_PATH, path) == 0); 572 VERIFY(nvlist_add_uint64(file, ZPOOL_CONFIG_ASHIFT, ashift) == 0); 573 574 return (file); 575 } 576 577 static nvlist_t * 578 make_vdev_raidz(char *path, char *aux, size_t size, uint64_t ashift, int r) 579 { 580 nvlist_t *raidz, **child; 581 int c; 582 583 if (r < 2) 584 return (make_vdev_file(path, aux, size, ashift)); 585 child = umem_alloc(r * sizeof (nvlist_t *), UMEM_NOFAIL); 586 587 for (c = 0; c < r; c++) 588 child[c] = make_vdev_file(path, aux, size, ashift); 589 590 VERIFY(nvlist_alloc(&raidz, NV_UNIQUE_NAME, 0) == 0); 591 VERIFY(nvlist_add_string(raidz, ZPOOL_CONFIG_TYPE, 592 VDEV_TYPE_RAIDZ) == 0); 593 VERIFY(nvlist_add_uint64(raidz, ZPOOL_CONFIG_NPARITY, 594 zopt_raidz_parity) == 0); 595 VERIFY(nvlist_add_nvlist_array(raidz, ZPOOL_CONFIG_CHILDREN, 596 child, r) == 0); 597 598 for (c = 0; c < r; c++) 599 nvlist_free(child[c]); 600 601 umem_free(child, r * sizeof (nvlist_t *)); 602 603 return (raidz); 604 } 605 606 static nvlist_t * 607 make_vdev_mirror(char *path, char *aux, size_t size, uint64_t ashift, 608 int r, int m) 609 { 610 nvlist_t *mirror, **child; 611 int c; 612 613 if (m < 1) 614 return (make_vdev_raidz(path, aux, size, ashift, r)); 615 616 child = umem_alloc(m * sizeof (nvlist_t *), UMEM_NOFAIL); 617 618 for (c = 0; c < m; c++) 619 child[c] = make_vdev_raidz(path, aux, size, ashift, r); 620 621 VERIFY(nvlist_alloc(&mirror, NV_UNIQUE_NAME, 0) == 0); 622 VERIFY(nvlist_add_string(mirror, ZPOOL_CONFIG_TYPE, 623 VDEV_TYPE_MIRROR) == 0); 624 VERIFY(nvlist_add_nvlist_array(mirror, ZPOOL_CONFIG_CHILDREN, 625 child, m) == 0); 626 627 for (c = 0; c < m; c++) 628 nvlist_free(child[c]); 629 630 umem_free(child, m * sizeof (nvlist_t *)); 631 632 return (mirror); 633 } 634 635 static nvlist_t * 636 make_vdev_root(char *path, char *aux, size_t size, uint64_t ashift, 637 int log, int r, int m, int t) 638 { 639 nvlist_t *root, **child; 640 int c; 641 642 ASSERT(t > 0); 643 644 child = umem_alloc(t * sizeof (nvlist_t *), UMEM_NOFAIL); 645 646 for (c = 0; c < t; c++) { 647 child[c] = make_vdev_mirror(path, aux, size, ashift, r, m); 648 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_IS_LOG, 649 log) == 0); 650 } 651 652 VERIFY(nvlist_alloc(&root, NV_UNIQUE_NAME, 0) == 0); 653 VERIFY(nvlist_add_string(root, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) == 0); 654 VERIFY(nvlist_add_nvlist_array(root, aux ? aux : ZPOOL_CONFIG_CHILDREN, 655 child, t) == 0); 656 657 for (c = 0; c < t; c++) 658 nvlist_free(child[c]); 659 660 umem_free(child, t * sizeof (nvlist_t *)); 661 662 return (root); 663 } 664 665 static void 666 ztest_set_random_blocksize(objset_t *os, uint64_t object, dmu_tx_t *tx) 667 { 668 int bs = SPA_MINBLOCKSHIFT + 669 ztest_random(SPA_MAXBLOCKSHIFT - SPA_MINBLOCKSHIFT + 1); 670 int ibs = DN_MIN_INDBLKSHIFT + 671 ztest_random(DN_MAX_INDBLKSHIFT - DN_MIN_INDBLKSHIFT + 1); 672 int error; 673 674 error = dmu_object_set_blocksize(os, object, 1ULL << bs, ibs, tx); 675 if (error) { 676 char osname[300]; 677 dmu_objset_name(os, osname); 678 fatal(0, "dmu_object_set_blocksize('%s', %llu, %d, %d) = %d", 679 osname, object, 1 << bs, ibs, error); 680 } 681 } 682 683 static uint8_t 684 ztest_random_checksum(void) 685 { 686 uint8_t checksum; 687 688 do { 689 checksum = ztest_random(ZIO_CHECKSUM_FUNCTIONS); 690 } while (zio_checksum_table[checksum].ci_zbt); 691 692 if (checksum == ZIO_CHECKSUM_OFF) 693 checksum = ZIO_CHECKSUM_ON; 694 695 return (checksum); 696 } 697 698 static uint8_t 699 ztest_random_compress(void) 700 { 701 return ((uint8_t)ztest_random(ZIO_COMPRESS_FUNCTIONS)); 702 } 703 704 typedef struct ztest_replay { 705 objset_t *zr_os; 706 uint64_t zr_assign; 707 } ztest_replay_t; 708 709 static int 710 ztest_replay_create(ztest_replay_t *zr, lr_create_t *lr, boolean_t byteswap) 711 { 712 objset_t *os = zr->zr_os; 713 dmu_tx_t *tx; 714 int error; 715 716 if (byteswap) 717 byteswap_uint64_array(lr, sizeof (*lr)); 718 719 tx = dmu_tx_create(os); 720 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT); 721 error = dmu_tx_assign(tx, zr->zr_assign); 722 if (error) { 723 dmu_tx_abort(tx); 724 return (error); 725 } 726 727 error = dmu_object_claim(os, lr->lr_doid, lr->lr_mode, 0, 728 DMU_OT_NONE, 0, tx); 729 ASSERT3U(error, ==, 0); 730 dmu_tx_commit(tx); 731 732 if (zopt_verbose >= 5) { 733 char osname[MAXNAMELEN]; 734 dmu_objset_name(os, osname); 735 (void) printf("replay create of %s object %llu" 736 " in txg %llu = %d\n", 737 osname, (u_longlong_t)lr->lr_doid, 738 (u_longlong_t)zr->zr_assign, error); 739 } 740 741 return (error); 742 } 743 744 static int 745 ztest_replay_remove(ztest_replay_t *zr, lr_remove_t *lr, boolean_t byteswap) 746 { 747 objset_t *os = zr->zr_os; 748 dmu_tx_t *tx; 749 int error; 750 751 if (byteswap) 752 byteswap_uint64_array(lr, sizeof (*lr)); 753 754 tx = dmu_tx_create(os); 755 dmu_tx_hold_free(tx, lr->lr_doid, 0, DMU_OBJECT_END); 756 error = dmu_tx_assign(tx, zr->zr_assign); 757 if (error) { 758 dmu_tx_abort(tx); 759 return (error); 760 } 761 762 error = dmu_object_free(os, lr->lr_doid, tx); 763 dmu_tx_commit(tx); 764 765 return (error); 766 } 767 768 zil_replay_func_t *ztest_replay_vector[TX_MAX_TYPE] = { 769 NULL, /* 0 no such transaction type */ 770 ztest_replay_create, /* TX_CREATE */ 771 NULL, /* TX_MKDIR */ 772 NULL, /* TX_MKXATTR */ 773 NULL, /* TX_SYMLINK */ 774 ztest_replay_remove, /* TX_REMOVE */ 775 NULL, /* TX_RMDIR */ 776 NULL, /* TX_LINK */ 777 NULL, /* TX_RENAME */ 778 NULL, /* TX_WRITE */ 779 NULL, /* TX_TRUNCATE */ 780 NULL, /* TX_SETATTR */ 781 NULL, /* TX_ACL */ 782 }; 783 784 /* 785 * Verify that we can't destroy an active pool, create an existing pool, 786 * or create a pool with a bad vdev spec. 787 */ 788 void 789 ztest_spa_create_destroy(ztest_args_t *za) 790 { 791 int error; 792 spa_t *spa; 793 nvlist_t *nvroot; 794 795 /* 796 * Attempt to create using a bad file. 797 */ 798 nvroot = make_vdev_root("/dev/bogus", NULL, 0, 0, 0, 0, 0, 1); 799 error = spa_create("ztest_bad_file", nvroot, NULL, NULL, NULL); 800 nvlist_free(nvroot); 801 if (error != ENOENT) 802 fatal(0, "spa_create(bad_file) = %d", error); 803 804 /* 805 * Attempt to create using a bad mirror. 806 */ 807 nvroot = make_vdev_root("/dev/bogus", NULL, 0, 0, 0, 0, 2, 1); 808 error = spa_create("ztest_bad_mirror", nvroot, NULL, NULL, NULL); 809 nvlist_free(nvroot); 810 if (error != ENOENT) 811 fatal(0, "spa_create(bad_mirror) = %d", error); 812 813 /* 814 * Attempt to create an existing pool. It shouldn't matter 815 * what's in the nvroot; we should fail with EEXIST. 816 */ 817 (void) rw_rdlock(&ztest_shared->zs_name_lock); 818 nvroot = make_vdev_root("/dev/bogus", NULL, 0, 0, 0, 0, 0, 1); 819 error = spa_create(za->za_pool, nvroot, NULL, NULL, NULL); 820 nvlist_free(nvroot); 821 if (error != EEXIST) 822 fatal(0, "spa_create(whatever) = %d", error); 823 824 error = spa_open(za->za_pool, &spa, FTAG); 825 if (error) 826 fatal(0, "spa_open() = %d", error); 827 828 error = spa_destroy(za->za_pool); 829 if (error != EBUSY) 830 fatal(0, "spa_destroy() = %d", error); 831 832 spa_close(spa, FTAG); 833 (void) rw_unlock(&ztest_shared->zs_name_lock); 834 } 835 836 static vdev_t * 837 vdev_lookup_by_path(vdev_t *vd, const char *path) 838 { 839 vdev_t *mvd; 840 841 if (vd->vdev_path != NULL && strcmp(path, vd->vdev_path) == 0) 842 return (vd); 843 844 for (int c = 0; c < vd->vdev_children; c++) 845 if ((mvd = vdev_lookup_by_path(vd->vdev_child[c], path)) != 846 NULL) 847 return (mvd); 848 849 return (NULL); 850 } 851 852 /* 853 * Verify that vdev_add() works as expected. 854 */ 855 void 856 ztest_vdev_add_remove(ztest_args_t *za) 857 { 858 spa_t *spa = za->za_spa; 859 uint64_t leaves = MAX(zopt_mirrors, 1) * zopt_raidz; 860 nvlist_t *nvroot; 861 int error; 862 863 (void) mutex_lock(&ztest_shared->zs_vdev_lock); 864 865 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER); 866 867 ztest_shared->zs_vdev_primaries = 868 spa->spa_root_vdev->vdev_children * leaves; 869 870 spa_config_exit(spa, SCL_VDEV, FTAG); 871 872 /* 873 * Make 1/4 of the devices be log devices. 874 */ 875 nvroot = make_vdev_root(NULL, NULL, zopt_vdev_size, 0, 876 ztest_random(4) == 0, zopt_raidz, zopt_mirrors, 1); 877 878 error = spa_vdev_add(spa, nvroot); 879 nvlist_free(nvroot); 880 881 (void) mutex_unlock(&ztest_shared->zs_vdev_lock); 882 883 if (error == ENOSPC) 884 ztest_record_enospc("spa_vdev_add"); 885 else if (error != 0) 886 fatal(0, "spa_vdev_add() = %d", error); 887 } 888 889 /* 890 * Verify that adding/removing aux devices (l2arc, hot spare) works as expected. 891 */ 892 void 893 ztest_vdev_aux_add_remove(ztest_args_t *za) 894 { 895 spa_t *spa = za->za_spa; 896 vdev_t *rvd = spa->spa_root_vdev; 897 spa_aux_vdev_t *sav; 898 char *aux; 899 uint64_t guid = 0; 900 int error; 901 902 if (ztest_random(2) == 0) { 903 sav = &spa->spa_spares; 904 aux = ZPOOL_CONFIG_SPARES; 905 } else { 906 sav = &spa->spa_l2cache; 907 aux = ZPOOL_CONFIG_L2CACHE; 908 } 909 910 (void) mutex_lock(&ztest_shared->zs_vdev_lock); 911 912 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER); 913 914 if (sav->sav_count != 0 && ztest_random(4) == 0) { 915 /* 916 * Pick a random device to remove. 917 */ 918 guid = sav->sav_vdevs[ztest_random(sav->sav_count)]->vdev_guid; 919 } else { 920 /* 921 * Find an unused device we can add. 922 */ 923 ztest_shared->zs_vdev_aux = 0; 924 for (;;) { 925 char path[MAXPATHLEN]; 926 int c; 927 (void) sprintf(path, ztest_aux_template, zopt_dir, 928 zopt_pool, aux, ztest_shared->zs_vdev_aux); 929 for (c = 0; c < sav->sav_count; c++) 930 if (strcmp(sav->sav_vdevs[c]->vdev_path, 931 path) == 0) 932 break; 933 if (c == sav->sav_count && 934 vdev_lookup_by_path(rvd, path) == NULL) 935 break; 936 ztest_shared->zs_vdev_aux++; 937 } 938 } 939 940 spa_config_exit(spa, SCL_VDEV, FTAG); 941 942 if (guid == 0) { 943 /* 944 * Add a new device. 945 */ 946 nvlist_t *nvroot = make_vdev_root(NULL, aux, 947 (zopt_vdev_size * 5) / 4, 0, 0, 0, 0, 1); 948 error = spa_vdev_add(spa, nvroot); 949 if (error != 0) 950 fatal(0, "spa_vdev_add(%p) = %d", nvroot, error); 951 nvlist_free(nvroot); 952 } else { 953 /* 954 * Remove an existing device. Sometimes, dirty its 955 * vdev state first to make sure we handle removal 956 * of devices that have pending state changes. 957 */ 958 if (ztest_random(2) == 0) 959 (void) vdev_online(spa, guid, B_FALSE, NULL); 960 961 error = spa_vdev_remove(spa, guid, B_FALSE); 962 if (error != 0 && error != EBUSY) 963 fatal(0, "spa_vdev_remove(%llu) = %d", guid, error); 964 } 965 966 (void) mutex_unlock(&ztest_shared->zs_vdev_lock); 967 } 968 969 /* 970 * Verify that we can attach and detach devices. 971 */ 972 void 973 ztest_vdev_attach_detach(ztest_args_t *za) 974 { 975 spa_t *spa = za->za_spa; 976 spa_aux_vdev_t *sav = &spa->spa_spares; 977 vdev_t *rvd = spa->spa_root_vdev; 978 vdev_t *oldvd, *newvd, *pvd; 979 nvlist_t *root; 980 uint64_t leaves = MAX(zopt_mirrors, 1) * zopt_raidz; 981 uint64_t leaf, top; 982 uint64_t ashift = ztest_get_ashift(); 983 uint64_t oldguid; 984 size_t oldsize, newsize; 985 char oldpath[MAXPATHLEN], newpath[MAXPATHLEN]; 986 int replacing; 987 int newvd_is_spare = B_FALSE; 988 int oldvd_is_log; 989 int error, expected_error; 990 991 (void) mutex_lock(&ztest_shared->zs_vdev_lock); 992 993 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER); 994 995 /* 996 * Decide whether to do an attach or a replace. 997 */ 998 replacing = ztest_random(2); 999 1000 /* 1001 * Pick a random top-level vdev. 1002 */ 1003 top = ztest_random(rvd->vdev_children); 1004 1005 /* 1006 * Pick a random leaf within it. 1007 */ 1008 leaf = ztest_random(leaves); 1009 1010 /* 1011 * Locate this vdev. 1012 */ 1013 oldvd = rvd->vdev_child[top]; 1014 if (zopt_mirrors >= 1) 1015 oldvd = oldvd->vdev_child[leaf / zopt_raidz]; 1016 if (zopt_raidz > 1) 1017 oldvd = oldvd->vdev_child[leaf % zopt_raidz]; 1018 1019 /* 1020 * If we're already doing an attach or replace, oldvd may be a 1021 * mirror vdev -- in which case, pick a random child. 1022 */ 1023 while (oldvd->vdev_children != 0) { 1024 ASSERT(oldvd->vdev_children == 2); 1025 oldvd = oldvd->vdev_child[ztest_random(2)]; 1026 } 1027 1028 oldguid = oldvd->vdev_guid; 1029 oldsize = vdev_get_rsize(oldvd); 1030 oldvd_is_log = oldvd->vdev_top->vdev_islog; 1031 (void) strcpy(oldpath, oldvd->vdev_path); 1032 pvd = oldvd->vdev_parent; 1033 1034 /* 1035 * For the new vdev, choose with equal probability between the two 1036 * standard paths (ending in either 'a' or 'b') or a random hot spare. 1037 */ 1038 if (sav->sav_count != 0 && ztest_random(3) == 0) { 1039 newvd = sav->sav_vdevs[ztest_random(sav->sav_count)]; 1040 newvd_is_spare = B_TRUE; 1041 (void) strcpy(newpath, newvd->vdev_path); 1042 } else { 1043 (void) snprintf(newpath, sizeof (newpath), ztest_dev_template, 1044 zopt_dir, zopt_pool, top * leaves + leaf); 1045 if (ztest_random(2) == 0) 1046 newpath[strlen(newpath) - 1] = 'b'; 1047 newvd = vdev_lookup_by_path(rvd, newpath); 1048 } 1049 1050 if (newvd) { 1051 newsize = vdev_get_rsize(newvd); 1052 } else { 1053 /* 1054 * Make newsize a little bigger or smaller than oldsize. 1055 * If it's smaller, the attach should fail. 1056 * If it's larger, and we're doing a replace, 1057 * we should get dynamic LUN growth when we're done. 1058 */ 1059 newsize = 10 * oldsize / (9 + ztest_random(3)); 1060 } 1061 1062 /* 1063 * If pvd is not a mirror or root, the attach should fail with ENOTSUP, 1064 * unless it's a replace; in that case any non-replacing parent is OK. 1065 * 1066 * If newvd is already part of the pool, it should fail with EBUSY. 1067 * 1068 * If newvd is too small, it should fail with EOVERFLOW. 1069 */ 1070 if (pvd->vdev_ops != &vdev_mirror_ops && 1071 pvd->vdev_ops != &vdev_root_ops && (!replacing || 1072 pvd->vdev_ops == &vdev_replacing_ops || 1073 pvd->vdev_ops == &vdev_spare_ops)) 1074 expected_error = ENOTSUP; 1075 else if (newvd_is_spare && (!replacing || oldvd_is_log)) 1076 expected_error = ENOTSUP; 1077 else if (newvd == oldvd) 1078 expected_error = replacing ? 0 : EBUSY; 1079 else if (vdev_lookup_by_path(rvd, newpath) != NULL) 1080 expected_error = EBUSY; 1081 else if (newsize < oldsize) 1082 expected_error = EOVERFLOW; 1083 else if (ashift > oldvd->vdev_top->vdev_ashift) 1084 expected_error = EDOM; 1085 else 1086 expected_error = 0; 1087 1088 spa_config_exit(spa, SCL_VDEV, FTAG); 1089 1090 /* 1091 * Build the nvlist describing newpath. 1092 */ 1093 root = make_vdev_root(newpath, NULL, newvd == NULL ? newsize : 0, 1094 ashift, 0, 0, 0, 1); 1095 1096 error = spa_vdev_attach(spa, oldguid, root, replacing); 1097 1098 nvlist_free(root); 1099 1100 /* 1101 * If our parent was the replacing vdev, but the replace completed, 1102 * then instead of failing with ENOTSUP we may either succeed, 1103 * fail with ENODEV, or fail with EOVERFLOW. 1104 */ 1105 if (expected_error == ENOTSUP && 1106 (error == 0 || error == ENODEV || error == EOVERFLOW)) 1107 expected_error = error; 1108 1109 /* 1110 * If someone grew the LUN, the replacement may be too small. 1111 */ 1112 if (error == EOVERFLOW || error == EBUSY) 1113 expected_error = error; 1114 1115 /* XXX workaround 6690467 */ 1116 if (error != expected_error && expected_error != EBUSY) { 1117 fatal(0, "attach (%s %llu, %s %llu, %d) " 1118 "returned %d, expected %d", 1119 oldpath, (longlong_t)oldsize, newpath, 1120 (longlong_t)newsize, replacing, error, expected_error); 1121 } 1122 1123 (void) mutex_unlock(&ztest_shared->zs_vdev_lock); 1124 } 1125 1126 /* 1127 * Verify that dynamic LUN growth works as expected. 1128 */ 1129 /* ARGSUSED */ 1130 void 1131 ztest_vdev_LUN_growth(ztest_args_t *za) 1132 { 1133 spa_t *spa = za->za_spa; 1134 char dev_name[MAXPATHLEN]; 1135 uint64_t leaves = MAX(zopt_mirrors, 1) * zopt_raidz; 1136 uint64_t vdev; 1137 size_t fsize; 1138 int fd; 1139 1140 (void) mutex_lock(&ztest_shared->zs_vdev_lock); 1141 1142 /* 1143 * Pick a random leaf vdev. 1144 */ 1145 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER); 1146 vdev = ztest_random(spa->spa_root_vdev->vdev_children * leaves); 1147 spa_config_exit(spa, SCL_VDEV, FTAG); 1148 1149 (void) sprintf(dev_name, ztest_dev_template, zopt_dir, zopt_pool, vdev); 1150 1151 if ((fd = open(dev_name, O_RDWR)) != -1) { 1152 /* 1153 * Determine the size. 1154 */ 1155 fsize = lseek(fd, 0, SEEK_END); 1156 1157 /* 1158 * If it's less than 2x the original size, grow by around 3%. 1159 */ 1160 if (fsize < 2 * zopt_vdev_size) { 1161 size_t newsize = fsize + ztest_random(fsize / 32); 1162 (void) ftruncate(fd, newsize); 1163 if (zopt_verbose >= 6) { 1164 (void) printf("%s grew from %lu to %lu bytes\n", 1165 dev_name, (ulong_t)fsize, (ulong_t)newsize); 1166 } 1167 } 1168 (void) close(fd); 1169 } 1170 1171 (void) mutex_unlock(&ztest_shared->zs_vdev_lock); 1172 } 1173 1174 /* ARGSUSED */ 1175 static void 1176 ztest_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx) 1177 { 1178 /* 1179 * Create the directory object. 1180 */ 1181 VERIFY(dmu_object_claim(os, ZTEST_DIROBJ, 1182 DMU_OT_UINT64_OTHER, ZTEST_DIROBJ_BLOCKSIZE, 1183 DMU_OT_UINT64_OTHER, 5 * sizeof (ztest_block_tag_t), tx) == 0); 1184 1185 VERIFY(zap_create_claim(os, ZTEST_MICROZAP_OBJ, 1186 DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx) == 0); 1187 1188 VERIFY(zap_create_claim(os, ZTEST_FATZAP_OBJ, 1189 DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx) == 0); 1190 } 1191 1192 static int 1193 ztest_destroy_cb(char *name, void *arg) 1194 { 1195 ztest_args_t *za = arg; 1196 objset_t *os; 1197 dmu_object_info_t *doi = &za->za_doi; 1198 int error; 1199 1200 /* 1201 * Verify that the dataset contains a directory object. 1202 */ 1203 error = dmu_objset_open(name, DMU_OST_OTHER, 1204 DS_MODE_USER | DS_MODE_READONLY, &os); 1205 ASSERT3U(error, ==, 0); 1206 error = dmu_object_info(os, ZTEST_DIROBJ, doi); 1207 if (error != ENOENT) { 1208 /* We could have crashed in the middle of destroying it */ 1209 ASSERT3U(error, ==, 0); 1210 ASSERT3U(doi->doi_type, ==, DMU_OT_UINT64_OTHER); 1211 ASSERT3S(doi->doi_physical_blks, >=, 0); 1212 } 1213 dmu_objset_close(os); 1214 1215 /* 1216 * Destroy the dataset. 1217 */ 1218 error = dmu_objset_destroy(name); 1219 if (error) { 1220 (void) dmu_objset_open(name, DMU_OST_OTHER, 1221 DS_MODE_USER | DS_MODE_READONLY, &os); 1222 fatal(0, "dmu_objset_destroy(os=%p) = %d\n", &os, error); 1223 } 1224 return (0); 1225 } 1226 1227 /* 1228 * Verify that dmu_objset_{create,destroy,open,close} work as expected. 1229 */ 1230 static uint64_t 1231 ztest_log_create(zilog_t *zilog, dmu_tx_t *tx, uint64_t object, int mode) 1232 { 1233 itx_t *itx; 1234 lr_create_t *lr; 1235 size_t namesize; 1236 char name[24]; 1237 1238 (void) sprintf(name, "ZOBJ_%llu", (u_longlong_t)object); 1239 namesize = strlen(name) + 1; 1240 1241 itx = zil_itx_create(TX_CREATE, sizeof (*lr) + namesize + 1242 ztest_random(ZIL_MAX_BLKSZ)); 1243 lr = (lr_create_t *)&itx->itx_lr; 1244 bzero(lr + 1, lr->lr_common.lrc_reclen - sizeof (*lr)); 1245 lr->lr_doid = object; 1246 lr->lr_foid = 0; 1247 lr->lr_mode = mode; 1248 lr->lr_uid = 0; 1249 lr->lr_gid = 0; 1250 lr->lr_gen = dmu_tx_get_txg(tx); 1251 lr->lr_crtime[0] = time(NULL); 1252 lr->lr_crtime[1] = 0; 1253 lr->lr_rdev = 0; 1254 bcopy(name, (char *)(lr + 1), namesize); 1255 1256 return (zil_itx_assign(zilog, itx, tx)); 1257 } 1258 1259 void 1260 ztest_dmu_objset_create_destroy(ztest_args_t *za) 1261 { 1262 int error; 1263 objset_t *os, *os2; 1264 char name[100]; 1265 int basemode, expected_error; 1266 zilog_t *zilog; 1267 uint64_t seq; 1268 uint64_t objects; 1269 ztest_replay_t zr; 1270 1271 (void) rw_rdlock(&ztest_shared->zs_name_lock); 1272 (void) snprintf(name, 100, "%s/%s_temp_%llu", za->za_pool, za->za_pool, 1273 (u_longlong_t)za->za_instance); 1274 1275 basemode = DS_MODE_TYPE(za->za_instance); 1276 if (basemode != DS_MODE_USER && basemode != DS_MODE_OWNER) 1277 basemode = DS_MODE_USER; 1278 1279 /* 1280 * If this dataset exists from a previous run, process its replay log 1281 * half of the time. If we don't replay it, then dmu_objset_destroy() 1282 * (invoked from ztest_destroy_cb() below) should just throw it away. 1283 */ 1284 if (ztest_random(2) == 0 && 1285 dmu_objset_open(name, DMU_OST_OTHER, DS_MODE_OWNER, &os) == 0) { 1286 zr.zr_os = os; 1287 zil_replay(os, &zr, &zr.zr_assign, ztest_replay_vector, NULL); 1288 dmu_objset_close(os); 1289 } 1290 1291 /* 1292 * There may be an old instance of the dataset we're about to 1293 * create lying around from a previous run. If so, destroy it 1294 * and all of its snapshots. 1295 */ 1296 (void) dmu_objset_find(name, ztest_destroy_cb, za, 1297 DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS); 1298 1299 /* 1300 * Verify that the destroyed dataset is no longer in the namespace. 1301 */ 1302 error = dmu_objset_open(name, DMU_OST_OTHER, basemode, &os); 1303 if (error != ENOENT) 1304 fatal(1, "dmu_objset_open(%s) found destroyed dataset %p", 1305 name, os); 1306 1307 /* 1308 * Verify that we can create a new dataset. 1309 */ 1310 error = dmu_objset_create(name, DMU_OST_OTHER, NULL, 0, 1311 ztest_create_cb, NULL); 1312 if (error) { 1313 if (error == ENOSPC) { 1314 ztest_record_enospc("dmu_objset_create"); 1315 (void) rw_unlock(&ztest_shared->zs_name_lock); 1316 return; 1317 } 1318 fatal(0, "dmu_objset_create(%s) = %d", name, error); 1319 } 1320 1321 error = dmu_objset_open(name, DMU_OST_OTHER, basemode, &os); 1322 if (error) { 1323 fatal(0, "dmu_objset_open(%s) = %d", name, error); 1324 } 1325 1326 /* 1327 * Open the intent log for it. 1328 */ 1329 zilog = zil_open(os, NULL); 1330 1331 /* 1332 * Put a random number of objects in there. 1333 */ 1334 objects = ztest_random(20); 1335 seq = 0; 1336 while (objects-- != 0) { 1337 uint64_t object; 1338 dmu_tx_t *tx = dmu_tx_create(os); 1339 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, sizeof (name)); 1340 error = dmu_tx_assign(tx, TXG_WAIT); 1341 if (error) { 1342 dmu_tx_abort(tx); 1343 } else { 1344 object = dmu_object_alloc(os, DMU_OT_UINT64_OTHER, 0, 1345 DMU_OT_NONE, 0, tx); 1346 ztest_set_random_blocksize(os, object, tx); 1347 seq = ztest_log_create(zilog, tx, object, 1348 DMU_OT_UINT64_OTHER); 1349 dmu_write(os, object, 0, sizeof (name), name, tx); 1350 dmu_tx_commit(tx); 1351 } 1352 if (ztest_random(5) == 0) { 1353 zil_commit(zilog, seq, object); 1354 } 1355 if (ztest_random(100) == 0) { 1356 error = zil_suspend(zilog); 1357 if (error == 0) { 1358 zil_resume(zilog); 1359 } 1360 } 1361 } 1362 1363 /* 1364 * Verify that we cannot create an existing dataset. 1365 */ 1366 error = dmu_objset_create(name, DMU_OST_OTHER, NULL, 0, NULL, NULL); 1367 if (error != EEXIST) 1368 fatal(0, "created existing dataset, error = %d", error); 1369 1370 /* 1371 * Verify that multiple dataset holds are allowed, but only when 1372 * the new access mode is compatible with the base mode. 1373 */ 1374 if (basemode == DS_MODE_OWNER) { 1375 error = dmu_objset_open(name, DMU_OST_OTHER, DS_MODE_USER, 1376 &os2); 1377 if (error) 1378 fatal(0, "dmu_objset_open('%s') = %d", name, error); 1379 else 1380 dmu_objset_close(os2); 1381 } 1382 error = dmu_objset_open(name, DMU_OST_OTHER, DS_MODE_OWNER, &os2); 1383 expected_error = (basemode == DS_MODE_OWNER) ? EBUSY : 0; 1384 if (error != expected_error) 1385 fatal(0, "dmu_objset_open('%s') = %d, expected %d", 1386 name, error, expected_error); 1387 if (error == 0) 1388 dmu_objset_close(os2); 1389 1390 zil_close(zilog); 1391 dmu_objset_close(os); 1392 1393 error = dmu_objset_destroy(name); 1394 if (error) 1395 fatal(0, "dmu_objset_destroy(%s) = %d", name, error); 1396 1397 (void) rw_unlock(&ztest_shared->zs_name_lock); 1398 } 1399 1400 /* 1401 * Verify that dmu_snapshot_{create,destroy,open,close} work as expected. 1402 */ 1403 void 1404 ztest_dmu_snapshot_create_destroy(ztest_args_t *za) 1405 { 1406 int error; 1407 objset_t *os = za->za_os; 1408 char snapname[100]; 1409 char osname[MAXNAMELEN]; 1410 1411 (void) rw_rdlock(&ztest_shared->zs_name_lock); 1412 dmu_objset_name(os, osname); 1413 (void) snprintf(snapname, 100, "%s@%llu", osname, 1414 (u_longlong_t)za->za_instance); 1415 1416 error = dmu_objset_destroy(snapname); 1417 if (error != 0 && error != ENOENT) 1418 fatal(0, "dmu_objset_destroy() = %d", error); 1419 error = dmu_objset_snapshot(osname, strchr(snapname, '@')+1, FALSE); 1420 if (error == ENOSPC) 1421 ztest_record_enospc("dmu_take_snapshot"); 1422 else if (error != 0 && error != EEXIST) 1423 fatal(0, "dmu_take_snapshot() = %d", error); 1424 (void) rw_unlock(&ztest_shared->zs_name_lock); 1425 } 1426 1427 #define ZTEST_TRAVERSE_BLOCKS 1000 1428 1429 static int 1430 ztest_blk_cb(traverse_blk_cache_t *bc, spa_t *spa, void *arg) 1431 { 1432 ztest_args_t *za = arg; 1433 zbookmark_t *zb = &bc->bc_bookmark; 1434 blkptr_t *bp = &bc->bc_blkptr; 1435 dnode_phys_t *dnp = bc->bc_dnode; 1436 traverse_handle_t *th = za->za_th; 1437 uint64_t size = BP_GET_LSIZE(bp); 1438 1439 /* 1440 * Level -1 indicates the objset_phys_t or something in its intent log. 1441 */ 1442 if (zb->zb_level == -1) { 1443 if (BP_GET_TYPE(bp) == DMU_OT_OBJSET) { 1444 ASSERT3U(zb->zb_object, ==, 0); 1445 ASSERT3U(zb->zb_blkid, ==, 0); 1446 ASSERT3U(size, ==, sizeof (objset_phys_t)); 1447 za->za_zil_seq = 0; 1448 } else if (BP_GET_TYPE(bp) == DMU_OT_INTENT_LOG) { 1449 ASSERT3U(zb->zb_object, ==, 0); 1450 ASSERT3U(zb->zb_blkid, >, za->za_zil_seq); 1451 za->za_zil_seq = zb->zb_blkid; 1452 } else { 1453 ASSERT3U(zb->zb_object, !=, 0); /* lr_write_t */ 1454 } 1455 1456 return (0); 1457 } 1458 1459 ASSERT(dnp != NULL); 1460 1461 if (bc->bc_errno) 1462 return (ERESTART); 1463 1464 /* 1465 * Once in a while, abort the traverse. We only do this to odd 1466 * instance numbers to ensure that even ones can run to completion. 1467 */ 1468 if ((za->za_instance & 1) && ztest_random(10000) == 0) 1469 return (EINTR); 1470 1471 if (bp->blk_birth == 0) { 1472 ASSERT(th->th_advance & ADVANCE_HOLES); 1473 return (0); 1474 } 1475 1476 if (zb->zb_level == 0 && !(th->th_advance & ADVANCE_DATA) && 1477 bc == &th->th_cache[ZB_DN_CACHE][0]) { 1478 ASSERT(bc->bc_data == NULL); 1479 return (0); 1480 } 1481 1482 ASSERT(bc->bc_data != NULL); 1483 1484 /* 1485 * This is an expensive question, so don't ask it too often. 1486 */ 1487 if (((za->za_random ^ th->th_callbacks) & 0xff) == 0) { 1488 void *xbuf = umem_alloc(size, UMEM_NOFAIL); 1489 if (arc_tryread(spa, bp, xbuf) == 0) { 1490 ASSERT(bcmp(bc->bc_data, xbuf, size) == 0); 1491 } 1492 umem_free(xbuf, size); 1493 } 1494 1495 if (zb->zb_level > 0) { 1496 ASSERT3U(size, ==, 1ULL << dnp->dn_indblkshift); 1497 return (0); 1498 } 1499 1500 ASSERT(zb->zb_level == 0); 1501 ASSERT3U(size, ==, dnp->dn_datablkszsec << DEV_BSHIFT); 1502 1503 return (0); 1504 } 1505 1506 /* 1507 * Verify that live pool traversal works. 1508 */ 1509 void 1510 ztest_traverse(ztest_args_t *za) 1511 { 1512 spa_t *spa = za->za_spa; 1513 traverse_handle_t *th = za->za_th; 1514 int rc, advance; 1515 uint64_t cbstart, cblimit; 1516 1517 if (th == NULL) { 1518 advance = 0; 1519 1520 if (ztest_random(2) == 0) 1521 advance |= ADVANCE_PRE; 1522 1523 if (ztest_random(2) == 0) 1524 advance |= ADVANCE_PRUNE; 1525 1526 if (ztest_random(2) == 0) 1527 advance |= ADVANCE_DATA; 1528 1529 if (ztest_random(2) == 0) 1530 advance |= ADVANCE_HOLES; 1531 1532 if (ztest_random(2) == 0) 1533 advance |= ADVANCE_ZIL; 1534 1535 th = za->za_th = traverse_init(spa, ztest_blk_cb, za, advance, 1536 ZIO_FLAG_CANFAIL); 1537 1538 traverse_add_pool(th, 0, -1ULL); 1539 } 1540 1541 advance = th->th_advance; 1542 cbstart = th->th_callbacks; 1543 cblimit = cbstart + ((advance & ADVANCE_DATA) ? 100 : 1000); 1544 1545 while ((rc = traverse_more(th)) == EAGAIN && th->th_callbacks < cblimit) 1546 continue; 1547 1548 if (zopt_verbose >= 5) 1549 (void) printf("traverse %s%s%s%s %llu blocks to " 1550 "<%llu, %llu, %lld, %llx>%s\n", 1551 (advance & ADVANCE_PRE) ? "pre" : "post", 1552 (advance & ADVANCE_PRUNE) ? "|prune" : "", 1553 (advance & ADVANCE_DATA) ? "|data" : "", 1554 (advance & ADVANCE_HOLES) ? "|holes" : "", 1555 (u_longlong_t)(th->th_callbacks - cbstart), 1556 (u_longlong_t)th->th_lastcb.zb_objset, 1557 (u_longlong_t)th->th_lastcb.zb_object, 1558 (u_longlong_t)th->th_lastcb.zb_level, 1559 (u_longlong_t)th->th_lastcb.zb_blkid, 1560 rc == 0 ? " [done]" : 1561 rc == EINTR ? " [aborted]" : 1562 rc == EAGAIN ? "" : 1563 strerror(rc)); 1564 1565 if (rc != EAGAIN) { 1566 if (rc != 0 && rc != EINTR) 1567 fatal(0, "traverse_more(%p) = %d", th, rc); 1568 traverse_fini(th); 1569 za->za_th = NULL; 1570 } 1571 } 1572 1573 /* 1574 * Verify that dmu_object_{alloc,free} work as expected. 1575 */ 1576 void 1577 ztest_dmu_object_alloc_free(ztest_args_t *za) 1578 { 1579 objset_t *os = za->za_os; 1580 dmu_buf_t *db; 1581 dmu_tx_t *tx; 1582 uint64_t batchobj, object, batchsize, endoff, temp; 1583 int b, c, error, bonuslen; 1584 dmu_object_info_t *doi = &za->za_doi; 1585 char osname[MAXNAMELEN]; 1586 1587 dmu_objset_name(os, osname); 1588 1589 endoff = -8ULL; 1590 batchsize = 2; 1591 1592 /* 1593 * Create a batch object if necessary, and record it in the directory. 1594 */ 1595 VERIFY3U(0, ==, dmu_read(os, ZTEST_DIROBJ, za->za_diroff, 1596 sizeof (uint64_t), &batchobj)); 1597 if (batchobj == 0) { 1598 tx = dmu_tx_create(os); 1599 dmu_tx_hold_write(tx, ZTEST_DIROBJ, za->za_diroff, 1600 sizeof (uint64_t)); 1601 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT); 1602 error = dmu_tx_assign(tx, TXG_WAIT); 1603 if (error) { 1604 ztest_record_enospc("create a batch object"); 1605 dmu_tx_abort(tx); 1606 return; 1607 } 1608 batchobj = dmu_object_alloc(os, DMU_OT_UINT64_OTHER, 0, 1609 DMU_OT_NONE, 0, tx); 1610 ztest_set_random_blocksize(os, batchobj, tx); 1611 dmu_write(os, ZTEST_DIROBJ, za->za_diroff, 1612 sizeof (uint64_t), &batchobj, tx); 1613 dmu_tx_commit(tx); 1614 } 1615 1616 /* 1617 * Destroy the previous batch of objects. 1618 */ 1619 for (b = 0; b < batchsize; b++) { 1620 VERIFY3U(0, ==, dmu_read(os, batchobj, b * sizeof (uint64_t), 1621 sizeof (uint64_t), &object)); 1622 if (object == 0) 1623 continue; 1624 /* 1625 * Read and validate contents. 1626 * We expect the nth byte of the bonus buffer to be n. 1627 */ 1628 VERIFY(0 == dmu_bonus_hold(os, object, FTAG, &db)); 1629 za->za_dbuf = db; 1630 1631 dmu_object_info_from_db(db, doi); 1632 ASSERT(doi->doi_type == DMU_OT_UINT64_OTHER); 1633 ASSERT(doi->doi_bonus_type == DMU_OT_PLAIN_OTHER); 1634 ASSERT3S(doi->doi_physical_blks, >=, 0); 1635 1636 bonuslen = doi->doi_bonus_size; 1637 1638 for (c = 0; c < bonuslen; c++) { 1639 if (((uint8_t *)db->db_data)[c] != 1640 (uint8_t)(c + bonuslen)) { 1641 fatal(0, 1642 "bad bonus: %s, obj %llu, off %d: %u != %u", 1643 osname, object, c, 1644 ((uint8_t *)db->db_data)[c], 1645 (uint8_t)(c + bonuslen)); 1646 } 1647 } 1648 1649 dmu_buf_rele(db, FTAG); 1650 za->za_dbuf = NULL; 1651 1652 /* 1653 * We expect the word at endoff to be our object number. 1654 */ 1655 VERIFY(0 == dmu_read(os, object, endoff, 1656 sizeof (uint64_t), &temp)); 1657 1658 if (temp != object) { 1659 fatal(0, "bad data in %s, got %llu, expected %llu", 1660 osname, temp, object); 1661 } 1662 1663 /* 1664 * Destroy old object and clear batch entry. 1665 */ 1666 tx = dmu_tx_create(os); 1667 dmu_tx_hold_write(tx, batchobj, 1668 b * sizeof (uint64_t), sizeof (uint64_t)); 1669 dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END); 1670 error = dmu_tx_assign(tx, TXG_WAIT); 1671 if (error) { 1672 ztest_record_enospc("free object"); 1673 dmu_tx_abort(tx); 1674 return; 1675 } 1676 error = dmu_object_free(os, object, tx); 1677 if (error) { 1678 fatal(0, "dmu_object_free('%s', %llu) = %d", 1679 osname, object, error); 1680 } 1681 object = 0; 1682 1683 dmu_object_set_checksum(os, batchobj, 1684 ztest_random_checksum(), tx); 1685 dmu_object_set_compress(os, batchobj, 1686 ztest_random_compress(), tx); 1687 1688 dmu_write(os, batchobj, b * sizeof (uint64_t), 1689 sizeof (uint64_t), &object, tx); 1690 1691 dmu_tx_commit(tx); 1692 } 1693 1694 /* 1695 * Before creating the new batch of objects, generate a bunch of churn. 1696 */ 1697 for (b = ztest_random(100); b > 0; b--) { 1698 tx = dmu_tx_create(os); 1699 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT); 1700 error = dmu_tx_assign(tx, TXG_WAIT); 1701 if (error) { 1702 ztest_record_enospc("churn objects"); 1703 dmu_tx_abort(tx); 1704 return; 1705 } 1706 object = dmu_object_alloc(os, DMU_OT_UINT64_OTHER, 0, 1707 DMU_OT_NONE, 0, tx); 1708 ztest_set_random_blocksize(os, object, tx); 1709 error = dmu_object_free(os, object, tx); 1710 if (error) { 1711 fatal(0, "dmu_object_free('%s', %llu) = %d", 1712 osname, object, error); 1713 } 1714 dmu_tx_commit(tx); 1715 } 1716 1717 /* 1718 * Create a new batch of objects with randomly chosen 1719 * blocksizes and record them in the batch directory. 1720 */ 1721 for (b = 0; b < batchsize; b++) { 1722 uint32_t va_blksize; 1723 u_longlong_t va_nblocks; 1724 1725 tx = dmu_tx_create(os); 1726 dmu_tx_hold_write(tx, batchobj, b * sizeof (uint64_t), 1727 sizeof (uint64_t)); 1728 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT); 1729 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, endoff, 1730 sizeof (uint64_t)); 1731 error = dmu_tx_assign(tx, TXG_WAIT); 1732 if (error) { 1733 ztest_record_enospc("create batchobj"); 1734 dmu_tx_abort(tx); 1735 return; 1736 } 1737 bonuslen = (int)ztest_random(dmu_bonus_max()) + 1; 1738 1739 object = dmu_object_alloc(os, DMU_OT_UINT64_OTHER, 0, 1740 DMU_OT_PLAIN_OTHER, bonuslen, tx); 1741 1742 ztest_set_random_blocksize(os, object, tx); 1743 1744 dmu_object_set_checksum(os, object, 1745 ztest_random_checksum(), tx); 1746 dmu_object_set_compress(os, object, 1747 ztest_random_compress(), tx); 1748 1749 dmu_write(os, batchobj, b * sizeof (uint64_t), 1750 sizeof (uint64_t), &object, tx); 1751 1752 /* 1753 * Write to both the bonus buffer and the regular data. 1754 */ 1755 VERIFY(dmu_bonus_hold(os, object, FTAG, &db) == 0); 1756 za->za_dbuf = db; 1757 ASSERT3U(bonuslen, <=, db->db_size); 1758 1759 dmu_object_size_from_db(db, &va_blksize, &va_nblocks); 1760 ASSERT3S(va_nblocks, >=, 0); 1761 1762 dmu_buf_will_dirty(db, tx); 1763 1764 /* 1765 * See comments above regarding the contents of 1766 * the bonus buffer and the word at endoff. 1767 */ 1768 for (c = 0; c < bonuslen; c++) 1769 ((uint8_t *)db->db_data)[c] = (uint8_t)(c + bonuslen); 1770 1771 dmu_buf_rele(db, FTAG); 1772 za->za_dbuf = NULL; 1773 1774 /* 1775 * Write to a large offset to increase indirection. 1776 */ 1777 dmu_write(os, object, endoff, sizeof (uint64_t), &object, tx); 1778 1779 dmu_tx_commit(tx); 1780 } 1781 } 1782 1783 /* 1784 * Verify that dmu_{read,write} work as expected. 1785 */ 1786 typedef struct bufwad { 1787 uint64_t bw_index; 1788 uint64_t bw_txg; 1789 uint64_t bw_data; 1790 } bufwad_t; 1791 1792 typedef struct dmu_read_write_dir { 1793 uint64_t dd_packobj; 1794 uint64_t dd_bigobj; 1795 uint64_t dd_chunk; 1796 } dmu_read_write_dir_t; 1797 1798 void 1799 ztest_dmu_read_write(ztest_args_t *za) 1800 { 1801 objset_t *os = za->za_os; 1802 dmu_read_write_dir_t dd; 1803 dmu_tx_t *tx; 1804 int i, freeit, error; 1805 uint64_t n, s, txg; 1806 bufwad_t *packbuf, *bigbuf, *pack, *bigH, *bigT; 1807 uint64_t packoff, packsize, bigoff, bigsize; 1808 uint64_t regions = 997; 1809 uint64_t stride = 123456789ULL; 1810 uint64_t width = 40; 1811 int free_percent = 5; 1812 1813 /* 1814 * This test uses two objects, packobj and bigobj, that are always 1815 * updated together (i.e. in the same tx) so that their contents are 1816 * in sync and can be compared. Their contents relate to each other 1817 * in a simple way: packobj is a dense array of 'bufwad' structures, 1818 * while bigobj is a sparse array of the same bufwads. Specifically, 1819 * for any index n, there are three bufwads that should be identical: 1820 * 1821 * packobj, at offset n * sizeof (bufwad_t) 1822 * bigobj, at the head of the nth chunk 1823 * bigobj, at the tail of the nth chunk 1824 * 1825 * The chunk size is arbitrary. It doesn't have to be a power of two, 1826 * and it doesn't have any relation to the object blocksize. 1827 * The only requirement is that it can hold at least two bufwads. 1828 * 1829 * Normally, we write the bufwad to each of these locations. 1830 * However, free_percent of the time we instead write zeroes to 1831 * packobj and perform a dmu_free_range() on bigobj. By comparing 1832 * bigobj to packobj, we can verify that the DMU is correctly 1833 * tracking which parts of an object are allocated and free, 1834 * and that the contents of the allocated blocks are correct. 1835 */ 1836 1837 /* 1838 * Read the directory info. If it's the first time, set things up. 1839 */ 1840 VERIFY(0 == dmu_read(os, ZTEST_DIROBJ, za->za_diroff, 1841 sizeof (dd), &dd)); 1842 if (dd.dd_chunk == 0) { 1843 ASSERT(dd.dd_packobj == 0); 1844 ASSERT(dd.dd_bigobj == 0); 1845 tx = dmu_tx_create(os); 1846 dmu_tx_hold_write(tx, ZTEST_DIROBJ, za->za_diroff, sizeof (dd)); 1847 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT); 1848 error = dmu_tx_assign(tx, TXG_WAIT); 1849 if (error) { 1850 ztest_record_enospc("create r/w directory"); 1851 dmu_tx_abort(tx); 1852 return; 1853 } 1854 1855 dd.dd_packobj = dmu_object_alloc(os, DMU_OT_UINT64_OTHER, 0, 1856 DMU_OT_NONE, 0, tx); 1857 dd.dd_bigobj = dmu_object_alloc(os, DMU_OT_UINT64_OTHER, 0, 1858 DMU_OT_NONE, 0, tx); 1859 dd.dd_chunk = (1000 + ztest_random(1000)) * sizeof (uint64_t); 1860 1861 ztest_set_random_blocksize(os, dd.dd_packobj, tx); 1862 ztest_set_random_blocksize(os, dd.dd_bigobj, tx); 1863 1864 dmu_write(os, ZTEST_DIROBJ, za->za_diroff, sizeof (dd), &dd, 1865 tx); 1866 dmu_tx_commit(tx); 1867 } 1868 1869 /* 1870 * Prefetch a random chunk of the big object. 1871 * Our aim here is to get some async reads in flight 1872 * for blocks that we may free below; the DMU should 1873 * handle this race correctly. 1874 */ 1875 n = ztest_random(regions) * stride + ztest_random(width); 1876 s = 1 + ztest_random(2 * width - 1); 1877 dmu_prefetch(os, dd.dd_bigobj, n * dd.dd_chunk, s * dd.dd_chunk); 1878 1879 /* 1880 * Pick a random index and compute the offsets into packobj and bigobj. 1881 */ 1882 n = ztest_random(regions) * stride + ztest_random(width); 1883 s = 1 + ztest_random(width - 1); 1884 1885 packoff = n * sizeof (bufwad_t); 1886 packsize = s * sizeof (bufwad_t); 1887 1888 bigoff = n * dd.dd_chunk; 1889 bigsize = s * dd.dd_chunk; 1890 1891 packbuf = umem_alloc(packsize, UMEM_NOFAIL); 1892 bigbuf = umem_alloc(bigsize, UMEM_NOFAIL); 1893 1894 /* 1895 * free_percent of the time, free a range of bigobj rather than 1896 * overwriting it. 1897 */ 1898 freeit = (ztest_random(100) < free_percent); 1899 1900 /* 1901 * Read the current contents of our objects. 1902 */ 1903 error = dmu_read(os, dd.dd_packobj, packoff, packsize, packbuf); 1904 ASSERT3U(error, ==, 0); 1905 error = dmu_read(os, dd.dd_bigobj, bigoff, bigsize, bigbuf); 1906 ASSERT3U(error, ==, 0); 1907 1908 /* 1909 * Get a tx for the mods to both packobj and bigobj. 1910 */ 1911 tx = dmu_tx_create(os); 1912 1913 dmu_tx_hold_write(tx, dd.dd_packobj, packoff, packsize); 1914 1915 if (freeit) 1916 dmu_tx_hold_free(tx, dd.dd_bigobj, bigoff, bigsize); 1917 else 1918 dmu_tx_hold_write(tx, dd.dd_bigobj, bigoff, bigsize); 1919 1920 error = dmu_tx_assign(tx, TXG_WAIT); 1921 1922 if (error) { 1923 ztest_record_enospc("dmu r/w range"); 1924 dmu_tx_abort(tx); 1925 umem_free(packbuf, packsize); 1926 umem_free(bigbuf, bigsize); 1927 return; 1928 } 1929 1930 txg = dmu_tx_get_txg(tx); 1931 1932 /* 1933 * For each index from n to n + s, verify that the existing bufwad 1934 * in packobj matches the bufwads at the head and tail of the 1935 * corresponding chunk in bigobj. Then update all three bufwads 1936 * with the new values we want to write out. 1937 */ 1938 for (i = 0; i < s; i++) { 1939 /* LINTED */ 1940 pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t)); 1941 /* LINTED */ 1942 bigH = (bufwad_t *)((char *)bigbuf + i * dd.dd_chunk); 1943 /* LINTED */ 1944 bigT = (bufwad_t *)((char *)bigH + dd.dd_chunk) - 1; 1945 1946 ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize); 1947 ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize); 1948 1949 if (pack->bw_txg > txg) 1950 fatal(0, "future leak: got %llx, open txg is %llx", 1951 pack->bw_txg, txg); 1952 1953 if (pack->bw_data != 0 && pack->bw_index != n + i) 1954 fatal(0, "wrong index: got %llx, wanted %llx+%llx", 1955 pack->bw_index, n, i); 1956 1957 if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0) 1958 fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH); 1959 1960 if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0) 1961 fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT); 1962 1963 if (freeit) { 1964 bzero(pack, sizeof (bufwad_t)); 1965 } else { 1966 pack->bw_index = n + i; 1967 pack->bw_txg = txg; 1968 pack->bw_data = 1 + ztest_random(-2ULL); 1969 } 1970 *bigH = *pack; 1971 *bigT = *pack; 1972 } 1973 1974 /* 1975 * We've verified all the old bufwads, and made new ones. 1976 * Now write them out. 1977 */ 1978 dmu_write(os, dd.dd_packobj, packoff, packsize, packbuf, tx); 1979 1980 if (freeit) { 1981 if (zopt_verbose >= 6) { 1982 (void) printf("freeing offset %llx size %llx" 1983 " txg %llx\n", 1984 (u_longlong_t)bigoff, 1985 (u_longlong_t)bigsize, 1986 (u_longlong_t)txg); 1987 } 1988 VERIFY(0 == dmu_free_range(os, dd.dd_bigobj, bigoff, 1989 bigsize, tx)); 1990 } else { 1991 if (zopt_verbose >= 6) { 1992 (void) printf("writing offset %llx size %llx" 1993 " txg %llx\n", 1994 (u_longlong_t)bigoff, 1995 (u_longlong_t)bigsize, 1996 (u_longlong_t)txg); 1997 } 1998 dmu_write(os, dd.dd_bigobj, bigoff, bigsize, bigbuf, tx); 1999 } 2000 2001 dmu_tx_commit(tx); 2002 2003 /* 2004 * Sanity check the stuff we just wrote. 2005 */ 2006 { 2007 void *packcheck = umem_alloc(packsize, UMEM_NOFAIL); 2008 void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL); 2009 2010 VERIFY(0 == dmu_read(os, dd.dd_packobj, packoff, 2011 packsize, packcheck)); 2012 VERIFY(0 == dmu_read(os, dd.dd_bigobj, bigoff, 2013 bigsize, bigcheck)); 2014 2015 ASSERT(bcmp(packbuf, packcheck, packsize) == 0); 2016 ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0); 2017 2018 umem_free(packcheck, packsize); 2019 umem_free(bigcheck, bigsize); 2020 } 2021 2022 umem_free(packbuf, packsize); 2023 umem_free(bigbuf, bigsize); 2024 } 2025 2026 void 2027 ztest_dmu_check_future_leak(ztest_args_t *za) 2028 { 2029 objset_t *os = za->za_os; 2030 dmu_buf_t *db; 2031 ztest_block_tag_t *bt; 2032 dmu_object_info_t *doi = &za->za_doi; 2033 2034 /* 2035 * Make sure that, if there is a write record in the bonus buffer 2036 * of the ZTEST_DIROBJ, that the txg for this record is <= the 2037 * last synced txg of the pool. 2038 */ 2039 VERIFY(dmu_bonus_hold(os, ZTEST_DIROBJ, FTAG, &db) == 0); 2040 za->za_dbuf = db; 2041 VERIFY(dmu_object_info(os, ZTEST_DIROBJ, doi) == 0); 2042 ASSERT3U(doi->doi_bonus_size, >=, sizeof (*bt)); 2043 ASSERT3U(doi->doi_bonus_size, <=, db->db_size); 2044 ASSERT3U(doi->doi_bonus_size % sizeof (*bt), ==, 0); 2045 bt = (void *)((char *)db->db_data + doi->doi_bonus_size - sizeof (*bt)); 2046 if (bt->bt_objset != 0) { 2047 ASSERT3U(bt->bt_objset, ==, dmu_objset_id(os)); 2048 ASSERT3U(bt->bt_object, ==, ZTEST_DIROBJ); 2049 ASSERT3U(bt->bt_offset, ==, -1ULL); 2050 ASSERT3U(bt->bt_txg, <, spa_first_txg(za->za_spa)); 2051 } 2052 dmu_buf_rele(db, FTAG); 2053 za->za_dbuf = NULL; 2054 } 2055 2056 void 2057 ztest_dmu_write_parallel(ztest_args_t *za) 2058 { 2059 objset_t *os = za->za_os; 2060 ztest_block_tag_t *rbt = &za->za_rbt; 2061 ztest_block_tag_t *wbt = &za->za_wbt; 2062 const size_t btsize = sizeof (ztest_block_tag_t); 2063 dmu_buf_t *db; 2064 int b, error; 2065 int bs = ZTEST_DIROBJ_BLOCKSIZE; 2066 int do_free = 0; 2067 uint64_t off, txg, txg_how; 2068 mutex_t *lp; 2069 char osname[MAXNAMELEN]; 2070 char iobuf[SPA_MAXBLOCKSIZE]; 2071 blkptr_t blk = { 0 }; 2072 uint64_t blkoff; 2073 zbookmark_t zb; 2074 dmu_tx_t *tx = dmu_tx_create(os); 2075 2076 dmu_objset_name(os, osname); 2077 2078 /* 2079 * Have multiple threads write to large offsets in ZTEST_DIROBJ 2080 * to verify that having multiple threads writing to the same object 2081 * in parallel doesn't cause any trouble. 2082 */ 2083 if (ztest_random(4) == 0) { 2084 /* 2085 * Do the bonus buffer instead of a regular block. 2086 * We need a lock to serialize resize vs. others, 2087 * so we hash on the objset ID. 2088 */ 2089 b = dmu_objset_id(os) % ZTEST_SYNC_LOCKS; 2090 off = -1ULL; 2091 dmu_tx_hold_bonus(tx, ZTEST_DIROBJ); 2092 } else { 2093 b = ztest_random(ZTEST_SYNC_LOCKS); 2094 off = za->za_diroff_shared + (b << SPA_MAXBLOCKSHIFT); 2095 if (ztest_random(4) == 0) { 2096 do_free = 1; 2097 dmu_tx_hold_free(tx, ZTEST_DIROBJ, off, bs); 2098 } else { 2099 dmu_tx_hold_write(tx, ZTEST_DIROBJ, off, bs); 2100 } 2101 } 2102 2103 txg_how = ztest_random(2) == 0 ? TXG_WAIT : TXG_NOWAIT; 2104 error = dmu_tx_assign(tx, txg_how); 2105 if (error) { 2106 if (error == ERESTART) { 2107 ASSERT(txg_how == TXG_NOWAIT); 2108 dmu_tx_wait(tx); 2109 } else { 2110 ztest_record_enospc("dmu write parallel"); 2111 } 2112 dmu_tx_abort(tx); 2113 return; 2114 } 2115 txg = dmu_tx_get_txg(tx); 2116 2117 lp = &ztest_shared->zs_sync_lock[b]; 2118 (void) mutex_lock(lp); 2119 2120 wbt->bt_objset = dmu_objset_id(os); 2121 wbt->bt_object = ZTEST_DIROBJ; 2122 wbt->bt_offset = off; 2123 wbt->bt_txg = txg; 2124 wbt->bt_thread = za->za_instance; 2125 wbt->bt_seq = ztest_shared->zs_seq[b]++; /* protected by lp */ 2126 2127 /* 2128 * Occasionally, write an all-zero block to test the behavior 2129 * of blocks that compress into holes. 2130 */ 2131 if (off != -1ULL && ztest_random(8) == 0) 2132 bzero(wbt, btsize); 2133 2134 if (off == -1ULL) { 2135 dmu_object_info_t *doi = &za->za_doi; 2136 char *dboff; 2137 2138 VERIFY(dmu_bonus_hold(os, ZTEST_DIROBJ, FTAG, &db) == 0); 2139 za->za_dbuf = db; 2140 dmu_object_info_from_db(db, doi); 2141 ASSERT3U(doi->doi_bonus_size, <=, db->db_size); 2142 ASSERT3U(doi->doi_bonus_size, >=, btsize); 2143 ASSERT3U(doi->doi_bonus_size % btsize, ==, 0); 2144 dboff = (char *)db->db_data + doi->doi_bonus_size - btsize; 2145 bcopy(dboff, rbt, btsize); 2146 if (rbt->bt_objset != 0) { 2147 ASSERT3U(rbt->bt_objset, ==, wbt->bt_objset); 2148 ASSERT3U(rbt->bt_object, ==, wbt->bt_object); 2149 ASSERT3U(rbt->bt_offset, ==, wbt->bt_offset); 2150 ASSERT3U(rbt->bt_txg, <=, wbt->bt_txg); 2151 } 2152 if (ztest_random(10) == 0) { 2153 int newsize = (ztest_random(db->db_size / 2154 btsize) + 1) * btsize; 2155 2156 ASSERT3U(newsize, >=, btsize); 2157 ASSERT3U(newsize, <=, db->db_size); 2158 VERIFY3U(dmu_set_bonus(db, newsize, tx), ==, 0); 2159 dboff = (char *)db->db_data + newsize - btsize; 2160 } 2161 dmu_buf_will_dirty(db, tx); 2162 bcopy(wbt, dboff, btsize); 2163 dmu_buf_rele(db, FTAG); 2164 za->za_dbuf = NULL; 2165 } else if (do_free) { 2166 VERIFY(dmu_free_range(os, ZTEST_DIROBJ, off, bs, tx) == 0); 2167 } else { 2168 dmu_write(os, ZTEST_DIROBJ, off, btsize, wbt, tx); 2169 } 2170 2171 (void) mutex_unlock(lp); 2172 2173 if (ztest_random(1000) == 0) 2174 (void) poll(NULL, 0, 1); /* open dn_notxholds window */ 2175 2176 dmu_tx_commit(tx); 2177 2178 if (ztest_random(10000) == 0) 2179 txg_wait_synced(dmu_objset_pool(os), txg); 2180 2181 if (off == -1ULL || do_free) 2182 return; 2183 2184 if (ztest_random(2) != 0) 2185 return; 2186 2187 /* 2188 * dmu_sync() the block we just wrote. 2189 */ 2190 (void) mutex_lock(lp); 2191 2192 blkoff = P2ALIGN_TYPED(off, bs, uint64_t); 2193 error = dmu_buf_hold(os, ZTEST_DIROBJ, blkoff, FTAG, &db); 2194 za->za_dbuf = db; 2195 if (error) { 2196 dprintf("dmu_buf_hold(%s, %d, %llx) = %d\n", 2197 osname, ZTEST_DIROBJ, blkoff, error); 2198 (void) mutex_unlock(lp); 2199 return; 2200 } 2201 blkoff = off - blkoff; 2202 error = dmu_sync(NULL, db, &blk, txg, NULL, NULL); 2203 dmu_buf_rele(db, FTAG); 2204 za->za_dbuf = NULL; 2205 2206 (void) mutex_unlock(lp); 2207 2208 if (error) { 2209 dprintf("dmu_sync(%s, %d, %llx) = %d\n", 2210 osname, ZTEST_DIROBJ, off, error); 2211 return; 2212 } 2213 2214 if (blk.blk_birth == 0) /* concurrent free */ 2215 return; 2216 2217 txg_suspend(dmu_objset_pool(os)); 2218 2219 ASSERT(blk.blk_fill == 1); 2220 ASSERT3U(BP_GET_TYPE(&blk), ==, DMU_OT_UINT64_OTHER); 2221 ASSERT3U(BP_GET_LEVEL(&blk), ==, 0); 2222 ASSERT3U(BP_GET_LSIZE(&blk), ==, bs); 2223 2224 /* 2225 * Read the block that dmu_sync() returned to make sure its contents 2226 * match what we wrote. We do this while still txg_suspend()ed 2227 * to ensure that the block can't be reused before we read it. 2228 */ 2229 zb.zb_objset = dmu_objset_id(os); 2230 zb.zb_object = ZTEST_DIROBJ; 2231 zb.zb_level = 0; 2232 zb.zb_blkid = off / bs; 2233 error = zio_wait(zio_read(NULL, za->za_spa, &blk, iobuf, bs, 2234 NULL, NULL, ZIO_PRIORITY_SYNC_READ, ZIO_FLAG_MUSTSUCCEED, &zb)); 2235 ASSERT3U(error, ==, 0); 2236 2237 txg_resume(dmu_objset_pool(os)); 2238 2239 bcopy(&iobuf[blkoff], rbt, btsize); 2240 2241 if (rbt->bt_objset == 0) /* concurrent free */ 2242 return; 2243 2244 if (wbt->bt_objset == 0) /* all-zero overwrite */ 2245 return; 2246 2247 ASSERT3U(rbt->bt_objset, ==, wbt->bt_objset); 2248 ASSERT3U(rbt->bt_object, ==, wbt->bt_object); 2249 ASSERT3U(rbt->bt_offset, ==, wbt->bt_offset); 2250 2251 /* 2252 * The semantic of dmu_sync() is that we always push the most recent 2253 * version of the data, so in the face of concurrent updates we may 2254 * see a newer version of the block. That's OK. 2255 */ 2256 ASSERT3U(rbt->bt_txg, >=, wbt->bt_txg); 2257 if (rbt->bt_thread == wbt->bt_thread) 2258 ASSERT3U(rbt->bt_seq, ==, wbt->bt_seq); 2259 else 2260 ASSERT3U(rbt->bt_seq, >, wbt->bt_seq); 2261 } 2262 2263 /* 2264 * Verify that zap_{create,destroy,add,remove,update} work as expected. 2265 */ 2266 #define ZTEST_ZAP_MIN_INTS 1 2267 #define ZTEST_ZAP_MAX_INTS 4 2268 #define ZTEST_ZAP_MAX_PROPS 1000 2269 2270 void 2271 ztest_zap(ztest_args_t *za) 2272 { 2273 objset_t *os = za->za_os; 2274 uint64_t object; 2275 uint64_t txg, last_txg; 2276 uint64_t value[ZTEST_ZAP_MAX_INTS]; 2277 uint64_t zl_ints, zl_intsize, prop; 2278 int i, ints; 2279 dmu_tx_t *tx; 2280 char propname[100], txgname[100]; 2281 int error; 2282 char osname[MAXNAMELEN]; 2283 char *hc[2] = { "s.acl.h", ".s.open.h.hyLZlg" }; 2284 2285 dmu_objset_name(os, osname); 2286 2287 /* 2288 * Create a new object if necessary, and record it in the directory. 2289 */ 2290 VERIFY(0 == dmu_read(os, ZTEST_DIROBJ, za->za_diroff, 2291 sizeof (uint64_t), &object)); 2292 2293 if (object == 0) { 2294 tx = dmu_tx_create(os); 2295 dmu_tx_hold_write(tx, ZTEST_DIROBJ, za->za_diroff, 2296 sizeof (uint64_t)); 2297 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, TRUE, NULL); 2298 error = dmu_tx_assign(tx, TXG_WAIT); 2299 if (error) { 2300 ztest_record_enospc("create zap test obj"); 2301 dmu_tx_abort(tx); 2302 return; 2303 } 2304 object = zap_create(os, DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx); 2305 if (error) { 2306 fatal(0, "zap_create('%s', %llu) = %d", 2307 osname, object, error); 2308 } 2309 ASSERT(object != 0); 2310 dmu_write(os, ZTEST_DIROBJ, za->za_diroff, 2311 sizeof (uint64_t), &object, tx); 2312 /* 2313 * Generate a known hash collision, and verify that 2314 * we can lookup and remove both entries. 2315 */ 2316 for (i = 0; i < 2; i++) { 2317 value[i] = i; 2318 error = zap_add(os, object, hc[i], sizeof (uint64_t), 2319 1, &value[i], tx); 2320 ASSERT3U(error, ==, 0); 2321 } 2322 for (i = 0; i < 2; i++) { 2323 error = zap_add(os, object, hc[i], sizeof (uint64_t), 2324 1, &value[i], tx); 2325 ASSERT3U(error, ==, EEXIST); 2326 error = zap_length(os, object, hc[i], 2327 &zl_intsize, &zl_ints); 2328 ASSERT3U(error, ==, 0); 2329 ASSERT3U(zl_intsize, ==, sizeof (uint64_t)); 2330 ASSERT3U(zl_ints, ==, 1); 2331 } 2332 for (i = 0; i < 2; i++) { 2333 error = zap_remove(os, object, hc[i], tx); 2334 ASSERT3U(error, ==, 0); 2335 } 2336 2337 dmu_tx_commit(tx); 2338 } 2339 2340 ints = MAX(ZTEST_ZAP_MIN_INTS, object % ZTEST_ZAP_MAX_INTS); 2341 2342 prop = ztest_random(ZTEST_ZAP_MAX_PROPS); 2343 (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop); 2344 (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop); 2345 bzero(value, sizeof (value)); 2346 last_txg = 0; 2347 2348 /* 2349 * If these zap entries already exist, validate their contents. 2350 */ 2351 error = zap_length(os, object, txgname, &zl_intsize, &zl_ints); 2352 if (error == 0) { 2353 ASSERT3U(zl_intsize, ==, sizeof (uint64_t)); 2354 ASSERT3U(zl_ints, ==, 1); 2355 2356 VERIFY(zap_lookup(os, object, txgname, zl_intsize, 2357 zl_ints, &last_txg) == 0); 2358 2359 VERIFY(zap_length(os, object, propname, &zl_intsize, 2360 &zl_ints) == 0); 2361 2362 ASSERT3U(zl_intsize, ==, sizeof (uint64_t)); 2363 ASSERT3U(zl_ints, ==, ints); 2364 2365 VERIFY(zap_lookup(os, object, propname, zl_intsize, 2366 zl_ints, value) == 0); 2367 2368 for (i = 0; i < ints; i++) { 2369 ASSERT3U(value[i], ==, last_txg + object + i); 2370 } 2371 } else { 2372 ASSERT3U(error, ==, ENOENT); 2373 } 2374 2375 /* 2376 * Atomically update two entries in our zap object. 2377 * The first is named txg_%llu, and contains the txg 2378 * in which the property was last updated. The second 2379 * is named prop_%llu, and the nth element of its value 2380 * should be txg + object + n. 2381 */ 2382 tx = dmu_tx_create(os); 2383 dmu_tx_hold_zap(tx, object, TRUE, NULL); 2384 error = dmu_tx_assign(tx, TXG_WAIT); 2385 if (error) { 2386 ztest_record_enospc("create zap entry"); 2387 dmu_tx_abort(tx); 2388 return; 2389 } 2390 txg = dmu_tx_get_txg(tx); 2391 2392 if (last_txg > txg) 2393 fatal(0, "zap future leak: old %llu new %llu", last_txg, txg); 2394 2395 for (i = 0; i < ints; i++) 2396 value[i] = txg + object + i; 2397 2398 error = zap_update(os, object, txgname, sizeof (uint64_t), 1, &txg, tx); 2399 if (error) 2400 fatal(0, "zap_update('%s', %llu, '%s') = %d", 2401 osname, object, txgname, error); 2402 2403 error = zap_update(os, object, propname, sizeof (uint64_t), 2404 ints, value, tx); 2405 if (error) 2406 fatal(0, "zap_update('%s', %llu, '%s') = %d", 2407 osname, object, propname, error); 2408 2409 dmu_tx_commit(tx); 2410 2411 /* 2412 * Remove a random pair of entries. 2413 */ 2414 prop = ztest_random(ZTEST_ZAP_MAX_PROPS); 2415 (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop); 2416 (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop); 2417 2418 error = zap_length(os, object, txgname, &zl_intsize, &zl_ints); 2419 2420 if (error == ENOENT) 2421 return; 2422 2423 ASSERT3U(error, ==, 0); 2424 2425 tx = dmu_tx_create(os); 2426 dmu_tx_hold_zap(tx, object, TRUE, NULL); 2427 error = dmu_tx_assign(tx, TXG_WAIT); 2428 if (error) { 2429 ztest_record_enospc("remove zap entry"); 2430 dmu_tx_abort(tx); 2431 return; 2432 } 2433 error = zap_remove(os, object, txgname, tx); 2434 if (error) 2435 fatal(0, "zap_remove('%s', %llu, '%s') = %d", 2436 osname, object, txgname, error); 2437 2438 error = zap_remove(os, object, propname, tx); 2439 if (error) 2440 fatal(0, "zap_remove('%s', %llu, '%s') = %d", 2441 osname, object, propname, error); 2442 2443 dmu_tx_commit(tx); 2444 2445 /* 2446 * Once in a while, destroy the object. 2447 */ 2448 if (ztest_random(1000) != 0) 2449 return; 2450 2451 tx = dmu_tx_create(os); 2452 dmu_tx_hold_write(tx, ZTEST_DIROBJ, za->za_diroff, sizeof (uint64_t)); 2453 dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END); 2454 error = dmu_tx_assign(tx, TXG_WAIT); 2455 if (error) { 2456 ztest_record_enospc("destroy zap object"); 2457 dmu_tx_abort(tx); 2458 return; 2459 } 2460 error = zap_destroy(os, object, tx); 2461 if (error) 2462 fatal(0, "zap_destroy('%s', %llu) = %d", 2463 osname, object, error); 2464 object = 0; 2465 dmu_write(os, ZTEST_DIROBJ, za->za_diroff, sizeof (uint64_t), 2466 &object, tx); 2467 dmu_tx_commit(tx); 2468 } 2469 2470 void 2471 ztest_zap_parallel(ztest_args_t *za) 2472 { 2473 objset_t *os = za->za_os; 2474 uint64_t txg, object, count, wsize, wc, zl_wsize, zl_wc; 2475 dmu_tx_t *tx; 2476 int i, namelen, error; 2477 char name[20], string_value[20]; 2478 void *data; 2479 2480 /* 2481 * Generate a random name of the form 'xxx.....' where each 2482 * x is a random printable character and the dots are dots. 2483 * There are 94 such characters, and the name length goes from 2484 * 6 to 20, so there are 94^3 * 15 = 12,458,760 possible names. 2485 */ 2486 namelen = ztest_random(sizeof (name) - 5) + 5 + 1; 2487 2488 for (i = 0; i < 3; i++) 2489 name[i] = '!' + ztest_random('~' - '!' + 1); 2490 for (; i < namelen - 1; i++) 2491 name[i] = '.'; 2492 name[i] = '\0'; 2493 2494 if (ztest_random(2) == 0) 2495 object = ZTEST_MICROZAP_OBJ; 2496 else 2497 object = ZTEST_FATZAP_OBJ; 2498 2499 if ((namelen & 1) || object == ZTEST_MICROZAP_OBJ) { 2500 wsize = sizeof (txg); 2501 wc = 1; 2502 data = &txg; 2503 } else { 2504 wsize = 1; 2505 wc = namelen; 2506 data = string_value; 2507 } 2508 2509 count = -1ULL; 2510 VERIFY(zap_count(os, object, &count) == 0); 2511 ASSERT(count != -1ULL); 2512 2513 /* 2514 * Select an operation: length, lookup, add, update, remove. 2515 */ 2516 i = ztest_random(5); 2517 2518 if (i >= 2) { 2519 tx = dmu_tx_create(os); 2520 dmu_tx_hold_zap(tx, object, TRUE, NULL); 2521 error = dmu_tx_assign(tx, TXG_WAIT); 2522 if (error) { 2523 ztest_record_enospc("zap parallel"); 2524 dmu_tx_abort(tx); 2525 return; 2526 } 2527 txg = dmu_tx_get_txg(tx); 2528 bcopy(name, string_value, namelen); 2529 } else { 2530 tx = NULL; 2531 txg = 0; 2532 bzero(string_value, namelen); 2533 } 2534 2535 switch (i) { 2536 2537 case 0: 2538 error = zap_length(os, object, name, &zl_wsize, &zl_wc); 2539 if (error == 0) { 2540 ASSERT3U(wsize, ==, zl_wsize); 2541 ASSERT3U(wc, ==, zl_wc); 2542 } else { 2543 ASSERT3U(error, ==, ENOENT); 2544 } 2545 break; 2546 2547 case 1: 2548 error = zap_lookup(os, object, name, wsize, wc, data); 2549 if (error == 0) { 2550 if (data == string_value && 2551 bcmp(name, data, namelen) != 0) 2552 fatal(0, "name '%s' != val '%s' len %d", 2553 name, data, namelen); 2554 } else { 2555 ASSERT3U(error, ==, ENOENT); 2556 } 2557 break; 2558 2559 case 2: 2560 error = zap_add(os, object, name, wsize, wc, data, tx); 2561 ASSERT(error == 0 || error == EEXIST); 2562 break; 2563 2564 case 3: 2565 VERIFY(zap_update(os, object, name, wsize, wc, data, tx) == 0); 2566 break; 2567 2568 case 4: 2569 error = zap_remove(os, object, name, tx); 2570 ASSERT(error == 0 || error == ENOENT); 2571 break; 2572 } 2573 2574 if (tx != NULL) 2575 dmu_tx_commit(tx); 2576 } 2577 2578 void 2579 ztest_dsl_prop_get_set(ztest_args_t *za) 2580 { 2581 objset_t *os = za->za_os; 2582 int i, inherit; 2583 uint64_t value; 2584 const char *prop, *valname; 2585 char setpoint[MAXPATHLEN]; 2586 char osname[MAXNAMELEN]; 2587 int error; 2588 2589 (void) rw_rdlock(&ztest_shared->zs_name_lock); 2590 2591 dmu_objset_name(os, osname); 2592 2593 for (i = 0; i < 2; i++) { 2594 if (i == 0) { 2595 prop = "checksum"; 2596 value = ztest_random_checksum(); 2597 inherit = (value == ZIO_CHECKSUM_INHERIT); 2598 } else { 2599 prop = "compression"; 2600 value = ztest_random_compress(); 2601 inherit = (value == ZIO_COMPRESS_INHERIT); 2602 } 2603 2604 error = dsl_prop_set(osname, prop, sizeof (value), 2605 !inherit, &value); 2606 2607 if (error == ENOSPC) { 2608 ztest_record_enospc("dsl_prop_set"); 2609 break; 2610 } 2611 2612 ASSERT3U(error, ==, 0); 2613 2614 VERIFY3U(dsl_prop_get(osname, prop, sizeof (value), 2615 1, &value, setpoint), ==, 0); 2616 2617 if (i == 0) 2618 valname = zio_checksum_table[value].ci_name; 2619 else 2620 valname = zio_compress_table[value].ci_name; 2621 2622 if (zopt_verbose >= 6) { 2623 (void) printf("%s %s = %s for '%s'\n", 2624 osname, prop, valname, setpoint); 2625 } 2626 } 2627 2628 (void) rw_unlock(&ztest_shared->zs_name_lock); 2629 } 2630 2631 /* 2632 * Inject random faults into the on-disk data. 2633 */ 2634 void 2635 ztest_fault_inject(ztest_args_t *za) 2636 { 2637 int fd; 2638 uint64_t offset; 2639 uint64_t leaves = MAX(zopt_mirrors, 1) * zopt_raidz; 2640 uint64_t bad = 0x1990c0ffeedecade; 2641 uint64_t top, leaf; 2642 char path0[MAXPATHLEN]; 2643 char pathrand[MAXPATHLEN]; 2644 size_t fsize; 2645 spa_t *spa = za->za_spa; 2646 int bshift = SPA_MAXBLOCKSHIFT + 2; /* don't scrog all labels */ 2647 int iters = 1000; 2648 int maxfaults = zopt_maxfaults; 2649 vdev_t *vd0 = NULL; 2650 uint64_t guid0 = 0; 2651 2652 ASSERT(leaves >= 1); 2653 2654 /* 2655 * We need SCL_STATE here because we're going to look at vd0->vdev_tsd. 2656 */ 2657 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 2658 2659 if (ztest_random(2) == 0) { 2660 /* 2661 * Inject errors on a normal data device. 2662 */ 2663 top = ztest_random(spa->spa_root_vdev->vdev_children); 2664 leaf = ztest_random(leaves); 2665 2666 /* 2667 * Generate paths to the first leaf in this top-level vdev, 2668 * and to the random leaf we selected. We'll induce transient 2669 * write failures and random online/offline activity on leaf 0, 2670 * and we'll write random garbage to the randomly chosen leaf. 2671 */ 2672 (void) snprintf(path0, sizeof (path0), ztest_dev_template, 2673 zopt_dir, zopt_pool, top * leaves + 0); 2674 (void) snprintf(pathrand, sizeof (pathrand), ztest_dev_template, 2675 zopt_dir, zopt_pool, top * leaves + leaf); 2676 2677 vd0 = vdev_lookup_by_path(spa->spa_root_vdev, path0); 2678 if (vd0 != NULL && maxfaults != 1) { 2679 /* 2680 * Make vd0 explicitly claim to be unreadable, 2681 * or unwriteable, or reach behind its back 2682 * and close the underlying fd. We can do this if 2683 * maxfaults == 0 because we'll fail and reexecute, 2684 * and we can do it if maxfaults >= 2 because we'll 2685 * have enough redundancy. If maxfaults == 1, the 2686 * combination of this with injection of random data 2687 * corruption below exceeds the pool's fault tolerance. 2688 */ 2689 vdev_file_t *vf = vd0->vdev_tsd; 2690 2691 if (vf != NULL && ztest_random(3) == 0) { 2692 (void) close(vf->vf_vnode->v_fd); 2693 vf->vf_vnode->v_fd = -1; 2694 } else if (ztest_random(2) == 0) { 2695 vd0->vdev_cant_read = B_TRUE; 2696 } else { 2697 vd0->vdev_cant_write = B_TRUE; 2698 } 2699 guid0 = vd0->vdev_guid; 2700 } 2701 } else { 2702 /* 2703 * Inject errors on an l2cache device. 2704 */ 2705 spa_aux_vdev_t *sav = &spa->spa_l2cache; 2706 2707 if (sav->sav_count == 0) { 2708 spa_config_exit(spa, SCL_STATE, FTAG); 2709 return; 2710 } 2711 vd0 = sav->sav_vdevs[ztest_random(sav->sav_count)]; 2712 guid0 = vd0->vdev_guid; 2713 (void) strcpy(path0, vd0->vdev_path); 2714 (void) strcpy(pathrand, vd0->vdev_path); 2715 2716 leaf = 0; 2717 leaves = 1; 2718 maxfaults = INT_MAX; /* no limit on cache devices */ 2719 } 2720 2721 dprintf("damaging %s and %s\n", path0, pathrand); 2722 2723 spa_config_exit(spa, SCL_STATE, FTAG); 2724 2725 if (maxfaults == 0) 2726 return; 2727 2728 /* 2729 * If we can tolerate two or more faults, randomly online/offline vd0. 2730 */ 2731 if (maxfaults >= 2 && guid0 != 0) { 2732 if (ztest_random(10) < 6) 2733 (void) vdev_offline(spa, guid0, B_TRUE); 2734 else 2735 (void) vdev_online(spa, guid0, B_FALSE, NULL); 2736 } 2737 2738 /* 2739 * We have at least single-fault tolerance, so inject data corruption. 2740 */ 2741 fd = open(pathrand, O_RDWR); 2742 2743 if (fd == -1) /* we hit a gap in the device namespace */ 2744 return; 2745 2746 fsize = lseek(fd, 0, SEEK_END); 2747 2748 while (--iters != 0) { 2749 offset = ztest_random(fsize / (leaves << bshift)) * 2750 (leaves << bshift) + (leaf << bshift) + 2751 (ztest_random(1ULL << (bshift - 1)) & -8ULL); 2752 2753 if (offset >= fsize) 2754 continue; 2755 2756 if (zopt_verbose >= 6) 2757 (void) printf("injecting bad word into %s," 2758 " offset 0x%llx\n", pathrand, (u_longlong_t)offset); 2759 2760 if (pwrite(fd, &bad, sizeof (bad), offset) != sizeof (bad)) 2761 fatal(1, "can't inject bad word at 0x%llx in %s", 2762 offset, pathrand); 2763 } 2764 2765 (void) close(fd); 2766 } 2767 2768 /* 2769 * Scrub the pool. 2770 */ 2771 void 2772 ztest_scrub(ztest_args_t *za) 2773 { 2774 spa_t *spa = za->za_spa; 2775 2776 (void) spa_scrub(spa, POOL_SCRUB_EVERYTHING); 2777 (void) poll(NULL, 0, 1000); /* wait a second, then force a restart */ 2778 (void) spa_scrub(spa, POOL_SCRUB_EVERYTHING); 2779 } 2780 2781 /* 2782 * Rename the pool to a different name and then rename it back. 2783 */ 2784 void 2785 ztest_spa_rename(ztest_args_t *za) 2786 { 2787 char *oldname, *newname; 2788 int error; 2789 spa_t *spa; 2790 2791 (void) rw_wrlock(&ztest_shared->zs_name_lock); 2792 2793 oldname = za->za_pool; 2794 newname = umem_alloc(strlen(oldname) + 5, UMEM_NOFAIL); 2795 (void) strcpy(newname, oldname); 2796 (void) strcat(newname, "_tmp"); 2797 2798 /* 2799 * Do the rename 2800 */ 2801 error = spa_rename(oldname, newname); 2802 if (error) 2803 fatal(0, "spa_rename('%s', '%s') = %d", oldname, 2804 newname, error); 2805 2806 /* 2807 * Try to open it under the old name, which shouldn't exist 2808 */ 2809 error = spa_open(oldname, &spa, FTAG); 2810 if (error != ENOENT) 2811 fatal(0, "spa_open('%s') = %d", oldname, error); 2812 2813 /* 2814 * Open it under the new name and make sure it's still the same spa_t. 2815 */ 2816 error = spa_open(newname, &spa, FTAG); 2817 if (error != 0) 2818 fatal(0, "spa_open('%s') = %d", newname, error); 2819 2820 ASSERT(spa == za->za_spa); 2821 spa_close(spa, FTAG); 2822 2823 /* 2824 * Rename it back to the original 2825 */ 2826 error = spa_rename(newname, oldname); 2827 if (error) 2828 fatal(0, "spa_rename('%s', '%s') = %d", newname, 2829 oldname, error); 2830 2831 /* 2832 * Make sure it can still be opened 2833 */ 2834 error = spa_open(oldname, &spa, FTAG); 2835 if (error != 0) 2836 fatal(0, "spa_open('%s') = %d", oldname, error); 2837 2838 ASSERT(spa == za->za_spa); 2839 spa_close(spa, FTAG); 2840 2841 umem_free(newname, strlen(newname) + 1); 2842 2843 (void) rw_unlock(&ztest_shared->zs_name_lock); 2844 } 2845 2846 2847 /* 2848 * Completely obliterate one disk. 2849 */ 2850 static void 2851 ztest_obliterate_one_disk(uint64_t vdev) 2852 { 2853 int fd; 2854 char dev_name[MAXPATHLEN], copy_name[MAXPATHLEN]; 2855 size_t fsize; 2856 2857 if (zopt_maxfaults < 2) 2858 return; 2859 2860 (void) sprintf(dev_name, ztest_dev_template, zopt_dir, zopt_pool, vdev); 2861 (void) snprintf(copy_name, MAXPATHLEN, "%s.old", dev_name); 2862 2863 fd = open(dev_name, O_RDWR); 2864 2865 if (fd == -1) 2866 fatal(1, "can't open %s", dev_name); 2867 2868 /* 2869 * Determine the size. 2870 */ 2871 fsize = lseek(fd, 0, SEEK_END); 2872 2873 (void) close(fd); 2874 2875 /* 2876 * Rename the old device to dev_name.old (useful for debugging). 2877 */ 2878 VERIFY(rename(dev_name, copy_name) == 0); 2879 2880 /* 2881 * Create a new one. 2882 */ 2883 VERIFY((fd = open(dev_name, O_RDWR | O_CREAT | O_TRUNC, 0666)) >= 0); 2884 VERIFY(ftruncate(fd, fsize) == 0); 2885 (void) close(fd); 2886 } 2887 2888 static void 2889 ztest_replace_one_disk(spa_t *spa, uint64_t vdev) 2890 { 2891 char dev_name[MAXPATHLEN]; 2892 nvlist_t *root; 2893 int error; 2894 uint64_t guid; 2895 vdev_t *vd; 2896 2897 (void) sprintf(dev_name, ztest_dev_template, zopt_dir, zopt_pool, vdev); 2898 2899 /* 2900 * Build the nvlist describing dev_name. 2901 */ 2902 root = make_vdev_root(dev_name, NULL, 0, 0, 0, 0, 0, 1); 2903 2904 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER); 2905 if ((vd = vdev_lookup_by_path(spa->spa_root_vdev, dev_name)) == NULL) 2906 guid = 0; 2907 else 2908 guid = vd->vdev_guid; 2909 spa_config_exit(spa, SCL_VDEV, FTAG); 2910 error = spa_vdev_attach(spa, guid, root, B_TRUE); 2911 if (error != 0 && 2912 error != EBUSY && 2913 error != ENOTSUP && 2914 error != ENODEV && 2915 error != EDOM) 2916 fatal(0, "spa_vdev_attach(in-place) = %d", error); 2917 2918 nvlist_free(root); 2919 } 2920 2921 static void 2922 ztest_verify_blocks(char *pool) 2923 { 2924 int status; 2925 char zdb[MAXPATHLEN + MAXNAMELEN + 20]; 2926 char zbuf[1024]; 2927 char *bin; 2928 char *ztest; 2929 char *isa; 2930 int isalen; 2931 FILE *fp; 2932 2933 (void) realpath(getexecname(), zdb); 2934 2935 /* zdb lives in /usr/sbin, while ztest lives in /usr/bin */ 2936 bin = strstr(zdb, "/usr/bin/"); 2937 ztest = strstr(bin, "/ztest"); 2938 isa = bin + 8; 2939 isalen = ztest - isa; 2940 isa = strdup(isa); 2941 /* LINTED */ 2942 (void) sprintf(bin, 2943 "/usr/sbin%.*s/zdb -bc%s%s -U /tmp/zpool.cache -O %s %s", 2944 isalen, 2945 isa, 2946 zopt_verbose >= 3 ? "s" : "", 2947 zopt_verbose >= 4 ? "v" : "", 2948 ztest_random(2) == 0 ? "pre" : "post", pool); 2949 free(isa); 2950 2951 if (zopt_verbose >= 5) 2952 (void) printf("Executing %s\n", strstr(zdb, "zdb ")); 2953 2954 fp = popen(zdb, "r"); 2955 2956 while (fgets(zbuf, sizeof (zbuf), fp) != NULL) 2957 if (zopt_verbose >= 3) 2958 (void) printf("%s", zbuf); 2959 2960 status = pclose(fp); 2961 2962 if (status == 0) 2963 return; 2964 2965 ztest_dump_core = 0; 2966 if (WIFEXITED(status)) 2967 fatal(0, "'%s' exit code %d", zdb, WEXITSTATUS(status)); 2968 else 2969 fatal(0, "'%s' died with signal %d", zdb, WTERMSIG(status)); 2970 } 2971 2972 static void 2973 ztest_walk_pool_directory(char *header) 2974 { 2975 spa_t *spa = NULL; 2976 2977 if (zopt_verbose >= 6) 2978 (void) printf("%s\n", header); 2979 2980 mutex_enter(&spa_namespace_lock); 2981 while ((spa = spa_next(spa)) != NULL) 2982 if (zopt_verbose >= 6) 2983 (void) printf("\t%s\n", spa_name(spa)); 2984 mutex_exit(&spa_namespace_lock); 2985 } 2986 2987 static void 2988 ztest_spa_import_export(char *oldname, char *newname) 2989 { 2990 nvlist_t *config; 2991 uint64_t pool_guid; 2992 spa_t *spa; 2993 int error; 2994 2995 if (zopt_verbose >= 4) { 2996 (void) printf("import/export: old = %s, new = %s\n", 2997 oldname, newname); 2998 } 2999 3000 /* 3001 * Clean up from previous runs. 3002 */ 3003 (void) spa_destroy(newname); 3004 3005 /* 3006 * Get the pool's configuration and guid. 3007 */ 3008 error = spa_open(oldname, &spa, FTAG); 3009 if (error) 3010 fatal(0, "spa_open('%s') = %d", oldname, error); 3011 3012 pool_guid = spa_guid(spa); 3013 spa_close(spa, FTAG); 3014 3015 ztest_walk_pool_directory("pools before export"); 3016 3017 /* 3018 * Export it. 3019 */ 3020 error = spa_export(oldname, &config, B_FALSE); 3021 if (error) 3022 fatal(0, "spa_export('%s') = %d", oldname, error); 3023 3024 ztest_walk_pool_directory("pools after export"); 3025 3026 /* 3027 * Import it under the new name. 3028 */ 3029 error = spa_import(newname, config, NULL); 3030 if (error) 3031 fatal(0, "spa_import('%s') = %d", newname, error); 3032 3033 ztest_walk_pool_directory("pools after import"); 3034 3035 /* 3036 * Try to import it again -- should fail with EEXIST. 3037 */ 3038 error = spa_import(newname, config, NULL); 3039 if (error != EEXIST) 3040 fatal(0, "spa_import('%s') twice", newname); 3041 3042 /* 3043 * Try to import it under a different name -- should fail with EEXIST. 3044 */ 3045 error = spa_import(oldname, config, NULL); 3046 if (error != EEXIST) 3047 fatal(0, "spa_import('%s') under multiple names", newname); 3048 3049 /* 3050 * Verify that the pool is no longer visible under the old name. 3051 */ 3052 error = spa_open(oldname, &spa, FTAG); 3053 if (error != ENOENT) 3054 fatal(0, "spa_open('%s') = %d", newname, error); 3055 3056 /* 3057 * Verify that we can open and close the pool using the new name. 3058 */ 3059 error = spa_open(newname, &spa, FTAG); 3060 if (error) 3061 fatal(0, "spa_open('%s') = %d", newname, error); 3062 ASSERT(pool_guid == spa_guid(spa)); 3063 spa_close(spa, FTAG); 3064 3065 nvlist_free(config); 3066 } 3067 3068 static void * 3069 ztest_resume(void *arg) 3070 { 3071 spa_t *spa = arg; 3072 3073 while (!ztest_exiting) { 3074 (void) poll(NULL, 0, 1000); 3075 3076 if (!spa_suspended(spa)) 3077 continue; 3078 3079 spa_vdev_state_enter(spa); 3080 vdev_clear(spa, NULL); 3081 (void) spa_vdev_state_exit(spa, NULL, 0); 3082 3083 zio_resume(spa); 3084 } 3085 return (NULL); 3086 } 3087 3088 static void * 3089 ztest_thread(void *arg) 3090 { 3091 ztest_args_t *za = arg; 3092 ztest_shared_t *zs = ztest_shared; 3093 hrtime_t now, functime; 3094 ztest_info_t *zi; 3095 int f, i; 3096 3097 while ((now = gethrtime()) < za->za_stop) { 3098 /* 3099 * See if it's time to force a crash. 3100 */ 3101 if (now > za->za_kill) { 3102 zs->zs_alloc = spa_get_alloc(za->za_spa); 3103 zs->zs_space = spa_get_space(za->za_spa); 3104 (void) kill(getpid(), SIGKILL); 3105 } 3106 3107 /* 3108 * Pick a random function. 3109 */ 3110 f = ztest_random(ZTEST_FUNCS); 3111 zi = &zs->zs_info[f]; 3112 3113 /* 3114 * Decide whether to call it, based on the requested frequency. 3115 */ 3116 if (zi->zi_call_target == 0 || 3117 (double)zi->zi_call_total / zi->zi_call_target > 3118 (double)(now - zs->zs_start_time) / (zopt_time * NANOSEC)) 3119 continue; 3120 3121 atomic_add_64(&zi->zi_calls, 1); 3122 atomic_add_64(&zi->zi_call_total, 1); 3123 3124 za->za_diroff = (za->za_instance * ZTEST_FUNCS + f) * 3125 ZTEST_DIRSIZE; 3126 za->za_diroff_shared = (1ULL << 63); 3127 3128 for (i = 0; i < zi->zi_iters; i++) 3129 zi->zi_func(za); 3130 3131 functime = gethrtime() - now; 3132 3133 atomic_add_64(&zi->zi_call_time, functime); 3134 3135 if (zopt_verbose >= 4) { 3136 Dl_info dli; 3137 (void) dladdr((void *)zi->zi_func, &dli); 3138 (void) printf("%6.2f sec in %s\n", 3139 (double)functime / NANOSEC, dli.dli_sname); 3140 } 3141 3142 /* 3143 * If we're getting ENOSPC with some regularity, stop. 3144 */ 3145 if (zs->zs_enospc_count > 10) 3146 break; 3147 } 3148 3149 return (NULL); 3150 } 3151 3152 /* 3153 * Kick off threads to run tests on all datasets in parallel. 3154 */ 3155 static void 3156 ztest_run(char *pool) 3157 { 3158 int t, d, error; 3159 ztest_shared_t *zs = ztest_shared; 3160 ztest_args_t *za; 3161 spa_t *spa; 3162 char name[100]; 3163 thread_t resume_tid; 3164 3165 ztest_exiting = B_FALSE; 3166 3167 (void) _mutex_init(&zs->zs_vdev_lock, USYNC_THREAD, NULL); 3168 (void) rwlock_init(&zs->zs_name_lock, USYNC_THREAD, NULL); 3169 3170 for (t = 0; t < ZTEST_SYNC_LOCKS; t++) 3171 (void) _mutex_init(&zs->zs_sync_lock[t], USYNC_THREAD, NULL); 3172 3173 /* 3174 * Destroy one disk before we even start. 3175 * It's mirrored, so everything should work just fine. 3176 * This makes us exercise fault handling very early in spa_load(). 3177 */ 3178 ztest_obliterate_one_disk(0); 3179 3180 /* 3181 * Verify that the sum of the sizes of all blocks in the pool 3182 * equals the SPA's allocated space total. 3183 */ 3184 ztest_verify_blocks(pool); 3185 3186 /* 3187 * Kick off a replacement of the disk we just obliterated. 3188 */ 3189 kernel_init(FREAD | FWRITE); 3190 VERIFY(spa_open(pool, &spa, FTAG) == 0); 3191 ztest_replace_one_disk(spa, 0); 3192 if (zopt_verbose >= 5) 3193 show_pool_stats(spa); 3194 spa_close(spa, FTAG); 3195 kernel_fini(); 3196 3197 kernel_init(FREAD | FWRITE); 3198 3199 /* 3200 * Verify that we can export the pool and reimport it under a 3201 * different name. 3202 */ 3203 if (ztest_random(2) == 0) { 3204 (void) snprintf(name, 100, "%s_import", pool); 3205 ztest_spa_import_export(pool, name); 3206 ztest_spa_import_export(name, pool); 3207 } 3208 3209 /* 3210 * Verify that we can loop over all pools. 3211 */ 3212 mutex_enter(&spa_namespace_lock); 3213 for (spa = spa_next(NULL); spa != NULL; spa = spa_next(spa)) { 3214 if (zopt_verbose > 3) { 3215 (void) printf("spa_next: found %s\n", spa_name(spa)); 3216 } 3217 } 3218 mutex_exit(&spa_namespace_lock); 3219 3220 /* 3221 * Open our pool. 3222 */ 3223 VERIFY(spa_open(pool, &spa, FTAG) == 0); 3224 3225 /* 3226 * Create a thread to periodically resume suspended I/O. 3227 */ 3228 VERIFY(thr_create(0, 0, ztest_resume, spa, THR_BOUND, 3229 &resume_tid) == 0); 3230 3231 /* 3232 * Verify that we can safely inquire about about any object, 3233 * whether it's allocated or not. To make it interesting, 3234 * we probe a 5-wide window around each power of two. 3235 * This hits all edge cases, including zero and the max. 3236 */ 3237 for (t = 0; t < 64; t++) { 3238 for (d = -5; d <= 5; d++) { 3239 error = dmu_object_info(spa->spa_meta_objset, 3240 (1ULL << t) + d, NULL); 3241 ASSERT(error == 0 || error == ENOENT || 3242 error == EINVAL); 3243 } 3244 } 3245 3246 /* 3247 * Now kick off all the tests that run in parallel. 3248 */ 3249 zs->zs_enospc_count = 0; 3250 3251 za = umem_zalloc(zopt_threads * sizeof (ztest_args_t), UMEM_NOFAIL); 3252 3253 if (zopt_verbose >= 4) 3254 (void) printf("starting main threads...\n"); 3255 3256 za[0].za_start = gethrtime(); 3257 za[0].za_stop = za[0].za_start + zopt_passtime * NANOSEC; 3258 za[0].za_stop = MIN(za[0].za_stop, zs->zs_stop_time); 3259 za[0].za_kill = za[0].za_stop; 3260 if (ztest_random(100) < zopt_killrate) 3261 za[0].za_kill -= ztest_random(zopt_passtime * NANOSEC); 3262 3263 for (t = 0; t < zopt_threads; t++) { 3264 d = t % zopt_datasets; 3265 3266 (void) strcpy(za[t].za_pool, pool); 3267 za[t].za_os = za[d].za_os; 3268 za[t].za_spa = spa; 3269 za[t].za_zilog = za[d].za_zilog; 3270 za[t].za_instance = t; 3271 za[t].za_random = ztest_random(-1ULL); 3272 za[t].za_start = za[0].za_start; 3273 za[t].za_stop = za[0].za_stop; 3274 za[t].za_kill = za[0].za_kill; 3275 3276 if (t < zopt_datasets) { 3277 ztest_replay_t zr; 3278 int test_future = FALSE; 3279 (void) rw_rdlock(&ztest_shared->zs_name_lock); 3280 (void) snprintf(name, 100, "%s/%s_%d", pool, pool, d); 3281 error = dmu_objset_create(name, DMU_OST_OTHER, NULL, 0, 3282 ztest_create_cb, NULL); 3283 if (error == EEXIST) { 3284 test_future = TRUE; 3285 } else if (error == ENOSPC) { 3286 zs->zs_enospc_count++; 3287 (void) rw_unlock(&ztest_shared->zs_name_lock); 3288 break; 3289 } else if (error != 0) { 3290 fatal(0, "dmu_objset_create(%s) = %d", 3291 name, error); 3292 } 3293 error = dmu_objset_open(name, DMU_OST_OTHER, 3294 DS_MODE_USER, &za[d].za_os); 3295 if (error) 3296 fatal(0, "dmu_objset_open('%s') = %d", 3297 name, error); 3298 (void) rw_unlock(&ztest_shared->zs_name_lock); 3299 if (test_future) 3300 ztest_dmu_check_future_leak(&za[t]); 3301 zr.zr_os = za[d].za_os; 3302 zil_replay(zr.zr_os, &zr, &zr.zr_assign, 3303 ztest_replay_vector, NULL); 3304 za[d].za_zilog = zil_open(za[d].za_os, NULL); 3305 } 3306 3307 VERIFY(thr_create(0, 0, ztest_thread, &za[t], THR_BOUND, 3308 &za[t].za_thread) == 0); 3309 } 3310 3311 while (--t >= 0) { 3312 VERIFY(thr_join(za[t].za_thread, NULL, NULL) == 0); 3313 if (za[t].za_th) 3314 traverse_fini(za[t].za_th); 3315 if (t < zopt_datasets) { 3316 zil_close(za[t].za_zilog); 3317 dmu_objset_close(za[t].za_os); 3318 } 3319 } 3320 3321 if (zopt_verbose >= 3) 3322 show_pool_stats(spa); 3323 3324 txg_wait_synced(spa_get_dsl(spa), 0); 3325 3326 zs->zs_alloc = spa_get_alloc(spa); 3327 zs->zs_space = spa_get_space(spa); 3328 3329 /* 3330 * If we had out-of-space errors, destroy a random objset. 3331 */ 3332 if (zs->zs_enospc_count != 0) { 3333 (void) rw_rdlock(&ztest_shared->zs_name_lock); 3334 d = (int)ztest_random(zopt_datasets); 3335 (void) snprintf(name, 100, "%s/%s_%d", pool, pool, d); 3336 if (zopt_verbose >= 3) 3337 (void) printf("Destroying %s to free up space\n", name); 3338 (void) dmu_objset_find(name, ztest_destroy_cb, &za[d], 3339 DS_FIND_SNAPSHOTS | DS_FIND_CHILDREN); 3340 (void) rw_unlock(&ztest_shared->zs_name_lock); 3341 } 3342 3343 txg_wait_synced(spa_get_dsl(spa), 0); 3344 3345 umem_free(za, zopt_threads * sizeof (ztest_args_t)); 3346 3347 /* Kill the resume thread */ 3348 ztest_exiting = B_TRUE; 3349 VERIFY(thr_join(resume_tid, NULL, NULL) == 0); 3350 3351 /* 3352 * Right before closing the pool, kick off a bunch of async I/O; 3353 * spa_close() should wait for it to complete. 3354 */ 3355 for (t = 1; t < 50; t++) 3356 dmu_prefetch(spa->spa_meta_objset, t, 0, 1 << 15); 3357 3358 spa_close(spa, FTAG); 3359 3360 kernel_fini(); 3361 } 3362 3363 void 3364 print_time(hrtime_t t, char *timebuf) 3365 { 3366 hrtime_t s = t / NANOSEC; 3367 hrtime_t m = s / 60; 3368 hrtime_t h = m / 60; 3369 hrtime_t d = h / 24; 3370 3371 s -= m * 60; 3372 m -= h * 60; 3373 h -= d * 24; 3374 3375 timebuf[0] = '\0'; 3376 3377 if (d) 3378 (void) sprintf(timebuf, 3379 "%llud%02lluh%02llum%02llus", d, h, m, s); 3380 else if (h) 3381 (void) sprintf(timebuf, "%lluh%02llum%02llus", h, m, s); 3382 else if (m) 3383 (void) sprintf(timebuf, "%llum%02llus", m, s); 3384 else 3385 (void) sprintf(timebuf, "%llus", s); 3386 } 3387 3388 /* 3389 * Create a storage pool with the given name and initial vdev size. 3390 * Then create the specified number of datasets in the pool. 3391 */ 3392 static void 3393 ztest_init(char *pool) 3394 { 3395 spa_t *spa; 3396 int error; 3397 nvlist_t *nvroot; 3398 3399 kernel_init(FREAD | FWRITE); 3400 3401 /* 3402 * Create the storage pool. 3403 */ 3404 (void) spa_destroy(pool); 3405 ztest_shared->zs_vdev_primaries = 0; 3406 nvroot = make_vdev_root(NULL, NULL, zopt_vdev_size, 0, 3407 0, zopt_raidz, zopt_mirrors, 1); 3408 error = spa_create(pool, nvroot, NULL, NULL, NULL); 3409 nvlist_free(nvroot); 3410 3411 if (error) 3412 fatal(0, "spa_create() = %d", error); 3413 error = spa_open(pool, &spa, FTAG); 3414 if (error) 3415 fatal(0, "spa_open() = %d", error); 3416 3417 if (zopt_verbose >= 3) 3418 show_pool_stats(spa); 3419 3420 spa_close(spa, FTAG); 3421 3422 kernel_fini(); 3423 } 3424 3425 int 3426 main(int argc, char **argv) 3427 { 3428 int kills = 0; 3429 int iters = 0; 3430 int i, f; 3431 ztest_shared_t *zs; 3432 ztest_info_t *zi; 3433 char timebuf[100]; 3434 char numbuf[6]; 3435 3436 (void) setvbuf(stdout, NULL, _IOLBF, 0); 3437 3438 /* Override location of zpool.cache */ 3439 spa_config_path = "/tmp/zpool.cache"; 3440 3441 ztest_random_fd = open("/dev/urandom", O_RDONLY); 3442 3443 process_options(argc, argv); 3444 3445 argc -= optind; 3446 argv += optind; 3447 3448 dprintf_setup(&argc, argv); 3449 3450 /* 3451 * Blow away any existing copy of zpool.cache 3452 */ 3453 if (zopt_init != 0) 3454 (void) remove("/tmp/zpool.cache"); 3455 3456 zs = ztest_shared = (void *)mmap(0, 3457 P2ROUNDUP(sizeof (ztest_shared_t), getpagesize()), 3458 PROT_READ | PROT_WRITE, MAP_SHARED | MAP_ANON, -1, 0); 3459 3460 if (zopt_verbose >= 1) { 3461 (void) printf("%llu vdevs, %d datasets, %d threads," 3462 " %llu seconds...\n", 3463 (u_longlong_t)zopt_vdevs, zopt_datasets, zopt_threads, 3464 (u_longlong_t)zopt_time); 3465 } 3466 3467 /* 3468 * Create and initialize our storage pool. 3469 */ 3470 for (i = 1; i <= zopt_init; i++) { 3471 bzero(zs, sizeof (ztest_shared_t)); 3472 if (zopt_verbose >= 3 && zopt_init != 1) 3473 (void) printf("ztest_init(), pass %d\n", i); 3474 ztest_init(zopt_pool); 3475 } 3476 3477 /* 3478 * Initialize the call targets for each function. 3479 */ 3480 for (f = 0; f < ZTEST_FUNCS; f++) { 3481 zi = &zs->zs_info[f]; 3482 3483 *zi = ztest_info[f]; 3484 3485 if (*zi->zi_interval == 0) 3486 zi->zi_call_target = UINT64_MAX; 3487 else 3488 zi->zi_call_target = zopt_time / *zi->zi_interval; 3489 } 3490 3491 zs->zs_start_time = gethrtime(); 3492 zs->zs_stop_time = zs->zs_start_time + zopt_time * NANOSEC; 3493 3494 /* 3495 * Run the tests in a loop. These tests include fault injection 3496 * to verify that self-healing data works, and forced crashes 3497 * to verify that we never lose on-disk consistency. 3498 */ 3499 while (gethrtime() < zs->zs_stop_time) { 3500 int status; 3501 pid_t pid; 3502 char *tmp; 3503 3504 /* 3505 * Initialize the workload counters for each function. 3506 */ 3507 for (f = 0; f < ZTEST_FUNCS; f++) { 3508 zi = &zs->zs_info[f]; 3509 zi->zi_calls = 0; 3510 zi->zi_call_time = 0; 3511 } 3512 3513 pid = fork(); 3514 3515 if (pid == -1) 3516 fatal(1, "fork failed"); 3517 3518 if (pid == 0) { /* child */ 3519 struct rlimit rl = { 1024, 1024 }; 3520 (void) setrlimit(RLIMIT_NOFILE, &rl); 3521 (void) enable_extended_FILE_stdio(-1, -1); 3522 ztest_run(zopt_pool); 3523 exit(0); 3524 } 3525 3526 while (waitpid(pid, &status, 0) != pid) 3527 continue; 3528 3529 if (WIFEXITED(status)) { 3530 if (WEXITSTATUS(status) != 0) { 3531 (void) fprintf(stderr, 3532 "child exited with code %d\n", 3533 WEXITSTATUS(status)); 3534 exit(2); 3535 } 3536 } else if (WIFSIGNALED(status)) { 3537 if (WTERMSIG(status) != SIGKILL) { 3538 (void) fprintf(stderr, 3539 "child died with signal %d\n", 3540 WTERMSIG(status)); 3541 exit(3); 3542 } 3543 kills++; 3544 } else { 3545 (void) fprintf(stderr, "something strange happened " 3546 "to child\n"); 3547 exit(4); 3548 } 3549 3550 iters++; 3551 3552 if (zopt_verbose >= 1) { 3553 hrtime_t now = gethrtime(); 3554 3555 now = MIN(now, zs->zs_stop_time); 3556 print_time(zs->zs_stop_time - now, timebuf); 3557 nicenum(zs->zs_space, numbuf); 3558 3559 (void) printf("Pass %3d, %8s, %3llu ENOSPC, " 3560 "%4.1f%% of %5s used, %3.0f%% done, %8s to go\n", 3561 iters, 3562 WIFEXITED(status) ? "Complete" : "SIGKILL", 3563 (u_longlong_t)zs->zs_enospc_count, 3564 100.0 * zs->zs_alloc / zs->zs_space, 3565 numbuf, 3566 100.0 * (now - zs->zs_start_time) / 3567 (zopt_time * NANOSEC), timebuf); 3568 } 3569 3570 if (zopt_verbose >= 2) { 3571 (void) printf("\nWorkload summary:\n\n"); 3572 (void) printf("%7s %9s %s\n", 3573 "Calls", "Time", "Function"); 3574 (void) printf("%7s %9s %s\n", 3575 "-----", "----", "--------"); 3576 for (f = 0; f < ZTEST_FUNCS; f++) { 3577 Dl_info dli; 3578 3579 zi = &zs->zs_info[f]; 3580 print_time(zi->zi_call_time, timebuf); 3581 (void) dladdr((void *)zi->zi_func, &dli); 3582 (void) printf("%7llu %9s %s\n", 3583 (u_longlong_t)zi->zi_calls, timebuf, 3584 dli.dli_sname); 3585 } 3586 (void) printf("\n"); 3587 } 3588 3589 /* 3590 * It's possible that we killed a child during a rename test, in 3591 * which case we'll have a 'ztest_tmp' pool lying around instead 3592 * of 'ztest'. Do a blind rename in case this happened. 3593 */ 3594 tmp = umem_alloc(strlen(zopt_pool) + 5, UMEM_NOFAIL); 3595 (void) strcpy(tmp, zopt_pool); 3596 (void) strcat(tmp, "_tmp"); 3597 kernel_init(FREAD | FWRITE); 3598 (void) spa_rename(tmp, zopt_pool); 3599 kernel_fini(); 3600 umem_free(tmp, strlen(tmp) + 1); 3601 } 3602 3603 ztest_verify_blocks(zopt_pool); 3604 3605 if (zopt_verbose >= 1) { 3606 (void) printf("%d killed, %d completed, %.0f%% kill rate\n", 3607 kills, iters - kills, (100.0 * kills) / MAX(1, iters)); 3608 } 3609 3610 return (0); 3611 } 3612