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 2007 Sun Microsystems, Inc. All rights reserved. 23 * Use is subject to license terms. 24 */ 25 26 #pragma ident "%Z%%M% %I% %E% SMI" 27 28 /* 29 * The objective of this program is to provide a DMU/ZAP/SPA stress test 30 * that runs entirely in userland, is easy to use, and easy to extend. 31 * 32 * The overall design of the ztest program is as follows: 33 * 34 * (1) For each major functional area (e.g. adding vdevs to a pool, 35 * creating and destroying datasets, reading and writing objects, etc) 36 * we have a simple routine to test that functionality. These 37 * individual routines do not have to do anything "stressful". 38 * 39 * (2) We turn these simple functionality tests into a stress test by 40 * running them all in parallel, with as many threads as desired, 41 * and spread across as many datasets, objects, and vdevs as desired. 42 * 43 * (3) While all this is happening, we inject faults into the pool to 44 * verify that self-healing data really works. 45 * 46 * (4) Every time we open a dataset, we change its checksum and compression 47 * functions. Thus even individual objects vary from block to block 48 * in which checksum they use and whether they're compressed. 49 * 50 * (5) To verify that we never lose on-disk consistency after a crash, 51 * we run the entire test in a child of the main process. 52 * At random times, the child self-immolates with a SIGKILL. 53 * This is the software equivalent of pulling the power cord. 54 * The parent then runs the test again, using the existing 55 * storage pool, as many times as desired. 56 * 57 * (6) To verify that we don't have future leaks or temporal incursions, 58 * many of the functional tests record the transaction group number 59 * as part of their data. When reading old data, they verify that 60 * the transaction group number is less than the current, open txg. 61 * If you add a new test, please do this if applicable. 62 * 63 * When run with no arguments, ztest runs for about five minutes and 64 * produces no output if successful. To get a little bit of information, 65 * specify -V. To get more information, specify -VV, and so on. 66 * 67 * To turn this into an overnight stress test, use -T to specify run time. 68 * 69 * You can ask more more vdevs [-v], datasets [-d], or threads [-t] 70 * to increase the pool capacity, fanout, and overall stress level. 71 * 72 * The -N(okill) option will suppress kills, so each child runs to completion. 73 * This can be useful when you're trying to distinguish temporal incursions 74 * from plain old race conditions. 75 */ 76 77 #include <sys/zfs_context.h> 78 #include <sys/spa.h> 79 #include <sys/dmu.h> 80 #include <sys/txg.h> 81 #include <sys/zap.h> 82 #include <sys/dmu_traverse.h> 83 #include <sys/dmu_objset.h> 84 #include <sys/poll.h> 85 #include <sys/stat.h> 86 #include <sys/time.h> 87 #include <sys/wait.h> 88 #include <sys/mman.h> 89 #include <sys/resource.h> 90 #include <sys/zio.h> 91 #include <sys/zio_checksum.h> 92 #include <sys/zio_compress.h> 93 #include <sys/zil.h> 94 #include <sys/vdev_impl.h> 95 #include <sys/spa_impl.h> 96 #include <sys/dsl_prop.h> 97 #include <sys/refcount.h> 98 #include <stdio.h> 99 #include <stdio_ext.h> 100 #include <stdlib.h> 101 #include <unistd.h> 102 #include <signal.h> 103 #include <umem.h> 104 #include <dlfcn.h> 105 #include <ctype.h> 106 #include <math.h> 107 #include <sys/fs/zfs.h> 108 109 static char cmdname[] = "ztest"; 110 static char *zopt_pool = cmdname; 111 112 static uint64_t zopt_vdevs = 5; 113 static uint64_t zopt_vdevtime; 114 static int zopt_ashift = SPA_MINBLOCKSHIFT; 115 static int zopt_mirrors = 2; 116 static int zopt_raidz = 4; 117 static int zopt_raidz_parity = 1; 118 static size_t zopt_vdev_size = SPA_MINDEVSIZE; 119 static int zopt_datasets = 7; 120 static int zopt_threads = 23; 121 static uint64_t zopt_passtime = 60; /* 60 seconds */ 122 static uint64_t zopt_killrate = 70; /* 70% kill rate */ 123 static int zopt_verbose = 0; 124 static int zopt_init = 1; 125 static char *zopt_dir = "/tmp"; 126 static uint64_t zopt_time = 300; /* 5 minutes */ 127 static int zopt_maxfaults; 128 static uint16_t zopt_write_fail_shift = 5; 129 130 typedef struct ztest_args { 131 char *za_pool; 132 objset_t *za_os; 133 zilog_t *za_zilog; 134 thread_t za_thread; 135 uint64_t za_instance; 136 uint64_t za_random; 137 uint64_t za_diroff; 138 uint64_t za_diroff_shared; 139 uint64_t za_zil_seq; 140 hrtime_t za_start; 141 hrtime_t za_stop; 142 hrtime_t za_kill; 143 traverse_handle_t *za_th; 144 } ztest_args_t; 145 146 typedef void ztest_func_t(ztest_args_t *); 147 148 /* 149 * Note: these aren't static because we want dladdr() to work. 150 */ 151 ztest_func_t ztest_dmu_read_write; 152 ztest_func_t ztest_dmu_write_parallel; 153 ztest_func_t ztest_dmu_object_alloc_free; 154 ztest_func_t ztest_zap; 155 ztest_func_t ztest_zap_parallel; 156 ztest_func_t ztest_traverse; 157 ztest_func_t ztest_dsl_prop_get_set; 158 ztest_func_t ztest_dmu_objset_create_destroy; 159 ztest_func_t ztest_dmu_snapshot_create_destroy; 160 ztest_func_t ztest_spa_create_destroy; 161 ztest_func_t ztest_fault_inject; 162 ztest_func_t ztest_vdev_attach_detach; 163 ztest_func_t ztest_vdev_LUN_growth; 164 ztest_func_t ztest_vdev_add_remove; 165 ztest_func_t ztest_scrub; 166 ztest_func_t ztest_spa_rename; 167 168 typedef struct ztest_info { 169 ztest_func_t *zi_func; /* test function */ 170 uint64_t *zi_interval; /* execute every <interval> seconds */ 171 uint64_t zi_calls; /* per-pass count */ 172 uint64_t zi_call_time; /* per-pass time */ 173 uint64_t zi_call_total; /* cumulative total */ 174 uint64_t zi_call_target; /* target cumulative total */ 175 } ztest_info_t; 176 177 uint64_t zopt_always = 0; /* all the time */ 178 uint64_t zopt_often = 1; /* every second */ 179 uint64_t zopt_sometimes = 10; /* every 10 seconds */ 180 uint64_t zopt_rarely = 60; /* every 60 seconds */ 181 182 ztest_info_t ztest_info[] = { 183 { ztest_dmu_read_write, &zopt_always }, 184 { ztest_dmu_write_parallel, &zopt_always }, 185 { ztest_dmu_object_alloc_free, &zopt_always }, 186 { ztest_zap, &zopt_always }, 187 { ztest_zap_parallel, &zopt_always }, 188 { ztest_traverse, &zopt_often }, 189 { ztest_dsl_prop_get_set, &zopt_sometimes }, 190 { ztest_dmu_objset_create_destroy, &zopt_sometimes }, 191 { ztest_dmu_snapshot_create_destroy, &zopt_rarely }, 192 { ztest_spa_create_destroy, &zopt_sometimes }, 193 { ztest_fault_inject, &zopt_sometimes }, 194 { ztest_spa_rename, &zopt_rarely }, 195 { ztest_vdev_attach_detach, &zopt_rarely }, 196 { ztest_vdev_LUN_growth, &zopt_rarely }, 197 { ztest_vdev_add_remove, &zopt_vdevtime }, 198 { ztest_scrub, &zopt_vdevtime }, 199 }; 200 201 #define ZTEST_FUNCS (sizeof (ztest_info) / sizeof (ztest_info_t)) 202 203 #define ZTEST_SYNC_LOCKS 16 204 205 /* 206 * Stuff we need to share writably between parent and child. 207 */ 208 typedef struct ztest_shared { 209 mutex_t zs_vdev_lock; 210 rwlock_t zs_name_lock; 211 uint64_t zs_vdev_primaries; 212 uint64_t zs_enospc_count; 213 hrtime_t zs_start_time; 214 hrtime_t zs_stop_time; 215 uint64_t zs_alloc; 216 uint64_t zs_space; 217 uint64_t zs_txg; 218 ztest_info_t zs_info[ZTEST_FUNCS]; 219 mutex_t zs_sync_lock[ZTEST_SYNC_LOCKS]; 220 uint64_t zs_seq[ZTEST_SYNC_LOCKS]; 221 } ztest_shared_t; 222 223 typedef struct ztest_block_tag { 224 uint64_t bt_objset; 225 uint64_t bt_object; 226 uint64_t bt_offset; 227 uint64_t bt_txg; 228 uint64_t bt_thread; 229 uint64_t bt_seq; 230 } ztest_block_tag_t; 231 232 static char ztest_dev_template[] = "%s/%s.%llua"; 233 static ztest_shared_t *ztest_shared; 234 235 static int ztest_random_fd; 236 static int ztest_dump_core = 1; 237 238 static boolean_t ztest_exiting = B_FALSE; 239 240 extern uint64_t zio_gang_bang; 241 extern uint16_t zio_zil_fail_shift; 242 extern uint16_t zio_io_fail_shift; 243 244 #define ZTEST_DIROBJ 1 245 #define ZTEST_MICROZAP_OBJ 2 246 #define ZTEST_FATZAP_OBJ 3 247 248 #define ZTEST_DIROBJ_BLOCKSIZE (1 << 10) 249 #define ZTEST_DIRSIZE 256 250 251 static void usage(boolean_t) __NORETURN; 252 253 /* 254 * These libumem hooks provide a reasonable set of defaults for the allocator's 255 * debugging facilities. 256 */ 257 const char * 258 _umem_debug_init() 259 { 260 return ("default,verbose"); /* $UMEM_DEBUG setting */ 261 } 262 263 const char * 264 _umem_logging_init(void) 265 { 266 return ("fail,contents"); /* $UMEM_LOGGING setting */ 267 } 268 269 #define FATAL_MSG_SZ 1024 270 271 char *fatal_msg; 272 273 static void 274 fatal(int do_perror, char *message, ...) 275 { 276 va_list args; 277 int save_errno = errno; 278 char buf[FATAL_MSG_SZ]; 279 280 (void) fflush(stdout); 281 282 va_start(args, message); 283 (void) sprintf(buf, "ztest: "); 284 /* LINTED */ 285 (void) vsprintf(buf + strlen(buf), message, args); 286 va_end(args); 287 if (do_perror) { 288 (void) snprintf(buf + strlen(buf), FATAL_MSG_SZ - strlen(buf), 289 ": %s", strerror(save_errno)); 290 } 291 (void) fprintf(stderr, "%s\n", buf); 292 fatal_msg = buf; /* to ease debugging */ 293 if (ztest_dump_core) 294 abort(); 295 exit(3); 296 } 297 298 static int 299 str2shift(const char *buf) 300 { 301 const char *ends = "BKMGTPEZ"; 302 int i; 303 304 if (buf[0] == '\0') 305 return (0); 306 for (i = 0; i < strlen(ends); i++) { 307 if (toupper(buf[0]) == ends[i]) 308 break; 309 } 310 if (i == strlen(ends)) { 311 (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n", 312 buf); 313 usage(B_FALSE); 314 } 315 if (buf[1] == '\0' || (toupper(buf[1]) == 'B' && buf[2] == '\0')) { 316 return (10*i); 317 } 318 (void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n", buf); 319 usage(B_FALSE); 320 /* NOTREACHED */ 321 } 322 323 static uint64_t 324 nicenumtoull(const char *buf) 325 { 326 char *end; 327 uint64_t val; 328 329 val = strtoull(buf, &end, 0); 330 if (end == buf) { 331 (void) fprintf(stderr, "ztest: bad numeric value: %s\n", buf); 332 usage(B_FALSE); 333 } else if (end[0] == '.') { 334 double fval = strtod(buf, &end); 335 fval *= pow(2, str2shift(end)); 336 if (fval > UINT64_MAX) { 337 (void) fprintf(stderr, "ztest: value too large: %s\n", 338 buf); 339 usage(B_FALSE); 340 } 341 val = (uint64_t)fval; 342 } else { 343 int shift = str2shift(end); 344 if (shift >= 64 || (val << shift) >> shift != val) { 345 (void) fprintf(stderr, "ztest: value too large: %s\n", 346 buf); 347 usage(B_FALSE); 348 } 349 val <<= shift; 350 } 351 return (val); 352 } 353 354 static void 355 usage(boolean_t requested) 356 { 357 char nice_vdev_size[10]; 358 char nice_gang_bang[10]; 359 FILE *fp = requested ? stdout : stderr; 360 361 nicenum(zopt_vdev_size, nice_vdev_size); 362 nicenum(zio_gang_bang, nice_gang_bang); 363 364 (void) fprintf(fp, "Usage: %s\n" 365 "\t[-v vdevs (default: %llu)]\n" 366 "\t[-s size_of_each_vdev (default: %s)]\n" 367 "\t[-a alignment_shift (default: %d) (use 0 for random)]\n" 368 "\t[-m mirror_copies (default: %d)]\n" 369 "\t[-r raidz_disks (default: %d)]\n" 370 "\t[-R raidz_parity (default: %d)]\n" 371 "\t[-d datasets (default: %d)]\n" 372 "\t[-t threads (default: %d)]\n" 373 "\t[-g gang_block_threshold (default: %s)]\n" 374 "\t[-i initialize pool i times (default: %d)]\n" 375 "\t[-k kill percentage (default: %llu%%)]\n" 376 "\t[-p pool_name (default: %s)]\n" 377 "\t[-f file directory for vdev files (default: %s)]\n" 378 "\t[-V(erbose)] (use multiple times for ever more blather)\n" 379 "\t[-E(xisting)] (use existing pool instead of creating new one)\n" 380 "\t[-T time] total run time (default: %llu sec)\n" 381 "\t[-P passtime] time per pass (default: %llu sec)\n" 382 "\t[-z zil failure rate (default: fail every 2^%llu allocs)]\n" 383 "\t[-w write failure rate (default: fail every 2^%llu allocs)]\n" 384 "\t[-h] (print help)\n" 385 "", 386 cmdname, 387 (u_longlong_t)zopt_vdevs, /* -v */ 388 nice_vdev_size, /* -s */ 389 zopt_ashift, /* -a */ 390 zopt_mirrors, /* -m */ 391 zopt_raidz, /* -r */ 392 zopt_raidz_parity, /* -R */ 393 zopt_datasets, /* -d */ 394 zopt_threads, /* -t */ 395 nice_gang_bang, /* -g */ 396 zopt_init, /* -i */ 397 (u_longlong_t)zopt_killrate, /* -k */ 398 zopt_pool, /* -p */ 399 zopt_dir, /* -f */ 400 (u_longlong_t)zopt_time, /* -T */ 401 (u_longlong_t)zopt_passtime, /* -P */ 402 (u_longlong_t)zio_zil_fail_shift, /* -z */ 403 (u_longlong_t)zopt_write_fail_shift); /* -w */ 404 exit(requested ? 0 : 1); 405 } 406 407 static uint64_t 408 ztest_random(uint64_t range) 409 { 410 uint64_t r; 411 412 if (range == 0) 413 return (0); 414 415 if (read(ztest_random_fd, &r, sizeof (r)) != sizeof (r)) 416 fatal(1, "short read from /dev/urandom"); 417 418 return (r % range); 419 } 420 421 static void 422 ztest_record_enospc(char *s) 423 { 424 dprintf("ENOSPC doing: %s\n", s ? s : "<unknown>"); 425 ztest_shared->zs_enospc_count++; 426 } 427 428 static void 429 process_options(int argc, char **argv) 430 { 431 int opt; 432 uint64_t value; 433 434 /* By default, test gang blocks for blocks 32K and greater */ 435 zio_gang_bang = 32 << 10; 436 437 /* Default value, fail every 32nd allocation */ 438 zio_zil_fail_shift = 5; 439 440 while ((opt = getopt(argc, argv, 441 "v:s:a:m:r:R:d:t:g:i:k:p:f:VET:P:z:w:h")) != EOF) { 442 value = 0; 443 switch (opt) { 444 case 'v': 445 case 's': 446 case 'a': 447 case 'm': 448 case 'r': 449 case 'R': 450 case 'd': 451 case 't': 452 case 'g': 453 case 'i': 454 case 'k': 455 case 'T': 456 case 'P': 457 case 'z': 458 case 'w': 459 value = nicenumtoull(optarg); 460 } 461 switch (opt) { 462 case 'v': 463 zopt_vdevs = value; 464 break; 465 case 's': 466 zopt_vdev_size = MAX(SPA_MINDEVSIZE, value); 467 break; 468 case 'a': 469 zopt_ashift = value; 470 break; 471 case 'm': 472 zopt_mirrors = value; 473 break; 474 case 'r': 475 zopt_raidz = MAX(1, value); 476 break; 477 case 'R': 478 zopt_raidz_parity = MIN(MAX(value, 1), 2); 479 break; 480 case 'd': 481 zopt_datasets = MAX(1, value); 482 break; 483 case 't': 484 zopt_threads = MAX(1, value); 485 break; 486 case 'g': 487 zio_gang_bang = MAX(SPA_MINBLOCKSIZE << 1, value); 488 break; 489 case 'i': 490 zopt_init = value; 491 break; 492 case 'k': 493 zopt_killrate = value; 494 break; 495 case 'p': 496 zopt_pool = strdup(optarg); 497 break; 498 case 'f': 499 zopt_dir = strdup(optarg); 500 break; 501 case 'V': 502 zopt_verbose++; 503 break; 504 case 'E': 505 zopt_init = 0; 506 break; 507 case 'T': 508 zopt_time = value; 509 break; 510 case 'P': 511 zopt_passtime = MAX(1, value); 512 break; 513 case 'z': 514 zio_zil_fail_shift = MIN(value, 16); 515 break; 516 case 'w': 517 zopt_write_fail_shift = MIN(value, 16); 518 break; 519 case 'h': 520 usage(B_TRUE); 521 break; 522 case '?': 523 default: 524 usage(B_FALSE); 525 break; 526 } 527 } 528 529 zopt_raidz_parity = MIN(zopt_raidz_parity, zopt_raidz - 1); 530 531 zopt_vdevtime = (zopt_vdevs > 0 ? zopt_time / zopt_vdevs : UINT64_MAX); 532 zopt_maxfaults = MAX(zopt_mirrors, 1) * (zopt_raidz_parity + 1) - 1; 533 } 534 535 static uint64_t 536 ztest_get_ashift(void) 537 { 538 if (zopt_ashift == 0) 539 return (SPA_MINBLOCKSHIFT + ztest_random(3)); 540 return (zopt_ashift); 541 } 542 543 static nvlist_t * 544 make_vdev_file(size_t size) 545 { 546 char dev_name[MAXPATHLEN]; 547 uint64_t vdev; 548 uint64_t ashift = ztest_get_ashift(); 549 int fd; 550 nvlist_t *file; 551 552 if (size == 0) { 553 (void) snprintf(dev_name, sizeof (dev_name), "%s", 554 "/dev/bogus"); 555 } else { 556 vdev = ztest_shared->zs_vdev_primaries++; 557 (void) sprintf(dev_name, ztest_dev_template, 558 zopt_dir, zopt_pool, vdev); 559 560 fd = open(dev_name, O_RDWR | O_CREAT | O_TRUNC, 0666); 561 if (fd == -1) 562 fatal(1, "can't open %s", dev_name); 563 if (ftruncate(fd, size) != 0) 564 fatal(1, "can't ftruncate %s", dev_name); 565 (void) close(fd); 566 } 567 568 VERIFY(nvlist_alloc(&file, NV_UNIQUE_NAME, 0) == 0); 569 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_TYPE, VDEV_TYPE_FILE) == 0); 570 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_PATH, dev_name) == 0); 571 VERIFY(nvlist_add_uint64(file, ZPOOL_CONFIG_ASHIFT, ashift) == 0); 572 573 return (file); 574 } 575 576 static nvlist_t * 577 make_vdev_raidz(size_t size, int r) 578 { 579 nvlist_t *raidz, **child; 580 int c; 581 582 if (r < 2) 583 return (make_vdev_file(size)); 584 585 child = umem_alloc(r * sizeof (nvlist_t *), UMEM_NOFAIL); 586 587 for (c = 0; c < r; c++) 588 child[c] = make_vdev_file(size); 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(size_t size, int log, int r, int m) 608 { 609 nvlist_t *mirror, **child; 610 int c; 611 612 if (m < 1) 613 return (make_vdev_raidz(size, r)); 614 615 child = umem_alloc(m * sizeof (nvlist_t *), UMEM_NOFAIL); 616 617 for (c = 0; c < m; c++) 618 child[c] = make_vdev_raidz(size, r); 619 620 VERIFY(nvlist_alloc(&mirror, NV_UNIQUE_NAME, 0) == 0); 621 VERIFY(nvlist_add_string(mirror, ZPOOL_CONFIG_TYPE, 622 VDEV_TYPE_MIRROR) == 0); 623 VERIFY(nvlist_add_nvlist_array(mirror, ZPOOL_CONFIG_CHILDREN, 624 child, m) == 0); 625 VERIFY(nvlist_add_uint64(mirror, ZPOOL_CONFIG_IS_LOG, log) == 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(size_t size, int log, int r, int m, int t) 637 { 638 nvlist_t *root, **child; 639 int c; 640 641 ASSERT(t > 0); 642 643 child = umem_alloc(t * sizeof (nvlist_t *), UMEM_NOFAIL); 644 645 for (c = 0; c < t; c++) 646 child[c] = make_vdev_mirror(size, log, r, m); 647 648 VERIFY(nvlist_alloc(&root, NV_UNIQUE_NAME, 0) == 0); 649 VERIFY(nvlist_add_string(root, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) == 0); 650 VERIFY(nvlist_add_nvlist_array(root, ZPOOL_CONFIG_CHILDREN, 651 child, t) == 0); 652 653 for (c = 0; c < t; c++) 654 nvlist_free(child[c]); 655 656 umem_free(child, t * sizeof (nvlist_t *)); 657 658 return (root); 659 } 660 661 static void 662 ztest_set_random_blocksize(objset_t *os, uint64_t object, dmu_tx_t *tx) 663 { 664 int bs = SPA_MINBLOCKSHIFT + 665 ztest_random(SPA_MAXBLOCKSHIFT - SPA_MINBLOCKSHIFT + 1); 666 int ibs = DN_MIN_INDBLKSHIFT + 667 ztest_random(DN_MAX_INDBLKSHIFT - DN_MIN_INDBLKSHIFT + 1); 668 int error; 669 670 error = dmu_object_set_blocksize(os, object, 1ULL << bs, ibs, tx); 671 if (error) { 672 char osname[300]; 673 dmu_objset_name(os, osname); 674 fatal(0, "dmu_object_set_blocksize('%s', %llu, %d, %d) = %d", 675 osname, object, 1 << bs, ibs, error); 676 } 677 } 678 679 static uint8_t 680 ztest_random_checksum(void) 681 { 682 uint8_t checksum; 683 684 do { 685 checksum = ztest_random(ZIO_CHECKSUM_FUNCTIONS); 686 } while (zio_checksum_table[checksum].ci_zbt); 687 688 if (checksum == ZIO_CHECKSUM_OFF) 689 checksum = ZIO_CHECKSUM_ON; 690 691 return (checksum); 692 } 693 694 static uint8_t 695 ztest_random_compress(void) 696 { 697 return ((uint8_t)ztest_random(ZIO_COMPRESS_FUNCTIONS)); 698 } 699 700 typedef struct ztest_replay { 701 objset_t *zr_os; 702 uint64_t zr_assign; 703 } ztest_replay_t; 704 705 static int 706 ztest_replay_create(ztest_replay_t *zr, lr_create_t *lr, boolean_t byteswap) 707 { 708 objset_t *os = zr->zr_os; 709 dmu_tx_t *tx; 710 int error; 711 712 if (byteswap) 713 byteswap_uint64_array(lr, sizeof (*lr)); 714 715 tx = dmu_tx_create(os); 716 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT); 717 error = dmu_tx_assign(tx, zr->zr_assign); 718 if (error) { 719 dmu_tx_abort(tx); 720 return (error); 721 } 722 723 error = dmu_object_claim(os, lr->lr_doid, lr->lr_mode, 0, 724 DMU_OT_NONE, 0, tx); 725 ASSERT3U(error, ==, 0); 726 dmu_tx_commit(tx); 727 728 if (zopt_verbose >= 5) { 729 char osname[MAXNAMELEN]; 730 dmu_objset_name(os, osname); 731 (void) printf("replay create of %s object %llu" 732 " in txg %llu = %d\n", 733 osname, (u_longlong_t)lr->lr_doid, 734 (u_longlong_t)zr->zr_assign, error); 735 } 736 737 return (error); 738 } 739 740 static int 741 ztest_replay_remove(ztest_replay_t *zr, lr_remove_t *lr, boolean_t byteswap) 742 { 743 objset_t *os = zr->zr_os; 744 dmu_tx_t *tx; 745 int error; 746 747 if (byteswap) 748 byteswap_uint64_array(lr, sizeof (*lr)); 749 750 tx = dmu_tx_create(os); 751 dmu_tx_hold_free(tx, lr->lr_doid, 0, DMU_OBJECT_END); 752 error = dmu_tx_assign(tx, zr->zr_assign); 753 if (error) { 754 dmu_tx_abort(tx); 755 return (error); 756 } 757 758 error = dmu_object_free(os, lr->lr_doid, tx); 759 dmu_tx_commit(tx); 760 761 return (error); 762 } 763 764 zil_replay_func_t *ztest_replay_vector[TX_MAX_TYPE] = { 765 NULL, /* 0 no such transaction type */ 766 ztest_replay_create, /* TX_CREATE */ 767 NULL, /* TX_MKDIR */ 768 NULL, /* TX_MKXATTR */ 769 NULL, /* TX_SYMLINK */ 770 ztest_replay_remove, /* TX_REMOVE */ 771 NULL, /* TX_RMDIR */ 772 NULL, /* TX_LINK */ 773 NULL, /* TX_RENAME */ 774 NULL, /* TX_WRITE */ 775 NULL, /* TX_TRUNCATE */ 776 NULL, /* TX_SETATTR */ 777 NULL, /* TX_ACL */ 778 }; 779 780 /* 781 * Verify that we can't destroy an active pool, create an existing pool, 782 * or create a pool with a bad vdev spec. 783 */ 784 void 785 ztest_spa_create_destroy(ztest_args_t *za) 786 { 787 int error; 788 spa_t *spa; 789 nvlist_t *nvroot; 790 791 /* 792 * Attempt to create using a bad file. 793 */ 794 nvroot = make_vdev_root(0, 0, 0, 0, 1); 795 error = spa_create("ztest_bad_file", nvroot, NULL, NULL); 796 nvlist_free(nvroot); 797 if (error != ENOENT) 798 fatal(0, "spa_create(bad_file) = %d", error); 799 800 /* 801 * Attempt to create using a bad mirror. 802 */ 803 nvroot = make_vdev_root(0, 0, 0, 2, 1); 804 error = spa_create("ztest_bad_mirror", nvroot, NULL, NULL); 805 nvlist_free(nvroot); 806 if (error != ENOENT) 807 fatal(0, "spa_create(bad_mirror) = %d", error); 808 809 /* 810 * Attempt to create an existing pool. It shouldn't matter 811 * what's in the nvroot; we should fail with EEXIST. 812 */ 813 (void) rw_rdlock(&ztest_shared->zs_name_lock); 814 nvroot = make_vdev_root(0, 0, 0, 0, 1); 815 error = spa_create(za->za_pool, nvroot, NULL, NULL); 816 nvlist_free(nvroot); 817 if (error != EEXIST) 818 fatal(0, "spa_create(whatever) = %d", error); 819 820 error = spa_open(za->za_pool, &spa, FTAG); 821 if (error) 822 fatal(0, "spa_open() = %d", error); 823 824 error = spa_destroy(za->za_pool); 825 if (error != EBUSY) 826 fatal(0, "spa_destroy() = %d", error); 827 828 spa_close(spa, FTAG); 829 (void) rw_unlock(&ztest_shared->zs_name_lock); 830 } 831 832 /* 833 * Verify that vdev_add() works as expected. 834 */ 835 void 836 ztest_vdev_add_remove(ztest_args_t *za) 837 { 838 spa_t *spa = dmu_objset_spa(za->za_os); 839 uint64_t leaves = MAX(zopt_mirrors, 1) * zopt_raidz; 840 nvlist_t *nvroot; 841 int error; 842 843 if (zopt_verbose >= 6) 844 (void) printf("adding vdev\n"); 845 846 (void) mutex_lock(&ztest_shared->zs_vdev_lock); 847 848 spa_config_enter(spa, RW_READER, FTAG); 849 850 ztest_shared->zs_vdev_primaries = 851 spa->spa_root_vdev->vdev_children * leaves; 852 853 spa_config_exit(spa, FTAG); 854 855 /* 856 * Make 1/4 of the devices be log devices. 857 */ 858 nvroot = make_vdev_root(zopt_vdev_size, 859 ztest_random(4) == 0, zopt_raidz, zopt_mirrors, 1); 860 861 error = spa_vdev_add(spa, nvroot); 862 nvlist_free(nvroot); 863 864 (void) mutex_unlock(&ztest_shared->zs_vdev_lock); 865 866 if (error == ENOSPC) 867 ztest_record_enospc("spa_vdev_add"); 868 else if (error != 0) 869 fatal(0, "spa_vdev_add() = %d", error); 870 871 if (zopt_verbose >= 6) 872 (void) printf("spa_vdev_add = %d, as expected\n", error); 873 } 874 875 static vdev_t * 876 vdev_lookup_by_path(vdev_t *vd, const char *path) 877 { 878 int c; 879 vdev_t *mvd; 880 881 if (vd->vdev_path != NULL) { 882 if (vd->vdev_wholedisk == 1) { 883 /* 884 * For whole disks, the internal path has 's0', but the 885 * path passed in by the user doesn't. 886 */ 887 if (strlen(path) == strlen(vd->vdev_path) - 2 && 888 strncmp(path, vd->vdev_path, strlen(path)) == 0) 889 return (vd); 890 } else if (strcmp(path, vd->vdev_path) == 0) { 891 return (vd); 892 } 893 } 894 895 for (c = 0; c < vd->vdev_children; c++) 896 if ((mvd = vdev_lookup_by_path(vd->vdev_child[c], path)) != 897 NULL) 898 return (mvd); 899 900 return (NULL); 901 } 902 903 /* 904 * Verify that we can attach and detach devices. 905 */ 906 void 907 ztest_vdev_attach_detach(ztest_args_t *za) 908 { 909 spa_t *spa = dmu_objset_spa(za->za_os); 910 vdev_t *rvd = spa->spa_root_vdev; 911 vdev_t *oldvd, *newvd, *pvd; 912 nvlist_t *root, *file; 913 uint64_t leaves = MAX(zopt_mirrors, 1) * zopt_raidz; 914 uint64_t leaf, top; 915 uint64_t ashift = ztest_get_ashift(); 916 size_t oldsize, newsize; 917 char oldpath[MAXPATHLEN], newpath[MAXPATHLEN]; 918 int replacing; 919 int error, expected_error; 920 int fd; 921 922 (void) mutex_lock(&ztest_shared->zs_vdev_lock); 923 924 spa_config_enter(spa, RW_READER, FTAG); 925 926 /* 927 * Decide whether to do an attach or a replace. 928 */ 929 replacing = ztest_random(2); 930 931 /* 932 * Pick a random top-level vdev. 933 */ 934 top = ztest_random(rvd->vdev_children); 935 936 /* 937 * Pick a random leaf within it. 938 */ 939 leaf = ztest_random(leaves); 940 941 /* 942 * Generate the path to this leaf. The filename will end with 'a'. 943 * We'll alternate replacements with a filename that ends with 'b'. 944 */ 945 (void) snprintf(oldpath, sizeof (oldpath), 946 ztest_dev_template, zopt_dir, zopt_pool, top * leaves + leaf); 947 948 bcopy(oldpath, newpath, MAXPATHLEN); 949 950 /* 951 * If the 'a' file isn't part of the pool, the 'b' file must be. 952 */ 953 if (vdev_lookup_by_path(rvd, oldpath) == NULL) 954 oldpath[strlen(oldpath) - 1] = 'b'; 955 else 956 newpath[strlen(newpath) - 1] = 'b'; 957 958 /* 959 * Now oldpath represents something that's already in the pool, 960 * and newpath is the thing we'll try to attach. 961 */ 962 oldvd = vdev_lookup_by_path(rvd, oldpath); 963 newvd = vdev_lookup_by_path(rvd, newpath); 964 ASSERT(oldvd != NULL); 965 pvd = oldvd->vdev_parent; 966 967 /* 968 * Make newsize a little bigger or smaller than oldsize. 969 * If it's smaller, the attach should fail. 970 * If it's larger, and we're doing a replace, 971 * we should get dynamic LUN growth when we're done. 972 */ 973 oldsize = vdev_get_rsize(oldvd); 974 newsize = 10 * oldsize / (9 + ztest_random(3)); 975 976 /* 977 * If pvd is not a mirror or root, the attach should fail with ENOTSUP, 978 * unless it's a replace; in that case any non-replacing parent is OK. 979 * 980 * If newvd is already part of the pool, it should fail with EBUSY. 981 * 982 * If newvd is too small, it should fail with EOVERFLOW. 983 */ 984 if (newvd != NULL) 985 expected_error = EBUSY; 986 else if (pvd->vdev_ops != &vdev_mirror_ops && 987 pvd->vdev_ops != &vdev_root_ops && 988 (!replacing || pvd->vdev_ops == &vdev_replacing_ops)) 989 expected_error = ENOTSUP; 990 else if (newsize < oldsize) 991 expected_error = EOVERFLOW; 992 else if (ashift > oldvd->vdev_top->vdev_ashift) 993 expected_error = EDOM; 994 else 995 expected_error = 0; 996 997 /* 998 * If newvd isn't already part of the pool, create it. 999 */ 1000 if (newvd == NULL) { 1001 fd = open(newpath, O_RDWR | O_CREAT | O_TRUNC, 0666); 1002 if (fd == -1) 1003 fatal(1, "can't open %s", newpath); 1004 if (ftruncate(fd, newsize) != 0) 1005 fatal(1, "can't ftruncate %s", newpath); 1006 (void) close(fd); 1007 } 1008 1009 spa_config_exit(spa, FTAG); 1010 1011 /* 1012 * Build the nvlist describing newpath. 1013 */ 1014 VERIFY(nvlist_alloc(&file, NV_UNIQUE_NAME, 0) == 0); 1015 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_TYPE, VDEV_TYPE_FILE) == 0); 1016 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_PATH, newpath) == 0); 1017 VERIFY(nvlist_add_uint64(file, ZPOOL_CONFIG_ASHIFT, ashift) == 0); 1018 1019 VERIFY(nvlist_alloc(&root, NV_UNIQUE_NAME, 0) == 0); 1020 VERIFY(nvlist_add_string(root, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) == 0); 1021 VERIFY(nvlist_add_nvlist_array(root, ZPOOL_CONFIG_CHILDREN, 1022 &file, 1) == 0); 1023 1024 error = spa_vdev_attach(spa, oldvd->vdev_guid, root, replacing); 1025 1026 nvlist_free(file); 1027 nvlist_free(root); 1028 1029 /* 1030 * If our parent was the replacing vdev, but the replace completed, 1031 * then instead of failing with ENOTSUP we may either succeed, 1032 * fail with ENODEV, or fail with EOVERFLOW. 1033 */ 1034 if (expected_error == ENOTSUP && 1035 (error == 0 || error == ENODEV || error == EOVERFLOW)) 1036 expected_error = error; 1037 1038 /* 1039 * If someone grew the LUN, the replacement may be too small. 1040 */ 1041 if (error == EOVERFLOW) 1042 expected_error = error; 1043 1044 if (error != expected_error) { 1045 fatal(0, "attach (%s, %s, %d) returned %d, expected %d", 1046 oldpath, newpath, replacing, error, expected_error); 1047 } 1048 1049 (void) mutex_unlock(&ztest_shared->zs_vdev_lock); 1050 } 1051 1052 /* 1053 * Verify that dynamic LUN growth works as expected. 1054 */ 1055 /* ARGSUSED */ 1056 void 1057 ztest_vdev_LUN_growth(ztest_args_t *za) 1058 { 1059 spa_t *spa = dmu_objset_spa(za->za_os); 1060 char dev_name[MAXPATHLEN]; 1061 uint64_t leaves = MAX(zopt_mirrors, 1) * zopt_raidz; 1062 uint64_t vdev; 1063 size_t fsize; 1064 int fd; 1065 1066 (void) mutex_lock(&ztest_shared->zs_vdev_lock); 1067 1068 /* 1069 * Pick a random leaf vdev. 1070 */ 1071 spa_config_enter(spa, RW_READER, FTAG); 1072 vdev = ztest_random(spa->spa_root_vdev->vdev_children * leaves); 1073 spa_config_exit(spa, FTAG); 1074 1075 (void) sprintf(dev_name, ztest_dev_template, zopt_dir, zopt_pool, vdev); 1076 1077 if ((fd = open(dev_name, O_RDWR)) != -1) { 1078 /* 1079 * Determine the size. 1080 */ 1081 fsize = lseek(fd, 0, SEEK_END); 1082 1083 /* 1084 * If it's less than 2x the original size, grow by around 3%. 1085 */ 1086 if (fsize < 2 * zopt_vdev_size) { 1087 size_t newsize = fsize + ztest_random(fsize / 32); 1088 (void) ftruncate(fd, newsize); 1089 if (zopt_verbose >= 6) { 1090 (void) printf("%s grew from %lu to %lu bytes\n", 1091 dev_name, (ulong_t)fsize, (ulong_t)newsize); 1092 } 1093 } 1094 (void) close(fd); 1095 } 1096 1097 (void) mutex_unlock(&ztest_shared->zs_vdev_lock); 1098 } 1099 1100 /* ARGSUSED */ 1101 static void 1102 ztest_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx) 1103 { 1104 /* 1105 * Create the directory object. 1106 */ 1107 VERIFY(dmu_object_claim(os, ZTEST_DIROBJ, 1108 DMU_OT_UINT64_OTHER, ZTEST_DIROBJ_BLOCKSIZE, 1109 DMU_OT_UINT64_OTHER, sizeof (ztest_block_tag_t), tx) == 0); 1110 1111 VERIFY(zap_create_claim(os, ZTEST_MICROZAP_OBJ, 1112 DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx) == 0); 1113 1114 VERIFY(zap_create_claim(os, ZTEST_FATZAP_OBJ, 1115 DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx) == 0); 1116 } 1117 1118 /* ARGSUSED */ 1119 static int 1120 ztest_destroy_cb(char *name, void *arg) 1121 { 1122 objset_t *os; 1123 dmu_object_info_t doi; 1124 int error; 1125 1126 /* 1127 * Verify that the dataset contains a directory object. 1128 */ 1129 error = dmu_objset_open(name, DMU_OST_OTHER, 1130 DS_MODE_STANDARD | DS_MODE_READONLY, &os); 1131 ASSERT3U(error, ==, 0); 1132 error = dmu_object_info(os, ZTEST_DIROBJ, &doi); 1133 if (error != ENOENT) { 1134 /* We could have crashed in the middle of destroying it */ 1135 ASSERT3U(error, ==, 0); 1136 ASSERT3U(doi.doi_type, ==, DMU_OT_UINT64_OTHER); 1137 ASSERT3S(doi.doi_physical_blks, >=, 0); 1138 } 1139 dmu_objset_close(os); 1140 1141 /* 1142 * Destroy the dataset. 1143 */ 1144 error = dmu_objset_destroy(name); 1145 ASSERT3U(error, ==, 0); 1146 return (0); 1147 } 1148 1149 /* 1150 * Verify that dmu_objset_{create,destroy,open,close} work as expected. 1151 */ 1152 static uint64_t 1153 ztest_log_create(zilog_t *zilog, dmu_tx_t *tx, uint64_t object, int mode) 1154 { 1155 itx_t *itx; 1156 lr_create_t *lr; 1157 size_t namesize; 1158 char name[24]; 1159 1160 (void) sprintf(name, "ZOBJ_%llu", (u_longlong_t)object); 1161 namesize = strlen(name) + 1; 1162 1163 itx = zil_itx_create(TX_CREATE, sizeof (*lr) + namesize + 1164 ztest_random(ZIL_MAX_BLKSZ)); 1165 lr = (lr_create_t *)&itx->itx_lr; 1166 bzero(lr + 1, lr->lr_common.lrc_reclen - sizeof (*lr)); 1167 lr->lr_doid = object; 1168 lr->lr_foid = 0; 1169 lr->lr_mode = mode; 1170 lr->lr_uid = 0; 1171 lr->lr_gid = 0; 1172 lr->lr_gen = dmu_tx_get_txg(tx); 1173 lr->lr_crtime[0] = time(NULL); 1174 lr->lr_crtime[1] = 0; 1175 lr->lr_rdev = 0; 1176 bcopy(name, (char *)(lr + 1), namesize); 1177 1178 return (zil_itx_assign(zilog, itx, tx)); 1179 } 1180 1181 void 1182 ztest_dmu_objset_create_destroy(ztest_args_t *za) 1183 { 1184 int error; 1185 objset_t *os; 1186 char name[100]; 1187 int mode, basemode, expected_error; 1188 zilog_t *zilog; 1189 uint64_t seq; 1190 uint64_t objects; 1191 ztest_replay_t zr; 1192 1193 (void) rw_rdlock(&ztest_shared->zs_name_lock); 1194 (void) snprintf(name, 100, "%s/%s_temp_%llu", za->za_pool, za->za_pool, 1195 (u_longlong_t)za->za_instance); 1196 1197 basemode = DS_MODE_LEVEL(za->za_instance); 1198 if (basemode == DS_MODE_NONE) 1199 basemode++; 1200 1201 /* 1202 * If this dataset exists from a previous run, process its replay log 1203 * half of the time. If we don't replay it, then dmu_objset_destroy() 1204 * (invoked from ztest_destroy_cb() below) should just throw it away. 1205 */ 1206 if (ztest_random(2) == 0 && 1207 dmu_objset_open(name, DMU_OST_OTHER, DS_MODE_PRIMARY, &os) == 0) { 1208 zr.zr_os = os; 1209 zil_replay(os, &zr, &zr.zr_assign, ztest_replay_vector); 1210 dmu_objset_close(os); 1211 } 1212 1213 /* 1214 * There may be an old instance of the dataset we're about to 1215 * create lying around from a previous run. If so, destroy it 1216 * and all of its snapshots. 1217 */ 1218 (void) dmu_objset_find(name, ztest_destroy_cb, NULL, 1219 DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS); 1220 1221 /* 1222 * Verify that the destroyed dataset is no longer in the namespace. 1223 */ 1224 error = dmu_objset_open(name, DMU_OST_OTHER, basemode, &os); 1225 if (error != ENOENT) 1226 fatal(1, "dmu_objset_open(%s) found destroyed dataset %p", 1227 name, os); 1228 1229 /* 1230 * Verify that we can create a new dataset. 1231 */ 1232 error = dmu_objset_create(name, DMU_OST_OTHER, NULL, ztest_create_cb, 1233 NULL); 1234 if (error) { 1235 if (error == ENOSPC) { 1236 ztest_record_enospc("dmu_objset_create"); 1237 (void) rw_unlock(&ztest_shared->zs_name_lock); 1238 return; 1239 } 1240 fatal(0, "dmu_objset_create(%s) = %d", name, error); 1241 } 1242 1243 error = dmu_objset_open(name, DMU_OST_OTHER, basemode, &os); 1244 if (error) { 1245 fatal(0, "dmu_objset_open(%s) = %d", name, error); 1246 } 1247 1248 /* 1249 * Open the intent log for it. 1250 */ 1251 zilog = zil_open(os, NULL); 1252 1253 /* 1254 * Put a random number of objects in there. 1255 */ 1256 objects = ztest_random(20); 1257 seq = 0; 1258 while (objects-- != 0) { 1259 uint64_t object; 1260 dmu_tx_t *tx = dmu_tx_create(os); 1261 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, sizeof (name)); 1262 error = dmu_tx_assign(tx, TXG_WAIT); 1263 if (error) { 1264 dmu_tx_abort(tx); 1265 } else { 1266 object = dmu_object_alloc(os, DMU_OT_UINT64_OTHER, 0, 1267 DMU_OT_NONE, 0, tx); 1268 ztest_set_random_blocksize(os, object, tx); 1269 seq = ztest_log_create(zilog, tx, object, 1270 DMU_OT_UINT64_OTHER); 1271 dmu_write(os, object, 0, sizeof (name), name, tx); 1272 dmu_tx_commit(tx); 1273 } 1274 if (ztest_random(5) == 0) { 1275 zil_commit(zilog, seq, object); 1276 } 1277 if (ztest_random(100) == 0) { 1278 error = zil_suspend(zilog); 1279 if (error == 0) { 1280 zil_resume(zilog); 1281 } 1282 } 1283 } 1284 1285 /* 1286 * Verify that we cannot create an existing dataset. 1287 */ 1288 error = dmu_objset_create(name, DMU_OST_OTHER, NULL, NULL, NULL); 1289 if (error != EEXIST) 1290 fatal(0, "created existing dataset, error = %d", error); 1291 1292 /* 1293 * Verify that multiple dataset opens are allowed, but only when 1294 * the new access mode is compatible with the base mode. 1295 * We use a mixture of typed and typeless opens, and when the 1296 * open succeeds, verify that the discovered type is correct. 1297 */ 1298 for (mode = DS_MODE_STANDARD; mode < DS_MODE_LEVELS; mode++) { 1299 objset_t *os2; 1300 error = dmu_objset_open(name, DMU_OST_OTHER, mode, &os2); 1301 expected_error = (basemode + mode < DS_MODE_LEVELS) ? 0 : EBUSY; 1302 if (error != expected_error) 1303 fatal(0, "dmu_objset_open('%s') = %d, expected %d", 1304 name, error, expected_error); 1305 if (error == 0) 1306 dmu_objset_close(os2); 1307 } 1308 1309 zil_close(zilog); 1310 dmu_objset_close(os); 1311 1312 error = dmu_objset_destroy(name); 1313 if (error) 1314 fatal(0, "dmu_objset_destroy(%s) = %d", name, error); 1315 1316 (void) rw_unlock(&ztest_shared->zs_name_lock); 1317 } 1318 1319 /* 1320 * Verify that dmu_snapshot_{create,destroy,open,close} work as expected. 1321 */ 1322 void 1323 ztest_dmu_snapshot_create_destroy(ztest_args_t *za) 1324 { 1325 int error; 1326 objset_t *os = za->za_os; 1327 char snapname[100]; 1328 char osname[MAXNAMELEN]; 1329 1330 (void) rw_rdlock(&ztest_shared->zs_name_lock); 1331 dmu_objset_name(os, osname); 1332 (void) snprintf(snapname, 100, "%s@%llu", osname, 1333 (u_longlong_t)za->za_instance); 1334 1335 error = dmu_objset_destroy(snapname); 1336 if (error != 0 && error != ENOENT) 1337 fatal(0, "dmu_objset_destroy() = %d", error); 1338 error = dmu_objset_snapshot(osname, strchr(snapname, '@')+1, FALSE); 1339 if (error == ENOSPC) 1340 ztest_record_enospc("dmu_take_snapshot"); 1341 else if (error != 0 && error != EEXIST) 1342 fatal(0, "dmu_take_snapshot() = %d", error); 1343 (void) rw_unlock(&ztest_shared->zs_name_lock); 1344 } 1345 1346 #define ZTEST_TRAVERSE_BLOCKS 1000 1347 1348 static int 1349 ztest_blk_cb(traverse_blk_cache_t *bc, spa_t *spa, void *arg) 1350 { 1351 ztest_args_t *za = arg; 1352 zbookmark_t *zb = &bc->bc_bookmark; 1353 blkptr_t *bp = &bc->bc_blkptr; 1354 dnode_phys_t *dnp = bc->bc_dnode; 1355 traverse_handle_t *th = za->za_th; 1356 uint64_t size = BP_GET_LSIZE(bp); 1357 1358 /* 1359 * Level -1 indicates the objset_phys_t or something in its intent log. 1360 */ 1361 if (zb->zb_level == -1) { 1362 if (BP_GET_TYPE(bp) == DMU_OT_OBJSET) { 1363 ASSERT3U(zb->zb_object, ==, 0); 1364 ASSERT3U(zb->zb_blkid, ==, 0); 1365 ASSERT3U(size, ==, sizeof (objset_phys_t)); 1366 za->za_zil_seq = 0; 1367 } else if (BP_GET_TYPE(bp) == DMU_OT_INTENT_LOG) { 1368 ASSERT3U(zb->zb_object, ==, 0); 1369 ASSERT3U(zb->zb_blkid, >, za->za_zil_seq); 1370 za->za_zil_seq = zb->zb_blkid; 1371 } else { 1372 ASSERT3U(zb->zb_object, !=, 0); /* lr_write_t */ 1373 } 1374 1375 return (0); 1376 } 1377 1378 ASSERT(dnp != NULL); 1379 1380 if (bc->bc_errno) 1381 return (ERESTART); 1382 1383 /* 1384 * Once in a while, abort the traverse. We only do this to odd 1385 * instance numbers to ensure that even ones can run to completion. 1386 */ 1387 if ((za->za_instance & 1) && ztest_random(10000) == 0) 1388 return (EINTR); 1389 1390 if (bp->blk_birth == 0) { 1391 ASSERT(th->th_advance & ADVANCE_HOLES); 1392 return (0); 1393 } 1394 1395 if (zb->zb_level == 0 && !(th->th_advance & ADVANCE_DATA) && 1396 bc == &th->th_cache[ZB_DN_CACHE][0]) { 1397 ASSERT(bc->bc_data == NULL); 1398 return (0); 1399 } 1400 1401 ASSERT(bc->bc_data != NULL); 1402 1403 /* 1404 * This is an expensive question, so don't ask it too often. 1405 */ 1406 if (((za->za_random ^ th->th_callbacks) & 0xff) == 0) { 1407 void *xbuf = umem_alloc(size, UMEM_NOFAIL); 1408 if (arc_tryread(spa, bp, xbuf) == 0) { 1409 ASSERT(bcmp(bc->bc_data, xbuf, size) == 0); 1410 } 1411 umem_free(xbuf, size); 1412 } 1413 1414 if (zb->zb_level > 0) { 1415 ASSERT3U(size, ==, 1ULL << dnp->dn_indblkshift); 1416 return (0); 1417 } 1418 1419 ASSERT(zb->zb_level == 0); 1420 ASSERT3U(size, ==, dnp->dn_datablkszsec << DEV_BSHIFT); 1421 1422 return (0); 1423 } 1424 1425 /* 1426 * Verify that live pool traversal works. 1427 */ 1428 void 1429 ztest_traverse(ztest_args_t *za) 1430 { 1431 spa_t *spa = dmu_objset_spa(za->za_os); 1432 traverse_handle_t *th = za->za_th; 1433 int rc, advance; 1434 uint64_t cbstart, cblimit; 1435 1436 if (th == NULL) { 1437 advance = 0; 1438 1439 if (ztest_random(2) == 0) 1440 advance |= ADVANCE_PRE; 1441 1442 if (ztest_random(2) == 0) 1443 advance |= ADVANCE_PRUNE; 1444 1445 if (ztest_random(2) == 0) 1446 advance |= ADVANCE_DATA; 1447 1448 if (ztest_random(2) == 0) 1449 advance |= ADVANCE_HOLES; 1450 1451 if (ztest_random(2) == 0) 1452 advance |= ADVANCE_ZIL; 1453 1454 th = za->za_th = traverse_init(spa, ztest_blk_cb, za, advance, 1455 ZIO_FLAG_CANFAIL); 1456 1457 traverse_add_pool(th, 0, -1ULL); 1458 } 1459 1460 advance = th->th_advance; 1461 cbstart = th->th_callbacks; 1462 cblimit = cbstart + ((advance & ADVANCE_DATA) ? 100 : 1000); 1463 1464 while ((rc = traverse_more(th)) == EAGAIN && th->th_callbacks < cblimit) 1465 continue; 1466 1467 if (zopt_verbose >= 5) 1468 (void) printf("traverse %s%s%s%s %llu blocks to " 1469 "<%llu, %llu, %lld, %llx>%s\n", 1470 (advance & ADVANCE_PRE) ? "pre" : "post", 1471 (advance & ADVANCE_PRUNE) ? "|prune" : "", 1472 (advance & ADVANCE_DATA) ? "|data" : "", 1473 (advance & ADVANCE_HOLES) ? "|holes" : "", 1474 (u_longlong_t)(th->th_callbacks - cbstart), 1475 (u_longlong_t)th->th_lastcb.zb_objset, 1476 (u_longlong_t)th->th_lastcb.zb_object, 1477 (u_longlong_t)th->th_lastcb.zb_level, 1478 (u_longlong_t)th->th_lastcb.zb_blkid, 1479 rc == 0 ? " [done]" : 1480 rc == EINTR ? " [aborted]" : 1481 rc == EAGAIN ? "" : 1482 strerror(rc)); 1483 1484 if (rc != EAGAIN) { 1485 if (rc != 0 && rc != EINTR) 1486 fatal(0, "traverse_more(%p) = %d", th, rc); 1487 traverse_fini(th); 1488 za->za_th = NULL; 1489 } 1490 } 1491 1492 /* 1493 * Verify that dmu_object_{alloc,free} work as expected. 1494 */ 1495 void 1496 ztest_dmu_object_alloc_free(ztest_args_t *za) 1497 { 1498 objset_t *os = za->za_os; 1499 dmu_buf_t *db; 1500 dmu_tx_t *tx; 1501 uint64_t batchobj, object, batchsize, endoff, temp; 1502 int b, c, error, bonuslen; 1503 dmu_object_info_t doi; 1504 char osname[MAXNAMELEN]; 1505 1506 dmu_objset_name(os, osname); 1507 1508 endoff = -8ULL; 1509 batchsize = 2; 1510 1511 /* 1512 * Create a batch object if necessary, and record it in the directory. 1513 */ 1514 VERIFY(0 == dmu_read(os, ZTEST_DIROBJ, za->za_diroff, 1515 sizeof (uint64_t), &batchobj)); 1516 if (batchobj == 0) { 1517 tx = dmu_tx_create(os); 1518 dmu_tx_hold_write(tx, ZTEST_DIROBJ, za->za_diroff, 1519 sizeof (uint64_t)); 1520 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT); 1521 error = dmu_tx_assign(tx, TXG_WAIT); 1522 if (error) { 1523 ztest_record_enospc("create a batch object"); 1524 dmu_tx_abort(tx); 1525 return; 1526 } 1527 batchobj = dmu_object_alloc(os, DMU_OT_UINT64_OTHER, 0, 1528 DMU_OT_NONE, 0, tx); 1529 ztest_set_random_blocksize(os, batchobj, tx); 1530 dmu_write(os, ZTEST_DIROBJ, za->za_diroff, 1531 sizeof (uint64_t), &batchobj, tx); 1532 dmu_tx_commit(tx); 1533 } 1534 1535 /* 1536 * Destroy the previous batch of objects. 1537 */ 1538 for (b = 0; b < batchsize; b++) { 1539 VERIFY(0 == dmu_read(os, batchobj, b * sizeof (uint64_t), 1540 sizeof (uint64_t), &object)); 1541 if (object == 0) 1542 continue; 1543 /* 1544 * Read and validate contents. 1545 * We expect the nth byte of the bonus buffer to be n. 1546 */ 1547 VERIFY(0 == dmu_bonus_hold(os, object, FTAG, &db)); 1548 1549 dmu_object_info_from_db(db, &doi); 1550 ASSERT(doi.doi_type == DMU_OT_UINT64_OTHER); 1551 ASSERT(doi.doi_bonus_type == DMU_OT_PLAIN_OTHER); 1552 ASSERT3S(doi.doi_physical_blks, >=, 0); 1553 1554 bonuslen = doi.doi_bonus_size; 1555 1556 for (c = 0; c < bonuslen; c++) { 1557 if (((uint8_t *)db->db_data)[c] != 1558 (uint8_t)(c + bonuslen)) { 1559 fatal(0, 1560 "bad bonus: %s, obj %llu, off %d: %u != %u", 1561 osname, object, c, 1562 ((uint8_t *)db->db_data)[c], 1563 (uint8_t)(c + bonuslen)); 1564 } 1565 } 1566 1567 dmu_buf_rele(db, FTAG); 1568 1569 /* 1570 * We expect the word at endoff to be our object number. 1571 */ 1572 VERIFY(0 == dmu_read(os, object, endoff, 1573 sizeof (uint64_t), &temp)); 1574 1575 if (temp != object) { 1576 fatal(0, "bad data in %s, got %llu, expected %llu", 1577 osname, temp, object); 1578 } 1579 1580 /* 1581 * Destroy old object and clear batch entry. 1582 */ 1583 tx = dmu_tx_create(os); 1584 dmu_tx_hold_write(tx, batchobj, 1585 b * sizeof (uint64_t), sizeof (uint64_t)); 1586 dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END); 1587 error = dmu_tx_assign(tx, TXG_WAIT); 1588 if (error) { 1589 ztest_record_enospc("free object"); 1590 dmu_tx_abort(tx); 1591 return; 1592 } 1593 error = dmu_object_free(os, object, tx); 1594 if (error) { 1595 fatal(0, "dmu_object_free('%s', %llu) = %d", 1596 osname, object, error); 1597 } 1598 object = 0; 1599 1600 dmu_object_set_checksum(os, batchobj, 1601 ztest_random_checksum(), tx); 1602 dmu_object_set_compress(os, batchobj, 1603 ztest_random_compress(), tx); 1604 1605 dmu_write(os, batchobj, b * sizeof (uint64_t), 1606 sizeof (uint64_t), &object, tx); 1607 1608 dmu_tx_commit(tx); 1609 } 1610 1611 /* 1612 * Before creating the new batch of objects, generate a bunch of churn. 1613 */ 1614 for (b = ztest_random(100); b > 0; b--) { 1615 tx = dmu_tx_create(os); 1616 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT); 1617 error = dmu_tx_assign(tx, TXG_WAIT); 1618 if (error) { 1619 ztest_record_enospc("churn objects"); 1620 dmu_tx_abort(tx); 1621 return; 1622 } 1623 object = dmu_object_alloc(os, DMU_OT_UINT64_OTHER, 0, 1624 DMU_OT_NONE, 0, tx); 1625 ztest_set_random_blocksize(os, object, tx); 1626 error = dmu_object_free(os, object, tx); 1627 if (error) { 1628 fatal(0, "dmu_object_free('%s', %llu) = %d", 1629 osname, object, error); 1630 } 1631 dmu_tx_commit(tx); 1632 } 1633 1634 /* 1635 * Create a new batch of objects with randomly chosen 1636 * blocksizes and record them in the batch directory. 1637 */ 1638 for (b = 0; b < batchsize; b++) { 1639 uint32_t va_blksize; 1640 u_longlong_t va_nblocks; 1641 1642 tx = dmu_tx_create(os); 1643 dmu_tx_hold_write(tx, batchobj, b * sizeof (uint64_t), 1644 sizeof (uint64_t)); 1645 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT); 1646 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, endoff, 1647 sizeof (uint64_t)); 1648 error = dmu_tx_assign(tx, TXG_WAIT); 1649 if (error) { 1650 ztest_record_enospc("create batchobj"); 1651 dmu_tx_abort(tx); 1652 return; 1653 } 1654 bonuslen = (int)ztest_random(dmu_bonus_max()) + 1; 1655 1656 object = dmu_object_alloc(os, DMU_OT_UINT64_OTHER, 0, 1657 DMU_OT_PLAIN_OTHER, bonuslen, tx); 1658 1659 ztest_set_random_blocksize(os, object, tx); 1660 1661 dmu_object_set_checksum(os, object, 1662 ztest_random_checksum(), tx); 1663 dmu_object_set_compress(os, object, 1664 ztest_random_compress(), tx); 1665 1666 dmu_write(os, batchobj, b * sizeof (uint64_t), 1667 sizeof (uint64_t), &object, tx); 1668 1669 /* 1670 * Write to both the bonus buffer and the regular data. 1671 */ 1672 VERIFY(0 == dmu_bonus_hold(os, object, FTAG, &db)); 1673 ASSERT3U(bonuslen, <=, db->db_size); 1674 1675 dmu_object_size_from_db(db, &va_blksize, &va_nblocks); 1676 ASSERT3S(va_nblocks, >=, 0); 1677 1678 dmu_buf_will_dirty(db, tx); 1679 1680 /* 1681 * See comments above regarding the contents of 1682 * the bonus buffer and the word at endoff. 1683 */ 1684 for (c = 0; c < bonuslen; c++) 1685 ((uint8_t *)db->db_data)[c] = (uint8_t)(c + bonuslen); 1686 1687 dmu_buf_rele(db, FTAG); 1688 1689 /* 1690 * Write to a large offset to increase indirection. 1691 */ 1692 dmu_write(os, object, endoff, sizeof (uint64_t), &object, tx); 1693 1694 dmu_tx_commit(tx); 1695 } 1696 } 1697 1698 /* 1699 * Verify that dmu_{read,write} work as expected. 1700 */ 1701 typedef struct bufwad { 1702 uint64_t bw_index; 1703 uint64_t bw_txg; 1704 uint64_t bw_data; 1705 } bufwad_t; 1706 1707 typedef struct dmu_read_write_dir { 1708 uint64_t dd_packobj; 1709 uint64_t dd_bigobj; 1710 uint64_t dd_chunk; 1711 } dmu_read_write_dir_t; 1712 1713 void 1714 ztest_dmu_read_write(ztest_args_t *za) 1715 { 1716 objset_t *os = za->za_os; 1717 dmu_read_write_dir_t dd; 1718 dmu_tx_t *tx; 1719 int i, freeit, error; 1720 uint64_t n, s, txg; 1721 bufwad_t *packbuf, *bigbuf, *pack, *bigH, *bigT; 1722 uint64_t packoff, packsize, bigoff, bigsize; 1723 uint64_t regions = 997; 1724 uint64_t stride = 123456789ULL; 1725 uint64_t width = 40; 1726 int free_percent = 5; 1727 1728 /* 1729 * This test uses two objects, packobj and bigobj, that are always 1730 * updated together (i.e. in the same tx) so that their contents are 1731 * in sync and can be compared. Their contents relate to each other 1732 * in a simple way: packobj is a dense array of 'bufwad' structures, 1733 * while bigobj is a sparse array of the same bufwads. Specifically, 1734 * for any index n, there are three bufwads that should be identical: 1735 * 1736 * packobj, at offset n * sizeof (bufwad_t) 1737 * bigobj, at the head of the nth chunk 1738 * bigobj, at the tail of the nth chunk 1739 * 1740 * The chunk size is arbitrary. It doesn't have to be a power of two, 1741 * and it doesn't have any relation to the object blocksize. 1742 * The only requirement is that it can hold at least two bufwads. 1743 * 1744 * Normally, we write the bufwad to each of these locations. 1745 * However, free_percent of the time we instead write zeroes to 1746 * packobj and perform a dmu_free_range() on bigobj. By comparing 1747 * bigobj to packobj, we can verify that the DMU is correctly 1748 * tracking which parts of an object are allocated and free, 1749 * and that the contents of the allocated blocks are correct. 1750 */ 1751 1752 /* 1753 * Read the directory info. If it's the first time, set things up. 1754 */ 1755 VERIFY(0 == dmu_read(os, ZTEST_DIROBJ, za->za_diroff, 1756 sizeof (dd), &dd)); 1757 if (dd.dd_chunk == 0) { 1758 ASSERT(dd.dd_packobj == 0); 1759 ASSERT(dd.dd_bigobj == 0); 1760 tx = dmu_tx_create(os); 1761 dmu_tx_hold_write(tx, ZTEST_DIROBJ, za->za_diroff, sizeof (dd)); 1762 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT); 1763 error = dmu_tx_assign(tx, TXG_WAIT); 1764 if (error) { 1765 ztest_record_enospc("create r/w directory"); 1766 dmu_tx_abort(tx); 1767 return; 1768 } 1769 1770 dd.dd_packobj = dmu_object_alloc(os, DMU_OT_UINT64_OTHER, 0, 1771 DMU_OT_NONE, 0, tx); 1772 dd.dd_bigobj = dmu_object_alloc(os, DMU_OT_UINT64_OTHER, 0, 1773 DMU_OT_NONE, 0, tx); 1774 dd.dd_chunk = (1000 + ztest_random(1000)) * sizeof (uint64_t); 1775 1776 ztest_set_random_blocksize(os, dd.dd_packobj, tx); 1777 ztest_set_random_blocksize(os, dd.dd_bigobj, tx); 1778 1779 dmu_write(os, ZTEST_DIROBJ, za->za_diroff, sizeof (dd), &dd, 1780 tx); 1781 dmu_tx_commit(tx); 1782 } 1783 1784 /* 1785 * Prefetch a random chunk of the big object. 1786 * Our aim here is to get some async reads in flight 1787 * for blocks that we may free below; the DMU should 1788 * handle this race correctly. 1789 */ 1790 n = ztest_random(regions) * stride + ztest_random(width); 1791 s = 1 + ztest_random(2 * width - 1); 1792 dmu_prefetch(os, dd.dd_bigobj, n * dd.dd_chunk, s * dd.dd_chunk); 1793 1794 /* 1795 * Pick a random index and compute the offsets into packobj and bigobj. 1796 */ 1797 n = ztest_random(regions) * stride + ztest_random(width); 1798 s = 1 + ztest_random(width - 1); 1799 1800 packoff = n * sizeof (bufwad_t); 1801 packsize = s * sizeof (bufwad_t); 1802 1803 bigoff = n * dd.dd_chunk; 1804 bigsize = s * dd.dd_chunk; 1805 1806 packbuf = umem_alloc(packsize, UMEM_NOFAIL); 1807 bigbuf = umem_alloc(bigsize, UMEM_NOFAIL); 1808 1809 /* 1810 * free_percent of the time, free a range of bigobj rather than 1811 * overwriting it. 1812 */ 1813 freeit = (ztest_random(100) < free_percent); 1814 1815 /* 1816 * Read the current contents of our objects. 1817 */ 1818 error = dmu_read(os, dd.dd_packobj, packoff, packsize, packbuf); 1819 ASSERT3U(error, ==, 0); 1820 error = dmu_read(os, dd.dd_bigobj, bigoff, bigsize, bigbuf); 1821 ASSERT3U(error, ==, 0); 1822 1823 /* 1824 * Get a tx for the mods to both packobj and bigobj. 1825 */ 1826 tx = dmu_tx_create(os); 1827 1828 dmu_tx_hold_write(tx, dd.dd_packobj, packoff, packsize); 1829 1830 if (freeit) 1831 dmu_tx_hold_free(tx, dd.dd_bigobj, bigoff, bigsize); 1832 else 1833 dmu_tx_hold_write(tx, dd.dd_bigobj, bigoff, bigsize); 1834 1835 error = dmu_tx_assign(tx, TXG_WAIT); 1836 1837 if (error) { 1838 ztest_record_enospc("dmu r/w range"); 1839 dmu_tx_abort(tx); 1840 umem_free(packbuf, packsize); 1841 umem_free(bigbuf, bigsize); 1842 return; 1843 } 1844 1845 txg = dmu_tx_get_txg(tx); 1846 1847 /* 1848 * For each index from n to n + s, verify that the existing bufwad 1849 * in packobj matches the bufwads at the head and tail of the 1850 * corresponding chunk in bigobj. Then update all three bufwads 1851 * with the new values we want to write out. 1852 */ 1853 for (i = 0; i < s; i++) { 1854 /* LINTED */ 1855 pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t)); 1856 /* LINTED */ 1857 bigH = (bufwad_t *)((char *)bigbuf + i * dd.dd_chunk); 1858 /* LINTED */ 1859 bigT = (bufwad_t *)((char *)bigH + dd.dd_chunk) - 1; 1860 1861 ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize); 1862 ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize); 1863 1864 if (pack->bw_txg > txg) 1865 fatal(0, "future leak: got %llx, open txg is %llx", 1866 pack->bw_txg, txg); 1867 1868 if (pack->bw_data != 0 && pack->bw_index != n + i) 1869 fatal(0, "wrong index: got %llx, wanted %llx+%llx", 1870 pack->bw_index, n, i); 1871 1872 if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0) 1873 fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH); 1874 1875 if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0) 1876 fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT); 1877 1878 if (freeit) { 1879 bzero(pack, sizeof (bufwad_t)); 1880 } else { 1881 pack->bw_index = n + i; 1882 pack->bw_txg = txg; 1883 pack->bw_data = 1 + ztest_random(-2ULL); 1884 } 1885 *bigH = *pack; 1886 *bigT = *pack; 1887 } 1888 1889 /* 1890 * We've verified all the old bufwads, and made new ones. 1891 * Now write them out. 1892 */ 1893 dmu_write(os, dd.dd_packobj, packoff, packsize, packbuf, tx); 1894 1895 if (freeit) { 1896 if (zopt_verbose >= 6) { 1897 (void) printf("freeing offset %llx size %llx" 1898 " txg %llx\n", 1899 (u_longlong_t)bigoff, 1900 (u_longlong_t)bigsize, 1901 (u_longlong_t)txg); 1902 } 1903 VERIFY(0 == dmu_free_range(os, dd.dd_bigobj, bigoff, 1904 bigsize, tx)); 1905 } else { 1906 if (zopt_verbose >= 6) { 1907 (void) printf("writing offset %llx size %llx" 1908 " txg %llx\n", 1909 (u_longlong_t)bigoff, 1910 (u_longlong_t)bigsize, 1911 (u_longlong_t)txg); 1912 } 1913 dmu_write(os, dd.dd_bigobj, bigoff, bigsize, bigbuf, tx); 1914 } 1915 1916 dmu_tx_commit(tx); 1917 1918 /* 1919 * Sanity check the stuff we just wrote. 1920 */ 1921 { 1922 void *packcheck = umem_alloc(packsize, UMEM_NOFAIL); 1923 void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL); 1924 1925 VERIFY(0 == dmu_read(os, dd.dd_packobj, packoff, 1926 packsize, packcheck)); 1927 VERIFY(0 == dmu_read(os, dd.dd_bigobj, bigoff, 1928 bigsize, bigcheck)); 1929 1930 ASSERT(bcmp(packbuf, packcheck, packsize) == 0); 1931 ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0); 1932 1933 umem_free(packcheck, packsize); 1934 umem_free(bigcheck, bigsize); 1935 } 1936 1937 umem_free(packbuf, packsize); 1938 umem_free(bigbuf, bigsize); 1939 } 1940 1941 void 1942 ztest_dmu_check_future_leak(objset_t *os, uint64_t txg) 1943 { 1944 dmu_buf_t *db; 1945 ztest_block_tag_t rbt; 1946 1947 if (zopt_verbose >= 3) { 1948 char osname[MAXNAMELEN]; 1949 dmu_objset_name(os, osname); 1950 (void) printf("checking %s for future leaks in txg %lld...\n", 1951 osname, (u_longlong_t)txg); 1952 } 1953 1954 /* 1955 * Make sure that, if there is a write record in the bonus buffer 1956 * of the ZTEST_DIROBJ, that the txg for this record is <= the 1957 * last synced txg of the pool. 1958 */ 1959 1960 VERIFY(0 == dmu_bonus_hold(os, ZTEST_DIROBJ, FTAG, &db)); 1961 ASSERT3U(db->db_size, >=, sizeof (rbt)); 1962 bcopy(db->db_data, &rbt, sizeof (rbt)); 1963 if (rbt.bt_objset != 0) { 1964 ASSERT3U(rbt.bt_objset, ==, dmu_objset_id(os)); 1965 ASSERT3U(rbt.bt_object, ==, ZTEST_DIROBJ); 1966 ASSERT3U(rbt.bt_offset, ==, -1ULL); 1967 if (rbt.bt_txg > txg) { 1968 fatal(0, 1969 "future leak: got %llx, last synced txg is %llx", 1970 rbt.bt_txg, txg); 1971 } 1972 } 1973 dmu_buf_rele(db, FTAG); 1974 } 1975 1976 void 1977 ztest_dmu_write_parallel(ztest_args_t *za) 1978 { 1979 objset_t *os = za->za_os; 1980 dmu_tx_t *tx; 1981 dmu_buf_t *db; 1982 int i, b, error, do_free, bs; 1983 uint64_t off, txg_how, txg; 1984 mutex_t *lp; 1985 char osname[MAXNAMELEN]; 1986 char iobuf[SPA_MAXBLOCKSIZE]; 1987 ztest_block_tag_t rbt, wbt; 1988 1989 dmu_objset_name(os, osname); 1990 bs = ZTEST_DIROBJ_BLOCKSIZE; 1991 1992 /* 1993 * Have multiple threads write to large offsets in ZTEST_DIROBJ 1994 * to verify that having multiple threads writing to the same object 1995 * in parallel doesn't cause any trouble. 1996 * Also do parallel writes to the bonus buffer on occasion. 1997 */ 1998 for (i = 0; i < 50; i++) { 1999 b = ztest_random(ZTEST_SYNC_LOCKS); 2000 lp = &ztest_shared->zs_sync_lock[b]; 2001 2002 do_free = (ztest_random(4) == 0); 2003 2004 off = za->za_diroff_shared + ((uint64_t)b << SPA_MAXBLOCKSHIFT); 2005 2006 if (ztest_random(4) == 0) { 2007 /* 2008 * Do the bonus buffer instead of a regular block. 2009 */ 2010 do_free = 0; 2011 off = -1ULL; 2012 } 2013 2014 tx = dmu_tx_create(os); 2015 2016 if (off == -1ULL) 2017 dmu_tx_hold_bonus(tx, ZTEST_DIROBJ); 2018 else if (do_free) 2019 dmu_tx_hold_free(tx, ZTEST_DIROBJ, off, bs); 2020 else 2021 dmu_tx_hold_write(tx, ZTEST_DIROBJ, off, bs); 2022 2023 txg_how = ztest_random(2) == 0 ? TXG_WAIT : TXG_NOWAIT; 2024 error = dmu_tx_assign(tx, txg_how); 2025 if (error) { 2026 if (error == ERESTART) { 2027 ASSERT(txg_how == TXG_NOWAIT); 2028 dmu_tx_wait(tx); 2029 dmu_tx_abort(tx); 2030 continue; 2031 } 2032 dmu_tx_abort(tx); 2033 ztest_record_enospc("dmu write parallel"); 2034 return; 2035 } 2036 txg = dmu_tx_get_txg(tx); 2037 2038 if (do_free) { 2039 (void) mutex_lock(lp); 2040 VERIFY(0 == dmu_free_range(os, ZTEST_DIROBJ, off, 2041 bs, tx)); 2042 (void) mutex_unlock(lp); 2043 dmu_tx_commit(tx); 2044 continue; 2045 } 2046 2047 wbt.bt_objset = dmu_objset_id(os); 2048 wbt.bt_object = ZTEST_DIROBJ; 2049 wbt.bt_offset = off; 2050 wbt.bt_txg = txg; 2051 wbt.bt_thread = za->za_instance; 2052 2053 if (off == -1ULL) { 2054 dmu_object_info_t doi; 2055 char *off; 2056 2057 wbt.bt_seq = 0; 2058 VERIFY(0 == dmu_bonus_hold(os, ZTEST_DIROBJ, 2059 FTAG, &db)); 2060 dmu_object_info_from_db(db, &doi); 2061 ASSERT3U(doi.doi_bonus_size, >=, sizeof (wbt)); 2062 off = (char *)db->db_data + 2063 doi.doi_bonus_size - sizeof (wbt); 2064 bcopy(off, &rbt, sizeof (wbt)); 2065 if (rbt.bt_objset != 0) { 2066 ASSERT3U(rbt.bt_objset, ==, wbt.bt_objset); 2067 ASSERT3U(rbt.bt_object, ==, wbt.bt_object); 2068 ASSERT3U(rbt.bt_offset, ==, wbt.bt_offset); 2069 ASSERT3U(rbt.bt_txg, <=, wbt.bt_txg); 2070 } 2071 if (ztest_random(10) == 0) { 2072 int newsize = (ztest_random( 2073 db->db_size / sizeof (wbt)) + 1) * 2074 sizeof (wbt); 2075 2076 ASSERT3U(newsize, >=, sizeof (wbt)); 2077 ASSERT3U(newsize, <=, db->db_size); 2078 error = dmu_set_bonus(db, newsize, tx); 2079 ASSERT3U(error, ==, 0); 2080 off = (char *)db->db_data + newsize - 2081 sizeof (wbt); 2082 } 2083 dmu_buf_will_dirty(db, tx); 2084 bcopy(&wbt, off, db->db_size); 2085 dmu_buf_rele(db, FTAG); 2086 dmu_tx_commit(tx); 2087 continue; 2088 } 2089 2090 (void) mutex_lock(lp); 2091 2092 wbt.bt_seq = ztest_shared->zs_seq[b]++; 2093 2094 dmu_write(os, ZTEST_DIROBJ, off, sizeof (wbt), &wbt, tx); 2095 2096 (void) mutex_unlock(lp); 2097 2098 if (ztest_random(100) == 0) 2099 (void) poll(NULL, 0, 1); /* open dn_notxholds window */ 2100 2101 dmu_tx_commit(tx); 2102 2103 if (ztest_random(1000) == 0) 2104 txg_wait_synced(dmu_objset_pool(os), txg); 2105 2106 if (ztest_random(2) == 0) { 2107 blkptr_t blk = { 0 }; 2108 uint64_t blkoff; 2109 zbookmark_t zb; 2110 2111 (void) mutex_lock(lp); 2112 blkoff = P2ALIGN_TYPED(off, bs, uint64_t); 2113 error = dmu_buf_hold(os, 2114 ZTEST_DIROBJ, blkoff, FTAG, &db); 2115 if (error) { 2116 dprintf("dmu_buf_hold(%s, %d, %llx) = %d\n", 2117 osname, ZTEST_DIROBJ, blkoff, error); 2118 (void) mutex_unlock(lp); 2119 continue; 2120 } 2121 blkoff = off - blkoff; 2122 error = dmu_sync(NULL, db, &blk, txg, NULL, NULL); 2123 dmu_buf_rele(db, FTAG); 2124 (void) mutex_unlock(lp); 2125 if (error) { 2126 dprintf("dmu_sync(%s, %d, %llx) = %d\n", 2127 osname, ZTEST_DIROBJ, off, error); 2128 continue; 2129 } 2130 2131 if (blk.blk_birth == 0) { /* concurrent free */ 2132 continue; 2133 } 2134 txg_suspend(dmu_objset_pool(os)); 2135 2136 ASSERT(blk.blk_fill == 1); 2137 ASSERT3U(BP_GET_TYPE(&blk), ==, DMU_OT_UINT64_OTHER); 2138 ASSERT3U(BP_GET_LEVEL(&blk), ==, 0); 2139 ASSERT3U(BP_GET_LSIZE(&blk), ==, bs); 2140 2141 /* 2142 * Read the block that dmu_sync() returned to 2143 * make sure its contents match what we wrote. 2144 * We do this while still txg_suspend()ed to ensure 2145 * that the block can't be reused before we read it. 2146 */ 2147 zb.zb_objset = dmu_objset_id(os); 2148 zb.zb_object = ZTEST_DIROBJ; 2149 zb.zb_level = 0; 2150 zb.zb_blkid = off / bs; 2151 error = zio_wait(zio_read(NULL, dmu_objset_spa(os), 2152 &blk, iobuf, bs, NULL, NULL, 2153 ZIO_PRIORITY_SYNC_READ, ZIO_FLAG_MUSTSUCCEED, &zb)); 2154 ASSERT(error == 0); 2155 2156 txg_resume(dmu_objset_pool(os)); 2157 2158 bcopy(&iobuf[blkoff], &rbt, sizeof (rbt)); 2159 2160 if (rbt.bt_objset == 0) /* concurrent free */ 2161 continue; 2162 2163 ASSERT3U(rbt.bt_objset, ==, wbt.bt_objset); 2164 ASSERT3U(rbt.bt_object, ==, wbt.bt_object); 2165 ASSERT3U(rbt.bt_offset, ==, wbt.bt_offset); 2166 2167 /* 2168 * The semantic of dmu_sync() is that we always 2169 * push the most recent version of the data, 2170 * so in the face of concurrent updates we may 2171 * see a newer version of the block. That's OK. 2172 */ 2173 ASSERT3U(rbt.bt_txg, >=, wbt.bt_txg); 2174 if (rbt.bt_thread == wbt.bt_thread) 2175 ASSERT3U(rbt.bt_seq, ==, wbt.bt_seq); 2176 else 2177 ASSERT3U(rbt.bt_seq, >, wbt.bt_seq); 2178 } 2179 } 2180 } 2181 2182 /* 2183 * Verify that zap_{create,destroy,add,remove,update} work as expected. 2184 */ 2185 #define ZTEST_ZAP_MIN_INTS 1 2186 #define ZTEST_ZAP_MAX_INTS 4 2187 #define ZTEST_ZAP_MAX_PROPS 1000 2188 2189 void 2190 ztest_zap(ztest_args_t *za) 2191 { 2192 objset_t *os = za->za_os; 2193 uint64_t object; 2194 uint64_t txg, last_txg; 2195 uint64_t value[ZTEST_ZAP_MAX_INTS]; 2196 uint64_t zl_ints, zl_intsize, prop; 2197 int i, ints; 2198 int iters = 100; 2199 dmu_tx_t *tx; 2200 char propname[100], txgname[100]; 2201 int error; 2202 char osname[MAXNAMELEN]; 2203 char *hc[2] = { "s.acl.h", ".s.open.h.hyLZlg" }; 2204 2205 dmu_objset_name(os, osname); 2206 2207 /* 2208 * Create a new object if necessary, and record it in the directory. 2209 */ 2210 VERIFY(0 == dmu_read(os, ZTEST_DIROBJ, za->za_diroff, 2211 sizeof (uint64_t), &object)); 2212 2213 if (object == 0) { 2214 tx = dmu_tx_create(os); 2215 dmu_tx_hold_write(tx, ZTEST_DIROBJ, za->za_diroff, 2216 sizeof (uint64_t)); 2217 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, TRUE, NULL); 2218 error = dmu_tx_assign(tx, TXG_WAIT); 2219 if (error) { 2220 ztest_record_enospc("create zap test obj"); 2221 dmu_tx_abort(tx); 2222 return; 2223 } 2224 object = zap_create(os, DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx); 2225 if (error) { 2226 fatal(0, "zap_create('%s', %llu) = %d", 2227 osname, object, error); 2228 } 2229 ASSERT(object != 0); 2230 dmu_write(os, ZTEST_DIROBJ, za->za_diroff, 2231 sizeof (uint64_t), &object, tx); 2232 /* 2233 * Generate a known hash collision, and verify that 2234 * we can lookup and remove both entries. 2235 */ 2236 for (i = 0; i < 2; i++) { 2237 value[i] = i; 2238 error = zap_add(os, object, hc[i], sizeof (uint64_t), 2239 1, &value[i], tx); 2240 ASSERT3U(error, ==, 0); 2241 } 2242 for (i = 0; i < 2; i++) { 2243 error = zap_add(os, object, hc[i], sizeof (uint64_t), 2244 1, &value[i], tx); 2245 ASSERT3U(error, ==, EEXIST); 2246 error = zap_length(os, object, hc[i], 2247 &zl_intsize, &zl_ints); 2248 ASSERT3U(error, ==, 0); 2249 ASSERT3U(zl_intsize, ==, sizeof (uint64_t)); 2250 ASSERT3U(zl_ints, ==, 1); 2251 } 2252 for (i = 0; i < 2; i++) { 2253 error = zap_remove(os, object, hc[i], tx); 2254 ASSERT3U(error, ==, 0); 2255 } 2256 2257 dmu_tx_commit(tx); 2258 } 2259 2260 ints = MAX(ZTEST_ZAP_MIN_INTS, object % ZTEST_ZAP_MAX_INTS); 2261 2262 while (--iters >= 0) { 2263 prop = ztest_random(ZTEST_ZAP_MAX_PROPS); 2264 (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop); 2265 (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop); 2266 bzero(value, sizeof (value)); 2267 last_txg = 0; 2268 2269 /* 2270 * If these zap entries already exist, validate their contents. 2271 */ 2272 error = zap_length(os, object, txgname, &zl_intsize, &zl_ints); 2273 if (error == 0) { 2274 ASSERT3U(zl_intsize, ==, sizeof (uint64_t)); 2275 ASSERT3U(zl_ints, ==, 1); 2276 2277 error = zap_lookup(os, object, txgname, zl_intsize, 2278 zl_ints, &last_txg); 2279 2280 ASSERT3U(error, ==, 0); 2281 2282 error = zap_length(os, object, propname, &zl_intsize, 2283 &zl_ints); 2284 2285 ASSERT3U(error, ==, 0); 2286 ASSERT3U(zl_intsize, ==, sizeof (uint64_t)); 2287 ASSERT3U(zl_ints, ==, ints); 2288 2289 error = zap_lookup(os, object, propname, zl_intsize, 2290 zl_ints, value); 2291 2292 ASSERT3U(error, ==, 0); 2293 2294 for (i = 0; i < ints; i++) { 2295 ASSERT3U(value[i], ==, last_txg + object + i); 2296 } 2297 } else { 2298 ASSERT3U(error, ==, ENOENT); 2299 } 2300 2301 /* 2302 * Atomically update two entries in our zap object. 2303 * The first is named txg_%llu, and contains the txg 2304 * in which the property was last updated. The second 2305 * is named prop_%llu, and the nth element of its value 2306 * should be txg + object + n. 2307 */ 2308 tx = dmu_tx_create(os); 2309 dmu_tx_hold_zap(tx, object, TRUE, NULL); 2310 error = dmu_tx_assign(tx, TXG_WAIT); 2311 if (error) { 2312 ztest_record_enospc("create zap entry"); 2313 dmu_tx_abort(tx); 2314 return; 2315 } 2316 txg = dmu_tx_get_txg(tx); 2317 2318 if (last_txg > txg) 2319 fatal(0, "zap future leak: old %llu new %llu", 2320 last_txg, txg); 2321 2322 for (i = 0; i < ints; i++) 2323 value[i] = txg + object + i; 2324 2325 error = zap_update(os, object, txgname, sizeof (uint64_t), 2326 1, &txg, tx); 2327 if (error) 2328 fatal(0, "zap_update('%s', %llu, '%s') = %d", 2329 osname, object, txgname, error); 2330 2331 error = zap_update(os, object, propname, sizeof (uint64_t), 2332 ints, value, tx); 2333 if (error) 2334 fatal(0, "zap_update('%s', %llu, '%s') = %d", 2335 osname, object, propname, error); 2336 2337 dmu_tx_commit(tx); 2338 2339 /* 2340 * Remove a random pair of entries. 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 2346 error = zap_length(os, object, txgname, &zl_intsize, &zl_ints); 2347 2348 if (error == ENOENT) 2349 continue; 2350 2351 ASSERT3U(error, ==, 0); 2352 2353 tx = dmu_tx_create(os); 2354 dmu_tx_hold_zap(tx, object, TRUE, NULL); 2355 error = dmu_tx_assign(tx, TXG_WAIT); 2356 if (error) { 2357 ztest_record_enospc("remove zap entry"); 2358 dmu_tx_abort(tx); 2359 return; 2360 } 2361 error = zap_remove(os, object, txgname, tx); 2362 if (error) 2363 fatal(0, "zap_remove('%s', %llu, '%s') = %d", 2364 osname, object, txgname, error); 2365 2366 error = zap_remove(os, object, propname, tx); 2367 if (error) 2368 fatal(0, "zap_remove('%s', %llu, '%s') = %d", 2369 osname, object, propname, error); 2370 2371 dmu_tx_commit(tx); 2372 } 2373 2374 /* 2375 * Once in a while, destroy the object. 2376 */ 2377 if (ztest_random(100) != 0) 2378 return; 2379 2380 tx = dmu_tx_create(os); 2381 dmu_tx_hold_write(tx, ZTEST_DIROBJ, za->za_diroff, sizeof (uint64_t)); 2382 dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END); 2383 error = dmu_tx_assign(tx, TXG_WAIT); 2384 if (error) { 2385 ztest_record_enospc("destroy zap object"); 2386 dmu_tx_abort(tx); 2387 return; 2388 } 2389 error = zap_destroy(os, object, tx); 2390 if (error) 2391 fatal(0, "zap_destroy('%s', %llu) = %d", 2392 osname, object, error); 2393 object = 0; 2394 dmu_write(os, ZTEST_DIROBJ, za->za_diroff, sizeof (uint64_t), 2395 &object, tx); 2396 dmu_tx_commit(tx); 2397 } 2398 2399 void 2400 ztest_zap_parallel(ztest_args_t *za) 2401 { 2402 objset_t *os = za->za_os; 2403 uint64_t txg, object, count, wsize, wc, zl_wsize, zl_wc; 2404 int iters = 100; 2405 dmu_tx_t *tx; 2406 int i, namelen, error; 2407 char name[20], string_value[20]; 2408 void *data; 2409 2410 while (--iters >= 0) { 2411 /* 2412 * Generate a random name of the form 'xxx.....' where each 2413 * x is a random printable character and the dots are dots. 2414 * There are 94 such characters, and the name length goes from 2415 * 6 to 20, so there are 94^3 * 15 = 12,458,760 possible names. 2416 */ 2417 namelen = ztest_random(sizeof (name) - 5) + 5 + 1; 2418 2419 for (i = 0; i < 3; i++) 2420 name[i] = '!' + ztest_random('~' - '!' + 1); 2421 for (; i < namelen - 1; i++) 2422 name[i] = '.'; 2423 name[i] = '\0'; 2424 2425 if (ztest_random(2) == 0) 2426 object = ZTEST_MICROZAP_OBJ; 2427 else 2428 object = ZTEST_FATZAP_OBJ; 2429 2430 if ((namelen & 1) || object == ZTEST_MICROZAP_OBJ) { 2431 wsize = sizeof (txg); 2432 wc = 1; 2433 data = &txg; 2434 } else { 2435 wsize = 1; 2436 wc = namelen; 2437 data = string_value; 2438 } 2439 2440 count = -1ULL; 2441 VERIFY(zap_count(os, object, &count) == 0); 2442 ASSERT(count != -1ULL); 2443 2444 /* 2445 * Select an operation: length, lookup, add, update, remove. 2446 */ 2447 i = ztest_random(5); 2448 2449 if (i >= 2) { 2450 tx = dmu_tx_create(os); 2451 dmu_tx_hold_zap(tx, object, TRUE, NULL); 2452 error = dmu_tx_assign(tx, TXG_WAIT); 2453 if (error) { 2454 ztest_record_enospc("zap parallel"); 2455 dmu_tx_abort(tx); 2456 return; 2457 } 2458 txg = dmu_tx_get_txg(tx); 2459 bcopy(name, string_value, namelen); 2460 } else { 2461 tx = NULL; 2462 txg = 0; 2463 bzero(string_value, namelen); 2464 } 2465 2466 switch (i) { 2467 2468 case 0: 2469 error = zap_length(os, object, name, &zl_wsize, &zl_wc); 2470 if (error == 0) { 2471 ASSERT3U(wsize, ==, zl_wsize); 2472 ASSERT3U(wc, ==, zl_wc); 2473 } else { 2474 ASSERT3U(error, ==, ENOENT); 2475 } 2476 break; 2477 2478 case 1: 2479 error = zap_lookup(os, object, name, wsize, wc, data); 2480 if (error == 0) { 2481 if (data == string_value && 2482 bcmp(name, data, namelen) != 0) 2483 fatal(0, "name '%s' != val '%s' len %d", 2484 name, data, namelen); 2485 } else { 2486 ASSERT3U(error, ==, ENOENT); 2487 } 2488 break; 2489 2490 case 2: 2491 error = zap_add(os, object, name, wsize, wc, data, tx); 2492 ASSERT(error == 0 || error == EEXIST); 2493 break; 2494 2495 case 3: 2496 VERIFY(zap_update(os, object, name, wsize, wc, 2497 data, tx) == 0); 2498 break; 2499 2500 case 4: 2501 error = zap_remove(os, object, name, tx); 2502 ASSERT(error == 0 || error == ENOENT); 2503 break; 2504 } 2505 2506 if (tx != NULL) 2507 dmu_tx_commit(tx); 2508 } 2509 } 2510 2511 void 2512 ztest_dsl_prop_get_set(ztest_args_t *za) 2513 { 2514 objset_t *os = za->za_os; 2515 int i, inherit; 2516 uint64_t value; 2517 const char *prop, *valname; 2518 char setpoint[MAXPATHLEN]; 2519 char osname[MAXNAMELEN]; 2520 int error; 2521 2522 (void) rw_rdlock(&ztest_shared->zs_name_lock); 2523 2524 dmu_objset_name(os, osname); 2525 2526 for (i = 0; i < 2; i++) { 2527 if (i == 0) { 2528 prop = "checksum"; 2529 value = ztest_random_checksum(); 2530 inherit = (value == ZIO_CHECKSUM_INHERIT); 2531 } else { 2532 prop = "compression"; 2533 value = ztest_random_compress(); 2534 inherit = (value == ZIO_COMPRESS_INHERIT); 2535 } 2536 2537 error = dsl_prop_set(osname, prop, sizeof (value), 2538 !inherit, &value); 2539 2540 if (error == ENOSPC) { 2541 ztest_record_enospc("dsl_prop_set"); 2542 break; 2543 } 2544 2545 ASSERT3U(error, ==, 0); 2546 2547 VERIFY3U(dsl_prop_get(osname, prop, sizeof (value), 2548 1, &value, setpoint), ==, 0); 2549 2550 if (i == 0) 2551 valname = zio_checksum_table[value].ci_name; 2552 else 2553 valname = zio_compress_table[value].ci_name; 2554 2555 if (zopt_verbose >= 6) { 2556 (void) printf("%s %s = %s for '%s'\n", 2557 osname, prop, valname, setpoint); 2558 } 2559 } 2560 2561 (void) rw_unlock(&ztest_shared->zs_name_lock); 2562 } 2563 2564 static void 2565 ztest_error_setup(vdev_t *vd, int mode, int mask, uint64_t arg) 2566 { 2567 int c; 2568 2569 for (c = 0; c < vd->vdev_children; c++) 2570 ztest_error_setup(vd->vdev_child[c], mode, mask, arg); 2571 2572 if (vd->vdev_path != NULL) { 2573 vd->vdev_fault_mode = mode; 2574 vd->vdev_fault_mask = mask; 2575 vd->vdev_fault_arg = arg; 2576 } 2577 } 2578 2579 /* 2580 * Inject random faults into the on-disk data. 2581 */ 2582 void 2583 ztest_fault_inject(ztest_args_t *za) 2584 { 2585 int fd; 2586 uint64_t offset; 2587 uint64_t leaves = MAX(zopt_mirrors, 1) * zopt_raidz; 2588 uint64_t bad = 0x1990c0ffeedecade; 2589 uint64_t top, leaf; 2590 char path0[MAXPATHLEN]; 2591 char pathrand[MAXPATHLEN]; 2592 size_t fsize; 2593 spa_t *spa = dmu_objset_spa(za->za_os); 2594 int bshift = SPA_MAXBLOCKSHIFT + 2; /* don't scrog all labels */ 2595 int iters = 1000; 2596 vdev_t *vd0; 2597 uint64_t guid0 = 0; 2598 2599 /* 2600 * We can't inject faults when we have no fault tolerance. 2601 */ 2602 if (zopt_maxfaults == 0) 2603 return; 2604 2605 ASSERT(leaves >= 2); 2606 2607 /* 2608 * Pick a random top-level vdev. 2609 */ 2610 spa_config_enter(spa, RW_READER, FTAG); 2611 top = ztest_random(spa->spa_root_vdev->vdev_children); 2612 spa_config_exit(spa, FTAG); 2613 2614 /* 2615 * Pick a random leaf. 2616 */ 2617 leaf = ztest_random(leaves); 2618 2619 /* 2620 * Generate paths to the first two leaves in this top-level vdev, 2621 * and to the random leaf we selected. We'll induce transient 2622 * I/O errors and random online/offline activity on leaf 0, 2623 * and we'll write random garbage to the randomly chosen leaf. 2624 */ 2625 (void) snprintf(path0, sizeof (path0), 2626 ztest_dev_template, zopt_dir, zopt_pool, top * leaves + 0); 2627 (void) snprintf(pathrand, sizeof (pathrand), 2628 ztest_dev_template, zopt_dir, zopt_pool, top * leaves + leaf); 2629 2630 dprintf("damaging %s and %s\n", path0, pathrand); 2631 2632 spa_config_enter(spa, RW_READER, FTAG); 2633 2634 /* 2635 * If we can tolerate two or more faults, make vd0 fail randomly. 2636 */ 2637 vd0 = vdev_lookup_by_path(spa->spa_root_vdev, path0); 2638 if (vd0 != NULL && zopt_maxfaults >= 2) { 2639 guid0 = vd0->vdev_guid; 2640 ztest_error_setup(vd0, VDEV_FAULT_COUNT, 2641 (1U << ZIO_TYPE_READ) | (1U << ZIO_TYPE_WRITE), 100); 2642 } 2643 2644 spa_config_exit(spa, FTAG); 2645 2646 /* 2647 * If we can tolerate two or more faults, randomly online/offline vd0. 2648 */ 2649 if (zopt_maxfaults >= 2 && guid0 != 0) { 2650 if (ztest_random(10) < 6) 2651 (void) vdev_offline(spa, guid0, B_TRUE); 2652 else 2653 (void) vdev_online(spa, guid0, B_FALSE, NULL); 2654 } 2655 2656 /* 2657 * We have at least single-fault tolerance, so inject data corruption. 2658 */ 2659 fd = open(pathrand, O_RDWR); 2660 2661 if (fd == -1) /* we hit a gap in the device namespace */ 2662 return; 2663 2664 fsize = lseek(fd, 0, SEEK_END); 2665 2666 while (--iters != 0) { 2667 offset = ztest_random(fsize / (leaves << bshift)) * 2668 (leaves << bshift) + (leaf << bshift) + 2669 (ztest_random(1ULL << (bshift - 1)) & -8ULL); 2670 2671 if (offset >= fsize) 2672 continue; 2673 2674 if (zopt_verbose >= 6) 2675 (void) printf("injecting bad word into %s," 2676 " offset 0x%llx\n", pathrand, (u_longlong_t)offset); 2677 2678 if (pwrite(fd, &bad, sizeof (bad), offset) != sizeof (bad)) 2679 fatal(1, "can't inject bad word at 0x%llx in %s", 2680 offset, pathrand); 2681 } 2682 2683 (void) close(fd); 2684 } 2685 2686 /* 2687 * Scrub the pool. 2688 */ 2689 void 2690 ztest_scrub(ztest_args_t *za) 2691 { 2692 spa_t *spa = dmu_objset_spa(za->za_os); 2693 2694 mutex_enter(&spa_namespace_lock); 2695 (void) spa_scrub(spa, POOL_SCRUB_EVERYTHING, B_FALSE); 2696 mutex_exit(&spa_namespace_lock); 2697 (void) poll(NULL, 0, 1000); /* wait a second, then force a restart */ 2698 mutex_enter(&spa_namespace_lock); 2699 (void) spa_scrub(spa, POOL_SCRUB_EVERYTHING, B_FALSE); 2700 mutex_exit(&spa_namespace_lock); 2701 } 2702 2703 /* 2704 * Rename the pool to a different name and then rename it back. 2705 */ 2706 void 2707 ztest_spa_rename(ztest_args_t *za) 2708 { 2709 char *oldname, *newname; 2710 int error; 2711 spa_t *spa; 2712 2713 (void) rw_wrlock(&ztest_shared->zs_name_lock); 2714 2715 oldname = za->za_pool; 2716 newname = umem_alloc(strlen(oldname) + 5, UMEM_NOFAIL); 2717 (void) strcpy(newname, oldname); 2718 (void) strcat(newname, "_tmp"); 2719 2720 /* 2721 * Do the rename 2722 */ 2723 error = spa_rename(oldname, newname); 2724 if (error) 2725 fatal(0, "spa_rename('%s', '%s') = %d", oldname, 2726 newname, error); 2727 2728 /* 2729 * Try to open it under the old name, which shouldn't exist 2730 */ 2731 error = spa_open(oldname, &spa, FTAG); 2732 if (error != ENOENT) 2733 fatal(0, "spa_open('%s') = %d", oldname, error); 2734 2735 /* 2736 * Open it under the new name and make sure it's still the same spa_t. 2737 */ 2738 error = spa_open(newname, &spa, FTAG); 2739 if (error != 0) 2740 fatal(0, "spa_open('%s') = %d", newname, error); 2741 2742 ASSERT(spa == dmu_objset_spa(za->za_os)); 2743 spa_close(spa, FTAG); 2744 2745 /* 2746 * Rename it back to the original 2747 */ 2748 error = spa_rename(newname, oldname); 2749 if (error) 2750 fatal(0, "spa_rename('%s', '%s') = %d", newname, 2751 oldname, error); 2752 2753 /* 2754 * Make sure it can still be opened 2755 */ 2756 error = spa_open(oldname, &spa, FTAG); 2757 if (error != 0) 2758 fatal(0, "spa_open('%s') = %d", oldname, error); 2759 2760 ASSERT(spa == dmu_objset_spa(za->za_os)); 2761 spa_close(spa, FTAG); 2762 2763 umem_free(newname, strlen(newname) + 1); 2764 2765 (void) rw_unlock(&ztest_shared->zs_name_lock); 2766 } 2767 2768 2769 /* 2770 * Completely obliterate one disk. 2771 */ 2772 static void 2773 ztest_obliterate_one_disk(uint64_t vdev) 2774 { 2775 int fd; 2776 char dev_name[MAXPATHLEN], copy_name[MAXPATHLEN]; 2777 size_t fsize; 2778 2779 if (zopt_maxfaults < 2) 2780 return; 2781 2782 (void) sprintf(dev_name, ztest_dev_template, zopt_dir, zopt_pool, vdev); 2783 (void) snprintf(copy_name, MAXPATHLEN, "%s.old", dev_name); 2784 2785 fd = open(dev_name, O_RDWR); 2786 2787 if (fd == -1) 2788 fatal(1, "can't open %s", dev_name); 2789 2790 /* 2791 * Determine the size. 2792 */ 2793 fsize = lseek(fd, 0, SEEK_END); 2794 2795 (void) close(fd); 2796 2797 /* 2798 * Rename the old device to dev_name.old (useful for debugging). 2799 */ 2800 VERIFY(rename(dev_name, copy_name) == 0); 2801 2802 /* 2803 * Create a new one. 2804 */ 2805 VERIFY((fd = open(dev_name, O_RDWR | O_CREAT | O_TRUNC, 0666)) >= 0); 2806 VERIFY(ftruncate(fd, fsize) == 0); 2807 (void) close(fd); 2808 } 2809 2810 static void 2811 ztest_replace_one_disk(spa_t *spa, uint64_t vdev) 2812 { 2813 char dev_name[MAXPATHLEN]; 2814 nvlist_t *file, *root; 2815 int error; 2816 uint64_t guid; 2817 uint64_t ashift = ztest_get_ashift(); 2818 vdev_t *vd; 2819 2820 (void) sprintf(dev_name, ztest_dev_template, zopt_dir, zopt_pool, vdev); 2821 2822 /* 2823 * Build the nvlist describing dev_name. 2824 */ 2825 VERIFY(nvlist_alloc(&file, NV_UNIQUE_NAME, 0) == 0); 2826 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_TYPE, VDEV_TYPE_FILE) == 0); 2827 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_PATH, dev_name) == 0); 2828 VERIFY(nvlist_add_uint64(file, ZPOOL_CONFIG_ASHIFT, ashift) == 0); 2829 2830 VERIFY(nvlist_alloc(&root, NV_UNIQUE_NAME, 0) == 0); 2831 VERIFY(nvlist_add_string(root, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) == 0); 2832 VERIFY(nvlist_add_nvlist_array(root, ZPOOL_CONFIG_CHILDREN, 2833 &file, 1) == 0); 2834 2835 spa_config_enter(spa, RW_READER, FTAG); 2836 if ((vd = vdev_lookup_by_path(spa->spa_root_vdev, dev_name)) == NULL) 2837 guid = 0; 2838 else 2839 guid = vd->vdev_guid; 2840 spa_config_exit(spa, FTAG); 2841 error = spa_vdev_attach(spa, guid, root, B_TRUE); 2842 if (error != 0 && 2843 error != EBUSY && 2844 error != ENOTSUP && 2845 error != ENODEV && 2846 error != EDOM) 2847 fatal(0, "spa_vdev_attach(in-place) = %d", error); 2848 2849 nvlist_free(file); 2850 nvlist_free(root); 2851 } 2852 2853 static void 2854 ztest_verify_blocks(char *pool) 2855 { 2856 int status; 2857 char zdb[MAXPATHLEN + MAXNAMELEN + 20]; 2858 char zbuf[1024]; 2859 char *bin; 2860 char *ztest; 2861 char *isa; 2862 int isalen; 2863 FILE *fp; 2864 2865 (void) realpath(getexecname(), zdb); 2866 2867 /* zdb lives in /usr/sbin, while ztest lives in /usr/bin */ 2868 bin = strstr(zdb, "/usr/bin/"); 2869 ztest = strstr(bin, "/ztest"); 2870 isa = bin + 8; 2871 isalen = ztest - isa; 2872 isa = strdup(isa); 2873 /* LINTED */ 2874 (void) sprintf(bin, "/usr/sbin%.*s/zdb -bc%s%s -U -O %s %s", 2875 isalen, 2876 isa, 2877 zopt_verbose >= 3 ? "s" : "", 2878 zopt_verbose >= 4 ? "v" : "", 2879 ztest_random(2) == 0 ? "pre" : "post", pool); 2880 free(isa); 2881 2882 if (zopt_verbose >= 5) 2883 (void) printf("Executing %s\n", strstr(zdb, "zdb ")); 2884 2885 fp = popen(zdb, "r"); 2886 2887 while (fgets(zbuf, sizeof (zbuf), fp) != NULL) 2888 if (zopt_verbose >= 3) 2889 (void) printf("%s", zbuf); 2890 2891 status = pclose(fp); 2892 2893 if (status == 0) 2894 return; 2895 2896 ztest_dump_core = 0; 2897 if (WIFEXITED(status)) 2898 fatal(0, "'%s' exit code %d", zdb, WEXITSTATUS(status)); 2899 else 2900 fatal(0, "'%s' died with signal %d", zdb, WTERMSIG(status)); 2901 } 2902 2903 static void 2904 ztest_walk_pool_directory(char *header) 2905 { 2906 spa_t *spa = NULL; 2907 2908 if (zopt_verbose >= 6) 2909 (void) printf("%s\n", header); 2910 2911 mutex_enter(&spa_namespace_lock); 2912 while ((spa = spa_next(spa)) != NULL) 2913 if (zopt_verbose >= 6) 2914 (void) printf("\t%s\n", spa_name(spa)); 2915 mutex_exit(&spa_namespace_lock); 2916 } 2917 2918 static void 2919 ztest_spa_import_export(char *oldname, char *newname) 2920 { 2921 nvlist_t *config; 2922 uint64_t pool_guid; 2923 spa_t *spa; 2924 int error; 2925 2926 if (zopt_verbose >= 4) { 2927 (void) printf("import/export: old = %s, new = %s\n", 2928 oldname, newname); 2929 } 2930 2931 /* 2932 * Clean up from previous runs. 2933 */ 2934 (void) spa_destroy(newname); 2935 2936 /* 2937 * Get the pool's configuration and guid. 2938 */ 2939 error = spa_open(oldname, &spa, FTAG); 2940 if (error) 2941 fatal(0, "spa_open('%s') = %d", oldname, error); 2942 2943 pool_guid = spa_guid(spa); 2944 spa_close(spa, FTAG); 2945 2946 ztest_walk_pool_directory("pools before export"); 2947 2948 /* 2949 * Export it. 2950 */ 2951 error = spa_export(oldname, &config); 2952 if (error) 2953 fatal(0, "spa_export('%s') = %d", oldname, error); 2954 2955 ztest_walk_pool_directory("pools after export"); 2956 2957 /* 2958 * Import it under the new name. 2959 */ 2960 error = spa_import(newname, config, NULL); 2961 if (error) 2962 fatal(0, "spa_import('%s') = %d", newname, error); 2963 2964 ztest_walk_pool_directory("pools after import"); 2965 2966 /* 2967 * Try to import it again -- should fail with EEXIST. 2968 */ 2969 error = spa_import(newname, config, NULL); 2970 if (error != EEXIST) 2971 fatal(0, "spa_import('%s') twice", newname); 2972 2973 /* 2974 * Try to import it under a different name -- should fail with EEXIST. 2975 */ 2976 error = spa_import(oldname, config, NULL); 2977 if (error != EEXIST) 2978 fatal(0, "spa_import('%s') under multiple names", newname); 2979 2980 /* 2981 * Verify that the pool is no longer visible under the old name. 2982 */ 2983 error = spa_open(oldname, &spa, FTAG); 2984 if (error != ENOENT) 2985 fatal(0, "spa_open('%s') = %d", newname, error); 2986 2987 /* 2988 * Verify that we can open and close the pool using the new name. 2989 */ 2990 error = spa_open(newname, &spa, FTAG); 2991 if (error) 2992 fatal(0, "spa_open('%s') = %d", newname, error); 2993 ASSERT(pool_guid == spa_guid(spa)); 2994 spa_close(spa, FTAG); 2995 2996 nvlist_free(config); 2997 } 2998 2999 /* ARGSUSED */ 3000 static void * 3001 ztest_suspend_monitor(void *arg) 3002 { 3003 spa_t *spa; 3004 int error; 3005 3006 error = spa_open(zopt_pool, &spa, FTAG); 3007 if (error) { 3008 (void) printf("Unable to monitor pool '%s'\n", zopt_pool); 3009 return (NULL); 3010 } 3011 3012 while (!ztest_exiting) { 3013 mutex_enter(&spa->spa_zio_lock); 3014 while (!ztest_exiting && list_is_empty(&spa->spa_zio_list)) 3015 cv_wait(&spa->spa_zio_cv, &spa->spa_zio_lock); 3016 mutex_exit(&spa->spa_zio_lock); 3017 3018 (void) sleep(3); 3019 /* 3020 * We don't hold the spa_config_lock since the pool is in 3021 * complete failure mode and there is no way for us to 3022 * change the vdev config when we're in this state. 3023 */ 3024 while ((error = zio_vdev_resume_io(spa)) != 0) { 3025 (void) printf("I/O could not be resumed, %d\n", error); 3026 (void) sleep(1); 3027 } 3028 vdev_clear(spa, NULL, B_TRUE); 3029 } 3030 spa_close(spa, FTAG); 3031 return (NULL); 3032 } 3033 3034 static void * 3035 ztest_thread(void *arg) 3036 { 3037 ztest_args_t *za = arg; 3038 ztest_shared_t *zs = ztest_shared; 3039 hrtime_t now, functime; 3040 ztest_info_t *zi; 3041 int f; 3042 3043 while ((now = gethrtime()) < za->za_stop) { 3044 /* 3045 * See if it's time to force a crash. 3046 */ 3047 if (now > za->za_kill) { 3048 dmu_tx_t *tx; 3049 uint64_t txg; 3050 3051 mutex_enter(&spa_namespace_lock); 3052 tx = dmu_tx_create(za->za_os); 3053 VERIFY(0 == dmu_tx_assign(tx, TXG_WAIT)); 3054 txg = dmu_tx_get_txg(tx); 3055 dmu_tx_commit(tx); 3056 zs->zs_txg = txg; 3057 if (zopt_verbose >= 3) 3058 (void) printf( 3059 "killing process after txg %lld\n", 3060 (u_longlong_t)txg); 3061 txg_wait_synced(dmu_objset_pool(za->za_os), txg); 3062 zs->zs_alloc = spa_get_alloc(dmu_objset_spa(za->za_os)); 3063 zs->zs_space = spa_get_space(dmu_objset_spa(za->za_os)); 3064 (void) kill(getpid(), SIGKILL); 3065 } 3066 3067 /* 3068 * Pick a random function. 3069 */ 3070 f = ztest_random(ZTEST_FUNCS); 3071 zi = &zs->zs_info[f]; 3072 3073 /* 3074 * Decide whether to call it, based on the requested frequency. 3075 */ 3076 if (zi->zi_call_target == 0 || 3077 (double)zi->zi_call_total / zi->zi_call_target > 3078 (double)(now - zs->zs_start_time) / (zopt_time * NANOSEC)) 3079 continue; 3080 3081 atomic_add_64(&zi->zi_calls, 1); 3082 atomic_add_64(&zi->zi_call_total, 1); 3083 3084 za->za_diroff = (za->za_instance * ZTEST_FUNCS + f) * 3085 ZTEST_DIRSIZE; 3086 za->za_diroff_shared = (1ULL << 63); 3087 3088 ztest_dmu_write_parallel(za); 3089 3090 zi->zi_func(za); 3091 3092 functime = gethrtime() - now; 3093 3094 atomic_add_64(&zi->zi_call_time, functime); 3095 3096 if (zopt_verbose >= 4) { 3097 Dl_info dli; 3098 (void) dladdr((void *)zi->zi_func, &dli); 3099 (void) printf("%6.2f sec in %s\n", 3100 (double)functime / NANOSEC, dli.dli_sname); 3101 } 3102 3103 /* 3104 * If we're getting ENOSPC with some regularity, stop. 3105 */ 3106 if (zs->zs_enospc_count > 10) 3107 break; 3108 } 3109 3110 return (NULL); 3111 } 3112 3113 /* 3114 * Kick off threads to run tests on all datasets in parallel. 3115 */ 3116 static void 3117 ztest_run(char *pool) 3118 { 3119 int t, d, error; 3120 ztest_shared_t *zs = ztest_shared; 3121 ztest_args_t *za; 3122 spa_t *spa; 3123 char name[100]; 3124 thread_t tid; 3125 3126 (void) _mutex_init(&zs->zs_vdev_lock, USYNC_THREAD, NULL); 3127 (void) rwlock_init(&zs->zs_name_lock, USYNC_THREAD, NULL); 3128 3129 for (t = 0; t < ZTEST_SYNC_LOCKS; t++) 3130 (void) _mutex_init(&zs->zs_sync_lock[t], USYNC_THREAD, NULL); 3131 3132 /* 3133 * Destroy one disk before we even start. 3134 * It's mirrored, so everything should work just fine. 3135 * This makes us exercise fault handling very early in spa_load(). 3136 */ 3137 ztest_obliterate_one_disk(0); 3138 3139 /* 3140 * Verify that the sum of the sizes of all blocks in the pool 3141 * equals the SPA's allocated space total. 3142 */ 3143 ztest_verify_blocks(pool); 3144 3145 /* 3146 * Kick off a replacement of the disk we just obliterated. 3147 */ 3148 kernel_init(FREAD | FWRITE); 3149 error = spa_open(pool, &spa, FTAG); 3150 if (error) 3151 fatal(0, "spa_open(%s) = %d", pool, error); 3152 ztest_replace_one_disk(spa, 0); 3153 if (zopt_verbose >= 5) 3154 show_pool_stats(spa); 3155 spa_close(spa, FTAG); 3156 kernel_fini(); 3157 3158 kernel_init(FREAD | FWRITE); 3159 3160 /* 3161 * Verify that we can export the pool and reimport it under a 3162 * different name. 3163 */ 3164 if (ztest_random(2) == 0) { 3165 (void) snprintf(name, 100, "%s_import", pool); 3166 ztest_spa_import_export(pool, name); 3167 ztest_spa_import_export(name, pool); 3168 } 3169 3170 /* 3171 * Verify that we can loop over all pools. 3172 */ 3173 mutex_enter(&spa_namespace_lock); 3174 for (spa = spa_next(NULL); spa != NULL; spa = spa_next(spa)) { 3175 if (zopt_verbose > 3) { 3176 (void) printf("spa_next: found %s\n", spa_name(spa)); 3177 } 3178 } 3179 mutex_exit(&spa_namespace_lock); 3180 3181 /* 3182 * Create a thread to handling complete pool failures. This 3183 * thread will kickstart the I/Os when they suspend. We must 3184 * start the thread before setting the zio_io_fail_shift, which 3185 * will indicate our failure rate. 3186 */ 3187 error = thr_create(0, 0, ztest_suspend_monitor, NULL, THR_BOUND, &tid); 3188 if (error) { 3189 fatal(0, "can't create suspend monitor thread: error %d", 3190 t, error); 3191 } 3192 3193 /* 3194 * Open our pool. 3195 */ 3196 error = spa_open(pool, &spa, FTAG); 3197 if (error) 3198 fatal(0, "spa_open() = %d", error); 3199 3200 /* 3201 * Verify that we can safely inquire about about any object, 3202 * whether it's allocated or not. To make it interesting, 3203 * we probe a 5-wide window around each power of two. 3204 * This hits all edge cases, including zero and the max. 3205 */ 3206 for (t = 0; t < 64; t++) { 3207 for (d = -5; d <= 5; d++) { 3208 error = dmu_object_info(spa->spa_meta_objset, 3209 (1ULL << t) + d, NULL); 3210 ASSERT(error == 0 || error == ENOENT || 3211 error == EINVAL); 3212 } 3213 } 3214 3215 /* 3216 * Now kick off all the tests that run in parallel. 3217 */ 3218 zs->zs_enospc_count = 0; 3219 3220 za = umem_zalloc(zopt_threads * sizeof (ztest_args_t), UMEM_NOFAIL); 3221 3222 if (zopt_verbose >= 4) 3223 (void) printf("starting main threads...\n"); 3224 3225 /* Let failures begin */ 3226 zio_io_fail_shift = zopt_write_fail_shift; 3227 3228 za[0].za_start = gethrtime(); 3229 za[0].za_stop = za[0].za_start + zopt_passtime * NANOSEC; 3230 za[0].za_stop = MIN(za[0].za_stop, zs->zs_stop_time); 3231 za[0].za_kill = za[0].za_stop; 3232 if (ztest_random(100) < zopt_killrate) 3233 za[0].za_kill -= ztest_random(zopt_passtime * NANOSEC); 3234 3235 for (t = 0; t < zopt_threads; t++) { 3236 d = t % zopt_datasets; 3237 if (t < zopt_datasets) { 3238 ztest_replay_t zr; 3239 int test_future = FALSE; 3240 (void) rw_rdlock(&ztest_shared->zs_name_lock); 3241 (void) snprintf(name, 100, "%s/%s_%d", pool, pool, d); 3242 error = dmu_objset_create(name, DMU_OST_OTHER, NULL, 3243 ztest_create_cb, NULL); 3244 if (error == EEXIST) { 3245 test_future = TRUE; 3246 } else if (error != 0) { 3247 if (error == ENOSPC) { 3248 zs->zs_enospc_count++; 3249 (void) rw_unlock( 3250 &ztest_shared->zs_name_lock); 3251 break; 3252 } 3253 fatal(0, "dmu_objset_create(%s) = %d", 3254 name, error); 3255 } 3256 error = dmu_objset_open(name, DMU_OST_OTHER, 3257 DS_MODE_STANDARD, &za[d].za_os); 3258 if (error) 3259 fatal(0, "dmu_objset_open('%s') = %d", 3260 name, error); 3261 (void) rw_unlock(&ztest_shared->zs_name_lock); 3262 if (test_future && ztest_shared->zs_txg > 0) 3263 ztest_dmu_check_future_leak(za[d].za_os, 3264 ztest_shared->zs_txg); 3265 zr.zr_os = za[d].za_os; 3266 zil_replay(zr.zr_os, &zr, &zr.zr_assign, 3267 ztest_replay_vector); 3268 za[d].za_zilog = zil_open(za[d].za_os, NULL); 3269 } 3270 za[t].za_pool = spa_strdup(pool); 3271 za[t].za_os = za[d].za_os; 3272 za[t].za_zilog = za[d].za_zilog; 3273 za[t].za_instance = t; 3274 za[t].za_random = ztest_random(-1ULL); 3275 za[t].za_start = za[0].za_start; 3276 za[t].za_stop = za[0].za_stop; 3277 za[t].za_kill = za[0].za_kill; 3278 3279 error = thr_create(0, 0, ztest_thread, &za[t], THR_BOUND, 3280 &za[t].za_thread); 3281 if (error) 3282 fatal(0, "can't create thread %d: error %d", 3283 t, error); 3284 } 3285 ztest_shared->zs_txg = 0; 3286 3287 while (--t >= 0) { 3288 error = thr_join(za[t].za_thread, NULL, NULL); 3289 if (error) 3290 fatal(0, "thr_join(%d) = %d", t, error); 3291 if (za[t].za_th) 3292 traverse_fini(za[t].za_th); 3293 if (t < zopt_datasets) { 3294 zil_close(za[t].za_zilog); 3295 dmu_objset_close(za[t].za_os); 3296 } 3297 spa_strfree(za[t].za_pool); 3298 } 3299 3300 umem_free(za, zopt_threads * sizeof (ztest_args_t)); 3301 3302 if (zopt_verbose >= 3) 3303 show_pool_stats(spa); 3304 3305 txg_wait_synced(spa_get_dsl(spa), 0); 3306 3307 zs->zs_alloc = spa_get_alloc(spa); 3308 zs->zs_space = spa_get_space(spa); 3309 3310 /* 3311 * Did we have out-of-space errors? If so, destroy a random objset. 3312 */ 3313 if (zs->zs_enospc_count != 0) { 3314 (void) rw_rdlock(&ztest_shared->zs_name_lock); 3315 (void) snprintf(name, 100, "%s/%s_%d", pool, pool, 3316 (int)ztest_random(zopt_datasets)); 3317 if (zopt_verbose >= 3) 3318 (void) printf("Destroying %s to free up space\n", name); 3319 (void) dmu_objset_find(name, ztest_destroy_cb, NULL, 3320 DS_FIND_SNAPSHOTS | DS_FIND_CHILDREN); 3321 (void) rw_unlock(&ztest_shared->zs_name_lock); 3322 } 3323 3324 txg_wait_synced(spa_get_dsl(spa), 0); 3325 3326 /* 3327 * Right before closing the pool, kick off a bunch of async I/O; 3328 * spa_close() should wait for it to complete. 3329 */ 3330 for (t = 1; t < 50; t++) 3331 dmu_prefetch(spa->spa_meta_objset, t, 0, 1 << 15); 3332 3333 spa_close(spa, FTAG); 3334 3335 /* Shutdown the suspend monitor thread */ 3336 zio_io_fail_shift = 0; 3337 ztest_exiting = B_TRUE; 3338 mutex_enter(&spa->spa_zio_lock); 3339 cv_broadcast(&spa->spa_zio_cv); 3340 mutex_exit(&spa->spa_zio_lock); 3341 error = thr_join(tid, NULL, NULL); 3342 if (error) 3343 fatal(0, "thr_join(%d) = %d", tid, error); 3344 3345 kernel_fini(); 3346 } 3347 3348 void 3349 print_time(hrtime_t t, char *timebuf) 3350 { 3351 hrtime_t s = t / NANOSEC; 3352 hrtime_t m = s / 60; 3353 hrtime_t h = m / 60; 3354 hrtime_t d = h / 24; 3355 3356 s -= m * 60; 3357 m -= h * 60; 3358 h -= d * 24; 3359 3360 timebuf[0] = '\0'; 3361 3362 if (d) 3363 (void) sprintf(timebuf, 3364 "%llud%02lluh%02llum%02llus", d, h, m, s); 3365 else if (h) 3366 (void) sprintf(timebuf, "%lluh%02llum%02llus", h, m, s); 3367 else if (m) 3368 (void) sprintf(timebuf, "%llum%02llus", m, s); 3369 else 3370 (void) sprintf(timebuf, "%llus", s); 3371 } 3372 3373 /* 3374 * Create a storage pool with the given name and initial vdev size. 3375 * Then create the specified number of datasets in the pool. 3376 */ 3377 static void 3378 ztest_init(char *pool) 3379 { 3380 spa_t *spa; 3381 int error; 3382 nvlist_t *nvroot; 3383 3384 kernel_init(FREAD | FWRITE); 3385 3386 /* 3387 * Create the storage pool. 3388 */ 3389 (void) spa_destroy(pool); 3390 ztest_shared->zs_vdev_primaries = 0; 3391 nvroot = make_vdev_root(zopt_vdev_size, 0, zopt_raidz, zopt_mirrors, 1); 3392 error = spa_create(pool, nvroot, NULL, NULL); 3393 nvlist_free(nvroot); 3394 3395 if (error) 3396 fatal(0, "spa_create() = %d", error); 3397 error = spa_open(pool, &spa, FTAG); 3398 if (error) 3399 fatal(0, "spa_open() = %d", error); 3400 3401 if (zopt_verbose >= 3) 3402 show_pool_stats(spa); 3403 3404 spa_close(spa, FTAG); 3405 3406 kernel_fini(); 3407 } 3408 3409 int 3410 main(int argc, char **argv) 3411 { 3412 int kills = 0; 3413 int iters = 0; 3414 int i, f; 3415 ztest_shared_t *zs; 3416 ztest_info_t *zi; 3417 char timebuf[100]; 3418 char numbuf[6]; 3419 3420 (void) setvbuf(stdout, NULL, _IOLBF, 0); 3421 3422 /* Override location of zpool.cache */ 3423 spa_config_dir = "/tmp"; 3424 3425 ztest_random_fd = open("/dev/urandom", O_RDONLY); 3426 3427 process_options(argc, argv); 3428 3429 argc -= optind; 3430 argv += optind; 3431 3432 dprintf_setup(&argc, argv); 3433 3434 /* 3435 * Blow away any existing copy of zpool.cache 3436 */ 3437 if (zopt_init != 0) 3438 (void) remove("/tmp/zpool.cache"); 3439 3440 zs = ztest_shared = (void *)mmap(0, 3441 P2ROUNDUP(sizeof (ztest_shared_t), getpagesize()), 3442 PROT_READ | PROT_WRITE, MAP_SHARED | MAP_ANON, -1, 0); 3443 3444 if (zopt_verbose >= 1) { 3445 (void) printf("%llu vdevs, %d datasets, %d threads," 3446 " %llu seconds...\n", 3447 (u_longlong_t)zopt_vdevs, zopt_datasets, zopt_threads, 3448 (u_longlong_t)zopt_time); 3449 } 3450 3451 /* 3452 * Create and initialize our storage pool. 3453 */ 3454 for (i = 1; i <= zopt_init; i++) { 3455 bzero(zs, sizeof (ztest_shared_t)); 3456 if (zopt_verbose >= 3 && zopt_init != 1) 3457 (void) printf("ztest_init(), pass %d\n", i); 3458 ztest_init(zopt_pool); 3459 } 3460 3461 /* 3462 * Initialize the call targets for each function. 3463 */ 3464 for (f = 0; f < ZTEST_FUNCS; f++) { 3465 zi = &zs->zs_info[f]; 3466 3467 *zi = ztest_info[f]; 3468 3469 if (*zi->zi_interval == 0) 3470 zi->zi_call_target = UINT64_MAX; 3471 else 3472 zi->zi_call_target = zopt_time / *zi->zi_interval; 3473 } 3474 3475 zs->zs_start_time = gethrtime(); 3476 zs->zs_stop_time = zs->zs_start_time + zopt_time * NANOSEC; 3477 3478 /* 3479 * Run the tests in a loop. These tests include fault injection 3480 * to verify that self-healing data works, and forced crashes 3481 * to verify that we never lose on-disk consistency. 3482 */ 3483 while (gethrtime() < zs->zs_stop_time) { 3484 int status; 3485 pid_t pid; 3486 char *tmp; 3487 3488 /* 3489 * Initialize the workload counters for each function. 3490 */ 3491 for (f = 0; f < ZTEST_FUNCS; f++) { 3492 zi = &zs->zs_info[f]; 3493 zi->zi_calls = 0; 3494 zi->zi_call_time = 0; 3495 } 3496 3497 pid = fork(); 3498 3499 if (pid == -1) 3500 fatal(1, "fork failed"); 3501 3502 if (pid == 0) { /* child */ 3503 struct rlimit rl = { 1024, 1024 }; 3504 (void) setrlimit(RLIMIT_NOFILE, &rl); 3505 (void) enable_extended_FILE_stdio(-1, -1); 3506 ztest_run(zopt_pool); 3507 exit(0); 3508 } 3509 3510 while (waitpid(pid, &status, 0) != pid) 3511 continue; 3512 3513 if (WIFEXITED(status)) { 3514 if (WEXITSTATUS(status) != 0) { 3515 (void) fprintf(stderr, 3516 "child exited with code %d\n", 3517 WEXITSTATUS(status)); 3518 exit(2); 3519 } 3520 } else if (WIFSIGNALED(status)) { 3521 if (WTERMSIG(status) != SIGKILL) { 3522 (void) fprintf(stderr, 3523 "child died with signal %d\n", 3524 WTERMSIG(status)); 3525 exit(3); 3526 } 3527 kills++; 3528 } else { 3529 (void) fprintf(stderr, "something strange happened " 3530 "to child\n"); 3531 exit(4); 3532 } 3533 3534 iters++; 3535 3536 if (zopt_verbose >= 1) { 3537 hrtime_t now = gethrtime(); 3538 3539 now = MIN(now, zs->zs_stop_time); 3540 print_time(zs->zs_stop_time - now, timebuf); 3541 nicenum(zs->zs_space, numbuf); 3542 3543 (void) printf("Pass %3d, %8s, %3llu ENOSPC, " 3544 "%4.1f%% of %5s used, %3.0f%% done, %8s to go\n", 3545 iters, 3546 WIFEXITED(status) ? "Complete" : "SIGKILL", 3547 (u_longlong_t)zs->zs_enospc_count, 3548 100.0 * zs->zs_alloc / zs->zs_space, 3549 numbuf, 3550 100.0 * (now - zs->zs_start_time) / 3551 (zopt_time * NANOSEC), timebuf); 3552 } 3553 3554 if (zopt_verbose >= 2) { 3555 (void) printf("\nWorkload summary:\n\n"); 3556 (void) printf("%7s %9s %s\n", 3557 "Calls", "Time", "Function"); 3558 (void) printf("%7s %9s %s\n", 3559 "-----", "----", "--------"); 3560 for (f = 0; f < ZTEST_FUNCS; f++) { 3561 Dl_info dli; 3562 3563 zi = &zs->zs_info[f]; 3564 print_time(zi->zi_call_time, timebuf); 3565 (void) dladdr((void *)zi->zi_func, &dli); 3566 (void) printf("%7llu %9s %s\n", 3567 (u_longlong_t)zi->zi_calls, timebuf, 3568 dli.dli_sname); 3569 } 3570 (void) printf("\n"); 3571 } 3572 3573 /* 3574 * It's possible that we killed a child during a rename test, in 3575 * which case we'll have a 'ztest_tmp' pool lying around instead 3576 * of 'ztest'. Do a blind rename in case this happened. 3577 */ 3578 tmp = umem_alloc(strlen(zopt_pool) + 5, UMEM_NOFAIL); 3579 (void) strcpy(tmp, zopt_pool); 3580 (void) strcat(tmp, "_tmp"); 3581 kernel_init(FREAD | FWRITE); 3582 (void) spa_rename(tmp, zopt_pool); 3583 kernel_fini(); 3584 umem_free(tmp, strlen(tmp) + 1); 3585 } 3586 3587 ztest_verify_blocks(zopt_pool); 3588 3589 if (zopt_verbose >= 1) { 3590 (void) printf("%d killed, %d completed, %.0f%% kill rate\n", 3591 kills, iters - kills, (100.0 * kills) / MAX(1, iters)); 3592 } 3593 3594 return (0); 3595 } 3596