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 2006 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 size_t zopt_vdev_size = SPA_MINDEVSIZE; 118 static int zopt_datasets = 7; 119 static int zopt_threads = 23; 120 static uint64_t zopt_passtime = 60; /* 60 seconds */ 121 static uint64_t zopt_killrate = 70; /* 70% kill rate */ 122 static int zopt_verbose = 0; 123 static int zopt_init = 1; 124 static char *zopt_dir = "/tmp"; 125 static uint64_t zopt_time = 300; /* 5 minutes */ 126 static int zopt_maxfaults; 127 128 typedef struct ztest_args { 129 char *za_pool; 130 objset_t *za_os; 131 zilog_t *za_zilog; 132 thread_t za_thread; 133 uint64_t za_instance; 134 uint64_t za_random; 135 uint64_t za_diroff; 136 uint64_t za_diroff_shared; 137 uint64_t za_zil_seq; 138 hrtime_t za_start; 139 hrtime_t za_stop; 140 hrtime_t za_kill; 141 traverse_handle_t *za_th; 142 } ztest_args_t; 143 144 typedef void ztest_func_t(ztest_args_t *); 145 146 /* 147 * Note: these aren't static because we want dladdr() to work. 148 */ 149 ztest_func_t ztest_dmu_read_write; 150 ztest_func_t ztest_dmu_write_parallel; 151 ztest_func_t ztest_dmu_object_alloc_free; 152 ztest_func_t ztest_zap; 153 ztest_func_t ztest_zap_parallel; 154 ztest_func_t ztest_traverse; 155 ztest_func_t ztest_dsl_prop_get_set; 156 ztest_func_t ztest_dmu_objset_create_destroy; 157 ztest_func_t ztest_dmu_snapshot_create_destroy; 158 ztest_func_t ztest_spa_create_destroy; 159 ztest_func_t ztest_fault_inject; 160 ztest_func_t ztest_vdev_attach_detach; 161 ztest_func_t ztest_vdev_LUN_growth; 162 ztest_func_t ztest_vdev_add_remove; 163 ztest_func_t ztest_scrub; 164 ztest_func_t ztest_spa_rename; 165 166 typedef struct ztest_info { 167 ztest_func_t *zi_func; /* test function */ 168 uint64_t *zi_interval; /* execute every <interval> seconds */ 169 uint64_t zi_calls; /* per-pass count */ 170 uint64_t zi_call_time; /* per-pass time */ 171 uint64_t zi_call_total; /* cumulative total */ 172 uint64_t zi_call_target; /* target cumulative total */ 173 } ztest_info_t; 174 175 uint64_t zopt_always = 0; /* all the time */ 176 uint64_t zopt_often = 1; /* every second */ 177 uint64_t zopt_sometimes = 10; /* every 10 seconds */ 178 uint64_t zopt_rarely = 60; /* every 60 seconds */ 179 180 ztest_info_t ztest_info[] = { 181 { ztest_dmu_read_write, &zopt_always }, 182 { ztest_dmu_write_parallel, &zopt_always }, 183 { ztest_dmu_object_alloc_free, &zopt_always }, 184 { ztest_zap, &zopt_always }, 185 { ztest_zap_parallel, &zopt_always }, 186 { ztest_traverse, &zopt_often }, 187 { ztest_dsl_prop_get_set, &zopt_sometimes }, 188 { ztest_dmu_objset_create_destroy, &zopt_sometimes }, 189 { ztest_dmu_snapshot_create_destroy, &zopt_rarely }, 190 { ztest_spa_create_destroy, &zopt_sometimes }, 191 { ztest_fault_inject, &zopt_sometimes }, 192 { ztest_spa_rename, &zopt_rarely }, 193 { ztest_vdev_attach_detach, &zopt_rarely }, 194 { ztest_vdev_LUN_growth, &zopt_rarely }, 195 { ztest_vdev_add_remove, &zopt_vdevtime }, 196 { ztest_scrub, &zopt_vdevtime }, 197 }; 198 199 #define ZTEST_FUNCS (sizeof (ztest_info) / sizeof (ztest_info_t)) 200 201 #define ZTEST_SYNC_LOCKS 16 202 203 /* 204 * Stuff we need to share writably between parent and child. 205 */ 206 typedef struct ztest_shared { 207 mutex_t zs_vdev_lock; 208 rwlock_t zs_name_lock; 209 uint64_t zs_vdev_primaries; 210 uint64_t zs_enospc_count; 211 hrtime_t zs_start_time; 212 hrtime_t zs_stop_time; 213 uint64_t zs_alloc; 214 uint64_t zs_space; 215 ztest_info_t zs_info[ZTEST_FUNCS]; 216 mutex_t zs_sync_lock[ZTEST_SYNC_LOCKS]; 217 uint64_t zs_seq[ZTEST_SYNC_LOCKS]; 218 } ztest_shared_t; 219 220 typedef struct ztest_block_tag { 221 uint64_t bt_objset; 222 uint64_t bt_object; 223 uint64_t bt_offset; 224 uint64_t bt_txg; 225 uint64_t bt_thread; 226 uint64_t bt_seq; 227 } ztest_block_tag_t; 228 229 static char ztest_dev_template[] = "%s/%s.%llua"; 230 static ztest_shared_t *ztest_shared; 231 232 static int ztest_random_fd; 233 static int ztest_dump_core = 1; 234 235 extern uint64_t zio_gang_bang; 236 237 #define ZTEST_DIROBJ 1 238 #define ZTEST_MICROZAP_OBJ 2 239 #define ZTEST_FATZAP_OBJ 3 240 241 #define ZTEST_DIROBJ_BLOCKSIZE (1 << 10) 242 #define ZTEST_DIRSIZE 256 243 244 /* 245 * These libumem hooks provide a reasonable set of defaults for the allocator's 246 * debugging facilities. 247 */ 248 const char * 249 _umem_debug_init() 250 { 251 return ("default,verbose"); /* $UMEM_DEBUG setting */ 252 } 253 254 const char * 255 _umem_logging_init(void) 256 { 257 return ("fail,contents"); /* $UMEM_LOGGING setting */ 258 } 259 260 #define FATAL_MSG_SZ 1024 261 262 char *fatal_msg; 263 264 static void 265 fatal(int do_perror, char *message, ...) 266 { 267 va_list args; 268 int save_errno = errno; 269 char buf[FATAL_MSG_SZ]; 270 271 (void) fflush(stdout); 272 273 va_start(args, message); 274 (void) sprintf(buf, "ztest: "); 275 /* LINTED */ 276 (void) vsprintf(buf + strlen(buf), message, args); 277 va_end(args); 278 if (do_perror) { 279 (void) snprintf(buf + strlen(buf), FATAL_MSG_SZ - strlen(buf), 280 ": %s", strerror(save_errno)); 281 } 282 (void) fprintf(stderr, "%s\n", buf); 283 fatal_msg = buf; /* to ease debugging */ 284 if (ztest_dump_core) 285 abort(); 286 exit(3); 287 } 288 289 static int 290 str2shift(const char *buf) 291 { 292 const char *ends = "BKMGTPEZ"; 293 int i; 294 295 if (buf[0] == '\0') 296 return (0); 297 for (i = 0; i < strlen(ends); i++) { 298 if (toupper(buf[0]) == ends[i]) 299 break; 300 } 301 if (i == strlen(ends)) 302 fatal(0, "invalid bytes suffix: %s", buf); 303 if (buf[1] == '\0' || (toupper(buf[1]) == 'B' && buf[2] == '\0')) { 304 return (10*i); 305 } 306 fatal(0, "invalid bytes suffix: %s", buf); 307 return (-1); 308 } 309 310 static uint64_t 311 nicenumtoull(const char *buf) 312 { 313 char *end; 314 uint64_t val; 315 316 val = strtoull(buf, &end, 0); 317 if (end == buf) { 318 fatal(0, "bad numeric value: %s", buf); 319 } else if (end[0] == '.') { 320 double fval = strtod(buf, &end); 321 fval *= pow(2, str2shift(end)); 322 if (fval > UINT64_MAX) 323 fatal(0, "value too large: %s", buf); 324 val = (uint64_t)fval; 325 } else { 326 int shift = str2shift(end); 327 if (shift >= 64 || (val << shift) >> shift != val) 328 fatal(0, "value too large: %s", buf); 329 val <<= shift; 330 } 331 return (val); 332 } 333 334 static void 335 usage(void) 336 { 337 char nice_vdev_size[10]; 338 char nice_gang_bang[10]; 339 340 nicenum(zopt_vdev_size, nice_vdev_size); 341 nicenum(zio_gang_bang, nice_gang_bang); 342 343 (void) printf("Usage: %s\n" 344 "\t[-v vdevs (default: %llu)]\n" 345 "\t[-s size_of_each_vdev (default: %s)]\n" 346 "\t[-a alignment_shift (default: %d) (use 0 for random)]\n" 347 "\t[-m mirror_copies (default: %d)]\n" 348 "\t[-r raidz_disks (default: %d)]\n" 349 "\t[-d datasets (default: %d)]\n" 350 "\t[-t threads (default: %d)]\n" 351 "\t[-g gang_block_threshold (default: %s)]\n" 352 "\t[-i initialize pool i times (default: %d)]\n" 353 "\t[-k kill percentage (default: %llu%%)]\n" 354 "\t[-p pool_name (default: %s)]\n" 355 "\t[-f file directory for vdev files (default: %s)]\n" 356 "\t[-V(erbose)] (use multiple times for ever more blather)\n" 357 "\t[-E(xisting)] (use existing pool instead of creating new one)\n" 358 "\t[-T time] total run time (default: %llu sec)\n" 359 "\t[-P passtime] time per pass (default: %llu sec)\n" 360 "", 361 cmdname, 362 (u_longlong_t)zopt_vdevs, /* -v */ 363 nice_vdev_size, /* -s */ 364 zopt_ashift, /* -a */ 365 zopt_mirrors, /* -m */ 366 zopt_raidz, /* -r */ 367 zopt_datasets, /* -d */ 368 zopt_threads, /* -t */ 369 nice_gang_bang, /* -g */ 370 zopt_init, /* -i */ 371 (u_longlong_t)zopt_killrate, /* -k */ 372 zopt_pool, /* -p */ 373 zopt_dir, /* -f */ 374 (u_longlong_t)zopt_time, /* -T */ 375 (u_longlong_t)zopt_passtime); /* -P */ 376 exit(1); 377 } 378 379 static uint64_t 380 ztest_random(uint64_t range) 381 { 382 uint64_t r; 383 384 if (range == 0) 385 return (0); 386 387 if (read(ztest_random_fd, &r, sizeof (r)) != sizeof (r)) 388 fatal(1, "short read from /dev/urandom"); 389 390 return (r % range); 391 } 392 393 static void 394 ztest_record_enospc(char *s) 395 { 396 dprintf("ENOSPC doing: %s\n", s ? s : "<unknown>"); 397 ztest_shared->zs_enospc_count++; 398 } 399 400 static void 401 process_options(int argc, char **argv) 402 { 403 int opt; 404 uint64_t value; 405 406 /* By default, test gang blocks for blocks 32K and greater */ 407 zio_gang_bang = 32 << 10; 408 409 while ((opt = getopt(argc, argv, 410 "v:s:a:m:r:d:t:g:i:k:p:f:VET:P:")) != EOF) { 411 value = 0; 412 switch (opt) { 413 case 'v': 414 case 's': 415 case 'a': 416 case 'm': 417 case 'r': 418 case 'd': 419 case 't': 420 case 'g': 421 case 'i': 422 case 'k': 423 case 'T': 424 case 'P': 425 value = nicenumtoull(optarg); 426 } 427 switch (opt) { 428 case 'v': 429 zopt_vdevs = value; 430 break; 431 case 's': 432 zopt_vdev_size = MAX(SPA_MINDEVSIZE, value); 433 break; 434 case 'a': 435 zopt_ashift = value; 436 break; 437 case 'm': 438 zopt_mirrors = value; 439 break; 440 case 'r': 441 zopt_raidz = MAX(1, value); 442 break; 443 case 'd': 444 zopt_datasets = MAX(1, value); 445 break; 446 case 't': 447 zopt_threads = MAX(1, value); 448 break; 449 case 'g': 450 zio_gang_bang = MAX(SPA_MINBLOCKSIZE << 1, value); 451 break; 452 case 'i': 453 zopt_init = value; 454 break; 455 case 'k': 456 zopt_killrate = value; 457 break; 458 case 'p': 459 zopt_pool = strdup(optarg); 460 break; 461 case 'f': 462 zopt_dir = strdup(optarg); 463 break; 464 case 'V': 465 zopt_verbose++; 466 break; 467 case 'E': 468 zopt_init = 0; 469 break; 470 case 'T': 471 zopt_time = value; 472 break; 473 case 'P': 474 zopt_passtime = MAX(1, value); 475 break; 476 case '?': 477 default: 478 usage(); 479 break; 480 } 481 } 482 483 zopt_vdevtime = (zopt_vdevs > 0 ? zopt_time / zopt_vdevs : UINT64_MAX); 484 zopt_maxfaults = MAX(zopt_mirrors, 1) * (zopt_raidz >= 2 ? 2 : 1) - 1; 485 } 486 487 static uint64_t 488 ztest_get_ashift(void) 489 { 490 if (zopt_ashift == 0) 491 return (SPA_MINBLOCKSHIFT + ztest_random(3)); 492 return (zopt_ashift); 493 } 494 495 static nvlist_t * 496 make_vdev_file(size_t size) 497 { 498 char dev_name[MAXPATHLEN]; 499 uint64_t vdev; 500 uint64_t ashift = ztest_get_ashift(); 501 int fd; 502 nvlist_t *file; 503 504 if (size == 0) { 505 (void) snprintf(dev_name, sizeof (dev_name), "%s", 506 "/dev/bogus"); 507 } else { 508 vdev = ztest_shared->zs_vdev_primaries++; 509 (void) sprintf(dev_name, ztest_dev_template, 510 zopt_dir, zopt_pool, vdev); 511 512 fd = open(dev_name, O_RDWR | O_CREAT | O_TRUNC, 0666); 513 if (fd == -1) 514 fatal(1, "can't open %s", dev_name); 515 if (ftruncate(fd, size) != 0) 516 fatal(1, "can't ftruncate %s", dev_name); 517 (void) close(fd); 518 } 519 520 VERIFY(nvlist_alloc(&file, NV_UNIQUE_NAME, 0) == 0); 521 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_TYPE, VDEV_TYPE_FILE) == 0); 522 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_PATH, dev_name) == 0); 523 VERIFY(nvlist_add_uint64(file, ZPOOL_CONFIG_ASHIFT, ashift) == 0); 524 525 return (file); 526 } 527 528 static nvlist_t * 529 make_vdev_raidz(size_t size, int r) 530 { 531 nvlist_t *raidz, **child; 532 int c; 533 534 if (r < 2) 535 return (make_vdev_file(size)); 536 537 child = umem_alloc(r * sizeof (nvlist_t *), UMEM_NOFAIL); 538 539 for (c = 0; c < r; c++) 540 child[c] = make_vdev_file(size); 541 542 VERIFY(nvlist_alloc(&raidz, NV_UNIQUE_NAME, 0) == 0); 543 VERIFY(nvlist_add_string(raidz, ZPOOL_CONFIG_TYPE, 544 VDEV_TYPE_RAIDZ) == 0); 545 VERIFY(nvlist_add_nvlist_array(raidz, ZPOOL_CONFIG_CHILDREN, 546 child, r) == 0); 547 548 for (c = 0; c < r; c++) 549 nvlist_free(child[c]); 550 551 umem_free(child, r * sizeof (nvlist_t *)); 552 553 return (raidz); 554 } 555 556 static nvlist_t * 557 make_vdev_mirror(size_t size, int r, int m) 558 { 559 nvlist_t *mirror, **child; 560 int c; 561 562 if (m < 1) 563 return (make_vdev_raidz(size, r)); 564 565 child = umem_alloc(m * sizeof (nvlist_t *), UMEM_NOFAIL); 566 567 for (c = 0; c < m; c++) 568 child[c] = make_vdev_raidz(size, r); 569 570 VERIFY(nvlist_alloc(&mirror, NV_UNIQUE_NAME, 0) == 0); 571 VERIFY(nvlist_add_string(mirror, ZPOOL_CONFIG_TYPE, 572 VDEV_TYPE_MIRROR) == 0); 573 VERIFY(nvlist_add_nvlist_array(mirror, ZPOOL_CONFIG_CHILDREN, 574 child, m) == 0); 575 576 for (c = 0; c < m; c++) 577 nvlist_free(child[c]); 578 579 umem_free(child, m * sizeof (nvlist_t *)); 580 581 return (mirror); 582 } 583 584 static nvlist_t * 585 make_vdev_root(size_t size, int r, int m, int t) 586 { 587 nvlist_t *root, **child; 588 int c; 589 590 ASSERT(t > 0); 591 592 child = umem_alloc(t * sizeof (nvlist_t *), UMEM_NOFAIL); 593 594 for (c = 0; c < t; c++) 595 child[c] = make_vdev_mirror(size, r, m); 596 597 VERIFY(nvlist_alloc(&root, NV_UNIQUE_NAME, 0) == 0); 598 VERIFY(nvlist_add_string(root, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) == 0); 599 VERIFY(nvlist_add_nvlist_array(root, ZPOOL_CONFIG_CHILDREN, 600 child, t) == 0); 601 602 for (c = 0; c < t; c++) 603 nvlist_free(child[c]); 604 605 umem_free(child, t * sizeof (nvlist_t *)); 606 607 return (root); 608 } 609 610 static void 611 ztest_set_random_blocksize(objset_t *os, uint64_t object, dmu_tx_t *tx) 612 { 613 int bs = SPA_MINBLOCKSHIFT + 614 ztest_random(SPA_MAXBLOCKSHIFT - SPA_MINBLOCKSHIFT + 1); 615 int ibs = DN_MIN_INDBLKSHIFT + 616 ztest_random(DN_MAX_INDBLKSHIFT - DN_MIN_INDBLKSHIFT + 1); 617 int error; 618 619 error = dmu_object_set_blocksize(os, object, 1ULL << bs, ibs, tx); 620 if (error) { 621 char osname[300]; 622 dmu_objset_name(os, osname); 623 fatal(0, "dmu_object_set_blocksize('%s', %llu, %d, %d) = %d", 624 osname, object, 1 << bs, ibs, error); 625 } 626 } 627 628 static uint8_t 629 ztest_random_checksum(void) 630 { 631 uint8_t checksum; 632 633 do { 634 checksum = ztest_random(ZIO_CHECKSUM_FUNCTIONS); 635 } while (zio_checksum_table[checksum].ci_zbt); 636 637 if (checksum == ZIO_CHECKSUM_OFF) 638 checksum = ZIO_CHECKSUM_ON; 639 640 return (checksum); 641 } 642 643 static uint8_t 644 ztest_random_compress(void) 645 { 646 return ((uint8_t)ztest_random(ZIO_COMPRESS_FUNCTIONS)); 647 } 648 649 typedef struct ztest_replay { 650 objset_t *zr_os; 651 uint64_t zr_assign; 652 } ztest_replay_t; 653 654 static int 655 ztest_replay_create(ztest_replay_t *zr, lr_create_t *lr, boolean_t byteswap) 656 { 657 objset_t *os = zr->zr_os; 658 dmu_tx_t *tx; 659 int error; 660 661 if (byteswap) 662 byteswap_uint64_array(lr, sizeof (*lr)); 663 664 tx = dmu_tx_create(os); 665 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT); 666 error = dmu_tx_assign(tx, zr->zr_assign); 667 if (error) { 668 dmu_tx_abort(tx); 669 return (error); 670 } 671 672 error = dmu_object_claim(os, lr->lr_doid, lr->lr_mode, 0, 673 DMU_OT_NONE, 0, tx); 674 ASSERT(error == 0); 675 dmu_tx_commit(tx); 676 677 if (zopt_verbose >= 5) { 678 char osname[MAXNAMELEN]; 679 dmu_objset_name(os, osname); 680 (void) printf("replay create of %s object %llu" 681 " in txg %llu = %d\n", 682 osname, (u_longlong_t)lr->lr_doid, 683 (u_longlong_t)zr->zr_assign, error); 684 } 685 686 return (error); 687 } 688 689 static int 690 ztest_replay_remove(ztest_replay_t *zr, lr_remove_t *lr, boolean_t byteswap) 691 { 692 objset_t *os = zr->zr_os; 693 dmu_tx_t *tx; 694 int error; 695 696 if (byteswap) 697 byteswap_uint64_array(lr, sizeof (*lr)); 698 699 tx = dmu_tx_create(os); 700 dmu_tx_hold_free(tx, lr->lr_doid, 0, DMU_OBJECT_END); 701 error = dmu_tx_assign(tx, zr->zr_assign); 702 if (error) { 703 dmu_tx_abort(tx); 704 return (error); 705 } 706 707 error = dmu_object_free(os, lr->lr_doid, tx); 708 dmu_tx_commit(tx); 709 710 return (error); 711 } 712 713 zil_replay_func_t *ztest_replay_vector[TX_MAX_TYPE] = { 714 NULL, /* 0 no such transaction type */ 715 ztest_replay_create, /* TX_CREATE */ 716 NULL, /* TX_MKDIR */ 717 NULL, /* TX_MKXATTR */ 718 NULL, /* TX_SYMLINK */ 719 ztest_replay_remove, /* TX_REMOVE */ 720 NULL, /* TX_RMDIR */ 721 NULL, /* TX_LINK */ 722 NULL, /* TX_RENAME */ 723 NULL, /* TX_WRITE */ 724 NULL, /* TX_TRUNCATE */ 725 NULL, /* TX_SETATTR */ 726 NULL, /* TX_ACL */ 727 }; 728 729 /* 730 * Verify that we can't destroy an active pool, create an existing pool, 731 * or create a pool with a bad vdev spec. 732 */ 733 void 734 ztest_spa_create_destroy(ztest_args_t *za) 735 { 736 int error; 737 spa_t *spa; 738 nvlist_t *nvroot; 739 740 /* 741 * Attempt to create using a bad file. 742 */ 743 nvroot = make_vdev_root(0, 0, 0, 1); 744 error = spa_create("ztest_bad_file", nvroot, NULL); 745 nvlist_free(nvroot); 746 if (error != ENOENT) 747 fatal(0, "spa_create(bad_file) = %d", error); 748 749 /* 750 * Attempt to create using a bad mirror. 751 */ 752 nvroot = make_vdev_root(0, 0, 2, 1); 753 error = spa_create("ztest_bad_mirror", nvroot, NULL); 754 nvlist_free(nvroot); 755 if (error != ENOENT) 756 fatal(0, "spa_create(bad_mirror) = %d", error); 757 758 /* 759 * Attempt to create an existing pool. It shouldn't matter 760 * what's in the nvroot; we should fail with EEXIST. 761 */ 762 (void) rw_rdlock(&ztest_shared->zs_name_lock); 763 nvroot = make_vdev_root(0, 0, 0, 1); 764 error = spa_create(za->za_pool, nvroot, NULL); 765 nvlist_free(nvroot); 766 if (error != EEXIST) 767 fatal(0, "spa_create(whatever) = %d", error); 768 769 error = spa_open(za->za_pool, &spa, FTAG); 770 if (error) 771 fatal(0, "spa_open() = %d", error); 772 773 error = spa_destroy(za->za_pool); 774 if (error != EBUSY) 775 fatal(0, "spa_destroy() = %d", error); 776 777 spa_close(spa, FTAG); 778 (void) rw_unlock(&ztest_shared->zs_name_lock); 779 } 780 781 /* 782 * Verify that vdev_add() works as expected. 783 */ 784 void 785 ztest_vdev_add_remove(ztest_args_t *za) 786 { 787 spa_t *spa = dmu_objset_spa(za->za_os); 788 uint64_t leaves = MAX(zopt_mirrors, 1) * zopt_raidz; 789 nvlist_t *nvroot; 790 int error; 791 792 if (zopt_verbose >= 6) 793 (void) printf("adding vdev\n"); 794 795 (void) mutex_lock(&ztest_shared->zs_vdev_lock); 796 797 spa_config_enter(spa, RW_READER, FTAG); 798 799 ztest_shared->zs_vdev_primaries = 800 spa->spa_root_vdev->vdev_children * leaves; 801 802 spa_config_exit(spa, FTAG); 803 804 nvroot = make_vdev_root(zopt_vdev_size, zopt_raidz, zopt_mirrors, 1); 805 error = spa_vdev_add(spa, nvroot); 806 nvlist_free(nvroot); 807 808 (void) mutex_unlock(&ztest_shared->zs_vdev_lock); 809 810 if (error == ENOSPC) 811 ztest_record_enospc("spa_vdev_add"); 812 else if (error != 0) 813 fatal(0, "spa_vdev_add() = %d", error); 814 815 if (zopt_verbose >= 6) 816 (void) printf("spa_vdev_add = %d, as expected\n", error); 817 } 818 819 static vdev_t * 820 vdev_lookup_by_path(vdev_t *vd, const char *path) 821 { 822 int c; 823 vdev_t *mvd; 824 825 if (vd->vdev_path != NULL) { 826 if (vd->vdev_wholedisk == 1) { 827 /* 828 * For whole disks, the internal path has 's0', but the 829 * path passed in by the user doesn't. 830 */ 831 if (strlen(path) == strlen(vd->vdev_path) - 2 && 832 strncmp(path, vd->vdev_path, strlen(path)) == 0) 833 return (vd); 834 } else if (strcmp(path, vd->vdev_path) == 0) { 835 return (vd); 836 } 837 } 838 839 for (c = 0; c < vd->vdev_children; c++) 840 if ((mvd = vdev_lookup_by_path(vd->vdev_child[c], path)) != 841 NULL) 842 return (mvd); 843 844 return (NULL); 845 } 846 847 /* 848 * Verify that we can attach and detach devices. 849 */ 850 void 851 ztest_vdev_attach_detach(ztest_args_t *za) 852 { 853 spa_t *spa = dmu_objset_spa(za->za_os); 854 vdev_t *rvd = spa->spa_root_vdev; 855 vdev_t *oldvd, *newvd, *pvd; 856 nvlist_t *root, *file; 857 uint64_t leaves = MAX(zopt_mirrors, 1) * zopt_raidz; 858 uint64_t leaf, top; 859 uint64_t ashift = ztest_get_ashift(); 860 size_t oldsize, newsize; 861 char oldpath[MAXPATHLEN], newpath[MAXPATHLEN]; 862 int replacing; 863 int error, expected_error; 864 int fd; 865 866 (void) mutex_lock(&ztest_shared->zs_vdev_lock); 867 868 spa_config_enter(spa, RW_READER, FTAG); 869 870 /* 871 * Decide whether to do an attach or a replace. 872 */ 873 replacing = ztest_random(2); 874 875 /* 876 * Pick a random top-level vdev. 877 */ 878 top = ztest_random(rvd->vdev_children); 879 880 /* 881 * Pick a random leaf within it. 882 */ 883 leaf = ztest_random(leaves); 884 885 /* 886 * Generate the path to this leaf. The filename will end with 'a'. 887 * We'll alternate replacements with a filename that ends with 'b'. 888 */ 889 (void) snprintf(oldpath, sizeof (oldpath), 890 ztest_dev_template, zopt_dir, zopt_pool, top * leaves + leaf); 891 892 bcopy(oldpath, newpath, MAXPATHLEN); 893 894 /* 895 * If the 'a' file isn't part of the pool, the 'b' file must be. 896 */ 897 if (vdev_lookup_by_path(rvd, oldpath) == NULL) 898 oldpath[strlen(oldpath) - 1] = 'b'; 899 else 900 newpath[strlen(newpath) - 1] = 'b'; 901 902 /* 903 * Now oldpath represents something that's already in the pool, 904 * and newpath is the thing we'll try to attach. 905 */ 906 oldvd = vdev_lookup_by_path(rvd, oldpath); 907 newvd = vdev_lookup_by_path(rvd, newpath); 908 ASSERT(oldvd != NULL); 909 pvd = oldvd->vdev_parent; 910 911 /* 912 * Make newsize a little bigger or smaller than oldsize. 913 * If it's smaller, the attach should fail. 914 * If it's larger, and we're doing a replace, 915 * we should get dynamic LUN growth when we're done. 916 */ 917 oldsize = vdev_get_rsize(oldvd); 918 newsize = 10 * oldsize / (9 + ztest_random(3)); 919 920 /* 921 * If pvd is not a mirror or root, the attach should fail with ENOTSUP, 922 * unless it's a replace; in that case any non-replacing parent is OK. 923 * 924 * If newvd is already part of the pool, it should fail with EBUSY. 925 * 926 * If newvd is too small, it should fail with EOVERFLOW. 927 */ 928 if (pvd->vdev_ops != &vdev_mirror_ops && 929 pvd->vdev_ops != &vdev_root_ops && 930 (!replacing || pvd->vdev_ops == &vdev_replacing_ops)) 931 expected_error = ENOTSUP; 932 else if (newvd != NULL) 933 expected_error = EBUSY; 934 else if (newsize < oldsize) 935 expected_error = EOVERFLOW; 936 else if (ashift > oldvd->vdev_top->vdev_ashift) 937 expected_error = EDOM; 938 else 939 expected_error = 0; 940 941 /* 942 * If newvd isn't already part of the pool, create it. 943 */ 944 if (newvd == NULL) { 945 fd = open(newpath, O_RDWR | O_CREAT | O_TRUNC, 0666); 946 if (fd == -1) 947 fatal(1, "can't open %s", newpath); 948 if (ftruncate(fd, newsize) != 0) 949 fatal(1, "can't ftruncate %s", newpath); 950 (void) close(fd); 951 } 952 953 spa_config_exit(spa, FTAG); 954 955 /* 956 * Build the nvlist describing newpath. 957 */ 958 VERIFY(nvlist_alloc(&file, NV_UNIQUE_NAME, 0) == 0); 959 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_TYPE, VDEV_TYPE_FILE) == 0); 960 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_PATH, newpath) == 0); 961 VERIFY(nvlist_add_uint64(file, ZPOOL_CONFIG_ASHIFT, ashift) == 0); 962 963 VERIFY(nvlist_alloc(&root, NV_UNIQUE_NAME, 0) == 0); 964 VERIFY(nvlist_add_string(root, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) == 0); 965 VERIFY(nvlist_add_nvlist_array(root, ZPOOL_CONFIG_CHILDREN, 966 &file, 1) == 0); 967 968 error = spa_vdev_attach(spa, oldvd->vdev_guid, root, replacing); 969 970 nvlist_free(file); 971 nvlist_free(root); 972 973 /* 974 * If our parent was the replacing vdev, but the replace completed, 975 * then instead of failing with ENOTSUP we may either succeed, 976 * fail with ENODEV, or fail with EOVERFLOW. 977 */ 978 if (expected_error == ENOTSUP && 979 (error == 0 || error == ENODEV || error == EOVERFLOW)) 980 expected_error = error; 981 982 /* 983 * If someone grew the LUN, the replacement may be too small. 984 */ 985 if (error == EOVERFLOW) 986 expected_error = error; 987 988 if (error != expected_error) { 989 fatal(0, "attach (%s, %s, %d) returned %d, expected %d", 990 oldpath, newpath, replacing, error, expected_error); 991 } 992 993 (void) mutex_unlock(&ztest_shared->zs_vdev_lock); 994 } 995 996 /* 997 * Verify that dynamic LUN growth works as expected. 998 */ 999 /* ARGSUSED */ 1000 void 1001 ztest_vdev_LUN_growth(ztest_args_t *za) 1002 { 1003 spa_t *spa = dmu_objset_spa(za->za_os); 1004 char dev_name[MAXPATHLEN]; 1005 uint64_t leaves = MAX(zopt_mirrors, 1) * zopt_raidz; 1006 uint64_t vdev; 1007 size_t fsize; 1008 int fd; 1009 1010 (void) mutex_lock(&ztest_shared->zs_vdev_lock); 1011 1012 /* 1013 * Pick a random leaf vdev. 1014 */ 1015 spa_config_enter(spa, RW_READER, FTAG); 1016 vdev = ztest_random(spa->spa_root_vdev->vdev_children * leaves); 1017 spa_config_exit(spa, FTAG); 1018 1019 (void) sprintf(dev_name, ztest_dev_template, zopt_dir, zopt_pool, vdev); 1020 1021 if ((fd = open(dev_name, O_RDWR)) != -1) { 1022 /* 1023 * Determine the size. 1024 */ 1025 fsize = lseek(fd, 0, SEEK_END); 1026 1027 /* 1028 * If it's less than 2x the original size, grow by around 3%. 1029 */ 1030 if (fsize < 2 * zopt_vdev_size) { 1031 size_t newsize = fsize + ztest_random(fsize / 32); 1032 (void) ftruncate(fd, newsize); 1033 if (zopt_verbose >= 6) { 1034 (void) printf("%s grew from %lu to %lu bytes\n", 1035 dev_name, (ulong_t)fsize, (ulong_t)newsize); 1036 } 1037 } 1038 (void) close(fd); 1039 } 1040 1041 (void) mutex_unlock(&ztest_shared->zs_vdev_lock); 1042 } 1043 1044 /* ARGSUSED */ 1045 static void 1046 ztest_create_cb(objset_t *os, void *arg, dmu_tx_t *tx) 1047 { 1048 /* 1049 * Create the directory object. 1050 */ 1051 VERIFY(dmu_object_claim(os, ZTEST_DIROBJ, 1052 DMU_OT_UINT64_OTHER, ZTEST_DIROBJ_BLOCKSIZE, 1053 DMU_OT_UINT64_OTHER, sizeof (ztest_block_tag_t), tx) == 0); 1054 1055 VERIFY(zap_create_claim(os, ZTEST_MICROZAP_OBJ, 1056 DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx) == 0); 1057 1058 VERIFY(zap_create_claim(os, ZTEST_FATZAP_OBJ, 1059 DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx) == 0); 1060 } 1061 1062 /* ARGSUSED */ 1063 static void 1064 ztest_destroy_cb(char *name, void *arg) 1065 { 1066 objset_t *os; 1067 dmu_object_info_t doi; 1068 int error; 1069 1070 /* 1071 * Verify that the dataset contains a directory object. 1072 */ 1073 error = dmu_objset_open(name, DMU_OST_OTHER, 1074 DS_MODE_STANDARD | DS_MODE_READONLY, &os); 1075 ASSERT3U(error, ==, 0); 1076 error = dmu_object_info(os, ZTEST_DIROBJ, &doi); 1077 if (error != ENOENT) { 1078 /* We could have crashed in the middle of destroying it */ 1079 ASSERT3U(error, ==, 0); 1080 ASSERT3U(doi.doi_type, ==, DMU_OT_UINT64_OTHER); 1081 ASSERT3S(doi.doi_physical_blks, >=, 0); 1082 } 1083 dmu_objset_close(os); 1084 1085 /* 1086 * Destroy the dataset. 1087 */ 1088 error = dmu_objset_destroy(name); 1089 ASSERT3U(error, ==, 0); 1090 } 1091 1092 /* 1093 * Verify that dmu_objset_{create,destroy,open,close} work as expected. 1094 */ 1095 static uint64_t 1096 ztest_log_create(zilog_t *zilog, dmu_tx_t *tx, uint64_t object, int mode) 1097 { 1098 itx_t *itx; 1099 lr_create_t *lr; 1100 size_t namesize; 1101 char name[24]; 1102 1103 (void) sprintf(name, "ZOBJ_%llu", (u_longlong_t)object); 1104 namesize = strlen(name) + 1; 1105 1106 itx = zil_itx_create(TX_CREATE, sizeof (*lr) + namesize + 1107 ztest_random(ZIL_MAX_BLKSZ)); 1108 lr = (lr_create_t *)&itx->itx_lr; 1109 bzero(lr + 1, lr->lr_common.lrc_reclen - sizeof (*lr)); 1110 lr->lr_doid = object; 1111 lr->lr_foid = 0; 1112 lr->lr_mode = mode; 1113 lr->lr_uid = 0; 1114 lr->lr_gid = 0; 1115 lr->lr_gen = dmu_tx_get_txg(tx); 1116 lr->lr_crtime[0] = time(NULL); 1117 lr->lr_crtime[1] = 0; 1118 lr->lr_rdev = 0; 1119 bcopy(name, (char *)(lr + 1), namesize); 1120 1121 return (zil_itx_assign(zilog, itx, tx)); 1122 } 1123 1124 #ifndef lint 1125 static uint64_t 1126 ztest_log_remove(zilog_t *zilog, dmu_tx_t *tx, uint64_t object) 1127 { 1128 itx_t *itx; 1129 lr_remove_t *lr; 1130 size_t namesize; 1131 char name[24]; 1132 1133 (void) sprintf(name, "ZOBJ_%llu", (u_longlong_t)object); 1134 namesize = strlen(name) + 1; 1135 1136 itx = zil_itx_create(TX_REMOVE, sizeof (*lr) + namesize + 1137 ztest_random(8000)); 1138 lr = (lr_remove_t *)&itx->itx_lr; 1139 lr->lr_doid = object; 1140 bcopy(name, (char *)(lr + 1), namesize); 1141 1142 return (zil_itx_assign(zilog, itx, tx)); 1143 } 1144 #endif /* lint */ 1145 1146 void 1147 ztest_dmu_objset_create_destroy(ztest_args_t *za) 1148 { 1149 int error; 1150 objset_t *os; 1151 char name[100]; 1152 int mode, basemode, expected_error; 1153 zilog_t *zilog; 1154 uint64_t seq; 1155 uint64_t objects; 1156 ztest_replay_t zr; 1157 1158 (void) rw_rdlock(&ztest_shared->zs_name_lock); 1159 (void) snprintf(name, 100, "%s/%s_temp_%llu", za->za_pool, za->za_pool, 1160 (u_longlong_t)za->za_instance); 1161 1162 basemode = DS_MODE_LEVEL(za->za_instance); 1163 if (basemode == DS_MODE_NONE) 1164 basemode++; 1165 1166 /* 1167 * If this dataset exists from a previous run, process its replay log 1168 * half of the time. If we don't replay it, then dmu_objset_destroy() 1169 * (invoked from ztest_destroy_cb() below) should just throw it away. 1170 */ 1171 if (ztest_random(2) == 0 && 1172 dmu_objset_open(name, DMU_OST_OTHER, DS_MODE_PRIMARY, &os) == 0) { 1173 zr.zr_os = os; 1174 zil_replay(os, &zr, &zr.zr_assign, ztest_replay_vector, NULL); 1175 dmu_objset_close(os); 1176 } 1177 1178 /* 1179 * There may be an old instance of the dataset we're about to 1180 * create lying around from a previous run. If so, destroy it 1181 * and all of its snapshots. 1182 */ 1183 dmu_objset_find(name, ztest_destroy_cb, NULL, DS_FIND_SNAPSHOTS); 1184 1185 /* 1186 * Verify that the destroyed dataset is no longer in the namespace. 1187 */ 1188 error = dmu_objset_open(name, DMU_OST_OTHER, basemode, &os); 1189 if (error != ENOENT) 1190 fatal(1, "dmu_objset_open(%s) found destroyed dataset %p", 1191 name, os); 1192 1193 /* 1194 * Verify that we can create a new dataset. 1195 */ 1196 error = dmu_objset_create(name, DMU_OST_OTHER, NULL, ztest_create_cb, 1197 NULL); 1198 if (error) { 1199 if (error == ENOSPC) { 1200 ztest_record_enospc("dmu_objset_create"); 1201 (void) rw_unlock(&ztest_shared->zs_name_lock); 1202 return; 1203 } 1204 fatal(0, "dmu_objset_create(%s) = %d", name, error); 1205 } 1206 1207 error = dmu_objset_open(name, DMU_OST_OTHER, basemode, &os); 1208 if (error) { 1209 fatal(0, "dmu_objset_open(%s) = %d", name, error); 1210 } 1211 1212 /* 1213 * Open the intent log for it. 1214 */ 1215 zilog = zil_open(os, NULL); 1216 1217 /* 1218 * Put a random number of objects in there. 1219 */ 1220 objects = ztest_random(20); 1221 seq = 0; 1222 while (objects-- != 0) { 1223 uint64_t object; 1224 dmu_tx_t *tx = dmu_tx_create(os); 1225 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, sizeof (name)); 1226 error = dmu_tx_assign(tx, TXG_WAIT); 1227 if (error) { 1228 dmu_tx_abort(tx); 1229 } else { 1230 object = dmu_object_alloc(os, DMU_OT_UINT64_OTHER, 0, 1231 DMU_OT_NONE, 0, tx); 1232 ztest_set_random_blocksize(os, object, tx); 1233 seq = ztest_log_create(zilog, tx, object, 1234 DMU_OT_UINT64_OTHER); 1235 dmu_write(os, object, 0, sizeof (name), name, tx); 1236 dmu_tx_commit(tx); 1237 } 1238 if (ztest_random(5) == 0) { 1239 zil_commit(zilog, seq, FSYNC); 1240 } 1241 if (ztest_random(100) == 0) { 1242 error = zil_suspend(zilog); 1243 if (error == 0) { 1244 zil_resume(zilog); 1245 } 1246 } 1247 } 1248 1249 /* 1250 * Verify that we cannot create an existing dataset. 1251 */ 1252 error = dmu_objset_create(name, DMU_OST_OTHER, NULL, NULL, NULL); 1253 if (error != EEXIST) 1254 fatal(0, "created existing dataset, error = %d", error); 1255 1256 /* 1257 * Verify that multiple dataset opens are allowed, but only when 1258 * the new access mode is compatible with the base mode. 1259 * We use a mixture of typed and typeless opens, and when the 1260 * open succeeds, verify that the discovered type is correct. 1261 */ 1262 for (mode = DS_MODE_STANDARD; mode < DS_MODE_LEVELS; mode++) { 1263 objset_t *os2; 1264 error = dmu_objset_open(name, DMU_OST_OTHER, mode, &os2); 1265 expected_error = (basemode + mode < DS_MODE_LEVELS) ? 0 : EBUSY; 1266 if (error != expected_error) 1267 fatal(0, "dmu_objset_open('%s') = %d, expected %d", 1268 name, error, expected_error); 1269 if (error == 0) 1270 dmu_objset_close(os2); 1271 } 1272 1273 zil_close(zilog); 1274 dmu_objset_close(os); 1275 1276 error = dmu_objset_destroy(name); 1277 if (error) 1278 fatal(0, "dmu_objset_destroy(%s) = %d", name, error); 1279 1280 (void) rw_unlock(&ztest_shared->zs_name_lock); 1281 } 1282 1283 /* 1284 * Verify that dmu_snapshot_{create,destroy,open,close} work as expected. 1285 */ 1286 void 1287 ztest_dmu_snapshot_create_destroy(ztest_args_t *za) 1288 { 1289 int error; 1290 objset_t *os = za->za_os; 1291 char snapname[100]; 1292 char osname[MAXNAMELEN]; 1293 1294 (void) rw_rdlock(&ztest_shared->zs_name_lock); 1295 dmu_objset_name(os, osname); 1296 (void) snprintf(snapname, 100, "%s@%llu", osname, 1297 (u_longlong_t)za->za_instance); 1298 1299 error = dmu_objset_destroy(snapname); 1300 if (error != 0 && error != ENOENT) 1301 fatal(0, "dmu_objset_destroy() = %d", error); 1302 error = dmu_objset_create(snapname, DMU_OST_OTHER, NULL, NULL, NULL); 1303 if (error == ENOSPC) 1304 ztest_record_enospc("dmu_take_snapshot"); 1305 else if (error != 0 && error != EEXIST) 1306 fatal(0, "dmu_take_snapshot() = %d", error); 1307 (void) rw_unlock(&ztest_shared->zs_name_lock); 1308 } 1309 1310 #define ZTEST_TRAVERSE_BLOCKS 1000 1311 1312 static int 1313 ztest_blk_cb(traverse_blk_cache_t *bc, spa_t *spa, void *arg) 1314 { 1315 ztest_args_t *za = arg; 1316 zbookmark_t *zb = &bc->bc_bookmark; 1317 blkptr_t *bp = &bc->bc_blkptr; 1318 dnode_phys_t *dnp = bc->bc_dnode; 1319 traverse_handle_t *th = za->za_th; 1320 uint64_t size = BP_GET_LSIZE(bp); 1321 1322 /* 1323 * Level -1 indicates the objset_phys_t or something in its intent log. 1324 */ 1325 if (zb->zb_level == -1) { 1326 if (BP_GET_TYPE(bp) == DMU_OT_OBJSET) { 1327 ASSERT3U(zb->zb_object, ==, 0); 1328 ASSERT3U(zb->zb_blkid, ==, 0); 1329 ASSERT3U(size, ==, sizeof (objset_phys_t)); 1330 za->za_zil_seq = 0; 1331 } else if (BP_GET_TYPE(bp) == DMU_OT_INTENT_LOG) { 1332 ASSERT3U(zb->zb_object, ==, 0); 1333 ASSERT3U(zb->zb_blkid, >, za->za_zil_seq); 1334 za->za_zil_seq = zb->zb_blkid; 1335 } else { 1336 ASSERT3U(zb->zb_object, !=, 0); /* lr_write_t */ 1337 } 1338 1339 return (0); 1340 } 1341 1342 ASSERT(dnp != NULL); 1343 1344 if (bc->bc_errno) 1345 return (ERESTART); 1346 1347 /* 1348 * Once in a while, abort the traverse. We only do this to odd 1349 * instance numbers to ensure that even ones can run to completion. 1350 */ 1351 if ((za->za_instance & 1) && ztest_random(10000) == 0) 1352 return (EINTR); 1353 1354 if (bp->blk_birth == 0) { 1355 ASSERT(th->th_advance & ADVANCE_HOLES); 1356 return (0); 1357 } 1358 1359 if (zb->zb_level == 0 && !(th->th_advance & ADVANCE_DATA) && 1360 bc == &th->th_cache[ZB_DN_CACHE][0]) { 1361 ASSERT(bc->bc_data == NULL); 1362 return (0); 1363 } 1364 1365 ASSERT(bc->bc_data != NULL); 1366 1367 /* 1368 * This is an expensive question, so don't ask it too often. 1369 */ 1370 if (((za->za_random ^ th->th_callbacks) & 0xff) == 0) { 1371 void *xbuf = umem_alloc(size, UMEM_NOFAIL); 1372 if (arc_tryread(spa, bp, xbuf) == 0) { 1373 ASSERT(bcmp(bc->bc_data, xbuf, size) == 0); 1374 } 1375 umem_free(xbuf, size); 1376 } 1377 1378 if (zb->zb_level > 0) { 1379 ASSERT3U(size, ==, 1ULL << dnp->dn_indblkshift); 1380 return (0); 1381 } 1382 1383 ASSERT(zb->zb_level == 0); 1384 ASSERT3U(size, ==, dnp->dn_datablkszsec << DEV_BSHIFT); 1385 1386 return (0); 1387 } 1388 1389 /* 1390 * Verify that live pool traversal works. 1391 */ 1392 void 1393 ztest_traverse(ztest_args_t *za) 1394 { 1395 spa_t *spa = dmu_objset_spa(za->za_os); 1396 traverse_handle_t *th = za->za_th; 1397 int rc, advance; 1398 uint64_t cbstart, cblimit; 1399 1400 if (th == NULL) { 1401 advance = 0; 1402 1403 if (ztest_random(2) == 0) 1404 advance |= ADVANCE_PRE; 1405 1406 if (ztest_random(2) == 0) 1407 advance |= ADVANCE_PRUNE; 1408 1409 if (ztest_random(2) == 0) 1410 advance |= ADVANCE_DATA; 1411 1412 if (ztest_random(2) == 0) 1413 advance |= ADVANCE_HOLES; 1414 1415 if (ztest_random(2) == 0) 1416 advance |= ADVANCE_ZIL; 1417 1418 th = za->za_th = traverse_init(spa, ztest_blk_cb, za, advance, 1419 ZIO_FLAG_CANFAIL); 1420 1421 traverse_add_pool(th, 0, -1ULL); 1422 } 1423 1424 advance = th->th_advance; 1425 cbstart = th->th_callbacks; 1426 cblimit = cbstart + ((advance & ADVANCE_DATA) ? 100 : 1000); 1427 1428 while ((rc = traverse_more(th)) == EAGAIN && th->th_callbacks < cblimit) 1429 continue; 1430 1431 if (zopt_verbose >= 5) 1432 (void) printf("traverse %s%s%s%s %llu blocks to " 1433 "<%llu, %llu, %lld, %llx>%s\n", 1434 (advance & ADVANCE_PRE) ? "pre" : "post", 1435 (advance & ADVANCE_PRUNE) ? "|prune" : "", 1436 (advance & ADVANCE_DATA) ? "|data" : "", 1437 (advance & ADVANCE_HOLES) ? "|holes" : "", 1438 (u_longlong_t)(th->th_callbacks - cbstart), 1439 (u_longlong_t)th->th_lastcb.zb_objset, 1440 (u_longlong_t)th->th_lastcb.zb_object, 1441 (u_longlong_t)th->th_lastcb.zb_level, 1442 (u_longlong_t)th->th_lastcb.zb_blkid, 1443 rc == 0 ? " [done]" : 1444 rc == EINTR ? " [aborted]" : 1445 rc == EAGAIN ? "" : 1446 strerror(rc)); 1447 1448 if (rc != EAGAIN) { 1449 if (rc != 0 && rc != EINTR) 1450 fatal(0, "traverse_more(%p) = %d", th, rc); 1451 traverse_fini(th); 1452 za->za_th = NULL; 1453 } 1454 } 1455 1456 /* 1457 * Verify that dmu_object_{alloc,free} work as expected. 1458 */ 1459 void 1460 ztest_dmu_object_alloc_free(ztest_args_t *za) 1461 { 1462 objset_t *os = za->za_os; 1463 dmu_buf_t *db; 1464 dmu_tx_t *tx; 1465 uint64_t batchobj, object, batchsize, endoff, temp; 1466 int b, c, error, bonuslen; 1467 dmu_object_info_t doi; 1468 char osname[MAXNAMELEN]; 1469 1470 dmu_objset_name(os, osname); 1471 1472 endoff = -8ULL; 1473 batchsize = 2; 1474 1475 /* 1476 * Create a batch object if necessary, and record it in the directory. 1477 */ 1478 VERIFY(0 == dmu_read(os, ZTEST_DIROBJ, za->za_diroff, 1479 sizeof (uint64_t), &batchobj)); 1480 if (batchobj == 0) { 1481 tx = dmu_tx_create(os); 1482 dmu_tx_hold_write(tx, ZTEST_DIROBJ, za->za_diroff, 1483 sizeof (uint64_t)); 1484 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT); 1485 error = dmu_tx_assign(tx, TXG_WAIT); 1486 if (error) { 1487 ztest_record_enospc("create a batch object"); 1488 dmu_tx_abort(tx); 1489 return; 1490 } 1491 batchobj = dmu_object_alloc(os, DMU_OT_UINT64_OTHER, 0, 1492 DMU_OT_NONE, 0, tx); 1493 ztest_set_random_blocksize(os, batchobj, tx); 1494 dmu_write(os, ZTEST_DIROBJ, za->za_diroff, 1495 sizeof (uint64_t), &batchobj, tx); 1496 dmu_tx_commit(tx); 1497 } 1498 1499 /* 1500 * Destroy the previous batch of objects. 1501 */ 1502 for (b = 0; b < batchsize; b++) { 1503 VERIFY(0 == dmu_read(os, batchobj, b * sizeof (uint64_t), 1504 sizeof (uint64_t), &object)); 1505 if (object == 0) 1506 continue; 1507 /* 1508 * Read and validate contents. 1509 * We expect the nth byte of the bonus buffer to be n. 1510 */ 1511 VERIFY(0 == dmu_bonus_hold(os, object, FTAG, &db)); 1512 1513 dmu_object_info_from_db(db, &doi); 1514 ASSERT(doi.doi_type == DMU_OT_UINT64_OTHER); 1515 ASSERT(doi.doi_bonus_type == DMU_OT_PLAIN_OTHER); 1516 ASSERT3S(doi.doi_physical_blks, >=, 0); 1517 1518 bonuslen = db->db_size; 1519 1520 for (c = 0; c < bonuslen; c++) { 1521 if (((uint8_t *)db->db_data)[c] != 1522 (uint8_t)(c + bonuslen)) { 1523 fatal(0, 1524 "bad bonus: %s, obj %llu, off %d: %u != %u", 1525 osname, object, c, 1526 ((uint8_t *)db->db_data)[c], 1527 (uint8_t)(c + bonuslen)); 1528 } 1529 } 1530 1531 dmu_buf_rele(db, FTAG); 1532 1533 /* 1534 * We expect the word at endoff to be our object number. 1535 */ 1536 VERIFY(0 == dmu_read(os, object, endoff, 1537 sizeof (uint64_t), &temp)); 1538 1539 if (temp != object) { 1540 fatal(0, "bad data in %s, got %llu, expected %llu", 1541 osname, temp, object); 1542 } 1543 1544 /* 1545 * Destroy old object and clear batch entry. 1546 */ 1547 tx = dmu_tx_create(os); 1548 dmu_tx_hold_write(tx, batchobj, 1549 b * sizeof (uint64_t), sizeof (uint64_t)); 1550 dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END); 1551 error = dmu_tx_assign(tx, TXG_WAIT); 1552 if (error) { 1553 ztest_record_enospc("free object"); 1554 dmu_tx_abort(tx); 1555 return; 1556 } 1557 error = dmu_object_free(os, object, tx); 1558 if (error) { 1559 fatal(0, "dmu_object_free('%s', %llu) = %d", 1560 osname, object, error); 1561 } 1562 object = 0; 1563 1564 dmu_object_set_checksum(os, batchobj, 1565 ztest_random_checksum(), tx); 1566 dmu_object_set_compress(os, batchobj, 1567 ztest_random_compress(), tx); 1568 1569 dmu_write(os, batchobj, b * sizeof (uint64_t), 1570 sizeof (uint64_t), &object, tx); 1571 1572 dmu_tx_commit(tx); 1573 } 1574 1575 /* 1576 * Before creating the new batch of objects, generate a bunch of churn. 1577 */ 1578 for (b = ztest_random(100); b > 0; b--) { 1579 tx = dmu_tx_create(os); 1580 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT); 1581 error = dmu_tx_assign(tx, TXG_WAIT); 1582 if (error) { 1583 ztest_record_enospc("churn objects"); 1584 dmu_tx_abort(tx); 1585 return; 1586 } 1587 object = dmu_object_alloc(os, DMU_OT_UINT64_OTHER, 0, 1588 DMU_OT_NONE, 0, tx); 1589 ztest_set_random_blocksize(os, object, tx); 1590 error = dmu_object_free(os, object, tx); 1591 if (error) { 1592 fatal(0, "dmu_object_free('%s', %llu) = %d", 1593 osname, object, error); 1594 } 1595 dmu_tx_commit(tx); 1596 } 1597 1598 /* 1599 * Create a new batch of objects with randomly chosen 1600 * blocksizes and record them in the batch directory. 1601 */ 1602 for (b = 0; b < batchsize; b++) { 1603 uint32_t va_blksize; 1604 u_longlong_t va_nblocks; 1605 1606 tx = dmu_tx_create(os); 1607 dmu_tx_hold_write(tx, batchobj, b * sizeof (uint64_t), 1608 sizeof (uint64_t)); 1609 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT); 1610 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, endoff, 1611 sizeof (uint64_t)); 1612 error = dmu_tx_assign(tx, TXG_WAIT); 1613 if (error) { 1614 ztest_record_enospc("create batchobj"); 1615 dmu_tx_abort(tx); 1616 return; 1617 } 1618 bonuslen = (int)ztest_random(dmu_bonus_max()) + 1; 1619 1620 object = dmu_object_alloc(os, DMU_OT_UINT64_OTHER, 0, 1621 DMU_OT_PLAIN_OTHER, bonuslen, tx); 1622 1623 ztest_set_random_blocksize(os, object, tx); 1624 1625 dmu_object_set_checksum(os, object, 1626 ztest_random_checksum(), tx); 1627 dmu_object_set_compress(os, object, 1628 ztest_random_compress(), tx); 1629 1630 dmu_write(os, batchobj, b * sizeof (uint64_t), 1631 sizeof (uint64_t), &object, tx); 1632 1633 /* 1634 * Write to both the bonus buffer and the regular data. 1635 */ 1636 VERIFY(0 == dmu_bonus_hold(os, object, FTAG, &db)); 1637 ASSERT3U(bonuslen, ==, db->db_size); 1638 1639 dmu_object_size_from_db(db, &va_blksize, &va_nblocks); 1640 ASSERT3S(va_nblocks, >=, 0); 1641 1642 dmu_buf_will_dirty(db, tx); 1643 1644 /* 1645 * See comments above regarding the contents of 1646 * the bonus buffer and the word at endoff. 1647 */ 1648 for (c = 0; c < db->db_size; c++) 1649 ((uint8_t *)db->db_data)[c] = (uint8_t)(c + bonuslen); 1650 1651 dmu_buf_rele(db, FTAG); 1652 1653 /* 1654 * Write to a large offset to increase indirection. 1655 */ 1656 dmu_write(os, object, endoff, sizeof (uint64_t), &object, tx); 1657 1658 dmu_tx_commit(tx); 1659 } 1660 } 1661 1662 /* 1663 * Verify that dmu_{read,write} work as expected. 1664 */ 1665 typedef struct bufwad { 1666 uint64_t bw_index; 1667 uint64_t bw_txg; 1668 uint64_t bw_data; 1669 } bufwad_t; 1670 1671 typedef struct dmu_read_write_dir { 1672 uint64_t dd_packobj; 1673 uint64_t dd_bigobj; 1674 uint64_t dd_chunk; 1675 } dmu_read_write_dir_t; 1676 1677 void 1678 ztest_dmu_read_write(ztest_args_t *za) 1679 { 1680 objset_t *os = za->za_os; 1681 dmu_read_write_dir_t dd; 1682 dmu_tx_t *tx; 1683 int i, freeit, error; 1684 uint64_t n, s, txg; 1685 bufwad_t *packbuf, *bigbuf, *pack, *bigH, *bigT; 1686 uint64_t packoff, packsize, bigoff, bigsize; 1687 uint64_t regions = 997; 1688 uint64_t stride = 123456789ULL; 1689 uint64_t width = 40; 1690 int free_percent = 5; 1691 1692 /* 1693 * This test uses two objects, packobj and bigobj, that are always 1694 * updated together (i.e. in the same tx) so that their contents are 1695 * in sync and can be compared. Their contents relate to each other 1696 * in a simple way: packobj is a dense array of 'bufwad' structures, 1697 * while bigobj is a sparse array of the same bufwads. Specifically, 1698 * for any index n, there are three bufwads that should be identical: 1699 * 1700 * packobj, at offset n * sizeof (bufwad_t) 1701 * bigobj, at the head of the nth chunk 1702 * bigobj, at the tail of the nth chunk 1703 * 1704 * The chunk size is arbitrary. It doesn't have to be a power of two, 1705 * and it doesn't have any relation to the object blocksize. 1706 * The only requirement is that it can hold at least two bufwads. 1707 * 1708 * Normally, we write the bufwad to each of these locations. 1709 * However, free_percent of the time we instead write zeroes to 1710 * packobj and perform a dmu_free_range() on bigobj. By comparing 1711 * bigobj to packobj, we can verify that the DMU is correctly 1712 * tracking which parts of an object are allocated and free, 1713 * and that the contents of the allocated blocks are correct. 1714 */ 1715 1716 /* 1717 * Read the directory info. If it's the first time, set things up. 1718 */ 1719 VERIFY(0 == dmu_read(os, ZTEST_DIROBJ, za->za_diroff, 1720 sizeof (dd), &dd)); 1721 if (dd.dd_chunk == 0) { 1722 ASSERT(dd.dd_packobj == 0); 1723 ASSERT(dd.dd_bigobj == 0); 1724 tx = dmu_tx_create(os); 1725 dmu_tx_hold_write(tx, ZTEST_DIROBJ, za->za_diroff, sizeof (dd)); 1726 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT); 1727 error = dmu_tx_assign(tx, TXG_WAIT); 1728 if (error) { 1729 ztest_record_enospc("create r/w directory"); 1730 dmu_tx_abort(tx); 1731 return; 1732 } 1733 1734 dd.dd_packobj = dmu_object_alloc(os, DMU_OT_UINT64_OTHER, 0, 1735 DMU_OT_NONE, 0, tx); 1736 dd.dd_bigobj = dmu_object_alloc(os, DMU_OT_UINT64_OTHER, 0, 1737 DMU_OT_NONE, 0, tx); 1738 dd.dd_chunk = (1000 + ztest_random(1000)) * sizeof (uint64_t); 1739 1740 ztest_set_random_blocksize(os, dd.dd_packobj, tx); 1741 ztest_set_random_blocksize(os, dd.dd_bigobj, tx); 1742 1743 dmu_write(os, ZTEST_DIROBJ, za->za_diroff, sizeof (dd), &dd, 1744 tx); 1745 dmu_tx_commit(tx); 1746 } 1747 1748 /* 1749 * Prefetch a random chunk of the big object. 1750 * Our aim here is to get some async reads in flight 1751 * for blocks that we may free below; the DMU should 1752 * handle this race correctly. 1753 */ 1754 n = ztest_random(regions) * stride + ztest_random(width); 1755 s = 1 + ztest_random(2 * width - 1); 1756 dmu_prefetch(os, dd.dd_bigobj, n * dd.dd_chunk, s * dd.dd_chunk); 1757 1758 /* 1759 * Pick a random index and compute the offsets into packobj and bigobj. 1760 */ 1761 n = ztest_random(regions) * stride + ztest_random(width); 1762 s = 1 + ztest_random(width - 1); 1763 1764 packoff = n * sizeof (bufwad_t); 1765 packsize = s * sizeof (bufwad_t); 1766 1767 bigoff = n * dd.dd_chunk; 1768 bigsize = s * dd.dd_chunk; 1769 1770 packbuf = umem_alloc(packsize, UMEM_NOFAIL); 1771 bigbuf = umem_alloc(bigsize, UMEM_NOFAIL); 1772 1773 /* 1774 * free_percent of the time, free a range of bigobj rather than 1775 * overwriting it. 1776 */ 1777 freeit = (ztest_random(100) < free_percent); 1778 1779 /* 1780 * Read the current contents of our objects. 1781 */ 1782 error = dmu_read(os, dd.dd_packobj, packoff, packsize, packbuf); 1783 ASSERT3U(error, ==, 0); 1784 error = dmu_read(os, dd.dd_bigobj, bigoff, bigsize, bigbuf); 1785 ASSERT3U(error, ==, 0); 1786 1787 /* 1788 * Get a tx for the mods to both packobj and bigobj. 1789 */ 1790 tx = dmu_tx_create(os); 1791 1792 dmu_tx_hold_write(tx, dd.dd_packobj, packoff, packsize); 1793 1794 if (freeit) 1795 dmu_tx_hold_free(tx, dd.dd_bigobj, bigoff, bigsize); 1796 else 1797 dmu_tx_hold_write(tx, dd.dd_bigobj, bigoff, bigsize); 1798 1799 error = dmu_tx_assign(tx, TXG_WAIT); 1800 1801 if (error) { 1802 ztest_record_enospc("dmu r/w range"); 1803 dmu_tx_abort(tx); 1804 umem_free(packbuf, packsize); 1805 umem_free(bigbuf, bigsize); 1806 return; 1807 } 1808 1809 txg = dmu_tx_get_txg(tx); 1810 1811 /* 1812 * For each index from n to n + s, verify that the existing bufwad 1813 * in packobj matches the bufwads at the head and tail of the 1814 * corresponding chunk in bigobj. Then update all three bufwads 1815 * with the new values we want to write out. 1816 */ 1817 for (i = 0; i < s; i++) { 1818 /* LINTED */ 1819 pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t)); 1820 /* LINTED */ 1821 bigH = (bufwad_t *)((char *)bigbuf + i * dd.dd_chunk); 1822 /* LINTED */ 1823 bigT = (bufwad_t *)((char *)bigH + dd.dd_chunk) - 1; 1824 1825 ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize); 1826 ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize); 1827 1828 if (pack->bw_txg > txg) 1829 fatal(0, "future leak: got %llx, open txg is %llx", 1830 pack->bw_txg, txg); 1831 1832 if (pack->bw_data != 0 && pack->bw_index != n + i) 1833 fatal(0, "wrong index: got %llx, wanted %llx+%llx", 1834 pack->bw_index, n, i); 1835 1836 if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0) 1837 fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH); 1838 1839 if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0) 1840 fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT); 1841 1842 if (freeit) { 1843 bzero(pack, sizeof (bufwad_t)); 1844 } else { 1845 pack->bw_index = n + i; 1846 pack->bw_txg = txg; 1847 pack->bw_data = 1 + ztest_random(-2ULL); 1848 } 1849 *bigH = *pack; 1850 *bigT = *pack; 1851 } 1852 1853 /* 1854 * We've verified all the old bufwads, and made new ones. 1855 * Now write them out. 1856 */ 1857 dmu_write(os, dd.dd_packobj, packoff, packsize, packbuf, tx); 1858 1859 if (freeit) { 1860 if (zopt_verbose >= 6) { 1861 (void) printf("freeing offset %llx size %llx" 1862 " txg %llx\n", 1863 (u_longlong_t)bigoff, 1864 (u_longlong_t)bigsize, 1865 (u_longlong_t)txg); 1866 } 1867 VERIFY(0 == dmu_free_range(os, dd.dd_bigobj, bigoff, 1868 bigsize, tx)); 1869 } else { 1870 if (zopt_verbose >= 6) { 1871 (void) printf("writing offset %llx size %llx" 1872 " txg %llx\n", 1873 (u_longlong_t)bigoff, 1874 (u_longlong_t)bigsize, 1875 (u_longlong_t)txg); 1876 } 1877 dmu_write(os, dd.dd_bigobj, bigoff, bigsize, bigbuf, tx); 1878 } 1879 1880 dmu_tx_commit(tx); 1881 1882 /* 1883 * Sanity check the stuff we just wrote. 1884 */ 1885 { 1886 void *packcheck = umem_alloc(packsize, UMEM_NOFAIL); 1887 void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL); 1888 1889 VERIFY(0 == dmu_read(os, dd.dd_packobj, packoff, 1890 packsize, packcheck)); 1891 VERIFY(0 == dmu_read(os, dd.dd_bigobj, bigoff, 1892 bigsize, bigcheck)); 1893 1894 ASSERT(bcmp(packbuf, packcheck, packsize) == 0); 1895 ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0); 1896 1897 umem_free(packcheck, packsize); 1898 umem_free(bigcheck, bigsize); 1899 } 1900 1901 umem_free(packbuf, packsize); 1902 umem_free(bigbuf, bigsize); 1903 } 1904 1905 void 1906 ztest_dmu_write_parallel(ztest_args_t *za) 1907 { 1908 objset_t *os = za->za_os; 1909 dmu_tx_t *tx; 1910 dmu_buf_t *db; 1911 int i, b, error, do_free, bs; 1912 uint64_t off, txg_how, txg; 1913 mutex_t *lp; 1914 char osname[MAXNAMELEN]; 1915 char iobuf[SPA_MAXBLOCKSIZE]; 1916 ztest_block_tag_t rbt, wbt; 1917 1918 dmu_objset_name(os, osname); 1919 bs = ZTEST_DIROBJ_BLOCKSIZE; 1920 1921 /* 1922 * Have multiple threads write to large offsets in ZTEST_DIROBJ 1923 * to verify that having multiple threads writing to the same object 1924 * in parallel doesn't cause any trouble. 1925 * Also do parallel writes to the bonus buffer on occasion. 1926 */ 1927 for (i = 0; i < 50; i++) { 1928 b = ztest_random(ZTEST_SYNC_LOCKS); 1929 lp = &ztest_shared->zs_sync_lock[b]; 1930 1931 do_free = (ztest_random(4) == 0); 1932 1933 off = za->za_diroff_shared + ((uint64_t)b << SPA_MAXBLOCKSHIFT); 1934 1935 if (ztest_random(4) == 0) { 1936 /* 1937 * Do the bonus buffer instead of a regular block. 1938 */ 1939 do_free = 0; 1940 off = -1ULL; 1941 } 1942 1943 tx = dmu_tx_create(os); 1944 1945 if (off == -1ULL) 1946 dmu_tx_hold_bonus(tx, ZTEST_DIROBJ); 1947 else if (do_free) 1948 dmu_tx_hold_free(tx, ZTEST_DIROBJ, off, bs); 1949 else 1950 dmu_tx_hold_write(tx, ZTEST_DIROBJ, off, bs); 1951 1952 txg_how = ztest_random(2) == 0 ? TXG_WAIT : TXG_NOWAIT; 1953 error = dmu_tx_assign(tx, txg_how); 1954 if (error) { 1955 dmu_tx_abort(tx); 1956 if (error == ERESTART) { 1957 ASSERT(txg_how == TXG_NOWAIT); 1958 txg_wait_open(dmu_objset_pool(os), 0); 1959 continue; 1960 } 1961 ztest_record_enospc("dmu write parallel"); 1962 return; 1963 } 1964 txg = dmu_tx_get_txg(tx); 1965 1966 if (do_free) { 1967 (void) mutex_lock(lp); 1968 VERIFY(0 == dmu_free_range(os, ZTEST_DIROBJ, off, 1969 bs, tx)); 1970 (void) mutex_unlock(lp); 1971 dmu_tx_commit(tx); 1972 continue; 1973 } 1974 1975 wbt.bt_objset = dmu_objset_id(os); 1976 wbt.bt_object = ZTEST_DIROBJ; 1977 wbt.bt_offset = off; 1978 wbt.bt_txg = txg; 1979 wbt.bt_thread = za->za_instance; 1980 1981 if (off == -1ULL) { 1982 wbt.bt_seq = 0; 1983 VERIFY(0 == dmu_bonus_hold(os, ZTEST_DIROBJ, 1984 FTAG, &db)); 1985 ASSERT3U(db->db_size, ==, sizeof (wbt)); 1986 bcopy(db->db_data, &rbt, db->db_size); 1987 if (rbt.bt_objset != 0) { 1988 ASSERT3U(rbt.bt_objset, ==, wbt.bt_objset); 1989 ASSERT3U(rbt.bt_object, ==, wbt.bt_object); 1990 ASSERT3U(rbt.bt_offset, ==, wbt.bt_offset); 1991 ASSERT3U(rbt.bt_txg, <=, wbt.bt_txg); 1992 } 1993 dmu_buf_will_dirty(db, tx); 1994 bcopy(&wbt, db->db_data, db->db_size); 1995 dmu_buf_rele(db, FTAG); 1996 dmu_tx_commit(tx); 1997 continue; 1998 } 1999 2000 (void) mutex_lock(lp); 2001 2002 wbt.bt_seq = ztest_shared->zs_seq[b]++; 2003 2004 dmu_write(os, ZTEST_DIROBJ, off, sizeof (wbt), &wbt, tx); 2005 2006 (void) mutex_unlock(lp); 2007 2008 if (ztest_random(100) == 0) 2009 (void) poll(NULL, 0, 1); /* open dn_notxholds window */ 2010 2011 dmu_tx_commit(tx); 2012 2013 if (ztest_random(1000) == 0) 2014 txg_wait_synced(dmu_objset_pool(os), txg); 2015 2016 if (ztest_random(2) == 0) { 2017 blkptr_t blk = { 0 }; 2018 uint64_t blkoff; 2019 zbookmark_t zb; 2020 2021 txg_suspend(dmu_objset_pool(os)); 2022 (void) mutex_lock(lp); 2023 error = dmu_sync(os, ZTEST_DIROBJ, off, &blkoff, &blk, 2024 txg); 2025 (void) mutex_unlock(lp); 2026 if (error) { 2027 txg_resume(dmu_objset_pool(os)); 2028 dprintf("dmu_sync(%s, %d, %llx) = %d\n", 2029 osname, ZTEST_DIROBJ, off, error); 2030 continue; 2031 } 2032 2033 if (blk.blk_birth == 0) { /* concurrent free */ 2034 txg_resume(dmu_objset_pool(os)); 2035 continue; 2036 } 2037 2038 ASSERT(blk.blk_fill == 1); 2039 ASSERT3U(BP_GET_TYPE(&blk), ==, DMU_OT_UINT64_OTHER); 2040 ASSERT3U(BP_GET_LEVEL(&blk), ==, 0); 2041 ASSERT3U(BP_GET_LSIZE(&blk), ==, bs); 2042 2043 /* 2044 * Read the block that dmu_sync() returned to 2045 * make sure its contents match what we wrote. 2046 * We do this while still txg_suspend()ed to ensure 2047 * that the block can't be reused before we read it. 2048 */ 2049 zb.zb_objset = dmu_objset_id(os); 2050 zb.zb_object = ZTEST_DIROBJ; 2051 zb.zb_level = 0; 2052 zb.zb_blkid = off / bs; 2053 error = zio_wait(zio_read(NULL, dmu_objset_spa(os), 2054 &blk, iobuf, bs, NULL, NULL, 2055 ZIO_PRIORITY_SYNC_READ, ZIO_FLAG_MUSTSUCCEED, &zb)); 2056 ASSERT(error == 0); 2057 2058 txg_resume(dmu_objset_pool(os)); 2059 2060 bcopy(&iobuf[blkoff], &rbt, sizeof (rbt)); 2061 2062 if (rbt.bt_objset == 0) /* concurrent free */ 2063 continue; 2064 2065 ASSERT3U(rbt.bt_objset, ==, wbt.bt_objset); 2066 ASSERT3U(rbt.bt_object, ==, wbt.bt_object); 2067 ASSERT3U(rbt.bt_offset, ==, wbt.bt_offset); 2068 2069 /* 2070 * The semantic of dmu_sync() is that we always 2071 * push the most recent version of the data, 2072 * so in the face of concurrent updates we may 2073 * see a newer version of the block. That's OK. 2074 */ 2075 ASSERT3U(rbt.bt_txg, >=, wbt.bt_txg); 2076 if (rbt.bt_thread == wbt.bt_thread) 2077 ASSERT3U(rbt.bt_seq, ==, wbt.bt_seq); 2078 else 2079 ASSERT3U(rbt.bt_seq, >, wbt.bt_seq); 2080 } 2081 } 2082 } 2083 2084 /* 2085 * Verify that zap_{create,destroy,add,remove,update} work as expected. 2086 */ 2087 #define ZTEST_ZAP_MIN_INTS 1 2088 #define ZTEST_ZAP_MAX_INTS 4 2089 #define ZTEST_ZAP_MAX_PROPS 1000 2090 2091 void 2092 ztest_zap(ztest_args_t *za) 2093 { 2094 objset_t *os = za->za_os; 2095 uint64_t object; 2096 uint64_t txg, last_txg; 2097 uint64_t value[ZTEST_ZAP_MAX_INTS]; 2098 uint64_t zl_ints, zl_intsize, prop; 2099 int i, ints; 2100 int iters = 100; 2101 dmu_tx_t *tx; 2102 char propname[100], txgname[100]; 2103 int error; 2104 char osname[MAXNAMELEN]; 2105 char *hc[2] = { "s.acl.h", ".s.open.h.hyLZlg" }; 2106 2107 dmu_objset_name(os, osname); 2108 2109 /* 2110 * Create a new object if necessary, and record it in the directory. 2111 */ 2112 VERIFY(0 == dmu_read(os, ZTEST_DIROBJ, za->za_diroff, 2113 sizeof (uint64_t), &object)); 2114 2115 if (object == 0) { 2116 tx = dmu_tx_create(os); 2117 dmu_tx_hold_write(tx, ZTEST_DIROBJ, za->za_diroff, 2118 sizeof (uint64_t)); 2119 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, TRUE, NULL); 2120 error = dmu_tx_assign(tx, TXG_WAIT); 2121 if (error) { 2122 ztest_record_enospc("create zap test obj"); 2123 dmu_tx_abort(tx); 2124 return; 2125 } 2126 object = zap_create(os, DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx); 2127 if (error) { 2128 fatal(0, "zap_create('%s', %llu) = %d", 2129 osname, object, error); 2130 } 2131 ASSERT(object != 0); 2132 dmu_write(os, ZTEST_DIROBJ, za->za_diroff, 2133 sizeof (uint64_t), &object, tx); 2134 /* 2135 * Generate a known hash collision, and verify that 2136 * we can lookup and remove both entries. 2137 */ 2138 for (i = 0; i < 2; i++) { 2139 value[i] = i; 2140 error = zap_add(os, object, hc[i], sizeof (uint64_t), 2141 1, &value[i], tx); 2142 ASSERT3U(error, ==, 0); 2143 } 2144 for (i = 0; i < 2; i++) { 2145 error = zap_add(os, object, hc[i], sizeof (uint64_t), 2146 1, &value[i], tx); 2147 ASSERT3U(error, ==, EEXIST); 2148 error = zap_length(os, object, hc[i], 2149 &zl_intsize, &zl_ints); 2150 ASSERT3U(error, ==, 0); 2151 ASSERT3U(zl_intsize, ==, sizeof (uint64_t)); 2152 ASSERT3U(zl_ints, ==, 1); 2153 } 2154 for (i = 0; i < 2; i++) { 2155 error = zap_remove(os, object, hc[i], tx); 2156 ASSERT3U(error, ==, 0); 2157 } 2158 2159 dmu_tx_commit(tx); 2160 } 2161 2162 ints = MAX(ZTEST_ZAP_MIN_INTS, object % ZTEST_ZAP_MAX_INTS); 2163 2164 while (--iters >= 0) { 2165 prop = ztest_random(ZTEST_ZAP_MAX_PROPS); 2166 (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop); 2167 (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop); 2168 bzero(value, sizeof (value)); 2169 last_txg = 0; 2170 2171 /* 2172 * If these zap entries already exist, validate their contents. 2173 */ 2174 error = zap_length(os, object, txgname, &zl_intsize, &zl_ints); 2175 if (error == 0) { 2176 ASSERT3U(zl_intsize, ==, sizeof (uint64_t)); 2177 ASSERT3U(zl_ints, ==, 1); 2178 2179 error = zap_lookup(os, object, txgname, zl_intsize, 2180 zl_ints, &last_txg); 2181 2182 ASSERT3U(error, ==, 0); 2183 2184 error = zap_length(os, object, propname, &zl_intsize, 2185 &zl_ints); 2186 2187 ASSERT3U(error, ==, 0); 2188 ASSERT3U(zl_intsize, ==, sizeof (uint64_t)); 2189 ASSERT3U(zl_ints, ==, ints); 2190 2191 error = zap_lookup(os, object, propname, zl_intsize, 2192 zl_ints, value); 2193 2194 ASSERT3U(error, ==, 0); 2195 2196 for (i = 0; i < ints; i++) { 2197 ASSERT3U(value[i], ==, last_txg + object + i); 2198 } 2199 } else { 2200 ASSERT3U(error, ==, ENOENT); 2201 } 2202 2203 /* 2204 * Atomically update two entries in our zap object. 2205 * The first is named txg_%llu, and contains the txg 2206 * in which the property was last updated. The second 2207 * is named prop_%llu, and the nth element of its value 2208 * should be txg + object + n. 2209 */ 2210 tx = dmu_tx_create(os); 2211 dmu_tx_hold_zap(tx, object, TRUE, NULL); 2212 error = dmu_tx_assign(tx, TXG_WAIT); 2213 if (error) { 2214 ztest_record_enospc("create zap entry"); 2215 dmu_tx_abort(tx); 2216 return; 2217 } 2218 txg = dmu_tx_get_txg(tx); 2219 2220 if (last_txg > txg) 2221 fatal(0, "zap future leak: old %llu new %llu", 2222 last_txg, txg); 2223 2224 for (i = 0; i < ints; i++) 2225 value[i] = txg + object + i; 2226 2227 error = zap_update(os, object, txgname, sizeof (uint64_t), 2228 1, &txg, tx); 2229 if (error) 2230 fatal(0, "zap_update('%s', %llu, '%s') = %d", 2231 osname, object, txgname, error); 2232 2233 error = zap_update(os, object, propname, sizeof (uint64_t), 2234 ints, value, tx); 2235 if (error) 2236 fatal(0, "zap_update('%s', %llu, '%s') = %d", 2237 osname, object, propname, error); 2238 2239 dmu_tx_commit(tx); 2240 2241 /* 2242 * Remove a random pair of entries. 2243 */ 2244 prop = ztest_random(ZTEST_ZAP_MAX_PROPS); 2245 (void) sprintf(propname, "prop_%llu", (u_longlong_t)prop); 2246 (void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop); 2247 2248 error = zap_length(os, object, txgname, &zl_intsize, &zl_ints); 2249 2250 if (error == ENOENT) 2251 continue; 2252 2253 ASSERT3U(error, ==, 0); 2254 2255 tx = dmu_tx_create(os); 2256 dmu_tx_hold_zap(tx, object, TRUE, NULL); 2257 error = dmu_tx_assign(tx, TXG_WAIT); 2258 if (error) { 2259 ztest_record_enospc("remove zap entry"); 2260 dmu_tx_abort(tx); 2261 return; 2262 } 2263 error = zap_remove(os, object, txgname, tx); 2264 if (error) 2265 fatal(0, "zap_remove('%s', %llu, '%s') = %d", 2266 osname, object, txgname, error); 2267 2268 error = zap_remove(os, object, propname, tx); 2269 if (error) 2270 fatal(0, "zap_remove('%s', %llu, '%s') = %d", 2271 osname, object, propname, error); 2272 2273 dmu_tx_commit(tx); 2274 } 2275 2276 /* 2277 * Once in a while, destroy the object. 2278 */ 2279 if (ztest_random(100) != 0) 2280 return; 2281 2282 tx = dmu_tx_create(os); 2283 dmu_tx_hold_write(tx, ZTEST_DIROBJ, za->za_diroff, sizeof (uint64_t)); 2284 dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END); 2285 error = dmu_tx_assign(tx, TXG_WAIT); 2286 if (error) { 2287 ztest_record_enospc("destroy zap object"); 2288 dmu_tx_abort(tx); 2289 return; 2290 } 2291 error = zap_destroy(os, object, tx); 2292 if (error) 2293 fatal(0, "zap_destroy('%s', %llu) = %d", 2294 osname, object, error); 2295 object = 0; 2296 dmu_write(os, ZTEST_DIROBJ, za->za_diroff, sizeof (uint64_t), 2297 &object, tx); 2298 dmu_tx_commit(tx); 2299 } 2300 2301 void 2302 ztest_zap_parallel(ztest_args_t *za) 2303 { 2304 objset_t *os = za->za_os; 2305 uint64_t txg, object, count, wsize, wc, zl_wsize, zl_wc; 2306 int iters = 100; 2307 dmu_tx_t *tx; 2308 int i, namelen, error; 2309 char name[20], string_value[20]; 2310 void *data; 2311 2312 while (--iters >= 0) { 2313 /* 2314 * Generate a random name of the form 'xxx.....' where each 2315 * x is a random printable character and the dots are dots. 2316 * There are 94 such characters, and the name length goes from 2317 * 6 to 20, so there are 94^3 * 15 = 12,458,760 possible names. 2318 */ 2319 namelen = ztest_random(sizeof (name) - 5) + 5 + 1; 2320 2321 for (i = 0; i < 3; i++) 2322 name[i] = '!' + ztest_random('~' - '!' + 1); 2323 for (; i < namelen - 1; i++) 2324 name[i] = '.'; 2325 name[i] = '\0'; 2326 2327 if (ztest_random(2) == 0) 2328 object = ZTEST_MICROZAP_OBJ; 2329 else 2330 object = ZTEST_FATZAP_OBJ; 2331 2332 if ((namelen & 1) || object == ZTEST_MICROZAP_OBJ) { 2333 wsize = sizeof (txg); 2334 wc = 1; 2335 data = &txg; 2336 } else { 2337 wsize = 1; 2338 wc = namelen; 2339 data = string_value; 2340 } 2341 2342 count = -1ULL; 2343 VERIFY(zap_count(os, object, &count) == 0); 2344 ASSERT(count != -1ULL); 2345 2346 /* 2347 * Select an operation: length, lookup, add, update, remove. 2348 */ 2349 i = ztest_random(5); 2350 2351 if (i >= 2) { 2352 tx = dmu_tx_create(os); 2353 dmu_tx_hold_zap(tx, object, TRUE, NULL); 2354 error = dmu_tx_assign(tx, TXG_WAIT); 2355 if (error) { 2356 ztest_record_enospc("zap parallel"); 2357 dmu_tx_abort(tx); 2358 return; 2359 } 2360 txg = dmu_tx_get_txg(tx); 2361 bcopy(name, string_value, namelen); 2362 } else { 2363 tx = NULL; 2364 txg = 0; 2365 bzero(string_value, namelen); 2366 } 2367 2368 switch (i) { 2369 2370 case 0: 2371 error = zap_length(os, object, name, &zl_wsize, &zl_wc); 2372 if (error == 0) { 2373 ASSERT3U(wsize, ==, zl_wsize); 2374 ASSERT3U(wc, ==, zl_wc); 2375 } else { 2376 ASSERT3U(error, ==, ENOENT); 2377 } 2378 break; 2379 2380 case 1: 2381 error = zap_lookup(os, object, name, wsize, wc, data); 2382 if (error == 0) { 2383 if (data == string_value && 2384 bcmp(name, data, namelen) != 0) 2385 fatal(0, "name '%s' != val '%s' len %d", 2386 name, data, namelen); 2387 } else { 2388 ASSERT3U(error, ==, ENOENT); 2389 } 2390 break; 2391 2392 case 2: 2393 error = zap_add(os, object, name, wsize, wc, data, tx); 2394 ASSERT(error == 0 || error == EEXIST); 2395 break; 2396 2397 case 3: 2398 VERIFY(zap_update(os, object, name, wsize, wc, 2399 data, tx) == 0); 2400 break; 2401 2402 case 4: 2403 error = zap_remove(os, object, name, tx); 2404 ASSERT(error == 0 || error == ENOENT); 2405 break; 2406 } 2407 2408 if (tx != NULL) 2409 dmu_tx_commit(tx); 2410 } 2411 } 2412 2413 void 2414 ztest_dsl_prop_get_set(ztest_args_t *za) 2415 { 2416 objset_t *os = za->za_os; 2417 int i, inherit; 2418 uint64_t value; 2419 const char *prop, *valname; 2420 char setpoint[MAXPATHLEN]; 2421 char osname[MAXNAMELEN]; 2422 int error; 2423 2424 (void) rw_rdlock(&ztest_shared->zs_name_lock); 2425 2426 dmu_objset_name(os, osname); 2427 2428 for (i = 0; i < 2; i++) { 2429 if (i == 0) { 2430 prop = "checksum"; 2431 value = ztest_random_checksum(); 2432 inherit = (value == ZIO_CHECKSUM_INHERIT); 2433 } else { 2434 prop = "compression"; 2435 value = ztest_random_compress(); 2436 inherit = (value == ZIO_COMPRESS_INHERIT); 2437 } 2438 2439 error = dsl_prop_set(osname, prop, sizeof (value), 2440 !inherit, &value); 2441 2442 if (error == ENOSPC) { 2443 ztest_record_enospc("dsl_prop_set"); 2444 break; 2445 } 2446 2447 ASSERT3U(error, ==, 0); 2448 2449 VERIFY3U(dsl_prop_get(osname, prop, sizeof (value), 2450 1, &value, setpoint), ==, 0); 2451 2452 if (i == 0) 2453 valname = zio_checksum_table[value].ci_name; 2454 else 2455 valname = zio_compress_table[value].ci_name; 2456 2457 if (zopt_verbose >= 6) { 2458 (void) printf("%s %s = %s for '%s'\n", 2459 osname, prop, valname, setpoint); 2460 } 2461 } 2462 2463 (void) rw_unlock(&ztest_shared->zs_name_lock); 2464 } 2465 2466 static void 2467 ztest_error_setup(vdev_t *vd, int mode, int mask, uint64_t arg) 2468 { 2469 int c; 2470 2471 for (c = 0; c < vd->vdev_children; c++) 2472 ztest_error_setup(vd->vdev_child[c], mode, mask, arg); 2473 2474 if (vd->vdev_path != NULL) { 2475 vd->vdev_fault_mode = mode; 2476 vd->vdev_fault_mask = mask; 2477 vd->vdev_fault_arg = arg; 2478 } 2479 } 2480 2481 /* 2482 * Inject random faults into the on-disk data. 2483 */ 2484 void 2485 ztest_fault_inject(ztest_args_t *za) 2486 { 2487 int fd; 2488 uint64_t offset; 2489 uint64_t leaves = MAX(zopt_mirrors, 1) * zopt_raidz; 2490 uint64_t bad = 0x1990c0ffeedecade; 2491 uint64_t top, leaf; 2492 char path0[MAXPATHLEN]; 2493 char pathrand[MAXPATHLEN]; 2494 size_t fsize; 2495 spa_t *spa = dmu_objset_spa(za->za_os); 2496 int bshift = SPA_MAXBLOCKSHIFT + 2; /* don't scrog all labels */ 2497 int iters = 1000; 2498 vdev_t *vd0; 2499 uint64_t guid0 = 0; 2500 2501 /* 2502 * We can't inject faults when we have no fault tolerance. 2503 */ 2504 if (zopt_maxfaults == 0) 2505 return; 2506 2507 ASSERT(leaves >= 2); 2508 2509 /* 2510 * Pick a random top-level vdev. 2511 */ 2512 spa_config_enter(spa, RW_READER, FTAG); 2513 top = ztest_random(spa->spa_root_vdev->vdev_children); 2514 spa_config_exit(spa, FTAG); 2515 2516 /* 2517 * Pick a random leaf. 2518 */ 2519 leaf = ztest_random(leaves); 2520 2521 /* 2522 * Generate paths to the first two leaves in this top-level vdev, 2523 * and to the random leaf we selected. We'll induce transient 2524 * I/O errors and random online/offline activity on leaf 0, 2525 * and we'll write random garbage to the randomly chosen leaf. 2526 */ 2527 (void) snprintf(path0, sizeof (path0), 2528 ztest_dev_template, zopt_dir, zopt_pool, top * leaves + 0); 2529 (void) snprintf(pathrand, sizeof (pathrand), 2530 ztest_dev_template, zopt_dir, zopt_pool, top * leaves + leaf); 2531 2532 dprintf("damaging %s and %s\n", path0, pathrand); 2533 2534 spa_config_enter(spa, RW_READER, FTAG); 2535 2536 /* 2537 * If we can tolerate two or more faults, make vd0 fail randomly. 2538 */ 2539 vd0 = vdev_lookup_by_path(spa->spa_root_vdev, path0); 2540 if (vd0 != NULL && zopt_maxfaults >= 2) { 2541 guid0 = vd0->vdev_guid; 2542 ztest_error_setup(vd0, VDEV_FAULT_COUNT, 2543 (1U << ZIO_TYPE_READ) | (1U << ZIO_TYPE_WRITE), 100); 2544 } 2545 2546 spa_config_exit(spa, FTAG); 2547 2548 /* 2549 * If we can tolerate two or more faults, randomly online/offline vd0. 2550 */ 2551 if (zopt_maxfaults >= 2 && guid0 != 0) { 2552 if (ztest_random(10) < 6) 2553 (void) vdev_offline(spa, guid0, B_TRUE); 2554 else 2555 (void) vdev_online(spa, guid0); 2556 } 2557 2558 /* 2559 * We have at least single-fault tolerance, so inject data corruption. 2560 */ 2561 fd = open(pathrand, O_RDWR); 2562 2563 if (fd == -1) /* we hit a gap in the device namespace */ 2564 return; 2565 2566 fsize = lseek(fd, 0, SEEK_END); 2567 2568 while (--iters != 0) { 2569 offset = ztest_random(fsize / (leaves << bshift)) * 2570 (leaves << bshift) + (leaf << bshift) + 2571 (ztest_random(1ULL << (bshift - 1)) & -8ULL); 2572 2573 if (offset >= fsize) 2574 continue; 2575 2576 if (zopt_verbose >= 6) 2577 (void) printf("injecting bad word into %s," 2578 " offset 0x%llx\n", pathrand, (u_longlong_t)offset); 2579 2580 if (pwrite(fd, &bad, sizeof (bad), offset) != sizeof (bad)) 2581 fatal(1, "can't inject bad word at 0x%llx in %s", 2582 offset, pathrand); 2583 } 2584 2585 (void) close(fd); 2586 } 2587 2588 /* 2589 * Scrub the pool. 2590 */ 2591 void 2592 ztest_scrub(ztest_args_t *za) 2593 { 2594 spa_t *spa = dmu_objset_spa(za->za_os); 2595 2596 (void) spa_scrub(spa, POOL_SCRUB_EVERYTHING, B_FALSE); 2597 (void) poll(NULL, 0, 1000); /* wait a second, then force a restart */ 2598 (void) spa_scrub(spa, POOL_SCRUB_EVERYTHING, B_FALSE); 2599 } 2600 2601 /* 2602 * Rename the pool to a different name and then rename it back. 2603 */ 2604 void 2605 ztest_spa_rename(ztest_args_t *za) 2606 { 2607 char *oldname, *newname; 2608 int error; 2609 spa_t *spa; 2610 2611 (void) rw_wrlock(&ztest_shared->zs_name_lock); 2612 2613 oldname = za->za_pool; 2614 newname = umem_alloc(strlen(oldname) + 5, UMEM_NOFAIL); 2615 (void) strcpy(newname, oldname); 2616 (void) strcat(newname, "_tmp"); 2617 2618 /* 2619 * Do the rename 2620 */ 2621 error = spa_rename(oldname, newname); 2622 if (error) 2623 fatal(0, "spa_rename('%s', '%s') = %d", oldname, 2624 newname, error); 2625 2626 /* 2627 * Try to open it under the old name, which shouldn't exist 2628 */ 2629 error = spa_open(oldname, &spa, FTAG); 2630 if (error != ENOENT) 2631 fatal(0, "spa_open('%s') = %d", oldname, error); 2632 2633 /* 2634 * Open it under the new name and make sure it's still the same spa_t. 2635 */ 2636 error = spa_open(newname, &spa, FTAG); 2637 if (error != 0) 2638 fatal(0, "spa_open('%s') = %d", newname, error); 2639 2640 ASSERT(spa == dmu_objset_spa(za->za_os)); 2641 spa_close(spa, FTAG); 2642 2643 /* 2644 * Rename it back to the original 2645 */ 2646 error = spa_rename(newname, oldname); 2647 if (error) 2648 fatal(0, "spa_rename('%s', '%s') = %d", newname, 2649 oldname, error); 2650 2651 /* 2652 * Make sure it can still be opened 2653 */ 2654 error = spa_open(oldname, &spa, FTAG); 2655 if (error != 0) 2656 fatal(0, "spa_open('%s') = %d", oldname, error); 2657 2658 ASSERT(spa == dmu_objset_spa(za->za_os)); 2659 spa_close(spa, FTAG); 2660 2661 umem_free(newname, strlen(newname) + 1); 2662 2663 (void) rw_unlock(&ztest_shared->zs_name_lock); 2664 } 2665 2666 2667 /* 2668 * Completely obliterate one disk. 2669 */ 2670 static void 2671 ztest_obliterate_one_disk(uint64_t vdev) 2672 { 2673 int fd; 2674 char dev_name[MAXPATHLEN], copy_name[MAXPATHLEN]; 2675 size_t fsize; 2676 2677 if (zopt_maxfaults < 2) 2678 return; 2679 2680 (void) sprintf(dev_name, ztest_dev_template, zopt_dir, zopt_pool, vdev); 2681 (void) snprintf(copy_name, MAXPATHLEN, "%s.old", dev_name); 2682 2683 fd = open(dev_name, O_RDWR); 2684 2685 if (fd == -1) 2686 fatal(1, "can't open %s", dev_name); 2687 2688 /* 2689 * Determine the size. 2690 */ 2691 fsize = lseek(fd, 0, SEEK_END); 2692 2693 (void) close(fd); 2694 2695 /* 2696 * Rename the old device to dev_name.old (useful for debugging). 2697 */ 2698 VERIFY(rename(dev_name, copy_name) == 0); 2699 2700 /* 2701 * Create a new one. 2702 */ 2703 VERIFY((fd = open(dev_name, O_RDWR | O_CREAT | O_TRUNC, 0666)) >= 0); 2704 VERIFY(ftruncate(fd, fsize) == 0); 2705 (void) close(fd); 2706 } 2707 2708 static void 2709 ztest_replace_one_disk(spa_t *spa, uint64_t vdev) 2710 { 2711 char dev_name[MAXPATHLEN]; 2712 nvlist_t *file, *root; 2713 int error; 2714 uint64_t guid; 2715 uint64_t ashift = ztest_get_ashift(); 2716 vdev_t *vd; 2717 2718 (void) sprintf(dev_name, ztest_dev_template, zopt_dir, zopt_pool, vdev); 2719 2720 /* 2721 * Build the nvlist describing dev_name. 2722 */ 2723 VERIFY(nvlist_alloc(&file, NV_UNIQUE_NAME, 0) == 0); 2724 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_TYPE, VDEV_TYPE_FILE) == 0); 2725 VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_PATH, dev_name) == 0); 2726 VERIFY(nvlist_add_uint64(file, ZPOOL_CONFIG_ASHIFT, ashift) == 0); 2727 2728 VERIFY(nvlist_alloc(&root, NV_UNIQUE_NAME, 0) == 0); 2729 VERIFY(nvlist_add_string(root, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) == 0); 2730 VERIFY(nvlist_add_nvlist_array(root, ZPOOL_CONFIG_CHILDREN, 2731 &file, 1) == 0); 2732 2733 spa_config_enter(spa, RW_READER, FTAG); 2734 if ((vd = vdev_lookup_by_path(spa->spa_root_vdev, dev_name)) == NULL) 2735 guid = 0; 2736 else 2737 guid = vd->vdev_guid; 2738 spa_config_exit(spa, FTAG); 2739 error = spa_vdev_attach(spa, guid, root, B_TRUE); 2740 if (error != 0 && 2741 error != EBUSY && 2742 error != ENOTSUP && 2743 error != ENODEV && 2744 error != EDOM) 2745 fatal(0, "spa_vdev_attach(in-place) = %d", error); 2746 2747 nvlist_free(file); 2748 nvlist_free(root); 2749 } 2750 2751 static void 2752 ztest_verify_blocks(char *pool) 2753 { 2754 int status; 2755 char zdb[MAXPATHLEN + MAXNAMELEN + 20]; 2756 char zbuf[1024]; 2757 char *bin; 2758 FILE *fp; 2759 2760 (void) realpath(getexecname(), zdb); 2761 2762 /* zdb lives in /usr/sbin, while ztest lives in /usr/bin */ 2763 bin = strstr(zdb, "/usr/bin/"); 2764 /* LINTED */ 2765 (void) sprintf(bin, "/usr/sbin/zdb -bc%s%s -U -O %s %s", 2766 zopt_verbose >= 3 ? "s" : "", 2767 zopt_verbose >= 4 ? "v" : "", 2768 ztest_random(2) == 0 ? "pre" : "post", pool); 2769 2770 if (zopt_verbose >= 5) 2771 (void) printf("Executing %s\n", strstr(zdb, "zdb ")); 2772 2773 fp = popen(zdb, "r"); 2774 2775 while (fgets(zbuf, sizeof (zbuf), fp) != NULL) 2776 if (zopt_verbose >= 3) 2777 (void) printf("%s", zbuf); 2778 2779 status = pclose(fp); 2780 2781 if (status == 0) 2782 return; 2783 2784 ztest_dump_core = 0; 2785 if (WIFEXITED(status)) 2786 fatal(0, "'%s' exit code %d", zdb, WEXITSTATUS(status)); 2787 else 2788 fatal(0, "'%s' died with signal %d", zdb, WTERMSIG(status)); 2789 } 2790 2791 static void 2792 ztest_walk_pool_directory(char *header) 2793 { 2794 spa_t *spa = NULL; 2795 2796 if (zopt_verbose >= 6) 2797 (void) printf("%s\n", header); 2798 2799 mutex_enter(&spa_namespace_lock); 2800 while ((spa = spa_next(spa)) != NULL) 2801 if (zopt_verbose >= 6) 2802 (void) printf("\t%s\n", spa_name(spa)); 2803 mutex_exit(&spa_namespace_lock); 2804 } 2805 2806 static void 2807 ztest_spa_import_export(char *oldname, char *newname) 2808 { 2809 nvlist_t *config; 2810 uint64_t pool_guid; 2811 spa_t *spa; 2812 int error; 2813 2814 if (zopt_verbose >= 4) { 2815 (void) printf("import/export: old = %s, new = %s\n", 2816 oldname, newname); 2817 } 2818 2819 /* 2820 * Clean up from previous runs. 2821 */ 2822 (void) spa_destroy(newname); 2823 2824 /* 2825 * Get the pool's configuration and guid. 2826 */ 2827 error = spa_open(oldname, &spa, FTAG); 2828 if (error) 2829 fatal(0, "spa_open('%s') = %d", oldname, error); 2830 2831 pool_guid = spa_guid(spa); 2832 spa_close(spa, FTAG); 2833 2834 ztest_walk_pool_directory("pools before export"); 2835 2836 /* 2837 * Export it. 2838 */ 2839 error = spa_export(oldname, &config); 2840 if (error) 2841 fatal(0, "spa_export('%s') = %d", oldname, error); 2842 2843 ztest_walk_pool_directory("pools after export"); 2844 2845 /* 2846 * Import it under the new name. 2847 */ 2848 error = spa_import(newname, config, NULL); 2849 if (error) 2850 fatal(0, "spa_import('%s') = %d", newname, error); 2851 2852 ztest_walk_pool_directory("pools after import"); 2853 2854 /* 2855 * Try to import it again -- should fail with EEXIST. 2856 */ 2857 error = spa_import(newname, config, NULL); 2858 if (error != EEXIST) 2859 fatal(0, "spa_import('%s') twice", newname); 2860 2861 /* 2862 * Try to import it under a different name -- should fail with EEXIST. 2863 */ 2864 error = spa_import(oldname, config, NULL); 2865 if (error != EEXIST) 2866 fatal(0, "spa_import('%s') under multiple names", newname); 2867 2868 /* 2869 * Verify that the pool is no longer visible under the old name. 2870 */ 2871 error = spa_open(oldname, &spa, FTAG); 2872 if (error != ENOENT) 2873 fatal(0, "spa_open('%s') = %d", newname, error); 2874 2875 /* 2876 * Verify that we can open and close the pool using the new name. 2877 */ 2878 error = spa_open(newname, &spa, FTAG); 2879 if (error) 2880 fatal(0, "spa_open('%s') = %d", newname, error); 2881 ASSERT(pool_guid == spa_guid(spa)); 2882 spa_close(spa, FTAG); 2883 2884 nvlist_free(config); 2885 } 2886 2887 static void * 2888 ztest_thread(void *arg) 2889 { 2890 ztest_args_t *za = arg; 2891 ztest_shared_t *zs = ztest_shared; 2892 hrtime_t now, functime; 2893 ztest_info_t *zi; 2894 int f; 2895 2896 while ((now = gethrtime()) < za->za_stop) { 2897 /* 2898 * See if it's time to force a crash. 2899 */ 2900 if (now > za->za_kill) { 2901 zs->zs_alloc = spa_get_alloc(dmu_objset_spa(za->za_os)); 2902 zs->zs_space = spa_get_space(dmu_objset_spa(za->za_os)); 2903 (void) kill(getpid(), SIGKILL); 2904 } 2905 2906 /* 2907 * Pick a random function. 2908 */ 2909 f = ztest_random(ZTEST_FUNCS); 2910 zi = &zs->zs_info[f]; 2911 2912 /* 2913 * Decide whether to call it, based on the requested frequency. 2914 */ 2915 if (zi->zi_call_target == 0 || 2916 (double)zi->zi_call_total / zi->zi_call_target > 2917 (double)(now - zs->zs_start_time) / (zopt_time * NANOSEC)) 2918 continue; 2919 2920 atomic_add_64(&zi->zi_calls, 1); 2921 atomic_add_64(&zi->zi_call_total, 1); 2922 2923 za->za_diroff = (za->za_instance * ZTEST_FUNCS + f) * 2924 ZTEST_DIRSIZE; 2925 za->za_diroff_shared = (1ULL << 63); 2926 2927 ztest_dmu_write_parallel(za); 2928 2929 zi->zi_func(za); 2930 2931 functime = gethrtime() - now; 2932 2933 atomic_add_64(&zi->zi_call_time, functime); 2934 2935 if (zopt_verbose >= 4) { 2936 Dl_info dli; 2937 (void) dladdr((void *)zi->zi_func, &dli); 2938 (void) printf("%6.2f sec in %s\n", 2939 (double)functime / NANOSEC, dli.dli_sname); 2940 } 2941 2942 /* 2943 * If we're getting ENOSPC with some regularity, stop. 2944 */ 2945 if (zs->zs_enospc_count > 10) 2946 break; 2947 } 2948 2949 return (NULL); 2950 } 2951 2952 /* 2953 * Kick off threads to run tests on all datasets in parallel. 2954 */ 2955 static void 2956 ztest_run(char *pool) 2957 { 2958 int t, d, error; 2959 ztest_shared_t *zs = ztest_shared; 2960 ztest_args_t *za; 2961 spa_t *spa; 2962 char name[100]; 2963 2964 (void) _mutex_init(&zs->zs_vdev_lock, USYNC_THREAD, NULL); 2965 (void) rwlock_init(&zs->zs_name_lock, USYNC_THREAD, NULL); 2966 2967 for (t = 0; t < ZTEST_SYNC_LOCKS; t++) 2968 (void) _mutex_init(&zs->zs_sync_lock[t], USYNC_THREAD, NULL); 2969 2970 /* 2971 * Destroy one disk before we even start. 2972 * It's mirrored, so everything should work just fine. 2973 * This makes us exercise fault handling very early in spa_load(). 2974 */ 2975 ztest_obliterate_one_disk(0); 2976 2977 /* 2978 * Verify that the sum of the sizes of all blocks in the pool 2979 * equals the SPA's allocated space total. 2980 */ 2981 ztest_verify_blocks(pool); 2982 2983 /* 2984 * Kick off a replacement of the disk we just obliterated. 2985 */ 2986 kernel_init(FREAD | FWRITE); 2987 error = spa_open(pool, &spa, FTAG); 2988 if (error) 2989 fatal(0, "spa_open(%s) = %d", pool, error); 2990 ztest_replace_one_disk(spa, 0); 2991 if (zopt_verbose >= 5) 2992 show_pool_stats(spa); 2993 spa_close(spa, FTAG); 2994 kernel_fini(); 2995 2996 kernel_init(FREAD | FWRITE); 2997 2998 /* 2999 * Verify that we can export the pool and reimport it under a 3000 * different name. 3001 */ 3002 if (ztest_random(2) == 0) { 3003 (void) snprintf(name, 100, "%s_import", pool); 3004 ztest_spa_import_export(pool, name); 3005 ztest_spa_import_export(name, pool); 3006 } 3007 3008 /* 3009 * Verify that we can loop over all pools. 3010 */ 3011 mutex_enter(&spa_namespace_lock); 3012 for (spa = spa_next(NULL); spa != NULL; spa = spa_next(spa)) { 3013 if (zopt_verbose > 3) { 3014 (void) printf("spa_next: found %s\n", spa_name(spa)); 3015 } 3016 } 3017 mutex_exit(&spa_namespace_lock); 3018 3019 /* 3020 * Open our pool. 3021 */ 3022 error = spa_open(pool, &spa, FTAG); 3023 if (error) 3024 fatal(0, "spa_open() = %d", error); 3025 3026 /* 3027 * Verify that we can safely inquire about about any object, 3028 * whether it's allocated or not. To make it interesting, 3029 * we probe a 5-wide window around each power of two. 3030 * This hits all edge cases, including zero and the max. 3031 */ 3032 for (t = 0; t < 64; t++) { 3033 for (d = -5; d <= 5; d++) { 3034 error = dmu_object_info(spa->spa_meta_objset, 3035 (1ULL << t) + d, NULL); 3036 ASSERT(error == 0 || error == ENOENT || 3037 error == EINVAL); 3038 } 3039 } 3040 3041 /* 3042 * Now kick off all the tests that run in parallel. 3043 */ 3044 zs->zs_enospc_count = 0; 3045 3046 za = umem_zalloc(zopt_threads * sizeof (ztest_args_t), UMEM_NOFAIL); 3047 3048 if (zopt_verbose >= 4) 3049 (void) printf("starting main threads...\n"); 3050 3051 za[0].za_start = gethrtime(); 3052 za[0].za_stop = za[0].za_start + zopt_passtime * NANOSEC; 3053 za[0].za_stop = MIN(za[0].za_stop, zs->zs_stop_time); 3054 za[0].za_kill = za[0].za_stop; 3055 if (ztest_random(100) < zopt_killrate) 3056 za[0].za_kill -= ztest_random(zopt_passtime * NANOSEC); 3057 3058 for (t = 0; t < zopt_threads; t++) { 3059 d = t % zopt_datasets; 3060 if (t < zopt_datasets) { 3061 ztest_replay_t zr; 3062 (void) rw_rdlock(&ztest_shared->zs_name_lock); 3063 (void) snprintf(name, 100, "%s/%s_%d", pool, pool, d); 3064 error = dmu_objset_create(name, DMU_OST_OTHER, NULL, 3065 ztest_create_cb, NULL); 3066 if (error != 0 && error != EEXIST) { 3067 if (error == ENOSPC) { 3068 zs->zs_enospc_count++; 3069 (void) rw_unlock( 3070 &ztest_shared->zs_name_lock); 3071 break; 3072 } 3073 fatal(0, "dmu_objset_create(%s) = %d", 3074 name, error); 3075 } 3076 error = dmu_objset_open(name, DMU_OST_OTHER, 3077 DS_MODE_STANDARD, &za[d].za_os); 3078 if (error) 3079 fatal(0, "dmu_objset_open('%s') = %d", 3080 name, error); 3081 (void) rw_unlock(&ztest_shared->zs_name_lock); 3082 zr.zr_os = za[d].za_os; 3083 zil_replay(zr.zr_os, &zr, &zr.zr_assign, 3084 ztest_replay_vector, NULL); 3085 za[d].za_zilog = zil_open(za[d].za_os, NULL); 3086 } 3087 za[t].za_pool = spa_strdup(pool); 3088 za[t].za_os = za[d].za_os; 3089 za[t].za_zilog = za[d].za_zilog; 3090 za[t].za_instance = t; 3091 za[t].za_random = ztest_random(-1ULL); 3092 za[t].za_start = za[0].za_start; 3093 za[t].za_stop = za[0].za_stop; 3094 za[t].za_kill = za[0].za_kill; 3095 3096 error = thr_create(0, 0, ztest_thread, &za[t], THR_BOUND, 3097 &za[t].za_thread); 3098 if (error) 3099 fatal(0, "can't create thread %d: error %d", 3100 t, error); 3101 } 3102 3103 while (--t >= 0) { 3104 error = thr_join(za[t].za_thread, NULL, NULL); 3105 if (error) 3106 fatal(0, "thr_join(%d) = %d", t, error); 3107 if (za[t].za_th) 3108 traverse_fini(za[t].za_th); 3109 if (t < zopt_datasets) { 3110 zil_close(za[t].za_zilog); 3111 dmu_objset_close(za[t].za_os); 3112 } 3113 spa_strfree(za[t].za_pool); 3114 } 3115 3116 umem_free(za, zopt_threads * sizeof (ztest_args_t)); 3117 3118 if (zopt_verbose >= 3) 3119 show_pool_stats(spa); 3120 3121 txg_wait_synced(spa_get_dsl(spa), 0); 3122 3123 zs->zs_alloc = spa_get_alloc(spa); 3124 zs->zs_space = spa_get_space(spa); 3125 3126 /* 3127 * Did we have out-of-space errors? If so, destroy a random objset. 3128 */ 3129 if (zs->zs_enospc_count != 0) { 3130 (void) rw_rdlock(&ztest_shared->zs_name_lock); 3131 (void) snprintf(name, 100, "%s/%s_%d", pool, pool, 3132 (int)ztest_random(zopt_datasets)); 3133 if (zopt_verbose >= 3) 3134 (void) printf("Destroying %s to free up space\n", name); 3135 dmu_objset_find(name, ztest_destroy_cb, NULL, 3136 DS_FIND_SNAPSHOTS); 3137 (void) rw_unlock(&ztest_shared->zs_name_lock); 3138 } 3139 3140 txg_wait_synced(spa_get_dsl(spa), 0); 3141 3142 /* 3143 * Right before closing the pool, kick off a bunch of async I/O; 3144 * spa_close() should wait for it to complete. 3145 */ 3146 for (t = 1; t < 50; t++) 3147 dmu_prefetch(spa->spa_meta_objset, t, 0, 1 << 15); 3148 3149 spa_close(spa, FTAG); 3150 3151 kernel_fini(); 3152 } 3153 3154 void 3155 print_time(hrtime_t t, char *timebuf) 3156 { 3157 hrtime_t s = t / NANOSEC; 3158 hrtime_t m = s / 60; 3159 hrtime_t h = m / 60; 3160 hrtime_t d = h / 24; 3161 3162 s -= m * 60; 3163 m -= h * 60; 3164 h -= d * 24; 3165 3166 timebuf[0] = '\0'; 3167 3168 if (d) 3169 (void) sprintf(timebuf, 3170 "%llud%02lluh%02llum%02llus", d, h, m, s); 3171 else if (h) 3172 (void) sprintf(timebuf, "%lluh%02llum%02llus", h, m, s); 3173 else if (m) 3174 (void) sprintf(timebuf, "%llum%02llus", m, s); 3175 else 3176 (void) sprintf(timebuf, "%llus", s); 3177 } 3178 3179 /* 3180 * Create a storage pool with the given name and initial vdev size. 3181 * Then create the specified number of datasets in the pool. 3182 */ 3183 static void 3184 ztest_init(char *pool) 3185 { 3186 spa_t *spa; 3187 int error; 3188 nvlist_t *nvroot; 3189 3190 kernel_init(FREAD | FWRITE); 3191 3192 /* 3193 * Create the storage pool. 3194 */ 3195 (void) spa_destroy(pool); 3196 ztest_shared->zs_vdev_primaries = 0; 3197 nvroot = make_vdev_root(zopt_vdev_size, zopt_raidz, zopt_mirrors, 1); 3198 error = spa_create(pool, nvroot, NULL); 3199 nvlist_free(nvroot); 3200 3201 if (error) 3202 fatal(0, "spa_create() = %d", error); 3203 error = spa_open(pool, &spa, FTAG); 3204 if (error) 3205 fatal(0, "spa_open() = %d", error); 3206 3207 if (zopt_verbose >= 3) 3208 show_pool_stats(spa); 3209 3210 spa_close(spa, FTAG); 3211 3212 kernel_fini(); 3213 } 3214 3215 int 3216 main(int argc, char **argv) 3217 { 3218 int kills = 0; 3219 int iters = 0; 3220 int i, f; 3221 ztest_shared_t *zs; 3222 ztest_info_t *zi; 3223 char timebuf[100]; 3224 char numbuf[6]; 3225 3226 (void) setvbuf(stdout, NULL, _IOLBF, 0); 3227 3228 /* Override location of zpool.cache */ 3229 spa_config_dir = "/tmp"; 3230 3231 ztest_random_fd = open("/dev/urandom", O_RDONLY); 3232 3233 process_options(argc, argv); 3234 3235 argc -= optind; 3236 argv += optind; 3237 3238 dprintf_setup(&argc, argv); 3239 3240 /* 3241 * Blow away any existing copy of zpool.cache 3242 */ 3243 if (zopt_init != 0) 3244 (void) remove("/tmp/zpool.cache"); 3245 3246 zs = ztest_shared = (void *)mmap(0, 3247 P2ROUNDUP(sizeof (ztest_shared_t), getpagesize()), 3248 PROT_READ | PROT_WRITE, MAP_SHARED | MAP_ANON, -1, 0); 3249 3250 if (zopt_verbose >= 1) { 3251 (void) printf("%llu vdevs, %d datasets, %d threads," 3252 " %llu seconds...\n", 3253 (u_longlong_t)zopt_vdevs, zopt_datasets, zopt_threads, 3254 (u_longlong_t)zopt_time); 3255 } 3256 3257 /* 3258 * Create and initialize our storage pool. 3259 */ 3260 for (i = 1; i <= zopt_init; i++) { 3261 bzero(zs, sizeof (ztest_shared_t)); 3262 if (zopt_verbose >= 3 && zopt_init != 1) 3263 (void) printf("ztest_init(), pass %d\n", i); 3264 ztest_init(zopt_pool); 3265 } 3266 3267 /* 3268 * Initialize the call targets for each function. 3269 */ 3270 for (f = 0; f < ZTEST_FUNCS; f++) { 3271 zi = &zs->zs_info[f]; 3272 3273 *zi = ztest_info[f]; 3274 3275 if (*zi->zi_interval == 0) 3276 zi->zi_call_target = UINT64_MAX; 3277 else 3278 zi->zi_call_target = zopt_time / *zi->zi_interval; 3279 } 3280 3281 zs->zs_start_time = gethrtime(); 3282 zs->zs_stop_time = zs->zs_start_time + zopt_time * NANOSEC; 3283 3284 /* 3285 * Run the tests in a loop. These tests include fault injection 3286 * to verify that self-healing data works, and forced crashes 3287 * to verify that we never lose on-disk consistency. 3288 */ 3289 while (gethrtime() < zs->zs_stop_time) { 3290 int status; 3291 pid_t pid; 3292 char *tmp; 3293 3294 /* 3295 * Initialize the workload counters for each function. 3296 */ 3297 for (f = 0; f < ZTEST_FUNCS; f++) { 3298 zi = &zs->zs_info[f]; 3299 zi->zi_calls = 0; 3300 zi->zi_call_time = 0; 3301 } 3302 3303 pid = fork(); 3304 3305 if (pid == -1) 3306 fatal(1, "fork failed"); 3307 3308 if (pid == 0) { /* child */ 3309 struct rlimit rl = { 1024, 1024 }; 3310 (void) setrlimit(RLIMIT_NOFILE, &rl); 3311 (void) enable_extended_FILE_stdio(-1, -1); 3312 ztest_run(zopt_pool); 3313 exit(0); 3314 } 3315 3316 while (waitpid(pid, &status, WEXITED) != pid) 3317 continue; 3318 3319 if (WIFEXITED(status)) { 3320 if (WEXITSTATUS(status) != 0) { 3321 (void) fprintf(stderr, 3322 "child exited with code %d\n", 3323 WEXITSTATUS(status)); 3324 exit(2); 3325 } 3326 } else { 3327 if (WTERMSIG(status) != SIGKILL) { 3328 (void) fprintf(stderr, 3329 "child died with signal %d\n", 3330 WTERMSIG(status)); 3331 exit(3); 3332 } 3333 kills++; 3334 } 3335 3336 iters++; 3337 3338 if (zopt_verbose >= 1) { 3339 hrtime_t now = gethrtime(); 3340 3341 now = MIN(now, zs->zs_stop_time); 3342 print_time(zs->zs_stop_time - now, timebuf); 3343 nicenum(zs->zs_space, numbuf); 3344 3345 (void) printf("Pass %3d, %8s, %3llu ENOSPC, " 3346 "%4.1f%% of %5s used, %3.0f%% done, %8s to go\n", 3347 iters, 3348 WIFEXITED(status) ? "Complete" : "SIGKILL", 3349 (u_longlong_t)zs->zs_enospc_count, 3350 100.0 * zs->zs_alloc / zs->zs_space, 3351 numbuf, 3352 100.0 * (now - zs->zs_start_time) / 3353 (zopt_time * NANOSEC), timebuf); 3354 } 3355 3356 if (zopt_verbose >= 2) { 3357 (void) printf("\nWorkload summary:\n\n"); 3358 (void) printf("%7s %9s %s\n", 3359 "Calls", "Time", "Function"); 3360 (void) printf("%7s %9s %s\n", 3361 "-----", "----", "--------"); 3362 for (f = 0; f < ZTEST_FUNCS; f++) { 3363 Dl_info dli; 3364 3365 zi = &zs->zs_info[f]; 3366 print_time(zi->zi_call_time, timebuf); 3367 (void) dladdr((void *)zi->zi_func, &dli); 3368 (void) printf("%7llu %9s %s\n", 3369 (u_longlong_t)zi->zi_calls, timebuf, 3370 dli.dli_sname); 3371 } 3372 (void) printf("\n"); 3373 } 3374 3375 /* 3376 * It's possible that we killed a child during a rename test, in 3377 * which case we'll have a 'ztest_tmp' pool lying around instead 3378 * of 'ztest'. Do a blind rename in case this happened. 3379 */ 3380 tmp = umem_alloc(strlen(zopt_pool) + 5, UMEM_NOFAIL); 3381 (void) strcpy(tmp, zopt_pool); 3382 (void) strcat(tmp, "_tmp"); 3383 kernel_init(FREAD | FWRITE); 3384 (void) spa_rename(tmp, zopt_pool); 3385 kernel_fini(); 3386 umem_free(tmp, strlen(tmp) + 1); 3387 } 3388 3389 ztest_verify_blocks(zopt_pool); 3390 3391 if (zopt_verbose >= 1) { 3392 (void) printf("%d killed, %d completed, %.0f%% kill rate\n", 3393 kills, iters - kills, (100.0 * kills) / MAX(1, iters)); 3394 } 3395 3396 return (0); 3397 } 3398